xref: /openbmc/linux/fs/btrfs/inode.c (revision aac5987a)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
51 #include "xattr.h"
52 #include "tree-log.h"
53 #include "volumes.h"
54 #include "compression.h"
55 #include "locking.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
58 #include "backref.h"
59 #include "hash.h"
60 #include "props.h"
61 #include "qgroup.h"
62 #include "dedupe.h"
63 
64 struct btrfs_iget_args {
65 	struct btrfs_key *location;
66 	struct btrfs_root *root;
67 };
68 
69 struct btrfs_dio_data {
70 	u64 outstanding_extents;
71 	u64 reserve;
72 	u64 unsubmitted_oe_range_start;
73 	u64 unsubmitted_oe_range_end;
74 	int overwrite;
75 };
76 
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
86 
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_transaction_cachep;
90 struct kmem_cache *btrfs_path_cachep;
91 struct kmem_cache *btrfs_free_space_cachep;
92 
93 #define S_SHIFT 12
94 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
95 	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
96 	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
97 	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
98 	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
99 	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
100 	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
101 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
102 };
103 
104 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
105 static int btrfs_truncate(struct inode *inode);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
107 static noinline int cow_file_range(struct inode *inode,
108 				   struct page *locked_page,
109 				   u64 start, u64 end, u64 delalloc_end,
110 				   int *page_started, unsigned long *nr_written,
111 				   int unlock, struct btrfs_dedupe_hash *hash);
112 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
113 				       u64 orig_start, u64 block_start,
114 				       u64 block_len, u64 orig_block_len,
115 				       u64 ram_bytes, int compress_type,
116 				       int type);
117 
118 static int btrfs_dirty_inode(struct inode *inode);
119 
120 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
121 void btrfs_test_inode_set_ops(struct inode *inode)
122 {
123 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
124 }
125 #endif
126 
127 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
128 				     struct inode *inode,  struct inode *dir,
129 				     const struct qstr *qstr)
130 {
131 	int err;
132 
133 	err = btrfs_init_acl(trans, inode, dir);
134 	if (!err)
135 		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
136 	return err;
137 }
138 
139 /*
140  * this does all the hard work for inserting an inline extent into
141  * the btree.  The caller should have done a btrfs_drop_extents so that
142  * no overlapping inline items exist in the btree
143  */
144 static int insert_inline_extent(struct btrfs_trans_handle *trans,
145 				struct btrfs_path *path, int extent_inserted,
146 				struct btrfs_root *root, struct inode *inode,
147 				u64 start, size_t size, size_t compressed_size,
148 				int compress_type,
149 				struct page **compressed_pages)
150 {
151 	struct extent_buffer *leaf;
152 	struct page *page = NULL;
153 	char *kaddr;
154 	unsigned long ptr;
155 	struct btrfs_file_extent_item *ei;
156 	int err = 0;
157 	int ret;
158 	size_t cur_size = size;
159 	unsigned long offset;
160 
161 	if (compressed_size && compressed_pages)
162 		cur_size = compressed_size;
163 
164 	inode_add_bytes(inode, size);
165 
166 	if (!extent_inserted) {
167 		struct btrfs_key key;
168 		size_t datasize;
169 
170 		key.objectid = btrfs_ino(BTRFS_I(inode));
171 		key.offset = start;
172 		key.type = BTRFS_EXTENT_DATA_KEY;
173 
174 		datasize = btrfs_file_extent_calc_inline_size(cur_size);
175 		path->leave_spinning = 1;
176 		ret = btrfs_insert_empty_item(trans, root, path, &key,
177 					      datasize);
178 		if (ret) {
179 			err = ret;
180 			goto fail;
181 		}
182 	}
183 	leaf = path->nodes[0];
184 	ei = btrfs_item_ptr(leaf, path->slots[0],
185 			    struct btrfs_file_extent_item);
186 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
187 	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
188 	btrfs_set_file_extent_encryption(leaf, ei, 0);
189 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
190 	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
191 	ptr = btrfs_file_extent_inline_start(ei);
192 
193 	if (compress_type != BTRFS_COMPRESS_NONE) {
194 		struct page *cpage;
195 		int i = 0;
196 		while (compressed_size > 0) {
197 			cpage = compressed_pages[i];
198 			cur_size = min_t(unsigned long, compressed_size,
199 				       PAGE_SIZE);
200 
201 			kaddr = kmap_atomic(cpage);
202 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
203 			kunmap_atomic(kaddr);
204 
205 			i++;
206 			ptr += cur_size;
207 			compressed_size -= cur_size;
208 		}
209 		btrfs_set_file_extent_compression(leaf, ei,
210 						  compress_type);
211 	} else {
212 		page = find_get_page(inode->i_mapping,
213 				     start >> PAGE_SHIFT);
214 		btrfs_set_file_extent_compression(leaf, ei, 0);
215 		kaddr = kmap_atomic(page);
216 		offset = start & (PAGE_SIZE - 1);
217 		write_extent_buffer(leaf, kaddr + offset, ptr, size);
218 		kunmap_atomic(kaddr);
219 		put_page(page);
220 	}
221 	btrfs_mark_buffer_dirty(leaf);
222 	btrfs_release_path(path);
223 
224 	/*
225 	 * we're an inline extent, so nobody can
226 	 * extend the file past i_size without locking
227 	 * a page we already have locked.
228 	 *
229 	 * We must do any isize and inode updates
230 	 * before we unlock the pages.  Otherwise we
231 	 * could end up racing with unlink.
232 	 */
233 	BTRFS_I(inode)->disk_i_size = inode->i_size;
234 	ret = btrfs_update_inode(trans, root, inode);
235 
236 	return ret;
237 fail:
238 	return err;
239 }
240 
241 
242 /*
243  * conditionally insert an inline extent into the file.  This
244  * does the checks required to make sure the data is small enough
245  * to fit as an inline extent.
246  */
247 static noinline int cow_file_range_inline(struct btrfs_root *root,
248 					  struct inode *inode, u64 start,
249 					  u64 end, size_t compressed_size,
250 					  int compress_type,
251 					  struct page **compressed_pages)
252 {
253 	struct btrfs_fs_info *fs_info = root->fs_info;
254 	struct btrfs_trans_handle *trans;
255 	u64 isize = i_size_read(inode);
256 	u64 actual_end = min(end + 1, isize);
257 	u64 inline_len = actual_end - start;
258 	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
259 	u64 data_len = inline_len;
260 	int ret;
261 	struct btrfs_path *path;
262 	int extent_inserted = 0;
263 	u32 extent_item_size;
264 
265 	if (compressed_size)
266 		data_len = compressed_size;
267 
268 	if (start > 0 ||
269 	    actual_end > fs_info->sectorsize ||
270 	    data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
271 	    (!compressed_size &&
272 	    (actual_end & (fs_info->sectorsize - 1)) == 0) ||
273 	    end + 1 < isize ||
274 	    data_len > fs_info->max_inline) {
275 		return 1;
276 	}
277 
278 	path = btrfs_alloc_path();
279 	if (!path)
280 		return -ENOMEM;
281 
282 	trans = btrfs_join_transaction(root);
283 	if (IS_ERR(trans)) {
284 		btrfs_free_path(path);
285 		return PTR_ERR(trans);
286 	}
287 	trans->block_rsv = &fs_info->delalloc_block_rsv;
288 
289 	if (compressed_size && compressed_pages)
290 		extent_item_size = btrfs_file_extent_calc_inline_size(
291 		   compressed_size);
292 	else
293 		extent_item_size = btrfs_file_extent_calc_inline_size(
294 		    inline_len);
295 
296 	ret = __btrfs_drop_extents(trans, root, inode, path,
297 				   start, aligned_end, NULL,
298 				   1, 1, extent_item_size, &extent_inserted);
299 	if (ret) {
300 		btrfs_abort_transaction(trans, ret);
301 		goto out;
302 	}
303 
304 	if (isize > actual_end)
305 		inline_len = min_t(u64, isize, actual_end);
306 	ret = insert_inline_extent(trans, path, extent_inserted,
307 				   root, inode, start,
308 				   inline_len, compressed_size,
309 				   compress_type, compressed_pages);
310 	if (ret && ret != -ENOSPC) {
311 		btrfs_abort_transaction(trans, ret);
312 		goto out;
313 	} else if (ret == -ENOSPC) {
314 		ret = 1;
315 		goto out;
316 	}
317 
318 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
319 	btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
320 	btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
321 out:
322 	/*
323 	 * Don't forget to free the reserved space, as for inlined extent
324 	 * it won't count as data extent, free them directly here.
325 	 * And at reserve time, it's always aligned to page size, so
326 	 * just free one page here.
327 	 */
328 	btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
329 	btrfs_free_path(path);
330 	btrfs_end_transaction(trans);
331 	return ret;
332 }
333 
334 struct async_extent {
335 	u64 start;
336 	u64 ram_size;
337 	u64 compressed_size;
338 	struct page **pages;
339 	unsigned long nr_pages;
340 	int compress_type;
341 	struct list_head list;
342 };
343 
344 struct async_cow {
345 	struct inode *inode;
346 	struct btrfs_root *root;
347 	struct page *locked_page;
348 	u64 start;
349 	u64 end;
350 	struct list_head extents;
351 	struct btrfs_work work;
352 };
353 
354 static noinline int add_async_extent(struct async_cow *cow,
355 				     u64 start, u64 ram_size,
356 				     u64 compressed_size,
357 				     struct page **pages,
358 				     unsigned long nr_pages,
359 				     int compress_type)
360 {
361 	struct async_extent *async_extent;
362 
363 	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
364 	BUG_ON(!async_extent); /* -ENOMEM */
365 	async_extent->start = start;
366 	async_extent->ram_size = ram_size;
367 	async_extent->compressed_size = compressed_size;
368 	async_extent->pages = pages;
369 	async_extent->nr_pages = nr_pages;
370 	async_extent->compress_type = compress_type;
371 	list_add_tail(&async_extent->list, &cow->extents);
372 	return 0;
373 }
374 
375 static inline int inode_need_compress(struct inode *inode)
376 {
377 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
378 
379 	/* force compress */
380 	if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
381 		return 1;
382 	/* bad compression ratios */
383 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
384 		return 0;
385 	if (btrfs_test_opt(fs_info, COMPRESS) ||
386 	    BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
387 	    BTRFS_I(inode)->force_compress)
388 		return 1;
389 	return 0;
390 }
391 
392 static inline void inode_should_defrag(struct btrfs_inode *inode,
393 		u64 start, u64 end, u64 num_bytes, u64 small_write)
394 {
395 	/* If this is a small write inside eof, kick off a defrag */
396 	if (num_bytes < small_write &&
397 	    (start > 0 || end + 1 < inode->disk_i_size))
398 		btrfs_add_inode_defrag(NULL, inode);
399 }
400 
401 /*
402  * we create compressed extents in two phases.  The first
403  * phase compresses a range of pages that have already been
404  * locked (both pages and state bits are locked).
405  *
406  * This is done inside an ordered work queue, and the compression
407  * is spread across many cpus.  The actual IO submission is step
408  * two, and the ordered work queue takes care of making sure that
409  * happens in the same order things were put onto the queue by
410  * writepages and friends.
411  *
412  * If this code finds it can't get good compression, it puts an
413  * entry onto the work queue to write the uncompressed bytes.  This
414  * makes sure that both compressed inodes and uncompressed inodes
415  * are written in the same order that the flusher thread sent them
416  * down.
417  */
418 static noinline void compress_file_range(struct inode *inode,
419 					struct page *locked_page,
420 					u64 start, u64 end,
421 					struct async_cow *async_cow,
422 					int *num_added)
423 {
424 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
425 	struct btrfs_root *root = BTRFS_I(inode)->root;
426 	u64 num_bytes;
427 	u64 blocksize = fs_info->sectorsize;
428 	u64 actual_end;
429 	u64 isize = i_size_read(inode);
430 	int ret = 0;
431 	struct page **pages = NULL;
432 	unsigned long nr_pages;
433 	unsigned long total_compressed = 0;
434 	unsigned long total_in = 0;
435 	int i;
436 	int will_compress;
437 	int compress_type = fs_info->compress_type;
438 	int redirty = 0;
439 
440 	inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
441 			SZ_16K);
442 
443 	actual_end = min_t(u64, isize, end + 1);
444 again:
445 	will_compress = 0;
446 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
447 	BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
448 	nr_pages = min_t(unsigned long, nr_pages,
449 			BTRFS_MAX_COMPRESSED / PAGE_SIZE);
450 
451 	/*
452 	 * we don't want to send crud past the end of i_size through
453 	 * compression, that's just a waste of CPU time.  So, if the
454 	 * end of the file is before the start of our current
455 	 * requested range of bytes, we bail out to the uncompressed
456 	 * cleanup code that can deal with all of this.
457 	 *
458 	 * It isn't really the fastest way to fix things, but this is a
459 	 * very uncommon corner.
460 	 */
461 	if (actual_end <= start)
462 		goto cleanup_and_bail_uncompressed;
463 
464 	total_compressed = actual_end - start;
465 
466 	/*
467 	 * skip compression for a small file range(<=blocksize) that
468 	 * isn't an inline extent, since it doesn't save disk space at all.
469 	 */
470 	if (total_compressed <= blocksize &&
471 	   (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
472 		goto cleanup_and_bail_uncompressed;
473 
474 	total_compressed = min_t(unsigned long, total_compressed,
475 			BTRFS_MAX_UNCOMPRESSED);
476 	num_bytes = ALIGN(end - start + 1, blocksize);
477 	num_bytes = max(blocksize,  num_bytes);
478 	total_in = 0;
479 	ret = 0;
480 
481 	/*
482 	 * we do compression for mount -o compress and when the
483 	 * inode has not been flagged as nocompress.  This flag can
484 	 * change at any time if we discover bad compression ratios.
485 	 */
486 	if (inode_need_compress(inode)) {
487 		WARN_ON(pages);
488 		pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 		if (!pages) {
490 			/* just bail out to the uncompressed code */
491 			goto cont;
492 		}
493 
494 		if (BTRFS_I(inode)->force_compress)
495 			compress_type = BTRFS_I(inode)->force_compress;
496 
497 		/*
498 		 * we need to call clear_page_dirty_for_io on each
499 		 * page in the range.  Otherwise applications with the file
500 		 * mmap'd can wander in and change the page contents while
501 		 * we are compressing them.
502 		 *
503 		 * If the compression fails for any reason, we set the pages
504 		 * dirty again later on.
505 		 */
506 		extent_range_clear_dirty_for_io(inode, start, end);
507 		redirty = 1;
508 		ret = btrfs_compress_pages(compress_type,
509 					   inode->i_mapping, start,
510 					   pages,
511 					   &nr_pages,
512 					   &total_in,
513 					   &total_compressed);
514 
515 		if (!ret) {
516 			unsigned long offset = total_compressed &
517 				(PAGE_SIZE - 1);
518 			struct page *page = pages[nr_pages - 1];
519 			char *kaddr;
520 
521 			/* zero the tail end of the last page, we might be
522 			 * sending it down to disk
523 			 */
524 			if (offset) {
525 				kaddr = kmap_atomic(page);
526 				memset(kaddr + offset, 0,
527 				       PAGE_SIZE - offset);
528 				kunmap_atomic(kaddr);
529 			}
530 			will_compress = 1;
531 		}
532 	}
533 cont:
534 	if (start == 0) {
535 		/* lets try to make an inline extent */
536 		if (ret || total_in < (actual_end - start)) {
537 			/* we didn't compress the entire range, try
538 			 * to make an uncompressed inline extent.
539 			 */
540 			ret = cow_file_range_inline(root, inode, start, end,
541 					    0, BTRFS_COMPRESS_NONE, NULL);
542 		} else {
543 			/* try making a compressed inline extent */
544 			ret = cow_file_range_inline(root, inode, start, end,
545 						    total_compressed,
546 						    compress_type, pages);
547 		}
548 		if (ret <= 0) {
549 			unsigned long clear_flags = EXTENT_DELALLOC |
550 				EXTENT_DEFRAG;
551 			unsigned long page_error_op;
552 
553 			clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 			page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
555 
556 			/*
557 			 * inline extent creation worked or returned error,
558 			 * we don't need to create any more async work items.
559 			 * Unlock and free up our temp pages.
560 			 */
561 			extent_clear_unlock_delalloc(inode, start, end, end,
562 						     NULL, clear_flags,
563 						     PAGE_UNLOCK |
564 						     PAGE_CLEAR_DIRTY |
565 						     PAGE_SET_WRITEBACK |
566 						     page_error_op |
567 						     PAGE_END_WRITEBACK);
568 			btrfs_free_reserved_data_space_noquota(inode, start,
569 						end - start + 1);
570 			goto free_pages_out;
571 		}
572 	}
573 
574 	if (will_compress) {
575 		/*
576 		 * we aren't doing an inline extent round the compressed size
577 		 * up to a block size boundary so the allocator does sane
578 		 * things
579 		 */
580 		total_compressed = ALIGN(total_compressed, blocksize);
581 
582 		/*
583 		 * one last check to make sure the compression is really a
584 		 * win, compare the page count read with the blocks on disk
585 		 */
586 		total_in = ALIGN(total_in, PAGE_SIZE);
587 		if (total_compressed >= total_in) {
588 			will_compress = 0;
589 		} else {
590 			num_bytes = total_in;
591 			*num_added += 1;
592 
593 			/*
594 			 * The async work queues will take care of doing actual
595 			 * allocation on disk for these compressed pages, and
596 			 * will submit them to the elevator.
597 			 */
598 			add_async_extent(async_cow, start, num_bytes,
599 					total_compressed, pages, nr_pages,
600 					compress_type);
601 
602 			if (start + num_bytes < end) {
603 				start += num_bytes;
604 				pages = NULL;
605 				cond_resched();
606 				goto again;
607 			}
608 			return;
609 		}
610 	}
611 	if (pages) {
612 		/*
613 		 * the compression code ran but failed to make things smaller,
614 		 * free any pages it allocated and our page pointer array
615 		 */
616 		for (i = 0; i < nr_pages; i++) {
617 			WARN_ON(pages[i]->mapping);
618 			put_page(pages[i]);
619 		}
620 		kfree(pages);
621 		pages = NULL;
622 		total_compressed = 0;
623 		nr_pages = 0;
624 
625 		/* flag the file so we don't compress in the future */
626 		if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
627 		    !(BTRFS_I(inode)->force_compress)) {
628 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
629 		}
630 	}
631 cleanup_and_bail_uncompressed:
632 	/*
633 	 * No compression, but we still need to write the pages in the file
634 	 * we've been given so far.  redirty the locked page if it corresponds
635 	 * to our extent and set things up for the async work queue to run
636 	 * cow_file_range to do the normal delalloc dance.
637 	 */
638 	if (page_offset(locked_page) >= start &&
639 	    page_offset(locked_page) <= end)
640 		__set_page_dirty_nobuffers(locked_page);
641 		/* unlocked later on in the async handlers */
642 
643 	if (redirty)
644 		extent_range_redirty_for_io(inode, start, end);
645 	add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
646 			 BTRFS_COMPRESS_NONE);
647 	*num_added += 1;
648 
649 	return;
650 
651 free_pages_out:
652 	for (i = 0; i < nr_pages; i++) {
653 		WARN_ON(pages[i]->mapping);
654 		put_page(pages[i]);
655 	}
656 	kfree(pages);
657 }
658 
659 static void free_async_extent_pages(struct async_extent *async_extent)
660 {
661 	int i;
662 
663 	if (!async_extent->pages)
664 		return;
665 
666 	for (i = 0; i < async_extent->nr_pages; i++) {
667 		WARN_ON(async_extent->pages[i]->mapping);
668 		put_page(async_extent->pages[i]);
669 	}
670 	kfree(async_extent->pages);
671 	async_extent->nr_pages = 0;
672 	async_extent->pages = NULL;
673 }
674 
675 /*
676  * phase two of compressed writeback.  This is the ordered portion
677  * of the code, which only gets called in the order the work was
678  * queued.  We walk all the async extents created by compress_file_range
679  * and send them down to the disk.
680  */
681 static noinline void submit_compressed_extents(struct inode *inode,
682 					      struct async_cow *async_cow)
683 {
684 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
685 	struct async_extent *async_extent;
686 	u64 alloc_hint = 0;
687 	struct btrfs_key ins;
688 	struct extent_map *em;
689 	struct btrfs_root *root = BTRFS_I(inode)->root;
690 	struct extent_io_tree *io_tree;
691 	int ret = 0;
692 
693 again:
694 	while (!list_empty(&async_cow->extents)) {
695 		async_extent = list_entry(async_cow->extents.next,
696 					  struct async_extent, list);
697 		list_del(&async_extent->list);
698 
699 		io_tree = &BTRFS_I(inode)->io_tree;
700 
701 retry:
702 		/* did the compression code fall back to uncompressed IO? */
703 		if (!async_extent->pages) {
704 			int page_started = 0;
705 			unsigned long nr_written = 0;
706 
707 			lock_extent(io_tree, async_extent->start,
708 					 async_extent->start +
709 					 async_extent->ram_size - 1);
710 
711 			/* allocate blocks */
712 			ret = cow_file_range(inode, async_cow->locked_page,
713 					     async_extent->start,
714 					     async_extent->start +
715 					     async_extent->ram_size - 1,
716 					     async_extent->start +
717 					     async_extent->ram_size - 1,
718 					     &page_started, &nr_written, 0,
719 					     NULL);
720 
721 			/* JDM XXX */
722 
723 			/*
724 			 * if page_started, cow_file_range inserted an
725 			 * inline extent and took care of all the unlocking
726 			 * and IO for us.  Otherwise, we need to submit
727 			 * all those pages down to the drive.
728 			 */
729 			if (!page_started && !ret)
730 				extent_write_locked_range(io_tree,
731 						  inode, async_extent->start,
732 						  async_extent->start +
733 						  async_extent->ram_size - 1,
734 						  btrfs_get_extent,
735 						  WB_SYNC_ALL);
736 			else if (ret)
737 				unlock_page(async_cow->locked_page);
738 			kfree(async_extent);
739 			cond_resched();
740 			continue;
741 		}
742 
743 		lock_extent(io_tree, async_extent->start,
744 			    async_extent->start + async_extent->ram_size - 1);
745 
746 		ret = btrfs_reserve_extent(root, async_extent->ram_size,
747 					   async_extent->compressed_size,
748 					   async_extent->compressed_size,
749 					   0, alloc_hint, &ins, 1, 1);
750 		if (ret) {
751 			free_async_extent_pages(async_extent);
752 
753 			if (ret == -ENOSPC) {
754 				unlock_extent(io_tree, async_extent->start,
755 					      async_extent->start +
756 					      async_extent->ram_size - 1);
757 
758 				/*
759 				 * we need to redirty the pages if we decide to
760 				 * fallback to uncompressed IO, otherwise we
761 				 * will not submit these pages down to lower
762 				 * layers.
763 				 */
764 				extent_range_redirty_for_io(inode,
765 						async_extent->start,
766 						async_extent->start +
767 						async_extent->ram_size - 1);
768 
769 				goto retry;
770 			}
771 			goto out_free;
772 		}
773 		/*
774 		 * here we're doing allocation and writeback of the
775 		 * compressed pages
776 		 */
777 		em = create_io_em(inode, async_extent->start,
778 				  async_extent->ram_size, /* len */
779 				  async_extent->start, /* orig_start */
780 				  ins.objectid, /* block_start */
781 				  ins.offset, /* block_len */
782 				  ins.offset, /* orig_block_len */
783 				  async_extent->ram_size, /* ram_bytes */
784 				  async_extent->compress_type,
785 				  BTRFS_ORDERED_COMPRESSED);
786 		if (IS_ERR(em))
787 			/* ret value is not necessary due to void function */
788 			goto out_free_reserve;
789 		free_extent_map(em);
790 
791 		ret = btrfs_add_ordered_extent_compress(inode,
792 						async_extent->start,
793 						ins.objectid,
794 						async_extent->ram_size,
795 						ins.offset,
796 						BTRFS_ORDERED_COMPRESSED,
797 						async_extent->compress_type);
798 		if (ret) {
799 			btrfs_drop_extent_cache(BTRFS_I(inode),
800 						async_extent->start,
801 						async_extent->start +
802 						async_extent->ram_size - 1, 0);
803 			goto out_free_reserve;
804 		}
805 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
806 
807 		/*
808 		 * clear dirty, set writeback and unlock the pages.
809 		 */
810 		extent_clear_unlock_delalloc(inode, async_extent->start,
811 				async_extent->start +
812 				async_extent->ram_size - 1,
813 				async_extent->start +
814 				async_extent->ram_size - 1,
815 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
816 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
817 				PAGE_SET_WRITEBACK);
818 		ret = btrfs_submit_compressed_write(inode,
819 				    async_extent->start,
820 				    async_extent->ram_size,
821 				    ins.objectid,
822 				    ins.offset, async_extent->pages,
823 				    async_extent->nr_pages);
824 		if (ret) {
825 			struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
826 			struct page *p = async_extent->pages[0];
827 			const u64 start = async_extent->start;
828 			const u64 end = start + async_extent->ram_size - 1;
829 
830 			p->mapping = inode->i_mapping;
831 			tree->ops->writepage_end_io_hook(p, start, end,
832 							 NULL, 0);
833 			p->mapping = NULL;
834 			extent_clear_unlock_delalloc(inode, start, end, end,
835 						     NULL, 0,
836 						     PAGE_END_WRITEBACK |
837 						     PAGE_SET_ERROR);
838 			free_async_extent_pages(async_extent);
839 		}
840 		alloc_hint = ins.objectid + ins.offset;
841 		kfree(async_extent);
842 		cond_resched();
843 	}
844 	return;
845 out_free_reserve:
846 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
847 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
848 out_free:
849 	extent_clear_unlock_delalloc(inode, async_extent->start,
850 				     async_extent->start +
851 				     async_extent->ram_size - 1,
852 				     async_extent->start +
853 				     async_extent->ram_size - 1,
854 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
855 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
856 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
857 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
858 				     PAGE_SET_ERROR);
859 	free_async_extent_pages(async_extent);
860 	kfree(async_extent);
861 	goto again;
862 }
863 
864 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
865 				      u64 num_bytes)
866 {
867 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
868 	struct extent_map *em;
869 	u64 alloc_hint = 0;
870 
871 	read_lock(&em_tree->lock);
872 	em = search_extent_mapping(em_tree, start, num_bytes);
873 	if (em) {
874 		/*
875 		 * if block start isn't an actual block number then find the
876 		 * first block in this inode and use that as a hint.  If that
877 		 * block is also bogus then just don't worry about it.
878 		 */
879 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
880 			free_extent_map(em);
881 			em = search_extent_mapping(em_tree, 0, 0);
882 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
883 				alloc_hint = em->block_start;
884 			if (em)
885 				free_extent_map(em);
886 		} else {
887 			alloc_hint = em->block_start;
888 			free_extent_map(em);
889 		}
890 	}
891 	read_unlock(&em_tree->lock);
892 
893 	return alloc_hint;
894 }
895 
896 /*
897  * when extent_io.c finds a delayed allocation range in the file,
898  * the call backs end up in this code.  The basic idea is to
899  * allocate extents on disk for the range, and create ordered data structs
900  * in ram to track those extents.
901  *
902  * locked_page is the page that writepage had locked already.  We use
903  * it to make sure we don't do extra locks or unlocks.
904  *
905  * *page_started is set to one if we unlock locked_page and do everything
906  * required to start IO on it.  It may be clean and already done with
907  * IO when we return.
908  */
909 static noinline int cow_file_range(struct inode *inode,
910 				   struct page *locked_page,
911 				   u64 start, u64 end, u64 delalloc_end,
912 				   int *page_started, unsigned long *nr_written,
913 				   int unlock, struct btrfs_dedupe_hash *hash)
914 {
915 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
916 	struct btrfs_root *root = BTRFS_I(inode)->root;
917 	u64 alloc_hint = 0;
918 	u64 num_bytes;
919 	unsigned long ram_size;
920 	u64 disk_num_bytes;
921 	u64 cur_alloc_size;
922 	u64 blocksize = fs_info->sectorsize;
923 	struct btrfs_key ins;
924 	struct extent_map *em;
925 	int ret = 0;
926 
927 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
928 		WARN_ON_ONCE(1);
929 		ret = -EINVAL;
930 		goto out_unlock;
931 	}
932 
933 	num_bytes = ALIGN(end - start + 1, blocksize);
934 	num_bytes = max(blocksize,  num_bytes);
935 	disk_num_bytes = num_bytes;
936 
937 	inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
938 
939 	if (start == 0) {
940 		/* lets try to make an inline extent */
941 		ret = cow_file_range_inline(root, inode, start, end, 0,
942 					BTRFS_COMPRESS_NONE, NULL);
943 		if (ret == 0) {
944 			extent_clear_unlock_delalloc(inode, start, end,
945 				     delalloc_end, NULL,
946 				     EXTENT_LOCKED | EXTENT_DELALLOC |
947 				     EXTENT_DEFRAG, PAGE_UNLOCK |
948 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
949 				     PAGE_END_WRITEBACK);
950 			btrfs_free_reserved_data_space_noquota(inode, start,
951 						end - start + 1);
952 			*nr_written = *nr_written +
953 			     (end - start + PAGE_SIZE) / PAGE_SIZE;
954 			*page_started = 1;
955 			goto out;
956 		} else if (ret < 0) {
957 			goto out_unlock;
958 		}
959 	}
960 
961 	BUG_ON(disk_num_bytes >
962 	       btrfs_super_total_bytes(fs_info->super_copy));
963 
964 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
965 	btrfs_drop_extent_cache(BTRFS_I(inode), start,
966 			start + num_bytes - 1, 0);
967 
968 	while (disk_num_bytes > 0) {
969 		unsigned long op;
970 
971 		cur_alloc_size = disk_num_bytes;
972 		ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
973 					   fs_info->sectorsize, 0, alloc_hint,
974 					   &ins, 1, 1);
975 		if (ret < 0)
976 			goto out_unlock;
977 
978 		ram_size = ins.offset;
979 		em = create_io_em(inode, start, ins.offset, /* len */
980 				  start, /* orig_start */
981 				  ins.objectid, /* block_start */
982 				  ins.offset, /* block_len */
983 				  ins.offset, /* orig_block_len */
984 				  ram_size, /* ram_bytes */
985 				  BTRFS_COMPRESS_NONE, /* compress_type */
986 				  BTRFS_ORDERED_REGULAR /* type */);
987 		if (IS_ERR(em))
988 			goto out_reserve;
989 		free_extent_map(em);
990 
991 		cur_alloc_size = ins.offset;
992 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
993 					       ram_size, cur_alloc_size, 0);
994 		if (ret)
995 			goto out_drop_extent_cache;
996 
997 		if (root->root_key.objectid ==
998 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
999 			ret = btrfs_reloc_clone_csums(inode, start,
1000 						      cur_alloc_size);
1001 			if (ret)
1002 				goto out_drop_extent_cache;
1003 		}
1004 
1005 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1006 
1007 		if (disk_num_bytes < cur_alloc_size)
1008 			break;
1009 
1010 		/* we're not doing compressed IO, don't unlock the first
1011 		 * page (which the caller expects to stay locked), don't
1012 		 * clear any dirty bits and don't set any writeback bits
1013 		 *
1014 		 * Do set the Private2 bit so we know this page was properly
1015 		 * setup for writepage
1016 		 */
1017 		op = unlock ? PAGE_UNLOCK : 0;
1018 		op |= PAGE_SET_PRIVATE2;
1019 
1020 		extent_clear_unlock_delalloc(inode, start,
1021 					     start + ram_size - 1,
1022 					     delalloc_end, locked_page,
1023 					     EXTENT_LOCKED | EXTENT_DELALLOC,
1024 					     op);
1025 		disk_num_bytes -= cur_alloc_size;
1026 		num_bytes -= cur_alloc_size;
1027 		alloc_hint = ins.objectid + ins.offset;
1028 		start += cur_alloc_size;
1029 	}
1030 out:
1031 	return ret;
1032 
1033 out_drop_extent_cache:
1034 	btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1035 out_reserve:
1036 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1037 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1038 out_unlock:
1039 	extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1040 				     locked_page,
1041 				     EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1042 				     EXTENT_DELALLOC | EXTENT_DEFRAG,
1043 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1044 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1045 	goto out;
1046 }
1047 
1048 /*
1049  * work queue call back to started compression on a file and pages
1050  */
1051 static noinline void async_cow_start(struct btrfs_work *work)
1052 {
1053 	struct async_cow *async_cow;
1054 	int num_added = 0;
1055 	async_cow = container_of(work, struct async_cow, work);
1056 
1057 	compress_file_range(async_cow->inode, async_cow->locked_page,
1058 			    async_cow->start, async_cow->end, async_cow,
1059 			    &num_added);
1060 	if (num_added == 0) {
1061 		btrfs_add_delayed_iput(async_cow->inode);
1062 		async_cow->inode = NULL;
1063 	}
1064 }
1065 
1066 /*
1067  * work queue call back to submit previously compressed pages
1068  */
1069 static noinline void async_cow_submit(struct btrfs_work *work)
1070 {
1071 	struct btrfs_fs_info *fs_info;
1072 	struct async_cow *async_cow;
1073 	struct btrfs_root *root;
1074 	unsigned long nr_pages;
1075 
1076 	async_cow = container_of(work, struct async_cow, work);
1077 
1078 	root = async_cow->root;
1079 	fs_info = root->fs_info;
1080 	nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1081 		PAGE_SHIFT;
1082 
1083 	/*
1084 	 * atomic_sub_return implies a barrier for waitqueue_active
1085 	 */
1086 	if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1087 	    5 * SZ_1M &&
1088 	    waitqueue_active(&fs_info->async_submit_wait))
1089 		wake_up(&fs_info->async_submit_wait);
1090 
1091 	if (async_cow->inode)
1092 		submit_compressed_extents(async_cow->inode, async_cow);
1093 }
1094 
1095 static noinline void async_cow_free(struct btrfs_work *work)
1096 {
1097 	struct async_cow *async_cow;
1098 	async_cow = container_of(work, struct async_cow, work);
1099 	if (async_cow->inode)
1100 		btrfs_add_delayed_iput(async_cow->inode);
1101 	kfree(async_cow);
1102 }
1103 
1104 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1105 				u64 start, u64 end, int *page_started,
1106 				unsigned long *nr_written)
1107 {
1108 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1109 	struct async_cow *async_cow;
1110 	struct btrfs_root *root = BTRFS_I(inode)->root;
1111 	unsigned long nr_pages;
1112 	u64 cur_end;
1113 
1114 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1115 			 1, 0, NULL, GFP_NOFS);
1116 	while (start < end) {
1117 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1118 		BUG_ON(!async_cow); /* -ENOMEM */
1119 		async_cow->inode = igrab(inode);
1120 		async_cow->root = root;
1121 		async_cow->locked_page = locked_page;
1122 		async_cow->start = start;
1123 
1124 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1125 		    !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1126 			cur_end = end;
1127 		else
1128 			cur_end = min(end, start + SZ_512K - 1);
1129 
1130 		async_cow->end = cur_end;
1131 		INIT_LIST_HEAD(&async_cow->extents);
1132 
1133 		btrfs_init_work(&async_cow->work,
1134 				btrfs_delalloc_helper,
1135 				async_cow_start, async_cow_submit,
1136 				async_cow_free);
1137 
1138 		nr_pages = (cur_end - start + PAGE_SIZE) >>
1139 			PAGE_SHIFT;
1140 		atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1141 
1142 		btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1143 
1144 		while (atomic_read(&fs_info->async_submit_draining) &&
1145 		       atomic_read(&fs_info->async_delalloc_pages)) {
1146 			wait_event(fs_info->async_submit_wait,
1147 				   (atomic_read(&fs_info->async_delalloc_pages) ==
1148 				    0));
1149 		}
1150 
1151 		*nr_written += nr_pages;
1152 		start = cur_end + 1;
1153 	}
1154 	*page_started = 1;
1155 	return 0;
1156 }
1157 
1158 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1159 					u64 bytenr, u64 num_bytes)
1160 {
1161 	int ret;
1162 	struct btrfs_ordered_sum *sums;
1163 	LIST_HEAD(list);
1164 
1165 	ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1166 				       bytenr + num_bytes - 1, &list, 0);
1167 	if (ret == 0 && list_empty(&list))
1168 		return 0;
1169 
1170 	while (!list_empty(&list)) {
1171 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1172 		list_del(&sums->list);
1173 		kfree(sums);
1174 	}
1175 	return 1;
1176 }
1177 
1178 /*
1179  * when nowcow writeback call back.  This checks for snapshots or COW copies
1180  * of the extents that exist in the file, and COWs the file as required.
1181  *
1182  * If no cow copies or snapshots exist, we write directly to the existing
1183  * blocks on disk
1184  */
1185 static noinline int run_delalloc_nocow(struct inode *inode,
1186 				       struct page *locked_page,
1187 			      u64 start, u64 end, int *page_started, int force,
1188 			      unsigned long *nr_written)
1189 {
1190 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1191 	struct btrfs_root *root = BTRFS_I(inode)->root;
1192 	struct extent_buffer *leaf;
1193 	struct btrfs_path *path;
1194 	struct btrfs_file_extent_item *fi;
1195 	struct btrfs_key found_key;
1196 	struct extent_map *em;
1197 	u64 cow_start;
1198 	u64 cur_offset;
1199 	u64 extent_end;
1200 	u64 extent_offset;
1201 	u64 disk_bytenr;
1202 	u64 num_bytes;
1203 	u64 disk_num_bytes;
1204 	u64 ram_bytes;
1205 	int extent_type;
1206 	int ret, err;
1207 	int type;
1208 	int nocow;
1209 	int check_prev = 1;
1210 	bool nolock;
1211 	u64 ino = btrfs_ino(BTRFS_I(inode));
1212 
1213 	path = btrfs_alloc_path();
1214 	if (!path) {
1215 		extent_clear_unlock_delalloc(inode, start, end, end,
1216 					     locked_page,
1217 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1218 					     EXTENT_DO_ACCOUNTING |
1219 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1220 					     PAGE_CLEAR_DIRTY |
1221 					     PAGE_SET_WRITEBACK |
1222 					     PAGE_END_WRITEBACK);
1223 		return -ENOMEM;
1224 	}
1225 
1226 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1227 
1228 	cow_start = (u64)-1;
1229 	cur_offset = start;
1230 	while (1) {
1231 		ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1232 					       cur_offset, 0);
1233 		if (ret < 0)
1234 			goto error;
1235 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1236 			leaf = path->nodes[0];
1237 			btrfs_item_key_to_cpu(leaf, &found_key,
1238 					      path->slots[0] - 1);
1239 			if (found_key.objectid == ino &&
1240 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1241 				path->slots[0]--;
1242 		}
1243 		check_prev = 0;
1244 next_slot:
1245 		leaf = path->nodes[0];
1246 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1247 			ret = btrfs_next_leaf(root, path);
1248 			if (ret < 0)
1249 				goto error;
1250 			if (ret > 0)
1251 				break;
1252 			leaf = path->nodes[0];
1253 		}
1254 
1255 		nocow = 0;
1256 		disk_bytenr = 0;
1257 		num_bytes = 0;
1258 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1259 
1260 		if (found_key.objectid > ino)
1261 			break;
1262 		if (WARN_ON_ONCE(found_key.objectid < ino) ||
1263 		    found_key.type < BTRFS_EXTENT_DATA_KEY) {
1264 			path->slots[0]++;
1265 			goto next_slot;
1266 		}
1267 		if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1268 		    found_key.offset > end)
1269 			break;
1270 
1271 		if (found_key.offset > cur_offset) {
1272 			extent_end = found_key.offset;
1273 			extent_type = 0;
1274 			goto out_check;
1275 		}
1276 
1277 		fi = btrfs_item_ptr(leaf, path->slots[0],
1278 				    struct btrfs_file_extent_item);
1279 		extent_type = btrfs_file_extent_type(leaf, fi);
1280 
1281 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1282 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1283 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1284 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1285 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1286 			extent_end = found_key.offset +
1287 				btrfs_file_extent_num_bytes(leaf, fi);
1288 			disk_num_bytes =
1289 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1290 			if (extent_end <= start) {
1291 				path->slots[0]++;
1292 				goto next_slot;
1293 			}
1294 			if (disk_bytenr == 0)
1295 				goto out_check;
1296 			if (btrfs_file_extent_compression(leaf, fi) ||
1297 			    btrfs_file_extent_encryption(leaf, fi) ||
1298 			    btrfs_file_extent_other_encoding(leaf, fi))
1299 				goto out_check;
1300 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1301 				goto out_check;
1302 			if (btrfs_extent_readonly(fs_info, disk_bytenr))
1303 				goto out_check;
1304 			if (btrfs_cross_ref_exist(root, ino,
1305 						  found_key.offset -
1306 						  extent_offset, disk_bytenr))
1307 				goto out_check;
1308 			disk_bytenr += extent_offset;
1309 			disk_bytenr += cur_offset - found_key.offset;
1310 			num_bytes = min(end + 1, extent_end) - cur_offset;
1311 			/*
1312 			 * if there are pending snapshots for this root,
1313 			 * we fall into common COW way.
1314 			 */
1315 			if (!nolock) {
1316 				err = btrfs_start_write_no_snapshoting(root);
1317 				if (!err)
1318 					goto out_check;
1319 			}
1320 			/*
1321 			 * force cow if csum exists in the range.
1322 			 * this ensure that csum for a given extent are
1323 			 * either valid or do not exist.
1324 			 */
1325 			if (csum_exist_in_range(fs_info, disk_bytenr,
1326 						num_bytes)) {
1327 				if (!nolock)
1328 					btrfs_end_write_no_snapshoting(root);
1329 				goto out_check;
1330 			}
1331 			if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1332 				if (!nolock)
1333 					btrfs_end_write_no_snapshoting(root);
1334 				goto out_check;
1335 			}
1336 			nocow = 1;
1337 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1338 			extent_end = found_key.offset +
1339 				btrfs_file_extent_inline_len(leaf,
1340 						     path->slots[0], fi);
1341 			extent_end = ALIGN(extent_end,
1342 					   fs_info->sectorsize);
1343 		} else {
1344 			BUG_ON(1);
1345 		}
1346 out_check:
1347 		if (extent_end <= start) {
1348 			path->slots[0]++;
1349 			if (!nolock && nocow)
1350 				btrfs_end_write_no_snapshoting(root);
1351 			if (nocow)
1352 				btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1353 			goto next_slot;
1354 		}
1355 		if (!nocow) {
1356 			if (cow_start == (u64)-1)
1357 				cow_start = cur_offset;
1358 			cur_offset = extent_end;
1359 			if (cur_offset > end)
1360 				break;
1361 			path->slots[0]++;
1362 			goto next_slot;
1363 		}
1364 
1365 		btrfs_release_path(path);
1366 		if (cow_start != (u64)-1) {
1367 			ret = cow_file_range(inode, locked_page,
1368 					     cow_start, found_key.offset - 1,
1369 					     end, page_started, nr_written, 1,
1370 					     NULL);
1371 			if (ret) {
1372 				if (!nolock && nocow)
1373 					btrfs_end_write_no_snapshoting(root);
1374 				if (nocow)
1375 					btrfs_dec_nocow_writers(fs_info,
1376 								disk_bytenr);
1377 				goto error;
1378 			}
1379 			cow_start = (u64)-1;
1380 		}
1381 
1382 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1383 			u64 orig_start = found_key.offset - extent_offset;
1384 
1385 			em = create_io_em(inode, cur_offset, num_bytes,
1386 					  orig_start,
1387 					  disk_bytenr, /* block_start */
1388 					  num_bytes, /* block_len */
1389 					  disk_num_bytes, /* orig_block_len */
1390 					  ram_bytes, BTRFS_COMPRESS_NONE,
1391 					  BTRFS_ORDERED_PREALLOC);
1392 			if (IS_ERR(em)) {
1393 				if (!nolock && nocow)
1394 					btrfs_end_write_no_snapshoting(root);
1395 				if (nocow)
1396 					btrfs_dec_nocow_writers(fs_info,
1397 								disk_bytenr);
1398 				ret = PTR_ERR(em);
1399 				goto error;
1400 			}
1401 			free_extent_map(em);
1402 		}
1403 
1404 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1405 			type = BTRFS_ORDERED_PREALLOC;
1406 		} else {
1407 			type = BTRFS_ORDERED_NOCOW;
1408 		}
1409 
1410 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1411 					       num_bytes, num_bytes, type);
1412 		if (nocow)
1413 			btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1414 		BUG_ON(ret); /* -ENOMEM */
1415 
1416 		if (root->root_key.objectid ==
1417 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1418 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1419 						      num_bytes);
1420 			if (ret) {
1421 				if (!nolock && nocow)
1422 					btrfs_end_write_no_snapshoting(root);
1423 				goto error;
1424 			}
1425 		}
1426 
1427 		extent_clear_unlock_delalloc(inode, cur_offset,
1428 					     cur_offset + num_bytes - 1, end,
1429 					     locked_page, EXTENT_LOCKED |
1430 					     EXTENT_DELALLOC |
1431 					     EXTENT_CLEAR_DATA_RESV,
1432 					     PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1433 
1434 		if (!nolock && nocow)
1435 			btrfs_end_write_no_snapshoting(root);
1436 		cur_offset = extent_end;
1437 		if (cur_offset > end)
1438 			break;
1439 	}
1440 	btrfs_release_path(path);
1441 
1442 	if (cur_offset <= end && cow_start == (u64)-1) {
1443 		cow_start = cur_offset;
1444 		cur_offset = end;
1445 	}
1446 
1447 	if (cow_start != (u64)-1) {
1448 		ret = cow_file_range(inode, locked_page, cow_start, end, end,
1449 				     page_started, nr_written, 1, NULL);
1450 		if (ret)
1451 			goto error;
1452 	}
1453 
1454 error:
1455 	if (ret && cur_offset < end)
1456 		extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1457 					     locked_page, EXTENT_LOCKED |
1458 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1459 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1460 					     PAGE_CLEAR_DIRTY |
1461 					     PAGE_SET_WRITEBACK |
1462 					     PAGE_END_WRITEBACK);
1463 	btrfs_free_path(path);
1464 	return ret;
1465 }
1466 
1467 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1468 {
1469 
1470 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1471 	    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1472 		return 0;
1473 
1474 	/*
1475 	 * @defrag_bytes is a hint value, no spinlock held here,
1476 	 * if is not zero, it means the file is defragging.
1477 	 * Force cow if given extent needs to be defragged.
1478 	 */
1479 	if (BTRFS_I(inode)->defrag_bytes &&
1480 	    test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1481 			   EXTENT_DEFRAG, 0, NULL))
1482 		return 1;
1483 
1484 	return 0;
1485 }
1486 
1487 /*
1488  * extent_io.c call back to do delayed allocation processing
1489  */
1490 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1491 			      u64 start, u64 end, int *page_started,
1492 			      unsigned long *nr_written)
1493 {
1494 	int ret;
1495 	int force_cow = need_force_cow(inode, start, end);
1496 
1497 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1498 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1499 					 page_started, 1, nr_written);
1500 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1501 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1502 					 page_started, 0, nr_written);
1503 	} else if (!inode_need_compress(inode)) {
1504 		ret = cow_file_range(inode, locked_page, start, end, end,
1505 				      page_started, nr_written, 1, NULL);
1506 	} else {
1507 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1508 			&BTRFS_I(inode)->runtime_flags);
1509 		ret = cow_file_range_async(inode, locked_page, start, end,
1510 					   page_started, nr_written);
1511 	}
1512 	return ret;
1513 }
1514 
1515 static void btrfs_split_extent_hook(struct inode *inode,
1516 				    struct extent_state *orig, u64 split)
1517 {
1518 	u64 size;
1519 
1520 	/* not delalloc, ignore it */
1521 	if (!(orig->state & EXTENT_DELALLOC))
1522 		return;
1523 
1524 	size = orig->end - orig->start + 1;
1525 	if (size > BTRFS_MAX_EXTENT_SIZE) {
1526 		u32 num_extents;
1527 		u64 new_size;
1528 
1529 		/*
1530 		 * See the explanation in btrfs_merge_extent_hook, the same
1531 		 * applies here, just in reverse.
1532 		 */
1533 		new_size = orig->end - split + 1;
1534 		num_extents = count_max_extents(new_size);
1535 		new_size = split - orig->start;
1536 		num_extents += count_max_extents(new_size);
1537 		if (count_max_extents(size) >= num_extents)
1538 			return;
1539 	}
1540 
1541 	spin_lock(&BTRFS_I(inode)->lock);
1542 	BTRFS_I(inode)->outstanding_extents++;
1543 	spin_unlock(&BTRFS_I(inode)->lock);
1544 }
1545 
1546 /*
1547  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1548  * extents so we can keep track of new extents that are just merged onto old
1549  * extents, such as when we are doing sequential writes, so we can properly
1550  * account for the metadata space we'll need.
1551  */
1552 static void btrfs_merge_extent_hook(struct inode *inode,
1553 				    struct extent_state *new,
1554 				    struct extent_state *other)
1555 {
1556 	u64 new_size, old_size;
1557 	u32 num_extents;
1558 
1559 	/* not delalloc, ignore it */
1560 	if (!(other->state & EXTENT_DELALLOC))
1561 		return;
1562 
1563 	if (new->start > other->start)
1564 		new_size = new->end - other->start + 1;
1565 	else
1566 		new_size = other->end - new->start + 1;
1567 
1568 	/* we're not bigger than the max, unreserve the space and go */
1569 	if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1570 		spin_lock(&BTRFS_I(inode)->lock);
1571 		BTRFS_I(inode)->outstanding_extents--;
1572 		spin_unlock(&BTRFS_I(inode)->lock);
1573 		return;
1574 	}
1575 
1576 	/*
1577 	 * We have to add up either side to figure out how many extents were
1578 	 * accounted for before we merged into one big extent.  If the number of
1579 	 * extents we accounted for is <= the amount we need for the new range
1580 	 * then we can return, otherwise drop.  Think of it like this
1581 	 *
1582 	 * [ 4k][MAX_SIZE]
1583 	 *
1584 	 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1585 	 * need 2 outstanding extents, on one side we have 1 and the other side
1586 	 * we have 1 so they are == and we can return.  But in this case
1587 	 *
1588 	 * [MAX_SIZE+4k][MAX_SIZE+4k]
1589 	 *
1590 	 * Each range on their own accounts for 2 extents, but merged together
1591 	 * they are only 3 extents worth of accounting, so we need to drop in
1592 	 * this case.
1593 	 */
1594 	old_size = other->end - other->start + 1;
1595 	num_extents = count_max_extents(old_size);
1596 	old_size = new->end - new->start + 1;
1597 	num_extents += count_max_extents(old_size);
1598 	if (count_max_extents(new_size) >= num_extents)
1599 		return;
1600 
1601 	spin_lock(&BTRFS_I(inode)->lock);
1602 	BTRFS_I(inode)->outstanding_extents--;
1603 	spin_unlock(&BTRFS_I(inode)->lock);
1604 }
1605 
1606 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1607 				      struct inode *inode)
1608 {
1609 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1610 
1611 	spin_lock(&root->delalloc_lock);
1612 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1613 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1614 			      &root->delalloc_inodes);
1615 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1616 			&BTRFS_I(inode)->runtime_flags);
1617 		root->nr_delalloc_inodes++;
1618 		if (root->nr_delalloc_inodes == 1) {
1619 			spin_lock(&fs_info->delalloc_root_lock);
1620 			BUG_ON(!list_empty(&root->delalloc_root));
1621 			list_add_tail(&root->delalloc_root,
1622 				      &fs_info->delalloc_roots);
1623 			spin_unlock(&fs_info->delalloc_root_lock);
1624 		}
1625 	}
1626 	spin_unlock(&root->delalloc_lock);
1627 }
1628 
1629 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1630 				     struct btrfs_inode *inode)
1631 {
1632 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1633 
1634 	spin_lock(&root->delalloc_lock);
1635 	if (!list_empty(&inode->delalloc_inodes)) {
1636 		list_del_init(&inode->delalloc_inodes);
1637 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1638 			  &inode->runtime_flags);
1639 		root->nr_delalloc_inodes--;
1640 		if (!root->nr_delalloc_inodes) {
1641 			spin_lock(&fs_info->delalloc_root_lock);
1642 			BUG_ON(list_empty(&root->delalloc_root));
1643 			list_del_init(&root->delalloc_root);
1644 			spin_unlock(&fs_info->delalloc_root_lock);
1645 		}
1646 	}
1647 	spin_unlock(&root->delalloc_lock);
1648 }
1649 
1650 /*
1651  * extent_io.c set_bit_hook, used to track delayed allocation
1652  * bytes in this file, and to maintain the list of inodes that
1653  * have pending delalloc work to be done.
1654  */
1655 static void btrfs_set_bit_hook(struct inode *inode,
1656 			       struct extent_state *state, unsigned *bits)
1657 {
1658 
1659 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1660 
1661 	if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1662 		WARN_ON(1);
1663 	/*
1664 	 * set_bit and clear bit hooks normally require _irqsave/restore
1665 	 * but in this case, we are only testing for the DELALLOC
1666 	 * bit, which is only set or cleared with irqs on
1667 	 */
1668 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1669 		struct btrfs_root *root = BTRFS_I(inode)->root;
1670 		u64 len = state->end + 1 - state->start;
1671 		bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1672 
1673 		if (*bits & EXTENT_FIRST_DELALLOC) {
1674 			*bits &= ~EXTENT_FIRST_DELALLOC;
1675 		} else {
1676 			spin_lock(&BTRFS_I(inode)->lock);
1677 			BTRFS_I(inode)->outstanding_extents++;
1678 			spin_unlock(&BTRFS_I(inode)->lock);
1679 		}
1680 
1681 		/* For sanity tests */
1682 		if (btrfs_is_testing(fs_info))
1683 			return;
1684 
1685 		__percpu_counter_add(&fs_info->delalloc_bytes, len,
1686 				     fs_info->delalloc_batch);
1687 		spin_lock(&BTRFS_I(inode)->lock);
1688 		BTRFS_I(inode)->delalloc_bytes += len;
1689 		if (*bits & EXTENT_DEFRAG)
1690 			BTRFS_I(inode)->defrag_bytes += len;
1691 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1692 					 &BTRFS_I(inode)->runtime_flags))
1693 			btrfs_add_delalloc_inodes(root, inode);
1694 		spin_unlock(&BTRFS_I(inode)->lock);
1695 	}
1696 }
1697 
1698 /*
1699  * extent_io.c clear_bit_hook, see set_bit_hook for why
1700  */
1701 static void btrfs_clear_bit_hook(struct btrfs_inode *inode,
1702 				 struct extent_state *state,
1703 				 unsigned *bits)
1704 {
1705 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1706 	u64 len = state->end + 1 - state->start;
1707 	u32 num_extents = count_max_extents(len);
1708 
1709 	spin_lock(&inode->lock);
1710 	if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1711 		inode->defrag_bytes -= len;
1712 	spin_unlock(&inode->lock);
1713 
1714 	/*
1715 	 * set_bit and clear bit hooks normally require _irqsave/restore
1716 	 * but in this case, we are only testing for the DELALLOC
1717 	 * bit, which is only set or cleared with irqs on
1718 	 */
1719 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1720 		struct btrfs_root *root = inode->root;
1721 		bool do_list = !btrfs_is_free_space_inode(inode);
1722 
1723 		if (*bits & EXTENT_FIRST_DELALLOC) {
1724 			*bits &= ~EXTENT_FIRST_DELALLOC;
1725 		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1726 			spin_lock(&inode->lock);
1727 			inode->outstanding_extents -= num_extents;
1728 			spin_unlock(&inode->lock);
1729 		}
1730 
1731 		/*
1732 		 * We don't reserve metadata space for space cache inodes so we
1733 		 * don't need to call dellalloc_release_metadata if there is an
1734 		 * error.
1735 		 */
1736 		if (*bits & EXTENT_DO_ACCOUNTING &&
1737 		    root != fs_info->tree_root)
1738 			btrfs_delalloc_release_metadata(inode, len);
1739 
1740 		/* For sanity tests. */
1741 		if (btrfs_is_testing(fs_info))
1742 			return;
1743 
1744 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1745 		    && do_list && !(state->state & EXTENT_NORESERVE)
1746 		    && (*bits & (EXTENT_DO_ACCOUNTING |
1747 		    EXTENT_CLEAR_DATA_RESV)))
1748 			btrfs_free_reserved_data_space_noquota(
1749 					&inode->vfs_inode,
1750 					state->start, len);
1751 
1752 		__percpu_counter_add(&fs_info->delalloc_bytes, -len,
1753 				     fs_info->delalloc_batch);
1754 		spin_lock(&inode->lock);
1755 		inode->delalloc_bytes -= len;
1756 		if (do_list && inode->delalloc_bytes == 0 &&
1757 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1758 					&inode->runtime_flags))
1759 			btrfs_del_delalloc_inode(root, inode);
1760 		spin_unlock(&inode->lock);
1761 	}
1762 }
1763 
1764 /*
1765  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1766  * we don't create bios that span stripes or chunks
1767  *
1768  * return 1 if page cannot be merged to bio
1769  * return 0 if page can be merged to bio
1770  * return error otherwise
1771  */
1772 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1773 			 size_t size, struct bio *bio,
1774 			 unsigned long bio_flags)
1775 {
1776 	struct inode *inode = page->mapping->host;
1777 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1778 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1779 	u64 length = 0;
1780 	u64 map_length;
1781 	int ret;
1782 
1783 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1784 		return 0;
1785 
1786 	length = bio->bi_iter.bi_size;
1787 	map_length = length;
1788 	ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1789 			      NULL, 0);
1790 	if (ret < 0)
1791 		return ret;
1792 	if (map_length < length + size)
1793 		return 1;
1794 	return 0;
1795 }
1796 
1797 /*
1798  * in order to insert checksums into the metadata in large chunks,
1799  * we wait until bio submission time.   All the pages in the bio are
1800  * checksummed and sums are attached onto the ordered extent record.
1801  *
1802  * At IO completion time the cums attached on the ordered extent record
1803  * are inserted into the btree
1804  */
1805 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1806 				    int mirror_num, unsigned long bio_flags,
1807 				    u64 bio_offset)
1808 {
1809 	int ret = 0;
1810 
1811 	ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1812 	BUG_ON(ret); /* -ENOMEM */
1813 	return 0;
1814 }
1815 
1816 /*
1817  * in order to insert checksums into the metadata in large chunks,
1818  * we wait until bio submission time.   All the pages in the bio are
1819  * checksummed and sums are attached onto the ordered extent record.
1820  *
1821  * At IO completion time the cums attached on the ordered extent record
1822  * are inserted into the btree
1823  */
1824 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1825 			  int mirror_num, unsigned long bio_flags,
1826 			  u64 bio_offset)
1827 {
1828 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1829 	int ret;
1830 
1831 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1832 	if (ret) {
1833 		bio->bi_error = ret;
1834 		bio_endio(bio);
1835 	}
1836 	return ret;
1837 }
1838 
1839 /*
1840  * extent_io.c submission hook. This does the right thing for csum calculation
1841  * on write, or reading the csums from the tree before a read
1842  */
1843 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1844 			  int mirror_num, unsigned long bio_flags,
1845 			  u64 bio_offset)
1846 {
1847 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1848 	struct btrfs_root *root = BTRFS_I(inode)->root;
1849 	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1850 	int ret = 0;
1851 	int skip_sum;
1852 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1853 
1854 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1855 
1856 	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1857 		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1858 
1859 	if (bio_op(bio) != REQ_OP_WRITE) {
1860 		ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1861 		if (ret)
1862 			goto out;
1863 
1864 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1865 			ret = btrfs_submit_compressed_read(inode, bio,
1866 							   mirror_num,
1867 							   bio_flags);
1868 			goto out;
1869 		} else if (!skip_sum) {
1870 			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1871 			if (ret)
1872 				goto out;
1873 		}
1874 		goto mapit;
1875 	} else if (async && !skip_sum) {
1876 		/* csum items have already been cloned */
1877 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1878 			goto mapit;
1879 		/* we're doing a write, do the async checksumming */
1880 		ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num,
1881 					  bio_flags, bio_offset,
1882 					  __btrfs_submit_bio_start,
1883 					  __btrfs_submit_bio_done);
1884 		goto out;
1885 	} else if (!skip_sum) {
1886 		ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1887 		if (ret)
1888 			goto out;
1889 	}
1890 
1891 mapit:
1892 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1893 
1894 out:
1895 	if (ret < 0) {
1896 		bio->bi_error = ret;
1897 		bio_endio(bio);
1898 	}
1899 	return ret;
1900 }
1901 
1902 /*
1903  * given a list of ordered sums record them in the inode.  This happens
1904  * at IO completion time based on sums calculated at bio submission time.
1905  */
1906 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1907 			     struct inode *inode, struct list_head *list)
1908 {
1909 	struct btrfs_ordered_sum *sum;
1910 
1911 	list_for_each_entry(sum, list, list) {
1912 		trans->adding_csums = 1;
1913 		btrfs_csum_file_blocks(trans,
1914 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
1915 		trans->adding_csums = 0;
1916 	}
1917 	return 0;
1918 }
1919 
1920 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1921 			      struct extent_state **cached_state, int dedupe)
1922 {
1923 	WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1924 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1925 				   cached_state);
1926 }
1927 
1928 /* see btrfs_writepage_start_hook for details on why this is required */
1929 struct btrfs_writepage_fixup {
1930 	struct page *page;
1931 	struct btrfs_work work;
1932 };
1933 
1934 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1935 {
1936 	struct btrfs_writepage_fixup *fixup;
1937 	struct btrfs_ordered_extent *ordered;
1938 	struct extent_state *cached_state = NULL;
1939 	struct page *page;
1940 	struct inode *inode;
1941 	u64 page_start;
1942 	u64 page_end;
1943 	int ret;
1944 
1945 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
1946 	page = fixup->page;
1947 again:
1948 	lock_page(page);
1949 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1950 		ClearPageChecked(page);
1951 		goto out_page;
1952 	}
1953 
1954 	inode = page->mapping->host;
1955 	page_start = page_offset(page);
1956 	page_end = page_offset(page) + PAGE_SIZE - 1;
1957 
1958 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1959 			 &cached_state);
1960 
1961 	/* already ordered? We're done */
1962 	if (PagePrivate2(page))
1963 		goto out;
1964 
1965 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
1966 					PAGE_SIZE);
1967 	if (ordered) {
1968 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1969 				     page_end, &cached_state, GFP_NOFS);
1970 		unlock_page(page);
1971 		btrfs_start_ordered_extent(inode, ordered, 1);
1972 		btrfs_put_ordered_extent(ordered);
1973 		goto again;
1974 	}
1975 
1976 	ret = btrfs_delalloc_reserve_space(inode, page_start,
1977 					   PAGE_SIZE);
1978 	if (ret) {
1979 		mapping_set_error(page->mapping, ret);
1980 		end_extent_writepage(page, ret, page_start, page_end);
1981 		ClearPageChecked(page);
1982 		goto out;
1983 	 }
1984 
1985 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
1986 				  0);
1987 	ClearPageChecked(page);
1988 	set_page_dirty(page);
1989 out:
1990 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1991 			     &cached_state, GFP_NOFS);
1992 out_page:
1993 	unlock_page(page);
1994 	put_page(page);
1995 	kfree(fixup);
1996 }
1997 
1998 /*
1999  * There are a few paths in the higher layers of the kernel that directly
2000  * set the page dirty bit without asking the filesystem if it is a
2001  * good idea.  This causes problems because we want to make sure COW
2002  * properly happens and the data=ordered rules are followed.
2003  *
2004  * In our case any range that doesn't have the ORDERED bit set
2005  * hasn't been properly setup for IO.  We kick off an async process
2006  * to fix it up.  The async helper will wait for ordered extents, set
2007  * the delalloc bit and make it safe to write the page.
2008  */
2009 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2010 {
2011 	struct inode *inode = page->mapping->host;
2012 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2013 	struct btrfs_writepage_fixup *fixup;
2014 
2015 	/* this page is properly in the ordered list */
2016 	if (TestClearPagePrivate2(page))
2017 		return 0;
2018 
2019 	if (PageChecked(page))
2020 		return -EAGAIN;
2021 
2022 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2023 	if (!fixup)
2024 		return -EAGAIN;
2025 
2026 	SetPageChecked(page);
2027 	get_page(page);
2028 	btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2029 			btrfs_writepage_fixup_worker, NULL, NULL);
2030 	fixup->page = page;
2031 	btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2032 	return -EBUSY;
2033 }
2034 
2035 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2036 				       struct inode *inode, u64 file_pos,
2037 				       u64 disk_bytenr, u64 disk_num_bytes,
2038 				       u64 num_bytes, u64 ram_bytes,
2039 				       u8 compression, u8 encryption,
2040 				       u16 other_encoding, int extent_type)
2041 {
2042 	struct btrfs_root *root = BTRFS_I(inode)->root;
2043 	struct btrfs_file_extent_item *fi;
2044 	struct btrfs_path *path;
2045 	struct extent_buffer *leaf;
2046 	struct btrfs_key ins;
2047 	int extent_inserted = 0;
2048 	int ret;
2049 
2050 	path = btrfs_alloc_path();
2051 	if (!path)
2052 		return -ENOMEM;
2053 
2054 	/*
2055 	 * we may be replacing one extent in the tree with another.
2056 	 * The new extent is pinned in the extent map, and we don't want
2057 	 * to drop it from the cache until it is completely in the btree.
2058 	 *
2059 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
2060 	 * the caller is expected to unpin it and allow it to be merged
2061 	 * with the others.
2062 	 */
2063 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2064 				   file_pos + num_bytes, NULL, 0,
2065 				   1, sizeof(*fi), &extent_inserted);
2066 	if (ret)
2067 		goto out;
2068 
2069 	if (!extent_inserted) {
2070 		ins.objectid = btrfs_ino(BTRFS_I(inode));
2071 		ins.offset = file_pos;
2072 		ins.type = BTRFS_EXTENT_DATA_KEY;
2073 
2074 		path->leave_spinning = 1;
2075 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
2076 					      sizeof(*fi));
2077 		if (ret)
2078 			goto out;
2079 	}
2080 	leaf = path->nodes[0];
2081 	fi = btrfs_item_ptr(leaf, path->slots[0],
2082 			    struct btrfs_file_extent_item);
2083 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2084 	btrfs_set_file_extent_type(leaf, fi, extent_type);
2085 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2086 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2087 	btrfs_set_file_extent_offset(leaf, fi, 0);
2088 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2089 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2090 	btrfs_set_file_extent_compression(leaf, fi, compression);
2091 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
2092 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2093 
2094 	btrfs_mark_buffer_dirty(leaf);
2095 	btrfs_release_path(path);
2096 
2097 	inode_add_bytes(inode, num_bytes);
2098 
2099 	ins.objectid = disk_bytenr;
2100 	ins.offset = disk_num_bytes;
2101 	ins.type = BTRFS_EXTENT_ITEM_KEY;
2102 	ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2103 			btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins);
2104 	/*
2105 	 * Release the reserved range from inode dirty range map, as it is
2106 	 * already moved into delayed_ref_head
2107 	 */
2108 	btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2109 out:
2110 	btrfs_free_path(path);
2111 
2112 	return ret;
2113 }
2114 
2115 /* snapshot-aware defrag */
2116 struct sa_defrag_extent_backref {
2117 	struct rb_node node;
2118 	struct old_sa_defrag_extent *old;
2119 	u64 root_id;
2120 	u64 inum;
2121 	u64 file_pos;
2122 	u64 extent_offset;
2123 	u64 num_bytes;
2124 	u64 generation;
2125 };
2126 
2127 struct old_sa_defrag_extent {
2128 	struct list_head list;
2129 	struct new_sa_defrag_extent *new;
2130 
2131 	u64 extent_offset;
2132 	u64 bytenr;
2133 	u64 offset;
2134 	u64 len;
2135 	int count;
2136 };
2137 
2138 struct new_sa_defrag_extent {
2139 	struct rb_root root;
2140 	struct list_head head;
2141 	struct btrfs_path *path;
2142 	struct inode *inode;
2143 	u64 file_pos;
2144 	u64 len;
2145 	u64 bytenr;
2146 	u64 disk_len;
2147 	u8 compress_type;
2148 };
2149 
2150 static int backref_comp(struct sa_defrag_extent_backref *b1,
2151 			struct sa_defrag_extent_backref *b2)
2152 {
2153 	if (b1->root_id < b2->root_id)
2154 		return -1;
2155 	else if (b1->root_id > b2->root_id)
2156 		return 1;
2157 
2158 	if (b1->inum < b2->inum)
2159 		return -1;
2160 	else if (b1->inum > b2->inum)
2161 		return 1;
2162 
2163 	if (b1->file_pos < b2->file_pos)
2164 		return -1;
2165 	else if (b1->file_pos > b2->file_pos)
2166 		return 1;
2167 
2168 	/*
2169 	 * [------------------------------] ===> (a range of space)
2170 	 *     |<--->|   |<---->| =============> (fs/file tree A)
2171 	 * |<---------------------------->| ===> (fs/file tree B)
2172 	 *
2173 	 * A range of space can refer to two file extents in one tree while
2174 	 * refer to only one file extent in another tree.
2175 	 *
2176 	 * So we may process a disk offset more than one time(two extents in A)
2177 	 * and locate at the same extent(one extent in B), then insert two same
2178 	 * backrefs(both refer to the extent in B).
2179 	 */
2180 	return 0;
2181 }
2182 
2183 static void backref_insert(struct rb_root *root,
2184 			   struct sa_defrag_extent_backref *backref)
2185 {
2186 	struct rb_node **p = &root->rb_node;
2187 	struct rb_node *parent = NULL;
2188 	struct sa_defrag_extent_backref *entry;
2189 	int ret;
2190 
2191 	while (*p) {
2192 		parent = *p;
2193 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2194 
2195 		ret = backref_comp(backref, entry);
2196 		if (ret < 0)
2197 			p = &(*p)->rb_left;
2198 		else
2199 			p = &(*p)->rb_right;
2200 	}
2201 
2202 	rb_link_node(&backref->node, parent, p);
2203 	rb_insert_color(&backref->node, root);
2204 }
2205 
2206 /*
2207  * Note the backref might has changed, and in this case we just return 0.
2208  */
2209 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2210 				       void *ctx)
2211 {
2212 	struct btrfs_file_extent_item *extent;
2213 	struct old_sa_defrag_extent *old = ctx;
2214 	struct new_sa_defrag_extent *new = old->new;
2215 	struct btrfs_path *path = new->path;
2216 	struct btrfs_key key;
2217 	struct btrfs_root *root;
2218 	struct sa_defrag_extent_backref *backref;
2219 	struct extent_buffer *leaf;
2220 	struct inode *inode = new->inode;
2221 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2222 	int slot;
2223 	int ret;
2224 	u64 extent_offset;
2225 	u64 num_bytes;
2226 
2227 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2228 	    inum == btrfs_ino(BTRFS_I(inode)))
2229 		return 0;
2230 
2231 	key.objectid = root_id;
2232 	key.type = BTRFS_ROOT_ITEM_KEY;
2233 	key.offset = (u64)-1;
2234 
2235 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2236 	if (IS_ERR(root)) {
2237 		if (PTR_ERR(root) == -ENOENT)
2238 			return 0;
2239 		WARN_ON(1);
2240 		btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2241 			 inum, offset, root_id);
2242 		return PTR_ERR(root);
2243 	}
2244 
2245 	key.objectid = inum;
2246 	key.type = BTRFS_EXTENT_DATA_KEY;
2247 	if (offset > (u64)-1 << 32)
2248 		key.offset = 0;
2249 	else
2250 		key.offset = offset;
2251 
2252 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2253 	if (WARN_ON(ret < 0))
2254 		return ret;
2255 	ret = 0;
2256 
2257 	while (1) {
2258 		cond_resched();
2259 
2260 		leaf = path->nodes[0];
2261 		slot = path->slots[0];
2262 
2263 		if (slot >= btrfs_header_nritems(leaf)) {
2264 			ret = btrfs_next_leaf(root, path);
2265 			if (ret < 0) {
2266 				goto out;
2267 			} else if (ret > 0) {
2268 				ret = 0;
2269 				goto out;
2270 			}
2271 			continue;
2272 		}
2273 
2274 		path->slots[0]++;
2275 
2276 		btrfs_item_key_to_cpu(leaf, &key, slot);
2277 
2278 		if (key.objectid > inum)
2279 			goto out;
2280 
2281 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2282 			continue;
2283 
2284 		extent = btrfs_item_ptr(leaf, slot,
2285 					struct btrfs_file_extent_item);
2286 
2287 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2288 			continue;
2289 
2290 		/*
2291 		 * 'offset' refers to the exact key.offset,
2292 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2293 		 * (key.offset - extent_offset).
2294 		 */
2295 		if (key.offset != offset)
2296 			continue;
2297 
2298 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2299 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2300 
2301 		if (extent_offset >= old->extent_offset + old->offset +
2302 		    old->len || extent_offset + num_bytes <=
2303 		    old->extent_offset + old->offset)
2304 			continue;
2305 		break;
2306 	}
2307 
2308 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2309 	if (!backref) {
2310 		ret = -ENOENT;
2311 		goto out;
2312 	}
2313 
2314 	backref->root_id = root_id;
2315 	backref->inum = inum;
2316 	backref->file_pos = offset;
2317 	backref->num_bytes = num_bytes;
2318 	backref->extent_offset = extent_offset;
2319 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2320 	backref->old = old;
2321 	backref_insert(&new->root, backref);
2322 	old->count++;
2323 out:
2324 	btrfs_release_path(path);
2325 	WARN_ON(ret);
2326 	return ret;
2327 }
2328 
2329 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2330 				   struct new_sa_defrag_extent *new)
2331 {
2332 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2333 	struct old_sa_defrag_extent *old, *tmp;
2334 	int ret;
2335 
2336 	new->path = path;
2337 
2338 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2339 		ret = iterate_inodes_from_logical(old->bytenr +
2340 						  old->extent_offset, fs_info,
2341 						  path, record_one_backref,
2342 						  old);
2343 		if (ret < 0 && ret != -ENOENT)
2344 			return false;
2345 
2346 		/* no backref to be processed for this extent */
2347 		if (!old->count) {
2348 			list_del(&old->list);
2349 			kfree(old);
2350 		}
2351 	}
2352 
2353 	if (list_empty(&new->head))
2354 		return false;
2355 
2356 	return true;
2357 }
2358 
2359 static int relink_is_mergable(struct extent_buffer *leaf,
2360 			      struct btrfs_file_extent_item *fi,
2361 			      struct new_sa_defrag_extent *new)
2362 {
2363 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2364 		return 0;
2365 
2366 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2367 		return 0;
2368 
2369 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2370 		return 0;
2371 
2372 	if (btrfs_file_extent_encryption(leaf, fi) ||
2373 	    btrfs_file_extent_other_encoding(leaf, fi))
2374 		return 0;
2375 
2376 	return 1;
2377 }
2378 
2379 /*
2380  * Note the backref might has changed, and in this case we just return 0.
2381  */
2382 static noinline int relink_extent_backref(struct btrfs_path *path,
2383 				 struct sa_defrag_extent_backref *prev,
2384 				 struct sa_defrag_extent_backref *backref)
2385 {
2386 	struct btrfs_file_extent_item *extent;
2387 	struct btrfs_file_extent_item *item;
2388 	struct btrfs_ordered_extent *ordered;
2389 	struct btrfs_trans_handle *trans;
2390 	struct btrfs_root *root;
2391 	struct btrfs_key key;
2392 	struct extent_buffer *leaf;
2393 	struct old_sa_defrag_extent *old = backref->old;
2394 	struct new_sa_defrag_extent *new = old->new;
2395 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2396 	struct inode *inode;
2397 	struct extent_state *cached = NULL;
2398 	int ret = 0;
2399 	u64 start;
2400 	u64 len;
2401 	u64 lock_start;
2402 	u64 lock_end;
2403 	bool merge = false;
2404 	int index;
2405 
2406 	if (prev && prev->root_id == backref->root_id &&
2407 	    prev->inum == backref->inum &&
2408 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2409 		merge = true;
2410 
2411 	/* step 1: get root */
2412 	key.objectid = backref->root_id;
2413 	key.type = BTRFS_ROOT_ITEM_KEY;
2414 	key.offset = (u64)-1;
2415 
2416 	index = srcu_read_lock(&fs_info->subvol_srcu);
2417 
2418 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2419 	if (IS_ERR(root)) {
2420 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2421 		if (PTR_ERR(root) == -ENOENT)
2422 			return 0;
2423 		return PTR_ERR(root);
2424 	}
2425 
2426 	if (btrfs_root_readonly(root)) {
2427 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2428 		return 0;
2429 	}
2430 
2431 	/* step 2: get inode */
2432 	key.objectid = backref->inum;
2433 	key.type = BTRFS_INODE_ITEM_KEY;
2434 	key.offset = 0;
2435 
2436 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2437 	if (IS_ERR(inode)) {
2438 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2439 		return 0;
2440 	}
2441 
2442 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2443 
2444 	/* step 3: relink backref */
2445 	lock_start = backref->file_pos;
2446 	lock_end = backref->file_pos + backref->num_bytes - 1;
2447 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2448 			 &cached);
2449 
2450 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2451 	if (ordered) {
2452 		btrfs_put_ordered_extent(ordered);
2453 		goto out_unlock;
2454 	}
2455 
2456 	trans = btrfs_join_transaction(root);
2457 	if (IS_ERR(trans)) {
2458 		ret = PTR_ERR(trans);
2459 		goto out_unlock;
2460 	}
2461 
2462 	key.objectid = backref->inum;
2463 	key.type = BTRFS_EXTENT_DATA_KEY;
2464 	key.offset = backref->file_pos;
2465 
2466 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2467 	if (ret < 0) {
2468 		goto out_free_path;
2469 	} else if (ret > 0) {
2470 		ret = 0;
2471 		goto out_free_path;
2472 	}
2473 
2474 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2475 				struct btrfs_file_extent_item);
2476 
2477 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2478 	    backref->generation)
2479 		goto out_free_path;
2480 
2481 	btrfs_release_path(path);
2482 
2483 	start = backref->file_pos;
2484 	if (backref->extent_offset < old->extent_offset + old->offset)
2485 		start += old->extent_offset + old->offset -
2486 			 backref->extent_offset;
2487 
2488 	len = min(backref->extent_offset + backref->num_bytes,
2489 		  old->extent_offset + old->offset + old->len);
2490 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2491 
2492 	ret = btrfs_drop_extents(trans, root, inode, start,
2493 				 start + len, 1);
2494 	if (ret)
2495 		goto out_free_path;
2496 again:
2497 	key.objectid = btrfs_ino(BTRFS_I(inode));
2498 	key.type = BTRFS_EXTENT_DATA_KEY;
2499 	key.offset = start;
2500 
2501 	path->leave_spinning = 1;
2502 	if (merge) {
2503 		struct btrfs_file_extent_item *fi;
2504 		u64 extent_len;
2505 		struct btrfs_key found_key;
2506 
2507 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2508 		if (ret < 0)
2509 			goto out_free_path;
2510 
2511 		path->slots[0]--;
2512 		leaf = path->nodes[0];
2513 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2514 
2515 		fi = btrfs_item_ptr(leaf, path->slots[0],
2516 				    struct btrfs_file_extent_item);
2517 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2518 
2519 		if (extent_len + found_key.offset == start &&
2520 		    relink_is_mergable(leaf, fi, new)) {
2521 			btrfs_set_file_extent_num_bytes(leaf, fi,
2522 							extent_len + len);
2523 			btrfs_mark_buffer_dirty(leaf);
2524 			inode_add_bytes(inode, len);
2525 
2526 			ret = 1;
2527 			goto out_free_path;
2528 		} else {
2529 			merge = false;
2530 			btrfs_release_path(path);
2531 			goto again;
2532 		}
2533 	}
2534 
2535 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2536 					sizeof(*extent));
2537 	if (ret) {
2538 		btrfs_abort_transaction(trans, ret);
2539 		goto out_free_path;
2540 	}
2541 
2542 	leaf = path->nodes[0];
2543 	item = btrfs_item_ptr(leaf, path->slots[0],
2544 				struct btrfs_file_extent_item);
2545 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2546 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2547 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2548 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2549 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2550 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2551 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2552 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2553 	btrfs_set_file_extent_encryption(leaf, item, 0);
2554 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2555 
2556 	btrfs_mark_buffer_dirty(leaf);
2557 	inode_add_bytes(inode, len);
2558 	btrfs_release_path(path);
2559 
2560 	ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2561 			new->disk_len, 0,
2562 			backref->root_id, backref->inum,
2563 			new->file_pos);	/* start - extent_offset */
2564 	if (ret) {
2565 		btrfs_abort_transaction(trans, ret);
2566 		goto out_free_path;
2567 	}
2568 
2569 	ret = 1;
2570 out_free_path:
2571 	btrfs_release_path(path);
2572 	path->leave_spinning = 0;
2573 	btrfs_end_transaction(trans);
2574 out_unlock:
2575 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2576 			     &cached, GFP_NOFS);
2577 	iput(inode);
2578 	return ret;
2579 }
2580 
2581 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2582 {
2583 	struct old_sa_defrag_extent *old, *tmp;
2584 
2585 	if (!new)
2586 		return;
2587 
2588 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2589 		kfree(old);
2590 	}
2591 	kfree(new);
2592 }
2593 
2594 static void relink_file_extents(struct new_sa_defrag_extent *new)
2595 {
2596 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2597 	struct btrfs_path *path;
2598 	struct sa_defrag_extent_backref *backref;
2599 	struct sa_defrag_extent_backref *prev = NULL;
2600 	struct inode *inode;
2601 	struct btrfs_root *root;
2602 	struct rb_node *node;
2603 	int ret;
2604 
2605 	inode = new->inode;
2606 	root = BTRFS_I(inode)->root;
2607 
2608 	path = btrfs_alloc_path();
2609 	if (!path)
2610 		return;
2611 
2612 	if (!record_extent_backrefs(path, new)) {
2613 		btrfs_free_path(path);
2614 		goto out;
2615 	}
2616 	btrfs_release_path(path);
2617 
2618 	while (1) {
2619 		node = rb_first(&new->root);
2620 		if (!node)
2621 			break;
2622 		rb_erase(node, &new->root);
2623 
2624 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2625 
2626 		ret = relink_extent_backref(path, prev, backref);
2627 		WARN_ON(ret < 0);
2628 
2629 		kfree(prev);
2630 
2631 		if (ret == 1)
2632 			prev = backref;
2633 		else
2634 			prev = NULL;
2635 		cond_resched();
2636 	}
2637 	kfree(prev);
2638 
2639 	btrfs_free_path(path);
2640 out:
2641 	free_sa_defrag_extent(new);
2642 
2643 	atomic_dec(&fs_info->defrag_running);
2644 	wake_up(&fs_info->transaction_wait);
2645 }
2646 
2647 static struct new_sa_defrag_extent *
2648 record_old_file_extents(struct inode *inode,
2649 			struct btrfs_ordered_extent *ordered)
2650 {
2651 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2652 	struct btrfs_root *root = BTRFS_I(inode)->root;
2653 	struct btrfs_path *path;
2654 	struct btrfs_key key;
2655 	struct old_sa_defrag_extent *old;
2656 	struct new_sa_defrag_extent *new;
2657 	int ret;
2658 
2659 	new = kmalloc(sizeof(*new), GFP_NOFS);
2660 	if (!new)
2661 		return NULL;
2662 
2663 	new->inode = inode;
2664 	new->file_pos = ordered->file_offset;
2665 	new->len = ordered->len;
2666 	new->bytenr = ordered->start;
2667 	new->disk_len = ordered->disk_len;
2668 	new->compress_type = ordered->compress_type;
2669 	new->root = RB_ROOT;
2670 	INIT_LIST_HEAD(&new->head);
2671 
2672 	path = btrfs_alloc_path();
2673 	if (!path)
2674 		goto out_kfree;
2675 
2676 	key.objectid = btrfs_ino(BTRFS_I(inode));
2677 	key.type = BTRFS_EXTENT_DATA_KEY;
2678 	key.offset = new->file_pos;
2679 
2680 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2681 	if (ret < 0)
2682 		goto out_free_path;
2683 	if (ret > 0 && path->slots[0] > 0)
2684 		path->slots[0]--;
2685 
2686 	/* find out all the old extents for the file range */
2687 	while (1) {
2688 		struct btrfs_file_extent_item *extent;
2689 		struct extent_buffer *l;
2690 		int slot;
2691 		u64 num_bytes;
2692 		u64 offset;
2693 		u64 end;
2694 		u64 disk_bytenr;
2695 		u64 extent_offset;
2696 
2697 		l = path->nodes[0];
2698 		slot = path->slots[0];
2699 
2700 		if (slot >= btrfs_header_nritems(l)) {
2701 			ret = btrfs_next_leaf(root, path);
2702 			if (ret < 0)
2703 				goto out_free_path;
2704 			else if (ret > 0)
2705 				break;
2706 			continue;
2707 		}
2708 
2709 		btrfs_item_key_to_cpu(l, &key, slot);
2710 
2711 		if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2712 			break;
2713 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2714 			break;
2715 		if (key.offset >= new->file_pos + new->len)
2716 			break;
2717 
2718 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2719 
2720 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2721 		if (key.offset + num_bytes < new->file_pos)
2722 			goto next;
2723 
2724 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2725 		if (!disk_bytenr)
2726 			goto next;
2727 
2728 		extent_offset = btrfs_file_extent_offset(l, extent);
2729 
2730 		old = kmalloc(sizeof(*old), GFP_NOFS);
2731 		if (!old)
2732 			goto out_free_path;
2733 
2734 		offset = max(new->file_pos, key.offset);
2735 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2736 
2737 		old->bytenr = disk_bytenr;
2738 		old->extent_offset = extent_offset;
2739 		old->offset = offset - key.offset;
2740 		old->len = end - offset;
2741 		old->new = new;
2742 		old->count = 0;
2743 		list_add_tail(&old->list, &new->head);
2744 next:
2745 		path->slots[0]++;
2746 		cond_resched();
2747 	}
2748 
2749 	btrfs_free_path(path);
2750 	atomic_inc(&fs_info->defrag_running);
2751 
2752 	return new;
2753 
2754 out_free_path:
2755 	btrfs_free_path(path);
2756 out_kfree:
2757 	free_sa_defrag_extent(new);
2758 	return NULL;
2759 }
2760 
2761 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2762 					 u64 start, u64 len)
2763 {
2764 	struct btrfs_block_group_cache *cache;
2765 
2766 	cache = btrfs_lookup_block_group(fs_info, start);
2767 	ASSERT(cache);
2768 
2769 	spin_lock(&cache->lock);
2770 	cache->delalloc_bytes -= len;
2771 	spin_unlock(&cache->lock);
2772 
2773 	btrfs_put_block_group(cache);
2774 }
2775 
2776 /* as ordered data IO finishes, this gets called so we can finish
2777  * an ordered extent if the range of bytes in the file it covers are
2778  * fully written.
2779  */
2780 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2781 {
2782 	struct inode *inode = ordered_extent->inode;
2783 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2784 	struct btrfs_root *root = BTRFS_I(inode)->root;
2785 	struct btrfs_trans_handle *trans = NULL;
2786 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2787 	struct extent_state *cached_state = NULL;
2788 	struct new_sa_defrag_extent *new = NULL;
2789 	int compress_type = 0;
2790 	int ret = 0;
2791 	u64 logical_len = ordered_extent->len;
2792 	bool nolock;
2793 	bool truncated = false;
2794 
2795 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2796 
2797 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2798 		ret = -EIO;
2799 		goto out;
2800 	}
2801 
2802 	btrfs_free_io_failure_record(BTRFS_I(inode),
2803 			ordered_extent->file_offset,
2804 			ordered_extent->file_offset +
2805 			ordered_extent->len - 1);
2806 
2807 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2808 		truncated = true;
2809 		logical_len = ordered_extent->truncated_len;
2810 		/* Truncated the entire extent, don't bother adding */
2811 		if (!logical_len)
2812 			goto out;
2813 	}
2814 
2815 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2816 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2817 
2818 		/*
2819 		 * For mwrite(mmap + memset to write) case, we still reserve
2820 		 * space for NOCOW range.
2821 		 * As NOCOW won't cause a new delayed ref, just free the space
2822 		 */
2823 		btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2824 				       ordered_extent->len);
2825 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2826 		if (nolock)
2827 			trans = btrfs_join_transaction_nolock(root);
2828 		else
2829 			trans = btrfs_join_transaction(root);
2830 		if (IS_ERR(trans)) {
2831 			ret = PTR_ERR(trans);
2832 			trans = NULL;
2833 			goto out;
2834 		}
2835 		trans->block_rsv = &fs_info->delalloc_block_rsv;
2836 		ret = btrfs_update_inode_fallback(trans, root, inode);
2837 		if (ret) /* -ENOMEM or corruption */
2838 			btrfs_abort_transaction(trans, ret);
2839 		goto out;
2840 	}
2841 
2842 	lock_extent_bits(io_tree, ordered_extent->file_offset,
2843 			 ordered_extent->file_offset + ordered_extent->len - 1,
2844 			 &cached_state);
2845 
2846 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
2847 			ordered_extent->file_offset + ordered_extent->len - 1,
2848 			EXTENT_DEFRAG, 1, cached_state);
2849 	if (ret) {
2850 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2851 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2852 			/* the inode is shared */
2853 			new = record_old_file_extents(inode, ordered_extent);
2854 
2855 		clear_extent_bit(io_tree, ordered_extent->file_offset,
2856 			ordered_extent->file_offset + ordered_extent->len - 1,
2857 			EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2858 	}
2859 
2860 	if (nolock)
2861 		trans = btrfs_join_transaction_nolock(root);
2862 	else
2863 		trans = btrfs_join_transaction(root);
2864 	if (IS_ERR(trans)) {
2865 		ret = PTR_ERR(trans);
2866 		trans = NULL;
2867 		goto out_unlock;
2868 	}
2869 
2870 	trans->block_rsv = &fs_info->delalloc_block_rsv;
2871 
2872 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2873 		compress_type = ordered_extent->compress_type;
2874 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2875 		BUG_ON(compress_type);
2876 		ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2877 						ordered_extent->file_offset,
2878 						ordered_extent->file_offset +
2879 						logical_len);
2880 	} else {
2881 		BUG_ON(root == fs_info->tree_root);
2882 		ret = insert_reserved_file_extent(trans, inode,
2883 						ordered_extent->file_offset,
2884 						ordered_extent->start,
2885 						ordered_extent->disk_len,
2886 						logical_len, logical_len,
2887 						compress_type, 0, 0,
2888 						BTRFS_FILE_EXTENT_REG);
2889 		if (!ret)
2890 			btrfs_release_delalloc_bytes(fs_info,
2891 						     ordered_extent->start,
2892 						     ordered_extent->disk_len);
2893 	}
2894 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2895 			   ordered_extent->file_offset, ordered_extent->len,
2896 			   trans->transid);
2897 	if (ret < 0) {
2898 		btrfs_abort_transaction(trans, ret);
2899 		goto out_unlock;
2900 	}
2901 
2902 	add_pending_csums(trans, inode, &ordered_extent->list);
2903 
2904 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2905 	ret = btrfs_update_inode_fallback(trans, root, inode);
2906 	if (ret) { /* -ENOMEM or corruption */
2907 		btrfs_abort_transaction(trans, ret);
2908 		goto out_unlock;
2909 	}
2910 	ret = 0;
2911 out_unlock:
2912 	unlock_extent_cached(io_tree, ordered_extent->file_offset,
2913 			     ordered_extent->file_offset +
2914 			     ordered_extent->len - 1, &cached_state, GFP_NOFS);
2915 out:
2916 	if (root != fs_info->tree_root)
2917 		btrfs_delalloc_release_metadata(BTRFS_I(inode),
2918 				ordered_extent->len);
2919 	if (trans)
2920 		btrfs_end_transaction(trans);
2921 
2922 	if (ret || truncated) {
2923 		u64 start, end;
2924 
2925 		if (truncated)
2926 			start = ordered_extent->file_offset + logical_len;
2927 		else
2928 			start = ordered_extent->file_offset;
2929 		end = ordered_extent->file_offset + ordered_extent->len - 1;
2930 		clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2931 
2932 		/* Drop the cache for the part of the extent we didn't write. */
2933 		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2934 
2935 		/*
2936 		 * If the ordered extent had an IOERR or something else went
2937 		 * wrong we need to return the space for this ordered extent
2938 		 * back to the allocator.  We only free the extent in the
2939 		 * truncated case if we didn't write out the extent at all.
2940 		 */
2941 		if ((ret || !logical_len) &&
2942 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2943 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2944 			btrfs_free_reserved_extent(fs_info,
2945 						   ordered_extent->start,
2946 						   ordered_extent->disk_len, 1);
2947 	}
2948 
2949 
2950 	/*
2951 	 * This needs to be done to make sure anybody waiting knows we are done
2952 	 * updating everything for this ordered extent.
2953 	 */
2954 	btrfs_remove_ordered_extent(inode, ordered_extent);
2955 
2956 	/* for snapshot-aware defrag */
2957 	if (new) {
2958 		if (ret) {
2959 			free_sa_defrag_extent(new);
2960 			atomic_dec(&fs_info->defrag_running);
2961 		} else {
2962 			relink_file_extents(new);
2963 		}
2964 	}
2965 
2966 	/* once for us */
2967 	btrfs_put_ordered_extent(ordered_extent);
2968 	/* once for the tree */
2969 	btrfs_put_ordered_extent(ordered_extent);
2970 
2971 	return ret;
2972 }
2973 
2974 static void finish_ordered_fn(struct btrfs_work *work)
2975 {
2976 	struct btrfs_ordered_extent *ordered_extent;
2977 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2978 	btrfs_finish_ordered_io(ordered_extent);
2979 }
2980 
2981 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2982 				struct extent_state *state, int uptodate)
2983 {
2984 	struct inode *inode = page->mapping->host;
2985 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2986 	struct btrfs_ordered_extent *ordered_extent = NULL;
2987 	struct btrfs_workqueue *wq;
2988 	btrfs_work_func_t func;
2989 
2990 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2991 
2992 	ClearPagePrivate2(page);
2993 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2994 					    end - start + 1, uptodate))
2995 		return;
2996 
2997 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
2998 		wq = fs_info->endio_freespace_worker;
2999 		func = btrfs_freespace_write_helper;
3000 	} else {
3001 		wq = fs_info->endio_write_workers;
3002 		func = btrfs_endio_write_helper;
3003 	}
3004 
3005 	btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3006 			NULL);
3007 	btrfs_queue_work(wq, &ordered_extent->work);
3008 }
3009 
3010 static int __readpage_endio_check(struct inode *inode,
3011 				  struct btrfs_io_bio *io_bio,
3012 				  int icsum, struct page *page,
3013 				  int pgoff, u64 start, size_t len)
3014 {
3015 	char *kaddr;
3016 	u32 csum_expected;
3017 	u32 csum = ~(u32)0;
3018 
3019 	csum_expected = *(((u32 *)io_bio->csum) + icsum);
3020 
3021 	kaddr = kmap_atomic(page);
3022 	csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3023 	btrfs_csum_final(csum, (u8 *)&csum);
3024 	if (csum != csum_expected)
3025 		goto zeroit;
3026 
3027 	kunmap_atomic(kaddr);
3028 	return 0;
3029 zeroit:
3030 	btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3031 				    io_bio->mirror_num);
3032 	memset(kaddr + pgoff, 1, len);
3033 	flush_dcache_page(page);
3034 	kunmap_atomic(kaddr);
3035 	if (csum_expected == 0)
3036 		return 0;
3037 	return -EIO;
3038 }
3039 
3040 /*
3041  * when reads are done, we need to check csums to verify the data is correct
3042  * if there's a match, we allow the bio to finish.  If not, the code in
3043  * extent_io.c will try to find good copies for us.
3044  */
3045 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3046 				      u64 phy_offset, struct page *page,
3047 				      u64 start, u64 end, int mirror)
3048 {
3049 	size_t offset = start - page_offset(page);
3050 	struct inode *inode = page->mapping->host;
3051 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3052 	struct btrfs_root *root = BTRFS_I(inode)->root;
3053 
3054 	if (PageChecked(page)) {
3055 		ClearPageChecked(page);
3056 		return 0;
3057 	}
3058 
3059 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3060 		return 0;
3061 
3062 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3063 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3064 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3065 		return 0;
3066 	}
3067 
3068 	phy_offset >>= inode->i_sb->s_blocksize_bits;
3069 	return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3070 				      start, (size_t)(end - start + 1));
3071 }
3072 
3073 void btrfs_add_delayed_iput(struct inode *inode)
3074 {
3075 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3076 	struct btrfs_inode *binode = BTRFS_I(inode);
3077 
3078 	if (atomic_add_unless(&inode->i_count, -1, 1))
3079 		return;
3080 
3081 	spin_lock(&fs_info->delayed_iput_lock);
3082 	if (binode->delayed_iput_count == 0) {
3083 		ASSERT(list_empty(&binode->delayed_iput));
3084 		list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3085 	} else {
3086 		binode->delayed_iput_count++;
3087 	}
3088 	spin_unlock(&fs_info->delayed_iput_lock);
3089 }
3090 
3091 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3092 {
3093 
3094 	spin_lock(&fs_info->delayed_iput_lock);
3095 	while (!list_empty(&fs_info->delayed_iputs)) {
3096 		struct btrfs_inode *inode;
3097 
3098 		inode = list_first_entry(&fs_info->delayed_iputs,
3099 				struct btrfs_inode, delayed_iput);
3100 		if (inode->delayed_iput_count) {
3101 			inode->delayed_iput_count--;
3102 			list_move_tail(&inode->delayed_iput,
3103 					&fs_info->delayed_iputs);
3104 		} else {
3105 			list_del_init(&inode->delayed_iput);
3106 		}
3107 		spin_unlock(&fs_info->delayed_iput_lock);
3108 		iput(&inode->vfs_inode);
3109 		spin_lock(&fs_info->delayed_iput_lock);
3110 	}
3111 	spin_unlock(&fs_info->delayed_iput_lock);
3112 }
3113 
3114 /*
3115  * This is called in transaction commit time. If there are no orphan
3116  * files in the subvolume, it removes orphan item and frees block_rsv
3117  * structure.
3118  */
3119 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3120 			      struct btrfs_root *root)
3121 {
3122 	struct btrfs_fs_info *fs_info = root->fs_info;
3123 	struct btrfs_block_rsv *block_rsv;
3124 	int ret;
3125 
3126 	if (atomic_read(&root->orphan_inodes) ||
3127 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3128 		return;
3129 
3130 	spin_lock(&root->orphan_lock);
3131 	if (atomic_read(&root->orphan_inodes)) {
3132 		spin_unlock(&root->orphan_lock);
3133 		return;
3134 	}
3135 
3136 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3137 		spin_unlock(&root->orphan_lock);
3138 		return;
3139 	}
3140 
3141 	block_rsv = root->orphan_block_rsv;
3142 	root->orphan_block_rsv = NULL;
3143 	spin_unlock(&root->orphan_lock);
3144 
3145 	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3146 	    btrfs_root_refs(&root->root_item) > 0) {
3147 		ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3148 					    root->root_key.objectid);
3149 		if (ret)
3150 			btrfs_abort_transaction(trans, ret);
3151 		else
3152 			clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3153 				  &root->state);
3154 	}
3155 
3156 	if (block_rsv) {
3157 		WARN_ON(block_rsv->size > 0);
3158 		btrfs_free_block_rsv(fs_info, block_rsv);
3159 	}
3160 }
3161 
3162 /*
3163  * This creates an orphan entry for the given inode in case something goes
3164  * wrong in the middle of an unlink/truncate.
3165  *
3166  * NOTE: caller of this function should reserve 5 units of metadata for
3167  *	 this function.
3168  */
3169 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3170 		struct btrfs_inode *inode)
3171 {
3172 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3173 	struct btrfs_root *root = inode->root;
3174 	struct btrfs_block_rsv *block_rsv = NULL;
3175 	int reserve = 0;
3176 	int insert = 0;
3177 	int ret;
3178 
3179 	if (!root->orphan_block_rsv) {
3180 		block_rsv = btrfs_alloc_block_rsv(fs_info,
3181 						  BTRFS_BLOCK_RSV_TEMP);
3182 		if (!block_rsv)
3183 			return -ENOMEM;
3184 	}
3185 
3186 	spin_lock(&root->orphan_lock);
3187 	if (!root->orphan_block_rsv) {
3188 		root->orphan_block_rsv = block_rsv;
3189 	} else if (block_rsv) {
3190 		btrfs_free_block_rsv(fs_info, block_rsv);
3191 		block_rsv = NULL;
3192 	}
3193 
3194 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3195 			      &inode->runtime_flags)) {
3196 #if 0
3197 		/*
3198 		 * For proper ENOSPC handling, we should do orphan
3199 		 * cleanup when mounting. But this introduces backward
3200 		 * compatibility issue.
3201 		 */
3202 		if (!xchg(&root->orphan_item_inserted, 1))
3203 			insert = 2;
3204 		else
3205 			insert = 1;
3206 #endif
3207 		insert = 1;
3208 		atomic_inc(&root->orphan_inodes);
3209 	}
3210 
3211 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3212 			      &inode->runtime_flags))
3213 		reserve = 1;
3214 	spin_unlock(&root->orphan_lock);
3215 
3216 	/* grab metadata reservation from transaction handle */
3217 	if (reserve) {
3218 		ret = btrfs_orphan_reserve_metadata(trans, inode);
3219 		ASSERT(!ret);
3220 		if (ret) {
3221 			atomic_dec(&root->orphan_inodes);
3222 			clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3223 				  &inode->runtime_flags);
3224 			if (insert)
3225 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3226 					  &inode->runtime_flags);
3227 			return ret;
3228 		}
3229 	}
3230 
3231 	/* insert an orphan item to track this unlinked/truncated file */
3232 	if (insert >= 1) {
3233 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3234 		if (ret) {
3235 			atomic_dec(&root->orphan_inodes);
3236 			if (reserve) {
3237 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3238 					  &inode->runtime_flags);
3239 				btrfs_orphan_release_metadata(inode);
3240 			}
3241 			if (ret != -EEXIST) {
3242 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3243 					  &inode->runtime_flags);
3244 				btrfs_abort_transaction(trans, ret);
3245 				return ret;
3246 			}
3247 		}
3248 		ret = 0;
3249 	}
3250 
3251 	/* insert an orphan item to track subvolume contains orphan files */
3252 	if (insert >= 2) {
3253 		ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3254 					       root->root_key.objectid);
3255 		if (ret && ret != -EEXIST) {
3256 			btrfs_abort_transaction(trans, ret);
3257 			return ret;
3258 		}
3259 	}
3260 	return 0;
3261 }
3262 
3263 /*
3264  * We have done the truncate/delete so we can go ahead and remove the orphan
3265  * item for this particular inode.
3266  */
3267 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3268 			    struct btrfs_inode *inode)
3269 {
3270 	struct btrfs_root *root = inode->root;
3271 	int delete_item = 0;
3272 	int release_rsv = 0;
3273 	int ret = 0;
3274 
3275 	spin_lock(&root->orphan_lock);
3276 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3277 			       &inode->runtime_flags))
3278 		delete_item = 1;
3279 
3280 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3281 			       &inode->runtime_flags))
3282 		release_rsv = 1;
3283 	spin_unlock(&root->orphan_lock);
3284 
3285 	if (delete_item) {
3286 		atomic_dec(&root->orphan_inodes);
3287 		if (trans)
3288 			ret = btrfs_del_orphan_item(trans, root,
3289 						    btrfs_ino(inode));
3290 	}
3291 
3292 	if (release_rsv)
3293 		btrfs_orphan_release_metadata(inode);
3294 
3295 	return ret;
3296 }
3297 
3298 /*
3299  * this cleans up any orphans that may be left on the list from the last use
3300  * of this root.
3301  */
3302 int btrfs_orphan_cleanup(struct btrfs_root *root)
3303 {
3304 	struct btrfs_fs_info *fs_info = root->fs_info;
3305 	struct btrfs_path *path;
3306 	struct extent_buffer *leaf;
3307 	struct btrfs_key key, found_key;
3308 	struct btrfs_trans_handle *trans;
3309 	struct inode *inode;
3310 	u64 last_objectid = 0;
3311 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3312 
3313 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3314 		return 0;
3315 
3316 	path = btrfs_alloc_path();
3317 	if (!path) {
3318 		ret = -ENOMEM;
3319 		goto out;
3320 	}
3321 	path->reada = READA_BACK;
3322 
3323 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3324 	key.type = BTRFS_ORPHAN_ITEM_KEY;
3325 	key.offset = (u64)-1;
3326 
3327 	while (1) {
3328 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3329 		if (ret < 0)
3330 			goto out;
3331 
3332 		/*
3333 		 * if ret == 0 means we found what we were searching for, which
3334 		 * is weird, but possible, so only screw with path if we didn't
3335 		 * find the key and see if we have stuff that matches
3336 		 */
3337 		if (ret > 0) {
3338 			ret = 0;
3339 			if (path->slots[0] == 0)
3340 				break;
3341 			path->slots[0]--;
3342 		}
3343 
3344 		/* pull out the item */
3345 		leaf = path->nodes[0];
3346 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3347 
3348 		/* make sure the item matches what we want */
3349 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3350 			break;
3351 		if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3352 			break;
3353 
3354 		/* release the path since we're done with it */
3355 		btrfs_release_path(path);
3356 
3357 		/*
3358 		 * this is where we are basically btrfs_lookup, without the
3359 		 * crossing root thing.  we store the inode number in the
3360 		 * offset of the orphan item.
3361 		 */
3362 
3363 		if (found_key.offset == last_objectid) {
3364 			btrfs_err(fs_info,
3365 				  "Error removing orphan entry, stopping orphan cleanup");
3366 			ret = -EINVAL;
3367 			goto out;
3368 		}
3369 
3370 		last_objectid = found_key.offset;
3371 
3372 		found_key.objectid = found_key.offset;
3373 		found_key.type = BTRFS_INODE_ITEM_KEY;
3374 		found_key.offset = 0;
3375 		inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3376 		ret = PTR_ERR_OR_ZERO(inode);
3377 		if (ret && ret != -ENOENT)
3378 			goto out;
3379 
3380 		if (ret == -ENOENT && root == fs_info->tree_root) {
3381 			struct btrfs_root *dead_root;
3382 			struct btrfs_fs_info *fs_info = root->fs_info;
3383 			int is_dead_root = 0;
3384 
3385 			/*
3386 			 * this is an orphan in the tree root. Currently these
3387 			 * could come from 2 sources:
3388 			 *  a) a snapshot deletion in progress
3389 			 *  b) a free space cache inode
3390 			 * We need to distinguish those two, as the snapshot
3391 			 * orphan must not get deleted.
3392 			 * find_dead_roots already ran before us, so if this
3393 			 * is a snapshot deletion, we should find the root
3394 			 * in the dead_roots list
3395 			 */
3396 			spin_lock(&fs_info->trans_lock);
3397 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3398 					    root_list) {
3399 				if (dead_root->root_key.objectid ==
3400 				    found_key.objectid) {
3401 					is_dead_root = 1;
3402 					break;
3403 				}
3404 			}
3405 			spin_unlock(&fs_info->trans_lock);
3406 			if (is_dead_root) {
3407 				/* prevent this orphan from being found again */
3408 				key.offset = found_key.objectid - 1;
3409 				continue;
3410 			}
3411 		}
3412 		/*
3413 		 * Inode is already gone but the orphan item is still there,
3414 		 * kill the orphan item.
3415 		 */
3416 		if (ret == -ENOENT) {
3417 			trans = btrfs_start_transaction(root, 1);
3418 			if (IS_ERR(trans)) {
3419 				ret = PTR_ERR(trans);
3420 				goto out;
3421 			}
3422 			btrfs_debug(fs_info, "auto deleting %Lu",
3423 				    found_key.objectid);
3424 			ret = btrfs_del_orphan_item(trans, root,
3425 						    found_key.objectid);
3426 			btrfs_end_transaction(trans);
3427 			if (ret)
3428 				goto out;
3429 			continue;
3430 		}
3431 
3432 		/*
3433 		 * add this inode to the orphan list so btrfs_orphan_del does
3434 		 * the proper thing when we hit it
3435 		 */
3436 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3437 			&BTRFS_I(inode)->runtime_flags);
3438 		atomic_inc(&root->orphan_inodes);
3439 
3440 		/* if we have links, this was a truncate, lets do that */
3441 		if (inode->i_nlink) {
3442 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3443 				iput(inode);
3444 				continue;
3445 			}
3446 			nr_truncate++;
3447 
3448 			/* 1 for the orphan item deletion. */
3449 			trans = btrfs_start_transaction(root, 1);
3450 			if (IS_ERR(trans)) {
3451 				iput(inode);
3452 				ret = PTR_ERR(trans);
3453 				goto out;
3454 			}
3455 			ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3456 			btrfs_end_transaction(trans);
3457 			if (ret) {
3458 				iput(inode);
3459 				goto out;
3460 			}
3461 
3462 			ret = btrfs_truncate(inode);
3463 			if (ret)
3464 				btrfs_orphan_del(NULL, BTRFS_I(inode));
3465 		} else {
3466 			nr_unlink++;
3467 		}
3468 
3469 		/* this will do delete_inode and everything for us */
3470 		iput(inode);
3471 		if (ret)
3472 			goto out;
3473 	}
3474 	/* release the path since we're done with it */
3475 	btrfs_release_path(path);
3476 
3477 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3478 
3479 	if (root->orphan_block_rsv)
3480 		btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3481 					(u64)-1);
3482 
3483 	if (root->orphan_block_rsv ||
3484 	    test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3485 		trans = btrfs_join_transaction(root);
3486 		if (!IS_ERR(trans))
3487 			btrfs_end_transaction(trans);
3488 	}
3489 
3490 	if (nr_unlink)
3491 		btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3492 	if (nr_truncate)
3493 		btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3494 
3495 out:
3496 	if (ret)
3497 		btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3498 	btrfs_free_path(path);
3499 	return ret;
3500 }
3501 
3502 /*
3503  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3504  * don't find any xattrs, we know there can't be any acls.
3505  *
3506  * slot is the slot the inode is in, objectid is the objectid of the inode
3507  */
3508 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3509 					  int slot, u64 objectid,
3510 					  int *first_xattr_slot)
3511 {
3512 	u32 nritems = btrfs_header_nritems(leaf);
3513 	struct btrfs_key found_key;
3514 	static u64 xattr_access = 0;
3515 	static u64 xattr_default = 0;
3516 	int scanned = 0;
3517 
3518 	if (!xattr_access) {
3519 		xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3520 					strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3521 		xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3522 					strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3523 	}
3524 
3525 	slot++;
3526 	*first_xattr_slot = -1;
3527 	while (slot < nritems) {
3528 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3529 
3530 		/* we found a different objectid, there must not be acls */
3531 		if (found_key.objectid != objectid)
3532 			return 0;
3533 
3534 		/* we found an xattr, assume we've got an acl */
3535 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3536 			if (*first_xattr_slot == -1)
3537 				*first_xattr_slot = slot;
3538 			if (found_key.offset == xattr_access ||
3539 			    found_key.offset == xattr_default)
3540 				return 1;
3541 		}
3542 
3543 		/*
3544 		 * we found a key greater than an xattr key, there can't
3545 		 * be any acls later on
3546 		 */
3547 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3548 			return 0;
3549 
3550 		slot++;
3551 		scanned++;
3552 
3553 		/*
3554 		 * it goes inode, inode backrefs, xattrs, extents,
3555 		 * so if there are a ton of hard links to an inode there can
3556 		 * be a lot of backrefs.  Don't waste time searching too hard,
3557 		 * this is just an optimization
3558 		 */
3559 		if (scanned >= 8)
3560 			break;
3561 	}
3562 	/* we hit the end of the leaf before we found an xattr or
3563 	 * something larger than an xattr.  We have to assume the inode
3564 	 * has acls
3565 	 */
3566 	if (*first_xattr_slot == -1)
3567 		*first_xattr_slot = slot;
3568 	return 1;
3569 }
3570 
3571 /*
3572  * read an inode from the btree into the in-memory inode
3573  */
3574 static int btrfs_read_locked_inode(struct inode *inode)
3575 {
3576 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3577 	struct btrfs_path *path;
3578 	struct extent_buffer *leaf;
3579 	struct btrfs_inode_item *inode_item;
3580 	struct btrfs_root *root = BTRFS_I(inode)->root;
3581 	struct btrfs_key location;
3582 	unsigned long ptr;
3583 	int maybe_acls;
3584 	u32 rdev;
3585 	int ret;
3586 	bool filled = false;
3587 	int first_xattr_slot;
3588 
3589 	ret = btrfs_fill_inode(inode, &rdev);
3590 	if (!ret)
3591 		filled = true;
3592 
3593 	path = btrfs_alloc_path();
3594 	if (!path) {
3595 		ret = -ENOMEM;
3596 		goto make_bad;
3597 	}
3598 
3599 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3600 
3601 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3602 	if (ret) {
3603 		if (ret > 0)
3604 			ret = -ENOENT;
3605 		goto make_bad;
3606 	}
3607 
3608 	leaf = path->nodes[0];
3609 
3610 	if (filled)
3611 		goto cache_index;
3612 
3613 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3614 				    struct btrfs_inode_item);
3615 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3616 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3617 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3618 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3619 	btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3620 
3621 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3622 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3623 
3624 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3625 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3626 
3627 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3628 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3629 
3630 	BTRFS_I(inode)->i_otime.tv_sec =
3631 		btrfs_timespec_sec(leaf, &inode_item->otime);
3632 	BTRFS_I(inode)->i_otime.tv_nsec =
3633 		btrfs_timespec_nsec(leaf, &inode_item->otime);
3634 
3635 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3636 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3637 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3638 
3639 	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3640 	inode->i_generation = BTRFS_I(inode)->generation;
3641 	inode->i_rdev = 0;
3642 	rdev = btrfs_inode_rdev(leaf, inode_item);
3643 
3644 	BTRFS_I(inode)->index_cnt = (u64)-1;
3645 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646 
3647 cache_index:
3648 	/*
3649 	 * If we were modified in the current generation and evicted from memory
3650 	 * and then re-read we need to do a full sync since we don't have any
3651 	 * idea about which extents were modified before we were evicted from
3652 	 * cache.
3653 	 *
3654 	 * This is required for both inode re-read from disk and delayed inode
3655 	 * in delayed_nodes_tree.
3656 	 */
3657 	if (BTRFS_I(inode)->last_trans == fs_info->generation)
3658 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3659 			&BTRFS_I(inode)->runtime_flags);
3660 
3661 	/*
3662 	 * We don't persist the id of the transaction where an unlink operation
3663 	 * against the inode was last made. So here we assume the inode might
3664 	 * have been evicted, and therefore the exact value of last_unlink_trans
3665 	 * lost, and set it to last_trans to avoid metadata inconsistencies
3666 	 * between the inode and its parent if the inode is fsync'ed and the log
3667 	 * replayed. For example, in the scenario:
3668 	 *
3669 	 * touch mydir/foo
3670 	 * ln mydir/foo mydir/bar
3671 	 * sync
3672 	 * unlink mydir/bar
3673 	 * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3674 	 * xfs_io -c fsync mydir/foo
3675 	 * <power failure>
3676 	 * mount fs, triggers fsync log replay
3677 	 *
3678 	 * We must make sure that when we fsync our inode foo we also log its
3679 	 * parent inode, otherwise after log replay the parent still has the
3680 	 * dentry with the "bar" name but our inode foo has a link count of 1
3681 	 * and doesn't have an inode ref with the name "bar" anymore.
3682 	 *
3683 	 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3684 	 * but it guarantees correctness at the expense of occasional full
3685 	 * transaction commits on fsync if our inode is a directory, or if our
3686 	 * inode is not a directory, logging its parent unnecessarily.
3687 	 */
3688 	BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3689 
3690 	path->slots[0]++;
3691 	if (inode->i_nlink != 1 ||
3692 	    path->slots[0] >= btrfs_header_nritems(leaf))
3693 		goto cache_acl;
3694 
3695 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3696 	if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3697 		goto cache_acl;
3698 
3699 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3700 	if (location.type == BTRFS_INODE_REF_KEY) {
3701 		struct btrfs_inode_ref *ref;
3702 
3703 		ref = (struct btrfs_inode_ref *)ptr;
3704 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3705 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3706 		struct btrfs_inode_extref *extref;
3707 
3708 		extref = (struct btrfs_inode_extref *)ptr;
3709 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3710 								     extref);
3711 	}
3712 cache_acl:
3713 	/*
3714 	 * try to precache a NULL acl entry for files that don't have
3715 	 * any xattrs or acls
3716 	 */
3717 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3718 			btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3719 	if (first_xattr_slot != -1) {
3720 		path->slots[0] = first_xattr_slot;
3721 		ret = btrfs_load_inode_props(inode, path);
3722 		if (ret)
3723 			btrfs_err(fs_info,
3724 				  "error loading props for ino %llu (root %llu): %d",
3725 				  btrfs_ino(BTRFS_I(inode)),
3726 				  root->root_key.objectid, ret);
3727 	}
3728 	btrfs_free_path(path);
3729 
3730 	if (!maybe_acls)
3731 		cache_no_acl(inode);
3732 
3733 	switch (inode->i_mode & S_IFMT) {
3734 	case S_IFREG:
3735 		inode->i_mapping->a_ops = &btrfs_aops;
3736 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3737 		inode->i_fop = &btrfs_file_operations;
3738 		inode->i_op = &btrfs_file_inode_operations;
3739 		break;
3740 	case S_IFDIR:
3741 		inode->i_fop = &btrfs_dir_file_operations;
3742 		inode->i_op = &btrfs_dir_inode_operations;
3743 		break;
3744 	case S_IFLNK:
3745 		inode->i_op = &btrfs_symlink_inode_operations;
3746 		inode_nohighmem(inode);
3747 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3748 		break;
3749 	default:
3750 		inode->i_op = &btrfs_special_inode_operations;
3751 		init_special_inode(inode, inode->i_mode, rdev);
3752 		break;
3753 	}
3754 
3755 	btrfs_update_iflags(inode);
3756 	return 0;
3757 
3758 make_bad:
3759 	btrfs_free_path(path);
3760 	make_bad_inode(inode);
3761 	return ret;
3762 }
3763 
3764 /*
3765  * given a leaf and an inode, copy the inode fields into the leaf
3766  */
3767 static void fill_inode_item(struct btrfs_trans_handle *trans,
3768 			    struct extent_buffer *leaf,
3769 			    struct btrfs_inode_item *item,
3770 			    struct inode *inode)
3771 {
3772 	struct btrfs_map_token token;
3773 
3774 	btrfs_init_map_token(&token);
3775 
3776 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3777 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3778 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3779 				   &token);
3780 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3781 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3782 
3783 	btrfs_set_token_timespec_sec(leaf, &item->atime,
3784 				     inode->i_atime.tv_sec, &token);
3785 	btrfs_set_token_timespec_nsec(leaf, &item->atime,
3786 				      inode->i_atime.tv_nsec, &token);
3787 
3788 	btrfs_set_token_timespec_sec(leaf, &item->mtime,
3789 				     inode->i_mtime.tv_sec, &token);
3790 	btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3791 				      inode->i_mtime.tv_nsec, &token);
3792 
3793 	btrfs_set_token_timespec_sec(leaf, &item->ctime,
3794 				     inode->i_ctime.tv_sec, &token);
3795 	btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3796 				      inode->i_ctime.tv_nsec, &token);
3797 
3798 	btrfs_set_token_timespec_sec(leaf, &item->otime,
3799 				     BTRFS_I(inode)->i_otime.tv_sec, &token);
3800 	btrfs_set_token_timespec_nsec(leaf, &item->otime,
3801 				      BTRFS_I(inode)->i_otime.tv_nsec, &token);
3802 
3803 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3804 				     &token);
3805 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3806 					 &token);
3807 	btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3808 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3809 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3810 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3811 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3812 }
3813 
3814 /*
3815  * copy everything in the in-memory inode into the btree.
3816  */
3817 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3818 				struct btrfs_root *root, struct inode *inode)
3819 {
3820 	struct btrfs_inode_item *inode_item;
3821 	struct btrfs_path *path;
3822 	struct extent_buffer *leaf;
3823 	int ret;
3824 
3825 	path = btrfs_alloc_path();
3826 	if (!path)
3827 		return -ENOMEM;
3828 
3829 	path->leave_spinning = 1;
3830 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3831 				 1);
3832 	if (ret) {
3833 		if (ret > 0)
3834 			ret = -ENOENT;
3835 		goto failed;
3836 	}
3837 
3838 	leaf = path->nodes[0];
3839 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3840 				    struct btrfs_inode_item);
3841 
3842 	fill_inode_item(trans, leaf, inode_item, inode);
3843 	btrfs_mark_buffer_dirty(leaf);
3844 	btrfs_set_inode_last_trans(trans, inode);
3845 	ret = 0;
3846 failed:
3847 	btrfs_free_path(path);
3848 	return ret;
3849 }
3850 
3851 /*
3852  * copy everything in the in-memory inode into the btree.
3853  */
3854 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3855 				struct btrfs_root *root, struct inode *inode)
3856 {
3857 	struct btrfs_fs_info *fs_info = root->fs_info;
3858 	int ret;
3859 
3860 	/*
3861 	 * If the inode is a free space inode, we can deadlock during commit
3862 	 * if we put it into the delayed code.
3863 	 *
3864 	 * The data relocation inode should also be directly updated
3865 	 * without delay
3866 	 */
3867 	if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3868 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3869 	    && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3870 		btrfs_update_root_times(trans, root);
3871 
3872 		ret = btrfs_delayed_update_inode(trans, root, inode);
3873 		if (!ret)
3874 			btrfs_set_inode_last_trans(trans, inode);
3875 		return ret;
3876 	}
3877 
3878 	return btrfs_update_inode_item(trans, root, inode);
3879 }
3880 
3881 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3882 					 struct btrfs_root *root,
3883 					 struct inode *inode)
3884 {
3885 	int ret;
3886 
3887 	ret = btrfs_update_inode(trans, root, inode);
3888 	if (ret == -ENOSPC)
3889 		return btrfs_update_inode_item(trans, root, inode);
3890 	return ret;
3891 }
3892 
3893 /*
3894  * unlink helper that gets used here in inode.c and in the tree logging
3895  * recovery code.  It remove a link in a directory with a given name, and
3896  * also drops the back refs in the inode to the directory
3897  */
3898 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3899 				struct btrfs_root *root,
3900 				struct btrfs_inode *dir,
3901 				struct btrfs_inode *inode,
3902 				const char *name, int name_len)
3903 {
3904 	struct btrfs_fs_info *fs_info = root->fs_info;
3905 	struct btrfs_path *path;
3906 	int ret = 0;
3907 	struct extent_buffer *leaf;
3908 	struct btrfs_dir_item *di;
3909 	struct btrfs_key key;
3910 	u64 index;
3911 	u64 ino = btrfs_ino(inode);
3912 	u64 dir_ino = btrfs_ino(dir);
3913 
3914 	path = btrfs_alloc_path();
3915 	if (!path) {
3916 		ret = -ENOMEM;
3917 		goto out;
3918 	}
3919 
3920 	path->leave_spinning = 1;
3921 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3922 				    name, name_len, -1);
3923 	if (IS_ERR(di)) {
3924 		ret = PTR_ERR(di);
3925 		goto err;
3926 	}
3927 	if (!di) {
3928 		ret = -ENOENT;
3929 		goto err;
3930 	}
3931 	leaf = path->nodes[0];
3932 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3933 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3934 	if (ret)
3935 		goto err;
3936 	btrfs_release_path(path);
3937 
3938 	/*
3939 	 * If we don't have dir index, we have to get it by looking up
3940 	 * the inode ref, since we get the inode ref, remove it directly,
3941 	 * it is unnecessary to do delayed deletion.
3942 	 *
3943 	 * But if we have dir index, needn't search inode ref to get it.
3944 	 * Since the inode ref is close to the inode item, it is better
3945 	 * that we delay to delete it, and just do this deletion when
3946 	 * we update the inode item.
3947 	 */
3948 	if (inode->dir_index) {
3949 		ret = btrfs_delayed_delete_inode_ref(inode);
3950 		if (!ret) {
3951 			index = inode->dir_index;
3952 			goto skip_backref;
3953 		}
3954 	}
3955 
3956 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3957 				  dir_ino, &index);
3958 	if (ret) {
3959 		btrfs_info(fs_info,
3960 			"failed to delete reference to %.*s, inode %llu parent %llu",
3961 			name_len, name, ino, dir_ino);
3962 		btrfs_abort_transaction(trans, ret);
3963 		goto err;
3964 	}
3965 skip_backref:
3966 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
3967 	if (ret) {
3968 		btrfs_abort_transaction(trans, ret);
3969 		goto err;
3970 	}
3971 
3972 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3973 			dir_ino);
3974 	if (ret != 0 && ret != -ENOENT) {
3975 		btrfs_abort_transaction(trans, ret);
3976 		goto err;
3977 	}
3978 
3979 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3980 			index);
3981 	if (ret == -ENOENT)
3982 		ret = 0;
3983 	else if (ret)
3984 		btrfs_abort_transaction(trans, ret);
3985 err:
3986 	btrfs_free_path(path);
3987 	if (ret)
3988 		goto out;
3989 
3990 	btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3991 	inode_inc_iversion(&inode->vfs_inode);
3992 	inode_inc_iversion(&dir->vfs_inode);
3993 	inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3994 		dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3995 	ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3996 out:
3997 	return ret;
3998 }
3999 
4000 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4001 		       struct btrfs_root *root,
4002 		       struct btrfs_inode *dir, struct btrfs_inode *inode,
4003 		       const char *name, int name_len)
4004 {
4005 	int ret;
4006 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4007 	if (!ret) {
4008 		drop_nlink(&inode->vfs_inode);
4009 		ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4010 	}
4011 	return ret;
4012 }
4013 
4014 /*
4015  * helper to start transaction for unlink and rmdir.
4016  *
4017  * unlink and rmdir are special in btrfs, they do not always free space, so
4018  * if we cannot make our reservations the normal way try and see if there is
4019  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4020  * allow the unlink to occur.
4021  */
4022 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4023 {
4024 	struct btrfs_root *root = BTRFS_I(dir)->root;
4025 
4026 	/*
4027 	 * 1 for the possible orphan item
4028 	 * 1 for the dir item
4029 	 * 1 for the dir index
4030 	 * 1 for the inode ref
4031 	 * 1 for the inode
4032 	 */
4033 	return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4034 }
4035 
4036 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4037 {
4038 	struct btrfs_root *root = BTRFS_I(dir)->root;
4039 	struct btrfs_trans_handle *trans;
4040 	struct inode *inode = d_inode(dentry);
4041 	int ret;
4042 
4043 	trans = __unlink_start_trans(dir);
4044 	if (IS_ERR(trans))
4045 		return PTR_ERR(trans);
4046 
4047 	btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4048 			0);
4049 
4050 	ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4051 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4052 			dentry->d_name.len);
4053 	if (ret)
4054 		goto out;
4055 
4056 	if (inode->i_nlink == 0) {
4057 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4058 		if (ret)
4059 			goto out;
4060 	}
4061 
4062 out:
4063 	btrfs_end_transaction(trans);
4064 	btrfs_btree_balance_dirty(root->fs_info);
4065 	return ret;
4066 }
4067 
4068 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4069 			struct btrfs_root *root,
4070 			struct inode *dir, u64 objectid,
4071 			const char *name, int name_len)
4072 {
4073 	struct btrfs_fs_info *fs_info = root->fs_info;
4074 	struct btrfs_path *path;
4075 	struct extent_buffer *leaf;
4076 	struct btrfs_dir_item *di;
4077 	struct btrfs_key key;
4078 	u64 index;
4079 	int ret;
4080 	u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4081 
4082 	path = btrfs_alloc_path();
4083 	if (!path)
4084 		return -ENOMEM;
4085 
4086 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4087 				   name, name_len, -1);
4088 	if (IS_ERR_OR_NULL(di)) {
4089 		if (!di)
4090 			ret = -ENOENT;
4091 		else
4092 			ret = PTR_ERR(di);
4093 		goto out;
4094 	}
4095 
4096 	leaf = path->nodes[0];
4097 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4098 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4099 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4100 	if (ret) {
4101 		btrfs_abort_transaction(trans, ret);
4102 		goto out;
4103 	}
4104 	btrfs_release_path(path);
4105 
4106 	ret = btrfs_del_root_ref(trans, fs_info, objectid,
4107 				 root->root_key.objectid, dir_ino,
4108 				 &index, name, name_len);
4109 	if (ret < 0) {
4110 		if (ret != -ENOENT) {
4111 			btrfs_abort_transaction(trans, ret);
4112 			goto out;
4113 		}
4114 		di = btrfs_search_dir_index_item(root, path, dir_ino,
4115 						 name, name_len);
4116 		if (IS_ERR_OR_NULL(di)) {
4117 			if (!di)
4118 				ret = -ENOENT;
4119 			else
4120 				ret = PTR_ERR(di);
4121 			btrfs_abort_transaction(trans, ret);
4122 			goto out;
4123 		}
4124 
4125 		leaf = path->nodes[0];
4126 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4127 		btrfs_release_path(path);
4128 		index = key.offset;
4129 	}
4130 	btrfs_release_path(path);
4131 
4132 	ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4133 	if (ret) {
4134 		btrfs_abort_transaction(trans, ret);
4135 		goto out;
4136 	}
4137 
4138 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4139 	inode_inc_iversion(dir);
4140 	dir->i_mtime = dir->i_ctime = current_time(dir);
4141 	ret = btrfs_update_inode_fallback(trans, root, dir);
4142 	if (ret)
4143 		btrfs_abort_transaction(trans, ret);
4144 out:
4145 	btrfs_free_path(path);
4146 	return ret;
4147 }
4148 
4149 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4150 {
4151 	struct inode *inode = d_inode(dentry);
4152 	int err = 0;
4153 	struct btrfs_root *root = BTRFS_I(dir)->root;
4154 	struct btrfs_trans_handle *trans;
4155 	u64 last_unlink_trans;
4156 
4157 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4158 		return -ENOTEMPTY;
4159 	if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4160 		return -EPERM;
4161 
4162 	trans = __unlink_start_trans(dir);
4163 	if (IS_ERR(trans))
4164 		return PTR_ERR(trans);
4165 
4166 	if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4167 		err = btrfs_unlink_subvol(trans, root, dir,
4168 					  BTRFS_I(inode)->location.objectid,
4169 					  dentry->d_name.name,
4170 					  dentry->d_name.len);
4171 		goto out;
4172 	}
4173 
4174 	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4175 	if (err)
4176 		goto out;
4177 
4178 	last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4179 
4180 	/* now the directory is empty */
4181 	err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4182 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4183 			dentry->d_name.len);
4184 	if (!err) {
4185 		btrfs_i_size_write(BTRFS_I(inode), 0);
4186 		/*
4187 		 * Propagate the last_unlink_trans value of the deleted dir to
4188 		 * its parent directory. This is to prevent an unrecoverable
4189 		 * log tree in the case we do something like this:
4190 		 * 1) create dir foo
4191 		 * 2) create snapshot under dir foo
4192 		 * 3) delete the snapshot
4193 		 * 4) rmdir foo
4194 		 * 5) mkdir foo
4195 		 * 6) fsync foo or some file inside foo
4196 		 */
4197 		if (last_unlink_trans >= trans->transid)
4198 			BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4199 	}
4200 out:
4201 	btrfs_end_transaction(trans);
4202 	btrfs_btree_balance_dirty(root->fs_info);
4203 
4204 	return err;
4205 }
4206 
4207 static int truncate_space_check(struct btrfs_trans_handle *trans,
4208 				struct btrfs_root *root,
4209 				u64 bytes_deleted)
4210 {
4211 	struct btrfs_fs_info *fs_info = root->fs_info;
4212 	int ret;
4213 
4214 	/*
4215 	 * This is only used to apply pressure to the enospc system, we don't
4216 	 * intend to use this reservation at all.
4217 	 */
4218 	bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4219 	bytes_deleted *= fs_info->nodesize;
4220 	ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4221 				  bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4222 	if (!ret) {
4223 		trace_btrfs_space_reservation(fs_info, "transaction",
4224 					      trans->transid,
4225 					      bytes_deleted, 1);
4226 		trans->bytes_reserved += bytes_deleted;
4227 	}
4228 	return ret;
4229 
4230 }
4231 
4232 static int truncate_inline_extent(struct inode *inode,
4233 				  struct btrfs_path *path,
4234 				  struct btrfs_key *found_key,
4235 				  const u64 item_end,
4236 				  const u64 new_size)
4237 {
4238 	struct extent_buffer *leaf = path->nodes[0];
4239 	int slot = path->slots[0];
4240 	struct btrfs_file_extent_item *fi;
4241 	u32 size = (u32)(new_size - found_key->offset);
4242 	struct btrfs_root *root = BTRFS_I(inode)->root;
4243 
4244 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4245 
4246 	if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4247 		loff_t offset = new_size;
4248 		loff_t page_end = ALIGN(offset, PAGE_SIZE);
4249 
4250 		/*
4251 		 * Zero out the remaining of the last page of our inline extent,
4252 		 * instead of directly truncating our inline extent here - that
4253 		 * would be much more complex (decompressing all the data, then
4254 		 * compressing the truncated data, which might be bigger than
4255 		 * the size of the inline extent, resize the extent, etc).
4256 		 * We release the path because to get the page we might need to
4257 		 * read the extent item from disk (data not in the page cache).
4258 		 */
4259 		btrfs_release_path(path);
4260 		return btrfs_truncate_block(inode, offset, page_end - offset,
4261 					0);
4262 	}
4263 
4264 	btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4265 	size = btrfs_file_extent_calc_inline_size(size);
4266 	btrfs_truncate_item(root->fs_info, path, size, 1);
4267 
4268 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4269 		inode_sub_bytes(inode, item_end + 1 - new_size);
4270 
4271 	return 0;
4272 }
4273 
4274 /*
4275  * this can truncate away extent items, csum items and directory items.
4276  * It starts at a high offset and removes keys until it can't find
4277  * any higher than new_size
4278  *
4279  * csum items that cross the new i_size are truncated to the new size
4280  * as well.
4281  *
4282  * min_type is the minimum key type to truncate down to.  If set to 0, this
4283  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4284  */
4285 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4286 			       struct btrfs_root *root,
4287 			       struct inode *inode,
4288 			       u64 new_size, u32 min_type)
4289 {
4290 	struct btrfs_fs_info *fs_info = root->fs_info;
4291 	struct btrfs_path *path;
4292 	struct extent_buffer *leaf;
4293 	struct btrfs_file_extent_item *fi;
4294 	struct btrfs_key key;
4295 	struct btrfs_key found_key;
4296 	u64 extent_start = 0;
4297 	u64 extent_num_bytes = 0;
4298 	u64 extent_offset = 0;
4299 	u64 item_end = 0;
4300 	u64 last_size = new_size;
4301 	u32 found_type = (u8)-1;
4302 	int found_extent;
4303 	int del_item;
4304 	int pending_del_nr = 0;
4305 	int pending_del_slot = 0;
4306 	int extent_type = -1;
4307 	int ret;
4308 	int err = 0;
4309 	u64 ino = btrfs_ino(BTRFS_I(inode));
4310 	u64 bytes_deleted = 0;
4311 	bool be_nice = 0;
4312 	bool should_throttle = 0;
4313 	bool should_end = 0;
4314 
4315 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4316 
4317 	/*
4318 	 * for non-free space inodes and ref cows, we want to back off from
4319 	 * time to time
4320 	 */
4321 	if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4322 	    test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4323 		be_nice = 1;
4324 
4325 	path = btrfs_alloc_path();
4326 	if (!path)
4327 		return -ENOMEM;
4328 	path->reada = READA_BACK;
4329 
4330 	/*
4331 	 * We want to drop from the next block forward in case this new size is
4332 	 * not block aligned since we will be keeping the last block of the
4333 	 * extent just the way it is.
4334 	 */
4335 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4336 	    root == fs_info->tree_root)
4337 		btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4338 					fs_info->sectorsize),
4339 					(u64)-1, 0);
4340 
4341 	/*
4342 	 * This function is also used to drop the items in the log tree before
4343 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4344 	 * it is used to drop the loged items. So we shouldn't kill the delayed
4345 	 * items.
4346 	 */
4347 	if (min_type == 0 && root == BTRFS_I(inode)->root)
4348 		btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4349 
4350 	key.objectid = ino;
4351 	key.offset = (u64)-1;
4352 	key.type = (u8)-1;
4353 
4354 search_again:
4355 	/*
4356 	 * with a 16K leaf size and 128MB extents, you can actually queue
4357 	 * up a huge file in a single leaf.  Most of the time that
4358 	 * bytes_deleted is > 0, it will be huge by the time we get here
4359 	 */
4360 	if (be_nice && bytes_deleted > SZ_32M) {
4361 		if (btrfs_should_end_transaction(trans)) {
4362 			err = -EAGAIN;
4363 			goto error;
4364 		}
4365 	}
4366 
4367 
4368 	path->leave_spinning = 1;
4369 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4370 	if (ret < 0) {
4371 		err = ret;
4372 		goto out;
4373 	}
4374 
4375 	if (ret > 0) {
4376 		/* there are no items in the tree for us to truncate, we're
4377 		 * done
4378 		 */
4379 		if (path->slots[0] == 0)
4380 			goto out;
4381 		path->slots[0]--;
4382 	}
4383 
4384 	while (1) {
4385 		fi = NULL;
4386 		leaf = path->nodes[0];
4387 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4388 		found_type = found_key.type;
4389 
4390 		if (found_key.objectid != ino)
4391 			break;
4392 
4393 		if (found_type < min_type)
4394 			break;
4395 
4396 		item_end = found_key.offset;
4397 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4398 			fi = btrfs_item_ptr(leaf, path->slots[0],
4399 					    struct btrfs_file_extent_item);
4400 			extent_type = btrfs_file_extent_type(leaf, fi);
4401 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4402 				item_end +=
4403 				    btrfs_file_extent_num_bytes(leaf, fi);
4404 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4405 				item_end += btrfs_file_extent_inline_len(leaf,
4406 							 path->slots[0], fi);
4407 			}
4408 			item_end--;
4409 		}
4410 		if (found_type > min_type) {
4411 			del_item = 1;
4412 		} else {
4413 			if (item_end < new_size)
4414 				break;
4415 			if (found_key.offset >= new_size)
4416 				del_item = 1;
4417 			else
4418 				del_item = 0;
4419 		}
4420 		found_extent = 0;
4421 		/* FIXME, shrink the extent if the ref count is only 1 */
4422 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4423 			goto delete;
4424 
4425 		if (del_item)
4426 			last_size = found_key.offset;
4427 		else
4428 			last_size = new_size;
4429 
4430 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4431 			u64 num_dec;
4432 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4433 			if (!del_item) {
4434 				u64 orig_num_bytes =
4435 					btrfs_file_extent_num_bytes(leaf, fi);
4436 				extent_num_bytes = ALIGN(new_size -
4437 						found_key.offset,
4438 						fs_info->sectorsize);
4439 				btrfs_set_file_extent_num_bytes(leaf, fi,
4440 							 extent_num_bytes);
4441 				num_dec = (orig_num_bytes -
4442 					   extent_num_bytes);
4443 				if (test_bit(BTRFS_ROOT_REF_COWS,
4444 					     &root->state) &&
4445 				    extent_start != 0)
4446 					inode_sub_bytes(inode, num_dec);
4447 				btrfs_mark_buffer_dirty(leaf);
4448 			} else {
4449 				extent_num_bytes =
4450 					btrfs_file_extent_disk_num_bytes(leaf,
4451 									 fi);
4452 				extent_offset = found_key.offset -
4453 					btrfs_file_extent_offset(leaf, fi);
4454 
4455 				/* FIXME blocksize != 4096 */
4456 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4457 				if (extent_start != 0) {
4458 					found_extent = 1;
4459 					if (test_bit(BTRFS_ROOT_REF_COWS,
4460 						     &root->state))
4461 						inode_sub_bytes(inode, num_dec);
4462 				}
4463 			}
4464 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4465 			/*
4466 			 * we can't truncate inline items that have had
4467 			 * special encodings
4468 			 */
4469 			if (!del_item &&
4470 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4471 			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4472 
4473 				/*
4474 				 * Need to release path in order to truncate a
4475 				 * compressed extent. So delete any accumulated
4476 				 * extent items so far.
4477 				 */
4478 				if (btrfs_file_extent_compression(leaf, fi) !=
4479 				    BTRFS_COMPRESS_NONE && pending_del_nr) {
4480 					err = btrfs_del_items(trans, root, path,
4481 							      pending_del_slot,
4482 							      pending_del_nr);
4483 					if (err) {
4484 						btrfs_abort_transaction(trans,
4485 									err);
4486 						goto error;
4487 					}
4488 					pending_del_nr = 0;
4489 				}
4490 
4491 				err = truncate_inline_extent(inode, path,
4492 							     &found_key,
4493 							     item_end,
4494 							     new_size);
4495 				if (err) {
4496 					btrfs_abort_transaction(trans, err);
4497 					goto error;
4498 				}
4499 			} else if (test_bit(BTRFS_ROOT_REF_COWS,
4500 					    &root->state)) {
4501 				inode_sub_bytes(inode, item_end + 1 - new_size);
4502 			}
4503 		}
4504 delete:
4505 		if (del_item) {
4506 			if (!pending_del_nr) {
4507 				/* no pending yet, add ourselves */
4508 				pending_del_slot = path->slots[0];
4509 				pending_del_nr = 1;
4510 			} else if (pending_del_nr &&
4511 				   path->slots[0] + 1 == pending_del_slot) {
4512 				/* hop on the pending chunk */
4513 				pending_del_nr++;
4514 				pending_del_slot = path->slots[0];
4515 			} else {
4516 				BUG();
4517 			}
4518 		} else {
4519 			break;
4520 		}
4521 		should_throttle = 0;
4522 
4523 		if (found_extent &&
4524 		    (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4525 		     root == fs_info->tree_root)) {
4526 			btrfs_set_path_blocking(path);
4527 			bytes_deleted += extent_num_bytes;
4528 			ret = btrfs_free_extent(trans, fs_info, extent_start,
4529 						extent_num_bytes, 0,
4530 						btrfs_header_owner(leaf),
4531 						ino, extent_offset);
4532 			BUG_ON(ret);
4533 			if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4534 				btrfs_async_run_delayed_refs(fs_info,
4535 					trans->delayed_ref_updates * 2,
4536 					trans->transid, 0);
4537 			if (be_nice) {
4538 				if (truncate_space_check(trans, root,
4539 							 extent_num_bytes)) {
4540 					should_end = 1;
4541 				}
4542 				if (btrfs_should_throttle_delayed_refs(trans,
4543 								       fs_info))
4544 					should_throttle = 1;
4545 			}
4546 		}
4547 
4548 		if (found_type == BTRFS_INODE_ITEM_KEY)
4549 			break;
4550 
4551 		if (path->slots[0] == 0 ||
4552 		    path->slots[0] != pending_del_slot ||
4553 		    should_throttle || should_end) {
4554 			if (pending_del_nr) {
4555 				ret = btrfs_del_items(trans, root, path,
4556 						pending_del_slot,
4557 						pending_del_nr);
4558 				if (ret) {
4559 					btrfs_abort_transaction(trans, ret);
4560 					goto error;
4561 				}
4562 				pending_del_nr = 0;
4563 			}
4564 			btrfs_release_path(path);
4565 			if (should_throttle) {
4566 				unsigned long updates = trans->delayed_ref_updates;
4567 				if (updates) {
4568 					trans->delayed_ref_updates = 0;
4569 					ret = btrfs_run_delayed_refs(trans,
4570 								   fs_info,
4571 								   updates * 2);
4572 					if (ret && !err)
4573 						err = ret;
4574 				}
4575 			}
4576 			/*
4577 			 * if we failed to refill our space rsv, bail out
4578 			 * and let the transaction restart
4579 			 */
4580 			if (should_end) {
4581 				err = -EAGAIN;
4582 				goto error;
4583 			}
4584 			goto search_again;
4585 		} else {
4586 			path->slots[0]--;
4587 		}
4588 	}
4589 out:
4590 	if (pending_del_nr) {
4591 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4592 				      pending_del_nr);
4593 		if (ret)
4594 			btrfs_abort_transaction(trans, ret);
4595 	}
4596 error:
4597 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4598 		ASSERT(last_size >= new_size);
4599 		if (!err && last_size > new_size)
4600 			last_size = new_size;
4601 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4602 	}
4603 
4604 	btrfs_free_path(path);
4605 
4606 	if (err == 0) {
4607 		/* only inline file may have last_size != new_size */
4608 		if (new_size >= fs_info->sectorsize ||
4609 		    new_size > fs_info->max_inline)
4610 			ASSERT(last_size == new_size);
4611 	}
4612 
4613 	if (be_nice && bytes_deleted > SZ_32M) {
4614 		unsigned long updates = trans->delayed_ref_updates;
4615 		if (updates) {
4616 			trans->delayed_ref_updates = 0;
4617 			ret = btrfs_run_delayed_refs(trans, fs_info,
4618 						     updates * 2);
4619 			if (ret && !err)
4620 				err = ret;
4621 		}
4622 	}
4623 	return err;
4624 }
4625 
4626 /*
4627  * btrfs_truncate_block - read, zero a chunk and write a block
4628  * @inode - inode that we're zeroing
4629  * @from - the offset to start zeroing
4630  * @len - the length to zero, 0 to zero the entire range respective to the
4631  *	offset
4632  * @front - zero up to the offset instead of from the offset on
4633  *
4634  * This will find the block for the "from" offset and cow the block and zero the
4635  * part we want to zero.  This is used with truncate and hole punching.
4636  */
4637 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4638 			int front)
4639 {
4640 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4641 	struct address_space *mapping = inode->i_mapping;
4642 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4643 	struct btrfs_ordered_extent *ordered;
4644 	struct extent_state *cached_state = NULL;
4645 	char *kaddr;
4646 	u32 blocksize = fs_info->sectorsize;
4647 	pgoff_t index = from >> PAGE_SHIFT;
4648 	unsigned offset = from & (blocksize - 1);
4649 	struct page *page;
4650 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4651 	int ret = 0;
4652 	u64 block_start;
4653 	u64 block_end;
4654 
4655 	if ((offset & (blocksize - 1)) == 0 &&
4656 	    (!len || ((len & (blocksize - 1)) == 0)))
4657 		goto out;
4658 
4659 	ret = btrfs_delalloc_reserve_space(inode,
4660 			round_down(from, blocksize), blocksize);
4661 	if (ret)
4662 		goto out;
4663 
4664 again:
4665 	page = find_or_create_page(mapping, index, mask);
4666 	if (!page) {
4667 		btrfs_delalloc_release_space(inode,
4668 				round_down(from, blocksize),
4669 				blocksize);
4670 		ret = -ENOMEM;
4671 		goto out;
4672 	}
4673 
4674 	block_start = round_down(from, blocksize);
4675 	block_end = block_start + blocksize - 1;
4676 
4677 	if (!PageUptodate(page)) {
4678 		ret = btrfs_readpage(NULL, page);
4679 		lock_page(page);
4680 		if (page->mapping != mapping) {
4681 			unlock_page(page);
4682 			put_page(page);
4683 			goto again;
4684 		}
4685 		if (!PageUptodate(page)) {
4686 			ret = -EIO;
4687 			goto out_unlock;
4688 		}
4689 	}
4690 	wait_on_page_writeback(page);
4691 
4692 	lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4693 	set_page_extent_mapped(page);
4694 
4695 	ordered = btrfs_lookup_ordered_extent(inode, block_start);
4696 	if (ordered) {
4697 		unlock_extent_cached(io_tree, block_start, block_end,
4698 				     &cached_state, GFP_NOFS);
4699 		unlock_page(page);
4700 		put_page(page);
4701 		btrfs_start_ordered_extent(inode, ordered, 1);
4702 		btrfs_put_ordered_extent(ordered);
4703 		goto again;
4704 	}
4705 
4706 	clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4707 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4708 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4709 			  0, 0, &cached_state, GFP_NOFS);
4710 
4711 	ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4712 					&cached_state, 0);
4713 	if (ret) {
4714 		unlock_extent_cached(io_tree, block_start, block_end,
4715 				     &cached_state, GFP_NOFS);
4716 		goto out_unlock;
4717 	}
4718 
4719 	if (offset != blocksize) {
4720 		if (!len)
4721 			len = blocksize - offset;
4722 		kaddr = kmap(page);
4723 		if (front)
4724 			memset(kaddr + (block_start - page_offset(page)),
4725 				0, offset);
4726 		else
4727 			memset(kaddr + (block_start - page_offset(page)) +  offset,
4728 				0, len);
4729 		flush_dcache_page(page);
4730 		kunmap(page);
4731 	}
4732 	ClearPageChecked(page);
4733 	set_page_dirty(page);
4734 	unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4735 			     GFP_NOFS);
4736 
4737 out_unlock:
4738 	if (ret)
4739 		btrfs_delalloc_release_space(inode, block_start,
4740 					     blocksize);
4741 	unlock_page(page);
4742 	put_page(page);
4743 out:
4744 	return ret;
4745 }
4746 
4747 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4748 			     u64 offset, u64 len)
4749 {
4750 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4751 	struct btrfs_trans_handle *trans;
4752 	int ret;
4753 
4754 	/*
4755 	 * Still need to make sure the inode looks like it's been updated so
4756 	 * that any holes get logged if we fsync.
4757 	 */
4758 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4759 		BTRFS_I(inode)->last_trans = fs_info->generation;
4760 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4761 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4762 		return 0;
4763 	}
4764 
4765 	/*
4766 	 * 1 - for the one we're dropping
4767 	 * 1 - for the one we're adding
4768 	 * 1 - for updating the inode.
4769 	 */
4770 	trans = btrfs_start_transaction(root, 3);
4771 	if (IS_ERR(trans))
4772 		return PTR_ERR(trans);
4773 
4774 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4775 	if (ret) {
4776 		btrfs_abort_transaction(trans, ret);
4777 		btrfs_end_transaction(trans);
4778 		return ret;
4779 	}
4780 
4781 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4782 			offset, 0, 0, len, 0, len, 0, 0, 0);
4783 	if (ret)
4784 		btrfs_abort_transaction(trans, ret);
4785 	else
4786 		btrfs_update_inode(trans, root, inode);
4787 	btrfs_end_transaction(trans);
4788 	return ret;
4789 }
4790 
4791 /*
4792  * This function puts in dummy file extents for the area we're creating a hole
4793  * for.  So if we are truncating this file to a larger size we need to insert
4794  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4795  * the range between oldsize and size
4796  */
4797 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4798 {
4799 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4800 	struct btrfs_root *root = BTRFS_I(inode)->root;
4801 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4802 	struct extent_map *em = NULL;
4803 	struct extent_state *cached_state = NULL;
4804 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4805 	u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4806 	u64 block_end = ALIGN(size, fs_info->sectorsize);
4807 	u64 last_byte;
4808 	u64 cur_offset;
4809 	u64 hole_size;
4810 	int err = 0;
4811 
4812 	/*
4813 	 * If our size started in the middle of a block we need to zero out the
4814 	 * rest of the block before we expand the i_size, otherwise we could
4815 	 * expose stale data.
4816 	 */
4817 	err = btrfs_truncate_block(inode, oldsize, 0, 0);
4818 	if (err)
4819 		return err;
4820 
4821 	if (size <= hole_start)
4822 		return 0;
4823 
4824 	while (1) {
4825 		struct btrfs_ordered_extent *ordered;
4826 
4827 		lock_extent_bits(io_tree, hole_start, block_end - 1,
4828 				 &cached_state);
4829 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4830 						     block_end - hole_start);
4831 		if (!ordered)
4832 			break;
4833 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4834 				     &cached_state, GFP_NOFS);
4835 		btrfs_start_ordered_extent(inode, ordered, 1);
4836 		btrfs_put_ordered_extent(ordered);
4837 	}
4838 
4839 	cur_offset = hole_start;
4840 	while (1) {
4841 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4842 				block_end - cur_offset, 0);
4843 		if (IS_ERR(em)) {
4844 			err = PTR_ERR(em);
4845 			em = NULL;
4846 			break;
4847 		}
4848 		last_byte = min(extent_map_end(em), block_end);
4849 		last_byte = ALIGN(last_byte, fs_info->sectorsize);
4850 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4851 			struct extent_map *hole_em;
4852 			hole_size = last_byte - cur_offset;
4853 
4854 			err = maybe_insert_hole(root, inode, cur_offset,
4855 						hole_size);
4856 			if (err)
4857 				break;
4858 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4859 						cur_offset + hole_size - 1, 0);
4860 			hole_em = alloc_extent_map();
4861 			if (!hole_em) {
4862 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4863 					&BTRFS_I(inode)->runtime_flags);
4864 				goto next;
4865 			}
4866 			hole_em->start = cur_offset;
4867 			hole_em->len = hole_size;
4868 			hole_em->orig_start = cur_offset;
4869 
4870 			hole_em->block_start = EXTENT_MAP_HOLE;
4871 			hole_em->block_len = 0;
4872 			hole_em->orig_block_len = 0;
4873 			hole_em->ram_bytes = hole_size;
4874 			hole_em->bdev = fs_info->fs_devices->latest_bdev;
4875 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
4876 			hole_em->generation = fs_info->generation;
4877 
4878 			while (1) {
4879 				write_lock(&em_tree->lock);
4880 				err = add_extent_mapping(em_tree, hole_em, 1);
4881 				write_unlock(&em_tree->lock);
4882 				if (err != -EEXIST)
4883 					break;
4884 				btrfs_drop_extent_cache(BTRFS_I(inode),
4885 							cur_offset,
4886 							cur_offset +
4887 							hole_size - 1, 0);
4888 			}
4889 			free_extent_map(hole_em);
4890 		}
4891 next:
4892 		free_extent_map(em);
4893 		em = NULL;
4894 		cur_offset = last_byte;
4895 		if (cur_offset >= block_end)
4896 			break;
4897 	}
4898 	free_extent_map(em);
4899 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4900 			     GFP_NOFS);
4901 	return err;
4902 }
4903 
4904 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4905 {
4906 	struct btrfs_root *root = BTRFS_I(inode)->root;
4907 	struct btrfs_trans_handle *trans;
4908 	loff_t oldsize = i_size_read(inode);
4909 	loff_t newsize = attr->ia_size;
4910 	int mask = attr->ia_valid;
4911 	int ret;
4912 
4913 	/*
4914 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4915 	 * special case where we need to update the times despite not having
4916 	 * these flags set.  For all other operations the VFS set these flags
4917 	 * explicitly if it wants a timestamp update.
4918 	 */
4919 	if (newsize != oldsize) {
4920 		inode_inc_iversion(inode);
4921 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4922 			inode->i_ctime = inode->i_mtime =
4923 				current_time(inode);
4924 	}
4925 
4926 	if (newsize > oldsize) {
4927 		/*
4928 		 * Don't do an expanding truncate while snapshoting is ongoing.
4929 		 * This is to ensure the snapshot captures a fully consistent
4930 		 * state of this file - if the snapshot captures this expanding
4931 		 * truncation, it must capture all writes that happened before
4932 		 * this truncation.
4933 		 */
4934 		btrfs_wait_for_snapshot_creation(root);
4935 		ret = btrfs_cont_expand(inode, oldsize, newsize);
4936 		if (ret) {
4937 			btrfs_end_write_no_snapshoting(root);
4938 			return ret;
4939 		}
4940 
4941 		trans = btrfs_start_transaction(root, 1);
4942 		if (IS_ERR(trans)) {
4943 			btrfs_end_write_no_snapshoting(root);
4944 			return PTR_ERR(trans);
4945 		}
4946 
4947 		i_size_write(inode, newsize);
4948 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4949 		pagecache_isize_extended(inode, oldsize, newsize);
4950 		ret = btrfs_update_inode(trans, root, inode);
4951 		btrfs_end_write_no_snapshoting(root);
4952 		btrfs_end_transaction(trans);
4953 	} else {
4954 
4955 		/*
4956 		 * We're truncating a file that used to have good data down to
4957 		 * zero. Make sure it gets into the ordered flush list so that
4958 		 * any new writes get down to disk quickly.
4959 		 */
4960 		if (newsize == 0)
4961 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4962 				&BTRFS_I(inode)->runtime_flags);
4963 
4964 		/*
4965 		 * 1 for the orphan item we're going to add
4966 		 * 1 for the orphan item deletion.
4967 		 */
4968 		trans = btrfs_start_transaction(root, 2);
4969 		if (IS_ERR(trans))
4970 			return PTR_ERR(trans);
4971 
4972 		/*
4973 		 * We need to do this in case we fail at _any_ point during the
4974 		 * actual truncate.  Once we do the truncate_setsize we could
4975 		 * invalidate pages which forces any outstanding ordered io to
4976 		 * be instantly completed which will give us extents that need
4977 		 * to be truncated.  If we fail to get an orphan inode down we
4978 		 * could have left over extents that were never meant to live,
4979 		 * so we need to guarantee from this point on that everything
4980 		 * will be consistent.
4981 		 */
4982 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4983 		btrfs_end_transaction(trans);
4984 		if (ret)
4985 			return ret;
4986 
4987 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
4988 		truncate_setsize(inode, newsize);
4989 
4990 		/* Disable nonlocked read DIO to avoid the end less truncate */
4991 		btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
4992 		inode_dio_wait(inode);
4993 		btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
4994 
4995 		ret = btrfs_truncate(inode);
4996 		if (ret && inode->i_nlink) {
4997 			int err;
4998 
4999 			/* To get a stable disk_i_size */
5000 			err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5001 			if (err) {
5002 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5003 				return err;
5004 			}
5005 
5006 			/*
5007 			 * failed to truncate, disk_i_size is only adjusted down
5008 			 * as we remove extents, so it should represent the true
5009 			 * size of the inode, so reset the in memory size and
5010 			 * delete our orphan entry.
5011 			 */
5012 			trans = btrfs_join_transaction(root);
5013 			if (IS_ERR(trans)) {
5014 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5015 				return ret;
5016 			}
5017 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5018 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
5019 			if (err)
5020 				btrfs_abort_transaction(trans, err);
5021 			btrfs_end_transaction(trans);
5022 		}
5023 	}
5024 
5025 	return ret;
5026 }
5027 
5028 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5029 {
5030 	struct inode *inode = d_inode(dentry);
5031 	struct btrfs_root *root = BTRFS_I(inode)->root;
5032 	int err;
5033 
5034 	if (btrfs_root_readonly(root))
5035 		return -EROFS;
5036 
5037 	err = setattr_prepare(dentry, attr);
5038 	if (err)
5039 		return err;
5040 
5041 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5042 		err = btrfs_setsize(inode, attr);
5043 		if (err)
5044 			return err;
5045 	}
5046 
5047 	if (attr->ia_valid) {
5048 		setattr_copy(inode, attr);
5049 		inode_inc_iversion(inode);
5050 		err = btrfs_dirty_inode(inode);
5051 
5052 		if (!err && attr->ia_valid & ATTR_MODE)
5053 			err = posix_acl_chmod(inode, inode->i_mode);
5054 	}
5055 
5056 	return err;
5057 }
5058 
5059 /*
5060  * While truncating the inode pages during eviction, we get the VFS calling
5061  * btrfs_invalidatepage() against each page of the inode. This is slow because
5062  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5063  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5064  * extent_state structures over and over, wasting lots of time.
5065  *
5066  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5067  * those expensive operations on a per page basis and do only the ordered io
5068  * finishing, while we release here the extent_map and extent_state structures,
5069  * without the excessive merging and splitting.
5070  */
5071 static void evict_inode_truncate_pages(struct inode *inode)
5072 {
5073 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5074 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5075 	struct rb_node *node;
5076 
5077 	ASSERT(inode->i_state & I_FREEING);
5078 	truncate_inode_pages_final(&inode->i_data);
5079 
5080 	write_lock(&map_tree->lock);
5081 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
5082 		struct extent_map *em;
5083 
5084 		node = rb_first(&map_tree->map);
5085 		em = rb_entry(node, struct extent_map, rb_node);
5086 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5087 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5088 		remove_extent_mapping(map_tree, em);
5089 		free_extent_map(em);
5090 		if (need_resched()) {
5091 			write_unlock(&map_tree->lock);
5092 			cond_resched();
5093 			write_lock(&map_tree->lock);
5094 		}
5095 	}
5096 	write_unlock(&map_tree->lock);
5097 
5098 	/*
5099 	 * Keep looping until we have no more ranges in the io tree.
5100 	 * We can have ongoing bios started by readpages (called from readahead)
5101 	 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5102 	 * still in progress (unlocked the pages in the bio but did not yet
5103 	 * unlocked the ranges in the io tree). Therefore this means some
5104 	 * ranges can still be locked and eviction started because before
5105 	 * submitting those bios, which are executed by a separate task (work
5106 	 * queue kthread), inode references (inode->i_count) were not taken
5107 	 * (which would be dropped in the end io callback of each bio).
5108 	 * Therefore here we effectively end up waiting for those bios and
5109 	 * anyone else holding locked ranges without having bumped the inode's
5110 	 * reference count - if we don't do it, when they access the inode's
5111 	 * io_tree to unlock a range it may be too late, leading to an
5112 	 * use-after-free issue.
5113 	 */
5114 	spin_lock(&io_tree->lock);
5115 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
5116 		struct extent_state *state;
5117 		struct extent_state *cached_state = NULL;
5118 		u64 start;
5119 		u64 end;
5120 
5121 		node = rb_first(&io_tree->state);
5122 		state = rb_entry(node, struct extent_state, rb_node);
5123 		start = state->start;
5124 		end = state->end;
5125 		spin_unlock(&io_tree->lock);
5126 
5127 		lock_extent_bits(io_tree, start, end, &cached_state);
5128 
5129 		/*
5130 		 * If still has DELALLOC flag, the extent didn't reach disk,
5131 		 * and its reserved space won't be freed by delayed_ref.
5132 		 * So we need to free its reserved space here.
5133 		 * (Refer to comment in btrfs_invalidatepage, case 2)
5134 		 *
5135 		 * Note, end is the bytenr of last byte, so we need + 1 here.
5136 		 */
5137 		if (state->state & EXTENT_DELALLOC)
5138 			btrfs_qgroup_free_data(inode, start, end - start + 1);
5139 
5140 		clear_extent_bit(io_tree, start, end,
5141 				 EXTENT_LOCKED | EXTENT_DIRTY |
5142 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5143 				 EXTENT_DEFRAG, 1, 1,
5144 				 &cached_state, GFP_NOFS);
5145 
5146 		cond_resched();
5147 		spin_lock(&io_tree->lock);
5148 	}
5149 	spin_unlock(&io_tree->lock);
5150 }
5151 
5152 void btrfs_evict_inode(struct inode *inode)
5153 {
5154 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5155 	struct btrfs_trans_handle *trans;
5156 	struct btrfs_root *root = BTRFS_I(inode)->root;
5157 	struct btrfs_block_rsv *rsv, *global_rsv;
5158 	int steal_from_global = 0;
5159 	u64 min_size;
5160 	int ret;
5161 
5162 	trace_btrfs_inode_evict(inode);
5163 
5164 	if (!root) {
5165 		kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5166 		return;
5167 	}
5168 
5169 	min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5170 
5171 	evict_inode_truncate_pages(inode);
5172 
5173 	if (inode->i_nlink &&
5174 	    ((btrfs_root_refs(&root->root_item) != 0 &&
5175 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5176 	     btrfs_is_free_space_inode(BTRFS_I(inode))))
5177 		goto no_delete;
5178 
5179 	if (is_bad_inode(inode)) {
5180 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5181 		goto no_delete;
5182 	}
5183 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5184 	if (!special_file(inode->i_mode))
5185 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
5186 
5187 	btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5188 
5189 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5190 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5191 				 &BTRFS_I(inode)->runtime_flags));
5192 		goto no_delete;
5193 	}
5194 
5195 	if (inode->i_nlink > 0) {
5196 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5197 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5198 		goto no_delete;
5199 	}
5200 
5201 	ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5202 	if (ret) {
5203 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5204 		goto no_delete;
5205 	}
5206 
5207 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5208 	if (!rsv) {
5209 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5210 		goto no_delete;
5211 	}
5212 	rsv->size = min_size;
5213 	rsv->failfast = 1;
5214 	global_rsv = &fs_info->global_block_rsv;
5215 
5216 	btrfs_i_size_write(BTRFS_I(inode), 0);
5217 
5218 	/*
5219 	 * This is a bit simpler than btrfs_truncate since we've already
5220 	 * reserved our space for our orphan item in the unlink, so we just
5221 	 * need to reserve some slack space in case we add bytes and update
5222 	 * inode item when doing the truncate.
5223 	 */
5224 	while (1) {
5225 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
5226 					     BTRFS_RESERVE_FLUSH_LIMIT);
5227 
5228 		/*
5229 		 * Try and steal from the global reserve since we will
5230 		 * likely not use this space anyway, we want to try as
5231 		 * hard as possible to get this to work.
5232 		 */
5233 		if (ret)
5234 			steal_from_global++;
5235 		else
5236 			steal_from_global = 0;
5237 		ret = 0;
5238 
5239 		/*
5240 		 * steal_from_global == 0: we reserved stuff, hooray!
5241 		 * steal_from_global == 1: we didn't reserve stuff, boo!
5242 		 * steal_from_global == 2: we've committed, still not a lot of
5243 		 * room but maybe we'll have room in the global reserve this
5244 		 * time.
5245 		 * steal_from_global == 3: abandon all hope!
5246 		 */
5247 		if (steal_from_global > 2) {
5248 			btrfs_warn(fs_info,
5249 				   "Could not get space for a delete, will truncate on mount %d",
5250 				   ret);
5251 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5252 			btrfs_free_block_rsv(fs_info, rsv);
5253 			goto no_delete;
5254 		}
5255 
5256 		trans = btrfs_join_transaction(root);
5257 		if (IS_ERR(trans)) {
5258 			btrfs_orphan_del(NULL, BTRFS_I(inode));
5259 			btrfs_free_block_rsv(fs_info, rsv);
5260 			goto no_delete;
5261 		}
5262 
5263 		/*
5264 		 * We can't just steal from the global reserve, we need to make
5265 		 * sure there is room to do it, if not we need to commit and try
5266 		 * again.
5267 		 */
5268 		if (steal_from_global) {
5269 			if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5270 				ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5271 							      min_size, 0);
5272 			else
5273 				ret = -ENOSPC;
5274 		}
5275 
5276 		/*
5277 		 * Couldn't steal from the global reserve, we have too much
5278 		 * pending stuff built up, commit the transaction and try it
5279 		 * again.
5280 		 */
5281 		if (ret) {
5282 			ret = btrfs_commit_transaction(trans);
5283 			if (ret) {
5284 				btrfs_orphan_del(NULL, BTRFS_I(inode));
5285 				btrfs_free_block_rsv(fs_info, rsv);
5286 				goto no_delete;
5287 			}
5288 			continue;
5289 		} else {
5290 			steal_from_global = 0;
5291 		}
5292 
5293 		trans->block_rsv = rsv;
5294 
5295 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5296 		if (ret != -ENOSPC && ret != -EAGAIN)
5297 			break;
5298 
5299 		trans->block_rsv = &fs_info->trans_block_rsv;
5300 		btrfs_end_transaction(trans);
5301 		trans = NULL;
5302 		btrfs_btree_balance_dirty(fs_info);
5303 	}
5304 
5305 	btrfs_free_block_rsv(fs_info, rsv);
5306 
5307 	/*
5308 	 * Errors here aren't a big deal, it just means we leave orphan items
5309 	 * in the tree.  They will be cleaned up on the next mount.
5310 	 */
5311 	if (ret == 0) {
5312 		trans->block_rsv = root->orphan_block_rsv;
5313 		btrfs_orphan_del(trans, BTRFS_I(inode));
5314 	} else {
5315 		btrfs_orphan_del(NULL, BTRFS_I(inode));
5316 	}
5317 
5318 	trans->block_rsv = &fs_info->trans_block_rsv;
5319 	if (!(root == fs_info->tree_root ||
5320 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5321 		btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5322 
5323 	btrfs_end_transaction(trans);
5324 	btrfs_btree_balance_dirty(fs_info);
5325 no_delete:
5326 	btrfs_remove_delayed_node(BTRFS_I(inode));
5327 	clear_inode(inode);
5328 }
5329 
5330 /*
5331  * this returns the key found in the dir entry in the location pointer.
5332  * If no dir entries were found, location->objectid is 0.
5333  */
5334 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5335 			       struct btrfs_key *location)
5336 {
5337 	const char *name = dentry->d_name.name;
5338 	int namelen = dentry->d_name.len;
5339 	struct btrfs_dir_item *di;
5340 	struct btrfs_path *path;
5341 	struct btrfs_root *root = BTRFS_I(dir)->root;
5342 	int ret = 0;
5343 
5344 	path = btrfs_alloc_path();
5345 	if (!path)
5346 		return -ENOMEM;
5347 
5348 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5349 			name, namelen, 0);
5350 	if (IS_ERR(di))
5351 		ret = PTR_ERR(di);
5352 
5353 	if (IS_ERR_OR_NULL(di))
5354 		goto out_err;
5355 
5356 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5357 out:
5358 	btrfs_free_path(path);
5359 	return ret;
5360 out_err:
5361 	location->objectid = 0;
5362 	goto out;
5363 }
5364 
5365 /*
5366  * when we hit a tree root in a directory, the btrfs part of the inode
5367  * needs to be changed to reflect the root directory of the tree root.  This
5368  * is kind of like crossing a mount point.
5369  */
5370 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5371 				    struct inode *dir,
5372 				    struct dentry *dentry,
5373 				    struct btrfs_key *location,
5374 				    struct btrfs_root **sub_root)
5375 {
5376 	struct btrfs_path *path;
5377 	struct btrfs_root *new_root;
5378 	struct btrfs_root_ref *ref;
5379 	struct extent_buffer *leaf;
5380 	struct btrfs_key key;
5381 	int ret;
5382 	int err = 0;
5383 
5384 	path = btrfs_alloc_path();
5385 	if (!path) {
5386 		err = -ENOMEM;
5387 		goto out;
5388 	}
5389 
5390 	err = -ENOENT;
5391 	key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5392 	key.type = BTRFS_ROOT_REF_KEY;
5393 	key.offset = location->objectid;
5394 
5395 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5396 	if (ret) {
5397 		if (ret < 0)
5398 			err = ret;
5399 		goto out;
5400 	}
5401 
5402 	leaf = path->nodes[0];
5403 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5404 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5405 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5406 		goto out;
5407 
5408 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5409 				   (unsigned long)(ref + 1),
5410 				   dentry->d_name.len);
5411 	if (ret)
5412 		goto out;
5413 
5414 	btrfs_release_path(path);
5415 
5416 	new_root = btrfs_read_fs_root_no_name(fs_info, location);
5417 	if (IS_ERR(new_root)) {
5418 		err = PTR_ERR(new_root);
5419 		goto out;
5420 	}
5421 
5422 	*sub_root = new_root;
5423 	location->objectid = btrfs_root_dirid(&new_root->root_item);
5424 	location->type = BTRFS_INODE_ITEM_KEY;
5425 	location->offset = 0;
5426 	err = 0;
5427 out:
5428 	btrfs_free_path(path);
5429 	return err;
5430 }
5431 
5432 static void inode_tree_add(struct inode *inode)
5433 {
5434 	struct btrfs_root *root = BTRFS_I(inode)->root;
5435 	struct btrfs_inode *entry;
5436 	struct rb_node **p;
5437 	struct rb_node *parent;
5438 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
5439 	u64 ino = btrfs_ino(BTRFS_I(inode));
5440 
5441 	if (inode_unhashed(inode))
5442 		return;
5443 	parent = NULL;
5444 	spin_lock(&root->inode_lock);
5445 	p = &root->inode_tree.rb_node;
5446 	while (*p) {
5447 		parent = *p;
5448 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
5449 
5450 		if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5451 			p = &parent->rb_left;
5452 		else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5453 			p = &parent->rb_right;
5454 		else {
5455 			WARN_ON(!(entry->vfs_inode.i_state &
5456 				  (I_WILL_FREE | I_FREEING)));
5457 			rb_replace_node(parent, new, &root->inode_tree);
5458 			RB_CLEAR_NODE(parent);
5459 			spin_unlock(&root->inode_lock);
5460 			return;
5461 		}
5462 	}
5463 	rb_link_node(new, parent, p);
5464 	rb_insert_color(new, &root->inode_tree);
5465 	spin_unlock(&root->inode_lock);
5466 }
5467 
5468 static void inode_tree_del(struct inode *inode)
5469 {
5470 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5471 	struct btrfs_root *root = BTRFS_I(inode)->root;
5472 	int empty = 0;
5473 
5474 	spin_lock(&root->inode_lock);
5475 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5476 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5477 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5478 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5479 	}
5480 	spin_unlock(&root->inode_lock);
5481 
5482 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
5483 		synchronize_srcu(&fs_info->subvol_srcu);
5484 		spin_lock(&root->inode_lock);
5485 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5486 		spin_unlock(&root->inode_lock);
5487 		if (empty)
5488 			btrfs_add_dead_root(root);
5489 	}
5490 }
5491 
5492 void btrfs_invalidate_inodes(struct btrfs_root *root)
5493 {
5494 	struct btrfs_fs_info *fs_info = root->fs_info;
5495 	struct rb_node *node;
5496 	struct rb_node *prev;
5497 	struct btrfs_inode *entry;
5498 	struct inode *inode;
5499 	u64 objectid = 0;
5500 
5501 	if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5502 		WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5503 
5504 	spin_lock(&root->inode_lock);
5505 again:
5506 	node = root->inode_tree.rb_node;
5507 	prev = NULL;
5508 	while (node) {
5509 		prev = node;
5510 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5511 
5512 		if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5513 			node = node->rb_left;
5514 		else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5515 			node = node->rb_right;
5516 		else
5517 			break;
5518 	}
5519 	if (!node) {
5520 		while (prev) {
5521 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
5522 			if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5523 				node = prev;
5524 				break;
5525 			}
5526 			prev = rb_next(prev);
5527 		}
5528 	}
5529 	while (node) {
5530 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5531 		objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5532 		inode = igrab(&entry->vfs_inode);
5533 		if (inode) {
5534 			spin_unlock(&root->inode_lock);
5535 			if (atomic_read(&inode->i_count) > 1)
5536 				d_prune_aliases(inode);
5537 			/*
5538 			 * btrfs_drop_inode will have it removed from
5539 			 * the inode cache when its usage count
5540 			 * hits zero.
5541 			 */
5542 			iput(inode);
5543 			cond_resched();
5544 			spin_lock(&root->inode_lock);
5545 			goto again;
5546 		}
5547 
5548 		if (cond_resched_lock(&root->inode_lock))
5549 			goto again;
5550 
5551 		node = rb_next(node);
5552 	}
5553 	spin_unlock(&root->inode_lock);
5554 }
5555 
5556 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5557 {
5558 	struct btrfs_iget_args *args = p;
5559 	inode->i_ino = args->location->objectid;
5560 	memcpy(&BTRFS_I(inode)->location, args->location,
5561 	       sizeof(*args->location));
5562 	BTRFS_I(inode)->root = args->root;
5563 	return 0;
5564 }
5565 
5566 static int btrfs_find_actor(struct inode *inode, void *opaque)
5567 {
5568 	struct btrfs_iget_args *args = opaque;
5569 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5570 		args->root == BTRFS_I(inode)->root;
5571 }
5572 
5573 static struct inode *btrfs_iget_locked(struct super_block *s,
5574 				       struct btrfs_key *location,
5575 				       struct btrfs_root *root)
5576 {
5577 	struct inode *inode;
5578 	struct btrfs_iget_args args;
5579 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5580 
5581 	args.location = location;
5582 	args.root = root;
5583 
5584 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5585 			     btrfs_init_locked_inode,
5586 			     (void *)&args);
5587 	return inode;
5588 }
5589 
5590 /* Get an inode object given its location and corresponding root.
5591  * Returns in *is_new if the inode was read from disk
5592  */
5593 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5594 			 struct btrfs_root *root, int *new)
5595 {
5596 	struct inode *inode;
5597 
5598 	inode = btrfs_iget_locked(s, location, root);
5599 	if (!inode)
5600 		return ERR_PTR(-ENOMEM);
5601 
5602 	if (inode->i_state & I_NEW) {
5603 		int ret;
5604 
5605 		ret = btrfs_read_locked_inode(inode);
5606 		if (!is_bad_inode(inode)) {
5607 			inode_tree_add(inode);
5608 			unlock_new_inode(inode);
5609 			if (new)
5610 				*new = 1;
5611 		} else {
5612 			unlock_new_inode(inode);
5613 			iput(inode);
5614 			ASSERT(ret < 0);
5615 			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5616 		}
5617 	}
5618 
5619 	return inode;
5620 }
5621 
5622 static struct inode *new_simple_dir(struct super_block *s,
5623 				    struct btrfs_key *key,
5624 				    struct btrfs_root *root)
5625 {
5626 	struct inode *inode = new_inode(s);
5627 
5628 	if (!inode)
5629 		return ERR_PTR(-ENOMEM);
5630 
5631 	BTRFS_I(inode)->root = root;
5632 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5633 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5634 
5635 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5636 	inode->i_op = &btrfs_dir_ro_inode_operations;
5637 	inode->i_opflags &= ~IOP_XATTR;
5638 	inode->i_fop = &simple_dir_operations;
5639 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5640 	inode->i_mtime = current_time(inode);
5641 	inode->i_atime = inode->i_mtime;
5642 	inode->i_ctime = inode->i_mtime;
5643 	BTRFS_I(inode)->i_otime = inode->i_mtime;
5644 
5645 	return inode;
5646 }
5647 
5648 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5649 {
5650 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5651 	struct inode *inode;
5652 	struct btrfs_root *root = BTRFS_I(dir)->root;
5653 	struct btrfs_root *sub_root = root;
5654 	struct btrfs_key location;
5655 	int index;
5656 	int ret = 0;
5657 
5658 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5659 		return ERR_PTR(-ENAMETOOLONG);
5660 
5661 	ret = btrfs_inode_by_name(dir, dentry, &location);
5662 	if (ret < 0)
5663 		return ERR_PTR(ret);
5664 
5665 	if (location.objectid == 0)
5666 		return ERR_PTR(-ENOENT);
5667 
5668 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5669 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5670 		return inode;
5671 	}
5672 
5673 	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5674 
5675 	index = srcu_read_lock(&fs_info->subvol_srcu);
5676 	ret = fixup_tree_root_location(fs_info, dir, dentry,
5677 				       &location, &sub_root);
5678 	if (ret < 0) {
5679 		if (ret != -ENOENT)
5680 			inode = ERR_PTR(ret);
5681 		else
5682 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5683 	} else {
5684 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5685 	}
5686 	srcu_read_unlock(&fs_info->subvol_srcu, index);
5687 
5688 	if (!IS_ERR(inode) && root != sub_root) {
5689 		down_read(&fs_info->cleanup_work_sem);
5690 		if (!(inode->i_sb->s_flags & MS_RDONLY))
5691 			ret = btrfs_orphan_cleanup(sub_root);
5692 		up_read(&fs_info->cleanup_work_sem);
5693 		if (ret) {
5694 			iput(inode);
5695 			inode = ERR_PTR(ret);
5696 		}
5697 	}
5698 
5699 	return inode;
5700 }
5701 
5702 static int btrfs_dentry_delete(const struct dentry *dentry)
5703 {
5704 	struct btrfs_root *root;
5705 	struct inode *inode = d_inode(dentry);
5706 
5707 	if (!inode && !IS_ROOT(dentry))
5708 		inode = d_inode(dentry->d_parent);
5709 
5710 	if (inode) {
5711 		root = BTRFS_I(inode)->root;
5712 		if (btrfs_root_refs(&root->root_item) == 0)
5713 			return 1;
5714 
5715 		if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5716 			return 1;
5717 	}
5718 	return 0;
5719 }
5720 
5721 static void btrfs_dentry_release(struct dentry *dentry)
5722 {
5723 	kfree(dentry->d_fsdata);
5724 }
5725 
5726 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5727 				   unsigned int flags)
5728 {
5729 	struct inode *inode;
5730 
5731 	inode = btrfs_lookup_dentry(dir, dentry);
5732 	if (IS_ERR(inode)) {
5733 		if (PTR_ERR(inode) == -ENOENT)
5734 			inode = NULL;
5735 		else
5736 			return ERR_CAST(inode);
5737 	}
5738 
5739 	return d_splice_alias(inode, dentry);
5740 }
5741 
5742 unsigned char btrfs_filetype_table[] = {
5743 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5744 };
5745 
5746 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5747 {
5748 	struct inode *inode = file_inode(file);
5749 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5750 	struct btrfs_root *root = BTRFS_I(inode)->root;
5751 	struct btrfs_item *item;
5752 	struct btrfs_dir_item *di;
5753 	struct btrfs_key key;
5754 	struct btrfs_key found_key;
5755 	struct btrfs_path *path;
5756 	struct list_head ins_list;
5757 	struct list_head del_list;
5758 	int ret;
5759 	struct extent_buffer *leaf;
5760 	int slot;
5761 	unsigned char d_type;
5762 	int over = 0;
5763 	char tmp_name[32];
5764 	char *name_ptr;
5765 	int name_len;
5766 	bool put = false;
5767 	struct btrfs_key location;
5768 
5769 	if (!dir_emit_dots(file, ctx))
5770 		return 0;
5771 
5772 	path = btrfs_alloc_path();
5773 	if (!path)
5774 		return -ENOMEM;
5775 
5776 	path->reada = READA_FORWARD;
5777 
5778 	INIT_LIST_HEAD(&ins_list);
5779 	INIT_LIST_HEAD(&del_list);
5780 	put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5781 
5782 	key.type = BTRFS_DIR_INDEX_KEY;
5783 	key.offset = ctx->pos;
5784 	key.objectid = btrfs_ino(BTRFS_I(inode));
5785 
5786 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5787 	if (ret < 0)
5788 		goto err;
5789 
5790 	while (1) {
5791 		leaf = path->nodes[0];
5792 		slot = path->slots[0];
5793 		if (slot >= btrfs_header_nritems(leaf)) {
5794 			ret = btrfs_next_leaf(root, path);
5795 			if (ret < 0)
5796 				goto err;
5797 			else if (ret > 0)
5798 				break;
5799 			continue;
5800 		}
5801 
5802 		item = btrfs_item_nr(slot);
5803 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5804 
5805 		if (found_key.objectid != key.objectid)
5806 			break;
5807 		if (found_key.type != BTRFS_DIR_INDEX_KEY)
5808 			break;
5809 		if (found_key.offset < ctx->pos)
5810 			goto next;
5811 		if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5812 			goto next;
5813 
5814 		ctx->pos = found_key.offset;
5815 
5816 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5817 		if (verify_dir_item(fs_info, leaf, di))
5818 			goto next;
5819 
5820 		name_len = btrfs_dir_name_len(leaf, di);
5821 		if (name_len <= sizeof(tmp_name)) {
5822 			name_ptr = tmp_name;
5823 		} else {
5824 			name_ptr = kmalloc(name_len, GFP_KERNEL);
5825 			if (!name_ptr) {
5826 				ret = -ENOMEM;
5827 				goto err;
5828 			}
5829 		}
5830 		read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5831 				   name_len);
5832 
5833 		d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5834 		btrfs_dir_item_key_to_cpu(leaf, di, &location);
5835 
5836 		over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5837 				 d_type);
5838 
5839 		if (name_ptr != tmp_name)
5840 			kfree(name_ptr);
5841 
5842 		if (over)
5843 			goto nopos;
5844 		ctx->pos++;
5845 next:
5846 		path->slots[0]++;
5847 	}
5848 
5849 	ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5850 	if (ret)
5851 		goto nopos;
5852 
5853 	/*
5854 	 * Stop new entries from being returned after we return the last
5855 	 * entry.
5856 	 *
5857 	 * New directory entries are assigned a strictly increasing
5858 	 * offset.  This means that new entries created during readdir
5859 	 * are *guaranteed* to be seen in the future by that readdir.
5860 	 * This has broken buggy programs which operate on names as
5861 	 * they're returned by readdir.  Until we re-use freed offsets
5862 	 * we have this hack to stop new entries from being returned
5863 	 * under the assumption that they'll never reach this huge
5864 	 * offset.
5865 	 *
5866 	 * This is being careful not to overflow 32bit loff_t unless the
5867 	 * last entry requires it because doing so has broken 32bit apps
5868 	 * in the past.
5869 	 */
5870 	if (ctx->pos >= INT_MAX)
5871 		ctx->pos = LLONG_MAX;
5872 	else
5873 		ctx->pos = INT_MAX;
5874 nopos:
5875 	ret = 0;
5876 err:
5877 	if (put)
5878 		btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5879 	btrfs_free_path(path);
5880 	return ret;
5881 }
5882 
5883 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5884 {
5885 	struct btrfs_root *root = BTRFS_I(inode)->root;
5886 	struct btrfs_trans_handle *trans;
5887 	int ret = 0;
5888 	bool nolock = false;
5889 
5890 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5891 		return 0;
5892 
5893 	if (btrfs_fs_closing(root->fs_info) &&
5894 			btrfs_is_free_space_inode(BTRFS_I(inode)))
5895 		nolock = true;
5896 
5897 	if (wbc->sync_mode == WB_SYNC_ALL) {
5898 		if (nolock)
5899 			trans = btrfs_join_transaction_nolock(root);
5900 		else
5901 			trans = btrfs_join_transaction(root);
5902 		if (IS_ERR(trans))
5903 			return PTR_ERR(trans);
5904 		ret = btrfs_commit_transaction(trans);
5905 	}
5906 	return ret;
5907 }
5908 
5909 /*
5910  * This is somewhat expensive, updating the tree every time the
5911  * inode changes.  But, it is most likely to find the inode in cache.
5912  * FIXME, needs more benchmarking...there are no reasons other than performance
5913  * to keep or drop this code.
5914  */
5915 static int btrfs_dirty_inode(struct inode *inode)
5916 {
5917 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5918 	struct btrfs_root *root = BTRFS_I(inode)->root;
5919 	struct btrfs_trans_handle *trans;
5920 	int ret;
5921 
5922 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5923 		return 0;
5924 
5925 	trans = btrfs_join_transaction(root);
5926 	if (IS_ERR(trans))
5927 		return PTR_ERR(trans);
5928 
5929 	ret = btrfs_update_inode(trans, root, inode);
5930 	if (ret && ret == -ENOSPC) {
5931 		/* whoops, lets try again with the full transaction */
5932 		btrfs_end_transaction(trans);
5933 		trans = btrfs_start_transaction(root, 1);
5934 		if (IS_ERR(trans))
5935 			return PTR_ERR(trans);
5936 
5937 		ret = btrfs_update_inode(trans, root, inode);
5938 	}
5939 	btrfs_end_transaction(trans);
5940 	if (BTRFS_I(inode)->delayed_node)
5941 		btrfs_balance_delayed_items(fs_info);
5942 
5943 	return ret;
5944 }
5945 
5946 /*
5947  * This is a copy of file_update_time.  We need this so we can return error on
5948  * ENOSPC for updating the inode in the case of file write and mmap writes.
5949  */
5950 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5951 			     int flags)
5952 {
5953 	struct btrfs_root *root = BTRFS_I(inode)->root;
5954 
5955 	if (btrfs_root_readonly(root))
5956 		return -EROFS;
5957 
5958 	if (flags & S_VERSION)
5959 		inode_inc_iversion(inode);
5960 	if (flags & S_CTIME)
5961 		inode->i_ctime = *now;
5962 	if (flags & S_MTIME)
5963 		inode->i_mtime = *now;
5964 	if (flags & S_ATIME)
5965 		inode->i_atime = *now;
5966 	return btrfs_dirty_inode(inode);
5967 }
5968 
5969 /*
5970  * find the highest existing sequence number in a directory
5971  * and then set the in-memory index_cnt variable to reflect
5972  * free sequence numbers
5973  */
5974 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5975 {
5976 	struct btrfs_root *root = inode->root;
5977 	struct btrfs_key key, found_key;
5978 	struct btrfs_path *path;
5979 	struct extent_buffer *leaf;
5980 	int ret;
5981 
5982 	key.objectid = btrfs_ino(inode);
5983 	key.type = BTRFS_DIR_INDEX_KEY;
5984 	key.offset = (u64)-1;
5985 
5986 	path = btrfs_alloc_path();
5987 	if (!path)
5988 		return -ENOMEM;
5989 
5990 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5991 	if (ret < 0)
5992 		goto out;
5993 	/* FIXME: we should be able to handle this */
5994 	if (ret == 0)
5995 		goto out;
5996 	ret = 0;
5997 
5998 	/*
5999 	 * MAGIC NUMBER EXPLANATION:
6000 	 * since we search a directory based on f_pos we have to start at 2
6001 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6002 	 * else has to start at 2
6003 	 */
6004 	if (path->slots[0] == 0) {
6005 		inode->index_cnt = 2;
6006 		goto out;
6007 	}
6008 
6009 	path->slots[0]--;
6010 
6011 	leaf = path->nodes[0];
6012 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6013 
6014 	if (found_key.objectid != btrfs_ino(inode) ||
6015 	    found_key.type != BTRFS_DIR_INDEX_KEY) {
6016 		inode->index_cnt = 2;
6017 		goto out;
6018 	}
6019 
6020 	inode->index_cnt = found_key.offset + 1;
6021 out:
6022 	btrfs_free_path(path);
6023 	return ret;
6024 }
6025 
6026 /*
6027  * helper to find a free sequence number in a given directory.  This current
6028  * code is very simple, later versions will do smarter things in the btree
6029  */
6030 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6031 {
6032 	int ret = 0;
6033 
6034 	if (dir->index_cnt == (u64)-1) {
6035 		ret = btrfs_inode_delayed_dir_index_count(dir);
6036 		if (ret) {
6037 			ret = btrfs_set_inode_index_count(dir);
6038 			if (ret)
6039 				return ret;
6040 		}
6041 	}
6042 
6043 	*index = dir->index_cnt;
6044 	dir->index_cnt++;
6045 
6046 	return ret;
6047 }
6048 
6049 static int btrfs_insert_inode_locked(struct inode *inode)
6050 {
6051 	struct btrfs_iget_args args;
6052 	args.location = &BTRFS_I(inode)->location;
6053 	args.root = BTRFS_I(inode)->root;
6054 
6055 	return insert_inode_locked4(inode,
6056 		   btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6057 		   btrfs_find_actor, &args);
6058 }
6059 
6060 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6061 				     struct btrfs_root *root,
6062 				     struct inode *dir,
6063 				     const char *name, int name_len,
6064 				     u64 ref_objectid, u64 objectid,
6065 				     umode_t mode, u64 *index)
6066 {
6067 	struct btrfs_fs_info *fs_info = root->fs_info;
6068 	struct inode *inode;
6069 	struct btrfs_inode_item *inode_item;
6070 	struct btrfs_key *location;
6071 	struct btrfs_path *path;
6072 	struct btrfs_inode_ref *ref;
6073 	struct btrfs_key key[2];
6074 	u32 sizes[2];
6075 	int nitems = name ? 2 : 1;
6076 	unsigned long ptr;
6077 	int ret;
6078 
6079 	path = btrfs_alloc_path();
6080 	if (!path)
6081 		return ERR_PTR(-ENOMEM);
6082 
6083 	inode = new_inode(fs_info->sb);
6084 	if (!inode) {
6085 		btrfs_free_path(path);
6086 		return ERR_PTR(-ENOMEM);
6087 	}
6088 
6089 	/*
6090 	 * O_TMPFILE, set link count to 0, so that after this point,
6091 	 * we fill in an inode item with the correct link count.
6092 	 */
6093 	if (!name)
6094 		set_nlink(inode, 0);
6095 
6096 	/*
6097 	 * we have to initialize this early, so we can reclaim the inode
6098 	 * number if we fail afterwards in this function.
6099 	 */
6100 	inode->i_ino = objectid;
6101 
6102 	if (dir && name) {
6103 		trace_btrfs_inode_request(dir);
6104 
6105 		ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6106 		if (ret) {
6107 			btrfs_free_path(path);
6108 			iput(inode);
6109 			return ERR_PTR(ret);
6110 		}
6111 	} else if (dir) {
6112 		*index = 0;
6113 	}
6114 	/*
6115 	 * index_cnt is ignored for everything but a dir,
6116 	 * btrfs_get_inode_index_count has an explanation for the magic
6117 	 * number
6118 	 */
6119 	BTRFS_I(inode)->index_cnt = 2;
6120 	BTRFS_I(inode)->dir_index = *index;
6121 	BTRFS_I(inode)->root = root;
6122 	BTRFS_I(inode)->generation = trans->transid;
6123 	inode->i_generation = BTRFS_I(inode)->generation;
6124 
6125 	/*
6126 	 * We could have gotten an inode number from somebody who was fsynced
6127 	 * and then removed in this same transaction, so let's just set full
6128 	 * sync since it will be a full sync anyway and this will blow away the
6129 	 * old info in the log.
6130 	 */
6131 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6132 
6133 	key[0].objectid = objectid;
6134 	key[0].type = BTRFS_INODE_ITEM_KEY;
6135 	key[0].offset = 0;
6136 
6137 	sizes[0] = sizeof(struct btrfs_inode_item);
6138 
6139 	if (name) {
6140 		/*
6141 		 * Start new inodes with an inode_ref. This is slightly more
6142 		 * efficient for small numbers of hard links since they will
6143 		 * be packed into one item. Extended refs will kick in if we
6144 		 * add more hard links than can fit in the ref item.
6145 		 */
6146 		key[1].objectid = objectid;
6147 		key[1].type = BTRFS_INODE_REF_KEY;
6148 		key[1].offset = ref_objectid;
6149 
6150 		sizes[1] = name_len + sizeof(*ref);
6151 	}
6152 
6153 	location = &BTRFS_I(inode)->location;
6154 	location->objectid = objectid;
6155 	location->offset = 0;
6156 	location->type = BTRFS_INODE_ITEM_KEY;
6157 
6158 	ret = btrfs_insert_inode_locked(inode);
6159 	if (ret < 0)
6160 		goto fail;
6161 
6162 	path->leave_spinning = 1;
6163 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6164 	if (ret != 0)
6165 		goto fail_unlock;
6166 
6167 	inode_init_owner(inode, dir, mode);
6168 	inode_set_bytes(inode, 0);
6169 
6170 	inode->i_mtime = current_time(inode);
6171 	inode->i_atime = inode->i_mtime;
6172 	inode->i_ctime = inode->i_mtime;
6173 	BTRFS_I(inode)->i_otime = inode->i_mtime;
6174 
6175 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6176 				  struct btrfs_inode_item);
6177 	memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6178 			     sizeof(*inode_item));
6179 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
6180 
6181 	if (name) {
6182 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6183 				     struct btrfs_inode_ref);
6184 		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6185 		btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6186 		ptr = (unsigned long)(ref + 1);
6187 		write_extent_buffer(path->nodes[0], name, ptr, name_len);
6188 	}
6189 
6190 	btrfs_mark_buffer_dirty(path->nodes[0]);
6191 	btrfs_free_path(path);
6192 
6193 	btrfs_inherit_iflags(inode, dir);
6194 
6195 	if (S_ISREG(mode)) {
6196 		if (btrfs_test_opt(fs_info, NODATASUM))
6197 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6198 		if (btrfs_test_opt(fs_info, NODATACOW))
6199 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6200 				BTRFS_INODE_NODATASUM;
6201 	}
6202 
6203 	inode_tree_add(inode);
6204 
6205 	trace_btrfs_inode_new(inode);
6206 	btrfs_set_inode_last_trans(trans, inode);
6207 
6208 	btrfs_update_root_times(trans, root);
6209 
6210 	ret = btrfs_inode_inherit_props(trans, inode, dir);
6211 	if (ret)
6212 		btrfs_err(fs_info,
6213 			  "error inheriting props for ino %llu (root %llu): %d",
6214 			btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6215 
6216 	return inode;
6217 
6218 fail_unlock:
6219 	unlock_new_inode(inode);
6220 fail:
6221 	if (dir && name)
6222 		BTRFS_I(dir)->index_cnt--;
6223 	btrfs_free_path(path);
6224 	iput(inode);
6225 	return ERR_PTR(ret);
6226 }
6227 
6228 static inline u8 btrfs_inode_type(struct inode *inode)
6229 {
6230 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6231 }
6232 
6233 /*
6234  * utility function to add 'inode' into 'parent_inode' with
6235  * a give name and a given sequence number.
6236  * if 'add_backref' is true, also insert a backref from the
6237  * inode to the parent directory.
6238  */
6239 int btrfs_add_link(struct btrfs_trans_handle *trans,
6240 		   struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6241 		   const char *name, int name_len, int add_backref, u64 index)
6242 {
6243 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6244 	int ret = 0;
6245 	struct btrfs_key key;
6246 	struct btrfs_root *root = parent_inode->root;
6247 	u64 ino = btrfs_ino(inode);
6248 	u64 parent_ino = btrfs_ino(parent_inode);
6249 
6250 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6251 		memcpy(&key, &inode->root->root_key, sizeof(key));
6252 	} else {
6253 		key.objectid = ino;
6254 		key.type = BTRFS_INODE_ITEM_KEY;
6255 		key.offset = 0;
6256 	}
6257 
6258 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6259 		ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6260 					 root->root_key.objectid, parent_ino,
6261 					 index, name, name_len);
6262 	} else if (add_backref) {
6263 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6264 					     parent_ino, index);
6265 	}
6266 
6267 	/* Nothing to clean up yet */
6268 	if (ret)
6269 		return ret;
6270 
6271 	ret = btrfs_insert_dir_item(trans, root, name, name_len,
6272 				    parent_inode, &key,
6273 				    btrfs_inode_type(&inode->vfs_inode), index);
6274 	if (ret == -EEXIST || ret == -EOVERFLOW)
6275 		goto fail_dir_item;
6276 	else if (ret) {
6277 		btrfs_abort_transaction(trans, ret);
6278 		return ret;
6279 	}
6280 
6281 	btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6282 			   name_len * 2);
6283 	inode_inc_iversion(&parent_inode->vfs_inode);
6284 	parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6285 		current_time(&parent_inode->vfs_inode);
6286 	ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6287 	if (ret)
6288 		btrfs_abort_transaction(trans, ret);
6289 	return ret;
6290 
6291 fail_dir_item:
6292 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6293 		u64 local_index;
6294 		int err;
6295 		err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6296 					 root->root_key.objectid, parent_ino,
6297 					 &local_index, name, name_len);
6298 
6299 	} else if (add_backref) {
6300 		u64 local_index;
6301 		int err;
6302 
6303 		err = btrfs_del_inode_ref(trans, root, name, name_len,
6304 					  ino, parent_ino, &local_index);
6305 	}
6306 	return ret;
6307 }
6308 
6309 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6310 			    struct btrfs_inode *dir, struct dentry *dentry,
6311 			    struct btrfs_inode *inode, int backref, u64 index)
6312 {
6313 	int err = btrfs_add_link(trans, dir, inode,
6314 				 dentry->d_name.name, dentry->d_name.len,
6315 				 backref, index);
6316 	if (err > 0)
6317 		err = -EEXIST;
6318 	return err;
6319 }
6320 
6321 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6322 			umode_t mode, dev_t rdev)
6323 {
6324 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6325 	struct btrfs_trans_handle *trans;
6326 	struct btrfs_root *root = BTRFS_I(dir)->root;
6327 	struct inode *inode = NULL;
6328 	int err;
6329 	int drop_inode = 0;
6330 	u64 objectid;
6331 	u64 index = 0;
6332 
6333 	/*
6334 	 * 2 for inode item and ref
6335 	 * 2 for dir items
6336 	 * 1 for xattr if selinux is on
6337 	 */
6338 	trans = btrfs_start_transaction(root, 5);
6339 	if (IS_ERR(trans))
6340 		return PTR_ERR(trans);
6341 
6342 	err = btrfs_find_free_ino(root, &objectid);
6343 	if (err)
6344 		goto out_unlock;
6345 
6346 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6347 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6348 			mode, &index);
6349 	if (IS_ERR(inode)) {
6350 		err = PTR_ERR(inode);
6351 		goto out_unlock;
6352 	}
6353 
6354 	/*
6355 	* If the active LSM wants to access the inode during
6356 	* d_instantiate it needs these. Smack checks to see
6357 	* if the filesystem supports xattrs by looking at the
6358 	* ops vector.
6359 	*/
6360 	inode->i_op = &btrfs_special_inode_operations;
6361 	init_special_inode(inode, inode->i_mode, rdev);
6362 
6363 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6364 	if (err)
6365 		goto out_unlock_inode;
6366 
6367 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6368 			0, index);
6369 	if (err) {
6370 		goto out_unlock_inode;
6371 	} else {
6372 		btrfs_update_inode(trans, root, inode);
6373 		unlock_new_inode(inode);
6374 		d_instantiate(dentry, inode);
6375 	}
6376 
6377 out_unlock:
6378 	btrfs_end_transaction(trans);
6379 	btrfs_balance_delayed_items(fs_info);
6380 	btrfs_btree_balance_dirty(fs_info);
6381 	if (drop_inode) {
6382 		inode_dec_link_count(inode);
6383 		iput(inode);
6384 	}
6385 	return err;
6386 
6387 out_unlock_inode:
6388 	drop_inode = 1;
6389 	unlock_new_inode(inode);
6390 	goto out_unlock;
6391 
6392 }
6393 
6394 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6395 			umode_t mode, bool excl)
6396 {
6397 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6398 	struct btrfs_trans_handle *trans;
6399 	struct btrfs_root *root = BTRFS_I(dir)->root;
6400 	struct inode *inode = NULL;
6401 	int drop_inode_on_err = 0;
6402 	int err;
6403 	u64 objectid;
6404 	u64 index = 0;
6405 
6406 	/*
6407 	 * 2 for inode item and ref
6408 	 * 2 for dir items
6409 	 * 1 for xattr if selinux is on
6410 	 */
6411 	trans = btrfs_start_transaction(root, 5);
6412 	if (IS_ERR(trans))
6413 		return PTR_ERR(trans);
6414 
6415 	err = btrfs_find_free_ino(root, &objectid);
6416 	if (err)
6417 		goto out_unlock;
6418 
6419 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6420 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6421 			mode, &index);
6422 	if (IS_ERR(inode)) {
6423 		err = PTR_ERR(inode);
6424 		goto out_unlock;
6425 	}
6426 	drop_inode_on_err = 1;
6427 	/*
6428 	* If the active LSM wants to access the inode during
6429 	* d_instantiate it needs these. Smack checks to see
6430 	* if the filesystem supports xattrs by looking at the
6431 	* ops vector.
6432 	*/
6433 	inode->i_fop = &btrfs_file_operations;
6434 	inode->i_op = &btrfs_file_inode_operations;
6435 	inode->i_mapping->a_ops = &btrfs_aops;
6436 
6437 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6438 	if (err)
6439 		goto out_unlock_inode;
6440 
6441 	err = btrfs_update_inode(trans, root, inode);
6442 	if (err)
6443 		goto out_unlock_inode;
6444 
6445 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6446 			0, index);
6447 	if (err)
6448 		goto out_unlock_inode;
6449 
6450 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6451 	unlock_new_inode(inode);
6452 	d_instantiate(dentry, inode);
6453 
6454 out_unlock:
6455 	btrfs_end_transaction(trans);
6456 	if (err && drop_inode_on_err) {
6457 		inode_dec_link_count(inode);
6458 		iput(inode);
6459 	}
6460 	btrfs_balance_delayed_items(fs_info);
6461 	btrfs_btree_balance_dirty(fs_info);
6462 	return err;
6463 
6464 out_unlock_inode:
6465 	unlock_new_inode(inode);
6466 	goto out_unlock;
6467 
6468 }
6469 
6470 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6471 		      struct dentry *dentry)
6472 {
6473 	struct btrfs_trans_handle *trans = NULL;
6474 	struct btrfs_root *root = BTRFS_I(dir)->root;
6475 	struct inode *inode = d_inode(old_dentry);
6476 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6477 	u64 index;
6478 	int err;
6479 	int drop_inode = 0;
6480 
6481 	/* do not allow sys_link's with other subvols of the same device */
6482 	if (root->objectid != BTRFS_I(inode)->root->objectid)
6483 		return -EXDEV;
6484 
6485 	if (inode->i_nlink >= BTRFS_LINK_MAX)
6486 		return -EMLINK;
6487 
6488 	err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6489 	if (err)
6490 		goto fail;
6491 
6492 	/*
6493 	 * 2 items for inode and inode ref
6494 	 * 2 items for dir items
6495 	 * 1 item for parent inode
6496 	 */
6497 	trans = btrfs_start_transaction(root, 5);
6498 	if (IS_ERR(trans)) {
6499 		err = PTR_ERR(trans);
6500 		trans = NULL;
6501 		goto fail;
6502 	}
6503 
6504 	/* There are several dir indexes for this inode, clear the cache. */
6505 	BTRFS_I(inode)->dir_index = 0ULL;
6506 	inc_nlink(inode);
6507 	inode_inc_iversion(inode);
6508 	inode->i_ctime = current_time(inode);
6509 	ihold(inode);
6510 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6511 
6512 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6513 			1, index);
6514 
6515 	if (err) {
6516 		drop_inode = 1;
6517 	} else {
6518 		struct dentry *parent = dentry->d_parent;
6519 		err = btrfs_update_inode(trans, root, inode);
6520 		if (err)
6521 			goto fail;
6522 		if (inode->i_nlink == 1) {
6523 			/*
6524 			 * If new hard link count is 1, it's a file created
6525 			 * with open(2) O_TMPFILE flag.
6526 			 */
6527 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
6528 			if (err)
6529 				goto fail;
6530 		}
6531 		d_instantiate(dentry, inode);
6532 		btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6533 	}
6534 
6535 	btrfs_balance_delayed_items(fs_info);
6536 fail:
6537 	if (trans)
6538 		btrfs_end_transaction(trans);
6539 	if (drop_inode) {
6540 		inode_dec_link_count(inode);
6541 		iput(inode);
6542 	}
6543 	btrfs_btree_balance_dirty(fs_info);
6544 	return err;
6545 }
6546 
6547 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6548 {
6549 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6550 	struct inode *inode = NULL;
6551 	struct btrfs_trans_handle *trans;
6552 	struct btrfs_root *root = BTRFS_I(dir)->root;
6553 	int err = 0;
6554 	int drop_on_err = 0;
6555 	u64 objectid = 0;
6556 	u64 index = 0;
6557 
6558 	/*
6559 	 * 2 items for inode and ref
6560 	 * 2 items for dir items
6561 	 * 1 for xattr if selinux is on
6562 	 */
6563 	trans = btrfs_start_transaction(root, 5);
6564 	if (IS_ERR(trans))
6565 		return PTR_ERR(trans);
6566 
6567 	err = btrfs_find_free_ino(root, &objectid);
6568 	if (err)
6569 		goto out_fail;
6570 
6571 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6572 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6573 			S_IFDIR | mode, &index);
6574 	if (IS_ERR(inode)) {
6575 		err = PTR_ERR(inode);
6576 		goto out_fail;
6577 	}
6578 
6579 	drop_on_err = 1;
6580 	/* these must be set before we unlock the inode */
6581 	inode->i_op = &btrfs_dir_inode_operations;
6582 	inode->i_fop = &btrfs_dir_file_operations;
6583 
6584 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6585 	if (err)
6586 		goto out_fail_inode;
6587 
6588 	btrfs_i_size_write(BTRFS_I(inode), 0);
6589 	err = btrfs_update_inode(trans, root, inode);
6590 	if (err)
6591 		goto out_fail_inode;
6592 
6593 	err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6594 			dentry->d_name.name,
6595 			dentry->d_name.len, 0, index);
6596 	if (err)
6597 		goto out_fail_inode;
6598 
6599 	d_instantiate(dentry, inode);
6600 	/*
6601 	 * mkdir is special.  We're unlocking after we call d_instantiate
6602 	 * to avoid a race with nfsd calling d_instantiate.
6603 	 */
6604 	unlock_new_inode(inode);
6605 	drop_on_err = 0;
6606 
6607 out_fail:
6608 	btrfs_end_transaction(trans);
6609 	if (drop_on_err) {
6610 		inode_dec_link_count(inode);
6611 		iput(inode);
6612 	}
6613 	btrfs_balance_delayed_items(fs_info);
6614 	btrfs_btree_balance_dirty(fs_info);
6615 	return err;
6616 
6617 out_fail_inode:
6618 	unlock_new_inode(inode);
6619 	goto out_fail;
6620 }
6621 
6622 /* Find next extent map of a given extent map, caller needs to ensure locks */
6623 static struct extent_map *next_extent_map(struct extent_map *em)
6624 {
6625 	struct rb_node *next;
6626 
6627 	next = rb_next(&em->rb_node);
6628 	if (!next)
6629 		return NULL;
6630 	return container_of(next, struct extent_map, rb_node);
6631 }
6632 
6633 static struct extent_map *prev_extent_map(struct extent_map *em)
6634 {
6635 	struct rb_node *prev;
6636 
6637 	prev = rb_prev(&em->rb_node);
6638 	if (!prev)
6639 		return NULL;
6640 	return container_of(prev, struct extent_map, rb_node);
6641 }
6642 
6643 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6644  * the existing extent is the nearest extent to map_start,
6645  * and an extent that you want to insert, deal with overlap and insert
6646  * the best fitted new extent into the tree.
6647  */
6648 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6649 				struct extent_map *existing,
6650 				struct extent_map *em,
6651 				u64 map_start)
6652 {
6653 	struct extent_map *prev;
6654 	struct extent_map *next;
6655 	u64 start;
6656 	u64 end;
6657 	u64 start_diff;
6658 
6659 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6660 
6661 	if (existing->start > map_start) {
6662 		next = existing;
6663 		prev = prev_extent_map(next);
6664 	} else {
6665 		prev = existing;
6666 		next = next_extent_map(prev);
6667 	}
6668 
6669 	start = prev ? extent_map_end(prev) : em->start;
6670 	start = max_t(u64, start, em->start);
6671 	end = next ? next->start : extent_map_end(em);
6672 	end = min_t(u64, end, extent_map_end(em));
6673 	start_diff = start - em->start;
6674 	em->start = start;
6675 	em->len = end - start;
6676 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6677 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6678 		em->block_start += start_diff;
6679 		em->block_len -= start_diff;
6680 	}
6681 	return add_extent_mapping(em_tree, em, 0);
6682 }
6683 
6684 static noinline int uncompress_inline(struct btrfs_path *path,
6685 				      struct page *page,
6686 				      size_t pg_offset, u64 extent_offset,
6687 				      struct btrfs_file_extent_item *item)
6688 {
6689 	int ret;
6690 	struct extent_buffer *leaf = path->nodes[0];
6691 	char *tmp;
6692 	size_t max_size;
6693 	unsigned long inline_size;
6694 	unsigned long ptr;
6695 	int compress_type;
6696 
6697 	WARN_ON(pg_offset != 0);
6698 	compress_type = btrfs_file_extent_compression(leaf, item);
6699 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6700 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6701 					btrfs_item_nr(path->slots[0]));
6702 	tmp = kmalloc(inline_size, GFP_NOFS);
6703 	if (!tmp)
6704 		return -ENOMEM;
6705 	ptr = btrfs_file_extent_inline_start(item);
6706 
6707 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6708 
6709 	max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6710 	ret = btrfs_decompress(compress_type, tmp, page,
6711 			       extent_offset, inline_size, max_size);
6712 	kfree(tmp);
6713 	return ret;
6714 }
6715 
6716 /*
6717  * a bit scary, this does extent mapping from logical file offset to the disk.
6718  * the ugly parts come from merging extents from the disk with the in-ram
6719  * representation.  This gets more complex because of the data=ordered code,
6720  * where the in-ram extents might be locked pending data=ordered completion.
6721  *
6722  * This also copies inline extents directly into the page.
6723  */
6724 
6725 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6726 		struct page *page,
6727 	    size_t pg_offset, u64 start, u64 len,
6728 		int create)
6729 {
6730 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6731 	int ret;
6732 	int err = 0;
6733 	u64 extent_start = 0;
6734 	u64 extent_end = 0;
6735 	u64 objectid = btrfs_ino(inode);
6736 	u32 found_type;
6737 	struct btrfs_path *path = NULL;
6738 	struct btrfs_root *root = inode->root;
6739 	struct btrfs_file_extent_item *item;
6740 	struct extent_buffer *leaf;
6741 	struct btrfs_key found_key;
6742 	struct extent_map *em = NULL;
6743 	struct extent_map_tree *em_tree = &inode->extent_tree;
6744 	struct extent_io_tree *io_tree = &inode->io_tree;
6745 	struct btrfs_trans_handle *trans = NULL;
6746 	const bool new_inline = !page || create;
6747 
6748 again:
6749 	read_lock(&em_tree->lock);
6750 	em = lookup_extent_mapping(em_tree, start, len);
6751 	if (em)
6752 		em->bdev = fs_info->fs_devices->latest_bdev;
6753 	read_unlock(&em_tree->lock);
6754 
6755 	if (em) {
6756 		if (em->start > start || em->start + em->len <= start)
6757 			free_extent_map(em);
6758 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6759 			free_extent_map(em);
6760 		else
6761 			goto out;
6762 	}
6763 	em = alloc_extent_map();
6764 	if (!em) {
6765 		err = -ENOMEM;
6766 		goto out;
6767 	}
6768 	em->bdev = fs_info->fs_devices->latest_bdev;
6769 	em->start = EXTENT_MAP_HOLE;
6770 	em->orig_start = EXTENT_MAP_HOLE;
6771 	em->len = (u64)-1;
6772 	em->block_len = (u64)-1;
6773 
6774 	if (!path) {
6775 		path = btrfs_alloc_path();
6776 		if (!path) {
6777 			err = -ENOMEM;
6778 			goto out;
6779 		}
6780 		/*
6781 		 * Chances are we'll be called again, so go ahead and do
6782 		 * readahead
6783 		 */
6784 		path->reada = READA_FORWARD;
6785 	}
6786 
6787 	ret = btrfs_lookup_file_extent(trans, root, path,
6788 				       objectid, start, trans != NULL);
6789 	if (ret < 0) {
6790 		err = ret;
6791 		goto out;
6792 	}
6793 
6794 	if (ret != 0) {
6795 		if (path->slots[0] == 0)
6796 			goto not_found;
6797 		path->slots[0]--;
6798 	}
6799 
6800 	leaf = path->nodes[0];
6801 	item = btrfs_item_ptr(leaf, path->slots[0],
6802 			      struct btrfs_file_extent_item);
6803 	/* are we inside the extent that was found? */
6804 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6805 	found_type = found_key.type;
6806 	if (found_key.objectid != objectid ||
6807 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6808 		/*
6809 		 * If we backup past the first extent we want to move forward
6810 		 * and see if there is an extent in front of us, otherwise we'll
6811 		 * say there is a hole for our whole search range which can
6812 		 * cause problems.
6813 		 */
6814 		extent_end = start;
6815 		goto next;
6816 	}
6817 
6818 	found_type = btrfs_file_extent_type(leaf, item);
6819 	extent_start = found_key.offset;
6820 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6821 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6822 		extent_end = extent_start +
6823 		       btrfs_file_extent_num_bytes(leaf, item);
6824 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6825 		size_t size;
6826 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6827 		extent_end = ALIGN(extent_start + size,
6828 				   fs_info->sectorsize);
6829 	}
6830 next:
6831 	if (start >= extent_end) {
6832 		path->slots[0]++;
6833 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6834 			ret = btrfs_next_leaf(root, path);
6835 			if (ret < 0) {
6836 				err = ret;
6837 				goto out;
6838 			}
6839 			if (ret > 0)
6840 				goto not_found;
6841 			leaf = path->nodes[0];
6842 		}
6843 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6844 		if (found_key.objectid != objectid ||
6845 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6846 			goto not_found;
6847 		if (start + len <= found_key.offset)
6848 			goto not_found;
6849 		if (start > found_key.offset)
6850 			goto next;
6851 		em->start = start;
6852 		em->orig_start = start;
6853 		em->len = found_key.offset - start;
6854 		goto not_found_em;
6855 	}
6856 
6857 	btrfs_extent_item_to_extent_map(inode, path, item,
6858 			new_inline, em);
6859 
6860 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6861 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6862 		goto insert;
6863 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6864 		unsigned long ptr;
6865 		char *map;
6866 		size_t size;
6867 		size_t extent_offset;
6868 		size_t copy_size;
6869 
6870 		if (new_inline)
6871 			goto out;
6872 
6873 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6874 		extent_offset = page_offset(page) + pg_offset - extent_start;
6875 		copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6876 				  size - extent_offset);
6877 		em->start = extent_start + extent_offset;
6878 		em->len = ALIGN(copy_size, fs_info->sectorsize);
6879 		em->orig_block_len = em->len;
6880 		em->orig_start = em->start;
6881 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6882 		if (create == 0 && !PageUptodate(page)) {
6883 			if (btrfs_file_extent_compression(leaf, item) !=
6884 			    BTRFS_COMPRESS_NONE) {
6885 				ret = uncompress_inline(path, page, pg_offset,
6886 							extent_offset, item);
6887 				if (ret) {
6888 					err = ret;
6889 					goto out;
6890 				}
6891 			} else {
6892 				map = kmap(page);
6893 				read_extent_buffer(leaf, map + pg_offset, ptr,
6894 						   copy_size);
6895 				if (pg_offset + copy_size < PAGE_SIZE) {
6896 					memset(map + pg_offset + copy_size, 0,
6897 					       PAGE_SIZE - pg_offset -
6898 					       copy_size);
6899 				}
6900 				kunmap(page);
6901 			}
6902 			flush_dcache_page(page);
6903 		} else if (create && PageUptodate(page)) {
6904 			BUG();
6905 			if (!trans) {
6906 				kunmap(page);
6907 				free_extent_map(em);
6908 				em = NULL;
6909 
6910 				btrfs_release_path(path);
6911 				trans = btrfs_join_transaction(root);
6912 
6913 				if (IS_ERR(trans))
6914 					return ERR_CAST(trans);
6915 				goto again;
6916 			}
6917 			map = kmap(page);
6918 			write_extent_buffer(leaf, map + pg_offset, ptr,
6919 					    copy_size);
6920 			kunmap(page);
6921 			btrfs_mark_buffer_dirty(leaf);
6922 		}
6923 		set_extent_uptodate(io_tree, em->start,
6924 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
6925 		goto insert;
6926 	}
6927 not_found:
6928 	em->start = start;
6929 	em->orig_start = start;
6930 	em->len = len;
6931 not_found_em:
6932 	em->block_start = EXTENT_MAP_HOLE;
6933 	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6934 insert:
6935 	btrfs_release_path(path);
6936 	if (em->start > start || extent_map_end(em) <= start) {
6937 		btrfs_err(fs_info,
6938 			  "bad extent! em: [%llu %llu] passed [%llu %llu]",
6939 			  em->start, em->len, start, len);
6940 		err = -EIO;
6941 		goto out;
6942 	}
6943 
6944 	err = 0;
6945 	write_lock(&em_tree->lock);
6946 	ret = add_extent_mapping(em_tree, em, 0);
6947 	/* it is possible that someone inserted the extent into the tree
6948 	 * while we had the lock dropped.  It is also possible that
6949 	 * an overlapping map exists in the tree
6950 	 */
6951 	if (ret == -EEXIST) {
6952 		struct extent_map *existing;
6953 
6954 		ret = 0;
6955 
6956 		existing = search_extent_mapping(em_tree, start, len);
6957 		/*
6958 		 * existing will always be non-NULL, since there must be
6959 		 * extent causing the -EEXIST.
6960 		 */
6961 		if (existing->start == em->start &&
6962 		    extent_map_end(existing) >= extent_map_end(em) &&
6963 		    em->block_start == existing->block_start) {
6964 			/*
6965 			 * The existing extent map already encompasses the
6966 			 * entire extent map we tried to add.
6967 			 */
6968 			free_extent_map(em);
6969 			em = existing;
6970 			err = 0;
6971 
6972 		} else if (start >= extent_map_end(existing) ||
6973 		    start <= existing->start) {
6974 			/*
6975 			 * The existing extent map is the one nearest to
6976 			 * the [start, start + len) range which overlaps
6977 			 */
6978 			err = merge_extent_mapping(em_tree, existing,
6979 						   em, start);
6980 			free_extent_map(existing);
6981 			if (err) {
6982 				free_extent_map(em);
6983 				em = NULL;
6984 			}
6985 		} else {
6986 			free_extent_map(em);
6987 			em = existing;
6988 			err = 0;
6989 		}
6990 	}
6991 	write_unlock(&em_tree->lock);
6992 out:
6993 
6994 	trace_btrfs_get_extent(root, inode, em);
6995 
6996 	btrfs_free_path(path);
6997 	if (trans) {
6998 		ret = btrfs_end_transaction(trans);
6999 		if (!err)
7000 			err = ret;
7001 	}
7002 	if (err) {
7003 		free_extent_map(em);
7004 		return ERR_PTR(err);
7005 	}
7006 	BUG_ON(!em); /* Error is always set */
7007 	return em;
7008 }
7009 
7010 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7011 		struct page *page,
7012 		size_t pg_offset, u64 start, u64 len,
7013 		int create)
7014 {
7015 	struct extent_map *em;
7016 	struct extent_map *hole_em = NULL;
7017 	u64 range_start = start;
7018 	u64 end;
7019 	u64 found;
7020 	u64 found_end;
7021 	int err = 0;
7022 
7023 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7024 	if (IS_ERR(em))
7025 		return em;
7026 	if (em) {
7027 		/*
7028 		 * if our em maps to
7029 		 * -  a hole or
7030 		 * -  a pre-alloc extent,
7031 		 * there might actually be delalloc bytes behind it.
7032 		 */
7033 		if (em->block_start != EXTENT_MAP_HOLE &&
7034 		    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7035 			return em;
7036 		else
7037 			hole_em = em;
7038 	}
7039 
7040 	/* check to see if we've wrapped (len == -1 or similar) */
7041 	end = start + len;
7042 	if (end < start)
7043 		end = (u64)-1;
7044 	else
7045 		end -= 1;
7046 
7047 	em = NULL;
7048 
7049 	/* ok, we didn't find anything, lets look for delalloc */
7050 	found = count_range_bits(&inode->io_tree, &range_start,
7051 				 end, len, EXTENT_DELALLOC, 1);
7052 	found_end = range_start + found;
7053 	if (found_end < range_start)
7054 		found_end = (u64)-1;
7055 
7056 	/*
7057 	 * we didn't find anything useful, return
7058 	 * the original results from get_extent()
7059 	 */
7060 	if (range_start > end || found_end <= start) {
7061 		em = hole_em;
7062 		hole_em = NULL;
7063 		goto out;
7064 	}
7065 
7066 	/* adjust the range_start to make sure it doesn't
7067 	 * go backwards from the start they passed in
7068 	 */
7069 	range_start = max(start, range_start);
7070 	found = found_end - range_start;
7071 
7072 	if (found > 0) {
7073 		u64 hole_start = start;
7074 		u64 hole_len = len;
7075 
7076 		em = alloc_extent_map();
7077 		if (!em) {
7078 			err = -ENOMEM;
7079 			goto out;
7080 		}
7081 		/*
7082 		 * when btrfs_get_extent can't find anything it
7083 		 * returns one huge hole
7084 		 *
7085 		 * make sure what it found really fits our range, and
7086 		 * adjust to make sure it is based on the start from
7087 		 * the caller
7088 		 */
7089 		if (hole_em) {
7090 			u64 calc_end = extent_map_end(hole_em);
7091 
7092 			if (calc_end <= start || (hole_em->start > end)) {
7093 				free_extent_map(hole_em);
7094 				hole_em = NULL;
7095 			} else {
7096 				hole_start = max(hole_em->start, start);
7097 				hole_len = calc_end - hole_start;
7098 			}
7099 		}
7100 		em->bdev = NULL;
7101 		if (hole_em && range_start > hole_start) {
7102 			/* our hole starts before our delalloc, so we
7103 			 * have to return just the parts of the hole
7104 			 * that go until  the delalloc starts
7105 			 */
7106 			em->len = min(hole_len,
7107 				      range_start - hole_start);
7108 			em->start = hole_start;
7109 			em->orig_start = hole_start;
7110 			/*
7111 			 * don't adjust block start at all,
7112 			 * it is fixed at EXTENT_MAP_HOLE
7113 			 */
7114 			em->block_start = hole_em->block_start;
7115 			em->block_len = hole_len;
7116 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7117 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7118 		} else {
7119 			em->start = range_start;
7120 			em->len = found;
7121 			em->orig_start = range_start;
7122 			em->block_start = EXTENT_MAP_DELALLOC;
7123 			em->block_len = found;
7124 		}
7125 	} else if (hole_em) {
7126 		return hole_em;
7127 	}
7128 out:
7129 
7130 	free_extent_map(hole_em);
7131 	if (err) {
7132 		free_extent_map(em);
7133 		return ERR_PTR(err);
7134 	}
7135 	return em;
7136 }
7137 
7138 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7139 						  const u64 start,
7140 						  const u64 len,
7141 						  const u64 orig_start,
7142 						  const u64 block_start,
7143 						  const u64 block_len,
7144 						  const u64 orig_block_len,
7145 						  const u64 ram_bytes,
7146 						  const int type)
7147 {
7148 	struct extent_map *em = NULL;
7149 	int ret;
7150 
7151 	if (type != BTRFS_ORDERED_NOCOW) {
7152 		em = create_io_em(inode, start, len, orig_start,
7153 				  block_start, block_len, orig_block_len,
7154 				  ram_bytes,
7155 				  BTRFS_COMPRESS_NONE, /* compress_type */
7156 				  type);
7157 		if (IS_ERR(em))
7158 			goto out;
7159 	}
7160 	ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7161 					   len, block_len, type);
7162 	if (ret) {
7163 		if (em) {
7164 			free_extent_map(em);
7165 			btrfs_drop_extent_cache(BTRFS_I(inode), start,
7166 						start + len - 1, 0);
7167 		}
7168 		em = ERR_PTR(ret);
7169 	}
7170  out:
7171 
7172 	return em;
7173 }
7174 
7175 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7176 						  u64 start, u64 len)
7177 {
7178 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7179 	struct btrfs_root *root = BTRFS_I(inode)->root;
7180 	struct extent_map *em;
7181 	struct btrfs_key ins;
7182 	u64 alloc_hint;
7183 	int ret;
7184 
7185 	alloc_hint = get_extent_allocation_hint(inode, start, len);
7186 	ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7187 				   0, alloc_hint, &ins, 1, 1);
7188 	if (ret)
7189 		return ERR_PTR(ret);
7190 
7191 	em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7192 				     ins.objectid, ins.offset, ins.offset,
7193 				     ins.offset, BTRFS_ORDERED_REGULAR);
7194 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7195 	if (IS_ERR(em))
7196 		btrfs_free_reserved_extent(fs_info, ins.objectid,
7197 					   ins.offset, 1);
7198 
7199 	return em;
7200 }
7201 
7202 /*
7203  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7204  * block must be cow'd
7205  */
7206 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7207 			      u64 *orig_start, u64 *orig_block_len,
7208 			      u64 *ram_bytes)
7209 {
7210 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7211 	struct btrfs_path *path;
7212 	int ret;
7213 	struct extent_buffer *leaf;
7214 	struct btrfs_root *root = BTRFS_I(inode)->root;
7215 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7216 	struct btrfs_file_extent_item *fi;
7217 	struct btrfs_key key;
7218 	u64 disk_bytenr;
7219 	u64 backref_offset;
7220 	u64 extent_end;
7221 	u64 num_bytes;
7222 	int slot;
7223 	int found_type;
7224 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7225 
7226 	path = btrfs_alloc_path();
7227 	if (!path)
7228 		return -ENOMEM;
7229 
7230 	ret = btrfs_lookup_file_extent(NULL, root, path,
7231 			btrfs_ino(BTRFS_I(inode)), offset, 0);
7232 	if (ret < 0)
7233 		goto out;
7234 
7235 	slot = path->slots[0];
7236 	if (ret == 1) {
7237 		if (slot == 0) {
7238 			/* can't find the item, must cow */
7239 			ret = 0;
7240 			goto out;
7241 		}
7242 		slot--;
7243 	}
7244 	ret = 0;
7245 	leaf = path->nodes[0];
7246 	btrfs_item_key_to_cpu(leaf, &key, slot);
7247 	if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7248 	    key.type != BTRFS_EXTENT_DATA_KEY) {
7249 		/* not our file or wrong item type, must cow */
7250 		goto out;
7251 	}
7252 
7253 	if (key.offset > offset) {
7254 		/* Wrong offset, must cow */
7255 		goto out;
7256 	}
7257 
7258 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7259 	found_type = btrfs_file_extent_type(leaf, fi);
7260 	if (found_type != BTRFS_FILE_EXTENT_REG &&
7261 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7262 		/* not a regular extent, must cow */
7263 		goto out;
7264 	}
7265 
7266 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7267 		goto out;
7268 
7269 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7270 	if (extent_end <= offset)
7271 		goto out;
7272 
7273 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7274 	if (disk_bytenr == 0)
7275 		goto out;
7276 
7277 	if (btrfs_file_extent_compression(leaf, fi) ||
7278 	    btrfs_file_extent_encryption(leaf, fi) ||
7279 	    btrfs_file_extent_other_encoding(leaf, fi))
7280 		goto out;
7281 
7282 	backref_offset = btrfs_file_extent_offset(leaf, fi);
7283 
7284 	if (orig_start) {
7285 		*orig_start = key.offset - backref_offset;
7286 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7287 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7288 	}
7289 
7290 	if (btrfs_extent_readonly(fs_info, disk_bytenr))
7291 		goto out;
7292 
7293 	num_bytes = min(offset + *len, extent_end) - offset;
7294 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7295 		u64 range_end;
7296 
7297 		range_end = round_up(offset + num_bytes,
7298 				     root->fs_info->sectorsize) - 1;
7299 		ret = test_range_bit(io_tree, offset, range_end,
7300 				     EXTENT_DELALLOC, 0, NULL);
7301 		if (ret) {
7302 			ret = -EAGAIN;
7303 			goto out;
7304 		}
7305 	}
7306 
7307 	btrfs_release_path(path);
7308 
7309 	/*
7310 	 * look for other files referencing this extent, if we
7311 	 * find any we must cow
7312 	 */
7313 
7314 	ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7315 				    key.offset - backref_offset, disk_bytenr);
7316 	if (ret) {
7317 		ret = 0;
7318 		goto out;
7319 	}
7320 
7321 	/*
7322 	 * adjust disk_bytenr and num_bytes to cover just the bytes
7323 	 * in this extent we are about to write.  If there
7324 	 * are any csums in that range we have to cow in order
7325 	 * to keep the csums correct
7326 	 */
7327 	disk_bytenr += backref_offset;
7328 	disk_bytenr += offset - key.offset;
7329 	if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7330 		goto out;
7331 	/*
7332 	 * all of the above have passed, it is safe to overwrite this extent
7333 	 * without cow
7334 	 */
7335 	*len = num_bytes;
7336 	ret = 1;
7337 out:
7338 	btrfs_free_path(path);
7339 	return ret;
7340 }
7341 
7342 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7343 {
7344 	struct radix_tree_root *root = &inode->i_mapping->page_tree;
7345 	int found = false;
7346 	void **pagep = NULL;
7347 	struct page *page = NULL;
7348 	int start_idx;
7349 	int end_idx;
7350 
7351 	start_idx = start >> PAGE_SHIFT;
7352 
7353 	/*
7354 	 * end is the last byte in the last page.  end == start is legal
7355 	 */
7356 	end_idx = end >> PAGE_SHIFT;
7357 
7358 	rcu_read_lock();
7359 
7360 	/* Most of the code in this while loop is lifted from
7361 	 * find_get_page.  It's been modified to begin searching from a
7362 	 * page and return just the first page found in that range.  If the
7363 	 * found idx is less than or equal to the end idx then we know that
7364 	 * a page exists.  If no pages are found or if those pages are
7365 	 * outside of the range then we're fine (yay!) */
7366 	while (page == NULL &&
7367 	       radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7368 		page = radix_tree_deref_slot(pagep);
7369 		if (unlikely(!page))
7370 			break;
7371 
7372 		if (radix_tree_exception(page)) {
7373 			if (radix_tree_deref_retry(page)) {
7374 				page = NULL;
7375 				continue;
7376 			}
7377 			/*
7378 			 * Otherwise, shmem/tmpfs must be storing a swap entry
7379 			 * here as an exceptional entry: so return it without
7380 			 * attempting to raise page count.
7381 			 */
7382 			page = NULL;
7383 			break; /* TODO: Is this relevant for this use case? */
7384 		}
7385 
7386 		if (!page_cache_get_speculative(page)) {
7387 			page = NULL;
7388 			continue;
7389 		}
7390 
7391 		/*
7392 		 * Has the page moved?
7393 		 * This is part of the lockless pagecache protocol. See
7394 		 * include/linux/pagemap.h for details.
7395 		 */
7396 		if (unlikely(page != *pagep)) {
7397 			put_page(page);
7398 			page = NULL;
7399 		}
7400 	}
7401 
7402 	if (page) {
7403 		if (page->index <= end_idx)
7404 			found = true;
7405 		put_page(page);
7406 	}
7407 
7408 	rcu_read_unlock();
7409 	return found;
7410 }
7411 
7412 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7413 			      struct extent_state **cached_state, int writing)
7414 {
7415 	struct btrfs_ordered_extent *ordered;
7416 	int ret = 0;
7417 
7418 	while (1) {
7419 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7420 				 cached_state);
7421 		/*
7422 		 * We're concerned with the entire range that we're going to be
7423 		 * doing DIO to, so we need to make sure there's no ordered
7424 		 * extents in this range.
7425 		 */
7426 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7427 						     lockend - lockstart + 1);
7428 
7429 		/*
7430 		 * We need to make sure there are no buffered pages in this
7431 		 * range either, we could have raced between the invalidate in
7432 		 * generic_file_direct_write and locking the extent.  The
7433 		 * invalidate needs to happen so that reads after a write do not
7434 		 * get stale data.
7435 		 */
7436 		if (!ordered &&
7437 		    (!writing ||
7438 		     !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7439 			break;
7440 
7441 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7442 				     cached_state, GFP_NOFS);
7443 
7444 		if (ordered) {
7445 			/*
7446 			 * If we are doing a DIO read and the ordered extent we
7447 			 * found is for a buffered write, we can not wait for it
7448 			 * to complete and retry, because if we do so we can
7449 			 * deadlock with concurrent buffered writes on page
7450 			 * locks. This happens only if our DIO read covers more
7451 			 * than one extent map, if at this point has already
7452 			 * created an ordered extent for a previous extent map
7453 			 * and locked its range in the inode's io tree, and a
7454 			 * concurrent write against that previous extent map's
7455 			 * range and this range started (we unlock the ranges
7456 			 * in the io tree only when the bios complete and
7457 			 * buffered writes always lock pages before attempting
7458 			 * to lock range in the io tree).
7459 			 */
7460 			if (writing ||
7461 			    test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7462 				btrfs_start_ordered_extent(inode, ordered, 1);
7463 			else
7464 				ret = -ENOTBLK;
7465 			btrfs_put_ordered_extent(ordered);
7466 		} else {
7467 			/*
7468 			 * We could trigger writeback for this range (and wait
7469 			 * for it to complete) and then invalidate the pages for
7470 			 * this range (through invalidate_inode_pages2_range()),
7471 			 * but that can lead us to a deadlock with a concurrent
7472 			 * call to readpages() (a buffered read or a defrag call
7473 			 * triggered a readahead) on a page lock due to an
7474 			 * ordered dio extent we created before but did not have
7475 			 * yet a corresponding bio submitted (whence it can not
7476 			 * complete), which makes readpages() wait for that
7477 			 * ordered extent to complete while holding a lock on
7478 			 * that page.
7479 			 */
7480 			ret = -ENOTBLK;
7481 		}
7482 
7483 		if (ret)
7484 			break;
7485 
7486 		cond_resched();
7487 	}
7488 
7489 	return ret;
7490 }
7491 
7492 /* The callers of this must take lock_extent() */
7493 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7494 				       u64 orig_start, u64 block_start,
7495 				       u64 block_len, u64 orig_block_len,
7496 				       u64 ram_bytes, int compress_type,
7497 				       int type)
7498 {
7499 	struct extent_map_tree *em_tree;
7500 	struct extent_map *em;
7501 	struct btrfs_root *root = BTRFS_I(inode)->root;
7502 	int ret;
7503 
7504 	ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7505 	       type == BTRFS_ORDERED_COMPRESSED ||
7506 	       type == BTRFS_ORDERED_NOCOW ||
7507 	       type == BTRFS_ORDERED_REGULAR);
7508 
7509 	em_tree = &BTRFS_I(inode)->extent_tree;
7510 	em = alloc_extent_map();
7511 	if (!em)
7512 		return ERR_PTR(-ENOMEM);
7513 
7514 	em->start = start;
7515 	em->orig_start = orig_start;
7516 	em->len = len;
7517 	em->block_len = block_len;
7518 	em->block_start = block_start;
7519 	em->bdev = root->fs_info->fs_devices->latest_bdev;
7520 	em->orig_block_len = orig_block_len;
7521 	em->ram_bytes = ram_bytes;
7522 	em->generation = -1;
7523 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
7524 	if (type == BTRFS_ORDERED_PREALLOC) {
7525 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
7526 	} else if (type == BTRFS_ORDERED_COMPRESSED) {
7527 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7528 		em->compress_type = compress_type;
7529 	}
7530 
7531 	do {
7532 		btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7533 				em->start + em->len - 1, 0);
7534 		write_lock(&em_tree->lock);
7535 		ret = add_extent_mapping(em_tree, em, 1);
7536 		write_unlock(&em_tree->lock);
7537 		/*
7538 		 * The caller has taken lock_extent(), who could race with us
7539 		 * to add em?
7540 		 */
7541 	} while (ret == -EEXIST);
7542 
7543 	if (ret) {
7544 		free_extent_map(em);
7545 		return ERR_PTR(ret);
7546 	}
7547 
7548 	/* em got 2 refs now, callers needs to do free_extent_map once. */
7549 	return em;
7550 }
7551 
7552 static void adjust_dio_outstanding_extents(struct inode *inode,
7553 					   struct btrfs_dio_data *dio_data,
7554 					   const u64 len)
7555 {
7556 	unsigned num_extents = count_max_extents(len);
7557 
7558 	/*
7559 	 * If we have an outstanding_extents count still set then we're
7560 	 * within our reservation, otherwise we need to adjust our inode
7561 	 * counter appropriately.
7562 	 */
7563 	if (dio_data->outstanding_extents >= num_extents) {
7564 		dio_data->outstanding_extents -= num_extents;
7565 	} else {
7566 		/*
7567 		 * If dio write length has been split due to no large enough
7568 		 * contiguous space, we need to compensate our inode counter
7569 		 * appropriately.
7570 		 */
7571 		u64 num_needed = num_extents - dio_data->outstanding_extents;
7572 
7573 		spin_lock(&BTRFS_I(inode)->lock);
7574 		BTRFS_I(inode)->outstanding_extents += num_needed;
7575 		spin_unlock(&BTRFS_I(inode)->lock);
7576 	}
7577 }
7578 
7579 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7580 				   struct buffer_head *bh_result, int create)
7581 {
7582 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7583 	struct extent_map *em;
7584 	struct extent_state *cached_state = NULL;
7585 	struct btrfs_dio_data *dio_data = NULL;
7586 	u64 start = iblock << inode->i_blkbits;
7587 	u64 lockstart, lockend;
7588 	u64 len = bh_result->b_size;
7589 	int unlock_bits = EXTENT_LOCKED;
7590 	int ret = 0;
7591 
7592 	if (create)
7593 		unlock_bits |= EXTENT_DIRTY;
7594 	else
7595 		len = min_t(u64, len, fs_info->sectorsize);
7596 
7597 	lockstart = start;
7598 	lockend = start + len - 1;
7599 
7600 	if (current->journal_info) {
7601 		/*
7602 		 * Need to pull our outstanding extents and set journal_info to NULL so
7603 		 * that anything that needs to check if there's a transaction doesn't get
7604 		 * confused.
7605 		 */
7606 		dio_data = current->journal_info;
7607 		current->journal_info = NULL;
7608 	}
7609 
7610 	/*
7611 	 * If this errors out it's because we couldn't invalidate pagecache for
7612 	 * this range and we need to fallback to buffered.
7613 	 */
7614 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7615 			       create)) {
7616 		ret = -ENOTBLK;
7617 		goto err;
7618 	}
7619 
7620 	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7621 	if (IS_ERR(em)) {
7622 		ret = PTR_ERR(em);
7623 		goto unlock_err;
7624 	}
7625 
7626 	/*
7627 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7628 	 * io.  INLINE is special, and we could probably kludge it in here, but
7629 	 * it's still buffered so for safety lets just fall back to the generic
7630 	 * buffered path.
7631 	 *
7632 	 * For COMPRESSED we _have_ to read the entire extent in so we can
7633 	 * decompress it, so there will be buffering required no matter what we
7634 	 * do, so go ahead and fallback to buffered.
7635 	 *
7636 	 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7637 	 * to buffered IO.  Don't blame me, this is the price we pay for using
7638 	 * the generic code.
7639 	 */
7640 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7641 	    em->block_start == EXTENT_MAP_INLINE) {
7642 		free_extent_map(em);
7643 		ret = -ENOTBLK;
7644 		goto unlock_err;
7645 	}
7646 
7647 	/* Just a good old fashioned hole, return */
7648 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7649 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7650 		free_extent_map(em);
7651 		goto unlock_err;
7652 	}
7653 
7654 	/*
7655 	 * We don't allocate a new extent in the following cases
7656 	 *
7657 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7658 	 * existing extent.
7659 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
7660 	 * just use the extent.
7661 	 *
7662 	 */
7663 	if (!create) {
7664 		len = min(len, em->len - (start - em->start));
7665 		lockstart = start + len;
7666 		goto unlock;
7667 	}
7668 
7669 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7670 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7671 	     em->block_start != EXTENT_MAP_HOLE)) {
7672 		int type;
7673 		u64 block_start, orig_start, orig_block_len, ram_bytes;
7674 
7675 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7676 			type = BTRFS_ORDERED_PREALLOC;
7677 		else
7678 			type = BTRFS_ORDERED_NOCOW;
7679 		len = min(len, em->len - (start - em->start));
7680 		block_start = em->block_start + (start - em->start);
7681 
7682 		if (can_nocow_extent(inode, start, &len, &orig_start,
7683 				     &orig_block_len, &ram_bytes) == 1 &&
7684 		    btrfs_inc_nocow_writers(fs_info, block_start)) {
7685 			struct extent_map *em2;
7686 
7687 			em2 = btrfs_create_dio_extent(inode, start, len,
7688 						      orig_start, block_start,
7689 						      len, orig_block_len,
7690 						      ram_bytes, type);
7691 			btrfs_dec_nocow_writers(fs_info, block_start);
7692 			if (type == BTRFS_ORDERED_PREALLOC) {
7693 				free_extent_map(em);
7694 				em = em2;
7695 			}
7696 			if (em2 && IS_ERR(em2)) {
7697 				ret = PTR_ERR(em2);
7698 				goto unlock_err;
7699 			}
7700 			/*
7701 			 * For inode marked NODATACOW or extent marked PREALLOC,
7702 			 * use the existing or preallocated extent, so does not
7703 			 * need to adjust btrfs_space_info's bytes_may_use.
7704 			 */
7705 			btrfs_free_reserved_data_space_noquota(inode,
7706 					start, len);
7707 			goto unlock;
7708 		}
7709 	}
7710 
7711 	/*
7712 	 * this will cow the extent, reset the len in case we changed
7713 	 * it above
7714 	 */
7715 	len = bh_result->b_size;
7716 	free_extent_map(em);
7717 	em = btrfs_new_extent_direct(inode, start, len);
7718 	if (IS_ERR(em)) {
7719 		ret = PTR_ERR(em);
7720 		goto unlock_err;
7721 	}
7722 	len = min(len, em->len - (start - em->start));
7723 unlock:
7724 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7725 		inode->i_blkbits;
7726 	bh_result->b_size = len;
7727 	bh_result->b_bdev = em->bdev;
7728 	set_buffer_mapped(bh_result);
7729 	if (create) {
7730 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7731 			set_buffer_new(bh_result);
7732 
7733 		/*
7734 		 * Need to update the i_size under the extent lock so buffered
7735 		 * readers will get the updated i_size when we unlock.
7736 		 */
7737 		if (!dio_data->overwrite && start + len > i_size_read(inode))
7738 			i_size_write(inode, start + len);
7739 
7740 		adjust_dio_outstanding_extents(inode, dio_data, len);
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, GFP_NOFS);
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, GFP_NOFS);
7767 err:
7768 	if (dio_data)
7769 		current->journal_info = dio_data;
7770 	/*
7771 	 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7772 	 * write less data then expected, so that we don't underflow our inode's
7773 	 * outstanding extents counter.
7774 	 */
7775 	if (create && dio_data)
7776 		adjust_dio_outstanding_extents(inode, dio_data, len);
7777 
7778 	return ret;
7779 }
7780 
7781 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7782 					int mirror_num)
7783 {
7784 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7785 	int ret;
7786 
7787 	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7788 
7789 	bio_get(bio);
7790 
7791 	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7792 	if (ret)
7793 		goto err;
7794 
7795 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7796 err:
7797 	bio_put(bio);
7798 	return ret;
7799 }
7800 
7801 static int btrfs_check_dio_repairable(struct inode *inode,
7802 				      struct bio *failed_bio,
7803 				      struct io_failure_record *failrec,
7804 				      int failed_mirror)
7805 {
7806 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7807 	int num_copies;
7808 
7809 	num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7810 	if (num_copies == 1) {
7811 		/*
7812 		 * we only have a single copy of the data, so don't bother with
7813 		 * all the retry and error correction code that follows. no
7814 		 * matter what the error is, it is very likely to persist.
7815 		 */
7816 		btrfs_debug(fs_info,
7817 			"Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7818 			num_copies, failrec->this_mirror, failed_mirror);
7819 		return 0;
7820 	}
7821 
7822 	failrec->failed_mirror = failed_mirror;
7823 	failrec->this_mirror++;
7824 	if (failrec->this_mirror == failed_mirror)
7825 		failrec->this_mirror++;
7826 
7827 	if (failrec->this_mirror > num_copies) {
7828 		btrfs_debug(fs_info,
7829 			"Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7830 			num_copies, failrec->this_mirror, failed_mirror);
7831 		return 0;
7832 	}
7833 
7834 	return 1;
7835 }
7836 
7837 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7838 			struct page *page, unsigned int pgoff,
7839 			u64 start, u64 end, int failed_mirror,
7840 			bio_end_io_t *repair_endio, void *repair_arg)
7841 {
7842 	struct io_failure_record *failrec;
7843 	struct bio *bio;
7844 	int isector;
7845 	int read_mode = 0;
7846 	int ret;
7847 
7848 	BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7849 
7850 	ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7851 	if (ret)
7852 		return ret;
7853 
7854 	ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7855 					 failed_mirror);
7856 	if (!ret) {
7857 		free_io_failure(BTRFS_I(inode), failrec);
7858 		return -EIO;
7859 	}
7860 
7861 	if ((failed_bio->bi_vcnt > 1)
7862 		|| (failed_bio->bi_io_vec->bv_len
7863 			> btrfs_inode_sectorsize(inode)))
7864 		read_mode |= REQ_FAILFAST_DEV;
7865 
7866 	isector = start - btrfs_io_bio(failed_bio)->logical;
7867 	isector >>= inode->i_sb->s_blocksize_bits;
7868 	bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7869 				pgoff, isector, repair_endio, repair_arg);
7870 	if (!bio) {
7871 		free_io_failure(BTRFS_I(inode), failrec);
7872 		return -EIO;
7873 	}
7874 	bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7875 
7876 	btrfs_debug(BTRFS_I(inode)->root->fs_info,
7877 		    "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7878 		    read_mode, failrec->this_mirror, failrec->in_validation);
7879 
7880 	ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7881 	if (ret) {
7882 		free_io_failure(BTRFS_I(inode), failrec);
7883 		bio_put(bio);
7884 	}
7885 
7886 	return ret;
7887 }
7888 
7889 struct btrfs_retry_complete {
7890 	struct completion done;
7891 	struct inode *inode;
7892 	u64 start;
7893 	int uptodate;
7894 };
7895 
7896 static void btrfs_retry_endio_nocsum(struct bio *bio)
7897 {
7898 	struct btrfs_retry_complete *done = bio->bi_private;
7899 	struct inode *inode;
7900 	struct bio_vec *bvec;
7901 	int i;
7902 
7903 	if (bio->bi_error)
7904 		goto end;
7905 
7906 	ASSERT(bio->bi_vcnt == 1);
7907 	inode = bio->bi_io_vec->bv_page->mapping->host;
7908 	ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7909 
7910 	done->uptodate = 1;
7911 	bio_for_each_segment_all(bvec, bio, i)
7912 	clean_io_failure(BTRFS_I(done->inode), done->start, bvec->bv_page, 0);
7913 end:
7914 	complete(&done->done);
7915 	bio_put(bio);
7916 }
7917 
7918 static int __btrfs_correct_data_nocsum(struct inode *inode,
7919 				       struct btrfs_io_bio *io_bio)
7920 {
7921 	struct btrfs_fs_info *fs_info;
7922 	struct bio_vec *bvec;
7923 	struct btrfs_retry_complete done;
7924 	u64 start;
7925 	unsigned int pgoff;
7926 	u32 sectorsize;
7927 	int nr_sectors;
7928 	int i;
7929 	int ret;
7930 
7931 	fs_info = BTRFS_I(inode)->root->fs_info;
7932 	sectorsize = fs_info->sectorsize;
7933 
7934 	start = io_bio->logical;
7935 	done.inode = inode;
7936 
7937 	bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7938 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7939 		pgoff = bvec->bv_offset;
7940 
7941 next_block_or_try_again:
7942 		done.uptodate = 0;
7943 		done.start = start;
7944 		init_completion(&done.done);
7945 
7946 		ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7947 				pgoff, start, start + sectorsize - 1,
7948 				io_bio->mirror_num,
7949 				btrfs_retry_endio_nocsum, &done);
7950 		if (ret)
7951 			return ret;
7952 
7953 		wait_for_completion(&done.done);
7954 
7955 		if (!done.uptodate) {
7956 			/* We might have another mirror, so try again */
7957 			goto next_block_or_try_again;
7958 		}
7959 
7960 		start += sectorsize;
7961 
7962 		if (nr_sectors--) {
7963 			pgoff += sectorsize;
7964 			goto next_block_or_try_again;
7965 		}
7966 	}
7967 
7968 	return 0;
7969 }
7970 
7971 static void btrfs_retry_endio(struct bio *bio)
7972 {
7973 	struct btrfs_retry_complete *done = bio->bi_private;
7974 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7975 	struct inode *inode;
7976 	struct bio_vec *bvec;
7977 	u64 start;
7978 	int uptodate;
7979 	int ret;
7980 	int i;
7981 
7982 	if (bio->bi_error)
7983 		goto end;
7984 
7985 	uptodate = 1;
7986 
7987 	start = done->start;
7988 
7989 	ASSERT(bio->bi_vcnt == 1);
7990 	inode = bio->bi_io_vec->bv_page->mapping->host;
7991 	ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7992 
7993 	bio_for_each_segment_all(bvec, bio, i) {
7994 		ret = __readpage_endio_check(done->inode, io_bio, i,
7995 					bvec->bv_page, bvec->bv_offset,
7996 					done->start, bvec->bv_len);
7997 		if (!ret)
7998 			clean_io_failure(BTRFS_I(done->inode), done->start,
7999 					bvec->bv_page, bvec->bv_offset);
8000 		else
8001 			uptodate = 0;
8002 	}
8003 
8004 	done->uptodate = uptodate;
8005 end:
8006 	complete(&done->done);
8007 	bio_put(bio);
8008 }
8009 
8010 static int __btrfs_subio_endio_read(struct inode *inode,
8011 				    struct btrfs_io_bio *io_bio, int err)
8012 {
8013 	struct btrfs_fs_info *fs_info;
8014 	struct bio_vec *bvec;
8015 	struct btrfs_retry_complete done;
8016 	u64 start;
8017 	u64 offset = 0;
8018 	u32 sectorsize;
8019 	int nr_sectors;
8020 	unsigned int pgoff;
8021 	int csum_pos;
8022 	int i;
8023 	int ret;
8024 
8025 	fs_info = BTRFS_I(inode)->root->fs_info;
8026 	sectorsize = fs_info->sectorsize;
8027 
8028 	err = 0;
8029 	start = io_bio->logical;
8030 	done.inode = inode;
8031 
8032 	bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8033 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8034 
8035 		pgoff = bvec->bv_offset;
8036 next_block:
8037 		csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8038 		ret = __readpage_endio_check(inode, io_bio, csum_pos,
8039 					bvec->bv_page, pgoff, start,
8040 					sectorsize);
8041 		if (likely(!ret))
8042 			goto next;
8043 try_again:
8044 		done.uptodate = 0;
8045 		done.start = start;
8046 		init_completion(&done.done);
8047 
8048 		ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8049 				pgoff, start, start + sectorsize - 1,
8050 				io_bio->mirror_num,
8051 				btrfs_retry_endio, &done);
8052 		if (ret) {
8053 			err = ret;
8054 			goto next;
8055 		}
8056 
8057 		wait_for_completion(&done.done);
8058 
8059 		if (!done.uptodate) {
8060 			/* We might have another mirror, so try again */
8061 			goto try_again;
8062 		}
8063 next:
8064 		offset += sectorsize;
8065 		start += sectorsize;
8066 
8067 		ASSERT(nr_sectors);
8068 
8069 		if (--nr_sectors) {
8070 			pgoff += sectorsize;
8071 			goto next_block;
8072 		}
8073 	}
8074 
8075 	return err;
8076 }
8077 
8078 static int btrfs_subio_endio_read(struct inode *inode,
8079 				  struct btrfs_io_bio *io_bio, int err)
8080 {
8081 	bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8082 
8083 	if (skip_csum) {
8084 		if (unlikely(err))
8085 			return __btrfs_correct_data_nocsum(inode, io_bio);
8086 		else
8087 			return 0;
8088 	} else {
8089 		return __btrfs_subio_endio_read(inode, io_bio, err);
8090 	}
8091 }
8092 
8093 static void btrfs_endio_direct_read(struct bio *bio)
8094 {
8095 	struct btrfs_dio_private *dip = bio->bi_private;
8096 	struct inode *inode = dip->inode;
8097 	struct bio *dio_bio;
8098 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8099 	int err = bio->bi_error;
8100 
8101 	if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8102 		err = btrfs_subio_endio_read(inode, io_bio, err);
8103 
8104 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8105 		      dip->logical_offset + dip->bytes - 1);
8106 	dio_bio = dip->dio_bio;
8107 
8108 	kfree(dip);
8109 
8110 	dio_bio->bi_error = bio->bi_error;
8111 	dio_end_io(dio_bio, bio->bi_error);
8112 
8113 	if (io_bio->end_io)
8114 		io_bio->end_io(io_bio, err);
8115 	bio_put(bio);
8116 }
8117 
8118 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8119 						    const u64 offset,
8120 						    const u64 bytes,
8121 						    const int uptodate)
8122 {
8123 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8124 	struct btrfs_ordered_extent *ordered = NULL;
8125 	u64 ordered_offset = offset;
8126 	u64 ordered_bytes = bytes;
8127 	int ret;
8128 
8129 again:
8130 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8131 						   &ordered_offset,
8132 						   ordered_bytes,
8133 						   uptodate);
8134 	if (!ret)
8135 		goto out_test;
8136 
8137 	btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8138 			finish_ordered_fn, NULL, NULL);
8139 	btrfs_queue_work(fs_info->endio_write_workers, &ordered->work);
8140 out_test:
8141 	/*
8142 	 * our bio might span multiple ordered extents.  If we haven't
8143 	 * completed the accounting for the whole dio, go back and try again
8144 	 */
8145 	if (ordered_offset < offset + bytes) {
8146 		ordered_bytes = offset + bytes - ordered_offset;
8147 		ordered = NULL;
8148 		goto again;
8149 	}
8150 }
8151 
8152 static void btrfs_endio_direct_write(struct bio *bio)
8153 {
8154 	struct btrfs_dio_private *dip = bio->bi_private;
8155 	struct bio *dio_bio = dip->dio_bio;
8156 
8157 	btrfs_endio_direct_write_update_ordered(dip->inode,
8158 						dip->logical_offset,
8159 						dip->bytes,
8160 						!bio->bi_error);
8161 
8162 	kfree(dip);
8163 
8164 	dio_bio->bi_error = bio->bi_error;
8165 	dio_end_io(dio_bio, bio->bi_error);
8166 	bio_put(bio);
8167 }
8168 
8169 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8170 				    struct bio *bio, int mirror_num,
8171 				    unsigned long bio_flags, u64 offset)
8172 {
8173 	int ret;
8174 	ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8175 	BUG_ON(ret); /* -ENOMEM */
8176 	return 0;
8177 }
8178 
8179 static void btrfs_end_dio_bio(struct bio *bio)
8180 {
8181 	struct btrfs_dio_private *dip = bio->bi_private;
8182 	int err = bio->bi_error;
8183 
8184 	if (err)
8185 		btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8186 			   "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8187 			   btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8188 			   bio->bi_opf,
8189 			   (unsigned long long)bio->bi_iter.bi_sector,
8190 			   bio->bi_iter.bi_size, err);
8191 
8192 	if (dip->subio_endio)
8193 		err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8194 
8195 	if (err) {
8196 		dip->errors = 1;
8197 
8198 		/*
8199 		 * before atomic variable goto zero, we must make sure
8200 		 * dip->errors is perceived to be set.
8201 		 */
8202 		smp_mb__before_atomic();
8203 	}
8204 
8205 	/* if there are more bios still pending for this dio, just exit */
8206 	if (!atomic_dec_and_test(&dip->pending_bios))
8207 		goto out;
8208 
8209 	if (dip->errors) {
8210 		bio_io_error(dip->orig_bio);
8211 	} else {
8212 		dip->dio_bio->bi_error = 0;
8213 		bio_endio(dip->orig_bio);
8214 	}
8215 out:
8216 	bio_put(bio);
8217 }
8218 
8219 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8220 				       u64 first_sector, gfp_t gfp_flags)
8221 {
8222 	struct bio *bio;
8223 	bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8224 	if (bio)
8225 		bio_associate_current(bio);
8226 	return bio;
8227 }
8228 
8229 static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8230 						 struct btrfs_dio_private *dip,
8231 						 struct bio *bio,
8232 						 u64 file_offset)
8233 {
8234 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8235 	struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8236 	int ret;
8237 
8238 	/*
8239 	 * We load all the csum data we need when we submit
8240 	 * the first bio to reduce the csum tree search and
8241 	 * contention.
8242 	 */
8243 	if (dip->logical_offset == file_offset) {
8244 		ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8245 						file_offset);
8246 		if (ret)
8247 			return ret;
8248 	}
8249 
8250 	if (bio == dip->orig_bio)
8251 		return 0;
8252 
8253 	file_offset -= dip->logical_offset;
8254 	file_offset >>= inode->i_sb->s_blocksize_bits;
8255 	io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8256 
8257 	return 0;
8258 }
8259 
8260 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8261 					 u64 file_offset, int skip_sum,
8262 					 int async_submit)
8263 {
8264 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8265 	struct btrfs_dio_private *dip = bio->bi_private;
8266 	bool write = bio_op(bio) == REQ_OP_WRITE;
8267 	int ret;
8268 
8269 	if (async_submit)
8270 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8271 
8272 	bio_get(bio);
8273 
8274 	if (!write) {
8275 		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8276 		if (ret)
8277 			goto err;
8278 	}
8279 
8280 	if (skip_sum)
8281 		goto map;
8282 
8283 	if (write && async_submit) {
8284 		ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0,
8285 					  file_offset,
8286 					  __btrfs_submit_bio_start_direct_io,
8287 					  __btrfs_submit_bio_done);
8288 		goto err;
8289 	} else if (write) {
8290 		/*
8291 		 * If we aren't doing async submit, calculate the csum of the
8292 		 * bio now.
8293 		 */
8294 		ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8295 		if (ret)
8296 			goto err;
8297 	} else {
8298 		ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8299 						     file_offset);
8300 		if (ret)
8301 			goto err;
8302 	}
8303 map:
8304 	ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8305 err:
8306 	bio_put(bio);
8307 	return ret;
8308 }
8309 
8310 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8311 				    int skip_sum)
8312 {
8313 	struct inode *inode = dip->inode;
8314 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8315 	struct btrfs_root *root = BTRFS_I(inode)->root;
8316 	struct bio *bio;
8317 	struct bio *orig_bio = dip->orig_bio;
8318 	struct bio_vec *bvec;
8319 	u64 start_sector = orig_bio->bi_iter.bi_sector;
8320 	u64 file_offset = dip->logical_offset;
8321 	u64 submit_len = 0;
8322 	u64 map_length;
8323 	u32 blocksize = fs_info->sectorsize;
8324 	int async_submit = 0;
8325 	int nr_sectors;
8326 	int ret;
8327 	int i, j;
8328 
8329 	map_length = orig_bio->bi_iter.bi_size;
8330 	ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8331 			      &map_length, NULL, 0);
8332 	if (ret)
8333 		return -EIO;
8334 
8335 	if (map_length >= orig_bio->bi_iter.bi_size) {
8336 		bio = orig_bio;
8337 		dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8338 		goto submit;
8339 	}
8340 
8341 	/* async crcs make it difficult to collect full stripe writes. */
8342 	if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8343 		async_submit = 0;
8344 	else
8345 		async_submit = 1;
8346 
8347 	bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8348 	if (!bio)
8349 		return -ENOMEM;
8350 
8351 	bio->bi_opf = orig_bio->bi_opf;
8352 	bio->bi_private = dip;
8353 	bio->bi_end_io = btrfs_end_dio_bio;
8354 	btrfs_io_bio(bio)->logical = file_offset;
8355 	atomic_inc(&dip->pending_bios);
8356 
8357 	bio_for_each_segment_all(bvec, orig_bio, j) {
8358 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8359 		i = 0;
8360 next_block:
8361 		if (unlikely(map_length < submit_len + blocksize ||
8362 		    bio_add_page(bio, bvec->bv_page, blocksize,
8363 			    bvec->bv_offset + (i * blocksize)) < blocksize)) {
8364 			/*
8365 			 * inc the count before we submit the bio so
8366 			 * we know the end IO handler won't happen before
8367 			 * we inc the count. Otherwise, the dip might get freed
8368 			 * before we're done setting it up
8369 			 */
8370 			atomic_inc(&dip->pending_bios);
8371 			ret = __btrfs_submit_dio_bio(bio, inode,
8372 						     file_offset, skip_sum,
8373 						     async_submit);
8374 			if (ret) {
8375 				bio_put(bio);
8376 				atomic_dec(&dip->pending_bios);
8377 				goto out_err;
8378 			}
8379 
8380 			start_sector += submit_len >> 9;
8381 			file_offset += submit_len;
8382 
8383 			submit_len = 0;
8384 
8385 			bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8386 						  start_sector, GFP_NOFS);
8387 			if (!bio)
8388 				goto out_err;
8389 			bio->bi_opf = orig_bio->bi_opf;
8390 			bio->bi_private = dip;
8391 			bio->bi_end_io = btrfs_end_dio_bio;
8392 			btrfs_io_bio(bio)->logical = file_offset;
8393 
8394 			map_length = orig_bio->bi_iter.bi_size;
8395 			ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8396 					      start_sector << 9,
8397 					      &map_length, NULL, 0);
8398 			if (ret) {
8399 				bio_put(bio);
8400 				goto out_err;
8401 			}
8402 
8403 			goto next_block;
8404 		} else {
8405 			submit_len += blocksize;
8406 			if (--nr_sectors) {
8407 				i++;
8408 				goto next_block;
8409 			}
8410 		}
8411 	}
8412 
8413 submit:
8414 	ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8415 				     async_submit);
8416 	if (!ret)
8417 		return 0;
8418 
8419 	bio_put(bio);
8420 out_err:
8421 	dip->errors = 1;
8422 	/*
8423 	 * before atomic variable goto zero, we must
8424 	 * make sure dip->errors is perceived to be set.
8425 	 */
8426 	smp_mb__before_atomic();
8427 	if (atomic_dec_and_test(&dip->pending_bios))
8428 		bio_io_error(dip->orig_bio);
8429 
8430 	/* bio_end_io() will handle error, so we needn't return it */
8431 	return 0;
8432 }
8433 
8434 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8435 				loff_t file_offset)
8436 {
8437 	struct btrfs_dio_private *dip = NULL;
8438 	struct bio *io_bio = NULL;
8439 	struct btrfs_io_bio *btrfs_bio;
8440 	int skip_sum;
8441 	bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8442 	int ret = 0;
8443 
8444 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8445 
8446 	io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8447 	if (!io_bio) {
8448 		ret = -ENOMEM;
8449 		goto free_ordered;
8450 	}
8451 
8452 	dip = kzalloc(sizeof(*dip), GFP_NOFS);
8453 	if (!dip) {
8454 		ret = -ENOMEM;
8455 		goto free_ordered;
8456 	}
8457 
8458 	dip->private = dio_bio->bi_private;
8459 	dip->inode = inode;
8460 	dip->logical_offset = file_offset;
8461 	dip->bytes = dio_bio->bi_iter.bi_size;
8462 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8463 	io_bio->bi_private = dip;
8464 	dip->orig_bio = io_bio;
8465 	dip->dio_bio = dio_bio;
8466 	atomic_set(&dip->pending_bios, 0);
8467 	btrfs_bio = btrfs_io_bio(io_bio);
8468 	btrfs_bio->logical = file_offset;
8469 
8470 	if (write) {
8471 		io_bio->bi_end_io = btrfs_endio_direct_write;
8472 	} else {
8473 		io_bio->bi_end_io = btrfs_endio_direct_read;
8474 		dip->subio_endio = btrfs_subio_endio_read;
8475 	}
8476 
8477 	/*
8478 	 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8479 	 * even if we fail to submit a bio, because in such case we do the
8480 	 * corresponding error handling below and it must not be done a second
8481 	 * time by btrfs_direct_IO().
8482 	 */
8483 	if (write) {
8484 		struct btrfs_dio_data *dio_data = current->journal_info;
8485 
8486 		dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8487 			dip->bytes;
8488 		dio_data->unsubmitted_oe_range_start =
8489 			dio_data->unsubmitted_oe_range_end;
8490 	}
8491 
8492 	ret = btrfs_submit_direct_hook(dip, skip_sum);
8493 	if (!ret)
8494 		return;
8495 
8496 	if (btrfs_bio->end_io)
8497 		btrfs_bio->end_io(btrfs_bio, ret);
8498 
8499 free_ordered:
8500 	/*
8501 	 * If we arrived here it means either we failed to submit the dip
8502 	 * or we either failed to clone the dio_bio or failed to allocate the
8503 	 * dip. If we cloned the dio_bio and allocated the dip, we can just
8504 	 * call bio_endio against our io_bio so that we get proper resource
8505 	 * cleanup if we fail to submit the dip, otherwise, we must do the
8506 	 * same as btrfs_endio_direct_[write|read] because we can't call these
8507 	 * callbacks - they require an allocated dip and a clone of dio_bio.
8508 	 */
8509 	if (io_bio && dip) {
8510 		io_bio->bi_error = -EIO;
8511 		bio_endio(io_bio);
8512 		/*
8513 		 * The end io callbacks free our dip, do the final put on io_bio
8514 		 * and all the cleanup and final put for dio_bio (through
8515 		 * dio_end_io()).
8516 		 */
8517 		dip = NULL;
8518 		io_bio = NULL;
8519 	} else {
8520 		if (write)
8521 			btrfs_endio_direct_write_update_ordered(inode,
8522 						file_offset,
8523 						dio_bio->bi_iter.bi_size,
8524 						0);
8525 		else
8526 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8527 			      file_offset + dio_bio->bi_iter.bi_size - 1);
8528 
8529 		dio_bio->bi_error = -EIO;
8530 		/*
8531 		 * Releases and cleans up our dio_bio, no need to bio_put()
8532 		 * nor bio_endio()/bio_io_error() against dio_bio.
8533 		 */
8534 		dio_end_io(dio_bio, ret);
8535 	}
8536 	if (io_bio)
8537 		bio_put(io_bio);
8538 	kfree(dip);
8539 }
8540 
8541 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8542 			       struct kiocb *iocb,
8543 			       const struct iov_iter *iter, loff_t offset)
8544 {
8545 	int seg;
8546 	int i;
8547 	unsigned int blocksize_mask = fs_info->sectorsize - 1;
8548 	ssize_t retval = -EINVAL;
8549 
8550 	if (offset & blocksize_mask)
8551 		goto out;
8552 
8553 	if (iov_iter_alignment(iter) & blocksize_mask)
8554 		goto out;
8555 
8556 	/* If this is a write we don't need to check anymore */
8557 	if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8558 		return 0;
8559 	/*
8560 	 * Check to make sure we don't have duplicate iov_base's in this
8561 	 * iovec, if so return EINVAL, otherwise we'll get csum errors
8562 	 * when reading back.
8563 	 */
8564 	for (seg = 0; seg < iter->nr_segs; seg++) {
8565 		for (i = seg + 1; i < iter->nr_segs; i++) {
8566 			if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8567 				goto out;
8568 		}
8569 	}
8570 	retval = 0;
8571 out:
8572 	return retval;
8573 }
8574 
8575 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8576 {
8577 	struct file *file = iocb->ki_filp;
8578 	struct inode *inode = file->f_mapping->host;
8579 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8580 	struct btrfs_dio_data dio_data = { 0 };
8581 	loff_t offset = iocb->ki_pos;
8582 	size_t count = 0;
8583 	int flags = 0;
8584 	bool wakeup = true;
8585 	bool relock = false;
8586 	ssize_t ret;
8587 
8588 	if (check_direct_IO(fs_info, iocb, iter, offset))
8589 		return 0;
8590 
8591 	inode_dio_begin(inode);
8592 	smp_mb__after_atomic();
8593 
8594 	/*
8595 	 * The generic stuff only does filemap_write_and_wait_range, which
8596 	 * isn't enough if we've written compressed pages to this area, so
8597 	 * we need to flush the dirty pages again to make absolutely sure
8598 	 * that any outstanding dirty pages are on disk.
8599 	 */
8600 	count = iov_iter_count(iter);
8601 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8602 		     &BTRFS_I(inode)->runtime_flags))
8603 		filemap_fdatawrite_range(inode->i_mapping, offset,
8604 					 offset + count - 1);
8605 
8606 	if (iov_iter_rw(iter) == WRITE) {
8607 		/*
8608 		 * If the write DIO is beyond the EOF, we need update
8609 		 * the isize, but it is protected by i_mutex. So we can
8610 		 * not unlock the i_mutex at this case.
8611 		 */
8612 		if (offset + count <= inode->i_size) {
8613 			dio_data.overwrite = 1;
8614 			inode_unlock(inode);
8615 			relock = true;
8616 		}
8617 		ret = btrfs_delalloc_reserve_space(inode, offset, count);
8618 		if (ret)
8619 			goto out;
8620 		dio_data.outstanding_extents = count_max_extents(count);
8621 
8622 		/*
8623 		 * We need to know how many extents we reserved so that we can
8624 		 * do the accounting properly if we go over the number we
8625 		 * originally calculated.  Abuse current->journal_info for this.
8626 		 */
8627 		dio_data.reserve = round_up(count,
8628 					    fs_info->sectorsize);
8629 		dio_data.unsubmitted_oe_range_start = (u64)offset;
8630 		dio_data.unsubmitted_oe_range_end = (u64)offset;
8631 		current->journal_info = &dio_data;
8632 		down_read(&BTRFS_I(inode)->dio_sem);
8633 	} else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8634 				     &BTRFS_I(inode)->runtime_flags)) {
8635 		inode_dio_end(inode);
8636 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
8637 		wakeup = false;
8638 	}
8639 
8640 	ret = __blockdev_direct_IO(iocb, inode,
8641 				   fs_info->fs_devices->latest_bdev,
8642 				   iter, btrfs_get_blocks_direct, NULL,
8643 				   btrfs_submit_direct, flags);
8644 	if (iov_iter_rw(iter) == WRITE) {
8645 		up_read(&BTRFS_I(inode)->dio_sem);
8646 		current->journal_info = NULL;
8647 		if (ret < 0 && ret != -EIOCBQUEUED) {
8648 			if (dio_data.reserve)
8649 				btrfs_delalloc_release_space(inode, offset,
8650 							     dio_data.reserve);
8651 			/*
8652 			 * On error we might have left some ordered extents
8653 			 * without submitting corresponding bios for them, so
8654 			 * cleanup them up to avoid other tasks getting them
8655 			 * and waiting for them to complete forever.
8656 			 */
8657 			if (dio_data.unsubmitted_oe_range_start <
8658 			    dio_data.unsubmitted_oe_range_end)
8659 				btrfs_endio_direct_write_update_ordered(inode,
8660 					dio_data.unsubmitted_oe_range_start,
8661 					dio_data.unsubmitted_oe_range_end -
8662 					dio_data.unsubmitted_oe_range_start,
8663 					0);
8664 		} else if (ret >= 0 && (size_t)ret < count)
8665 			btrfs_delalloc_release_space(inode, offset,
8666 						     count - (size_t)ret);
8667 	}
8668 out:
8669 	if (wakeup)
8670 		inode_dio_end(inode);
8671 	if (relock)
8672 		inode_lock(inode);
8673 
8674 	return ret;
8675 }
8676 
8677 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
8678 
8679 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8680 		__u64 start, __u64 len)
8681 {
8682 	int	ret;
8683 
8684 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8685 	if (ret)
8686 		return ret;
8687 
8688 	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8689 }
8690 
8691 int btrfs_readpage(struct file *file, struct page *page)
8692 {
8693 	struct extent_io_tree *tree;
8694 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8695 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8696 }
8697 
8698 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8699 {
8700 	struct extent_io_tree *tree;
8701 	struct inode *inode = page->mapping->host;
8702 	int ret;
8703 
8704 	if (current->flags & PF_MEMALLOC) {
8705 		redirty_page_for_writepage(wbc, page);
8706 		unlock_page(page);
8707 		return 0;
8708 	}
8709 
8710 	/*
8711 	 * If we are under memory pressure we will call this directly from the
8712 	 * VM, we need to make sure we have the inode referenced for the ordered
8713 	 * extent.  If not just return like we didn't do anything.
8714 	 */
8715 	if (!igrab(inode)) {
8716 		redirty_page_for_writepage(wbc, page);
8717 		return AOP_WRITEPAGE_ACTIVATE;
8718 	}
8719 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8720 	ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8721 	btrfs_add_delayed_iput(inode);
8722 	return ret;
8723 }
8724 
8725 static int btrfs_writepages(struct address_space *mapping,
8726 			    struct writeback_control *wbc)
8727 {
8728 	struct extent_io_tree *tree;
8729 
8730 	tree = &BTRFS_I(mapping->host)->io_tree;
8731 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8732 }
8733 
8734 static int
8735 btrfs_readpages(struct file *file, struct address_space *mapping,
8736 		struct list_head *pages, unsigned nr_pages)
8737 {
8738 	struct extent_io_tree *tree;
8739 	tree = &BTRFS_I(mapping->host)->io_tree;
8740 	return extent_readpages(tree, mapping, pages, nr_pages,
8741 				btrfs_get_extent);
8742 }
8743 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8744 {
8745 	struct extent_io_tree *tree;
8746 	struct extent_map_tree *map;
8747 	int ret;
8748 
8749 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8750 	map = &BTRFS_I(page->mapping->host)->extent_tree;
8751 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8752 	if (ret == 1) {
8753 		ClearPagePrivate(page);
8754 		set_page_private(page, 0);
8755 		put_page(page);
8756 	}
8757 	return ret;
8758 }
8759 
8760 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8761 {
8762 	if (PageWriteback(page) || PageDirty(page))
8763 		return 0;
8764 	return __btrfs_releasepage(page, gfp_flags);
8765 }
8766 
8767 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8768 				 unsigned int length)
8769 {
8770 	struct inode *inode = page->mapping->host;
8771 	struct extent_io_tree *tree;
8772 	struct btrfs_ordered_extent *ordered;
8773 	struct extent_state *cached_state = NULL;
8774 	u64 page_start = page_offset(page);
8775 	u64 page_end = page_start + PAGE_SIZE - 1;
8776 	u64 start;
8777 	u64 end;
8778 	int inode_evicting = inode->i_state & I_FREEING;
8779 
8780 	/*
8781 	 * we have the page locked, so new writeback can't start,
8782 	 * and the dirty bit won't be cleared while we are here.
8783 	 *
8784 	 * Wait for IO on this page so that we can safely clear
8785 	 * the PagePrivate2 bit and do ordered accounting
8786 	 */
8787 	wait_on_page_writeback(page);
8788 
8789 	tree = &BTRFS_I(inode)->io_tree;
8790 	if (offset) {
8791 		btrfs_releasepage(page, GFP_NOFS);
8792 		return;
8793 	}
8794 
8795 	if (!inode_evicting)
8796 		lock_extent_bits(tree, page_start, page_end, &cached_state);
8797 again:
8798 	start = page_start;
8799 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8800 					page_end - start + 1);
8801 	if (ordered) {
8802 		end = min(page_end, ordered->file_offset + ordered->len - 1);
8803 		/*
8804 		 * IO on this page will never be started, so we need
8805 		 * to account for any ordered extents now
8806 		 */
8807 		if (!inode_evicting)
8808 			clear_extent_bit(tree, start, end,
8809 					 EXTENT_DIRTY | EXTENT_DELALLOC |
8810 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8811 					 EXTENT_DEFRAG, 1, 0, &cached_state,
8812 					 GFP_NOFS);
8813 		/*
8814 		 * whoever cleared the private bit is responsible
8815 		 * for the finish_ordered_io
8816 		 */
8817 		if (TestClearPagePrivate2(page)) {
8818 			struct btrfs_ordered_inode_tree *tree;
8819 			u64 new_len;
8820 
8821 			tree = &BTRFS_I(inode)->ordered_tree;
8822 
8823 			spin_lock_irq(&tree->lock);
8824 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8825 			new_len = start - ordered->file_offset;
8826 			if (new_len < ordered->truncated_len)
8827 				ordered->truncated_len = new_len;
8828 			spin_unlock_irq(&tree->lock);
8829 
8830 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
8831 							   start,
8832 							   end - start + 1, 1))
8833 				btrfs_finish_ordered_io(ordered);
8834 		}
8835 		btrfs_put_ordered_extent(ordered);
8836 		if (!inode_evicting) {
8837 			cached_state = NULL;
8838 			lock_extent_bits(tree, start, end,
8839 					 &cached_state);
8840 		}
8841 
8842 		start = end + 1;
8843 		if (start < page_end)
8844 			goto again;
8845 	}
8846 
8847 	/*
8848 	 * Qgroup reserved space handler
8849 	 * Page here will be either
8850 	 * 1) Already written to disk
8851 	 *    In this case, its reserved space is released from data rsv map
8852 	 *    and will be freed by delayed_ref handler finally.
8853 	 *    So even we call qgroup_free_data(), it won't decrease reserved
8854 	 *    space.
8855 	 * 2) Not written to disk
8856 	 *    This means the reserved space should be freed here. However,
8857 	 *    if a truncate invalidates the page (by clearing PageDirty)
8858 	 *    and the page is accounted for while allocating extent
8859 	 *    in btrfs_check_data_free_space() we let delayed_ref to
8860 	 *    free the entire extent.
8861 	 */
8862 	if (PageDirty(page))
8863 		btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8864 	if (!inode_evicting) {
8865 		clear_extent_bit(tree, page_start, page_end,
8866 				 EXTENT_LOCKED | EXTENT_DIRTY |
8867 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8868 				 EXTENT_DEFRAG, 1, 1,
8869 				 &cached_state, GFP_NOFS);
8870 
8871 		__btrfs_releasepage(page, GFP_NOFS);
8872 	}
8873 
8874 	ClearPageChecked(page);
8875 	if (PagePrivate(page)) {
8876 		ClearPagePrivate(page);
8877 		set_page_private(page, 0);
8878 		put_page(page);
8879 	}
8880 }
8881 
8882 /*
8883  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8884  * called from a page fault handler when a page is first dirtied. Hence we must
8885  * be careful to check for EOF conditions here. We set the page up correctly
8886  * for a written page which means we get ENOSPC checking when writing into
8887  * holes and correct delalloc and unwritten extent mapping on filesystems that
8888  * support these features.
8889  *
8890  * We are not allowed to take the i_mutex here so we have to play games to
8891  * protect against truncate races as the page could now be beyond EOF.  Because
8892  * vmtruncate() writes the inode size before removing pages, once we have the
8893  * page lock we can determine safely if the page is beyond EOF. If it is not
8894  * beyond EOF, then the page is guaranteed safe against truncation until we
8895  * unlock the page.
8896  */
8897 int btrfs_page_mkwrite(struct vm_fault *vmf)
8898 {
8899 	struct page *page = vmf->page;
8900 	struct inode *inode = file_inode(vmf->vma->vm_file);
8901 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8902 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8903 	struct btrfs_ordered_extent *ordered;
8904 	struct extent_state *cached_state = NULL;
8905 	char *kaddr;
8906 	unsigned long zero_start;
8907 	loff_t size;
8908 	int ret;
8909 	int reserved = 0;
8910 	u64 reserved_space;
8911 	u64 page_start;
8912 	u64 page_end;
8913 	u64 end;
8914 
8915 	reserved_space = PAGE_SIZE;
8916 
8917 	sb_start_pagefault(inode->i_sb);
8918 	page_start = page_offset(page);
8919 	page_end = page_start + PAGE_SIZE - 1;
8920 	end = page_end;
8921 
8922 	/*
8923 	 * Reserving delalloc space after obtaining the page lock can lead to
8924 	 * deadlock. For example, if a dirty page is locked by this function
8925 	 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8926 	 * dirty page write out, then the btrfs_writepage() function could
8927 	 * end up waiting indefinitely to get a lock on the page currently
8928 	 * being processed by btrfs_page_mkwrite() function.
8929 	 */
8930 	ret = btrfs_delalloc_reserve_space(inode, page_start,
8931 					   reserved_space);
8932 	if (!ret) {
8933 		ret = file_update_time(vmf->vma->vm_file);
8934 		reserved = 1;
8935 	}
8936 	if (ret) {
8937 		if (ret == -ENOMEM)
8938 			ret = VM_FAULT_OOM;
8939 		else /* -ENOSPC, -EIO, etc */
8940 			ret = VM_FAULT_SIGBUS;
8941 		if (reserved)
8942 			goto out;
8943 		goto out_noreserve;
8944 	}
8945 
8946 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8947 again:
8948 	lock_page(page);
8949 	size = i_size_read(inode);
8950 
8951 	if ((page->mapping != inode->i_mapping) ||
8952 	    (page_start >= size)) {
8953 		/* page got truncated out from underneath us */
8954 		goto out_unlock;
8955 	}
8956 	wait_on_page_writeback(page);
8957 
8958 	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8959 	set_page_extent_mapped(page);
8960 
8961 	/*
8962 	 * we can't set the delalloc bits if there are pending ordered
8963 	 * extents.  Drop our locks and wait for them to finish
8964 	 */
8965 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8966 			PAGE_SIZE);
8967 	if (ordered) {
8968 		unlock_extent_cached(io_tree, page_start, page_end,
8969 				     &cached_state, GFP_NOFS);
8970 		unlock_page(page);
8971 		btrfs_start_ordered_extent(inode, ordered, 1);
8972 		btrfs_put_ordered_extent(ordered);
8973 		goto again;
8974 	}
8975 
8976 	if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8977 		reserved_space = round_up(size - page_start,
8978 					  fs_info->sectorsize);
8979 		if (reserved_space < PAGE_SIZE) {
8980 			end = page_start + reserved_space - 1;
8981 			spin_lock(&BTRFS_I(inode)->lock);
8982 			BTRFS_I(inode)->outstanding_extents++;
8983 			spin_unlock(&BTRFS_I(inode)->lock);
8984 			btrfs_delalloc_release_space(inode, page_start,
8985 						PAGE_SIZE - reserved_space);
8986 		}
8987 	}
8988 
8989 	/*
8990 	 * page_mkwrite gets called when the page is firstly dirtied after it's
8991 	 * faulted in, but write(2) could also dirty a page and set delalloc
8992 	 * bits, thus in this case for space account reason, we still need to
8993 	 * clear any delalloc bits within this page range since we have to
8994 	 * reserve data&meta space before lock_page() (see above comments).
8995 	 */
8996 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8997 			  EXTENT_DIRTY | EXTENT_DELALLOC |
8998 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8999 			  0, 0, &cached_state, GFP_NOFS);
9000 
9001 	ret = btrfs_set_extent_delalloc(inode, page_start, end,
9002 					&cached_state, 0);
9003 	if (ret) {
9004 		unlock_extent_cached(io_tree, page_start, page_end,
9005 				     &cached_state, GFP_NOFS);
9006 		ret = VM_FAULT_SIGBUS;
9007 		goto out_unlock;
9008 	}
9009 	ret = 0;
9010 
9011 	/* page is wholly or partially inside EOF */
9012 	if (page_start + PAGE_SIZE > size)
9013 		zero_start = size & ~PAGE_MASK;
9014 	else
9015 		zero_start = PAGE_SIZE;
9016 
9017 	if (zero_start != PAGE_SIZE) {
9018 		kaddr = kmap(page);
9019 		memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9020 		flush_dcache_page(page);
9021 		kunmap(page);
9022 	}
9023 	ClearPageChecked(page);
9024 	set_page_dirty(page);
9025 	SetPageUptodate(page);
9026 
9027 	BTRFS_I(inode)->last_trans = fs_info->generation;
9028 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9029 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9030 
9031 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9032 
9033 out_unlock:
9034 	if (!ret) {
9035 		sb_end_pagefault(inode->i_sb);
9036 		return VM_FAULT_LOCKED;
9037 	}
9038 	unlock_page(page);
9039 out:
9040 	btrfs_delalloc_release_space(inode, page_start, reserved_space);
9041 out_noreserve:
9042 	sb_end_pagefault(inode->i_sb);
9043 	return ret;
9044 }
9045 
9046 static int btrfs_truncate(struct inode *inode)
9047 {
9048 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9049 	struct btrfs_root *root = BTRFS_I(inode)->root;
9050 	struct btrfs_block_rsv *rsv;
9051 	int ret = 0;
9052 	int err = 0;
9053 	struct btrfs_trans_handle *trans;
9054 	u64 mask = fs_info->sectorsize - 1;
9055 	u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9056 
9057 	ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9058 				       (u64)-1);
9059 	if (ret)
9060 		return ret;
9061 
9062 	/*
9063 	 * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9064 	 * 3 things going on here
9065 	 *
9066 	 * 1) We need to reserve space for our orphan item and the space to
9067 	 * delete our orphan item.  Lord knows we don't want to have a dangling
9068 	 * orphan item because we didn't reserve space to remove it.
9069 	 *
9070 	 * 2) We need to reserve space to update our inode.
9071 	 *
9072 	 * 3) We need to have something to cache all the space that is going to
9073 	 * be free'd up by the truncate operation, but also have some slack
9074 	 * space reserved in case it uses space during the truncate (thank you
9075 	 * very much snapshotting).
9076 	 *
9077 	 * And we need these to all be separate.  The fact is we can use a lot of
9078 	 * space doing the truncate, and we have no earthly idea how much space
9079 	 * we will use, so we need the truncate reservation to be separate so it
9080 	 * doesn't end up using space reserved for updating the inode or
9081 	 * removing the orphan item.  We also need to be able to stop the
9082 	 * transaction and start a new one, which means we need to be able to
9083 	 * update the inode several times, and we have no idea of knowing how
9084 	 * many times that will be, so we can't just reserve 1 item for the
9085 	 * entirety of the operation, so that has to be done separately as well.
9086 	 * Then there is the orphan item, which does indeed need to be held on
9087 	 * to for the whole operation, and we need nobody to touch this reserved
9088 	 * space except the orphan code.
9089 	 *
9090 	 * So that leaves us with
9091 	 *
9092 	 * 1) root->orphan_block_rsv - for the orphan deletion.
9093 	 * 2) rsv - for the truncate reservation, which we will steal from the
9094 	 * transaction reservation.
9095 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9096 	 * updating the inode.
9097 	 */
9098 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9099 	if (!rsv)
9100 		return -ENOMEM;
9101 	rsv->size = min_size;
9102 	rsv->failfast = 1;
9103 
9104 	/*
9105 	 * 1 for the truncate slack space
9106 	 * 1 for updating the inode.
9107 	 */
9108 	trans = btrfs_start_transaction(root, 2);
9109 	if (IS_ERR(trans)) {
9110 		err = PTR_ERR(trans);
9111 		goto out;
9112 	}
9113 
9114 	/* Migrate the slack space for the truncate to our reserve */
9115 	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9116 				      min_size, 0);
9117 	BUG_ON(ret);
9118 
9119 	/*
9120 	 * So if we truncate and then write and fsync we normally would just
9121 	 * write the extents that changed, which is a problem if we need to
9122 	 * first truncate that entire inode.  So set this flag so we write out
9123 	 * all of the extents in the inode to the sync log so we're completely
9124 	 * safe.
9125 	 */
9126 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9127 	trans->block_rsv = rsv;
9128 
9129 	while (1) {
9130 		ret = btrfs_truncate_inode_items(trans, root, inode,
9131 						 inode->i_size,
9132 						 BTRFS_EXTENT_DATA_KEY);
9133 		if (ret != -ENOSPC && ret != -EAGAIN) {
9134 			err = ret;
9135 			break;
9136 		}
9137 
9138 		trans->block_rsv = &fs_info->trans_block_rsv;
9139 		ret = btrfs_update_inode(trans, root, inode);
9140 		if (ret) {
9141 			err = ret;
9142 			break;
9143 		}
9144 
9145 		btrfs_end_transaction(trans);
9146 		btrfs_btree_balance_dirty(fs_info);
9147 
9148 		trans = btrfs_start_transaction(root, 2);
9149 		if (IS_ERR(trans)) {
9150 			ret = err = PTR_ERR(trans);
9151 			trans = NULL;
9152 			break;
9153 		}
9154 
9155 		btrfs_block_rsv_release(fs_info, rsv, -1);
9156 		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9157 					      rsv, min_size, 0);
9158 		BUG_ON(ret);	/* shouldn't happen */
9159 		trans->block_rsv = rsv;
9160 	}
9161 
9162 	if (ret == 0 && inode->i_nlink > 0) {
9163 		trans->block_rsv = root->orphan_block_rsv;
9164 		ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9165 		if (ret)
9166 			err = ret;
9167 	}
9168 
9169 	if (trans) {
9170 		trans->block_rsv = &fs_info->trans_block_rsv;
9171 		ret = btrfs_update_inode(trans, root, inode);
9172 		if (ret && !err)
9173 			err = ret;
9174 
9175 		ret = btrfs_end_transaction(trans);
9176 		btrfs_btree_balance_dirty(fs_info);
9177 	}
9178 out:
9179 	btrfs_free_block_rsv(fs_info, rsv);
9180 
9181 	if (ret && !err)
9182 		err = ret;
9183 
9184 	return err;
9185 }
9186 
9187 /*
9188  * create a new subvolume directory/inode (helper for the ioctl).
9189  */
9190 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9191 			     struct btrfs_root *new_root,
9192 			     struct btrfs_root *parent_root,
9193 			     u64 new_dirid)
9194 {
9195 	struct inode *inode;
9196 	int err;
9197 	u64 index = 0;
9198 
9199 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9200 				new_dirid, new_dirid,
9201 				S_IFDIR | (~current_umask() & S_IRWXUGO),
9202 				&index);
9203 	if (IS_ERR(inode))
9204 		return PTR_ERR(inode);
9205 	inode->i_op = &btrfs_dir_inode_operations;
9206 	inode->i_fop = &btrfs_dir_file_operations;
9207 
9208 	set_nlink(inode, 1);
9209 	btrfs_i_size_write(BTRFS_I(inode), 0);
9210 	unlock_new_inode(inode);
9211 
9212 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9213 	if (err)
9214 		btrfs_err(new_root->fs_info,
9215 			  "error inheriting subvolume %llu properties: %d",
9216 			  new_root->root_key.objectid, err);
9217 
9218 	err = btrfs_update_inode(trans, new_root, inode);
9219 
9220 	iput(inode);
9221 	return err;
9222 }
9223 
9224 struct inode *btrfs_alloc_inode(struct super_block *sb)
9225 {
9226 	struct btrfs_inode *ei;
9227 	struct inode *inode;
9228 
9229 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9230 	if (!ei)
9231 		return NULL;
9232 
9233 	ei->root = NULL;
9234 	ei->generation = 0;
9235 	ei->last_trans = 0;
9236 	ei->last_sub_trans = 0;
9237 	ei->logged_trans = 0;
9238 	ei->delalloc_bytes = 0;
9239 	ei->defrag_bytes = 0;
9240 	ei->disk_i_size = 0;
9241 	ei->flags = 0;
9242 	ei->csum_bytes = 0;
9243 	ei->index_cnt = (u64)-1;
9244 	ei->dir_index = 0;
9245 	ei->last_unlink_trans = 0;
9246 	ei->last_log_commit = 0;
9247 	ei->delayed_iput_count = 0;
9248 
9249 	spin_lock_init(&ei->lock);
9250 	ei->outstanding_extents = 0;
9251 	ei->reserved_extents = 0;
9252 
9253 	ei->runtime_flags = 0;
9254 	ei->force_compress = BTRFS_COMPRESS_NONE;
9255 
9256 	ei->delayed_node = NULL;
9257 
9258 	ei->i_otime.tv_sec = 0;
9259 	ei->i_otime.tv_nsec = 0;
9260 
9261 	inode = &ei->vfs_inode;
9262 	extent_map_tree_init(&ei->extent_tree);
9263 	extent_io_tree_init(&ei->io_tree, &inode->i_data);
9264 	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9265 	ei->io_tree.track_uptodate = 1;
9266 	ei->io_failure_tree.track_uptodate = 1;
9267 	atomic_set(&ei->sync_writers, 0);
9268 	mutex_init(&ei->log_mutex);
9269 	mutex_init(&ei->delalloc_mutex);
9270 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9271 	INIT_LIST_HEAD(&ei->delalloc_inodes);
9272 	INIT_LIST_HEAD(&ei->delayed_iput);
9273 	RB_CLEAR_NODE(&ei->rb_node);
9274 	init_rwsem(&ei->dio_sem);
9275 
9276 	return inode;
9277 }
9278 
9279 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9280 void btrfs_test_destroy_inode(struct inode *inode)
9281 {
9282 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9283 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9284 }
9285 #endif
9286 
9287 static void btrfs_i_callback(struct rcu_head *head)
9288 {
9289 	struct inode *inode = container_of(head, struct inode, i_rcu);
9290 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9291 }
9292 
9293 void btrfs_destroy_inode(struct inode *inode)
9294 {
9295 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9296 	struct btrfs_ordered_extent *ordered;
9297 	struct btrfs_root *root = BTRFS_I(inode)->root;
9298 
9299 	WARN_ON(!hlist_empty(&inode->i_dentry));
9300 	WARN_ON(inode->i_data.nrpages);
9301 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
9302 	WARN_ON(BTRFS_I(inode)->reserved_extents);
9303 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9304 	WARN_ON(BTRFS_I(inode)->csum_bytes);
9305 	WARN_ON(BTRFS_I(inode)->defrag_bytes);
9306 
9307 	/*
9308 	 * This can happen where we create an inode, but somebody else also
9309 	 * created the same inode and we need to destroy the one we already
9310 	 * created.
9311 	 */
9312 	if (!root)
9313 		goto free;
9314 
9315 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9316 		     &BTRFS_I(inode)->runtime_flags)) {
9317 		btrfs_info(fs_info, "inode %llu still on the orphan list",
9318 			   btrfs_ino(BTRFS_I(inode)));
9319 		atomic_dec(&root->orphan_inodes);
9320 	}
9321 
9322 	while (1) {
9323 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9324 		if (!ordered)
9325 			break;
9326 		else {
9327 			btrfs_err(fs_info,
9328 				  "found ordered extent %llu %llu on inode cleanup",
9329 				  ordered->file_offset, ordered->len);
9330 			btrfs_remove_ordered_extent(inode, ordered);
9331 			btrfs_put_ordered_extent(ordered);
9332 			btrfs_put_ordered_extent(ordered);
9333 		}
9334 	}
9335 	btrfs_qgroup_check_reserved_leak(inode);
9336 	inode_tree_del(inode);
9337 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9338 free:
9339 	call_rcu(&inode->i_rcu, btrfs_i_callback);
9340 }
9341 
9342 int btrfs_drop_inode(struct inode *inode)
9343 {
9344 	struct btrfs_root *root = BTRFS_I(inode)->root;
9345 
9346 	if (root == NULL)
9347 		return 1;
9348 
9349 	/* the snap/subvol tree is on deleting */
9350 	if (btrfs_root_refs(&root->root_item) == 0)
9351 		return 1;
9352 	else
9353 		return generic_drop_inode(inode);
9354 }
9355 
9356 static void init_once(void *foo)
9357 {
9358 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9359 
9360 	inode_init_once(&ei->vfs_inode);
9361 }
9362 
9363 void btrfs_destroy_cachep(void)
9364 {
9365 	/*
9366 	 * Make sure all delayed rcu free inodes are flushed before we
9367 	 * destroy cache.
9368 	 */
9369 	rcu_barrier();
9370 	kmem_cache_destroy(btrfs_inode_cachep);
9371 	kmem_cache_destroy(btrfs_trans_handle_cachep);
9372 	kmem_cache_destroy(btrfs_transaction_cachep);
9373 	kmem_cache_destroy(btrfs_path_cachep);
9374 	kmem_cache_destroy(btrfs_free_space_cachep);
9375 }
9376 
9377 int btrfs_init_cachep(void)
9378 {
9379 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9380 			sizeof(struct btrfs_inode), 0,
9381 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9382 			init_once);
9383 	if (!btrfs_inode_cachep)
9384 		goto fail;
9385 
9386 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9387 			sizeof(struct btrfs_trans_handle), 0,
9388 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9389 	if (!btrfs_trans_handle_cachep)
9390 		goto fail;
9391 
9392 	btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9393 			sizeof(struct btrfs_transaction), 0,
9394 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9395 	if (!btrfs_transaction_cachep)
9396 		goto fail;
9397 
9398 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
9399 			sizeof(struct btrfs_path), 0,
9400 			SLAB_MEM_SPREAD, NULL);
9401 	if (!btrfs_path_cachep)
9402 		goto fail;
9403 
9404 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9405 			sizeof(struct btrfs_free_space), 0,
9406 			SLAB_MEM_SPREAD, NULL);
9407 	if (!btrfs_free_space_cachep)
9408 		goto fail;
9409 
9410 	return 0;
9411 fail:
9412 	btrfs_destroy_cachep();
9413 	return -ENOMEM;
9414 }
9415 
9416 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9417 			 u32 request_mask, unsigned int flags)
9418 {
9419 	u64 delalloc_bytes;
9420 	struct inode *inode = d_inode(path->dentry);
9421 	u32 blocksize = inode->i_sb->s_blocksize;
9422 
9423 	generic_fillattr(inode, stat);
9424 	stat->dev = BTRFS_I(inode)->root->anon_dev;
9425 
9426 	spin_lock(&BTRFS_I(inode)->lock);
9427 	delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9428 	spin_unlock(&BTRFS_I(inode)->lock);
9429 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9430 			ALIGN(delalloc_bytes, blocksize)) >> 9;
9431 	return 0;
9432 }
9433 
9434 static int btrfs_rename_exchange(struct inode *old_dir,
9435 			      struct dentry *old_dentry,
9436 			      struct inode *new_dir,
9437 			      struct dentry *new_dentry)
9438 {
9439 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9440 	struct btrfs_trans_handle *trans;
9441 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9442 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9443 	struct inode *new_inode = new_dentry->d_inode;
9444 	struct inode *old_inode = old_dentry->d_inode;
9445 	struct timespec ctime = current_time(old_inode);
9446 	struct dentry *parent;
9447 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9448 	u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9449 	u64 old_idx = 0;
9450 	u64 new_idx = 0;
9451 	u64 root_objectid;
9452 	int ret;
9453 	bool root_log_pinned = false;
9454 	bool dest_log_pinned = false;
9455 
9456 	/* we only allow rename subvolume link between subvolumes */
9457 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9458 		return -EXDEV;
9459 
9460 	/* close the race window with snapshot create/destroy ioctl */
9461 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9462 		down_read(&fs_info->subvol_sem);
9463 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9464 		down_read(&fs_info->subvol_sem);
9465 
9466 	/*
9467 	 * We want to reserve the absolute worst case amount of items.  So if
9468 	 * both inodes are subvols and we need to unlink them then that would
9469 	 * require 4 item modifications, but if they are both normal inodes it
9470 	 * would require 5 item modifications, so we'll assume their normal
9471 	 * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9472 	 * should cover the worst case number of items we'll modify.
9473 	 */
9474 	trans = btrfs_start_transaction(root, 12);
9475 	if (IS_ERR(trans)) {
9476 		ret = PTR_ERR(trans);
9477 		goto out_notrans;
9478 	}
9479 
9480 	/*
9481 	 * We need to find a free sequence number both in the source and
9482 	 * in the destination directory for the exchange.
9483 	 */
9484 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9485 	if (ret)
9486 		goto out_fail;
9487 	ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9488 	if (ret)
9489 		goto out_fail;
9490 
9491 	BTRFS_I(old_inode)->dir_index = 0ULL;
9492 	BTRFS_I(new_inode)->dir_index = 0ULL;
9493 
9494 	/* Reference for the source. */
9495 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9496 		/* force full log commit if subvolume involved. */
9497 		btrfs_set_log_full_commit(fs_info, trans);
9498 	} else {
9499 		btrfs_pin_log_trans(root);
9500 		root_log_pinned = true;
9501 		ret = btrfs_insert_inode_ref(trans, dest,
9502 					     new_dentry->d_name.name,
9503 					     new_dentry->d_name.len,
9504 					     old_ino,
9505 					     btrfs_ino(BTRFS_I(new_dir)),
9506 					     old_idx);
9507 		if (ret)
9508 			goto out_fail;
9509 	}
9510 
9511 	/* And now for the dest. */
9512 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9513 		/* force full log commit if subvolume involved. */
9514 		btrfs_set_log_full_commit(fs_info, trans);
9515 	} else {
9516 		btrfs_pin_log_trans(dest);
9517 		dest_log_pinned = true;
9518 		ret = btrfs_insert_inode_ref(trans, root,
9519 					     old_dentry->d_name.name,
9520 					     old_dentry->d_name.len,
9521 					     new_ino,
9522 					     btrfs_ino(BTRFS_I(old_dir)),
9523 					     new_idx);
9524 		if (ret)
9525 			goto out_fail;
9526 	}
9527 
9528 	/* Update inode version and ctime/mtime. */
9529 	inode_inc_iversion(old_dir);
9530 	inode_inc_iversion(new_dir);
9531 	inode_inc_iversion(old_inode);
9532 	inode_inc_iversion(new_inode);
9533 	old_dir->i_ctime = old_dir->i_mtime = ctime;
9534 	new_dir->i_ctime = new_dir->i_mtime = ctime;
9535 	old_inode->i_ctime = ctime;
9536 	new_inode->i_ctime = ctime;
9537 
9538 	if (old_dentry->d_parent != new_dentry->d_parent) {
9539 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9540 				BTRFS_I(old_inode), 1);
9541 		btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9542 				BTRFS_I(new_inode), 1);
9543 	}
9544 
9545 	/* src is a subvolume */
9546 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9547 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9548 		ret = btrfs_unlink_subvol(trans, root, old_dir,
9549 					  root_objectid,
9550 					  old_dentry->d_name.name,
9551 					  old_dentry->d_name.len);
9552 	} else { /* src is an inode */
9553 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9554 					   BTRFS_I(old_dentry->d_inode),
9555 					   old_dentry->d_name.name,
9556 					   old_dentry->d_name.len);
9557 		if (!ret)
9558 			ret = btrfs_update_inode(trans, root, old_inode);
9559 	}
9560 	if (ret) {
9561 		btrfs_abort_transaction(trans, ret);
9562 		goto out_fail;
9563 	}
9564 
9565 	/* dest is a subvolume */
9566 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9567 		root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9568 		ret = btrfs_unlink_subvol(trans, dest, new_dir,
9569 					  root_objectid,
9570 					  new_dentry->d_name.name,
9571 					  new_dentry->d_name.len);
9572 	} else { /* dest is an inode */
9573 		ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9574 					   BTRFS_I(new_dentry->d_inode),
9575 					   new_dentry->d_name.name,
9576 					   new_dentry->d_name.len);
9577 		if (!ret)
9578 			ret = btrfs_update_inode(trans, dest, new_inode);
9579 	}
9580 	if (ret) {
9581 		btrfs_abort_transaction(trans, ret);
9582 		goto out_fail;
9583 	}
9584 
9585 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9586 			     new_dentry->d_name.name,
9587 			     new_dentry->d_name.len, 0, old_idx);
9588 	if (ret) {
9589 		btrfs_abort_transaction(trans, ret);
9590 		goto out_fail;
9591 	}
9592 
9593 	ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9594 			     old_dentry->d_name.name,
9595 			     old_dentry->d_name.len, 0, new_idx);
9596 	if (ret) {
9597 		btrfs_abort_transaction(trans, ret);
9598 		goto out_fail;
9599 	}
9600 
9601 	if (old_inode->i_nlink == 1)
9602 		BTRFS_I(old_inode)->dir_index = old_idx;
9603 	if (new_inode->i_nlink == 1)
9604 		BTRFS_I(new_inode)->dir_index = new_idx;
9605 
9606 	if (root_log_pinned) {
9607 		parent = new_dentry->d_parent;
9608 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9609 				parent);
9610 		btrfs_end_log_trans(root);
9611 		root_log_pinned = false;
9612 	}
9613 	if (dest_log_pinned) {
9614 		parent = old_dentry->d_parent;
9615 		btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9616 				parent);
9617 		btrfs_end_log_trans(dest);
9618 		dest_log_pinned = false;
9619 	}
9620 out_fail:
9621 	/*
9622 	 * If we have pinned a log and an error happened, we unpin tasks
9623 	 * trying to sync the log and force them to fallback to a transaction
9624 	 * commit if the log currently contains any of the inodes involved in
9625 	 * this rename operation (to ensure we do not persist a log with an
9626 	 * inconsistent state for any of these inodes or leading to any
9627 	 * inconsistencies when replayed). If the transaction was aborted, the
9628 	 * abortion reason is propagated to userspace when attempting to commit
9629 	 * the transaction. If the log does not contain any of these inodes, we
9630 	 * allow the tasks to sync it.
9631 	 */
9632 	if (ret && (root_log_pinned || dest_log_pinned)) {
9633 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9634 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9635 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9636 		    (new_inode &&
9637 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9638 			btrfs_set_log_full_commit(fs_info, trans);
9639 
9640 		if (root_log_pinned) {
9641 			btrfs_end_log_trans(root);
9642 			root_log_pinned = false;
9643 		}
9644 		if (dest_log_pinned) {
9645 			btrfs_end_log_trans(dest);
9646 			dest_log_pinned = false;
9647 		}
9648 	}
9649 	ret = btrfs_end_transaction(trans);
9650 out_notrans:
9651 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9652 		up_read(&fs_info->subvol_sem);
9653 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9654 		up_read(&fs_info->subvol_sem);
9655 
9656 	return ret;
9657 }
9658 
9659 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9660 				     struct btrfs_root *root,
9661 				     struct inode *dir,
9662 				     struct dentry *dentry)
9663 {
9664 	int ret;
9665 	struct inode *inode;
9666 	u64 objectid;
9667 	u64 index;
9668 
9669 	ret = btrfs_find_free_ino(root, &objectid);
9670 	if (ret)
9671 		return ret;
9672 
9673 	inode = btrfs_new_inode(trans, root, dir,
9674 				dentry->d_name.name,
9675 				dentry->d_name.len,
9676 				btrfs_ino(BTRFS_I(dir)),
9677 				objectid,
9678 				S_IFCHR | WHITEOUT_MODE,
9679 				&index);
9680 
9681 	if (IS_ERR(inode)) {
9682 		ret = PTR_ERR(inode);
9683 		return ret;
9684 	}
9685 
9686 	inode->i_op = &btrfs_special_inode_operations;
9687 	init_special_inode(inode, inode->i_mode,
9688 		WHITEOUT_DEV);
9689 
9690 	ret = btrfs_init_inode_security(trans, inode, dir,
9691 				&dentry->d_name);
9692 	if (ret)
9693 		goto out;
9694 
9695 	ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9696 				BTRFS_I(inode), 0, index);
9697 	if (ret)
9698 		goto out;
9699 
9700 	ret = btrfs_update_inode(trans, root, inode);
9701 out:
9702 	unlock_new_inode(inode);
9703 	if (ret)
9704 		inode_dec_link_count(inode);
9705 	iput(inode);
9706 
9707 	return ret;
9708 }
9709 
9710 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9711 			   struct inode *new_dir, struct dentry *new_dentry,
9712 			   unsigned int flags)
9713 {
9714 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9715 	struct btrfs_trans_handle *trans;
9716 	unsigned int trans_num_items;
9717 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9718 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9719 	struct inode *new_inode = d_inode(new_dentry);
9720 	struct inode *old_inode = d_inode(old_dentry);
9721 	u64 index = 0;
9722 	u64 root_objectid;
9723 	int ret;
9724 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9725 	bool log_pinned = false;
9726 
9727 	if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9728 		return -EPERM;
9729 
9730 	/* we only allow rename subvolume link between subvolumes */
9731 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9732 		return -EXDEV;
9733 
9734 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9735 	    (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9736 		return -ENOTEMPTY;
9737 
9738 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
9739 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9740 		return -ENOTEMPTY;
9741 
9742 
9743 	/* check for collisions, even if the  name isn't there */
9744 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9745 			     new_dentry->d_name.name,
9746 			     new_dentry->d_name.len);
9747 
9748 	if (ret) {
9749 		if (ret == -EEXIST) {
9750 			/* we shouldn't get
9751 			 * eexist without a new_inode */
9752 			if (WARN_ON(!new_inode)) {
9753 				return ret;
9754 			}
9755 		} else {
9756 			/* maybe -EOVERFLOW */
9757 			return ret;
9758 		}
9759 	}
9760 	ret = 0;
9761 
9762 	/*
9763 	 * we're using rename to replace one file with another.  Start IO on it
9764 	 * now so  we don't add too much work to the end of the transaction
9765 	 */
9766 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9767 		filemap_flush(old_inode->i_mapping);
9768 
9769 	/* close the racy window with snapshot create/destroy ioctl */
9770 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9771 		down_read(&fs_info->subvol_sem);
9772 	/*
9773 	 * We want to reserve the absolute worst case amount of items.  So if
9774 	 * both inodes are subvols and we need to unlink them then that would
9775 	 * require 4 item modifications, but if they are both normal inodes it
9776 	 * would require 5 item modifications, so we'll assume they are normal
9777 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9778 	 * should cover the worst case number of items we'll modify.
9779 	 * If our rename has the whiteout flag, we need more 5 units for the
9780 	 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9781 	 * when selinux is enabled).
9782 	 */
9783 	trans_num_items = 11;
9784 	if (flags & RENAME_WHITEOUT)
9785 		trans_num_items += 5;
9786 	trans = btrfs_start_transaction(root, trans_num_items);
9787 	if (IS_ERR(trans)) {
9788 		ret = PTR_ERR(trans);
9789 		goto out_notrans;
9790 	}
9791 
9792 	if (dest != root)
9793 		btrfs_record_root_in_trans(trans, dest);
9794 
9795 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9796 	if (ret)
9797 		goto out_fail;
9798 
9799 	BTRFS_I(old_inode)->dir_index = 0ULL;
9800 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9801 		/* force full log commit if subvolume involved. */
9802 		btrfs_set_log_full_commit(fs_info, trans);
9803 	} else {
9804 		btrfs_pin_log_trans(root);
9805 		log_pinned = true;
9806 		ret = btrfs_insert_inode_ref(trans, dest,
9807 					     new_dentry->d_name.name,
9808 					     new_dentry->d_name.len,
9809 					     old_ino,
9810 					     btrfs_ino(BTRFS_I(new_dir)), index);
9811 		if (ret)
9812 			goto out_fail;
9813 	}
9814 
9815 	inode_inc_iversion(old_dir);
9816 	inode_inc_iversion(new_dir);
9817 	inode_inc_iversion(old_inode);
9818 	old_dir->i_ctime = old_dir->i_mtime =
9819 	new_dir->i_ctime = new_dir->i_mtime =
9820 	old_inode->i_ctime = current_time(old_dir);
9821 
9822 	if (old_dentry->d_parent != new_dentry->d_parent)
9823 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9824 				BTRFS_I(old_inode), 1);
9825 
9826 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9827 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9828 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9829 					old_dentry->d_name.name,
9830 					old_dentry->d_name.len);
9831 	} else {
9832 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9833 					BTRFS_I(d_inode(old_dentry)),
9834 					old_dentry->d_name.name,
9835 					old_dentry->d_name.len);
9836 		if (!ret)
9837 			ret = btrfs_update_inode(trans, root, old_inode);
9838 	}
9839 	if (ret) {
9840 		btrfs_abort_transaction(trans, ret);
9841 		goto out_fail;
9842 	}
9843 
9844 	if (new_inode) {
9845 		inode_inc_iversion(new_inode);
9846 		new_inode->i_ctime = current_time(new_inode);
9847 		if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9848 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9849 			root_objectid = BTRFS_I(new_inode)->location.objectid;
9850 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
9851 						root_objectid,
9852 						new_dentry->d_name.name,
9853 						new_dentry->d_name.len);
9854 			BUG_ON(new_inode->i_nlink == 0);
9855 		} else {
9856 			ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9857 						 BTRFS_I(d_inode(new_dentry)),
9858 						 new_dentry->d_name.name,
9859 						 new_dentry->d_name.len);
9860 		}
9861 		if (!ret && new_inode->i_nlink == 0)
9862 			ret = btrfs_orphan_add(trans,
9863 					BTRFS_I(d_inode(new_dentry)));
9864 		if (ret) {
9865 			btrfs_abort_transaction(trans, ret);
9866 			goto out_fail;
9867 		}
9868 	}
9869 
9870 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9871 			     new_dentry->d_name.name,
9872 			     new_dentry->d_name.len, 0, index);
9873 	if (ret) {
9874 		btrfs_abort_transaction(trans, ret);
9875 		goto out_fail;
9876 	}
9877 
9878 	if (old_inode->i_nlink == 1)
9879 		BTRFS_I(old_inode)->dir_index = index;
9880 
9881 	if (log_pinned) {
9882 		struct dentry *parent = new_dentry->d_parent;
9883 
9884 		btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9885 				parent);
9886 		btrfs_end_log_trans(root);
9887 		log_pinned = false;
9888 	}
9889 
9890 	if (flags & RENAME_WHITEOUT) {
9891 		ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9892 						old_dentry);
9893 
9894 		if (ret) {
9895 			btrfs_abort_transaction(trans, ret);
9896 			goto out_fail;
9897 		}
9898 	}
9899 out_fail:
9900 	/*
9901 	 * If we have pinned the log and an error happened, we unpin tasks
9902 	 * trying to sync the log and force them to fallback to a transaction
9903 	 * commit if the log currently contains any of the inodes involved in
9904 	 * this rename operation (to ensure we do not persist a log with an
9905 	 * inconsistent state for any of these inodes or leading to any
9906 	 * inconsistencies when replayed). If the transaction was aborted, the
9907 	 * abortion reason is propagated to userspace when attempting to commit
9908 	 * the transaction. If the log does not contain any of these inodes, we
9909 	 * allow the tasks to sync it.
9910 	 */
9911 	if (ret && log_pinned) {
9912 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9913 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9914 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9915 		    (new_inode &&
9916 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9917 			btrfs_set_log_full_commit(fs_info, trans);
9918 
9919 		btrfs_end_log_trans(root);
9920 		log_pinned = false;
9921 	}
9922 	btrfs_end_transaction(trans);
9923 out_notrans:
9924 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9925 		up_read(&fs_info->subvol_sem);
9926 
9927 	return ret;
9928 }
9929 
9930 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9931 			 struct inode *new_dir, struct dentry *new_dentry,
9932 			 unsigned int flags)
9933 {
9934 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9935 		return -EINVAL;
9936 
9937 	if (flags & RENAME_EXCHANGE)
9938 		return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9939 					  new_dentry);
9940 
9941 	return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9942 }
9943 
9944 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9945 {
9946 	struct btrfs_delalloc_work *delalloc_work;
9947 	struct inode *inode;
9948 
9949 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
9950 				     work);
9951 	inode = delalloc_work->inode;
9952 	filemap_flush(inode->i_mapping);
9953 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9954 				&BTRFS_I(inode)->runtime_flags))
9955 		filemap_flush(inode->i_mapping);
9956 
9957 	if (delalloc_work->delay_iput)
9958 		btrfs_add_delayed_iput(inode);
9959 	else
9960 		iput(inode);
9961 	complete(&delalloc_work->completion);
9962 }
9963 
9964 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9965 						    int delay_iput)
9966 {
9967 	struct btrfs_delalloc_work *work;
9968 
9969 	work = kmalloc(sizeof(*work), GFP_NOFS);
9970 	if (!work)
9971 		return NULL;
9972 
9973 	init_completion(&work->completion);
9974 	INIT_LIST_HEAD(&work->list);
9975 	work->inode = inode;
9976 	work->delay_iput = delay_iput;
9977 	WARN_ON_ONCE(!inode);
9978 	btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9979 			btrfs_run_delalloc_work, NULL, NULL);
9980 
9981 	return work;
9982 }
9983 
9984 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9985 {
9986 	wait_for_completion(&work->completion);
9987 	kfree(work);
9988 }
9989 
9990 /*
9991  * some fairly slow code that needs optimization. This walks the list
9992  * of all the inodes with pending delalloc and forces them to disk.
9993  */
9994 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9995 				   int nr)
9996 {
9997 	struct btrfs_inode *binode;
9998 	struct inode *inode;
9999 	struct btrfs_delalloc_work *work, *next;
10000 	struct list_head works;
10001 	struct list_head splice;
10002 	int ret = 0;
10003 
10004 	INIT_LIST_HEAD(&works);
10005 	INIT_LIST_HEAD(&splice);
10006 
10007 	mutex_lock(&root->delalloc_mutex);
10008 	spin_lock(&root->delalloc_lock);
10009 	list_splice_init(&root->delalloc_inodes, &splice);
10010 	while (!list_empty(&splice)) {
10011 		binode = list_entry(splice.next, struct btrfs_inode,
10012 				    delalloc_inodes);
10013 
10014 		list_move_tail(&binode->delalloc_inodes,
10015 			       &root->delalloc_inodes);
10016 		inode = igrab(&binode->vfs_inode);
10017 		if (!inode) {
10018 			cond_resched_lock(&root->delalloc_lock);
10019 			continue;
10020 		}
10021 		spin_unlock(&root->delalloc_lock);
10022 
10023 		work = btrfs_alloc_delalloc_work(inode, delay_iput);
10024 		if (!work) {
10025 			if (delay_iput)
10026 				btrfs_add_delayed_iput(inode);
10027 			else
10028 				iput(inode);
10029 			ret = -ENOMEM;
10030 			goto out;
10031 		}
10032 		list_add_tail(&work->list, &works);
10033 		btrfs_queue_work(root->fs_info->flush_workers,
10034 				 &work->work);
10035 		ret++;
10036 		if (nr != -1 && ret >= nr)
10037 			goto out;
10038 		cond_resched();
10039 		spin_lock(&root->delalloc_lock);
10040 	}
10041 	spin_unlock(&root->delalloc_lock);
10042 
10043 out:
10044 	list_for_each_entry_safe(work, next, &works, list) {
10045 		list_del_init(&work->list);
10046 		btrfs_wait_and_free_delalloc_work(work);
10047 	}
10048 
10049 	if (!list_empty_careful(&splice)) {
10050 		spin_lock(&root->delalloc_lock);
10051 		list_splice_tail(&splice, &root->delalloc_inodes);
10052 		spin_unlock(&root->delalloc_lock);
10053 	}
10054 	mutex_unlock(&root->delalloc_mutex);
10055 	return ret;
10056 }
10057 
10058 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10059 {
10060 	struct btrfs_fs_info *fs_info = root->fs_info;
10061 	int ret;
10062 
10063 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10064 		return -EROFS;
10065 
10066 	ret = __start_delalloc_inodes(root, delay_iput, -1);
10067 	if (ret > 0)
10068 		ret = 0;
10069 	/*
10070 	 * the filemap_flush will queue IO into the worker threads, but
10071 	 * we have to make sure the IO is actually started and that
10072 	 * ordered extents get created before we return
10073 	 */
10074 	atomic_inc(&fs_info->async_submit_draining);
10075 	while (atomic_read(&fs_info->nr_async_submits) ||
10076 	       atomic_read(&fs_info->async_delalloc_pages)) {
10077 		wait_event(fs_info->async_submit_wait,
10078 			   (atomic_read(&fs_info->nr_async_submits) == 0 &&
10079 			    atomic_read(&fs_info->async_delalloc_pages) == 0));
10080 	}
10081 	atomic_dec(&fs_info->async_submit_draining);
10082 	return ret;
10083 }
10084 
10085 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10086 			       int nr)
10087 {
10088 	struct btrfs_root *root;
10089 	struct list_head splice;
10090 	int ret;
10091 
10092 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10093 		return -EROFS;
10094 
10095 	INIT_LIST_HEAD(&splice);
10096 
10097 	mutex_lock(&fs_info->delalloc_root_mutex);
10098 	spin_lock(&fs_info->delalloc_root_lock);
10099 	list_splice_init(&fs_info->delalloc_roots, &splice);
10100 	while (!list_empty(&splice) && nr) {
10101 		root = list_first_entry(&splice, struct btrfs_root,
10102 					delalloc_root);
10103 		root = btrfs_grab_fs_root(root);
10104 		BUG_ON(!root);
10105 		list_move_tail(&root->delalloc_root,
10106 			       &fs_info->delalloc_roots);
10107 		spin_unlock(&fs_info->delalloc_root_lock);
10108 
10109 		ret = __start_delalloc_inodes(root, delay_iput, nr);
10110 		btrfs_put_fs_root(root);
10111 		if (ret < 0)
10112 			goto out;
10113 
10114 		if (nr != -1) {
10115 			nr -= ret;
10116 			WARN_ON(nr < 0);
10117 		}
10118 		spin_lock(&fs_info->delalloc_root_lock);
10119 	}
10120 	spin_unlock(&fs_info->delalloc_root_lock);
10121 
10122 	ret = 0;
10123 	atomic_inc(&fs_info->async_submit_draining);
10124 	while (atomic_read(&fs_info->nr_async_submits) ||
10125 	      atomic_read(&fs_info->async_delalloc_pages)) {
10126 		wait_event(fs_info->async_submit_wait,
10127 		   (atomic_read(&fs_info->nr_async_submits) == 0 &&
10128 		    atomic_read(&fs_info->async_delalloc_pages) == 0));
10129 	}
10130 	atomic_dec(&fs_info->async_submit_draining);
10131 out:
10132 	if (!list_empty_careful(&splice)) {
10133 		spin_lock(&fs_info->delalloc_root_lock);
10134 		list_splice_tail(&splice, &fs_info->delalloc_roots);
10135 		spin_unlock(&fs_info->delalloc_root_lock);
10136 	}
10137 	mutex_unlock(&fs_info->delalloc_root_mutex);
10138 	return ret;
10139 }
10140 
10141 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10142 			 const char *symname)
10143 {
10144 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10145 	struct btrfs_trans_handle *trans;
10146 	struct btrfs_root *root = BTRFS_I(dir)->root;
10147 	struct btrfs_path *path;
10148 	struct btrfs_key key;
10149 	struct inode *inode = NULL;
10150 	int err;
10151 	int drop_inode = 0;
10152 	u64 objectid;
10153 	u64 index = 0;
10154 	int name_len;
10155 	int datasize;
10156 	unsigned long ptr;
10157 	struct btrfs_file_extent_item *ei;
10158 	struct extent_buffer *leaf;
10159 
10160 	name_len = strlen(symname);
10161 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10162 		return -ENAMETOOLONG;
10163 
10164 	/*
10165 	 * 2 items for inode item and ref
10166 	 * 2 items for dir items
10167 	 * 1 item for updating parent inode item
10168 	 * 1 item for the inline extent item
10169 	 * 1 item for xattr if selinux is on
10170 	 */
10171 	trans = btrfs_start_transaction(root, 7);
10172 	if (IS_ERR(trans))
10173 		return PTR_ERR(trans);
10174 
10175 	err = btrfs_find_free_ino(root, &objectid);
10176 	if (err)
10177 		goto out_unlock;
10178 
10179 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10180 				dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10181 				objectid, S_IFLNK|S_IRWXUGO, &index);
10182 	if (IS_ERR(inode)) {
10183 		err = PTR_ERR(inode);
10184 		goto out_unlock;
10185 	}
10186 
10187 	/*
10188 	* If the active LSM wants to access the inode during
10189 	* d_instantiate it needs these. Smack checks to see
10190 	* if the filesystem supports xattrs by looking at the
10191 	* ops vector.
10192 	*/
10193 	inode->i_fop = &btrfs_file_operations;
10194 	inode->i_op = &btrfs_file_inode_operations;
10195 	inode->i_mapping->a_ops = &btrfs_aops;
10196 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10197 
10198 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10199 	if (err)
10200 		goto out_unlock_inode;
10201 
10202 	path = btrfs_alloc_path();
10203 	if (!path) {
10204 		err = -ENOMEM;
10205 		goto out_unlock_inode;
10206 	}
10207 	key.objectid = btrfs_ino(BTRFS_I(inode));
10208 	key.offset = 0;
10209 	key.type = BTRFS_EXTENT_DATA_KEY;
10210 	datasize = btrfs_file_extent_calc_inline_size(name_len);
10211 	err = btrfs_insert_empty_item(trans, root, path, &key,
10212 				      datasize);
10213 	if (err) {
10214 		btrfs_free_path(path);
10215 		goto out_unlock_inode;
10216 	}
10217 	leaf = path->nodes[0];
10218 	ei = btrfs_item_ptr(leaf, path->slots[0],
10219 			    struct btrfs_file_extent_item);
10220 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10221 	btrfs_set_file_extent_type(leaf, ei,
10222 				   BTRFS_FILE_EXTENT_INLINE);
10223 	btrfs_set_file_extent_encryption(leaf, ei, 0);
10224 	btrfs_set_file_extent_compression(leaf, ei, 0);
10225 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10226 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10227 
10228 	ptr = btrfs_file_extent_inline_start(ei);
10229 	write_extent_buffer(leaf, symname, ptr, name_len);
10230 	btrfs_mark_buffer_dirty(leaf);
10231 	btrfs_free_path(path);
10232 
10233 	inode->i_op = &btrfs_symlink_inode_operations;
10234 	inode_nohighmem(inode);
10235 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
10236 	inode_set_bytes(inode, name_len);
10237 	btrfs_i_size_write(BTRFS_I(inode), name_len);
10238 	err = btrfs_update_inode(trans, root, inode);
10239 	/*
10240 	 * Last step, add directory indexes for our symlink inode. This is the
10241 	 * last step to avoid extra cleanup of these indexes if an error happens
10242 	 * elsewhere above.
10243 	 */
10244 	if (!err)
10245 		err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10246 				BTRFS_I(inode), 0, index);
10247 	if (err) {
10248 		drop_inode = 1;
10249 		goto out_unlock_inode;
10250 	}
10251 
10252 	unlock_new_inode(inode);
10253 	d_instantiate(dentry, inode);
10254 
10255 out_unlock:
10256 	btrfs_end_transaction(trans);
10257 	if (drop_inode) {
10258 		inode_dec_link_count(inode);
10259 		iput(inode);
10260 	}
10261 	btrfs_btree_balance_dirty(fs_info);
10262 	return err;
10263 
10264 out_unlock_inode:
10265 	drop_inode = 1;
10266 	unlock_new_inode(inode);
10267 	goto out_unlock;
10268 }
10269 
10270 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10271 				       u64 start, u64 num_bytes, u64 min_size,
10272 				       loff_t actual_len, u64 *alloc_hint,
10273 				       struct btrfs_trans_handle *trans)
10274 {
10275 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10276 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10277 	struct extent_map *em;
10278 	struct btrfs_root *root = BTRFS_I(inode)->root;
10279 	struct btrfs_key ins;
10280 	u64 cur_offset = start;
10281 	u64 i_size;
10282 	u64 cur_bytes;
10283 	u64 last_alloc = (u64)-1;
10284 	int ret = 0;
10285 	bool own_trans = true;
10286 	u64 end = start + num_bytes - 1;
10287 
10288 	if (trans)
10289 		own_trans = false;
10290 	while (num_bytes > 0) {
10291 		if (own_trans) {
10292 			trans = btrfs_start_transaction(root, 3);
10293 			if (IS_ERR(trans)) {
10294 				ret = PTR_ERR(trans);
10295 				break;
10296 			}
10297 		}
10298 
10299 		cur_bytes = min_t(u64, num_bytes, SZ_256M);
10300 		cur_bytes = max(cur_bytes, min_size);
10301 		/*
10302 		 * If we are severely fragmented we could end up with really
10303 		 * small allocations, so if the allocator is returning small
10304 		 * chunks lets make its job easier by only searching for those
10305 		 * sized chunks.
10306 		 */
10307 		cur_bytes = min(cur_bytes, last_alloc);
10308 		ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10309 				min_size, 0, *alloc_hint, &ins, 1, 0);
10310 		if (ret) {
10311 			if (own_trans)
10312 				btrfs_end_transaction(trans);
10313 			break;
10314 		}
10315 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10316 
10317 		last_alloc = ins.offset;
10318 		ret = insert_reserved_file_extent(trans, inode,
10319 						  cur_offset, ins.objectid,
10320 						  ins.offset, ins.offset,
10321 						  ins.offset, 0, 0, 0,
10322 						  BTRFS_FILE_EXTENT_PREALLOC);
10323 		if (ret) {
10324 			btrfs_free_reserved_extent(fs_info, ins.objectid,
10325 						   ins.offset, 0);
10326 			btrfs_abort_transaction(trans, ret);
10327 			if (own_trans)
10328 				btrfs_end_transaction(trans);
10329 			break;
10330 		}
10331 
10332 		btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10333 					cur_offset + ins.offset -1, 0);
10334 
10335 		em = alloc_extent_map();
10336 		if (!em) {
10337 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10338 				&BTRFS_I(inode)->runtime_flags);
10339 			goto next;
10340 		}
10341 
10342 		em->start = cur_offset;
10343 		em->orig_start = cur_offset;
10344 		em->len = ins.offset;
10345 		em->block_start = ins.objectid;
10346 		em->block_len = ins.offset;
10347 		em->orig_block_len = ins.offset;
10348 		em->ram_bytes = ins.offset;
10349 		em->bdev = fs_info->fs_devices->latest_bdev;
10350 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10351 		em->generation = trans->transid;
10352 
10353 		while (1) {
10354 			write_lock(&em_tree->lock);
10355 			ret = add_extent_mapping(em_tree, em, 1);
10356 			write_unlock(&em_tree->lock);
10357 			if (ret != -EEXIST)
10358 				break;
10359 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10360 						cur_offset + ins.offset - 1,
10361 						0);
10362 		}
10363 		free_extent_map(em);
10364 next:
10365 		num_bytes -= ins.offset;
10366 		cur_offset += ins.offset;
10367 		*alloc_hint = ins.objectid + ins.offset;
10368 
10369 		inode_inc_iversion(inode);
10370 		inode->i_ctime = current_time(inode);
10371 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10372 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10373 		    (actual_len > inode->i_size) &&
10374 		    (cur_offset > inode->i_size)) {
10375 			if (cur_offset > actual_len)
10376 				i_size = actual_len;
10377 			else
10378 				i_size = cur_offset;
10379 			i_size_write(inode, i_size);
10380 			btrfs_ordered_update_i_size(inode, i_size, NULL);
10381 		}
10382 
10383 		ret = btrfs_update_inode(trans, root, inode);
10384 
10385 		if (ret) {
10386 			btrfs_abort_transaction(trans, ret);
10387 			if (own_trans)
10388 				btrfs_end_transaction(trans);
10389 			break;
10390 		}
10391 
10392 		if (own_trans)
10393 			btrfs_end_transaction(trans);
10394 	}
10395 	if (cur_offset < end)
10396 		btrfs_free_reserved_data_space(inode, cur_offset,
10397 			end - cur_offset + 1);
10398 	return ret;
10399 }
10400 
10401 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10402 			      u64 start, u64 num_bytes, u64 min_size,
10403 			      loff_t actual_len, u64 *alloc_hint)
10404 {
10405 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10406 					   min_size, actual_len, alloc_hint,
10407 					   NULL);
10408 }
10409 
10410 int btrfs_prealloc_file_range_trans(struct inode *inode,
10411 				    struct btrfs_trans_handle *trans, 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, trans);
10417 }
10418 
10419 static int btrfs_set_page_dirty(struct page *page)
10420 {
10421 	return __set_page_dirty_nobuffers(page);
10422 }
10423 
10424 static int btrfs_permission(struct inode *inode, int mask)
10425 {
10426 	struct btrfs_root *root = BTRFS_I(inode)->root;
10427 	umode_t mode = inode->i_mode;
10428 
10429 	if (mask & MAY_WRITE &&
10430 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10431 		if (btrfs_root_readonly(root))
10432 			return -EROFS;
10433 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10434 			return -EACCES;
10435 	}
10436 	return generic_permission(inode, mask);
10437 }
10438 
10439 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10440 {
10441 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10442 	struct btrfs_trans_handle *trans;
10443 	struct btrfs_root *root = BTRFS_I(dir)->root;
10444 	struct inode *inode = NULL;
10445 	u64 objectid;
10446 	u64 index;
10447 	int ret = 0;
10448 
10449 	/*
10450 	 * 5 units required for adding orphan entry
10451 	 */
10452 	trans = btrfs_start_transaction(root, 5);
10453 	if (IS_ERR(trans))
10454 		return PTR_ERR(trans);
10455 
10456 	ret = btrfs_find_free_ino(root, &objectid);
10457 	if (ret)
10458 		goto out;
10459 
10460 	inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10461 			btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10462 	if (IS_ERR(inode)) {
10463 		ret = PTR_ERR(inode);
10464 		inode = NULL;
10465 		goto out;
10466 	}
10467 
10468 	inode->i_fop = &btrfs_file_operations;
10469 	inode->i_op = &btrfs_file_inode_operations;
10470 
10471 	inode->i_mapping->a_ops = &btrfs_aops;
10472 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10473 
10474 	ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10475 	if (ret)
10476 		goto out_inode;
10477 
10478 	ret = btrfs_update_inode(trans, root, inode);
10479 	if (ret)
10480 		goto out_inode;
10481 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10482 	if (ret)
10483 		goto out_inode;
10484 
10485 	/*
10486 	 * We set number of links to 0 in btrfs_new_inode(), and here we set
10487 	 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10488 	 * through:
10489 	 *
10490 	 *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10491 	 */
10492 	set_nlink(inode, 1);
10493 	unlock_new_inode(inode);
10494 	d_tmpfile(dentry, inode);
10495 	mark_inode_dirty(inode);
10496 
10497 out:
10498 	btrfs_end_transaction(trans);
10499 	if (ret)
10500 		iput(inode);
10501 	btrfs_balance_delayed_items(fs_info);
10502 	btrfs_btree_balance_dirty(fs_info);
10503 	return ret;
10504 
10505 out_inode:
10506 	unlock_new_inode(inode);
10507 	goto out;
10508 
10509 }
10510 
10511 __attribute__((const))
10512 static int dummy_readpage_io_failed_hook(struct page *page, int failed_mirror)
10513 {
10514 	return 0;
10515 }
10516 
10517 static const struct inode_operations btrfs_dir_inode_operations = {
10518 	.getattr	= btrfs_getattr,
10519 	.lookup		= btrfs_lookup,
10520 	.create		= btrfs_create,
10521 	.unlink		= btrfs_unlink,
10522 	.link		= btrfs_link,
10523 	.mkdir		= btrfs_mkdir,
10524 	.rmdir		= btrfs_rmdir,
10525 	.rename		= btrfs_rename2,
10526 	.symlink	= btrfs_symlink,
10527 	.setattr	= btrfs_setattr,
10528 	.mknod		= btrfs_mknod,
10529 	.listxattr	= btrfs_listxattr,
10530 	.permission	= btrfs_permission,
10531 	.get_acl	= btrfs_get_acl,
10532 	.set_acl	= btrfs_set_acl,
10533 	.update_time	= btrfs_update_time,
10534 	.tmpfile        = btrfs_tmpfile,
10535 };
10536 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10537 	.lookup		= btrfs_lookup,
10538 	.permission	= btrfs_permission,
10539 	.update_time	= btrfs_update_time,
10540 };
10541 
10542 static const struct file_operations btrfs_dir_file_operations = {
10543 	.llseek		= generic_file_llseek,
10544 	.read		= generic_read_dir,
10545 	.iterate_shared	= btrfs_real_readdir,
10546 	.unlocked_ioctl	= btrfs_ioctl,
10547 #ifdef CONFIG_COMPAT
10548 	.compat_ioctl	= btrfs_compat_ioctl,
10549 #endif
10550 	.release        = btrfs_release_file,
10551 	.fsync		= btrfs_sync_file,
10552 };
10553 
10554 static const struct extent_io_ops btrfs_extent_io_ops = {
10555 	/* mandatory callbacks */
10556 	.submit_bio_hook = btrfs_submit_bio_hook,
10557 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
10558 	.merge_bio_hook = btrfs_merge_bio_hook,
10559 	.readpage_io_failed_hook = dummy_readpage_io_failed_hook,
10560 
10561 	/* optional callbacks */
10562 	.fill_delalloc = run_delalloc_range,
10563 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
10564 	.writepage_start_hook = btrfs_writepage_start_hook,
10565 	.set_bit_hook = btrfs_set_bit_hook,
10566 	.clear_bit_hook = btrfs_clear_bit_hook,
10567 	.merge_extent_hook = btrfs_merge_extent_hook,
10568 	.split_extent_hook = btrfs_split_extent_hook,
10569 };
10570 
10571 /*
10572  * btrfs doesn't support the bmap operation because swapfiles
10573  * use bmap to make a mapping of extents in the file.  They assume
10574  * these extents won't change over the life of the file and they
10575  * use the bmap result to do IO directly to the drive.
10576  *
10577  * the btrfs bmap call would return logical addresses that aren't
10578  * suitable for IO and they also will change frequently as COW
10579  * operations happen.  So, swapfile + btrfs == corruption.
10580  *
10581  * For now we're avoiding this by dropping bmap.
10582  */
10583 static const struct address_space_operations btrfs_aops = {
10584 	.readpage	= btrfs_readpage,
10585 	.writepage	= btrfs_writepage,
10586 	.writepages	= btrfs_writepages,
10587 	.readpages	= btrfs_readpages,
10588 	.direct_IO	= btrfs_direct_IO,
10589 	.invalidatepage = btrfs_invalidatepage,
10590 	.releasepage	= btrfs_releasepage,
10591 	.set_page_dirty	= btrfs_set_page_dirty,
10592 	.error_remove_page = generic_error_remove_page,
10593 };
10594 
10595 static const struct address_space_operations btrfs_symlink_aops = {
10596 	.readpage	= btrfs_readpage,
10597 	.writepage	= btrfs_writepage,
10598 	.invalidatepage = btrfs_invalidatepage,
10599 	.releasepage	= btrfs_releasepage,
10600 };
10601 
10602 static const struct inode_operations btrfs_file_inode_operations = {
10603 	.getattr	= btrfs_getattr,
10604 	.setattr	= btrfs_setattr,
10605 	.listxattr      = btrfs_listxattr,
10606 	.permission	= btrfs_permission,
10607 	.fiemap		= btrfs_fiemap,
10608 	.get_acl	= btrfs_get_acl,
10609 	.set_acl	= btrfs_set_acl,
10610 	.update_time	= btrfs_update_time,
10611 };
10612 static const struct inode_operations btrfs_special_inode_operations = {
10613 	.getattr	= btrfs_getattr,
10614 	.setattr	= btrfs_setattr,
10615 	.permission	= btrfs_permission,
10616 	.listxattr	= btrfs_listxattr,
10617 	.get_acl	= btrfs_get_acl,
10618 	.set_acl	= btrfs_set_acl,
10619 	.update_time	= btrfs_update_time,
10620 };
10621 static const struct inode_operations btrfs_symlink_inode_operations = {
10622 	.get_link	= page_get_link,
10623 	.getattr	= btrfs_getattr,
10624 	.setattr	= btrfs_setattr,
10625 	.permission	= btrfs_permission,
10626 	.listxattr	= btrfs_listxattr,
10627 	.update_time	= btrfs_update_time,
10628 };
10629 
10630 const struct dentry_operations btrfs_dentry_operations = {
10631 	.d_delete	= btrfs_dentry_delete,
10632 	.d_release	= btrfs_dentry_release,
10633 };
10634