xref: /openbmc/linux/fs/btrfs/inode.c (revision 5d0e4d78)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63 
64 struct btrfs_iget_args {
65 	struct btrfs_key *location;
66 	struct btrfs_root *root;
67 };
68 
69 struct btrfs_dio_data {
70 	u64 outstanding_extents;
71 	u64 reserve;
72 	u64 unsubmitted_oe_range_start;
73 	u64 unsubmitted_oe_range_end;
74 	int overwrite;
75 };
76 
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86 
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
91 
92 #define S_SHIFT 12
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
95 	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
96 	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
97 	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
98 	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
99 	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
100 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
101 };
102 
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 				   struct page *locked_page,
108 				   u64 start, u64 end, u64 delalloc_end,
109 				   int *page_started, unsigned long *nr_written,
110 				   int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 				       u64 orig_start, u64 block_start,
113 				       u64 block_len, u64 orig_block_len,
114 				       u64 ram_bytes, int compress_type,
115 				       int type);
116 
117 static void __endio_write_update_ordered(struct inode *inode,
118 					 const u64 offset, const u64 bytes,
119 					 const bool uptodate);
120 
121 /*
122  * Cleanup all submitted ordered extents in specified range to handle errors
123  * from the fill_dellaloc() callback.
124  *
125  * NOTE: caller must ensure that when an error happens, it can not call
126  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128  * to be released, which we want to happen only when finishing the ordered
129  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130  * fill_delalloc() callback already does proper cleanup for the first page of
131  * the range, that is, it invokes the callback writepage_end_io_hook() for the
132  * range of the first page.
133  */
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
135 						 const u64 offset,
136 						 const u64 bytes)
137 {
138 	return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 					    bytes - PAGE_SIZE, false);
140 }
141 
142 static int btrfs_dirty_inode(struct inode *inode);
143 
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
146 {
147 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
148 }
149 #endif
150 
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 				     struct inode *inode,  struct inode *dir,
153 				     const struct qstr *qstr)
154 {
155 	int err;
156 
157 	err = btrfs_init_acl(trans, inode, dir);
158 	if (!err)
159 		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160 	return err;
161 }
162 
163 /*
164  * this does all the hard work for inserting an inline extent into
165  * the btree.  The caller should have done a btrfs_drop_extents so that
166  * no overlapping inline items exist in the btree
167  */
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 				struct btrfs_path *path, int extent_inserted,
170 				struct btrfs_root *root, struct inode *inode,
171 				u64 start, size_t size, size_t compressed_size,
172 				int compress_type,
173 				struct page **compressed_pages)
174 {
175 	struct extent_buffer *leaf;
176 	struct page *page = NULL;
177 	char *kaddr;
178 	unsigned long ptr;
179 	struct btrfs_file_extent_item *ei;
180 	int ret;
181 	size_t cur_size = size;
182 	unsigned long offset;
183 
184 	if (compressed_size && compressed_pages)
185 		cur_size = compressed_size;
186 
187 	inode_add_bytes(inode, size);
188 
189 	if (!extent_inserted) {
190 		struct btrfs_key key;
191 		size_t datasize;
192 
193 		key.objectid = btrfs_ino(BTRFS_I(inode));
194 		key.offset = start;
195 		key.type = BTRFS_EXTENT_DATA_KEY;
196 
197 		datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 		path->leave_spinning = 1;
199 		ret = btrfs_insert_empty_item(trans, root, path, &key,
200 					      datasize);
201 		if (ret)
202 			goto fail;
203 	}
204 	leaf = path->nodes[0];
205 	ei = btrfs_item_ptr(leaf, path->slots[0],
206 			    struct btrfs_file_extent_item);
207 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 	btrfs_set_file_extent_encryption(leaf, ei, 0);
210 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 	ptr = btrfs_file_extent_inline_start(ei);
213 
214 	if (compress_type != BTRFS_COMPRESS_NONE) {
215 		struct page *cpage;
216 		int i = 0;
217 		while (compressed_size > 0) {
218 			cpage = compressed_pages[i];
219 			cur_size = min_t(unsigned long, compressed_size,
220 				       PAGE_SIZE);
221 
222 			kaddr = kmap_atomic(cpage);
223 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 			kunmap_atomic(kaddr);
225 
226 			i++;
227 			ptr += cur_size;
228 			compressed_size -= cur_size;
229 		}
230 		btrfs_set_file_extent_compression(leaf, ei,
231 						  compress_type);
232 	} else {
233 		page = find_get_page(inode->i_mapping,
234 				     start >> PAGE_SHIFT);
235 		btrfs_set_file_extent_compression(leaf, ei, 0);
236 		kaddr = kmap_atomic(page);
237 		offset = start & (PAGE_SIZE - 1);
238 		write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 		kunmap_atomic(kaddr);
240 		put_page(page);
241 	}
242 	btrfs_mark_buffer_dirty(leaf);
243 	btrfs_release_path(path);
244 
245 	/*
246 	 * we're an inline extent, so nobody can
247 	 * extend the file past i_size without locking
248 	 * a page we already have locked.
249 	 *
250 	 * We must do any isize and inode updates
251 	 * before we unlock the pages.  Otherwise we
252 	 * could end up racing with unlink.
253 	 */
254 	BTRFS_I(inode)->disk_i_size = inode->i_size;
255 	ret = btrfs_update_inode(trans, root, inode);
256 
257 fail:
258 	return ret;
259 }
260 
261 
262 /*
263  * conditionally insert an inline extent into the file.  This
264  * does the checks required to make sure the data is small enough
265  * to fit as an inline extent.
266  */
267 static noinline int cow_file_range_inline(struct btrfs_root *root,
268 					  struct inode *inode, u64 start,
269 					  u64 end, size_t compressed_size,
270 					  int compress_type,
271 					  struct page **compressed_pages)
272 {
273 	struct btrfs_fs_info *fs_info = root->fs_info;
274 	struct btrfs_trans_handle *trans;
275 	u64 isize = i_size_read(inode);
276 	u64 actual_end = min(end + 1, isize);
277 	u64 inline_len = actual_end - start;
278 	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 	u64 data_len = inline_len;
280 	int ret;
281 	struct btrfs_path *path;
282 	int extent_inserted = 0;
283 	u32 extent_item_size;
284 
285 	if (compressed_size)
286 		data_len = compressed_size;
287 
288 	if (start > 0 ||
289 	    actual_end > fs_info->sectorsize ||
290 	    data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
291 	    (!compressed_size &&
292 	    (actual_end & (fs_info->sectorsize - 1)) == 0) ||
293 	    end + 1 < isize ||
294 	    data_len > fs_info->max_inline) {
295 		return 1;
296 	}
297 
298 	path = btrfs_alloc_path();
299 	if (!path)
300 		return -ENOMEM;
301 
302 	trans = btrfs_join_transaction(root);
303 	if (IS_ERR(trans)) {
304 		btrfs_free_path(path);
305 		return PTR_ERR(trans);
306 	}
307 	trans->block_rsv = &fs_info->delalloc_block_rsv;
308 
309 	if (compressed_size && compressed_pages)
310 		extent_item_size = btrfs_file_extent_calc_inline_size(
311 		   compressed_size);
312 	else
313 		extent_item_size = btrfs_file_extent_calc_inline_size(
314 		    inline_len);
315 
316 	ret = __btrfs_drop_extents(trans, root, inode, path,
317 				   start, aligned_end, NULL,
318 				   1, 1, extent_item_size, &extent_inserted);
319 	if (ret) {
320 		btrfs_abort_transaction(trans, ret);
321 		goto out;
322 	}
323 
324 	if (isize > actual_end)
325 		inline_len = min_t(u64, isize, actual_end);
326 	ret = insert_inline_extent(trans, path, extent_inserted,
327 				   root, inode, start,
328 				   inline_len, compressed_size,
329 				   compress_type, compressed_pages);
330 	if (ret && ret != -ENOSPC) {
331 		btrfs_abort_transaction(trans, ret);
332 		goto out;
333 	} else if (ret == -ENOSPC) {
334 		ret = 1;
335 		goto out;
336 	}
337 
338 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 	btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
340 	btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
341 out:
342 	/*
343 	 * Don't forget to free the reserved space, as for inlined extent
344 	 * it won't count as data extent, free them directly here.
345 	 * And at reserve time, it's always aligned to page size, so
346 	 * just free one page here.
347 	 */
348 	btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
349 	btrfs_free_path(path);
350 	btrfs_end_transaction(trans);
351 	return ret;
352 }
353 
354 struct async_extent {
355 	u64 start;
356 	u64 ram_size;
357 	u64 compressed_size;
358 	struct page **pages;
359 	unsigned long nr_pages;
360 	int compress_type;
361 	struct list_head list;
362 };
363 
364 struct async_cow {
365 	struct inode *inode;
366 	struct btrfs_root *root;
367 	struct page *locked_page;
368 	u64 start;
369 	u64 end;
370 	struct list_head extents;
371 	struct btrfs_work work;
372 };
373 
374 static noinline int add_async_extent(struct async_cow *cow,
375 				     u64 start, u64 ram_size,
376 				     u64 compressed_size,
377 				     struct page **pages,
378 				     unsigned long nr_pages,
379 				     int compress_type)
380 {
381 	struct async_extent *async_extent;
382 
383 	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 	BUG_ON(!async_extent); /* -ENOMEM */
385 	async_extent->start = start;
386 	async_extent->ram_size = ram_size;
387 	async_extent->compressed_size = compressed_size;
388 	async_extent->pages = pages;
389 	async_extent->nr_pages = nr_pages;
390 	async_extent->compress_type = compress_type;
391 	list_add_tail(&async_extent->list, &cow->extents);
392 	return 0;
393 }
394 
395 static inline int inode_need_compress(struct inode *inode)
396 {
397 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
398 
399 	/* force compress */
400 	if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
401 		return 1;
402 	/* bad compression ratios */
403 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
404 		return 0;
405 	if (btrfs_test_opt(fs_info, COMPRESS) ||
406 	    BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
407 	    BTRFS_I(inode)->force_compress)
408 		return 1;
409 	return 0;
410 }
411 
412 static inline void inode_should_defrag(struct btrfs_inode *inode,
413 		u64 start, u64 end, u64 num_bytes, u64 small_write)
414 {
415 	/* If this is a small write inside eof, kick off a defrag */
416 	if (num_bytes < small_write &&
417 	    (start > 0 || end + 1 < inode->disk_i_size))
418 		btrfs_add_inode_defrag(NULL, inode);
419 }
420 
421 /*
422  * we create compressed extents in two phases.  The first
423  * phase compresses a range of pages that have already been
424  * locked (both pages and state bits are locked).
425  *
426  * This is done inside an ordered work queue, and the compression
427  * is spread across many cpus.  The actual IO submission is step
428  * two, and the ordered work queue takes care of making sure that
429  * happens in the same order things were put onto the queue by
430  * writepages and friends.
431  *
432  * If this code finds it can't get good compression, it puts an
433  * entry onto the work queue to write the uncompressed bytes.  This
434  * makes sure that both compressed inodes and uncompressed inodes
435  * are written in the same order that the flusher thread sent them
436  * down.
437  */
438 static noinline void compress_file_range(struct inode *inode,
439 					struct page *locked_page,
440 					u64 start, u64 end,
441 					struct async_cow *async_cow,
442 					int *num_added)
443 {
444 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
445 	struct btrfs_root *root = BTRFS_I(inode)->root;
446 	u64 num_bytes;
447 	u64 blocksize = fs_info->sectorsize;
448 	u64 actual_end;
449 	u64 isize = i_size_read(inode);
450 	int ret = 0;
451 	struct page **pages = NULL;
452 	unsigned long nr_pages;
453 	unsigned long total_compressed = 0;
454 	unsigned long total_in = 0;
455 	int i;
456 	int will_compress;
457 	int compress_type = fs_info->compress_type;
458 	int redirty = 0;
459 
460 	inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
461 			SZ_16K);
462 
463 	actual_end = min_t(u64, isize, end + 1);
464 again:
465 	will_compress = 0;
466 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
467 	BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
468 	nr_pages = min_t(unsigned long, nr_pages,
469 			BTRFS_MAX_COMPRESSED / PAGE_SIZE);
470 
471 	/*
472 	 * we don't want to send crud past the end of i_size through
473 	 * compression, that's just a waste of CPU time.  So, if the
474 	 * end of the file is before the start of our current
475 	 * requested range of bytes, we bail out to the uncompressed
476 	 * cleanup code that can deal with all of this.
477 	 *
478 	 * It isn't really the fastest way to fix things, but this is a
479 	 * very uncommon corner.
480 	 */
481 	if (actual_end <= start)
482 		goto cleanup_and_bail_uncompressed;
483 
484 	total_compressed = actual_end - start;
485 
486 	/*
487 	 * skip compression for a small file range(<=blocksize) that
488 	 * isn't an inline extent, since it doesn't save disk space at all.
489 	 */
490 	if (total_compressed <= blocksize &&
491 	   (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
492 		goto cleanup_and_bail_uncompressed;
493 
494 	total_compressed = min_t(unsigned long, total_compressed,
495 			BTRFS_MAX_UNCOMPRESSED);
496 	num_bytes = ALIGN(end - start + 1, blocksize);
497 	num_bytes = max(blocksize,  num_bytes);
498 	total_in = 0;
499 	ret = 0;
500 
501 	/*
502 	 * we do compression for mount -o compress and when the
503 	 * inode has not been flagged as nocompress.  This flag can
504 	 * change at any time if we discover bad compression ratios.
505 	 */
506 	if (inode_need_compress(inode)) {
507 		WARN_ON(pages);
508 		pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
509 		if (!pages) {
510 			/* just bail out to the uncompressed code */
511 			goto cont;
512 		}
513 
514 		if (BTRFS_I(inode)->force_compress)
515 			compress_type = BTRFS_I(inode)->force_compress;
516 
517 		/*
518 		 * we need to call clear_page_dirty_for_io on each
519 		 * page in the range.  Otherwise applications with the file
520 		 * mmap'd can wander in and change the page contents while
521 		 * we are compressing them.
522 		 *
523 		 * If the compression fails for any reason, we set the pages
524 		 * dirty again later on.
525 		 */
526 		extent_range_clear_dirty_for_io(inode, start, end);
527 		redirty = 1;
528 		ret = btrfs_compress_pages(compress_type,
529 					   inode->i_mapping, start,
530 					   pages,
531 					   &nr_pages,
532 					   &total_in,
533 					   &total_compressed);
534 
535 		if (!ret) {
536 			unsigned long offset = total_compressed &
537 				(PAGE_SIZE - 1);
538 			struct page *page = pages[nr_pages - 1];
539 			char *kaddr;
540 
541 			/* zero the tail end of the last page, we might be
542 			 * sending it down to disk
543 			 */
544 			if (offset) {
545 				kaddr = kmap_atomic(page);
546 				memset(kaddr + offset, 0,
547 				       PAGE_SIZE - offset);
548 				kunmap_atomic(kaddr);
549 			}
550 			will_compress = 1;
551 		}
552 	}
553 cont:
554 	if (start == 0) {
555 		/* lets try to make an inline extent */
556 		if (ret || total_in < (actual_end - start)) {
557 			/* we didn't compress the entire range, try
558 			 * to make an uncompressed inline extent.
559 			 */
560 			ret = cow_file_range_inline(root, inode, start, end,
561 					    0, BTRFS_COMPRESS_NONE, NULL);
562 		} else {
563 			/* try making a compressed inline extent */
564 			ret = cow_file_range_inline(root, inode, start, end,
565 						    total_compressed,
566 						    compress_type, pages);
567 		}
568 		if (ret <= 0) {
569 			unsigned long clear_flags = EXTENT_DELALLOC |
570 				EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
571 			unsigned long page_error_op;
572 
573 			clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
574 			page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
575 
576 			/*
577 			 * inline extent creation worked or returned error,
578 			 * we don't need to create any more async work items.
579 			 * Unlock and free up our temp pages.
580 			 */
581 			extent_clear_unlock_delalloc(inode, start, end, end,
582 						     NULL, clear_flags,
583 						     PAGE_UNLOCK |
584 						     PAGE_CLEAR_DIRTY |
585 						     PAGE_SET_WRITEBACK |
586 						     page_error_op |
587 						     PAGE_END_WRITEBACK);
588 			if (ret == 0)
589 				btrfs_free_reserved_data_space_noquota(inode,
590 							       start,
591 							       end - start + 1);
592 			goto free_pages_out;
593 		}
594 	}
595 
596 	if (will_compress) {
597 		/*
598 		 * we aren't doing an inline extent round the compressed size
599 		 * up to a block size boundary so the allocator does sane
600 		 * things
601 		 */
602 		total_compressed = ALIGN(total_compressed, blocksize);
603 
604 		/*
605 		 * one last check to make sure the compression is really a
606 		 * win, compare the page count read with the blocks on disk,
607 		 * compression must free at least one sector size
608 		 */
609 		total_in = ALIGN(total_in, PAGE_SIZE);
610 		if (total_compressed + blocksize <= total_in) {
611 			num_bytes = total_in;
612 			*num_added += 1;
613 
614 			/*
615 			 * The async work queues will take care of doing actual
616 			 * allocation on disk for these compressed pages, and
617 			 * will submit them to the elevator.
618 			 */
619 			add_async_extent(async_cow, start, num_bytes,
620 					total_compressed, pages, nr_pages,
621 					compress_type);
622 
623 			if (start + num_bytes < end) {
624 				start += num_bytes;
625 				pages = NULL;
626 				cond_resched();
627 				goto again;
628 			}
629 			return;
630 		}
631 	}
632 	if (pages) {
633 		/*
634 		 * the compression code ran but failed to make things smaller,
635 		 * free any pages it allocated and our page pointer array
636 		 */
637 		for (i = 0; i < nr_pages; i++) {
638 			WARN_ON(pages[i]->mapping);
639 			put_page(pages[i]);
640 		}
641 		kfree(pages);
642 		pages = NULL;
643 		total_compressed = 0;
644 		nr_pages = 0;
645 
646 		/* flag the file so we don't compress in the future */
647 		if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
648 		    !(BTRFS_I(inode)->force_compress)) {
649 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
650 		}
651 	}
652 cleanup_and_bail_uncompressed:
653 	/*
654 	 * No compression, but we still need to write the pages in the file
655 	 * we've been given so far.  redirty the locked page if it corresponds
656 	 * to our extent and set things up for the async work queue to run
657 	 * cow_file_range to do the normal delalloc dance.
658 	 */
659 	if (page_offset(locked_page) >= start &&
660 	    page_offset(locked_page) <= end)
661 		__set_page_dirty_nobuffers(locked_page);
662 		/* unlocked later on in the async handlers */
663 
664 	if (redirty)
665 		extent_range_redirty_for_io(inode, start, end);
666 	add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
667 			 BTRFS_COMPRESS_NONE);
668 	*num_added += 1;
669 
670 	return;
671 
672 free_pages_out:
673 	for (i = 0; i < nr_pages; i++) {
674 		WARN_ON(pages[i]->mapping);
675 		put_page(pages[i]);
676 	}
677 	kfree(pages);
678 }
679 
680 static void free_async_extent_pages(struct async_extent *async_extent)
681 {
682 	int i;
683 
684 	if (!async_extent->pages)
685 		return;
686 
687 	for (i = 0; i < async_extent->nr_pages; i++) {
688 		WARN_ON(async_extent->pages[i]->mapping);
689 		put_page(async_extent->pages[i]);
690 	}
691 	kfree(async_extent->pages);
692 	async_extent->nr_pages = 0;
693 	async_extent->pages = NULL;
694 }
695 
696 /*
697  * phase two of compressed writeback.  This is the ordered portion
698  * of the code, which only gets called in the order the work was
699  * queued.  We walk all the async extents created by compress_file_range
700  * and send them down to the disk.
701  */
702 static noinline void submit_compressed_extents(struct inode *inode,
703 					      struct async_cow *async_cow)
704 {
705 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
706 	struct async_extent *async_extent;
707 	u64 alloc_hint = 0;
708 	struct btrfs_key ins;
709 	struct extent_map *em;
710 	struct btrfs_root *root = BTRFS_I(inode)->root;
711 	struct extent_io_tree *io_tree;
712 	int ret = 0;
713 
714 again:
715 	while (!list_empty(&async_cow->extents)) {
716 		async_extent = list_entry(async_cow->extents.next,
717 					  struct async_extent, list);
718 		list_del(&async_extent->list);
719 
720 		io_tree = &BTRFS_I(inode)->io_tree;
721 
722 retry:
723 		/* did the compression code fall back to uncompressed IO? */
724 		if (!async_extent->pages) {
725 			int page_started = 0;
726 			unsigned long nr_written = 0;
727 
728 			lock_extent(io_tree, async_extent->start,
729 					 async_extent->start +
730 					 async_extent->ram_size - 1);
731 
732 			/* allocate blocks */
733 			ret = cow_file_range(inode, async_cow->locked_page,
734 					     async_extent->start,
735 					     async_extent->start +
736 					     async_extent->ram_size - 1,
737 					     async_extent->start +
738 					     async_extent->ram_size - 1,
739 					     &page_started, &nr_written, 0,
740 					     NULL);
741 
742 			/* JDM XXX */
743 
744 			/*
745 			 * if page_started, cow_file_range inserted an
746 			 * inline extent and took care of all the unlocking
747 			 * and IO for us.  Otherwise, we need to submit
748 			 * all those pages down to the drive.
749 			 */
750 			if (!page_started && !ret)
751 				extent_write_locked_range(io_tree,
752 						  inode, async_extent->start,
753 						  async_extent->start +
754 						  async_extent->ram_size - 1,
755 						  btrfs_get_extent,
756 						  WB_SYNC_ALL);
757 			else if (ret)
758 				unlock_page(async_cow->locked_page);
759 			kfree(async_extent);
760 			cond_resched();
761 			continue;
762 		}
763 
764 		lock_extent(io_tree, async_extent->start,
765 			    async_extent->start + async_extent->ram_size - 1);
766 
767 		ret = btrfs_reserve_extent(root, async_extent->ram_size,
768 					   async_extent->compressed_size,
769 					   async_extent->compressed_size,
770 					   0, alloc_hint, &ins, 1, 1);
771 		if (ret) {
772 			free_async_extent_pages(async_extent);
773 
774 			if (ret == -ENOSPC) {
775 				unlock_extent(io_tree, async_extent->start,
776 					      async_extent->start +
777 					      async_extent->ram_size - 1);
778 
779 				/*
780 				 * we need to redirty the pages if we decide to
781 				 * fallback to uncompressed IO, otherwise we
782 				 * will not submit these pages down to lower
783 				 * layers.
784 				 */
785 				extent_range_redirty_for_io(inode,
786 						async_extent->start,
787 						async_extent->start +
788 						async_extent->ram_size - 1);
789 
790 				goto retry;
791 			}
792 			goto out_free;
793 		}
794 		/*
795 		 * here we're doing allocation and writeback of the
796 		 * compressed pages
797 		 */
798 		em = create_io_em(inode, async_extent->start,
799 				  async_extent->ram_size, /* len */
800 				  async_extent->start, /* orig_start */
801 				  ins.objectid, /* block_start */
802 				  ins.offset, /* block_len */
803 				  ins.offset, /* orig_block_len */
804 				  async_extent->ram_size, /* ram_bytes */
805 				  async_extent->compress_type,
806 				  BTRFS_ORDERED_COMPRESSED);
807 		if (IS_ERR(em))
808 			/* ret value is not necessary due to void function */
809 			goto out_free_reserve;
810 		free_extent_map(em);
811 
812 		ret = btrfs_add_ordered_extent_compress(inode,
813 						async_extent->start,
814 						ins.objectid,
815 						async_extent->ram_size,
816 						ins.offset,
817 						BTRFS_ORDERED_COMPRESSED,
818 						async_extent->compress_type);
819 		if (ret) {
820 			btrfs_drop_extent_cache(BTRFS_I(inode),
821 						async_extent->start,
822 						async_extent->start +
823 						async_extent->ram_size - 1, 0);
824 			goto out_free_reserve;
825 		}
826 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
827 
828 		/*
829 		 * clear dirty, set writeback and unlock the pages.
830 		 */
831 		extent_clear_unlock_delalloc(inode, async_extent->start,
832 				async_extent->start +
833 				async_extent->ram_size - 1,
834 				async_extent->start +
835 				async_extent->ram_size - 1,
836 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
837 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
838 				PAGE_SET_WRITEBACK);
839 		if (btrfs_submit_compressed_write(inode,
840 				    async_extent->start,
841 				    async_extent->ram_size,
842 				    ins.objectid,
843 				    ins.offset, async_extent->pages,
844 				    async_extent->nr_pages)) {
845 			struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 			struct page *p = async_extent->pages[0];
847 			const u64 start = async_extent->start;
848 			const u64 end = start + async_extent->ram_size - 1;
849 
850 			p->mapping = inode->i_mapping;
851 			tree->ops->writepage_end_io_hook(p, start, end,
852 							 NULL, 0);
853 			p->mapping = NULL;
854 			extent_clear_unlock_delalloc(inode, start, end, end,
855 						     NULL, 0,
856 						     PAGE_END_WRITEBACK |
857 						     PAGE_SET_ERROR);
858 			free_async_extent_pages(async_extent);
859 		}
860 		alloc_hint = ins.objectid + ins.offset;
861 		kfree(async_extent);
862 		cond_resched();
863 	}
864 	return;
865 out_free_reserve:
866 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
867 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
868 out_free:
869 	extent_clear_unlock_delalloc(inode, async_extent->start,
870 				     async_extent->start +
871 				     async_extent->ram_size - 1,
872 				     async_extent->start +
873 				     async_extent->ram_size - 1,
874 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
875 				     EXTENT_DELALLOC_NEW |
876 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
877 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
878 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
879 				     PAGE_SET_ERROR);
880 	free_async_extent_pages(async_extent);
881 	kfree(async_extent);
882 	goto again;
883 }
884 
885 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
886 				      u64 num_bytes)
887 {
888 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
889 	struct extent_map *em;
890 	u64 alloc_hint = 0;
891 
892 	read_lock(&em_tree->lock);
893 	em = search_extent_mapping(em_tree, start, num_bytes);
894 	if (em) {
895 		/*
896 		 * if block start isn't an actual block number then find the
897 		 * first block in this inode and use that as a hint.  If that
898 		 * block is also bogus then just don't worry about it.
899 		 */
900 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
901 			free_extent_map(em);
902 			em = search_extent_mapping(em_tree, 0, 0);
903 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
904 				alloc_hint = em->block_start;
905 			if (em)
906 				free_extent_map(em);
907 		} else {
908 			alloc_hint = em->block_start;
909 			free_extent_map(em);
910 		}
911 	}
912 	read_unlock(&em_tree->lock);
913 
914 	return alloc_hint;
915 }
916 
917 /*
918  * when extent_io.c finds a delayed allocation range in the file,
919  * the call backs end up in this code.  The basic idea is to
920  * allocate extents on disk for the range, and create ordered data structs
921  * in ram to track those extents.
922  *
923  * locked_page is the page that writepage had locked already.  We use
924  * it to make sure we don't do extra locks or unlocks.
925  *
926  * *page_started is set to one if we unlock locked_page and do everything
927  * required to start IO on it.  It may be clean and already done with
928  * IO when we return.
929  */
930 static noinline int cow_file_range(struct inode *inode,
931 				   struct page *locked_page,
932 				   u64 start, u64 end, u64 delalloc_end,
933 				   int *page_started, unsigned long *nr_written,
934 				   int unlock, struct btrfs_dedupe_hash *hash)
935 {
936 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
937 	struct btrfs_root *root = BTRFS_I(inode)->root;
938 	u64 alloc_hint = 0;
939 	u64 num_bytes;
940 	unsigned long ram_size;
941 	u64 disk_num_bytes;
942 	u64 cur_alloc_size = 0;
943 	u64 blocksize = fs_info->sectorsize;
944 	struct btrfs_key ins;
945 	struct extent_map *em;
946 	unsigned clear_bits;
947 	unsigned long page_ops;
948 	bool extent_reserved = false;
949 	int ret = 0;
950 
951 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
952 		WARN_ON_ONCE(1);
953 		ret = -EINVAL;
954 		goto out_unlock;
955 	}
956 
957 	num_bytes = ALIGN(end - start + 1, blocksize);
958 	num_bytes = max(blocksize,  num_bytes);
959 	disk_num_bytes = num_bytes;
960 
961 	inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
962 
963 	if (start == 0) {
964 		/* lets try to make an inline extent */
965 		ret = cow_file_range_inline(root, inode, start, end, 0,
966 					BTRFS_COMPRESS_NONE, NULL);
967 		if (ret == 0) {
968 			extent_clear_unlock_delalloc(inode, start, end,
969 				     delalloc_end, NULL,
970 				     EXTENT_LOCKED | EXTENT_DELALLOC |
971 				     EXTENT_DELALLOC_NEW |
972 				     EXTENT_DEFRAG, PAGE_UNLOCK |
973 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
974 				     PAGE_END_WRITEBACK);
975 			btrfs_free_reserved_data_space_noquota(inode, start,
976 						end - start + 1);
977 			*nr_written = *nr_written +
978 			     (end - start + PAGE_SIZE) / PAGE_SIZE;
979 			*page_started = 1;
980 			goto out;
981 		} else if (ret < 0) {
982 			goto out_unlock;
983 		}
984 	}
985 
986 	BUG_ON(disk_num_bytes >
987 	       btrfs_super_total_bytes(fs_info->super_copy));
988 
989 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
990 	btrfs_drop_extent_cache(BTRFS_I(inode), start,
991 			start + num_bytes - 1, 0);
992 
993 	while (disk_num_bytes > 0) {
994 		cur_alloc_size = disk_num_bytes;
995 		ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
996 					   fs_info->sectorsize, 0, alloc_hint,
997 					   &ins, 1, 1);
998 		if (ret < 0)
999 			goto out_unlock;
1000 		cur_alloc_size = ins.offset;
1001 		extent_reserved = true;
1002 
1003 		ram_size = ins.offset;
1004 		em = create_io_em(inode, start, ins.offset, /* len */
1005 				  start, /* orig_start */
1006 				  ins.objectid, /* block_start */
1007 				  ins.offset, /* block_len */
1008 				  ins.offset, /* orig_block_len */
1009 				  ram_size, /* ram_bytes */
1010 				  BTRFS_COMPRESS_NONE, /* compress_type */
1011 				  BTRFS_ORDERED_REGULAR /* type */);
1012 		if (IS_ERR(em))
1013 			goto out_reserve;
1014 		free_extent_map(em);
1015 
1016 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1017 					       ram_size, cur_alloc_size, 0);
1018 		if (ret)
1019 			goto out_drop_extent_cache;
1020 
1021 		if (root->root_key.objectid ==
1022 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1023 			ret = btrfs_reloc_clone_csums(inode, start,
1024 						      cur_alloc_size);
1025 			/*
1026 			 * Only drop cache here, and process as normal.
1027 			 *
1028 			 * We must not allow extent_clear_unlock_delalloc()
1029 			 * at out_unlock label to free meta of this ordered
1030 			 * extent, as its meta should be freed by
1031 			 * btrfs_finish_ordered_io().
1032 			 *
1033 			 * So we must continue until @start is increased to
1034 			 * skip current ordered extent.
1035 			 */
1036 			if (ret)
1037 				btrfs_drop_extent_cache(BTRFS_I(inode), start,
1038 						start + ram_size - 1, 0);
1039 		}
1040 
1041 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1042 
1043 		/* we're not doing compressed IO, don't unlock the first
1044 		 * page (which the caller expects to stay locked), don't
1045 		 * clear any dirty bits and don't set any writeback bits
1046 		 *
1047 		 * Do set the Private2 bit so we know this page was properly
1048 		 * setup for writepage
1049 		 */
1050 		page_ops = unlock ? PAGE_UNLOCK : 0;
1051 		page_ops |= PAGE_SET_PRIVATE2;
1052 
1053 		extent_clear_unlock_delalloc(inode, start,
1054 					     start + ram_size - 1,
1055 					     delalloc_end, locked_page,
1056 					     EXTENT_LOCKED | EXTENT_DELALLOC,
1057 					     page_ops);
1058 		if (disk_num_bytes < cur_alloc_size)
1059 			disk_num_bytes = 0;
1060 		else
1061 			disk_num_bytes -= cur_alloc_size;
1062 		num_bytes -= cur_alloc_size;
1063 		alloc_hint = ins.objectid + ins.offset;
1064 		start += cur_alloc_size;
1065 		extent_reserved = false;
1066 
1067 		/*
1068 		 * btrfs_reloc_clone_csums() error, since start is increased
1069 		 * extent_clear_unlock_delalloc() at out_unlock label won't
1070 		 * free metadata of current ordered extent, we're OK to exit.
1071 		 */
1072 		if (ret)
1073 			goto out_unlock;
1074 	}
1075 out:
1076 	return ret;
1077 
1078 out_drop_extent_cache:
1079 	btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1080 out_reserve:
1081 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1082 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1083 out_unlock:
1084 	clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1085 		EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1086 	page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1087 		PAGE_END_WRITEBACK;
1088 	/*
1089 	 * If we reserved an extent for our delalloc range (or a subrange) and
1090 	 * failed to create the respective ordered extent, then it means that
1091 	 * when we reserved the extent we decremented the extent's size from
1092 	 * the data space_info's bytes_may_use counter and incremented the
1093 	 * space_info's bytes_reserved counter by the same amount. We must make
1094 	 * sure extent_clear_unlock_delalloc() does not try to decrement again
1095 	 * the data space_info's bytes_may_use counter, therefore we do not pass
1096 	 * it the flag EXTENT_CLEAR_DATA_RESV.
1097 	 */
1098 	if (extent_reserved) {
1099 		extent_clear_unlock_delalloc(inode, start,
1100 					     start + cur_alloc_size,
1101 					     start + cur_alloc_size,
1102 					     locked_page,
1103 					     clear_bits,
1104 					     page_ops);
1105 		start += cur_alloc_size;
1106 		if (start >= end)
1107 			goto out;
1108 	}
1109 	extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1110 				     locked_page,
1111 				     clear_bits | EXTENT_CLEAR_DATA_RESV,
1112 				     page_ops);
1113 	goto out;
1114 }
1115 
1116 /*
1117  * work queue call back to started compression on a file and pages
1118  */
1119 static noinline void async_cow_start(struct btrfs_work *work)
1120 {
1121 	struct async_cow *async_cow;
1122 	int num_added = 0;
1123 	async_cow = container_of(work, struct async_cow, work);
1124 
1125 	compress_file_range(async_cow->inode, async_cow->locked_page,
1126 			    async_cow->start, async_cow->end, async_cow,
1127 			    &num_added);
1128 	if (num_added == 0) {
1129 		btrfs_add_delayed_iput(async_cow->inode);
1130 		async_cow->inode = NULL;
1131 	}
1132 }
1133 
1134 /*
1135  * work queue call back to submit previously compressed pages
1136  */
1137 static noinline void async_cow_submit(struct btrfs_work *work)
1138 {
1139 	struct btrfs_fs_info *fs_info;
1140 	struct async_cow *async_cow;
1141 	struct btrfs_root *root;
1142 	unsigned long nr_pages;
1143 
1144 	async_cow = container_of(work, struct async_cow, work);
1145 
1146 	root = async_cow->root;
1147 	fs_info = root->fs_info;
1148 	nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1149 		PAGE_SHIFT;
1150 
1151 	/*
1152 	 * atomic_sub_return implies a barrier for waitqueue_active
1153 	 */
1154 	if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1155 	    5 * SZ_1M &&
1156 	    waitqueue_active(&fs_info->async_submit_wait))
1157 		wake_up(&fs_info->async_submit_wait);
1158 
1159 	if (async_cow->inode)
1160 		submit_compressed_extents(async_cow->inode, async_cow);
1161 }
1162 
1163 static noinline void async_cow_free(struct btrfs_work *work)
1164 {
1165 	struct async_cow *async_cow;
1166 	async_cow = container_of(work, struct async_cow, work);
1167 	if (async_cow->inode)
1168 		btrfs_add_delayed_iput(async_cow->inode);
1169 	kfree(async_cow);
1170 }
1171 
1172 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1173 				u64 start, u64 end, int *page_started,
1174 				unsigned long *nr_written)
1175 {
1176 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1177 	struct async_cow *async_cow;
1178 	struct btrfs_root *root = BTRFS_I(inode)->root;
1179 	unsigned long nr_pages;
1180 	u64 cur_end;
1181 
1182 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1183 			 1, 0, NULL, GFP_NOFS);
1184 	while (start < end) {
1185 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1186 		BUG_ON(!async_cow); /* -ENOMEM */
1187 		async_cow->inode = igrab(inode);
1188 		async_cow->root = root;
1189 		async_cow->locked_page = locked_page;
1190 		async_cow->start = start;
1191 
1192 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1193 		    !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1194 			cur_end = end;
1195 		else
1196 			cur_end = min(end, start + SZ_512K - 1);
1197 
1198 		async_cow->end = cur_end;
1199 		INIT_LIST_HEAD(&async_cow->extents);
1200 
1201 		btrfs_init_work(&async_cow->work,
1202 				btrfs_delalloc_helper,
1203 				async_cow_start, async_cow_submit,
1204 				async_cow_free);
1205 
1206 		nr_pages = (cur_end - start + PAGE_SIZE) >>
1207 			PAGE_SHIFT;
1208 		atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1209 
1210 		btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1211 
1212 		while (atomic_read(&fs_info->async_submit_draining) &&
1213 		       atomic_read(&fs_info->async_delalloc_pages)) {
1214 			wait_event(fs_info->async_submit_wait,
1215 				   (atomic_read(&fs_info->async_delalloc_pages) ==
1216 				    0));
1217 		}
1218 
1219 		*nr_written += nr_pages;
1220 		start = cur_end + 1;
1221 	}
1222 	*page_started = 1;
1223 	return 0;
1224 }
1225 
1226 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1227 					u64 bytenr, u64 num_bytes)
1228 {
1229 	int ret;
1230 	struct btrfs_ordered_sum *sums;
1231 	LIST_HEAD(list);
1232 
1233 	ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1234 				       bytenr + num_bytes - 1, &list, 0);
1235 	if (ret == 0 && list_empty(&list))
1236 		return 0;
1237 
1238 	while (!list_empty(&list)) {
1239 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1240 		list_del(&sums->list);
1241 		kfree(sums);
1242 	}
1243 	return 1;
1244 }
1245 
1246 /*
1247  * when nowcow writeback call back.  This checks for snapshots or COW copies
1248  * of the extents that exist in the file, and COWs the file as required.
1249  *
1250  * If no cow copies or snapshots exist, we write directly to the existing
1251  * blocks on disk
1252  */
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 				       struct page *locked_page,
1255 			      u64 start, u64 end, int *page_started, int force,
1256 			      unsigned long *nr_written)
1257 {
1258 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 	struct btrfs_root *root = BTRFS_I(inode)->root;
1260 	struct extent_buffer *leaf;
1261 	struct btrfs_path *path;
1262 	struct btrfs_file_extent_item *fi;
1263 	struct btrfs_key found_key;
1264 	struct extent_map *em;
1265 	u64 cow_start;
1266 	u64 cur_offset;
1267 	u64 extent_end;
1268 	u64 extent_offset;
1269 	u64 disk_bytenr;
1270 	u64 num_bytes;
1271 	u64 disk_num_bytes;
1272 	u64 ram_bytes;
1273 	int extent_type;
1274 	int ret, err;
1275 	int type;
1276 	int nocow;
1277 	int check_prev = 1;
1278 	bool nolock;
1279 	u64 ino = btrfs_ino(BTRFS_I(inode));
1280 
1281 	path = btrfs_alloc_path();
1282 	if (!path) {
1283 		extent_clear_unlock_delalloc(inode, start, end, end,
1284 					     locked_page,
1285 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1286 					     EXTENT_DO_ACCOUNTING |
1287 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1288 					     PAGE_CLEAR_DIRTY |
1289 					     PAGE_SET_WRITEBACK |
1290 					     PAGE_END_WRITEBACK);
1291 		return -ENOMEM;
1292 	}
1293 
1294 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1295 
1296 	cow_start = (u64)-1;
1297 	cur_offset = start;
1298 	while (1) {
1299 		ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1300 					       cur_offset, 0);
1301 		if (ret < 0)
1302 			goto error;
1303 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 			leaf = path->nodes[0];
1305 			btrfs_item_key_to_cpu(leaf, &found_key,
1306 					      path->slots[0] - 1);
1307 			if (found_key.objectid == ino &&
1308 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1309 				path->slots[0]--;
1310 		}
1311 		check_prev = 0;
1312 next_slot:
1313 		leaf = path->nodes[0];
1314 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 			ret = btrfs_next_leaf(root, path);
1316 			if (ret < 0)
1317 				goto error;
1318 			if (ret > 0)
1319 				break;
1320 			leaf = path->nodes[0];
1321 		}
1322 
1323 		nocow = 0;
1324 		disk_bytenr = 0;
1325 		num_bytes = 0;
1326 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1327 
1328 		if (found_key.objectid > ino)
1329 			break;
1330 		if (WARN_ON_ONCE(found_key.objectid < ino) ||
1331 		    found_key.type < BTRFS_EXTENT_DATA_KEY) {
1332 			path->slots[0]++;
1333 			goto next_slot;
1334 		}
1335 		if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1336 		    found_key.offset > end)
1337 			break;
1338 
1339 		if (found_key.offset > cur_offset) {
1340 			extent_end = found_key.offset;
1341 			extent_type = 0;
1342 			goto out_check;
1343 		}
1344 
1345 		fi = btrfs_item_ptr(leaf, path->slots[0],
1346 				    struct btrfs_file_extent_item);
1347 		extent_type = btrfs_file_extent_type(leaf, fi);
1348 
1349 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1350 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1351 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1353 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1354 			extent_end = found_key.offset +
1355 				btrfs_file_extent_num_bytes(leaf, fi);
1356 			disk_num_bytes =
1357 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1358 			if (extent_end <= start) {
1359 				path->slots[0]++;
1360 				goto next_slot;
1361 			}
1362 			if (disk_bytenr == 0)
1363 				goto out_check;
1364 			if (btrfs_file_extent_compression(leaf, fi) ||
1365 			    btrfs_file_extent_encryption(leaf, fi) ||
1366 			    btrfs_file_extent_other_encoding(leaf, fi))
1367 				goto out_check;
1368 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1369 				goto out_check;
1370 			if (btrfs_extent_readonly(fs_info, disk_bytenr))
1371 				goto out_check;
1372 			if (btrfs_cross_ref_exist(root, ino,
1373 						  found_key.offset -
1374 						  extent_offset, disk_bytenr))
1375 				goto out_check;
1376 			disk_bytenr += extent_offset;
1377 			disk_bytenr += cur_offset - found_key.offset;
1378 			num_bytes = min(end + 1, extent_end) - cur_offset;
1379 			/*
1380 			 * if there are pending snapshots for this root,
1381 			 * we fall into common COW way.
1382 			 */
1383 			if (!nolock) {
1384 				err = btrfs_start_write_no_snapshoting(root);
1385 				if (!err)
1386 					goto out_check;
1387 			}
1388 			/*
1389 			 * force cow if csum exists in the range.
1390 			 * this ensure that csum for a given extent are
1391 			 * either valid or do not exist.
1392 			 */
1393 			if (csum_exist_in_range(fs_info, disk_bytenr,
1394 						num_bytes)) {
1395 				if (!nolock)
1396 					btrfs_end_write_no_snapshoting(root);
1397 				goto out_check;
1398 			}
1399 			if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1400 				if (!nolock)
1401 					btrfs_end_write_no_snapshoting(root);
1402 				goto out_check;
1403 			}
1404 			nocow = 1;
1405 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1406 			extent_end = found_key.offset +
1407 				btrfs_file_extent_inline_len(leaf,
1408 						     path->slots[0], fi);
1409 			extent_end = ALIGN(extent_end,
1410 					   fs_info->sectorsize);
1411 		} else {
1412 			BUG_ON(1);
1413 		}
1414 out_check:
1415 		if (extent_end <= start) {
1416 			path->slots[0]++;
1417 			if (!nolock && nocow)
1418 				btrfs_end_write_no_snapshoting(root);
1419 			if (nocow)
1420 				btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1421 			goto next_slot;
1422 		}
1423 		if (!nocow) {
1424 			if (cow_start == (u64)-1)
1425 				cow_start = cur_offset;
1426 			cur_offset = extent_end;
1427 			if (cur_offset > end)
1428 				break;
1429 			path->slots[0]++;
1430 			goto next_slot;
1431 		}
1432 
1433 		btrfs_release_path(path);
1434 		if (cow_start != (u64)-1) {
1435 			ret = cow_file_range(inode, locked_page,
1436 					     cow_start, found_key.offset - 1,
1437 					     end, page_started, nr_written, 1,
1438 					     NULL);
1439 			if (ret) {
1440 				if (!nolock && nocow)
1441 					btrfs_end_write_no_snapshoting(root);
1442 				if (nocow)
1443 					btrfs_dec_nocow_writers(fs_info,
1444 								disk_bytenr);
1445 				goto error;
1446 			}
1447 			cow_start = (u64)-1;
1448 		}
1449 
1450 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1451 			u64 orig_start = found_key.offset - extent_offset;
1452 
1453 			em = create_io_em(inode, cur_offset, num_bytes,
1454 					  orig_start,
1455 					  disk_bytenr, /* block_start */
1456 					  num_bytes, /* block_len */
1457 					  disk_num_bytes, /* orig_block_len */
1458 					  ram_bytes, BTRFS_COMPRESS_NONE,
1459 					  BTRFS_ORDERED_PREALLOC);
1460 			if (IS_ERR(em)) {
1461 				if (!nolock && nocow)
1462 					btrfs_end_write_no_snapshoting(root);
1463 				if (nocow)
1464 					btrfs_dec_nocow_writers(fs_info,
1465 								disk_bytenr);
1466 				ret = PTR_ERR(em);
1467 				goto error;
1468 			}
1469 			free_extent_map(em);
1470 		}
1471 
1472 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1473 			type = BTRFS_ORDERED_PREALLOC;
1474 		} else {
1475 			type = BTRFS_ORDERED_NOCOW;
1476 		}
1477 
1478 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1479 					       num_bytes, num_bytes, type);
1480 		if (nocow)
1481 			btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1482 		BUG_ON(ret); /* -ENOMEM */
1483 
1484 		if (root->root_key.objectid ==
1485 		    BTRFS_DATA_RELOC_TREE_OBJECTID)
1486 			/*
1487 			 * Error handled later, as we must prevent
1488 			 * extent_clear_unlock_delalloc() in error handler
1489 			 * from freeing metadata of created ordered extent.
1490 			 */
1491 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1492 						      num_bytes);
1493 
1494 		extent_clear_unlock_delalloc(inode, cur_offset,
1495 					     cur_offset + num_bytes - 1, end,
1496 					     locked_page, EXTENT_LOCKED |
1497 					     EXTENT_DELALLOC |
1498 					     EXTENT_CLEAR_DATA_RESV,
1499 					     PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1500 
1501 		if (!nolock && nocow)
1502 			btrfs_end_write_no_snapshoting(root);
1503 		cur_offset = extent_end;
1504 
1505 		/*
1506 		 * btrfs_reloc_clone_csums() error, now we're OK to call error
1507 		 * handler, as metadata for created ordered extent will only
1508 		 * be freed by btrfs_finish_ordered_io().
1509 		 */
1510 		if (ret)
1511 			goto error;
1512 		if (cur_offset > end)
1513 			break;
1514 	}
1515 	btrfs_release_path(path);
1516 
1517 	if (cur_offset <= end && cow_start == (u64)-1) {
1518 		cow_start = cur_offset;
1519 		cur_offset = end;
1520 	}
1521 
1522 	if (cow_start != (u64)-1) {
1523 		ret = cow_file_range(inode, locked_page, cow_start, end, end,
1524 				     page_started, nr_written, 1, NULL);
1525 		if (ret)
1526 			goto error;
1527 	}
1528 
1529 error:
1530 	if (ret && cur_offset < end)
1531 		extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1532 					     locked_page, EXTENT_LOCKED |
1533 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1534 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1535 					     PAGE_CLEAR_DIRTY |
1536 					     PAGE_SET_WRITEBACK |
1537 					     PAGE_END_WRITEBACK);
1538 	btrfs_free_path(path);
1539 	return ret;
1540 }
1541 
1542 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1543 {
1544 
1545 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1546 	    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1547 		return 0;
1548 
1549 	/*
1550 	 * @defrag_bytes is a hint value, no spinlock held here,
1551 	 * if is not zero, it means the file is defragging.
1552 	 * Force cow if given extent needs to be defragged.
1553 	 */
1554 	if (BTRFS_I(inode)->defrag_bytes &&
1555 	    test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1556 			   EXTENT_DEFRAG, 0, NULL))
1557 		return 1;
1558 
1559 	return 0;
1560 }
1561 
1562 /*
1563  * extent_io.c call back to do delayed allocation processing
1564  */
1565 static int run_delalloc_range(void *private_data, struct page *locked_page,
1566 			      u64 start, u64 end, int *page_started,
1567 			      unsigned long *nr_written)
1568 {
1569 	struct inode *inode = private_data;
1570 	int ret;
1571 	int force_cow = need_force_cow(inode, start, end);
1572 
1573 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1574 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1575 					 page_started, 1, nr_written);
1576 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1577 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1578 					 page_started, 0, nr_written);
1579 	} else if (!inode_need_compress(inode)) {
1580 		ret = cow_file_range(inode, locked_page, start, end, end,
1581 				      page_started, nr_written, 1, NULL);
1582 	} else {
1583 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1584 			&BTRFS_I(inode)->runtime_flags);
1585 		ret = cow_file_range_async(inode, locked_page, start, end,
1586 					   page_started, nr_written);
1587 	}
1588 	if (ret)
1589 		btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1590 	return ret;
1591 }
1592 
1593 static void btrfs_split_extent_hook(void *private_data,
1594 				    struct extent_state *orig, u64 split)
1595 {
1596 	struct inode *inode = private_data;
1597 	u64 size;
1598 
1599 	/* not delalloc, ignore it */
1600 	if (!(orig->state & EXTENT_DELALLOC))
1601 		return;
1602 
1603 	size = orig->end - orig->start + 1;
1604 	if (size > BTRFS_MAX_EXTENT_SIZE) {
1605 		u32 num_extents;
1606 		u64 new_size;
1607 
1608 		/*
1609 		 * See the explanation in btrfs_merge_extent_hook, the same
1610 		 * applies here, just in reverse.
1611 		 */
1612 		new_size = orig->end - split + 1;
1613 		num_extents = count_max_extents(new_size);
1614 		new_size = split - orig->start;
1615 		num_extents += count_max_extents(new_size);
1616 		if (count_max_extents(size) >= num_extents)
1617 			return;
1618 	}
1619 
1620 	spin_lock(&BTRFS_I(inode)->lock);
1621 	BTRFS_I(inode)->outstanding_extents++;
1622 	spin_unlock(&BTRFS_I(inode)->lock);
1623 }
1624 
1625 /*
1626  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1627  * extents so we can keep track of new extents that are just merged onto old
1628  * extents, such as when we are doing sequential writes, so we can properly
1629  * account for the metadata space we'll need.
1630  */
1631 static void btrfs_merge_extent_hook(void *private_data,
1632 				    struct extent_state *new,
1633 				    struct extent_state *other)
1634 {
1635 	struct inode *inode = private_data;
1636 	u64 new_size, old_size;
1637 	u32 num_extents;
1638 
1639 	/* not delalloc, ignore it */
1640 	if (!(other->state & EXTENT_DELALLOC))
1641 		return;
1642 
1643 	if (new->start > other->start)
1644 		new_size = new->end - other->start + 1;
1645 	else
1646 		new_size = other->end - new->start + 1;
1647 
1648 	/* we're not bigger than the max, unreserve the space and go */
1649 	if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1650 		spin_lock(&BTRFS_I(inode)->lock);
1651 		BTRFS_I(inode)->outstanding_extents--;
1652 		spin_unlock(&BTRFS_I(inode)->lock);
1653 		return;
1654 	}
1655 
1656 	/*
1657 	 * We have to add up either side to figure out how many extents were
1658 	 * accounted for before we merged into one big extent.  If the number of
1659 	 * extents we accounted for is <= the amount we need for the new range
1660 	 * then we can return, otherwise drop.  Think of it like this
1661 	 *
1662 	 * [ 4k][MAX_SIZE]
1663 	 *
1664 	 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1665 	 * need 2 outstanding extents, on one side we have 1 and the other side
1666 	 * we have 1 so they are == and we can return.  But in this case
1667 	 *
1668 	 * [MAX_SIZE+4k][MAX_SIZE+4k]
1669 	 *
1670 	 * Each range on their own accounts for 2 extents, but merged together
1671 	 * they are only 3 extents worth of accounting, so we need to drop in
1672 	 * this case.
1673 	 */
1674 	old_size = other->end - other->start + 1;
1675 	num_extents = count_max_extents(old_size);
1676 	old_size = new->end - new->start + 1;
1677 	num_extents += count_max_extents(old_size);
1678 	if (count_max_extents(new_size) >= num_extents)
1679 		return;
1680 
1681 	spin_lock(&BTRFS_I(inode)->lock);
1682 	BTRFS_I(inode)->outstanding_extents--;
1683 	spin_unlock(&BTRFS_I(inode)->lock);
1684 }
1685 
1686 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1687 				      struct inode *inode)
1688 {
1689 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1690 
1691 	spin_lock(&root->delalloc_lock);
1692 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1693 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1694 			      &root->delalloc_inodes);
1695 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1696 			&BTRFS_I(inode)->runtime_flags);
1697 		root->nr_delalloc_inodes++;
1698 		if (root->nr_delalloc_inodes == 1) {
1699 			spin_lock(&fs_info->delalloc_root_lock);
1700 			BUG_ON(!list_empty(&root->delalloc_root));
1701 			list_add_tail(&root->delalloc_root,
1702 				      &fs_info->delalloc_roots);
1703 			spin_unlock(&fs_info->delalloc_root_lock);
1704 		}
1705 	}
1706 	spin_unlock(&root->delalloc_lock);
1707 }
1708 
1709 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1710 				     struct btrfs_inode *inode)
1711 {
1712 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1713 
1714 	spin_lock(&root->delalloc_lock);
1715 	if (!list_empty(&inode->delalloc_inodes)) {
1716 		list_del_init(&inode->delalloc_inodes);
1717 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1718 			  &inode->runtime_flags);
1719 		root->nr_delalloc_inodes--;
1720 		if (!root->nr_delalloc_inodes) {
1721 			spin_lock(&fs_info->delalloc_root_lock);
1722 			BUG_ON(list_empty(&root->delalloc_root));
1723 			list_del_init(&root->delalloc_root);
1724 			spin_unlock(&fs_info->delalloc_root_lock);
1725 		}
1726 	}
1727 	spin_unlock(&root->delalloc_lock);
1728 }
1729 
1730 /*
1731  * extent_io.c set_bit_hook, used to track delayed allocation
1732  * bytes in this file, and to maintain the list of inodes that
1733  * have pending delalloc work to be done.
1734  */
1735 static void btrfs_set_bit_hook(void *private_data,
1736 			       struct extent_state *state, unsigned *bits)
1737 {
1738 	struct inode *inode = private_data;
1739 
1740 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1741 
1742 	if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1743 		WARN_ON(1);
1744 	/*
1745 	 * set_bit and clear bit hooks normally require _irqsave/restore
1746 	 * but in this case, we are only testing for the DELALLOC
1747 	 * bit, which is only set or cleared with irqs on
1748 	 */
1749 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1750 		struct btrfs_root *root = BTRFS_I(inode)->root;
1751 		u64 len = state->end + 1 - state->start;
1752 		bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1753 
1754 		if (*bits & EXTENT_FIRST_DELALLOC) {
1755 			*bits &= ~EXTENT_FIRST_DELALLOC;
1756 		} else {
1757 			spin_lock(&BTRFS_I(inode)->lock);
1758 			BTRFS_I(inode)->outstanding_extents++;
1759 			spin_unlock(&BTRFS_I(inode)->lock);
1760 		}
1761 
1762 		/* For sanity tests */
1763 		if (btrfs_is_testing(fs_info))
1764 			return;
1765 
1766 		percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1767 					 fs_info->delalloc_batch);
1768 		spin_lock(&BTRFS_I(inode)->lock);
1769 		BTRFS_I(inode)->delalloc_bytes += len;
1770 		if (*bits & EXTENT_DEFRAG)
1771 			BTRFS_I(inode)->defrag_bytes += len;
1772 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1773 					 &BTRFS_I(inode)->runtime_flags))
1774 			btrfs_add_delalloc_inodes(root, inode);
1775 		spin_unlock(&BTRFS_I(inode)->lock);
1776 	}
1777 
1778 	if (!(state->state & EXTENT_DELALLOC_NEW) &&
1779 	    (*bits & EXTENT_DELALLOC_NEW)) {
1780 		spin_lock(&BTRFS_I(inode)->lock);
1781 		BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1782 			state->start;
1783 		spin_unlock(&BTRFS_I(inode)->lock);
1784 	}
1785 }
1786 
1787 /*
1788  * extent_io.c clear_bit_hook, see set_bit_hook for why
1789  */
1790 static void btrfs_clear_bit_hook(void *private_data,
1791 				 struct extent_state *state,
1792 				 unsigned *bits)
1793 {
1794 	struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1795 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1796 	u64 len = state->end + 1 - state->start;
1797 	u32 num_extents = count_max_extents(len);
1798 
1799 	spin_lock(&inode->lock);
1800 	if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1801 		inode->defrag_bytes -= len;
1802 	spin_unlock(&inode->lock);
1803 
1804 	/*
1805 	 * set_bit and clear bit hooks normally require _irqsave/restore
1806 	 * but in this case, we are only testing for the DELALLOC
1807 	 * bit, which is only set or cleared with irqs on
1808 	 */
1809 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1810 		struct btrfs_root *root = inode->root;
1811 		bool do_list = !btrfs_is_free_space_inode(inode);
1812 
1813 		if (*bits & EXTENT_FIRST_DELALLOC) {
1814 			*bits &= ~EXTENT_FIRST_DELALLOC;
1815 		} else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1816 			spin_lock(&inode->lock);
1817 			inode->outstanding_extents -= num_extents;
1818 			spin_unlock(&inode->lock);
1819 		}
1820 
1821 		/*
1822 		 * We don't reserve metadata space for space cache inodes so we
1823 		 * don't need to call dellalloc_release_metadata if there is an
1824 		 * error.
1825 		 */
1826 		if (*bits & EXTENT_CLEAR_META_RESV &&
1827 		    root != fs_info->tree_root)
1828 			btrfs_delalloc_release_metadata(inode, len);
1829 
1830 		/* For sanity tests. */
1831 		if (btrfs_is_testing(fs_info))
1832 			return;
1833 
1834 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1835 		    do_list && !(state->state & EXTENT_NORESERVE) &&
1836 		    (*bits & EXTENT_CLEAR_DATA_RESV))
1837 			btrfs_free_reserved_data_space_noquota(
1838 					&inode->vfs_inode,
1839 					state->start, len);
1840 
1841 		percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1842 					 fs_info->delalloc_batch);
1843 		spin_lock(&inode->lock);
1844 		inode->delalloc_bytes -= len;
1845 		if (do_list && inode->delalloc_bytes == 0 &&
1846 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1847 					&inode->runtime_flags))
1848 			btrfs_del_delalloc_inode(root, inode);
1849 		spin_unlock(&inode->lock);
1850 	}
1851 
1852 	if ((state->state & EXTENT_DELALLOC_NEW) &&
1853 	    (*bits & EXTENT_DELALLOC_NEW)) {
1854 		spin_lock(&inode->lock);
1855 		ASSERT(inode->new_delalloc_bytes >= len);
1856 		inode->new_delalloc_bytes -= len;
1857 		spin_unlock(&inode->lock);
1858 	}
1859 }
1860 
1861 /*
1862  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1863  * we don't create bios that span stripes or chunks
1864  *
1865  * return 1 if page cannot be merged to bio
1866  * return 0 if page can be merged to bio
1867  * return error otherwise
1868  */
1869 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1870 			 size_t size, struct bio *bio,
1871 			 unsigned long bio_flags)
1872 {
1873 	struct inode *inode = page->mapping->host;
1874 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1875 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1876 	u64 length = 0;
1877 	u64 map_length;
1878 	int ret;
1879 
1880 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1881 		return 0;
1882 
1883 	length = bio->bi_iter.bi_size;
1884 	map_length = length;
1885 	ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1886 			      NULL, 0);
1887 	if (ret < 0)
1888 		return ret;
1889 	if (map_length < length + size)
1890 		return 1;
1891 	return 0;
1892 }
1893 
1894 /*
1895  * in order to insert checksums into the metadata in large chunks,
1896  * we wait until bio submission time.   All the pages in the bio are
1897  * checksummed and sums are attached onto the ordered extent record.
1898  *
1899  * At IO completion time the cums attached on the ordered extent record
1900  * are inserted into the btree
1901  */
1902 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1903 				    int mirror_num, unsigned long bio_flags,
1904 				    u64 bio_offset)
1905 {
1906 	struct inode *inode = private_data;
1907 	blk_status_t ret = 0;
1908 
1909 	ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1910 	BUG_ON(ret); /* -ENOMEM */
1911 	return 0;
1912 }
1913 
1914 /*
1915  * in order to insert checksums into the metadata in large chunks,
1916  * we wait until bio submission time.   All the pages in the bio are
1917  * checksummed and sums are attached onto the ordered extent record.
1918  *
1919  * At IO completion time the cums attached on the ordered extent record
1920  * are inserted into the btree
1921  */
1922 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1923 			  int mirror_num, unsigned long bio_flags,
1924 			  u64 bio_offset)
1925 {
1926 	struct inode *inode = private_data;
1927 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1928 	blk_status_t ret;
1929 
1930 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1931 	if (ret) {
1932 		bio->bi_status = ret;
1933 		bio_endio(bio);
1934 	}
1935 	return ret;
1936 }
1937 
1938 /*
1939  * extent_io.c submission hook. This does the right thing for csum calculation
1940  * on write, or reading the csums from the tree before a read
1941  */
1942 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1943 				 int mirror_num, unsigned long bio_flags,
1944 				 u64 bio_offset)
1945 {
1946 	struct inode *inode = private_data;
1947 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1948 	struct btrfs_root *root = BTRFS_I(inode)->root;
1949 	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1950 	blk_status_t ret = 0;
1951 	int skip_sum;
1952 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1953 
1954 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1955 
1956 	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1957 		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1958 
1959 	if (bio_op(bio) != REQ_OP_WRITE) {
1960 		ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1961 		if (ret)
1962 			goto out;
1963 
1964 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1965 			ret = btrfs_submit_compressed_read(inode, bio,
1966 							   mirror_num,
1967 							   bio_flags);
1968 			goto out;
1969 		} else if (!skip_sum) {
1970 			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1971 			if (ret)
1972 				goto out;
1973 		}
1974 		goto mapit;
1975 	} else if (async && !skip_sum) {
1976 		/* csum items have already been cloned */
1977 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1978 			goto mapit;
1979 		/* we're doing a write, do the async checksumming */
1980 		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1981 					  bio_offset, inode,
1982 					  __btrfs_submit_bio_start,
1983 					  __btrfs_submit_bio_done);
1984 		goto out;
1985 	} else if (!skip_sum) {
1986 		ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1987 		if (ret)
1988 			goto out;
1989 	}
1990 
1991 mapit:
1992 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1993 
1994 out:
1995 	if (ret) {
1996 		bio->bi_status = ret;
1997 		bio_endio(bio);
1998 	}
1999 	return ret;
2000 }
2001 
2002 /*
2003  * given a list of ordered sums record them in the inode.  This happens
2004  * at IO completion time based on sums calculated at bio submission time.
2005  */
2006 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2007 			     struct inode *inode, struct list_head *list)
2008 {
2009 	struct btrfs_ordered_sum *sum;
2010 
2011 	list_for_each_entry(sum, list, list) {
2012 		trans->adding_csums = 1;
2013 		btrfs_csum_file_blocks(trans,
2014 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
2015 		trans->adding_csums = 0;
2016 	}
2017 	return 0;
2018 }
2019 
2020 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2021 			      struct extent_state **cached_state, int dedupe)
2022 {
2023 	WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2024 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2025 				   cached_state);
2026 }
2027 
2028 /* see btrfs_writepage_start_hook for details on why this is required */
2029 struct btrfs_writepage_fixup {
2030 	struct page *page;
2031 	struct btrfs_work work;
2032 };
2033 
2034 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2035 {
2036 	struct btrfs_writepage_fixup *fixup;
2037 	struct btrfs_ordered_extent *ordered;
2038 	struct extent_state *cached_state = NULL;
2039 	struct extent_changeset *data_reserved = NULL;
2040 	struct page *page;
2041 	struct inode *inode;
2042 	u64 page_start;
2043 	u64 page_end;
2044 	int ret;
2045 
2046 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
2047 	page = fixup->page;
2048 again:
2049 	lock_page(page);
2050 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2051 		ClearPageChecked(page);
2052 		goto out_page;
2053 	}
2054 
2055 	inode = page->mapping->host;
2056 	page_start = page_offset(page);
2057 	page_end = page_offset(page) + PAGE_SIZE - 1;
2058 
2059 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2060 			 &cached_state);
2061 
2062 	/* already ordered? We're done */
2063 	if (PagePrivate2(page))
2064 		goto out;
2065 
2066 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2067 					PAGE_SIZE);
2068 	if (ordered) {
2069 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2070 				     page_end, &cached_state, GFP_NOFS);
2071 		unlock_page(page);
2072 		btrfs_start_ordered_extent(inode, ordered, 1);
2073 		btrfs_put_ordered_extent(ordered);
2074 		goto again;
2075 	}
2076 
2077 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2078 					   PAGE_SIZE);
2079 	if (ret) {
2080 		mapping_set_error(page->mapping, ret);
2081 		end_extent_writepage(page, ret, page_start, page_end);
2082 		ClearPageChecked(page);
2083 		goto out;
2084 	 }
2085 
2086 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2087 				  0);
2088 	ClearPageChecked(page);
2089 	set_page_dirty(page);
2090 out:
2091 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2092 			     &cached_state, GFP_NOFS);
2093 out_page:
2094 	unlock_page(page);
2095 	put_page(page);
2096 	kfree(fixup);
2097 	extent_changeset_free(data_reserved);
2098 }
2099 
2100 /*
2101  * There are a few paths in the higher layers of the kernel that directly
2102  * set the page dirty bit without asking the filesystem if it is a
2103  * good idea.  This causes problems because we want to make sure COW
2104  * properly happens and the data=ordered rules are followed.
2105  *
2106  * In our case any range that doesn't have the ORDERED bit set
2107  * hasn't been properly setup for IO.  We kick off an async process
2108  * to fix it up.  The async helper will wait for ordered extents, set
2109  * the delalloc bit and make it safe to write the page.
2110  */
2111 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2112 {
2113 	struct inode *inode = page->mapping->host;
2114 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2115 	struct btrfs_writepage_fixup *fixup;
2116 
2117 	/* this page is properly in the ordered list */
2118 	if (TestClearPagePrivate2(page))
2119 		return 0;
2120 
2121 	if (PageChecked(page))
2122 		return -EAGAIN;
2123 
2124 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2125 	if (!fixup)
2126 		return -EAGAIN;
2127 
2128 	SetPageChecked(page);
2129 	get_page(page);
2130 	btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2131 			btrfs_writepage_fixup_worker, NULL, NULL);
2132 	fixup->page = page;
2133 	btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2134 	return -EBUSY;
2135 }
2136 
2137 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2138 				       struct inode *inode, u64 file_pos,
2139 				       u64 disk_bytenr, u64 disk_num_bytes,
2140 				       u64 num_bytes, u64 ram_bytes,
2141 				       u8 compression, u8 encryption,
2142 				       u16 other_encoding, int extent_type)
2143 {
2144 	struct btrfs_root *root = BTRFS_I(inode)->root;
2145 	struct btrfs_file_extent_item *fi;
2146 	struct btrfs_path *path;
2147 	struct extent_buffer *leaf;
2148 	struct btrfs_key ins;
2149 	u64 qg_released;
2150 	int extent_inserted = 0;
2151 	int ret;
2152 
2153 	path = btrfs_alloc_path();
2154 	if (!path)
2155 		return -ENOMEM;
2156 
2157 	/*
2158 	 * we may be replacing one extent in the tree with another.
2159 	 * The new extent is pinned in the extent map, and we don't want
2160 	 * to drop it from the cache until it is completely in the btree.
2161 	 *
2162 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
2163 	 * the caller is expected to unpin it and allow it to be merged
2164 	 * with the others.
2165 	 */
2166 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2167 				   file_pos + num_bytes, NULL, 0,
2168 				   1, sizeof(*fi), &extent_inserted);
2169 	if (ret)
2170 		goto out;
2171 
2172 	if (!extent_inserted) {
2173 		ins.objectid = btrfs_ino(BTRFS_I(inode));
2174 		ins.offset = file_pos;
2175 		ins.type = BTRFS_EXTENT_DATA_KEY;
2176 
2177 		path->leave_spinning = 1;
2178 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
2179 					      sizeof(*fi));
2180 		if (ret)
2181 			goto out;
2182 	}
2183 	leaf = path->nodes[0];
2184 	fi = btrfs_item_ptr(leaf, path->slots[0],
2185 			    struct btrfs_file_extent_item);
2186 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2187 	btrfs_set_file_extent_type(leaf, fi, extent_type);
2188 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2189 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2190 	btrfs_set_file_extent_offset(leaf, fi, 0);
2191 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2192 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2193 	btrfs_set_file_extent_compression(leaf, fi, compression);
2194 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
2195 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2196 
2197 	btrfs_mark_buffer_dirty(leaf);
2198 	btrfs_release_path(path);
2199 
2200 	inode_add_bytes(inode, num_bytes);
2201 
2202 	ins.objectid = disk_bytenr;
2203 	ins.offset = disk_num_bytes;
2204 	ins.type = BTRFS_EXTENT_ITEM_KEY;
2205 
2206 	/*
2207 	 * Release the reserved range from inode dirty range map, as it is
2208 	 * already moved into delayed_ref_head
2209 	 */
2210 	ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2211 	if (ret < 0)
2212 		goto out;
2213 	qg_released = ret;
2214 	ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2215 			btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins);
2216 out:
2217 	btrfs_free_path(path);
2218 
2219 	return ret;
2220 }
2221 
2222 /* snapshot-aware defrag */
2223 struct sa_defrag_extent_backref {
2224 	struct rb_node node;
2225 	struct old_sa_defrag_extent *old;
2226 	u64 root_id;
2227 	u64 inum;
2228 	u64 file_pos;
2229 	u64 extent_offset;
2230 	u64 num_bytes;
2231 	u64 generation;
2232 };
2233 
2234 struct old_sa_defrag_extent {
2235 	struct list_head list;
2236 	struct new_sa_defrag_extent *new;
2237 
2238 	u64 extent_offset;
2239 	u64 bytenr;
2240 	u64 offset;
2241 	u64 len;
2242 	int count;
2243 };
2244 
2245 struct new_sa_defrag_extent {
2246 	struct rb_root root;
2247 	struct list_head head;
2248 	struct btrfs_path *path;
2249 	struct inode *inode;
2250 	u64 file_pos;
2251 	u64 len;
2252 	u64 bytenr;
2253 	u64 disk_len;
2254 	u8 compress_type;
2255 };
2256 
2257 static int backref_comp(struct sa_defrag_extent_backref *b1,
2258 			struct sa_defrag_extent_backref *b2)
2259 {
2260 	if (b1->root_id < b2->root_id)
2261 		return -1;
2262 	else if (b1->root_id > b2->root_id)
2263 		return 1;
2264 
2265 	if (b1->inum < b2->inum)
2266 		return -1;
2267 	else if (b1->inum > b2->inum)
2268 		return 1;
2269 
2270 	if (b1->file_pos < b2->file_pos)
2271 		return -1;
2272 	else if (b1->file_pos > b2->file_pos)
2273 		return 1;
2274 
2275 	/*
2276 	 * [------------------------------] ===> (a range of space)
2277 	 *     |<--->|   |<---->| =============> (fs/file tree A)
2278 	 * |<---------------------------->| ===> (fs/file tree B)
2279 	 *
2280 	 * A range of space can refer to two file extents in one tree while
2281 	 * refer to only one file extent in another tree.
2282 	 *
2283 	 * So we may process a disk offset more than one time(two extents in A)
2284 	 * and locate at the same extent(one extent in B), then insert two same
2285 	 * backrefs(both refer to the extent in B).
2286 	 */
2287 	return 0;
2288 }
2289 
2290 static void backref_insert(struct rb_root *root,
2291 			   struct sa_defrag_extent_backref *backref)
2292 {
2293 	struct rb_node **p = &root->rb_node;
2294 	struct rb_node *parent = NULL;
2295 	struct sa_defrag_extent_backref *entry;
2296 	int ret;
2297 
2298 	while (*p) {
2299 		parent = *p;
2300 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2301 
2302 		ret = backref_comp(backref, entry);
2303 		if (ret < 0)
2304 			p = &(*p)->rb_left;
2305 		else
2306 			p = &(*p)->rb_right;
2307 	}
2308 
2309 	rb_link_node(&backref->node, parent, p);
2310 	rb_insert_color(&backref->node, root);
2311 }
2312 
2313 /*
2314  * Note the backref might has changed, and in this case we just return 0.
2315  */
2316 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2317 				       void *ctx)
2318 {
2319 	struct btrfs_file_extent_item *extent;
2320 	struct old_sa_defrag_extent *old = ctx;
2321 	struct new_sa_defrag_extent *new = old->new;
2322 	struct btrfs_path *path = new->path;
2323 	struct btrfs_key key;
2324 	struct btrfs_root *root;
2325 	struct sa_defrag_extent_backref *backref;
2326 	struct extent_buffer *leaf;
2327 	struct inode *inode = new->inode;
2328 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2329 	int slot;
2330 	int ret;
2331 	u64 extent_offset;
2332 	u64 num_bytes;
2333 
2334 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2335 	    inum == btrfs_ino(BTRFS_I(inode)))
2336 		return 0;
2337 
2338 	key.objectid = root_id;
2339 	key.type = BTRFS_ROOT_ITEM_KEY;
2340 	key.offset = (u64)-1;
2341 
2342 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2343 	if (IS_ERR(root)) {
2344 		if (PTR_ERR(root) == -ENOENT)
2345 			return 0;
2346 		WARN_ON(1);
2347 		btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2348 			 inum, offset, root_id);
2349 		return PTR_ERR(root);
2350 	}
2351 
2352 	key.objectid = inum;
2353 	key.type = BTRFS_EXTENT_DATA_KEY;
2354 	if (offset > (u64)-1 << 32)
2355 		key.offset = 0;
2356 	else
2357 		key.offset = offset;
2358 
2359 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2360 	if (WARN_ON(ret < 0))
2361 		return ret;
2362 	ret = 0;
2363 
2364 	while (1) {
2365 		cond_resched();
2366 
2367 		leaf = path->nodes[0];
2368 		slot = path->slots[0];
2369 
2370 		if (slot >= btrfs_header_nritems(leaf)) {
2371 			ret = btrfs_next_leaf(root, path);
2372 			if (ret < 0) {
2373 				goto out;
2374 			} else if (ret > 0) {
2375 				ret = 0;
2376 				goto out;
2377 			}
2378 			continue;
2379 		}
2380 
2381 		path->slots[0]++;
2382 
2383 		btrfs_item_key_to_cpu(leaf, &key, slot);
2384 
2385 		if (key.objectid > inum)
2386 			goto out;
2387 
2388 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2389 			continue;
2390 
2391 		extent = btrfs_item_ptr(leaf, slot,
2392 					struct btrfs_file_extent_item);
2393 
2394 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2395 			continue;
2396 
2397 		/*
2398 		 * 'offset' refers to the exact key.offset,
2399 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2400 		 * (key.offset - extent_offset).
2401 		 */
2402 		if (key.offset != offset)
2403 			continue;
2404 
2405 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2406 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2407 
2408 		if (extent_offset >= old->extent_offset + old->offset +
2409 		    old->len || extent_offset + num_bytes <=
2410 		    old->extent_offset + old->offset)
2411 			continue;
2412 		break;
2413 	}
2414 
2415 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2416 	if (!backref) {
2417 		ret = -ENOENT;
2418 		goto out;
2419 	}
2420 
2421 	backref->root_id = root_id;
2422 	backref->inum = inum;
2423 	backref->file_pos = offset;
2424 	backref->num_bytes = num_bytes;
2425 	backref->extent_offset = extent_offset;
2426 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2427 	backref->old = old;
2428 	backref_insert(&new->root, backref);
2429 	old->count++;
2430 out:
2431 	btrfs_release_path(path);
2432 	WARN_ON(ret);
2433 	return ret;
2434 }
2435 
2436 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2437 				   struct new_sa_defrag_extent *new)
2438 {
2439 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2440 	struct old_sa_defrag_extent *old, *tmp;
2441 	int ret;
2442 
2443 	new->path = path;
2444 
2445 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2446 		ret = iterate_inodes_from_logical(old->bytenr +
2447 						  old->extent_offset, fs_info,
2448 						  path, record_one_backref,
2449 						  old);
2450 		if (ret < 0 && ret != -ENOENT)
2451 			return false;
2452 
2453 		/* no backref to be processed for this extent */
2454 		if (!old->count) {
2455 			list_del(&old->list);
2456 			kfree(old);
2457 		}
2458 	}
2459 
2460 	if (list_empty(&new->head))
2461 		return false;
2462 
2463 	return true;
2464 }
2465 
2466 static int relink_is_mergable(struct extent_buffer *leaf,
2467 			      struct btrfs_file_extent_item *fi,
2468 			      struct new_sa_defrag_extent *new)
2469 {
2470 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2471 		return 0;
2472 
2473 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2474 		return 0;
2475 
2476 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2477 		return 0;
2478 
2479 	if (btrfs_file_extent_encryption(leaf, fi) ||
2480 	    btrfs_file_extent_other_encoding(leaf, fi))
2481 		return 0;
2482 
2483 	return 1;
2484 }
2485 
2486 /*
2487  * Note the backref might has changed, and in this case we just return 0.
2488  */
2489 static noinline int relink_extent_backref(struct btrfs_path *path,
2490 				 struct sa_defrag_extent_backref *prev,
2491 				 struct sa_defrag_extent_backref *backref)
2492 {
2493 	struct btrfs_file_extent_item *extent;
2494 	struct btrfs_file_extent_item *item;
2495 	struct btrfs_ordered_extent *ordered;
2496 	struct btrfs_trans_handle *trans;
2497 	struct btrfs_root *root;
2498 	struct btrfs_key key;
2499 	struct extent_buffer *leaf;
2500 	struct old_sa_defrag_extent *old = backref->old;
2501 	struct new_sa_defrag_extent *new = old->new;
2502 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2503 	struct inode *inode;
2504 	struct extent_state *cached = NULL;
2505 	int ret = 0;
2506 	u64 start;
2507 	u64 len;
2508 	u64 lock_start;
2509 	u64 lock_end;
2510 	bool merge = false;
2511 	int index;
2512 
2513 	if (prev && prev->root_id == backref->root_id &&
2514 	    prev->inum == backref->inum &&
2515 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2516 		merge = true;
2517 
2518 	/* step 1: get root */
2519 	key.objectid = backref->root_id;
2520 	key.type = BTRFS_ROOT_ITEM_KEY;
2521 	key.offset = (u64)-1;
2522 
2523 	index = srcu_read_lock(&fs_info->subvol_srcu);
2524 
2525 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2526 	if (IS_ERR(root)) {
2527 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2528 		if (PTR_ERR(root) == -ENOENT)
2529 			return 0;
2530 		return PTR_ERR(root);
2531 	}
2532 
2533 	if (btrfs_root_readonly(root)) {
2534 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2535 		return 0;
2536 	}
2537 
2538 	/* step 2: get inode */
2539 	key.objectid = backref->inum;
2540 	key.type = BTRFS_INODE_ITEM_KEY;
2541 	key.offset = 0;
2542 
2543 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2544 	if (IS_ERR(inode)) {
2545 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2546 		return 0;
2547 	}
2548 
2549 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2550 
2551 	/* step 3: relink backref */
2552 	lock_start = backref->file_pos;
2553 	lock_end = backref->file_pos + backref->num_bytes - 1;
2554 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2555 			 &cached);
2556 
2557 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2558 	if (ordered) {
2559 		btrfs_put_ordered_extent(ordered);
2560 		goto out_unlock;
2561 	}
2562 
2563 	trans = btrfs_join_transaction(root);
2564 	if (IS_ERR(trans)) {
2565 		ret = PTR_ERR(trans);
2566 		goto out_unlock;
2567 	}
2568 
2569 	key.objectid = backref->inum;
2570 	key.type = BTRFS_EXTENT_DATA_KEY;
2571 	key.offset = backref->file_pos;
2572 
2573 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2574 	if (ret < 0) {
2575 		goto out_free_path;
2576 	} else if (ret > 0) {
2577 		ret = 0;
2578 		goto out_free_path;
2579 	}
2580 
2581 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2582 				struct btrfs_file_extent_item);
2583 
2584 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2585 	    backref->generation)
2586 		goto out_free_path;
2587 
2588 	btrfs_release_path(path);
2589 
2590 	start = backref->file_pos;
2591 	if (backref->extent_offset < old->extent_offset + old->offset)
2592 		start += old->extent_offset + old->offset -
2593 			 backref->extent_offset;
2594 
2595 	len = min(backref->extent_offset + backref->num_bytes,
2596 		  old->extent_offset + old->offset + old->len);
2597 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2598 
2599 	ret = btrfs_drop_extents(trans, root, inode, start,
2600 				 start + len, 1);
2601 	if (ret)
2602 		goto out_free_path;
2603 again:
2604 	key.objectid = btrfs_ino(BTRFS_I(inode));
2605 	key.type = BTRFS_EXTENT_DATA_KEY;
2606 	key.offset = start;
2607 
2608 	path->leave_spinning = 1;
2609 	if (merge) {
2610 		struct btrfs_file_extent_item *fi;
2611 		u64 extent_len;
2612 		struct btrfs_key found_key;
2613 
2614 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2615 		if (ret < 0)
2616 			goto out_free_path;
2617 
2618 		path->slots[0]--;
2619 		leaf = path->nodes[0];
2620 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2621 
2622 		fi = btrfs_item_ptr(leaf, path->slots[0],
2623 				    struct btrfs_file_extent_item);
2624 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2625 
2626 		if (extent_len + found_key.offset == start &&
2627 		    relink_is_mergable(leaf, fi, new)) {
2628 			btrfs_set_file_extent_num_bytes(leaf, fi,
2629 							extent_len + len);
2630 			btrfs_mark_buffer_dirty(leaf);
2631 			inode_add_bytes(inode, len);
2632 
2633 			ret = 1;
2634 			goto out_free_path;
2635 		} else {
2636 			merge = false;
2637 			btrfs_release_path(path);
2638 			goto again;
2639 		}
2640 	}
2641 
2642 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2643 					sizeof(*extent));
2644 	if (ret) {
2645 		btrfs_abort_transaction(trans, ret);
2646 		goto out_free_path;
2647 	}
2648 
2649 	leaf = path->nodes[0];
2650 	item = btrfs_item_ptr(leaf, path->slots[0],
2651 				struct btrfs_file_extent_item);
2652 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2653 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2654 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2655 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2656 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2657 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2658 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2659 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2660 	btrfs_set_file_extent_encryption(leaf, item, 0);
2661 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2662 
2663 	btrfs_mark_buffer_dirty(leaf);
2664 	inode_add_bytes(inode, len);
2665 	btrfs_release_path(path);
2666 
2667 	ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2668 			new->disk_len, 0,
2669 			backref->root_id, backref->inum,
2670 			new->file_pos);	/* start - extent_offset */
2671 	if (ret) {
2672 		btrfs_abort_transaction(trans, ret);
2673 		goto out_free_path;
2674 	}
2675 
2676 	ret = 1;
2677 out_free_path:
2678 	btrfs_release_path(path);
2679 	path->leave_spinning = 0;
2680 	btrfs_end_transaction(trans);
2681 out_unlock:
2682 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2683 			     &cached, GFP_NOFS);
2684 	iput(inode);
2685 	return ret;
2686 }
2687 
2688 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2689 {
2690 	struct old_sa_defrag_extent *old, *tmp;
2691 
2692 	if (!new)
2693 		return;
2694 
2695 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2696 		kfree(old);
2697 	}
2698 	kfree(new);
2699 }
2700 
2701 static void relink_file_extents(struct new_sa_defrag_extent *new)
2702 {
2703 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2704 	struct btrfs_path *path;
2705 	struct sa_defrag_extent_backref *backref;
2706 	struct sa_defrag_extent_backref *prev = NULL;
2707 	struct inode *inode;
2708 	struct btrfs_root *root;
2709 	struct rb_node *node;
2710 	int ret;
2711 
2712 	inode = new->inode;
2713 	root = BTRFS_I(inode)->root;
2714 
2715 	path = btrfs_alloc_path();
2716 	if (!path)
2717 		return;
2718 
2719 	if (!record_extent_backrefs(path, new)) {
2720 		btrfs_free_path(path);
2721 		goto out;
2722 	}
2723 	btrfs_release_path(path);
2724 
2725 	while (1) {
2726 		node = rb_first(&new->root);
2727 		if (!node)
2728 			break;
2729 		rb_erase(node, &new->root);
2730 
2731 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2732 
2733 		ret = relink_extent_backref(path, prev, backref);
2734 		WARN_ON(ret < 0);
2735 
2736 		kfree(prev);
2737 
2738 		if (ret == 1)
2739 			prev = backref;
2740 		else
2741 			prev = NULL;
2742 		cond_resched();
2743 	}
2744 	kfree(prev);
2745 
2746 	btrfs_free_path(path);
2747 out:
2748 	free_sa_defrag_extent(new);
2749 
2750 	atomic_dec(&fs_info->defrag_running);
2751 	wake_up(&fs_info->transaction_wait);
2752 }
2753 
2754 static struct new_sa_defrag_extent *
2755 record_old_file_extents(struct inode *inode,
2756 			struct btrfs_ordered_extent *ordered)
2757 {
2758 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2759 	struct btrfs_root *root = BTRFS_I(inode)->root;
2760 	struct btrfs_path *path;
2761 	struct btrfs_key key;
2762 	struct old_sa_defrag_extent *old;
2763 	struct new_sa_defrag_extent *new;
2764 	int ret;
2765 
2766 	new = kmalloc(sizeof(*new), GFP_NOFS);
2767 	if (!new)
2768 		return NULL;
2769 
2770 	new->inode = inode;
2771 	new->file_pos = ordered->file_offset;
2772 	new->len = ordered->len;
2773 	new->bytenr = ordered->start;
2774 	new->disk_len = ordered->disk_len;
2775 	new->compress_type = ordered->compress_type;
2776 	new->root = RB_ROOT;
2777 	INIT_LIST_HEAD(&new->head);
2778 
2779 	path = btrfs_alloc_path();
2780 	if (!path)
2781 		goto out_kfree;
2782 
2783 	key.objectid = btrfs_ino(BTRFS_I(inode));
2784 	key.type = BTRFS_EXTENT_DATA_KEY;
2785 	key.offset = new->file_pos;
2786 
2787 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2788 	if (ret < 0)
2789 		goto out_free_path;
2790 	if (ret > 0 && path->slots[0] > 0)
2791 		path->slots[0]--;
2792 
2793 	/* find out all the old extents for the file range */
2794 	while (1) {
2795 		struct btrfs_file_extent_item *extent;
2796 		struct extent_buffer *l;
2797 		int slot;
2798 		u64 num_bytes;
2799 		u64 offset;
2800 		u64 end;
2801 		u64 disk_bytenr;
2802 		u64 extent_offset;
2803 
2804 		l = path->nodes[0];
2805 		slot = path->slots[0];
2806 
2807 		if (slot >= btrfs_header_nritems(l)) {
2808 			ret = btrfs_next_leaf(root, path);
2809 			if (ret < 0)
2810 				goto out_free_path;
2811 			else if (ret > 0)
2812 				break;
2813 			continue;
2814 		}
2815 
2816 		btrfs_item_key_to_cpu(l, &key, slot);
2817 
2818 		if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2819 			break;
2820 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2821 			break;
2822 		if (key.offset >= new->file_pos + new->len)
2823 			break;
2824 
2825 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2826 
2827 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2828 		if (key.offset + num_bytes < new->file_pos)
2829 			goto next;
2830 
2831 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2832 		if (!disk_bytenr)
2833 			goto next;
2834 
2835 		extent_offset = btrfs_file_extent_offset(l, extent);
2836 
2837 		old = kmalloc(sizeof(*old), GFP_NOFS);
2838 		if (!old)
2839 			goto out_free_path;
2840 
2841 		offset = max(new->file_pos, key.offset);
2842 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2843 
2844 		old->bytenr = disk_bytenr;
2845 		old->extent_offset = extent_offset;
2846 		old->offset = offset - key.offset;
2847 		old->len = end - offset;
2848 		old->new = new;
2849 		old->count = 0;
2850 		list_add_tail(&old->list, &new->head);
2851 next:
2852 		path->slots[0]++;
2853 		cond_resched();
2854 	}
2855 
2856 	btrfs_free_path(path);
2857 	atomic_inc(&fs_info->defrag_running);
2858 
2859 	return new;
2860 
2861 out_free_path:
2862 	btrfs_free_path(path);
2863 out_kfree:
2864 	free_sa_defrag_extent(new);
2865 	return NULL;
2866 }
2867 
2868 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2869 					 u64 start, u64 len)
2870 {
2871 	struct btrfs_block_group_cache *cache;
2872 
2873 	cache = btrfs_lookup_block_group(fs_info, start);
2874 	ASSERT(cache);
2875 
2876 	spin_lock(&cache->lock);
2877 	cache->delalloc_bytes -= len;
2878 	spin_unlock(&cache->lock);
2879 
2880 	btrfs_put_block_group(cache);
2881 }
2882 
2883 /* as ordered data IO finishes, this gets called so we can finish
2884  * an ordered extent if the range of bytes in the file it covers are
2885  * fully written.
2886  */
2887 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2888 {
2889 	struct inode *inode = ordered_extent->inode;
2890 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2891 	struct btrfs_root *root = BTRFS_I(inode)->root;
2892 	struct btrfs_trans_handle *trans = NULL;
2893 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2894 	struct extent_state *cached_state = NULL;
2895 	struct new_sa_defrag_extent *new = NULL;
2896 	int compress_type = 0;
2897 	int ret = 0;
2898 	u64 logical_len = ordered_extent->len;
2899 	bool nolock;
2900 	bool truncated = false;
2901 	bool range_locked = false;
2902 	bool clear_new_delalloc_bytes = false;
2903 
2904 	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2905 	    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2906 	    !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2907 		clear_new_delalloc_bytes = true;
2908 
2909 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2910 
2911 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2912 		ret = -EIO;
2913 		goto out;
2914 	}
2915 
2916 	btrfs_free_io_failure_record(BTRFS_I(inode),
2917 			ordered_extent->file_offset,
2918 			ordered_extent->file_offset +
2919 			ordered_extent->len - 1);
2920 
2921 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2922 		truncated = true;
2923 		logical_len = ordered_extent->truncated_len;
2924 		/* Truncated the entire extent, don't bother adding */
2925 		if (!logical_len)
2926 			goto out;
2927 	}
2928 
2929 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2930 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2931 
2932 		/*
2933 		 * For mwrite(mmap + memset to write) case, we still reserve
2934 		 * space for NOCOW range.
2935 		 * As NOCOW won't cause a new delayed ref, just free the space
2936 		 */
2937 		btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2938 				       ordered_extent->len);
2939 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2940 		if (nolock)
2941 			trans = btrfs_join_transaction_nolock(root);
2942 		else
2943 			trans = btrfs_join_transaction(root);
2944 		if (IS_ERR(trans)) {
2945 			ret = PTR_ERR(trans);
2946 			trans = NULL;
2947 			goto out;
2948 		}
2949 		trans->block_rsv = &fs_info->delalloc_block_rsv;
2950 		ret = btrfs_update_inode_fallback(trans, root, inode);
2951 		if (ret) /* -ENOMEM or corruption */
2952 			btrfs_abort_transaction(trans, ret);
2953 		goto out;
2954 	}
2955 
2956 	range_locked = true;
2957 	lock_extent_bits(io_tree, ordered_extent->file_offset,
2958 			 ordered_extent->file_offset + ordered_extent->len - 1,
2959 			 &cached_state);
2960 
2961 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
2962 			ordered_extent->file_offset + ordered_extent->len - 1,
2963 			EXTENT_DEFRAG, 0, cached_state);
2964 	if (ret) {
2965 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2966 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2967 			/* the inode is shared */
2968 			new = record_old_file_extents(inode, ordered_extent);
2969 
2970 		clear_extent_bit(io_tree, ordered_extent->file_offset,
2971 			ordered_extent->file_offset + ordered_extent->len - 1,
2972 			EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2973 	}
2974 
2975 	if (nolock)
2976 		trans = btrfs_join_transaction_nolock(root);
2977 	else
2978 		trans = btrfs_join_transaction(root);
2979 	if (IS_ERR(trans)) {
2980 		ret = PTR_ERR(trans);
2981 		trans = NULL;
2982 		goto out;
2983 	}
2984 
2985 	trans->block_rsv = &fs_info->delalloc_block_rsv;
2986 
2987 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2988 		compress_type = ordered_extent->compress_type;
2989 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2990 		BUG_ON(compress_type);
2991 		ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2992 						ordered_extent->file_offset,
2993 						ordered_extent->file_offset +
2994 						logical_len);
2995 	} else {
2996 		BUG_ON(root == fs_info->tree_root);
2997 		ret = insert_reserved_file_extent(trans, inode,
2998 						ordered_extent->file_offset,
2999 						ordered_extent->start,
3000 						ordered_extent->disk_len,
3001 						logical_len, logical_len,
3002 						compress_type, 0, 0,
3003 						BTRFS_FILE_EXTENT_REG);
3004 		if (!ret)
3005 			btrfs_release_delalloc_bytes(fs_info,
3006 						     ordered_extent->start,
3007 						     ordered_extent->disk_len);
3008 	}
3009 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3010 			   ordered_extent->file_offset, ordered_extent->len,
3011 			   trans->transid);
3012 	if (ret < 0) {
3013 		btrfs_abort_transaction(trans, ret);
3014 		goto out;
3015 	}
3016 
3017 	add_pending_csums(trans, inode, &ordered_extent->list);
3018 
3019 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3020 	ret = btrfs_update_inode_fallback(trans, root, inode);
3021 	if (ret) { /* -ENOMEM or corruption */
3022 		btrfs_abort_transaction(trans, ret);
3023 		goto out;
3024 	}
3025 	ret = 0;
3026 out:
3027 	if (range_locked || clear_new_delalloc_bytes) {
3028 		unsigned int clear_bits = 0;
3029 
3030 		if (range_locked)
3031 			clear_bits |= EXTENT_LOCKED;
3032 		if (clear_new_delalloc_bytes)
3033 			clear_bits |= EXTENT_DELALLOC_NEW;
3034 		clear_extent_bit(&BTRFS_I(inode)->io_tree,
3035 				 ordered_extent->file_offset,
3036 				 ordered_extent->file_offset +
3037 				 ordered_extent->len - 1,
3038 				 clear_bits,
3039 				 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3040 				 0, &cached_state, GFP_NOFS);
3041 	}
3042 
3043 	if (root != fs_info->tree_root)
3044 		btrfs_delalloc_release_metadata(BTRFS_I(inode),
3045 				ordered_extent->len);
3046 	if (trans)
3047 		btrfs_end_transaction(trans);
3048 
3049 	if (ret || truncated) {
3050 		u64 start, end;
3051 
3052 		if (truncated)
3053 			start = ordered_extent->file_offset + logical_len;
3054 		else
3055 			start = ordered_extent->file_offset;
3056 		end = ordered_extent->file_offset + ordered_extent->len - 1;
3057 		clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3058 
3059 		/* Drop the cache for the part of the extent we didn't write. */
3060 		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3061 
3062 		/*
3063 		 * If the ordered extent had an IOERR or something else went
3064 		 * wrong we need to return the space for this ordered extent
3065 		 * back to the allocator.  We only free the extent in the
3066 		 * truncated case if we didn't write out the extent at all.
3067 		 */
3068 		if ((ret || !logical_len) &&
3069 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3070 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3071 			btrfs_free_reserved_extent(fs_info,
3072 						   ordered_extent->start,
3073 						   ordered_extent->disk_len, 1);
3074 	}
3075 
3076 
3077 	/*
3078 	 * This needs to be done to make sure anybody waiting knows we are done
3079 	 * updating everything for this ordered extent.
3080 	 */
3081 	btrfs_remove_ordered_extent(inode, ordered_extent);
3082 
3083 	/* for snapshot-aware defrag */
3084 	if (new) {
3085 		if (ret) {
3086 			free_sa_defrag_extent(new);
3087 			atomic_dec(&fs_info->defrag_running);
3088 		} else {
3089 			relink_file_extents(new);
3090 		}
3091 	}
3092 
3093 	/* once for us */
3094 	btrfs_put_ordered_extent(ordered_extent);
3095 	/* once for the tree */
3096 	btrfs_put_ordered_extent(ordered_extent);
3097 
3098 	return ret;
3099 }
3100 
3101 static void finish_ordered_fn(struct btrfs_work *work)
3102 {
3103 	struct btrfs_ordered_extent *ordered_extent;
3104 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3105 	btrfs_finish_ordered_io(ordered_extent);
3106 }
3107 
3108 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3109 				struct extent_state *state, int uptodate)
3110 {
3111 	struct inode *inode = page->mapping->host;
3112 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3113 	struct btrfs_ordered_extent *ordered_extent = NULL;
3114 	struct btrfs_workqueue *wq;
3115 	btrfs_work_func_t func;
3116 
3117 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3118 
3119 	ClearPagePrivate2(page);
3120 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3121 					    end - start + 1, uptodate))
3122 		return;
3123 
3124 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3125 		wq = fs_info->endio_freespace_worker;
3126 		func = btrfs_freespace_write_helper;
3127 	} else {
3128 		wq = fs_info->endio_write_workers;
3129 		func = btrfs_endio_write_helper;
3130 	}
3131 
3132 	btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3133 			NULL);
3134 	btrfs_queue_work(wq, &ordered_extent->work);
3135 }
3136 
3137 static int __readpage_endio_check(struct inode *inode,
3138 				  struct btrfs_io_bio *io_bio,
3139 				  int icsum, struct page *page,
3140 				  int pgoff, u64 start, size_t len)
3141 {
3142 	char *kaddr;
3143 	u32 csum_expected;
3144 	u32 csum = ~(u32)0;
3145 
3146 	csum_expected = *(((u32 *)io_bio->csum) + icsum);
3147 
3148 	kaddr = kmap_atomic(page);
3149 	csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3150 	btrfs_csum_final(csum, (u8 *)&csum);
3151 	if (csum != csum_expected)
3152 		goto zeroit;
3153 
3154 	kunmap_atomic(kaddr);
3155 	return 0;
3156 zeroit:
3157 	btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3158 				    io_bio->mirror_num);
3159 	memset(kaddr + pgoff, 1, len);
3160 	flush_dcache_page(page);
3161 	kunmap_atomic(kaddr);
3162 	if (csum_expected == 0)
3163 		return 0;
3164 	return -EIO;
3165 }
3166 
3167 /*
3168  * when reads are done, we need to check csums to verify the data is correct
3169  * if there's a match, we allow the bio to finish.  If not, the code in
3170  * extent_io.c will try to find good copies for us.
3171  */
3172 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3173 				      u64 phy_offset, struct page *page,
3174 				      u64 start, u64 end, int mirror)
3175 {
3176 	size_t offset = start - page_offset(page);
3177 	struct inode *inode = page->mapping->host;
3178 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3179 	struct btrfs_root *root = BTRFS_I(inode)->root;
3180 
3181 	if (PageChecked(page)) {
3182 		ClearPageChecked(page);
3183 		return 0;
3184 	}
3185 
3186 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3187 		return 0;
3188 
3189 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3190 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3191 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3192 		return 0;
3193 	}
3194 
3195 	phy_offset >>= inode->i_sb->s_blocksize_bits;
3196 	return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3197 				      start, (size_t)(end - start + 1));
3198 }
3199 
3200 void btrfs_add_delayed_iput(struct inode *inode)
3201 {
3202 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3203 	struct btrfs_inode *binode = BTRFS_I(inode);
3204 
3205 	if (atomic_add_unless(&inode->i_count, -1, 1))
3206 		return;
3207 
3208 	spin_lock(&fs_info->delayed_iput_lock);
3209 	if (binode->delayed_iput_count == 0) {
3210 		ASSERT(list_empty(&binode->delayed_iput));
3211 		list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3212 	} else {
3213 		binode->delayed_iput_count++;
3214 	}
3215 	spin_unlock(&fs_info->delayed_iput_lock);
3216 }
3217 
3218 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3219 {
3220 
3221 	spin_lock(&fs_info->delayed_iput_lock);
3222 	while (!list_empty(&fs_info->delayed_iputs)) {
3223 		struct btrfs_inode *inode;
3224 
3225 		inode = list_first_entry(&fs_info->delayed_iputs,
3226 				struct btrfs_inode, delayed_iput);
3227 		if (inode->delayed_iput_count) {
3228 			inode->delayed_iput_count--;
3229 			list_move_tail(&inode->delayed_iput,
3230 					&fs_info->delayed_iputs);
3231 		} else {
3232 			list_del_init(&inode->delayed_iput);
3233 		}
3234 		spin_unlock(&fs_info->delayed_iput_lock);
3235 		iput(&inode->vfs_inode);
3236 		spin_lock(&fs_info->delayed_iput_lock);
3237 	}
3238 	spin_unlock(&fs_info->delayed_iput_lock);
3239 }
3240 
3241 /*
3242  * This is called in transaction commit time. If there are no orphan
3243  * files in the subvolume, it removes orphan item and frees block_rsv
3244  * structure.
3245  */
3246 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3247 			      struct btrfs_root *root)
3248 {
3249 	struct btrfs_fs_info *fs_info = root->fs_info;
3250 	struct btrfs_block_rsv *block_rsv;
3251 	int ret;
3252 
3253 	if (atomic_read(&root->orphan_inodes) ||
3254 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3255 		return;
3256 
3257 	spin_lock(&root->orphan_lock);
3258 	if (atomic_read(&root->orphan_inodes)) {
3259 		spin_unlock(&root->orphan_lock);
3260 		return;
3261 	}
3262 
3263 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3264 		spin_unlock(&root->orphan_lock);
3265 		return;
3266 	}
3267 
3268 	block_rsv = root->orphan_block_rsv;
3269 	root->orphan_block_rsv = NULL;
3270 	spin_unlock(&root->orphan_lock);
3271 
3272 	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3273 	    btrfs_root_refs(&root->root_item) > 0) {
3274 		ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3275 					    root->root_key.objectid);
3276 		if (ret)
3277 			btrfs_abort_transaction(trans, ret);
3278 		else
3279 			clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3280 				  &root->state);
3281 	}
3282 
3283 	if (block_rsv) {
3284 		WARN_ON(block_rsv->size > 0);
3285 		btrfs_free_block_rsv(fs_info, block_rsv);
3286 	}
3287 }
3288 
3289 /*
3290  * This creates an orphan entry for the given inode in case something goes
3291  * wrong in the middle of an unlink/truncate.
3292  *
3293  * NOTE: caller of this function should reserve 5 units of metadata for
3294  *	 this function.
3295  */
3296 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3297 		struct btrfs_inode *inode)
3298 {
3299 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3300 	struct btrfs_root *root = inode->root;
3301 	struct btrfs_block_rsv *block_rsv = NULL;
3302 	int reserve = 0;
3303 	int insert = 0;
3304 	int ret;
3305 
3306 	if (!root->orphan_block_rsv) {
3307 		block_rsv = btrfs_alloc_block_rsv(fs_info,
3308 						  BTRFS_BLOCK_RSV_TEMP);
3309 		if (!block_rsv)
3310 			return -ENOMEM;
3311 	}
3312 
3313 	spin_lock(&root->orphan_lock);
3314 	if (!root->orphan_block_rsv) {
3315 		root->orphan_block_rsv = block_rsv;
3316 	} else if (block_rsv) {
3317 		btrfs_free_block_rsv(fs_info, block_rsv);
3318 		block_rsv = NULL;
3319 	}
3320 
3321 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3322 			      &inode->runtime_flags)) {
3323 #if 0
3324 		/*
3325 		 * For proper ENOSPC handling, we should do orphan
3326 		 * cleanup when mounting. But this introduces backward
3327 		 * compatibility issue.
3328 		 */
3329 		if (!xchg(&root->orphan_item_inserted, 1))
3330 			insert = 2;
3331 		else
3332 			insert = 1;
3333 #endif
3334 		insert = 1;
3335 		atomic_inc(&root->orphan_inodes);
3336 	}
3337 
3338 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3339 			      &inode->runtime_flags))
3340 		reserve = 1;
3341 	spin_unlock(&root->orphan_lock);
3342 
3343 	/* grab metadata reservation from transaction handle */
3344 	if (reserve) {
3345 		ret = btrfs_orphan_reserve_metadata(trans, inode);
3346 		ASSERT(!ret);
3347 		if (ret) {
3348 			atomic_dec(&root->orphan_inodes);
3349 			clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3350 				  &inode->runtime_flags);
3351 			if (insert)
3352 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3353 					  &inode->runtime_flags);
3354 			return ret;
3355 		}
3356 	}
3357 
3358 	/* insert an orphan item to track this unlinked/truncated file */
3359 	if (insert >= 1) {
3360 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3361 		if (ret) {
3362 			atomic_dec(&root->orphan_inodes);
3363 			if (reserve) {
3364 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3365 					  &inode->runtime_flags);
3366 				btrfs_orphan_release_metadata(inode);
3367 			}
3368 			if (ret != -EEXIST) {
3369 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3370 					  &inode->runtime_flags);
3371 				btrfs_abort_transaction(trans, ret);
3372 				return ret;
3373 			}
3374 		}
3375 		ret = 0;
3376 	}
3377 
3378 	/* insert an orphan item to track subvolume contains orphan files */
3379 	if (insert >= 2) {
3380 		ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3381 					       root->root_key.objectid);
3382 		if (ret && ret != -EEXIST) {
3383 			btrfs_abort_transaction(trans, ret);
3384 			return ret;
3385 		}
3386 	}
3387 	return 0;
3388 }
3389 
3390 /*
3391  * We have done the truncate/delete so we can go ahead and remove the orphan
3392  * item for this particular inode.
3393  */
3394 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3395 			    struct btrfs_inode *inode)
3396 {
3397 	struct btrfs_root *root = inode->root;
3398 	int delete_item = 0;
3399 	int release_rsv = 0;
3400 	int ret = 0;
3401 
3402 	spin_lock(&root->orphan_lock);
3403 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3404 			       &inode->runtime_flags))
3405 		delete_item = 1;
3406 
3407 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3408 			       &inode->runtime_flags))
3409 		release_rsv = 1;
3410 	spin_unlock(&root->orphan_lock);
3411 
3412 	if (delete_item) {
3413 		atomic_dec(&root->orphan_inodes);
3414 		if (trans)
3415 			ret = btrfs_del_orphan_item(trans, root,
3416 						    btrfs_ino(inode));
3417 	}
3418 
3419 	if (release_rsv)
3420 		btrfs_orphan_release_metadata(inode);
3421 
3422 	return ret;
3423 }
3424 
3425 /*
3426  * this cleans up any orphans that may be left on the list from the last use
3427  * of this root.
3428  */
3429 int btrfs_orphan_cleanup(struct btrfs_root *root)
3430 {
3431 	struct btrfs_fs_info *fs_info = root->fs_info;
3432 	struct btrfs_path *path;
3433 	struct extent_buffer *leaf;
3434 	struct btrfs_key key, found_key;
3435 	struct btrfs_trans_handle *trans;
3436 	struct inode *inode;
3437 	u64 last_objectid = 0;
3438 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3439 
3440 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3441 		return 0;
3442 
3443 	path = btrfs_alloc_path();
3444 	if (!path) {
3445 		ret = -ENOMEM;
3446 		goto out;
3447 	}
3448 	path->reada = READA_BACK;
3449 
3450 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3451 	key.type = BTRFS_ORPHAN_ITEM_KEY;
3452 	key.offset = (u64)-1;
3453 
3454 	while (1) {
3455 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3456 		if (ret < 0)
3457 			goto out;
3458 
3459 		/*
3460 		 * if ret == 0 means we found what we were searching for, which
3461 		 * is weird, but possible, so only screw with path if we didn't
3462 		 * find the key and see if we have stuff that matches
3463 		 */
3464 		if (ret > 0) {
3465 			ret = 0;
3466 			if (path->slots[0] == 0)
3467 				break;
3468 			path->slots[0]--;
3469 		}
3470 
3471 		/* pull out the item */
3472 		leaf = path->nodes[0];
3473 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3474 
3475 		/* make sure the item matches what we want */
3476 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3477 			break;
3478 		if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3479 			break;
3480 
3481 		/* release the path since we're done with it */
3482 		btrfs_release_path(path);
3483 
3484 		/*
3485 		 * this is where we are basically btrfs_lookup, without the
3486 		 * crossing root thing.  we store the inode number in the
3487 		 * offset of the orphan item.
3488 		 */
3489 
3490 		if (found_key.offset == last_objectid) {
3491 			btrfs_err(fs_info,
3492 				  "Error removing orphan entry, stopping orphan cleanup");
3493 			ret = -EINVAL;
3494 			goto out;
3495 		}
3496 
3497 		last_objectid = found_key.offset;
3498 
3499 		found_key.objectid = found_key.offset;
3500 		found_key.type = BTRFS_INODE_ITEM_KEY;
3501 		found_key.offset = 0;
3502 		inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3503 		ret = PTR_ERR_OR_ZERO(inode);
3504 		if (ret && ret != -ENOENT)
3505 			goto out;
3506 
3507 		if (ret == -ENOENT && root == fs_info->tree_root) {
3508 			struct btrfs_root *dead_root;
3509 			struct btrfs_fs_info *fs_info = root->fs_info;
3510 			int is_dead_root = 0;
3511 
3512 			/*
3513 			 * this is an orphan in the tree root. Currently these
3514 			 * could come from 2 sources:
3515 			 *  a) a snapshot deletion in progress
3516 			 *  b) a free space cache inode
3517 			 * We need to distinguish those two, as the snapshot
3518 			 * orphan must not get deleted.
3519 			 * find_dead_roots already ran before us, so if this
3520 			 * is a snapshot deletion, we should find the root
3521 			 * in the dead_roots list
3522 			 */
3523 			spin_lock(&fs_info->trans_lock);
3524 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3525 					    root_list) {
3526 				if (dead_root->root_key.objectid ==
3527 				    found_key.objectid) {
3528 					is_dead_root = 1;
3529 					break;
3530 				}
3531 			}
3532 			spin_unlock(&fs_info->trans_lock);
3533 			if (is_dead_root) {
3534 				/* prevent this orphan from being found again */
3535 				key.offset = found_key.objectid - 1;
3536 				continue;
3537 			}
3538 		}
3539 		/*
3540 		 * Inode is already gone but the orphan item is still there,
3541 		 * kill the orphan item.
3542 		 */
3543 		if (ret == -ENOENT) {
3544 			trans = btrfs_start_transaction(root, 1);
3545 			if (IS_ERR(trans)) {
3546 				ret = PTR_ERR(trans);
3547 				goto out;
3548 			}
3549 			btrfs_debug(fs_info, "auto deleting %Lu",
3550 				    found_key.objectid);
3551 			ret = btrfs_del_orphan_item(trans, root,
3552 						    found_key.objectid);
3553 			btrfs_end_transaction(trans);
3554 			if (ret)
3555 				goto out;
3556 			continue;
3557 		}
3558 
3559 		/*
3560 		 * add this inode to the orphan list so btrfs_orphan_del does
3561 		 * the proper thing when we hit it
3562 		 */
3563 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3564 			&BTRFS_I(inode)->runtime_flags);
3565 		atomic_inc(&root->orphan_inodes);
3566 
3567 		/* if we have links, this was a truncate, lets do that */
3568 		if (inode->i_nlink) {
3569 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3570 				iput(inode);
3571 				continue;
3572 			}
3573 			nr_truncate++;
3574 
3575 			/* 1 for the orphan item deletion. */
3576 			trans = btrfs_start_transaction(root, 1);
3577 			if (IS_ERR(trans)) {
3578 				iput(inode);
3579 				ret = PTR_ERR(trans);
3580 				goto out;
3581 			}
3582 			ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3583 			btrfs_end_transaction(trans);
3584 			if (ret) {
3585 				iput(inode);
3586 				goto out;
3587 			}
3588 
3589 			ret = btrfs_truncate(inode);
3590 			if (ret)
3591 				btrfs_orphan_del(NULL, BTRFS_I(inode));
3592 		} else {
3593 			nr_unlink++;
3594 		}
3595 
3596 		/* this will do delete_inode and everything for us */
3597 		iput(inode);
3598 		if (ret)
3599 			goto out;
3600 	}
3601 	/* release the path since we're done with it */
3602 	btrfs_release_path(path);
3603 
3604 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3605 
3606 	if (root->orphan_block_rsv)
3607 		btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3608 					(u64)-1);
3609 
3610 	if (root->orphan_block_rsv ||
3611 	    test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3612 		trans = btrfs_join_transaction(root);
3613 		if (!IS_ERR(trans))
3614 			btrfs_end_transaction(trans);
3615 	}
3616 
3617 	if (nr_unlink)
3618 		btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3619 	if (nr_truncate)
3620 		btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3621 
3622 out:
3623 	if (ret)
3624 		btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3625 	btrfs_free_path(path);
3626 	return ret;
3627 }
3628 
3629 /*
3630  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3631  * don't find any xattrs, we know there can't be any acls.
3632  *
3633  * slot is the slot the inode is in, objectid is the objectid of the inode
3634  */
3635 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3636 					  int slot, u64 objectid,
3637 					  int *first_xattr_slot)
3638 {
3639 	u32 nritems = btrfs_header_nritems(leaf);
3640 	struct btrfs_key found_key;
3641 	static u64 xattr_access = 0;
3642 	static u64 xattr_default = 0;
3643 	int scanned = 0;
3644 
3645 	if (!xattr_access) {
3646 		xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3647 					strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3648 		xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3649 					strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3650 	}
3651 
3652 	slot++;
3653 	*first_xattr_slot = -1;
3654 	while (slot < nritems) {
3655 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3656 
3657 		/* we found a different objectid, there must not be acls */
3658 		if (found_key.objectid != objectid)
3659 			return 0;
3660 
3661 		/* we found an xattr, assume we've got an acl */
3662 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3663 			if (*first_xattr_slot == -1)
3664 				*first_xattr_slot = slot;
3665 			if (found_key.offset == xattr_access ||
3666 			    found_key.offset == xattr_default)
3667 				return 1;
3668 		}
3669 
3670 		/*
3671 		 * we found a key greater than an xattr key, there can't
3672 		 * be any acls later on
3673 		 */
3674 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3675 			return 0;
3676 
3677 		slot++;
3678 		scanned++;
3679 
3680 		/*
3681 		 * it goes inode, inode backrefs, xattrs, extents,
3682 		 * so if there are a ton of hard links to an inode there can
3683 		 * be a lot of backrefs.  Don't waste time searching too hard,
3684 		 * this is just an optimization
3685 		 */
3686 		if (scanned >= 8)
3687 			break;
3688 	}
3689 	/* we hit the end of the leaf before we found an xattr or
3690 	 * something larger than an xattr.  We have to assume the inode
3691 	 * has acls
3692 	 */
3693 	if (*first_xattr_slot == -1)
3694 		*first_xattr_slot = slot;
3695 	return 1;
3696 }
3697 
3698 /*
3699  * read an inode from the btree into the in-memory inode
3700  */
3701 static int btrfs_read_locked_inode(struct inode *inode)
3702 {
3703 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3704 	struct btrfs_path *path;
3705 	struct extent_buffer *leaf;
3706 	struct btrfs_inode_item *inode_item;
3707 	struct btrfs_root *root = BTRFS_I(inode)->root;
3708 	struct btrfs_key location;
3709 	unsigned long ptr;
3710 	int maybe_acls;
3711 	u32 rdev;
3712 	int ret;
3713 	bool filled = false;
3714 	int first_xattr_slot;
3715 
3716 	ret = btrfs_fill_inode(inode, &rdev);
3717 	if (!ret)
3718 		filled = true;
3719 
3720 	path = btrfs_alloc_path();
3721 	if (!path) {
3722 		ret = -ENOMEM;
3723 		goto make_bad;
3724 	}
3725 
3726 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3727 
3728 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3729 	if (ret) {
3730 		if (ret > 0)
3731 			ret = -ENOENT;
3732 		goto make_bad;
3733 	}
3734 
3735 	leaf = path->nodes[0];
3736 
3737 	if (filled)
3738 		goto cache_index;
3739 
3740 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3741 				    struct btrfs_inode_item);
3742 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3743 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3744 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3745 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3746 	btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3747 
3748 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3749 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3750 
3751 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3752 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3753 
3754 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3755 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3756 
3757 	BTRFS_I(inode)->i_otime.tv_sec =
3758 		btrfs_timespec_sec(leaf, &inode_item->otime);
3759 	BTRFS_I(inode)->i_otime.tv_nsec =
3760 		btrfs_timespec_nsec(leaf, &inode_item->otime);
3761 
3762 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3763 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3764 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3765 
3766 	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3767 	inode->i_generation = BTRFS_I(inode)->generation;
3768 	inode->i_rdev = 0;
3769 	rdev = btrfs_inode_rdev(leaf, inode_item);
3770 
3771 	BTRFS_I(inode)->index_cnt = (u64)-1;
3772 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3773 
3774 cache_index:
3775 	/*
3776 	 * If we were modified in the current generation and evicted from memory
3777 	 * and then re-read we need to do a full sync since we don't have any
3778 	 * idea about which extents were modified before we were evicted from
3779 	 * cache.
3780 	 *
3781 	 * This is required for both inode re-read from disk and delayed inode
3782 	 * in delayed_nodes_tree.
3783 	 */
3784 	if (BTRFS_I(inode)->last_trans == fs_info->generation)
3785 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3786 			&BTRFS_I(inode)->runtime_flags);
3787 
3788 	/*
3789 	 * We don't persist the id of the transaction where an unlink operation
3790 	 * against the inode was last made. So here we assume the inode might
3791 	 * have been evicted, and therefore the exact value of last_unlink_trans
3792 	 * lost, and set it to last_trans to avoid metadata inconsistencies
3793 	 * between the inode and its parent if the inode is fsync'ed and the log
3794 	 * replayed. For example, in the scenario:
3795 	 *
3796 	 * touch mydir/foo
3797 	 * ln mydir/foo mydir/bar
3798 	 * sync
3799 	 * unlink mydir/bar
3800 	 * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3801 	 * xfs_io -c fsync mydir/foo
3802 	 * <power failure>
3803 	 * mount fs, triggers fsync log replay
3804 	 *
3805 	 * We must make sure that when we fsync our inode foo we also log its
3806 	 * parent inode, otherwise after log replay the parent still has the
3807 	 * dentry with the "bar" name but our inode foo has a link count of 1
3808 	 * and doesn't have an inode ref with the name "bar" anymore.
3809 	 *
3810 	 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3811 	 * but it guarantees correctness at the expense of occasional full
3812 	 * transaction commits on fsync if our inode is a directory, or if our
3813 	 * inode is not a directory, logging its parent unnecessarily.
3814 	 */
3815 	BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3816 
3817 	path->slots[0]++;
3818 	if (inode->i_nlink != 1 ||
3819 	    path->slots[0] >= btrfs_header_nritems(leaf))
3820 		goto cache_acl;
3821 
3822 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3823 	if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3824 		goto cache_acl;
3825 
3826 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3827 	if (location.type == BTRFS_INODE_REF_KEY) {
3828 		struct btrfs_inode_ref *ref;
3829 
3830 		ref = (struct btrfs_inode_ref *)ptr;
3831 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3832 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3833 		struct btrfs_inode_extref *extref;
3834 
3835 		extref = (struct btrfs_inode_extref *)ptr;
3836 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3837 								     extref);
3838 	}
3839 cache_acl:
3840 	/*
3841 	 * try to precache a NULL acl entry for files that don't have
3842 	 * any xattrs or acls
3843 	 */
3844 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3845 			btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3846 	if (first_xattr_slot != -1) {
3847 		path->slots[0] = first_xattr_slot;
3848 		ret = btrfs_load_inode_props(inode, path);
3849 		if (ret)
3850 			btrfs_err(fs_info,
3851 				  "error loading props for ino %llu (root %llu): %d",
3852 				  btrfs_ino(BTRFS_I(inode)),
3853 				  root->root_key.objectid, ret);
3854 	}
3855 	btrfs_free_path(path);
3856 
3857 	if (!maybe_acls)
3858 		cache_no_acl(inode);
3859 
3860 	switch (inode->i_mode & S_IFMT) {
3861 	case S_IFREG:
3862 		inode->i_mapping->a_ops = &btrfs_aops;
3863 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3864 		inode->i_fop = &btrfs_file_operations;
3865 		inode->i_op = &btrfs_file_inode_operations;
3866 		break;
3867 	case S_IFDIR:
3868 		inode->i_fop = &btrfs_dir_file_operations;
3869 		inode->i_op = &btrfs_dir_inode_operations;
3870 		break;
3871 	case S_IFLNK:
3872 		inode->i_op = &btrfs_symlink_inode_operations;
3873 		inode_nohighmem(inode);
3874 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3875 		break;
3876 	default:
3877 		inode->i_op = &btrfs_special_inode_operations;
3878 		init_special_inode(inode, inode->i_mode, rdev);
3879 		break;
3880 	}
3881 
3882 	btrfs_update_iflags(inode);
3883 	return 0;
3884 
3885 make_bad:
3886 	btrfs_free_path(path);
3887 	make_bad_inode(inode);
3888 	return ret;
3889 }
3890 
3891 /*
3892  * given a leaf and an inode, copy the inode fields into the leaf
3893  */
3894 static void fill_inode_item(struct btrfs_trans_handle *trans,
3895 			    struct extent_buffer *leaf,
3896 			    struct btrfs_inode_item *item,
3897 			    struct inode *inode)
3898 {
3899 	struct btrfs_map_token token;
3900 
3901 	btrfs_init_map_token(&token);
3902 
3903 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3904 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3905 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3906 				   &token);
3907 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3908 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3909 
3910 	btrfs_set_token_timespec_sec(leaf, &item->atime,
3911 				     inode->i_atime.tv_sec, &token);
3912 	btrfs_set_token_timespec_nsec(leaf, &item->atime,
3913 				      inode->i_atime.tv_nsec, &token);
3914 
3915 	btrfs_set_token_timespec_sec(leaf, &item->mtime,
3916 				     inode->i_mtime.tv_sec, &token);
3917 	btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3918 				      inode->i_mtime.tv_nsec, &token);
3919 
3920 	btrfs_set_token_timespec_sec(leaf, &item->ctime,
3921 				     inode->i_ctime.tv_sec, &token);
3922 	btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3923 				      inode->i_ctime.tv_nsec, &token);
3924 
3925 	btrfs_set_token_timespec_sec(leaf, &item->otime,
3926 				     BTRFS_I(inode)->i_otime.tv_sec, &token);
3927 	btrfs_set_token_timespec_nsec(leaf, &item->otime,
3928 				      BTRFS_I(inode)->i_otime.tv_nsec, &token);
3929 
3930 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3931 				     &token);
3932 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3933 					 &token);
3934 	btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3935 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3936 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3937 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3938 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3939 }
3940 
3941 /*
3942  * copy everything in the in-memory inode into the btree.
3943  */
3944 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3945 				struct btrfs_root *root, struct inode *inode)
3946 {
3947 	struct btrfs_inode_item *inode_item;
3948 	struct btrfs_path *path;
3949 	struct extent_buffer *leaf;
3950 	int ret;
3951 
3952 	path = btrfs_alloc_path();
3953 	if (!path)
3954 		return -ENOMEM;
3955 
3956 	path->leave_spinning = 1;
3957 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3958 				 1);
3959 	if (ret) {
3960 		if (ret > 0)
3961 			ret = -ENOENT;
3962 		goto failed;
3963 	}
3964 
3965 	leaf = path->nodes[0];
3966 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3967 				    struct btrfs_inode_item);
3968 
3969 	fill_inode_item(trans, leaf, inode_item, inode);
3970 	btrfs_mark_buffer_dirty(leaf);
3971 	btrfs_set_inode_last_trans(trans, inode);
3972 	ret = 0;
3973 failed:
3974 	btrfs_free_path(path);
3975 	return ret;
3976 }
3977 
3978 /*
3979  * copy everything in the in-memory inode into the btree.
3980  */
3981 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3982 				struct btrfs_root *root, struct inode *inode)
3983 {
3984 	struct btrfs_fs_info *fs_info = root->fs_info;
3985 	int ret;
3986 
3987 	/*
3988 	 * If the inode is a free space inode, we can deadlock during commit
3989 	 * if we put it into the delayed code.
3990 	 *
3991 	 * The data relocation inode should also be directly updated
3992 	 * without delay
3993 	 */
3994 	if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3995 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3996 	    && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3997 		btrfs_update_root_times(trans, root);
3998 
3999 		ret = btrfs_delayed_update_inode(trans, root, inode);
4000 		if (!ret)
4001 			btrfs_set_inode_last_trans(trans, inode);
4002 		return ret;
4003 	}
4004 
4005 	return btrfs_update_inode_item(trans, root, inode);
4006 }
4007 
4008 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4009 					 struct btrfs_root *root,
4010 					 struct inode *inode)
4011 {
4012 	int ret;
4013 
4014 	ret = btrfs_update_inode(trans, root, inode);
4015 	if (ret == -ENOSPC)
4016 		return btrfs_update_inode_item(trans, root, inode);
4017 	return ret;
4018 }
4019 
4020 /*
4021  * unlink helper that gets used here in inode.c and in the tree logging
4022  * recovery code.  It remove a link in a directory with a given name, and
4023  * also drops the back refs in the inode to the directory
4024  */
4025 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4026 				struct btrfs_root *root,
4027 				struct btrfs_inode *dir,
4028 				struct btrfs_inode *inode,
4029 				const char *name, int name_len)
4030 {
4031 	struct btrfs_fs_info *fs_info = root->fs_info;
4032 	struct btrfs_path *path;
4033 	int ret = 0;
4034 	struct extent_buffer *leaf;
4035 	struct btrfs_dir_item *di;
4036 	struct btrfs_key key;
4037 	u64 index;
4038 	u64 ino = btrfs_ino(inode);
4039 	u64 dir_ino = btrfs_ino(dir);
4040 
4041 	path = btrfs_alloc_path();
4042 	if (!path) {
4043 		ret = -ENOMEM;
4044 		goto out;
4045 	}
4046 
4047 	path->leave_spinning = 1;
4048 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4049 				    name, name_len, -1);
4050 	if (IS_ERR(di)) {
4051 		ret = PTR_ERR(di);
4052 		goto err;
4053 	}
4054 	if (!di) {
4055 		ret = -ENOENT;
4056 		goto err;
4057 	}
4058 	leaf = path->nodes[0];
4059 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4060 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4061 	if (ret)
4062 		goto err;
4063 	btrfs_release_path(path);
4064 
4065 	/*
4066 	 * If we don't have dir index, we have to get it by looking up
4067 	 * the inode ref, since we get the inode ref, remove it directly,
4068 	 * it is unnecessary to do delayed deletion.
4069 	 *
4070 	 * But if we have dir index, needn't search inode ref to get it.
4071 	 * Since the inode ref is close to the inode item, it is better
4072 	 * that we delay to delete it, and just do this deletion when
4073 	 * we update the inode item.
4074 	 */
4075 	if (inode->dir_index) {
4076 		ret = btrfs_delayed_delete_inode_ref(inode);
4077 		if (!ret) {
4078 			index = inode->dir_index;
4079 			goto skip_backref;
4080 		}
4081 	}
4082 
4083 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4084 				  dir_ino, &index);
4085 	if (ret) {
4086 		btrfs_info(fs_info,
4087 			"failed to delete reference to %.*s, inode %llu parent %llu",
4088 			name_len, name, ino, dir_ino);
4089 		btrfs_abort_transaction(trans, ret);
4090 		goto err;
4091 	}
4092 skip_backref:
4093 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4094 	if (ret) {
4095 		btrfs_abort_transaction(trans, ret);
4096 		goto err;
4097 	}
4098 
4099 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4100 			dir_ino);
4101 	if (ret != 0 && ret != -ENOENT) {
4102 		btrfs_abort_transaction(trans, ret);
4103 		goto err;
4104 	}
4105 
4106 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4107 			index);
4108 	if (ret == -ENOENT)
4109 		ret = 0;
4110 	else if (ret)
4111 		btrfs_abort_transaction(trans, ret);
4112 err:
4113 	btrfs_free_path(path);
4114 	if (ret)
4115 		goto out;
4116 
4117 	btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4118 	inode_inc_iversion(&inode->vfs_inode);
4119 	inode_inc_iversion(&dir->vfs_inode);
4120 	inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4121 		dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4122 	ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4123 out:
4124 	return ret;
4125 }
4126 
4127 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4128 		       struct btrfs_root *root,
4129 		       struct btrfs_inode *dir, struct btrfs_inode *inode,
4130 		       const char *name, int name_len)
4131 {
4132 	int ret;
4133 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4134 	if (!ret) {
4135 		drop_nlink(&inode->vfs_inode);
4136 		ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4137 	}
4138 	return ret;
4139 }
4140 
4141 /*
4142  * helper to start transaction for unlink and rmdir.
4143  *
4144  * unlink and rmdir are special in btrfs, they do not always free space, so
4145  * if we cannot make our reservations the normal way try and see if there is
4146  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4147  * allow the unlink to occur.
4148  */
4149 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4150 {
4151 	struct btrfs_root *root = BTRFS_I(dir)->root;
4152 
4153 	/*
4154 	 * 1 for the possible orphan item
4155 	 * 1 for the dir item
4156 	 * 1 for the dir index
4157 	 * 1 for the inode ref
4158 	 * 1 for the inode
4159 	 */
4160 	return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4161 }
4162 
4163 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4164 {
4165 	struct btrfs_root *root = BTRFS_I(dir)->root;
4166 	struct btrfs_trans_handle *trans;
4167 	struct inode *inode = d_inode(dentry);
4168 	int ret;
4169 
4170 	trans = __unlink_start_trans(dir);
4171 	if (IS_ERR(trans))
4172 		return PTR_ERR(trans);
4173 
4174 	btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4175 			0);
4176 
4177 	ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4178 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4179 			dentry->d_name.len);
4180 	if (ret)
4181 		goto out;
4182 
4183 	if (inode->i_nlink == 0) {
4184 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4185 		if (ret)
4186 			goto out;
4187 	}
4188 
4189 out:
4190 	btrfs_end_transaction(trans);
4191 	btrfs_btree_balance_dirty(root->fs_info);
4192 	return ret;
4193 }
4194 
4195 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4196 			struct btrfs_root *root,
4197 			struct inode *dir, u64 objectid,
4198 			const char *name, int name_len)
4199 {
4200 	struct btrfs_fs_info *fs_info = root->fs_info;
4201 	struct btrfs_path *path;
4202 	struct extent_buffer *leaf;
4203 	struct btrfs_dir_item *di;
4204 	struct btrfs_key key;
4205 	u64 index;
4206 	int ret;
4207 	u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4208 
4209 	path = btrfs_alloc_path();
4210 	if (!path)
4211 		return -ENOMEM;
4212 
4213 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4214 				   name, name_len, -1);
4215 	if (IS_ERR_OR_NULL(di)) {
4216 		if (!di)
4217 			ret = -ENOENT;
4218 		else
4219 			ret = PTR_ERR(di);
4220 		goto out;
4221 	}
4222 
4223 	leaf = path->nodes[0];
4224 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4225 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4226 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4227 	if (ret) {
4228 		btrfs_abort_transaction(trans, ret);
4229 		goto out;
4230 	}
4231 	btrfs_release_path(path);
4232 
4233 	ret = btrfs_del_root_ref(trans, fs_info, objectid,
4234 				 root->root_key.objectid, dir_ino,
4235 				 &index, name, name_len);
4236 	if (ret < 0) {
4237 		if (ret != -ENOENT) {
4238 			btrfs_abort_transaction(trans, ret);
4239 			goto out;
4240 		}
4241 		di = btrfs_search_dir_index_item(root, path, dir_ino,
4242 						 name, name_len);
4243 		if (IS_ERR_OR_NULL(di)) {
4244 			if (!di)
4245 				ret = -ENOENT;
4246 			else
4247 				ret = PTR_ERR(di);
4248 			btrfs_abort_transaction(trans, ret);
4249 			goto out;
4250 		}
4251 
4252 		leaf = path->nodes[0];
4253 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4254 		btrfs_release_path(path);
4255 		index = key.offset;
4256 	}
4257 	btrfs_release_path(path);
4258 
4259 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4260 	if (ret) {
4261 		btrfs_abort_transaction(trans, ret);
4262 		goto out;
4263 	}
4264 
4265 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4266 	inode_inc_iversion(dir);
4267 	dir->i_mtime = dir->i_ctime = current_time(dir);
4268 	ret = btrfs_update_inode_fallback(trans, root, dir);
4269 	if (ret)
4270 		btrfs_abort_transaction(trans, ret);
4271 out:
4272 	btrfs_free_path(path);
4273 	return ret;
4274 }
4275 
4276 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4277 {
4278 	struct inode *inode = d_inode(dentry);
4279 	int err = 0;
4280 	struct btrfs_root *root = BTRFS_I(dir)->root;
4281 	struct btrfs_trans_handle *trans;
4282 	u64 last_unlink_trans;
4283 
4284 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4285 		return -ENOTEMPTY;
4286 	if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4287 		return -EPERM;
4288 
4289 	trans = __unlink_start_trans(dir);
4290 	if (IS_ERR(trans))
4291 		return PTR_ERR(trans);
4292 
4293 	if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4294 		err = btrfs_unlink_subvol(trans, root, dir,
4295 					  BTRFS_I(inode)->location.objectid,
4296 					  dentry->d_name.name,
4297 					  dentry->d_name.len);
4298 		goto out;
4299 	}
4300 
4301 	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4302 	if (err)
4303 		goto out;
4304 
4305 	last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4306 
4307 	/* now the directory is empty */
4308 	err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4309 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4310 			dentry->d_name.len);
4311 	if (!err) {
4312 		btrfs_i_size_write(BTRFS_I(inode), 0);
4313 		/*
4314 		 * Propagate the last_unlink_trans value of the deleted dir to
4315 		 * its parent directory. This is to prevent an unrecoverable
4316 		 * log tree in the case we do something like this:
4317 		 * 1) create dir foo
4318 		 * 2) create snapshot under dir foo
4319 		 * 3) delete the snapshot
4320 		 * 4) rmdir foo
4321 		 * 5) mkdir foo
4322 		 * 6) fsync foo or some file inside foo
4323 		 */
4324 		if (last_unlink_trans >= trans->transid)
4325 			BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4326 	}
4327 out:
4328 	btrfs_end_transaction(trans);
4329 	btrfs_btree_balance_dirty(root->fs_info);
4330 
4331 	return err;
4332 }
4333 
4334 static int truncate_space_check(struct btrfs_trans_handle *trans,
4335 				struct btrfs_root *root,
4336 				u64 bytes_deleted)
4337 {
4338 	struct btrfs_fs_info *fs_info = root->fs_info;
4339 	int ret;
4340 
4341 	/*
4342 	 * This is only used to apply pressure to the enospc system, we don't
4343 	 * intend to use this reservation at all.
4344 	 */
4345 	bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4346 	bytes_deleted *= fs_info->nodesize;
4347 	ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4348 				  bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4349 	if (!ret) {
4350 		trace_btrfs_space_reservation(fs_info, "transaction",
4351 					      trans->transid,
4352 					      bytes_deleted, 1);
4353 		trans->bytes_reserved += bytes_deleted;
4354 	}
4355 	return ret;
4356 
4357 }
4358 
4359 static int truncate_inline_extent(struct inode *inode,
4360 				  struct btrfs_path *path,
4361 				  struct btrfs_key *found_key,
4362 				  const u64 item_end,
4363 				  const u64 new_size)
4364 {
4365 	struct extent_buffer *leaf = path->nodes[0];
4366 	int slot = path->slots[0];
4367 	struct btrfs_file_extent_item *fi;
4368 	u32 size = (u32)(new_size - found_key->offset);
4369 	struct btrfs_root *root = BTRFS_I(inode)->root;
4370 
4371 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4372 
4373 	if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4374 		loff_t offset = new_size;
4375 		loff_t page_end = ALIGN(offset, PAGE_SIZE);
4376 
4377 		/*
4378 		 * Zero out the remaining of the last page of our inline extent,
4379 		 * instead of directly truncating our inline extent here - that
4380 		 * would be much more complex (decompressing all the data, then
4381 		 * compressing the truncated data, which might be bigger than
4382 		 * the size of the inline extent, resize the extent, etc).
4383 		 * We release the path because to get the page we might need to
4384 		 * read the extent item from disk (data not in the page cache).
4385 		 */
4386 		btrfs_release_path(path);
4387 		return btrfs_truncate_block(inode, offset, page_end - offset,
4388 					0);
4389 	}
4390 
4391 	btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4392 	size = btrfs_file_extent_calc_inline_size(size);
4393 	btrfs_truncate_item(root->fs_info, path, size, 1);
4394 
4395 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4396 		inode_sub_bytes(inode, item_end + 1 - new_size);
4397 
4398 	return 0;
4399 }
4400 
4401 /*
4402  * this can truncate away extent items, csum items and directory items.
4403  * It starts at a high offset and removes keys until it can't find
4404  * any higher than new_size
4405  *
4406  * csum items that cross the new i_size are truncated to the new size
4407  * as well.
4408  *
4409  * min_type is the minimum key type to truncate down to.  If set to 0, this
4410  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4411  */
4412 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4413 			       struct btrfs_root *root,
4414 			       struct inode *inode,
4415 			       u64 new_size, u32 min_type)
4416 {
4417 	struct btrfs_fs_info *fs_info = root->fs_info;
4418 	struct btrfs_path *path;
4419 	struct extent_buffer *leaf;
4420 	struct btrfs_file_extent_item *fi;
4421 	struct btrfs_key key;
4422 	struct btrfs_key found_key;
4423 	u64 extent_start = 0;
4424 	u64 extent_num_bytes = 0;
4425 	u64 extent_offset = 0;
4426 	u64 item_end = 0;
4427 	u64 last_size = new_size;
4428 	u32 found_type = (u8)-1;
4429 	int found_extent;
4430 	int del_item;
4431 	int pending_del_nr = 0;
4432 	int pending_del_slot = 0;
4433 	int extent_type = -1;
4434 	int ret;
4435 	int err = 0;
4436 	u64 ino = btrfs_ino(BTRFS_I(inode));
4437 	u64 bytes_deleted = 0;
4438 	bool be_nice = 0;
4439 	bool should_throttle = 0;
4440 	bool should_end = 0;
4441 
4442 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4443 
4444 	/*
4445 	 * for non-free space inodes and ref cows, we want to back off from
4446 	 * time to time
4447 	 */
4448 	if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4449 	    test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4450 		be_nice = 1;
4451 
4452 	path = btrfs_alloc_path();
4453 	if (!path)
4454 		return -ENOMEM;
4455 	path->reada = READA_BACK;
4456 
4457 	/*
4458 	 * We want to drop from the next block forward in case this new size is
4459 	 * not block aligned since we will be keeping the last block of the
4460 	 * extent just the way it is.
4461 	 */
4462 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4463 	    root == fs_info->tree_root)
4464 		btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4465 					fs_info->sectorsize),
4466 					(u64)-1, 0);
4467 
4468 	/*
4469 	 * This function is also used to drop the items in the log tree before
4470 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4471 	 * it is used to drop the loged items. So we shouldn't kill the delayed
4472 	 * items.
4473 	 */
4474 	if (min_type == 0 && root == BTRFS_I(inode)->root)
4475 		btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4476 
4477 	key.objectid = ino;
4478 	key.offset = (u64)-1;
4479 	key.type = (u8)-1;
4480 
4481 search_again:
4482 	/*
4483 	 * with a 16K leaf size and 128MB extents, you can actually queue
4484 	 * up a huge file in a single leaf.  Most of the time that
4485 	 * bytes_deleted is > 0, it will be huge by the time we get here
4486 	 */
4487 	if (be_nice && bytes_deleted > SZ_32M) {
4488 		if (btrfs_should_end_transaction(trans)) {
4489 			err = -EAGAIN;
4490 			goto error;
4491 		}
4492 	}
4493 
4494 
4495 	path->leave_spinning = 1;
4496 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4497 	if (ret < 0) {
4498 		err = ret;
4499 		goto out;
4500 	}
4501 
4502 	if (ret > 0) {
4503 		/* there are no items in the tree for us to truncate, we're
4504 		 * done
4505 		 */
4506 		if (path->slots[0] == 0)
4507 			goto out;
4508 		path->slots[0]--;
4509 	}
4510 
4511 	while (1) {
4512 		fi = NULL;
4513 		leaf = path->nodes[0];
4514 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4515 		found_type = found_key.type;
4516 
4517 		if (found_key.objectid != ino)
4518 			break;
4519 
4520 		if (found_type < min_type)
4521 			break;
4522 
4523 		item_end = found_key.offset;
4524 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4525 			fi = btrfs_item_ptr(leaf, path->slots[0],
4526 					    struct btrfs_file_extent_item);
4527 			extent_type = btrfs_file_extent_type(leaf, fi);
4528 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4529 				item_end +=
4530 				    btrfs_file_extent_num_bytes(leaf, fi);
4531 
4532 				trace_btrfs_truncate_show_fi_regular(
4533 					BTRFS_I(inode), leaf, fi,
4534 					found_key.offset);
4535 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4536 				item_end += btrfs_file_extent_inline_len(leaf,
4537 							 path->slots[0], fi);
4538 
4539 				trace_btrfs_truncate_show_fi_inline(
4540 					BTRFS_I(inode), leaf, fi, path->slots[0],
4541 					found_key.offset);
4542 			}
4543 			item_end--;
4544 		}
4545 		if (found_type > min_type) {
4546 			del_item = 1;
4547 		} else {
4548 			if (item_end < new_size)
4549 				break;
4550 			if (found_key.offset >= new_size)
4551 				del_item = 1;
4552 			else
4553 				del_item = 0;
4554 		}
4555 		found_extent = 0;
4556 		/* FIXME, shrink the extent if the ref count is only 1 */
4557 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4558 			goto delete;
4559 
4560 		if (del_item)
4561 			last_size = found_key.offset;
4562 		else
4563 			last_size = new_size;
4564 
4565 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4566 			u64 num_dec;
4567 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4568 			if (!del_item) {
4569 				u64 orig_num_bytes =
4570 					btrfs_file_extent_num_bytes(leaf, fi);
4571 				extent_num_bytes = ALIGN(new_size -
4572 						found_key.offset,
4573 						fs_info->sectorsize);
4574 				btrfs_set_file_extent_num_bytes(leaf, fi,
4575 							 extent_num_bytes);
4576 				num_dec = (orig_num_bytes -
4577 					   extent_num_bytes);
4578 				if (test_bit(BTRFS_ROOT_REF_COWS,
4579 					     &root->state) &&
4580 				    extent_start != 0)
4581 					inode_sub_bytes(inode, num_dec);
4582 				btrfs_mark_buffer_dirty(leaf);
4583 			} else {
4584 				extent_num_bytes =
4585 					btrfs_file_extent_disk_num_bytes(leaf,
4586 									 fi);
4587 				extent_offset = found_key.offset -
4588 					btrfs_file_extent_offset(leaf, fi);
4589 
4590 				/* FIXME blocksize != 4096 */
4591 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4592 				if (extent_start != 0) {
4593 					found_extent = 1;
4594 					if (test_bit(BTRFS_ROOT_REF_COWS,
4595 						     &root->state))
4596 						inode_sub_bytes(inode, num_dec);
4597 				}
4598 			}
4599 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4600 			/*
4601 			 * we can't truncate inline items that have had
4602 			 * special encodings
4603 			 */
4604 			if (!del_item &&
4605 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4606 			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4607 
4608 				/*
4609 				 * Need to release path in order to truncate a
4610 				 * compressed extent. So delete any accumulated
4611 				 * extent items so far.
4612 				 */
4613 				if (btrfs_file_extent_compression(leaf, fi) !=
4614 				    BTRFS_COMPRESS_NONE && pending_del_nr) {
4615 					err = btrfs_del_items(trans, root, path,
4616 							      pending_del_slot,
4617 							      pending_del_nr);
4618 					if (err) {
4619 						btrfs_abort_transaction(trans,
4620 									err);
4621 						goto error;
4622 					}
4623 					pending_del_nr = 0;
4624 				}
4625 
4626 				err = truncate_inline_extent(inode, path,
4627 							     &found_key,
4628 							     item_end,
4629 							     new_size);
4630 				if (err) {
4631 					btrfs_abort_transaction(trans, err);
4632 					goto error;
4633 				}
4634 			} else if (test_bit(BTRFS_ROOT_REF_COWS,
4635 					    &root->state)) {
4636 				inode_sub_bytes(inode, item_end + 1 - new_size);
4637 			}
4638 		}
4639 delete:
4640 		if (del_item) {
4641 			if (!pending_del_nr) {
4642 				/* no pending yet, add ourselves */
4643 				pending_del_slot = path->slots[0];
4644 				pending_del_nr = 1;
4645 			} else if (pending_del_nr &&
4646 				   path->slots[0] + 1 == pending_del_slot) {
4647 				/* hop on the pending chunk */
4648 				pending_del_nr++;
4649 				pending_del_slot = path->slots[0];
4650 			} else {
4651 				BUG();
4652 			}
4653 		} else {
4654 			break;
4655 		}
4656 		should_throttle = 0;
4657 
4658 		if (found_extent &&
4659 		    (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4660 		     root == fs_info->tree_root)) {
4661 			btrfs_set_path_blocking(path);
4662 			bytes_deleted += extent_num_bytes;
4663 			ret = btrfs_free_extent(trans, fs_info, extent_start,
4664 						extent_num_bytes, 0,
4665 						btrfs_header_owner(leaf),
4666 						ino, extent_offset);
4667 			BUG_ON(ret);
4668 			if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4669 				btrfs_async_run_delayed_refs(fs_info,
4670 					trans->delayed_ref_updates * 2,
4671 					trans->transid, 0);
4672 			if (be_nice) {
4673 				if (truncate_space_check(trans, root,
4674 							 extent_num_bytes)) {
4675 					should_end = 1;
4676 				}
4677 				if (btrfs_should_throttle_delayed_refs(trans,
4678 								       fs_info))
4679 					should_throttle = 1;
4680 			}
4681 		}
4682 
4683 		if (found_type == BTRFS_INODE_ITEM_KEY)
4684 			break;
4685 
4686 		if (path->slots[0] == 0 ||
4687 		    path->slots[0] != pending_del_slot ||
4688 		    should_throttle || should_end) {
4689 			if (pending_del_nr) {
4690 				ret = btrfs_del_items(trans, root, path,
4691 						pending_del_slot,
4692 						pending_del_nr);
4693 				if (ret) {
4694 					btrfs_abort_transaction(trans, ret);
4695 					goto error;
4696 				}
4697 				pending_del_nr = 0;
4698 			}
4699 			btrfs_release_path(path);
4700 			if (should_throttle) {
4701 				unsigned long updates = trans->delayed_ref_updates;
4702 				if (updates) {
4703 					trans->delayed_ref_updates = 0;
4704 					ret = btrfs_run_delayed_refs(trans,
4705 								   fs_info,
4706 								   updates * 2);
4707 					if (ret && !err)
4708 						err = ret;
4709 				}
4710 			}
4711 			/*
4712 			 * if we failed to refill our space rsv, bail out
4713 			 * and let the transaction restart
4714 			 */
4715 			if (should_end) {
4716 				err = -EAGAIN;
4717 				goto error;
4718 			}
4719 			goto search_again;
4720 		} else {
4721 			path->slots[0]--;
4722 		}
4723 	}
4724 out:
4725 	if (pending_del_nr) {
4726 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4727 				      pending_del_nr);
4728 		if (ret)
4729 			btrfs_abort_transaction(trans, ret);
4730 	}
4731 error:
4732 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4733 		ASSERT(last_size >= new_size);
4734 		if (!err && last_size > new_size)
4735 			last_size = new_size;
4736 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4737 	}
4738 
4739 	btrfs_free_path(path);
4740 
4741 	if (be_nice && bytes_deleted > SZ_32M) {
4742 		unsigned long updates = trans->delayed_ref_updates;
4743 		if (updates) {
4744 			trans->delayed_ref_updates = 0;
4745 			ret = btrfs_run_delayed_refs(trans, fs_info,
4746 						     updates * 2);
4747 			if (ret && !err)
4748 				err = ret;
4749 		}
4750 	}
4751 	return err;
4752 }
4753 
4754 /*
4755  * btrfs_truncate_block - read, zero a chunk and write a block
4756  * @inode - inode that we're zeroing
4757  * @from - the offset to start zeroing
4758  * @len - the length to zero, 0 to zero the entire range respective to the
4759  *	offset
4760  * @front - zero up to the offset instead of from the offset on
4761  *
4762  * This will find the block for the "from" offset and cow the block and zero the
4763  * part we want to zero.  This is used with truncate and hole punching.
4764  */
4765 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4766 			int front)
4767 {
4768 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4769 	struct address_space *mapping = inode->i_mapping;
4770 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4771 	struct btrfs_ordered_extent *ordered;
4772 	struct extent_state *cached_state = NULL;
4773 	struct extent_changeset *data_reserved = NULL;
4774 	char *kaddr;
4775 	u32 blocksize = fs_info->sectorsize;
4776 	pgoff_t index = from >> PAGE_SHIFT;
4777 	unsigned offset = from & (blocksize - 1);
4778 	struct page *page;
4779 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4780 	int ret = 0;
4781 	u64 block_start;
4782 	u64 block_end;
4783 
4784 	if ((offset & (blocksize - 1)) == 0 &&
4785 	    (!len || ((len & (blocksize - 1)) == 0)))
4786 		goto out;
4787 
4788 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4789 			round_down(from, blocksize), blocksize);
4790 	if (ret)
4791 		goto out;
4792 
4793 again:
4794 	page = find_or_create_page(mapping, index, mask);
4795 	if (!page) {
4796 		btrfs_delalloc_release_space(inode, data_reserved,
4797 				round_down(from, blocksize),
4798 				blocksize);
4799 		ret = -ENOMEM;
4800 		goto out;
4801 	}
4802 
4803 	block_start = round_down(from, blocksize);
4804 	block_end = block_start + blocksize - 1;
4805 
4806 	if (!PageUptodate(page)) {
4807 		ret = btrfs_readpage(NULL, page);
4808 		lock_page(page);
4809 		if (page->mapping != mapping) {
4810 			unlock_page(page);
4811 			put_page(page);
4812 			goto again;
4813 		}
4814 		if (!PageUptodate(page)) {
4815 			ret = -EIO;
4816 			goto out_unlock;
4817 		}
4818 	}
4819 	wait_on_page_writeback(page);
4820 
4821 	lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4822 	set_page_extent_mapped(page);
4823 
4824 	ordered = btrfs_lookup_ordered_extent(inode, block_start);
4825 	if (ordered) {
4826 		unlock_extent_cached(io_tree, block_start, block_end,
4827 				     &cached_state, GFP_NOFS);
4828 		unlock_page(page);
4829 		put_page(page);
4830 		btrfs_start_ordered_extent(inode, ordered, 1);
4831 		btrfs_put_ordered_extent(ordered);
4832 		goto again;
4833 	}
4834 
4835 	clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4836 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4837 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4838 			  0, 0, &cached_state, GFP_NOFS);
4839 
4840 	ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4841 					&cached_state, 0);
4842 	if (ret) {
4843 		unlock_extent_cached(io_tree, block_start, block_end,
4844 				     &cached_state, GFP_NOFS);
4845 		goto out_unlock;
4846 	}
4847 
4848 	if (offset != blocksize) {
4849 		if (!len)
4850 			len = blocksize - offset;
4851 		kaddr = kmap(page);
4852 		if (front)
4853 			memset(kaddr + (block_start - page_offset(page)),
4854 				0, offset);
4855 		else
4856 			memset(kaddr + (block_start - page_offset(page)) +  offset,
4857 				0, len);
4858 		flush_dcache_page(page);
4859 		kunmap(page);
4860 	}
4861 	ClearPageChecked(page);
4862 	set_page_dirty(page);
4863 	unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4864 			     GFP_NOFS);
4865 
4866 out_unlock:
4867 	if (ret)
4868 		btrfs_delalloc_release_space(inode, data_reserved, block_start,
4869 					     blocksize);
4870 	unlock_page(page);
4871 	put_page(page);
4872 out:
4873 	extent_changeset_free(data_reserved);
4874 	return ret;
4875 }
4876 
4877 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4878 			     u64 offset, u64 len)
4879 {
4880 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4881 	struct btrfs_trans_handle *trans;
4882 	int ret;
4883 
4884 	/*
4885 	 * Still need to make sure the inode looks like it's been updated so
4886 	 * that any holes get logged if we fsync.
4887 	 */
4888 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4889 		BTRFS_I(inode)->last_trans = fs_info->generation;
4890 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4891 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4892 		return 0;
4893 	}
4894 
4895 	/*
4896 	 * 1 - for the one we're dropping
4897 	 * 1 - for the one we're adding
4898 	 * 1 - for updating the inode.
4899 	 */
4900 	trans = btrfs_start_transaction(root, 3);
4901 	if (IS_ERR(trans))
4902 		return PTR_ERR(trans);
4903 
4904 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4905 	if (ret) {
4906 		btrfs_abort_transaction(trans, ret);
4907 		btrfs_end_transaction(trans);
4908 		return ret;
4909 	}
4910 
4911 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4912 			offset, 0, 0, len, 0, len, 0, 0, 0);
4913 	if (ret)
4914 		btrfs_abort_transaction(trans, ret);
4915 	else
4916 		btrfs_update_inode(trans, root, inode);
4917 	btrfs_end_transaction(trans);
4918 	return ret;
4919 }
4920 
4921 /*
4922  * This function puts in dummy file extents for the area we're creating a hole
4923  * for.  So if we are truncating this file to a larger size we need to insert
4924  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4925  * the range between oldsize and size
4926  */
4927 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4928 {
4929 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4930 	struct btrfs_root *root = BTRFS_I(inode)->root;
4931 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4932 	struct extent_map *em = NULL;
4933 	struct extent_state *cached_state = NULL;
4934 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4935 	u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4936 	u64 block_end = ALIGN(size, fs_info->sectorsize);
4937 	u64 last_byte;
4938 	u64 cur_offset;
4939 	u64 hole_size;
4940 	int err = 0;
4941 
4942 	/*
4943 	 * If our size started in the middle of a block we need to zero out the
4944 	 * rest of the block before we expand the i_size, otherwise we could
4945 	 * expose stale data.
4946 	 */
4947 	err = btrfs_truncate_block(inode, oldsize, 0, 0);
4948 	if (err)
4949 		return err;
4950 
4951 	if (size <= hole_start)
4952 		return 0;
4953 
4954 	while (1) {
4955 		struct btrfs_ordered_extent *ordered;
4956 
4957 		lock_extent_bits(io_tree, hole_start, block_end - 1,
4958 				 &cached_state);
4959 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4960 						     block_end - hole_start);
4961 		if (!ordered)
4962 			break;
4963 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4964 				     &cached_state, GFP_NOFS);
4965 		btrfs_start_ordered_extent(inode, ordered, 1);
4966 		btrfs_put_ordered_extent(ordered);
4967 	}
4968 
4969 	cur_offset = hole_start;
4970 	while (1) {
4971 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4972 				block_end - cur_offset, 0);
4973 		if (IS_ERR(em)) {
4974 			err = PTR_ERR(em);
4975 			em = NULL;
4976 			break;
4977 		}
4978 		last_byte = min(extent_map_end(em), block_end);
4979 		last_byte = ALIGN(last_byte, fs_info->sectorsize);
4980 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4981 			struct extent_map *hole_em;
4982 			hole_size = last_byte - cur_offset;
4983 
4984 			err = maybe_insert_hole(root, inode, cur_offset,
4985 						hole_size);
4986 			if (err)
4987 				break;
4988 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4989 						cur_offset + hole_size - 1, 0);
4990 			hole_em = alloc_extent_map();
4991 			if (!hole_em) {
4992 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4993 					&BTRFS_I(inode)->runtime_flags);
4994 				goto next;
4995 			}
4996 			hole_em->start = cur_offset;
4997 			hole_em->len = hole_size;
4998 			hole_em->orig_start = cur_offset;
4999 
5000 			hole_em->block_start = EXTENT_MAP_HOLE;
5001 			hole_em->block_len = 0;
5002 			hole_em->orig_block_len = 0;
5003 			hole_em->ram_bytes = hole_size;
5004 			hole_em->bdev = fs_info->fs_devices->latest_bdev;
5005 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
5006 			hole_em->generation = fs_info->generation;
5007 
5008 			while (1) {
5009 				write_lock(&em_tree->lock);
5010 				err = add_extent_mapping(em_tree, hole_em, 1);
5011 				write_unlock(&em_tree->lock);
5012 				if (err != -EEXIST)
5013 					break;
5014 				btrfs_drop_extent_cache(BTRFS_I(inode),
5015 							cur_offset,
5016 							cur_offset +
5017 							hole_size - 1, 0);
5018 			}
5019 			free_extent_map(hole_em);
5020 		}
5021 next:
5022 		free_extent_map(em);
5023 		em = NULL;
5024 		cur_offset = last_byte;
5025 		if (cur_offset >= block_end)
5026 			break;
5027 	}
5028 	free_extent_map(em);
5029 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5030 			     GFP_NOFS);
5031 	return err;
5032 }
5033 
5034 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5035 {
5036 	struct btrfs_root *root = BTRFS_I(inode)->root;
5037 	struct btrfs_trans_handle *trans;
5038 	loff_t oldsize = i_size_read(inode);
5039 	loff_t newsize = attr->ia_size;
5040 	int mask = attr->ia_valid;
5041 	int ret;
5042 
5043 	/*
5044 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5045 	 * special case where we need to update the times despite not having
5046 	 * these flags set.  For all other operations the VFS set these flags
5047 	 * explicitly if it wants a timestamp update.
5048 	 */
5049 	if (newsize != oldsize) {
5050 		inode_inc_iversion(inode);
5051 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5052 			inode->i_ctime = inode->i_mtime =
5053 				current_time(inode);
5054 	}
5055 
5056 	if (newsize > oldsize) {
5057 		/*
5058 		 * Don't do an expanding truncate while snapshoting is ongoing.
5059 		 * This is to ensure the snapshot captures a fully consistent
5060 		 * state of this file - if the snapshot captures this expanding
5061 		 * truncation, it must capture all writes that happened before
5062 		 * this truncation.
5063 		 */
5064 		btrfs_wait_for_snapshot_creation(root);
5065 		ret = btrfs_cont_expand(inode, oldsize, newsize);
5066 		if (ret) {
5067 			btrfs_end_write_no_snapshoting(root);
5068 			return ret;
5069 		}
5070 
5071 		trans = btrfs_start_transaction(root, 1);
5072 		if (IS_ERR(trans)) {
5073 			btrfs_end_write_no_snapshoting(root);
5074 			return PTR_ERR(trans);
5075 		}
5076 
5077 		i_size_write(inode, newsize);
5078 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5079 		pagecache_isize_extended(inode, oldsize, newsize);
5080 		ret = btrfs_update_inode(trans, root, inode);
5081 		btrfs_end_write_no_snapshoting(root);
5082 		btrfs_end_transaction(trans);
5083 	} else {
5084 
5085 		/*
5086 		 * We're truncating a file that used to have good data down to
5087 		 * zero. Make sure it gets into the ordered flush list so that
5088 		 * any new writes get down to disk quickly.
5089 		 */
5090 		if (newsize == 0)
5091 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5092 				&BTRFS_I(inode)->runtime_flags);
5093 
5094 		/*
5095 		 * 1 for the orphan item we're going to add
5096 		 * 1 for the orphan item deletion.
5097 		 */
5098 		trans = btrfs_start_transaction(root, 2);
5099 		if (IS_ERR(trans))
5100 			return PTR_ERR(trans);
5101 
5102 		/*
5103 		 * We need to do this in case we fail at _any_ point during the
5104 		 * actual truncate.  Once we do the truncate_setsize we could
5105 		 * invalidate pages which forces any outstanding ordered io to
5106 		 * be instantly completed which will give us extents that need
5107 		 * to be truncated.  If we fail to get an orphan inode down we
5108 		 * could have left over extents that were never meant to live,
5109 		 * so we need to guarantee from this point on that everything
5110 		 * will be consistent.
5111 		 */
5112 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5113 		btrfs_end_transaction(trans);
5114 		if (ret)
5115 			return ret;
5116 
5117 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
5118 		truncate_setsize(inode, newsize);
5119 
5120 		/* Disable nonlocked read DIO to avoid the end less truncate */
5121 		btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5122 		inode_dio_wait(inode);
5123 		btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5124 
5125 		ret = btrfs_truncate(inode);
5126 		if (ret && inode->i_nlink) {
5127 			int err;
5128 
5129 			/* To get a stable disk_i_size */
5130 			err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5131 			if (err) {
5132 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5133 				return err;
5134 			}
5135 
5136 			/*
5137 			 * failed to truncate, disk_i_size is only adjusted down
5138 			 * as we remove extents, so it should represent the true
5139 			 * size of the inode, so reset the in memory size and
5140 			 * delete our orphan entry.
5141 			 */
5142 			trans = btrfs_join_transaction(root);
5143 			if (IS_ERR(trans)) {
5144 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5145 				return ret;
5146 			}
5147 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5148 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
5149 			if (err)
5150 				btrfs_abort_transaction(trans, err);
5151 			btrfs_end_transaction(trans);
5152 		}
5153 	}
5154 
5155 	return ret;
5156 }
5157 
5158 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5159 {
5160 	struct inode *inode = d_inode(dentry);
5161 	struct btrfs_root *root = BTRFS_I(inode)->root;
5162 	int err;
5163 
5164 	if (btrfs_root_readonly(root))
5165 		return -EROFS;
5166 
5167 	err = setattr_prepare(dentry, attr);
5168 	if (err)
5169 		return err;
5170 
5171 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5172 		err = btrfs_setsize(inode, attr);
5173 		if (err)
5174 			return err;
5175 	}
5176 
5177 	if (attr->ia_valid) {
5178 		setattr_copy(inode, attr);
5179 		inode_inc_iversion(inode);
5180 		err = btrfs_dirty_inode(inode);
5181 
5182 		if (!err && attr->ia_valid & ATTR_MODE)
5183 			err = posix_acl_chmod(inode, inode->i_mode);
5184 	}
5185 
5186 	return err;
5187 }
5188 
5189 /*
5190  * While truncating the inode pages during eviction, we get the VFS calling
5191  * btrfs_invalidatepage() against each page of the inode. This is slow because
5192  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5193  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5194  * extent_state structures over and over, wasting lots of time.
5195  *
5196  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5197  * those expensive operations on a per page basis and do only the ordered io
5198  * finishing, while we release here the extent_map and extent_state structures,
5199  * without the excessive merging and splitting.
5200  */
5201 static void evict_inode_truncate_pages(struct inode *inode)
5202 {
5203 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5204 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5205 	struct rb_node *node;
5206 
5207 	ASSERT(inode->i_state & I_FREEING);
5208 	truncate_inode_pages_final(&inode->i_data);
5209 
5210 	write_lock(&map_tree->lock);
5211 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
5212 		struct extent_map *em;
5213 
5214 		node = rb_first(&map_tree->map);
5215 		em = rb_entry(node, struct extent_map, rb_node);
5216 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5217 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5218 		remove_extent_mapping(map_tree, em);
5219 		free_extent_map(em);
5220 		if (need_resched()) {
5221 			write_unlock(&map_tree->lock);
5222 			cond_resched();
5223 			write_lock(&map_tree->lock);
5224 		}
5225 	}
5226 	write_unlock(&map_tree->lock);
5227 
5228 	/*
5229 	 * Keep looping until we have no more ranges in the io tree.
5230 	 * We can have ongoing bios started by readpages (called from readahead)
5231 	 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5232 	 * still in progress (unlocked the pages in the bio but did not yet
5233 	 * unlocked the ranges in the io tree). Therefore this means some
5234 	 * ranges can still be locked and eviction started because before
5235 	 * submitting those bios, which are executed by a separate task (work
5236 	 * queue kthread), inode references (inode->i_count) were not taken
5237 	 * (which would be dropped in the end io callback of each bio).
5238 	 * Therefore here we effectively end up waiting for those bios and
5239 	 * anyone else holding locked ranges without having bumped the inode's
5240 	 * reference count - if we don't do it, when they access the inode's
5241 	 * io_tree to unlock a range it may be too late, leading to an
5242 	 * use-after-free issue.
5243 	 */
5244 	spin_lock(&io_tree->lock);
5245 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
5246 		struct extent_state *state;
5247 		struct extent_state *cached_state = NULL;
5248 		u64 start;
5249 		u64 end;
5250 
5251 		node = rb_first(&io_tree->state);
5252 		state = rb_entry(node, struct extent_state, rb_node);
5253 		start = state->start;
5254 		end = state->end;
5255 		spin_unlock(&io_tree->lock);
5256 
5257 		lock_extent_bits(io_tree, start, end, &cached_state);
5258 
5259 		/*
5260 		 * If still has DELALLOC flag, the extent didn't reach disk,
5261 		 * and its reserved space won't be freed by delayed_ref.
5262 		 * So we need to free its reserved space here.
5263 		 * (Refer to comment in btrfs_invalidatepage, case 2)
5264 		 *
5265 		 * Note, end is the bytenr of last byte, so we need + 1 here.
5266 		 */
5267 		if (state->state & EXTENT_DELALLOC)
5268 			btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5269 
5270 		clear_extent_bit(io_tree, start, end,
5271 				 EXTENT_LOCKED | EXTENT_DIRTY |
5272 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5273 				 EXTENT_DEFRAG, 1, 1,
5274 				 &cached_state, GFP_NOFS);
5275 
5276 		cond_resched();
5277 		spin_lock(&io_tree->lock);
5278 	}
5279 	spin_unlock(&io_tree->lock);
5280 }
5281 
5282 void btrfs_evict_inode(struct inode *inode)
5283 {
5284 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5285 	struct btrfs_trans_handle *trans;
5286 	struct btrfs_root *root = BTRFS_I(inode)->root;
5287 	struct btrfs_block_rsv *rsv, *global_rsv;
5288 	int steal_from_global = 0;
5289 	u64 min_size;
5290 	int ret;
5291 
5292 	trace_btrfs_inode_evict(inode);
5293 
5294 	if (!root) {
5295 		kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5296 		return;
5297 	}
5298 
5299 	min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5300 
5301 	evict_inode_truncate_pages(inode);
5302 
5303 	if (inode->i_nlink &&
5304 	    ((btrfs_root_refs(&root->root_item) != 0 &&
5305 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5306 	     btrfs_is_free_space_inode(BTRFS_I(inode))))
5307 		goto no_delete;
5308 
5309 	if (is_bad_inode(inode)) {
5310 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5311 		goto no_delete;
5312 	}
5313 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5314 	if (!special_file(inode->i_mode))
5315 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
5316 
5317 	btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5318 
5319 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5320 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5321 				 &BTRFS_I(inode)->runtime_flags));
5322 		goto no_delete;
5323 	}
5324 
5325 	if (inode->i_nlink > 0) {
5326 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5327 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5328 		goto no_delete;
5329 	}
5330 
5331 	ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5332 	if (ret) {
5333 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5334 		goto no_delete;
5335 	}
5336 
5337 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5338 	if (!rsv) {
5339 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5340 		goto no_delete;
5341 	}
5342 	rsv->size = min_size;
5343 	rsv->failfast = 1;
5344 	global_rsv = &fs_info->global_block_rsv;
5345 
5346 	btrfs_i_size_write(BTRFS_I(inode), 0);
5347 
5348 	/*
5349 	 * This is a bit simpler than btrfs_truncate since we've already
5350 	 * reserved our space for our orphan item in the unlink, so we just
5351 	 * need to reserve some slack space in case we add bytes and update
5352 	 * inode item when doing the truncate.
5353 	 */
5354 	while (1) {
5355 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
5356 					     BTRFS_RESERVE_FLUSH_LIMIT);
5357 
5358 		/*
5359 		 * Try and steal from the global reserve since we will
5360 		 * likely not use this space anyway, we want to try as
5361 		 * hard as possible to get this to work.
5362 		 */
5363 		if (ret)
5364 			steal_from_global++;
5365 		else
5366 			steal_from_global = 0;
5367 		ret = 0;
5368 
5369 		/*
5370 		 * steal_from_global == 0: we reserved stuff, hooray!
5371 		 * steal_from_global == 1: we didn't reserve stuff, boo!
5372 		 * steal_from_global == 2: we've committed, still not a lot of
5373 		 * room but maybe we'll have room in the global reserve this
5374 		 * time.
5375 		 * steal_from_global == 3: abandon all hope!
5376 		 */
5377 		if (steal_from_global > 2) {
5378 			btrfs_warn(fs_info,
5379 				   "Could not get space for a delete, will truncate on mount %d",
5380 				   ret);
5381 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5382 			btrfs_free_block_rsv(fs_info, rsv);
5383 			goto no_delete;
5384 		}
5385 
5386 		trans = btrfs_join_transaction(root);
5387 		if (IS_ERR(trans)) {
5388 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5389 			btrfs_free_block_rsv(fs_info, rsv);
5390 			goto no_delete;
5391 		}
5392 
5393 		/*
5394 		 * We can't just steal from the global reserve, we need to make
5395 		 * sure there is room to do it, if not we need to commit and try
5396 		 * again.
5397 		 */
5398 		if (steal_from_global) {
5399 			if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5400 				ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5401 							      min_size, 0);
5402 			else
5403 				ret = -ENOSPC;
5404 		}
5405 
5406 		/*
5407 		 * Couldn't steal from the global reserve, we have too much
5408 		 * pending stuff built up, commit the transaction and try it
5409 		 * again.
5410 		 */
5411 		if (ret) {
5412 			ret = btrfs_commit_transaction(trans);
5413 			if (ret) {
5414 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5415 				btrfs_free_block_rsv(fs_info, rsv);
5416 				goto no_delete;
5417 			}
5418 			continue;
5419 		} else {
5420 			steal_from_global = 0;
5421 		}
5422 
5423 		trans->block_rsv = rsv;
5424 
5425 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5426 		if (ret != -ENOSPC && ret != -EAGAIN)
5427 			break;
5428 
5429 		trans->block_rsv = &fs_info->trans_block_rsv;
5430 		btrfs_end_transaction(trans);
5431 		trans = NULL;
5432 		btrfs_btree_balance_dirty(fs_info);
5433 	}
5434 
5435 	btrfs_free_block_rsv(fs_info, rsv);
5436 
5437 	/*
5438 	 * Errors here aren't a big deal, it just means we leave orphan items
5439 	 * in the tree.  They will be cleaned up on the next mount.
5440 	 */
5441 	if (ret == 0) {
5442 		trans->block_rsv = root->orphan_block_rsv;
5443 		btrfs_orphan_del(trans, BTRFS_I(inode));
5444 	} else {
5445 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5446 	}
5447 
5448 	trans->block_rsv = &fs_info->trans_block_rsv;
5449 	if (!(root == fs_info->tree_root ||
5450 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5451 		btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5452 
5453 	btrfs_end_transaction(trans);
5454 	btrfs_btree_balance_dirty(fs_info);
5455 no_delete:
5456 	btrfs_remove_delayed_node(BTRFS_I(inode));
5457 	clear_inode(inode);
5458 }
5459 
5460 /*
5461  * this returns the key found in the dir entry in the location pointer.
5462  * If no dir entries were found, location->objectid is 0.
5463  */
5464 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5465 			       struct btrfs_key *location)
5466 {
5467 	const char *name = dentry->d_name.name;
5468 	int namelen = dentry->d_name.len;
5469 	struct btrfs_dir_item *di;
5470 	struct btrfs_path *path;
5471 	struct btrfs_root *root = BTRFS_I(dir)->root;
5472 	int ret = 0;
5473 
5474 	path = btrfs_alloc_path();
5475 	if (!path)
5476 		return -ENOMEM;
5477 
5478 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5479 			name, namelen, 0);
5480 	if (IS_ERR(di))
5481 		ret = PTR_ERR(di);
5482 
5483 	if (IS_ERR_OR_NULL(di))
5484 		goto out_err;
5485 
5486 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5487 out:
5488 	btrfs_free_path(path);
5489 	return ret;
5490 out_err:
5491 	location->objectid = 0;
5492 	goto out;
5493 }
5494 
5495 /*
5496  * when we hit a tree root in a directory, the btrfs part of the inode
5497  * needs to be changed to reflect the root directory of the tree root.  This
5498  * is kind of like crossing a mount point.
5499  */
5500 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5501 				    struct inode *dir,
5502 				    struct dentry *dentry,
5503 				    struct btrfs_key *location,
5504 				    struct btrfs_root **sub_root)
5505 {
5506 	struct btrfs_path *path;
5507 	struct btrfs_root *new_root;
5508 	struct btrfs_root_ref *ref;
5509 	struct extent_buffer *leaf;
5510 	struct btrfs_key key;
5511 	int ret;
5512 	int err = 0;
5513 
5514 	path = btrfs_alloc_path();
5515 	if (!path) {
5516 		err = -ENOMEM;
5517 		goto out;
5518 	}
5519 
5520 	err = -ENOENT;
5521 	key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5522 	key.type = BTRFS_ROOT_REF_KEY;
5523 	key.offset = location->objectid;
5524 
5525 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5526 	if (ret) {
5527 		if (ret < 0)
5528 			err = ret;
5529 		goto out;
5530 	}
5531 
5532 	leaf = path->nodes[0];
5533 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5534 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5535 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5536 		goto out;
5537 
5538 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5539 				   (unsigned long)(ref + 1),
5540 				   dentry->d_name.len);
5541 	if (ret)
5542 		goto out;
5543 
5544 	btrfs_release_path(path);
5545 
5546 	new_root = btrfs_read_fs_root_no_name(fs_info, location);
5547 	if (IS_ERR(new_root)) {
5548 		err = PTR_ERR(new_root);
5549 		goto out;
5550 	}
5551 
5552 	*sub_root = new_root;
5553 	location->objectid = btrfs_root_dirid(&new_root->root_item);
5554 	location->type = BTRFS_INODE_ITEM_KEY;
5555 	location->offset = 0;
5556 	err = 0;
5557 out:
5558 	btrfs_free_path(path);
5559 	return err;
5560 }
5561 
5562 static void inode_tree_add(struct inode *inode)
5563 {
5564 	struct btrfs_root *root = BTRFS_I(inode)->root;
5565 	struct btrfs_inode *entry;
5566 	struct rb_node **p;
5567 	struct rb_node *parent;
5568 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
5569 	u64 ino = btrfs_ino(BTRFS_I(inode));
5570 
5571 	if (inode_unhashed(inode))
5572 		return;
5573 	parent = NULL;
5574 	spin_lock(&root->inode_lock);
5575 	p = &root->inode_tree.rb_node;
5576 	while (*p) {
5577 		parent = *p;
5578 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
5579 
5580 		if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5581 			p = &parent->rb_left;
5582 		else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5583 			p = &parent->rb_right;
5584 		else {
5585 			WARN_ON(!(entry->vfs_inode.i_state &
5586 				  (I_WILL_FREE | I_FREEING)));
5587 			rb_replace_node(parent, new, &root->inode_tree);
5588 			RB_CLEAR_NODE(parent);
5589 			spin_unlock(&root->inode_lock);
5590 			return;
5591 		}
5592 	}
5593 	rb_link_node(new, parent, p);
5594 	rb_insert_color(new, &root->inode_tree);
5595 	spin_unlock(&root->inode_lock);
5596 }
5597 
5598 static void inode_tree_del(struct inode *inode)
5599 {
5600 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5601 	struct btrfs_root *root = BTRFS_I(inode)->root;
5602 	int empty = 0;
5603 
5604 	spin_lock(&root->inode_lock);
5605 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5606 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5607 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5608 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5609 	}
5610 	spin_unlock(&root->inode_lock);
5611 
5612 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
5613 		synchronize_srcu(&fs_info->subvol_srcu);
5614 		spin_lock(&root->inode_lock);
5615 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5616 		spin_unlock(&root->inode_lock);
5617 		if (empty)
5618 			btrfs_add_dead_root(root);
5619 	}
5620 }
5621 
5622 void btrfs_invalidate_inodes(struct btrfs_root *root)
5623 {
5624 	struct btrfs_fs_info *fs_info = root->fs_info;
5625 	struct rb_node *node;
5626 	struct rb_node *prev;
5627 	struct btrfs_inode *entry;
5628 	struct inode *inode;
5629 	u64 objectid = 0;
5630 
5631 	if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5632 		WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5633 
5634 	spin_lock(&root->inode_lock);
5635 again:
5636 	node = root->inode_tree.rb_node;
5637 	prev = NULL;
5638 	while (node) {
5639 		prev = node;
5640 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5641 
5642 		if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5643 			node = node->rb_left;
5644 		else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5645 			node = node->rb_right;
5646 		else
5647 			break;
5648 	}
5649 	if (!node) {
5650 		while (prev) {
5651 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
5652 			if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5653 				node = prev;
5654 				break;
5655 			}
5656 			prev = rb_next(prev);
5657 		}
5658 	}
5659 	while (node) {
5660 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5661 		objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5662 		inode = igrab(&entry->vfs_inode);
5663 		if (inode) {
5664 			spin_unlock(&root->inode_lock);
5665 			if (atomic_read(&inode->i_count) > 1)
5666 				d_prune_aliases(inode);
5667 			/*
5668 			 * btrfs_drop_inode will have it removed from
5669 			 * the inode cache when its usage count
5670 			 * hits zero.
5671 			 */
5672 			iput(inode);
5673 			cond_resched();
5674 			spin_lock(&root->inode_lock);
5675 			goto again;
5676 		}
5677 
5678 		if (cond_resched_lock(&root->inode_lock))
5679 			goto again;
5680 
5681 		node = rb_next(node);
5682 	}
5683 	spin_unlock(&root->inode_lock);
5684 }
5685 
5686 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5687 {
5688 	struct btrfs_iget_args *args = p;
5689 	inode->i_ino = args->location->objectid;
5690 	memcpy(&BTRFS_I(inode)->location, args->location,
5691 	       sizeof(*args->location));
5692 	BTRFS_I(inode)->root = args->root;
5693 	return 0;
5694 }
5695 
5696 static int btrfs_find_actor(struct inode *inode, void *opaque)
5697 {
5698 	struct btrfs_iget_args *args = opaque;
5699 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5700 		args->root == BTRFS_I(inode)->root;
5701 }
5702 
5703 static struct inode *btrfs_iget_locked(struct super_block *s,
5704 				       struct btrfs_key *location,
5705 				       struct btrfs_root *root)
5706 {
5707 	struct inode *inode;
5708 	struct btrfs_iget_args args;
5709 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5710 
5711 	args.location = location;
5712 	args.root = root;
5713 
5714 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5715 			     btrfs_init_locked_inode,
5716 			     (void *)&args);
5717 	return inode;
5718 }
5719 
5720 /* Get an inode object given its location and corresponding root.
5721  * Returns in *is_new if the inode was read from disk
5722  */
5723 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5724 			 struct btrfs_root *root, int *new)
5725 {
5726 	struct inode *inode;
5727 
5728 	inode = btrfs_iget_locked(s, location, root);
5729 	if (!inode)
5730 		return ERR_PTR(-ENOMEM);
5731 
5732 	if (inode->i_state & I_NEW) {
5733 		int ret;
5734 
5735 		ret = btrfs_read_locked_inode(inode);
5736 		if (!is_bad_inode(inode)) {
5737 			inode_tree_add(inode);
5738 			unlock_new_inode(inode);
5739 			if (new)
5740 				*new = 1;
5741 		} else {
5742 			unlock_new_inode(inode);
5743 			iput(inode);
5744 			ASSERT(ret < 0);
5745 			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5746 		}
5747 	}
5748 
5749 	return inode;
5750 }
5751 
5752 static struct inode *new_simple_dir(struct super_block *s,
5753 				    struct btrfs_key *key,
5754 				    struct btrfs_root *root)
5755 {
5756 	struct inode *inode = new_inode(s);
5757 
5758 	if (!inode)
5759 		return ERR_PTR(-ENOMEM);
5760 
5761 	BTRFS_I(inode)->root = root;
5762 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5763 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5764 
5765 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5766 	inode->i_op = &btrfs_dir_ro_inode_operations;
5767 	inode->i_opflags &= ~IOP_XATTR;
5768 	inode->i_fop = &simple_dir_operations;
5769 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5770 	inode->i_mtime = current_time(inode);
5771 	inode->i_atime = inode->i_mtime;
5772 	inode->i_ctime = inode->i_mtime;
5773 	BTRFS_I(inode)->i_otime = inode->i_mtime;
5774 
5775 	return inode;
5776 }
5777 
5778 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5779 {
5780 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5781 	struct inode *inode;
5782 	struct btrfs_root *root = BTRFS_I(dir)->root;
5783 	struct btrfs_root *sub_root = root;
5784 	struct btrfs_key location;
5785 	int index;
5786 	int ret = 0;
5787 
5788 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5789 		return ERR_PTR(-ENAMETOOLONG);
5790 
5791 	ret = btrfs_inode_by_name(dir, dentry, &location);
5792 	if (ret < 0)
5793 		return ERR_PTR(ret);
5794 
5795 	if (location.objectid == 0)
5796 		return ERR_PTR(-ENOENT);
5797 
5798 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5799 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5800 		return inode;
5801 	}
5802 
5803 	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5804 
5805 	index = srcu_read_lock(&fs_info->subvol_srcu);
5806 	ret = fixup_tree_root_location(fs_info, dir, dentry,
5807 				       &location, &sub_root);
5808 	if (ret < 0) {
5809 		if (ret != -ENOENT)
5810 			inode = ERR_PTR(ret);
5811 		else
5812 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5813 	} else {
5814 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5815 	}
5816 	srcu_read_unlock(&fs_info->subvol_srcu, index);
5817 
5818 	if (!IS_ERR(inode) && root != sub_root) {
5819 		down_read(&fs_info->cleanup_work_sem);
5820 		if (!(inode->i_sb->s_flags & MS_RDONLY))
5821 			ret = btrfs_orphan_cleanup(sub_root);
5822 		up_read(&fs_info->cleanup_work_sem);
5823 		if (ret) {
5824 			iput(inode);
5825 			inode = ERR_PTR(ret);
5826 		}
5827 	}
5828 
5829 	return inode;
5830 }
5831 
5832 static int btrfs_dentry_delete(const struct dentry *dentry)
5833 {
5834 	struct btrfs_root *root;
5835 	struct inode *inode = d_inode(dentry);
5836 
5837 	if (!inode && !IS_ROOT(dentry))
5838 		inode = d_inode(dentry->d_parent);
5839 
5840 	if (inode) {
5841 		root = BTRFS_I(inode)->root;
5842 		if (btrfs_root_refs(&root->root_item) == 0)
5843 			return 1;
5844 
5845 		if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5846 			return 1;
5847 	}
5848 	return 0;
5849 }
5850 
5851 static void btrfs_dentry_release(struct dentry *dentry)
5852 {
5853 	kfree(dentry->d_fsdata);
5854 }
5855 
5856 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5857 				   unsigned int flags)
5858 {
5859 	struct inode *inode;
5860 
5861 	inode = btrfs_lookup_dentry(dir, dentry);
5862 	if (IS_ERR(inode)) {
5863 		if (PTR_ERR(inode) == -ENOENT)
5864 			inode = NULL;
5865 		else
5866 			return ERR_CAST(inode);
5867 	}
5868 
5869 	return d_splice_alias(inode, dentry);
5870 }
5871 
5872 unsigned char btrfs_filetype_table[] = {
5873 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5874 };
5875 
5876 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5877 {
5878 	struct inode *inode = file_inode(file);
5879 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5880 	struct btrfs_root *root = BTRFS_I(inode)->root;
5881 	struct btrfs_dir_item *di;
5882 	struct btrfs_key key;
5883 	struct btrfs_key found_key;
5884 	struct btrfs_path *path;
5885 	struct list_head ins_list;
5886 	struct list_head del_list;
5887 	int ret;
5888 	struct extent_buffer *leaf;
5889 	int slot;
5890 	unsigned char d_type;
5891 	int over = 0;
5892 	char tmp_name[32];
5893 	char *name_ptr;
5894 	int name_len;
5895 	bool put = false;
5896 	struct btrfs_key location;
5897 
5898 	if (!dir_emit_dots(file, ctx))
5899 		return 0;
5900 
5901 	path = btrfs_alloc_path();
5902 	if (!path)
5903 		return -ENOMEM;
5904 
5905 	path->reada = READA_FORWARD;
5906 
5907 	INIT_LIST_HEAD(&ins_list);
5908 	INIT_LIST_HEAD(&del_list);
5909 	put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5910 
5911 	key.type = BTRFS_DIR_INDEX_KEY;
5912 	key.offset = ctx->pos;
5913 	key.objectid = btrfs_ino(BTRFS_I(inode));
5914 
5915 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5916 	if (ret < 0)
5917 		goto err;
5918 
5919 	while (1) {
5920 		leaf = path->nodes[0];
5921 		slot = path->slots[0];
5922 		if (slot >= btrfs_header_nritems(leaf)) {
5923 			ret = btrfs_next_leaf(root, path);
5924 			if (ret < 0)
5925 				goto err;
5926 			else if (ret > 0)
5927 				break;
5928 			continue;
5929 		}
5930 
5931 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5932 
5933 		if (found_key.objectid != key.objectid)
5934 			break;
5935 		if (found_key.type != BTRFS_DIR_INDEX_KEY)
5936 			break;
5937 		if (found_key.offset < ctx->pos)
5938 			goto next;
5939 		if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5940 			goto next;
5941 
5942 		ctx->pos = found_key.offset;
5943 
5944 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5945 		if (verify_dir_item(fs_info, leaf, slot, di))
5946 			goto next;
5947 
5948 		name_len = btrfs_dir_name_len(leaf, di);
5949 		if (name_len <= sizeof(tmp_name)) {
5950 			name_ptr = tmp_name;
5951 		} else {
5952 			name_ptr = kmalloc(name_len, GFP_KERNEL);
5953 			if (!name_ptr) {
5954 				ret = -ENOMEM;
5955 				goto err;
5956 			}
5957 		}
5958 		read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5959 				   name_len);
5960 
5961 		d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5962 		btrfs_dir_item_key_to_cpu(leaf, di, &location);
5963 
5964 		over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5965 				 d_type);
5966 
5967 		if (name_ptr != tmp_name)
5968 			kfree(name_ptr);
5969 
5970 		if (over)
5971 			goto nopos;
5972 		ctx->pos++;
5973 next:
5974 		path->slots[0]++;
5975 	}
5976 
5977 	ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5978 	if (ret)
5979 		goto nopos;
5980 
5981 	/*
5982 	 * Stop new entries from being returned after we return the last
5983 	 * entry.
5984 	 *
5985 	 * New directory entries are assigned a strictly increasing
5986 	 * offset.  This means that new entries created during readdir
5987 	 * are *guaranteed* to be seen in the future by that readdir.
5988 	 * This has broken buggy programs which operate on names as
5989 	 * they're returned by readdir.  Until we re-use freed offsets
5990 	 * we have this hack to stop new entries from being returned
5991 	 * under the assumption that they'll never reach this huge
5992 	 * offset.
5993 	 *
5994 	 * This is being careful not to overflow 32bit loff_t unless the
5995 	 * last entry requires it because doing so has broken 32bit apps
5996 	 * in the past.
5997 	 */
5998 	if (ctx->pos >= INT_MAX)
5999 		ctx->pos = LLONG_MAX;
6000 	else
6001 		ctx->pos = INT_MAX;
6002 nopos:
6003 	ret = 0;
6004 err:
6005 	if (put)
6006 		btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6007 	btrfs_free_path(path);
6008 	return ret;
6009 }
6010 
6011 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6012 {
6013 	struct btrfs_root *root = BTRFS_I(inode)->root;
6014 	struct btrfs_trans_handle *trans;
6015 	int ret = 0;
6016 	bool nolock = false;
6017 
6018 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6019 		return 0;
6020 
6021 	if (btrfs_fs_closing(root->fs_info) &&
6022 			btrfs_is_free_space_inode(BTRFS_I(inode)))
6023 		nolock = true;
6024 
6025 	if (wbc->sync_mode == WB_SYNC_ALL) {
6026 		if (nolock)
6027 			trans = btrfs_join_transaction_nolock(root);
6028 		else
6029 			trans = btrfs_join_transaction(root);
6030 		if (IS_ERR(trans))
6031 			return PTR_ERR(trans);
6032 		ret = btrfs_commit_transaction(trans);
6033 	}
6034 	return ret;
6035 }
6036 
6037 /*
6038  * This is somewhat expensive, updating the tree every time the
6039  * inode changes.  But, it is most likely to find the inode in cache.
6040  * FIXME, needs more benchmarking...there are no reasons other than performance
6041  * to keep or drop this code.
6042  */
6043 static int btrfs_dirty_inode(struct inode *inode)
6044 {
6045 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6046 	struct btrfs_root *root = BTRFS_I(inode)->root;
6047 	struct btrfs_trans_handle *trans;
6048 	int ret;
6049 
6050 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6051 		return 0;
6052 
6053 	trans = btrfs_join_transaction(root);
6054 	if (IS_ERR(trans))
6055 		return PTR_ERR(trans);
6056 
6057 	ret = btrfs_update_inode(trans, root, inode);
6058 	if (ret && ret == -ENOSPC) {
6059 		/* whoops, lets try again with the full transaction */
6060 		btrfs_end_transaction(trans);
6061 		trans = btrfs_start_transaction(root, 1);
6062 		if (IS_ERR(trans))
6063 			return PTR_ERR(trans);
6064 
6065 		ret = btrfs_update_inode(trans, root, inode);
6066 	}
6067 	btrfs_end_transaction(trans);
6068 	if (BTRFS_I(inode)->delayed_node)
6069 		btrfs_balance_delayed_items(fs_info);
6070 
6071 	return ret;
6072 }
6073 
6074 /*
6075  * This is a copy of file_update_time.  We need this so we can return error on
6076  * ENOSPC for updating the inode in the case of file write and mmap writes.
6077  */
6078 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6079 			     int flags)
6080 {
6081 	struct btrfs_root *root = BTRFS_I(inode)->root;
6082 
6083 	if (btrfs_root_readonly(root))
6084 		return -EROFS;
6085 
6086 	if (flags & S_VERSION)
6087 		inode_inc_iversion(inode);
6088 	if (flags & S_CTIME)
6089 		inode->i_ctime = *now;
6090 	if (flags & S_MTIME)
6091 		inode->i_mtime = *now;
6092 	if (flags & S_ATIME)
6093 		inode->i_atime = *now;
6094 	return btrfs_dirty_inode(inode);
6095 }
6096 
6097 /*
6098  * find the highest existing sequence number in a directory
6099  * and then set the in-memory index_cnt variable to reflect
6100  * free sequence numbers
6101  */
6102 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6103 {
6104 	struct btrfs_root *root = inode->root;
6105 	struct btrfs_key key, found_key;
6106 	struct btrfs_path *path;
6107 	struct extent_buffer *leaf;
6108 	int ret;
6109 
6110 	key.objectid = btrfs_ino(inode);
6111 	key.type = BTRFS_DIR_INDEX_KEY;
6112 	key.offset = (u64)-1;
6113 
6114 	path = btrfs_alloc_path();
6115 	if (!path)
6116 		return -ENOMEM;
6117 
6118 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6119 	if (ret < 0)
6120 		goto out;
6121 	/* FIXME: we should be able to handle this */
6122 	if (ret == 0)
6123 		goto out;
6124 	ret = 0;
6125 
6126 	/*
6127 	 * MAGIC NUMBER EXPLANATION:
6128 	 * since we search a directory based on f_pos we have to start at 2
6129 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6130 	 * else has to start at 2
6131 	 */
6132 	if (path->slots[0] == 0) {
6133 		inode->index_cnt = 2;
6134 		goto out;
6135 	}
6136 
6137 	path->slots[0]--;
6138 
6139 	leaf = path->nodes[0];
6140 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6141 
6142 	if (found_key.objectid != btrfs_ino(inode) ||
6143 	    found_key.type != BTRFS_DIR_INDEX_KEY) {
6144 		inode->index_cnt = 2;
6145 		goto out;
6146 	}
6147 
6148 	inode->index_cnt = found_key.offset + 1;
6149 out:
6150 	btrfs_free_path(path);
6151 	return ret;
6152 }
6153 
6154 /*
6155  * helper to find a free sequence number in a given directory.  This current
6156  * code is very simple, later versions will do smarter things in the btree
6157  */
6158 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6159 {
6160 	int ret = 0;
6161 
6162 	if (dir->index_cnt == (u64)-1) {
6163 		ret = btrfs_inode_delayed_dir_index_count(dir);
6164 		if (ret) {
6165 			ret = btrfs_set_inode_index_count(dir);
6166 			if (ret)
6167 				return ret;
6168 		}
6169 	}
6170 
6171 	*index = dir->index_cnt;
6172 	dir->index_cnt++;
6173 
6174 	return ret;
6175 }
6176 
6177 static int btrfs_insert_inode_locked(struct inode *inode)
6178 {
6179 	struct btrfs_iget_args args;
6180 	args.location = &BTRFS_I(inode)->location;
6181 	args.root = BTRFS_I(inode)->root;
6182 
6183 	return insert_inode_locked4(inode,
6184 		   btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6185 		   btrfs_find_actor, &args);
6186 }
6187 
6188 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6189 				     struct btrfs_root *root,
6190 				     struct inode *dir,
6191 				     const char *name, int name_len,
6192 				     u64 ref_objectid, u64 objectid,
6193 				     umode_t mode, u64 *index)
6194 {
6195 	struct btrfs_fs_info *fs_info = root->fs_info;
6196 	struct inode *inode;
6197 	struct btrfs_inode_item *inode_item;
6198 	struct btrfs_key *location;
6199 	struct btrfs_path *path;
6200 	struct btrfs_inode_ref *ref;
6201 	struct btrfs_key key[2];
6202 	u32 sizes[2];
6203 	int nitems = name ? 2 : 1;
6204 	unsigned long ptr;
6205 	int ret;
6206 
6207 	path = btrfs_alloc_path();
6208 	if (!path)
6209 		return ERR_PTR(-ENOMEM);
6210 
6211 	inode = new_inode(fs_info->sb);
6212 	if (!inode) {
6213 		btrfs_free_path(path);
6214 		return ERR_PTR(-ENOMEM);
6215 	}
6216 
6217 	/*
6218 	 * O_TMPFILE, set link count to 0, so that after this point,
6219 	 * we fill in an inode item with the correct link count.
6220 	 */
6221 	if (!name)
6222 		set_nlink(inode, 0);
6223 
6224 	/*
6225 	 * we have to initialize this early, so we can reclaim the inode
6226 	 * number if we fail afterwards in this function.
6227 	 */
6228 	inode->i_ino = objectid;
6229 
6230 	if (dir && name) {
6231 		trace_btrfs_inode_request(dir);
6232 
6233 		ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6234 		if (ret) {
6235 			btrfs_free_path(path);
6236 			iput(inode);
6237 			return ERR_PTR(ret);
6238 		}
6239 	} else if (dir) {
6240 		*index = 0;
6241 	}
6242 	/*
6243 	 * index_cnt is ignored for everything but a dir,
6244 	 * btrfs_get_inode_index_count has an explanation for the magic
6245 	 * number
6246 	 */
6247 	BTRFS_I(inode)->index_cnt = 2;
6248 	BTRFS_I(inode)->dir_index = *index;
6249 	BTRFS_I(inode)->root = root;
6250 	BTRFS_I(inode)->generation = trans->transid;
6251 	inode->i_generation = BTRFS_I(inode)->generation;
6252 
6253 	/*
6254 	 * We could have gotten an inode number from somebody who was fsynced
6255 	 * and then removed in this same transaction, so let's just set full
6256 	 * sync since it will be a full sync anyway and this will blow away the
6257 	 * old info in the log.
6258 	 */
6259 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6260 
6261 	key[0].objectid = objectid;
6262 	key[0].type = BTRFS_INODE_ITEM_KEY;
6263 	key[0].offset = 0;
6264 
6265 	sizes[0] = sizeof(struct btrfs_inode_item);
6266 
6267 	if (name) {
6268 		/*
6269 		 * Start new inodes with an inode_ref. This is slightly more
6270 		 * efficient for small numbers of hard links since they will
6271 		 * be packed into one item. Extended refs will kick in if we
6272 		 * add more hard links than can fit in the ref item.
6273 		 */
6274 		key[1].objectid = objectid;
6275 		key[1].type = BTRFS_INODE_REF_KEY;
6276 		key[1].offset = ref_objectid;
6277 
6278 		sizes[1] = name_len + sizeof(*ref);
6279 	}
6280 
6281 	location = &BTRFS_I(inode)->location;
6282 	location->objectid = objectid;
6283 	location->offset = 0;
6284 	location->type = BTRFS_INODE_ITEM_KEY;
6285 
6286 	ret = btrfs_insert_inode_locked(inode);
6287 	if (ret < 0)
6288 		goto fail;
6289 
6290 	path->leave_spinning = 1;
6291 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6292 	if (ret != 0)
6293 		goto fail_unlock;
6294 
6295 	inode_init_owner(inode, dir, mode);
6296 	inode_set_bytes(inode, 0);
6297 
6298 	inode->i_mtime = current_time(inode);
6299 	inode->i_atime = inode->i_mtime;
6300 	inode->i_ctime = inode->i_mtime;
6301 	BTRFS_I(inode)->i_otime = inode->i_mtime;
6302 
6303 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6304 				  struct btrfs_inode_item);
6305 	memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6306 			     sizeof(*inode_item));
6307 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
6308 
6309 	if (name) {
6310 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6311 				     struct btrfs_inode_ref);
6312 		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6313 		btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6314 		ptr = (unsigned long)(ref + 1);
6315 		write_extent_buffer(path->nodes[0], name, ptr, name_len);
6316 	}
6317 
6318 	btrfs_mark_buffer_dirty(path->nodes[0]);
6319 	btrfs_free_path(path);
6320 
6321 	btrfs_inherit_iflags(inode, dir);
6322 
6323 	if (S_ISREG(mode)) {
6324 		if (btrfs_test_opt(fs_info, NODATASUM))
6325 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6326 		if (btrfs_test_opt(fs_info, NODATACOW))
6327 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6328 				BTRFS_INODE_NODATASUM;
6329 	}
6330 
6331 	inode_tree_add(inode);
6332 
6333 	trace_btrfs_inode_new(inode);
6334 	btrfs_set_inode_last_trans(trans, inode);
6335 
6336 	btrfs_update_root_times(trans, root);
6337 
6338 	ret = btrfs_inode_inherit_props(trans, inode, dir);
6339 	if (ret)
6340 		btrfs_err(fs_info,
6341 			  "error inheriting props for ino %llu (root %llu): %d",
6342 			btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6343 
6344 	return inode;
6345 
6346 fail_unlock:
6347 	unlock_new_inode(inode);
6348 fail:
6349 	if (dir && name)
6350 		BTRFS_I(dir)->index_cnt--;
6351 	btrfs_free_path(path);
6352 	iput(inode);
6353 	return ERR_PTR(ret);
6354 }
6355 
6356 static inline u8 btrfs_inode_type(struct inode *inode)
6357 {
6358 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6359 }
6360 
6361 /*
6362  * utility function to add 'inode' into 'parent_inode' with
6363  * a give name and a given sequence number.
6364  * if 'add_backref' is true, also insert a backref from the
6365  * inode to the parent directory.
6366  */
6367 int btrfs_add_link(struct btrfs_trans_handle *trans,
6368 		   struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6369 		   const char *name, int name_len, int add_backref, u64 index)
6370 {
6371 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6372 	int ret = 0;
6373 	struct btrfs_key key;
6374 	struct btrfs_root *root = parent_inode->root;
6375 	u64 ino = btrfs_ino(inode);
6376 	u64 parent_ino = btrfs_ino(parent_inode);
6377 
6378 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6379 		memcpy(&key, &inode->root->root_key, sizeof(key));
6380 	} else {
6381 		key.objectid = ino;
6382 		key.type = BTRFS_INODE_ITEM_KEY;
6383 		key.offset = 0;
6384 	}
6385 
6386 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6387 		ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6388 					 root->root_key.objectid, parent_ino,
6389 					 index, name, name_len);
6390 	} else if (add_backref) {
6391 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6392 					     parent_ino, index);
6393 	}
6394 
6395 	/* Nothing to clean up yet */
6396 	if (ret)
6397 		return ret;
6398 
6399 	ret = btrfs_insert_dir_item(trans, root, name, name_len,
6400 				    parent_inode, &key,
6401 				    btrfs_inode_type(&inode->vfs_inode), index);
6402 	if (ret == -EEXIST || ret == -EOVERFLOW)
6403 		goto fail_dir_item;
6404 	else if (ret) {
6405 		btrfs_abort_transaction(trans, ret);
6406 		return ret;
6407 	}
6408 
6409 	btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6410 			   name_len * 2);
6411 	inode_inc_iversion(&parent_inode->vfs_inode);
6412 	parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6413 		current_time(&parent_inode->vfs_inode);
6414 	ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6415 	if (ret)
6416 		btrfs_abort_transaction(trans, ret);
6417 	return ret;
6418 
6419 fail_dir_item:
6420 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6421 		u64 local_index;
6422 		int err;
6423 		err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6424 					 root->root_key.objectid, parent_ino,
6425 					 &local_index, name, name_len);
6426 
6427 	} else if (add_backref) {
6428 		u64 local_index;
6429 		int err;
6430 
6431 		err = btrfs_del_inode_ref(trans, root, name, name_len,
6432 					  ino, parent_ino, &local_index);
6433 	}
6434 	return ret;
6435 }
6436 
6437 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6438 			    struct btrfs_inode *dir, struct dentry *dentry,
6439 			    struct btrfs_inode *inode, int backref, u64 index)
6440 {
6441 	int err = btrfs_add_link(trans, dir, inode,
6442 				 dentry->d_name.name, dentry->d_name.len,
6443 				 backref, index);
6444 	if (err > 0)
6445 		err = -EEXIST;
6446 	return err;
6447 }
6448 
6449 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6450 			umode_t mode, dev_t rdev)
6451 {
6452 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6453 	struct btrfs_trans_handle *trans;
6454 	struct btrfs_root *root = BTRFS_I(dir)->root;
6455 	struct inode *inode = NULL;
6456 	int err;
6457 	int drop_inode = 0;
6458 	u64 objectid;
6459 	u64 index = 0;
6460 
6461 	/*
6462 	 * 2 for inode item and ref
6463 	 * 2 for dir items
6464 	 * 1 for xattr if selinux is on
6465 	 */
6466 	trans = btrfs_start_transaction(root, 5);
6467 	if (IS_ERR(trans))
6468 		return PTR_ERR(trans);
6469 
6470 	err = btrfs_find_free_ino(root, &objectid);
6471 	if (err)
6472 		goto out_unlock;
6473 
6474 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6475 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6476 			mode, &index);
6477 	if (IS_ERR(inode)) {
6478 		err = PTR_ERR(inode);
6479 		goto out_unlock;
6480 	}
6481 
6482 	/*
6483 	* If the active LSM wants to access the inode during
6484 	* d_instantiate it needs these. Smack checks to see
6485 	* if the filesystem supports xattrs by looking at the
6486 	* ops vector.
6487 	*/
6488 	inode->i_op = &btrfs_special_inode_operations;
6489 	init_special_inode(inode, inode->i_mode, rdev);
6490 
6491 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6492 	if (err)
6493 		goto out_unlock_inode;
6494 
6495 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6496 			0, index);
6497 	if (err) {
6498 		goto out_unlock_inode;
6499 	} else {
6500 		btrfs_update_inode(trans, root, inode);
6501 		unlock_new_inode(inode);
6502 		d_instantiate(dentry, inode);
6503 	}
6504 
6505 out_unlock:
6506 	btrfs_end_transaction(trans);
6507 	btrfs_balance_delayed_items(fs_info);
6508 	btrfs_btree_balance_dirty(fs_info);
6509 	if (drop_inode) {
6510 		inode_dec_link_count(inode);
6511 		iput(inode);
6512 	}
6513 	return err;
6514 
6515 out_unlock_inode:
6516 	drop_inode = 1;
6517 	unlock_new_inode(inode);
6518 	goto out_unlock;
6519 
6520 }
6521 
6522 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6523 			umode_t mode, bool excl)
6524 {
6525 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6526 	struct btrfs_trans_handle *trans;
6527 	struct btrfs_root *root = BTRFS_I(dir)->root;
6528 	struct inode *inode = NULL;
6529 	int drop_inode_on_err = 0;
6530 	int err;
6531 	u64 objectid;
6532 	u64 index = 0;
6533 
6534 	/*
6535 	 * 2 for inode item and ref
6536 	 * 2 for dir items
6537 	 * 1 for xattr if selinux is on
6538 	 */
6539 	trans = btrfs_start_transaction(root, 5);
6540 	if (IS_ERR(trans))
6541 		return PTR_ERR(trans);
6542 
6543 	err = btrfs_find_free_ino(root, &objectid);
6544 	if (err)
6545 		goto out_unlock;
6546 
6547 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6548 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6549 			mode, &index);
6550 	if (IS_ERR(inode)) {
6551 		err = PTR_ERR(inode);
6552 		goto out_unlock;
6553 	}
6554 	drop_inode_on_err = 1;
6555 	/*
6556 	* If the active LSM wants to access the inode during
6557 	* d_instantiate it needs these. Smack checks to see
6558 	* if the filesystem supports xattrs by looking at the
6559 	* ops vector.
6560 	*/
6561 	inode->i_fop = &btrfs_file_operations;
6562 	inode->i_op = &btrfs_file_inode_operations;
6563 	inode->i_mapping->a_ops = &btrfs_aops;
6564 
6565 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6566 	if (err)
6567 		goto out_unlock_inode;
6568 
6569 	err = btrfs_update_inode(trans, root, inode);
6570 	if (err)
6571 		goto out_unlock_inode;
6572 
6573 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6574 			0, index);
6575 	if (err)
6576 		goto out_unlock_inode;
6577 
6578 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6579 	unlock_new_inode(inode);
6580 	d_instantiate(dentry, inode);
6581 
6582 out_unlock:
6583 	btrfs_end_transaction(trans);
6584 	if (err && drop_inode_on_err) {
6585 		inode_dec_link_count(inode);
6586 		iput(inode);
6587 	}
6588 	btrfs_balance_delayed_items(fs_info);
6589 	btrfs_btree_balance_dirty(fs_info);
6590 	return err;
6591 
6592 out_unlock_inode:
6593 	unlock_new_inode(inode);
6594 	goto out_unlock;
6595 
6596 }
6597 
6598 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6599 		      struct dentry *dentry)
6600 {
6601 	struct btrfs_trans_handle *trans = NULL;
6602 	struct btrfs_root *root = BTRFS_I(dir)->root;
6603 	struct inode *inode = d_inode(old_dentry);
6604 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6605 	u64 index;
6606 	int err;
6607 	int drop_inode = 0;
6608 
6609 	/* do not allow sys_link's with other subvols of the same device */
6610 	if (root->objectid != BTRFS_I(inode)->root->objectid)
6611 		return -EXDEV;
6612 
6613 	if (inode->i_nlink >= BTRFS_LINK_MAX)
6614 		return -EMLINK;
6615 
6616 	err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6617 	if (err)
6618 		goto fail;
6619 
6620 	/*
6621 	 * 2 items for inode and inode ref
6622 	 * 2 items for dir items
6623 	 * 1 item for parent inode
6624 	 */
6625 	trans = btrfs_start_transaction(root, 5);
6626 	if (IS_ERR(trans)) {
6627 		err = PTR_ERR(trans);
6628 		trans = NULL;
6629 		goto fail;
6630 	}
6631 
6632 	/* There are several dir indexes for this inode, clear the cache. */
6633 	BTRFS_I(inode)->dir_index = 0ULL;
6634 	inc_nlink(inode);
6635 	inode_inc_iversion(inode);
6636 	inode->i_ctime = current_time(inode);
6637 	ihold(inode);
6638 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6639 
6640 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6641 			1, index);
6642 
6643 	if (err) {
6644 		drop_inode = 1;
6645 	} else {
6646 		struct dentry *parent = dentry->d_parent;
6647 		err = btrfs_update_inode(trans, root, inode);
6648 		if (err)
6649 			goto fail;
6650 		if (inode->i_nlink == 1) {
6651 			/*
6652 			 * If new hard link count is 1, it's a file created
6653 			 * with open(2) O_TMPFILE flag.
6654 			 */
6655 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
6656 			if (err)
6657 				goto fail;
6658 		}
6659 		d_instantiate(dentry, inode);
6660 		btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6661 	}
6662 
6663 	btrfs_balance_delayed_items(fs_info);
6664 fail:
6665 	if (trans)
6666 		btrfs_end_transaction(trans);
6667 	if (drop_inode) {
6668 		inode_dec_link_count(inode);
6669 		iput(inode);
6670 	}
6671 	btrfs_btree_balance_dirty(fs_info);
6672 	return err;
6673 }
6674 
6675 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6676 {
6677 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6678 	struct inode *inode = NULL;
6679 	struct btrfs_trans_handle *trans;
6680 	struct btrfs_root *root = BTRFS_I(dir)->root;
6681 	int err = 0;
6682 	int drop_on_err = 0;
6683 	u64 objectid = 0;
6684 	u64 index = 0;
6685 
6686 	/*
6687 	 * 2 items for inode and ref
6688 	 * 2 items for dir items
6689 	 * 1 for xattr if selinux is on
6690 	 */
6691 	trans = btrfs_start_transaction(root, 5);
6692 	if (IS_ERR(trans))
6693 		return PTR_ERR(trans);
6694 
6695 	err = btrfs_find_free_ino(root, &objectid);
6696 	if (err)
6697 		goto out_fail;
6698 
6699 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6700 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6701 			S_IFDIR | mode, &index);
6702 	if (IS_ERR(inode)) {
6703 		err = PTR_ERR(inode);
6704 		goto out_fail;
6705 	}
6706 
6707 	drop_on_err = 1;
6708 	/* these must be set before we unlock the inode */
6709 	inode->i_op = &btrfs_dir_inode_operations;
6710 	inode->i_fop = &btrfs_dir_file_operations;
6711 
6712 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6713 	if (err)
6714 		goto out_fail_inode;
6715 
6716 	btrfs_i_size_write(BTRFS_I(inode), 0);
6717 	err = btrfs_update_inode(trans, root, inode);
6718 	if (err)
6719 		goto out_fail_inode;
6720 
6721 	err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6722 			dentry->d_name.name,
6723 			dentry->d_name.len, 0, index);
6724 	if (err)
6725 		goto out_fail_inode;
6726 
6727 	d_instantiate(dentry, inode);
6728 	/*
6729 	 * mkdir is special.  We're unlocking after we call d_instantiate
6730 	 * to avoid a race with nfsd calling d_instantiate.
6731 	 */
6732 	unlock_new_inode(inode);
6733 	drop_on_err = 0;
6734 
6735 out_fail:
6736 	btrfs_end_transaction(trans);
6737 	if (drop_on_err) {
6738 		inode_dec_link_count(inode);
6739 		iput(inode);
6740 	}
6741 	btrfs_balance_delayed_items(fs_info);
6742 	btrfs_btree_balance_dirty(fs_info);
6743 	return err;
6744 
6745 out_fail_inode:
6746 	unlock_new_inode(inode);
6747 	goto out_fail;
6748 }
6749 
6750 /* Find next extent map of a given extent map, caller needs to ensure locks */
6751 static struct extent_map *next_extent_map(struct extent_map *em)
6752 {
6753 	struct rb_node *next;
6754 
6755 	next = rb_next(&em->rb_node);
6756 	if (!next)
6757 		return NULL;
6758 	return container_of(next, struct extent_map, rb_node);
6759 }
6760 
6761 static struct extent_map *prev_extent_map(struct extent_map *em)
6762 {
6763 	struct rb_node *prev;
6764 
6765 	prev = rb_prev(&em->rb_node);
6766 	if (!prev)
6767 		return NULL;
6768 	return container_of(prev, struct extent_map, rb_node);
6769 }
6770 
6771 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6772  * the existing extent is the nearest extent to map_start,
6773  * and an extent that you want to insert, deal with overlap and insert
6774  * the best fitted new extent into the tree.
6775  */
6776 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6777 				struct extent_map *existing,
6778 				struct extent_map *em,
6779 				u64 map_start)
6780 {
6781 	struct extent_map *prev;
6782 	struct extent_map *next;
6783 	u64 start;
6784 	u64 end;
6785 	u64 start_diff;
6786 
6787 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6788 
6789 	if (existing->start > map_start) {
6790 		next = existing;
6791 		prev = prev_extent_map(next);
6792 	} else {
6793 		prev = existing;
6794 		next = next_extent_map(prev);
6795 	}
6796 
6797 	start = prev ? extent_map_end(prev) : em->start;
6798 	start = max_t(u64, start, em->start);
6799 	end = next ? next->start : extent_map_end(em);
6800 	end = min_t(u64, end, extent_map_end(em));
6801 	start_diff = start - em->start;
6802 	em->start = start;
6803 	em->len = end - start;
6804 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6805 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6806 		em->block_start += start_diff;
6807 		em->block_len -= start_diff;
6808 	}
6809 	return add_extent_mapping(em_tree, em, 0);
6810 }
6811 
6812 static noinline int uncompress_inline(struct btrfs_path *path,
6813 				      struct page *page,
6814 				      size_t pg_offset, u64 extent_offset,
6815 				      struct btrfs_file_extent_item *item)
6816 {
6817 	int ret;
6818 	struct extent_buffer *leaf = path->nodes[0];
6819 	char *tmp;
6820 	size_t max_size;
6821 	unsigned long inline_size;
6822 	unsigned long ptr;
6823 	int compress_type;
6824 
6825 	WARN_ON(pg_offset != 0);
6826 	compress_type = btrfs_file_extent_compression(leaf, item);
6827 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6828 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6829 					btrfs_item_nr(path->slots[0]));
6830 	tmp = kmalloc(inline_size, GFP_NOFS);
6831 	if (!tmp)
6832 		return -ENOMEM;
6833 	ptr = btrfs_file_extent_inline_start(item);
6834 
6835 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6836 
6837 	max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6838 	ret = btrfs_decompress(compress_type, tmp, page,
6839 			       extent_offset, inline_size, max_size);
6840 
6841 	/*
6842 	 * decompression code contains a memset to fill in any space between the end
6843 	 * of the uncompressed data and the end of max_size in case the decompressed
6844 	 * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6845 	 * the end of an inline extent and the beginning of the next block, so we
6846 	 * cover that region here.
6847 	 */
6848 
6849 	if (max_size + pg_offset < PAGE_SIZE) {
6850 		char *map = kmap(page);
6851 		memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6852 		kunmap(page);
6853 	}
6854 	kfree(tmp);
6855 	return ret;
6856 }
6857 
6858 /*
6859  * a bit scary, this does extent mapping from logical file offset to the disk.
6860  * the ugly parts come from merging extents from the disk with the in-ram
6861  * representation.  This gets more complex because of the data=ordered code,
6862  * where the in-ram extents might be locked pending data=ordered completion.
6863  *
6864  * This also copies inline extents directly into the page.
6865  */
6866 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6867 		struct page *page,
6868 	    size_t pg_offset, u64 start, u64 len,
6869 		int create)
6870 {
6871 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6872 	int ret;
6873 	int err = 0;
6874 	u64 extent_start = 0;
6875 	u64 extent_end = 0;
6876 	u64 objectid = btrfs_ino(inode);
6877 	u32 found_type;
6878 	struct btrfs_path *path = NULL;
6879 	struct btrfs_root *root = inode->root;
6880 	struct btrfs_file_extent_item *item;
6881 	struct extent_buffer *leaf;
6882 	struct btrfs_key found_key;
6883 	struct extent_map *em = NULL;
6884 	struct extent_map_tree *em_tree = &inode->extent_tree;
6885 	struct extent_io_tree *io_tree = &inode->io_tree;
6886 	struct btrfs_trans_handle *trans = NULL;
6887 	const bool new_inline = !page || create;
6888 
6889 again:
6890 	read_lock(&em_tree->lock);
6891 	em = lookup_extent_mapping(em_tree, start, len);
6892 	if (em)
6893 		em->bdev = fs_info->fs_devices->latest_bdev;
6894 	read_unlock(&em_tree->lock);
6895 
6896 	if (em) {
6897 		if (em->start > start || em->start + em->len <= start)
6898 			free_extent_map(em);
6899 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6900 			free_extent_map(em);
6901 		else
6902 			goto out;
6903 	}
6904 	em = alloc_extent_map();
6905 	if (!em) {
6906 		err = -ENOMEM;
6907 		goto out;
6908 	}
6909 	em->bdev = fs_info->fs_devices->latest_bdev;
6910 	em->start = EXTENT_MAP_HOLE;
6911 	em->orig_start = EXTENT_MAP_HOLE;
6912 	em->len = (u64)-1;
6913 	em->block_len = (u64)-1;
6914 
6915 	if (!path) {
6916 		path = btrfs_alloc_path();
6917 		if (!path) {
6918 			err = -ENOMEM;
6919 			goto out;
6920 		}
6921 		/*
6922 		 * Chances are we'll be called again, so go ahead and do
6923 		 * readahead
6924 		 */
6925 		path->reada = READA_FORWARD;
6926 	}
6927 
6928 	ret = btrfs_lookup_file_extent(trans, root, path,
6929 				       objectid, start, trans != NULL);
6930 	if (ret < 0) {
6931 		err = ret;
6932 		goto out;
6933 	}
6934 
6935 	if (ret != 0) {
6936 		if (path->slots[0] == 0)
6937 			goto not_found;
6938 		path->slots[0]--;
6939 	}
6940 
6941 	leaf = path->nodes[0];
6942 	item = btrfs_item_ptr(leaf, path->slots[0],
6943 			      struct btrfs_file_extent_item);
6944 	/* are we inside the extent that was found? */
6945 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6946 	found_type = found_key.type;
6947 	if (found_key.objectid != objectid ||
6948 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6949 		/*
6950 		 * If we backup past the first extent we want to move forward
6951 		 * and see if there is an extent in front of us, otherwise we'll
6952 		 * say there is a hole for our whole search range which can
6953 		 * cause problems.
6954 		 */
6955 		extent_end = start;
6956 		goto next;
6957 	}
6958 
6959 	found_type = btrfs_file_extent_type(leaf, item);
6960 	extent_start = found_key.offset;
6961 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6962 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6963 		extent_end = extent_start +
6964 		       btrfs_file_extent_num_bytes(leaf, item);
6965 
6966 		trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6967 						       extent_start);
6968 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6969 		size_t size;
6970 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6971 		extent_end = ALIGN(extent_start + size,
6972 				   fs_info->sectorsize);
6973 
6974 		trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6975 						      path->slots[0],
6976 						      extent_start);
6977 	}
6978 next:
6979 	if (start >= extent_end) {
6980 		path->slots[0]++;
6981 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6982 			ret = btrfs_next_leaf(root, path);
6983 			if (ret < 0) {
6984 				err = ret;
6985 				goto out;
6986 			}
6987 			if (ret > 0)
6988 				goto not_found;
6989 			leaf = path->nodes[0];
6990 		}
6991 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6992 		if (found_key.objectid != objectid ||
6993 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6994 			goto not_found;
6995 		if (start + len <= found_key.offset)
6996 			goto not_found;
6997 		if (start > found_key.offset)
6998 			goto next;
6999 		em->start = start;
7000 		em->orig_start = start;
7001 		em->len = found_key.offset - start;
7002 		goto not_found_em;
7003 	}
7004 
7005 	btrfs_extent_item_to_extent_map(inode, path, item,
7006 			new_inline, em);
7007 
7008 	if (found_type == BTRFS_FILE_EXTENT_REG ||
7009 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7010 		goto insert;
7011 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7012 		unsigned long ptr;
7013 		char *map;
7014 		size_t size;
7015 		size_t extent_offset;
7016 		size_t copy_size;
7017 
7018 		if (new_inline)
7019 			goto out;
7020 
7021 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7022 		extent_offset = page_offset(page) + pg_offset - extent_start;
7023 		copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7024 				  size - extent_offset);
7025 		em->start = extent_start + extent_offset;
7026 		em->len = ALIGN(copy_size, fs_info->sectorsize);
7027 		em->orig_block_len = em->len;
7028 		em->orig_start = em->start;
7029 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7030 		if (create == 0 && !PageUptodate(page)) {
7031 			if (btrfs_file_extent_compression(leaf, item) !=
7032 			    BTRFS_COMPRESS_NONE) {
7033 				ret = uncompress_inline(path, page, pg_offset,
7034 							extent_offset, item);
7035 				if (ret) {
7036 					err = ret;
7037 					goto out;
7038 				}
7039 			} else {
7040 				map = kmap(page);
7041 				read_extent_buffer(leaf, map + pg_offset, ptr,
7042 						   copy_size);
7043 				if (pg_offset + copy_size < PAGE_SIZE) {
7044 					memset(map + pg_offset + copy_size, 0,
7045 					       PAGE_SIZE - pg_offset -
7046 					       copy_size);
7047 				}
7048 				kunmap(page);
7049 			}
7050 			flush_dcache_page(page);
7051 		} else if (create && PageUptodate(page)) {
7052 			BUG();
7053 			if (!trans) {
7054 				kunmap(page);
7055 				free_extent_map(em);
7056 				em = NULL;
7057 
7058 				btrfs_release_path(path);
7059 				trans = btrfs_join_transaction(root);
7060 
7061 				if (IS_ERR(trans))
7062 					return ERR_CAST(trans);
7063 				goto again;
7064 			}
7065 			map = kmap(page);
7066 			write_extent_buffer(leaf, map + pg_offset, ptr,
7067 					    copy_size);
7068 			kunmap(page);
7069 			btrfs_mark_buffer_dirty(leaf);
7070 		}
7071 		set_extent_uptodate(io_tree, em->start,
7072 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
7073 		goto insert;
7074 	}
7075 not_found:
7076 	em->start = start;
7077 	em->orig_start = start;
7078 	em->len = len;
7079 not_found_em:
7080 	em->block_start = EXTENT_MAP_HOLE;
7081 	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7082 insert:
7083 	btrfs_release_path(path);
7084 	if (em->start > start || extent_map_end(em) <= start) {
7085 		btrfs_err(fs_info,
7086 			  "bad extent! em: [%llu %llu] passed [%llu %llu]",
7087 			  em->start, em->len, start, len);
7088 		err = -EIO;
7089 		goto out;
7090 	}
7091 
7092 	err = 0;
7093 	write_lock(&em_tree->lock);
7094 	ret = add_extent_mapping(em_tree, em, 0);
7095 	/* it is possible that someone inserted the extent into the tree
7096 	 * while we had the lock dropped.  It is also possible that
7097 	 * an overlapping map exists in the tree
7098 	 */
7099 	if (ret == -EEXIST) {
7100 		struct extent_map *existing;
7101 
7102 		ret = 0;
7103 
7104 		existing = search_extent_mapping(em_tree, start, len);
7105 		/*
7106 		 * existing will always be non-NULL, since there must be
7107 		 * extent causing the -EEXIST.
7108 		 */
7109 		if (existing->start == em->start &&
7110 		    extent_map_end(existing) >= extent_map_end(em) &&
7111 		    em->block_start == existing->block_start) {
7112 			/*
7113 			 * The existing extent map already encompasses the
7114 			 * entire extent map we tried to add.
7115 			 */
7116 			free_extent_map(em);
7117 			em = existing;
7118 			err = 0;
7119 
7120 		} else if (start >= extent_map_end(existing) ||
7121 		    start <= existing->start) {
7122 			/*
7123 			 * The existing extent map is the one nearest to
7124 			 * the [start, start + len) range which overlaps
7125 			 */
7126 			err = merge_extent_mapping(em_tree, existing,
7127 						   em, start);
7128 			free_extent_map(existing);
7129 			if (err) {
7130 				free_extent_map(em);
7131 				em = NULL;
7132 			}
7133 		} else {
7134 			free_extent_map(em);
7135 			em = existing;
7136 			err = 0;
7137 		}
7138 	}
7139 	write_unlock(&em_tree->lock);
7140 out:
7141 
7142 	trace_btrfs_get_extent(root, inode, em);
7143 
7144 	btrfs_free_path(path);
7145 	if (trans) {
7146 		ret = btrfs_end_transaction(trans);
7147 		if (!err)
7148 			err = ret;
7149 	}
7150 	if (err) {
7151 		free_extent_map(em);
7152 		return ERR_PTR(err);
7153 	}
7154 	BUG_ON(!em); /* Error is always set */
7155 	return em;
7156 }
7157 
7158 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7159 		struct page *page,
7160 		size_t pg_offset, u64 start, u64 len,
7161 		int create)
7162 {
7163 	struct extent_map *em;
7164 	struct extent_map *hole_em = NULL;
7165 	u64 range_start = start;
7166 	u64 end;
7167 	u64 found;
7168 	u64 found_end;
7169 	int err = 0;
7170 
7171 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7172 	if (IS_ERR(em))
7173 		return em;
7174 	/*
7175 	 * If our em maps to:
7176 	 * - a hole or
7177 	 * - a pre-alloc extent,
7178 	 * there might actually be delalloc bytes behind it.
7179 	 */
7180 	if (em->block_start != EXTENT_MAP_HOLE &&
7181 	    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7182 		return em;
7183 	else
7184 		hole_em = em;
7185 
7186 	/* check to see if we've wrapped (len == -1 or similar) */
7187 	end = start + len;
7188 	if (end < start)
7189 		end = (u64)-1;
7190 	else
7191 		end -= 1;
7192 
7193 	em = NULL;
7194 
7195 	/* ok, we didn't find anything, lets look for delalloc */
7196 	found = count_range_bits(&inode->io_tree, &range_start,
7197 				 end, len, EXTENT_DELALLOC, 1);
7198 	found_end = range_start + found;
7199 	if (found_end < range_start)
7200 		found_end = (u64)-1;
7201 
7202 	/*
7203 	 * we didn't find anything useful, return
7204 	 * the original results from get_extent()
7205 	 */
7206 	if (range_start > end || found_end <= start) {
7207 		em = hole_em;
7208 		hole_em = NULL;
7209 		goto out;
7210 	}
7211 
7212 	/* adjust the range_start to make sure it doesn't
7213 	 * go backwards from the start they passed in
7214 	 */
7215 	range_start = max(start, range_start);
7216 	found = found_end - range_start;
7217 
7218 	if (found > 0) {
7219 		u64 hole_start = start;
7220 		u64 hole_len = len;
7221 
7222 		em = alloc_extent_map();
7223 		if (!em) {
7224 			err = -ENOMEM;
7225 			goto out;
7226 		}
7227 		/*
7228 		 * when btrfs_get_extent can't find anything it
7229 		 * returns one huge hole
7230 		 *
7231 		 * make sure what it found really fits our range, and
7232 		 * adjust to make sure it is based on the start from
7233 		 * the caller
7234 		 */
7235 		if (hole_em) {
7236 			u64 calc_end = extent_map_end(hole_em);
7237 
7238 			if (calc_end <= start || (hole_em->start > end)) {
7239 				free_extent_map(hole_em);
7240 				hole_em = NULL;
7241 			} else {
7242 				hole_start = max(hole_em->start, start);
7243 				hole_len = calc_end - hole_start;
7244 			}
7245 		}
7246 		em->bdev = NULL;
7247 		if (hole_em && range_start > hole_start) {
7248 			/* our hole starts before our delalloc, so we
7249 			 * have to return just the parts of the hole
7250 			 * that go until  the delalloc starts
7251 			 */
7252 			em->len = min(hole_len,
7253 				      range_start - hole_start);
7254 			em->start = hole_start;
7255 			em->orig_start = hole_start;
7256 			/*
7257 			 * don't adjust block start at all,
7258 			 * it is fixed at EXTENT_MAP_HOLE
7259 			 */
7260 			em->block_start = hole_em->block_start;
7261 			em->block_len = hole_len;
7262 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7263 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7264 		} else {
7265 			em->start = range_start;
7266 			em->len = found;
7267 			em->orig_start = range_start;
7268 			em->block_start = EXTENT_MAP_DELALLOC;
7269 			em->block_len = found;
7270 		}
7271 	} else if (hole_em) {
7272 		return hole_em;
7273 	}
7274 out:
7275 
7276 	free_extent_map(hole_em);
7277 	if (err) {
7278 		free_extent_map(em);
7279 		return ERR_PTR(err);
7280 	}
7281 	return em;
7282 }
7283 
7284 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7285 						  const u64 start,
7286 						  const u64 len,
7287 						  const u64 orig_start,
7288 						  const u64 block_start,
7289 						  const u64 block_len,
7290 						  const u64 orig_block_len,
7291 						  const u64 ram_bytes,
7292 						  const int type)
7293 {
7294 	struct extent_map *em = NULL;
7295 	int ret;
7296 
7297 	if (type != BTRFS_ORDERED_NOCOW) {
7298 		em = create_io_em(inode, start, len, orig_start,
7299 				  block_start, block_len, orig_block_len,
7300 				  ram_bytes,
7301 				  BTRFS_COMPRESS_NONE, /* compress_type */
7302 				  type);
7303 		if (IS_ERR(em))
7304 			goto out;
7305 	}
7306 	ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7307 					   len, block_len, type);
7308 	if (ret) {
7309 		if (em) {
7310 			free_extent_map(em);
7311 			btrfs_drop_extent_cache(BTRFS_I(inode), start,
7312 						start + len - 1, 0);
7313 		}
7314 		em = ERR_PTR(ret);
7315 	}
7316  out:
7317 
7318 	return em;
7319 }
7320 
7321 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7322 						  u64 start, u64 len)
7323 {
7324 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7325 	struct btrfs_root *root = BTRFS_I(inode)->root;
7326 	struct extent_map *em;
7327 	struct btrfs_key ins;
7328 	u64 alloc_hint;
7329 	int ret;
7330 
7331 	alloc_hint = get_extent_allocation_hint(inode, start, len);
7332 	ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7333 				   0, alloc_hint, &ins, 1, 1);
7334 	if (ret)
7335 		return ERR_PTR(ret);
7336 
7337 	em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7338 				     ins.objectid, ins.offset, ins.offset,
7339 				     ins.offset, BTRFS_ORDERED_REGULAR);
7340 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7341 	if (IS_ERR(em))
7342 		btrfs_free_reserved_extent(fs_info, ins.objectid,
7343 					   ins.offset, 1);
7344 
7345 	return em;
7346 }
7347 
7348 /*
7349  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7350  * block must be cow'd
7351  */
7352 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7353 			      u64 *orig_start, u64 *orig_block_len,
7354 			      u64 *ram_bytes)
7355 {
7356 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7357 	struct btrfs_path *path;
7358 	int ret;
7359 	struct extent_buffer *leaf;
7360 	struct btrfs_root *root = BTRFS_I(inode)->root;
7361 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7362 	struct btrfs_file_extent_item *fi;
7363 	struct btrfs_key key;
7364 	u64 disk_bytenr;
7365 	u64 backref_offset;
7366 	u64 extent_end;
7367 	u64 num_bytes;
7368 	int slot;
7369 	int found_type;
7370 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7371 
7372 	path = btrfs_alloc_path();
7373 	if (!path)
7374 		return -ENOMEM;
7375 
7376 	ret = btrfs_lookup_file_extent(NULL, root, path,
7377 			btrfs_ino(BTRFS_I(inode)), offset, 0);
7378 	if (ret < 0)
7379 		goto out;
7380 
7381 	slot = path->slots[0];
7382 	if (ret == 1) {
7383 		if (slot == 0) {
7384 			/* can't find the item, must cow */
7385 			ret = 0;
7386 			goto out;
7387 		}
7388 		slot--;
7389 	}
7390 	ret = 0;
7391 	leaf = path->nodes[0];
7392 	btrfs_item_key_to_cpu(leaf, &key, slot);
7393 	if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7394 	    key.type != BTRFS_EXTENT_DATA_KEY) {
7395 		/* not our file or wrong item type, must cow */
7396 		goto out;
7397 	}
7398 
7399 	if (key.offset > offset) {
7400 		/* Wrong offset, must cow */
7401 		goto out;
7402 	}
7403 
7404 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7405 	found_type = btrfs_file_extent_type(leaf, fi);
7406 	if (found_type != BTRFS_FILE_EXTENT_REG &&
7407 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7408 		/* not a regular extent, must cow */
7409 		goto out;
7410 	}
7411 
7412 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7413 		goto out;
7414 
7415 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7416 	if (extent_end <= offset)
7417 		goto out;
7418 
7419 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7420 	if (disk_bytenr == 0)
7421 		goto out;
7422 
7423 	if (btrfs_file_extent_compression(leaf, fi) ||
7424 	    btrfs_file_extent_encryption(leaf, fi) ||
7425 	    btrfs_file_extent_other_encoding(leaf, fi))
7426 		goto out;
7427 
7428 	backref_offset = btrfs_file_extent_offset(leaf, fi);
7429 
7430 	if (orig_start) {
7431 		*orig_start = key.offset - backref_offset;
7432 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7433 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7434 	}
7435 
7436 	if (btrfs_extent_readonly(fs_info, disk_bytenr))
7437 		goto out;
7438 
7439 	num_bytes = min(offset + *len, extent_end) - offset;
7440 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7441 		u64 range_end;
7442 
7443 		range_end = round_up(offset + num_bytes,
7444 				     root->fs_info->sectorsize) - 1;
7445 		ret = test_range_bit(io_tree, offset, range_end,
7446 				     EXTENT_DELALLOC, 0, NULL);
7447 		if (ret) {
7448 			ret = -EAGAIN;
7449 			goto out;
7450 		}
7451 	}
7452 
7453 	btrfs_release_path(path);
7454 
7455 	/*
7456 	 * look for other files referencing this extent, if we
7457 	 * find any we must cow
7458 	 */
7459 
7460 	ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7461 				    key.offset - backref_offset, disk_bytenr);
7462 	if (ret) {
7463 		ret = 0;
7464 		goto out;
7465 	}
7466 
7467 	/*
7468 	 * adjust disk_bytenr and num_bytes to cover just the bytes
7469 	 * in this extent we are about to write.  If there
7470 	 * are any csums in that range we have to cow in order
7471 	 * to keep the csums correct
7472 	 */
7473 	disk_bytenr += backref_offset;
7474 	disk_bytenr += offset - key.offset;
7475 	if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7476 		goto out;
7477 	/*
7478 	 * all of the above have passed, it is safe to overwrite this extent
7479 	 * without cow
7480 	 */
7481 	*len = num_bytes;
7482 	ret = 1;
7483 out:
7484 	btrfs_free_path(path);
7485 	return ret;
7486 }
7487 
7488 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7489 {
7490 	struct radix_tree_root *root = &inode->i_mapping->page_tree;
7491 	bool found = false;
7492 	void **pagep = NULL;
7493 	struct page *page = NULL;
7494 	unsigned long start_idx;
7495 	unsigned long end_idx;
7496 
7497 	start_idx = start >> PAGE_SHIFT;
7498 
7499 	/*
7500 	 * end is the last byte in the last page.  end == start is legal
7501 	 */
7502 	end_idx = end >> PAGE_SHIFT;
7503 
7504 	rcu_read_lock();
7505 
7506 	/* Most of the code in this while loop is lifted from
7507 	 * find_get_page.  It's been modified to begin searching from a
7508 	 * page and return just the first page found in that range.  If the
7509 	 * found idx is less than or equal to the end idx then we know that
7510 	 * a page exists.  If no pages are found or if those pages are
7511 	 * outside of the range then we're fine (yay!) */
7512 	while (page == NULL &&
7513 	       radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7514 		page = radix_tree_deref_slot(pagep);
7515 		if (unlikely(!page))
7516 			break;
7517 
7518 		if (radix_tree_exception(page)) {
7519 			if (radix_tree_deref_retry(page)) {
7520 				page = NULL;
7521 				continue;
7522 			}
7523 			/*
7524 			 * Otherwise, shmem/tmpfs must be storing a swap entry
7525 			 * here as an exceptional entry: so return it without
7526 			 * attempting to raise page count.
7527 			 */
7528 			page = NULL;
7529 			break; /* TODO: Is this relevant for this use case? */
7530 		}
7531 
7532 		if (!page_cache_get_speculative(page)) {
7533 			page = NULL;
7534 			continue;
7535 		}
7536 
7537 		/*
7538 		 * Has the page moved?
7539 		 * This is part of the lockless pagecache protocol. See
7540 		 * include/linux/pagemap.h for details.
7541 		 */
7542 		if (unlikely(page != *pagep)) {
7543 			put_page(page);
7544 			page = NULL;
7545 		}
7546 	}
7547 
7548 	if (page) {
7549 		if (page->index <= end_idx)
7550 			found = true;
7551 		put_page(page);
7552 	}
7553 
7554 	rcu_read_unlock();
7555 	return found;
7556 }
7557 
7558 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7559 			      struct extent_state **cached_state, int writing)
7560 {
7561 	struct btrfs_ordered_extent *ordered;
7562 	int ret = 0;
7563 
7564 	while (1) {
7565 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7566 				 cached_state);
7567 		/*
7568 		 * We're concerned with the entire range that we're going to be
7569 		 * doing DIO to, so we need to make sure there's no ordered
7570 		 * extents in this range.
7571 		 */
7572 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7573 						     lockend - lockstart + 1);
7574 
7575 		/*
7576 		 * We need to make sure there are no buffered pages in this
7577 		 * range either, we could have raced between the invalidate in
7578 		 * generic_file_direct_write and locking the extent.  The
7579 		 * invalidate needs to happen so that reads after a write do not
7580 		 * get stale data.
7581 		 */
7582 		if (!ordered &&
7583 		    (!writing ||
7584 		     !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7585 			break;
7586 
7587 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7588 				     cached_state, GFP_NOFS);
7589 
7590 		if (ordered) {
7591 			/*
7592 			 * If we are doing a DIO read and the ordered extent we
7593 			 * found is for a buffered write, we can not wait for it
7594 			 * to complete and retry, because if we do so we can
7595 			 * deadlock with concurrent buffered writes on page
7596 			 * locks. This happens only if our DIO read covers more
7597 			 * than one extent map, if at this point has already
7598 			 * created an ordered extent for a previous extent map
7599 			 * and locked its range in the inode's io tree, and a
7600 			 * concurrent write against that previous extent map's
7601 			 * range and this range started (we unlock the ranges
7602 			 * in the io tree only when the bios complete and
7603 			 * buffered writes always lock pages before attempting
7604 			 * to lock range in the io tree).
7605 			 */
7606 			if (writing ||
7607 			    test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7608 				btrfs_start_ordered_extent(inode, ordered, 1);
7609 			else
7610 				ret = -ENOTBLK;
7611 			btrfs_put_ordered_extent(ordered);
7612 		} else {
7613 			/*
7614 			 * We could trigger writeback for this range (and wait
7615 			 * for it to complete) and then invalidate the pages for
7616 			 * this range (through invalidate_inode_pages2_range()),
7617 			 * but that can lead us to a deadlock with a concurrent
7618 			 * call to readpages() (a buffered read or a defrag call
7619 			 * triggered a readahead) on a page lock due to an
7620 			 * ordered dio extent we created before but did not have
7621 			 * yet a corresponding bio submitted (whence it can not
7622 			 * complete), which makes readpages() wait for that
7623 			 * ordered extent to complete while holding a lock on
7624 			 * that page.
7625 			 */
7626 			ret = -ENOTBLK;
7627 		}
7628 
7629 		if (ret)
7630 			break;
7631 
7632 		cond_resched();
7633 	}
7634 
7635 	return ret;
7636 }
7637 
7638 /* The callers of this must take lock_extent() */
7639 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7640 				       u64 orig_start, u64 block_start,
7641 				       u64 block_len, u64 orig_block_len,
7642 				       u64 ram_bytes, int compress_type,
7643 				       int type)
7644 {
7645 	struct extent_map_tree *em_tree;
7646 	struct extent_map *em;
7647 	struct btrfs_root *root = BTRFS_I(inode)->root;
7648 	int ret;
7649 
7650 	ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7651 	       type == BTRFS_ORDERED_COMPRESSED ||
7652 	       type == BTRFS_ORDERED_NOCOW ||
7653 	       type == BTRFS_ORDERED_REGULAR);
7654 
7655 	em_tree = &BTRFS_I(inode)->extent_tree;
7656 	em = alloc_extent_map();
7657 	if (!em)
7658 		return ERR_PTR(-ENOMEM);
7659 
7660 	em->start = start;
7661 	em->orig_start = orig_start;
7662 	em->len = len;
7663 	em->block_len = block_len;
7664 	em->block_start = block_start;
7665 	em->bdev = root->fs_info->fs_devices->latest_bdev;
7666 	em->orig_block_len = orig_block_len;
7667 	em->ram_bytes = ram_bytes;
7668 	em->generation = -1;
7669 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
7670 	if (type == BTRFS_ORDERED_PREALLOC) {
7671 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
7672 	} else if (type == BTRFS_ORDERED_COMPRESSED) {
7673 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7674 		em->compress_type = compress_type;
7675 	}
7676 
7677 	do {
7678 		btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7679 				em->start + em->len - 1, 0);
7680 		write_lock(&em_tree->lock);
7681 		ret = add_extent_mapping(em_tree, em, 1);
7682 		write_unlock(&em_tree->lock);
7683 		/*
7684 		 * The caller has taken lock_extent(), who could race with us
7685 		 * to add em?
7686 		 */
7687 	} while (ret == -EEXIST);
7688 
7689 	if (ret) {
7690 		free_extent_map(em);
7691 		return ERR_PTR(ret);
7692 	}
7693 
7694 	/* em got 2 refs now, callers needs to do free_extent_map once. */
7695 	return em;
7696 }
7697 
7698 static void adjust_dio_outstanding_extents(struct inode *inode,
7699 					   struct btrfs_dio_data *dio_data,
7700 					   const u64 len)
7701 {
7702 	unsigned num_extents = count_max_extents(len);
7703 
7704 	/*
7705 	 * If we have an outstanding_extents count still set then we're
7706 	 * within our reservation, otherwise we need to adjust our inode
7707 	 * counter appropriately.
7708 	 */
7709 	if (dio_data->outstanding_extents >= num_extents) {
7710 		dio_data->outstanding_extents -= num_extents;
7711 	} else {
7712 		/*
7713 		 * If dio write length has been split due to no large enough
7714 		 * contiguous space, we need to compensate our inode counter
7715 		 * appropriately.
7716 		 */
7717 		u64 num_needed = num_extents - dio_data->outstanding_extents;
7718 
7719 		spin_lock(&BTRFS_I(inode)->lock);
7720 		BTRFS_I(inode)->outstanding_extents += num_needed;
7721 		spin_unlock(&BTRFS_I(inode)->lock);
7722 	}
7723 }
7724 
7725 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7726 				   struct buffer_head *bh_result, int create)
7727 {
7728 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7729 	struct extent_map *em;
7730 	struct extent_state *cached_state = NULL;
7731 	struct btrfs_dio_data *dio_data = NULL;
7732 	u64 start = iblock << inode->i_blkbits;
7733 	u64 lockstart, lockend;
7734 	u64 len = bh_result->b_size;
7735 	int unlock_bits = EXTENT_LOCKED;
7736 	int ret = 0;
7737 
7738 	if (create)
7739 		unlock_bits |= EXTENT_DIRTY;
7740 	else
7741 		len = min_t(u64, len, fs_info->sectorsize);
7742 
7743 	lockstart = start;
7744 	lockend = start + len - 1;
7745 
7746 	if (current->journal_info) {
7747 		/*
7748 		 * Need to pull our outstanding extents and set journal_info to NULL so
7749 		 * that anything that needs to check if there's a transaction doesn't get
7750 		 * confused.
7751 		 */
7752 		dio_data = current->journal_info;
7753 		current->journal_info = NULL;
7754 	}
7755 
7756 	/*
7757 	 * If this errors out it's because we couldn't invalidate pagecache for
7758 	 * this range and we need to fallback to buffered.
7759 	 */
7760 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7761 			       create)) {
7762 		ret = -ENOTBLK;
7763 		goto err;
7764 	}
7765 
7766 	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7767 	if (IS_ERR(em)) {
7768 		ret = PTR_ERR(em);
7769 		goto unlock_err;
7770 	}
7771 
7772 	/*
7773 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7774 	 * io.  INLINE is special, and we could probably kludge it in here, but
7775 	 * it's still buffered so for safety lets just fall back to the generic
7776 	 * buffered path.
7777 	 *
7778 	 * For COMPRESSED we _have_ to read the entire extent in so we can
7779 	 * decompress it, so there will be buffering required no matter what we
7780 	 * do, so go ahead and fallback to buffered.
7781 	 *
7782 	 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7783 	 * to buffered IO.  Don't blame me, this is the price we pay for using
7784 	 * the generic code.
7785 	 */
7786 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7787 	    em->block_start == EXTENT_MAP_INLINE) {
7788 		free_extent_map(em);
7789 		ret = -ENOTBLK;
7790 		goto unlock_err;
7791 	}
7792 
7793 	/* Just a good old fashioned hole, return */
7794 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7795 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7796 		free_extent_map(em);
7797 		goto unlock_err;
7798 	}
7799 
7800 	/*
7801 	 * We don't allocate a new extent in the following cases
7802 	 *
7803 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7804 	 * existing extent.
7805 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
7806 	 * just use the extent.
7807 	 *
7808 	 */
7809 	if (!create) {
7810 		len = min(len, em->len - (start - em->start));
7811 		lockstart = start + len;
7812 		goto unlock;
7813 	}
7814 
7815 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7816 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7817 	     em->block_start != EXTENT_MAP_HOLE)) {
7818 		int type;
7819 		u64 block_start, orig_start, orig_block_len, ram_bytes;
7820 
7821 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7822 			type = BTRFS_ORDERED_PREALLOC;
7823 		else
7824 			type = BTRFS_ORDERED_NOCOW;
7825 		len = min(len, em->len - (start - em->start));
7826 		block_start = em->block_start + (start - em->start);
7827 
7828 		if (can_nocow_extent(inode, start, &len, &orig_start,
7829 				     &orig_block_len, &ram_bytes) == 1 &&
7830 		    btrfs_inc_nocow_writers(fs_info, block_start)) {
7831 			struct extent_map *em2;
7832 
7833 			em2 = btrfs_create_dio_extent(inode, start, len,
7834 						      orig_start, block_start,
7835 						      len, orig_block_len,
7836 						      ram_bytes, type);
7837 			btrfs_dec_nocow_writers(fs_info, block_start);
7838 			if (type == BTRFS_ORDERED_PREALLOC) {
7839 				free_extent_map(em);
7840 				em = em2;
7841 			}
7842 			if (em2 && IS_ERR(em2)) {
7843 				ret = PTR_ERR(em2);
7844 				goto unlock_err;
7845 			}
7846 			/*
7847 			 * For inode marked NODATACOW or extent marked PREALLOC,
7848 			 * use the existing or preallocated extent, so does not
7849 			 * need to adjust btrfs_space_info's bytes_may_use.
7850 			 */
7851 			btrfs_free_reserved_data_space_noquota(inode,
7852 					start, len);
7853 			goto unlock;
7854 		}
7855 	}
7856 
7857 	/*
7858 	 * this will cow the extent, reset the len in case we changed
7859 	 * it above
7860 	 */
7861 	len = bh_result->b_size;
7862 	free_extent_map(em);
7863 	em = btrfs_new_extent_direct(inode, start, len);
7864 	if (IS_ERR(em)) {
7865 		ret = PTR_ERR(em);
7866 		goto unlock_err;
7867 	}
7868 	len = min(len, em->len - (start - em->start));
7869 unlock:
7870 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7871 		inode->i_blkbits;
7872 	bh_result->b_size = len;
7873 	bh_result->b_bdev = em->bdev;
7874 	set_buffer_mapped(bh_result);
7875 	if (create) {
7876 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7877 			set_buffer_new(bh_result);
7878 
7879 		/*
7880 		 * Need to update the i_size under the extent lock so buffered
7881 		 * readers will get the updated i_size when we unlock.
7882 		 */
7883 		if (!dio_data->overwrite && start + len > i_size_read(inode))
7884 			i_size_write(inode, start + len);
7885 
7886 		adjust_dio_outstanding_extents(inode, dio_data, len);
7887 		WARN_ON(dio_data->reserve < len);
7888 		dio_data->reserve -= len;
7889 		dio_data->unsubmitted_oe_range_end = start + len;
7890 		current->journal_info = dio_data;
7891 	}
7892 
7893 	/*
7894 	 * In the case of write we need to clear and unlock the entire range,
7895 	 * in the case of read we need to unlock only the end area that we
7896 	 * aren't using if there is any left over space.
7897 	 */
7898 	if (lockstart < lockend) {
7899 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7900 				 lockend, unlock_bits, 1, 0,
7901 				 &cached_state, GFP_NOFS);
7902 	} else {
7903 		free_extent_state(cached_state);
7904 	}
7905 
7906 	free_extent_map(em);
7907 
7908 	return 0;
7909 
7910 unlock_err:
7911 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7912 			 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7913 err:
7914 	if (dio_data)
7915 		current->journal_info = dio_data;
7916 	/*
7917 	 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7918 	 * write less data then expected, so that we don't underflow our inode's
7919 	 * outstanding extents counter.
7920 	 */
7921 	if (create && dio_data)
7922 		adjust_dio_outstanding_extents(inode, dio_data, len);
7923 
7924 	return ret;
7925 }
7926 
7927 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7928 					int mirror_num)
7929 {
7930 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7931 	int ret;
7932 
7933 	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7934 
7935 	bio_get(bio);
7936 
7937 	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7938 	if (ret)
7939 		goto err;
7940 
7941 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7942 err:
7943 	bio_put(bio);
7944 	return ret;
7945 }
7946 
7947 static int btrfs_check_dio_repairable(struct inode *inode,
7948 				      struct bio *failed_bio,
7949 				      struct io_failure_record *failrec,
7950 				      int failed_mirror)
7951 {
7952 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7953 	int num_copies;
7954 
7955 	num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7956 	if (num_copies == 1) {
7957 		/*
7958 		 * we only have a single copy of the data, so don't bother with
7959 		 * all the retry and error correction code that follows. no
7960 		 * matter what the error is, it is very likely to persist.
7961 		 */
7962 		btrfs_debug(fs_info,
7963 			"Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7964 			num_copies, failrec->this_mirror, failed_mirror);
7965 		return 0;
7966 	}
7967 
7968 	failrec->failed_mirror = failed_mirror;
7969 	failrec->this_mirror++;
7970 	if (failrec->this_mirror == failed_mirror)
7971 		failrec->this_mirror++;
7972 
7973 	if (failrec->this_mirror > num_copies) {
7974 		btrfs_debug(fs_info,
7975 			"Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7976 			num_copies, failrec->this_mirror, failed_mirror);
7977 		return 0;
7978 	}
7979 
7980 	return 1;
7981 }
7982 
7983 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7984 			struct page *page, unsigned int pgoff,
7985 			u64 start, u64 end, int failed_mirror,
7986 			bio_end_io_t *repair_endio, void *repair_arg)
7987 {
7988 	struct io_failure_record *failrec;
7989 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7990 	struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7991 	struct bio *bio;
7992 	int isector;
7993 	int read_mode = 0;
7994 	int segs;
7995 	int ret;
7996 
7997 	BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7998 
7999 	ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8000 	if (ret)
8001 		return ret;
8002 
8003 	ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8004 					 failed_mirror);
8005 	if (!ret) {
8006 		free_io_failure(failure_tree, io_tree, failrec);
8007 		return -EIO;
8008 	}
8009 
8010 	segs = bio_segments(failed_bio);
8011 	if (segs > 1 ||
8012 	    (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8013 		read_mode |= REQ_FAILFAST_DEV;
8014 
8015 	isector = start - btrfs_io_bio(failed_bio)->logical;
8016 	isector >>= inode->i_sb->s_blocksize_bits;
8017 	bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8018 				pgoff, isector, repair_endio, repair_arg);
8019 	bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8020 
8021 	btrfs_debug(BTRFS_I(inode)->root->fs_info,
8022 		    "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
8023 		    read_mode, failrec->this_mirror, failrec->in_validation);
8024 
8025 	ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8026 	if (ret) {
8027 		free_io_failure(failure_tree, io_tree, failrec);
8028 		bio_put(bio);
8029 	}
8030 
8031 	return ret;
8032 }
8033 
8034 struct btrfs_retry_complete {
8035 	struct completion done;
8036 	struct inode *inode;
8037 	u64 start;
8038 	int uptodate;
8039 };
8040 
8041 static void btrfs_retry_endio_nocsum(struct bio *bio)
8042 {
8043 	struct btrfs_retry_complete *done = bio->bi_private;
8044 	struct inode *inode = done->inode;
8045 	struct bio_vec *bvec;
8046 	struct extent_io_tree *io_tree, *failure_tree;
8047 	int i;
8048 
8049 	if (bio->bi_status)
8050 		goto end;
8051 
8052 	ASSERT(bio->bi_vcnt == 1);
8053 	io_tree = &BTRFS_I(inode)->io_tree;
8054 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
8055 	ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8056 
8057 	done->uptodate = 1;
8058 	ASSERT(!bio_flagged(bio, BIO_CLONED));
8059 	bio_for_each_segment_all(bvec, bio, i)
8060 		clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8061 				 io_tree, done->start, bvec->bv_page,
8062 				 btrfs_ino(BTRFS_I(inode)), 0);
8063 end:
8064 	complete(&done->done);
8065 	bio_put(bio);
8066 }
8067 
8068 static int __btrfs_correct_data_nocsum(struct inode *inode,
8069 				       struct btrfs_io_bio *io_bio)
8070 {
8071 	struct btrfs_fs_info *fs_info;
8072 	struct bio_vec bvec;
8073 	struct bvec_iter iter;
8074 	struct btrfs_retry_complete done;
8075 	u64 start;
8076 	unsigned int pgoff;
8077 	u32 sectorsize;
8078 	int nr_sectors;
8079 	int ret;
8080 	int err = 0;
8081 
8082 	fs_info = BTRFS_I(inode)->root->fs_info;
8083 	sectorsize = fs_info->sectorsize;
8084 
8085 	start = io_bio->logical;
8086 	done.inode = inode;
8087 	io_bio->bio.bi_iter = io_bio->iter;
8088 
8089 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
8090 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8091 		pgoff = bvec.bv_offset;
8092 
8093 next_block_or_try_again:
8094 		done.uptodate = 0;
8095 		done.start = start;
8096 		init_completion(&done.done);
8097 
8098 		ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8099 				pgoff, start, start + sectorsize - 1,
8100 				io_bio->mirror_num,
8101 				btrfs_retry_endio_nocsum, &done);
8102 		if (ret) {
8103 			err = ret;
8104 			goto next;
8105 		}
8106 
8107 		wait_for_completion(&done.done);
8108 
8109 		if (!done.uptodate) {
8110 			/* We might have another mirror, so try again */
8111 			goto next_block_or_try_again;
8112 		}
8113 
8114 next:
8115 		start += sectorsize;
8116 
8117 		nr_sectors--;
8118 		if (nr_sectors) {
8119 			pgoff += sectorsize;
8120 			ASSERT(pgoff < PAGE_SIZE);
8121 			goto next_block_or_try_again;
8122 		}
8123 	}
8124 
8125 	return err;
8126 }
8127 
8128 static void btrfs_retry_endio(struct bio *bio)
8129 {
8130 	struct btrfs_retry_complete *done = bio->bi_private;
8131 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8132 	struct extent_io_tree *io_tree, *failure_tree;
8133 	struct inode *inode = done->inode;
8134 	struct bio_vec *bvec;
8135 	int uptodate;
8136 	int ret;
8137 	int i;
8138 
8139 	if (bio->bi_status)
8140 		goto end;
8141 
8142 	uptodate = 1;
8143 
8144 	ASSERT(bio->bi_vcnt == 1);
8145 	ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8146 
8147 	io_tree = &BTRFS_I(inode)->io_tree;
8148 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
8149 
8150 	ASSERT(!bio_flagged(bio, BIO_CLONED));
8151 	bio_for_each_segment_all(bvec, bio, i) {
8152 		ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8153 					     bvec->bv_offset, done->start,
8154 					     bvec->bv_len);
8155 		if (!ret)
8156 			clean_io_failure(BTRFS_I(inode)->root->fs_info,
8157 					 failure_tree, io_tree, done->start,
8158 					 bvec->bv_page,
8159 					 btrfs_ino(BTRFS_I(inode)),
8160 					 bvec->bv_offset);
8161 		else
8162 			uptodate = 0;
8163 	}
8164 
8165 	done->uptodate = uptodate;
8166 end:
8167 	complete(&done->done);
8168 	bio_put(bio);
8169 }
8170 
8171 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8172 		struct btrfs_io_bio *io_bio, blk_status_t err)
8173 {
8174 	struct btrfs_fs_info *fs_info;
8175 	struct bio_vec bvec;
8176 	struct bvec_iter iter;
8177 	struct btrfs_retry_complete done;
8178 	u64 start;
8179 	u64 offset = 0;
8180 	u32 sectorsize;
8181 	int nr_sectors;
8182 	unsigned int pgoff;
8183 	int csum_pos;
8184 	bool uptodate = (err == 0);
8185 	int ret;
8186 
8187 	fs_info = BTRFS_I(inode)->root->fs_info;
8188 	sectorsize = fs_info->sectorsize;
8189 
8190 	err = 0;
8191 	start = io_bio->logical;
8192 	done.inode = inode;
8193 	io_bio->bio.bi_iter = io_bio->iter;
8194 
8195 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
8196 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8197 
8198 		pgoff = bvec.bv_offset;
8199 next_block:
8200 		if (uptodate) {
8201 			csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8202 			ret = __readpage_endio_check(inode, io_bio, csum_pos,
8203 					bvec.bv_page, pgoff, start, sectorsize);
8204 			if (likely(!ret))
8205 				goto next;
8206 		}
8207 try_again:
8208 		done.uptodate = 0;
8209 		done.start = start;
8210 		init_completion(&done.done);
8211 
8212 		ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8213 				pgoff, start, start + sectorsize - 1,
8214 				io_bio->mirror_num,
8215 				btrfs_retry_endio, &done);
8216 		if (ret) {
8217 			err = errno_to_blk_status(ret);
8218 			goto next;
8219 		}
8220 
8221 		wait_for_completion(&done.done);
8222 
8223 		if (!done.uptodate) {
8224 			/* We might have another mirror, so try again */
8225 			goto try_again;
8226 		}
8227 next:
8228 		offset += sectorsize;
8229 		start += sectorsize;
8230 
8231 		ASSERT(nr_sectors);
8232 
8233 		nr_sectors--;
8234 		if (nr_sectors) {
8235 			pgoff += sectorsize;
8236 			ASSERT(pgoff < PAGE_SIZE);
8237 			goto next_block;
8238 		}
8239 	}
8240 
8241 	return err;
8242 }
8243 
8244 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8245 		struct btrfs_io_bio *io_bio, blk_status_t err)
8246 {
8247 	bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8248 
8249 	if (skip_csum) {
8250 		if (unlikely(err))
8251 			return __btrfs_correct_data_nocsum(inode, io_bio);
8252 		else
8253 			return 0;
8254 	} else {
8255 		return __btrfs_subio_endio_read(inode, io_bio, err);
8256 	}
8257 }
8258 
8259 static void btrfs_endio_direct_read(struct bio *bio)
8260 {
8261 	struct btrfs_dio_private *dip = bio->bi_private;
8262 	struct inode *inode = dip->inode;
8263 	struct bio *dio_bio;
8264 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8265 	blk_status_t err = bio->bi_status;
8266 
8267 	if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) {
8268 		err = btrfs_subio_endio_read(inode, io_bio, err);
8269 		if (!err)
8270 			bio->bi_status = 0;
8271 	}
8272 
8273 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8274 		      dip->logical_offset + dip->bytes - 1);
8275 	dio_bio = dip->dio_bio;
8276 
8277 	kfree(dip);
8278 
8279 	dio_bio->bi_status = bio->bi_status;
8280 	dio_end_io(dio_bio);
8281 
8282 	if (io_bio->end_io)
8283 		io_bio->end_io(io_bio, blk_status_to_errno(err));
8284 	bio_put(bio);
8285 }
8286 
8287 static void __endio_write_update_ordered(struct inode *inode,
8288 					 const u64 offset, const u64 bytes,
8289 					 const bool uptodate)
8290 {
8291 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8292 	struct btrfs_ordered_extent *ordered = NULL;
8293 	struct btrfs_workqueue *wq;
8294 	btrfs_work_func_t func;
8295 	u64 ordered_offset = offset;
8296 	u64 ordered_bytes = bytes;
8297 	int ret;
8298 
8299 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8300 		wq = fs_info->endio_freespace_worker;
8301 		func = btrfs_freespace_write_helper;
8302 	} else {
8303 		wq = fs_info->endio_write_workers;
8304 		func = btrfs_endio_write_helper;
8305 	}
8306 
8307 again:
8308 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8309 						   &ordered_offset,
8310 						   ordered_bytes,
8311 						   uptodate);
8312 	if (!ret)
8313 		goto out_test;
8314 
8315 	btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8316 	btrfs_queue_work(wq, &ordered->work);
8317 out_test:
8318 	/*
8319 	 * our bio might span multiple ordered extents.  If we haven't
8320 	 * completed the accounting for the whole dio, go back and try again
8321 	 */
8322 	if (ordered_offset < offset + bytes) {
8323 		ordered_bytes = offset + bytes - ordered_offset;
8324 		ordered = NULL;
8325 		goto again;
8326 	}
8327 }
8328 
8329 static void btrfs_endio_direct_write(struct bio *bio)
8330 {
8331 	struct btrfs_dio_private *dip = bio->bi_private;
8332 	struct bio *dio_bio = dip->dio_bio;
8333 
8334 	__endio_write_update_ordered(dip->inode, dip->logical_offset,
8335 				     dip->bytes, !bio->bi_status);
8336 
8337 	kfree(dip);
8338 
8339 	dio_bio->bi_status = bio->bi_status;
8340 	dio_end_io(dio_bio);
8341 	bio_put(bio);
8342 }
8343 
8344 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8345 				    struct bio *bio, int mirror_num,
8346 				    unsigned long bio_flags, u64 offset)
8347 {
8348 	struct inode *inode = private_data;
8349 	blk_status_t ret;
8350 	ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8351 	BUG_ON(ret); /* -ENOMEM */
8352 	return 0;
8353 }
8354 
8355 static void btrfs_end_dio_bio(struct bio *bio)
8356 {
8357 	struct btrfs_dio_private *dip = bio->bi_private;
8358 	blk_status_t err = bio->bi_status;
8359 
8360 	if (err)
8361 		btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8362 			   "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8363 			   btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8364 			   bio->bi_opf,
8365 			   (unsigned long long)bio->bi_iter.bi_sector,
8366 			   bio->bi_iter.bi_size, err);
8367 
8368 	if (dip->subio_endio)
8369 		err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8370 
8371 	if (err) {
8372 		dip->errors = 1;
8373 
8374 		/*
8375 		 * before atomic variable goto zero, we must make sure
8376 		 * dip->errors is perceived to be set.
8377 		 */
8378 		smp_mb__before_atomic();
8379 	}
8380 
8381 	/* if there are more bios still pending for this dio, just exit */
8382 	if (!atomic_dec_and_test(&dip->pending_bios))
8383 		goto out;
8384 
8385 	if (dip->errors) {
8386 		bio_io_error(dip->orig_bio);
8387 	} else {
8388 		dip->dio_bio->bi_status = 0;
8389 		bio_endio(dip->orig_bio);
8390 	}
8391 out:
8392 	bio_put(bio);
8393 }
8394 
8395 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8396 						 struct btrfs_dio_private *dip,
8397 						 struct bio *bio,
8398 						 u64 file_offset)
8399 {
8400 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8401 	struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8402 	blk_status_t ret;
8403 
8404 	/*
8405 	 * We load all the csum data we need when we submit
8406 	 * the first bio to reduce the csum tree search and
8407 	 * contention.
8408 	 */
8409 	if (dip->logical_offset == file_offset) {
8410 		ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8411 						file_offset);
8412 		if (ret)
8413 			return ret;
8414 	}
8415 
8416 	if (bio == dip->orig_bio)
8417 		return 0;
8418 
8419 	file_offset -= dip->logical_offset;
8420 	file_offset >>= inode->i_sb->s_blocksize_bits;
8421 	io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8422 
8423 	return 0;
8424 }
8425 
8426 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8427 					 u64 file_offset, int skip_sum,
8428 					 int async_submit)
8429 {
8430 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8431 	struct btrfs_dio_private *dip = bio->bi_private;
8432 	bool write = bio_op(bio) == REQ_OP_WRITE;
8433 	blk_status_t ret;
8434 
8435 	if (async_submit)
8436 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8437 
8438 	bio_get(bio);
8439 
8440 	if (!write) {
8441 		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8442 		if (ret)
8443 			goto err;
8444 	}
8445 
8446 	if (skip_sum)
8447 		goto map;
8448 
8449 	if (write && async_submit) {
8450 		ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8451 					  file_offset, inode,
8452 					  __btrfs_submit_bio_start_direct_io,
8453 					  __btrfs_submit_bio_done);
8454 		goto err;
8455 	} else if (write) {
8456 		/*
8457 		 * If we aren't doing async submit, calculate the csum of the
8458 		 * bio now.
8459 		 */
8460 		ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8461 		if (ret)
8462 			goto err;
8463 	} else {
8464 		ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8465 						     file_offset);
8466 		if (ret)
8467 			goto err;
8468 	}
8469 map:
8470 	ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8471 err:
8472 	bio_put(bio);
8473 	return ret;
8474 }
8475 
8476 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8477 				    int skip_sum)
8478 {
8479 	struct inode *inode = dip->inode;
8480 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8481 	struct bio *bio;
8482 	struct bio *orig_bio = dip->orig_bio;
8483 	u64 start_sector = orig_bio->bi_iter.bi_sector;
8484 	u64 file_offset = dip->logical_offset;
8485 	u64 map_length;
8486 	int async_submit = 0;
8487 	u64 submit_len;
8488 	int clone_offset = 0;
8489 	int clone_len;
8490 	int ret;
8491 
8492 	map_length = orig_bio->bi_iter.bi_size;
8493 	submit_len = map_length;
8494 	ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8495 			      &map_length, NULL, 0);
8496 	if (ret)
8497 		return -EIO;
8498 
8499 	if (map_length >= submit_len) {
8500 		bio = orig_bio;
8501 		dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8502 		goto submit;
8503 	}
8504 
8505 	/* async crcs make it difficult to collect full stripe writes. */
8506 	if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8507 		async_submit = 0;
8508 	else
8509 		async_submit = 1;
8510 
8511 	/* bio split */
8512 	ASSERT(map_length <= INT_MAX);
8513 	atomic_inc(&dip->pending_bios);
8514 	do {
8515 		clone_len = min_t(int, submit_len, map_length);
8516 
8517 		/*
8518 		 * This will never fail as it's passing GPF_NOFS and
8519 		 * the allocation is backed by btrfs_bioset.
8520 		 */
8521 		bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8522 					      clone_len);
8523 		bio->bi_private = dip;
8524 		bio->bi_end_io = btrfs_end_dio_bio;
8525 		btrfs_io_bio(bio)->logical = file_offset;
8526 
8527 		ASSERT(submit_len >= clone_len);
8528 		submit_len -= clone_len;
8529 		if (submit_len == 0)
8530 			break;
8531 
8532 		/*
8533 		 * Increase the count before we submit the bio so we know
8534 		 * the end IO handler won't happen before we increase the
8535 		 * count. Otherwise, the dip might get freed before we're
8536 		 * done setting it up.
8537 		 */
8538 		atomic_inc(&dip->pending_bios);
8539 
8540 		ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8541 					     async_submit);
8542 		if (ret) {
8543 			bio_put(bio);
8544 			atomic_dec(&dip->pending_bios);
8545 			goto out_err;
8546 		}
8547 
8548 		clone_offset += clone_len;
8549 		start_sector += clone_len >> 9;
8550 		file_offset += clone_len;
8551 
8552 		map_length = submit_len;
8553 		ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8554 				      start_sector << 9, &map_length, NULL, 0);
8555 		if (ret)
8556 			goto out_err;
8557 	} while (submit_len > 0);
8558 
8559 submit:
8560 	ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8561 				     async_submit);
8562 	if (!ret)
8563 		return 0;
8564 
8565 	bio_put(bio);
8566 out_err:
8567 	dip->errors = 1;
8568 	/*
8569 	 * before atomic variable goto zero, we must
8570 	 * make sure dip->errors is perceived to be set.
8571 	 */
8572 	smp_mb__before_atomic();
8573 	if (atomic_dec_and_test(&dip->pending_bios))
8574 		bio_io_error(dip->orig_bio);
8575 
8576 	/* bio_end_io() will handle error, so we needn't return it */
8577 	return 0;
8578 }
8579 
8580 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8581 				loff_t file_offset)
8582 {
8583 	struct btrfs_dio_private *dip = NULL;
8584 	struct bio *bio = NULL;
8585 	struct btrfs_io_bio *io_bio;
8586 	int skip_sum;
8587 	bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8588 	int ret = 0;
8589 
8590 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8591 
8592 	bio = btrfs_bio_clone(dio_bio);
8593 
8594 	dip = kzalloc(sizeof(*dip), GFP_NOFS);
8595 	if (!dip) {
8596 		ret = -ENOMEM;
8597 		goto free_ordered;
8598 	}
8599 
8600 	dip->private = dio_bio->bi_private;
8601 	dip->inode = inode;
8602 	dip->logical_offset = file_offset;
8603 	dip->bytes = dio_bio->bi_iter.bi_size;
8604 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8605 	bio->bi_private = dip;
8606 	dip->orig_bio = bio;
8607 	dip->dio_bio = dio_bio;
8608 	atomic_set(&dip->pending_bios, 0);
8609 	io_bio = btrfs_io_bio(bio);
8610 	io_bio->logical = file_offset;
8611 
8612 	if (write) {
8613 		bio->bi_end_io = btrfs_endio_direct_write;
8614 	} else {
8615 		bio->bi_end_io = btrfs_endio_direct_read;
8616 		dip->subio_endio = btrfs_subio_endio_read;
8617 	}
8618 
8619 	/*
8620 	 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8621 	 * even if we fail to submit a bio, because in such case we do the
8622 	 * corresponding error handling below and it must not be done a second
8623 	 * time by btrfs_direct_IO().
8624 	 */
8625 	if (write) {
8626 		struct btrfs_dio_data *dio_data = current->journal_info;
8627 
8628 		dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8629 			dip->bytes;
8630 		dio_data->unsubmitted_oe_range_start =
8631 			dio_data->unsubmitted_oe_range_end;
8632 	}
8633 
8634 	ret = btrfs_submit_direct_hook(dip, skip_sum);
8635 	if (!ret)
8636 		return;
8637 
8638 	if (io_bio->end_io)
8639 		io_bio->end_io(io_bio, ret);
8640 
8641 free_ordered:
8642 	/*
8643 	 * If we arrived here it means either we failed to submit the dip
8644 	 * or we either failed to clone the dio_bio or failed to allocate the
8645 	 * dip. If we cloned the dio_bio and allocated the dip, we can just
8646 	 * call bio_endio against our io_bio so that we get proper resource
8647 	 * cleanup if we fail to submit the dip, otherwise, we must do the
8648 	 * same as btrfs_endio_direct_[write|read] because we can't call these
8649 	 * callbacks - they require an allocated dip and a clone of dio_bio.
8650 	 */
8651 	if (bio && dip) {
8652 		bio_io_error(bio);
8653 		/*
8654 		 * The end io callbacks free our dip, do the final put on bio
8655 		 * and all the cleanup and final put for dio_bio (through
8656 		 * dio_end_io()).
8657 		 */
8658 		dip = NULL;
8659 		bio = NULL;
8660 	} else {
8661 		if (write)
8662 			__endio_write_update_ordered(inode,
8663 						file_offset,
8664 						dio_bio->bi_iter.bi_size,
8665 						false);
8666 		else
8667 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8668 			      file_offset + dio_bio->bi_iter.bi_size - 1);
8669 
8670 		dio_bio->bi_status = BLK_STS_IOERR;
8671 		/*
8672 		 * Releases and cleans up our dio_bio, no need to bio_put()
8673 		 * nor bio_endio()/bio_io_error() against dio_bio.
8674 		 */
8675 		dio_end_io(dio_bio);
8676 	}
8677 	if (bio)
8678 		bio_put(bio);
8679 	kfree(dip);
8680 }
8681 
8682 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8683 			       struct kiocb *iocb,
8684 			       const struct iov_iter *iter, loff_t offset)
8685 {
8686 	int seg;
8687 	int i;
8688 	unsigned int blocksize_mask = fs_info->sectorsize - 1;
8689 	ssize_t retval = -EINVAL;
8690 
8691 	if (offset & blocksize_mask)
8692 		goto out;
8693 
8694 	if (iov_iter_alignment(iter) & blocksize_mask)
8695 		goto out;
8696 
8697 	/* If this is a write we don't need to check anymore */
8698 	if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8699 		return 0;
8700 	/*
8701 	 * Check to make sure we don't have duplicate iov_base's in this
8702 	 * iovec, if so return EINVAL, otherwise we'll get csum errors
8703 	 * when reading back.
8704 	 */
8705 	for (seg = 0; seg < iter->nr_segs; seg++) {
8706 		for (i = seg + 1; i < iter->nr_segs; i++) {
8707 			if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8708 				goto out;
8709 		}
8710 	}
8711 	retval = 0;
8712 out:
8713 	return retval;
8714 }
8715 
8716 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8717 {
8718 	struct file *file = iocb->ki_filp;
8719 	struct inode *inode = file->f_mapping->host;
8720 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8721 	struct btrfs_dio_data dio_data = { 0 };
8722 	struct extent_changeset *data_reserved = NULL;
8723 	loff_t offset = iocb->ki_pos;
8724 	size_t count = 0;
8725 	int flags = 0;
8726 	bool wakeup = true;
8727 	bool relock = false;
8728 	ssize_t ret;
8729 
8730 	if (check_direct_IO(fs_info, iocb, iter, offset))
8731 		return 0;
8732 
8733 	inode_dio_begin(inode);
8734 	smp_mb__after_atomic();
8735 
8736 	/*
8737 	 * The generic stuff only does filemap_write_and_wait_range, which
8738 	 * isn't enough if we've written compressed pages to this area, so
8739 	 * we need to flush the dirty pages again to make absolutely sure
8740 	 * that any outstanding dirty pages are on disk.
8741 	 */
8742 	count = iov_iter_count(iter);
8743 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8744 		     &BTRFS_I(inode)->runtime_flags))
8745 		filemap_fdatawrite_range(inode->i_mapping, offset,
8746 					 offset + count - 1);
8747 
8748 	if (iov_iter_rw(iter) == WRITE) {
8749 		/*
8750 		 * If the write DIO is beyond the EOF, we need update
8751 		 * the isize, but it is protected by i_mutex. So we can
8752 		 * not unlock the i_mutex at this case.
8753 		 */
8754 		if (offset + count <= inode->i_size) {
8755 			dio_data.overwrite = 1;
8756 			inode_unlock(inode);
8757 			relock = true;
8758 		} else if (iocb->ki_flags & IOCB_NOWAIT) {
8759 			ret = -EAGAIN;
8760 			goto out;
8761 		}
8762 		ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8763 						   offset, count);
8764 		if (ret)
8765 			goto out;
8766 		dio_data.outstanding_extents = count_max_extents(count);
8767 
8768 		/*
8769 		 * We need to know how many extents we reserved so that we can
8770 		 * do the accounting properly if we go over the number we
8771 		 * originally calculated.  Abuse current->journal_info for this.
8772 		 */
8773 		dio_data.reserve = round_up(count,
8774 					    fs_info->sectorsize);
8775 		dio_data.unsubmitted_oe_range_start = (u64)offset;
8776 		dio_data.unsubmitted_oe_range_end = (u64)offset;
8777 		current->journal_info = &dio_data;
8778 		down_read(&BTRFS_I(inode)->dio_sem);
8779 	} else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8780 				     &BTRFS_I(inode)->runtime_flags)) {
8781 		inode_dio_end(inode);
8782 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
8783 		wakeup = false;
8784 	}
8785 
8786 	ret = __blockdev_direct_IO(iocb, inode,
8787 				   fs_info->fs_devices->latest_bdev,
8788 				   iter, btrfs_get_blocks_direct, NULL,
8789 				   btrfs_submit_direct, flags);
8790 	if (iov_iter_rw(iter) == WRITE) {
8791 		up_read(&BTRFS_I(inode)->dio_sem);
8792 		current->journal_info = NULL;
8793 		if (ret < 0 && ret != -EIOCBQUEUED) {
8794 			if (dio_data.reserve)
8795 				btrfs_delalloc_release_space(inode, data_reserved,
8796 					offset, dio_data.reserve);
8797 			/*
8798 			 * On error we might have left some ordered extents
8799 			 * without submitting corresponding bios for them, so
8800 			 * cleanup them up to avoid other tasks getting them
8801 			 * and waiting for them to complete forever.
8802 			 */
8803 			if (dio_data.unsubmitted_oe_range_start <
8804 			    dio_data.unsubmitted_oe_range_end)
8805 				__endio_write_update_ordered(inode,
8806 					dio_data.unsubmitted_oe_range_start,
8807 					dio_data.unsubmitted_oe_range_end -
8808 					dio_data.unsubmitted_oe_range_start,
8809 					false);
8810 		} else if (ret >= 0 && (size_t)ret < count)
8811 			btrfs_delalloc_release_space(inode, data_reserved,
8812 					offset, count - (size_t)ret);
8813 	}
8814 out:
8815 	if (wakeup)
8816 		inode_dio_end(inode);
8817 	if (relock)
8818 		inode_lock(inode);
8819 
8820 	extent_changeset_free(data_reserved);
8821 	return ret;
8822 }
8823 
8824 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
8825 
8826 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8827 		__u64 start, __u64 len)
8828 {
8829 	int	ret;
8830 
8831 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8832 	if (ret)
8833 		return ret;
8834 
8835 	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8836 }
8837 
8838 int btrfs_readpage(struct file *file, struct page *page)
8839 {
8840 	struct extent_io_tree *tree;
8841 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8842 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8843 }
8844 
8845 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8846 {
8847 	struct extent_io_tree *tree;
8848 	struct inode *inode = page->mapping->host;
8849 	int ret;
8850 
8851 	if (current->flags & PF_MEMALLOC) {
8852 		redirty_page_for_writepage(wbc, page);
8853 		unlock_page(page);
8854 		return 0;
8855 	}
8856 
8857 	/*
8858 	 * If we are under memory pressure we will call this directly from the
8859 	 * VM, we need to make sure we have the inode referenced for the ordered
8860 	 * extent.  If not just return like we didn't do anything.
8861 	 */
8862 	if (!igrab(inode)) {
8863 		redirty_page_for_writepage(wbc, page);
8864 		return AOP_WRITEPAGE_ACTIVATE;
8865 	}
8866 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8867 	ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8868 	btrfs_add_delayed_iput(inode);
8869 	return ret;
8870 }
8871 
8872 static int btrfs_writepages(struct address_space *mapping,
8873 			    struct writeback_control *wbc)
8874 {
8875 	struct extent_io_tree *tree;
8876 
8877 	tree = &BTRFS_I(mapping->host)->io_tree;
8878 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8879 }
8880 
8881 static int
8882 btrfs_readpages(struct file *file, struct address_space *mapping,
8883 		struct list_head *pages, unsigned nr_pages)
8884 {
8885 	struct extent_io_tree *tree;
8886 	tree = &BTRFS_I(mapping->host)->io_tree;
8887 	return extent_readpages(tree, mapping, pages, nr_pages,
8888 				btrfs_get_extent);
8889 }
8890 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8891 {
8892 	struct extent_io_tree *tree;
8893 	struct extent_map_tree *map;
8894 	int ret;
8895 
8896 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8897 	map = &BTRFS_I(page->mapping->host)->extent_tree;
8898 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8899 	if (ret == 1) {
8900 		ClearPagePrivate(page);
8901 		set_page_private(page, 0);
8902 		put_page(page);
8903 	}
8904 	return ret;
8905 }
8906 
8907 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8908 {
8909 	if (PageWriteback(page) || PageDirty(page))
8910 		return 0;
8911 	return __btrfs_releasepage(page, gfp_flags);
8912 }
8913 
8914 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8915 				 unsigned int length)
8916 {
8917 	struct inode *inode = page->mapping->host;
8918 	struct extent_io_tree *tree;
8919 	struct btrfs_ordered_extent *ordered;
8920 	struct extent_state *cached_state = NULL;
8921 	u64 page_start = page_offset(page);
8922 	u64 page_end = page_start + PAGE_SIZE - 1;
8923 	u64 start;
8924 	u64 end;
8925 	int inode_evicting = inode->i_state & I_FREEING;
8926 
8927 	/*
8928 	 * we have the page locked, so new writeback can't start,
8929 	 * and the dirty bit won't be cleared while we are here.
8930 	 *
8931 	 * Wait for IO on this page so that we can safely clear
8932 	 * the PagePrivate2 bit and do ordered accounting
8933 	 */
8934 	wait_on_page_writeback(page);
8935 
8936 	tree = &BTRFS_I(inode)->io_tree;
8937 	if (offset) {
8938 		btrfs_releasepage(page, GFP_NOFS);
8939 		return;
8940 	}
8941 
8942 	if (!inode_evicting)
8943 		lock_extent_bits(tree, page_start, page_end, &cached_state);
8944 again:
8945 	start = page_start;
8946 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8947 					page_end - start + 1);
8948 	if (ordered) {
8949 		end = min(page_end, ordered->file_offset + ordered->len - 1);
8950 		/*
8951 		 * IO on this page will never be started, so we need
8952 		 * to account for any ordered extents now
8953 		 */
8954 		if (!inode_evicting)
8955 			clear_extent_bit(tree, start, end,
8956 					 EXTENT_DIRTY | EXTENT_DELALLOC |
8957 					 EXTENT_DELALLOC_NEW |
8958 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8959 					 EXTENT_DEFRAG, 1, 0, &cached_state,
8960 					 GFP_NOFS);
8961 		/*
8962 		 * whoever cleared the private bit is responsible
8963 		 * for the finish_ordered_io
8964 		 */
8965 		if (TestClearPagePrivate2(page)) {
8966 			struct btrfs_ordered_inode_tree *tree;
8967 			u64 new_len;
8968 
8969 			tree = &BTRFS_I(inode)->ordered_tree;
8970 
8971 			spin_lock_irq(&tree->lock);
8972 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8973 			new_len = start - ordered->file_offset;
8974 			if (new_len < ordered->truncated_len)
8975 				ordered->truncated_len = new_len;
8976 			spin_unlock_irq(&tree->lock);
8977 
8978 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
8979 							   start,
8980 							   end - start + 1, 1))
8981 				btrfs_finish_ordered_io(ordered);
8982 		}
8983 		btrfs_put_ordered_extent(ordered);
8984 		if (!inode_evicting) {
8985 			cached_state = NULL;
8986 			lock_extent_bits(tree, start, end,
8987 					 &cached_state);
8988 		}
8989 
8990 		start = end + 1;
8991 		if (start < page_end)
8992 			goto again;
8993 	}
8994 
8995 	/*
8996 	 * Qgroup reserved space handler
8997 	 * Page here will be either
8998 	 * 1) Already written to disk
8999 	 *    In this case, its reserved space is released from data rsv map
9000 	 *    and will be freed by delayed_ref handler finally.
9001 	 *    So even we call qgroup_free_data(), it won't decrease reserved
9002 	 *    space.
9003 	 * 2) Not written to disk
9004 	 *    This means the reserved space should be freed here. However,
9005 	 *    if a truncate invalidates the page (by clearing PageDirty)
9006 	 *    and the page is accounted for while allocating extent
9007 	 *    in btrfs_check_data_free_space() we let delayed_ref to
9008 	 *    free the entire extent.
9009 	 */
9010 	if (PageDirty(page))
9011 		btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9012 	if (!inode_evicting) {
9013 		clear_extent_bit(tree, page_start, page_end,
9014 				 EXTENT_LOCKED | EXTENT_DIRTY |
9015 				 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9016 				 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9017 				 &cached_state, GFP_NOFS);
9018 
9019 		__btrfs_releasepage(page, GFP_NOFS);
9020 	}
9021 
9022 	ClearPageChecked(page);
9023 	if (PagePrivate(page)) {
9024 		ClearPagePrivate(page);
9025 		set_page_private(page, 0);
9026 		put_page(page);
9027 	}
9028 }
9029 
9030 /*
9031  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9032  * called from a page fault handler when a page is first dirtied. Hence we must
9033  * be careful to check for EOF conditions here. We set the page up correctly
9034  * for a written page which means we get ENOSPC checking when writing into
9035  * holes and correct delalloc and unwritten extent mapping on filesystems that
9036  * support these features.
9037  *
9038  * We are not allowed to take the i_mutex here so we have to play games to
9039  * protect against truncate races as the page could now be beyond EOF.  Because
9040  * vmtruncate() writes the inode size before removing pages, once we have the
9041  * page lock we can determine safely if the page is beyond EOF. If it is not
9042  * beyond EOF, then the page is guaranteed safe against truncation until we
9043  * unlock the page.
9044  */
9045 int btrfs_page_mkwrite(struct vm_fault *vmf)
9046 {
9047 	struct page *page = vmf->page;
9048 	struct inode *inode = file_inode(vmf->vma->vm_file);
9049 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9050 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9051 	struct btrfs_ordered_extent *ordered;
9052 	struct extent_state *cached_state = NULL;
9053 	struct extent_changeset *data_reserved = NULL;
9054 	char *kaddr;
9055 	unsigned long zero_start;
9056 	loff_t size;
9057 	int ret;
9058 	int reserved = 0;
9059 	u64 reserved_space;
9060 	u64 page_start;
9061 	u64 page_end;
9062 	u64 end;
9063 
9064 	reserved_space = PAGE_SIZE;
9065 
9066 	sb_start_pagefault(inode->i_sb);
9067 	page_start = page_offset(page);
9068 	page_end = page_start + PAGE_SIZE - 1;
9069 	end = page_end;
9070 
9071 	/*
9072 	 * Reserving delalloc space after obtaining the page lock can lead to
9073 	 * deadlock. For example, if a dirty page is locked by this function
9074 	 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9075 	 * dirty page write out, then the btrfs_writepage() function could
9076 	 * end up waiting indefinitely to get a lock on the page currently
9077 	 * being processed by btrfs_page_mkwrite() function.
9078 	 */
9079 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9080 					   reserved_space);
9081 	if (!ret) {
9082 		ret = file_update_time(vmf->vma->vm_file);
9083 		reserved = 1;
9084 	}
9085 	if (ret) {
9086 		if (ret == -ENOMEM)
9087 			ret = VM_FAULT_OOM;
9088 		else /* -ENOSPC, -EIO, etc */
9089 			ret = VM_FAULT_SIGBUS;
9090 		if (reserved)
9091 			goto out;
9092 		goto out_noreserve;
9093 	}
9094 
9095 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9096 again:
9097 	lock_page(page);
9098 	size = i_size_read(inode);
9099 
9100 	if ((page->mapping != inode->i_mapping) ||
9101 	    (page_start >= size)) {
9102 		/* page got truncated out from underneath us */
9103 		goto out_unlock;
9104 	}
9105 	wait_on_page_writeback(page);
9106 
9107 	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9108 	set_page_extent_mapped(page);
9109 
9110 	/*
9111 	 * we can't set the delalloc bits if there are pending ordered
9112 	 * extents.  Drop our locks and wait for them to finish
9113 	 */
9114 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9115 			PAGE_SIZE);
9116 	if (ordered) {
9117 		unlock_extent_cached(io_tree, page_start, page_end,
9118 				     &cached_state, GFP_NOFS);
9119 		unlock_page(page);
9120 		btrfs_start_ordered_extent(inode, ordered, 1);
9121 		btrfs_put_ordered_extent(ordered);
9122 		goto again;
9123 	}
9124 
9125 	if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9126 		reserved_space = round_up(size - page_start,
9127 					  fs_info->sectorsize);
9128 		if (reserved_space < PAGE_SIZE) {
9129 			end = page_start + reserved_space - 1;
9130 			spin_lock(&BTRFS_I(inode)->lock);
9131 			BTRFS_I(inode)->outstanding_extents++;
9132 			spin_unlock(&BTRFS_I(inode)->lock);
9133 			btrfs_delalloc_release_space(inode, data_reserved,
9134 					page_start, PAGE_SIZE - reserved_space);
9135 		}
9136 	}
9137 
9138 	/*
9139 	 * page_mkwrite gets called when the page is firstly dirtied after it's
9140 	 * faulted in, but write(2) could also dirty a page and set delalloc
9141 	 * bits, thus in this case for space account reason, we still need to
9142 	 * clear any delalloc bits within this page range since we have to
9143 	 * reserve data&meta space before lock_page() (see above comments).
9144 	 */
9145 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9146 			  EXTENT_DIRTY | EXTENT_DELALLOC |
9147 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9148 			  0, 0, &cached_state, GFP_NOFS);
9149 
9150 	ret = btrfs_set_extent_delalloc(inode, page_start, end,
9151 					&cached_state, 0);
9152 	if (ret) {
9153 		unlock_extent_cached(io_tree, page_start, page_end,
9154 				     &cached_state, GFP_NOFS);
9155 		ret = VM_FAULT_SIGBUS;
9156 		goto out_unlock;
9157 	}
9158 	ret = 0;
9159 
9160 	/* page is wholly or partially inside EOF */
9161 	if (page_start + PAGE_SIZE > size)
9162 		zero_start = size & ~PAGE_MASK;
9163 	else
9164 		zero_start = PAGE_SIZE;
9165 
9166 	if (zero_start != PAGE_SIZE) {
9167 		kaddr = kmap(page);
9168 		memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9169 		flush_dcache_page(page);
9170 		kunmap(page);
9171 	}
9172 	ClearPageChecked(page);
9173 	set_page_dirty(page);
9174 	SetPageUptodate(page);
9175 
9176 	BTRFS_I(inode)->last_trans = fs_info->generation;
9177 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9178 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9179 
9180 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9181 
9182 out_unlock:
9183 	if (!ret) {
9184 		sb_end_pagefault(inode->i_sb);
9185 		extent_changeset_free(data_reserved);
9186 		return VM_FAULT_LOCKED;
9187 	}
9188 	unlock_page(page);
9189 out:
9190 	btrfs_delalloc_release_space(inode, data_reserved, page_start,
9191 				     reserved_space);
9192 out_noreserve:
9193 	sb_end_pagefault(inode->i_sb);
9194 	extent_changeset_free(data_reserved);
9195 	return ret;
9196 }
9197 
9198 static int btrfs_truncate(struct inode *inode)
9199 {
9200 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9201 	struct btrfs_root *root = BTRFS_I(inode)->root;
9202 	struct btrfs_block_rsv *rsv;
9203 	int ret = 0;
9204 	int err = 0;
9205 	struct btrfs_trans_handle *trans;
9206 	u64 mask = fs_info->sectorsize - 1;
9207 	u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9208 
9209 	ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9210 				       (u64)-1);
9211 	if (ret)
9212 		return ret;
9213 
9214 	/*
9215 	 * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9216 	 * 3 things going on here
9217 	 *
9218 	 * 1) We need to reserve space for our orphan item and the space to
9219 	 * delete our orphan item.  Lord knows we don't want to have a dangling
9220 	 * orphan item because we didn't reserve space to remove it.
9221 	 *
9222 	 * 2) We need to reserve space to update our inode.
9223 	 *
9224 	 * 3) We need to have something to cache all the space that is going to
9225 	 * be free'd up by the truncate operation, but also have some slack
9226 	 * space reserved in case it uses space during the truncate (thank you
9227 	 * very much snapshotting).
9228 	 *
9229 	 * And we need these to all be separate.  The fact is we can use a lot of
9230 	 * space doing the truncate, and we have no earthly idea how much space
9231 	 * we will use, so we need the truncate reservation to be separate so it
9232 	 * doesn't end up using space reserved for updating the inode or
9233 	 * removing the orphan item.  We also need to be able to stop the
9234 	 * transaction and start a new one, which means we need to be able to
9235 	 * update the inode several times, and we have no idea of knowing how
9236 	 * many times that will be, so we can't just reserve 1 item for the
9237 	 * entirety of the operation, so that has to be done separately as well.
9238 	 * Then there is the orphan item, which does indeed need to be held on
9239 	 * to for the whole operation, and we need nobody to touch this reserved
9240 	 * space except the orphan code.
9241 	 *
9242 	 * So that leaves us with
9243 	 *
9244 	 * 1) root->orphan_block_rsv - for the orphan deletion.
9245 	 * 2) rsv - for the truncate reservation, which we will steal from the
9246 	 * transaction reservation.
9247 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9248 	 * updating the inode.
9249 	 */
9250 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9251 	if (!rsv)
9252 		return -ENOMEM;
9253 	rsv->size = min_size;
9254 	rsv->failfast = 1;
9255 
9256 	/*
9257 	 * 1 for the truncate slack space
9258 	 * 1 for updating the inode.
9259 	 */
9260 	trans = btrfs_start_transaction(root, 2);
9261 	if (IS_ERR(trans)) {
9262 		err = PTR_ERR(trans);
9263 		goto out;
9264 	}
9265 
9266 	/* Migrate the slack space for the truncate to our reserve */
9267 	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9268 				      min_size, 0);
9269 	BUG_ON(ret);
9270 
9271 	/*
9272 	 * So if we truncate and then write and fsync we normally would just
9273 	 * write the extents that changed, which is a problem if we need to
9274 	 * first truncate that entire inode.  So set this flag so we write out
9275 	 * all of the extents in the inode to the sync log so we're completely
9276 	 * safe.
9277 	 */
9278 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9279 	trans->block_rsv = rsv;
9280 
9281 	while (1) {
9282 		ret = btrfs_truncate_inode_items(trans, root, inode,
9283 						 inode->i_size,
9284 						 BTRFS_EXTENT_DATA_KEY);
9285 		if (ret != -ENOSPC && ret != -EAGAIN) {
9286 			err = ret;
9287 			break;
9288 		}
9289 
9290 		trans->block_rsv = &fs_info->trans_block_rsv;
9291 		ret = btrfs_update_inode(trans, root, inode);
9292 		if (ret) {
9293 			err = ret;
9294 			break;
9295 		}
9296 
9297 		btrfs_end_transaction(trans);
9298 		btrfs_btree_balance_dirty(fs_info);
9299 
9300 		trans = btrfs_start_transaction(root, 2);
9301 		if (IS_ERR(trans)) {
9302 			ret = err = PTR_ERR(trans);
9303 			trans = NULL;
9304 			break;
9305 		}
9306 
9307 		btrfs_block_rsv_release(fs_info, rsv, -1);
9308 		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9309 					      rsv, min_size, 0);
9310 		BUG_ON(ret);	/* shouldn't happen */
9311 		trans->block_rsv = rsv;
9312 	}
9313 
9314 	if (ret == 0 && inode->i_nlink > 0) {
9315 		trans->block_rsv = root->orphan_block_rsv;
9316 		ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9317 		if (ret)
9318 			err = ret;
9319 	}
9320 
9321 	if (trans) {
9322 		trans->block_rsv = &fs_info->trans_block_rsv;
9323 		ret = btrfs_update_inode(trans, root, inode);
9324 		if (ret && !err)
9325 			err = ret;
9326 
9327 		ret = btrfs_end_transaction(trans);
9328 		btrfs_btree_balance_dirty(fs_info);
9329 	}
9330 out:
9331 	btrfs_free_block_rsv(fs_info, rsv);
9332 
9333 	if (ret && !err)
9334 		err = ret;
9335 
9336 	return err;
9337 }
9338 
9339 /*
9340  * create a new subvolume directory/inode (helper for the ioctl).
9341  */
9342 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9343 			     struct btrfs_root *new_root,
9344 			     struct btrfs_root *parent_root,
9345 			     u64 new_dirid)
9346 {
9347 	struct inode *inode;
9348 	int err;
9349 	u64 index = 0;
9350 
9351 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9352 				new_dirid, new_dirid,
9353 				S_IFDIR | (~current_umask() & S_IRWXUGO),
9354 				&index);
9355 	if (IS_ERR(inode))
9356 		return PTR_ERR(inode);
9357 	inode->i_op = &btrfs_dir_inode_operations;
9358 	inode->i_fop = &btrfs_dir_file_operations;
9359 
9360 	set_nlink(inode, 1);
9361 	btrfs_i_size_write(BTRFS_I(inode), 0);
9362 	unlock_new_inode(inode);
9363 
9364 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9365 	if (err)
9366 		btrfs_err(new_root->fs_info,
9367 			  "error inheriting subvolume %llu properties: %d",
9368 			  new_root->root_key.objectid, err);
9369 
9370 	err = btrfs_update_inode(trans, new_root, inode);
9371 
9372 	iput(inode);
9373 	return err;
9374 }
9375 
9376 struct inode *btrfs_alloc_inode(struct super_block *sb)
9377 {
9378 	struct btrfs_inode *ei;
9379 	struct inode *inode;
9380 
9381 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9382 	if (!ei)
9383 		return NULL;
9384 
9385 	ei->root = NULL;
9386 	ei->generation = 0;
9387 	ei->last_trans = 0;
9388 	ei->last_sub_trans = 0;
9389 	ei->logged_trans = 0;
9390 	ei->delalloc_bytes = 0;
9391 	ei->new_delalloc_bytes = 0;
9392 	ei->defrag_bytes = 0;
9393 	ei->disk_i_size = 0;
9394 	ei->flags = 0;
9395 	ei->csum_bytes = 0;
9396 	ei->index_cnt = (u64)-1;
9397 	ei->dir_index = 0;
9398 	ei->last_unlink_trans = 0;
9399 	ei->last_log_commit = 0;
9400 	ei->delayed_iput_count = 0;
9401 
9402 	spin_lock_init(&ei->lock);
9403 	ei->outstanding_extents = 0;
9404 	ei->reserved_extents = 0;
9405 
9406 	ei->runtime_flags = 0;
9407 	ei->force_compress = BTRFS_COMPRESS_NONE;
9408 
9409 	ei->delayed_node = NULL;
9410 
9411 	ei->i_otime.tv_sec = 0;
9412 	ei->i_otime.tv_nsec = 0;
9413 
9414 	inode = &ei->vfs_inode;
9415 	extent_map_tree_init(&ei->extent_tree);
9416 	extent_io_tree_init(&ei->io_tree, inode);
9417 	extent_io_tree_init(&ei->io_failure_tree, inode);
9418 	ei->io_tree.track_uptodate = 1;
9419 	ei->io_failure_tree.track_uptodate = 1;
9420 	atomic_set(&ei->sync_writers, 0);
9421 	mutex_init(&ei->log_mutex);
9422 	mutex_init(&ei->delalloc_mutex);
9423 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9424 	INIT_LIST_HEAD(&ei->delalloc_inodes);
9425 	INIT_LIST_HEAD(&ei->delayed_iput);
9426 	RB_CLEAR_NODE(&ei->rb_node);
9427 	init_rwsem(&ei->dio_sem);
9428 
9429 	return inode;
9430 }
9431 
9432 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9433 void btrfs_test_destroy_inode(struct inode *inode)
9434 {
9435 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9436 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9437 }
9438 #endif
9439 
9440 static void btrfs_i_callback(struct rcu_head *head)
9441 {
9442 	struct inode *inode = container_of(head, struct inode, i_rcu);
9443 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9444 }
9445 
9446 void btrfs_destroy_inode(struct inode *inode)
9447 {
9448 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9449 	struct btrfs_ordered_extent *ordered;
9450 	struct btrfs_root *root = BTRFS_I(inode)->root;
9451 
9452 	WARN_ON(!hlist_empty(&inode->i_dentry));
9453 	WARN_ON(inode->i_data.nrpages);
9454 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
9455 	WARN_ON(BTRFS_I(inode)->reserved_extents);
9456 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9457 	WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9458 	WARN_ON(BTRFS_I(inode)->csum_bytes);
9459 	WARN_ON(BTRFS_I(inode)->defrag_bytes);
9460 
9461 	/*
9462 	 * This can happen where we create an inode, but somebody else also
9463 	 * created the same inode and we need to destroy the one we already
9464 	 * created.
9465 	 */
9466 	if (!root)
9467 		goto free;
9468 
9469 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9470 		     &BTRFS_I(inode)->runtime_flags)) {
9471 		btrfs_info(fs_info, "inode %llu still on the orphan list",
9472 			   btrfs_ino(BTRFS_I(inode)));
9473 		atomic_dec(&root->orphan_inodes);
9474 	}
9475 
9476 	while (1) {
9477 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9478 		if (!ordered)
9479 			break;
9480 		else {
9481 			btrfs_err(fs_info,
9482 				  "found ordered extent %llu %llu on inode cleanup",
9483 				  ordered->file_offset, ordered->len);
9484 			btrfs_remove_ordered_extent(inode, ordered);
9485 			btrfs_put_ordered_extent(ordered);
9486 			btrfs_put_ordered_extent(ordered);
9487 		}
9488 	}
9489 	btrfs_qgroup_check_reserved_leak(inode);
9490 	inode_tree_del(inode);
9491 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9492 free:
9493 	call_rcu(&inode->i_rcu, btrfs_i_callback);
9494 }
9495 
9496 int btrfs_drop_inode(struct inode *inode)
9497 {
9498 	struct btrfs_root *root = BTRFS_I(inode)->root;
9499 
9500 	if (root == NULL)
9501 		return 1;
9502 
9503 	/* the snap/subvol tree is on deleting */
9504 	if (btrfs_root_refs(&root->root_item) == 0)
9505 		return 1;
9506 	else
9507 		return generic_drop_inode(inode);
9508 }
9509 
9510 static void init_once(void *foo)
9511 {
9512 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9513 
9514 	inode_init_once(&ei->vfs_inode);
9515 }
9516 
9517 void btrfs_destroy_cachep(void)
9518 {
9519 	/*
9520 	 * Make sure all delayed rcu free inodes are flushed before we
9521 	 * destroy cache.
9522 	 */
9523 	rcu_barrier();
9524 	kmem_cache_destroy(btrfs_inode_cachep);
9525 	kmem_cache_destroy(btrfs_trans_handle_cachep);
9526 	kmem_cache_destroy(btrfs_path_cachep);
9527 	kmem_cache_destroy(btrfs_free_space_cachep);
9528 }
9529 
9530 int btrfs_init_cachep(void)
9531 {
9532 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9533 			sizeof(struct btrfs_inode), 0,
9534 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9535 			init_once);
9536 	if (!btrfs_inode_cachep)
9537 		goto fail;
9538 
9539 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9540 			sizeof(struct btrfs_trans_handle), 0,
9541 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9542 	if (!btrfs_trans_handle_cachep)
9543 		goto fail;
9544 
9545 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
9546 			sizeof(struct btrfs_path), 0,
9547 			SLAB_MEM_SPREAD, NULL);
9548 	if (!btrfs_path_cachep)
9549 		goto fail;
9550 
9551 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9552 			sizeof(struct btrfs_free_space), 0,
9553 			SLAB_MEM_SPREAD, NULL);
9554 	if (!btrfs_free_space_cachep)
9555 		goto fail;
9556 
9557 	return 0;
9558 fail:
9559 	btrfs_destroy_cachep();
9560 	return -ENOMEM;
9561 }
9562 
9563 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9564 			 u32 request_mask, unsigned int flags)
9565 {
9566 	u64 delalloc_bytes;
9567 	struct inode *inode = d_inode(path->dentry);
9568 	u32 blocksize = inode->i_sb->s_blocksize;
9569 	u32 bi_flags = BTRFS_I(inode)->flags;
9570 
9571 	stat->result_mask |= STATX_BTIME;
9572 	stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9573 	stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9574 	if (bi_flags & BTRFS_INODE_APPEND)
9575 		stat->attributes |= STATX_ATTR_APPEND;
9576 	if (bi_flags & BTRFS_INODE_COMPRESS)
9577 		stat->attributes |= STATX_ATTR_COMPRESSED;
9578 	if (bi_flags & BTRFS_INODE_IMMUTABLE)
9579 		stat->attributes |= STATX_ATTR_IMMUTABLE;
9580 	if (bi_flags & BTRFS_INODE_NODUMP)
9581 		stat->attributes |= STATX_ATTR_NODUMP;
9582 
9583 	stat->attributes_mask |= (STATX_ATTR_APPEND |
9584 				  STATX_ATTR_COMPRESSED |
9585 				  STATX_ATTR_IMMUTABLE |
9586 				  STATX_ATTR_NODUMP);
9587 
9588 	generic_fillattr(inode, stat);
9589 	stat->dev = BTRFS_I(inode)->root->anon_dev;
9590 
9591 	spin_lock(&BTRFS_I(inode)->lock);
9592 	delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9593 	spin_unlock(&BTRFS_I(inode)->lock);
9594 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9595 			ALIGN(delalloc_bytes, blocksize)) >> 9;
9596 	return 0;
9597 }
9598 
9599 static int btrfs_rename_exchange(struct inode *old_dir,
9600 			      struct dentry *old_dentry,
9601 			      struct inode *new_dir,
9602 			      struct dentry *new_dentry)
9603 {
9604 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9605 	struct btrfs_trans_handle *trans;
9606 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9607 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9608 	struct inode *new_inode = new_dentry->d_inode;
9609 	struct inode *old_inode = old_dentry->d_inode;
9610 	struct timespec ctime = current_time(old_inode);
9611 	struct dentry *parent;
9612 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9613 	u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9614 	u64 old_idx = 0;
9615 	u64 new_idx = 0;
9616 	u64 root_objectid;
9617 	int ret;
9618 	bool root_log_pinned = false;
9619 	bool dest_log_pinned = false;
9620 
9621 	/* we only allow rename subvolume link between subvolumes */
9622 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9623 		return -EXDEV;
9624 
9625 	/* close the race window with snapshot create/destroy ioctl */
9626 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9627 		down_read(&fs_info->subvol_sem);
9628 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9629 		down_read(&fs_info->subvol_sem);
9630 
9631 	/*
9632 	 * We want to reserve the absolute worst case amount of items.  So if
9633 	 * both inodes are subvols and we need to unlink them then that would
9634 	 * require 4 item modifications, but if they are both normal inodes it
9635 	 * would require 5 item modifications, so we'll assume their normal
9636 	 * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9637 	 * should cover the worst case number of items we'll modify.
9638 	 */
9639 	trans = btrfs_start_transaction(root, 12);
9640 	if (IS_ERR(trans)) {
9641 		ret = PTR_ERR(trans);
9642 		goto out_notrans;
9643 	}
9644 
9645 	/*
9646 	 * We need to find a free sequence number both in the source and
9647 	 * in the destination directory for the exchange.
9648 	 */
9649 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9650 	if (ret)
9651 		goto out_fail;
9652 	ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9653 	if (ret)
9654 		goto out_fail;
9655 
9656 	BTRFS_I(old_inode)->dir_index = 0ULL;
9657 	BTRFS_I(new_inode)->dir_index = 0ULL;
9658 
9659 	/* Reference for the source. */
9660 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9661 		/* force full log commit if subvolume involved. */
9662 		btrfs_set_log_full_commit(fs_info, trans);
9663 	} else {
9664 		btrfs_pin_log_trans(root);
9665 		root_log_pinned = true;
9666 		ret = btrfs_insert_inode_ref(trans, dest,
9667 					     new_dentry->d_name.name,
9668 					     new_dentry->d_name.len,
9669 					     old_ino,
9670 					     btrfs_ino(BTRFS_I(new_dir)),
9671 					     old_idx);
9672 		if (ret)
9673 			goto out_fail;
9674 	}
9675 
9676 	/* And now for the dest. */
9677 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9678 		/* force full log commit if subvolume involved. */
9679 		btrfs_set_log_full_commit(fs_info, trans);
9680 	} else {
9681 		btrfs_pin_log_trans(dest);
9682 		dest_log_pinned = true;
9683 		ret = btrfs_insert_inode_ref(trans, root,
9684 					     old_dentry->d_name.name,
9685 					     old_dentry->d_name.len,
9686 					     new_ino,
9687 					     btrfs_ino(BTRFS_I(old_dir)),
9688 					     new_idx);
9689 		if (ret)
9690 			goto out_fail;
9691 	}
9692 
9693 	/* Update inode version and ctime/mtime. */
9694 	inode_inc_iversion(old_dir);
9695 	inode_inc_iversion(new_dir);
9696 	inode_inc_iversion(old_inode);
9697 	inode_inc_iversion(new_inode);
9698 	old_dir->i_ctime = old_dir->i_mtime = ctime;
9699 	new_dir->i_ctime = new_dir->i_mtime = ctime;
9700 	old_inode->i_ctime = ctime;
9701 	new_inode->i_ctime = ctime;
9702 
9703 	if (old_dentry->d_parent != new_dentry->d_parent) {
9704 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9705 				BTRFS_I(old_inode), 1);
9706 		btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9707 				BTRFS_I(new_inode), 1);
9708 	}
9709 
9710 	/* src is a subvolume */
9711 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9712 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9713 		ret = btrfs_unlink_subvol(trans, root, old_dir,
9714 					  root_objectid,
9715 					  old_dentry->d_name.name,
9716 					  old_dentry->d_name.len);
9717 	} else { /* src is an inode */
9718 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9719 					   BTRFS_I(old_dentry->d_inode),
9720 					   old_dentry->d_name.name,
9721 					   old_dentry->d_name.len);
9722 		if (!ret)
9723 			ret = btrfs_update_inode(trans, root, old_inode);
9724 	}
9725 	if (ret) {
9726 		btrfs_abort_transaction(trans, ret);
9727 		goto out_fail;
9728 	}
9729 
9730 	/* dest is a subvolume */
9731 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9732 		root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9733 		ret = btrfs_unlink_subvol(trans, dest, new_dir,
9734 					  root_objectid,
9735 					  new_dentry->d_name.name,
9736 					  new_dentry->d_name.len);
9737 	} else { /* dest is an inode */
9738 		ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9739 					   BTRFS_I(new_dentry->d_inode),
9740 					   new_dentry->d_name.name,
9741 					   new_dentry->d_name.len);
9742 		if (!ret)
9743 			ret = btrfs_update_inode(trans, dest, new_inode);
9744 	}
9745 	if (ret) {
9746 		btrfs_abort_transaction(trans, ret);
9747 		goto out_fail;
9748 	}
9749 
9750 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9751 			     new_dentry->d_name.name,
9752 			     new_dentry->d_name.len, 0, old_idx);
9753 	if (ret) {
9754 		btrfs_abort_transaction(trans, ret);
9755 		goto out_fail;
9756 	}
9757 
9758 	ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9759 			     old_dentry->d_name.name,
9760 			     old_dentry->d_name.len, 0, new_idx);
9761 	if (ret) {
9762 		btrfs_abort_transaction(trans, ret);
9763 		goto out_fail;
9764 	}
9765 
9766 	if (old_inode->i_nlink == 1)
9767 		BTRFS_I(old_inode)->dir_index = old_idx;
9768 	if (new_inode->i_nlink == 1)
9769 		BTRFS_I(new_inode)->dir_index = new_idx;
9770 
9771 	if (root_log_pinned) {
9772 		parent = new_dentry->d_parent;
9773 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9774 				parent);
9775 		btrfs_end_log_trans(root);
9776 		root_log_pinned = false;
9777 	}
9778 	if (dest_log_pinned) {
9779 		parent = old_dentry->d_parent;
9780 		btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9781 				parent);
9782 		btrfs_end_log_trans(dest);
9783 		dest_log_pinned = false;
9784 	}
9785 out_fail:
9786 	/*
9787 	 * If we have pinned a log and an error happened, we unpin tasks
9788 	 * trying to sync the log and force them to fallback to a transaction
9789 	 * commit if the log currently contains any of the inodes involved in
9790 	 * this rename operation (to ensure we do not persist a log with an
9791 	 * inconsistent state for any of these inodes or leading to any
9792 	 * inconsistencies when replayed). If the transaction was aborted, the
9793 	 * abortion reason is propagated to userspace when attempting to commit
9794 	 * the transaction. If the log does not contain any of these inodes, we
9795 	 * allow the tasks to sync it.
9796 	 */
9797 	if (ret && (root_log_pinned || dest_log_pinned)) {
9798 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9799 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9800 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9801 		    (new_inode &&
9802 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9803 			btrfs_set_log_full_commit(fs_info, trans);
9804 
9805 		if (root_log_pinned) {
9806 			btrfs_end_log_trans(root);
9807 			root_log_pinned = false;
9808 		}
9809 		if (dest_log_pinned) {
9810 			btrfs_end_log_trans(dest);
9811 			dest_log_pinned = false;
9812 		}
9813 	}
9814 	ret = btrfs_end_transaction(trans);
9815 out_notrans:
9816 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9817 		up_read(&fs_info->subvol_sem);
9818 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9819 		up_read(&fs_info->subvol_sem);
9820 
9821 	return ret;
9822 }
9823 
9824 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9825 				     struct btrfs_root *root,
9826 				     struct inode *dir,
9827 				     struct dentry *dentry)
9828 {
9829 	int ret;
9830 	struct inode *inode;
9831 	u64 objectid;
9832 	u64 index;
9833 
9834 	ret = btrfs_find_free_ino(root, &objectid);
9835 	if (ret)
9836 		return ret;
9837 
9838 	inode = btrfs_new_inode(trans, root, dir,
9839 				dentry->d_name.name,
9840 				dentry->d_name.len,
9841 				btrfs_ino(BTRFS_I(dir)),
9842 				objectid,
9843 				S_IFCHR | WHITEOUT_MODE,
9844 				&index);
9845 
9846 	if (IS_ERR(inode)) {
9847 		ret = PTR_ERR(inode);
9848 		return ret;
9849 	}
9850 
9851 	inode->i_op = &btrfs_special_inode_operations;
9852 	init_special_inode(inode, inode->i_mode,
9853 		WHITEOUT_DEV);
9854 
9855 	ret = btrfs_init_inode_security(trans, inode, dir,
9856 				&dentry->d_name);
9857 	if (ret)
9858 		goto out;
9859 
9860 	ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9861 				BTRFS_I(inode), 0, index);
9862 	if (ret)
9863 		goto out;
9864 
9865 	ret = btrfs_update_inode(trans, root, inode);
9866 out:
9867 	unlock_new_inode(inode);
9868 	if (ret)
9869 		inode_dec_link_count(inode);
9870 	iput(inode);
9871 
9872 	return ret;
9873 }
9874 
9875 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9876 			   struct inode *new_dir, struct dentry *new_dentry,
9877 			   unsigned int flags)
9878 {
9879 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9880 	struct btrfs_trans_handle *trans;
9881 	unsigned int trans_num_items;
9882 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9883 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9884 	struct inode *new_inode = d_inode(new_dentry);
9885 	struct inode *old_inode = d_inode(old_dentry);
9886 	u64 index = 0;
9887 	u64 root_objectid;
9888 	int ret;
9889 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9890 	bool log_pinned = false;
9891 
9892 	if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9893 		return -EPERM;
9894 
9895 	/* we only allow rename subvolume link between subvolumes */
9896 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9897 		return -EXDEV;
9898 
9899 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9900 	    (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9901 		return -ENOTEMPTY;
9902 
9903 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
9904 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9905 		return -ENOTEMPTY;
9906 
9907 
9908 	/* check for collisions, even if the  name isn't there */
9909 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9910 			     new_dentry->d_name.name,
9911 			     new_dentry->d_name.len);
9912 
9913 	if (ret) {
9914 		if (ret == -EEXIST) {
9915 			/* we shouldn't get
9916 			 * eexist without a new_inode */
9917 			if (WARN_ON(!new_inode)) {
9918 				return ret;
9919 			}
9920 		} else {
9921 			/* maybe -EOVERFLOW */
9922 			return ret;
9923 		}
9924 	}
9925 	ret = 0;
9926 
9927 	/*
9928 	 * we're using rename to replace one file with another.  Start IO on it
9929 	 * now so  we don't add too much work to the end of the transaction
9930 	 */
9931 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9932 		filemap_flush(old_inode->i_mapping);
9933 
9934 	/* close the racy window with snapshot create/destroy ioctl */
9935 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9936 		down_read(&fs_info->subvol_sem);
9937 	/*
9938 	 * We want to reserve the absolute worst case amount of items.  So if
9939 	 * both inodes are subvols and we need to unlink them then that would
9940 	 * require 4 item modifications, but if they are both normal inodes it
9941 	 * would require 5 item modifications, so we'll assume they are normal
9942 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9943 	 * should cover the worst case number of items we'll modify.
9944 	 * If our rename has the whiteout flag, we need more 5 units for the
9945 	 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9946 	 * when selinux is enabled).
9947 	 */
9948 	trans_num_items = 11;
9949 	if (flags & RENAME_WHITEOUT)
9950 		trans_num_items += 5;
9951 	trans = btrfs_start_transaction(root, trans_num_items);
9952 	if (IS_ERR(trans)) {
9953 		ret = PTR_ERR(trans);
9954 		goto out_notrans;
9955 	}
9956 
9957 	if (dest != root)
9958 		btrfs_record_root_in_trans(trans, dest);
9959 
9960 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9961 	if (ret)
9962 		goto out_fail;
9963 
9964 	BTRFS_I(old_inode)->dir_index = 0ULL;
9965 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9966 		/* force full log commit if subvolume involved. */
9967 		btrfs_set_log_full_commit(fs_info, trans);
9968 	} else {
9969 		btrfs_pin_log_trans(root);
9970 		log_pinned = true;
9971 		ret = btrfs_insert_inode_ref(trans, dest,
9972 					     new_dentry->d_name.name,
9973 					     new_dentry->d_name.len,
9974 					     old_ino,
9975 					     btrfs_ino(BTRFS_I(new_dir)), index);
9976 		if (ret)
9977 			goto out_fail;
9978 	}
9979 
9980 	inode_inc_iversion(old_dir);
9981 	inode_inc_iversion(new_dir);
9982 	inode_inc_iversion(old_inode);
9983 	old_dir->i_ctime = old_dir->i_mtime =
9984 	new_dir->i_ctime = new_dir->i_mtime =
9985 	old_inode->i_ctime = current_time(old_dir);
9986 
9987 	if (old_dentry->d_parent != new_dentry->d_parent)
9988 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9989 				BTRFS_I(old_inode), 1);
9990 
9991 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9992 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9993 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9994 					old_dentry->d_name.name,
9995 					old_dentry->d_name.len);
9996 	} else {
9997 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9998 					BTRFS_I(d_inode(old_dentry)),
9999 					old_dentry->d_name.name,
10000 					old_dentry->d_name.len);
10001 		if (!ret)
10002 			ret = btrfs_update_inode(trans, root, old_inode);
10003 	}
10004 	if (ret) {
10005 		btrfs_abort_transaction(trans, ret);
10006 		goto out_fail;
10007 	}
10008 
10009 	if (new_inode) {
10010 		inode_inc_iversion(new_inode);
10011 		new_inode->i_ctime = current_time(new_inode);
10012 		if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10013 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10014 			root_objectid = BTRFS_I(new_inode)->location.objectid;
10015 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
10016 						root_objectid,
10017 						new_dentry->d_name.name,
10018 						new_dentry->d_name.len);
10019 			BUG_ON(new_inode->i_nlink == 0);
10020 		} else {
10021 			ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10022 						 BTRFS_I(d_inode(new_dentry)),
10023 						 new_dentry->d_name.name,
10024 						 new_dentry->d_name.len);
10025 		}
10026 		if (!ret && new_inode->i_nlink == 0)
10027 			ret = btrfs_orphan_add(trans,
10028 					BTRFS_I(d_inode(new_dentry)));
10029 		if (ret) {
10030 			btrfs_abort_transaction(trans, ret);
10031 			goto out_fail;
10032 		}
10033 	}
10034 
10035 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10036 			     new_dentry->d_name.name,
10037 			     new_dentry->d_name.len, 0, index);
10038 	if (ret) {
10039 		btrfs_abort_transaction(trans, ret);
10040 		goto out_fail;
10041 	}
10042 
10043 	if (old_inode->i_nlink == 1)
10044 		BTRFS_I(old_inode)->dir_index = index;
10045 
10046 	if (log_pinned) {
10047 		struct dentry *parent = new_dentry->d_parent;
10048 
10049 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10050 				parent);
10051 		btrfs_end_log_trans(root);
10052 		log_pinned = false;
10053 	}
10054 
10055 	if (flags & RENAME_WHITEOUT) {
10056 		ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10057 						old_dentry);
10058 
10059 		if (ret) {
10060 			btrfs_abort_transaction(trans, ret);
10061 			goto out_fail;
10062 		}
10063 	}
10064 out_fail:
10065 	/*
10066 	 * If we have pinned the log and an error happened, we unpin tasks
10067 	 * trying to sync the log and force them to fallback to a transaction
10068 	 * commit if the log currently contains any of the inodes involved in
10069 	 * this rename operation (to ensure we do not persist a log with an
10070 	 * inconsistent state for any of these inodes or leading to any
10071 	 * inconsistencies when replayed). If the transaction was aborted, the
10072 	 * abortion reason is propagated to userspace when attempting to commit
10073 	 * the transaction. If the log does not contain any of these inodes, we
10074 	 * allow the tasks to sync it.
10075 	 */
10076 	if (ret && log_pinned) {
10077 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10078 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10079 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10080 		    (new_inode &&
10081 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10082 			btrfs_set_log_full_commit(fs_info, trans);
10083 
10084 		btrfs_end_log_trans(root);
10085 		log_pinned = false;
10086 	}
10087 	btrfs_end_transaction(trans);
10088 out_notrans:
10089 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10090 		up_read(&fs_info->subvol_sem);
10091 
10092 	return ret;
10093 }
10094 
10095 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10096 			 struct inode *new_dir, struct dentry *new_dentry,
10097 			 unsigned int flags)
10098 {
10099 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10100 		return -EINVAL;
10101 
10102 	if (flags & RENAME_EXCHANGE)
10103 		return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10104 					  new_dentry);
10105 
10106 	return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10107 }
10108 
10109 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10110 {
10111 	struct btrfs_delalloc_work *delalloc_work;
10112 	struct inode *inode;
10113 
10114 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
10115 				     work);
10116 	inode = delalloc_work->inode;
10117 	filemap_flush(inode->i_mapping);
10118 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10119 				&BTRFS_I(inode)->runtime_flags))
10120 		filemap_flush(inode->i_mapping);
10121 
10122 	if (delalloc_work->delay_iput)
10123 		btrfs_add_delayed_iput(inode);
10124 	else
10125 		iput(inode);
10126 	complete(&delalloc_work->completion);
10127 }
10128 
10129 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10130 						    int delay_iput)
10131 {
10132 	struct btrfs_delalloc_work *work;
10133 
10134 	work = kmalloc(sizeof(*work), GFP_NOFS);
10135 	if (!work)
10136 		return NULL;
10137 
10138 	init_completion(&work->completion);
10139 	INIT_LIST_HEAD(&work->list);
10140 	work->inode = inode;
10141 	work->delay_iput = delay_iput;
10142 	WARN_ON_ONCE(!inode);
10143 	btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10144 			btrfs_run_delalloc_work, NULL, NULL);
10145 
10146 	return work;
10147 }
10148 
10149 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10150 {
10151 	wait_for_completion(&work->completion);
10152 	kfree(work);
10153 }
10154 
10155 /*
10156  * some fairly slow code that needs optimization. This walks the list
10157  * of all the inodes with pending delalloc and forces them to disk.
10158  */
10159 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10160 				   int nr)
10161 {
10162 	struct btrfs_inode *binode;
10163 	struct inode *inode;
10164 	struct btrfs_delalloc_work *work, *next;
10165 	struct list_head works;
10166 	struct list_head splice;
10167 	int ret = 0;
10168 
10169 	INIT_LIST_HEAD(&works);
10170 	INIT_LIST_HEAD(&splice);
10171 
10172 	mutex_lock(&root->delalloc_mutex);
10173 	spin_lock(&root->delalloc_lock);
10174 	list_splice_init(&root->delalloc_inodes, &splice);
10175 	while (!list_empty(&splice)) {
10176 		binode = list_entry(splice.next, struct btrfs_inode,
10177 				    delalloc_inodes);
10178 
10179 		list_move_tail(&binode->delalloc_inodes,
10180 			       &root->delalloc_inodes);
10181 		inode = igrab(&binode->vfs_inode);
10182 		if (!inode) {
10183 			cond_resched_lock(&root->delalloc_lock);
10184 			continue;
10185 		}
10186 		spin_unlock(&root->delalloc_lock);
10187 
10188 		work = btrfs_alloc_delalloc_work(inode, delay_iput);
10189 		if (!work) {
10190 			if (delay_iput)
10191 				btrfs_add_delayed_iput(inode);
10192 			else
10193 				iput(inode);
10194 			ret = -ENOMEM;
10195 			goto out;
10196 		}
10197 		list_add_tail(&work->list, &works);
10198 		btrfs_queue_work(root->fs_info->flush_workers,
10199 				 &work->work);
10200 		ret++;
10201 		if (nr != -1 && ret >= nr)
10202 			goto out;
10203 		cond_resched();
10204 		spin_lock(&root->delalloc_lock);
10205 	}
10206 	spin_unlock(&root->delalloc_lock);
10207 
10208 out:
10209 	list_for_each_entry_safe(work, next, &works, list) {
10210 		list_del_init(&work->list);
10211 		btrfs_wait_and_free_delalloc_work(work);
10212 	}
10213 
10214 	if (!list_empty_careful(&splice)) {
10215 		spin_lock(&root->delalloc_lock);
10216 		list_splice_tail(&splice, &root->delalloc_inodes);
10217 		spin_unlock(&root->delalloc_lock);
10218 	}
10219 	mutex_unlock(&root->delalloc_mutex);
10220 	return ret;
10221 }
10222 
10223 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10224 {
10225 	struct btrfs_fs_info *fs_info = root->fs_info;
10226 	int ret;
10227 
10228 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10229 		return -EROFS;
10230 
10231 	ret = __start_delalloc_inodes(root, delay_iput, -1);
10232 	if (ret > 0)
10233 		ret = 0;
10234 	/*
10235 	 * the filemap_flush will queue IO into the worker threads, but
10236 	 * we have to make sure the IO is actually started and that
10237 	 * ordered extents get created before we return
10238 	 */
10239 	atomic_inc(&fs_info->async_submit_draining);
10240 	while (atomic_read(&fs_info->nr_async_submits) ||
10241 	       atomic_read(&fs_info->async_delalloc_pages)) {
10242 		wait_event(fs_info->async_submit_wait,
10243 			   (atomic_read(&fs_info->nr_async_submits) == 0 &&
10244 			    atomic_read(&fs_info->async_delalloc_pages) == 0));
10245 	}
10246 	atomic_dec(&fs_info->async_submit_draining);
10247 	return ret;
10248 }
10249 
10250 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10251 			       int nr)
10252 {
10253 	struct btrfs_root *root;
10254 	struct list_head splice;
10255 	int ret;
10256 
10257 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10258 		return -EROFS;
10259 
10260 	INIT_LIST_HEAD(&splice);
10261 
10262 	mutex_lock(&fs_info->delalloc_root_mutex);
10263 	spin_lock(&fs_info->delalloc_root_lock);
10264 	list_splice_init(&fs_info->delalloc_roots, &splice);
10265 	while (!list_empty(&splice) && nr) {
10266 		root = list_first_entry(&splice, struct btrfs_root,
10267 					delalloc_root);
10268 		root = btrfs_grab_fs_root(root);
10269 		BUG_ON(!root);
10270 		list_move_tail(&root->delalloc_root,
10271 			       &fs_info->delalloc_roots);
10272 		spin_unlock(&fs_info->delalloc_root_lock);
10273 
10274 		ret = __start_delalloc_inodes(root, delay_iput, nr);
10275 		btrfs_put_fs_root(root);
10276 		if (ret < 0)
10277 			goto out;
10278 
10279 		if (nr != -1) {
10280 			nr -= ret;
10281 			WARN_ON(nr < 0);
10282 		}
10283 		spin_lock(&fs_info->delalloc_root_lock);
10284 	}
10285 	spin_unlock(&fs_info->delalloc_root_lock);
10286 
10287 	ret = 0;
10288 	atomic_inc(&fs_info->async_submit_draining);
10289 	while (atomic_read(&fs_info->nr_async_submits) ||
10290 	      atomic_read(&fs_info->async_delalloc_pages)) {
10291 		wait_event(fs_info->async_submit_wait,
10292 		   (atomic_read(&fs_info->nr_async_submits) == 0 &&
10293 		    atomic_read(&fs_info->async_delalloc_pages) == 0));
10294 	}
10295 	atomic_dec(&fs_info->async_submit_draining);
10296 out:
10297 	if (!list_empty_careful(&splice)) {
10298 		spin_lock(&fs_info->delalloc_root_lock);
10299 		list_splice_tail(&splice, &fs_info->delalloc_roots);
10300 		spin_unlock(&fs_info->delalloc_root_lock);
10301 	}
10302 	mutex_unlock(&fs_info->delalloc_root_mutex);
10303 	return ret;
10304 }
10305 
10306 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10307 			 const char *symname)
10308 {
10309 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10310 	struct btrfs_trans_handle *trans;
10311 	struct btrfs_root *root = BTRFS_I(dir)->root;
10312 	struct btrfs_path *path;
10313 	struct btrfs_key key;
10314 	struct inode *inode = NULL;
10315 	int err;
10316 	int drop_inode = 0;
10317 	u64 objectid;
10318 	u64 index = 0;
10319 	int name_len;
10320 	int datasize;
10321 	unsigned long ptr;
10322 	struct btrfs_file_extent_item *ei;
10323 	struct extent_buffer *leaf;
10324 
10325 	name_len = strlen(symname);
10326 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10327 		return -ENAMETOOLONG;
10328 
10329 	/*
10330 	 * 2 items for inode item and ref
10331 	 * 2 items for dir items
10332 	 * 1 item for updating parent inode item
10333 	 * 1 item for the inline extent item
10334 	 * 1 item for xattr if selinux is on
10335 	 */
10336 	trans = btrfs_start_transaction(root, 7);
10337 	if (IS_ERR(trans))
10338 		return PTR_ERR(trans);
10339 
10340 	err = btrfs_find_free_ino(root, &objectid);
10341 	if (err)
10342 		goto out_unlock;
10343 
10344 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10345 				dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10346 				objectid, S_IFLNK|S_IRWXUGO, &index);
10347 	if (IS_ERR(inode)) {
10348 		err = PTR_ERR(inode);
10349 		goto out_unlock;
10350 	}
10351 
10352 	/*
10353 	* If the active LSM wants to access the inode during
10354 	* d_instantiate it needs these. Smack checks to see
10355 	* if the filesystem supports xattrs by looking at the
10356 	* ops vector.
10357 	*/
10358 	inode->i_fop = &btrfs_file_operations;
10359 	inode->i_op = &btrfs_file_inode_operations;
10360 	inode->i_mapping->a_ops = &btrfs_aops;
10361 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10362 
10363 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10364 	if (err)
10365 		goto out_unlock_inode;
10366 
10367 	path = btrfs_alloc_path();
10368 	if (!path) {
10369 		err = -ENOMEM;
10370 		goto out_unlock_inode;
10371 	}
10372 	key.objectid = btrfs_ino(BTRFS_I(inode));
10373 	key.offset = 0;
10374 	key.type = BTRFS_EXTENT_DATA_KEY;
10375 	datasize = btrfs_file_extent_calc_inline_size(name_len);
10376 	err = btrfs_insert_empty_item(trans, root, path, &key,
10377 				      datasize);
10378 	if (err) {
10379 		btrfs_free_path(path);
10380 		goto out_unlock_inode;
10381 	}
10382 	leaf = path->nodes[0];
10383 	ei = btrfs_item_ptr(leaf, path->slots[0],
10384 			    struct btrfs_file_extent_item);
10385 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10386 	btrfs_set_file_extent_type(leaf, ei,
10387 				   BTRFS_FILE_EXTENT_INLINE);
10388 	btrfs_set_file_extent_encryption(leaf, ei, 0);
10389 	btrfs_set_file_extent_compression(leaf, ei, 0);
10390 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10391 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10392 
10393 	ptr = btrfs_file_extent_inline_start(ei);
10394 	write_extent_buffer(leaf, symname, ptr, name_len);
10395 	btrfs_mark_buffer_dirty(leaf);
10396 	btrfs_free_path(path);
10397 
10398 	inode->i_op = &btrfs_symlink_inode_operations;
10399 	inode_nohighmem(inode);
10400 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
10401 	inode_set_bytes(inode, name_len);
10402 	btrfs_i_size_write(BTRFS_I(inode), name_len);
10403 	err = btrfs_update_inode(trans, root, inode);
10404 	/*
10405 	 * Last step, add directory indexes for our symlink inode. This is the
10406 	 * last step to avoid extra cleanup of these indexes if an error happens
10407 	 * elsewhere above.
10408 	 */
10409 	if (!err)
10410 		err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10411 				BTRFS_I(inode), 0, index);
10412 	if (err) {
10413 		drop_inode = 1;
10414 		goto out_unlock_inode;
10415 	}
10416 
10417 	unlock_new_inode(inode);
10418 	d_instantiate(dentry, inode);
10419 
10420 out_unlock:
10421 	btrfs_end_transaction(trans);
10422 	if (drop_inode) {
10423 		inode_dec_link_count(inode);
10424 		iput(inode);
10425 	}
10426 	btrfs_btree_balance_dirty(fs_info);
10427 	return err;
10428 
10429 out_unlock_inode:
10430 	drop_inode = 1;
10431 	unlock_new_inode(inode);
10432 	goto out_unlock;
10433 }
10434 
10435 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10436 				       u64 start, u64 num_bytes, u64 min_size,
10437 				       loff_t actual_len, u64 *alloc_hint,
10438 				       struct btrfs_trans_handle *trans)
10439 {
10440 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10441 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10442 	struct extent_map *em;
10443 	struct btrfs_root *root = BTRFS_I(inode)->root;
10444 	struct btrfs_key ins;
10445 	u64 cur_offset = start;
10446 	u64 i_size;
10447 	u64 cur_bytes;
10448 	u64 last_alloc = (u64)-1;
10449 	int ret = 0;
10450 	bool own_trans = true;
10451 	u64 end = start + num_bytes - 1;
10452 
10453 	if (trans)
10454 		own_trans = false;
10455 	while (num_bytes > 0) {
10456 		if (own_trans) {
10457 			trans = btrfs_start_transaction(root, 3);
10458 			if (IS_ERR(trans)) {
10459 				ret = PTR_ERR(trans);
10460 				break;
10461 			}
10462 		}
10463 
10464 		cur_bytes = min_t(u64, num_bytes, SZ_256M);
10465 		cur_bytes = max(cur_bytes, min_size);
10466 		/*
10467 		 * If we are severely fragmented we could end up with really
10468 		 * small allocations, so if the allocator is returning small
10469 		 * chunks lets make its job easier by only searching for those
10470 		 * sized chunks.
10471 		 */
10472 		cur_bytes = min(cur_bytes, last_alloc);
10473 		ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10474 				min_size, 0, *alloc_hint, &ins, 1, 0);
10475 		if (ret) {
10476 			if (own_trans)
10477 				btrfs_end_transaction(trans);
10478 			break;
10479 		}
10480 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10481 
10482 		last_alloc = ins.offset;
10483 		ret = insert_reserved_file_extent(trans, inode,
10484 						  cur_offset, ins.objectid,
10485 						  ins.offset, ins.offset,
10486 						  ins.offset, 0, 0, 0,
10487 						  BTRFS_FILE_EXTENT_PREALLOC);
10488 		if (ret) {
10489 			btrfs_free_reserved_extent(fs_info, ins.objectid,
10490 						   ins.offset, 0);
10491 			btrfs_abort_transaction(trans, ret);
10492 			if (own_trans)
10493 				btrfs_end_transaction(trans);
10494 			break;
10495 		}
10496 
10497 		btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10498 					cur_offset + ins.offset -1, 0);
10499 
10500 		em = alloc_extent_map();
10501 		if (!em) {
10502 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10503 				&BTRFS_I(inode)->runtime_flags);
10504 			goto next;
10505 		}
10506 
10507 		em->start = cur_offset;
10508 		em->orig_start = cur_offset;
10509 		em->len = ins.offset;
10510 		em->block_start = ins.objectid;
10511 		em->block_len = ins.offset;
10512 		em->orig_block_len = ins.offset;
10513 		em->ram_bytes = ins.offset;
10514 		em->bdev = fs_info->fs_devices->latest_bdev;
10515 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10516 		em->generation = trans->transid;
10517 
10518 		while (1) {
10519 			write_lock(&em_tree->lock);
10520 			ret = add_extent_mapping(em_tree, em, 1);
10521 			write_unlock(&em_tree->lock);
10522 			if (ret != -EEXIST)
10523 				break;
10524 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10525 						cur_offset + ins.offset - 1,
10526 						0);
10527 		}
10528 		free_extent_map(em);
10529 next:
10530 		num_bytes -= ins.offset;
10531 		cur_offset += ins.offset;
10532 		*alloc_hint = ins.objectid + ins.offset;
10533 
10534 		inode_inc_iversion(inode);
10535 		inode->i_ctime = current_time(inode);
10536 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10537 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10538 		    (actual_len > inode->i_size) &&
10539 		    (cur_offset > inode->i_size)) {
10540 			if (cur_offset > actual_len)
10541 				i_size = actual_len;
10542 			else
10543 				i_size = cur_offset;
10544 			i_size_write(inode, i_size);
10545 			btrfs_ordered_update_i_size(inode, i_size, NULL);
10546 		}
10547 
10548 		ret = btrfs_update_inode(trans, root, inode);
10549 
10550 		if (ret) {
10551 			btrfs_abort_transaction(trans, ret);
10552 			if (own_trans)
10553 				btrfs_end_transaction(trans);
10554 			break;
10555 		}
10556 
10557 		if (own_trans)
10558 			btrfs_end_transaction(trans);
10559 	}
10560 	if (cur_offset < end)
10561 		btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10562 			end - cur_offset + 1);
10563 	return ret;
10564 }
10565 
10566 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10567 			      u64 start, u64 num_bytes, u64 min_size,
10568 			      loff_t actual_len, u64 *alloc_hint)
10569 {
10570 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10571 					   min_size, actual_len, alloc_hint,
10572 					   NULL);
10573 }
10574 
10575 int btrfs_prealloc_file_range_trans(struct inode *inode,
10576 				    struct btrfs_trans_handle *trans, int mode,
10577 				    u64 start, u64 num_bytes, u64 min_size,
10578 				    loff_t actual_len, u64 *alloc_hint)
10579 {
10580 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10581 					   min_size, actual_len, alloc_hint, trans);
10582 }
10583 
10584 static int btrfs_set_page_dirty(struct page *page)
10585 {
10586 	return __set_page_dirty_nobuffers(page);
10587 }
10588 
10589 static int btrfs_permission(struct inode *inode, int mask)
10590 {
10591 	struct btrfs_root *root = BTRFS_I(inode)->root;
10592 	umode_t mode = inode->i_mode;
10593 
10594 	if (mask & MAY_WRITE &&
10595 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10596 		if (btrfs_root_readonly(root))
10597 			return -EROFS;
10598 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10599 			return -EACCES;
10600 	}
10601 	return generic_permission(inode, mask);
10602 }
10603 
10604 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10605 {
10606 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10607 	struct btrfs_trans_handle *trans;
10608 	struct btrfs_root *root = BTRFS_I(dir)->root;
10609 	struct inode *inode = NULL;
10610 	u64 objectid;
10611 	u64 index;
10612 	int ret = 0;
10613 
10614 	/*
10615 	 * 5 units required for adding orphan entry
10616 	 */
10617 	trans = btrfs_start_transaction(root, 5);
10618 	if (IS_ERR(trans))
10619 		return PTR_ERR(trans);
10620 
10621 	ret = btrfs_find_free_ino(root, &objectid);
10622 	if (ret)
10623 		goto out;
10624 
10625 	inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10626 			btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10627 	if (IS_ERR(inode)) {
10628 		ret = PTR_ERR(inode);
10629 		inode = NULL;
10630 		goto out;
10631 	}
10632 
10633 	inode->i_fop = &btrfs_file_operations;
10634 	inode->i_op = &btrfs_file_inode_operations;
10635 
10636 	inode->i_mapping->a_ops = &btrfs_aops;
10637 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10638 
10639 	ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10640 	if (ret)
10641 		goto out_inode;
10642 
10643 	ret = btrfs_update_inode(trans, root, inode);
10644 	if (ret)
10645 		goto out_inode;
10646 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10647 	if (ret)
10648 		goto out_inode;
10649 
10650 	/*
10651 	 * We set number of links to 0 in btrfs_new_inode(), and here we set
10652 	 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10653 	 * through:
10654 	 *
10655 	 *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10656 	 */
10657 	set_nlink(inode, 1);
10658 	unlock_new_inode(inode);
10659 	d_tmpfile(dentry, inode);
10660 	mark_inode_dirty(inode);
10661 
10662 out:
10663 	btrfs_end_transaction(trans);
10664 	if (ret)
10665 		iput(inode);
10666 	btrfs_balance_delayed_items(fs_info);
10667 	btrfs_btree_balance_dirty(fs_info);
10668 	return ret;
10669 
10670 out_inode:
10671 	unlock_new_inode(inode);
10672 	goto out;
10673 
10674 }
10675 
10676 __attribute__((const))
10677 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10678 {
10679 	return -EAGAIN;
10680 }
10681 
10682 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10683 {
10684 	struct inode *inode = private_data;
10685 	return btrfs_sb(inode->i_sb);
10686 }
10687 
10688 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10689 					u64 start, u64 end)
10690 {
10691 	struct inode *inode = private_data;
10692 	u64 isize;
10693 
10694 	isize = i_size_read(inode);
10695 	if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10696 		btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10697 		    "%s: ino %llu isize %llu odd range [%llu,%llu]",
10698 			caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10699 	}
10700 }
10701 
10702 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10703 {
10704 	struct inode *inode = private_data;
10705 	unsigned long index = start >> PAGE_SHIFT;
10706 	unsigned long end_index = end >> PAGE_SHIFT;
10707 	struct page *page;
10708 
10709 	while (index <= end_index) {
10710 		page = find_get_page(inode->i_mapping, index);
10711 		ASSERT(page); /* Pages should be in the extent_io_tree */
10712 		set_page_writeback(page);
10713 		put_page(page);
10714 		index++;
10715 	}
10716 }
10717 
10718 static const struct inode_operations btrfs_dir_inode_operations = {
10719 	.getattr	= btrfs_getattr,
10720 	.lookup		= btrfs_lookup,
10721 	.create		= btrfs_create,
10722 	.unlink		= btrfs_unlink,
10723 	.link		= btrfs_link,
10724 	.mkdir		= btrfs_mkdir,
10725 	.rmdir		= btrfs_rmdir,
10726 	.rename		= btrfs_rename2,
10727 	.symlink	= btrfs_symlink,
10728 	.setattr	= btrfs_setattr,
10729 	.mknod		= btrfs_mknod,
10730 	.listxattr	= btrfs_listxattr,
10731 	.permission	= btrfs_permission,
10732 	.get_acl	= btrfs_get_acl,
10733 	.set_acl	= btrfs_set_acl,
10734 	.update_time	= btrfs_update_time,
10735 	.tmpfile        = btrfs_tmpfile,
10736 };
10737 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10738 	.lookup		= btrfs_lookup,
10739 	.permission	= btrfs_permission,
10740 	.update_time	= btrfs_update_time,
10741 };
10742 
10743 static const struct file_operations btrfs_dir_file_operations = {
10744 	.llseek		= generic_file_llseek,
10745 	.read		= generic_read_dir,
10746 	.iterate_shared	= btrfs_real_readdir,
10747 	.unlocked_ioctl	= btrfs_ioctl,
10748 #ifdef CONFIG_COMPAT
10749 	.compat_ioctl	= btrfs_compat_ioctl,
10750 #endif
10751 	.release        = btrfs_release_file,
10752 	.fsync		= btrfs_sync_file,
10753 };
10754 
10755 static const struct extent_io_ops btrfs_extent_io_ops = {
10756 	/* mandatory callbacks */
10757 	.submit_bio_hook = btrfs_submit_bio_hook,
10758 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
10759 	.merge_bio_hook = btrfs_merge_bio_hook,
10760 	.readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10761 	.tree_fs_info = iotree_fs_info,
10762 	.set_range_writeback = btrfs_set_range_writeback,
10763 
10764 	/* optional callbacks */
10765 	.fill_delalloc = run_delalloc_range,
10766 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
10767 	.writepage_start_hook = btrfs_writepage_start_hook,
10768 	.set_bit_hook = btrfs_set_bit_hook,
10769 	.clear_bit_hook = btrfs_clear_bit_hook,
10770 	.merge_extent_hook = btrfs_merge_extent_hook,
10771 	.split_extent_hook = btrfs_split_extent_hook,
10772 	.check_extent_io_range = btrfs_check_extent_io_range,
10773 };
10774 
10775 /*
10776  * btrfs doesn't support the bmap operation because swapfiles
10777  * use bmap to make a mapping of extents in the file.  They assume
10778  * these extents won't change over the life of the file and they
10779  * use the bmap result to do IO directly to the drive.
10780  *
10781  * the btrfs bmap call would return logical addresses that aren't
10782  * suitable for IO and they also will change frequently as COW
10783  * operations happen.  So, swapfile + btrfs == corruption.
10784  *
10785  * For now we're avoiding this by dropping bmap.
10786  */
10787 static const struct address_space_operations btrfs_aops = {
10788 	.readpage	= btrfs_readpage,
10789 	.writepage	= btrfs_writepage,
10790 	.writepages	= btrfs_writepages,
10791 	.readpages	= btrfs_readpages,
10792 	.direct_IO	= btrfs_direct_IO,
10793 	.invalidatepage = btrfs_invalidatepage,
10794 	.releasepage	= btrfs_releasepage,
10795 	.set_page_dirty	= btrfs_set_page_dirty,
10796 	.error_remove_page = generic_error_remove_page,
10797 };
10798 
10799 static const struct address_space_operations btrfs_symlink_aops = {
10800 	.readpage	= btrfs_readpage,
10801 	.writepage	= btrfs_writepage,
10802 	.invalidatepage = btrfs_invalidatepage,
10803 	.releasepage	= btrfs_releasepage,
10804 };
10805 
10806 static const struct inode_operations btrfs_file_inode_operations = {
10807 	.getattr	= btrfs_getattr,
10808 	.setattr	= btrfs_setattr,
10809 	.listxattr      = btrfs_listxattr,
10810 	.permission	= btrfs_permission,
10811 	.fiemap		= btrfs_fiemap,
10812 	.get_acl	= btrfs_get_acl,
10813 	.set_acl	= btrfs_set_acl,
10814 	.update_time	= btrfs_update_time,
10815 };
10816 static const struct inode_operations btrfs_special_inode_operations = {
10817 	.getattr	= btrfs_getattr,
10818 	.setattr	= btrfs_setattr,
10819 	.permission	= btrfs_permission,
10820 	.listxattr	= btrfs_listxattr,
10821 	.get_acl	= btrfs_get_acl,
10822 	.set_acl	= btrfs_set_acl,
10823 	.update_time	= btrfs_update_time,
10824 };
10825 static const struct inode_operations btrfs_symlink_inode_operations = {
10826 	.get_link	= page_get_link,
10827 	.getattr	= btrfs_getattr,
10828 	.setattr	= btrfs_setattr,
10829 	.permission	= btrfs_permission,
10830 	.listxattr	= btrfs_listxattr,
10831 	.update_time	= btrfs_update_time,
10832 };
10833 
10834 const struct dentry_operations btrfs_dentry_operations = {
10835 	.d_delete	= btrfs_dentry_delete,
10836 	.d_release	= btrfs_dentry_release,
10837 };
10838