xref: /openbmc/linux/fs/btrfs/inode.c (revision b240b419db5d624ce7a5a397d6f62a1a686009ec)
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 <linux/magic.h>
46 #include <linux/iversion.h>
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
53 #include "xattr.h"
54 #include "tree-log.h"
55 #include "volumes.h"
56 #include "compression.h"
57 #include "locking.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
60 #include "backref.h"
61 #include "props.h"
62 #include "qgroup.h"
63 #include "dedupe.h"
64 
65 struct btrfs_iget_args {
66 	struct btrfs_key *location;
67 	struct btrfs_root *root;
68 };
69 
70 struct btrfs_dio_data {
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, bool skip_writeback);
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 	unsigned long index = offset >> PAGE_SHIFT;
139 	unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
140 	struct page *page;
141 
142 	while (index <= end_index) {
143 		page = find_get_page(inode->i_mapping, index);
144 		index++;
145 		if (!page)
146 			continue;
147 		ClearPagePrivate2(page);
148 		put_page(page);
149 	}
150 	return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
151 					    bytes - PAGE_SIZE, false);
152 }
153 
154 static int btrfs_dirty_inode(struct inode *inode);
155 
156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157 void btrfs_test_inode_set_ops(struct inode *inode)
158 {
159 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
160 }
161 #endif
162 
163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
164 				     struct inode *inode,  struct inode *dir,
165 				     const struct qstr *qstr)
166 {
167 	int err;
168 
169 	err = btrfs_init_acl(trans, inode, dir);
170 	if (!err)
171 		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
172 	return err;
173 }
174 
175 /*
176  * this does all the hard work for inserting an inline extent into
177  * the btree.  The caller should have done a btrfs_drop_extents so that
178  * no overlapping inline items exist in the btree
179  */
180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
181 				struct btrfs_path *path, int extent_inserted,
182 				struct btrfs_root *root, struct inode *inode,
183 				u64 start, size_t size, size_t compressed_size,
184 				int compress_type,
185 				struct page **compressed_pages)
186 {
187 	struct extent_buffer *leaf;
188 	struct page *page = NULL;
189 	char *kaddr;
190 	unsigned long ptr;
191 	struct btrfs_file_extent_item *ei;
192 	int ret;
193 	size_t cur_size = size;
194 	unsigned long offset;
195 
196 	if (compressed_size && compressed_pages)
197 		cur_size = compressed_size;
198 
199 	inode_add_bytes(inode, size);
200 
201 	if (!extent_inserted) {
202 		struct btrfs_key key;
203 		size_t datasize;
204 
205 		key.objectid = btrfs_ino(BTRFS_I(inode));
206 		key.offset = start;
207 		key.type = BTRFS_EXTENT_DATA_KEY;
208 
209 		datasize = btrfs_file_extent_calc_inline_size(cur_size);
210 		path->leave_spinning = 1;
211 		ret = btrfs_insert_empty_item(trans, root, path, &key,
212 					      datasize);
213 		if (ret)
214 			goto fail;
215 	}
216 	leaf = path->nodes[0];
217 	ei = btrfs_item_ptr(leaf, path->slots[0],
218 			    struct btrfs_file_extent_item);
219 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
220 	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
221 	btrfs_set_file_extent_encryption(leaf, ei, 0);
222 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
223 	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
224 	ptr = btrfs_file_extent_inline_start(ei);
225 
226 	if (compress_type != BTRFS_COMPRESS_NONE) {
227 		struct page *cpage;
228 		int i = 0;
229 		while (compressed_size > 0) {
230 			cpage = compressed_pages[i];
231 			cur_size = min_t(unsigned long, compressed_size,
232 				       PAGE_SIZE);
233 
234 			kaddr = kmap_atomic(cpage);
235 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
236 			kunmap_atomic(kaddr);
237 
238 			i++;
239 			ptr += cur_size;
240 			compressed_size -= cur_size;
241 		}
242 		btrfs_set_file_extent_compression(leaf, ei,
243 						  compress_type);
244 	} else {
245 		page = find_get_page(inode->i_mapping,
246 				     start >> PAGE_SHIFT);
247 		btrfs_set_file_extent_compression(leaf, ei, 0);
248 		kaddr = kmap_atomic(page);
249 		offset = start & (PAGE_SIZE - 1);
250 		write_extent_buffer(leaf, kaddr + offset, ptr, size);
251 		kunmap_atomic(kaddr);
252 		put_page(page);
253 	}
254 	btrfs_mark_buffer_dirty(leaf);
255 	btrfs_release_path(path);
256 
257 	/*
258 	 * we're an inline extent, so nobody can
259 	 * extend the file past i_size without locking
260 	 * a page we already have locked.
261 	 *
262 	 * We must do any isize and inode updates
263 	 * before we unlock the pages.  Otherwise we
264 	 * could end up racing with unlink.
265 	 */
266 	BTRFS_I(inode)->disk_i_size = inode->i_size;
267 	ret = btrfs_update_inode(trans, root, inode);
268 
269 fail:
270 	return ret;
271 }
272 
273 
274 /*
275  * conditionally insert an inline extent into the file.  This
276  * does the checks required to make sure the data is small enough
277  * to fit as an inline extent.
278  */
279 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
280 					  u64 end, size_t compressed_size,
281 					  int compress_type,
282 					  struct page **compressed_pages)
283 {
284 	struct btrfs_root *root = BTRFS_I(inode)->root;
285 	struct btrfs_fs_info *fs_info = root->fs_info;
286 	struct btrfs_trans_handle *trans;
287 	u64 isize = i_size_read(inode);
288 	u64 actual_end = min(end + 1, isize);
289 	u64 inline_len = actual_end - start;
290 	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
291 	u64 data_len = inline_len;
292 	int ret;
293 	struct btrfs_path *path;
294 	int extent_inserted = 0;
295 	u32 extent_item_size;
296 
297 	if (compressed_size)
298 		data_len = compressed_size;
299 
300 	if (start > 0 ||
301 	    actual_end > fs_info->sectorsize ||
302 	    data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
303 	    (!compressed_size &&
304 	    (actual_end & (fs_info->sectorsize - 1)) == 0) ||
305 	    end + 1 < isize ||
306 	    data_len > fs_info->max_inline) {
307 		return 1;
308 	}
309 
310 	path = btrfs_alloc_path();
311 	if (!path)
312 		return -ENOMEM;
313 
314 	trans = btrfs_join_transaction(root);
315 	if (IS_ERR(trans)) {
316 		btrfs_free_path(path);
317 		return PTR_ERR(trans);
318 	}
319 	trans->block_rsv = &BTRFS_I(inode)->block_rsv;
320 
321 	if (compressed_size && compressed_pages)
322 		extent_item_size = btrfs_file_extent_calc_inline_size(
323 		   compressed_size);
324 	else
325 		extent_item_size = btrfs_file_extent_calc_inline_size(
326 		    inline_len);
327 
328 	ret = __btrfs_drop_extents(trans, root, inode, path,
329 				   start, aligned_end, NULL,
330 				   1, 1, extent_item_size, &extent_inserted);
331 	if (ret) {
332 		btrfs_abort_transaction(trans, ret);
333 		goto out;
334 	}
335 
336 	if (isize > actual_end)
337 		inline_len = min_t(u64, isize, actual_end);
338 	ret = insert_inline_extent(trans, path, extent_inserted,
339 				   root, inode, start,
340 				   inline_len, compressed_size,
341 				   compress_type, compressed_pages);
342 	if (ret && ret != -ENOSPC) {
343 		btrfs_abort_transaction(trans, ret);
344 		goto out;
345 	} else if (ret == -ENOSPC) {
346 		ret = 1;
347 		goto out;
348 	}
349 
350 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
351 	btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
352 out:
353 	/*
354 	 * Don't forget to free the reserved space, as for inlined extent
355 	 * it won't count as data extent, free them directly here.
356 	 * And at reserve time, it's always aligned to page size, so
357 	 * just free one page here.
358 	 */
359 	btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
360 	btrfs_free_path(path);
361 	btrfs_end_transaction(trans);
362 	return ret;
363 }
364 
365 struct async_extent {
366 	u64 start;
367 	u64 ram_size;
368 	u64 compressed_size;
369 	struct page **pages;
370 	unsigned long nr_pages;
371 	int compress_type;
372 	struct list_head list;
373 };
374 
375 struct async_cow {
376 	struct inode *inode;
377 	struct btrfs_root *root;
378 	struct page *locked_page;
379 	u64 start;
380 	u64 end;
381 	unsigned int write_flags;
382 	struct list_head extents;
383 	struct btrfs_work work;
384 };
385 
386 static noinline int add_async_extent(struct async_cow *cow,
387 				     u64 start, u64 ram_size,
388 				     u64 compressed_size,
389 				     struct page **pages,
390 				     unsigned long nr_pages,
391 				     int compress_type)
392 {
393 	struct async_extent *async_extent;
394 
395 	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
396 	BUG_ON(!async_extent); /* -ENOMEM */
397 	async_extent->start = start;
398 	async_extent->ram_size = ram_size;
399 	async_extent->compressed_size = compressed_size;
400 	async_extent->pages = pages;
401 	async_extent->nr_pages = nr_pages;
402 	async_extent->compress_type = compress_type;
403 	list_add_tail(&async_extent->list, &cow->extents);
404 	return 0;
405 }
406 
407 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
408 {
409 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
410 
411 	/* force compress */
412 	if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
413 		return 1;
414 	/* defrag ioctl */
415 	if (BTRFS_I(inode)->defrag_compress)
416 		return 1;
417 	/* bad compression ratios */
418 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
419 		return 0;
420 	if (btrfs_test_opt(fs_info, COMPRESS) ||
421 	    BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
422 	    BTRFS_I(inode)->prop_compress)
423 		return btrfs_compress_heuristic(inode, start, end);
424 	return 0;
425 }
426 
427 static inline void inode_should_defrag(struct btrfs_inode *inode,
428 		u64 start, u64 end, u64 num_bytes, u64 small_write)
429 {
430 	/* If this is a small write inside eof, kick off a defrag */
431 	if (num_bytes < small_write &&
432 	    (start > 0 || end + 1 < inode->disk_i_size))
433 		btrfs_add_inode_defrag(NULL, inode);
434 }
435 
436 /*
437  * we create compressed extents in two phases.  The first
438  * phase compresses a range of pages that have already been
439  * locked (both pages and state bits are locked).
440  *
441  * This is done inside an ordered work queue, and the compression
442  * is spread across many cpus.  The actual IO submission is step
443  * two, and the ordered work queue takes care of making sure that
444  * happens in the same order things were put onto the queue by
445  * writepages and friends.
446  *
447  * If this code finds it can't get good compression, it puts an
448  * entry onto the work queue to write the uncompressed bytes.  This
449  * makes sure that both compressed inodes and uncompressed inodes
450  * are written in the same order that the flusher thread sent them
451  * down.
452  */
453 static noinline void compress_file_range(struct inode *inode,
454 					struct page *locked_page,
455 					u64 start, u64 end,
456 					struct async_cow *async_cow,
457 					int *num_added)
458 {
459 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
460 	u64 blocksize = fs_info->sectorsize;
461 	u64 actual_end;
462 	u64 isize = i_size_read(inode);
463 	int ret = 0;
464 	struct page **pages = NULL;
465 	unsigned long nr_pages;
466 	unsigned long total_compressed = 0;
467 	unsigned long total_in = 0;
468 	int i;
469 	int will_compress;
470 	int compress_type = fs_info->compress_type;
471 	int redirty = 0;
472 
473 	inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
474 			SZ_16K);
475 
476 	actual_end = min_t(u64, isize, end + 1);
477 again:
478 	will_compress = 0;
479 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
480 	BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
481 	nr_pages = min_t(unsigned long, nr_pages,
482 			BTRFS_MAX_COMPRESSED / PAGE_SIZE);
483 
484 	/*
485 	 * we don't want to send crud past the end of i_size through
486 	 * compression, that's just a waste of CPU time.  So, if the
487 	 * end of the file is before the start of our current
488 	 * requested range of bytes, we bail out to the uncompressed
489 	 * cleanup code that can deal with all of this.
490 	 *
491 	 * It isn't really the fastest way to fix things, but this is a
492 	 * very uncommon corner.
493 	 */
494 	if (actual_end <= start)
495 		goto cleanup_and_bail_uncompressed;
496 
497 	total_compressed = actual_end - start;
498 
499 	/*
500 	 * skip compression for a small file range(<=blocksize) that
501 	 * isn't an inline extent, since it doesn't save disk space at all.
502 	 */
503 	if (total_compressed <= blocksize &&
504 	   (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
505 		goto cleanup_and_bail_uncompressed;
506 
507 	total_compressed = min_t(unsigned long, total_compressed,
508 			BTRFS_MAX_UNCOMPRESSED);
509 	total_in = 0;
510 	ret = 0;
511 
512 	/*
513 	 * we do compression for mount -o compress and when the
514 	 * inode has not been flagged as nocompress.  This flag can
515 	 * change at any time if we discover bad compression ratios.
516 	 */
517 	if (inode_need_compress(inode, start, end)) {
518 		WARN_ON(pages);
519 		pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
520 		if (!pages) {
521 			/* just bail out to the uncompressed code */
522 			goto cont;
523 		}
524 
525 		if (BTRFS_I(inode)->defrag_compress)
526 			compress_type = BTRFS_I(inode)->defrag_compress;
527 		else if (BTRFS_I(inode)->prop_compress)
528 			compress_type = BTRFS_I(inode)->prop_compress;
529 
530 		/*
531 		 * we need to call clear_page_dirty_for_io on each
532 		 * page in the range.  Otherwise applications with the file
533 		 * mmap'd can wander in and change the page contents while
534 		 * we are compressing them.
535 		 *
536 		 * If the compression fails for any reason, we set the pages
537 		 * dirty again later on.
538 		 *
539 		 * Note that the remaining part is redirtied, the start pointer
540 		 * has moved, the end is the original one.
541 		 */
542 		if (!redirty) {
543 			extent_range_clear_dirty_for_io(inode, start, end);
544 			redirty = 1;
545 		}
546 
547 		/* Compression level is applied here and only here */
548 		ret = btrfs_compress_pages(
549 			compress_type | (fs_info->compress_level << 4),
550 					   inode->i_mapping, start,
551 					   pages,
552 					   &nr_pages,
553 					   &total_in,
554 					   &total_compressed);
555 
556 		if (!ret) {
557 			unsigned long offset = total_compressed &
558 				(PAGE_SIZE - 1);
559 			struct page *page = pages[nr_pages - 1];
560 			char *kaddr;
561 
562 			/* zero the tail end of the last page, we might be
563 			 * sending it down to disk
564 			 */
565 			if (offset) {
566 				kaddr = kmap_atomic(page);
567 				memset(kaddr + offset, 0,
568 				       PAGE_SIZE - offset);
569 				kunmap_atomic(kaddr);
570 			}
571 			will_compress = 1;
572 		}
573 	}
574 cont:
575 	if (start == 0) {
576 		/* lets try to make an inline extent */
577 		if (ret || total_in < actual_end) {
578 			/* we didn't compress the entire range, try
579 			 * to make an uncompressed inline extent.
580 			 */
581 			ret = cow_file_range_inline(inode, start, end, 0,
582 						    BTRFS_COMPRESS_NONE, NULL);
583 		} else {
584 			/* try making a compressed inline extent */
585 			ret = cow_file_range_inline(inode, start, end,
586 						    total_compressed,
587 						    compress_type, pages);
588 		}
589 		if (ret <= 0) {
590 			unsigned long clear_flags = EXTENT_DELALLOC |
591 				EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
592 				EXTENT_DO_ACCOUNTING;
593 			unsigned long page_error_op;
594 
595 			page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
596 
597 			/*
598 			 * inline extent creation worked or returned error,
599 			 * we don't need to create any more async work items.
600 			 * Unlock and free up our temp pages.
601 			 *
602 			 * We use DO_ACCOUNTING here because we need the
603 			 * delalloc_release_metadata to be done _after_ we drop
604 			 * our outstanding extent for clearing delalloc for this
605 			 * range.
606 			 */
607 			extent_clear_unlock_delalloc(inode, start, end, end,
608 						     NULL, clear_flags,
609 						     PAGE_UNLOCK |
610 						     PAGE_CLEAR_DIRTY |
611 						     PAGE_SET_WRITEBACK |
612 						     page_error_op |
613 						     PAGE_END_WRITEBACK);
614 			goto free_pages_out;
615 		}
616 	}
617 
618 	if (will_compress) {
619 		/*
620 		 * we aren't doing an inline extent round the compressed size
621 		 * up to a block size boundary so the allocator does sane
622 		 * things
623 		 */
624 		total_compressed = ALIGN(total_compressed, blocksize);
625 
626 		/*
627 		 * one last check to make sure the compression is really a
628 		 * win, compare the page count read with the blocks on disk,
629 		 * compression must free at least one sector size
630 		 */
631 		total_in = ALIGN(total_in, PAGE_SIZE);
632 		if (total_compressed + blocksize <= total_in) {
633 			*num_added += 1;
634 
635 			/*
636 			 * The async work queues will take care of doing actual
637 			 * allocation on disk for these compressed pages, and
638 			 * will submit them to the elevator.
639 			 */
640 			add_async_extent(async_cow, start, total_in,
641 					total_compressed, pages, nr_pages,
642 					compress_type);
643 
644 			if (start + total_in < end) {
645 				start += total_in;
646 				pages = NULL;
647 				cond_resched();
648 				goto again;
649 			}
650 			return;
651 		}
652 	}
653 	if (pages) {
654 		/*
655 		 * the compression code ran but failed to make things smaller,
656 		 * free any pages it allocated and our page pointer array
657 		 */
658 		for (i = 0; i < nr_pages; i++) {
659 			WARN_ON(pages[i]->mapping);
660 			put_page(pages[i]);
661 		}
662 		kfree(pages);
663 		pages = NULL;
664 		total_compressed = 0;
665 		nr_pages = 0;
666 
667 		/* flag the file so we don't compress in the future */
668 		if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
669 		    !(BTRFS_I(inode)->prop_compress)) {
670 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
671 		}
672 	}
673 cleanup_and_bail_uncompressed:
674 	/*
675 	 * No compression, but we still need to write the pages in the file
676 	 * we've been given so far.  redirty the locked page if it corresponds
677 	 * to our extent and set things up for the async work queue to run
678 	 * cow_file_range to do the normal delalloc dance.
679 	 */
680 	if (page_offset(locked_page) >= start &&
681 	    page_offset(locked_page) <= end)
682 		__set_page_dirty_nobuffers(locked_page);
683 		/* unlocked later on in the async handlers */
684 
685 	if (redirty)
686 		extent_range_redirty_for_io(inode, start, end);
687 	add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
688 			 BTRFS_COMPRESS_NONE);
689 	*num_added += 1;
690 
691 	return;
692 
693 free_pages_out:
694 	for (i = 0; i < nr_pages; i++) {
695 		WARN_ON(pages[i]->mapping);
696 		put_page(pages[i]);
697 	}
698 	kfree(pages);
699 }
700 
701 static void free_async_extent_pages(struct async_extent *async_extent)
702 {
703 	int i;
704 
705 	if (!async_extent->pages)
706 		return;
707 
708 	for (i = 0; i < async_extent->nr_pages; i++) {
709 		WARN_ON(async_extent->pages[i]->mapping);
710 		put_page(async_extent->pages[i]);
711 	}
712 	kfree(async_extent->pages);
713 	async_extent->nr_pages = 0;
714 	async_extent->pages = NULL;
715 }
716 
717 /*
718  * phase two of compressed writeback.  This is the ordered portion
719  * of the code, which only gets called in the order the work was
720  * queued.  We walk all the async extents created by compress_file_range
721  * and send them down to the disk.
722  */
723 static noinline void submit_compressed_extents(struct inode *inode,
724 					      struct async_cow *async_cow)
725 {
726 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
727 	struct async_extent *async_extent;
728 	u64 alloc_hint = 0;
729 	struct btrfs_key ins;
730 	struct extent_map *em;
731 	struct btrfs_root *root = BTRFS_I(inode)->root;
732 	struct extent_io_tree *io_tree;
733 	int ret = 0;
734 
735 again:
736 	while (!list_empty(&async_cow->extents)) {
737 		async_extent = list_entry(async_cow->extents.next,
738 					  struct async_extent, list);
739 		list_del(&async_extent->list);
740 
741 		io_tree = &BTRFS_I(inode)->io_tree;
742 
743 retry:
744 		/* did the compression code fall back to uncompressed IO? */
745 		if (!async_extent->pages) {
746 			int page_started = 0;
747 			unsigned long nr_written = 0;
748 
749 			lock_extent(io_tree, async_extent->start,
750 					 async_extent->start +
751 					 async_extent->ram_size - 1);
752 
753 			/* allocate blocks */
754 			ret = cow_file_range(inode, async_cow->locked_page,
755 					     async_extent->start,
756 					     async_extent->start +
757 					     async_extent->ram_size - 1,
758 					     async_extent->start +
759 					     async_extent->ram_size - 1,
760 					     &page_started, &nr_written, 0,
761 					     NULL);
762 
763 			/* JDM XXX */
764 
765 			/*
766 			 * if page_started, cow_file_range inserted an
767 			 * inline extent and took care of all the unlocking
768 			 * and IO for us.  Otherwise, we need to submit
769 			 * all those pages down to the drive.
770 			 */
771 			if (!page_started && !ret)
772 				extent_write_locked_range(inode,
773 						  async_extent->start,
774 						  async_extent->start +
775 						  async_extent->ram_size - 1,
776 						  WB_SYNC_ALL);
777 			else if (ret)
778 				unlock_page(async_cow->locked_page);
779 			kfree(async_extent);
780 			cond_resched();
781 			continue;
782 		}
783 
784 		lock_extent(io_tree, async_extent->start,
785 			    async_extent->start + async_extent->ram_size - 1);
786 
787 		ret = btrfs_reserve_extent(root, async_extent->ram_size,
788 					   async_extent->compressed_size,
789 					   async_extent->compressed_size,
790 					   0, alloc_hint, &ins, 1, 1);
791 		if (ret) {
792 			free_async_extent_pages(async_extent);
793 
794 			if (ret == -ENOSPC) {
795 				unlock_extent(io_tree, async_extent->start,
796 					      async_extent->start +
797 					      async_extent->ram_size - 1);
798 
799 				/*
800 				 * we need to redirty the pages if we decide to
801 				 * fallback to uncompressed IO, otherwise we
802 				 * will not submit these pages down to lower
803 				 * layers.
804 				 */
805 				extent_range_redirty_for_io(inode,
806 						async_extent->start,
807 						async_extent->start +
808 						async_extent->ram_size - 1);
809 
810 				goto retry;
811 			}
812 			goto out_free;
813 		}
814 		/*
815 		 * here we're doing allocation and writeback of the
816 		 * compressed pages
817 		 */
818 		em = create_io_em(inode, async_extent->start,
819 				  async_extent->ram_size, /* len */
820 				  async_extent->start, /* orig_start */
821 				  ins.objectid, /* block_start */
822 				  ins.offset, /* block_len */
823 				  ins.offset, /* orig_block_len */
824 				  async_extent->ram_size, /* ram_bytes */
825 				  async_extent->compress_type,
826 				  BTRFS_ORDERED_COMPRESSED);
827 		if (IS_ERR(em))
828 			/* ret value is not necessary due to void function */
829 			goto out_free_reserve;
830 		free_extent_map(em);
831 
832 		ret = btrfs_add_ordered_extent_compress(inode,
833 						async_extent->start,
834 						ins.objectid,
835 						async_extent->ram_size,
836 						ins.offset,
837 						BTRFS_ORDERED_COMPRESSED,
838 						async_extent->compress_type);
839 		if (ret) {
840 			btrfs_drop_extent_cache(BTRFS_I(inode),
841 						async_extent->start,
842 						async_extent->start +
843 						async_extent->ram_size - 1, 0);
844 			goto out_free_reserve;
845 		}
846 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
847 
848 		/*
849 		 * clear dirty, set writeback and unlock the pages.
850 		 */
851 		extent_clear_unlock_delalloc(inode, async_extent->start,
852 				async_extent->start +
853 				async_extent->ram_size - 1,
854 				async_extent->start +
855 				async_extent->ram_size - 1,
856 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
857 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
858 				PAGE_SET_WRITEBACK);
859 		if (btrfs_submit_compressed_write(inode,
860 				    async_extent->start,
861 				    async_extent->ram_size,
862 				    ins.objectid,
863 				    ins.offset, async_extent->pages,
864 				    async_extent->nr_pages,
865 				    async_cow->write_flags)) {
866 			struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
867 			struct page *p = async_extent->pages[0];
868 			const u64 start = async_extent->start;
869 			const u64 end = start + async_extent->ram_size - 1;
870 
871 			p->mapping = inode->i_mapping;
872 			tree->ops->writepage_end_io_hook(p, start, end,
873 							 NULL, 0);
874 			p->mapping = NULL;
875 			extent_clear_unlock_delalloc(inode, start, end, end,
876 						     NULL, 0,
877 						     PAGE_END_WRITEBACK |
878 						     PAGE_SET_ERROR);
879 			free_async_extent_pages(async_extent);
880 		}
881 		alloc_hint = ins.objectid + ins.offset;
882 		kfree(async_extent);
883 		cond_resched();
884 	}
885 	return;
886 out_free_reserve:
887 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
888 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
889 out_free:
890 	extent_clear_unlock_delalloc(inode, async_extent->start,
891 				     async_extent->start +
892 				     async_extent->ram_size - 1,
893 				     async_extent->start +
894 				     async_extent->ram_size - 1,
895 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
896 				     EXTENT_DELALLOC_NEW |
897 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
898 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
899 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
900 				     PAGE_SET_ERROR);
901 	free_async_extent_pages(async_extent);
902 	kfree(async_extent);
903 	goto again;
904 }
905 
906 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
907 				      u64 num_bytes)
908 {
909 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
910 	struct extent_map *em;
911 	u64 alloc_hint = 0;
912 
913 	read_lock(&em_tree->lock);
914 	em = search_extent_mapping(em_tree, start, num_bytes);
915 	if (em) {
916 		/*
917 		 * if block start isn't an actual block number then find the
918 		 * first block in this inode and use that as a hint.  If that
919 		 * block is also bogus then just don't worry about it.
920 		 */
921 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
922 			free_extent_map(em);
923 			em = search_extent_mapping(em_tree, 0, 0);
924 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
925 				alloc_hint = em->block_start;
926 			if (em)
927 				free_extent_map(em);
928 		} else {
929 			alloc_hint = em->block_start;
930 			free_extent_map(em);
931 		}
932 	}
933 	read_unlock(&em_tree->lock);
934 
935 	return alloc_hint;
936 }
937 
938 /*
939  * when extent_io.c finds a delayed allocation range in the file,
940  * the call backs end up in this code.  The basic idea is to
941  * allocate extents on disk for the range, and create ordered data structs
942  * in ram to track those extents.
943  *
944  * locked_page is the page that writepage had locked already.  We use
945  * it to make sure we don't do extra locks or unlocks.
946  *
947  * *page_started is set to one if we unlock locked_page and do everything
948  * required to start IO on it.  It may be clean and already done with
949  * IO when we return.
950  */
951 static noinline int cow_file_range(struct inode *inode,
952 				   struct page *locked_page,
953 				   u64 start, u64 end, u64 delalloc_end,
954 				   int *page_started, unsigned long *nr_written,
955 				   int unlock, struct btrfs_dedupe_hash *hash)
956 {
957 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
958 	struct btrfs_root *root = BTRFS_I(inode)->root;
959 	u64 alloc_hint = 0;
960 	u64 num_bytes;
961 	unsigned long ram_size;
962 	u64 cur_alloc_size = 0;
963 	u64 blocksize = fs_info->sectorsize;
964 	struct btrfs_key ins;
965 	struct extent_map *em;
966 	unsigned clear_bits;
967 	unsigned long page_ops;
968 	bool extent_reserved = false;
969 	int ret = 0;
970 
971 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
972 		WARN_ON_ONCE(1);
973 		ret = -EINVAL;
974 		goto out_unlock;
975 	}
976 
977 	num_bytes = ALIGN(end - start + 1, blocksize);
978 	num_bytes = max(blocksize,  num_bytes);
979 	ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
980 
981 	inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
982 
983 	if (start == 0) {
984 		/* lets try to make an inline extent */
985 		ret = cow_file_range_inline(inode, start, end, 0,
986 					    BTRFS_COMPRESS_NONE, NULL);
987 		if (ret == 0) {
988 			/*
989 			 * We use DO_ACCOUNTING here because we need the
990 			 * delalloc_release_metadata to be run _after_ we drop
991 			 * our outstanding extent for clearing delalloc for this
992 			 * range.
993 			 */
994 			extent_clear_unlock_delalloc(inode, start, end,
995 				     delalloc_end, NULL,
996 				     EXTENT_LOCKED | EXTENT_DELALLOC |
997 				     EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
998 				     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
999 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1000 				     PAGE_END_WRITEBACK);
1001 			*nr_written = *nr_written +
1002 			     (end - start + PAGE_SIZE) / PAGE_SIZE;
1003 			*page_started = 1;
1004 			goto out;
1005 		} else if (ret < 0) {
1006 			goto out_unlock;
1007 		}
1008 	}
1009 
1010 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1011 	btrfs_drop_extent_cache(BTRFS_I(inode), start,
1012 			start + num_bytes - 1, 0);
1013 
1014 	while (num_bytes > 0) {
1015 		cur_alloc_size = num_bytes;
1016 		ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1017 					   fs_info->sectorsize, 0, alloc_hint,
1018 					   &ins, 1, 1);
1019 		if (ret < 0)
1020 			goto out_unlock;
1021 		cur_alloc_size = ins.offset;
1022 		extent_reserved = true;
1023 
1024 		ram_size = ins.offset;
1025 		em = create_io_em(inode, start, ins.offset, /* len */
1026 				  start, /* orig_start */
1027 				  ins.objectid, /* block_start */
1028 				  ins.offset, /* block_len */
1029 				  ins.offset, /* orig_block_len */
1030 				  ram_size, /* ram_bytes */
1031 				  BTRFS_COMPRESS_NONE, /* compress_type */
1032 				  BTRFS_ORDERED_REGULAR /* type */);
1033 		if (IS_ERR(em))
1034 			goto out_reserve;
1035 		free_extent_map(em);
1036 
1037 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1038 					       ram_size, cur_alloc_size, 0);
1039 		if (ret)
1040 			goto out_drop_extent_cache;
1041 
1042 		if (root->root_key.objectid ==
1043 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1044 			ret = btrfs_reloc_clone_csums(inode, start,
1045 						      cur_alloc_size);
1046 			/*
1047 			 * Only drop cache here, and process as normal.
1048 			 *
1049 			 * We must not allow extent_clear_unlock_delalloc()
1050 			 * at out_unlock label to free meta of this ordered
1051 			 * extent, as its meta should be freed by
1052 			 * btrfs_finish_ordered_io().
1053 			 *
1054 			 * So we must continue until @start is increased to
1055 			 * skip current ordered extent.
1056 			 */
1057 			if (ret)
1058 				btrfs_drop_extent_cache(BTRFS_I(inode), start,
1059 						start + ram_size - 1, 0);
1060 		}
1061 
1062 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1063 
1064 		/* we're not doing compressed IO, don't unlock the first
1065 		 * page (which the caller expects to stay locked), don't
1066 		 * clear any dirty bits and don't set any writeback bits
1067 		 *
1068 		 * Do set the Private2 bit so we know this page was properly
1069 		 * setup for writepage
1070 		 */
1071 		page_ops = unlock ? PAGE_UNLOCK : 0;
1072 		page_ops |= PAGE_SET_PRIVATE2;
1073 
1074 		extent_clear_unlock_delalloc(inode, start,
1075 					     start + ram_size - 1,
1076 					     delalloc_end, locked_page,
1077 					     EXTENT_LOCKED | EXTENT_DELALLOC,
1078 					     page_ops);
1079 		if (num_bytes < cur_alloc_size)
1080 			num_bytes = 0;
1081 		else
1082 			num_bytes -= cur_alloc_size;
1083 		alloc_hint = ins.objectid + ins.offset;
1084 		start += cur_alloc_size;
1085 		extent_reserved = false;
1086 
1087 		/*
1088 		 * btrfs_reloc_clone_csums() error, since start is increased
1089 		 * extent_clear_unlock_delalloc() at out_unlock label won't
1090 		 * free metadata of current ordered extent, we're OK to exit.
1091 		 */
1092 		if (ret)
1093 			goto out_unlock;
1094 	}
1095 out:
1096 	return ret;
1097 
1098 out_drop_extent_cache:
1099 	btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1100 out_reserve:
1101 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1102 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1103 out_unlock:
1104 	clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1105 		EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1106 	page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1107 		PAGE_END_WRITEBACK;
1108 	/*
1109 	 * If we reserved an extent for our delalloc range (or a subrange) and
1110 	 * failed to create the respective ordered extent, then it means that
1111 	 * when we reserved the extent we decremented the extent's size from
1112 	 * the data space_info's bytes_may_use counter and incremented the
1113 	 * space_info's bytes_reserved counter by the same amount. We must make
1114 	 * sure extent_clear_unlock_delalloc() does not try to decrement again
1115 	 * the data space_info's bytes_may_use counter, therefore we do not pass
1116 	 * it the flag EXTENT_CLEAR_DATA_RESV.
1117 	 */
1118 	if (extent_reserved) {
1119 		extent_clear_unlock_delalloc(inode, start,
1120 					     start + cur_alloc_size,
1121 					     start + cur_alloc_size,
1122 					     locked_page,
1123 					     clear_bits,
1124 					     page_ops);
1125 		start += cur_alloc_size;
1126 		if (start >= end)
1127 			goto out;
1128 	}
1129 	extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1130 				     locked_page,
1131 				     clear_bits | EXTENT_CLEAR_DATA_RESV,
1132 				     page_ops);
1133 	goto out;
1134 }
1135 
1136 /*
1137  * work queue call back to started compression on a file and pages
1138  */
1139 static noinline void async_cow_start(struct btrfs_work *work)
1140 {
1141 	struct async_cow *async_cow;
1142 	int num_added = 0;
1143 	async_cow = container_of(work, struct async_cow, work);
1144 
1145 	compress_file_range(async_cow->inode, async_cow->locked_page,
1146 			    async_cow->start, async_cow->end, async_cow,
1147 			    &num_added);
1148 	if (num_added == 0) {
1149 		btrfs_add_delayed_iput(async_cow->inode);
1150 		async_cow->inode = NULL;
1151 	}
1152 }
1153 
1154 /*
1155  * work queue call back to submit previously compressed pages
1156  */
1157 static noinline void async_cow_submit(struct btrfs_work *work)
1158 {
1159 	struct btrfs_fs_info *fs_info;
1160 	struct async_cow *async_cow;
1161 	struct btrfs_root *root;
1162 	unsigned long nr_pages;
1163 
1164 	async_cow = container_of(work, struct async_cow, work);
1165 
1166 	root = async_cow->root;
1167 	fs_info = root->fs_info;
1168 	nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1169 		PAGE_SHIFT;
1170 
1171 	/*
1172 	 * atomic_sub_return implies a barrier for waitqueue_active
1173 	 */
1174 	if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1175 	    5 * SZ_1M &&
1176 	    waitqueue_active(&fs_info->async_submit_wait))
1177 		wake_up(&fs_info->async_submit_wait);
1178 
1179 	if (async_cow->inode)
1180 		submit_compressed_extents(async_cow->inode, async_cow);
1181 }
1182 
1183 static noinline void async_cow_free(struct btrfs_work *work)
1184 {
1185 	struct async_cow *async_cow;
1186 	async_cow = container_of(work, struct async_cow, work);
1187 	if (async_cow->inode)
1188 		btrfs_add_delayed_iput(async_cow->inode);
1189 	kfree(async_cow);
1190 }
1191 
1192 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1193 				u64 start, u64 end, int *page_started,
1194 				unsigned long *nr_written,
1195 				unsigned int write_flags)
1196 {
1197 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1198 	struct async_cow *async_cow;
1199 	struct btrfs_root *root = BTRFS_I(inode)->root;
1200 	unsigned long nr_pages;
1201 	u64 cur_end;
1202 
1203 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1204 			 1, 0, NULL);
1205 	while (start < end) {
1206 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1207 		BUG_ON(!async_cow); /* -ENOMEM */
1208 		async_cow->inode = igrab(inode);
1209 		async_cow->root = root;
1210 		async_cow->locked_page = locked_page;
1211 		async_cow->start = start;
1212 		async_cow->write_flags = write_flags;
1213 
1214 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1215 		    !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1216 			cur_end = end;
1217 		else
1218 			cur_end = min(end, start + SZ_512K - 1);
1219 
1220 		async_cow->end = cur_end;
1221 		INIT_LIST_HEAD(&async_cow->extents);
1222 
1223 		btrfs_init_work(&async_cow->work,
1224 				btrfs_delalloc_helper,
1225 				async_cow_start, async_cow_submit,
1226 				async_cow_free);
1227 
1228 		nr_pages = (cur_end - start + PAGE_SIZE) >>
1229 			PAGE_SHIFT;
1230 		atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1231 
1232 		btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1233 
1234 		*nr_written += nr_pages;
1235 		start = cur_end + 1;
1236 	}
1237 	*page_started = 1;
1238 	return 0;
1239 }
1240 
1241 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1242 					u64 bytenr, u64 num_bytes)
1243 {
1244 	int ret;
1245 	struct btrfs_ordered_sum *sums;
1246 	LIST_HEAD(list);
1247 
1248 	ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1249 				       bytenr + num_bytes - 1, &list, 0);
1250 	if (ret == 0 && list_empty(&list))
1251 		return 0;
1252 
1253 	while (!list_empty(&list)) {
1254 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1255 		list_del(&sums->list);
1256 		kfree(sums);
1257 	}
1258 	if (ret < 0)
1259 		return ret;
1260 	return 1;
1261 }
1262 
1263 /*
1264  * when nowcow writeback call back.  This checks for snapshots or COW copies
1265  * of the extents that exist in the file, and COWs the file as required.
1266  *
1267  * If no cow copies or snapshots exist, we write directly to the existing
1268  * blocks on disk
1269  */
1270 static noinline int run_delalloc_nocow(struct inode *inode,
1271 				       struct page *locked_page,
1272 			      u64 start, u64 end, int *page_started, int force,
1273 			      unsigned long *nr_written)
1274 {
1275 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1276 	struct btrfs_root *root = BTRFS_I(inode)->root;
1277 	struct extent_buffer *leaf;
1278 	struct btrfs_path *path;
1279 	struct btrfs_file_extent_item *fi;
1280 	struct btrfs_key found_key;
1281 	struct extent_map *em;
1282 	u64 cow_start;
1283 	u64 cur_offset;
1284 	u64 extent_end;
1285 	u64 extent_offset;
1286 	u64 disk_bytenr;
1287 	u64 num_bytes;
1288 	u64 disk_num_bytes;
1289 	u64 ram_bytes;
1290 	int extent_type;
1291 	int ret, err;
1292 	int type;
1293 	int nocow;
1294 	int check_prev = 1;
1295 	bool nolock;
1296 	u64 ino = btrfs_ino(BTRFS_I(inode));
1297 
1298 	path = btrfs_alloc_path();
1299 	if (!path) {
1300 		extent_clear_unlock_delalloc(inode, start, end, end,
1301 					     locked_page,
1302 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1303 					     EXTENT_DO_ACCOUNTING |
1304 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1305 					     PAGE_CLEAR_DIRTY |
1306 					     PAGE_SET_WRITEBACK |
1307 					     PAGE_END_WRITEBACK);
1308 		return -ENOMEM;
1309 	}
1310 
1311 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1312 
1313 	cow_start = (u64)-1;
1314 	cur_offset = start;
1315 	while (1) {
1316 		ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1317 					       cur_offset, 0);
1318 		if (ret < 0)
1319 			goto error;
1320 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1321 			leaf = path->nodes[0];
1322 			btrfs_item_key_to_cpu(leaf, &found_key,
1323 					      path->slots[0] - 1);
1324 			if (found_key.objectid == ino &&
1325 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1326 				path->slots[0]--;
1327 		}
1328 		check_prev = 0;
1329 next_slot:
1330 		leaf = path->nodes[0];
1331 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 			ret = btrfs_next_leaf(root, path);
1333 			if (ret < 0) {
1334 				if (cow_start != (u64)-1)
1335 					cur_offset = cow_start;
1336 				goto error;
1337 			}
1338 			if (ret > 0)
1339 				break;
1340 			leaf = path->nodes[0];
1341 		}
1342 
1343 		nocow = 0;
1344 		disk_bytenr = 0;
1345 		num_bytes = 0;
1346 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1347 
1348 		if (found_key.objectid > ino)
1349 			break;
1350 		if (WARN_ON_ONCE(found_key.objectid < ino) ||
1351 		    found_key.type < BTRFS_EXTENT_DATA_KEY) {
1352 			path->slots[0]++;
1353 			goto next_slot;
1354 		}
1355 		if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1356 		    found_key.offset > end)
1357 			break;
1358 
1359 		if (found_key.offset > cur_offset) {
1360 			extent_end = found_key.offset;
1361 			extent_type = 0;
1362 			goto out_check;
1363 		}
1364 
1365 		fi = btrfs_item_ptr(leaf, path->slots[0],
1366 				    struct btrfs_file_extent_item);
1367 		extent_type = btrfs_file_extent_type(leaf, fi);
1368 
1369 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1370 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1371 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1372 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1373 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1374 			extent_end = found_key.offset +
1375 				btrfs_file_extent_num_bytes(leaf, fi);
1376 			disk_num_bytes =
1377 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1378 			if (extent_end <= start) {
1379 				path->slots[0]++;
1380 				goto next_slot;
1381 			}
1382 			if (disk_bytenr == 0)
1383 				goto out_check;
1384 			if (btrfs_file_extent_compression(leaf, fi) ||
1385 			    btrfs_file_extent_encryption(leaf, fi) ||
1386 			    btrfs_file_extent_other_encoding(leaf, fi))
1387 				goto out_check;
1388 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1389 				goto out_check;
1390 			if (btrfs_extent_readonly(fs_info, disk_bytenr))
1391 				goto out_check;
1392 			ret = btrfs_cross_ref_exist(root, ino,
1393 						    found_key.offset -
1394 						    extent_offset, disk_bytenr);
1395 			if (ret) {
1396 				/*
1397 				 * ret could be -EIO if the above fails to read
1398 				 * metadata.
1399 				 */
1400 				if (ret < 0) {
1401 					if (cow_start != (u64)-1)
1402 						cur_offset = cow_start;
1403 					goto error;
1404 				}
1405 
1406 				WARN_ON_ONCE(nolock);
1407 				goto out_check;
1408 			}
1409 			disk_bytenr += extent_offset;
1410 			disk_bytenr += cur_offset - found_key.offset;
1411 			num_bytes = min(end + 1, extent_end) - cur_offset;
1412 			/*
1413 			 * if there are pending snapshots for this root,
1414 			 * we fall into common COW way.
1415 			 */
1416 			if (!nolock) {
1417 				err = btrfs_start_write_no_snapshotting(root);
1418 				if (!err)
1419 					goto out_check;
1420 			}
1421 			/*
1422 			 * force cow if csum exists in the range.
1423 			 * this ensure that csum for a given extent are
1424 			 * either valid or do not exist.
1425 			 */
1426 			ret = csum_exist_in_range(fs_info, disk_bytenr,
1427 						  num_bytes);
1428 			if (ret) {
1429 				if (!nolock)
1430 					btrfs_end_write_no_snapshotting(root);
1431 
1432 				/*
1433 				 * ret could be -EIO if the above fails to read
1434 				 * metadata.
1435 				 */
1436 				if (ret < 0) {
1437 					if (cow_start != (u64)-1)
1438 						cur_offset = cow_start;
1439 					goto error;
1440 				}
1441 				WARN_ON_ONCE(nolock);
1442 				goto out_check;
1443 			}
1444 			if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1445 				if (!nolock)
1446 					btrfs_end_write_no_snapshotting(root);
1447 				goto out_check;
1448 			}
1449 			nocow = 1;
1450 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1451 			extent_end = found_key.offset +
1452 				btrfs_file_extent_inline_len(leaf,
1453 						     path->slots[0], fi);
1454 			extent_end = ALIGN(extent_end,
1455 					   fs_info->sectorsize);
1456 		} else {
1457 			BUG_ON(1);
1458 		}
1459 out_check:
1460 		if (extent_end <= start) {
1461 			path->slots[0]++;
1462 			if (!nolock && nocow)
1463 				btrfs_end_write_no_snapshotting(root);
1464 			if (nocow)
1465 				btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1466 			goto next_slot;
1467 		}
1468 		if (!nocow) {
1469 			if (cow_start == (u64)-1)
1470 				cow_start = cur_offset;
1471 			cur_offset = extent_end;
1472 			if (cur_offset > end)
1473 				break;
1474 			path->slots[0]++;
1475 			goto next_slot;
1476 		}
1477 
1478 		btrfs_release_path(path);
1479 		if (cow_start != (u64)-1) {
1480 			ret = cow_file_range(inode, locked_page,
1481 					     cow_start, found_key.offset - 1,
1482 					     end, page_started, nr_written, 1,
1483 					     NULL);
1484 			if (ret) {
1485 				if (!nolock && nocow)
1486 					btrfs_end_write_no_snapshotting(root);
1487 				if (nocow)
1488 					btrfs_dec_nocow_writers(fs_info,
1489 								disk_bytenr);
1490 				goto error;
1491 			}
1492 			cow_start = (u64)-1;
1493 		}
1494 
1495 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1496 			u64 orig_start = found_key.offset - extent_offset;
1497 
1498 			em = create_io_em(inode, cur_offset, num_bytes,
1499 					  orig_start,
1500 					  disk_bytenr, /* block_start */
1501 					  num_bytes, /* block_len */
1502 					  disk_num_bytes, /* orig_block_len */
1503 					  ram_bytes, BTRFS_COMPRESS_NONE,
1504 					  BTRFS_ORDERED_PREALLOC);
1505 			if (IS_ERR(em)) {
1506 				if (!nolock && nocow)
1507 					btrfs_end_write_no_snapshotting(root);
1508 				if (nocow)
1509 					btrfs_dec_nocow_writers(fs_info,
1510 								disk_bytenr);
1511 				ret = PTR_ERR(em);
1512 				goto error;
1513 			}
1514 			free_extent_map(em);
1515 		}
1516 
1517 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1518 			type = BTRFS_ORDERED_PREALLOC;
1519 		} else {
1520 			type = BTRFS_ORDERED_NOCOW;
1521 		}
1522 
1523 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1524 					       num_bytes, num_bytes, type);
1525 		if (nocow)
1526 			btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1527 		BUG_ON(ret); /* -ENOMEM */
1528 
1529 		if (root->root_key.objectid ==
1530 		    BTRFS_DATA_RELOC_TREE_OBJECTID)
1531 			/*
1532 			 * Error handled later, as we must prevent
1533 			 * extent_clear_unlock_delalloc() in error handler
1534 			 * from freeing metadata of created ordered extent.
1535 			 */
1536 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1537 						      num_bytes);
1538 
1539 		extent_clear_unlock_delalloc(inode, cur_offset,
1540 					     cur_offset + num_bytes - 1, end,
1541 					     locked_page, EXTENT_LOCKED |
1542 					     EXTENT_DELALLOC |
1543 					     EXTENT_CLEAR_DATA_RESV,
1544 					     PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1545 
1546 		if (!nolock && nocow)
1547 			btrfs_end_write_no_snapshotting(root);
1548 		cur_offset = extent_end;
1549 
1550 		/*
1551 		 * btrfs_reloc_clone_csums() error, now we're OK to call error
1552 		 * handler, as metadata for created ordered extent will only
1553 		 * be freed by btrfs_finish_ordered_io().
1554 		 */
1555 		if (ret)
1556 			goto error;
1557 		if (cur_offset > end)
1558 			break;
1559 	}
1560 	btrfs_release_path(path);
1561 
1562 	if (cur_offset <= end && cow_start == (u64)-1) {
1563 		cow_start = cur_offset;
1564 		cur_offset = end;
1565 	}
1566 
1567 	if (cow_start != (u64)-1) {
1568 		ret = cow_file_range(inode, locked_page, cow_start, end, end,
1569 				     page_started, nr_written, 1, NULL);
1570 		if (ret)
1571 			goto error;
1572 	}
1573 
1574 error:
1575 	if (ret && cur_offset < end)
1576 		extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1577 					     locked_page, EXTENT_LOCKED |
1578 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1579 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1580 					     PAGE_CLEAR_DIRTY |
1581 					     PAGE_SET_WRITEBACK |
1582 					     PAGE_END_WRITEBACK);
1583 	btrfs_free_path(path);
1584 	return ret;
1585 }
1586 
1587 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1588 {
1589 
1590 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1591 	    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1592 		return 0;
1593 
1594 	/*
1595 	 * @defrag_bytes is a hint value, no spinlock held here,
1596 	 * if is not zero, it means the file is defragging.
1597 	 * Force cow if given extent needs to be defragged.
1598 	 */
1599 	if (BTRFS_I(inode)->defrag_bytes &&
1600 	    test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1601 			   EXTENT_DEFRAG, 0, NULL))
1602 		return 1;
1603 
1604 	return 0;
1605 }
1606 
1607 /*
1608  * extent_io.c call back to do delayed allocation processing
1609  */
1610 static int run_delalloc_range(void *private_data, struct page *locked_page,
1611 			      u64 start, u64 end, int *page_started,
1612 			      unsigned long *nr_written,
1613 			      struct writeback_control *wbc)
1614 {
1615 	struct inode *inode = private_data;
1616 	int ret;
1617 	int force_cow = need_force_cow(inode, start, end);
1618 	unsigned int write_flags = wbc_to_write_flags(wbc);
1619 
1620 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1621 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1622 					 page_started, 1, nr_written);
1623 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1624 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1625 					 page_started, 0, nr_written);
1626 	} else if (!inode_need_compress(inode, start, end)) {
1627 		ret = cow_file_range(inode, locked_page, start, end, end,
1628 				      page_started, nr_written, 1, NULL);
1629 	} else {
1630 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1631 			&BTRFS_I(inode)->runtime_flags);
1632 		ret = cow_file_range_async(inode, locked_page, start, end,
1633 					   page_started, nr_written,
1634 					   write_flags);
1635 	}
1636 	if (ret)
1637 		btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1638 	return ret;
1639 }
1640 
1641 static void btrfs_split_extent_hook(void *private_data,
1642 				    struct extent_state *orig, u64 split)
1643 {
1644 	struct inode *inode = private_data;
1645 	u64 size;
1646 
1647 	/* not delalloc, ignore it */
1648 	if (!(orig->state & EXTENT_DELALLOC))
1649 		return;
1650 
1651 	size = orig->end - orig->start + 1;
1652 	if (size > BTRFS_MAX_EXTENT_SIZE) {
1653 		u32 num_extents;
1654 		u64 new_size;
1655 
1656 		/*
1657 		 * See the explanation in btrfs_merge_extent_hook, the same
1658 		 * applies here, just in reverse.
1659 		 */
1660 		new_size = orig->end - split + 1;
1661 		num_extents = count_max_extents(new_size);
1662 		new_size = split - orig->start;
1663 		num_extents += count_max_extents(new_size);
1664 		if (count_max_extents(size) >= num_extents)
1665 			return;
1666 	}
1667 
1668 	spin_lock(&BTRFS_I(inode)->lock);
1669 	btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1670 	spin_unlock(&BTRFS_I(inode)->lock);
1671 }
1672 
1673 /*
1674  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1675  * extents so we can keep track of new extents that are just merged onto old
1676  * extents, such as when we are doing sequential writes, so we can properly
1677  * account for the metadata space we'll need.
1678  */
1679 static void btrfs_merge_extent_hook(void *private_data,
1680 				    struct extent_state *new,
1681 				    struct extent_state *other)
1682 {
1683 	struct inode *inode = private_data;
1684 	u64 new_size, old_size;
1685 	u32 num_extents;
1686 
1687 	/* not delalloc, ignore it */
1688 	if (!(other->state & EXTENT_DELALLOC))
1689 		return;
1690 
1691 	if (new->start > other->start)
1692 		new_size = new->end - other->start + 1;
1693 	else
1694 		new_size = other->end - new->start + 1;
1695 
1696 	/* we're not bigger than the max, unreserve the space and go */
1697 	if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1698 		spin_lock(&BTRFS_I(inode)->lock);
1699 		btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1700 		spin_unlock(&BTRFS_I(inode)->lock);
1701 		return;
1702 	}
1703 
1704 	/*
1705 	 * We have to add up either side to figure out how many extents were
1706 	 * accounted for before we merged into one big extent.  If the number of
1707 	 * extents we accounted for is <= the amount we need for the new range
1708 	 * then we can return, otherwise drop.  Think of it like this
1709 	 *
1710 	 * [ 4k][MAX_SIZE]
1711 	 *
1712 	 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1713 	 * need 2 outstanding extents, on one side we have 1 and the other side
1714 	 * we have 1 so they are == and we can return.  But in this case
1715 	 *
1716 	 * [MAX_SIZE+4k][MAX_SIZE+4k]
1717 	 *
1718 	 * Each range on their own accounts for 2 extents, but merged together
1719 	 * they are only 3 extents worth of accounting, so we need to drop in
1720 	 * this case.
1721 	 */
1722 	old_size = other->end - other->start + 1;
1723 	num_extents = count_max_extents(old_size);
1724 	old_size = new->end - new->start + 1;
1725 	num_extents += count_max_extents(old_size);
1726 	if (count_max_extents(new_size) >= num_extents)
1727 		return;
1728 
1729 	spin_lock(&BTRFS_I(inode)->lock);
1730 	btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1731 	spin_unlock(&BTRFS_I(inode)->lock);
1732 }
1733 
1734 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1735 				      struct inode *inode)
1736 {
1737 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1738 
1739 	spin_lock(&root->delalloc_lock);
1740 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1741 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1742 			      &root->delalloc_inodes);
1743 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1744 			&BTRFS_I(inode)->runtime_flags);
1745 		root->nr_delalloc_inodes++;
1746 		if (root->nr_delalloc_inodes == 1) {
1747 			spin_lock(&fs_info->delalloc_root_lock);
1748 			BUG_ON(!list_empty(&root->delalloc_root));
1749 			list_add_tail(&root->delalloc_root,
1750 				      &fs_info->delalloc_roots);
1751 			spin_unlock(&fs_info->delalloc_root_lock);
1752 		}
1753 	}
1754 	spin_unlock(&root->delalloc_lock);
1755 }
1756 
1757 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1758 				     struct btrfs_inode *inode)
1759 {
1760 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1761 
1762 	spin_lock(&root->delalloc_lock);
1763 	if (!list_empty(&inode->delalloc_inodes)) {
1764 		list_del_init(&inode->delalloc_inodes);
1765 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1766 			  &inode->runtime_flags);
1767 		root->nr_delalloc_inodes--;
1768 		if (!root->nr_delalloc_inodes) {
1769 			spin_lock(&fs_info->delalloc_root_lock);
1770 			BUG_ON(list_empty(&root->delalloc_root));
1771 			list_del_init(&root->delalloc_root);
1772 			spin_unlock(&fs_info->delalloc_root_lock);
1773 		}
1774 	}
1775 	spin_unlock(&root->delalloc_lock);
1776 }
1777 
1778 /*
1779  * extent_io.c set_bit_hook, used to track delayed allocation
1780  * bytes in this file, and to maintain the list of inodes that
1781  * have pending delalloc work to be done.
1782  */
1783 static void btrfs_set_bit_hook(void *private_data,
1784 			       struct extent_state *state, unsigned *bits)
1785 {
1786 	struct inode *inode = private_data;
1787 
1788 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1789 
1790 	if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1791 		WARN_ON(1);
1792 	/*
1793 	 * set_bit and clear bit hooks normally require _irqsave/restore
1794 	 * but in this case, we are only testing for the DELALLOC
1795 	 * bit, which is only set or cleared with irqs on
1796 	 */
1797 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1798 		struct btrfs_root *root = BTRFS_I(inode)->root;
1799 		u64 len = state->end + 1 - state->start;
1800 		u32 num_extents = count_max_extents(len);
1801 		bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1802 
1803 		spin_lock(&BTRFS_I(inode)->lock);
1804 		btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1805 		spin_unlock(&BTRFS_I(inode)->lock);
1806 
1807 		/* For sanity tests */
1808 		if (btrfs_is_testing(fs_info))
1809 			return;
1810 
1811 		percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1812 					 fs_info->delalloc_batch);
1813 		spin_lock(&BTRFS_I(inode)->lock);
1814 		BTRFS_I(inode)->delalloc_bytes += len;
1815 		if (*bits & EXTENT_DEFRAG)
1816 			BTRFS_I(inode)->defrag_bytes += len;
1817 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1818 					 &BTRFS_I(inode)->runtime_flags))
1819 			btrfs_add_delalloc_inodes(root, inode);
1820 		spin_unlock(&BTRFS_I(inode)->lock);
1821 	}
1822 
1823 	if (!(state->state & EXTENT_DELALLOC_NEW) &&
1824 	    (*bits & EXTENT_DELALLOC_NEW)) {
1825 		spin_lock(&BTRFS_I(inode)->lock);
1826 		BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1827 			state->start;
1828 		spin_unlock(&BTRFS_I(inode)->lock);
1829 	}
1830 }
1831 
1832 /*
1833  * extent_io.c clear_bit_hook, see set_bit_hook for why
1834  */
1835 static void btrfs_clear_bit_hook(void *private_data,
1836 				 struct extent_state *state,
1837 				 unsigned *bits)
1838 {
1839 	struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1840 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1841 	u64 len = state->end + 1 - state->start;
1842 	u32 num_extents = count_max_extents(len);
1843 
1844 	if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1845 		spin_lock(&inode->lock);
1846 		inode->defrag_bytes -= len;
1847 		spin_unlock(&inode->lock);
1848 	}
1849 
1850 	/*
1851 	 * set_bit and clear bit hooks normally require _irqsave/restore
1852 	 * but in this case, we are only testing for the DELALLOC
1853 	 * bit, which is only set or cleared with irqs on
1854 	 */
1855 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1856 		struct btrfs_root *root = inode->root;
1857 		bool do_list = !btrfs_is_free_space_inode(inode);
1858 
1859 		spin_lock(&inode->lock);
1860 		btrfs_mod_outstanding_extents(inode, -num_extents);
1861 		spin_unlock(&inode->lock);
1862 
1863 		/*
1864 		 * We don't reserve metadata space for space cache inodes so we
1865 		 * don't need to call dellalloc_release_metadata if there is an
1866 		 * error.
1867 		 */
1868 		if (*bits & EXTENT_CLEAR_META_RESV &&
1869 		    root != fs_info->tree_root)
1870 			btrfs_delalloc_release_metadata(inode, len, false);
1871 
1872 		/* For sanity tests. */
1873 		if (btrfs_is_testing(fs_info))
1874 			return;
1875 
1876 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1877 		    do_list && !(state->state & EXTENT_NORESERVE) &&
1878 		    (*bits & EXTENT_CLEAR_DATA_RESV))
1879 			btrfs_free_reserved_data_space_noquota(
1880 					&inode->vfs_inode,
1881 					state->start, len);
1882 
1883 		percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1884 					 fs_info->delalloc_batch);
1885 		spin_lock(&inode->lock);
1886 		inode->delalloc_bytes -= len;
1887 		if (do_list && inode->delalloc_bytes == 0 &&
1888 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1889 					&inode->runtime_flags))
1890 			btrfs_del_delalloc_inode(root, inode);
1891 		spin_unlock(&inode->lock);
1892 	}
1893 
1894 	if ((state->state & EXTENT_DELALLOC_NEW) &&
1895 	    (*bits & EXTENT_DELALLOC_NEW)) {
1896 		spin_lock(&inode->lock);
1897 		ASSERT(inode->new_delalloc_bytes >= len);
1898 		inode->new_delalloc_bytes -= len;
1899 		spin_unlock(&inode->lock);
1900 	}
1901 }
1902 
1903 /*
1904  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1905  * we don't create bios that span stripes or chunks
1906  *
1907  * return 1 if page cannot be merged to bio
1908  * return 0 if page can be merged to bio
1909  * return error otherwise
1910  */
1911 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1912 			 size_t size, struct bio *bio,
1913 			 unsigned long bio_flags)
1914 {
1915 	struct inode *inode = page->mapping->host;
1916 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1917 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1918 	u64 length = 0;
1919 	u64 map_length;
1920 	int ret;
1921 
1922 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1923 		return 0;
1924 
1925 	length = bio->bi_iter.bi_size;
1926 	map_length = length;
1927 	ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1928 			      NULL, 0);
1929 	if (ret < 0)
1930 		return ret;
1931 	if (map_length < length + size)
1932 		return 1;
1933 	return 0;
1934 }
1935 
1936 /*
1937  * in order to insert checksums into the metadata in large chunks,
1938  * we wait until bio submission time.   All the pages in the bio are
1939  * checksummed and sums are attached onto the ordered extent record.
1940  *
1941  * At IO completion time the cums attached on the ordered extent record
1942  * are inserted into the btree
1943  */
1944 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1945 				    u64 bio_offset)
1946 {
1947 	struct inode *inode = private_data;
1948 	blk_status_t ret = 0;
1949 
1950 	ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1951 	BUG_ON(ret); /* -ENOMEM */
1952 	return 0;
1953 }
1954 
1955 /*
1956  * in order to insert checksums into the metadata in large chunks,
1957  * we wait until bio submission time.   All the pages in the bio are
1958  * checksummed and sums are attached onto the ordered extent record.
1959  *
1960  * At IO completion time the cums attached on the ordered extent record
1961  * are inserted into the btree
1962  */
1963 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1964 			  int mirror_num)
1965 {
1966 	struct inode *inode = private_data;
1967 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1968 	blk_status_t ret;
1969 
1970 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1971 	if (ret) {
1972 		bio->bi_status = ret;
1973 		bio_endio(bio);
1974 	}
1975 	return ret;
1976 }
1977 
1978 /*
1979  * extent_io.c submission hook. This does the right thing for csum calculation
1980  * on write, or reading the csums from the tree before a read.
1981  *
1982  * Rules about async/sync submit,
1983  * a) read:				sync submit
1984  *
1985  * b) write without checksum:		sync submit
1986  *
1987  * c) write with checksum:
1988  *    c-1) if bio is issued by fsync:	sync submit
1989  *         (sync_writers != 0)
1990  *
1991  *    c-2) if root is reloc root:	sync submit
1992  *         (only in case of buffered IO)
1993  *
1994  *    c-3) otherwise:			async submit
1995  */
1996 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1997 				 int mirror_num, unsigned long bio_flags,
1998 				 u64 bio_offset)
1999 {
2000 	struct inode *inode = private_data;
2001 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2002 	struct btrfs_root *root = BTRFS_I(inode)->root;
2003 	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2004 	blk_status_t ret = 0;
2005 	int skip_sum;
2006 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2007 
2008 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2009 
2010 	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2011 		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2012 
2013 	if (bio_op(bio) != REQ_OP_WRITE) {
2014 		ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2015 		if (ret)
2016 			goto out;
2017 
2018 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
2019 			ret = btrfs_submit_compressed_read(inode, bio,
2020 							   mirror_num,
2021 							   bio_flags);
2022 			goto out;
2023 		} else if (!skip_sum) {
2024 			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2025 			if (ret)
2026 				goto out;
2027 		}
2028 		goto mapit;
2029 	} else if (async && !skip_sum) {
2030 		/* csum items have already been cloned */
2031 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2032 			goto mapit;
2033 		/* we're doing a write, do the async checksumming */
2034 		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2035 					  bio_offset, inode,
2036 					  btrfs_submit_bio_start,
2037 					  btrfs_submit_bio_done);
2038 		goto out;
2039 	} else if (!skip_sum) {
2040 		ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2041 		if (ret)
2042 			goto out;
2043 	}
2044 
2045 mapit:
2046 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2047 
2048 out:
2049 	if (ret) {
2050 		bio->bi_status = ret;
2051 		bio_endio(bio);
2052 	}
2053 	return ret;
2054 }
2055 
2056 /*
2057  * given a list of ordered sums record them in the inode.  This happens
2058  * at IO completion time based on sums calculated at bio submission time.
2059  */
2060 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2061 			     struct inode *inode, struct list_head *list)
2062 {
2063 	struct btrfs_ordered_sum *sum;
2064 	int ret;
2065 
2066 	list_for_each_entry(sum, list, list) {
2067 		trans->adding_csums = true;
2068 		ret = btrfs_csum_file_blocks(trans,
2069 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
2070 		trans->adding_csums = false;
2071 		if (ret)
2072 			return ret;
2073 	}
2074 	return 0;
2075 }
2076 
2077 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2078 			      unsigned int extra_bits,
2079 			      struct extent_state **cached_state, int dedupe)
2080 {
2081 	WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2082 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2083 				   extra_bits, cached_state);
2084 }
2085 
2086 /* see btrfs_writepage_start_hook for details on why this is required */
2087 struct btrfs_writepage_fixup {
2088 	struct page *page;
2089 	struct btrfs_work work;
2090 };
2091 
2092 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2093 {
2094 	struct btrfs_writepage_fixup *fixup;
2095 	struct btrfs_ordered_extent *ordered;
2096 	struct extent_state *cached_state = NULL;
2097 	struct extent_changeset *data_reserved = NULL;
2098 	struct page *page;
2099 	struct inode *inode;
2100 	u64 page_start;
2101 	u64 page_end;
2102 	int ret;
2103 
2104 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
2105 	page = fixup->page;
2106 again:
2107 	lock_page(page);
2108 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2109 		ClearPageChecked(page);
2110 		goto out_page;
2111 	}
2112 
2113 	inode = page->mapping->host;
2114 	page_start = page_offset(page);
2115 	page_end = page_offset(page) + PAGE_SIZE - 1;
2116 
2117 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2118 			 &cached_state);
2119 
2120 	/* already ordered? We're done */
2121 	if (PagePrivate2(page))
2122 		goto out;
2123 
2124 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2125 					PAGE_SIZE);
2126 	if (ordered) {
2127 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2128 				     page_end, &cached_state);
2129 		unlock_page(page);
2130 		btrfs_start_ordered_extent(inode, ordered, 1);
2131 		btrfs_put_ordered_extent(ordered);
2132 		goto again;
2133 	}
2134 
2135 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2136 					   PAGE_SIZE);
2137 	if (ret) {
2138 		mapping_set_error(page->mapping, ret);
2139 		end_extent_writepage(page, ret, page_start, page_end);
2140 		ClearPageChecked(page);
2141 		goto out;
2142 	 }
2143 
2144 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2145 					&cached_state, 0);
2146 	if (ret) {
2147 		mapping_set_error(page->mapping, ret);
2148 		end_extent_writepage(page, ret, page_start, page_end);
2149 		ClearPageChecked(page);
2150 		goto out;
2151 	}
2152 
2153 	ClearPageChecked(page);
2154 	set_page_dirty(page);
2155 	btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2156 out:
2157 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2158 			     &cached_state);
2159 out_page:
2160 	unlock_page(page);
2161 	put_page(page);
2162 	kfree(fixup);
2163 	extent_changeset_free(data_reserved);
2164 }
2165 
2166 /*
2167  * There are a few paths in the higher layers of the kernel that directly
2168  * set the page dirty bit without asking the filesystem if it is a
2169  * good idea.  This causes problems because we want to make sure COW
2170  * properly happens and the data=ordered rules are followed.
2171  *
2172  * In our case any range that doesn't have the ORDERED bit set
2173  * hasn't been properly setup for IO.  We kick off an async process
2174  * to fix it up.  The async helper will wait for ordered extents, set
2175  * the delalloc bit and make it safe to write the page.
2176  */
2177 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2178 {
2179 	struct inode *inode = page->mapping->host;
2180 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2181 	struct btrfs_writepage_fixup *fixup;
2182 
2183 	/* this page is properly in the ordered list */
2184 	if (TestClearPagePrivate2(page))
2185 		return 0;
2186 
2187 	if (PageChecked(page))
2188 		return -EAGAIN;
2189 
2190 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2191 	if (!fixup)
2192 		return -EAGAIN;
2193 
2194 	SetPageChecked(page);
2195 	get_page(page);
2196 	btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2197 			btrfs_writepage_fixup_worker, NULL, NULL);
2198 	fixup->page = page;
2199 	btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2200 	return -EBUSY;
2201 }
2202 
2203 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2204 				       struct inode *inode, u64 file_pos,
2205 				       u64 disk_bytenr, u64 disk_num_bytes,
2206 				       u64 num_bytes, u64 ram_bytes,
2207 				       u8 compression, u8 encryption,
2208 				       u16 other_encoding, int extent_type)
2209 {
2210 	struct btrfs_root *root = BTRFS_I(inode)->root;
2211 	struct btrfs_file_extent_item *fi;
2212 	struct btrfs_path *path;
2213 	struct extent_buffer *leaf;
2214 	struct btrfs_key ins;
2215 	u64 qg_released;
2216 	int extent_inserted = 0;
2217 	int ret;
2218 
2219 	path = btrfs_alloc_path();
2220 	if (!path)
2221 		return -ENOMEM;
2222 
2223 	/*
2224 	 * we may be replacing one extent in the tree with another.
2225 	 * The new extent is pinned in the extent map, and we don't want
2226 	 * to drop it from the cache until it is completely in the btree.
2227 	 *
2228 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
2229 	 * the caller is expected to unpin it and allow it to be merged
2230 	 * with the others.
2231 	 */
2232 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2233 				   file_pos + num_bytes, NULL, 0,
2234 				   1, sizeof(*fi), &extent_inserted);
2235 	if (ret)
2236 		goto out;
2237 
2238 	if (!extent_inserted) {
2239 		ins.objectid = btrfs_ino(BTRFS_I(inode));
2240 		ins.offset = file_pos;
2241 		ins.type = BTRFS_EXTENT_DATA_KEY;
2242 
2243 		path->leave_spinning = 1;
2244 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
2245 					      sizeof(*fi));
2246 		if (ret)
2247 			goto out;
2248 	}
2249 	leaf = path->nodes[0];
2250 	fi = btrfs_item_ptr(leaf, path->slots[0],
2251 			    struct btrfs_file_extent_item);
2252 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2253 	btrfs_set_file_extent_type(leaf, fi, extent_type);
2254 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2255 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2256 	btrfs_set_file_extent_offset(leaf, fi, 0);
2257 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2258 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2259 	btrfs_set_file_extent_compression(leaf, fi, compression);
2260 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
2261 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2262 
2263 	btrfs_mark_buffer_dirty(leaf);
2264 	btrfs_release_path(path);
2265 
2266 	inode_add_bytes(inode, num_bytes);
2267 
2268 	ins.objectid = disk_bytenr;
2269 	ins.offset = disk_num_bytes;
2270 	ins.type = BTRFS_EXTENT_ITEM_KEY;
2271 
2272 	/*
2273 	 * Release the reserved range from inode dirty range map, as it is
2274 	 * already moved into delayed_ref_head
2275 	 */
2276 	ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2277 	if (ret < 0)
2278 		goto out;
2279 	qg_released = ret;
2280 	ret = btrfs_alloc_reserved_file_extent(trans, root,
2281 					       btrfs_ino(BTRFS_I(inode)),
2282 					       file_pos, qg_released, &ins);
2283 out:
2284 	btrfs_free_path(path);
2285 
2286 	return ret;
2287 }
2288 
2289 /* snapshot-aware defrag */
2290 struct sa_defrag_extent_backref {
2291 	struct rb_node node;
2292 	struct old_sa_defrag_extent *old;
2293 	u64 root_id;
2294 	u64 inum;
2295 	u64 file_pos;
2296 	u64 extent_offset;
2297 	u64 num_bytes;
2298 	u64 generation;
2299 };
2300 
2301 struct old_sa_defrag_extent {
2302 	struct list_head list;
2303 	struct new_sa_defrag_extent *new;
2304 
2305 	u64 extent_offset;
2306 	u64 bytenr;
2307 	u64 offset;
2308 	u64 len;
2309 	int count;
2310 };
2311 
2312 struct new_sa_defrag_extent {
2313 	struct rb_root root;
2314 	struct list_head head;
2315 	struct btrfs_path *path;
2316 	struct inode *inode;
2317 	u64 file_pos;
2318 	u64 len;
2319 	u64 bytenr;
2320 	u64 disk_len;
2321 	u8 compress_type;
2322 };
2323 
2324 static int backref_comp(struct sa_defrag_extent_backref *b1,
2325 			struct sa_defrag_extent_backref *b2)
2326 {
2327 	if (b1->root_id < b2->root_id)
2328 		return -1;
2329 	else if (b1->root_id > b2->root_id)
2330 		return 1;
2331 
2332 	if (b1->inum < b2->inum)
2333 		return -1;
2334 	else if (b1->inum > b2->inum)
2335 		return 1;
2336 
2337 	if (b1->file_pos < b2->file_pos)
2338 		return -1;
2339 	else if (b1->file_pos > b2->file_pos)
2340 		return 1;
2341 
2342 	/*
2343 	 * [------------------------------] ===> (a range of space)
2344 	 *     |<--->|   |<---->| =============> (fs/file tree A)
2345 	 * |<---------------------------->| ===> (fs/file tree B)
2346 	 *
2347 	 * A range of space can refer to two file extents in one tree while
2348 	 * refer to only one file extent in another tree.
2349 	 *
2350 	 * So we may process a disk offset more than one time(two extents in A)
2351 	 * and locate at the same extent(one extent in B), then insert two same
2352 	 * backrefs(both refer to the extent in B).
2353 	 */
2354 	return 0;
2355 }
2356 
2357 static void backref_insert(struct rb_root *root,
2358 			   struct sa_defrag_extent_backref *backref)
2359 {
2360 	struct rb_node **p = &root->rb_node;
2361 	struct rb_node *parent = NULL;
2362 	struct sa_defrag_extent_backref *entry;
2363 	int ret;
2364 
2365 	while (*p) {
2366 		parent = *p;
2367 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2368 
2369 		ret = backref_comp(backref, entry);
2370 		if (ret < 0)
2371 			p = &(*p)->rb_left;
2372 		else
2373 			p = &(*p)->rb_right;
2374 	}
2375 
2376 	rb_link_node(&backref->node, parent, p);
2377 	rb_insert_color(&backref->node, root);
2378 }
2379 
2380 /*
2381  * Note the backref might has changed, and in this case we just return 0.
2382  */
2383 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2384 				       void *ctx)
2385 {
2386 	struct btrfs_file_extent_item *extent;
2387 	struct old_sa_defrag_extent *old = ctx;
2388 	struct new_sa_defrag_extent *new = old->new;
2389 	struct btrfs_path *path = new->path;
2390 	struct btrfs_key key;
2391 	struct btrfs_root *root;
2392 	struct sa_defrag_extent_backref *backref;
2393 	struct extent_buffer *leaf;
2394 	struct inode *inode = new->inode;
2395 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2396 	int slot;
2397 	int ret;
2398 	u64 extent_offset;
2399 	u64 num_bytes;
2400 
2401 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2402 	    inum == btrfs_ino(BTRFS_I(inode)))
2403 		return 0;
2404 
2405 	key.objectid = root_id;
2406 	key.type = BTRFS_ROOT_ITEM_KEY;
2407 	key.offset = (u64)-1;
2408 
2409 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2410 	if (IS_ERR(root)) {
2411 		if (PTR_ERR(root) == -ENOENT)
2412 			return 0;
2413 		WARN_ON(1);
2414 		btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2415 			 inum, offset, root_id);
2416 		return PTR_ERR(root);
2417 	}
2418 
2419 	key.objectid = inum;
2420 	key.type = BTRFS_EXTENT_DATA_KEY;
2421 	if (offset > (u64)-1 << 32)
2422 		key.offset = 0;
2423 	else
2424 		key.offset = offset;
2425 
2426 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2427 	if (WARN_ON(ret < 0))
2428 		return ret;
2429 	ret = 0;
2430 
2431 	while (1) {
2432 		cond_resched();
2433 
2434 		leaf = path->nodes[0];
2435 		slot = path->slots[0];
2436 
2437 		if (slot >= btrfs_header_nritems(leaf)) {
2438 			ret = btrfs_next_leaf(root, path);
2439 			if (ret < 0) {
2440 				goto out;
2441 			} else if (ret > 0) {
2442 				ret = 0;
2443 				goto out;
2444 			}
2445 			continue;
2446 		}
2447 
2448 		path->slots[0]++;
2449 
2450 		btrfs_item_key_to_cpu(leaf, &key, slot);
2451 
2452 		if (key.objectid > inum)
2453 			goto out;
2454 
2455 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2456 			continue;
2457 
2458 		extent = btrfs_item_ptr(leaf, slot,
2459 					struct btrfs_file_extent_item);
2460 
2461 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2462 			continue;
2463 
2464 		/*
2465 		 * 'offset' refers to the exact key.offset,
2466 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2467 		 * (key.offset - extent_offset).
2468 		 */
2469 		if (key.offset != offset)
2470 			continue;
2471 
2472 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2473 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2474 
2475 		if (extent_offset >= old->extent_offset + old->offset +
2476 		    old->len || extent_offset + num_bytes <=
2477 		    old->extent_offset + old->offset)
2478 			continue;
2479 		break;
2480 	}
2481 
2482 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2483 	if (!backref) {
2484 		ret = -ENOENT;
2485 		goto out;
2486 	}
2487 
2488 	backref->root_id = root_id;
2489 	backref->inum = inum;
2490 	backref->file_pos = offset;
2491 	backref->num_bytes = num_bytes;
2492 	backref->extent_offset = extent_offset;
2493 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2494 	backref->old = old;
2495 	backref_insert(&new->root, backref);
2496 	old->count++;
2497 out:
2498 	btrfs_release_path(path);
2499 	WARN_ON(ret);
2500 	return ret;
2501 }
2502 
2503 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2504 				   struct new_sa_defrag_extent *new)
2505 {
2506 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2507 	struct old_sa_defrag_extent *old, *tmp;
2508 	int ret;
2509 
2510 	new->path = path;
2511 
2512 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2513 		ret = iterate_inodes_from_logical(old->bytenr +
2514 						  old->extent_offset, fs_info,
2515 						  path, record_one_backref,
2516 						  old, false);
2517 		if (ret < 0 && ret != -ENOENT)
2518 			return false;
2519 
2520 		/* no backref to be processed for this extent */
2521 		if (!old->count) {
2522 			list_del(&old->list);
2523 			kfree(old);
2524 		}
2525 	}
2526 
2527 	if (list_empty(&new->head))
2528 		return false;
2529 
2530 	return true;
2531 }
2532 
2533 static int relink_is_mergable(struct extent_buffer *leaf,
2534 			      struct btrfs_file_extent_item *fi,
2535 			      struct new_sa_defrag_extent *new)
2536 {
2537 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2538 		return 0;
2539 
2540 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2541 		return 0;
2542 
2543 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2544 		return 0;
2545 
2546 	if (btrfs_file_extent_encryption(leaf, fi) ||
2547 	    btrfs_file_extent_other_encoding(leaf, fi))
2548 		return 0;
2549 
2550 	return 1;
2551 }
2552 
2553 /*
2554  * Note the backref might has changed, and in this case we just return 0.
2555  */
2556 static noinline int relink_extent_backref(struct btrfs_path *path,
2557 				 struct sa_defrag_extent_backref *prev,
2558 				 struct sa_defrag_extent_backref *backref)
2559 {
2560 	struct btrfs_file_extent_item *extent;
2561 	struct btrfs_file_extent_item *item;
2562 	struct btrfs_ordered_extent *ordered;
2563 	struct btrfs_trans_handle *trans;
2564 	struct btrfs_root *root;
2565 	struct btrfs_key key;
2566 	struct extent_buffer *leaf;
2567 	struct old_sa_defrag_extent *old = backref->old;
2568 	struct new_sa_defrag_extent *new = old->new;
2569 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2570 	struct inode *inode;
2571 	struct extent_state *cached = NULL;
2572 	int ret = 0;
2573 	u64 start;
2574 	u64 len;
2575 	u64 lock_start;
2576 	u64 lock_end;
2577 	bool merge = false;
2578 	int index;
2579 
2580 	if (prev && prev->root_id == backref->root_id &&
2581 	    prev->inum == backref->inum &&
2582 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2583 		merge = true;
2584 
2585 	/* step 1: get root */
2586 	key.objectid = backref->root_id;
2587 	key.type = BTRFS_ROOT_ITEM_KEY;
2588 	key.offset = (u64)-1;
2589 
2590 	index = srcu_read_lock(&fs_info->subvol_srcu);
2591 
2592 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2593 	if (IS_ERR(root)) {
2594 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2595 		if (PTR_ERR(root) == -ENOENT)
2596 			return 0;
2597 		return PTR_ERR(root);
2598 	}
2599 
2600 	if (btrfs_root_readonly(root)) {
2601 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2602 		return 0;
2603 	}
2604 
2605 	/* step 2: get inode */
2606 	key.objectid = backref->inum;
2607 	key.type = BTRFS_INODE_ITEM_KEY;
2608 	key.offset = 0;
2609 
2610 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2611 	if (IS_ERR(inode)) {
2612 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2613 		return 0;
2614 	}
2615 
2616 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2617 
2618 	/* step 3: relink backref */
2619 	lock_start = backref->file_pos;
2620 	lock_end = backref->file_pos + backref->num_bytes - 1;
2621 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2622 			 &cached);
2623 
2624 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2625 	if (ordered) {
2626 		btrfs_put_ordered_extent(ordered);
2627 		goto out_unlock;
2628 	}
2629 
2630 	trans = btrfs_join_transaction(root);
2631 	if (IS_ERR(trans)) {
2632 		ret = PTR_ERR(trans);
2633 		goto out_unlock;
2634 	}
2635 
2636 	key.objectid = backref->inum;
2637 	key.type = BTRFS_EXTENT_DATA_KEY;
2638 	key.offset = backref->file_pos;
2639 
2640 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2641 	if (ret < 0) {
2642 		goto out_free_path;
2643 	} else if (ret > 0) {
2644 		ret = 0;
2645 		goto out_free_path;
2646 	}
2647 
2648 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2649 				struct btrfs_file_extent_item);
2650 
2651 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2652 	    backref->generation)
2653 		goto out_free_path;
2654 
2655 	btrfs_release_path(path);
2656 
2657 	start = backref->file_pos;
2658 	if (backref->extent_offset < old->extent_offset + old->offset)
2659 		start += old->extent_offset + old->offset -
2660 			 backref->extent_offset;
2661 
2662 	len = min(backref->extent_offset + backref->num_bytes,
2663 		  old->extent_offset + old->offset + old->len);
2664 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2665 
2666 	ret = btrfs_drop_extents(trans, root, inode, start,
2667 				 start + len, 1);
2668 	if (ret)
2669 		goto out_free_path;
2670 again:
2671 	key.objectid = btrfs_ino(BTRFS_I(inode));
2672 	key.type = BTRFS_EXTENT_DATA_KEY;
2673 	key.offset = start;
2674 
2675 	path->leave_spinning = 1;
2676 	if (merge) {
2677 		struct btrfs_file_extent_item *fi;
2678 		u64 extent_len;
2679 		struct btrfs_key found_key;
2680 
2681 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2682 		if (ret < 0)
2683 			goto out_free_path;
2684 
2685 		path->slots[0]--;
2686 		leaf = path->nodes[0];
2687 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2688 
2689 		fi = btrfs_item_ptr(leaf, path->slots[0],
2690 				    struct btrfs_file_extent_item);
2691 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2692 
2693 		if (extent_len + found_key.offset == start &&
2694 		    relink_is_mergable(leaf, fi, new)) {
2695 			btrfs_set_file_extent_num_bytes(leaf, fi,
2696 							extent_len + len);
2697 			btrfs_mark_buffer_dirty(leaf);
2698 			inode_add_bytes(inode, len);
2699 
2700 			ret = 1;
2701 			goto out_free_path;
2702 		} else {
2703 			merge = false;
2704 			btrfs_release_path(path);
2705 			goto again;
2706 		}
2707 	}
2708 
2709 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2710 					sizeof(*extent));
2711 	if (ret) {
2712 		btrfs_abort_transaction(trans, ret);
2713 		goto out_free_path;
2714 	}
2715 
2716 	leaf = path->nodes[0];
2717 	item = btrfs_item_ptr(leaf, path->slots[0],
2718 				struct btrfs_file_extent_item);
2719 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2720 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2721 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2722 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2723 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2724 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2725 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2726 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2727 	btrfs_set_file_extent_encryption(leaf, item, 0);
2728 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2729 
2730 	btrfs_mark_buffer_dirty(leaf);
2731 	inode_add_bytes(inode, len);
2732 	btrfs_release_path(path);
2733 
2734 	ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2735 			new->disk_len, 0,
2736 			backref->root_id, backref->inum,
2737 			new->file_pos);	/* start - extent_offset */
2738 	if (ret) {
2739 		btrfs_abort_transaction(trans, ret);
2740 		goto out_free_path;
2741 	}
2742 
2743 	ret = 1;
2744 out_free_path:
2745 	btrfs_release_path(path);
2746 	path->leave_spinning = 0;
2747 	btrfs_end_transaction(trans);
2748 out_unlock:
2749 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2750 			     &cached);
2751 	iput(inode);
2752 	return ret;
2753 }
2754 
2755 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2756 {
2757 	struct old_sa_defrag_extent *old, *tmp;
2758 
2759 	if (!new)
2760 		return;
2761 
2762 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2763 		kfree(old);
2764 	}
2765 	kfree(new);
2766 }
2767 
2768 static void relink_file_extents(struct new_sa_defrag_extent *new)
2769 {
2770 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2771 	struct btrfs_path *path;
2772 	struct sa_defrag_extent_backref *backref;
2773 	struct sa_defrag_extent_backref *prev = NULL;
2774 	struct inode *inode;
2775 	struct rb_node *node;
2776 	int ret;
2777 
2778 	inode = new->inode;
2779 
2780 	path = btrfs_alloc_path();
2781 	if (!path)
2782 		return;
2783 
2784 	if (!record_extent_backrefs(path, new)) {
2785 		btrfs_free_path(path);
2786 		goto out;
2787 	}
2788 	btrfs_release_path(path);
2789 
2790 	while (1) {
2791 		node = rb_first(&new->root);
2792 		if (!node)
2793 			break;
2794 		rb_erase(node, &new->root);
2795 
2796 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2797 
2798 		ret = relink_extent_backref(path, prev, backref);
2799 		WARN_ON(ret < 0);
2800 
2801 		kfree(prev);
2802 
2803 		if (ret == 1)
2804 			prev = backref;
2805 		else
2806 			prev = NULL;
2807 		cond_resched();
2808 	}
2809 	kfree(prev);
2810 
2811 	btrfs_free_path(path);
2812 out:
2813 	free_sa_defrag_extent(new);
2814 
2815 	atomic_dec(&fs_info->defrag_running);
2816 	wake_up(&fs_info->transaction_wait);
2817 }
2818 
2819 static struct new_sa_defrag_extent *
2820 record_old_file_extents(struct inode *inode,
2821 			struct btrfs_ordered_extent *ordered)
2822 {
2823 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2824 	struct btrfs_root *root = BTRFS_I(inode)->root;
2825 	struct btrfs_path *path;
2826 	struct btrfs_key key;
2827 	struct old_sa_defrag_extent *old;
2828 	struct new_sa_defrag_extent *new;
2829 	int ret;
2830 
2831 	new = kmalloc(sizeof(*new), GFP_NOFS);
2832 	if (!new)
2833 		return NULL;
2834 
2835 	new->inode = inode;
2836 	new->file_pos = ordered->file_offset;
2837 	new->len = ordered->len;
2838 	new->bytenr = ordered->start;
2839 	new->disk_len = ordered->disk_len;
2840 	new->compress_type = ordered->compress_type;
2841 	new->root = RB_ROOT;
2842 	INIT_LIST_HEAD(&new->head);
2843 
2844 	path = btrfs_alloc_path();
2845 	if (!path)
2846 		goto out_kfree;
2847 
2848 	key.objectid = btrfs_ino(BTRFS_I(inode));
2849 	key.type = BTRFS_EXTENT_DATA_KEY;
2850 	key.offset = new->file_pos;
2851 
2852 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2853 	if (ret < 0)
2854 		goto out_free_path;
2855 	if (ret > 0 && path->slots[0] > 0)
2856 		path->slots[0]--;
2857 
2858 	/* find out all the old extents for the file range */
2859 	while (1) {
2860 		struct btrfs_file_extent_item *extent;
2861 		struct extent_buffer *l;
2862 		int slot;
2863 		u64 num_bytes;
2864 		u64 offset;
2865 		u64 end;
2866 		u64 disk_bytenr;
2867 		u64 extent_offset;
2868 
2869 		l = path->nodes[0];
2870 		slot = path->slots[0];
2871 
2872 		if (slot >= btrfs_header_nritems(l)) {
2873 			ret = btrfs_next_leaf(root, path);
2874 			if (ret < 0)
2875 				goto out_free_path;
2876 			else if (ret > 0)
2877 				break;
2878 			continue;
2879 		}
2880 
2881 		btrfs_item_key_to_cpu(l, &key, slot);
2882 
2883 		if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2884 			break;
2885 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2886 			break;
2887 		if (key.offset >= new->file_pos + new->len)
2888 			break;
2889 
2890 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2891 
2892 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2893 		if (key.offset + num_bytes < new->file_pos)
2894 			goto next;
2895 
2896 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2897 		if (!disk_bytenr)
2898 			goto next;
2899 
2900 		extent_offset = btrfs_file_extent_offset(l, extent);
2901 
2902 		old = kmalloc(sizeof(*old), GFP_NOFS);
2903 		if (!old)
2904 			goto out_free_path;
2905 
2906 		offset = max(new->file_pos, key.offset);
2907 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2908 
2909 		old->bytenr = disk_bytenr;
2910 		old->extent_offset = extent_offset;
2911 		old->offset = offset - key.offset;
2912 		old->len = end - offset;
2913 		old->new = new;
2914 		old->count = 0;
2915 		list_add_tail(&old->list, &new->head);
2916 next:
2917 		path->slots[0]++;
2918 		cond_resched();
2919 	}
2920 
2921 	btrfs_free_path(path);
2922 	atomic_inc(&fs_info->defrag_running);
2923 
2924 	return new;
2925 
2926 out_free_path:
2927 	btrfs_free_path(path);
2928 out_kfree:
2929 	free_sa_defrag_extent(new);
2930 	return NULL;
2931 }
2932 
2933 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2934 					 u64 start, u64 len)
2935 {
2936 	struct btrfs_block_group_cache *cache;
2937 
2938 	cache = btrfs_lookup_block_group(fs_info, start);
2939 	ASSERT(cache);
2940 
2941 	spin_lock(&cache->lock);
2942 	cache->delalloc_bytes -= len;
2943 	spin_unlock(&cache->lock);
2944 
2945 	btrfs_put_block_group(cache);
2946 }
2947 
2948 /* as ordered data IO finishes, this gets called so we can finish
2949  * an ordered extent if the range of bytes in the file it covers are
2950  * fully written.
2951  */
2952 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2953 {
2954 	struct inode *inode = ordered_extent->inode;
2955 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2956 	struct btrfs_root *root = BTRFS_I(inode)->root;
2957 	struct btrfs_trans_handle *trans = NULL;
2958 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2959 	struct extent_state *cached_state = NULL;
2960 	struct new_sa_defrag_extent *new = NULL;
2961 	int compress_type = 0;
2962 	int ret = 0;
2963 	u64 logical_len = ordered_extent->len;
2964 	bool nolock;
2965 	bool truncated = false;
2966 	bool range_locked = false;
2967 	bool clear_new_delalloc_bytes = false;
2968 
2969 	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2970 	    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2971 	    !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2972 		clear_new_delalloc_bytes = true;
2973 
2974 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2975 
2976 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2977 		ret = -EIO;
2978 		goto out;
2979 	}
2980 
2981 	btrfs_free_io_failure_record(BTRFS_I(inode),
2982 			ordered_extent->file_offset,
2983 			ordered_extent->file_offset +
2984 			ordered_extent->len - 1);
2985 
2986 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2987 		truncated = true;
2988 		logical_len = ordered_extent->truncated_len;
2989 		/* Truncated the entire extent, don't bother adding */
2990 		if (!logical_len)
2991 			goto out;
2992 	}
2993 
2994 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2995 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2996 
2997 		/*
2998 		 * For mwrite(mmap + memset to write) case, we still reserve
2999 		 * space for NOCOW range.
3000 		 * As NOCOW won't cause a new delayed ref, just free the space
3001 		 */
3002 		btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3003 				       ordered_extent->len);
3004 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3005 		if (nolock)
3006 			trans = btrfs_join_transaction_nolock(root);
3007 		else
3008 			trans = btrfs_join_transaction(root);
3009 		if (IS_ERR(trans)) {
3010 			ret = PTR_ERR(trans);
3011 			trans = NULL;
3012 			goto out;
3013 		}
3014 		trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3015 		ret = btrfs_update_inode_fallback(trans, root, inode);
3016 		if (ret) /* -ENOMEM or corruption */
3017 			btrfs_abort_transaction(trans, ret);
3018 		goto out;
3019 	}
3020 
3021 	range_locked = true;
3022 	lock_extent_bits(io_tree, ordered_extent->file_offset,
3023 			 ordered_extent->file_offset + ordered_extent->len - 1,
3024 			 &cached_state);
3025 
3026 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
3027 			ordered_extent->file_offset + ordered_extent->len - 1,
3028 			EXTENT_DEFRAG, 0, cached_state);
3029 	if (ret) {
3030 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3031 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3032 			/* the inode is shared */
3033 			new = record_old_file_extents(inode, ordered_extent);
3034 
3035 		clear_extent_bit(io_tree, ordered_extent->file_offset,
3036 			ordered_extent->file_offset + ordered_extent->len - 1,
3037 			EXTENT_DEFRAG, 0, 0, &cached_state);
3038 	}
3039 
3040 	if (nolock)
3041 		trans = btrfs_join_transaction_nolock(root);
3042 	else
3043 		trans = btrfs_join_transaction(root);
3044 	if (IS_ERR(trans)) {
3045 		ret = PTR_ERR(trans);
3046 		trans = NULL;
3047 		goto out;
3048 	}
3049 
3050 	trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3051 
3052 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3053 		compress_type = ordered_extent->compress_type;
3054 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3055 		BUG_ON(compress_type);
3056 		btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3057 				       ordered_extent->len);
3058 		ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3059 						ordered_extent->file_offset,
3060 						ordered_extent->file_offset +
3061 						logical_len);
3062 	} else {
3063 		BUG_ON(root == fs_info->tree_root);
3064 		ret = insert_reserved_file_extent(trans, inode,
3065 						ordered_extent->file_offset,
3066 						ordered_extent->start,
3067 						ordered_extent->disk_len,
3068 						logical_len, logical_len,
3069 						compress_type, 0, 0,
3070 						BTRFS_FILE_EXTENT_REG);
3071 		if (!ret)
3072 			btrfs_release_delalloc_bytes(fs_info,
3073 						     ordered_extent->start,
3074 						     ordered_extent->disk_len);
3075 	}
3076 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3077 			   ordered_extent->file_offset, ordered_extent->len,
3078 			   trans->transid);
3079 	if (ret < 0) {
3080 		btrfs_abort_transaction(trans, ret);
3081 		goto out;
3082 	}
3083 
3084 	ret = add_pending_csums(trans, inode, &ordered_extent->list);
3085 	if (ret) {
3086 		btrfs_abort_transaction(trans, ret);
3087 		goto out;
3088 	}
3089 
3090 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3091 	ret = btrfs_update_inode_fallback(trans, root, inode);
3092 	if (ret) { /* -ENOMEM or corruption */
3093 		btrfs_abort_transaction(trans, ret);
3094 		goto out;
3095 	}
3096 	ret = 0;
3097 out:
3098 	if (range_locked || clear_new_delalloc_bytes) {
3099 		unsigned int clear_bits = 0;
3100 
3101 		if (range_locked)
3102 			clear_bits |= EXTENT_LOCKED;
3103 		if (clear_new_delalloc_bytes)
3104 			clear_bits |= EXTENT_DELALLOC_NEW;
3105 		clear_extent_bit(&BTRFS_I(inode)->io_tree,
3106 				 ordered_extent->file_offset,
3107 				 ordered_extent->file_offset +
3108 				 ordered_extent->len - 1,
3109 				 clear_bits,
3110 				 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3111 				 0, &cached_state);
3112 	}
3113 
3114 	if (trans)
3115 		btrfs_end_transaction(trans);
3116 
3117 	if (ret || truncated) {
3118 		u64 start, end;
3119 
3120 		if (truncated)
3121 			start = ordered_extent->file_offset + logical_len;
3122 		else
3123 			start = ordered_extent->file_offset;
3124 		end = ordered_extent->file_offset + ordered_extent->len - 1;
3125 		clear_extent_uptodate(io_tree, start, end, NULL);
3126 
3127 		/* Drop the cache for the part of the extent we didn't write. */
3128 		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3129 
3130 		/*
3131 		 * If the ordered extent had an IOERR or something else went
3132 		 * wrong we need to return the space for this ordered extent
3133 		 * back to the allocator.  We only free the extent in the
3134 		 * truncated case if we didn't write out the extent at all.
3135 		 */
3136 		if ((ret || !logical_len) &&
3137 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3138 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3139 			btrfs_free_reserved_extent(fs_info,
3140 						   ordered_extent->start,
3141 						   ordered_extent->disk_len, 1);
3142 	}
3143 
3144 
3145 	/*
3146 	 * This needs to be done to make sure anybody waiting knows we are done
3147 	 * updating everything for this ordered extent.
3148 	 */
3149 	btrfs_remove_ordered_extent(inode, ordered_extent);
3150 
3151 	/* for snapshot-aware defrag */
3152 	if (new) {
3153 		if (ret) {
3154 			free_sa_defrag_extent(new);
3155 			atomic_dec(&fs_info->defrag_running);
3156 		} else {
3157 			relink_file_extents(new);
3158 		}
3159 	}
3160 
3161 	/* once for us */
3162 	btrfs_put_ordered_extent(ordered_extent);
3163 	/* once for the tree */
3164 	btrfs_put_ordered_extent(ordered_extent);
3165 
3166 	return ret;
3167 }
3168 
3169 static void finish_ordered_fn(struct btrfs_work *work)
3170 {
3171 	struct btrfs_ordered_extent *ordered_extent;
3172 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3173 	btrfs_finish_ordered_io(ordered_extent);
3174 }
3175 
3176 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3177 				struct extent_state *state, int uptodate)
3178 {
3179 	struct inode *inode = page->mapping->host;
3180 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3181 	struct btrfs_ordered_extent *ordered_extent = NULL;
3182 	struct btrfs_workqueue *wq;
3183 	btrfs_work_func_t func;
3184 
3185 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3186 
3187 	ClearPagePrivate2(page);
3188 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3189 					    end - start + 1, uptodate))
3190 		return;
3191 
3192 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3193 		wq = fs_info->endio_freespace_worker;
3194 		func = btrfs_freespace_write_helper;
3195 	} else {
3196 		wq = fs_info->endio_write_workers;
3197 		func = btrfs_endio_write_helper;
3198 	}
3199 
3200 	btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3201 			NULL);
3202 	btrfs_queue_work(wq, &ordered_extent->work);
3203 }
3204 
3205 static int __readpage_endio_check(struct inode *inode,
3206 				  struct btrfs_io_bio *io_bio,
3207 				  int icsum, struct page *page,
3208 				  int pgoff, u64 start, size_t len)
3209 {
3210 	char *kaddr;
3211 	u32 csum_expected;
3212 	u32 csum = ~(u32)0;
3213 
3214 	csum_expected = *(((u32 *)io_bio->csum) + icsum);
3215 
3216 	kaddr = kmap_atomic(page);
3217 	csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3218 	btrfs_csum_final(csum, (u8 *)&csum);
3219 	if (csum != csum_expected)
3220 		goto zeroit;
3221 
3222 	kunmap_atomic(kaddr);
3223 	return 0;
3224 zeroit:
3225 	btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3226 				    io_bio->mirror_num);
3227 	memset(kaddr + pgoff, 1, len);
3228 	flush_dcache_page(page);
3229 	kunmap_atomic(kaddr);
3230 	return -EIO;
3231 }
3232 
3233 /*
3234  * when reads are done, we need to check csums to verify the data is correct
3235  * if there's a match, we allow the bio to finish.  If not, the code in
3236  * extent_io.c will try to find good copies for us.
3237  */
3238 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3239 				      u64 phy_offset, struct page *page,
3240 				      u64 start, u64 end, int mirror)
3241 {
3242 	size_t offset = start - page_offset(page);
3243 	struct inode *inode = page->mapping->host;
3244 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3245 	struct btrfs_root *root = BTRFS_I(inode)->root;
3246 
3247 	if (PageChecked(page)) {
3248 		ClearPageChecked(page);
3249 		return 0;
3250 	}
3251 
3252 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3253 		return 0;
3254 
3255 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3256 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3257 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3258 		return 0;
3259 	}
3260 
3261 	phy_offset >>= inode->i_sb->s_blocksize_bits;
3262 	return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3263 				      start, (size_t)(end - start + 1));
3264 }
3265 
3266 /*
3267  * btrfs_add_delayed_iput - perform a delayed iput on @inode
3268  *
3269  * @inode: The inode we want to perform iput on
3270  *
3271  * This function uses the generic vfs_inode::i_count to track whether we should
3272  * just decrement it (in case it's > 1) or if this is the last iput then link
3273  * the inode to the delayed iput machinery. Delayed iputs are processed at
3274  * transaction commit time/superblock commit/cleaner kthread.
3275  */
3276 void btrfs_add_delayed_iput(struct inode *inode)
3277 {
3278 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3279 	struct btrfs_inode *binode = BTRFS_I(inode);
3280 
3281 	if (atomic_add_unless(&inode->i_count, -1, 1))
3282 		return;
3283 
3284 	spin_lock(&fs_info->delayed_iput_lock);
3285 	ASSERT(list_empty(&binode->delayed_iput));
3286 	list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3287 	spin_unlock(&fs_info->delayed_iput_lock);
3288 }
3289 
3290 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3291 {
3292 
3293 	spin_lock(&fs_info->delayed_iput_lock);
3294 	while (!list_empty(&fs_info->delayed_iputs)) {
3295 		struct btrfs_inode *inode;
3296 
3297 		inode = list_first_entry(&fs_info->delayed_iputs,
3298 				struct btrfs_inode, delayed_iput);
3299 		list_del_init(&inode->delayed_iput);
3300 		spin_unlock(&fs_info->delayed_iput_lock);
3301 		iput(&inode->vfs_inode);
3302 		spin_lock(&fs_info->delayed_iput_lock);
3303 	}
3304 	spin_unlock(&fs_info->delayed_iput_lock);
3305 }
3306 
3307 /*
3308  * This is called in transaction commit time. If there are no orphan
3309  * files in the subvolume, it removes orphan item and frees block_rsv
3310  * structure.
3311  */
3312 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3313 			      struct btrfs_root *root)
3314 {
3315 	struct btrfs_fs_info *fs_info = root->fs_info;
3316 	struct btrfs_block_rsv *block_rsv;
3317 	int ret;
3318 
3319 	if (atomic_read(&root->orphan_inodes) ||
3320 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3321 		return;
3322 
3323 	spin_lock(&root->orphan_lock);
3324 	if (atomic_read(&root->orphan_inodes)) {
3325 		spin_unlock(&root->orphan_lock);
3326 		return;
3327 	}
3328 
3329 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3330 		spin_unlock(&root->orphan_lock);
3331 		return;
3332 	}
3333 
3334 	block_rsv = root->orphan_block_rsv;
3335 	root->orphan_block_rsv = NULL;
3336 	spin_unlock(&root->orphan_lock);
3337 
3338 	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3339 	    btrfs_root_refs(&root->root_item) > 0) {
3340 		ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3341 					    root->root_key.objectid);
3342 		if (ret)
3343 			btrfs_abort_transaction(trans, ret);
3344 		else
3345 			clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3346 				  &root->state);
3347 	}
3348 
3349 	if (block_rsv) {
3350 		WARN_ON(block_rsv->size > 0);
3351 		btrfs_free_block_rsv(fs_info, block_rsv);
3352 	}
3353 }
3354 
3355 /*
3356  * This creates an orphan entry for the given inode in case something goes
3357  * wrong in the middle of an unlink/truncate.
3358  *
3359  * NOTE: caller of this function should reserve 5 units of metadata for
3360  *	 this function.
3361  */
3362 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3363 		struct btrfs_inode *inode)
3364 {
3365 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3366 	struct btrfs_root *root = inode->root;
3367 	struct btrfs_block_rsv *block_rsv = NULL;
3368 	int reserve = 0;
3369 	bool insert = false;
3370 	int ret;
3371 
3372 	if (!root->orphan_block_rsv) {
3373 		block_rsv = btrfs_alloc_block_rsv(fs_info,
3374 						  BTRFS_BLOCK_RSV_TEMP);
3375 		if (!block_rsv)
3376 			return -ENOMEM;
3377 	}
3378 
3379 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3380 			      &inode->runtime_flags))
3381 		insert = true;
3382 
3383 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3384 			      &inode->runtime_flags))
3385 		reserve = 1;
3386 
3387 	spin_lock(&root->orphan_lock);
3388 	/* If someone has created ->orphan_block_rsv, be happy to use it. */
3389 	if (!root->orphan_block_rsv) {
3390 		root->orphan_block_rsv = block_rsv;
3391 	} else if (block_rsv) {
3392 		btrfs_free_block_rsv(fs_info, block_rsv);
3393 		block_rsv = NULL;
3394 	}
3395 
3396 	if (insert)
3397 		atomic_inc(&root->orphan_inodes);
3398 	spin_unlock(&root->orphan_lock);
3399 
3400 	/* grab metadata reservation from transaction handle */
3401 	if (reserve) {
3402 		ret = btrfs_orphan_reserve_metadata(trans, inode);
3403 		ASSERT(!ret);
3404 		if (ret) {
3405 			/*
3406 			 * dec doesn't need spin_lock as ->orphan_block_rsv
3407 			 * would be released only if ->orphan_inodes is
3408 			 * zero.
3409 			 */
3410 			atomic_dec(&root->orphan_inodes);
3411 			clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3412 				  &inode->runtime_flags);
3413 			if (insert)
3414 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3415 					  &inode->runtime_flags);
3416 			return ret;
3417 		}
3418 	}
3419 
3420 	/* insert an orphan item to track this unlinked/truncated file */
3421 	if (insert) {
3422 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3423 		if (ret) {
3424 			if (reserve) {
3425 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3426 					  &inode->runtime_flags);
3427 				btrfs_orphan_release_metadata(inode);
3428 			}
3429 			/*
3430 			 * btrfs_orphan_commit_root may race with us and set
3431 			 * ->orphan_block_rsv to zero, in order to avoid that,
3432 			 * decrease ->orphan_inodes after everything is done.
3433 			 */
3434 			atomic_dec(&root->orphan_inodes);
3435 			if (ret != -EEXIST) {
3436 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3437 					  &inode->runtime_flags);
3438 				btrfs_abort_transaction(trans, ret);
3439 				return ret;
3440 			}
3441 		}
3442 		ret = 0;
3443 	}
3444 
3445 	return 0;
3446 }
3447 
3448 /*
3449  * We have done the truncate/delete so we can go ahead and remove the orphan
3450  * item for this particular inode.
3451  */
3452 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3453 			    struct btrfs_inode *inode)
3454 {
3455 	struct btrfs_root *root = inode->root;
3456 	int delete_item = 0;
3457 	int ret = 0;
3458 
3459 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3460 			       &inode->runtime_flags))
3461 		delete_item = 1;
3462 
3463 	if (delete_item && trans)
3464 		ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3465 
3466 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3467 			       &inode->runtime_flags))
3468 		btrfs_orphan_release_metadata(inode);
3469 
3470 	/*
3471 	 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3472 	 * to zero, in order to avoid that, decrease ->orphan_inodes after
3473 	 * everything is done.
3474 	 */
3475 	if (delete_item)
3476 		atomic_dec(&root->orphan_inodes);
3477 
3478 	return ret;
3479 }
3480 
3481 /*
3482  * this cleans up any orphans that may be left on the list from the last use
3483  * of this root.
3484  */
3485 int btrfs_orphan_cleanup(struct btrfs_root *root)
3486 {
3487 	struct btrfs_fs_info *fs_info = root->fs_info;
3488 	struct btrfs_path *path;
3489 	struct extent_buffer *leaf;
3490 	struct btrfs_key key, found_key;
3491 	struct btrfs_trans_handle *trans;
3492 	struct inode *inode;
3493 	u64 last_objectid = 0;
3494 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3495 
3496 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3497 		return 0;
3498 
3499 	path = btrfs_alloc_path();
3500 	if (!path) {
3501 		ret = -ENOMEM;
3502 		goto out;
3503 	}
3504 	path->reada = READA_BACK;
3505 
3506 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3507 	key.type = BTRFS_ORPHAN_ITEM_KEY;
3508 	key.offset = (u64)-1;
3509 
3510 	while (1) {
3511 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3512 		if (ret < 0)
3513 			goto out;
3514 
3515 		/*
3516 		 * if ret == 0 means we found what we were searching for, which
3517 		 * is weird, but possible, so only screw with path if we didn't
3518 		 * find the key and see if we have stuff that matches
3519 		 */
3520 		if (ret > 0) {
3521 			ret = 0;
3522 			if (path->slots[0] == 0)
3523 				break;
3524 			path->slots[0]--;
3525 		}
3526 
3527 		/* pull out the item */
3528 		leaf = path->nodes[0];
3529 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3530 
3531 		/* make sure the item matches what we want */
3532 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3533 			break;
3534 		if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3535 			break;
3536 
3537 		/* release the path since we're done with it */
3538 		btrfs_release_path(path);
3539 
3540 		/*
3541 		 * this is where we are basically btrfs_lookup, without the
3542 		 * crossing root thing.  we store the inode number in the
3543 		 * offset of the orphan item.
3544 		 */
3545 
3546 		if (found_key.offset == last_objectid) {
3547 			btrfs_err(fs_info,
3548 				  "Error removing orphan entry, stopping orphan cleanup");
3549 			ret = -EINVAL;
3550 			goto out;
3551 		}
3552 
3553 		last_objectid = found_key.offset;
3554 
3555 		found_key.objectid = found_key.offset;
3556 		found_key.type = BTRFS_INODE_ITEM_KEY;
3557 		found_key.offset = 0;
3558 		inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3559 		ret = PTR_ERR_OR_ZERO(inode);
3560 		if (ret && ret != -ENOENT)
3561 			goto out;
3562 
3563 		if (ret == -ENOENT && root == fs_info->tree_root) {
3564 			struct btrfs_root *dead_root;
3565 			struct btrfs_fs_info *fs_info = root->fs_info;
3566 			int is_dead_root = 0;
3567 
3568 			/*
3569 			 * this is an orphan in the tree root. Currently these
3570 			 * could come from 2 sources:
3571 			 *  a) a snapshot deletion in progress
3572 			 *  b) a free space cache inode
3573 			 * We need to distinguish those two, as the snapshot
3574 			 * orphan must not get deleted.
3575 			 * find_dead_roots already ran before us, so if this
3576 			 * is a snapshot deletion, we should find the root
3577 			 * in the dead_roots list
3578 			 */
3579 			spin_lock(&fs_info->trans_lock);
3580 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3581 					    root_list) {
3582 				if (dead_root->root_key.objectid ==
3583 				    found_key.objectid) {
3584 					is_dead_root = 1;
3585 					break;
3586 				}
3587 			}
3588 			spin_unlock(&fs_info->trans_lock);
3589 			if (is_dead_root) {
3590 				/* prevent this orphan from being found again */
3591 				key.offset = found_key.objectid - 1;
3592 				continue;
3593 			}
3594 		}
3595 		/*
3596 		 * Inode is already gone but the orphan item is still there,
3597 		 * kill the orphan item.
3598 		 */
3599 		if (ret == -ENOENT) {
3600 			trans = btrfs_start_transaction(root, 1);
3601 			if (IS_ERR(trans)) {
3602 				ret = PTR_ERR(trans);
3603 				goto out;
3604 			}
3605 			btrfs_debug(fs_info, "auto deleting %Lu",
3606 				    found_key.objectid);
3607 			ret = btrfs_del_orphan_item(trans, root,
3608 						    found_key.objectid);
3609 			btrfs_end_transaction(trans);
3610 			if (ret)
3611 				goto out;
3612 			continue;
3613 		}
3614 
3615 		/*
3616 		 * add this inode to the orphan list so btrfs_orphan_del does
3617 		 * the proper thing when we hit it
3618 		 */
3619 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3620 			&BTRFS_I(inode)->runtime_flags);
3621 		atomic_inc(&root->orphan_inodes);
3622 
3623 		/* if we have links, this was a truncate, lets do that */
3624 		if (inode->i_nlink) {
3625 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3626 				iput(inode);
3627 				continue;
3628 			}
3629 			nr_truncate++;
3630 
3631 			/* 1 for the orphan item deletion. */
3632 			trans = btrfs_start_transaction(root, 1);
3633 			if (IS_ERR(trans)) {
3634 				iput(inode);
3635 				ret = PTR_ERR(trans);
3636 				goto out;
3637 			}
3638 			ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3639 			btrfs_end_transaction(trans);
3640 			if (ret) {
3641 				iput(inode);
3642 				goto out;
3643 			}
3644 
3645 			ret = btrfs_truncate(inode, false);
3646 			if (ret)
3647 				btrfs_orphan_del(NULL, BTRFS_I(inode));
3648 		} else {
3649 			nr_unlink++;
3650 		}
3651 
3652 		/* this will do delete_inode and everything for us */
3653 		iput(inode);
3654 		if (ret)
3655 			goto out;
3656 	}
3657 	/* release the path since we're done with it */
3658 	btrfs_release_path(path);
3659 
3660 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3661 
3662 	if (root->orphan_block_rsv)
3663 		btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3664 					(u64)-1);
3665 
3666 	if (root->orphan_block_rsv ||
3667 	    test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3668 		trans = btrfs_join_transaction(root);
3669 		if (!IS_ERR(trans))
3670 			btrfs_end_transaction(trans);
3671 	}
3672 
3673 	if (nr_unlink)
3674 		btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3675 	if (nr_truncate)
3676 		btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3677 
3678 out:
3679 	if (ret)
3680 		btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3681 	btrfs_free_path(path);
3682 	return ret;
3683 }
3684 
3685 /*
3686  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3687  * don't find any xattrs, we know there can't be any acls.
3688  *
3689  * slot is the slot the inode is in, objectid is the objectid of the inode
3690  */
3691 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3692 					  int slot, u64 objectid,
3693 					  int *first_xattr_slot)
3694 {
3695 	u32 nritems = btrfs_header_nritems(leaf);
3696 	struct btrfs_key found_key;
3697 	static u64 xattr_access = 0;
3698 	static u64 xattr_default = 0;
3699 	int scanned = 0;
3700 
3701 	if (!xattr_access) {
3702 		xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3703 					strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3704 		xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3705 					strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3706 	}
3707 
3708 	slot++;
3709 	*first_xattr_slot = -1;
3710 	while (slot < nritems) {
3711 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3712 
3713 		/* we found a different objectid, there must not be acls */
3714 		if (found_key.objectid != objectid)
3715 			return 0;
3716 
3717 		/* we found an xattr, assume we've got an acl */
3718 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3719 			if (*first_xattr_slot == -1)
3720 				*first_xattr_slot = slot;
3721 			if (found_key.offset == xattr_access ||
3722 			    found_key.offset == xattr_default)
3723 				return 1;
3724 		}
3725 
3726 		/*
3727 		 * we found a key greater than an xattr key, there can't
3728 		 * be any acls later on
3729 		 */
3730 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3731 			return 0;
3732 
3733 		slot++;
3734 		scanned++;
3735 
3736 		/*
3737 		 * it goes inode, inode backrefs, xattrs, extents,
3738 		 * so if there are a ton of hard links to an inode there can
3739 		 * be a lot of backrefs.  Don't waste time searching too hard,
3740 		 * this is just an optimization
3741 		 */
3742 		if (scanned >= 8)
3743 			break;
3744 	}
3745 	/* we hit the end of the leaf before we found an xattr or
3746 	 * something larger than an xattr.  We have to assume the inode
3747 	 * has acls
3748 	 */
3749 	if (*first_xattr_slot == -1)
3750 		*first_xattr_slot = slot;
3751 	return 1;
3752 }
3753 
3754 /*
3755  * read an inode from the btree into the in-memory inode
3756  */
3757 static int btrfs_read_locked_inode(struct inode *inode)
3758 {
3759 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3760 	struct btrfs_path *path;
3761 	struct extent_buffer *leaf;
3762 	struct btrfs_inode_item *inode_item;
3763 	struct btrfs_root *root = BTRFS_I(inode)->root;
3764 	struct btrfs_key location;
3765 	unsigned long ptr;
3766 	int maybe_acls;
3767 	u32 rdev;
3768 	int ret;
3769 	bool filled = false;
3770 	int first_xattr_slot;
3771 
3772 	ret = btrfs_fill_inode(inode, &rdev);
3773 	if (!ret)
3774 		filled = true;
3775 
3776 	path = btrfs_alloc_path();
3777 	if (!path) {
3778 		ret = -ENOMEM;
3779 		goto make_bad;
3780 	}
3781 
3782 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3783 
3784 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3785 	if (ret) {
3786 		if (ret > 0)
3787 			ret = -ENOENT;
3788 		goto make_bad;
3789 	}
3790 
3791 	leaf = path->nodes[0];
3792 
3793 	if (filled)
3794 		goto cache_index;
3795 
3796 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3797 				    struct btrfs_inode_item);
3798 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3799 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3800 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3801 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3802 	btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3803 
3804 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3805 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3806 
3807 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3808 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3809 
3810 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3811 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3812 
3813 	BTRFS_I(inode)->i_otime.tv_sec =
3814 		btrfs_timespec_sec(leaf, &inode_item->otime);
3815 	BTRFS_I(inode)->i_otime.tv_nsec =
3816 		btrfs_timespec_nsec(leaf, &inode_item->otime);
3817 
3818 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3819 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3820 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3821 
3822 	inode_set_iversion_queried(inode,
3823 				   btrfs_inode_sequence(leaf, inode_item));
3824 	inode->i_generation = BTRFS_I(inode)->generation;
3825 	inode->i_rdev = 0;
3826 	rdev = btrfs_inode_rdev(leaf, inode_item);
3827 
3828 	BTRFS_I(inode)->index_cnt = (u64)-1;
3829 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3830 
3831 cache_index:
3832 	/*
3833 	 * If we were modified in the current generation and evicted from memory
3834 	 * and then re-read we need to do a full sync since we don't have any
3835 	 * idea about which extents were modified before we were evicted from
3836 	 * cache.
3837 	 *
3838 	 * This is required for both inode re-read from disk and delayed inode
3839 	 * in delayed_nodes_tree.
3840 	 */
3841 	if (BTRFS_I(inode)->last_trans == fs_info->generation)
3842 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3843 			&BTRFS_I(inode)->runtime_flags);
3844 
3845 	/*
3846 	 * We don't persist the id of the transaction where an unlink operation
3847 	 * against the inode was last made. So here we assume the inode might
3848 	 * have been evicted, and therefore the exact value of last_unlink_trans
3849 	 * lost, and set it to last_trans to avoid metadata inconsistencies
3850 	 * between the inode and its parent if the inode is fsync'ed and the log
3851 	 * replayed. For example, in the scenario:
3852 	 *
3853 	 * touch mydir/foo
3854 	 * ln mydir/foo mydir/bar
3855 	 * sync
3856 	 * unlink mydir/bar
3857 	 * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3858 	 * xfs_io -c fsync mydir/foo
3859 	 * <power failure>
3860 	 * mount fs, triggers fsync log replay
3861 	 *
3862 	 * We must make sure that when we fsync our inode foo we also log its
3863 	 * parent inode, otherwise after log replay the parent still has the
3864 	 * dentry with the "bar" name but our inode foo has a link count of 1
3865 	 * and doesn't have an inode ref with the name "bar" anymore.
3866 	 *
3867 	 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3868 	 * but it guarantees correctness at the expense of occasional full
3869 	 * transaction commits on fsync if our inode is a directory, or if our
3870 	 * inode is not a directory, logging its parent unnecessarily.
3871 	 */
3872 	BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3873 
3874 	path->slots[0]++;
3875 	if (inode->i_nlink != 1 ||
3876 	    path->slots[0] >= btrfs_header_nritems(leaf))
3877 		goto cache_acl;
3878 
3879 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3880 	if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3881 		goto cache_acl;
3882 
3883 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3884 	if (location.type == BTRFS_INODE_REF_KEY) {
3885 		struct btrfs_inode_ref *ref;
3886 
3887 		ref = (struct btrfs_inode_ref *)ptr;
3888 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3889 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3890 		struct btrfs_inode_extref *extref;
3891 
3892 		extref = (struct btrfs_inode_extref *)ptr;
3893 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3894 								     extref);
3895 	}
3896 cache_acl:
3897 	/*
3898 	 * try to precache a NULL acl entry for files that don't have
3899 	 * any xattrs or acls
3900 	 */
3901 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3902 			btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3903 	if (first_xattr_slot != -1) {
3904 		path->slots[0] = first_xattr_slot;
3905 		ret = btrfs_load_inode_props(inode, path);
3906 		if (ret)
3907 			btrfs_err(fs_info,
3908 				  "error loading props for ino %llu (root %llu): %d",
3909 				  btrfs_ino(BTRFS_I(inode)),
3910 				  root->root_key.objectid, ret);
3911 	}
3912 	btrfs_free_path(path);
3913 
3914 	if (!maybe_acls)
3915 		cache_no_acl(inode);
3916 
3917 	switch (inode->i_mode & S_IFMT) {
3918 	case S_IFREG:
3919 		inode->i_mapping->a_ops = &btrfs_aops;
3920 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3921 		inode->i_fop = &btrfs_file_operations;
3922 		inode->i_op = &btrfs_file_inode_operations;
3923 		break;
3924 	case S_IFDIR:
3925 		inode->i_fop = &btrfs_dir_file_operations;
3926 		inode->i_op = &btrfs_dir_inode_operations;
3927 		break;
3928 	case S_IFLNK:
3929 		inode->i_op = &btrfs_symlink_inode_operations;
3930 		inode_nohighmem(inode);
3931 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3932 		break;
3933 	default:
3934 		inode->i_op = &btrfs_special_inode_operations;
3935 		init_special_inode(inode, inode->i_mode, rdev);
3936 		break;
3937 	}
3938 
3939 	btrfs_update_iflags(inode);
3940 	return 0;
3941 
3942 make_bad:
3943 	btrfs_free_path(path);
3944 	make_bad_inode(inode);
3945 	return ret;
3946 }
3947 
3948 /*
3949  * given a leaf and an inode, copy the inode fields into the leaf
3950  */
3951 static void fill_inode_item(struct btrfs_trans_handle *trans,
3952 			    struct extent_buffer *leaf,
3953 			    struct btrfs_inode_item *item,
3954 			    struct inode *inode)
3955 {
3956 	struct btrfs_map_token token;
3957 
3958 	btrfs_init_map_token(&token);
3959 
3960 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3961 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3962 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3963 				   &token);
3964 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3965 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3966 
3967 	btrfs_set_token_timespec_sec(leaf, &item->atime,
3968 				     inode->i_atime.tv_sec, &token);
3969 	btrfs_set_token_timespec_nsec(leaf, &item->atime,
3970 				      inode->i_atime.tv_nsec, &token);
3971 
3972 	btrfs_set_token_timespec_sec(leaf, &item->mtime,
3973 				     inode->i_mtime.tv_sec, &token);
3974 	btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3975 				      inode->i_mtime.tv_nsec, &token);
3976 
3977 	btrfs_set_token_timespec_sec(leaf, &item->ctime,
3978 				     inode->i_ctime.tv_sec, &token);
3979 	btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3980 				      inode->i_ctime.tv_nsec, &token);
3981 
3982 	btrfs_set_token_timespec_sec(leaf, &item->otime,
3983 				     BTRFS_I(inode)->i_otime.tv_sec, &token);
3984 	btrfs_set_token_timespec_nsec(leaf, &item->otime,
3985 				      BTRFS_I(inode)->i_otime.tv_nsec, &token);
3986 
3987 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3988 				     &token);
3989 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3990 					 &token);
3991 	btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3992 				       &token);
3993 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3994 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3995 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3996 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3997 }
3998 
3999 /*
4000  * copy everything in the in-memory inode into the btree.
4001  */
4002 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4003 				struct btrfs_root *root, struct inode *inode)
4004 {
4005 	struct btrfs_inode_item *inode_item;
4006 	struct btrfs_path *path;
4007 	struct extent_buffer *leaf;
4008 	int ret;
4009 
4010 	path = btrfs_alloc_path();
4011 	if (!path)
4012 		return -ENOMEM;
4013 
4014 	path->leave_spinning = 1;
4015 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4016 				 1);
4017 	if (ret) {
4018 		if (ret > 0)
4019 			ret = -ENOENT;
4020 		goto failed;
4021 	}
4022 
4023 	leaf = path->nodes[0];
4024 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
4025 				    struct btrfs_inode_item);
4026 
4027 	fill_inode_item(trans, leaf, inode_item, inode);
4028 	btrfs_mark_buffer_dirty(leaf);
4029 	btrfs_set_inode_last_trans(trans, inode);
4030 	ret = 0;
4031 failed:
4032 	btrfs_free_path(path);
4033 	return ret;
4034 }
4035 
4036 /*
4037  * copy everything in the in-memory inode into the btree.
4038  */
4039 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4040 				struct btrfs_root *root, struct inode *inode)
4041 {
4042 	struct btrfs_fs_info *fs_info = root->fs_info;
4043 	int ret;
4044 
4045 	/*
4046 	 * If the inode is a free space inode, we can deadlock during commit
4047 	 * if we put it into the delayed code.
4048 	 *
4049 	 * The data relocation inode should also be directly updated
4050 	 * without delay
4051 	 */
4052 	if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4053 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4054 	    && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4055 		btrfs_update_root_times(trans, root);
4056 
4057 		ret = btrfs_delayed_update_inode(trans, root, inode);
4058 		if (!ret)
4059 			btrfs_set_inode_last_trans(trans, inode);
4060 		return ret;
4061 	}
4062 
4063 	return btrfs_update_inode_item(trans, root, inode);
4064 }
4065 
4066 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4067 					 struct btrfs_root *root,
4068 					 struct inode *inode)
4069 {
4070 	int ret;
4071 
4072 	ret = btrfs_update_inode(trans, root, inode);
4073 	if (ret == -ENOSPC)
4074 		return btrfs_update_inode_item(trans, root, inode);
4075 	return ret;
4076 }
4077 
4078 /*
4079  * unlink helper that gets used here in inode.c and in the tree logging
4080  * recovery code.  It remove a link in a directory with a given name, and
4081  * also drops the back refs in the inode to the directory
4082  */
4083 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4084 				struct btrfs_root *root,
4085 				struct btrfs_inode *dir,
4086 				struct btrfs_inode *inode,
4087 				const char *name, int name_len)
4088 {
4089 	struct btrfs_fs_info *fs_info = root->fs_info;
4090 	struct btrfs_path *path;
4091 	int ret = 0;
4092 	struct extent_buffer *leaf;
4093 	struct btrfs_dir_item *di;
4094 	struct btrfs_key key;
4095 	u64 index;
4096 	u64 ino = btrfs_ino(inode);
4097 	u64 dir_ino = btrfs_ino(dir);
4098 
4099 	path = btrfs_alloc_path();
4100 	if (!path) {
4101 		ret = -ENOMEM;
4102 		goto out;
4103 	}
4104 
4105 	path->leave_spinning = 1;
4106 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4107 				    name, name_len, -1);
4108 	if (IS_ERR(di)) {
4109 		ret = PTR_ERR(di);
4110 		goto err;
4111 	}
4112 	if (!di) {
4113 		ret = -ENOENT;
4114 		goto err;
4115 	}
4116 	leaf = path->nodes[0];
4117 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4118 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119 	if (ret)
4120 		goto err;
4121 	btrfs_release_path(path);
4122 
4123 	/*
4124 	 * If we don't have dir index, we have to get it by looking up
4125 	 * the inode ref, since we get the inode ref, remove it directly,
4126 	 * it is unnecessary to do delayed deletion.
4127 	 *
4128 	 * But if we have dir index, needn't search inode ref to get it.
4129 	 * Since the inode ref is close to the inode item, it is better
4130 	 * that we delay to delete it, and just do this deletion when
4131 	 * we update the inode item.
4132 	 */
4133 	if (inode->dir_index) {
4134 		ret = btrfs_delayed_delete_inode_ref(inode);
4135 		if (!ret) {
4136 			index = inode->dir_index;
4137 			goto skip_backref;
4138 		}
4139 	}
4140 
4141 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4142 				  dir_ino, &index);
4143 	if (ret) {
4144 		btrfs_info(fs_info,
4145 			"failed to delete reference to %.*s, inode %llu parent %llu",
4146 			name_len, name, ino, dir_ino);
4147 		btrfs_abort_transaction(trans, ret);
4148 		goto err;
4149 	}
4150 skip_backref:
4151 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4152 	if (ret) {
4153 		btrfs_abort_transaction(trans, ret);
4154 		goto err;
4155 	}
4156 
4157 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4158 			dir_ino);
4159 	if (ret != 0 && ret != -ENOENT) {
4160 		btrfs_abort_transaction(trans, ret);
4161 		goto err;
4162 	}
4163 
4164 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4165 			index);
4166 	if (ret == -ENOENT)
4167 		ret = 0;
4168 	else if (ret)
4169 		btrfs_abort_transaction(trans, ret);
4170 err:
4171 	btrfs_free_path(path);
4172 	if (ret)
4173 		goto out;
4174 
4175 	btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4176 	inode_inc_iversion(&inode->vfs_inode);
4177 	inode_inc_iversion(&dir->vfs_inode);
4178 	inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4179 		dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4180 	ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4181 out:
4182 	return ret;
4183 }
4184 
4185 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4186 		       struct btrfs_root *root,
4187 		       struct btrfs_inode *dir, struct btrfs_inode *inode,
4188 		       const char *name, int name_len)
4189 {
4190 	int ret;
4191 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4192 	if (!ret) {
4193 		drop_nlink(&inode->vfs_inode);
4194 		ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4195 	}
4196 	return ret;
4197 }
4198 
4199 /*
4200  * helper to start transaction for unlink and rmdir.
4201  *
4202  * unlink and rmdir are special in btrfs, they do not always free space, so
4203  * if we cannot make our reservations the normal way try and see if there is
4204  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4205  * allow the unlink to occur.
4206  */
4207 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4208 {
4209 	struct btrfs_root *root = BTRFS_I(dir)->root;
4210 
4211 	/*
4212 	 * 1 for the possible orphan item
4213 	 * 1 for the dir item
4214 	 * 1 for the dir index
4215 	 * 1 for the inode ref
4216 	 * 1 for the inode
4217 	 */
4218 	return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4219 }
4220 
4221 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4222 {
4223 	struct btrfs_root *root = BTRFS_I(dir)->root;
4224 	struct btrfs_trans_handle *trans;
4225 	struct inode *inode = d_inode(dentry);
4226 	int ret;
4227 
4228 	trans = __unlink_start_trans(dir);
4229 	if (IS_ERR(trans))
4230 		return PTR_ERR(trans);
4231 
4232 	btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4233 			0);
4234 
4235 	ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4236 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4237 			dentry->d_name.len);
4238 	if (ret)
4239 		goto out;
4240 
4241 	if (inode->i_nlink == 0) {
4242 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4243 		if (ret)
4244 			goto out;
4245 	}
4246 
4247 out:
4248 	btrfs_end_transaction(trans);
4249 	btrfs_btree_balance_dirty(root->fs_info);
4250 	return ret;
4251 }
4252 
4253 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4254 			struct btrfs_root *root,
4255 			struct inode *dir, u64 objectid,
4256 			const char *name, int name_len)
4257 {
4258 	struct btrfs_fs_info *fs_info = root->fs_info;
4259 	struct btrfs_path *path;
4260 	struct extent_buffer *leaf;
4261 	struct btrfs_dir_item *di;
4262 	struct btrfs_key key;
4263 	u64 index;
4264 	int ret;
4265 	u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4266 
4267 	path = btrfs_alloc_path();
4268 	if (!path)
4269 		return -ENOMEM;
4270 
4271 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4272 				   name, name_len, -1);
4273 	if (IS_ERR_OR_NULL(di)) {
4274 		if (!di)
4275 			ret = -ENOENT;
4276 		else
4277 			ret = PTR_ERR(di);
4278 		goto out;
4279 	}
4280 
4281 	leaf = path->nodes[0];
4282 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4283 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4284 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4285 	if (ret) {
4286 		btrfs_abort_transaction(trans, ret);
4287 		goto out;
4288 	}
4289 	btrfs_release_path(path);
4290 
4291 	ret = btrfs_del_root_ref(trans, fs_info, objectid,
4292 				 root->root_key.objectid, dir_ino,
4293 				 &index, name, name_len);
4294 	if (ret < 0) {
4295 		if (ret != -ENOENT) {
4296 			btrfs_abort_transaction(trans, ret);
4297 			goto out;
4298 		}
4299 		di = btrfs_search_dir_index_item(root, path, dir_ino,
4300 						 name, name_len);
4301 		if (IS_ERR_OR_NULL(di)) {
4302 			if (!di)
4303 				ret = -ENOENT;
4304 			else
4305 				ret = PTR_ERR(di);
4306 			btrfs_abort_transaction(trans, ret);
4307 			goto out;
4308 		}
4309 
4310 		leaf = path->nodes[0];
4311 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4312 		btrfs_release_path(path);
4313 		index = key.offset;
4314 	}
4315 	btrfs_release_path(path);
4316 
4317 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4318 	if (ret) {
4319 		btrfs_abort_transaction(trans, ret);
4320 		goto out;
4321 	}
4322 
4323 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4324 	inode_inc_iversion(dir);
4325 	dir->i_mtime = dir->i_ctime = current_time(dir);
4326 	ret = btrfs_update_inode_fallback(trans, root, dir);
4327 	if (ret)
4328 		btrfs_abort_transaction(trans, ret);
4329 out:
4330 	btrfs_free_path(path);
4331 	return ret;
4332 }
4333 
4334 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4335 {
4336 	struct inode *inode = d_inode(dentry);
4337 	int err = 0;
4338 	struct btrfs_root *root = BTRFS_I(dir)->root;
4339 	struct btrfs_trans_handle *trans;
4340 	u64 last_unlink_trans;
4341 
4342 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4343 		return -ENOTEMPTY;
4344 	if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4345 		return -EPERM;
4346 
4347 	trans = __unlink_start_trans(dir);
4348 	if (IS_ERR(trans))
4349 		return PTR_ERR(trans);
4350 
4351 	if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4352 		err = btrfs_unlink_subvol(trans, root, dir,
4353 					  BTRFS_I(inode)->location.objectid,
4354 					  dentry->d_name.name,
4355 					  dentry->d_name.len);
4356 		goto out;
4357 	}
4358 
4359 	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4360 	if (err)
4361 		goto out;
4362 
4363 	last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4364 
4365 	/* now the directory is empty */
4366 	err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4367 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4368 			dentry->d_name.len);
4369 	if (!err) {
4370 		btrfs_i_size_write(BTRFS_I(inode), 0);
4371 		/*
4372 		 * Propagate the last_unlink_trans value of the deleted dir to
4373 		 * its parent directory. This is to prevent an unrecoverable
4374 		 * log tree in the case we do something like this:
4375 		 * 1) create dir foo
4376 		 * 2) create snapshot under dir foo
4377 		 * 3) delete the snapshot
4378 		 * 4) rmdir foo
4379 		 * 5) mkdir foo
4380 		 * 6) fsync foo or some file inside foo
4381 		 */
4382 		if (last_unlink_trans >= trans->transid)
4383 			BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4384 	}
4385 out:
4386 	btrfs_end_transaction(trans);
4387 	btrfs_btree_balance_dirty(root->fs_info);
4388 
4389 	return err;
4390 }
4391 
4392 static int truncate_space_check(struct btrfs_trans_handle *trans,
4393 				struct btrfs_root *root,
4394 				u64 bytes_deleted)
4395 {
4396 	struct btrfs_fs_info *fs_info = root->fs_info;
4397 	int ret;
4398 
4399 	/*
4400 	 * This is only used to apply pressure to the enospc system, we don't
4401 	 * intend to use this reservation at all.
4402 	 */
4403 	bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4404 	bytes_deleted *= fs_info->nodesize;
4405 	ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4406 				  bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4407 	if (!ret) {
4408 		trace_btrfs_space_reservation(fs_info, "transaction",
4409 					      trans->transid,
4410 					      bytes_deleted, 1);
4411 		trans->bytes_reserved += bytes_deleted;
4412 	}
4413 	return ret;
4414 
4415 }
4416 
4417 /*
4418  * Return this if we need to call truncate_block for the last bit of the
4419  * truncate.
4420  */
4421 #define NEED_TRUNCATE_BLOCK 1
4422 
4423 /*
4424  * this can truncate away extent items, csum items and directory items.
4425  * It starts at a high offset and removes keys until it can't find
4426  * any higher than new_size
4427  *
4428  * csum items that cross the new i_size are truncated to the new size
4429  * as well.
4430  *
4431  * min_type is the minimum key type to truncate down to.  If set to 0, this
4432  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4433  */
4434 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4435 			       struct btrfs_root *root,
4436 			       struct inode *inode,
4437 			       u64 new_size, u32 min_type)
4438 {
4439 	struct btrfs_fs_info *fs_info = root->fs_info;
4440 	struct btrfs_path *path;
4441 	struct extent_buffer *leaf;
4442 	struct btrfs_file_extent_item *fi;
4443 	struct btrfs_key key;
4444 	struct btrfs_key found_key;
4445 	u64 extent_start = 0;
4446 	u64 extent_num_bytes = 0;
4447 	u64 extent_offset = 0;
4448 	u64 item_end = 0;
4449 	u64 last_size = new_size;
4450 	u32 found_type = (u8)-1;
4451 	int found_extent;
4452 	int del_item;
4453 	int pending_del_nr = 0;
4454 	int pending_del_slot = 0;
4455 	int extent_type = -1;
4456 	int ret;
4457 	int err = 0;
4458 	u64 ino = btrfs_ino(BTRFS_I(inode));
4459 	u64 bytes_deleted = 0;
4460 	bool be_nice = false;
4461 	bool should_throttle = false;
4462 	bool should_end = false;
4463 
4464 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4465 
4466 	/*
4467 	 * for non-free space inodes and ref cows, we want to back off from
4468 	 * time to time
4469 	 */
4470 	if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4471 	    test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4472 		be_nice = true;
4473 
4474 	path = btrfs_alloc_path();
4475 	if (!path)
4476 		return -ENOMEM;
4477 	path->reada = READA_BACK;
4478 
4479 	/*
4480 	 * We want to drop from the next block forward in case this new size is
4481 	 * not block aligned since we will be keeping the last block of the
4482 	 * extent just the way it is.
4483 	 */
4484 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4485 	    root == fs_info->tree_root)
4486 		btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4487 					fs_info->sectorsize),
4488 					(u64)-1, 0);
4489 
4490 	/*
4491 	 * This function is also used to drop the items in the log tree before
4492 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4493 	 * it is used to drop the loged items. So we shouldn't kill the delayed
4494 	 * items.
4495 	 */
4496 	if (min_type == 0 && root == BTRFS_I(inode)->root)
4497 		btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4498 
4499 	key.objectid = ino;
4500 	key.offset = (u64)-1;
4501 	key.type = (u8)-1;
4502 
4503 search_again:
4504 	/*
4505 	 * with a 16K leaf size and 128MB extents, you can actually queue
4506 	 * up a huge file in a single leaf.  Most of the time that
4507 	 * bytes_deleted is > 0, it will be huge by the time we get here
4508 	 */
4509 	if (be_nice && bytes_deleted > SZ_32M) {
4510 		if (btrfs_should_end_transaction(trans)) {
4511 			err = -EAGAIN;
4512 			goto error;
4513 		}
4514 	}
4515 
4516 
4517 	path->leave_spinning = 1;
4518 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4519 	if (ret < 0) {
4520 		err = ret;
4521 		goto out;
4522 	}
4523 
4524 	if (ret > 0) {
4525 		/* there are no items in the tree for us to truncate, we're
4526 		 * done
4527 		 */
4528 		if (path->slots[0] == 0)
4529 			goto out;
4530 		path->slots[0]--;
4531 	}
4532 
4533 	while (1) {
4534 		fi = NULL;
4535 		leaf = path->nodes[0];
4536 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4537 		found_type = found_key.type;
4538 
4539 		if (found_key.objectid != ino)
4540 			break;
4541 
4542 		if (found_type < min_type)
4543 			break;
4544 
4545 		item_end = found_key.offset;
4546 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4547 			fi = btrfs_item_ptr(leaf, path->slots[0],
4548 					    struct btrfs_file_extent_item);
4549 			extent_type = btrfs_file_extent_type(leaf, fi);
4550 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4551 				item_end +=
4552 				    btrfs_file_extent_num_bytes(leaf, fi);
4553 
4554 				trace_btrfs_truncate_show_fi_regular(
4555 					BTRFS_I(inode), leaf, fi,
4556 					found_key.offset);
4557 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4558 				item_end += btrfs_file_extent_inline_len(leaf,
4559 							 path->slots[0], fi);
4560 
4561 				trace_btrfs_truncate_show_fi_inline(
4562 					BTRFS_I(inode), leaf, fi, path->slots[0],
4563 					found_key.offset);
4564 			}
4565 			item_end--;
4566 		}
4567 		if (found_type > min_type) {
4568 			del_item = 1;
4569 		} else {
4570 			if (item_end < new_size)
4571 				break;
4572 			if (found_key.offset >= new_size)
4573 				del_item = 1;
4574 			else
4575 				del_item = 0;
4576 		}
4577 		found_extent = 0;
4578 		/* FIXME, shrink the extent if the ref count is only 1 */
4579 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4580 			goto delete;
4581 
4582 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4583 			u64 num_dec;
4584 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4585 			if (!del_item) {
4586 				u64 orig_num_bytes =
4587 					btrfs_file_extent_num_bytes(leaf, fi);
4588 				extent_num_bytes = ALIGN(new_size -
4589 						found_key.offset,
4590 						fs_info->sectorsize);
4591 				btrfs_set_file_extent_num_bytes(leaf, fi,
4592 							 extent_num_bytes);
4593 				num_dec = (orig_num_bytes -
4594 					   extent_num_bytes);
4595 				if (test_bit(BTRFS_ROOT_REF_COWS,
4596 					     &root->state) &&
4597 				    extent_start != 0)
4598 					inode_sub_bytes(inode, num_dec);
4599 				btrfs_mark_buffer_dirty(leaf);
4600 			} else {
4601 				extent_num_bytes =
4602 					btrfs_file_extent_disk_num_bytes(leaf,
4603 									 fi);
4604 				extent_offset = found_key.offset -
4605 					btrfs_file_extent_offset(leaf, fi);
4606 
4607 				/* FIXME blocksize != 4096 */
4608 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4609 				if (extent_start != 0) {
4610 					found_extent = 1;
4611 					if (test_bit(BTRFS_ROOT_REF_COWS,
4612 						     &root->state))
4613 						inode_sub_bytes(inode, num_dec);
4614 				}
4615 			}
4616 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4617 			/*
4618 			 * we can't truncate inline items that have had
4619 			 * special encodings
4620 			 */
4621 			if (!del_item &&
4622 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4623 			    btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4624 			    btrfs_file_extent_compression(leaf, fi) == 0) {
4625 				u32 size = (u32)(new_size - found_key.offset);
4626 
4627 				btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4628 				size = btrfs_file_extent_calc_inline_size(size);
4629 				btrfs_truncate_item(root->fs_info, path, size, 1);
4630 			} else if (!del_item) {
4631 				/*
4632 				 * We have to bail so the last_size is set to
4633 				 * just before this extent.
4634 				 */
4635 				err = NEED_TRUNCATE_BLOCK;
4636 				break;
4637 			}
4638 
4639 			if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4640 				inode_sub_bytes(inode, item_end + 1 - new_size);
4641 		}
4642 delete:
4643 		if (del_item)
4644 			last_size = found_key.offset;
4645 		else
4646 			last_size = new_size;
4647 		if (del_item) {
4648 			if (!pending_del_nr) {
4649 				/* no pending yet, add ourselves */
4650 				pending_del_slot = path->slots[0];
4651 				pending_del_nr = 1;
4652 			} else if (pending_del_nr &&
4653 				   path->slots[0] + 1 == pending_del_slot) {
4654 				/* hop on the pending chunk */
4655 				pending_del_nr++;
4656 				pending_del_slot = path->slots[0];
4657 			} else {
4658 				BUG();
4659 			}
4660 		} else {
4661 			break;
4662 		}
4663 		should_throttle = false;
4664 
4665 		if (found_extent &&
4666 		    (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4667 		     root == fs_info->tree_root)) {
4668 			btrfs_set_path_blocking(path);
4669 			bytes_deleted += extent_num_bytes;
4670 			ret = btrfs_free_extent(trans, root, extent_start,
4671 						extent_num_bytes, 0,
4672 						btrfs_header_owner(leaf),
4673 						ino, extent_offset);
4674 			BUG_ON(ret);
4675 			if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4676 				btrfs_async_run_delayed_refs(fs_info,
4677 					trans->delayed_ref_updates * 2,
4678 					trans->transid, 0);
4679 			if (be_nice) {
4680 				if (truncate_space_check(trans, root,
4681 							 extent_num_bytes)) {
4682 					should_end = true;
4683 				}
4684 				if (btrfs_should_throttle_delayed_refs(trans,
4685 								       fs_info))
4686 					should_throttle = true;
4687 			}
4688 		}
4689 
4690 		if (found_type == BTRFS_INODE_ITEM_KEY)
4691 			break;
4692 
4693 		if (path->slots[0] == 0 ||
4694 		    path->slots[0] != pending_del_slot ||
4695 		    should_throttle || should_end) {
4696 			if (pending_del_nr) {
4697 				ret = btrfs_del_items(trans, root, path,
4698 						pending_del_slot,
4699 						pending_del_nr);
4700 				if (ret) {
4701 					btrfs_abort_transaction(trans, ret);
4702 					goto error;
4703 				}
4704 				pending_del_nr = 0;
4705 			}
4706 			btrfs_release_path(path);
4707 			if (should_throttle) {
4708 				unsigned long updates = trans->delayed_ref_updates;
4709 				if (updates) {
4710 					trans->delayed_ref_updates = 0;
4711 					ret = btrfs_run_delayed_refs(trans,
4712 								   updates * 2);
4713 					if (ret && !err)
4714 						err = ret;
4715 				}
4716 			}
4717 			/*
4718 			 * if we failed to refill our space rsv, bail out
4719 			 * and let the transaction restart
4720 			 */
4721 			if (should_end) {
4722 				err = -EAGAIN;
4723 				goto error;
4724 			}
4725 			goto search_again;
4726 		} else {
4727 			path->slots[0]--;
4728 		}
4729 	}
4730 out:
4731 	if (pending_del_nr) {
4732 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4733 				      pending_del_nr);
4734 		if (ret)
4735 			btrfs_abort_transaction(trans, ret);
4736 	}
4737 error:
4738 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4739 		ASSERT(last_size >= new_size);
4740 		if (!err && last_size > new_size)
4741 			last_size = new_size;
4742 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4743 	}
4744 
4745 	btrfs_free_path(path);
4746 
4747 	if (be_nice && bytes_deleted > SZ_32M) {
4748 		unsigned long updates = trans->delayed_ref_updates;
4749 		if (updates) {
4750 			trans->delayed_ref_updates = 0;
4751 			ret = btrfs_run_delayed_refs(trans, updates * 2);
4752 			if (ret && !err)
4753 				err = ret;
4754 		}
4755 	}
4756 	return err;
4757 }
4758 
4759 /*
4760  * btrfs_truncate_block - read, zero a chunk and write a block
4761  * @inode - inode that we're zeroing
4762  * @from - the offset to start zeroing
4763  * @len - the length to zero, 0 to zero the entire range respective to the
4764  *	offset
4765  * @front - zero up to the offset instead of from the offset on
4766  *
4767  * This will find the block for the "from" offset and cow the block and zero the
4768  * part we want to zero.  This is used with truncate and hole punching.
4769  */
4770 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4771 			int front)
4772 {
4773 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4774 	struct address_space *mapping = inode->i_mapping;
4775 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4776 	struct btrfs_ordered_extent *ordered;
4777 	struct extent_state *cached_state = NULL;
4778 	struct extent_changeset *data_reserved = NULL;
4779 	char *kaddr;
4780 	u32 blocksize = fs_info->sectorsize;
4781 	pgoff_t index = from >> PAGE_SHIFT;
4782 	unsigned offset = from & (blocksize - 1);
4783 	struct page *page;
4784 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4785 	int ret = 0;
4786 	u64 block_start;
4787 	u64 block_end;
4788 
4789 	if (IS_ALIGNED(offset, blocksize) &&
4790 	    (!len || IS_ALIGNED(len, blocksize)))
4791 		goto out;
4792 
4793 	block_start = round_down(from, blocksize);
4794 	block_end = block_start + blocksize - 1;
4795 
4796 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4797 					   block_start, blocksize);
4798 	if (ret)
4799 		goto out;
4800 
4801 again:
4802 	page = find_or_create_page(mapping, index, mask);
4803 	if (!page) {
4804 		btrfs_delalloc_release_space(inode, data_reserved,
4805 					     block_start, blocksize, true);
4806 		btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4807 		ret = -ENOMEM;
4808 		goto out;
4809 	}
4810 
4811 	if (!PageUptodate(page)) {
4812 		ret = btrfs_readpage(NULL, page);
4813 		lock_page(page);
4814 		if (page->mapping != mapping) {
4815 			unlock_page(page);
4816 			put_page(page);
4817 			goto again;
4818 		}
4819 		if (!PageUptodate(page)) {
4820 			ret = -EIO;
4821 			goto out_unlock;
4822 		}
4823 	}
4824 	wait_on_page_writeback(page);
4825 
4826 	lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4827 	set_page_extent_mapped(page);
4828 
4829 	ordered = btrfs_lookup_ordered_extent(inode, block_start);
4830 	if (ordered) {
4831 		unlock_extent_cached(io_tree, block_start, block_end,
4832 				     &cached_state);
4833 		unlock_page(page);
4834 		put_page(page);
4835 		btrfs_start_ordered_extent(inode, ordered, 1);
4836 		btrfs_put_ordered_extent(ordered);
4837 		goto again;
4838 	}
4839 
4840 	clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4841 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4842 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4843 			  0, 0, &cached_state);
4844 
4845 	ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4846 					&cached_state, 0);
4847 	if (ret) {
4848 		unlock_extent_cached(io_tree, block_start, block_end,
4849 				     &cached_state);
4850 		goto out_unlock;
4851 	}
4852 
4853 	if (offset != blocksize) {
4854 		if (!len)
4855 			len = blocksize - offset;
4856 		kaddr = kmap(page);
4857 		if (front)
4858 			memset(kaddr + (block_start - page_offset(page)),
4859 				0, offset);
4860 		else
4861 			memset(kaddr + (block_start - page_offset(page)) +  offset,
4862 				0, len);
4863 		flush_dcache_page(page);
4864 		kunmap(page);
4865 	}
4866 	ClearPageChecked(page);
4867 	set_page_dirty(page);
4868 	unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4869 
4870 out_unlock:
4871 	if (ret)
4872 		btrfs_delalloc_release_space(inode, data_reserved, block_start,
4873 					     blocksize, true);
4874 	btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4875 	unlock_page(page);
4876 	put_page(page);
4877 out:
4878 	extent_changeset_free(data_reserved);
4879 	return ret;
4880 }
4881 
4882 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4883 			     u64 offset, u64 len)
4884 {
4885 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4886 	struct btrfs_trans_handle *trans;
4887 	int ret;
4888 
4889 	/*
4890 	 * Still need to make sure the inode looks like it's been updated so
4891 	 * that any holes get logged if we fsync.
4892 	 */
4893 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4894 		BTRFS_I(inode)->last_trans = fs_info->generation;
4895 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4896 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4897 		return 0;
4898 	}
4899 
4900 	/*
4901 	 * 1 - for the one we're dropping
4902 	 * 1 - for the one we're adding
4903 	 * 1 - for updating the inode.
4904 	 */
4905 	trans = btrfs_start_transaction(root, 3);
4906 	if (IS_ERR(trans))
4907 		return PTR_ERR(trans);
4908 
4909 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4910 	if (ret) {
4911 		btrfs_abort_transaction(trans, ret);
4912 		btrfs_end_transaction(trans);
4913 		return ret;
4914 	}
4915 
4916 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4917 			offset, 0, 0, len, 0, len, 0, 0, 0);
4918 	if (ret)
4919 		btrfs_abort_transaction(trans, ret);
4920 	else
4921 		btrfs_update_inode(trans, root, inode);
4922 	btrfs_end_transaction(trans);
4923 	return ret;
4924 }
4925 
4926 /*
4927  * This function puts in dummy file extents for the area we're creating a hole
4928  * for.  So if we are truncating this file to a larger size we need to insert
4929  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4930  * the range between oldsize and size
4931  */
4932 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4933 {
4934 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4935 	struct btrfs_root *root = BTRFS_I(inode)->root;
4936 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4937 	struct extent_map *em = NULL;
4938 	struct extent_state *cached_state = NULL;
4939 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4940 	u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4941 	u64 block_end = ALIGN(size, fs_info->sectorsize);
4942 	u64 last_byte;
4943 	u64 cur_offset;
4944 	u64 hole_size;
4945 	int err = 0;
4946 
4947 	/*
4948 	 * If our size started in the middle of a block we need to zero out the
4949 	 * rest of the block before we expand the i_size, otherwise we could
4950 	 * expose stale data.
4951 	 */
4952 	err = btrfs_truncate_block(inode, oldsize, 0, 0);
4953 	if (err)
4954 		return err;
4955 
4956 	if (size <= hole_start)
4957 		return 0;
4958 
4959 	while (1) {
4960 		struct btrfs_ordered_extent *ordered;
4961 
4962 		lock_extent_bits(io_tree, hole_start, block_end - 1,
4963 				 &cached_state);
4964 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4965 						     block_end - hole_start);
4966 		if (!ordered)
4967 			break;
4968 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4969 				     &cached_state);
4970 		btrfs_start_ordered_extent(inode, ordered, 1);
4971 		btrfs_put_ordered_extent(ordered);
4972 	}
4973 
4974 	cur_offset = hole_start;
4975 	while (1) {
4976 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4977 				block_end - cur_offset, 0);
4978 		if (IS_ERR(em)) {
4979 			err = PTR_ERR(em);
4980 			em = NULL;
4981 			break;
4982 		}
4983 		last_byte = min(extent_map_end(em), block_end);
4984 		last_byte = ALIGN(last_byte, fs_info->sectorsize);
4985 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4986 			struct extent_map *hole_em;
4987 			hole_size = last_byte - cur_offset;
4988 
4989 			err = maybe_insert_hole(root, inode, cur_offset,
4990 						hole_size);
4991 			if (err)
4992 				break;
4993 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4994 						cur_offset + hole_size - 1, 0);
4995 			hole_em = alloc_extent_map();
4996 			if (!hole_em) {
4997 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4998 					&BTRFS_I(inode)->runtime_flags);
4999 				goto next;
5000 			}
5001 			hole_em->start = cur_offset;
5002 			hole_em->len = hole_size;
5003 			hole_em->orig_start = cur_offset;
5004 
5005 			hole_em->block_start = EXTENT_MAP_HOLE;
5006 			hole_em->block_len = 0;
5007 			hole_em->orig_block_len = 0;
5008 			hole_em->ram_bytes = hole_size;
5009 			hole_em->bdev = fs_info->fs_devices->latest_bdev;
5010 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
5011 			hole_em->generation = fs_info->generation;
5012 
5013 			while (1) {
5014 				write_lock(&em_tree->lock);
5015 				err = add_extent_mapping(em_tree, hole_em, 1);
5016 				write_unlock(&em_tree->lock);
5017 				if (err != -EEXIST)
5018 					break;
5019 				btrfs_drop_extent_cache(BTRFS_I(inode),
5020 							cur_offset,
5021 							cur_offset +
5022 							hole_size - 1, 0);
5023 			}
5024 			free_extent_map(hole_em);
5025 		}
5026 next:
5027 		free_extent_map(em);
5028 		em = NULL;
5029 		cur_offset = last_byte;
5030 		if (cur_offset >= block_end)
5031 			break;
5032 	}
5033 	free_extent_map(em);
5034 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5035 	return err;
5036 }
5037 
5038 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5039 {
5040 	struct btrfs_root *root = BTRFS_I(inode)->root;
5041 	struct btrfs_trans_handle *trans;
5042 	loff_t oldsize = i_size_read(inode);
5043 	loff_t newsize = attr->ia_size;
5044 	int mask = attr->ia_valid;
5045 	int ret;
5046 
5047 	/*
5048 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5049 	 * special case where we need to update the times despite not having
5050 	 * these flags set.  For all other operations the VFS set these flags
5051 	 * explicitly if it wants a timestamp update.
5052 	 */
5053 	if (newsize != oldsize) {
5054 		inode_inc_iversion(inode);
5055 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5056 			inode->i_ctime = inode->i_mtime =
5057 				current_time(inode);
5058 	}
5059 
5060 	if (newsize > oldsize) {
5061 		/*
5062 		 * Don't do an expanding truncate while snapshotting is ongoing.
5063 		 * This is to ensure the snapshot captures a fully consistent
5064 		 * state of this file - if the snapshot captures this expanding
5065 		 * truncation, it must capture all writes that happened before
5066 		 * this truncation.
5067 		 */
5068 		btrfs_wait_for_snapshot_creation(root);
5069 		ret = btrfs_cont_expand(inode, oldsize, newsize);
5070 		if (ret) {
5071 			btrfs_end_write_no_snapshotting(root);
5072 			return ret;
5073 		}
5074 
5075 		trans = btrfs_start_transaction(root, 1);
5076 		if (IS_ERR(trans)) {
5077 			btrfs_end_write_no_snapshotting(root);
5078 			return PTR_ERR(trans);
5079 		}
5080 
5081 		i_size_write(inode, newsize);
5082 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5083 		pagecache_isize_extended(inode, oldsize, newsize);
5084 		ret = btrfs_update_inode(trans, root, inode);
5085 		btrfs_end_write_no_snapshotting(root);
5086 		btrfs_end_transaction(trans);
5087 	} else {
5088 
5089 		/*
5090 		 * We're truncating a file that used to have good data down to
5091 		 * zero. Make sure it gets into the ordered flush list so that
5092 		 * any new writes get down to disk quickly.
5093 		 */
5094 		if (newsize == 0)
5095 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5096 				&BTRFS_I(inode)->runtime_flags);
5097 
5098 		/*
5099 		 * 1 for the orphan item we're going to add
5100 		 * 1 for the orphan item deletion.
5101 		 */
5102 		trans = btrfs_start_transaction(root, 2);
5103 		if (IS_ERR(trans))
5104 			return PTR_ERR(trans);
5105 
5106 		/*
5107 		 * We need to do this in case we fail at _any_ point during the
5108 		 * actual truncate.  Once we do the truncate_setsize we could
5109 		 * invalidate pages which forces any outstanding ordered io to
5110 		 * be instantly completed which will give us extents that need
5111 		 * to be truncated.  If we fail to get an orphan inode down we
5112 		 * could have left over extents that were never meant to live,
5113 		 * so we need to guarantee from this point on that everything
5114 		 * will be consistent.
5115 		 */
5116 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5117 		btrfs_end_transaction(trans);
5118 		if (ret)
5119 			return ret;
5120 
5121 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
5122 		truncate_setsize(inode, newsize);
5123 
5124 		/* Disable nonlocked read DIO to avoid the end less truncate */
5125 		btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5126 		inode_dio_wait(inode);
5127 		btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5128 
5129 		ret = btrfs_truncate(inode, newsize == oldsize);
5130 		if (ret && inode->i_nlink) {
5131 			int err;
5132 
5133 			/* To get a stable disk_i_size */
5134 			err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5135 			if (err) {
5136 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5137 				return err;
5138 			}
5139 
5140 			/*
5141 			 * failed to truncate, disk_i_size is only adjusted down
5142 			 * as we remove extents, so it should represent the true
5143 			 * size of the inode, so reset the in memory size and
5144 			 * delete our orphan entry.
5145 			 */
5146 			trans = btrfs_join_transaction(root);
5147 			if (IS_ERR(trans)) {
5148 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5149 				return ret;
5150 			}
5151 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5152 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
5153 			if (err)
5154 				btrfs_abort_transaction(trans, err);
5155 			btrfs_end_transaction(trans);
5156 		}
5157 	}
5158 
5159 	return ret;
5160 }
5161 
5162 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5163 {
5164 	struct inode *inode = d_inode(dentry);
5165 	struct btrfs_root *root = BTRFS_I(inode)->root;
5166 	int err;
5167 
5168 	if (btrfs_root_readonly(root))
5169 		return -EROFS;
5170 
5171 	err = setattr_prepare(dentry, attr);
5172 	if (err)
5173 		return err;
5174 
5175 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5176 		err = btrfs_setsize(inode, attr);
5177 		if (err)
5178 			return err;
5179 	}
5180 
5181 	if (attr->ia_valid) {
5182 		setattr_copy(inode, attr);
5183 		inode_inc_iversion(inode);
5184 		err = btrfs_dirty_inode(inode);
5185 
5186 		if (!err && attr->ia_valid & ATTR_MODE)
5187 			err = posix_acl_chmod(inode, inode->i_mode);
5188 	}
5189 
5190 	return err;
5191 }
5192 
5193 /*
5194  * While truncating the inode pages during eviction, we get the VFS calling
5195  * btrfs_invalidatepage() against each page of the inode. This is slow because
5196  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5197  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5198  * extent_state structures over and over, wasting lots of time.
5199  *
5200  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5201  * those expensive operations on a per page basis and do only the ordered io
5202  * finishing, while we release here the extent_map and extent_state structures,
5203  * without the excessive merging and splitting.
5204  */
5205 static void evict_inode_truncate_pages(struct inode *inode)
5206 {
5207 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5208 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5209 	struct rb_node *node;
5210 
5211 	ASSERT(inode->i_state & I_FREEING);
5212 	truncate_inode_pages_final(&inode->i_data);
5213 
5214 	write_lock(&map_tree->lock);
5215 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
5216 		struct extent_map *em;
5217 
5218 		node = rb_first(&map_tree->map);
5219 		em = rb_entry(node, struct extent_map, rb_node);
5220 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5221 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5222 		remove_extent_mapping(map_tree, em);
5223 		free_extent_map(em);
5224 		if (need_resched()) {
5225 			write_unlock(&map_tree->lock);
5226 			cond_resched();
5227 			write_lock(&map_tree->lock);
5228 		}
5229 	}
5230 	write_unlock(&map_tree->lock);
5231 
5232 	/*
5233 	 * Keep looping until we have no more ranges in the io tree.
5234 	 * We can have ongoing bios started by readpages (called from readahead)
5235 	 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5236 	 * still in progress (unlocked the pages in the bio but did not yet
5237 	 * unlocked the ranges in the io tree). Therefore this means some
5238 	 * ranges can still be locked and eviction started because before
5239 	 * submitting those bios, which are executed by a separate task (work
5240 	 * queue kthread), inode references (inode->i_count) were not taken
5241 	 * (which would be dropped in the end io callback of each bio).
5242 	 * Therefore here we effectively end up waiting for those bios and
5243 	 * anyone else holding locked ranges without having bumped the inode's
5244 	 * reference count - if we don't do it, when they access the inode's
5245 	 * io_tree to unlock a range it may be too late, leading to an
5246 	 * use-after-free issue.
5247 	 */
5248 	spin_lock(&io_tree->lock);
5249 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
5250 		struct extent_state *state;
5251 		struct extent_state *cached_state = NULL;
5252 		u64 start;
5253 		u64 end;
5254 
5255 		node = rb_first(&io_tree->state);
5256 		state = rb_entry(node, struct extent_state, rb_node);
5257 		start = state->start;
5258 		end = state->end;
5259 		spin_unlock(&io_tree->lock);
5260 
5261 		lock_extent_bits(io_tree, start, end, &cached_state);
5262 
5263 		/*
5264 		 * If still has DELALLOC flag, the extent didn't reach disk,
5265 		 * and its reserved space won't be freed by delayed_ref.
5266 		 * So we need to free its reserved space here.
5267 		 * (Refer to comment in btrfs_invalidatepage, case 2)
5268 		 *
5269 		 * Note, end is the bytenr of last byte, so we need + 1 here.
5270 		 */
5271 		if (state->state & EXTENT_DELALLOC)
5272 			btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5273 
5274 		clear_extent_bit(io_tree, start, end,
5275 				 EXTENT_LOCKED | EXTENT_DIRTY |
5276 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5277 				 EXTENT_DEFRAG, 1, 1, &cached_state);
5278 
5279 		cond_resched();
5280 		spin_lock(&io_tree->lock);
5281 	}
5282 	spin_unlock(&io_tree->lock);
5283 }
5284 
5285 void btrfs_evict_inode(struct inode *inode)
5286 {
5287 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5288 	struct btrfs_trans_handle *trans;
5289 	struct btrfs_root *root = BTRFS_I(inode)->root;
5290 	struct btrfs_block_rsv *rsv, *global_rsv;
5291 	int steal_from_global = 0;
5292 	u64 min_size;
5293 	int ret;
5294 
5295 	trace_btrfs_inode_evict(inode);
5296 
5297 	if (!root) {
5298 		clear_inode(inode);
5299 		return;
5300 	}
5301 
5302 	min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5303 
5304 	evict_inode_truncate_pages(inode);
5305 
5306 	if (inode->i_nlink &&
5307 	    ((btrfs_root_refs(&root->root_item) != 0 &&
5308 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5309 	     btrfs_is_free_space_inode(BTRFS_I(inode))))
5310 		goto no_delete;
5311 
5312 	if (is_bad_inode(inode)) {
5313 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5314 		goto no_delete;
5315 	}
5316 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5317 	if (!special_file(inode->i_mode))
5318 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
5319 
5320 	btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5321 
5322 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5323 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5324 				 &BTRFS_I(inode)->runtime_flags));
5325 		goto no_delete;
5326 	}
5327 
5328 	if (inode->i_nlink > 0) {
5329 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5330 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5331 		goto no_delete;
5332 	}
5333 
5334 	ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5335 	if (ret) {
5336 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5337 		goto no_delete;
5338 	}
5339 
5340 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5341 	if (!rsv) {
5342 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5343 		goto no_delete;
5344 	}
5345 	rsv->size = min_size;
5346 	rsv->failfast = 1;
5347 	global_rsv = &fs_info->global_block_rsv;
5348 
5349 	btrfs_i_size_write(BTRFS_I(inode), 0);
5350 
5351 	/*
5352 	 * This is a bit simpler than btrfs_truncate since we've already
5353 	 * reserved our space for our orphan item in the unlink, so we just
5354 	 * need to reserve some slack space in case we add bytes and update
5355 	 * inode item when doing the truncate.
5356 	 */
5357 	while (1) {
5358 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
5359 					     BTRFS_RESERVE_FLUSH_LIMIT);
5360 
5361 		/*
5362 		 * Try and steal from the global reserve since we will
5363 		 * likely not use this space anyway, we want to try as
5364 		 * hard as possible to get this to work.
5365 		 */
5366 		if (ret)
5367 			steal_from_global++;
5368 		else
5369 			steal_from_global = 0;
5370 		ret = 0;
5371 
5372 		/*
5373 		 * steal_from_global == 0: we reserved stuff, hooray!
5374 		 * steal_from_global == 1: we didn't reserve stuff, boo!
5375 		 * steal_from_global == 2: we've committed, still not a lot of
5376 		 * room but maybe we'll have room in the global reserve this
5377 		 * time.
5378 		 * steal_from_global == 3: abandon all hope!
5379 		 */
5380 		if (steal_from_global > 2) {
5381 			btrfs_warn(fs_info,
5382 				   "Could not get space for a delete, will truncate on mount %d",
5383 				   ret);
5384 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5385 			btrfs_free_block_rsv(fs_info, rsv);
5386 			goto no_delete;
5387 		}
5388 
5389 		trans = btrfs_join_transaction(root);
5390 		if (IS_ERR(trans)) {
5391 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5392 			btrfs_free_block_rsv(fs_info, rsv);
5393 			goto no_delete;
5394 		}
5395 
5396 		/*
5397 		 * We can't just steal from the global reserve, we need to make
5398 		 * sure there is room to do it, if not we need to commit and try
5399 		 * again.
5400 		 */
5401 		if (steal_from_global) {
5402 			if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5403 				ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5404 							      min_size, 0);
5405 			else
5406 				ret = -ENOSPC;
5407 		}
5408 
5409 		/*
5410 		 * Couldn't steal from the global reserve, we have too much
5411 		 * pending stuff built up, commit the transaction and try it
5412 		 * again.
5413 		 */
5414 		if (ret) {
5415 			ret = btrfs_commit_transaction(trans);
5416 			if (ret) {
5417 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5418 				btrfs_free_block_rsv(fs_info, rsv);
5419 				goto no_delete;
5420 			}
5421 			continue;
5422 		} else {
5423 			steal_from_global = 0;
5424 		}
5425 
5426 		trans->block_rsv = rsv;
5427 
5428 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5429 		if (ret != -ENOSPC && ret != -EAGAIN)
5430 			break;
5431 
5432 		trans->block_rsv = &fs_info->trans_block_rsv;
5433 		btrfs_end_transaction(trans);
5434 		trans = NULL;
5435 		btrfs_btree_balance_dirty(fs_info);
5436 	}
5437 
5438 	btrfs_free_block_rsv(fs_info, rsv);
5439 
5440 	/*
5441 	 * Errors here aren't a big deal, it just means we leave orphan items
5442 	 * in the tree.  They will be cleaned up on the next mount.
5443 	 */
5444 	if (ret == 0) {
5445 		trans->block_rsv = root->orphan_block_rsv;
5446 		btrfs_orphan_del(trans, BTRFS_I(inode));
5447 	} else {
5448 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5449 	}
5450 
5451 	trans->block_rsv = &fs_info->trans_block_rsv;
5452 	if (!(root == fs_info->tree_root ||
5453 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5454 		btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5455 
5456 	btrfs_end_transaction(trans);
5457 	btrfs_btree_balance_dirty(fs_info);
5458 no_delete:
5459 	btrfs_remove_delayed_node(BTRFS_I(inode));
5460 	clear_inode(inode);
5461 }
5462 
5463 /*
5464  * this returns the key found in the dir entry in the location pointer.
5465  * If no dir entries were found, returns -ENOENT.
5466  * If found a corrupted location in dir entry, returns -EUCLEAN.
5467  */
5468 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5469 			       struct btrfs_key *location)
5470 {
5471 	const char *name = dentry->d_name.name;
5472 	int namelen = dentry->d_name.len;
5473 	struct btrfs_dir_item *di;
5474 	struct btrfs_path *path;
5475 	struct btrfs_root *root = BTRFS_I(dir)->root;
5476 	int ret = 0;
5477 
5478 	path = btrfs_alloc_path();
5479 	if (!path)
5480 		return -ENOMEM;
5481 
5482 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5483 			name, namelen, 0);
5484 	if (!di) {
5485 		ret = -ENOENT;
5486 		goto out;
5487 	}
5488 	if (IS_ERR(di)) {
5489 		ret = PTR_ERR(di);
5490 		goto out;
5491 	}
5492 
5493 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5494 	if (location->type != BTRFS_INODE_ITEM_KEY &&
5495 	    location->type != BTRFS_ROOT_ITEM_KEY) {
5496 		ret = -EUCLEAN;
5497 		btrfs_warn(root->fs_info,
5498 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5499 			   __func__, name, btrfs_ino(BTRFS_I(dir)),
5500 			   location->objectid, location->type, location->offset);
5501 	}
5502 out:
5503 	btrfs_free_path(path);
5504 	return ret;
5505 }
5506 
5507 /*
5508  * when we hit a tree root in a directory, the btrfs part of the inode
5509  * needs to be changed to reflect the root directory of the tree root.  This
5510  * is kind of like crossing a mount point.
5511  */
5512 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5513 				    struct inode *dir,
5514 				    struct dentry *dentry,
5515 				    struct btrfs_key *location,
5516 				    struct btrfs_root **sub_root)
5517 {
5518 	struct btrfs_path *path;
5519 	struct btrfs_root *new_root;
5520 	struct btrfs_root_ref *ref;
5521 	struct extent_buffer *leaf;
5522 	struct btrfs_key key;
5523 	int ret;
5524 	int err = 0;
5525 
5526 	path = btrfs_alloc_path();
5527 	if (!path) {
5528 		err = -ENOMEM;
5529 		goto out;
5530 	}
5531 
5532 	err = -ENOENT;
5533 	key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5534 	key.type = BTRFS_ROOT_REF_KEY;
5535 	key.offset = location->objectid;
5536 
5537 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5538 	if (ret) {
5539 		if (ret < 0)
5540 			err = ret;
5541 		goto out;
5542 	}
5543 
5544 	leaf = path->nodes[0];
5545 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5546 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5547 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5548 		goto out;
5549 
5550 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5551 				   (unsigned long)(ref + 1),
5552 				   dentry->d_name.len);
5553 	if (ret)
5554 		goto out;
5555 
5556 	btrfs_release_path(path);
5557 
5558 	new_root = btrfs_read_fs_root_no_name(fs_info, location);
5559 	if (IS_ERR(new_root)) {
5560 		err = PTR_ERR(new_root);
5561 		goto out;
5562 	}
5563 
5564 	*sub_root = new_root;
5565 	location->objectid = btrfs_root_dirid(&new_root->root_item);
5566 	location->type = BTRFS_INODE_ITEM_KEY;
5567 	location->offset = 0;
5568 	err = 0;
5569 out:
5570 	btrfs_free_path(path);
5571 	return err;
5572 }
5573 
5574 static void inode_tree_add(struct inode *inode)
5575 {
5576 	struct btrfs_root *root = BTRFS_I(inode)->root;
5577 	struct btrfs_inode *entry;
5578 	struct rb_node **p;
5579 	struct rb_node *parent;
5580 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
5581 	u64 ino = btrfs_ino(BTRFS_I(inode));
5582 
5583 	if (inode_unhashed(inode))
5584 		return;
5585 	parent = NULL;
5586 	spin_lock(&root->inode_lock);
5587 	p = &root->inode_tree.rb_node;
5588 	while (*p) {
5589 		parent = *p;
5590 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
5591 
5592 		if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5593 			p = &parent->rb_left;
5594 		else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5595 			p = &parent->rb_right;
5596 		else {
5597 			WARN_ON(!(entry->vfs_inode.i_state &
5598 				  (I_WILL_FREE | I_FREEING)));
5599 			rb_replace_node(parent, new, &root->inode_tree);
5600 			RB_CLEAR_NODE(parent);
5601 			spin_unlock(&root->inode_lock);
5602 			return;
5603 		}
5604 	}
5605 	rb_link_node(new, parent, p);
5606 	rb_insert_color(new, &root->inode_tree);
5607 	spin_unlock(&root->inode_lock);
5608 }
5609 
5610 static void inode_tree_del(struct inode *inode)
5611 {
5612 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5613 	struct btrfs_root *root = BTRFS_I(inode)->root;
5614 	int empty = 0;
5615 
5616 	spin_lock(&root->inode_lock);
5617 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5618 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5619 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5620 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5621 	}
5622 	spin_unlock(&root->inode_lock);
5623 
5624 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
5625 		synchronize_srcu(&fs_info->subvol_srcu);
5626 		spin_lock(&root->inode_lock);
5627 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5628 		spin_unlock(&root->inode_lock);
5629 		if (empty)
5630 			btrfs_add_dead_root(root);
5631 	}
5632 }
5633 
5634 void btrfs_invalidate_inodes(struct btrfs_root *root)
5635 {
5636 	struct btrfs_fs_info *fs_info = root->fs_info;
5637 	struct rb_node *node;
5638 	struct rb_node *prev;
5639 	struct btrfs_inode *entry;
5640 	struct inode *inode;
5641 	u64 objectid = 0;
5642 
5643 	if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5644 		WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5645 
5646 	spin_lock(&root->inode_lock);
5647 again:
5648 	node = root->inode_tree.rb_node;
5649 	prev = NULL;
5650 	while (node) {
5651 		prev = node;
5652 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5653 
5654 		if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5655 			node = node->rb_left;
5656 		else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5657 			node = node->rb_right;
5658 		else
5659 			break;
5660 	}
5661 	if (!node) {
5662 		while (prev) {
5663 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
5664 			if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5665 				node = prev;
5666 				break;
5667 			}
5668 			prev = rb_next(prev);
5669 		}
5670 	}
5671 	while (node) {
5672 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5673 		objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5674 		inode = igrab(&entry->vfs_inode);
5675 		if (inode) {
5676 			spin_unlock(&root->inode_lock);
5677 			if (atomic_read(&inode->i_count) > 1)
5678 				d_prune_aliases(inode);
5679 			/*
5680 			 * btrfs_drop_inode will have it removed from
5681 			 * the inode cache when its usage count
5682 			 * hits zero.
5683 			 */
5684 			iput(inode);
5685 			cond_resched();
5686 			spin_lock(&root->inode_lock);
5687 			goto again;
5688 		}
5689 
5690 		if (cond_resched_lock(&root->inode_lock))
5691 			goto again;
5692 
5693 		node = rb_next(node);
5694 	}
5695 	spin_unlock(&root->inode_lock);
5696 }
5697 
5698 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5699 {
5700 	struct btrfs_iget_args *args = p;
5701 	inode->i_ino = args->location->objectid;
5702 	memcpy(&BTRFS_I(inode)->location, args->location,
5703 	       sizeof(*args->location));
5704 	BTRFS_I(inode)->root = args->root;
5705 	return 0;
5706 }
5707 
5708 static int btrfs_find_actor(struct inode *inode, void *opaque)
5709 {
5710 	struct btrfs_iget_args *args = opaque;
5711 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5712 		args->root == BTRFS_I(inode)->root;
5713 }
5714 
5715 static struct inode *btrfs_iget_locked(struct super_block *s,
5716 				       struct btrfs_key *location,
5717 				       struct btrfs_root *root)
5718 {
5719 	struct inode *inode;
5720 	struct btrfs_iget_args args;
5721 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5722 
5723 	args.location = location;
5724 	args.root = root;
5725 
5726 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5727 			     btrfs_init_locked_inode,
5728 			     (void *)&args);
5729 	return inode;
5730 }
5731 
5732 /* Get an inode object given its location and corresponding root.
5733  * Returns in *is_new if the inode was read from disk
5734  */
5735 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5736 			 struct btrfs_root *root, int *new)
5737 {
5738 	struct inode *inode;
5739 
5740 	inode = btrfs_iget_locked(s, location, root);
5741 	if (!inode)
5742 		return ERR_PTR(-ENOMEM);
5743 
5744 	if (inode->i_state & I_NEW) {
5745 		int ret;
5746 
5747 		ret = btrfs_read_locked_inode(inode);
5748 		if (!is_bad_inode(inode)) {
5749 			inode_tree_add(inode);
5750 			unlock_new_inode(inode);
5751 			if (new)
5752 				*new = 1;
5753 		} else {
5754 			unlock_new_inode(inode);
5755 			iput(inode);
5756 			ASSERT(ret < 0);
5757 			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5758 		}
5759 	}
5760 
5761 	return inode;
5762 }
5763 
5764 static struct inode *new_simple_dir(struct super_block *s,
5765 				    struct btrfs_key *key,
5766 				    struct btrfs_root *root)
5767 {
5768 	struct inode *inode = new_inode(s);
5769 
5770 	if (!inode)
5771 		return ERR_PTR(-ENOMEM);
5772 
5773 	BTRFS_I(inode)->root = root;
5774 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5775 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5776 
5777 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5778 	inode->i_op = &btrfs_dir_ro_inode_operations;
5779 	inode->i_opflags &= ~IOP_XATTR;
5780 	inode->i_fop = &simple_dir_operations;
5781 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5782 	inode->i_mtime = current_time(inode);
5783 	inode->i_atime = inode->i_mtime;
5784 	inode->i_ctime = inode->i_mtime;
5785 	BTRFS_I(inode)->i_otime = inode->i_mtime;
5786 
5787 	return inode;
5788 }
5789 
5790 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5791 {
5792 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5793 	struct inode *inode;
5794 	struct btrfs_root *root = BTRFS_I(dir)->root;
5795 	struct btrfs_root *sub_root = root;
5796 	struct btrfs_key location;
5797 	int index;
5798 	int ret = 0;
5799 
5800 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5801 		return ERR_PTR(-ENAMETOOLONG);
5802 
5803 	ret = btrfs_inode_by_name(dir, dentry, &location);
5804 	if (ret < 0)
5805 		return ERR_PTR(ret);
5806 
5807 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5808 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5809 		return inode;
5810 	}
5811 
5812 	index = srcu_read_lock(&fs_info->subvol_srcu);
5813 	ret = fixup_tree_root_location(fs_info, dir, dentry,
5814 				       &location, &sub_root);
5815 	if (ret < 0) {
5816 		if (ret != -ENOENT)
5817 			inode = ERR_PTR(ret);
5818 		else
5819 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5820 	} else {
5821 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5822 	}
5823 	srcu_read_unlock(&fs_info->subvol_srcu, index);
5824 
5825 	if (!IS_ERR(inode) && root != sub_root) {
5826 		down_read(&fs_info->cleanup_work_sem);
5827 		if (!sb_rdonly(inode->i_sb))
5828 			ret = btrfs_orphan_cleanup(sub_root);
5829 		up_read(&fs_info->cleanup_work_sem);
5830 		if (ret) {
5831 			iput(inode);
5832 			inode = ERR_PTR(ret);
5833 		}
5834 	}
5835 
5836 	return inode;
5837 }
5838 
5839 static int btrfs_dentry_delete(const struct dentry *dentry)
5840 {
5841 	struct btrfs_root *root;
5842 	struct inode *inode = d_inode(dentry);
5843 
5844 	if (!inode && !IS_ROOT(dentry))
5845 		inode = d_inode(dentry->d_parent);
5846 
5847 	if (inode) {
5848 		root = BTRFS_I(inode)->root;
5849 		if (btrfs_root_refs(&root->root_item) == 0)
5850 			return 1;
5851 
5852 		if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5853 			return 1;
5854 	}
5855 	return 0;
5856 }
5857 
5858 static void btrfs_dentry_release(struct dentry *dentry)
5859 {
5860 	kfree(dentry->d_fsdata);
5861 }
5862 
5863 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5864 				   unsigned int flags)
5865 {
5866 	struct inode *inode;
5867 
5868 	inode = btrfs_lookup_dentry(dir, dentry);
5869 	if (IS_ERR(inode)) {
5870 		if (PTR_ERR(inode) == -ENOENT)
5871 			inode = NULL;
5872 		else
5873 			return ERR_CAST(inode);
5874 	}
5875 
5876 	return d_splice_alias(inode, dentry);
5877 }
5878 
5879 unsigned char btrfs_filetype_table[] = {
5880 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5881 };
5882 
5883 /*
5884  * All this infrastructure exists because dir_emit can fault, and we are holding
5885  * the tree lock when doing readdir.  For now just allocate a buffer and copy
5886  * our information into that, and then dir_emit from the buffer.  This is
5887  * similar to what NFS does, only we don't keep the buffer around in pagecache
5888  * because I'm afraid I'll mess that up.  Long term we need to make filldir do
5889  * copy_to_user_inatomic so we don't have to worry about page faulting under the
5890  * tree lock.
5891  */
5892 static int btrfs_opendir(struct inode *inode, struct file *file)
5893 {
5894 	struct btrfs_file_private *private;
5895 
5896 	private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5897 	if (!private)
5898 		return -ENOMEM;
5899 	private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5900 	if (!private->filldir_buf) {
5901 		kfree(private);
5902 		return -ENOMEM;
5903 	}
5904 	file->private_data = private;
5905 	return 0;
5906 }
5907 
5908 struct dir_entry {
5909 	u64 ino;
5910 	u64 offset;
5911 	unsigned type;
5912 	int name_len;
5913 };
5914 
5915 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5916 {
5917 	while (entries--) {
5918 		struct dir_entry *entry = addr;
5919 		char *name = (char *)(entry + 1);
5920 
5921 		ctx->pos = entry->offset;
5922 		if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5923 			      entry->type))
5924 			return 1;
5925 		addr += sizeof(struct dir_entry) + entry->name_len;
5926 		ctx->pos++;
5927 	}
5928 	return 0;
5929 }
5930 
5931 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5932 {
5933 	struct inode *inode = file_inode(file);
5934 	struct btrfs_root *root = BTRFS_I(inode)->root;
5935 	struct btrfs_file_private *private = file->private_data;
5936 	struct btrfs_dir_item *di;
5937 	struct btrfs_key key;
5938 	struct btrfs_key found_key;
5939 	struct btrfs_path *path;
5940 	void *addr;
5941 	struct list_head ins_list;
5942 	struct list_head del_list;
5943 	int ret;
5944 	struct extent_buffer *leaf;
5945 	int slot;
5946 	char *name_ptr;
5947 	int name_len;
5948 	int entries = 0;
5949 	int total_len = 0;
5950 	bool put = false;
5951 	struct btrfs_key location;
5952 
5953 	if (!dir_emit_dots(file, ctx))
5954 		return 0;
5955 
5956 	path = btrfs_alloc_path();
5957 	if (!path)
5958 		return -ENOMEM;
5959 
5960 	addr = private->filldir_buf;
5961 	path->reada = READA_FORWARD;
5962 
5963 	INIT_LIST_HEAD(&ins_list);
5964 	INIT_LIST_HEAD(&del_list);
5965 	put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5966 
5967 again:
5968 	key.type = BTRFS_DIR_INDEX_KEY;
5969 	key.offset = ctx->pos;
5970 	key.objectid = btrfs_ino(BTRFS_I(inode));
5971 
5972 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5973 	if (ret < 0)
5974 		goto err;
5975 
5976 	while (1) {
5977 		struct dir_entry *entry;
5978 
5979 		leaf = path->nodes[0];
5980 		slot = path->slots[0];
5981 		if (slot >= btrfs_header_nritems(leaf)) {
5982 			ret = btrfs_next_leaf(root, path);
5983 			if (ret < 0)
5984 				goto err;
5985 			else if (ret > 0)
5986 				break;
5987 			continue;
5988 		}
5989 
5990 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5991 
5992 		if (found_key.objectid != key.objectid)
5993 			break;
5994 		if (found_key.type != BTRFS_DIR_INDEX_KEY)
5995 			break;
5996 		if (found_key.offset < ctx->pos)
5997 			goto next;
5998 		if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5999 			goto next;
6000 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6001 		name_len = btrfs_dir_name_len(leaf, di);
6002 		if ((total_len + sizeof(struct dir_entry) + name_len) >=
6003 		    PAGE_SIZE) {
6004 			btrfs_release_path(path);
6005 			ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6006 			if (ret)
6007 				goto nopos;
6008 			addr = private->filldir_buf;
6009 			entries = 0;
6010 			total_len = 0;
6011 			goto again;
6012 		}
6013 
6014 		entry = addr;
6015 		entry->name_len = name_len;
6016 		name_ptr = (char *)(entry + 1);
6017 		read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6018 				   name_len);
6019 		entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6020 		btrfs_dir_item_key_to_cpu(leaf, di, &location);
6021 		entry->ino = location.objectid;
6022 		entry->offset = found_key.offset;
6023 		entries++;
6024 		addr += sizeof(struct dir_entry) + name_len;
6025 		total_len += sizeof(struct dir_entry) + name_len;
6026 next:
6027 		path->slots[0]++;
6028 	}
6029 	btrfs_release_path(path);
6030 
6031 	ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6032 	if (ret)
6033 		goto nopos;
6034 
6035 	ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6036 	if (ret)
6037 		goto nopos;
6038 
6039 	/*
6040 	 * Stop new entries from being returned after we return the last
6041 	 * entry.
6042 	 *
6043 	 * New directory entries are assigned a strictly increasing
6044 	 * offset.  This means that new entries created during readdir
6045 	 * are *guaranteed* to be seen in the future by that readdir.
6046 	 * This has broken buggy programs which operate on names as
6047 	 * they're returned by readdir.  Until we re-use freed offsets
6048 	 * we have this hack to stop new entries from being returned
6049 	 * under the assumption that they'll never reach this huge
6050 	 * offset.
6051 	 *
6052 	 * This is being careful not to overflow 32bit loff_t unless the
6053 	 * last entry requires it because doing so has broken 32bit apps
6054 	 * in the past.
6055 	 */
6056 	if (ctx->pos >= INT_MAX)
6057 		ctx->pos = LLONG_MAX;
6058 	else
6059 		ctx->pos = INT_MAX;
6060 nopos:
6061 	ret = 0;
6062 err:
6063 	if (put)
6064 		btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6065 	btrfs_free_path(path);
6066 	return ret;
6067 }
6068 
6069 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6070 {
6071 	struct btrfs_root *root = BTRFS_I(inode)->root;
6072 	struct btrfs_trans_handle *trans;
6073 	int ret = 0;
6074 	bool nolock = false;
6075 
6076 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6077 		return 0;
6078 
6079 	if (btrfs_fs_closing(root->fs_info) &&
6080 			btrfs_is_free_space_inode(BTRFS_I(inode)))
6081 		nolock = true;
6082 
6083 	if (wbc->sync_mode == WB_SYNC_ALL) {
6084 		if (nolock)
6085 			trans = btrfs_join_transaction_nolock(root);
6086 		else
6087 			trans = btrfs_join_transaction(root);
6088 		if (IS_ERR(trans))
6089 			return PTR_ERR(trans);
6090 		ret = btrfs_commit_transaction(trans);
6091 	}
6092 	return ret;
6093 }
6094 
6095 /*
6096  * This is somewhat expensive, updating the tree every time the
6097  * inode changes.  But, it is most likely to find the inode in cache.
6098  * FIXME, needs more benchmarking...there are no reasons other than performance
6099  * to keep or drop this code.
6100  */
6101 static int btrfs_dirty_inode(struct inode *inode)
6102 {
6103 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6104 	struct btrfs_root *root = BTRFS_I(inode)->root;
6105 	struct btrfs_trans_handle *trans;
6106 	int ret;
6107 
6108 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6109 		return 0;
6110 
6111 	trans = btrfs_join_transaction(root);
6112 	if (IS_ERR(trans))
6113 		return PTR_ERR(trans);
6114 
6115 	ret = btrfs_update_inode(trans, root, inode);
6116 	if (ret && ret == -ENOSPC) {
6117 		/* whoops, lets try again with the full transaction */
6118 		btrfs_end_transaction(trans);
6119 		trans = btrfs_start_transaction(root, 1);
6120 		if (IS_ERR(trans))
6121 			return PTR_ERR(trans);
6122 
6123 		ret = btrfs_update_inode(trans, root, inode);
6124 	}
6125 	btrfs_end_transaction(trans);
6126 	if (BTRFS_I(inode)->delayed_node)
6127 		btrfs_balance_delayed_items(fs_info);
6128 
6129 	return ret;
6130 }
6131 
6132 /*
6133  * This is a copy of file_update_time.  We need this so we can return error on
6134  * ENOSPC for updating the inode in the case of file write and mmap writes.
6135  */
6136 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6137 			     int flags)
6138 {
6139 	struct btrfs_root *root = BTRFS_I(inode)->root;
6140 	bool dirty = flags & ~S_VERSION;
6141 
6142 	if (btrfs_root_readonly(root))
6143 		return -EROFS;
6144 
6145 	if (flags & S_VERSION)
6146 		dirty |= inode_maybe_inc_iversion(inode, dirty);
6147 	if (flags & S_CTIME)
6148 		inode->i_ctime = *now;
6149 	if (flags & S_MTIME)
6150 		inode->i_mtime = *now;
6151 	if (flags & S_ATIME)
6152 		inode->i_atime = *now;
6153 	return dirty ? btrfs_dirty_inode(inode) : 0;
6154 }
6155 
6156 /*
6157  * find the highest existing sequence number in a directory
6158  * and then set the in-memory index_cnt variable to reflect
6159  * free sequence numbers
6160  */
6161 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6162 {
6163 	struct btrfs_root *root = inode->root;
6164 	struct btrfs_key key, found_key;
6165 	struct btrfs_path *path;
6166 	struct extent_buffer *leaf;
6167 	int ret;
6168 
6169 	key.objectid = btrfs_ino(inode);
6170 	key.type = BTRFS_DIR_INDEX_KEY;
6171 	key.offset = (u64)-1;
6172 
6173 	path = btrfs_alloc_path();
6174 	if (!path)
6175 		return -ENOMEM;
6176 
6177 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6178 	if (ret < 0)
6179 		goto out;
6180 	/* FIXME: we should be able to handle this */
6181 	if (ret == 0)
6182 		goto out;
6183 	ret = 0;
6184 
6185 	/*
6186 	 * MAGIC NUMBER EXPLANATION:
6187 	 * since we search a directory based on f_pos we have to start at 2
6188 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6189 	 * else has to start at 2
6190 	 */
6191 	if (path->slots[0] == 0) {
6192 		inode->index_cnt = 2;
6193 		goto out;
6194 	}
6195 
6196 	path->slots[0]--;
6197 
6198 	leaf = path->nodes[0];
6199 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6200 
6201 	if (found_key.objectid != btrfs_ino(inode) ||
6202 	    found_key.type != BTRFS_DIR_INDEX_KEY) {
6203 		inode->index_cnt = 2;
6204 		goto out;
6205 	}
6206 
6207 	inode->index_cnt = found_key.offset + 1;
6208 out:
6209 	btrfs_free_path(path);
6210 	return ret;
6211 }
6212 
6213 /*
6214  * helper to find a free sequence number in a given directory.  This current
6215  * code is very simple, later versions will do smarter things in the btree
6216  */
6217 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6218 {
6219 	int ret = 0;
6220 
6221 	if (dir->index_cnt == (u64)-1) {
6222 		ret = btrfs_inode_delayed_dir_index_count(dir);
6223 		if (ret) {
6224 			ret = btrfs_set_inode_index_count(dir);
6225 			if (ret)
6226 				return ret;
6227 		}
6228 	}
6229 
6230 	*index = dir->index_cnt;
6231 	dir->index_cnt++;
6232 
6233 	return ret;
6234 }
6235 
6236 static int btrfs_insert_inode_locked(struct inode *inode)
6237 {
6238 	struct btrfs_iget_args args;
6239 	args.location = &BTRFS_I(inode)->location;
6240 	args.root = BTRFS_I(inode)->root;
6241 
6242 	return insert_inode_locked4(inode,
6243 		   btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6244 		   btrfs_find_actor, &args);
6245 }
6246 
6247 /*
6248  * Inherit flags from the parent inode.
6249  *
6250  * Currently only the compression flags and the cow flags are inherited.
6251  */
6252 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6253 {
6254 	unsigned int flags;
6255 
6256 	if (!dir)
6257 		return;
6258 
6259 	flags = BTRFS_I(dir)->flags;
6260 
6261 	if (flags & BTRFS_INODE_NOCOMPRESS) {
6262 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6263 		BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6264 	} else if (flags & BTRFS_INODE_COMPRESS) {
6265 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6266 		BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6267 	}
6268 
6269 	if (flags & BTRFS_INODE_NODATACOW) {
6270 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6271 		if (S_ISREG(inode->i_mode))
6272 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6273 	}
6274 
6275 	btrfs_update_iflags(inode);
6276 }
6277 
6278 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6279 				     struct btrfs_root *root,
6280 				     struct inode *dir,
6281 				     const char *name, int name_len,
6282 				     u64 ref_objectid, u64 objectid,
6283 				     umode_t mode, u64 *index)
6284 {
6285 	struct btrfs_fs_info *fs_info = root->fs_info;
6286 	struct inode *inode;
6287 	struct btrfs_inode_item *inode_item;
6288 	struct btrfs_key *location;
6289 	struct btrfs_path *path;
6290 	struct btrfs_inode_ref *ref;
6291 	struct btrfs_key key[2];
6292 	u32 sizes[2];
6293 	int nitems = name ? 2 : 1;
6294 	unsigned long ptr;
6295 	int ret;
6296 
6297 	path = btrfs_alloc_path();
6298 	if (!path)
6299 		return ERR_PTR(-ENOMEM);
6300 
6301 	inode = new_inode(fs_info->sb);
6302 	if (!inode) {
6303 		btrfs_free_path(path);
6304 		return ERR_PTR(-ENOMEM);
6305 	}
6306 
6307 	/*
6308 	 * O_TMPFILE, set link count to 0, so that after this point,
6309 	 * we fill in an inode item with the correct link count.
6310 	 */
6311 	if (!name)
6312 		set_nlink(inode, 0);
6313 
6314 	/*
6315 	 * we have to initialize this early, so we can reclaim the inode
6316 	 * number if we fail afterwards in this function.
6317 	 */
6318 	inode->i_ino = objectid;
6319 
6320 	if (dir && name) {
6321 		trace_btrfs_inode_request(dir);
6322 
6323 		ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6324 		if (ret) {
6325 			btrfs_free_path(path);
6326 			iput(inode);
6327 			return ERR_PTR(ret);
6328 		}
6329 	} else if (dir) {
6330 		*index = 0;
6331 	}
6332 	/*
6333 	 * index_cnt is ignored for everything but a dir,
6334 	 * btrfs_set_inode_index_count has an explanation for the magic
6335 	 * number
6336 	 */
6337 	BTRFS_I(inode)->index_cnt = 2;
6338 	BTRFS_I(inode)->dir_index = *index;
6339 	BTRFS_I(inode)->root = root;
6340 	BTRFS_I(inode)->generation = trans->transid;
6341 	inode->i_generation = BTRFS_I(inode)->generation;
6342 
6343 	/*
6344 	 * We could have gotten an inode number from somebody who was fsynced
6345 	 * and then removed in this same transaction, so let's just set full
6346 	 * sync since it will be a full sync anyway and this will blow away the
6347 	 * old info in the log.
6348 	 */
6349 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6350 
6351 	key[0].objectid = objectid;
6352 	key[0].type = BTRFS_INODE_ITEM_KEY;
6353 	key[0].offset = 0;
6354 
6355 	sizes[0] = sizeof(struct btrfs_inode_item);
6356 
6357 	if (name) {
6358 		/*
6359 		 * Start new inodes with an inode_ref. This is slightly more
6360 		 * efficient for small numbers of hard links since they will
6361 		 * be packed into one item. Extended refs will kick in if we
6362 		 * add more hard links than can fit in the ref item.
6363 		 */
6364 		key[1].objectid = objectid;
6365 		key[1].type = BTRFS_INODE_REF_KEY;
6366 		key[1].offset = ref_objectid;
6367 
6368 		sizes[1] = name_len + sizeof(*ref);
6369 	}
6370 
6371 	location = &BTRFS_I(inode)->location;
6372 	location->objectid = objectid;
6373 	location->offset = 0;
6374 	location->type = BTRFS_INODE_ITEM_KEY;
6375 
6376 	ret = btrfs_insert_inode_locked(inode);
6377 	if (ret < 0)
6378 		goto fail;
6379 
6380 	path->leave_spinning = 1;
6381 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6382 	if (ret != 0)
6383 		goto fail_unlock;
6384 
6385 	inode_init_owner(inode, dir, mode);
6386 	inode_set_bytes(inode, 0);
6387 
6388 	inode->i_mtime = current_time(inode);
6389 	inode->i_atime = inode->i_mtime;
6390 	inode->i_ctime = inode->i_mtime;
6391 	BTRFS_I(inode)->i_otime = inode->i_mtime;
6392 
6393 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6394 				  struct btrfs_inode_item);
6395 	memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6396 			     sizeof(*inode_item));
6397 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
6398 
6399 	if (name) {
6400 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6401 				     struct btrfs_inode_ref);
6402 		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6403 		btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6404 		ptr = (unsigned long)(ref + 1);
6405 		write_extent_buffer(path->nodes[0], name, ptr, name_len);
6406 	}
6407 
6408 	btrfs_mark_buffer_dirty(path->nodes[0]);
6409 	btrfs_free_path(path);
6410 
6411 	btrfs_inherit_iflags(inode, dir);
6412 
6413 	if (S_ISREG(mode)) {
6414 		if (btrfs_test_opt(fs_info, NODATASUM))
6415 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6416 		if (btrfs_test_opt(fs_info, NODATACOW))
6417 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6418 				BTRFS_INODE_NODATASUM;
6419 	}
6420 
6421 	inode_tree_add(inode);
6422 
6423 	trace_btrfs_inode_new(inode);
6424 	btrfs_set_inode_last_trans(trans, inode);
6425 
6426 	btrfs_update_root_times(trans, root);
6427 
6428 	ret = btrfs_inode_inherit_props(trans, inode, dir);
6429 	if (ret)
6430 		btrfs_err(fs_info,
6431 			  "error inheriting props for ino %llu (root %llu): %d",
6432 			btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6433 
6434 	return inode;
6435 
6436 fail_unlock:
6437 	unlock_new_inode(inode);
6438 fail:
6439 	if (dir && name)
6440 		BTRFS_I(dir)->index_cnt--;
6441 	btrfs_free_path(path);
6442 	iput(inode);
6443 	return ERR_PTR(ret);
6444 }
6445 
6446 static inline u8 btrfs_inode_type(struct inode *inode)
6447 {
6448 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6449 }
6450 
6451 /*
6452  * utility function to add 'inode' into 'parent_inode' with
6453  * a give name and a given sequence number.
6454  * if 'add_backref' is true, also insert a backref from the
6455  * inode to the parent directory.
6456  */
6457 int btrfs_add_link(struct btrfs_trans_handle *trans,
6458 		   struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6459 		   const char *name, int name_len, int add_backref, u64 index)
6460 {
6461 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6462 	int ret = 0;
6463 	struct btrfs_key key;
6464 	struct btrfs_root *root = parent_inode->root;
6465 	u64 ino = btrfs_ino(inode);
6466 	u64 parent_ino = btrfs_ino(parent_inode);
6467 
6468 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6469 		memcpy(&key, &inode->root->root_key, sizeof(key));
6470 	} else {
6471 		key.objectid = ino;
6472 		key.type = BTRFS_INODE_ITEM_KEY;
6473 		key.offset = 0;
6474 	}
6475 
6476 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6477 		ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6478 					 root->root_key.objectid, parent_ino,
6479 					 index, name, name_len);
6480 	} else if (add_backref) {
6481 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6482 					     parent_ino, index);
6483 	}
6484 
6485 	/* Nothing to clean up yet */
6486 	if (ret)
6487 		return ret;
6488 
6489 	ret = btrfs_insert_dir_item(trans, root, name, name_len,
6490 				    parent_inode, &key,
6491 				    btrfs_inode_type(&inode->vfs_inode), index);
6492 	if (ret == -EEXIST || ret == -EOVERFLOW)
6493 		goto fail_dir_item;
6494 	else if (ret) {
6495 		btrfs_abort_transaction(trans, ret);
6496 		return ret;
6497 	}
6498 
6499 	btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6500 			   name_len * 2);
6501 	inode_inc_iversion(&parent_inode->vfs_inode);
6502 	parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6503 		current_time(&parent_inode->vfs_inode);
6504 	ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6505 	if (ret)
6506 		btrfs_abort_transaction(trans, ret);
6507 	return ret;
6508 
6509 fail_dir_item:
6510 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6511 		u64 local_index;
6512 		int err;
6513 		err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6514 					 root->root_key.objectid, parent_ino,
6515 					 &local_index, name, name_len);
6516 
6517 	} else if (add_backref) {
6518 		u64 local_index;
6519 		int err;
6520 
6521 		err = btrfs_del_inode_ref(trans, root, name, name_len,
6522 					  ino, parent_ino, &local_index);
6523 	}
6524 	return ret;
6525 }
6526 
6527 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6528 			    struct btrfs_inode *dir, struct dentry *dentry,
6529 			    struct btrfs_inode *inode, int backref, u64 index)
6530 {
6531 	int err = btrfs_add_link(trans, dir, inode,
6532 				 dentry->d_name.name, dentry->d_name.len,
6533 				 backref, index);
6534 	if (err > 0)
6535 		err = -EEXIST;
6536 	return err;
6537 }
6538 
6539 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6540 			umode_t mode, dev_t rdev)
6541 {
6542 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6543 	struct btrfs_trans_handle *trans;
6544 	struct btrfs_root *root = BTRFS_I(dir)->root;
6545 	struct inode *inode = NULL;
6546 	int err;
6547 	int drop_inode = 0;
6548 	u64 objectid;
6549 	u64 index = 0;
6550 
6551 	/*
6552 	 * 2 for inode item and ref
6553 	 * 2 for dir items
6554 	 * 1 for xattr if selinux is on
6555 	 */
6556 	trans = btrfs_start_transaction(root, 5);
6557 	if (IS_ERR(trans))
6558 		return PTR_ERR(trans);
6559 
6560 	err = btrfs_find_free_ino(root, &objectid);
6561 	if (err)
6562 		goto out_unlock;
6563 
6564 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6565 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6566 			mode, &index);
6567 	if (IS_ERR(inode)) {
6568 		err = PTR_ERR(inode);
6569 		goto out_unlock;
6570 	}
6571 
6572 	/*
6573 	* If the active LSM wants to access the inode during
6574 	* d_instantiate it needs these. Smack checks to see
6575 	* if the filesystem supports xattrs by looking at the
6576 	* ops vector.
6577 	*/
6578 	inode->i_op = &btrfs_special_inode_operations;
6579 	init_special_inode(inode, inode->i_mode, rdev);
6580 
6581 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6582 	if (err)
6583 		goto out_unlock_inode;
6584 
6585 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6586 			0, index);
6587 	if (err) {
6588 		goto out_unlock_inode;
6589 	} else {
6590 		btrfs_update_inode(trans, root, inode);
6591 		unlock_new_inode(inode);
6592 		d_instantiate(dentry, inode);
6593 	}
6594 
6595 out_unlock:
6596 	btrfs_end_transaction(trans);
6597 	btrfs_btree_balance_dirty(fs_info);
6598 	if (drop_inode) {
6599 		inode_dec_link_count(inode);
6600 		iput(inode);
6601 	}
6602 	return err;
6603 
6604 out_unlock_inode:
6605 	drop_inode = 1;
6606 	unlock_new_inode(inode);
6607 	goto out_unlock;
6608 
6609 }
6610 
6611 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6612 			umode_t mode, bool excl)
6613 {
6614 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6615 	struct btrfs_trans_handle *trans;
6616 	struct btrfs_root *root = BTRFS_I(dir)->root;
6617 	struct inode *inode = NULL;
6618 	int drop_inode_on_err = 0;
6619 	int err;
6620 	u64 objectid;
6621 	u64 index = 0;
6622 
6623 	/*
6624 	 * 2 for inode item and ref
6625 	 * 2 for dir items
6626 	 * 1 for xattr if selinux is on
6627 	 */
6628 	trans = btrfs_start_transaction(root, 5);
6629 	if (IS_ERR(trans))
6630 		return PTR_ERR(trans);
6631 
6632 	err = btrfs_find_free_ino(root, &objectid);
6633 	if (err)
6634 		goto out_unlock;
6635 
6636 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6637 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6638 			mode, &index);
6639 	if (IS_ERR(inode)) {
6640 		err = PTR_ERR(inode);
6641 		goto out_unlock;
6642 	}
6643 	drop_inode_on_err = 1;
6644 	/*
6645 	* If the active LSM wants to access the inode during
6646 	* d_instantiate it needs these. Smack checks to see
6647 	* if the filesystem supports xattrs by looking at the
6648 	* ops vector.
6649 	*/
6650 	inode->i_fop = &btrfs_file_operations;
6651 	inode->i_op = &btrfs_file_inode_operations;
6652 	inode->i_mapping->a_ops = &btrfs_aops;
6653 
6654 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6655 	if (err)
6656 		goto out_unlock_inode;
6657 
6658 	err = btrfs_update_inode(trans, root, inode);
6659 	if (err)
6660 		goto out_unlock_inode;
6661 
6662 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6663 			0, index);
6664 	if (err)
6665 		goto out_unlock_inode;
6666 
6667 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6668 	unlock_new_inode(inode);
6669 	d_instantiate(dentry, inode);
6670 
6671 out_unlock:
6672 	btrfs_end_transaction(trans);
6673 	if (err && drop_inode_on_err) {
6674 		inode_dec_link_count(inode);
6675 		iput(inode);
6676 	}
6677 	btrfs_btree_balance_dirty(fs_info);
6678 	return err;
6679 
6680 out_unlock_inode:
6681 	unlock_new_inode(inode);
6682 	goto out_unlock;
6683 
6684 }
6685 
6686 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6687 		      struct dentry *dentry)
6688 {
6689 	struct btrfs_trans_handle *trans = NULL;
6690 	struct btrfs_root *root = BTRFS_I(dir)->root;
6691 	struct inode *inode = d_inode(old_dentry);
6692 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6693 	u64 index;
6694 	int err;
6695 	int drop_inode = 0;
6696 
6697 	/* do not allow sys_link's with other subvols of the same device */
6698 	if (root->objectid != BTRFS_I(inode)->root->objectid)
6699 		return -EXDEV;
6700 
6701 	if (inode->i_nlink >= BTRFS_LINK_MAX)
6702 		return -EMLINK;
6703 
6704 	err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6705 	if (err)
6706 		goto fail;
6707 
6708 	/*
6709 	 * 2 items for inode and inode ref
6710 	 * 2 items for dir items
6711 	 * 1 item for parent inode
6712 	 */
6713 	trans = btrfs_start_transaction(root, 5);
6714 	if (IS_ERR(trans)) {
6715 		err = PTR_ERR(trans);
6716 		trans = NULL;
6717 		goto fail;
6718 	}
6719 
6720 	/* There are several dir indexes for this inode, clear the cache. */
6721 	BTRFS_I(inode)->dir_index = 0ULL;
6722 	inc_nlink(inode);
6723 	inode_inc_iversion(inode);
6724 	inode->i_ctime = current_time(inode);
6725 	ihold(inode);
6726 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6727 
6728 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6729 			1, index);
6730 
6731 	if (err) {
6732 		drop_inode = 1;
6733 	} else {
6734 		struct dentry *parent = dentry->d_parent;
6735 		err = btrfs_update_inode(trans, root, inode);
6736 		if (err)
6737 			goto fail;
6738 		if (inode->i_nlink == 1) {
6739 			/*
6740 			 * If new hard link count is 1, it's a file created
6741 			 * with open(2) O_TMPFILE flag.
6742 			 */
6743 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
6744 			if (err)
6745 				goto fail;
6746 		}
6747 		d_instantiate(dentry, inode);
6748 		btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6749 	}
6750 
6751 fail:
6752 	if (trans)
6753 		btrfs_end_transaction(trans);
6754 	if (drop_inode) {
6755 		inode_dec_link_count(inode);
6756 		iput(inode);
6757 	}
6758 	btrfs_btree_balance_dirty(fs_info);
6759 	return err;
6760 }
6761 
6762 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6763 {
6764 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6765 	struct inode *inode = NULL;
6766 	struct btrfs_trans_handle *trans;
6767 	struct btrfs_root *root = BTRFS_I(dir)->root;
6768 	int err = 0;
6769 	int drop_on_err = 0;
6770 	u64 objectid = 0;
6771 	u64 index = 0;
6772 
6773 	/*
6774 	 * 2 items for inode and ref
6775 	 * 2 items for dir items
6776 	 * 1 for xattr if selinux is on
6777 	 */
6778 	trans = btrfs_start_transaction(root, 5);
6779 	if (IS_ERR(trans))
6780 		return PTR_ERR(trans);
6781 
6782 	err = btrfs_find_free_ino(root, &objectid);
6783 	if (err)
6784 		goto out_fail;
6785 
6786 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6787 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6788 			S_IFDIR | mode, &index);
6789 	if (IS_ERR(inode)) {
6790 		err = PTR_ERR(inode);
6791 		goto out_fail;
6792 	}
6793 
6794 	drop_on_err = 1;
6795 	/* these must be set before we unlock the inode */
6796 	inode->i_op = &btrfs_dir_inode_operations;
6797 	inode->i_fop = &btrfs_dir_file_operations;
6798 
6799 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6800 	if (err)
6801 		goto out_fail_inode;
6802 
6803 	btrfs_i_size_write(BTRFS_I(inode), 0);
6804 	err = btrfs_update_inode(trans, root, inode);
6805 	if (err)
6806 		goto out_fail_inode;
6807 
6808 	err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6809 			dentry->d_name.name,
6810 			dentry->d_name.len, 0, index);
6811 	if (err)
6812 		goto out_fail_inode;
6813 
6814 	d_instantiate(dentry, inode);
6815 	/*
6816 	 * mkdir is special.  We're unlocking after we call d_instantiate
6817 	 * to avoid a race with nfsd calling d_instantiate.
6818 	 */
6819 	unlock_new_inode(inode);
6820 	drop_on_err = 0;
6821 
6822 out_fail:
6823 	btrfs_end_transaction(trans);
6824 	if (drop_on_err) {
6825 		inode_dec_link_count(inode);
6826 		iput(inode);
6827 	}
6828 	btrfs_btree_balance_dirty(fs_info);
6829 	return err;
6830 
6831 out_fail_inode:
6832 	unlock_new_inode(inode);
6833 	goto out_fail;
6834 }
6835 
6836 static noinline int uncompress_inline(struct btrfs_path *path,
6837 				      struct page *page,
6838 				      size_t pg_offset, u64 extent_offset,
6839 				      struct btrfs_file_extent_item *item)
6840 {
6841 	int ret;
6842 	struct extent_buffer *leaf = path->nodes[0];
6843 	char *tmp;
6844 	size_t max_size;
6845 	unsigned long inline_size;
6846 	unsigned long ptr;
6847 	int compress_type;
6848 
6849 	WARN_ON(pg_offset != 0);
6850 	compress_type = btrfs_file_extent_compression(leaf, item);
6851 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6852 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6853 					btrfs_item_nr(path->slots[0]));
6854 	tmp = kmalloc(inline_size, GFP_NOFS);
6855 	if (!tmp)
6856 		return -ENOMEM;
6857 	ptr = btrfs_file_extent_inline_start(item);
6858 
6859 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6860 
6861 	max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6862 	ret = btrfs_decompress(compress_type, tmp, page,
6863 			       extent_offset, inline_size, max_size);
6864 
6865 	/*
6866 	 * decompression code contains a memset to fill in any space between the end
6867 	 * of the uncompressed data and the end of max_size in case the decompressed
6868 	 * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6869 	 * the end of an inline extent and the beginning of the next block, so we
6870 	 * cover that region here.
6871 	 */
6872 
6873 	if (max_size + pg_offset < PAGE_SIZE) {
6874 		char *map = kmap(page);
6875 		memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6876 		kunmap(page);
6877 	}
6878 	kfree(tmp);
6879 	return ret;
6880 }
6881 
6882 /*
6883  * a bit scary, this does extent mapping from logical file offset to the disk.
6884  * the ugly parts come from merging extents from the disk with the in-ram
6885  * representation.  This gets more complex because of the data=ordered code,
6886  * where the in-ram extents might be locked pending data=ordered completion.
6887  *
6888  * This also copies inline extents directly into the page.
6889  */
6890 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6891 		struct page *page,
6892 	    size_t pg_offset, u64 start, u64 len,
6893 		int create)
6894 {
6895 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6896 	int ret;
6897 	int err = 0;
6898 	u64 extent_start = 0;
6899 	u64 extent_end = 0;
6900 	u64 objectid = btrfs_ino(inode);
6901 	u32 found_type;
6902 	struct btrfs_path *path = NULL;
6903 	struct btrfs_root *root = inode->root;
6904 	struct btrfs_file_extent_item *item;
6905 	struct extent_buffer *leaf;
6906 	struct btrfs_key found_key;
6907 	struct extent_map *em = NULL;
6908 	struct extent_map_tree *em_tree = &inode->extent_tree;
6909 	struct extent_io_tree *io_tree = &inode->io_tree;
6910 	const bool new_inline = !page || create;
6911 
6912 	read_lock(&em_tree->lock);
6913 	em = lookup_extent_mapping(em_tree, start, len);
6914 	if (em)
6915 		em->bdev = fs_info->fs_devices->latest_bdev;
6916 	read_unlock(&em_tree->lock);
6917 
6918 	if (em) {
6919 		if (em->start > start || em->start + em->len <= start)
6920 			free_extent_map(em);
6921 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6922 			free_extent_map(em);
6923 		else
6924 			goto out;
6925 	}
6926 	em = alloc_extent_map();
6927 	if (!em) {
6928 		err = -ENOMEM;
6929 		goto out;
6930 	}
6931 	em->bdev = fs_info->fs_devices->latest_bdev;
6932 	em->start = EXTENT_MAP_HOLE;
6933 	em->orig_start = EXTENT_MAP_HOLE;
6934 	em->len = (u64)-1;
6935 	em->block_len = (u64)-1;
6936 
6937 	if (!path) {
6938 		path = btrfs_alloc_path();
6939 		if (!path) {
6940 			err = -ENOMEM;
6941 			goto out;
6942 		}
6943 		/*
6944 		 * Chances are we'll be called again, so go ahead and do
6945 		 * readahead
6946 		 */
6947 		path->reada = READA_FORWARD;
6948 	}
6949 
6950 	ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6951 	if (ret < 0) {
6952 		err = ret;
6953 		goto out;
6954 	}
6955 
6956 	if (ret != 0) {
6957 		if (path->slots[0] == 0)
6958 			goto not_found;
6959 		path->slots[0]--;
6960 	}
6961 
6962 	leaf = path->nodes[0];
6963 	item = btrfs_item_ptr(leaf, path->slots[0],
6964 			      struct btrfs_file_extent_item);
6965 	/* are we inside the extent that was found? */
6966 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6967 	found_type = found_key.type;
6968 	if (found_key.objectid != objectid ||
6969 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6970 		/*
6971 		 * If we backup past the first extent we want to move forward
6972 		 * and see if there is an extent in front of us, otherwise we'll
6973 		 * say there is a hole for our whole search range which can
6974 		 * cause problems.
6975 		 */
6976 		extent_end = start;
6977 		goto next;
6978 	}
6979 
6980 	found_type = btrfs_file_extent_type(leaf, item);
6981 	extent_start = found_key.offset;
6982 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6983 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6984 		extent_end = extent_start +
6985 		       btrfs_file_extent_num_bytes(leaf, item);
6986 
6987 		trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6988 						       extent_start);
6989 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6990 		size_t size;
6991 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6992 		extent_end = ALIGN(extent_start + size,
6993 				   fs_info->sectorsize);
6994 
6995 		trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6996 						      path->slots[0],
6997 						      extent_start);
6998 	}
6999 next:
7000 	if (start >= extent_end) {
7001 		path->slots[0]++;
7002 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7003 			ret = btrfs_next_leaf(root, path);
7004 			if (ret < 0) {
7005 				err = ret;
7006 				goto out;
7007 			}
7008 			if (ret > 0)
7009 				goto not_found;
7010 			leaf = path->nodes[0];
7011 		}
7012 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7013 		if (found_key.objectid != objectid ||
7014 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
7015 			goto not_found;
7016 		if (start + len <= found_key.offset)
7017 			goto not_found;
7018 		if (start > found_key.offset)
7019 			goto next;
7020 		em->start = start;
7021 		em->orig_start = start;
7022 		em->len = found_key.offset - start;
7023 		goto not_found_em;
7024 	}
7025 
7026 	btrfs_extent_item_to_extent_map(inode, path, item,
7027 			new_inline, em);
7028 
7029 	if (found_type == BTRFS_FILE_EXTENT_REG ||
7030 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7031 		goto insert;
7032 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7033 		unsigned long ptr;
7034 		char *map;
7035 		size_t size;
7036 		size_t extent_offset;
7037 		size_t copy_size;
7038 
7039 		if (new_inline)
7040 			goto out;
7041 
7042 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7043 		extent_offset = page_offset(page) + pg_offset - extent_start;
7044 		copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7045 				  size - extent_offset);
7046 		em->start = extent_start + extent_offset;
7047 		em->len = ALIGN(copy_size, fs_info->sectorsize);
7048 		em->orig_block_len = em->len;
7049 		em->orig_start = em->start;
7050 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7051 		if (!PageUptodate(page)) {
7052 			if (btrfs_file_extent_compression(leaf, item) !=
7053 			    BTRFS_COMPRESS_NONE) {
7054 				ret = uncompress_inline(path, page, pg_offset,
7055 							extent_offset, item);
7056 				if (ret) {
7057 					err = ret;
7058 					goto out;
7059 				}
7060 			} else {
7061 				map = kmap(page);
7062 				read_extent_buffer(leaf, map + pg_offset, ptr,
7063 						   copy_size);
7064 				if (pg_offset + copy_size < PAGE_SIZE) {
7065 					memset(map + pg_offset + copy_size, 0,
7066 					       PAGE_SIZE - pg_offset -
7067 					       copy_size);
7068 				}
7069 				kunmap(page);
7070 			}
7071 			flush_dcache_page(page);
7072 		}
7073 		set_extent_uptodate(io_tree, em->start,
7074 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
7075 		goto insert;
7076 	}
7077 not_found:
7078 	em->start = start;
7079 	em->orig_start = start;
7080 	em->len = len;
7081 not_found_em:
7082 	em->block_start = EXTENT_MAP_HOLE;
7083 insert:
7084 	btrfs_release_path(path);
7085 	if (em->start > start || extent_map_end(em) <= start) {
7086 		btrfs_err(fs_info,
7087 			  "bad extent! em: [%llu %llu] passed [%llu %llu]",
7088 			  em->start, em->len, start, len);
7089 		err = -EIO;
7090 		goto out;
7091 	}
7092 
7093 	err = 0;
7094 	write_lock(&em_tree->lock);
7095 	err = btrfs_add_extent_mapping(em_tree, &em, start, len);
7096 	write_unlock(&em_tree->lock);
7097 out:
7098 
7099 	trace_btrfs_get_extent(root, inode, em);
7100 
7101 	btrfs_free_path(path);
7102 	if (err) {
7103 		free_extent_map(em);
7104 		return ERR_PTR(err);
7105 	}
7106 	BUG_ON(!em); /* Error is always set */
7107 	return em;
7108 }
7109 
7110 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7111 		struct page *page,
7112 		size_t pg_offset, u64 start, u64 len,
7113 		int create)
7114 {
7115 	struct extent_map *em;
7116 	struct extent_map *hole_em = NULL;
7117 	u64 range_start = start;
7118 	u64 end;
7119 	u64 found;
7120 	u64 found_end;
7121 	int err = 0;
7122 
7123 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7124 	if (IS_ERR(em))
7125 		return em;
7126 	/*
7127 	 * If our em maps to:
7128 	 * - a hole or
7129 	 * - a pre-alloc extent,
7130 	 * there might actually be delalloc bytes behind it.
7131 	 */
7132 	if (em->block_start != EXTENT_MAP_HOLE &&
7133 	    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7134 		return em;
7135 	else
7136 		hole_em = em;
7137 
7138 	/* check to see if we've wrapped (len == -1 or similar) */
7139 	end = start + len;
7140 	if (end < start)
7141 		end = (u64)-1;
7142 	else
7143 		end -= 1;
7144 
7145 	em = NULL;
7146 
7147 	/* ok, we didn't find anything, lets look for delalloc */
7148 	found = count_range_bits(&inode->io_tree, &range_start,
7149 				 end, len, EXTENT_DELALLOC, 1);
7150 	found_end = range_start + found;
7151 	if (found_end < range_start)
7152 		found_end = (u64)-1;
7153 
7154 	/*
7155 	 * we didn't find anything useful, return
7156 	 * the original results from get_extent()
7157 	 */
7158 	if (range_start > end || found_end <= start) {
7159 		em = hole_em;
7160 		hole_em = NULL;
7161 		goto out;
7162 	}
7163 
7164 	/* adjust the range_start to make sure it doesn't
7165 	 * go backwards from the start they passed in
7166 	 */
7167 	range_start = max(start, range_start);
7168 	found = found_end - range_start;
7169 
7170 	if (found > 0) {
7171 		u64 hole_start = start;
7172 		u64 hole_len = len;
7173 
7174 		em = alloc_extent_map();
7175 		if (!em) {
7176 			err = -ENOMEM;
7177 			goto out;
7178 		}
7179 		/*
7180 		 * when btrfs_get_extent can't find anything it
7181 		 * returns one huge hole
7182 		 *
7183 		 * make sure what it found really fits our range, and
7184 		 * adjust to make sure it is based on the start from
7185 		 * the caller
7186 		 */
7187 		if (hole_em) {
7188 			u64 calc_end = extent_map_end(hole_em);
7189 
7190 			if (calc_end <= start || (hole_em->start > end)) {
7191 				free_extent_map(hole_em);
7192 				hole_em = NULL;
7193 			} else {
7194 				hole_start = max(hole_em->start, start);
7195 				hole_len = calc_end - hole_start;
7196 			}
7197 		}
7198 		em->bdev = NULL;
7199 		if (hole_em && range_start > hole_start) {
7200 			/* our hole starts before our delalloc, so we
7201 			 * have to return just the parts of the hole
7202 			 * that go until  the delalloc starts
7203 			 */
7204 			em->len = min(hole_len,
7205 				      range_start - hole_start);
7206 			em->start = hole_start;
7207 			em->orig_start = hole_start;
7208 			/*
7209 			 * don't adjust block start at all,
7210 			 * it is fixed at EXTENT_MAP_HOLE
7211 			 */
7212 			em->block_start = hole_em->block_start;
7213 			em->block_len = hole_len;
7214 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7215 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7216 		} else {
7217 			em->start = range_start;
7218 			em->len = found;
7219 			em->orig_start = range_start;
7220 			em->block_start = EXTENT_MAP_DELALLOC;
7221 			em->block_len = found;
7222 		}
7223 	} else {
7224 		return hole_em;
7225 	}
7226 out:
7227 
7228 	free_extent_map(hole_em);
7229 	if (err) {
7230 		free_extent_map(em);
7231 		return ERR_PTR(err);
7232 	}
7233 	return em;
7234 }
7235 
7236 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7237 						  const u64 start,
7238 						  const u64 len,
7239 						  const u64 orig_start,
7240 						  const u64 block_start,
7241 						  const u64 block_len,
7242 						  const u64 orig_block_len,
7243 						  const u64 ram_bytes,
7244 						  const int type)
7245 {
7246 	struct extent_map *em = NULL;
7247 	int ret;
7248 
7249 	if (type != BTRFS_ORDERED_NOCOW) {
7250 		em = create_io_em(inode, start, len, orig_start,
7251 				  block_start, block_len, orig_block_len,
7252 				  ram_bytes,
7253 				  BTRFS_COMPRESS_NONE, /* compress_type */
7254 				  type);
7255 		if (IS_ERR(em))
7256 			goto out;
7257 	}
7258 	ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7259 					   len, block_len, type);
7260 	if (ret) {
7261 		if (em) {
7262 			free_extent_map(em);
7263 			btrfs_drop_extent_cache(BTRFS_I(inode), start,
7264 						start + len - 1, 0);
7265 		}
7266 		em = ERR_PTR(ret);
7267 	}
7268  out:
7269 
7270 	return em;
7271 }
7272 
7273 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7274 						  u64 start, u64 len)
7275 {
7276 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7277 	struct btrfs_root *root = BTRFS_I(inode)->root;
7278 	struct extent_map *em;
7279 	struct btrfs_key ins;
7280 	u64 alloc_hint;
7281 	int ret;
7282 
7283 	alloc_hint = get_extent_allocation_hint(inode, start, len);
7284 	ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7285 				   0, alloc_hint, &ins, 1, 1);
7286 	if (ret)
7287 		return ERR_PTR(ret);
7288 
7289 	em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7290 				     ins.objectid, ins.offset, ins.offset,
7291 				     ins.offset, BTRFS_ORDERED_REGULAR);
7292 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7293 	if (IS_ERR(em))
7294 		btrfs_free_reserved_extent(fs_info, ins.objectid,
7295 					   ins.offset, 1);
7296 
7297 	return em;
7298 }
7299 
7300 /*
7301  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7302  * block must be cow'd
7303  */
7304 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7305 			      u64 *orig_start, u64 *orig_block_len,
7306 			      u64 *ram_bytes)
7307 {
7308 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7309 	struct btrfs_path *path;
7310 	int ret;
7311 	struct extent_buffer *leaf;
7312 	struct btrfs_root *root = BTRFS_I(inode)->root;
7313 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7314 	struct btrfs_file_extent_item *fi;
7315 	struct btrfs_key key;
7316 	u64 disk_bytenr;
7317 	u64 backref_offset;
7318 	u64 extent_end;
7319 	u64 num_bytes;
7320 	int slot;
7321 	int found_type;
7322 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7323 
7324 	path = btrfs_alloc_path();
7325 	if (!path)
7326 		return -ENOMEM;
7327 
7328 	ret = btrfs_lookup_file_extent(NULL, root, path,
7329 			btrfs_ino(BTRFS_I(inode)), offset, 0);
7330 	if (ret < 0)
7331 		goto out;
7332 
7333 	slot = path->slots[0];
7334 	if (ret == 1) {
7335 		if (slot == 0) {
7336 			/* can't find the item, must cow */
7337 			ret = 0;
7338 			goto out;
7339 		}
7340 		slot--;
7341 	}
7342 	ret = 0;
7343 	leaf = path->nodes[0];
7344 	btrfs_item_key_to_cpu(leaf, &key, slot);
7345 	if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7346 	    key.type != BTRFS_EXTENT_DATA_KEY) {
7347 		/* not our file or wrong item type, must cow */
7348 		goto out;
7349 	}
7350 
7351 	if (key.offset > offset) {
7352 		/* Wrong offset, must cow */
7353 		goto out;
7354 	}
7355 
7356 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7357 	found_type = btrfs_file_extent_type(leaf, fi);
7358 	if (found_type != BTRFS_FILE_EXTENT_REG &&
7359 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7360 		/* not a regular extent, must cow */
7361 		goto out;
7362 	}
7363 
7364 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7365 		goto out;
7366 
7367 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7368 	if (extent_end <= offset)
7369 		goto out;
7370 
7371 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7372 	if (disk_bytenr == 0)
7373 		goto out;
7374 
7375 	if (btrfs_file_extent_compression(leaf, fi) ||
7376 	    btrfs_file_extent_encryption(leaf, fi) ||
7377 	    btrfs_file_extent_other_encoding(leaf, fi))
7378 		goto out;
7379 
7380 	backref_offset = btrfs_file_extent_offset(leaf, fi);
7381 
7382 	if (orig_start) {
7383 		*orig_start = key.offset - backref_offset;
7384 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7385 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7386 	}
7387 
7388 	if (btrfs_extent_readonly(fs_info, disk_bytenr))
7389 		goto out;
7390 
7391 	num_bytes = min(offset + *len, extent_end) - offset;
7392 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7393 		u64 range_end;
7394 
7395 		range_end = round_up(offset + num_bytes,
7396 				     root->fs_info->sectorsize) - 1;
7397 		ret = test_range_bit(io_tree, offset, range_end,
7398 				     EXTENT_DELALLOC, 0, NULL);
7399 		if (ret) {
7400 			ret = -EAGAIN;
7401 			goto out;
7402 		}
7403 	}
7404 
7405 	btrfs_release_path(path);
7406 
7407 	/*
7408 	 * look for other files referencing this extent, if we
7409 	 * find any we must cow
7410 	 */
7411 
7412 	ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7413 				    key.offset - backref_offset, disk_bytenr);
7414 	if (ret) {
7415 		ret = 0;
7416 		goto out;
7417 	}
7418 
7419 	/*
7420 	 * adjust disk_bytenr and num_bytes to cover just the bytes
7421 	 * in this extent we are about to write.  If there
7422 	 * are any csums in that range we have to cow in order
7423 	 * to keep the csums correct
7424 	 */
7425 	disk_bytenr += backref_offset;
7426 	disk_bytenr += offset - key.offset;
7427 	if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7428 		goto out;
7429 	/*
7430 	 * all of the above have passed, it is safe to overwrite this extent
7431 	 * without cow
7432 	 */
7433 	*len = num_bytes;
7434 	ret = 1;
7435 out:
7436 	btrfs_free_path(path);
7437 	return ret;
7438 }
7439 
7440 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7441 			      struct extent_state **cached_state, int writing)
7442 {
7443 	struct btrfs_ordered_extent *ordered;
7444 	int ret = 0;
7445 
7446 	while (1) {
7447 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7448 				 cached_state);
7449 		/*
7450 		 * We're concerned with the entire range that we're going to be
7451 		 * doing DIO to, so we need to make sure there's no ordered
7452 		 * extents in this range.
7453 		 */
7454 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7455 						     lockend - lockstart + 1);
7456 
7457 		/*
7458 		 * We need to make sure there are no buffered pages in this
7459 		 * range either, we could have raced between the invalidate in
7460 		 * generic_file_direct_write and locking the extent.  The
7461 		 * invalidate needs to happen so that reads after a write do not
7462 		 * get stale data.
7463 		 */
7464 		if (!ordered &&
7465 		    (!writing || !filemap_range_has_page(inode->i_mapping,
7466 							 lockstart, lockend)))
7467 			break;
7468 
7469 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7470 				     cached_state);
7471 
7472 		if (ordered) {
7473 			/*
7474 			 * If we are doing a DIO read and the ordered extent we
7475 			 * found is for a buffered write, we can not wait for it
7476 			 * to complete and retry, because if we do so we can
7477 			 * deadlock with concurrent buffered writes on page
7478 			 * locks. This happens only if our DIO read covers more
7479 			 * than one extent map, if at this point has already
7480 			 * created an ordered extent for a previous extent map
7481 			 * and locked its range in the inode's io tree, and a
7482 			 * concurrent write against that previous extent map's
7483 			 * range and this range started (we unlock the ranges
7484 			 * in the io tree only when the bios complete and
7485 			 * buffered writes always lock pages before attempting
7486 			 * to lock range in the io tree).
7487 			 */
7488 			if (writing ||
7489 			    test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7490 				btrfs_start_ordered_extent(inode, ordered, 1);
7491 			else
7492 				ret = -ENOTBLK;
7493 			btrfs_put_ordered_extent(ordered);
7494 		} else {
7495 			/*
7496 			 * We could trigger writeback for this range (and wait
7497 			 * for it to complete) and then invalidate the pages for
7498 			 * this range (through invalidate_inode_pages2_range()),
7499 			 * but that can lead us to a deadlock with a concurrent
7500 			 * call to readpages() (a buffered read or a defrag call
7501 			 * triggered a readahead) on a page lock due to an
7502 			 * ordered dio extent we created before but did not have
7503 			 * yet a corresponding bio submitted (whence it can not
7504 			 * complete), which makes readpages() wait for that
7505 			 * ordered extent to complete while holding a lock on
7506 			 * that page.
7507 			 */
7508 			ret = -ENOTBLK;
7509 		}
7510 
7511 		if (ret)
7512 			break;
7513 
7514 		cond_resched();
7515 	}
7516 
7517 	return ret;
7518 }
7519 
7520 /* The callers of this must take lock_extent() */
7521 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7522 				       u64 orig_start, u64 block_start,
7523 				       u64 block_len, u64 orig_block_len,
7524 				       u64 ram_bytes, int compress_type,
7525 				       int type)
7526 {
7527 	struct extent_map_tree *em_tree;
7528 	struct extent_map *em;
7529 	struct btrfs_root *root = BTRFS_I(inode)->root;
7530 	int ret;
7531 
7532 	ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7533 	       type == BTRFS_ORDERED_COMPRESSED ||
7534 	       type == BTRFS_ORDERED_NOCOW ||
7535 	       type == BTRFS_ORDERED_REGULAR);
7536 
7537 	em_tree = &BTRFS_I(inode)->extent_tree;
7538 	em = alloc_extent_map();
7539 	if (!em)
7540 		return ERR_PTR(-ENOMEM);
7541 
7542 	em->start = start;
7543 	em->orig_start = orig_start;
7544 	em->len = len;
7545 	em->block_len = block_len;
7546 	em->block_start = block_start;
7547 	em->bdev = root->fs_info->fs_devices->latest_bdev;
7548 	em->orig_block_len = orig_block_len;
7549 	em->ram_bytes = ram_bytes;
7550 	em->generation = -1;
7551 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
7552 	if (type == BTRFS_ORDERED_PREALLOC) {
7553 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
7554 	} else if (type == BTRFS_ORDERED_COMPRESSED) {
7555 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7556 		em->compress_type = compress_type;
7557 	}
7558 
7559 	do {
7560 		btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7561 				em->start + em->len - 1, 0);
7562 		write_lock(&em_tree->lock);
7563 		ret = add_extent_mapping(em_tree, em, 1);
7564 		write_unlock(&em_tree->lock);
7565 		/*
7566 		 * The caller has taken lock_extent(), who could race with us
7567 		 * to add em?
7568 		 */
7569 	} while (ret == -EEXIST);
7570 
7571 	if (ret) {
7572 		free_extent_map(em);
7573 		return ERR_PTR(ret);
7574 	}
7575 
7576 	/* em got 2 refs now, callers needs to do free_extent_map once. */
7577 	return em;
7578 }
7579 
7580 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7581 				   struct buffer_head *bh_result, int create)
7582 {
7583 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7584 	struct extent_map *em;
7585 	struct extent_state *cached_state = NULL;
7586 	struct btrfs_dio_data *dio_data = NULL;
7587 	u64 start = iblock << inode->i_blkbits;
7588 	u64 lockstart, lockend;
7589 	u64 len = bh_result->b_size;
7590 	int unlock_bits = EXTENT_LOCKED;
7591 	int ret = 0;
7592 
7593 	if (create)
7594 		unlock_bits |= EXTENT_DIRTY;
7595 	else
7596 		len = min_t(u64, len, fs_info->sectorsize);
7597 
7598 	lockstart = start;
7599 	lockend = start + len - 1;
7600 
7601 	if (current->journal_info) {
7602 		/*
7603 		 * Need to pull our outstanding extents and set journal_info to NULL so
7604 		 * that anything that needs to check if there's a transaction doesn't get
7605 		 * confused.
7606 		 */
7607 		dio_data = current->journal_info;
7608 		current->journal_info = NULL;
7609 	}
7610 
7611 	/*
7612 	 * If this errors out it's because we couldn't invalidate pagecache for
7613 	 * this range and we need to fallback to buffered.
7614 	 */
7615 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7616 			       create)) {
7617 		ret = -ENOTBLK;
7618 		goto err;
7619 	}
7620 
7621 	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7622 	if (IS_ERR(em)) {
7623 		ret = PTR_ERR(em);
7624 		goto unlock_err;
7625 	}
7626 
7627 	/*
7628 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7629 	 * io.  INLINE is special, and we could probably kludge it in here, but
7630 	 * it's still buffered so for safety lets just fall back to the generic
7631 	 * buffered path.
7632 	 *
7633 	 * For COMPRESSED we _have_ to read the entire extent in so we can
7634 	 * decompress it, so there will be buffering required no matter what we
7635 	 * do, so go ahead and fallback to buffered.
7636 	 *
7637 	 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7638 	 * to buffered IO.  Don't blame me, this is the price we pay for using
7639 	 * the generic code.
7640 	 */
7641 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7642 	    em->block_start == EXTENT_MAP_INLINE) {
7643 		free_extent_map(em);
7644 		ret = -ENOTBLK;
7645 		goto unlock_err;
7646 	}
7647 
7648 	/* Just a good old fashioned hole, return */
7649 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7650 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7651 		free_extent_map(em);
7652 		goto unlock_err;
7653 	}
7654 
7655 	/*
7656 	 * We don't allocate a new extent in the following cases
7657 	 *
7658 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7659 	 * existing extent.
7660 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
7661 	 * just use the extent.
7662 	 *
7663 	 */
7664 	if (!create) {
7665 		len = min(len, em->len - (start - em->start));
7666 		lockstart = start + len;
7667 		goto unlock;
7668 	}
7669 
7670 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7671 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7672 	     em->block_start != EXTENT_MAP_HOLE)) {
7673 		int type;
7674 		u64 block_start, orig_start, orig_block_len, ram_bytes;
7675 
7676 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7677 			type = BTRFS_ORDERED_PREALLOC;
7678 		else
7679 			type = BTRFS_ORDERED_NOCOW;
7680 		len = min(len, em->len - (start - em->start));
7681 		block_start = em->block_start + (start - em->start);
7682 
7683 		if (can_nocow_extent(inode, start, &len, &orig_start,
7684 				     &orig_block_len, &ram_bytes) == 1 &&
7685 		    btrfs_inc_nocow_writers(fs_info, block_start)) {
7686 			struct extent_map *em2;
7687 
7688 			em2 = btrfs_create_dio_extent(inode, start, len,
7689 						      orig_start, block_start,
7690 						      len, orig_block_len,
7691 						      ram_bytes, type);
7692 			btrfs_dec_nocow_writers(fs_info, block_start);
7693 			if (type == BTRFS_ORDERED_PREALLOC) {
7694 				free_extent_map(em);
7695 				em = em2;
7696 			}
7697 			if (em2 && IS_ERR(em2)) {
7698 				ret = PTR_ERR(em2);
7699 				goto unlock_err;
7700 			}
7701 			/*
7702 			 * For inode marked NODATACOW or extent marked PREALLOC,
7703 			 * use the existing or preallocated extent, so does not
7704 			 * need to adjust btrfs_space_info's bytes_may_use.
7705 			 */
7706 			btrfs_free_reserved_data_space_noquota(inode,
7707 					start, len);
7708 			goto unlock;
7709 		}
7710 	}
7711 
7712 	/*
7713 	 * this will cow the extent, reset the len in case we changed
7714 	 * it above
7715 	 */
7716 	len = bh_result->b_size;
7717 	free_extent_map(em);
7718 	em = btrfs_new_extent_direct(inode, start, len);
7719 	if (IS_ERR(em)) {
7720 		ret = PTR_ERR(em);
7721 		goto unlock_err;
7722 	}
7723 	len = min(len, em->len - (start - em->start));
7724 unlock:
7725 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7726 		inode->i_blkbits;
7727 	bh_result->b_size = len;
7728 	bh_result->b_bdev = em->bdev;
7729 	set_buffer_mapped(bh_result);
7730 	if (create) {
7731 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7732 			set_buffer_new(bh_result);
7733 
7734 		/*
7735 		 * Need to update the i_size under the extent lock so buffered
7736 		 * readers will get the updated i_size when we unlock.
7737 		 */
7738 		if (!dio_data->overwrite && start + len > i_size_read(inode))
7739 			i_size_write(inode, start + len);
7740 
7741 		WARN_ON(dio_data->reserve < len);
7742 		dio_data->reserve -= len;
7743 		dio_data->unsubmitted_oe_range_end = start + len;
7744 		current->journal_info = dio_data;
7745 	}
7746 
7747 	/*
7748 	 * In the case of write we need to clear and unlock the entire range,
7749 	 * in the case of read we need to unlock only the end area that we
7750 	 * aren't using if there is any left over space.
7751 	 */
7752 	if (lockstart < lockend) {
7753 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7754 				 lockend, unlock_bits, 1, 0,
7755 				 &cached_state);
7756 	} else {
7757 		free_extent_state(cached_state);
7758 	}
7759 
7760 	free_extent_map(em);
7761 
7762 	return 0;
7763 
7764 unlock_err:
7765 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7766 			 unlock_bits, 1, 0, &cached_state);
7767 err:
7768 	if (dio_data)
7769 		current->journal_info = dio_data;
7770 	return ret;
7771 }
7772 
7773 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7774 						 struct bio *bio,
7775 						 int mirror_num)
7776 {
7777 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7778 	blk_status_t ret;
7779 
7780 	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7781 
7782 	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7783 	if (ret)
7784 		return ret;
7785 
7786 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7787 
7788 	return ret;
7789 }
7790 
7791 static int btrfs_check_dio_repairable(struct inode *inode,
7792 				      struct bio *failed_bio,
7793 				      struct io_failure_record *failrec,
7794 				      int failed_mirror)
7795 {
7796 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7797 	int num_copies;
7798 
7799 	num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7800 	if (num_copies == 1) {
7801 		/*
7802 		 * we only have a single copy of the data, so don't bother with
7803 		 * all the retry and error correction code that follows. no
7804 		 * matter what the error is, it is very likely to persist.
7805 		 */
7806 		btrfs_debug(fs_info,
7807 			"Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7808 			num_copies, failrec->this_mirror, failed_mirror);
7809 		return 0;
7810 	}
7811 
7812 	failrec->failed_mirror = failed_mirror;
7813 	failrec->this_mirror++;
7814 	if (failrec->this_mirror == failed_mirror)
7815 		failrec->this_mirror++;
7816 
7817 	if (failrec->this_mirror > num_copies) {
7818 		btrfs_debug(fs_info,
7819 			"Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7820 			num_copies, failrec->this_mirror, failed_mirror);
7821 		return 0;
7822 	}
7823 
7824 	return 1;
7825 }
7826 
7827 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7828 				   struct page *page, unsigned int pgoff,
7829 				   u64 start, u64 end, int failed_mirror,
7830 				   bio_end_io_t *repair_endio, void *repair_arg)
7831 {
7832 	struct io_failure_record *failrec;
7833 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7834 	struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7835 	struct bio *bio;
7836 	int isector;
7837 	unsigned int read_mode = 0;
7838 	int segs;
7839 	int ret;
7840 	blk_status_t status;
7841 	struct bio_vec bvec;
7842 
7843 	BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7844 
7845 	ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7846 	if (ret)
7847 		return errno_to_blk_status(ret);
7848 
7849 	ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7850 					 failed_mirror);
7851 	if (!ret) {
7852 		free_io_failure(failure_tree, io_tree, failrec);
7853 		return BLK_STS_IOERR;
7854 	}
7855 
7856 	segs = bio_segments(failed_bio);
7857 	bio_get_first_bvec(failed_bio, &bvec);
7858 	if (segs > 1 ||
7859 	    (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7860 		read_mode |= REQ_FAILFAST_DEV;
7861 
7862 	isector = start - btrfs_io_bio(failed_bio)->logical;
7863 	isector >>= inode->i_sb->s_blocksize_bits;
7864 	bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7865 				pgoff, isector, repair_endio, repair_arg);
7866 	bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7867 
7868 	btrfs_debug(BTRFS_I(inode)->root->fs_info,
7869 		    "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7870 		    read_mode, failrec->this_mirror, failrec->in_validation);
7871 
7872 	status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7873 	if (status) {
7874 		free_io_failure(failure_tree, io_tree, failrec);
7875 		bio_put(bio);
7876 	}
7877 
7878 	return status;
7879 }
7880 
7881 struct btrfs_retry_complete {
7882 	struct completion done;
7883 	struct inode *inode;
7884 	u64 start;
7885 	int uptodate;
7886 };
7887 
7888 static void btrfs_retry_endio_nocsum(struct bio *bio)
7889 {
7890 	struct btrfs_retry_complete *done = bio->bi_private;
7891 	struct inode *inode = done->inode;
7892 	struct bio_vec *bvec;
7893 	struct extent_io_tree *io_tree, *failure_tree;
7894 	int i;
7895 
7896 	if (bio->bi_status)
7897 		goto end;
7898 
7899 	ASSERT(bio->bi_vcnt == 1);
7900 	io_tree = &BTRFS_I(inode)->io_tree;
7901 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
7902 	ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7903 
7904 	done->uptodate = 1;
7905 	ASSERT(!bio_flagged(bio, BIO_CLONED));
7906 	bio_for_each_segment_all(bvec, bio, i)
7907 		clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7908 				 io_tree, done->start, bvec->bv_page,
7909 				 btrfs_ino(BTRFS_I(inode)), 0);
7910 end:
7911 	complete(&done->done);
7912 	bio_put(bio);
7913 }
7914 
7915 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7916 						struct btrfs_io_bio *io_bio)
7917 {
7918 	struct btrfs_fs_info *fs_info;
7919 	struct bio_vec bvec;
7920 	struct bvec_iter iter;
7921 	struct btrfs_retry_complete done;
7922 	u64 start;
7923 	unsigned int pgoff;
7924 	u32 sectorsize;
7925 	int nr_sectors;
7926 	blk_status_t ret;
7927 	blk_status_t err = BLK_STS_OK;
7928 
7929 	fs_info = BTRFS_I(inode)->root->fs_info;
7930 	sectorsize = fs_info->sectorsize;
7931 
7932 	start = io_bio->logical;
7933 	done.inode = inode;
7934 	io_bio->bio.bi_iter = io_bio->iter;
7935 
7936 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
7937 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7938 		pgoff = bvec.bv_offset;
7939 
7940 next_block_or_try_again:
7941 		done.uptodate = 0;
7942 		done.start = start;
7943 		init_completion(&done.done);
7944 
7945 		ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7946 				pgoff, start, start + sectorsize - 1,
7947 				io_bio->mirror_num,
7948 				btrfs_retry_endio_nocsum, &done);
7949 		if (ret) {
7950 			err = ret;
7951 			goto next;
7952 		}
7953 
7954 		wait_for_completion_io(&done.done);
7955 
7956 		if (!done.uptodate) {
7957 			/* We might have another mirror, so try again */
7958 			goto next_block_or_try_again;
7959 		}
7960 
7961 next:
7962 		start += sectorsize;
7963 
7964 		nr_sectors--;
7965 		if (nr_sectors) {
7966 			pgoff += sectorsize;
7967 			ASSERT(pgoff < PAGE_SIZE);
7968 			goto next_block_or_try_again;
7969 		}
7970 	}
7971 
7972 	return err;
7973 }
7974 
7975 static void btrfs_retry_endio(struct bio *bio)
7976 {
7977 	struct btrfs_retry_complete *done = bio->bi_private;
7978 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7979 	struct extent_io_tree *io_tree, *failure_tree;
7980 	struct inode *inode = done->inode;
7981 	struct bio_vec *bvec;
7982 	int uptodate;
7983 	int ret;
7984 	int i;
7985 
7986 	if (bio->bi_status)
7987 		goto end;
7988 
7989 	uptodate = 1;
7990 
7991 	ASSERT(bio->bi_vcnt == 1);
7992 	ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7993 
7994 	io_tree = &BTRFS_I(inode)->io_tree;
7995 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
7996 
7997 	ASSERT(!bio_flagged(bio, BIO_CLONED));
7998 	bio_for_each_segment_all(bvec, bio, i) {
7999 		ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8000 					     bvec->bv_offset, done->start,
8001 					     bvec->bv_len);
8002 		if (!ret)
8003 			clean_io_failure(BTRFS_I(inode)->root->fs_info,
8004 					 failure_tree, io_tree, done->start,
8005 					 bvec->bv_page,
8006 					 btrfs_ino(BTRFS_I(inode)),
8007 					 bvec->bv_offset);
8008 		else
8009 			uptodate = 0;
8010 	}
8011 
8012 	done->uptodate = uptodate;
8013 end:
8014 	complete(&done->done);
8015 	bio_put(bio);
8016 }
8017 
8018 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8019 		struct btrfs_io_bio *io_bio, blk_status_t err)
8020 {
8021 	struct btrfs_fs_info *fs_info;
8022 	struct bio_vec bvec;
8023 	struct bvec_iter iter;
8024 	struct btrfs_retry_complete done;
8025 	u64 start;
8026 	u64 offset = 0;
8027 	u32 sectorsize;
8028 	int nr_sectors;
8029 	unsigned int pgoff;
8030 	int csum_pos;
8031 	bool uptodate = (err == 0);
8032 	int ret;
8033 	blk_status_t status;
8034 
8035 	fs_info = BTRFS_I(inode)->root->fs_info;
8036 	sectorsize = fs_info->sectorsize;
8037 
8038 	err = BLK_STS_OK;
8039 	start = io_bio->logical;
8040 	done.inode = inode;
8041 	io_bio->bio.bi_iter = io_bio->iter;
8042 
8043 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
8044 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8045 
8046 		pgoff = bvec.bv_offset;
8047 next_block:
8048 		if (uptodate) {
8049 			csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8050 			ret = __readpage_endio_check(inode, io_bio, csum_pos,
8051 					bvec.bv_page, pgoff, start, sectorsize);
8052 			if (likely(!ret))
8053 				goto next;
8054 		}
8055 try_again:
8056 		done.uptodate = 0;
8057 		done.start = start;
8058 		init_completion(&done.done);
8059 
8060 		status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8061 					pgoff, start, start + sectorsize - 1,
8062 					io_bio->mirror_num, btrfs_retry_endio,
8063 					&done);
8064 		if (status) {
8065 			err = status;
8066 			goto next;
8067 		}
8068 
8069 		wait_for_completion_io(&done.done);
8070 
8071 		if (!done.uptodate) {
8072 			/* We might have another mirror, so try again */
8073 			goto try_again;
8074 		}
8075 next:
8076 		offset += sectorsize;
8077 		start += sectorsize;
8078 
8079 		ASSERT(nr_sectors);
8080 
8081 		nr_sectors--;
8082 		if (nr_sectors) {
8083 			pgoff += sectorsize;
8084 			ASSERT(pgoff < PAGE_SIZE);
8085 			goto next_block;
8086 		}
8087 	}
8088 
8089 	return err;
8090 }
8091 
8092 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8093 		struct btrfs_io_bio *io_bio, blk_status_t err)
8094 {
8095 	bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8096 
8097 	if (skip_csum) {
8098 		if (unlikely(err))
8099 			return __btrfs_correct_data_nocsum(inode, io_bio);
8100 		else
8101 			return BLK_STS_OK;
8102 	} else {
8103 		return __btrfs_subio_endio_read(inode, io_bio, err);
8104 	}
8105 }
8106 
8107 static void btrfs_endio_direct_read(struct bio *bio)
8108 {
8109 	struct btrfs_dio_private *dip = bio->bi_private;
8110 	struct inode *inode = dip->inode;
8111 	struct bio *dio_bio;
8112 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8113 	blk_status_t err = bio->bi_status;
8114 
8115 	if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8116 		err = btrfs_subio_endio_read(inode, io_bio, err);
8117 
8118 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8119 		      dip->logical_offset + dip->bytes - 1);
8120 	dio_bio = dip->dio_bio;
8121 
8122 	kfree(dip);
8123 
8124 	dio_bio->bi_status = err;
8125 	dio_end_io(dio_bio);
8126 
8127 	if (io_bio->end_io)
8128 		io_bio->end_io(io_bio, blk_status_to_errno(err));
8129 	bio_put(bio);
8130 }
8131 
8132 static void __endio_write_update_ordered(struct inode *inode,
8133 					 const u64 offset, const u64 bytes,
8134 					 const bool uptodate)
8135 {
8136 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8137 	struct btrfs_ordered_extent *ordered = NULL;
8138 	struct btrfs_workqueue *wq;
8139 	btrfs_work_func_t func;
8140 	u64 ordered_offset = offset;
8141 	u64 ordered_bytes = bytes;
8142 	u64 last_offset;
8143 	int ret;
8144 
8145 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8146 		wq = fs_info->endio_freespace_worker;
8147 		func = btrfs_freespace_write_helper;
8148 	} else {
8149 		wq = fs_info->endio_write_workers;
8150 		func = btrfs_endio_write_helper;
8151 	}
8152 
8153 again:
8154 	last_offset = ordered_offset;
8155 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8156 						   &ordered_offset,
8157 						   ordered_bytes,
8158 						   uptodate);
8159 	if (!ret)
8160 		goto out_test;
8161 
8162 	btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8163 	btrfs_queue_work(wq, &ordered->work);
8164 out_test:
8165 	/*
8166 	 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8167 	 * in the range, we can exit.
8168 	 */
8169 	if (ordered_offset == last_offset)
8170 		return;
8171 	/*
8172 	 * our bio might span multiple ordered extents.  If we haven't
8173 	 * completed the accounting for the whole dio, go back and try again
8174 	 */
8175 	if (ordered_offset < offset + bytes) {
8176 		ordered_bytes = offset + bytes - ordered_offset;
8177 		ordered = NULL;
8178 		goto again;
8179 	}
8180 }
8181 
8182 static void btrfs_endio_direct_write(struct bio *bio)
8183 {
8184 	struct btrfs_dio_private *dip = bio->bi_private;
8185 	struct bio *dio_bio = dip->dio_bio;
8186 
8187 	__endio_write_update_ordered(dip->inode, dip->logical_offset,
8188 				     dip->bytes, !bio->bi_status);
8189 
8190 	kfree(dip);
8191 
8192 	dio_bio->bi_status = bio->bi_status;
8193 	dio_end_io(dio_bio);
8194 	bio_put(bio);
8195 }
8196 
8197 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8198 				    struct bio *bio, u64 offset)
8199 {
8200 	struct inode *inode = private_data;
8201 	blk_status_t ret;
8202 	ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8203 	BUG_ON(ret); /* -ENOMEM */
8204 	return 0;
8205 }
8206 
8207 static void btrfs_end_dio_bio(struct bio *bio)
8208 {
8209 	struct btrfs_dio_private *dip = bio->bi_private;
8210 	blk_status_t err = bio->bi_status;
8211 
8212 	if (err)
8213 		btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8214 			   "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8215 			   btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8216 			   bio->bi_opf,
8217 			   (unsigned long long)bio->bi_iter.bi_sector,
8218 			   bio->bi_iter.bi_size, err);
8219 
8220 	if (dip->subio_endio)
8221 		err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8222 
8223 	if (err) {
8224 		/*
8225 		 * We want to perceive the errors flag being set before
8226 		 * decrementing the reference count. We don't need a barrier
8227 		 * since atomic operations with a return value are fully
8228 		 * ordered as per atomic_t.txt
8229 		 */
8230 		dip->errors = 1;
8231 	}
8232 
8233 	/* if there are more bios still pending for this dio, just exit */
8234 	if (!atomic_dec_and_test(&dip->pending_bios))
8235 		goto out;
8236 
8237 	if (dip->errors) {
8238 		bio_io_error(dip->orig_bio);
8239 	} else {
8240 		dip->dio_bio->bi_status = BLK_STS_OK;
8241 		bio_endio(dip->orig_bio);
8242 	}
8243 out:
8244 	bio_put(bio);
8245 }
8246 
8247 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8248 						 struct btrfs_dio_private *dip,
8249 						 struct bio *bio,
8250 						 u64 file_offset)
8251 {
8252 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8253 	struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8254 	blk_status_t ret;
8255 
8256 	/*
8257 	 * We load all the csum data we need when we submit
8258 	 * the first bio to reduce the csum tree search and
8259 	 * contention.
8260 	 */
8261 	if (dip->logical_offset == file_offset) {
8262 		ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8263 						file_offset);
8264 		if (ret)
8265 			return ret;
8266 	}
8267 
8268 	if (bio == dip->orig_bio)
8269 		return 0;
8270 
8271 	file_offset -= dip->logical_offset;
8272 	file_offset >>= inode->i_sb->s_blocksize_bits;
8273 	io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8274 
8275 	return 0;
8276 }
8277 
8278 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8279 		struct inode *inode, u64 file_offset, int async_submit)
8280 {
8281 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8282 	struct btrfs_dio_private *dip = bio->bi_private;
8283 	bool write = bio_op(bio) == REQ_OP_WRITE;
8284 	blk_status_t ret;
8285 
8286 	/* Check btrfs_submit_bio_hook() for rules about async submit. */
8287 	if (async_submit)
8288 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8289 
8290 	if (!write) {
8291 		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8292 		if (ret)
8293 			goto err;
8294 	}
8295 
8296 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8297 		goto map;
8298 
8299 	if (write && async_submit) {
8300 		ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8301 					  file_offset, inode,
8302 					  btrfs_submit_bio_start_direct_io,
8303 					  btrfs_submit_bio_done);
8304 		goto err;
8305 	} else if (write) {
8306 		/*
8307 		 * If we aren't doing async submit, calculate the csum of the
8308 		 * bio now.
8309 		 */
8310 		ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8311 		if (ret)
8312 			goto err;
8313 	} else {
8314 		ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8315 						     file_offset);
8316 		if (ret)
8317 			goto err;
8318 	}
8319 map:
8320 	ret = btrfs_map_bio(fs_info, bio, 0, 0);
8321 err:
8322 	return ret;
8323 }
8324 
8325 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8326 {
8327 	struct inode *inode = dip->inode;
8328 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8329 	struct bio *bio;
8330 	struct bio *orig_bio = dip->orig_bio;
8331 	u64 start_sector = orig_bio->bi_iter.bi_sector;
8332 	u64 file_offset = dip->logical_offset;
8333 	u64 map_length;
8334 	int async_submit = 0;
8335 	u64 submit_len;
8336 	int clone_offset = 0;
8337 	int clone_len;
8338 	int ret;
8339 	blk_status_t status;
8340 
8341 	map_length = orig_bio->bi_iter.bi_size;
8342 	submit_len = map_length;
8343 	ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8344 			      &map_length, NULL, 0);
8345 	if (ret)
8346 		return -EIO;
8347 
8348 	if (map_length >= submit_len) {
8349 		bio = orig_bio;
8350 		dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8351 		goto submit;
8352 	}
8353 
8354 	/* async crcs make it difficult to collect full stripe writes. */
8355 	if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8356 		async_submit = 0;
8357 	else
8358 		async_submit = 1;
8359 
8360 	/* bio split */
8361 	ASSERT(map_length <= INT_MAX);
8362 	atomic_inc(&dip->pending_bios);
8363 	do {
8364 		clone_len = min_t(int, submit_len, map_length);
8365 
8366 		/*
8367 		 * This will never fail as it's passing GPF_NOFS and
8368 		 * the allocation is backed by btrfs_bioset.
8369 		 */
8370 		bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8371 					      clone_len);
8372 		bio->bi_private = dip;
8373 		bio->bi_end_io = btrfs_end_dio_bio;
8374 		btrfs_io_bio(bio)->logical = file_offset;
8375 
8376 		ASSERT(submit_len >= clone_len);
8377 		submit_len -= clone_len;
8378 		if (submit_len == 0)
8379 			break;
8380 
8381 		/*
8382 		 * Increase the count before we submit the bio so we know
8383 		 * the end IO handler won't happen before we increase the
8384 		 * count. Otherwise, the dip might get freed before we're
8385 		 * done setting it up.
8386 		 */
8387 		atomic_inc(&dip->pending_bios);
8388 
8389 		status = btrfs_submit_dio_bio(bio, inode, file_offset,
8390 						async_submit);
8391 		if (status) {
8392 			bio_put(bio);
8393 			atomic_dec(&dip->pending_bios);
8394 			goto out_err;
8395 		}
8396 
8397 		clone_offset += clone_len;
8398 		start_sector += clone_len >> 9;
8399 		file_offset += clone_len;
8400 
8401 		map_length = submit_len;
8402 		ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8403 				      start_sector << 9, &map_length, NULL, 0);
8404 		if (ret)
8405 			goto out_err;
8406 	} while (submit_len > 0);
8407 
8408 submit:
8409 	status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8410 	if (!status)
8411 		return 0;
8412 
8413 	bio_put(bio);
8414 out_err:
8415 	dip->errors = 1;
8416 	/*
8417 	 * Before atomic variable goto zero, we must  make sure dip->errors is
8418 	 * perceived to be set. This ordering is ensured by the fact that an
8419 	 * atomic operations with a return value are fully ordered as per
8420 	 * atomic_t.txt
8421 	 */
8422 	if (atomic_dec_and_test(&dip->pending_bios))
8423 		bio_io_error(dip->orig_bio);
8424 
8425 	/* bio_end_io() will handle error, so we needn't return it */
8426 	return 0;
8427 }
8428 
8429 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8430 				loff_t file_offset)
8431 {
8432 	struct btrfs_dio_private *dip = NULL;
8433 	struct bio *bio = NULL;
8434 	struct btrfs_io_bio *io_bio;
8435 	bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8436 	int ret = 0;
8437 
8438 	bio = btrfs_bio_clone(dio_bio);
8439 
8440 	dip = kzalloc(sizeof(*dip), GFP_NOFS);
8441 	if (!dip) {
8442 		ret = -ENOMEM;
8443 		goto free_ordered;
8444 	}
8445 
8446 	dip->private = dio_bio->bi_private;
8447 	dip->inode = inode;
8448 	dip->logical_offset = file_offset;
8449 	dip->bytes = dio_bio->bi_iter.bi_size;
8450 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8451 	bio->bi_private = dip;
8452 	dip->orig_bio = bio;
8453 	dip->dio_bio = dio_bio;
8454 	atomic_set(&dip->pending_bios, 0);
8455 	io_bio = btrfs_io_bio(bio);
8456 	io_bio->logical = file_offset;
8457 
8458 	if (write) {
8459 		bio->bi_end_io = btrfs_endio_direct_write;
8460 	} else {
8461 		bio->bi_end_io = btrfs_endio_direct_read;
8462 		dip->subio_endio = btrfs_subio_endio_read;
8463 	}
8464 
8465 	/*
8466 	 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8467 	 * even if we fail to submit a bio, because in such case we do the
8468 	 * corresponding error handling below and it must not be done a second
8469 	 * time by btrfs_direct_IO().
8470 	 */
8471 	if (write) {
8472 		struct btrfs_dio_data *dio_data = current->journal_info;
8473 
8474 		dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8475 			dip->bytes;
8476 		dio_data->unsubmitted_oe_range_start =
8477 			dio_data->unsubmitted_oe_range_end;
8478 	}
8479 
8480 	ret = btrfs_submit_direct_hook(dip);
8481 	if (!ret)
8482 		return;
8483 
8484 	if (io_bio->end_io)
8485 		io_bio->end_io(io_bio, ret);
8486 
8487 free_ordered:
8488 	/*
8489 	 * If we arrived here it means either we failed to submit the dip
8490 	 * or we either failed to clone the dio_bio or failed to allocate the
8491 	 * dip. If we cloned the dio_bio and allocated the dip, we can just
8492 	 * call bio_endio against our io_bio so that we get proper resource
8493 	 * cleanup if we fail to submit the dip, otherwise, we must do the
8494 	 * same as btrfs_endio_direct_[write|read] because we can't call these
8495 	 * callbacks - they require an allocated dip and a clone of dio_bio.
8496 	 */
8497 	if (bio && dip) {
8498 		bio_io_error(bio);
8499 		/*
8500 		 * The end io callbacks free our dip, do the final put on bio
8501 		 * and all the cleanup and final put for dio_bio (through
8502 		 * dio_end_io()).
8503 		 */
8504 		dip = NULL;
8505 		bio = NULL;
8506 	} else {
8507 		if (write)
8508 			__endio_write_update_ordered(inode,
8509 						file_offset,
8510 						dio_bio->bi_iter.bi_size,
8511 						false);
8512 		else
8513 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8514 			      file_offset + dio_bio->bi_iter.bi_size - 1);
8515 
8516 		dio_bio->bi_status = BLK_STS_IOERR;
8517 		/*
8518 		 * Releases and cleans up our dio_bio, no need to bio_put()
8519 		 * nor bio_endio()/bio_io_error() against dio_bio.
8520 		 */
8521 		dio_end_io(dio_bio);
8522 	}
8523 	if (bio)
8524 		bio_put(bio);
8525 	kfree(dip);
8526 }
8527 
8528 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8529 			       const struct iov_iter *iter, loff_t offset)
8530 {
8531 	int seg;
8532 	int i;
8533 	unsigned int blocksize_mask = fs_info->sectorsize - 1;
8534 	ssize_t retval = -EINVAL;
8535 
8536 	if (offset & blocksize_mask)
8537 		goto out;
8538 
8539 	if (iov_iter_alignment(iter) & blocksize_mask)
8540 		goto out;
8541 
8542 	/* If this is a write we don't need to check anymore */
8543 	if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8544 		return 0;
8545 	/*
8546 	 * Check to make sure we don't have duplicate iov_base's in this
8547 	 * iovec, if so return EINVAL, otherwise we'll get csum errors
8548 	 * when reading back.
8549 	 */
8550 	for (seg = 0; seg < iter->nr_segs; seg++) {
8551 		for (i = seg + 1; i < iter->nr_segs; i++) {
8552 			if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8553 				goto out;
8554 		}
8555 	}
8556 	retval = 0;
8557 out:
8558 	return retval;
8559 }
8560 
8561 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8562 {
8563 	struct file *file = iocb->ki_filp;
8564 	struct inode *inode = file->f_mapping->host;
8565 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8566 	struct btrfs_dio_data dio_data = { 0 };
8567 	struct extent_changeset *data_reserved = NULL;
8568 	loff_t offset = iocb->ki_pos;
8569 	size_t count = 0;
8570 	int flags = 0;
8571 	bool wakeup = true;
8572 	bool relock = false;
8573 	ssize_t ret;
8574 
8575 	if (check_direct_IO(fs_info, iter, offset))
8576 		return 0;
8577 
8578 	inode_dio_begin(inode);
8579 
8580 	/*
8581 	 * The generic stuff only does filemap_write_and_wait_range, which
8582 	 * isn't enough if we've written compressed pages to this area, so
8583 	 * we need to flush the dirty pages again to make absolutely sure
8584 	 * that any outstanding dirty pages are on disk.
8585 	 */
8586 	count = iov_iter_count(iter);
8587 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8588 		     &BTRFS_I(inode)->runtime_flags))
8589 		filemap_fdatawrite_range(inode->i_mapping, offset,
8590 					 offset + count - 1);
8591 
8592 	if (iov_iter_rw(iter) == WRITE) {
8593 		/*
8594 		 * If the write DIO is beyond the EOF, we need update
8595 		 * the isize, but it is protected by i_mutex. So we can
8596 		 * not unlock the i_mutex at this case.
8597 		 */
8598 		if (offset + count <= inode->i_size) {
8599 			dio_data.overwrite = 1;
8600 			inode_unlock(inode);
8601 			relock = true;
8602 		} else if (iocb->ki_flags & IOCB_NOWAIT) {
8603 			ret = -EAGAIN;
8604 			goto out;
8605 		}
8606 		ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8607 						   offset, count);
8608 		if (ret)
8609 			goto out;
8610 
8611 		/*
8612 		 * We need to know how many extents we reserved so that we can
8613 		 * do the accounting properly if we go over the number we
8614 		 * originally calculated.  Abuse current->journal_info for this.
8615 		 */
8616 		dio_data.reserve = round_up(count,
8617 					    fs_info->sectorsize);
8618 		dio_data.unsubmitted_oe_range_start = (u64)offset;
8619 		dio_data.unsubmitted_oe_range_end = (u64)offset;
8620 		current->journal_info = &dio_data;
8621 		down_read(&BTRFS_I(inode)->dio_sem);
8622 	} else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8623 				     &BTRFS_I(inode)->runtime_flags)) {
8624 		inode_dio_end(inode);
8625 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
8626 		wakeup = false;
8627 	}
8628 
8629 	ret = __blockdev_direct_IO(iocb, inode,
8630 				   fs_info->fs_devices->latest_bdev,
8631 				   iter, btrfs_get_blocks_direct, NULL,
8632 				   btrfs_submit_direct, flags);
8633 	if (iov_iter_rw(iter) == WRITE) {
8634 		up_read(&BTRFS_I(inode)->dio_sem);
8635 		current->journal_info = NULL;
8636 		if (ret < 0 && ret != -EIOCBQUEUED) {
8637 			if (dio_data.reserve)
8638 				btrfs_delalloc_release_space(inode, data_reserved,
8639 					offset, dio_data.reserve, true);
8640 			/*
8641 			 * On error we might have left some ordered extents
8642 			 * without submitting corresponding bios for them, so
8643 			 * cleanup them up to avoid other tasks getting them
8644 			 * and waiting for them to complete forever.
8645 			 */
8646 			if (dio_data.unsubmitted_oe_range_start <
8647 			    dio_data.unsubmitted_oe_range_end)
8648 				__endio_write_update_ordered(inode,
8649 					dio_data.unsubmitted_oe_range_start,
8650 					dio_data.unsubmitted_oe_range_end -
8651 					dio_data.unsubmitted_oe_range_start,
8652 					false);
8653 		} else if (ret >= 0 && (size_t)ret < count)
8654 			btrfs_delalloc_release_space(inode, data_reserved,
8655 					offset, count - (size_t)ret, true);
8656 		btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8657 	}
8658 out:
8659 	if (wakeup)
8660 		inode_dio_end(inode);
8661 	if (relock)
8662 		inode_lock(inode);
8663 
8664 	extent_changeset_free(data_reserved);
8665 	return ret;
8666 }
8667 
8668 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
8669 
8670 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8671 		__u64 start, __u64 len)
8672 {
8673 	int	ret;
8674 
8675 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8676 	if (ret)
8677 		return ret;
8678 
8679 	return extent_fiemap(inode, fieinfo, start, len);
8680 }
8681 
8682 int btrfs_readpage(struct file *file, struct page *page)
8683 {
8684 	struct extent_io_tree *tree;
8685 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8686 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8687 }
8688 
8689 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8690 {
8691 	struct inode *inode = page->mapping->host;
8692 	int ret;
8693 
8694 	if (current->flags & PF_MEMALLOC) {
8695 		redirty_page_for_writepage(wbc, page);
8696 		unlock_page(page);
8697 		return 0;
8698 	}
8699 
8700 	/*
8701 	 * If we are under memory pressure we will call this directly from the
8702 	 * VM, we need to make sure we have the inode referenced for the ordered
8703 	 * extent.  If not just return like we didn't do anything.
8704 	 */
8705 	if (!igrab(inode)) {
8706 		redirty_page_for_writepage(wbc, page);
8707 		return AOP_WRITEPAGE_ACTIVATE;
8708 	}
8709 	ret = extent_write_full_page(page, wbc);
8710 	btrfs_add_delayed_iput(inode);
8711 	return ret;
8712 }
8713 
8714 static int btrfs_writepages(struct address_space *mapping,
8715 			    struct writeback_control *wbc)
8716 {
8717 	struct extent_io_tree *tree;
8718 
8719 	tree = &BTRFS_I(mapping->host)->io_tree;
8720 	return extent_writepages(tree, mapping, wbc);
8721 }
8722 
8723 static int
8724 btrfs_readpages(struct file *file, struct address_space *mapping,
8725 		struct list_head *pages, unsigned nr_pages)
8726 {
8727 	struct extent_io_tree *tree;
8728 	tree = &BTRFS_I(mapping->host)->io_tree;
8729 	return extent_readpages(tree, mapping, pages, nr_pages);
8730 }
8731 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8732 {
8733 	struct extent_io_tree *tree;
8734 	struct extent_map_tree *map;
8735 	int ret;
8736 
8737 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8738 	map = &BTRFS_I(page->mapping->host)->extent_tree;
8739 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8740 	if (ret == 1) {
8741 		ClearPagePrivate(page);
8742 		set_page_private(page, 0);
8743 		put_page(page);
8744 	}
8745 	return ret;
8746 }
8747 
8748 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8749 {
8750 	if (PageWriteback(page) || PageDirty(page))
8751 		return 0;
8752 	return __btrfs_releasepage(page, gfp_flags);
8753 }
8754 
8755 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8756 				 unsigned int length)
8757 {
8758 	struct inode *inode = page->mapping->host;
8759 	struct extent_io_tree *tree;
8760 	struct btrfs_ordered_extent *ordered;
8761 	struct extent_state *cached_state = NULL;
8762 	u64 page_start = page_offset(page);
8763 	u64 page_end = page_start + PAGE_SIZE - 1;
8764 	u64 start;
8765 	u64 end;
8766 	int inode_evicting = inode->i_state & I_FREEING;
8767 
8768 	/*
8769 	 * we have the page locked, so new writeback can't start,
8770 	 * and the dirty bit won't be cleared while we are here.
8771 	 *
8772 	 * Wait for IO on this page so that we can safely clear
8773 	 * the PagePrivate2 bit and do ordered accounting
8774 	 */
8775 	wait_on_page_writeback(page);
8776 
8777 	tree = &BTRFS_I(inode)->io_tree;
8778 	if (offset) {
8779 		btrfs_releasepage(page, GFP_NOFS);
8780 		return;
8781 	}
8782 
8783 	if (!inode_evicting)
8784 		lock_extent_bits(tree, page_start, page_end, &cached_state);
8785 again:
8786 	start = page_start;
8787 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8788 					page_end - start + 1);
8789 	if (ordered) {
8790 		end = min(page_end, ordered->file_offset + ordered->len - 1);
8791 		/*
8792 		 * IO on this page will never be started, so we need
8793 		 * to account for any ordered extents now
8794 		 */
8795 		if (!inode_evicting)
8796 			clear_extent_bit(tree, start, end,
8797 					 EXTENT_DIRTY | EXTENT_DELALLOC |
8798 					 EXTENT_DELALLOC_NEW |
8799 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8800 					 EXTENT_DEFRAG, 1, 0, &cached_state);
8801 		/*
8802 		 * whoever cleared the private bit is responsible
8803 		 * for the finish_ordered_io
8804 		 */
8805 		if (TestClearPagePrivate2(page)) {
8806 			struct btrfs_ordered_inode_tree *tree;
8807 			u64 new_len;
8808 
8809 			tree = &BTRFS_I(inode)->ordered_tree;
8810 
8811 			spin_lock_irq(&tree->lock);
8812 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8813 			new_len = start - ordered->file_offset;
8814 			if (new_len < ordered->truncated_len)
8815 				ordered->truncated_len = new_len;
8816 			spin_unlock_irq(&tree->lock);
8817 
8818 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
8819 							   start,
8820 							   end - start + 1, 1))
8821 				btrfs_finish_ordered_io(ordered);
8822 		}
8823 		btrfs_put_ordered_extent(ordered);
8824 		if (!inode_evicting) {
8825 			cached_state = NULL;
8826 			lock_extent_bits(tree, start, end,
8827 					 &cached_state);
8828 		}
8829 
8830 		start = end + 1;
8831 		if (start < page_end)
8832 			goto again;
8833 	}
8834 
8835 	/*
8836 	 * Qgroup reserved space handler
8837 	 * Page here will be either
8838 	 * 1) Already written to disk
8839 	 *    In this case, its reserved space is released from data rsv map
8840 	 *    and will be freed by delayed_ref handler finally.
8841 	 *    So even we call qgroup_free_data(), it won't decrease reserved
8842 	 *    space.
8843 	 * 2) Not written to disk
8844 	 *    This means the reserved space should be freed here. However,
8845 	 *    if a truncate invalidates the page (by clearing PageDirty)
8846 	 *    and the page is accounted for while allocating extent
8847 	 *    in btrfs_check_data_free_space() we let delayed_ref to
8848 	 *    free the entire extent.
8849 	 */
8850 	if (PageDirty(page))
8851 		btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8852 	if (!inode_evicting) {
8853 		clear_extent_bit(tree, page_start, page_end,
8854 				 EXTENT_LOCKED | EXTENT_DIRTY |
8855 				 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8856 				 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8857 				 &cached_state);
8858 
8859 		__btrfs_releasepage(page, GFP_NOFS);
8860 	}
8861 
8862 	ClearPageChecked(page);
8863 	if (PagePrivate(page)) {
8864 		ClearPagePrivate(page);
8865 		set_page_private(page, 0);
8866 		put_page(page);
8867 	}
8868 }
8869 
8870 /*
8871  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8872  * called from a page fault handler when a page is first dirtied. Hence we must
8873  * be careful to check for EOF conditions here. We set the page up correctly
8874  * for a written page which means we get ENOSPC checking when writing into
8875  * holes and correct delalloc and unwritten extent mapping on filesystems that
8876  * support these features.
8877  *
8878  * We are not allowed to take the i_mutex here so we have to play games to
8879  * protect against truncate races as the page could now be beyond EOF.  Because
8880  * vmtruncate() writes the inode size before removing pages, once we have the
8881  * page lock we can determine safely if the page is beyond EOF. If it is not
8882  * beyond EOF, then the page is guaranteed safe against truncation until we
8883  * unlock the page.
8884  */
8885 int btrfs_page_mkwrite(struct vm_fault *vmf)
8886 {
8887 	struct page *page = vmf->page;
8888 	struct inode *inode = file_inode(vmf->vma->vm_file);
8889 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8890 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8891 	struct btrfs_ordered_extent *ordered;
8892 	struct extent_state *cached_state = NULL;
8893 	struct extent_changeset *data_reserved = NULL;
8894 	char *kaddr;
8895 	unsigned long zero_start;
8896 	loff_t size;
8897 	int ret;
8898 	int reserved = 0;
8899 	u64 reserved_space;
8900 	u64 page_start;
8901 	u64 page_end;
8902 	u64 end;
8903 
8904 	reserved_space = PAGE_SIZE;
8905 
8906 	sb_start_pagefault(inode->i_sb);
8907 	page_start = page_offset(page);
8908 	page_end = page_start + PAGE_SIZE - 1;
8909 	end = page_end;
8910 
8911 	/*
8912 	 * Reserving delalloc space after obtaining the page lock can lead to
8913 	 * deadlock. For example, if a dirty page is locked by this function
8914 	 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8915 	 * dirty page write out, then the btrfs_writepage() function could
8916 	 * end up waiting indefinitely to get a lock on the page currently
8917 	 * being processed by btrfs_page_mkwrite() function.
8918 	 */
8919 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8920 					   reserved_space);
8921 	if (!ret) {
8922 		ret = file_update_time(vmf->vma->vm_file);
8923 		reserved = 1;
8924 	}
8925 	if (ret) {
8926 		if (ret == -ENOMEM)
8927 			ret = VM_FAULT_OOM;
8928 		else /* -ENOSPC, -EIO, etc */
8929 			ret = VM_FAULT_SIGBUS;
8930 		if (reserved)
8931 			goto out;
8932 		goto out_noreserve;
8933 	}
8934 
8935 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8936 again:
8937 	lock_page(page);
8938 	size = i_size_read(inode);
8939 
8940 	if ((page->mapping != inode->i_mapping) ||
8941 	    (page_start >= size)) {
8942 		/* page got truncated out from underneath us */
8943 		goto out_unlock;
8944 	}
8945 	wait_on_page_writeback(page);
8946 
8947 	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8948 	set_page_extent_mapped(page);
8949 
8950 	/*
8951 	 * we can't set the delalloc bits if there are pending ordered
8952 	 * extents.  Drop our locks and wait for them to finish
8953 	 */
8954 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8955 			PAGE_SIZE);
8956 	if (ordered) {
8957 		unlock_extent_cached(io_tree, page_start, page_end,
8958 				     &cached_state);
8959 		unlock_page(page);
8960 		btrfs_start_ordered_extent(inode, ordered, 1);
8961 		btrfs_put_ordered_extent(ordered);
8962 		goto again;
8963 	}
8964 
8965 	if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8966 		reserved_space = round_up(size - page_start,
8967 					  fs_info->sectorsize);
8968 		if (reserved_space < PAGE_SIZE) {
8969 			end = page_start + reserved_space - 1;
8970 			btrfs_delalloc_release_space(inode, data_reserved,
8971 					page_start, PAGE_SIZE - reserved_space,
8972 					true);
8973 		}
8974 	}
8975 
8976 	/*
8977 	 * page_mkwrite gets called when the page is firstly dirtied after it's
8978 	 * faulted in, but write(2) could also dirty a page and set delalloc
8979 	 * bits, thus in this case for space account reason, we still need to
8980 	 * clear any delalloc bits within this page range since we have to
8981 	 * reserve data&meta space before lock_page() (see above comments).
8982 	 */
8983 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8984 			  EXTENT_DIRTY | EXTENT_DELALLOC |
8985 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8986 			  0, 0, &cached_state);
8987 
8988 	ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8989 					&cached_state, 0);
8990 	if (ret) {
8991 		unlock_extent_cached(io_tree, page_start, page_end,
8992 				     &cached_state);
8993 		ret = VM_FAULT_SIGBUS;
8994 		goto out_unlock;
8995 	}
8996 	ret = 0;
8997 
8998 	/* page is wholly or partially inside EOF */
8999 	if (page_start + PAGE_SIZE > size)
9000 		zero_start = size & ~PAGE_MASK;
9001 	else
9002 		zero_start = PAGE_SIZE;
9003 
9004 	if (zero_start != PAGE_SIZE) {
9005 		kaddr = kmap(page);
9006 		memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9007 		flush_dcache_page(page);
9008 		kunmap(page);
9009 	}
9010 	ClearPageChecked(page);
9011 	set_page_dirty(page);
9012 	SetPageUptodate(page);
9013 
9014 	BTRFS_I(inode)->last_trans = fs_info->generation;
9015 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9016 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9017 
9018 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9019 
9020 out_unlock:
9021 	if (!ret) {
9022 		btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
9023 		sb_end_pagefault(inode->i_sb);
9024 		extent_changeset_free(data_reserved);
9025 		return VM_FAULT_LOCKED;
9026 	}
9027 	unlock_page(page);
9028 out:
9029 	btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
9030 	btrfs_delalloc_release_space(inode, data_reserved, page_start,
9031 				     reserved_space, (ret != 0));
9032 out_noreserve:
9033 	sb_end_pagefault(inode->i_sb);
9034 	extent_changeset_free(data_reserved);
9035 	return ret;
9036 }
9037 
9038 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9039 {
9040 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9041 	struct btrfs_root *root = BTRFS_I(inode)->root;
9042 	struct btrfs_block_rsv *rsv;
9043 	int ret = 0;
9044 	int err = 0;
9045 	struct btrfs_trans_handle *trans;
9046 	u64 mask = fs_info->sectorsize - 1;
9047 	u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9048 
9049 	if (!skip_writeback) {
9050 		ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9051 					       (u64)-1);
9052 		if (ret)
9053 			return ret;
9054 	}
9055 
9056 	/*
9057 	 * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9058 	 * 3 things going on here
9059 	 *
9060 	 * 1) We need to reserve space for our orphan item and the space to
9061 	 * delete our orphan item.  Lord knows we don't want to have a dangling
9062 	 * orphan item because we didn't reserve space to remove it.
9063 	 *
9064 	 * 2) We need to reserve space to update our inode.
9065 	 *
9066 	 * 3) We need to have something to cache all the space that is going to
9067 	 * be free'd up by the truncate operation, but also have some slack
9068 	 * space reserved in case it uses space during the truncate (thank you
9069 	 * very much snapshotting).
9070 	 *
9071 	 * And we need these to all be separate.  The fact is we can use a lot of
9072 	 * space doing the truncate, and we have no earthly idea how much space
9073 	 * we will use, so we need the truncate reservation to be separate so it
9074 	 * doesn't end up using space reserved for updating the inode or
9075 	 * removing the orphan item.  We also need to be able to stop the
9076 	 * transaction and start a new one, which means we need to be able to
9077 	 * update the inode several times, and we have no idea of knowing how
9078 	 * many times that will be, so we can't just reserve 1 item for the
9079 	 * entirety of the operation, so that has to be done separately as well.
9080 	 * Then there is the orphan item, which does indeed need to be held on
9081 	 * to for the whole operation, and we need nobody to touch this reserved
9082 	 * space except the orphan code.
9083 	 *
9084 	 * So that leaves us with
9085 	 *
9086 	 * 1) root->orphan_block_rsv - for the orphan deletion.
9087 	 * 2) rsv - for the truncate reservation, which we will steal from the
9088 	 * transaction reservation.
9089 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9090 	 * updating the inode.
9091 	 */
9092 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9093 	if (!rsv)
9094 		return -ENOMEM;
9095 	rsv->size = min_size;
9096 	rsv->failfast = 1;
9097 
9098 	/*
9099 	 * 1 for the truncate slack space
9100 	 * 1 for updating the inode.
9101 	 */
9102 	trans = btrfs_start_transaction(root, 2);
9103 	if (IS_ERR(trans)) {
9104 		err = PTR_ERR(trans);
9105 		goto out;
9106 	}
9107 
9108 	/* Migrate the slack space for the truncate to our reserve */
9109 	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9110 				      min_size, 0);
9111 	BUG_ON(ret);
9112 
9113 	/*
9114 	 * So if we truncate and then write and fsync we normally would just
9115 	 * write the extents that changed, which is a problem if we need to
9116 	 * first truncate that entire inode.  So set this flag so we write out
9117 	 * all of the extents in the inode to the sync log so we're completely
9118 	 * safe.
9119 	 */
9120 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9121 	trans->block_rsv = rsv;
9122 
9123 	while (1) {
9124 		ret = btrfs_truncate_inode_items(trans, root, inode,
9125 						 inode->i_size,
9126 						 BTRFS_EXTENT_DATA_KEY);
9127 		trans->block_rsv = &fs_info->trans_block_rsv;
9128 		if (ret != -ENOSPC && ret != -EAGAIN) {
9129 			err = ret;
9130 			break;
9131 		}
9132 
9133 		ret = btrfs_update_inode(trans, root, inode);
9134 		if (ret) {
9135 			err = ret;
9136 			break;
9137 		}
9138 
9139 		btrfs_end_transaction(trans);
9140 		btrfs_btree_balance_dirty(fs_info);
9141 
9142 		trans = btrfs_start_transaction(root, 2);
9143 		if (IS_ERR(trans)) {
9144 			ret = err = PTR_ERR(trans);
9145 			trans = NULL;
9146 			break;
9147 		}
9148 
9149 		btrfs_block_rsv_release(fs_info, rsv, -1);
9150 		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9151 					      rsv, min_size, 0);
9152 		BUG_ON(ret);	/* shouldn't happen */
9153 		trans->block_rsv = rsv;
9154 	}
9155 
9156 	/*
9157 	 * We can't call btrfs_truncate_block inside a trans handle as we could
9158 	 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9159 	 * we've truncated everything except the last little bit, and can do
9160 	 * btrfs_truncate_block and then update the disk_i_size.
9161 	 */
9162 	if (ret == NEED_TRUNCATE_BLOCK) {
9163 		btrfs_end_transaction(trans);
9164 		btrfs_btree_balance_dirty(fs_info);
9165 
9166 		ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9167 		if (ret)
9168 			goto out;
9169 		trans = btrfs_start_transaction(root, 1);
9170 		if (IS_ERR(trans)) {
9171 			ret = PTR_ERR(trans);
9172 			goto out;
9173 		}
9174 		btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9175 	}
9176 
9177 	if (ret == 0 && inode->i_nlink > 0) {
9178 		trans->block_rsv = root->orphan_block_rsv;
9179 		ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9180 		if (ret)
9181 			err = ret;
9182 	}
9183 
9184 	if (trans) {
9185 		trans->block_rsv = &fs_info->trans_block_rsv;
9186 		ret = btrfs_update_inode(trans, root, inode);
9187 		if (ret && !err)
9188 			err = ret;
9189 
9190 		ret = btrfs_end_transaction(trans);
9191 		btrfs_btree_balance_dirty(fs_info);
9192 	}
9193 out:
9194 	btrfs_free_block_rsv(fs_info, rsv);
9195 
9196 	if (ret && !err)
9197 		err = ret;
9198 
9199 	return err;
9200 }
9201 
9202 /*
9203  * create a new subvolume directory/inode (helper for the ioctl).
9204  */
9205 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9206 			     struct btrfs_root *new_root,
9207 			     struct btrfs_root *parent_root,
9208 			     u64 new_dirid)
9209 {
9210 	struct inode *inode;
9211 	int err;
9212 	u64 index = 0;
9213 
9214 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9215 				new_dirid, new_dirid,
9216 				S_IFDIR | (~current_umask() & S_IRWXUGO),
9217 				&index);
9218 	if (IS_ERR(inode))
9219 		return PTR_ERR(inode);
9220 	inode->i_op = &btrfs_dir_inode_operations;
9221 	inode->i_fop = &btrfs_dir_file_operations;
9222 
9223 	set_nlink(inode, 1);
9224 	btrfs_i_size_write(BTRFS_I(inode), 0);
9225 	unlock_new_inode(inode);
9226 
9227 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9228 	if (err)
9229 		btrfs_err(new_root->fs_info,
9230 			  "error inheriting subvolume %llu properties: %d",
9231 			  new_root->root_key.objectid, err);
9232 
9233 	err = btrfs_update_inode(trans, new_root, inode);
9234 
9235 	iput(inode);
9236 	return err;
9237 }
9238 
9239 struct inode *btrfs_alloc_inode(struct super_block *sb)
9240 {
9241 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9242 	struct btrfs_inode *ei;
9243 	struct inode *inode;
9244 
9245 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9246 	if (!ei)
9247 		return NULL;
9248 
9249 	ei->root = NULL;
9250 	ei->generation = 0;
9251 	ei->last_trans = 0;
9252 	ei->last_sub_trans = 0;
9253 	ei->logged_trans = 0;
9254 	ei->delalloc_bytes = 0;
9255 	ei->new_delalloc_bytes = 0;
9256 	ei->defrag_bytes = 0;
9257 	ei->disk_i_size = 0;
9258 	ei->flags = 0;
9259 	ei->csum_bytes = 0;
9260 	ei->index_cnt = (u64)-1;
9261 	ei->dir_index = 0;
9262 	ei->last_unlink_trans = 0;
9263 	ei->last_log_commit = 0;
9264 
9265 	spin_lock_init(&ei->lock);
9266 	ei->outstanding_extents = 0;
9267 	if (sb->s_magic != BTRFS_TEST_MAGIC)
9268 		btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9269 					      BTRFS_BLOCK_RSV_DELALLOC);
9270 	ei->runtime_flags = 0;
9271 	ei->prop_compress = BTRFS_COMPRESS_NONE;
9272 	ei->defrag_compress = BTRFS_COMPRESS_NONE;
9273 
9274 	ei->delayed_node = NULL;
9275 
9276 	ei->i_otime.tv_sec = 0;
9277 	ei->i_otime.tv_nsec = 0;
9278 
9279 	inode = &ei->vfs_inode;
9280 	extent_map_tree_init(&ei->extent_tree);
9281 	extent_io_tree_init(&ei->io_tree, inode);
9282 	extent_io_tree_init(&ei->io_failure_tree, inode);
9283 	ei->io_tree.track_uptodate = 1;
9284 	ei->io_failure_tree.track_uptodate = 1;
9285 	atomic_set(&ei->sync_writers, 0);
9286 	mutex_init(&ei->log_mutex);
9287 	mutex_init(&ei->delalloc_mutex);
9288 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9289 	INIT_LIST_HEAD(&ei->delalloc_inodes);
9290 	INIT_LIST_HEAD(&ei->delayed_iput);
9291 	RB_CLEAR_NODE(&ei->rb_node);
9292 	init_rwsem(&ei->dio_sem);
9293 
9294 	return inode;
9295 }
9296 
9297 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9298 void btrfs_test_destroy_inode(struct inode *inode)
9299 {
9300 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9301 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9302 }
9303 #endif
9304 
9305 static void btrfs_i_callback(struct rcu_head *head)
9306 {
9307 	struct inode *inode = container_of(head, struct inode, i_rcu);
9308 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9309 }
9310 
9311 void btrfs_destroy_inode(struct inode *inode)
9312 {
9313 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9314 	struct btrfs_ordered_extent *ordered;
9315 	struct btrfs_root *root = BTRFS_I(inode)->root;
9316 
9317 	WARN_ON(!hlist_empty(&inode->i_dentry));
9318 	WARN_ON(inode->i_data.nrpages);
9319 	WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9320 	WARN_ON(BTRFS_I(inode)->block_rsv.size);
9321 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
9322 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9323 	WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9324 	WARN_ON(BTRFS_I(inode)->csum_bytes);
9325 	WARN_ON(BTRFS_I(inode)->defrag_bytes);
9326 
9327 	/*
9328 	 * This can happen where we create an inode, but somebody else also
9329 	 * created the same inode and we need to destroy the one we already
9330 	 * created.
9331 	 */
9332 	if (!root)
9333 		goto free;
9334 
9335 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9336 		     &BTRFS_I(inode)->runtime_flags)) {
9337 		btrfs_info(fs_info, "inode %llu still on the orphan list",
9338 			   btrfs_ino(BTRFS_I(inode)));
9339 		atomic_dec(&root->orphan_inodes);
9340 	}
9341 
9342 	while (1) {
9343 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9344 		if (!ordered)
9345 			break;
9346 		else {
9347 			btrfs_err(fs_info,
9348 				  "found ordered extent %llu %llu on inode cleanup",
9349 				  ordered->file_offset, ordered->len);
9350 			btrfs_remove_ordered_extent(inode, ordered);
9351 			btrfs_put_ordered_extent(ordered);
9352 			btrfs_put_ordered_extent(ordered);
9353 		}
9354 	}
9355 	btrfs_qgroup_check_reserved_leak(inode);
9356 	inode_tree_del(inode);
9357 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9358 free:
9359 	call_rcu(&inode->i_rcu, btrfs_i_callback);
9360 }
9361 
9362 int btrfs_drop_inode(struct inode *inode)
9363 {
9364 	struct btrfs_root *root = BTRFS_I(inode)->root;
9365 
9366 	if (root == NULL)
9367 		return 1;
9368 
9369 	/* the snap/subvol tree is on deleting */
9370 	if (btrfs_root_refs(&root->root_item) == 0)
9371 		return 1;
9372 	else
9373 		return generic_drop_inode(inode);
9374 }
9375 
9376 static void init_once(void *foo)
9377 {
9378 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9379 
9380 	inode_init_once(&ei->vfs_inode);
9381 }
9382 
9383 void __cold btrfs_destroy_cachep(void)
9384 {
9385 	/*
9386 	 * Make sure all delayed rcu free inodes are flushed before we
9387 	 * destroy cache.
9388 	 */
9389 	rcu_barrier();
9390 	kmem_cache_destroy(btrfs_inode_cachep);
9391 	kmem_cache_destroy(btrfs_trans_handle_cachep);
9392 	kmem_cache_destroy(btrfs_path_cachep);
9393 	kmem_cache_destroy(btrfs_free_space_cachep);
9394 }
9395 
9396 int __init btrfs_init_cachep(void)
9397 {
9398 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9399 			sizeof(struct btrfs_inode), 0,
9400 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9401 			init_once);
9402 	if (!btrfs_inode_cachep)
9403 		goto fail;
9404 
9405 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9406 			sizeof(struct btrfs_trans_handle), 0,
9407 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9408 	if (!btrfs_trans_handle_cachep)
9409 		goto fail;
9410 
9411 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
9412 			sizeof(struct btrfs_path), 0,
9413 			SLAB_MEM_SPREAD, NULL);
9414 	if (!btrfs_path_cachep)
9415 		goto fail;
9416 
9417 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9418 			sizeof(struct btrfs_free_space), 0,
9419 			SLAB_MEM_SPREAD, NULL);
9420 	if (!btrfs_free_space_cachep)
9421 		goto fail;
9422 
9423 	return 0;
9424 fail:
9425 	btrfs_destroy_cachep();
9426 	return -ENOMEM;
9427 }
9428 
9429 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9430 			 u32 request_mask, unsigned int flags)
9431 {
9432 	u64 delalloc_bytes;
9433 	struct inode *inode = d_inode(path->dentry);
9434 	u32 blocksize = inode->i_sb->s_blocksize;
9435 	u32 bi_flags = BTRFS_I(inode)->flags;
9436 
9437 	stat->result_mask |= STATX_BTIME;
9438 	stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9439 	stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9440 	if (bi_flags & BTRFS_INODE_APPEND)
9441 		stat->attributes |= STATX_ATTR_APPEND;
9442 	if (bi_flags & BTRFS_INODE_COMPRESS)
9443 		stat->attributes |= STATX_ATTR_COMPRESSED;
9444 	if (bi_flags & BTRFS_INODE_IMMUTABLE)
9445 		stat->attributes |= STATX_ATTR_IMMUTABLE;
9446 	if (bi_flags & BTRFS_INODE_NODUMP)
9447 		stat->attributes |= STATX_ATTR_NODUMP;
9448 
9449 	stat->attributes_mask |= (STATX_ATTR_APPEND |
9450 				  STATX_ATTR_COMPRESSED |
9451 				  STATX_ATTR_IMMUTABLE |
9452 				  STATX_ATTR_NODUMP);
9453 
9454 	generic_fillattr(inode, stat);
9455 	stat->dev = BTRFS_I(inode)->root->anon_dev;
9456 
9457 	spin_lock(&BTRFS_I(inode)->lock);
9458 	delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9459 	spin_unlock(&BTRFS_I(inode)->lock);
9460 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9461 			ALIGN(delalloc_bytes, blocksize)) >> 9;
9462 	return 0;
9463 }
9464 
9465 static int btrfs_rename_exchange(struct inode *old_dir,
9466 			      struct dentry *old_dentry,
9467 			      struct inode *new_dir,
9468 			      struct dentry *new_dentry)
9469 {
9470 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9471 	struct btrfs_trans_handle *trans;
9472 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9473 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9474 	struct inode *new_inode = new_dentry->d_inode;
9475 	struct inode *old_inode = old_dentry->d_inode;
9476 	struct timespec ctime = current_time(old_inode);
9477 	struct dentry *parent;
9478 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9479 	u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9480 	u64 old_idx = 0;
9481 	u64 new_idx = 0;
9482 	u64 root_objectid;
9483 	int ret;
9484 	bool root_log_pinned = false;
9485 	bool dest_log_pinned = false;
9486 
9487 	/* we only allow rename subvolume link between subvolumes */
9488 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9489 		return -EXDEV;
9490 
9491 	/* close the race window with snapshot create/destroy ioctl */
9492 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9493 		down_read(&fs_info->subvol_sem);
9494 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9495 		down_read(&fs_info->subvol_sem);
9496 
9497 	/*
9498 	 * We want to reserve the absolute worst case amount of items.  So if
9499 	 * both inodes are subvols and we need to unlink them then that would
9500 	 * require 4 item modifications, but if they are both normal inodes it
9501 	 * would require 5 item modifications, so we'll assume their normal
9502 	 * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9503 	 * should cover the worst case number of items we'll modify.
9504 	 */
9505 	trans = btrfs_start_transaction(root, 12);
9506 	if (IS_ERR(trans)) {
9507 		ret = PTR_ERR(trans);
9508 		goto out_notrans;
9509 	}
9510 
9511 	/*
9512 	 * We need to find a free sequence number both in the source and
9513 	 * in the destination directory for the exchange.
9514 	 */
9515 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9516 	if (ret)
9517 		goto out_fail;
9518 	ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9519 	if (ret)
9520 		goto out_fail;
9521 
9522 	BTRFS_I(old_inode)->dir_index = 0ULL;
9523 	BTRFS_I(new_inode)->dir_index = 0ULL;
9524 
9525 	/* Reference for the source. */
9526 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9527 		/* force full log commit if subvolume involved. */
9528 		btrfs_set_log_full_commit(fs_info, trans);
9529 	} else {
9530 		btrfs_pin_log_trans(root);
9531 		root_log_pinned = true;
9532 		ret = btrfs_insert_inode_ref(trans, dest,
9533 					     new_dentry->d_name.name,
9534 					     new_dentry->d_name.len,
9535 					     old_ino,
9536 					     btrfs_ino(BTRFS_I(new_dir)),
9537 					     old_idx);
9538 		if (ret)
9539 			goto out_fail;
9540 	}
9541 
9542 	/* And now for the dest. */
9543 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9544 		/* force full log commit if subvolume involved. */
9545 		btrfs_set_log_full_commit(fs_info, trans);
9546 	} else {
9547 		btrfs_pin_log_trans(dest);
9548 		dest_log_pinned = true;
9549 		ret = btrfs_insert_inode_ref(trans, root,
9550 					     old_dentry->d_name.name,
9551 					     old_dentry->d_name.len,
9552 					     new_ino,
9553 					     btrfs_ino(BTRFS_I(old_dir)),
9554 					     new_idx);
9555 		if (ret)
9556 			goto out_fail;
9557 	}
9558 
9559 	/* Update inode version and ctime/mtime. */
9560 	inode_inc_iversion(old_dir);
9561 	inode_inc_iversion(new_dir);
9562 	inode_inc_iversion(old_inode);
9563 	inode_inc_iversion(new_inode);
9564 	old_dir->i_ctime = old_dir->i_mtime = ctime;
9565 	new_dir->i_ctime = new_dir->i_mtime = ctime;
9566 	old_inode->i_ctime = ctime;
9567 	new_inode->i_ctime = ctime;
9568 
9569 	if (old_dentry->d_parent != new_dentry->d_parent) {
9570 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9571 				BTRFS_I(old_inode), 1);
9572 		btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9573 				BTRFS_I(new_inode), 1);
9574 	}
9575 
9576 	/* src is a subvolume */
9577 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9578 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9579 		ret = btrfs_unlink_subvol(trans, root, old_dir,
9580 					  root_objectid,
9581 					  old_dentry->d_name.name,
9582 					  old_dentry->d_name.len);
9583 	} else { /* src is an inode */
9584 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9585 					   BTRFS_I(old_dentry->d_inode),
9586 					   old_dentry->d_name.name,
9587 					   old_dentry->d_name.len);
9588 		if (!ret)
9589 			ret = btrfs_update_inode(trans, root, old_inode);
9590 	}
9591 	if (ret) {
9592 		btrfs_abort_transaction(trans, ret);
9593 		goto out_fail;
9594 	}
9595 
9596 	/* dest is a subvolume */
9597 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9598 		root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9599 		ret = btrfs_unlink_subvol(trans, dest, new_dir,
9600 					  root_objectid,
9601 					  new_dentry->d_name.name,
9602 					  new_dentry->d_name.len);
9603 	} else { /* dest is an inode */
9604 		ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9605 					   BTRFS_I(new_dentry->d_inode),
9606 					   new_dentry->d_name.name,
9607 					   new_dentry->d_name.len);
9608 		if (!ret)
9609 			ret = btrfs_update_inode(trans, dest, new_inode);
9610 	}
9611 	if (ret) {
9612 		btrfs_abort_transaction(trans, ret);
9613 		goto out_fail;
9614 	}
9615 
9616 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9617 			     new_dentry->d_name.name,
9618 			     new_dentry->d_name.len, 0, old_idx);
9619 	if (ret) {
9620 		btrfs_abort_transaction(trans, ret);
9621 		goto out_fail;
9622 	}
9623 
9624 	ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9625 			     old_dentry->d_name.name,
9626 			     old_dentry->d_name.len, 0, new_idx);
9627 	if (ret) {
9628 		btrfs_abort_transaction(trans, ret);
9629 		goto out_fail;
9630 	}
9631 
9632 	if (old_inode->i_nlink == 1)
9633 		BTRFS_I(old_inode)->dir_index = old_idx;
9634 	if (new_inode->i_nlink == 1)
9635 		BTRFS_I(new_inode)->dir_index = new_idx;
9636 
9637 	if (root_log_pinned) {
9638 		parent = new_dentry->d_parent;
9639 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9640 				parent);
9641 		btrfs_end_log_trans(root);
9642 		root_log_pinned = false;
9643 	}
9644 	if (dest_log_pinned) {
9645 		parent = old_dentry->d_parent;
9646 		btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9647 				parent);
9648 		btrfs_end_log_trans(dest);
9649 		dest_log_pinned = false;
9650 	}
9651 out_fail:
9652 	/*
9653 	 * If we have pinned a log and an error happened, we unpin tasks
9654 	 * trying to sync the log and force them to fallback to a transaction
9655 	 * commit if the log currently contains any of the inodes involved in
9656 	 * this rename operation (to ensure we do not persist a log with an
9657 	 * inconsistent state for any of these inodes or leading to any
9658 	 * inconsistencies when replayed). If the transaction was aborted, the
9659 	 * abortion reason is propagated to userspace when attempting to commit
9660 	 * the transaction. If the log does not contain any of these inodes, we
9661 	 * allow the tasks to sync it.
9662 	 */
9663 	if (ret && (root_log_pinned || dest_log_pinned)) {
9664 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9665 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9666 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9667 		    (new_inode &&
9668 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9669 			btrfs_set_log_full_commit(fs_info, trans);
9670 
9671 		if (root_log_pinned) {
9672 			btrfs_end_log_trans(root);
9673 			root_log_pinned = false;
9674 		}
9675 		if (dest_log_pinned) {
9676 			btrfs_end_log_trans(dest);
9677 			dest_log_pinned = false;
9678 		}
9679 	}
9680 	ret = btrfs_end_transaction(trans);
9681 out_notrans:
9682 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9683 		up_read(&fs_info->subvol_sem);
9684 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9685 		up_read(&fs_info->subvol_sem);
9686 
9687 	return ret;
9688 }
9689 
9690 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9691 				     struct btrfs_root *root,
9692 				     struct inode *dir,
9693 				     struct dentry *dentry)
9694 {
9695 	int ret;
9696 	struct inode *inode;
9697 	u64 objectid;
9698 	u64 index;
9699 
9700 	ret = btrfs_find_free_ino(root, &objectid);
9701 	if (ret)
9702 		return ret;
9703 
9704 	inode = btrfs_new_inode(trans, root, dir,
9705 				dentry->d_name.name,
9706 				dentry->d_name.len,
9707 				btrfs_ino(BTRFS_I(dir)),
9708 				objectid,
9709 				S_IFCHR | WHITEOUT_MODE,
9710 				&index);
9711 
9712 	if (IS_ERR(inode)) {
9713 		ret = PTR_ERR(inode);
9714 		return ret;
9715 	}
9716 
9717 	inode->i_op = &btrfs_special_inode_operations;
9718 	init_special_inode(inode, inode->i_mode,
9719 		WHITEOUT_DEV);
9720 
9721 	ret = btrfs_init_inode_security(trans, inode, dir,
9722 				&dentry->d_name);
9723 	if (ret)
9724 		goto out;
9725 
9726 	ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9727 				BTRFS_I(inode), 0, index);
9728 	if (ret)
9729 		goto out;
9730 
9731 	ret = btrfs_update_inode(trans, root, inode);
9732 out:
9733 	unlock_new_inode(inode);
9734 	if (ret)
9735 		inode_dec_link_count(inode);
9736 	iput(inode);
9737 
9738 	return ret;
9739 }
9740 
9741 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9742 			   struct inode *new_dir, struct dentry *new_dentry,
9743 			   unsigned int flags)
9744 {
9745 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9746 	struct btrfs_trans_handle *trans;
9747 	unsigned int trans_num_items;
9748 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9749 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9750 	struct inode *new_inode = d_inode(new_dentry);
9751 	struct inode *old_inode = d_inode(old_dentry);
9752 	u64 index = 0;
9753 	u64 root_objectid;
9754 	int ret;
9755 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9756 	bool log_pinned = false;
9757 
9758 	if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9759 		return -EPERM;
9760 
9761 	/* we only allow rename subvolume link between subvolumes */
9762 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9763 		return -EXDEV;
9764 
9765 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9766 	    (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9767 		return -ENOTEMPTY;
9768 
9769 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
9770 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9771 		return -ENOTEMPTY;
9772 
9773 
9774 	/* check for collisions, even if the  name isn't there */
9775 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9776 			     new_dentry->d_name.name,
9777 			     new_dentry->d_name.len);
9778 
9779 	if (ret) {
9780 		if (ret == -EEXIST) {
9781 			/* we shouldn't get
9782 			 * eexist without a new_inode */
9783 			if (WARN_ON(!new_inode)) {
9784 				return ret;
9785 			}
9786 		} else {
9787 			/* maybe -EOVERFLOW */
9788 			return ret;
9789 		}
9790 	}
9791 	ret = 0;
9792 
9793 	/*
9794 	 * we're using rename to replace one file with another.  Start IO on it
9795 	 * now so  we don't add too much work to the end of the transaction
9796 	 */
9797 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9798 		filemap_flush(old_inode->i_mapping);
9799 
9800 	/* close the racy window with snapshot create/destroy ioctl */
9801 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9802 		down_read(&fs_info->subvol_sem);
9803 	/*
9804 	 * We want to reserve the absolute worst case amount of items.  So if
9805 	 * both inodes are subvols and we need to unlink them then that would
9806 	 * require 4 item modifications, but if they are both normal inodes it
9807 	 * would require 5 item modifications, so we'll assume they are normal
9808 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9809 	 * should cover the worst case number of items we'll modify.
9810 	 * If our rename has the whiteout flag, we need more 5 units for the
9811 	 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9812 	 * when selinux is enabled).
9813 	 */
9814 	trans_num_items = 11;
9815 	if (flags & RENAME_WHITEOUT)
9816 		trans_num_items += 5;
9817 	trans = btrfs_start_transaction(root, trans_num_items);
9818 	if (IS_ERR(trans)) {
9819 		ret = PTR_ERR(trans);
9820 		goto out_notrans;
9821 	}
9822 
9823 	if (dest != root)
9824 		btrfs_record_root_in_trans(trans, dest);
9825 
9826 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9827 	if (ret)
9828 		goto out_fail;
9829 
9830 	BTRFS_I(old_inode)->dir_index = 0ULL;
9831 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9832 		/* force full log commit if subvolume involved. */
9833 		btrfs_set_log_full_commit(fs_info, trans);
9834 	} else {
9835 		btrfs_pin_log_trans(root);
9836 		log_pinned = true;
9837 		ret = btrfs_insert_inode_ref(trans, dest,
9838 					     new_dentry->d_name.name,
9839 					     new_dentry->d_name.len,
9840 					     old_ino,
9841 					     btrfs_ino(BTRFS_I(new_dir)), index);
9842 		if (ret)
9843 			goto out_fail;
9844 	}
9845 
9846 	inode_inc_iversion(old_dir);
9847 	inode_inc_iversion(new_dir);
9848 	inode_inc_iversion(old_inode);
9849 	old_dir->i_ctime = old_dir->i_mtime =
9850 	new_dir->i_ctime = new_dir->i_mtime =
9851 	old_inode->i_ctime = current_time(old_dir);
9852 
9853 	if (old_dentry->d_parent != new_dentry->d_parent)
9854 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9855 				BTRFS_I(old_inode), 1);
9856 
9857 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9858 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9859 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9860 					old_dentry->d_name.name,
9861 					old_dentry->d_name.len);
9862 	} else {
9863 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9864 					BTRFS_I(d_inode(old_dentry)),
9865 					old_dentry->d_name.name,
9866 					old_dentry->d_name.len);
9867 		if (!ret)
9868 			ret = btrfs_update_inode(trans, root, old_inode);
9869 	}
9870 	if (ret) {
9871 		btrfs_abort_transaction(trans, ret);
9872 		goto out_fail;
9873 	}
9874 
9875 	if (new_inode) {
9876 		inode_inc_iversion(new_inode);
9877 		new_inode->i_ctime = current_time(new_inode);
9878 		if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9879 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9880 			root_objectid = BTRFS_I(new_inode)->location.objectid;
9881 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
9882 						root_objectid,
9883 						new_dentry->d_name.name,
9884 						new_dentry->d_name.len);
9885 			BUG_ON(new_inode->i_nlink == 0);
9886 		} else {
9887 			ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9888 						 BTRFS_I(d_inode(new_dentry)),
9889 						 new_dentry->d_name.name,
9890 						 new_dentry->d_name.len);
9891 		}
9892 		if (!ret && new_inode->i_nlink == 0)
9893 			ret = btrfs_orphan_add(trans,
9894 					BTRFS_I(d_inode(new_dentry)));
9895 		if (ret) {
9896 			btrfs_abort_transaction(trans, ret);
9897 			goto out_fail;
9898 		}
9899 	}
9900 
9901 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9902 			     new_dentry->d_name.name,
9903 			     new_dentry->d_name.len, 0, index);
9904 	if (ret) {
9905 		btrfs_abort_transaction(trans, ret);
9906 		goto out_fail;
9907 	}
9908 
9909 	if (old_inode->i_nlink == 1)
9910 		BTRFS_I(old_inode)->dir_index = index;
9911 
9912 	if (log_pinned) {
9913 		struct dentry *parent = new_dentry->d_parent;
9914 
9915 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9916 				parent);
9917 		btrfs_end_log_trans(root);
9918 		log_pinned = false;
9919 	}
9920 
9921 	if (flags & RENAME_WHITEOUT) {
9922 		ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9923 						old_dentry);
9924 
9925 		if (ret) {
9926 			btrfs_abort_transaction(trans, ret);
9927 			goto out_fail;
9928 		}
9929 	}
9930 out_fail:
9931 	/*
9932 	 * If we have pinned the log and an error happened, we unpin tasks
9933 	 * trying to sync the log and force them to fallback to a transaction
9934 	 * commit if the log currently contains any of the inodes involved in
9935 	 * this rename operation (to ensure we do not persist a log with an
9936 	 * inconsistent state for any of these inodes or leading to any
9937 	 * inconsistencies when replayed). If the transaction was aborted, the
9938 	 * abortion reason is propagated to userspace when attempting to commit
9939 	 * the transaction. If the log does not contain any of these inodes, we
9940 	 * allow the tasks to sync it.
9941 	 */
9942 	if (ret && log_pinned) {
9943 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9944 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9945 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9946 		    (new_inode &&
9947 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9948 			btrfs_set_log_full_commit(fs_info, trans);
9949 
9950 		btrfs_end_log_trans(root);
9951 		log_pinned = false;
9952 	}
9953 	btrfs_end_transaction(trans);
9954 out_notrans:
9955 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9956 		up_read(&fs_info->subvol_sem);
9957 
9958 	return ret;
9959 }
9960 
9961 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9962 			 struct inode *new_dir, struct dentry *new_dentry,
9963 			 unsigned int flags)
9964 {
9965 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9966 		return -EINVAL;
9967 
9968 	if (flags & RENAME_EXCHANGE)
9969 		return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9970 					  new_dentry);
9971 
9972 	return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9973 }
9974 
9975 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9976 {
9977 	struct btrfs_delalloc_work *delalloc_work;
9978 	struct inode *inode;
9979 
9980 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
9981 				     work);
9982 	inode = delalloc_work->inode;
9983 	filemap_flush(inode->i_mapping);
9984 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9985 				&BTRFS_I(inode)->runtime_flags))
9986 		filemap_flush(inode->i_mapping);
9987 
9988 	if (delalloc_work->delay_iput)
9989 		btrfs_add_delayed_iput(inode);
9990 	else
9991 		iput(inode);
9992 	complete(&delalloc_work->completion);
9993 }
9994 
9995 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9996 						    int delay_iput)
9997 {
9998 	struct btrfs_delalloc_work *work;
9999 
10000 	work = kmalloc(sizeof(*work), GFP_NOFS);
10001 	if (!work)
10002 		return NULL;
10003 
10004 	init_completion(&work->completion);
10005 	INIT_LIST_HEAD(&work->list);
10006 	work->inode = inode;
10007 	work->delay_iput = delay_iput;
10008 	WARN_ON_ONCE(!inode);
10009 	btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10010 			btrfs_run_delalloc_work, NULL, NULL);
10011 
10012 	return work;
10013 }
10014 
10015 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10016 {
10017 	wait_for_completion(&work->completion);
10018 	kfree(work);
10019 }
10020 
10021 /*
10022  * some fairly slow code that needs optimization. This walks the list
10023  * of all the inodes with pending delalloc and forces them to disk.
10024  */
10025 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10026 				   int nr)
10027 {
10028 	struct btrfs_inode *binode;
10029 	struct inode *inode;
10030 	struct btrfs_delalloc_work *work, *next;
10031 	struct list_head works;
10032 	struct list_head splice;
10033 	int ret = 0;
10034 
10035 	INIT_LIST_HEAD(&works);
10036 	INIT_LIST_HEAD(&splice);
10037 
10038 	mutex_lock(&root->delalloc_mutex);
10039 	spin_lock(&root->delalloc_lock);
10040 	list_splice_init(&root->delalloc_inodes, &splice);
10041 	while (!list_empty(&splice)) {
10042 		binode = list_entry(splice.next, struct btrfs_inode,
10043 				    delalloc_inodes);
10044 
10045 		list_move_tail(&binode->delalloc_inodes,
10046 			       &root->delalloc_inodes);
10047 		inode = igrab(&binode->vfs_inode);
10048 		if (!inode) {
10049 			cond_resched_lock(&root->delalloc_lock);
10050 			continue;
10051 		}
10052 		spin_unlock(&root->delalloc_lock);
10053 
10054 		work = btrfs_alloc_delalloc_work(inode, delay_iput);
10055 		if (!work) {
10056 			if (delay_iput)
10057 				btrfs_add_delayed_iput(inode);
10058 			else
10059 				iput(inode);
10060 			ret = -ENOMEM;
10061 			goto out;
10062 		}
10063 		list_add_tail(&work->list, &works);
10064 		btrfs_queue_work(root->fs_info->flush_workers,
10065 				 &work->work);
10066 		ret++;
10067 		if (nr != -1 && ret >= nr)
10068 			goto out;
10069 		cond_resched();
10070 		spin_lock(&root->delalloc_lock);
10071 	}
10072 	spin_unlock(&root->delalloc_lock);
10073 
10074 out:
10075 	list_for_each_entry_safe(work, next, &works, list) {
10076 		list_del_init(&work->list);
10077 		btrfs_wait_and_free_delalloc_work(work);
10078 	}
10079 
10080 	if (!list_empty_careful(&splice)) {
10081 		spin_lock(&root->delalloc_lock);
10082 		list_splice_tail(&splice, &root->delalloc_inodes);
10083 		spin_unlock(&root->delalloc_lock);
10084 	}
10085 	mutex_unlock(&root->delalloc_mutex);
10086 	return ret;
10087 }
10088 
10089 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10090 {
10091 	struct btrfs_fs_info *fs_info = root->fs_info;
10092 	int ret;
10093 
10094 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10095 		return -EROFS;
10096 
10097 	ret = __start_delalloc_inodes(root, delay_iput, -1);
10098 	if (ret > 0)
10099 		ret = 0;
10100 	return ret;
10101 }
10102 
10103 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10104 			       int nr)
10105 {
10106 	struct btrfs_root *root;
10107 	struct list_head splice;
10108 	int ret;
10109 
10110 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10111 		return -EROFS;
10112 
10113 	INIT_LIST_HEAD(&splice);
10114 
10115 	mutex_lock(&fs_info->delalloc_root_mutex);
10116 	spin_lock(&fs_info->delalloc_root_lock);
10117 	list_splice_init(&fs_info->delalloc_roots, &splice);
10118 	while (!list_empty(&splice) && nr) {
10119 		root = list_first_entry(&splice, struct btrfs_root,
10120 					delalloc_root);
10121 		root = btrfs_grab_fs_root(root);
10122 		BUG_ON(!root);
10123 		list_move_tail(&root->delalloc_root,
10124 			       &fs_info->delalloc_roots);
10125 		spin_unlock(&fs_info->delalloc_root_lock);
10126 
10127 		ret = __start_delalloc_inodes(root, delay_iput, nr);
10128 		btrfs_put_fs_root(root);
10129 		if (ret < 0)
10130 			goto out;
10131 
10132 		if (nr != -1) {
10133 			nr -= ret;
10134 			WARN_ON(nr < 0);
10135 		}
10136 		spin_lock(&fs_info->delalloc_root_lock);
10137 	}
10138 	spin_unlock(&fs_info->delalloc_root_lock);
10139 
10140 	ret = 0;
10141 out:
10142 	if (!list_empty_careful(&splice)) {
10143 		spin_lock(&fs_info->delalloc_root_lock);
10144 		list_splice_tail(&splice, &fs_info->delalloc_roots);
10145 		spin_unlock(&fs_info->delalloc_root_lock);
10146 	}
10147 	mutex_unlock(&fs_info->delalloc_root_mutex);
10148 	return ret;
10149 }
10150 
10151 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10152 			 const char *symname)
10153 {
10154 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10155 	struct btrfs_trans_handle *trans;
10156 	struct btrfs_root *root = BTRFS_I(dir)->root;
10157 	struct btrfs_path *path;
10158 	struct btrfs_key key;
10159 	struct inode *inode = NULL;
10160 	int err;
10161 	int drop_inode = 0;
10162 	u64 objectid;
10163 	u64 index = 0;
10164 	int name_len;
10165 	int datasize;
10166 	unsigned long ptr;
10167 	struct btrfs_file_extent_item *ei;
10168 	struct extent_buffer *leaf;
10169 
10170 	name_len = strlen(symname);
10171 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10172 		return -ENAMETOOLONG;
10173 
10174 	/*
10175 	 * 2 items for inode item and ref
10176 	 * 2 items for dir items
10177 	 * 1 item for updating parent inode item
10178 	 * 1 item for the inline extent item
10179 	 * 1 item for xattr if selinux is on
10180 	 */
10181 	trans = btrfs_start_transaction(root, 7);
10182 	if (IS_ERR(trans))
10183 		return PTR_ERR(trans);
10184 
10185 	err = btrfs_find_free_ino(root, &objectid);
10186 	if (err)
10187 		goto out_unlock;
10188 
10189 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10190 				dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10191 				objectid, S_IFLNK|S_IRWXUGO, &index);
10192 	if (IS_ERR(inode)) {
10193 		err = PTR_ERR(inode);
10194 		goto out_unlock;
10195 	}
10196 
10197 	/*
10198 	* If the active LSM wants to access the inode during
10199 	* d_instantiate it needs these. Smack checks to see
10200 	* if the filesystem supports xattrs by looking at the
10201 	* ops vector.
10202 	*/
10203 	inode->i_fop = &btrfs_file_operations;
10204 	inode->i_op = &btrfs_file_inode_operations;
10205 	inode->i_mapping->a_ops = &btrfs_aops;
10206 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10207 
10208 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10209 	if (err)
10210 		goto out_unlock_inode;
10211 
10212 	path = btrfs_alloc_path();
10213 	if (!path) {
10214 		err = -ENOMEM;
10215 		goto out_unlock_inode;
10216 	}
10217 	key.objectid = btrfs_ino(BTRFS_I(inode));
10218 	key.offset = 0;
10219 	key.type = BTRFS_EXTENT_DATA_KEY;
10220 	datasize = btrfs_file_extent_calc_inline_size(name_len);
10221 	err = btrfs_insert_empty_item(trans, root, path, &key,
10222 				      datasize);
10223 	if (err) {
10224 		btrfs_free_path(path);
10225 		goto out_unlock_inode;
10226 	}
10227 	leaf = path->nodes[0];
10228 	ei = btrfs_item_ptr(leaf, path->slots[0],
10229 			    struct btrfs_file_extent_item);
10230 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10231 	btrfs_set_file_extent_type(leaf, ei,
10232 				   BTRFS_FILE_EXTENT_INLINE);
10233 	btrfs_set_file_extent_encryption(leaf, ei, 0);
10234 	btrfs_set_file_extent_compression(leaf, ei, 0);
10235 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10236 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10237 
10238 	ptr = btrfs_file_extent_inline_start(ei);
10239 	write_extent_buffer(leaf, symname, ptr, name_len);
10240 	btrfs_mark_buffer_dirty(leaf);
10241 	btrfs_free_path(path);
10242 
10243 	inode->i_op = &btrfs_symlink_inode_operations;
10244 	inode_nohighmem(inode);
10245 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
10246 	inode_set_bytes(inode, name_len);
10247 	btrfs_i_size_write(BTRFS_I(inode), name_len);
10248 	err = btrfs_update_inode(trans, root, inode);
10249 	/*
10250 	 * Last step, add directory indexes for our symlink inode. This is the
10251 	 * last step to avoid extra cleanup of these indexes if an error happens
10252 	 * elsewhere above.
10253 	 */
10254 	if (!err)
10255 		err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10256 				BTRFS_I(inode), 0, index);
10257 	if (err) {
10258 		drop_inode = 1;
10259 		goto out_unlock_inode;
10260 	}
10261 
10262 	unlock_new_inode(inode);
10263 	d_instantiate(dentry, inode);
10264 
10265 out_unlock:
10266 	btrfs_end_transaction(trans);
10267 	if (drop_inode) {
10268 		inode_dec_link_count(inode);
10269 		iput(inode);
10270 	}
10271 	btrfs_btree_balance_dirty(fs_info);
10272 	return err;
10273 
10274 out_unlock_inode:
10275 	drop_inode = 1;
10276 	unlock_new_inode(inode);
10277 	goto out_unlock;
10278 }
10279 
10280 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10281 				       u64 start, u64 num_bytes, u64 min_size,
10282 				       loff_t actual_len, u64 *alloc_hint,
10283 				       struct btrfs_trans_handle *trans)
10284 {
10285 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10286 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10287 	struct extent_map *em;
10288 	struct btrfs_root *root = BTRFS_I(inode)->root;
10289 	struct btrfs_key ins;
10290 	u64 cur_offset = start;
10291 	u64 i_size;
10292 	u64 cur_bytes;
10293 	u64 last_alloc = (u64)-1;
10294 	int ret = 0;
10295 	bool own_trans = true;
10296 	u64 end = start + num_bytes - 1;
10297 
10298 	if (trans)
10299 		own_trans = false;
10300 	while (num_bytes > 0) {
10301 		if (own_trans) {
10302 			trans = btrfs_start_transaction(root, 3);
10303 			if (IS_ERR(trans)) {
10304 				ret = PTR_ERR(trans);
10305 				break;
10306 			}
10307 		}
10308 
10309 		cur_bytes = min_t(u64, num_bytes, SZ_256M);
10310 		cur_bytes = max(cur_bytes, min_size);
10311 		/*
10312 		 * If we are severely fragmented we could end up with really
10313 		 * small allocations, so if the allocator is returning small
10314 		 * chunks lets make its job easier by only searching for those
10315 		 * sized chunks.
10316 		 */
10317 		cur_bytes = min(cur_bytes, last_alloc);
10318 		ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10319 				min_size, 0, *alloc_hint, &ins, 1, 0);
10320 		if (ret) {
10321 			if (own_trans)
10322 				btrfs_end_transaction(trans);
10323 			break;
10324 		}
10325 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10326 
10327 		last_alloc = ins.offset;
10328 		ret = insert_reserved_file_extent(trans, inode,
10329 						  cur_offset, ins.objectid,
10330 						  ins.offset, ins.offset,
10331 						  ins.offset, 0, 0, 0,
10332 						  BTRFS_FILE_EXTENT_PREALLOC);
10333 		if (ret) {
10334 			btrfs_free_reserved_extent(fs_info, ins.objectid,
10335 						   ins.offset, 0);
10336 			btrfs_abort_transaction(trans, ret);
10337 			if (own_trans)
10338 				btrfs_end_transaction(trans);
10339 			break;
10340 		}
10341 
10342 		btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10343 					cur_offset + ins.offset -1, 0);
10344 
10345 		em = alloc_extent_map();
10346 		if (!em) {
10347 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10348 				&BTRFS_I(inode)->runtime_flags);
10349 			goto next;
10350 		}
10351 
10352 		em->start = cur_offset;
10353 		em->orig_start = cur_offset;
10354 		em->len = ins.offset;
10355 		em->block_start = ins.objectid;
10356 		em->block_len = ins.offset;
10357 		em->orig_block_len = ins.offset;
10358 		em->ram_bytes = ins.offset;
10359 		em->bdev = fs_info->fs_devices->latest_bdev;
10360 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10361 		em->generation = trans->transid;
10362 
10363 		while (1) {
10364 			write_lock(&em_tree->lock);
10365 			ret = add_extent_mapping(em_tree, em, 1);
10366 			write_unlock(&em_tree->lock);
10367 			if (ret != -EEXIST)
10368 				break;
10369 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10370 						cur_offset + ins.offset - 1,
10371 						0);
10372 		}
10373 		free_extent_map(em);
10374 next:
10375 		num_bytes -= ins.offset;
10376 		cur_offset += ins.offset;
10377 		*alloc_hint = ins.objectid + ins.offset;
10378 
10379 		inode_inc_iversion(inode);
10380 		inode->i_ctime = current_time(inode);
10381 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10382 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10383 		    (actual_len > inode->i_size) &&
10384 		    (cur_offset > inode->i_size)) {
10385 			if (cur_offset > actual_len)
10386 				i_size = actual_len;
10387 			else
10388 				i_size = cur_offset;
10389 			i_size_write(inode, i_size);
10390 			btrfs_ordered_update_i_size(inode, i_size, NULL);
10391 		}
10392 
10393 		ret = btrfs_update_inode(trans, root, inode);
10394 
10395 		if (ret) {
10396 			btrfs_abort_transaction(trans, ret);
10397 			if (own_trans)
10398 				btrfs_end_transaction(trans);
10399 			break;
10400 		}
10401 
10402 		if (own_trans)
10403 			btrfs_end_transaction(trans);
10404 	}
10405 	if (cur_offset < end)
10406 		btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10407 			end - cur_offset + 1);
10408 	return ret;
10409 }
10410 
10411 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10412 			      u64 start, u64 num_bytes, u64 min_size,
10413 			      loff_t actual_len, u64 *alloc_hint)
10414 {
10415 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10416 					   min_size, actual_len, alloc_hint,
10417 					   NULL);
10418 }
10419 
10420 int btrfs_prealloc_file_range_trans(struct inode *inode,
10421 				    struct btrfs_trans_handle *trans, int mode,
10422 				    u64 start, u64 num_bytes, u64 min_size,
10423 				    loff_t actual_len, u64 *alloc_hint)
10424 {
10425 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10426 					   min_size, actual_len, alloc_hint, trans);
10427 }
10428 
10429 static int btrfs_set_page_dirty(struct page *page)
10430 {
10431 	return __set_page_dirty_nobuffers(page);
10432 }
10433 
10434 static int btrfs_permission(struct inode *inode, int mask)
10435 {
10436 	struct btrfs_root *root = BTRFS_I(inode)->root;
10437 	umode_t mode = inode->i_mode;
10438 
10439 	if (mask & MAY_WRITE &&
10440 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10441 		if (btrfs_root_readonly(root))
10442 			return -EROFS;
10443 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10444 			return -EACCES;
10445 	}
10446 	return generic_permission(inode, mask);
10447 }
10448 
10449 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10450 {
10451 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10452 	struct btrfs_trans_handle *trans;
10453 	struct btrfs_root *root = BTRFS_I(dir)->root;
10454 	struct inode *inode = NULL;
10455 	u64 objectid;
10456 	u64 index;
10457 	int ret = 0;
10458 
10459 	/*
10460 	 * 5 units required for adding orphan entry
10461 	 */
10462 	trans = btrfs_start_transaction(root, 5);
10463 	if (IS_ERR(trans))
10464 		return PTR_ERR(trans);
10465 
10466 	ret = btrfs_find_free_ino(root, &objectid);
10467 	if (ret)
10468 		goto out;
10469 
10470 	inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10471 			btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10472 	if (IS_ERR(inode)) {
10473 		ret = PTR_ERR(inode);
10474 		inode = NULL;
10475 		goto out;
10476 	}
10477 
10478 	inode->i_fop = &btrfs_file_operations;
10479 	inode->i_op = &btrfs_file_inode_operations;
10480 
10481 	inode->i_mapping->a_ops = &btrfs_aops;
10482 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10483 
10484 	ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10485 	if (ret)
10486 		goto out_inode;
10487 
10488 	ret = btrfs_update_inode(trans, root, inode);
10489 	if (ret)
10490 		goto out_inode;
10491 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10492 	if (ret)
10493 		goto out_inode;
10494 
10495 	/*
10496 	 * We set number of links to 0 in btrfs_new_inode(), and here we set
10497 	 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10498 	 * through:
10499 	 *
10500 	 *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10501 	 */
10502 	set_nlink(inode, 1);
10503 	unlock_new_inode(inode);
10504 	d_tmpfile(dentry, inode);
10505 	mark_inode_dirty(inode);
10506 
10507 out:
10508 	btrfs_end_transaction(trans);
10509 	if (ret)
10510 		iput(inode);
10511 	btrfs_btree_balance_dirty(fs_info);
10512 	return ret;
10513 
10514 out_inode:
10515 	unlock_new_inode(inode);
10516 	goto out;
10517 
10518 }
10519 
10520 __attribute__((const))
10521 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10522 {
10523 	return -EAGAIN;
10524 }
10525 
10526 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10527 {
10528 	struct inode *inode = private_data;
10529 	return btrfs_sb(inode->i_sb);
10530 }
10531 
10532 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10533 					u64 start, u64 end)
10534 {
10535 	struct inode *inode = private_data;
10536 	u64 isize;
10537 
10538 	isize = i_size_read(inode);
10539 	if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10540 		btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10541 		    "%s: ino %llu isize %llu odd range [%llu,%llu]",
10542 			caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10543 	}
10544 }
10545 
10546 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10547 {
10548 	struct inode *inode = private_data;
10549 	unsigned long index = start >> PAGE_SHIFT;
10550 	unsigned long end_index = end >> PAGE_SHIFT;
10551 	struct page *page;
10552 
10553 	while (index <= end_index) {
10554 		page = find_get_page(inode->i_mapping, index);
10555 		ASSERT(page); /* Pages should be in the extent_io_tree */
10556 		set_page_writeback(page);
10557 		put_page(page);
10558 		index++;
10559 	}
10560 }
10561 
10562 static const struct inode_operations btrfs_dir_inode_operations = {
10563 	.getattr	= btrfs_getattr,
10564 	.lookup		= btrfs_lookup,
10565 	.create		= btrfs_create,
10566 	.unlink		= btrfs_unlink,
10567 	.link		= btrfs_link,
10568 	.mkdir		= btrfs_mkdir,
10569 	.rmdir		= btrfs_rmdir,
10570 	.rename		= btrfs_rename2,
10571 	.symlink	= btrfs_symlink,
10572 	.setattr	= btrfs_setattr,
10573 	.mknod		= btrfs_mknod,
10574 	.listxattr	= btrfs_listxattr,
10575 	.permission	= btrfs_permission,
10576 	.get_acl	= btrfs_get_acl,
10577 	.set_acl	= btrfs_set_acl,
10578 	.update_time	= btrfs_update_time,
10579 	.tmpfile        = btrfs_tmpfile,
10580 };
10581 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10582 	.lookup		= btrfs_lookup,
10583 	.permission	= btrfs_permission,
10584 	.update_time	= btrfs_update_time,
10585 };
10586 
10587 static const struct file_operations btrfs_dir_file_operations = {
10588 	.llseek		= generic_file_llseek,
10589 	.read		= generic_read_dir,
10590 	.iterate_shared	= btrfs_real_readdir,
10591 	.open		= btrfs_opendir,
10592 	.unlocked_ioctl	= btrfs_ioctl,
10593 #ifdef CONFIG_COMPAT
10594 	.compat_ioctl	= btrfs_compat_ioctl,
10595 #endif
10596 	.release        = btrfs_release_file,
10597 	.fsync		= btrfs_sync_file,
10598 };
10599 
10600 static const struct extent_io_ops btrfs_extent_io_ops = {
10601 	/* mandatory callbacks */
10602 	.submit_bio_hook = btrfs_submit_bio_hook,
10603 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
10604 	.merge_bio_hook = btrfs_merge_bio_hook,
10605 	.readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10606 	.tree_fs_info = iotree_fs_info,
10607 	.set_range_writeback = btrfs_set_range_writeback,
10608 
10609 	/* optional callbacks */
10610 	.fill_delalloc = run_delalloc_range,
10611 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
10612 	.writepage_start_hook = btrfs_writepage_start_hook,
10613 	.set_bit_hook = btrfs_set_bit_hook,
10614 	.clear_bit_hook = btrfs_clear_bit_hook,
10615 	.merge_extent_hook = btrfs_merge_extent_hook,
10616 	.split_extent_hook = btrfs_split_extent_hook,
10617 	.check_extent_io_range = btrfs_check_extent_io_range,
10618 };
10619 
10620 /*
10621  * btrfs doesn't support the bmap operation because swapfiles
10622  * use bmap to make a mapping of extents in the file.  They assume
10623  * these extents won't change over the life of the file and they
10624  * use the bmap result to do IO directly to the drive.
10625  *
10626  * the btrfs bmap call would return logical addresses that aren't
10627  * suitable for IO and they also will change frequently as COW
10628  * operations happen.  So, swapfile + btrfs == corruption.
10629  *
10630  * For now we're avoiding this by dropping bmap.
10631  */
10632 static const struct address_space_operations btrfs_aops = {
10633 	.readpage	= btrfs_readpage,
10634 	.writepage	= btrfs_writepage,
10635 	.writepages	= btrfs_writepages,
10636 	.readpages	= btrfs_readpages,
10637 	.direct_IO	= btrfs_direct_IO,
10638 	.invalidatepage = btrfs_invalidatepage,
10639 	.releasepage	= btrfs_releasepage,
10640 	.set_page_dirty	= btrfs_set_page_dirty,
10641 	.error_remove_page = generic_error_remove_page,
10642 };
10643 
10644 static const struct address_space_operations btrfs_symlink_aops = {
10645 	.readpage	= btrfs_readpage,
10646 	.writepage	= btrfs_writepage,
10647 	.invalidatepage = btrfs_invalidatepage,
10648 	.releasepage	= btrfs_releasepage,
10649 };
10650 
10651 static const struct inode_operations btrfs_file_inode_operations = {
10652 	.getattr	= btrfs_getattr,
10653 	.setattr	= btrfs_setattr,
10654 	.listxattr      = btrfs_listxattr,
10655 	.permission	= btrfs_permission,
10656 	.fiemap		= btrfs_fiemap,
10657 	.get_acl	= btrfs_get_acl,
10658 	.set_acl	= btrfs_set_acl,
10659 	.update_time	= btrfs_update_time,
10660 };
10661 static const struct inode_operations btrfs_special_inode_operations = {
10662 	.getattr	= btrfs_getattr,
10663 	.setattr	= btrfs_setattr,
10664 	.permission	= btrfs_permission,
10665 	.listxattr	= btrfs_listxattr,
10666 	.get_acl	= btrfs_get_acl,
10667 	.set_acl	= btrfs_set_acl,
10668 	.update_time	= btrfs_update_time,
10669 };
10670 static const struct inode_operations btrfs_symlink_inode_operations = {
10671 	.get_link	= page_get_link,
10672 	.getattr	= btrfs_getattr,
10673 	.setattr	= btrfs_setattr,
10674 	.permission	= btrfs_permission,
10675 	.listxattr	= btrfs_listxattr,
10676 	.update_time	= btrfs_update_time,
10677 };
10678 
10679 const struct dentry_operations btrfs_dentry_operations = {
10680 	.d_delete	= btrfs_dentry_delete,
10681 	.d_release	= btrfs_dentry_release,
10682 };
10683