xref: /openbmc/linux/fs/btrfs/inode.c (revision 2209fda3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
10 #include <linux/fs.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <asm/unaligned.h>
31 #include "ctree.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "ordered-data.h"
37 #include "xattr.h"
38 #include "tree-log.h"
39 #include "volumes.h"
40 #include "compression.h"
41 #include "locking.h"
42 #include "free-space-cache.h"
43 #include "inode-map.h"
44 #include "backref.h"
45 #include "props.h"
46 #include "qgroup.h"
47 #include "dedupe.h"
48 
49 struct btrfs_iget_args {
50 	struct btrfs_key *location;
51 	struct btrfs_root *root;
52 };
53 
54 struct btrfs_dio_data {
55 	u64 reserve;
56 	u64 unsubmitted_oe_range_start;
57 	u64 unsubmitted_oe_range_end;
58 	int overwrite;
59 };
60 
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static const struct extent_io_ops btrfs_extent_io_ops;
69 
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73 struct kmem_cache *btrfs_free_space_cachep;
74 
75 #define S_SHIFT 12
76 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
78 	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
79 	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
80 	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
81 	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
82 	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
83 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
84 };
85 
86 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
87 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
88 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
89 static noinline int cow_file_range(struct inode *inode,
90 				   struct page *locked_page,
91 				   u64 start, u64 end, u64 delalloc_end,
92 				   int *page_started, unsigned long *nr_written,
93 				   int unlock, struct btrfs_dedupe_hash *hash);
94 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
95 				       u64 orig_start, u64 block_start,
96 				       u64 block_len, u64 orig_block_len,
97 				       u64 ram_bytes, int compress_type,
98 				       int type);
99 
100 static void __endio_write_update_ordered(struct inode *inode,
101 					 const u64 offset, const u64 bytes,
102 					 const bool uptodate);
103 
104 /*
105  * Cleanup all submitted ordered extents in specified range to handle errors
106  * from the fill_dellaloc() callback.
107  *
108  * NOTE: caller must ensure that when an error happens, it can not call
109  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
110  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
111  * to be released, which we want to happen only when finishing the ordered
112  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
113  * fill_delalloc() callback already does proper cleanup for the first page of
114  * the range, that is, it invokes the callback writepage_end_io_hook() for the
115  * range of the first page.
116  */
117 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
118 						 const u64 offset,
119 						 const u64 bytes)
120 {
121 	unsigned long index = offset >> PAGE_SHIFT;
122 	unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
123 	struct page *page;
124 
125 	while (index <= end_index) {
126 		page = find_get_page(inode->i_mapping, index);
127 		index++;
128 		if (!page)
129 			continue;
130 		ClearPagePrivate2(page);
131 		put_page(page);
132 	}
133 	return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
134 					    bytes - PAGE_SIZE, false);
135 }
136 
137 static int btrfs_dirty_inode(struct inode *inode);
138 
139 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
140 void btrfs_test_inode_set_ops(struct inode *inode)
141 {
142 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
143 }
144 #endif
145 
146 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
147 				     struct inode *inode,  struct inode *dir,
148 				     const struct qstr *qstr)
149 {
150 	int err;
151 
152 	err = btrfs_init_acl(trans, inode, dir);
153 	if (!err)
154 		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
155 	return err;
156 }
157 
158 /*
159  * this does all the hard work for inserting an inline extent into
160  * the btree.  The caller should have done a btrfs_drop_extents so that
161  * no overlapping inline items exist in the btree
162  */
163 static int insert_inline_extent(struct btrfs_trans_handle *trans,
164 				struct btrfs_path *path, int extent_inserted,
165 				struct btrfs_root *root, struct inode *inode,
166 				u64 start, size_t size, size_t compressed_size,
167 				int compress_type,
168 				struct page **compressed_pages)
169 {
170 	struct extent_buffer *leaf;
171 	struct page *page = NULL;
172 	char *kaddr;
173 	unsigned long ptr;
174 	struct btrfs_file_extent_item *ei;
175 	int ret;
176 	size_t cur_size = size;
177 	unsigned long offset;
178 
179 	if (compressed_size && compressed_pages)
180 		cur_size = compressed_size;
181 
182 	inode_add_bytes(inode, size);
183 
184 	if (!extent_inserted) {
185 		struct btrfs_key key;
186 		size_t datasize;
187 
188 		key.objectid = btrfs_ino(BTRFS_I(inode));
189 		key.offset = start;
190 		key.type = BTRFS_EXTENT_DATA_KEY;
191 
192 		datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 		path->leave_spinning = 1;
194 		ret = btrfs_insert_empty_item(trans, root, path, &key,
195 					      datasize);
196 		if (ret)
197 			goto fail;
198 	}
199 	leaf = path->nodes[0];
200 	ei = btrfs_item_ptr(leaf, path->slots[0],
201 			    struct btrfs_file_extent_item);
202 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 	btrfs_set_file_extent_encryption(leaf, ei, 0);
205 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 	ptr = btrfs_file_extent_inline_start(ei);
208 
209 	if (compress_type != BTRFS_COMPRESS_NONE) {
210 		struct page *cpage;
211 		int i = 0;
212 		while (compressed_size > 0) {
213 			cpage = compressed_pages[i];
214 			cur_size = min_t(unsigned long, compressed_size,
215 				       PAGE_SIZE);
216 
217 			kaddr = kmap_atomic(cpage);
218 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 			kunmap_atomic(kaddr);
220 
221 			i++;
222 			ptr += cur_size;
223 			compressed_size -= cur_size;
224 		}
225 		btrfs_set_file_extent_compression(leaf, ei,
226 						  compress_type);
227 	} else {
228 		page = find_get_page(inode->i_mapping,
229 				     start >> PAGE_SHIFT);
230 		btrfs_set_file_extent_compression(leaf, ei, 0);
231 		kaddr = kmap_atomic(page);
232 		offset = start & (PAGE_SIZE - 1);
233 		write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 		kunmap_atomic(kaddr);
235 		put_page(page);
236 	}
237 	btrfs_mark_buffer_dirty(leaf);
238 	btrfs_release_path(path);
239 
240 	/*
241 	 * we're an inline extent, so nobody can
242 	 * extend the file past i_size without locking
243 	 * a page we already have locked.
244 	 *
245 	 * We must do any isize and inode updates
246 	 * before we unlock the pages.  Otherwise we
247 	 * could end up racing with unlink.
248 	 */
249 	BTRFS_I(inode)->disk_i_size = inode->i_size;
250 	ret = btrfs_update_inode(trans, root, inode);
251 
252 fail:
253 	return ret;
254 }
255 
256 
257 /*
258  * conditionally insert an inline extent into the file.  This
259  * does the checks required to make sure the data is small enough
260  * to fit as an inline extent.
261  */
262 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
263 					  u64 end, size_t compressed_size,
264 					  int compress_type,
265 					  struct page **compressed_pages)
266 {
267 	struct btrfs_root *root = BTRFS_I(inode)->root;
268 	struct btrfs_fs_info *fs_info = root->fs_info;
269 	struct btrfs_trans_handle *trans;
270 	u64 isize = i_size_read(inode);
271 	u64 actual_end = min(end + 1, isize);
272 	u64 inline_len = actual_end - start;
273 	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
274 	u64 data_len = inline_len;
275 	int ret;
276 	struct btrfs_path *path;
277 	int extent_inserted = 0;
278 	u32 extent_item_size;
279 
280 	if (compressed_size)
281 		data_len = compressed_size;
282 
283 	if (start > 0 ||
284 	    actual_end > fs_info->sectorsize ||
285 	    data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
286 	    (!compressed_size &&
287 	    (actual_end & (fs_info->sectorsize - 1)) == 0) ||
288 	    end + 1 < isize ||
289 	    data_len > fs_info->max_inline) {
290 		return 1;
291 	}
292 
293 	path = btrfs_alloc_path();
294 	if (!path)
295 		return -ENOMEM;
296 
297 	trans = btrfs_join_transaction(root);
298 	if (IS_ERR(trans)) {
299 		btrfs_free_path(path);
300 		return PTR_ERR(trans);
301 	}
302 	trans->block_rsv = &BTRFS_I(inode)->block_rsv;
303 
304 	if (compressed_size && compressed_pages)
305 		extent_item_size = btrfs_file_extent_calc_inline_size(
306 		   compressed_size);
307 	else
308 		extent_item_size = btrfs_file_extent_calc_inline_size(
309 		    inline_len);
310 
311 	ret = __btrfs_drop_extents(trans, root, inode, path,
312 				   start, aligned_end, NULL,
313 				   1, 1, extent_item_size, &extent_inserted);
314 	if (ret) {
315 		btrfs_abort_transaction(trans, ret);
316 		goto out;
317 	}
318 
319 	if (isize > actual_end)
320 		inline_len = min_t(u64, isize, actual_end);
321 	ret = insert_inline_extent(trans, path, extent_inserted,
322 				   root, inode, start,
323 				   inline_len, compressed_size,
324 				   compress_type, compressed_pages);
325 	if (ret && ret != -ENOSPC) {
326 		btrfs_abort_transaction(trans, ret);
327 		goto out;
328 	} else if (ret == -ENOSPC) {
329 		ret = 1;
330 		goto out;
331 	}
332 
333 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
334 	btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
335 out:
336 	/*
337 	 * Don't forget to free the reserved space, as for inlined extent
338 	 * it won't count as data extent, free them directly here.
339 	 * And at reserve time, it's always aligned to page size, so
340 	 * just free one page here.
341 	 */
342 	btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
343 	btrfs_free_path(path);
344 	btrfs_end_transaction(trans);
345 	return ret;
346 }
347 
348 struct async_extent {
349 	u64 start;
350 	u64 ram_size;
351 	u64 compressed_size;
352 	struct page **pages;
353 	unsigned long nr_pages;
354 	int compress_type;
355 	struct list_head list;
356 };
357 
358 struct async_cow {
359 	struct inode *inode;
360 	struct btrfs_root *root;
361 	struct page *locked_page;
362 	u64 start;
363 	u64 end;
364 	unsigned int write_flags;
365 	struct list_head extents;
366 	struct btrfs_work work;
367 };
368 
369 static noinline int add_async_extent(struct async_cow *cow,
370 				     u64 start, u64 ram_size,
371 				     u64 compressed_size,
372 				     struct page **pages,
373 				     unsigned long nr_pages,
374 				     int compress_type)
375 {
376 	struct async_extent *async_extent;
377 
378 	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
379 	BUG_ON(!async_extent); /* -ENOMEM */
380 	async_extent->start = start;
381 	async_extent->ram_size = ram_size;
382 	async_extent->compressed_size = compressed_size;
383 	async_extent->pages = pages;
384 	async_extent->nr_pages = nr_pages;
385 	async_extent->compress_type = compress_type;
386 	list_add_tail(&async_extent->list, &cow->extents);
387 	return 0;
388 }
389 
390 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
391 {
392 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
393 
394 	/* force compress */
395 	if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
396 		return 1;
397 	/* defrag ioctl */
398 	if (BTRFS_I(inode)->defrag_compress)
399 		return 1;
400 	/* bad compression ratios */
401 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
402 		return 0;
403 	if (btrfs_test_opt(fs_info, COMPRESS) ||
404 	    BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
405 	    BTRFS_I(inode)->prop_compress)
406 		return btrfs_compress_heuristic(inode, start, end);
407 	return 0;
408 }
409 
410 static inline void inode_should_defrag(struct btrfs_inode *inode,
411 		u64 start, u64 end, u64 num_bytes, u64 small_write)
412 {
413 	/* If this is a small write inside eof, kick off a defrag */
414 	if (num_bytes < small_write &&
415 	    (start > 0 || end + 1 < inode->disk_i_size))
416 		btrfs_add_inode_defrag(NULL, inode);
417 }
418 
419 /*
420  * we create compressed extents in two phases.  The first
421  * phase compresses a range of pages that have already been
422  * locked (both pages and state bits are locked).
423  *
424  * This is done inside an ordered work queue, and the compression
425  * is spread across many cpus.  The actual IO submission is step
426  * two, and the ordered work queue takes care of making sure that
427  * happens in the same order things were put onto the queue by
428  * writepages and friends.
429  *
430  * If this code finds it can't get good compression, it puts an
431  * entry onto the work queue to write the uncompressed bytes.  This
432  * makes sure that both compressed inodes and uncompressed inodes
433  * are written in the same order that the flusher thread sent them
434  * down.
435  */
436 static noinline void compress_file_range(struct inode *inode,
437 					struct page *locked_page,
438 					u64 start, u64 end,
439 					struct async_cow *async_cow,
440 					int *num_added)
441 {
442 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
443 	u64 blocksize = fs_info->sectorsize;
444 	u64 actual_end;
445 	u64 isize = i_size_read(inode);
446 	int ret = 0;
447 	struct page **pages = NULL;
448 	unsigned long nr_pages;
449 	unsigned long total_compressed = 0;
450 	unsigned long total_in = 0;
451 	int i;
452 	int will_compress;
453 	int compress_type = fs_info->compress_type;
454 	int redirty = 0;
455 
456 	inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
457 			SZ_16K);
458 
459 	actual_end = min_t(u64, isize, end + 1);
460 again:
461 	will_compress = 0;
462 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
463 	BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
464 	nr_pages = min_t(unsigned long, nr_pages,
465 			BTRFS_MAX_COMPRESSED / PAGE_SIZE);
466 
467 	/*
468 	 * we don't want to send crud past the end of i_size through
469 	 * compression, that's just a waste of CPU time.  So, if the
470 	 * end of the file is before the start of our current
471 	 * requested range of bytes, we bail out to the uncompressed
472 	 * cleanup code that can deal with all of this.
473 	 *
474 	 * It isn't really the fastest way to fix things, but this is a
475 	 * very uncommon corner.
476 	 */
477 	if (actual_end <= start)
478 		goto cleanup_and_bail_uncompressed;
479 
480 	total_compressed = actual_end - start;
481 
482 	/*
483 	 * skip compression for a small file range(<=blocksize) that
484 	 * isn't an inline extent, since it doesn't save disk space at all.
485 	 */
486 	if (total_compressed <= blocksize &&
487 	   (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
488 		goto cleanup_and_bail_uncompressed;
489 
490 	total_compressed = min_t(unsigned long, total_compressed,
491 			BTRFS_MAX_UNCOMPRESSED);
492 	total_in = 0;
493 	ret = 0;
494 
495 	/*
496 	 * we do compression for mount -o compress and when the
497 	 * inode has not been flagged as nocompress.  This flag can
498 	 * change at any time if we discover bad compression ratios.
499 	 */
500 	if (inode_need_compress(inode, start, end)) {
501 		WARN_ON(pages);
502 		pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
503 		if (!pages) {
504 			/* just bail out to the uncompressed code */
505 			nr_pages = 0;
506 			goto cont;
507 		}
508 
509 		if (BTRFS_I(inode)->defrag_compress)
510 			compress_type = BTRFS_I(inode)->defrag_compress;
511 		else if (BTRFS_I(inode)->prop_compress)
512 			compress_type = BTRFS_I(inode)->prop_compress;
513 
514 		/*
515 		 * we need to call clear_page_dirty_for_io on each
516 		 * page in the range.  Otherwise applications with the file
517 		 * mmap'd can wander in and change the page contents while
518 		 * we are compressing them.
519 		 *
520 		 * If the compression fails for any reason, we set the pages
521 		 * dirty again later on.
522 		 *
523 		 * Note that the remaining part is redirtied, the start pointer
524 		 * has moved, the end is the original one.
525 		 */
526 		if (!redirty) {
527 			extent_range_clear_dirty_for_io(inode, start, end);
528 			redirty = 1;
529 		}
530 
531 		/* Compression level is applied here and only here */
532 		ret = btrfs_compress_pages(
533 			compress_type | (fs_info->compress_level << 4),
534 					   inode->i_mapping, start,
535 					   pages,
536 					   &nr_pages,
537 					   &total_in,
538 					   &total_compressed);
539 
540 		if (!ret) {
541 			unsigned long offset = total_compressed &
542 				(PAGE_SIZE - 1);
543 			struct page *page = pages[nr_pages - 1];
544 			char *kaddr;
545 
546 			/* zero the tail end of the last page, we might be
547 			 * sending it down to disk
548 			 */
549 			if (offset) {
550 				kaddr = kmap_atomic(page);
551 				memset(kaddr + offset, 0,
552 				       PAGE_SIZE - offset);
553 				kunmap_atomic(kaddr);
554 			}
555 			will_compress = 1;
556 		}
557 	}
558 cont:
559 	if (start == 0) {
560 		/* lets try to make an inline extent */
561 		if (ret || total_in < actual_end) {
562 			/* we didn't compress the entire range, try
563 			 * to make an uncompressed inline extent.
564 			 */
565 			ret = cow_file_range_inline(inode, start, end, 0,
566 						    BTRFS_COMPRESS_NONE, NULL);
567 		} else {
568 			/* try making a compressed inline extent */
569 			ret = cow_file_range_inline(inode, start, end,
570 						    total_compressed,
571 						    compress_type, pages);
572 		}
573 		if (ret <= 0) {
574 			unsigned long clear_flags = EXTENT_DELALLOC |
575 				EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
576 				EXTENT_DO_ACCOUNTING;
577 			unsigned long page_error_op;
578 
579 			page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
580 
581 			/*
582 			 * inline extent creation worked or returned error,
583 			 * we don't need to create any more async work items.
584 			 * Unlock and free up our temp pages.
585 			 *
586 			 * We use DO_ACCOUNTING here because we need the
587 			 * delalloc_release_metadata to be done _after_ we drop
588 			 * our outstanding extent for clearing delalloc for this
589 			 * range.
590 			 */
591 			extent_clear_unlock_delalloc(inode, start, end, end,
592 						     NULL, clear_flags,
593 						     PAGE_UNLOCK |
594 						     PAGE_CLEAR_DIRTY |
595 						     PAGE_SET_WRITEBACK |
596 						     page_error_op |
597 						     PAGE_END_WRITEBACK);
598 			goto free_pages_out;
599 		}
600 	}
601 
602 	if (will_compress) {
603 		/*
604 		 * we aren't doing an inline extent round the compressed size
605 		 * up to a block size boundary so the allocator does sane
606 		 * things
607 		 */
608 		total_compressed = ALIGN(total_compressed, blocksize);
609 
610 		/*
611 		 * one last check to make sure the compression is really a
612 		 * win, compare the page count read with the blocks on disk,
613 		 * compression must free at least one sector size
614 		 */
615 		total_in = ALIGN(total_in, PAGE_SIZE);
616 		if (total_compressed + blocksize <= total_in) {
617 			*num_added += 1;
618 
619 			/*
620 			 * The async work queues will take care of doing actual
621 			 * allocation on disk for these compressed pages, and
622 			 * will submit them to the elevator.
623 			 */
624 			add_async_extent(async_cow, start, total_in,
625 					total_compressed, pages, nr_pages,
626 					compress_type);
627 
628 			if (start + total_in < end) {
629 				start += total_in;
630 				pages = NULL;
631 				cond_resched();
632 				goto again;
633 			}
634 			return;
635 		}
636 	}
637 	if (pages) {
638 		/*
639 		 * the compression code ran but failed to make things smaller,
640 		 * free any pages it allocated and our page pointer array
641 		 */
642 		for (i = 0; i < nr_pages; i++) {
643 			WARN_ON(pages[i]->mapping);
644 			put_page(pages[i]);
645 		}
646 		kfree(pages);
647 		pages = NULL;
648 		total_compressed = 0;
649 		nr_pages = 0;
650 
651 		/* flag the file so we don't compress in the future */
652 		if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
653 		    !(BTRFS_I(inode)->prop_compress)) {
654 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
655 		}
656 	}
657 cleanup_and_bail_uncompressed:
658 	/*
659 	 * No compression, but we still need to write the pages in the file
660 	 * we've been given so far.  redirty the locked page if it corresponds
661 	 * to our extent and set things up for the async work queue to run
662 	 * cow_file_range to do the normal delalloc dance.
663 	 */
664 	if (page_offset(locked_page) >= start &&
665 	    page_offset(locked_page) <= end)
666 		__set_page_dirty_nobuffers(locked_page);
667 		/* unlocked later on in the async handlers */
668 
669 	if (redirty)
670 		extent_range_redirty_for_io(inode, start, end);
671 	add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
672 			 BTRFS_COMPRESS_NONE);
673 	*num_added += 1;
674 
675 	return;
676 
677 free_pages_out:
678 	for (i = 0; i < nr_pages; i++) {
679 		WARN_ON(pages[i]->mapping);
680 		put_page(pages[i]);
681 	}
682 	kfree(pages);
683 }
684 
685 static void free_async_extent_pages(struct async_extent *async_extent)
686 {
687 	int i;
688 
689 	if (!async_extent->pages)
690 		return;
691 
692 	for (i = 0; i < async_extent->nr_pages; i++) {
693 		WARN_ON(async_extent->pages[i]->mapping);
694 		put_page(async_extent->pages[i]);
695 	}
696 	kfree(async_extent->pages);
697 	async_extent->nr_pages = 0;
698 	async_extent->pages = NULL;
699 }
700 
701 /*
702  * phase two of compressed writeback.  This is the ordered portion
703  * of the code, which only gets called in the order the work was
704  * queued.  We walk all the async extents created by compress_file_range
705  * and send them down to the disk.
706  */
707 static noinline void submit_compressed_extents(struct inode *inode,
708 					      struct async_cow *async_cow)
709 {
710 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
711 	struct async_extent *async_extent;
712 	u64 alloc_hint = 0;
713 	struct btrfs_key ins;
714 	struct extent_map *em;
715 	struct btrfs_root *root = BTRFS_I(inode)->root;
716 	struct extent_io_tree *io_tree;
717 	int ret = 0;
718 
719 again:
720 	while (!list_empty(&async_cow->extents)) {
721 		async_extent = list_entry(async_cow->extents.next,
722 					  struct async_extent, list);
723 		list_del(&async_extent->list);
724 
725 		io_tree = &BTRFS_I(inode)->io_tree;
726 
727 retry:
728 		/* did the compression code fall back to uncompressed IO? */
729 		if (!async_extent->pages) {
730 			int page_started = 0;
731 			unsigned long nr_written = 0;
732 
733 			lock_extent(io_tree, async_extent->start,
734 					 async_extent->start +
735 					 async_extent->ram_size - 1);
736 
737 			/* allocate blocks */
738 			ret = cow_file_range(inode, async_cow->locked_page,
739 					     async_extent->start,
740 					     async_extent->start +
741 					     async_extent->ram_size - 1,
742 					     async_extent->start +
743 					     async_extent->ram_size - 1,
744 					     &page_started, &nr_written, 0,
745 					     NULL);
746 
747 			/* JDM XXX */
748 
749 			/*
750 			 * if page_started, cow_file_range inserted an
751 			 * inline extent and took care of all the unlocking
752 			 * and IO for us.  Otherwise, we need to submit
753 			 * all those pages down to the drive.
754 			 */
755 			if (!page_started && !ret)
756 				extent_write_locked_range(inode,
757 						  async_extent->start,
758 						  async_extent->start +
759 						  async_extent->ram_size - 1,
760 						  WB_SYNC_ALL);
761 			else if (ret)
762 				unlock_page(async_cow->locked_page);
763 			kfree(async_extent);
764 			cond_resched();
765 			continue;
766 		}
767 
768 		lock_extent(io_tree, async_extent->start,
769 			    async_extent->start + async_extent->ram_size - 1);
770 
771 		ret = btrfs_reserve_extent(root, async_extent->ram_size,
772 					   async_extent->compressed_size,
773 					   async_extent->compressed_size,
774 					   0, alloc_hint, &ins, 1, 1);
775 		if (ret) {
776 			free_async_extent_pages(async_extent);
777 
778 			if (ret == -ENOSPC) {
779 				unlock_extent(io_tree, async_extent->start,
780 					      async_extent->start +
781 					      async_extent->ram_size - 1);
782 
783 				/*
784 				 * we need to redirty the pages if we decide to
785 				 * fallback to uncompressed IO, otherwise we
786 				 * will not submit these pages down to lower
787 				 * layers.
788 				 */
789 				extent_range_redirty_for_io(inode,
790 						async_extent->start,
791 						async_extent->start +
792 						async_extent->ram_size - 1);
793 
794 				goto retry;
795 			}
796 			goto out_free;
797 		}
798 		/*
799 		 * here we're doing allocation and writeback of the
800 		 * compressed pages
801 		 */
802 		em = create_io_em(inode, async_extent->start,
803 				  async_extent->ram_size, /* len */
804 				  async_extent->start, /* orig_start */
805 				  ins.objectid, /* block_start */
806 				  ins.offset, /* block_len */
807 				  ins.offset, /* orig_block_len */
808 				  async_extent->ram_size, /* ram_bytes */
809 				  async_extent->compress_type,
810 				  BTRFS_ORDERED_COMPRESSED);
811 		if (IS_ERR(em))
812 			/* ret value is not necessary due to void function */
813 			goto out_free_reserve;
814 		free_extent_map(em);
815 
816 		ret = btrfs_add_ordered_extent_compress(inode,
817 						async_extent->start,
818 						ins.objectid,
819 						async_extent->ram_size,
820 						ins.offset,
821 						BTRFS_ORDERED_COMPRESSED,
822 						async_extent->compress_type);
823 		if (ret) {
824 			btrfs_drop_extent_cache(BTRFS_I(inode),
825 						async_extent->start,
826 						async_extent->start +
827 						async_extent->ram_size - 1, 0);
828 			goto out_free_reserve;
829 		}
830 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
831 
832 		/*
833 		 * clear dirty, set writeback and unlock the pages.
834 		 */
835 		extent_clear_unlock_delalloc(inode, async_extent->start,
836 				async_extent->start +
837 				async_extent->ram_size - 1,
838 				async_extent->start +
839 				async_extent->ram_size - 1,
840 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
841 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
842 				PAGE_SET_WRITEBACK);
843 		if (btrfs_submit_compressed_write(inode,
844 				    async_extent->start,
845 				    async_extent->ram_size,
846 				    ins.objectid,
847 				    ins.offset, async_extent->pages,
848 				    async_extent->nr_pages,
849 				    async_cow->write_flags)) {
850 			struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
851 			struct page *p = async_extent->pages[0];
852 			const u64 start = async_extent->start;
853 			const u64 end = start + async_extent->ram_size - 1;
854 
855 			p->mapping = inode->i_mapping;
856 			tree->ops->writepage_end_io_hook(p, start, end,
857 							 NULL, 0);
858 			p->mapping = NULL;
859 			extent_clear_unlock_delalloc(inode, start, end, end,
860 						     NULL, 0,
861 						     PAGE_END_WRITEBACK |
862 						     PAGE_SET_ERROR);
863 			free_async_extent_pages(async_extent);
864 		}
865 		alloc_hint = ins.objectid + ins.offset;
866 		kfree(async_extent);
867 		cond_resched();
868 	}
869 	return;
870 out_free_reserve:
871 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
872 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
873 out_free:
874 	extent_clear_unlock_delalloc(inode, async_extent->start,
875 				     async_extent->start +
876 				     async_extent->ram_size - 1,
877 				     async_extent->start +
878 				     async_extent->ram_size - 1,
879 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
880 				     EXTENT_DELALLOC_NEW |
881 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
882 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
883 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
884 				     PAGE_SET_ERROR);
885 	free_async_extent_pages(async_extent);
886 	kfree(async_extent);
887 	goto again;
888 }
889 
890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
891 				      u64 num_bytes)
892 {
893 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
894 	struct extent_map *em;
895 	u64 alloc_hint = 0;
896 
897 	read_lock(&em_tree->lock);
898 	em = search_extent_mapping(em_tree, start, num_bytes);
899 	if (em) {
900 		/*
901 		 * if block start isn't an actual block number then find the
902 		 * first block in this inode and use that as a hint.  If that
903 		 * block is also bogus then just don't worry about it.
904 		 */
905 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
906 			free_extent_map(em);
907 			em = search_extent_mapping(em_tree, 0, 0);
908 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
909 				alloc_hint = em->block_start;
910 			if (em)
911 				free_extent_map(em);
912 		} else {
913 			alloc_hint = em->block_start;
914 			free_extent_map(em);
915 		}
916 	}
917 	read_unlock(&em_tree->lock);
918 
919 	return alloc_hint;
920 }
921 
922 /*
923  * when extent_io.c finds a delayed allocation range in the file,
924  * the call backs end up in this code.  The basic idea is to
925  * allocate extents on disk for the range, and create ordered data structs
926  * in ram to track those extents.
927  *
928  * locked_page is the page that writepage had locked already.  We use
929  * it to make sure we don't do extra locks or unlocks.
930  *
931  * *page_started is set to one if we unlock locked_page and do everything
932  * required to start IO on it.  It may be clean and already done with
933  * IO when we return.
934  */
935 static noinline int cow_file_range(struct inode *inode,
936 				   struct page *locked_page,
937 				   u64 start, u64 end, u64 delalloc_end,
938 				   int *page_started, unsigned long *nr_written,
939 				   int unlock, struct btrfs_dedupe_hash *hash)
940 {
941 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
942 	struct btrfs_root *root = BTRFS_I(inode)->root;
943 	u64 alloc_hint = 0;
944 	u64 num_bytes;
945 	unsigned long ram_size;
946 	u64 cur_alloc_size = 0;
947 	u64 blocksize = fs_info->sectorsize;
948 	struct btrfs_key ins;
949 	struct extent_map *em;
950 	unsigned clear_bits;
951 	unsigned long page_ops;
952 	bool extent_reserved = false;
953 	int ret = 0;
954 
955 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
956 		WARN_ON_ONCE(1);
957 		ret = -EINVAL;
958 		goto out_unlock;
959 	}
960 
961 	num_bytes = ALIGN(end - start + 1, blocksize);
962 	num_bytes = max(blocksize,  num_bytes);
963 	ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
964 
965 	inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
966 
967 	if (start == 0) {
968 		/* lets try to make an inline extent */
969 		ret = cow_file_range_inline(inode, start, end, 0,
970 					    BTRFS_COMPRESS_NONE, NULL);
971 		if (ret == 0) {
972 			/*
973 			 * We use DO_ACCOUNTING here because we need the
974 			 * delalloc_release_metadata to be run _after_ we drop
975 			 * our outstanding extent for clearing delalloc for this
976 			 * range.
977 			 */
978 			extent_clear_unlock_delalloc(inode, start, end,
979 				     delalloc_end, NULL,
980 				     EXTENT_LOCKED | EXTENT_DELALLOC |
981 				     EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
982 				     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
983 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
984 				     PAGE_END_WRITEBACK);
985 			*nr_written = *nr_written +
986 			     (end - start + PAGE_SIZE) / PAGE_SIZE;
987 			*page_started = 1;
988 			goto out;
989 		} else if (ret < 0) {
990 			goto out_unlock;
991 		}
992 	}
993 
994 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
995 	btrfs_drop_extent_cache(BTRFS_I(inode), start,
996 			start + num_bytes - 1, 0);
997 
998 	while (num_bytes > 0) {
999 		cur_alloc_size = num_bytes;
1000 		ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1001 					   fs_info->sectorsize, 0, alloc_hint,
1002 					   &ins, 1, 1);
1003 		if (ret < 0)
1004 			goto out_unlock;
1005 		cur_alloc_size = ins.offset;
1006 		extent_reserved = true;
1007 
1008 		ram_size = ins.offset;
1009 		em = create_io_em(inode, start, ins.offset, /* len */
1010 				  start, /* orig_start */
1011 				  ins.objectid, /* block_start */
1012 				  ins.offset, /* block_len */
1013 				  ins.offset, /* orig_block_len */
1014 				  ram_size, /* ram_bytes */
1015 				  BTRFS_COMPRESS_NONE, /* compress_type */
1016 				  BTRFS_ORDERED_REGULAR /* type */);
1017 		if (IS_ERR(em)) {
1018 			ret = PTR_ERR(em);
1019 			goto out_reserve;
1020 		}
1021 		free_extent_map(em);
1022 
1023 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024 					       ram_size, cur_alloc_size, 0);
1025 		if (ret)
1026 			goto out_drop_extent_cache;
1027 
1028 		if (root->root_key.objectid ==
1029 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 			ret = btrfs_reloc_clone_csums(inode, start,
1031 						      cur_alloc_size);
1032 			/*
1033 			 * Only drop cache here, and process as normal.
1034 			 *
1035 			 * We must not allow extent_clear_unlock_delalloc()
1036 			 * at out_unlock label to free meta of this ordered
1037 			 * extent, as its meta should be freed by
1038 			 * btrfs_finish_ordered_io().
1039 			 *
1040 			 * So we must continue until @start is increased to
1041 			 * skip current ordered extent.
1042 			 */
1043 			if (ret)
1044 				btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045 						start + ram_size - 1, 0);
1046 		}
1047 
1048 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1049 
1050 		/* we're not doing compressed IO, don't unlock the first
1051 		 * page (which the caller expects to stay locked), don't
1052 		 * clear any dirty bits and don't set any writeback bits
1053 		 *
1054 		 * Do set the Private2 bit so we know this page was properly
1055 		 * setup for writepage
1056 		 */
1057 		page_ops = unlock ? PAGE_UNLOCK : 0;
1058 		page_ops |= PAGE_SET_PRIVATE2;
1059 
1060 		extent_clear_unlock_delalloc(inode, start,
1061 					     start + ram_size - 1,
1062 					     delalloc_end, locked_page,
1063 					     EXTENT_LOCKED | EXTENT_DELALLOC,
1064 					     page_ops);
1065 		if (num_bytes < cur_alloc_size)
1066 			num_bytes = 0;
1067 		else
1068 			num_bytes -= cur_alloc_size;
1069 		alloc_hint = ins.objectid + ins.offset;
1070 		start += cur_alloc_size;
1071 		extent_reserved = false;
1072 
1073 		/*
1074 		 * btrfs_reloc_clone_csums() error, since start is increased
1075 		 * extent_clear_unlock_delalloc() at out_unlock label won't
1076 		 * free metadata of current ordered extent, we're OK to exit.
1077 		 */
1078 		if (ret)
1079 			goto out_unlock;
1080 	}
1081 out:
1082 	return ret;
1083 
1084 out_drop_extent_cache:
1085 	btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1086 out_reserve:
1087 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1088 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1089 out_unlock:
1090 	clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1091 		EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1092 	page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1093 		PAGE_END_WRITEBACK;
1094 	/*
1095 	 * If we reserved an extent for our delalloc range (or a subrange) and
1096 	 * failed to create the respective ordered extent, then it means that
1097 	 * when we reserved the extent we decremented the extent's size from
1098 	 * the data space_info's bytes_may_use counter and incremented the
1099 	 * space_info's bytes_reserved counter by the same amount. We must make
1100 	 * sure extent_clear_unlock_delalloc() does not try to decrement again
1101 	 * the data space_info's bytes_may_use counter, therefore we do not pass
1102 	 * it the flag EXTENT_CLEAR_DATA_RESV.
1103 	 */
1104 	if (extent_reserved) {
1105 		extent_clear_unlock_delalloc(inode, start,
1106 					     start + cur_alloc_size,
1107 					     start + cur_alloc_size,
1108 					     locked_page,
1109 					     clear_bits,
1110 					     page_ops);
1111 		start += cur_alloc_size;
1112 		if (start >= end)
1113 			goto out;
1114 	}
1115 	extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1116 				     locked_page,
1117 				     clear_bits | EXTENT_CLEAR_DATA_RESV,
1118 				     page_ops);
1119 	goto out;
1120 }
1121 
1122 /*
1123  * work queue call back to started compression on a file and pages
1124  */
1125 static noinline void async_cow_start(struct btrfs_work *work)
1126 {
1127 	struct async_cow *async_cow;
1128 	int num_added = 0;
1129 	async_cow = container_of(work, struct async_cow, work);
1130 
1131 	compress_file_range(async_cow->inode, async_cow->locked_page,
1132 			    async_cow->start, async_cow->end, async_cow,
1133 			    &num_added);
1134 	if (num_added == 0) {
1135 		btrfs_add_delayed_iput(async_cow->inode);
1136 		async_cow->inode = NULL;
1137 	}
1138 }
1139 
1140 /*
1141  * work queue call back to submit previously compressed pages
1142  */
1143 static noinline void async_cow_submit(struct btrfs_work *work)
1144 {
1145 	struct btrfs_fs_info *fs_info;
1146 	struct async_cow *async_cow;
1147 	struct btrfs_root *root;
1148 	unsigned long nr_pages;
1149 
1150 	async_cow = container_of(work, struct async_cow, work);
1151 
1152 	root = async_cow->root;
1153 	fs_info = root->fs_info;
1154 	nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1155 		PAGE_SHIFT;
1156 
1157 	/* atomic_sub_return implies a barrier */
1158 	if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1159 	    5 * SZ_1M)
1160 		cond_wake_up_nomb(&fs_info->async_submit_wait);
1161 
1162 	if (async_cow->inode)
1163 		submit_compressed_extents(async_cow->inode, async_cow);
1164 }
1165 
1166 static noinline void async_cow_free(struct btrfs_work *work)
1167 {
1168 	struct async_cow *async_cow;
1169 	async_cow = container_of(work, struct async_cow, work);
1170 	if (async_cow->inode)
1171 		btrfs_add_delayed_iput(async_cow->inode);
1172 	kfree(async_cow);
1173 }
1174 
1175 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1176 				u64 start, u64 end, int *page_started,
1177 				unsigned long *nr_written,
1178 				unsigned int write_flags)
1179 {
1180 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1181 	struct async_cow *async_cow;
1182 	struct btrfs_root *root = BTRFS_I(inode)->root;
1183 	unsigned long nr_pages;
1184 	u64 cur_end;
1185 
1186 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1187 			 1, 0, NULL);
1188 	while (start < end) {
1189 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1190 		BUG_ON(!async_cow); /* -ENOMEM */
1191 		async_cow->inode = igrab(inode);
1192 		async_cow->root = root;
1193 		async_cow->locked_page = locked_page;
1194 		async_cow->start = start;
1195 		async_cow->write_flags = write_flags;
1196 
1197 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1198 		    !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1199 			cur_end = end;
1200 		else
1201 			cur_end = min(end, start + SZ_512K - 1);
1202 
1203 		async_cow->end = cur_end;
1204 		INIT_LIST_HEAD(&async_cow->extents);
1205 
1206 		btrfs_init_work(&async_cow->work,
1207 				btrfs_delalloc_helper,
1208 				async_cow_start, async_cow_submit,
1209 				async_cow_free);
1210 
1211 		nr_pages = (cur_end - start + PAGE_SIZE) >>
1212 			PAGE_SHIFT;
1213 		atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1214 
1215 		btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1216 
1217 		*nr_written += nr_pages;
1218 		start = cur_end + 1;
1219 	}
1220 	*page_started = 1;
1221 	return 0;
1222 }
1223 
1224 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1225 					u64 bytenr, u64 num_bytes)
1226 {
1227 	int ret;
1228 	struct btrfs_ordered_sum *sums;
1229 	LIST_HEAD(list);
1230 
1231 	ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1232 				       bytenr + num_bytes - 1, &list, 0);
1233 	if (ret == 0 && list_empty(&list))
1234 		return 0;
1235 
1236 	while (!list_empty(&list)) {
1237 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1238 		list_del(&sums->list);
1239 		kfree(sums);
1240 	}
1241 	if (ret < 0)
1242 		return ret;
1243 	return 1;
1244 }
1245 
1246 /*
1247  * when nowcow writeback call back.  This checks for snapshots or COW copies
1248  * of the extents that exist in the file, and COWs the file as required.
1249  *
1250  * If no cow copies or snapshots exist, we write directly to the existing
1251  * blocks on disk
1252  */
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 				       struct page *locked_page,
1255 			      u64 start, u64 end, int *page_started, int force,
1256 			      unsigned long *nr_written)
1257 {
1258 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 	struct btrfs_root *root = BTRFS_I(inode)->root;
1260 	struct extent_buffer *leaf;
1261 	struct btrfs_path *path;
1262 	struct btrfs_file_extent_item *fi;
1263 	struct btrfs_key found_key;
1264 	struct extent_map *em;
1265 	u64 cow_start;
1266 	u64 cur_offset;
1267 	u64 extent_end;
1268 	u64 extent_offset;
1269 	u64 disk_bytenr;
1270 	u64 num_bytes;
1271 	u64 disk_num_bytes;
1272 	u64 ram_bytes;
1273 	int extent_type;
1274 	int ret;
1275 	int type;
1276 	int nocow;
1277 	int check_prev = 1;
1278 	bool nolock;
1279 	u64 ino = btrfs_ino(BTRFS_I(inode));
1280 
1281 	path = btrfs_alloc_path();
1282 	if (!path) {
1283 		extent_clear_unlock_delalloc(inode, start, end, end,
1284 					     locked_page,
1285 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1286 					     EXTENT_DO_ACCOUNTING |
1287 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1288 					     PAGE_CLEAR_DIRTY |
1289 					     PAGE_SET_WRITEBACK |
1290 					     PAGE_END_WRITEBACK);
1291 		return -ENOMEM;
1292 	}
1293 
1294 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1295 
1296 	cow_start = (u64)-1;
1297 	cur_offset = start;
1298 	while (1) {
1299 		ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1300 					       cur_offset, 0);
1301 		if (ret < 0)
1302 			goto error;
1303 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 			leaf = path->nodes[0];
1305 			btrfs_item_key_to_cpu(leaf, &found_key,
1306 					      path->slots[0] - 1);
1307 			if (found_key.objectid == ino &&
1308 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1309 				path->slots[0]--;
1310 		}
1311 		check_prev = 0;
1312 next_slot:
1313 		leaf = path->nodes[0];
1314 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 			ret = btrfs_next_leaf(root, path);
1316 			if (ret < 0) {
1317 				if (cow_start != (u64)-1)
1318 					cur_offset = cow_start;
1319 				goto error;
1320 			}
1321 			if (ret > 0)
1322 				break;
1323 			leaf = path->nodes[0];
1324 		}
1325 
1326 		nocow = 0;
1327 		disk_bytenr = 0;
1328 		num_bytes = 0;
1329 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1330 
1331 		if (found_key.objectid > ino)
1332 			break;
1333 		if (WARN_ON_ONCE(found_key.objectid < ino) ||
1334 		    found_key.type < BTRFS_EXTENT_DATA_KEY) {
1335 			path->slots[0]++;
1336 			goto next_slot;
1337 		}
1338 		if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1339 		    found_key.offset > end)
1340 			break;
1341 
1342 		if (found_key.offset > cur_offset) {
1343 			extent_end = found_key.offset;
1344 			extent_type = 0;
1345 			goto out_check;
1346 		}
1347 
1348 		fi = btrfs_item_ptr(leaf, path->slots[0],
1349 				    struct btrfs_file_extent_item);
1350 		extent_type = btrfs_file_extent_type(leaf, fi);
1351 
1352 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1353 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1354 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1355 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1356 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1357 			extent_end = found_key.offset +
1358 				btrfs_file_extent_num_bytes(leaf, fi);
1359 			disk_num_bytes =
1360 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1361 			if (extent_end <= start) {
1362 				path->slots[0]++;
1363 				goto next_slot;
1364 			}
1365 			if (disk_bytenr == 0)
1366 				goto out_check;
1367 			if (btrfs_file_extent_compression(leaf, fi) ||
1368 			    btrfs_file_extent_encryption(leaf, fi) ||
1369 			    btrfs_file_extent_other_encoding(leaf, fi))
1370 				goto out_check;
1371 			/*
1372 			 * Do the same check as in btrfs_cross_ref_exist but
1373 			 * without the unnecessary search.
1374 			 */
1375 			if (btrfs_file_extent_generation(leaf, fi) <=
1376 			    btrfs_root_last_snapshot(&root->root_item))
1377 				goto out_check;
1378 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1379 				goto out_check;
1380 			if (btrfs_extent_readonly(fs_info, disk_bytenr))
1381 				goto out_check;
1382 			ret = btrfs_cross_ref_exist(root, ino,
1383 						    found_key.offset -
1384 						    extent_offset, disk_bytenr);
1385 			if (ret) {
1386 				/*
1387 				 * ret could be -EIO if the above fails to read
1388 				 * metadata.
1389 				 */
1390 				if (ret < 0) {
1391 					if (cow_start != (u64)-1)
1392 						cur_offset = cow_start;
1393 					goto error;
1394 				}
1395 
1396 				WARN_ON_ONCE(nolock);
1397 				goto out_check;
1398 			}
1399 			disk_bytenr += extent_offset;
1400 			disk_bytenr += cur_offset - found_key.offset;
1401 			num_bytes = min(end + 1, extent_end) - cur_offset;
1402 			/*
1403 			 * if there are pending snapshots for this root,
1404 			 * we fall into common COW way.
1405 			 */
1406 			if (!nolock && atomic_read(&root->snapshot_force_cow))
1407 				goto out_check;
1408 			/*
1409 			 * force cow if csum exists in the range.
1410 			 * this ensure that csum for a given extent are
1411 			 * either valid or do not exist.
1412 			 */
1413 			ret = csum_exist_in_range(fs_info, disk_bytenr,
1414 						  num_bytes);
1415 			if (ret) {
1416 				/*
1417 				 * ret could be -EIO if the above fails to read
1418 				 * metadata.
1419 				 */
1420 				if (ret < 0) {
1421 					if (cow_start != (u64)-1)
1422 						cur_offset = cow_start;
1423 					goto error;
1424 				}
1425 				WARN_ON_ONCE(nolock);
1426 				goto out_check;
1427 			}
1428 			if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1429 				goto out_check;
1430 			nocow = 1;
1431 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1432 			extent_end = found_key.offset +
1433 				btrfs_file_extent_ram_bytes(leaf, fi);
1434 			extent_end = ALIGN(extent_end,
1435 					   fs_info->sectorsize);
1436 		} else {
1437 			BUG_ON(1);
1438 		}
1439 out_check:
1440 		if (extent_end <= start) {
1441 			path->slots[0]++;
1442 			if (nocow)
1443 				btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1444 			goto next_slot;
1445 		}
1446 		if (!nocow) {
1447 			if (cow_start == (u64)-1)
1448 				cow_start = cur_offset;
1449 			cur_offset = extent_end;
1450 			if (cur_offset > end)
1451 				break;
1452 			path->slots[0]++;
1453 			goto next_slot;
1454 		}
1455 
1456 		btrfs_release_path(path);
1457 		if (cow_start != (u64)-1) {
1458 			ret = cow_file_range(inode, locked_page,
1459 					     cow_start, found_key.offset - 1,
1460 					     end, page_started, nr_written, 1,
1461 					     NULL);
1462 			if (ret) {
1463 				if (nocow)
1464 					btrfs_dec_nocow_writers(fs_info,
1465 								disk_bytenr);
1466 				goto error;
1467 			}
1468 			cow_start = (u64)-1;
1469 		}
1470 
1471 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1472 			u64 orig_start = found_key.offset - extent_offset;
1473 
1474 			em = create_io_em(inode, cur_offset, num_bytes,
1475 					  orig_start,
1476 					  disk_bytenr, /* block_start */
1477 					  num_bytes, /* block_len */
1478 					  disk_num_bytes, /* orig_block_len */
1479 					  ram_bytes, BTRFS_COMPRESS_NONE,
1480 					  BTRFS_ORDERED_PREALLOC);
1481 			if (IS_ERR(em)) {
1482 				if (nocow)
1483 					btrfs_dec_nocow_writers(fs_info,
1484 								disk_bytenr);
1485 				ret = PTR_ERR(em);
1486 				goto error;
1487 			}
1488 			free_extent_map(em);
1489 		}
1490 
1491 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1492 			type = BTRFS_ORDERED_PREALLOC;
1493 		} else {
1494 			type = BTRFS_ORDERED_NOCOW;
1495 		}
1496 
1497 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1498 					       num_bytes, num_bytes, type);
1499 		if (nocow)
1500 			btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1501 		BUG_ON(ret); /* -ENOMEM */
1502 
1503 		if (root->root_key.objectid ==
1504 		    BTRFS_DATA_RELOC_TREE_OBJECTID)
1505 			/*
1506 			 * Error handled later, as we must prevent
1507 			 * extent_clear_unlock_delalloc() in error handler
1508 			 * from freeing metadata of created ordered extent.
1509 			 */
1510 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1511 						      num_bytes);
1512 
1513 		extent_clear_unlock_delalloc(inode, cur_offset,
1514 					     cur_offset + num_bytes - 1, end,
1515 					     locked_page, EXTENT_LOCKED |
1516 					     EXTENT_DELALLOC |
1517 					     EXTENT_CLEAR_DATA_RESV,
1518 					     PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1519 
1520 		cur_offset = extent_end;
1521 
1522 		/*
1523 		 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 		 * handler, as metadata for created ordered extent will only
1525 		 * be freed by btrfs_finish_ordered_io().
1526 		 */
1527 		if (ret)
1528 			goto error;
1529 		if (cur_offset > end)
1530 			break;
1531 	}
1532 	btrfs_release_path(path);
1533 
1534 	if (cur_offset <= end && cow_start == (u64)-1) {
1535 		cow_start = cur_offset;
1536 		cur_offset = end;
1537 	}
1538 
1539 	if (cow_start != (u64)-1) {
1540 		ret = cow_file_range(inode, locked_page, cow_start, end, end,
1541 				     page_started, nr_written, 1, NULL);
1542 		if (ret)
1543 			goto error;
1544 	}
1545 
1546 error:
1547 	if (ret && cur_offset < end)
1548 		extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1549 					     locked_page, EXTENT_LOCKED |
1550 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1551 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1552 					     PAGE_CLEAR_DIRTY |
1553 					     PAGE_SET_WRITEBACK |
1554 					     PAGE_END_WRITEBACK);
1555 	btrfs_free_path(path);
1556 	return ret;
1557 }
1558 
1559 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1560 {
1561 
1562 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1563 	    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1564 		return 0;
1565 
1566 	/*
1567 	 * @defrag_bytes is a hint value, no spinlock held here,
1568 	 * if is not zero, it means the file is defragging.
1569 	 * Force cow if given extent needs to be defragged.
1570 	 */
1571 	if (BTRFS_I(inode)->defrag_bytes &&
1572 	    test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1573 			   EXTENT_DEFRAG, 0, NULL))
1574 		return 1;
1575 
1576 	return 0;
1577 }
1578 
1579 /*
1580  * extent_io.c call back to do delayed allocation processing
1581  */
1582 static int run_delalloc_range(void *private_data, struct page *locked_page,
1583 			      u64 start, u64 end, int *page_started,
1584 			      unsigned long *nr_written,
1585 			      struct writeback_control *wbc)
1586 {
1587 	struct inode *inode = private_data;
1588 	int ret;
1589 	int force_cow = need_force_cow(inode, start, end);
1590 	unsigned int write_flags = wbc_to_write_flags(wbc);
1591 
1592 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1593 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1594 					 page_started, 1, nr_written);
1595 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1596 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1597 					 page_started, 0, nr_written);
1598 	} else if (!inode_need_compress(inode, start, end)) {
1599 		ret = cow_file_range(inode, locked_page, start, end, end,
1600 				      page_started, nr_written, 1, NULL);
1601 	} else {
1602 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1603 			&BTRFS_I(inode)->runtime_flags);
1604 		ret = cow_file_range_async(inode, locked_page, start, end,
1605 					   page_started, nr_written,
1606 					   write_flags);
1607 	}
1608 	if (ret)
1609 		btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1610 	return ret;
1611 }
1612 
1613 static void btrfs_split_extent_hook(void *private_data,
1614 				    struct extent_state *orig, u64 split)
1615 {
1616 	struct inode *inode = private_data;
1617 	u64 size;
1618 
1619 	/* not delalloc, ignore it */
1620 	if (!(orig->state & EXTENT_DELALLOC))
1621 		return;
1622 
1623 	size = orig->end - orig->start + 1;
1624 	if (size > BTRFS_MAX_EXTENT_SIZE) {
1625 		u32 num_extents;
1626 		u64 new_size;
1627 
1628 		/*
1629 		 * See the explanation in btrfs_merge_extent_hook, the same
1630 		 * applies here, just in reverse.
1631 		 */
1632 		new_size = orig->end - split + 1;
1633 		num_extents = count_max_extents(new_size);
1634 		new_size = split - orig->start;
1635 		num_extents += count_max_extents(new_size);
1636 		if (count_max_extents(size) >= num_extents)
1637 			return;
1638 	}
1639 
1640 	spin_lock(&BTRFS_I(inode)->lock);
1641 	btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1642 	spin_unlock(&BTRFS_I(inode)->lock);
1643 }
1644 
1645 /*
1646  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1647  * extents so we can keep track of new extents that are just merged onto old
1648  * extents, such as when we are doing sequential writes, so we can properly
1649  * account for the metadata space we'll need.
1650  */
1651 static void btrfs_merge_extent_hook(void *private_data,
1652 				    struct extent_state *new,
1653 				    struct extent_state *other)
1654 {
1655 	struct inode *inode = private_data;
1656 	u64 new_size, old_size;
1657 	u32 num_extents;
1658 
1659 	/* not delalloc, ignore it */
1660 	if (!(other->state & EXTENT_DELALLOC))
1661 		return;
1662 
1663 	if (new->start > other->start)
1664 		new_size = new->end - other->start + 1;
1665 	else
1666 		new_size = other->end - new->start + 1;
1667 
1668 	/* we're not bigger than the max, unreserve the space and go */
1669 	if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1670 		spin_lock(&BTRFS_I(inode)->lock);
1671 		btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1672 		spin_unlock(&BTRFS_I(inode)->lock);
1673 		return;
1674 	}
1675 
1676 	/*
1677 	 * We have to add up either side to figure out how many extents were
1678 	 * accounted for before we merged into one big extent.  If the number of
1679 	 * extents we accounted for is <= the amount we need for the new range
1680 	 * then we can return, otherwise drop.  Think of it like this
1681 	 *
1682 	 * [ 4k][MAX_SIZE]
1683 	 *
1684 	 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1685 	 * need 2 outstanding extents, on one side we have 1 and the other side
1686 	 * we have 1 so they are == and we can return.  But in this case
1687 	 *
1688 	 * [MAX_SIZE+4k][MAX_SIZE+4k]
1689 	 *
1690 	 * Each range on their own accounts for 2 extents, but merged together
1691 	 * they are only 3 extents worth of accounting, so we need to drop in
1692 	 * this case.
1693 	 */
1694 	old_size = other->end - other->start + 1;
1695 	num_extents = count_max_extents(old_size);
1696 	old_size = new->end - new->start + 1;
1697 	num_extents += count_max_extents(old_size);
1698 	if (count_max_extents(new_size) >= num_extents)
1699 		return;
1700 
1701 	spin_lock(&BTRFS_I(inode)->lock);
1702 	btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1703 	spin_unlock(&BTRFS_I(inode)->lock);
1704 }
1705 
1706 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1707 				      struct inode *inode)
1708 {
1709 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1710 
1711 	spin_lock(&root->delalloc_lock);
1712 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1713 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1714 			      &root->delalloc_inodes);
1715 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1716 			&BTRFS_I(inode)->runtime_flags);
1717 		root->nr_delalloc_inodes++;
1718 		if (root->nr_delalloc_inodes == 1) {
1719 			spin_lock(&fs_info->delalloc_root_lock);
1720 			BUG_ON(!list_empty(&root->delalloc_root));
1721 			list_add_tail(&root->delalloc_root,
1722 				      &fs_info->delalloc_roots);
1723 			spin_unlock(&fs_info->delalloc_root_lock);
1724 		}
1725 	}
1726 	spin_unlock(&root->delalloc_lock);
1727 }
1728 
1729 
1730 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1731 				struct btrfs_inode *inode)
1732 {
1733 	struct btrfs_fs_info *fs_info = root->fs_info;
1734 
1735 	if (!list_empty(&inode->delalloc_inodes)) {
1736 		list_del_init(&inode->delalloc_inodes);
1737 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1738 			  &inode->runtime_flags);
1739 		root->nr_delalloc_inodes--;
1740 		if (!root->nr_delalloc_inodes) {
1741 			ASSERT(list_empty(&root->delalloc_inodes));
1742 			spin_lock(&fs_info->delalloc_root_lock);
1743 			BUG_ON(list_empty(&root->delalloc_root));
1744 			list_del_init(&root->delalloc_root);
1745 			spin_unlock(&fs_info->delalloc_root_lock);
1746 		}
1747 	}
1748 }
1749 
1750 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1751 				     struct btrfs_inode *inode)
1752 {
1753 	spin_lock(&root->delalloc_lock);
1754 	__btrfs_del_delalloc_inode(root, inode);
1755 	spin_unlock(&root->delalloc_lock);
1756 }
1757 
1758 /*
1759  * extent_io.c set_bit_hook, used to track delayed allocation
1760  * bytes in this file, and to maintain the list of inodes that
1761  * have pending delalloc work to be done.
1762  */
1763 static void btrfs_set_bit_hook(void *private_data,
1764 			       struct extent_state *state, unsigned *bits)
1765 {
1766 	struct inode *inode = private_data;
1767 
1768 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1769 
1770 	if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1771 		WARN_ON(1);
1772 	/*
1773 	 * set_bit and clear bit hooks normally require _irqsave/restore
1774 	 * but in this case, we are only testing for the DELALLOC
1775 	 * bit, which is only set or cleared with irqs on
1776 	 */
1777 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1778 		struct btrfs_root *root = BTRFS_I(inode)->root;
1779 		u64 len = state->end + 1 - state->start;
1780 		u32 num_extents = count_max_extents(len);
1781 		bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1782 
1783 		spin_lock(&BTRFS_I(inode)->lock);
1784 		btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1785 		spin_unlock(&BTRFS_I(inode)->lock);
1786 
1787 		/* For sanity tests */
1788 		if (btrfs_is_testing(fs_info))
1789 			return;
1790 
1791 		percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1792 					 fs_info->delalloc_batch);
1793 		spin_lock(&BTRFS_I(inode)->lock);
1794 		BTRFS_I(inode)->delalloc_bytes += len;
1795 		if (*bits & EXTENT_DEFRAG)
1796 			BTRFS_I(inode)->defrag_bytes += len;
1797 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1798 					 &BTRFS_I(inode)->runtime_flags))
1799 			btrfs_add_delalloc_inodes(root, inode);
1800 		spin_unlock(&BTRFS_I(inode)->lock);
1801 	}
1802 
1803 	if (!(state->state & EXTENT_DELALLOC_NEW) &&
1804 	    (*bits & EXTENT_DELALLOC_NEW)) {
1805 		spin_lock(&BTRFS_I(inode)->lock);
1806 		BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1807 			state->start;
1808 		spin_unlock(&BTRFS_I(inode)->lock);
1809 	}
1810 }
1811 
1812 /*
1813  * extent_io.c clear_bit_hook, see set_bit_hook for why
1814  */
1815 static void btrfs_clear_bit_hook(void *private_data,
1816 				 struct extent_state *state,
1817 				 unsigned *bits)
1818 {
1819 	struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1820 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1821 	u64 len = state->end + 1 - state->start;
1822 	u32 num_extents = count_max_extents(len);
1823 
1824 	if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1825 		spin_lock(&inode->lock);
1826 		inode->defrag_bytes -= len;
1827 		spin_unlock(&inode->lock);
1828 	}
1829 
1830 	/*
1831 	 * set_bit and clear bit hooks normally require _irqsave/restore
1832 	 * but in this case, we are only testing for the DELALLOC
1833 	 * bit, which is only set or cleared with irqs on
1834 	 */
1835 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1836 		struct btrfs_root *root = inode->root;
1837 		bool do_list = !btrfs_is_free_space_inode(inode);
1838 
1839 		spin_lock(&inode->lock);
1840 		btrfs_mod_outstanding_extents(inode, -num_extents);
1841 		spin_unlock(&inode->lock);
1842 
1843 		/*
1844 		 * We don't reserve metadata space for space cache inodes so we
1845 		 * don't need to call dellalloc_release_metadata if there is an
1846 		 * error.
1847 		 */
1848 		if (*bits & EXTENT_CLEAR_META_RESV &&
1849 		    root != fs_info->tree_root)
1850 			btrfs_delalloc_release_metadata(inode, len, false);
1851 
1852 		/* For sanity tests. */
1853 		if (btrfs_is_testing(fs_info))
1854 			return;
1855 
1856 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1857 		    do_list && !(state->state & EXTENT_NORESERVE) &&
1858 		    (*bits & EXTENT_CLEAR_DATA_RESV))
1859 			btrfs_free_reserved_data_space_noquota(
1860 					&inode->vfs_inode,
1861 					state->start, len);
1862 
1863 		percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1864 					 fs_info->delalloc_batch);
1865 		spin_lock(&inode->lock);
1866 		inode->delalloc_bytes -= len;
1867 		if (do_list && inode->delalloc_bytes == 0 &&
1868 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1869 					&inode->runtime_flags))
1870 			btrfs_del_delalloc_inode(root, inode);
1871 		spin_unlock(&inode->lock);
1872 	}
1873 
1874 	if ((state->state & EXTENT_DELALLOC_NEW) &&
1875 	    (*bits & EXTENT_DELALLOC_NEW)) {
1876 		spin_lock(&inode->lock);
1877 		ASSERT(inode->new_delalloc_bytes >= len);
1878 		inode->new_delalloc_bytes -= len;
1879 		spin_unlock(&inode->lock);
1880 	}
1881 }
1882 
1883 /*
1884  * Merge bio hook, this must check the chunk tree to make sure we don't create
1885  * bios that span stripes or chunks
1886  *
1887  * return 1 if page cannot be merged to bio
1888  * return 0 if page can be merged to bio
1889  * return error otherwise
1890  */
1891 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1892 			 size_t size, struct bio *bio,
1893 			 unsigned long bio_flags)
1894 {
1895 	struct inode *inode = page->mapping->host;
1896 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1897 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1898 	u64 length = 0;
1899 	u64 map_length;
1900 	int ret;
1901 
1902 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1903 		return 0;
1904 
1905 	length = bio->bi_iter.bi_size;
1906 	map_length = length;
1907 	ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1908 			      NULL, 0);
1909 	if (ret < 0)
1910 		return ret;
1911 	if (map_length < length + size)
1912 		return 1;
1913 	return 0;
1914 }
1915 
1916 /*
1917  * in order to insert checksums into the metadata in large chunks,
1918  * we wait until bio submission time.   All the pages in the bio are
1919  * checksummed and sums are attached onto the ordered extent record.
1920  *
1921  * At IO completion time the cums attached on the ordered extent record
1922  * are inserted into the btree
1923  */
1924 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1925 				    u64 bio_offset)
1926 {
1927 	struct inode *inode = private_data;
1928 	blk_status_t ret = 0;
1929 
1930 	ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1931 	BUG_ON(ret); /* -ENOMEM */
1932 	return 0;
1933 }
1934 
1935 /*
1936  * in order to insert checksums into the metadata in large chunks,
1937  * we wait until bio submission time.   All the pages in the bio are
1938  * checksummed and sums are attached onto the ordered extent record.
1939  *
1940  * At IO completion time the cums attached on the ordered extent record
1941  * are inserted into the btree
1942  */
1943 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1944 			  int mirror_num)
1945 {
1946 	struct inode *inode = private_data;
1947 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1948 	blk_status_t ret;
1949 
1950 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1951 	if (ret) {
1952 		bio->bi_status = ret;
1953 		bio_endio(bio);
1954 	}
1955 	return ret;
1956 }
1957 
1958 /*
1959  * extent_io.c submission hook. This does the right thing for csum calculation
1960  * on write, or reading the csums from the tree before a read.
1961  *
1962  * Rules about async/sync submit,
1963  * a) read:				sync submit
1964  *
1965  * b) write without checksum:		sync submit
1966  *
1967  * c) write with checksum:
1968  *    c-1) if bio is issued by fsync:	sync submit
1969  *         (sync_writers != 0)
1970  *
1971  *    c-2) if root is reloc root:	sync submit
1972  *         (only in case of buffered IO)
1973  *
1974  *    c-3) otherwise:			async submit
1975  */
1976 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1977 				 int mirror_num, unsigned long bio_flags,
1978 				 u64 bio_offset)
1979 {
1980 	struct inode *inode = private_data;
1981 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1982 	struct btrfs_root *root = BTRFS_I(inode)->root;
1983 	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1984 	blk_status_t ret = 0;
1985 	int skip_sum;
1986 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1987 
1988 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1989 
1990 	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1991 		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1992 
1993 	if (bio_op(bio) != REQ_OP_WRITE) {
1994 		ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1995 		if (ret)
1996 			goto out;
1997 
1998 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1999 			ret = btrfs_submit_compressed_read(inode, bio,
2000 							   mirror_num,
2001 							   bio_flags);
2002 			goto out;
2003 		} else if (!skip_sum) {
2004 			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2005 			if (ret)
2006 				goto out;
2007 		}
2008 		goto mapit;
2009 	} else if (async && !skip_sum) {
2010 		/* csum items have already been cloned */
2011 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2012 			goto mapit;
2013 		/* we're doing a write, do the async checksumming */
2014 		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2015 					  bio_offset, inode,
2016 					  btrfs_submit_bio_start);
2017 		goto out;
2018 	} else if (!skip_sum) {
2019 		ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2020 		if (ret)
2021 			goto out;
2022 	}
2023 
2024 mapit:
2025 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2026 
2027 out:
2028 	if (ret) {
2029 		bio->bi_status = ret;
2030 		bio_endio(bio);
2031 	}
2032 	return ret;
2033 }
2034 
2035 /*
2036  * given a list of ordered sums record them in the inode.  This happens
2037  * at IO completion time based on sums calculated at bio submission time.
2038  */
2039 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2040 			     struct inode *inode, struct list_head *list)
2041 {
2042 	struct btrfs_ordered_sum *sum;
2043 	int ret;
2044 
2045 	list_for_each_entry(sum, list, list) {
2046 		trans->adding_csums = true;
2047 		ret = btrfs_csum_file_blocks(trans,
2048 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
2049 		trans->adding_csums = false;
2050 		if (ret)
2051 			return ret;
2052 	}
2053 	return 0;
2054 }
2055 
2056 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2057 			      unsigned int extra_bits,
2058 			      struct extent_state **cached_state, int dedupe)
2059 {
2060 	WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2061 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2062 				   extra_bits, cached_state);
2063 }
2064 
2065 /* see btrfs_writepage_start_hook for details on why this is required */
2066 struct btrfs_writepage_fixup {
2067 	struct page *page;
2068 	struct btrfs_work work;
2069 };
2070 
2071 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2072 {
2073 	struct btrfs_writepage_fixup *fixup;
2074 	struct btrfs_ordered_extent *ordered;
2075 	struct extent_state *cached_state = NULL;
2076 	struct extent_changeset *data_reserved = NULL;
2077 	struct page *page;
2078 	struct inode *inode;
2079 	u64 page_start;
2080 	u64 page_end;
2081 	int ret;
2082 
2083 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
2084 	page = fixup->page;
2085 again:
2086 	lock_page(page);
2087 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2088 		ClearPageChecked(page);
2089 		goto out_page;
2090 	}
2091 
2092 	inode = page->mapping->host;
2093 	page_start = page_offset(page);
2094 	page_end = page_offset(page) + PAGE_SIZE - 1;
2095 
2096 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2097 			 &cached_state);
2098 
2099 	/* already ordered? We're done */
2100 	if (PagePrivate2(page))
2101 		goto out;
2102 
2103 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2104 					PAGE_SIZE);
2105 	if (ordered) {
2106 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2107 				     page_end, &cached_state);
2108 		unlock_page(page);
2109 		btrfs_start_ordered_extent(inode, ordered, 1);
2110 		btrfs_put_ordered_extent(ordered);
2111 		goto again;
2112 	}
2113 
2114 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2115 					   PAGE_SIZE);
2116 	if (ret) {
2117 		mapping_set_error(page->mapping, ret);
2118 		end_extent_writepage(page, ret, page_start, page_end);
2119 		ClearPageChecked(page);
2120 		goto out;
2121 	 }
2122 
2123 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2124 					&cached_state, 0);
2125 	if (ret) {
2126 		mapping_set_error(page->mapping, ret);
2127 		end_extent_writepage(page, ret, page_start, page_end);
2128 		ClearPageChecked(page);
2129 		goto out;
2130 	}
2131 
2132 	ClearPageChecked(page);
2133 	set_page_dirty(page);
2134 	btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2135 out:
2136 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2137 			     &cached_state);
2138 out_page:
2139 	unlock_page(page);
2140 	put_page(page);
2141 	kfree(fixup);
2142 	extent_changeset_free(data_reserved);
2143 }
2144 
2145 /*
2146  * There are a few paths in the higher layers of the kernel that directly
2147  * set the page dirty bit without asking the filesystem if it is a
2148  * good idea.  This causes problems because we want to make sure COW
2149  * properly happens and the data=ordered rules are followed.
2150  *
2151  * In our case any range that doesn't have the ORDERED bit set
2152  * hasn't been properly setup for IO.  We kick off an async process
2153  * to fix it up.  The async helper will wait for ordered extents, set
2154  * the delalloc bit and make it safe to write the page.
2155  */
2156 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2157 {
2158 	struct inode *inode = page->mapping->host;
2159 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2160 	struct btrfs_writepage_fixup *fixup;
2161 
2162 	/* this page is properly in the ordered list */
2163 	if (TestClearPagePrivate2(page))
2164 		return 0;
2165 
2166 	if (PageChecked(page))
2167 		return -EAGAIN;
2168 
2169 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2170 	if (!fixup)
2171 		return -EAGAIN;
2172 
2173 	SetPageChecked(page);
2174 	get_page(page);
2175 	btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2176 			btrfs_writepage_fixup_worker, NULL, NULL);
2177 	fixup->page = page;
2178 	btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2179 	return -EBUSY;
2180 }
2181 
2182 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2183 				       struct inode *inode, u64 file_pos,
2184 				       u64 disk_bytenr, u64 disk_num_bytes,
2185 				       u64 num_bytes, u64 ram_bytes,
2186 				       u8 compression, u8 encryption,
2187 				       u16 other_encoding, int extent_type)
2188 {
2189 	struct btrfs_root *root = BTRFS_I(inode)->root;
2190 	struct btrfs_file_extent_item *fi;
2191 	struct btrfs_path *path;
2192 	struct extent_buffer *leaf;
2193 	struct btrfs_key ins;
2194 	u64 qg_released;
2195 	int extent_inserted = 0;
2196 	int ret;
2197 
2198 	path = btrfs_alloc_path();
2199 	if (!path)
2200 		return -ENOMEM;
2201 
2202 	/*
2203 	 * we may be replacing one extent in the tree with another.
2204 	 * The new extent is pinned in the extent map, and we don't want
2205 	 * to drop it from the cache until it is completely in the btree.
2206 	 *
2207 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
2208 	 * the caller is expected to unpin it and allow it to be merged
2209 	 * with the others.
2210 	 */
2211 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2212 				   file_pos + num_bytes, NULL, 0,
2213 				   1, sizeof(*fi), &extent_inserted);
2214 	if (ret)
2215 		goto out;
2216 
2217 	if (!extent_inserted) {
2218 		ins.objectid = btrfs_ino(BTRFS_I(inode));
2219 		ins.offset = file_pos;
2220 		ins.type = BTRFS_EXTENT_DATA_KEY;
2221 
2222 		path->leave_spinning = 1;
2223 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
2224 					      sizeof(*fi));
2225 		if (ret)
2226 			goto out;
2227 	}
2228 	leaf = path->nodes[0];
2229 	fi = btrfs_item_ptr(leaf, path->slots[0],
2230 			    struct btrfs_file_extent_item);
2231 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2232 	btrfs_set_file_extent_type(leaf, fi, extent_type);
2233 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2234 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2235 	btrfs_set_file_extent_offset(leaf, fi, 0);
2236 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2237 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2238 	btrfs_set_file_extent_compression(leaf, fi, compression);
2239 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
2240 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2241 
2242 	btrfs_mark_buffer_dirty(leaf);
2243 	btrfs_release_path(path);
2244 
2245 	inode_add_bytes(inode, num_bytes);
2246 
2247 	ins.objectid = disk_bytenr;
2248 	ins.offset = disk_num_bytes;
2249 	ins.type = BTRFS_EXTENT_ITEM_KEY;
2250 
2251 	/*
2252 	 * Release the reserved range from inode dirty range map, as it is
2253 	 * already moved into delayed_ref_head
2254 	 */
2255 	ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2256 	if (ret < 0)
2257 		goto out;
2258 	qg_released = ret;
2259 	ret = btrfs_alloc_reserved_file_extent(trans, root,
2260 					       btrfs_ino(BTRFS_I(inode)),
2261 					       file_pos, qg_released, &ins);
2262 out:
2263 	btrfs_free_path(path);
2264 
2265 	return ret;
2266 }
2267 
2268 /* snapshot-aware defrag */
2269 struct sa_defrag_extent_backref {
2270 	struct rb_node node;
2271 	struct old_sa_defrag_extent *old;
2272 	u64 root_id;
2273 	u64 inum;
2274 	u64 file_pos;
2275 	u64 extent_offset;
2276 	u64 num_bytes;
2277 	u64 generation;
2278 };
2279 
2280 struct old_sa_defrag_extent {
2281 	struct list_head list;
2282 	struct new_sa_defrag_extent *new;
2283 
2284 	u64 extent_offset;
2285 	u64 bytenr;
2286 	u64 offset;
2287 	u64 len;
2288 	int count;
2289 };
2290 
2291 struct new_sa_defrag_extent {
2292 	struct rb_root root;
2293 	struct list_head head;
2294 	struct btrfs_path *path;
2295 	struct inode *inode;
2296 	u64 file_pos;
2297 	u64 len;
2298 	u64 bytenr;
2299 	u64 disk_len;
2300 	u8 compress_type;
2301 };
2302 
2303 static int backref_comp(struct sa_defrag_extent_backref *b1,
2304 			struct sa_defrag_extent_backref *b2)
2305 {
2306 	if (b1->root_id < b2->root_id)
2307 		return -1;
2308 	else if (b1->root_id > b2->root_id)
2309 		return 1;
2310 
2311 	if (b1->inum < b2->inum)
2312 		return -1;
2313 	else if (b1->inum > b2->inum)
2314 		return 1;
2315 
2316 	if (b1->file_pos < b2->file_pos)
2317 		return -1;
2318 	else if (b1->file_pos > b2->file_pos)
2319 		return 1;
2320 
2321 	/*
2322 	 * [------------------------------] ===> (a range of space)
2323 	 *     |<--->|   |<---->| =============> (fs/file tree A)
2324 	 * |<---------------------------->| ===> (fs/file tree B)
2325 	 *
2326 	 * A range of space can refer to two file extents in one tree while
2327 	 * refer to only one file extent in another tree.
2328 	 *
2329 	 * So we may process a disk offset more than one time(two extents in A)
2330 	 * and locate at the same extent(one extent in B), then insert two same
2331 	 * backrefs(both refer to the extent in B).
2332 	 */
2333 	return 0;
2334 }
2335 
2336 static void backref_insert(struct rb_root *root,
2337 			   struct sa_defrag_extent_backref *backref)
2338 {
2339 	struct rb_node **p = &root->rb_node;
2340 	struct rb_node *parent = NULL;
2341 	struct sa_defrag_extent_backref *entry;
2342 	int ret;
2343 
2344 	while (*p) {
2345 		parent = *p;
2346 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2347 
2348 		ret = backref_comp(backref, entry);
2349 		if (ret < 0)
2350 			p = &(*p)->rb_left;
2351 		else
2352 			p = &(*p)->rb_right;
2353 	}
2354 
2355 	rb_link_node(&backref->node, parent, p);
2356 	rb_insert_color(&backref->node, root);
2357 }
2358 
2359 /*
2360  * Note the backref might has changed, and in this case we just return 0.
2361  */
2362 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2363 				       void *ctx)
2364 {
2365 	struct btrfs_file_extent_item *extent;
2366 	struct old_sa_defrag_extent *old = ctx;
2367 	struct new_sa_defrag_extent *new = old->new;
2368 	struct btrfs_path *path = new->path;
2369 	struct btrfs_key key;
2370 	struct btrfs_root *root;
2371 	struct sa_defrag_extent_backref *backref;
2372 	struct extent_buffer *leaf;
2373 	struct inode *inode = new->inode;
2374 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2375 	int slot;
2376 	int ret;
2377 	u64 extent_offset;
2378 	u64 num_bytes;
2379 
2380 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2381 	    inum == btrfs_ino(BTRFS_I(inode)))
2382 		return 0;
2383 
2384 	key.objectid = root_id;
2385 	key.type = BTRFS_ROOT_ITEM_KEY;
2386 	key.offset = (u64)-1;
2387 
2388 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2389 	if (IS_ERR(root)) {
2390 		if (PTR_ERR(root) == -ENOENT)
2391 			return 0;
2392 		WARN_ON(1);
2393 		btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2394 			 inum, offset, root_id);
2395 		return PTR_ERR(root);
2396 	}
2397 
2398 	key.objectid = inum;
2399 	key.type = BTRFS_EXTENT_DATA_KEY;
2400 	if (offset > (u64)-1 << 32)
2401 		key.offset = 0;
2402 	else
2403 		key.offset = offset;
2404 
2405 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2406 	if (WARN_ON(ret < 0))
2407 		return ret;
2408 	ret = 0;
2409 
2410 	while (1) {
2411 		cond_resched();
2412 
2413 		leaf = path->nodes[0];
2414 		slot = path->slots[0];
2415 
2416 		if (slot >= btrfs_header_nritems(leaf)) {
2417 			ret = btrfs_next_leaf(root, path);
2418 			if (ret < 0) {
2419 				goto out;
2420 			} else if (ret > 0) {
2421 				ret = 0;
2422 				goto out;
2423 			}
2424 			continue;
2425 		}
2426 
2427 		path->slots[0]++;
2428 
2429 		btrfs_item_key_to_cpu(leaf, &key, slot);
2430 
2431 		if (key.objectid > inum)
2432 			goto out;
2433 
2434 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2435 			continue;
2436 
2437 		extent = btrfs_item_ptr(leaf, slot,
2438 					struct btrfs_file_extent_item);
2439 
2440 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2441 			continue;
2442 
2443 		/*
2444 		 * 'offset' refers to the exact key.offset,
2445 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2446 		 * (key.offset - extent_offset).
2447 		 */
2448 		if (key.offset != offset)
2449 			continue;
2450 
2451 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2452 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2453 
2454 		if (extent_offset >= old->extent_offset + old->offset +
2455 		    old->len || extent_offset + num_bytes <=
2456 		    old->extent_offset + old->offset)
2457 			continue;
2458 		break;
2459 	}
2460 
2461 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2462 	if (!backref) {
2463 		ret = -ENOENT;
2464 		goto out;
2465 	}
2466 
2467 	backref->root_id = root_id;
2468 	backref->inum = inum;
2469 	backref->file_pos = offset;
2470 	backref->num_bytes = num_bytes;
2471 	backref->extent_offset = extent_offset;
2472 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2473 	backref->old = old;
2474 	backref_insert(&new->root, backref);
2475 	old->count++;
2476 out:
2477 	btrfs_release_path(path);
2478 	WARN_ON(ret);
2479 	return ret;
2480 }
2481 
2482 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2483 				   struct new_sa_defrag_extent *new)
2484 {
2485 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2486 	struct old_sa_defrag_extent *old, *tmp;
2487 	int ret;
2488 
2489 	new->path = path;
2490 
2491 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2492 		ret = iterate_inodes_from_logical(old->bytenr +
2493 						  old->extent_offset, fs_info,
2494 						  path, record_one_backref,
2495 						  old, false);
2496 		if (ret < 0 && ret != -ENOENT)
2497 			return false;
2498 
2499 		/* no backref to be processed for this extent */
2500 		if (!old->count) {
2501 			list_del(&old->list);
2502 			kfree(old);
2503 		}
2504 	}
2505 
2506 	if (list_empty(&new->head))
2507 		return false;
2508 
2509 	return true;
2510 }
2511 
2512 static int relink_is_mergable(struct extent_buffer *leaf,
2513 			      struct btrfs_file_extent_item *fi,
2514 			      struct new_sa_defrag_extent *new)
2515 {
2516 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2517 		return 0;
2518 
2519 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2520 		return 0;
2521 
2522 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2523 		return 0;
2524 
2525 	if (btrfs_file_extent_encryption(leaf, fi) ||
2526 	    btrfs_file_extent_other_encoding(leaf, fi))
2527 		return 0;
2528 
2529 	return 1;
2530 }
2531 
2532 /*
2533  * Note the backref might has changed, and in this case we just return 0.
2534  */
2535 static noinline int relink_extent_backref(struct btrfs_path *path,
2536 				 struct sa_defrag_extent_backref *prev,
2537 				 struct sa_defrag_extent_backref *backref)
2538 {
2539 	struct btrfs_file_extent_item *extent;
2540 	struct btrfs_file_extent_item *item;
2541 	struct btrfs_ordered_extent *ordered;
2542 	struct btrfs_trans_handle *trans;
2543 	struct btrfs_root *root;
2544 	struct btrfs_key key;
2545 	struct extent_buffer *leaf;
2546 	struct old_sa_defrag_extent *old = backref->old;
2547 	struct new_sa_defrag_extent *new = old->new;
2548 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2549 	struct inode *inode;
2550 	struct extent_state *cached = NULL;
2551 	int ret = 0;
2552 	u64 start;
2553 	u64 len;
2554 	u64 lock_start;
2555 	u64 lock_end;
2556 	bool merge = false;
2557 	int index;
2558 
2559 	if (prev && prev->root_id == backref->root_id &&
2560 	    prev->inum == backref->inum &&
2561 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2562 		merge = true;
2563 
2564 	/* step 1: get root */
2565 	key.objectid = backref->root_id;
2566 	key.type = BTRFS_ROOT_ITEM_KEY;
2567 	key.offset = (u64)-1;
2568 
2569 	index = srcu_read_lock(&fs_info->subvol_srcu);
2570 
2571 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2572 	if (IS_ERR(root)) {
2573 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2574 		if (PTR_ERR(root) == -ENOENT)
2575 			return 0;
2576 		return PTR_ERR(root);
2577 	}
2578 
2579 	if (btrfs_root_readonly(root)) {
2580 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2581 		return 0;
2582 	}
2583 
2584 	/* step 2: get inode */
2585 	key.objectid = backref->inum;
2586 	key.type = BTRFS_INODE_ITEM_KEY;
2587 	key.offset = 0;
2588 
2589 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2590 	if (IS_ERR(inode)) {
2591 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2592 		return 0;
2593 	}
2594 
2595 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2596 
2597 	/* step 3: relink backref */
2598 	lock_start = backref->file_pos;
2599 	lock_end = backref->file_pos + backref->num_bytes - 1;
2600 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2601 			 &cached);
2602 
2603 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2604 	if (ordered) {
2605 		btrfs_put_ordered_extent(ordered);
2606 		goto out_unlock;
2607 	}
2608 
2609 	trans = btrfs_join_transaction(root);
2610 	if (IS_ERR(trans)) {
2611 		ret = PTR_ERR(trans);
2612 		goto out_unlock;
2613 	}
2614 
2615 	key.objectid = backref->inum;
2616 	key.type = BTRFS_EXTENT_DATA_KEY;
2617 	key.offset = backref->file_pos;
2618 
2619 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2620 	if (ret < 0) {
2621 		goto out_free_path;
2622 	} else if (ret > 0) {
2623 		ret = 0;
2624 		goto out_free_path;
2625 	}
2626 
2627 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2628 				struct btrfs_file_extent_item);
2629 
2630 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2631 	    backref->generation)
2632 		goto out_free_path;
2633 
2634 	btrfs_release_path(path);
2635 
2636 	start = backref->file_pos;
2637 	if (backref->extent_offset < old->extent_offset + old->offset)
2638 		start += old->extent_offset + old->offset -
2639 			 backref->extent_offset;
2640 
2641 	len = min(backref->extent_offset + backref->num_bytes,
2642 		  old->extent_offset + old->offset + old->len);
2643 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2644 
2645 	ret = btrfs_drop_extents(trans, root, inode, start,
2646 				 start + len, 1);
2647 	if (ret)
2648 		goto out_free_path;
2649 again:
2650 	key.objectid = btrfs_ino(BTRFS_I(inode));
2651 	key.type = BTRFS_EXTENT_DATA_KEY;
2652 	key.offset = start;
2653 
2654 	path->leave_spinning = 1;
2655 	if (merge) {
2656 		struct btrfs_file_extent_item *fi;
2657 		u64 extent_len;
2658 		struct btrfs_key found_key;
2659 
2660 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2661 		if (ret < 0)
2662 			goto out_free_path;
2663 
2664 		path->slots[0]--;
2665 		leaf = path->nodes[0];
2666 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2667 
2668 		fi = btrfs_item_ptr(leaf, path->slots[0],
2669 				    struct btrfs_file_extent_item);
2670 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2671 
2672 		if (extent_len + found_key.offset == start &&
2673 		    relink_is_mergable(leaf, fi, new)) {
2674 			btrfs_set_file_extent_num_bytes(leaf, fi,
2675 							extent_len + len);
2676 			btrfs_mark_buffer_dirty(leaf);
2677 			inode_add_bytes(inode, len);
2678 
2679 			ret = 1;
2680 			goto out_free_path;
2681 		} else {
2682 			merge = false;
2683 			btrfs_release_path(path);
2684 			goto again;
2685 		}
2686 	}
2687 
2688 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2689 					sizeof(*extent));
2690 	if (ret) {
2691 		btrfs_abort_transaction(trans, ret);
2692 		goto out_free_path;
2693 	}
2694 
2695 	leaf = path->nodes[0];
2696 	item = btrfs_item_ptr(leaf, path->slots[0],
2697 				struct btrfs_file_extent_item);
2698 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2699 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2700 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2701 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2702 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2703 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2704 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2705 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2706 	btrfs_set_file_extent_encryption(leaf, item, 0);
2707 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2708 
2709 	btrfs_mark_buffer_dirty(leaf);
2710 	inode_add_bytes(inode, len);
2711 	btrfs_release_path(path);
2712 
2713 	ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2714 			new->disk_len, 0,
2715 			backref->root_id, backref->inum,
2716 			new->file_pos);	/* start - extent_offset */
2717 	if (ret) {
2718 		btrfs_abort_transaction(trans, ret);
2719 		goto out_free_path;
2720 	}
2721 
2722 	ret = 1;
2723 out_free_path:
2724 	btrfs_release_path(path);
2725 	path->leave_spinning = 0;
2726 	btrfs_end_transaction(trans);
2727 out_unlock:
2728 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2729 			     &cached);
2730 	iput(inode);
2731 	return ret;
2732 }
2733 
2734 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2735 {
2736 	struct old_sa_defrag_extent *old, *tmp;
2737 
2738 	if (!new)
2739 		return;
2740 
2741 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2742 		kfree(old);
2743 	}
2744 	kfree(new);
2745 }
2746 
2747 static void relink_file_extents(struct new_sa_defrag_extent *new)
2748 {
2749 	struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2750 	struct btrfs_path *path;
2751 	struct sa_defrag_extent_backref *backref;
2752 	struct sa_defrag_extent_backref *prev = NULL;
2753 	struct rb_node *node;
2754 	int ret;
2755 
2756 	path = btrfs_alloc_path();
2757 	if (!path)
2758 		return;
2759 
2760 	if (!record_extent_backrefs(path, new)) {
2761 		btrfs_free_path(path);
2762 		goto out;
2763 	}
2764 	btrfs_release_path(path);
2765 
2766 	while (1) {
2767 		node = rb_first(&new->root);
2768 		if (!node)
2769 			break;
2770 		rb_erase(node, &new->root);
2771 
2772 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2773 
2774 		ret = relink_extent_backref(path, prev, backref);
2775 		WARN_ON(ret < 0);
2776 
2777 		kfree(prev);
2778 
2779 		if (ret == 1)
2780 			prev = backref;
2781 		else
2782 			prev = NULL;
2783 		cond_resched();
2784 	}
2785 	kfree(prev);
2786 
2787 	btrfs_free_path(path);
2788 out:
2789 	free_sa_defrag_extent(new);
2790 
2791 	atomic_dec(&fs_info->defrag_running);
2792 	wake_up(&fs_info->transaction_wait);
2793 }
2794 
2795 static struct new_sa_defrag_extent *
2796 record_old_file_extents(struct inode *inode,
2797 			struct btrfs_ordered_extent *ordered)
2798 {
2799 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2800 	struct btrfs_root *root = BTRFS_I(inode)->root;
2801 	struct btrfs_path *path;
2802 	struct btrfs_key key;
2803 	struct old_sa_defrag_extent *old;
2804 	struct new_sa_defrag_extent *new;
2805 	int ret;
2806 
2807 	new = kmalloc(sizeof(*new), GFP_NOFS);
2808 	if (!new)
2809 		return NULL;
2810 
2811 	new->inode = inode;
2812 	new->file_pos = ordered->file_offset;
2813 	new->len = ordered->len;
2814 	new->bytenr = ordered->start;
2815 	new->disk_len = ordered->disk_len;
2816 	new->compress_type = ordered->compress_type;
2817 	new->root = RB_ROOT;
2818 	INIT_LIST_HEAD(&new->head);
2819 
2820 	path = btrfs_alloc_path();
2821 	if (!path)
2822 		goto out_kfree;
2823 
2824 	key.objectid = btrfs_ino(BTRFS_I(inode));
2825 	key.type = BTRFS_EXTENT_DATA_KEY;
2826 	key.offset = new->file_pos;
2827 
2828 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2829 	if (ret < 0)
2830 		goto out_free_path;
2831 	if (ret > 0 && path->slots[0] > 0)
2832 		path->slots[0]--;
2833 
2834 	/* find out all the old extents for the file range */
2835 	while (1) {
2836 		struct btrfs_file_extent_item *extent;
2837 		struct extent_buffer *l;
2838 		int slot;
2839 		u64 num_bytes;
2840 		u64 offset;
2841 		u64 end;
2842 		u64 disk_bytenr;
2843 		u64 extent_offset;
2844 
2845 		l = path->nodes[0];
2846 		slot = path->slots[0];
2847 
2848 		if (slot >= btrfs_header_nritems(l)) {
2849 			ret = btrfs_next_leaf(root, path);
2850 			if (ret < 0)
2851 				goto out_free_path;
2852 			else if (ret > 0)
2853 				break;
2854 			continue;
2855 		}
2856 
2857 		btrfs_item_key_to_cpu(l, &key, slot);
2858 
2859 		if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2860 			break;
2861 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2862 			break;
2863 		if (key.offset >= new->file_pos + new->len)
2864 			break;
2865 
2866 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2867 
2868 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2869 		if (key.offset + num_bytes < new->file_pos)
2870 			goto next;
2871 
2872 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2873 		if (!disk_bytenr)
2874 			goto next;
2875 
2876 		extent_offset = btrfs_file_extent_offset(l, extent);
2877 
2878 		old = kmalloc(sizeof(*old), GFP_NOFS);
2879 		if (!old)
2880 			goto out_free_path;
2881 
2882 		offset = max(new->file_pos, key.offset);
2883 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2884 
2885 		old->bytenr = disk_bytenr;
2886 		old->extent_offset = extent_offset;
2887 		old->offset = offset - key.offset;
2888 		old->len = end - offset;
2889 		old->new = new;
2890 		old->count = 0;
2891 		list_add_tail(&old->list, &new->head);
2892 next:
2893 		path->slots[0]++;
2894 		cond_resched();
2895 	}
2896 
2897 	btrfs_free_path(path);
2898 	atomic_inc(&fs_info->defrag_running);
2899 
2900 	return new;
2901 
2902 out_free_path:
2903 	btrfs_free_path(path);
2904 out_kfree:
2905 	free_sa_defrag_extent(new);
2906 	return NULL;
2907 }
2908 
2909 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2910 					 u64 start, u64 len)
2911 {
2912 	struct btrfs_block_group_cache *cache;
2913 
2914 	cache = btrfs_lookup_block_group(fs_info, start);
2915 	ASSERT(cache);
2916 
2917 	spin_lock(&cache->lock);
2918 	cache->delalloc_bytes -= len;
2919 	spin_unlock(&cache->lock);
2920 
2921 	btrfs_put_block_group(cache);
2922 }
2923 
2924 /* as ordered data IO finishes, this gets called so we can finish
2925  * an ordered extent if the range of bytes in the file it covers are
2926  * fully written.
2927  */
2928 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2929 {
2930 	struct inode *inode = ordered_extent->inode;
2931 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2932 	struct btrfs_root *root = BTRFS_I(inode)->root;
2933 	struct btrfs_trans_handle *trans = NULL;
2934 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2935 	struct extent_state *cached_state = NULL;
2936 	struct new_sa_defrag_extent *new = NULL;
2937 	int compress_type = 0;
2938 	int ret = 0;
2939 	u64 logical_len = ordered_extent->len;
2940 	bool nolock;
2941 	bool truncated = false;
2942 	bool range_locked = false;
2943 	bool clear_new_delalloc_bytes = false;
2944 	bool clear_reserved_extent = true;
2945 
2946 	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2947 	    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2948 	    !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2949 		clear_new_delalloc_bytes = true;
2950 
2951 	nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2952 
2953 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2954 		ret = -EIO;
2955 		goto out;
2956 	}
2957 
2958 	btrfs_free_io_failure_record(BTRFS_I(inode),
2959 			ordered_extent->file_offset,
2960 			ordered_extent->file_offset +
2961 			ordered_extent->len - 1);
2962 
2963 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2964 		truncated = true;
2965 		logical_len = ordered_extent->truncated_len;
2966 		/* Truncated the entire extent, don't bother adding */
2967 		if (!logical_len)
2968 			goto out;
2969 	}
2970 
2971 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2972 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2973 
2974 		/*
2975 		 * For mwrite(mmap + memset to write) case, we still reserve
2976 		 * space for NOCOW range.
2977 		 * As NOCOW won't cause a new delayed ref, just free the space
2978 		 */
2979 		btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2980 				       ordered_extent->len);
2981 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2982 		if (nolock)
2983 			trans = btrfs_join_transaction_nolock(root);
2984 		else
2985 			trans = btrfs_join_transaction(root);
2986 		if (IS_ERR(trans)) {
2987 			ret = PTR_ERR(trans);
2988 			trans = NULL;
2989 			goto out;
2990 		}
2991 		trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2992 		ret = btrfs_update_inode_fallback(trans, root, inode);
2993 		if (ret) /* -ENOMEM or corruption */
2994 			btrfs_abort_transaction(trans, ret);
2995 		goto out;
2996 	}
2997 
2998 	range_locked = true;
2999 	lock_extent_bits(io_tree, ordered_extent->file_offset,
3000 			 ordered_extent->file_offset + ordered_extent->len - 1,
3001 			 &cached_state);
3002 
3003 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
3004 			ordered_extent->file_offset + ordered_extent->len - 1,
3005 			EXTENT_DEFRAG, 0, cached_state);
3006 	if (ret) {
3007 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3008 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3009 			/* the inode is shared */
3010 			new = record_old_file_extents(inode, ordered_extent);
3011 
3012 		clear_extent_bit(io_tree, ordered_extent->file_offset,
3013 			ordered_extent->file_offset + ordered_extent->len - 1,
3014 			EXTENT_DEFRAG, 0, 0, &cached_state);
3015 	}
3016 
3017 	if (nolock)
3018 		trans = btrfs_join_transaction_nolock(root);
3019 	else
3020 		trans = btrfs_join_transaction(root);
3021 	if (IS_ERR(trans)) {
3022 		ret = PTR_ERR(trans);
3023 		trans = NULL;
3024 		goto out;
3025 	}
3026 
3027 	trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3028 
3029 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3030 		compress_type = ordered_extent->compress_type;
3031 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3032 		BUG_ON(compress_type);
3033 		btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3034 				       ordered_extent->len);
3035 		ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3036 						ordered_extent->file_offset,
3037 						ordered_extent->file_offset +
3038 						logical_len);
3039 	} else {
3040 		BUG_ON(root == fs_info->tree_root);
3041 		ret = insert_reserved_file_extent(trans, inode,
3042 						ordered_extent->file_offset,
3043 						ordered_extent->start,
3044 						ordered_extent->disk_len,
3045 						logical_len, logical_len,
3046 						compress_type, 0, 0,
3047 						BTRFS_FILE_EXTENT_REG);
3048 		if (!ret) {
3049 			clear_reserved_extent = false;
3050 			btrfs_release_delalloc_bytes(fs_info,
3051 						     ordered_extent->start,
3052 						     ordered_extent->disk_len);
3053 		}
3054 	}
3055 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3056 			   ordered_extent->file_offset, ordered_extent->len,
3057 			   trans->transid);
3058 	if (ret < 0) {
3059 		btrfs_abort_transaction(trans, ret);
3060 		goto out;
3061 	}
3062 
3063 	ret = add_pending_csums(trans, inode, &ordered_extent->list);
3064 	if (ret) {
3065 		btrfs_abort_transaction(trans, ret);
3066 		goto out;
3067 	}
3068 
3069 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3070 	ret = btrfs_update_inode_fallback(trans, root, inode);
3071 	if (ret) { /* -ENOMEM or corruption */
3072 		btrfs_abort_transaction(trans, ret);
3073 		goto out;
3074 	}
3075 	ret = 0;
3076 out:
3077 	if (range_locked || clear_new_delalloc_bytes) {
3078 		unsigned int clear_bits = 0;
3079 
3080 		if (range_locked)
3081 			clear_bits |= EXTENT_LOCKED;
3082 		if (clear_new_delalloc_bytes)
3083 			clear_bits |= EXTENT_DELALLOC_NEW;
3084 		clear_extent_bit(&BTRFS_I(inode)->io_tree,
3085 				 ordered_extent->file_offset,
3086 				 ordered_extent->file_offset +
3087 				 ordered_extent->len - 1,
3088 				 clear_bits,
3089 				 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3090 				 0, &cached_state);
3091 	}
3092 
3093 	if (trans)
3094 		btrfs_end_transaction(trans);
3095 
3096 	if (ret || truncated) {
3097 		u64 start, end;
3098 
3099 		if (truncated)
3100 			start = ordered_extent->file_offset + logical_len;
3101 		else
3102 			start = ordered_extent->file_offset;
3103 		end = ordered_extent->file_offset + ordered_extent->len - 1;
3104 		clear_extent_uptodate(io_tree, start, end, NULL);
3105 
3106 		/* Drop the cache for the part of the extent we didn't write. */
3107 		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3108 
3109 		/*
3110 		 * If the ordered extent had an IOERR or something else went
3111 		 * wrong we need to return the space for this ordered extent
3112 		 * back to the allocator.  We only free the extent in the
3113 		 * truncated case if we didn't write out the extent at all.
3114 		 *
3115 		 * If we made it past insert_reserved_file_extent before we
3116 		 * errored out then we don't need to do this as the accounting
3117 		 * has already been done.
3118 		 */
3119 		if ((ret || !logical_len) &&
3120 		    clear_reserved_extent &&
3121 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3122 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3123 			btrfs_free_reserved_extent(fs_info,
3124 						   ordered_extent->start,
3125 						   ordered_extent->disk_len, 1);
3126 	}
3127 
3128 
3129 	/*
3130 	 * This needs to be done to make sure anybody waiting knows we are done
3131 	 * updating everything for this ordered extent.
3132 	 */
3133 	btrfs_remove_ordered_extent(inode, ordered_extent);
3134 
3135 	/* for snapshot-aware defrag */
3136 	if (new) {
3137 		if (ret) {
3138 			free_sa_defrag_extent(new);
3139 			atomic_dec(&fs_info->defrag_running);
3140 		} else {
3141 			relink_file_extents(new);
3142 		}
3143 	}
3144 
3145 	/* once for us */
3146 	btrfs_put_ordered_extent(ordered_extent);
3147 	/* once for the tree */
3148 	btrfs_put_ordered_extent(ordered_extent);
3149 
3150 	/* Try to release some metadata so we don't get an OOM but don't wait */
3151 	btrfs_btree_balance_dirty_nodelay(fs_info);
3152 
3153 	return ret;
3154 }
3155 
3156 static void finish_ordered_fn(struct btrfs_work *work)
3157 {
3158 	struct btrfs_ordered_extent *ordered_extent;
3159 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3160 	btrfs_finish_ordered_io(ordered_extent);
3161 }
3162 
3163 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3164 				struct extent_state *state, int uptodate)
3165 {
3166 	struct inode *inode = page->mapping->host;
3167 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3168 	struct btrfs_ordered_extent *ordered_extent = NULL;
3169 	struct btrfs_workqueue *wq;
3170 	btrfs_work_func_t func;
3171 
3172 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3173 
3174 	ClearPagePrivate2(page);
3175 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3176 					    end - start + 1, uptodate))
3177 		return;
3178 
3179 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3180 		wq = fs_info->endio_freespace_worker;
3181 		func = btrfs_freespace_write_helper;
3182 	} else {
3183 		wq = fs_info->endio_write_workers;
3184 		func = btrfs_endio_write_helper;
3185 	}
3186 
3187 	btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3188 			NULL);
3189 	btrfs_queue_work(wq, &ordered_extent->work);
3190 }
3191 
3192 static int __readpage_endio_check(struct inode *inode,
3193 				  struct btrfs_io_bio *io_bio,
3194 				  int icsum, struct page *page,
3195 				  int pgoff, u64 start, size_t len)
3196 {
3197 	char *kaddr;
3198 	u32 csum_expected;
3199 	u32 csum = ~(u32)0;
3200 
3201 	csum_expected = *(((u32 *)io_bio->csum) + icsum);
3202 
3203 	kaddr = kmap_atomic(page);
3204 	csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3205 	btrfs_csum_final(csum, (u8 *)&csum);
3206 	if (csum != csum_expected)
3207 		goto zeroit;
3208 
3209 	kunmap_atomic(kaddr);
3210 	return 0;
3211 zeroit:
3212 	btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3213 				    io_bio->mirror_num);
3214 	memset(kaddr + pgoff, 1, len);
3215 	flush_dcache_page(page);
3216 	kunmap_atomic(kaddr);
3217 	return -EIO;
3218 }
3219 
3220 /*
3221  * when reads are done, we need to check csums to verify the data is correct
3222  * if there's a match, we allow the bio to finish.  If not, the code in
3223  * extent_io.c will try to find good copies for us.
3224  */
3225 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3226 				      u64 phy_offset, struct page *page,
3227 				      u64 start, u64 end, int mirror)
3228 {
3229 	size_t offset = start - page_offset(page);
3230 	struct inode *inode = page->mapping->host;
3231 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3232 	struct btrfs_root *root = BTRFS_I(inode)->root;
3233 
3234 	if (PageChecked(page)) {
3235 		ClearPageChecked(page);
3236 		return 0;
3237 	}
3238 
3239 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3240 		return 0;
3241 
3242 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3243 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3244 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3245 		return 0;
3246 	}
3247 
3248 	phy_offset >>= inode->i_sb->s_blocksize_bits;
3249 	return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3250 				      start, (size_t)(end - start + 1));
3251 }
3252 
3253 /*
3254  * btrfs_add_delayed_iput - perform a delayed iput on @inode
3255  *
3256  * @inode: The inode we want to perform iput on
3257  *
3258  * This function uses the generic vfs_inode::i_count to track whether we should
3259  * just decrement it (in case it's > 1) or if this is the last iput then link
3260  * the inode to the delayed iput machinery. Delayed iputs are processed at
3261  * transaction commit time/superblock commit/cleaner kthread.
3262  */
3263 void btrfs_add_delayed_iput(struct inode *inode)
3264 {
3265 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3266 	struct btrfs_inode *binode = BTRFS_I(inode);
3267 
3268 	if (atomic_add_unless(&inode->i_count, -1, 1))
3269 		return;
3270 
3271 	spin_lock(&fs_info->delayed_iput_lock);
3272 	ASSERT(list_empty(&binode->delayed_iput));
3273 	list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3274 	spin_unlock(&fs_info->delayed_iput_lock);
3275 }
3276 
3277 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3278 {
3279 
3280 	spin_lock(&fs_info->delayed_iput_lock);
3281 	while (!list_empty(&fs_info->delayed_iputs)) {
3282 		struct btrfs_inode *inode;
3283 
3284 		inode = list_first_entry(&fs_info->delayed_iputs,
3285 				struct btrfs_inode, delayed_iput);
3286 		list_del_init(&inode->delayed_iput);
3287 		spin_unlock(&fs_info->delayed_iput_lock);
3288 		iput(&inode->vfs_inode);
3289 		spin_lock(&fs_info->delayed_iput_lock);
3290 	}
3291 	spin_unlock(&fs_info->delayed_iput_lock);
3292 }
3293 
3294 /*
3295  * This creates an orphan entry for the given inode in case something goes wrong
3296  * in the middle of an unlink.
3297  */
3298 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3299 		     struct btrfs_inode *inode)
3300 {
3301 	int ret;
3302 
3303 	ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3304 	if (ret && ret != -EEXIST) {
3305 		btrfs_abort_transaction(trans, ret);
3306 		return ret;
3307 	}
3308 
3309 	return 0;
3310 }
3311 
3312 /*
3313  * We have done the delete so we can go ahead and remove the orphan item for
3314  * this particular inode.
3315  */
3316 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3317 			    struct btrfs_inode *inode)
3318 {
3319 	return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3320 }
3321 
3322 /*
3323  * this cleans up any orphans that may be left on the list from the last use
3324  * of this root.
3325  */
3326 int btrfs_orphan_cleanup(struct btrfs_root *root)
3327 {
3328 	struct btrfs_fs_info *fs_info = root->fs_info;
3329 	struct btrfs_path *path;
3330 	struct extent_buffer *leaf;
3331 	struct btrfs_key key, found_key;
3332 	struct btrfs_trans_handle *trans;
3333 	struct inode *inode;
3334 	u64 last_objectid = 0;
3335 	int ret = 0, nr_unlink = 0;
3336 
3337 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3338 		return 0;
3339 
3340 	path = btrfs_alloc_path();
3341 	if (!path) {
3342 		ret = -ENOMEM;
3343 		goto out;
3344 	}
3345 	path->reada = READA_BACK;
3346 
3347 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3348 	key.type = BTRFS_ORPHAN_ITEM_KEY;
3349 	key.offset = (u64)-1;
3350 
3351 	while (1) {
3352 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3353 		if (ret < 0)
3354 			goto out;
3355 
3356 		/*
3357 		 * if ret == 0 means we found what we were searching for, which
3358 		 * is weird, but possible, so only screw with path if we didn't
3359 		 * find the key and see if we have stuff that matches
3360 		 */
3361 		if (ret > 0) {
3362 			ret = 0;
3363 			if (path->slots[0] == 0)
3364 				break;
3365 			path->slots[0]--;
3366 		}
3367 
3368 		/* pull out the item */
3369 		leaf = path->nodes[0];
3370 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3371 
3372 		/* make sure the item matches what we want */
3373 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3374 			break;
3375 		if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3376 			break;
3377 
3378 		/* release the path since we're done with it */
3379 		btrfs_release_path(path);
3380 
3381 		/*
3382 		 * this is where we are basically btrfs_lookup, without the
3383 		 * crossing root thing.  we store the inode number in the
3384 		 * offset of the orphan item.
3385 		 */
3386 
3387 		if (found_key.offset == last_objectid) {
3388 			btrfs_err(fs_info,
3389 				  "Error removing orphan entry, stopping orphan cleanup");
3390 			ret = -EINVAL;
3391 			goto out;
3392 		}
3393 
3394 		last_objectid = found_key.offset;
3395 
3396 		found_key.objectid = found_key.offset;
3397 		found_key.type = BTRFS_INODE_ITEM_KEY;
3398 		found_key.offset = 0;
3399 		inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3400 		ret = PTR_ERR_OR_ZERO(inode);
3401 		if (ret && ret != -ENOENT)
3402 			goto out;
3403 
3404 		if (ret == -ENOENT && root == fs_info->tree_root) {
3405 			struct btrfs_root *dead_root;
3406 			struct btrfs_fs_info *fs_info = root->fs_info;
3407 			int is_dead_root = 0;
3408 
3409 			/*
3410 			 * this is an orphan in the tree root. Currently these
3411 			 * could come from 2 sources:
3412 			 *  a) a snapshot deletion in progress
3413 			 *  b) a free space cache inode
3414 			 * We need to distinguish those two, as the snapshot
3415 			 * orphan must not get deleted.
3416 			 * find_dead_roots already ran before us, so if this
3417 			 * is a snapshot deletion, we should find the root
3418 			 * in the dead_roots list
3419 			 */
3420 			spin_lock(&fs_info->trans_lock);
3421 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3422 					    root_list) {
3423 				if (dead_root->root_key.objectid ==
3424 				    found_key.objectid) {
3425 					is_dead_root = 1;
3426 					break;
3427 				}
3428 			}
3429 			spin_unlock(&fs_info->trans_lock);
3430 			if (is_dead_root) {
3431 				/* prevent this orphan from being found again */
3432 				key.offset = found_key.objectid - 1;
3433 				continue;
3434 			}
3435 
3436 		}
3437 
3438 		/*
3439 		 * If we have an inode with links, there are a couple of
3440 		 * possibilities. Old kernels (before v3.12) used to create an
3441 		 * orphan item for truncate indicating that there were possibly
3442 		 * extent items past i_size that needed to be deleted. In v3.12,
3443 		 * truncate was changed to update i_size in sync with the extent
3444 		 * items, but the (useless) orphan item was still created. Since
3445 		 * v4.18, we don't create the orphan item for truncate at all.
3446 		 *
3447 		 * So, this item could mean that we need to do a truncate, but
3448 		 * only if this filesystem was last used on a pre-v3.12 kernel
3449 		 * and was not cleanly unmounted. The odds of that are quite
3450 		 * slim, and it's a pain to do the truncate now, so just delete
3451 		 * the orphan item.
3452 		 *
3453 		 * It's also possible that this orphan item was supposed to be
3454 		 * deleted but wasn't. The inode number may have been reused,
3455 		 * but either way, we can delete the orphan item.
3456 		 */
3457 		if (ret == -ENOENT || inode->i_nlink) {
3458 			if (!ret)
3459 				iput(inode);
3460 			trans = btrfs_start_transaction(root, 1);
3461 			if (IS_ERR(trans)) {
3462 				ret = PTR_ERR(trans);
3463 				goto out;
3464 			}
3465 			btrfs_debug(fs_info, "auto deleting %Lu",
3466 				    found_key.objectid);
3467 			ret = btrfs_del_orphan_item(trans, root,
3468 						    found_key.objectid);
3469 			btrfs_end_transaction(trans);
3470 			if (ret)
3471 				goto out;
3472 			continue;
3473 		}
3474 
3475 		nr_unlink++;
3476 
3477 		/* this will do delete_inode and everything for us */
3478 		iput(inode);
3479 	}
3480 	/* release the path since we're done with it */
3481 	btrfs_release_path(path);
3482 
3483 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3484 
3485 	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3486 		trans = btrfs_join_transaction(root);
3487 		if (!IS_ERR(trans))
3488 			btrfs_end_transaction(trans);
3489 	}
3490 
3491 	if (nr_unlink)
3492 		btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3493 
3494 out:
3495 	if (ret)
3496 		btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3497 	btrfs_free_path(path);
3498 	return ret;
3499 }
3500 
3501 /*
3502  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3503  * don't find any xattrs, we know there can't be any acls.
3504  *
3505  * slot is the slot the inode is in, objectid is the objectid of the inode
3506  */
3507 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3508 					  int slot, u64 objectid,
3509 					  int *first_xattr_slot)
3510 {
3511 	u32 nritems = btrfs_header_nritems(leaf);
3512 	struct btrfs_key found_key;
3513 	static u64 xattr_access = 0;
3514 	static u64 xattr_default = 0;
3515 	int scanned = 0;
3516 
3517 	if (!xattr_access) {
3518 		xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3519 					strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3520 		xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3521 					strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3522 	}
3523 
3524 	slot++;
3525 	*first_xattr_slot = -1;
3526 	while (slot < nritems) {
3527 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3528 
3529 		/* we found a different objectid, there must not be acls */
3530 		if (found_key.objectid != objectid)
3531 			return 0;
3532 
3533 		/* we found an xattr, assume we've got an acl */
3534 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3535 			if (*first_xattr_slot == -1)
3536 				*first_xattr_slot = slot;
3537 			if (found_key.offset == xattr_access ||
3538 			    found_key.offset == xattr_default)
3539 				return 1;
3540 		}
3541 
3542 		/*
3543 		 * we found a key greater than an xattr key, there can't
3544 		 * be any acls later on
3545 		 */
3546 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3547 			return 0;
3548 
3549 		slot++;
3550 		scanned++;
3551 
3552 		/*
3553 		 * it goes inode, inode backrefs, xattrs, extents,
3554 		 * so if there are a ton of hard links to an inode there can
3555 		 * be a lot of backrefs.  Don't waste time searching too hard,
3556 		 * this is just an optimization
3557 		 */
3558 		if (scanned >= 8)
3559 			break;
3560 	}
3561 	/* we hit the end of the leaf before we found an xattr or
3562 	 * something larger than an xattr.  We have to assume the inode
3563 	 * has acls
3564 	 */
3565 	if (*first_xattr_slot == -1)
3566 		*first_xattr_slot = slot;
3567 	return 1;
3568 }
3569 
3570 /*
3571  * read an inode from the btree into the in-memory inode
3572  */
3573 static int btrfs_read_locked_inode(struct inode *inode)
3574 {
3575 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3576 	struct btrfs_path *path;
3577 	struct extent_buffer *leaf;
3578 	struct btrfs_inode_item *inode_item;
3579 	struct btrfs_root *root = BTRFS_I(inode)->root;
3580 	struct btrfs_key location;
3581 	unsigned long ptr;
3582 	int maybe_acls;
3583 	u32 rdev;
3584 	int ret;
3585 	bool filled = false;
3586 	int first_xattr_slot;
3587 
3588 	ret = btrfs_fill_inode(inode, &rdev);
3589 	if (!ret)
3590 		filled = true;
3591 
3592 	path = btrfs_alloc_path();
3593 	if (!path)
3594 		return -ENOMEM;
3595 
3596 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3597 
3598 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3599 	if (ret) {
3600 		btrfs_free_path(path);
3601 		return ret;
3602 	}
3603 
3604 	leaf = path->nodes[0];
3605 
3606 	if (filled)
3607 		goto cache_index;
3608 
3609 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610 				    struct btrfs_inode_item);
3611 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615 	btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3616 
3617 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3619 
3620 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3622 
3623 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3625 
3626 	BTRFS_I(inode)->i_otime.tv_sec =
3627 		btrfs_timespec_sec(leaf, &inode_item->otime);
3628 	BTRFS_I(inode)->i_otime.tv_nsec =
3629 		btrfs_timespec_nsec(leaf, &inode_item->otime);
3630 
3631 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3634 
3635 	inode_set_iversion_queried(inode,
3636 				   btrfs_inode_sequence(leaf, inode_item));
3637 	inode->i_generation = BTRFS_I(inode)->generation;
3638 	inode->i_rdev = 0;
3639 	rdev = btrfs_inode_rdev(leaf, inode_item);
3640 
3641 	BTRFS_I(inode)->index_cnt = (u64)-1;
3642 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3643 
3644 cache_index:
3645 	/*
3646 	 * If we were modified in the current generation and evicted from memory
3647 	 * and then re-read we need to do a full sync since we don't have any
3648 	 * idea about which extents were modified before we were evicted from
3649 	 * cache.
3650 	 *
3651 	 * This is required for both inode re-read from disk and delayed inode
3652 	 * in delayed_nodes_tree.
3653 	 */
3654 	if (BTRFS_I(inode)->last_trans == fs_info->generation)
3655 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3656 			&BTRFS_I(inode)->runtime_flags);
3657 
3658 	/*
3659 	 * We don't persist the id of the transaction where an unlink operation
3660 	 * against the inode was last made. So here we assume the inode might
3661 	 * have been evicted, and therefore the exact value of last_unlink_trans
3662 	 * lost, and set it to last_trans to avoid metadata inconsistencies
3663 	 * between the inode and its parent if the inode is fsync'ed and the log
3664 	 * replayed. For example, in the scenario:
3665 	 *
3666 	 * touch mydir/foo
3667 	 * ln mydir/foo mydir/bar
3668 	 * sync
3669 	 * unlink mydir/bar
3670 	 * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3671 	 * xfs_io -c fsync mydir/foo
3672 	 * <power failure>
3673 	 * mount fs, triggers fsync log replay
3674 	 *
3675 	 * We must make sure that when we fsync our inode foo we also log its
3676 	 * parent inode, otherwise after log replay the parent still has the
3677 	 * dentry with the "bar" name but our inode foo has a link count of 1
3678 	 * and doesn't have an inode ref with the name "bar" anymore.
3679 	 *
3680 	 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3681 	 * but it guarantees correctness at the expense of occasional full
3682 	 * transaction commits on fsync if our inode is a directory, or if our
3683 	 * inode is not a directory, logging its parent unnecessarily.
3684 	 */
3685 	BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3686 
3687 	path->slots[0]++;
3688 	if (inode->i_nlink != 1 ||
3689 	    path->slots[0] >= btrfs_header_nritems(leaf))
3690 		goto cache_acl;
3691 
3692 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3693 	if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3694 		goto cache_acl;
3695 
3696 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3697 	if (location.type == BTRFS_INODE_REF_KEY) {
3698 		struct btrfs_inode_ref *ref;
3699 
3700 		ref = (struct btrfs_inode_ref *)ptr;
3701 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3702 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3703 		struct btrfs_inode_extref *extref;
3704 
3705 		extref = (struct btrfs_inode_extref *)ptr;
3706 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3707 								     extref);
3708 	}
3709 cache_acl:
3710 	/*
3711 	 * try to precache a NULL acl entry for files that don't have
3712 	 * any xattrs or acls
3713 	 */
3714 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3715 			btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3716 	if (first_xattr_slot != -1) {
3717 		path->slots[0] = first_xattr_slot;
3718 		ret = btrfs_load_inode_props(inode, path);
3719 		if (ret)
3720 			btrfs_err(fs_info,
3721 				  "error loading props for ino %llu (root %llu): %d",
3722 				  btrfs_ino(BTRFS_I(inode)),
3723 				  root->root_key.objectid, ret);
3724 	}
3725 	btrfs_free_path(path);
3726 
3727 	if (!maybe_acls)
3728 		cache_no_acl(inode);
3729 
3730 	switch (inode->i_mode & S_IFMT) {
3731 	case S_IFREG:
3732 		inode->i_mapping->a_ops = &btrfs_aops;
3733 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3734 		inode->i_fop = &btrfs_file_operations;
3735 		inode->i_op = &btrfs_file_inode_operations;
3736 		break;
3737 	case S_IFDIR:
3738 		inode->i_fop = &btrfs_dir_file_operations;
3739 		inode->i_op = &btrfs_dir_inode_operations;
3740 		break;
3741 	case S_IFLNK:
3742 		inode->i_op = &btrfs_symlink_inode_operations;
3743 		inode_nohighmem(inode);
3744 		inode->i_mapping->a_ops = &btrfs_aops;
3745 		break;
3746 	default:
3747 		inode->i_op = &btrfs_special_inode_operations;
3748 		init_special_inode(inode, inode->i_mode, rdev);
3749 		break;
3750 	}
3751 
3752 	btrfs_sync_inode_flags_to_i_flags(inode);
3753 	return 0;
3754 }
3755 
3756 /*
3757  * given a leaf and an inode, copy the inode fields into the leaf
3758  */
3759 static void fill_inode_item(struct btrfs_trans_handle *trans,
3760 			    struct extent_buffer *leaf,
3761 			    struct btrfs_inode_item *item,
3762 			    struct inode *inode)
3763 {
3764 	struct btrfs_map_token token;
3765 
3766 	btrfs_init_map_token(&token);
3767 
3768 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3769 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3770 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3771 				   &token);
3772 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3773 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3774 
3775 	btrfs_set_token_timespec_sec(leaf, &item->atime,
3776 				     inode->i_atime.tv_sec, &token);
3777 	btrfs_set_token_timespec_nsec(leaf, &item->atime,
3778 				      inode->i_atime.tv_nsec, &token);
3779 
3780 	btrfs_set_token_timespec_sec(leaf, &item->mtime,
3781 				     inode->i_mtime.tv_sec, &token);
3782 	btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3783 				      inode->i_mtime.tv_nsec, &token);
3784 
3785 	btrfs_set_token_timespec_sec(leaf, &item->ctime,
3786 				     inode->i_ctime.tv_sec, &token);
3787 	btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3788 				      inode->i_ctime.tv_nsec, &token);
3789 
3790 	btrfs_set_token_timespec_sec(leaf, &item->otime,
3791 				     BTRFS_I(inode)->i_otime.tv_sec, &token);
3792 	btrfs_set_token_timespec_nsec(leaf, &item->otime,
3793 				      BTRFS_I(inode)->i_otime.tv_nsec, &token);
3794 
3795 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3796 				     &token);
3797 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3798 					 &token);
3799 	btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3800 				       &token);
3801 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3802 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3803 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3804 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3805 }
3806 
3807 /*
3808  * copy everything in the in-memory inode into the btree.
3809  */
3810 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3811 				struct btrfs_root *root, struct inode *inode)
3812 {
3813 	struct btrfs_inode_item *inode_item;
3814 	struct btrfs_path *path;
3815 	struct extent_buffer *leaf;
3816 	int ret;
3817 
3818 	path = btrfs_alloc_path();
3819 	if (!path)
3820 		return -ENOMEM;
3821 
3822 	path->leave_spinning = 1;
3823 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3824 				 1);
3825 	if (ret) {
3826 		if (ret > 0)
3827 			ret = -ENOENT;
3828 		goto failed;
3829 	}
3830 
3831 	leaf = path->nodes[0];
3832 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3833 				    struct btrfs_inode_item);
3834 
3835 	fill_inode_item(trans, leaf, inode_item, inode);
3836 	btrfs_mark_buffer_dirty(leaf);
3837 	btrfs_set_inode_last_trans(trans, inode);
3838 	ret = 0;
3839 failed:
3840 	btrfs_free_path(path);
3841 	return ret;
3842 }
3843 
3844 /*
3845  * copy everything in the in-memory inode into the btree.
3846  */
3847 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3848 				struct btrfs_root *root, struct inode *inode)
3849 {
3850 	struct btrfs_fs_info *fs_info = root->fs_info;
3851 	int ret;
3852 
3853 	/*
3854 	 * If the inode is a free space inode, we can deadlock during commit
3855 	 * if we put it into the delayed code.
3856 	 *
3857 	 * The data relocation inode should also be directly updated
3858 	 * without delay
3859 	 */
3860 	if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3861 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3862 	    && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3863 		btrfs_update_root_times(trans, root);
3864 
3865 		ret = btrfs_delayed_update_inode(trans, root, inode);
3866 		if (!ret)
3867 			btrfs_set_inode_last_trans(trans, inode);
3868 		return ret;
3869 	}
3870 
3871 	return btrfs_update_inode_item(trans, root, inode);
3872 }
3873 
3874 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3875 					 struct btrfs_root *root,
3876 					 struct inode *inode)
3877 {
3878 	int ret;
3879 
3880 	ret = btrfs_update_inode(trans, root, inode);
3881 	if (ret == -ENOSPC)
3882 		return btrfs_update_inode_item(trans, root, inode);
3883 	return ret;
3884 }
3885 
3886 /*
3887  * unlink helper that gets used here in inode.c and in the tree logging
3888  * recovery code.  It remove a link in a directory with a given name, and
3889  * also drops the back refs in the inode to the directory
3890  */
3891 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3892 				struct btrfs_root *root,
3893 				struct btrfs_inode *dir,
3894 				struct btrfs_inode *inode,
3895 				const char *name, int name_len)
3896 {
3897 	struct btrfs_fs_info *fs_info = root->fs_info;
3898 	struct btrfs_path *path;
3899 	int ret = 0;
3900 	struct extent_buffer *leaf;
3901 	struct btrfs_dir_item *di;
3902 	struct btrfs_key key;
3903 	u64 index;
3904 	u64 ino = btrfs_ino(inode);
3905 	u64 dir_ino = btrfs_ino(dir);
3906 
3907 	path = btrfs_alloc_path();
3908 	if (!path) {
3909 		ret = -ENOMEM;
3910 		goto out;
3911 	}
3912 
3913 	path->leave_spinning = 1;
3914 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3915 				    name, name_len, -1);
3916 	if (IS_ERR_OR_NULL(di)) {
3917 		ret = di ? PTR_ERR(di) : -ENOENT;
3918 		goto err;
3919 	}
3920 	leaf = path->nodes[0];
3921 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3922 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3923 	if (ret)
3924 		goto err;
3925 	btrfs_release_path(path);
3926 
3927 	/*
3928 	 * If we don't have dir index, we have to get it by looking up
3929 	 * the inode ref, since we get the inode ref, remove it directly,
3930 	 * it is unnecessary to do delayed deletion.
3931 	 *
3932 	 * But if we have dir index, needn't search inode ref to get it.
3933 	 * Since the inode ref is close to the inode item, it is better
3934 	 * that we delay to delete it, and just do this deletion when
3935 	 * we update the inode item.
3936 	 */
3937 	if (inode->dir_index) {
3938 		ret = btrfs_delayed_delete_inode_ref(inode);
3939 		if (!ret) {
3940 			index = inode->dir_index;
3941 			goto skip_backref;
3942 		}
3943 	}
3944 
3945 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3946 				  dir_ino, &index);
3947 	if (ret) {
3948 		btrfs_info(fs_info,
3949 			"failed to delete reference to %.*s, inode %llu parent %llu",
3950 			name_len, name, ino, dir_ino);
3951 		btrfs_abort_transaction(trans, ret);
3952 		goto err;
3953 	}
3954 skip_backref:
3955 	ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3956 	if (ret) {
3957 		btrfs_abort_transaction(trans, ret);
3958 		goto err;
3959 	}
3960 
3961 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3962 			dir_ino);
3963 	if (ret != 0 && ret != -ENOENT) {
3964 		btrfs_abort_transaction(trans, ret);
3965 		goto err;
3966 	}
3967 
3968 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3969 			index);
3970 	if (ret == -ENOENT)
3971 		ret = 0;
3972 	else if (ret)
3973 		btrfs_abort_transaction(trans, ret);
3974 err:
3975 	btrfs_free_path(path);
3976 	if (ret)
3977 		goto out;
3978 
3979 	btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3980 	inode_inc_iversion(&inode->vfs_inode);
3981 	inode_inc_iversion(&dir->vfs_inode);
3982 	inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3983 		dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3984 	ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3985 out:
3986 	return ret;
3987 }
3988 
3989 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3990 		       struct btrfs_root *root,
3991 		       struct btrfs_inode *dir, struct btrfs_inode *inode,
3992 		       const char *name, int name_len)
3993 {
3994 	int ret;
3995 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3996 	if (!ret) {
3997 		drop_nlink(&inode->vfs_inode);
3998 		ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3999 	}
4000 	return ret;
4001 }
4002 
4003 /*
4004  * helper to start transaction for unlink and rmdir.
4005  *
4006  * unlink and rmdir are special in btrfs, they do not always free space, so
4007  * if we cannot make our reservations the normal way try and see if there is
4008  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4009  * allow the unlink to occur.
4010  */
4011 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4012 {
4013 	struct btrfs_root *root = BTRFS_I(dir)->root;
4014 
4015 	/*
4016 	 * 1 for the possible orphan item
4017 	 * 1 for the dir item
4018 	 * 1 for the dir index
4019 	 * 1 for the inode ref
4020 	 * 1 for the inode
4021 	 */
4022 	return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4023 }
4024 
4025 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4026 {
4027 	struct btrfs_root *root = BTRFS_I(dir)->root;
4028 	struct btrfs_trans_handle *trans;
4029 	struct inode *inode = d_inode(dentry);
4030 	int ret;
4031 
4032 	trans = __unlink_start_trans(dir);
4033 	if (IS_ERR(trans))
4034 		return PTR_ERR(trans);
4035 
4036 	btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4037 			0);
4038 
4039 	ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4040 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4041 			dentry->d_name.len);
4042 	if (ret)
4043 		goto out;
4044 
4045 	if (inode->i_nlink == 0) {
4046 		ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4047 		if (ret)
4048 			goto out;
4049 	}
4050 
4051 out:
4052 	btrfs_end_transaction(trans);
4053 	btrfs_btree_balance_dirty(root->fs_info);
4054 	return ret;
4055 }
4056 
4057 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4058 			       struct inode *dir, u64 objectid,
4059 			       const char *name, int name_len)
4060 {
4061 	struct btrfs_root *root = BTRFS_I(dir)->root;
4062 	struct btrfs_path *path;
4063 	struct extent_buffer *leaf;
4064 	struct btrfs_dir_item *di;
4065 	struct btrfs_key key;
4066 	u64 index;
4067 	int ret;
4068 	u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4069 
4070 	path = btrfs_alloc_path();
4071 	if (!path)
4072 		return -ENOMEM;
4073 
4074 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4075 				   name, name_len, -1);
4076 	if (IS_ERR_OR_NULL(di)) {
4077 		ret = di ? PTR_ERR(di) : -ENOENT;
4078 		goto out;
4079 	}
4080 
4081 	leaf = path->nodes[0];
4082 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
4083 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4084 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4085 	if (ret) {
4086 		btrfs_abort_transaction(trans, ret);
4087 		goto out;
4088 	}
4089 	btrfs_release_path(path);
4090 
4091 	ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4092 				 dir_ino, &index, name, name_len);
4093 	if (ret < 0) {
4094 		if (ret != -ENOENT) {
4095 			btrfs_abort_transaction(trans, ret);
4096 			goto out;
4097 		}
4098 		di = btrfs_search_dir_index_item(root, path, dir_ino,
4099 						 name, name_len);
4100 		if (IS_ERR_OR_NULL(di)) {
4101 			if (!di)
4102 				ret = -ENOENT;
4103 			else
4104 				ret = PTR_ERR(di);
4105 			btrfs_abort_transaction(trans, ret);
4106 			goto out;
4107 		}
4108 
4109 		leaf = path->nodes[0];
4110 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4111 		index = key.offset;
4112 	}
4113 	btrfs_release_path(path);
4114 
4115 	ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4116 	if (ret) {
4117 		btrfs_abort_transaction(trans, ret);
4118 		goto out;
4119 	}
4120 
4121 	btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4122 	inode_inc_iversion(dir);
4123 	dir->i_mtime = dir->i_ctime = current_time(dir);
4124 	ret = btrfs_update_inode_fallback(trans, root, dir);
4125 	if (ret)
4126 		btrfs_abort_transaction(trans, ret);
4127 out:
4128 	btrfs_free_path(path);
4129 	return ret;
4130 }
4131 
4132 /*
4133  * Helper to check if the subvolume references other subvolumes or if it's
4134  * default.
4135  */
4136 static noinline int may_destroy_subvol(struct btrfs_root *root)
4137 {
4138 	struct btrfs_fs_info *fs_info = root->fs_info;
4139 	struct btrfs_path *path;
4140 	struct btrfs_dir_item *di;
4141 	struct btrfs_key key;
4142 	u64 dir_id;
4143 	int ret;
4144 
4145 	path = btrfs_alloc_path();
4146 	if (!path)
4147 		return -ENOMEM;
4148 
4149 	/* Make sure this root isn't set as the default subvol */
4150 	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4151 	di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4152 				   dir_id, "default", 7, 0);
4153 	if (di && !IS_ERR(di)) {
4154 		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4155 		if (key.objectid == root->root_key.objectid) {
4156 			ret = -EPERM;
4157 			btrfs_err(fs_info,
4158 				  "deleting default subvolume %llu is not allowed",
4159 				  key.objectid);
4160 			goto out;
4161 		}
4162 		btrfs_release_path(path);
4163 	}
4164 
4165 	key.objectid = root->root_key.objectid;
4166 	key.type = BTRFS_ROOT_REF_KEY;
4167 	key.offset = (u64)-1;
4168 
4169 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4170 	if (ret < 0)
4171 		goto out;
4172 	BUG_ON(ret == 0);
4173 
4174 	ret = 0;
4175 	if (path->slots[0] > 0) {
4176 		path->slots[0]--;
4177 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4178 		if (key.objectid == root->root_key.objectid &&
4179 		    key.type == BTRFS_ROOT_REF_KEY)
4180 			ret = -ENOTEMPTY;
4181 	}
4182 out:
4183 	btrfs_free_path(path);
4184 	return ret;
4185 }
4186 
4187 /* Delete all dentries for inodes belonging to the root */
4188 static void btrfs_prune_dentries(struct btrfs_root *root)
4189 {
4190 	struct btrfs_fs_info *fs_info = root->fs_info;
4191 	struct rb_node *node;
4192 	struct rb_node *prev;
4193 	struct btrfs_inode *entry;
4194 	struct inode *inode;
4195 	u64 objectid = 0;
4196 
4197 	if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4198 		WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4199 
4200 	spin_lock(&root->inode_lock);
4201 again:
4202 	node = root->inode_tree.rb_node;
4203 	prev = NULL;
4204 	while (node) {
4205 		prev = node;
4206 		entry = rb_entry(node, struct btrfs_inode, rb_node);
4207 
4208 		if (objectid < btrfs_ino(entry))
4209 			node = node->rb_left;
4210 		else if (objectid > btrfs_ino(entry))
4211 			node = node->rb_right;
4212 		else
4213 			break;
4214 	}
4215 	if (!node) {
4216 		while (prev) {
4217 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
4218 			if (objectid <= btrfs_ino(entry)) {
4219 				node = prev;
4220 				break;
4221 			}
4222 			prev = rb_next(prev);
4223 		}
4224 	}
4225 	while (node) {
4226 		entry = rb_entry(node, struct btrfs_inode, rb_node);
4227 		objectid = btrfs_ino(entry) + 1;
4228 		inode = igrab(&entry->vfs_inode);
4229 		if (inode) {
4230 			spin_unlock(&root->inode_lock);
4231 			if (atomic_read(&inode->i_count) > 1)
4232 				d_prune_aliases(inode);
4233 			/*
4234 			 * btrfs_drop_inode will have it removed from the inode
4235 			 * cache when its usage count hits zero.
4236 			 */
4237 			iput(inode);
4238 			cond_resched();
4239 			spin_lock(&root->inode_lock);
4240 			goto again;
4241 		}
4242 
4243 		if (cond_resched_lock(&root->inode_lock))
4244 			goto again;
4245 
4246 		node = rb_next(node);
4247 	}
4248 	spin_unlock(&root->inode_lock);
4249 }
4250 
4251 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4252 {
4253 	struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4254 	struct btrfs_root *root = BTRFS_I(dir)->root;
4255 	struct inode *inode = d_inode(dentry);
4256 	struct btrfs_root *dest = BTRFS_I(inode)->root;
4257 	struct btrfs_trans_handle *trans;
4258 	struct btrfs_block_rsv block_rsv;
4259 	u64 root_flags;
4260 	int ret;
4261 	int err;
4262 
4263 	/*
4264 	 * Don't allow to delete a subvolume with send in progress. This is
4265 	 * inside the inode lock so the error handling that has to drop the bit
4266 	 * again is not run concurrently.
4267 	 */
4268 	spin_lock(&dest->root_item_lock);
4269 	if (dest->send_in_progress) {
4270 		spin_unlock(&dest->root_item_lock);
4271 		btrfs_warn(fs_info,
4272 			   "attempt to delete subvolume %llu during send",
4273 			   dest->root_key.objectid);
4274 		return -EPERM;
4275 	}
4276 	root_flags = btrfs_root_flags(&dest->root_item);
4277 	btrfs_set_root_flags(&dest->root_item,
4278 			     root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4279 	spin_unlock(&dest->root_item_lock);
4280 
4281 	down_write(&fs_info->subvol_sem);
4282 
4283 	err = may_destroy_subvol(dest);
4284 	if (err)
4285 		goto out_up_write;
4286 
4287 	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4288 	/*
4289 	 * One for dir inode,
4290 	 * two for dir entries,
4291 	 * two for root ref/backref.
4292 	 */
4293 	err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4294 	if (err)
4295 		goto out_up_write;
4296 
4297 	trans = btrfs_start_transaction(root, 0);
4298 	if (IS_ERR(trans)) {
4299 		err = PTR_ERR(trans);
4300 		goto out_release;
4301 	}
4302 	trans->block_rsv = &block_rsv;
4303 	trans->bytes_reserved = block_rsv.size;
4304 
4305 	btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4306 
4307 	ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4308 				  dentry->d_name.name, dentry->d_name.len);
4309 	if (ret) {
4310 		err = ret;
4311 		btrfs_abort_transaction(trans, ret);
4312 		goto out_end_trans;
4313 	}
4314 
4315 	btrfs_record_root_in_trans(trans, dest);
4316 
4317 	memset(&dest->root_item.drop_progress, 0,
4318 		sizeof(dest->root_item.drop_progress));
4319 	dest->root_item.drop_level = 0;
4320 	btrfs_set_root_refs(&dest->root_item, 0);
4321 
4322 	if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4323 		ret = btrfs_insert_orphan_item(trans,
4324 					fs_info->tree_root,
4325 					dest->root_key.objectid);
4326 		if (ret) {
4327 			btrfs_abort_transaction(trans, ret);
4328 			err = ret;
4329 			goto out_end_trans;
4330 		}
4331 	}
4332 
4333 	ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4334 				  BTRFS_UUID_KEY_SUBVOL,
4335 				  dest->root_key.objectid);
4336 	if (ret && ret != -ENOENT) {
4337 		btrfs_abort_transaction(trans, ret);
4338 		err = ret;
4339 		goto out_end_trans;
4340 	}
4341 	if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4342 		ret = btrfs_uuid_tree_remove(trans,
4343 					  dest->root_item.received_uuid,
4344 					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4345 					  dest->root_key.objectid);
4346 		if (ret && ret != -ENOENT) {
4347 			btrfs_abort_transaction(trans, ret);
4348 			err = ret;
4349 			goto out_end_trans;
4350 		}
4351 	}
4352 
4353 out_end_trans:
4354 	trans->block_rsv = NULL;
4355 	trans->bytes_reserved = 0;
4356 	ret = btrfs_end_transaction(trans);
4357 	if (ret && !err)
4358 		err = ret;
4359 	inode->i_flags |= S_DEAD;
4360 out_release:
4361 	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4362 out_up_write:
4363 	up_write(&fs_info->subvol_sem);
4364 	if (err) {
4365 		spin_lock(&dest->root_item_lock);
4366 		root_flags = btrfs_root_flags(&dest->root_item);
4367 		btrfs_set_root_flags(&dest->root_item,
4368 				root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4369 		spin_unlock(&dest->root_item_lock);
4370 	} else {
4371 		d_invalidate(dentry);
4372 		btrfs_prune_dentries(dest);
4373 		ASSERT(dest->send_in_progress == 0);
4374 
4375 		/* the last ref */
4376 		if (dest->ino_cache_inode) {
4377 			iput(dest->ino_cache_inode);
4378 			dest->ino_cache_inode = NULL;
4379 		}
4380 	}
4381 
4382 	return err;
4383 }
4384 
4385 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4386 {
4387 	struct inode *inode = d_inode(dentry);
4388 	int err = 0;
4389 	struct btrfs_root *root = BTRFS_I(dir)->root;
4390 	struct btrfs_trans_handle *trans;
4391 	u64 last_unlink_trans;
4392 
4393 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4394 		return -ENOTEMPTY;
4395 	if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4396 		return btrfs_delete_subvolume(dir, dentry);
4397 
4398 	trans = __unlink_start_trans(dir);
4399 	if (IS_ERR(trans))
4400 		return PTR_ERR(trans);
4401 
4402 	if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4403 		err = btrfs_unlink_subvol(trans, dir,
4404 					  BTRFS_I(inode)->location.objectid,
4405 					  dentry->d_name.name,
4406 					  dentry->d_name.len);
4407 		goto out;
4408 	}
4409 
4410 	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4411 	if (err)
4412 		goto out;
4413 
4414 	last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4415 
4416 	/* now the directory is empty */
4417 	err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4418 			BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4419 			dentry->d_name.len);
4420 	if (!err) {
4421 		btrfs_i_size_write(BTRFS_I(inode), 0);
4422 		/*
4423 		 * Propagate the last_unlink_trans value of the deleted dir to
4424 		 * its parent directory. This is to prevent an unrecoverable
4425 		 * log tree in the case we do something like this:
4426 		 * 1) create dir foo
4427 		 * 2) create snapshot under dir foo
4428 		 * 3) delete the snapshot
4429 		 * 4) rmdir foo
4430 		 * 5) mkdir foo
4431 		 * 6) fsync foo or some file inside foo
4432 		 */
4433 		if (last_unlink_trans >= trans->transid)
4434 			BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4435 	}
4436 out:
4437 	btrfs_end_transaction(trans);
4438 	btrfs_btree_balance_dirty(root->fs_info);
4439 
4440 	return err;
4441 }
4442 
4443 static int truncate_space_check(struct btrfs_trans_handle *trans,
4444 				struct btrfs_root *root,
4445 				u64 bytes_deleted)
4446 {
4447 	struct btrfs_fs_info *fs_info = root->fs_info;
4448 	int ret;
4449 
4450 	/*
4451 	 * This is only used to apply pressure to the enospc system, we don't
4452 	 * intend to use this reservation at all.
4453 	 */
4454 	bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4455 	bytes_deleted *= fs_info->nodesize;
4456 	ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4457 				  bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4458 	if (!ret) {
4459 		trace_btrfs_space_reservation(fs_info, "transaction",
4460 					      trans->transid,
4461 					      bytes_deleted, 1);
4462 		trans->bytes_reserved += bytes_deleted;
4463 	}
4464 	return ret;
4465 
4466 }
4467 
4468 /*
4469  * Return this if we need to call truncate_block for the last bit of the
4470  * truncate.
4471  */
4472 #define NEED_TRUNCATE_BLOCK 1
4473 
4474 /*
4475  * this can truncate away extent items, csum items and directory items.
4476  * It starts at a high offset and removes keys until it can't find
4477  * any higher than new_size
4478  *
4479  * csum items that cross the new i_size are truncated to the new size
4480  * as well.
4481  *
4482  * min_type is the minimum key type to truncate down to.  If set to 0, this
4483  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4484  */
4485 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4486 			       struct btrfs_root *root,
4487 			       struct inode *inode,
4488 			       u64 new_size, u32 min_type)
4489 {
4490 	struct btrfs_fs_info *fs_info = root->fs_info;
4491 	struct btrfs_path *path;
4492 	struct extent_buffer *leaf;
4493 	struct btrfs_file_extent_item *fi;
4494 	struct btrfs_key key;
4495 	struct btrfs_key found_key;
4496 	u64 extent_start = 0;
4497 	u64 extent_num_bytes = 0;
4498 	u64 extent_offset = 0;
4499 	u64 item_end = 0;
4500 	u64 last_size = new_size;
4501 	u32 found_type = (u8)-1;
4502 	int found_extent;
4503 	int del_item;
4504 	int pending_del_nr = 0;
4505 	int pending_del_slot = 0;
4506 	int extent_type = -1;
4507 	int ret;
4508 	u64 ino = btrfs_ino(BTRFS_I(inode));
4509 	u64 bytes_deleted = 0;
4510 	bool be_nice = false;
4511 	bool should_throttle = false;
4512 	bool should_end = false;
4513 
4514 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4515 
4516 	/*
4517 	 * for non-free space inodes and ref cows, we want to back off from
4518 	 * time to time
4519 	 */
4520 	if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4521 	    test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4522 		be_nice = true;
4523 
4524 	path = btrfs_alloc_path();
4525 	if (!path)
4526 		return -ENOMEM;
4527 	path->reada = READA_BACK;
4528 
4529 	/*
4530 	 * We want to drop from the next block forward in case this new size is
4531 	 * not block aligned since we will be keeping the last block of the
4532 	 * extent just the way it is.
4533 	 */
4534 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4535 	    root == fs_info->tree_root)
4536 		btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4537 					fs_info->sectorsize),
4538 					(u64)-1, 0);
4539 
4540 	/*
4541 	 * This function is also used to drop the items in the log tree before
4542 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4543 	 * it is used to drop the loged items. So we shouldn't kill the delayed
4544 	 * items.
4545 	 */
4546 	if (min_type == 0 && root == BTRFS_I(inode)->root)
4547 		btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4548 
4549 	key.objectid = ino;
4550 	key.offset = (u64)-1;
4551 	key.type = (u8)-1;
4552 
4553 search_again:
4554 	/*
4555 	 * with a 16K leaf size and 128MB extents, you can actually queue
4556 	 * up a huge file in a single leaf.  Most of the time that
4557 	 * bytes_deleted is > 0, it will be huge by the time we get here
4558 	 */
4559 	if (be_nice && bytes_deleted > SZ_32M &&
4560 	    btrfs_should_end_transaction(trans)) {
4561 		ret = -EAGAIN;
4562 		goto out;
4563 	}
4564 
4565 	path->leave_spinning = 1;
4566 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4567 	if (ret < 0)
4568 		goto out;
4569 
4570 	if (ret > 0) {
4571 		ret = 0;
4572 		/* there are no items in the tree for us to truncate, we're
4573 		 * done
4574 		 */
4575 		if (path->slots[0] == 0)
4576 			goto out;
4577 		path->slots[0]--;
4578 	}
4579 
4580 	while (1) {
4581 		fi = NULL;
4582 		leaf = path->nodes[0];
4583 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4584 		found_type = found_key.type;
4585 
4586 		if (found_key.objectid != ino)
4587 			break;
4588 
4589 		if (found_type < min_type)
4590 			break;
4591 
4592 		item_end = found_key.offset;
4593 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4594 			fi = btrfs_item_ptr(leaf, path->slots[0],
4595 					    struct btrfs_file_extent_item);
4596 			extent_type = btrfs_file_extent_type(leaf, fi);
4597 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4598 				item_end +=
4599 				    btrfs_file_extent_num_bytes(leaf, fi);
4600 
4601 				trace_btrfs_truncate_show_fi_regular(
4602 					BTRFS_I(inode), leaf, fi,
4603 					found_key.offset);
4604 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4605 				item_end += btrfs_file_extent_ram_bytes(leaf,
4606 									fi);
4607 
4608 				trace_btrfs_truncate_show_fi_inline(
4609 					BTRFS_I(inode), leaf, fi, path->slots[0],
4610 					found_key.offset);
4611 			}
4612 			item_end--;
4613 		}
4614 		if (found_type > min_type) {
4615 			del_item = 1;
4616 		} else {
4617 			if (item_end < new_size)
4618 				break;
4619 			if (found_key.offset >= new_size)
4620 				del_item = 1;
4621 			else
4622 				del_item = 0;
4623 		}
4624 		found_extent = 0;
4625 		/* FIXME, shrink the extent if the ref count is only 1 */
4626 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4627 			goto delete;
4628 
4629 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4630 			u64 num_dec;
4631 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4632 			if (!del_item) {
4633 				u64 orig_num_bytes =
4634 					btrfs_file_extent_num_bytes(leaf, fi);
4635 				extent_num_bytes = ALIGN(new_size -
4636 						found_key.offset,
4637 						fs_info->sectorsize);
4638 				btrfs_set_file_extent_num_bytes(leaf, fi,
4639 							 extent_num_bytes);
4640 				num_dec = (orig_num_bytes -
4641 					   extent_num_bytes);
4642 				if (test_bit(BTRFS_ROOT_REF_COWS,
4643 					     &root->state) &&
4644 				    extent_start != 0)
4645 					inode_sub_bytes(inode, num_dec);
4646 				btrfs_mark_buffer_dirty(leaf);
4647 			} else {
4648 				extent_num_bytes =
4649 					btrfs_file_extent_disk_num_bytes(leaf,
4650 									 fi);
4651 				extent_offset = found_key.offset -
4652 					btrfs_file_extent_offset(leaf, fi);
4653 
4654 				/* FIXME blocksize != 4096 */
4655 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4656 				if (extent_start != 0) {
4657 					found_extent = 1;
4658 					if (test_bit(BTRFS_ROOT_REF_COWS,
4659 						     &root->state))
4660 						inode_sub_bytes(inode, num_dec);
4661 				}
4662 			}
4663 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4664 			/*
4665 			 * we can't truncate inline items that have had
4666 			 * special encodings
4667 			 */
4668 			if (!del_item &&
4669 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4670 			    btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4671 			    btrfs_file_extent_compression(leaf, fi) == 0) {
4672 				u32 size = (u32)(new_size - found_key.offset);
4673 
4674 				btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4675 				size = btrfs_file_extent_calc_inline_size(size);
4676 				btrfs_truncate_item(root->fs_info, path, size, 1);
4677 			} else if (!del_item) {
4678 				/*
4679 				 * We have to bail so the last_size is set to
4680 				 * just before this extent.
4681 				 */
4682 				ret = NEED_TRUNCATE_BLOCK;
4683 				break;
4684 			}
4685 
4686 			if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4687 				inode_sub_bytes(inode, item_end + 1 - new_size);
4688 		}
4689 delete:
4690 		if (del_item)
4691 			last_size = found_key.offset;
4692 		else
4693 			last_size = new_size;
4694 		if (del_item) {
4695 			if (!pending_del_nr) {
4696 				/* no pending yet, add ourselves */
4697 				pending_del_slot = path->slots[0];
4698 				pending_del_nr = 1;
4699 			} else if (pending_del_nr &&
4700 				   path->slots[0] + 1 == pending_del_slot) {
4701 				/* hop on the pending chunk */
4702 				pending_del_nr++;
4703 				pending_del_slot = path->slots[0];
4704 			} else {
4705 				BUG();
4706 			}
4707 		} else {
4708 			break;
4709 		}
4710 		should_throttle = false;
4711 
4712 		if (found_extent &&
4713 		    (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4714 		     root == fs_info->tree_root)) {
4715 			btrfs_set_path_blocking(path);
4716 			bytes_deleted += extent_num_bytes;
4717 			ret = btrfs_free_extent(trans, root, extent_start,
4718 						extent_num_bytes, 0,
4719 						btrfs_header_owner(leaf),
4720 						ino, extent_offset);
4721 			if (ret) {
4722 				btrfs_abort_transaction(trans, ret);
4723 				break;
4724 			}
4725 			if (btrfs_should_throttle_delayed_refs(trans))
4726 				btrfs_async_run_delayed_refs(fs_info,
4727 					trans->delayed_ref_updates * 2,
4728 					trans->transid, 0);
4729 			if (be_nice) {
4730 				if (truncate_space_check(trans, root,
4731 							 extent_num_bytes)) {
4732 					should_end = true;
4733 				}
4734 				if (btrfs_should_throttle_delayed_refs(trans))
4735 					should_throttle = true;
4736 			}
4737 		}
4738 
4739 		if (found_type == BTRFS_INODE_ITEM_KEY)
4740 			break;
4741 
4742 		if (path->slots[0] == 0 ||
4743 		    path->slots[0] != pending_del_slot ||
4744 		    should_throttle || should_end) {
4745 			if (pending_del_nr) {
4746 				ret = btrfs_del_items(trans, root, path,
4747 						pending_del_slot,
4748 						pending_del_nr);
4749 				if (ret) {
4750 					btrfs_abort_transaction(trans, ret);
4751 					break;
4752 				}
4753 				pending_del_nr = 0;
4754 			}
4755 			btrfs_release_path(path);
4756 			if (should_throttle) {
4757 				unsigned long updates = trans->delayed_ref_updates;
4758 				if (updates) {
4759 					trans->delayed_ref_updates = 0;
4760 					ret = btrfs_run_delayed_refs(trans,
4761 								   updates * 2);
4762 					if (ret)
4763 						break;
4764 				}
4765 			}
4766 			/*
4767 			 * if we failed to refill our space rsv, bail out
4768 			 * and let the transaction restart
4769 			 */
4770 			if (should_end) {
4771 				ret = -EAGAIN;
4772 				break;
4773 			}
4774 			goto search_again;
4775 		} else {
4776 			path->slots[0]--;
4777 		}
4778 	}
4779 out:
4780 	if (ret >= 0 && pending_del_nr) {
4781 		int err;
4782 
4783 		err = btrfs_del_items(trans, root, path, pending_del_slot,
4784 				      pending_del_nr);
4785 		if (err) {
4786 			btrfs_abort_transaction(trans, err);
4787 			ret = err;
4788 		}
4789 	}
4790 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4791 		ASSERT(last_size >= new_size);
4792 		if (!ret && last_size > new_size)
4793 			last_size = new_size;
4794 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4795 	}
4796 
4797 	btrfs_free_path(path);
4798 
4799 	if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4800 		unsigned long updates = trans->delayed_ref_updates;
4801 		int err;
4802 
4803 		if (updates) {
4804 			trans->delayed_ref_updates = 0;
4805 			err = btrfs_run_delayed_refs(trans, updates * 2);
4806 			if (err)
4807 				ret = err;
4808 		}
4809 	}
4810 	return ret;
4811 }
4812 
4813 /*
4814  * btrfs_truncate_block - read, zero a chunk and write a block
4815  * @inode - inode that we're zeroing
4816  * @from - the offset to start zeroing
4817  * @len - the length to zero, 0 to zero the entire range respective to the
4818  *	offset
4819  * @front - zero up to the offset instead of from the offset on
4820  *
4821  * This will find the block for the "from" offset and cow the block and zero the
4822  * part we want to zero.  This is used with truncate and hole punching.
4823  */
4824 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4825 			int front)
4826 {
4827 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4828 	struct address_space *mapping = inode->i_mapping;
4829 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4830 	struct btrfs_ordered_extent *ordered;
4831 	struct extent_state *cached_state = NULL;
4832 	struct extent_changeset *data_reserved = NULL;
4833 	char *kaddr;
4834 	u32 blocksize = fs_info->sectorsize;
4835 	pgoff_t index = from >> PAGE_SHIFT;
4836 	unsigned offset = from & (blocksize - 1);
4837 	struct page *page;
4838 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4839 	int ret = 0;
4840 	u64 block_start;
4841 	u64 block_end;
4842 
4843 	if (IS_ALIGNED(offset, blocksize) &&
4844 	    (!len || IS_ALIGNED(len, blocksize)))
4845 		goto out;
4846 
4847 	block_start = round_down(from, blocksize);
4848 	block_end = block_start + blocksize - 1;
4849 
4850 	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4851 					   block_start, blocksize);
4852 	if (ret)
4853 		goto out;
4854 
4855 again:
4856 	page = find_or_create_page(mapping, index, mask);
4857 	if (!page) {
4858 		btrfs_delalloc_release_space(inode, data_reserved,
4859 					     block_start, blocksize, true);
4860 		btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4861 		ret = -ENOMEM;
4862 		goto out;
4863 	}
4864 
4865 	if (!PageUptodate(page)) {
4866 		ret = btrfs_readpage(NULL, page);
4867 		lock_page(page);
4868 		if (page->mapping != mapping) {
4869 			unlock_page(page);
4870 			put_page(page);
4871 			goto again;
4872 		}
4873 		if (!PageUptodate(page)) {
4874 			ret = -EIO;
4875 			goto out_unlock;
4876 		}
4877 	}
4878 	wait_on_page_writeback(page);
4879 
4880 	lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4881 	set_page_extent_mapped(page);
4882 
4883 	ordered = btrfs_lookup_ordered_extent(inode, block_start);
4884 	if (ordered) {
4885 		unlock_extent_cached(io_tree, block_start, block_end,
4886 				     &cached_state);
4887 		unlock_page(page);
4888 		put_page(page);
4889 		btrfs_start_ordered_extent(inode, ordered, 1);
4890 		btrfs_put_ordered_extent(ordered);
4891 		goto again;
4892 	}
4893 
4894 	clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4895 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4896 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4897 			  0, 0, &cached_state);
4898 
4899 	ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4900 					&cached_state, 0);
4901 	if (ret) {
4902 		unlock_extent_cached(io_tree, block_start, block_end,
4903 				     &cached_state);
4904 		goto out_unlock;
4905 	}
4906 
4907 	if (offset != blocksize) {
4908 		if (!len)
4909 			len = blocksize - offset;
4910 		kaddr = kmap(page);
4911 		if (front)
4912 			memset(kaddr + (block_start - page_offset(page)),
4913 				0, offset);
4914 		else
4915 			memset(kaddr + (block_start - page_offset(page)) +  offset,
4916 				0, len);
4917 		flush_dcache_page(page);
4918 		kunmap(page);
4919 	}
4920 	ClearPageChecked(page);
4921 	set_page_dirty(page);
4922 	unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4923 
4924 out_unlock:
4925 	if (ret)
4926 		btrfs_delalloc_release_space(inode, data_reserved, block_start,
4927 					     blocksize, true);
4928 	btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4929 	unlock_page(page);
4930 	put_page(page);
4931 out:
4932 	extent_changeset_free(data_reserved);
4933 	return ret;
4934 }
4935 
4936 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4937 			     u64 offset, u64 len)
4938 {
4939 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4940 	struct btrfs_trans_handle *trans;
4941 	int ret;
4942 
4943 	/*
4944 	 * Still need to make sure the inode looks like it's been updated so
4945 	 * that any holes get logged if we fsync.
4946 	 */
4947 	if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4948 		BTRFS_I(inode)->last_trans = fs_info->generation;
4949 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4950 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4951 		return 0;
4952 	}
4953 
4954 	/*
4955 	 * 1 - for the one we're dropping
4956 	 * 1 - for the one we're adding
4957 	 * 1 - for updating the inode.
4958 	 */
4959 	trans = btrfs_start_transaction(root, 3);
4960 	if (IS_ERR(trans))
4961 		return PTR_ERR(trans);
4962 
4963 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4964 	if (ret) {
4965 		btrfs_abort_transaction(trans, ret);
4966 		btrfs_end_transaction(trans);
4967 		return ret;
4968 	}
4969 
4970 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4971 			offset, 0, 0, len, 0, len, 0, 0, 0);
4972 	if (ret)
4973 		btrfs_abort_transaction(trans, ret);
4974 	else
4975 		btrfs_update_inode(trans, root, inode);
4976 	btrfs_end_transaction(trans);
4977 	return ret;
4978 }
4979 
4980 /*
4981  * This function puts in dummy file extents for the area we're creating a hole
4982  * for.  So if we are truncating this file to a larger size we need to insert
4983  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4984  * the range between oldsize and size
4985  */
4986 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4987 {
4988 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4989 	struct btrfs_root *root = BTRFS_I(inode)->root;
4990 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4991 	struct extent_map *em = NULL;
4992 	struct extent_state *cached_state = NULL;
4993 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4994 	u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4995 	u64 block_end = ALIGN(size, fs_info->sectorsize);
4996 	u64 last_byte;
4997 	u64 cur_offset;
4998 	u64 hole_size;
4999 	int err = 0;
5000 
5001 	/*
5002 	 * If our size started in the middle of a block we need to zero out the
5003 	 * rest of the block before we expand the i_size, otherwise we could
5004 	 * expose stale data.
5005 	 */
5006 	err = btrfs_truncate_block(inode, oldsize, 0, 0);
5007 	if (err)
5008 		return err;
5009 
5010 	if (size <= hole_start)
5011 		return 0;
5012 
5013 	while (1) {
5014 		struct btrfs_ordered_extent *ordered;
5015 
5016 		lock_extent_bits(io_tree, hole_start, block_end - 1,
5017 				 &cached_state);
5018 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5019 						     block_end - hole_start);
5020 		if (!ordered)
5021 			break;
5022 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
5023 				     &cached_state);
5024 		btrfs_start_ordered_extent(inode, ordered, 1);
5025 		btrfs_put_ordered_extent(ordered);
5026 	}
5027 
5028 	cur_offset = hole_start;
5029 	while (1) {
5030 		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5031 				block_end - cur_offset, 0);
5032 		if (IS_ERR(em)) {
5033 			err = PTR_ERR(em);
5034 			em = NULL;
5035 			break;
5036 		}
5037 		last_byte = min(extent_map_end(em), block_end);
5038 		last_byte = ALIGN(last_byte, fs_info->sectorsize);
5039 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5040 			struct extent_map *hole_em;
5041 			hole_size = last_byte - cur_offset;
5042 
5043 			err = maybe_insert_hole(root, inode, cur_offset,
5044 						hole_size);
5045 			if (err)
5046 				break;
5047 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5048 						cur_offset + hole_size - 1, 0);
5049 			hole_em = alloc_extent_map();
5050 			if (!hole_em) {
5051 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5052 					&BTRFS_I(inode)->runtime_flags);
5053 				goto next;
5054 			}
5055 			hole_em->start = cur_offset;
5056 			hole_em->len = hole_size;
5057 			hole_em->orig_start = cur_offset;
5058 
5059 			hole_em->block_start = EXTENT_MAP_HOLE;
5060 			hole_em->block_len = 0;
5061 			hole_em->orig_block_len = 0;
5062 			hole_em->ram_bytes = hole_size;
5063 			hole_em->bdev = fs_info->fs_devices->latest_bdev;
5064 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
5065 			hole_em->generation = fs_info->generation;
5066 
5067 			while (1) {
5068 				write_lock(&em_tree->lock);
5069 				err = add_extent_mapping(em_tree, hole_em, 1);
5070 				write_unlock(&em_tree->lock);
5071 				if (err != -EEXIST)
5072 					break;
5073 				btrfs_drop_extent_cache(BTRFS_I(inode),
5074 							cur_offset,
5075 							cur_offset +
5076 							hole_size - 1, 0);
5077 			}
5078 			free_extent_map(hole_em);
5079 		}
5080 next:
5081 		free_extent_map(em);
5082 		em = NULL;
5083 		cur_offset = last_byte;
5084 		if (cur_offset >= block_end)
5085 			break;
5086 	}
5087 	free_extent_map(em);
5088 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5089 	return err;
5090 }
5091 
5092 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5093 {
5094 	struct btrfs_root *root = BTRFS_I(inode)->root;
5095 	struct btrfs_trans_handle *trans;
5096 	loff_t oldsize = i_size_read(inode);
5097 	loff_t newsize = attr->ia_size;
5098 	int mask = attr->ia_valid;
5099 	int ret;
5100 
5101 	/*
5102 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5103 	 * special case where we need to update the times despite not having
5104 	 * these flags set.  For all other operations the VFS set these flags
5105 	 * explicitly if it wants a timestamp update.
5106 	 */
5107 	if (newsize != oldsize) {
5108 		inode_inc_iversion(inode);
5109 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5110 			inode->i_ctime = inode->i_mtime =
5111 				current_time(inode);
5112 	}
5113 
5114 	if (newsize > oldsize) {
5115 		/*
5116 		 * Don't do an expanding truncate while snapshotting is ongoing.
5117 		 * This is to ensure the snapshot captures a fully consistent
5118 		 * state of this file - if the snapshot captures this expanding
5119 		 * truncation, it must capture all writes that happened before
5120 		 * this truncation.
5121 		 */
5122 		btrfs_wait_for_snapshot_creation(root);
5123 		ret = btrfs_cont_expand(inode, oldsize, newsize);
5124 		if (ret) {
5125 			btrfs_end_write_no_snapshotting(root);
5126 			return ret;
5127 		}
5128 
5129 		trans = btrfs_start_transaction(root, 1);
5130 		if (IS_ERR(trans)) {
5131 			btrfs_end_write_no_snapshotting(root);
5132 			return PTR_ERR(trans);
5133 		}
5134 
5135 		i_size_write(inode, newsize);
5136 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5137 		pagecache_isize_extended(inode, oldsize, newsize);
5138 		ret = btrfs_update_inode(trans, root, inode);
5139 		btrfs_end_write_no_snapshotting(root);
5140 		btrfs_end_transaction(trans);
5141 	} else {
5142 
5143 		/*
5144 		 * We're truncating a file that used to have good data down to
5145 		 * zero. Make sure it gets into the ordered flush list so that
5146 		 * any new writes get down to disk quickly.
5147 		 */
5148 		if (newsize == 0)
5149 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5150 				&BTRFS_I(inode)->runtime_flags);
5151 
5152 		truncate_setsize(inode, newsize);
5153 
5154 		/* Disable nonlocked read DIO to avoid the end less truncate */
5155 		btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5156 		inode_dio_wait(inode);
5157 		btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5158 
5159 		ret = btrfs_truncate(inode, newsize == oldsize);
5160 		if (ret && inode->i_nlink) {
5161 			int err;
5162 
5163 			/*
5164 			 * Truncate failed, so fix up the in-memory size. We
5165 			 * adjusted disk_i_size down as we removed extents, so
5166 			 * wait for disk_i_size to be stable and then update the
5167 			 * in-memory size to match.
5168 			 */
5169 			err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5170 			if (err)
5171 				return err;
5172 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5173 		}
5174 	}
5175 
5176 	return ret;
5177 }
5178 
5179 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5180 {
5181 	struct inode *inode = d_inode(dentry);
5182 	struct btrfs_root *root = BTRFS_I(inode)->root;
5183 	int err;
5184 
5185 	if (btrfs_root_readonly(root))
5186 		return -EROFS;
5187 
5188 	err = setattr_prepare(dentry, attr);
5189 	if (err)
5190 		return err;
5191 
5192 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5193 		err = btrfs_setsize(inode, attr);
5194 		if (err)
5195 			return err;
5196 	}
5197 
5198 	if (attr->ia_valid) {
5199 		setattr_copy(inode, attr);
5200 		inode_inc_iversion(inode);
5201 		err = btrfs_dirty_inode(inode);
5202 
5203 		if (!err && attr->ia_valid & ATTR_MODE)
5204 			err = posix_acl_chmod(inode, inode->i_mode);
5205 	}
5206 
5207 	return err;
5208 }
5209 
5210 /*
5211  * While truncating the inode pages during eviction, we get the VFS calling
5212  * btrfs_invalidatepage() against each page of the inode. This is slow because
5213  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5214  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5215  * extent_state structures over and over, wasting lots of time.
5216  *
5217  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5218  * those expensive operations on a per page basis and do only the ordered io
5219  * finishing, while we release here the extent_map and extent_state structures,
5220  * without the excessive merging and splitting.
5221  */
5222 static void evict_inode_truncate_pages(struct inode *inode)
5223 {
5224 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5225 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5226 	struct rb_node *node;
5227 
5228 	ASSERT(inode->i_state & I_FREEING);
5229 	truncate_inode_pages_final(&inode->i_data);
5230 
5231 	write_lock(&map_tree->lock);
5232 	while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5233 		struct extent_map *em;
5234 
5235 		node = rb_first_cached(&map_tree->map);
5236 		em = rb_entry(node, struct extent_map, rb_node);
5237 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5238 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5239 		remove_extent_mapping(map_tree, em);
5240 		free_extent_map(em);
5241 		if (need_resched()) {
5242 			write_unlock(&map_tree->lock);
5243 			cond_resched();
5244 			write_lock(&map_tree->lock);
5245 		}
5246 	}
5247 	write_unlock(&map_tree->lock);
5248 
5249 	/*
5250 	 * Keep looping until we have no more ranges in the io tree.
5251 	 * We can have ongoing bios started by readpages (called from readahead)
5252 	 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5253 	 * still in progress (unlocked the pages in the bio but did not yet
5254 	 * unlocked the ranges in the io tree). Therefore this means some
5255 	 * ranges can still be locked and eviction started because before
5256 	 * submitting those bios, which are executed by a separate task (work
5257 	 * queue kthread), inode references (inode->i_count) were not taken
5258 	 * (which would be dropped in the end io callback of each bio).
5259 	 * Therefore here we effectively end up waiting for those bios and
5260 	 * anyone else holding locked ranges without having bumped the inode's
5261 	 * reference count - if we don't do it, when they access the inode's
5262 	 * io_tree to unlock a range it may be too late, leading to an
5263 	 * use-after-free issue.
5264 	 */
5265 	spin_lock(&io_tree->lock);
5266 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
5267 		struct extent_state *state;
5268 		struct extent_state *cached_state = NULL;
5269 		u64 start;
5270 		u64 end;
5271 		unsigned state_flags;
5272 
5273 		node = rb_first(&io_tree->state);
5274 		state = rb_entry(node, struct extent_state, rb_node);
5275 		start = state->start;
5276 		end = state->end;
5277 		state_flags = state->state;
5278 		spin_unlock(&io_tree->lock);
5279 
5280 		lock_extent_bits(io_tree, start, end, &cached_state);
5281 
5282 		/*
5283 		 * If still has DELALLOC flag, the extent didn't reach disk,
5284 		 * and its reserved space won't be freed by delayed_ref.
5285 		 * So we need to free its reserved space here.
5286 		 * (Refer to comment in btrfs_invalidatepage, case 2)
5287 		 *
5288 		 * Note, end is the bytenr of last byte, so we need + 1 here.
5289 		 */
5290 		if (state_flags & EXTENT_DELALLOC)
5291 			btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5292 
5293 		clear_extent_bit(io_tree, start, end,
5294 				 EXTENT_LOCKED | EXTENT_DIRTY |
5295 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5296 				 EXTENT_DEFRAG, 1, 1, &cached_state);
5297 
5298 		cond_resched();
5299 		spin_lock(&io_tree->lock);
5300 	}
5301 	spin_unlock(&io_tree->lock);
5302 }
5303 
5304 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5305 							struct btrfs_block_rsv *rsv)
5306 {
5307 	struct btrfs_fs_info *fs_info = root->fs_info;
5308 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5309 	int failures = 0;
5310 
5311 	for (;;) {
5312 		struct btrfs_trans_handle *trans;
5313 		int ret;
5314 
5315 		ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5316 					     BTRFS_RESERVE_FLUSH_LIMIT);
5317 
5318 		if (ret && ++failures > 2) {
5319 			btrfs_warn(fs_info,
5320 				   "could not allocate space for a delete; will truncate on mount");
5321 			return ERR_PTR(-ENOSPC);
5322 		}
5323 
5324 		trans = btrfs_join_transaction(root);
5325 		if (IS_ERR(trans) || !ret)
5326 			return trans;
5327 
5328 		/*
5329 		 * Try to steal from the global reserve if there is space for
5330 		 * it.
5331 		 */
5332 		if (!btrfs_check_space_for_delayed_refs(trans) &&
5333 		    !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5334 			return trans;
5335 
5336 		/* If not, commit and try again. */
5337 		ret = btrfs_commit_transaction(trans);
5338 		if (ret)
5339 			return ERR_PTR(ret);
5340 	}
5341 }
5342 
5343 void btrfs_evict_inode(struct inode *inode)
5344 {
5345 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5346 	struct btrfs_trans_handle *trans;
5347 	struct btrfs_root *root = BTRFS_I(inode)->root;
5348 	struct btrfs_block_rsv *rsv;
5349 	int ret;
5350 
5351 	trace_btrfs_inode_evict(inode);
5352 
5353 	if (!root) {
5354 		clear_inode(inode);
5355 		return;
5356 	}
5357 
5358 	evict_inode_truncate_pages(inode);
5359 
5360 	if (inode->i_nlink &&
5361 	    ((btrfs_root_refs(&root->root_item) != 0 &&
5362 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5363 	     btrfs_is_free_space_inode(BTRFS_I(inode))))
5364 		goto no_delete;
5365 
5366 	if (is_bad_inode(inode))
5367 		goto no_delete;
5368 
5369 	btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5370 
5371 	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5372 		goto no_delete;
5373 
5374 	if (inode->i_nlink > 0) {
5375 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5376 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5377 		goto no_delete;
5378 	}
5379 
5380 	ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5381 	if (ret)
5382 		goto no_delete;
5383 
5384 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5385 	if (!rsv)
5386 		goto no_delete;
5387 	rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5388 	rsv->failfast = 1;
5389 
5390 	btrfs_i_size_write(BTRFS_I(inode), 0);
5391 
5392 	while (1) {
5393 		trans = evict_refill_and_join(root, rsv);
5394 		if (IS_ERR(trans))
5395 			goto free_rsv;
5396 
5397 		trans->block_rsv = rsv;
5398 
5399 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5400 		trans->block_rsv = &fs_info->trans_block_rsv;
5401 		btrfs_end_transaction(trans);
5402 		btrfs_btree_balance_dirty(fs_info);
5403 		if (ret && ret != -ENOSPC && ret != -EAGAIN)
5404 			goto free_rsv;
5405 		else if (!ret)
5406 			break;
5407 	}
5408 
5409 	/*
5410 	 * Errors here aren't a big deal, it just means we leave orphan items in
5411 	 * the tree. They will be cleaned up on the next mount. If the inode
5412 	 * number gets reused, cleanup deletes the orphan item without doing
5413 	 * anything, and unlink reuses the existing orphan item.
5414 	 *
5415 	 * If it turns out that we are dropping too many of these, we might want
5416 	 * to add a mechanism for retrying these after a commit.
5417 	 */
5418 	trans = evict_refill_and_join(root, rsv);
5419 	if (!IS_ERR(trans)) {
5420 		trans->block_rsv = rsv;
5421 		btrfs_orphan_del(trans, BTRFS_I(inode));
5422 		trans->block_rsv = &fs_info->trans_block_rsv;
5423 		btrfs_end_transaction(trans);
5424 	}
5425 
5426 	if (!(root == fs_info->tree_root ||
5427 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5428 		btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5429 
5430 free_rsv:
5431 	btrfs_free_block_rsv(fs_info, rsv);
5432 no_delete:
5433 	/*
5434 	 * If we didn't successfully delete, the orphan item will still be in
5435 	 * the tree and we'll retry on the next mount. Again, we might also want
5436 	 * to retry these periodically in the future.
5437 	 */
5438 	btrfs_remove_delayed_node(BTRFS_I(inode));
5439 	clear_inode(inode);
5440 }
5441 
5442 /*
5443  * this returns the key found in the dir entry in the location pointer.
5444  * If no dir entries were found, returns -ENOENT.
5445  * If found a corrupted location in dir entry, returns -EUCLEAN.
5446  */
5447 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5448 			       struct btrfs_key *location)
5449 {
5450 	const char *name = dentry->d_name.name;
5451 	int namelen = dentry->d_name.len;
5452 	struct btrfs_dir_item *di;
5453 	struct btrfs_path *path;
5454 	struct btrfs_root *root = BTRFS_I(dir)->root;
5455 	int ret = 0;
5456 
5457 	path = btrfs_alloc_path();
5458 	if (!path)
5459 		return -ENOMEM;
5460 
5461 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5462 			name, namelen, 0);
5463 	if (IS_ERR_OR_NULL(di)) {
5464 		ret = di ? PTR_ERR(di) : -ENOENT;
5465 		goto out;
5466 	}
5467 
5468 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5469 	if (location->type != BTRFS_INODE_ITEM_KEY &&
5470 	    location->type != BTRFS_ROOT_ITEM_KEY) {
5471 		ret = -EUCLEAN;
5472 		btrfs_warn(root->fs_info,
5473 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5474 			   __func__, name, btrfs_ino(BTRFS_I(dir)),
5475 			   location->objectid, location->type, location->offset);
5476 	}
5477 out:
5478 	btrfs_free_path(path);
5479 	return ret;
5480 }
5481 
5482 /*
5483  * when we hit a tree root in a directory, the btrfs part of the inode
5484  * needs to be changed to reflect the root directory of the tree root.  This
5485  * is kind of like crossing a mount point.
5486  */
5487 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5488 				    struct inode *dir,
5489 				    struct dentry *dentry,
5490 				    struct btrfs_key *location,
5491 				    struct btrfs_root **sub_root)
5492 {
5493 	struct btrfs_path *path;
5494 	struct btrfs_root *new_root;
5495 	struct btrfs_root_ref *ref;
5496 	struct extent_buffer *leaf;
5497 	struct btrfs_key key;
5498 	int ret;
5499 	int err = 0;
5500 
5501 	path = btrfs_alloc_path();
5502 	if (!path) {
5503 		err = -ENOMEM;
5504 		goto out;
5505 	}
5506 
5507 	err = -ENOENT;
5508 	key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5509 	key.type = BTRFS_ROOT_REF_KEY;
5510 	key.offset = location->objectid;
5511 
5512 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5513 	if (ret) {
5514 		if (ret < 0)
5515 			err = ret;
5516 		goto out;
5517 	}
5518 
5519 	leaf = path->nodes[0];
5520 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5521 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5522 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5523 		goto out;
5524 
5525 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5526 				   (unsigned long)(ref + 1),
5527 				   dentry->d_name.len);
5528 	if (ret)
5529 		goto out;
5530 
5531 	btrfs_release_path(path);
5532 
5533 	new_root = btrfs_read_fs_root_no_name(fs_info, location);
5534 	if (IS_ERR(new_root)) {
5535 		err = PTR_ERR(new_root);
5536 		goto out;
5537 	}
5538 
5539 	*sub_root = new_root;
5540 	location->objectid = btrfs_root_dirid(&new_root->root_item);
5541 	location->type = BTRFS_INODE_ITEM_KEY;
5542 	location->offset = 0;
5543 	err = 0;
5544 out:
5545 	btrfs_free_path(path);
5546 	return err;
5547 }
5548 
5549 static void inode_tree_add(struct inode *inode)
5550 {
5551 	struct btrfs_root *root = BTRFS_I(inode)->root;
5552 	struct btrfs_inode *entry;
5553 	struct rb_node **p;
5554 	struct rb_node *parent;
5555 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
5556 	u64 ino = btrfs_ino(BTRFS_I(inode));
5557 
5558 	if (inode_unhashed(inode))
5559 		return;
5560 	parent = NULL;
5561 	spin_lock(&root->inode_lock);
5562 	p = &root->inode_tree.rb_node;
5563 	while (*p) {
5564 		parent = *p;
5565 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
5566 
5567 		if (ino < btrfs_ino(entry))
5568 			p = &parent->rb_left;
5569 		else if (ino > btrfs_ino(entry))
5570 			p = &parent->rb_right;
5571 		else {
5572 			WARN_ON(!(entry->vfs_inode.i_state &
5573 				  (I_WILL_FREE | I_FREEING)));
5574 			rb_replace_node(parent, new, &root->inode_tree);
5575 			RB_CLEAR_NODE(parent);
5576 			spin_unlock(&root->inode_lock);
5577 			return;
5578 		}
5579 	}
5580 	rb_link_node(new, parent, p);
5581 	rb_insert_color(new, &root->inode_tree);
5582 	spin_unlock(&root->inode_lock);
5583 }
5584 
5585 static void inode_tree_del(struct inode *inode)
5586 {
5587 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5588 	struct btrfs_root *root = BTRFS_I(inode)->root;
5589 	int empty = 0;
5590 
5591 	spin_lock(&root->inode_lock);
5592 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5593 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5594 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5595 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5596 	}
5597 	spin_unlock(&root->inode_lock);
5598 
5599 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
5600 		synchronize_srcu(&fs_info->subvol_srcu);
5601 		spin_lock(&root->inode_lock);
5602 		empty = RB_EMPTY_ROOT(&root->inode_tree);
5603 		spin_unlock(&root->inode_lock);
5604 		if (empty)
5605 			btrfs_add_dead_root(root);
5606 	}
5607 }
5608 
5609 
5610 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5611 {
5612 	struct btrfs_iget_args *args = p;
5613 	inode->i_ino = args->location->objectid;
5614 	memcpy(&BTRFS_I(inode)->location, args->location,
5615 	       sizeof(*args->location));
5616 	BTRFS_I(inode)->root = args->root;
5617 	return 0;
5618 }
5619 
5620 static int btrfs_find_actor(struct inode *inode, void *opaque)
5621 {
5622 	struct btrfs_iget_args *args = opaque;
5623 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5624 		args->root == BTRFS_I(inode)->root;
5625 }
5626 
5627 static struct inode *btrfs_iget_locked(struct super_block *s,
5628 				       struct btrfs_key *location,
5629 				       struct btrfs_root *root)
5630 {
5631 	struct inode *inode;
5632 	struct btrfs_iget_args args;
5633 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5634 
5635 	args.location = location;
5636 	args.root = root;
5637 
5638 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5639 			     btrfs_init_locked_inode,
5640 			     (void *)&args);
5641 	return inode;
5642 }
5643 
5644 /* Get an inode object given its location and corresponding root.
5645  * Returns in *is_new if the inode was read from disk
5646  */
5647 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5648 			 struct btrfs_root *root, int *new)
5649 {
5650 	struct inode *inode;
5651 
5652 	inode = btrfs_iget_locked(s, location, root);
5653 	if (!inode)
5654 		return ERR_PTR(-ENOMEM);
5655 
5656 	if (inode->i_state & I_NEW) {
5657 		int ret;
5658 
5659 		ret = btrfs_read_locked_inode(inode);
5660 		if (!ret) {
5661 			inode_tree_add(inode);
5662 			unlock_new_inode(inode);
5663 			if (new)
5664 				*new = 1;
5665 		} else {
5666 			iget_failed(inode);
5667 			/*
5668 			 * ret > 0 can come from btrfs_search_slot called by
5669 			 * btrfs_read_locked_inode, this means the inode item
5670 			 * was not found.
5671 			 */
5672 			if (ret > 0)
5673 				ret = -ENOENT;
5674 			inode = ERR_PTR(ret);
5675 		}
5676 	}
5677 
5678 	return inode;
5679 }
5680 
5681 static struct inode *new_simple_dir(struct super_block *s,
5682 				    struct btrfs_key *key,
5683 				    struct btrfs_root *root)
5684 {
5685 	struct inode *inode = new_inode(s);
5686 
5687 	if (!inode)
5688 		return ERR_PTR(-ENOMEM);
5689 
5690 	BTRFS_I(inode)->root = root;
5691 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5692 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5693 
5694 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5695 	inode->i_op = &btrfs_dir_ro_inode_operations;
5696 	inode->i_opflags &= ~IOP_XATTR;
5697 	inode->i_fop = &simple_dir_operations;
5698 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5699 	inode->i_mtime = current_time(inode);
5700 	inode->i_atime = inode->i_mtime;
5701 	inode->i_ctime = inode->i_mtime;
5702 	BTRFS_I(inode)->i_otime = inode->i_mtime;
5703 
5704 	return inode;
5705 }
5706 
5707 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5708 {
5709 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5710 	struct inode *inode;
5711 	struct btrfs_root *root = BTRFS_I(dir)->root;
5712 	struct btrfs_root *sub_root = root;
5713 	struct btrfs_key location;
5714 	int index;
5715 	int ret = 0;
5716 
5717 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5718 		return ERR_PTR(-ENAMETOOLONG);
5719 
5720 	ret = btrfs_inode_by_name(dir, dentry, &location);
5721 	if (ret < 0)
5722 		return ERR_PTR(ret);
5723 
5724 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5725 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5726 		return inode;
5727 	}
5728 
5729 	index = srcu_read_lock(&fs_info->subvol_srcu);
5730 	ret = fixup_tree_root_location(fs_info, dir, dentry,
5731 				       &location, &sub_root);
5732 	if (ret < 0) {
5733 		if (ret != -ENOENT)
5734 			inode = ERR_PTR(ret);
5735 		else
5736 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5737 	} else {
5738 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5739 	}
5740 	srcu_read_unlock(&fs_info->subvol_srcu, index);
5741 
5742 	if (!IS_ERR(inode) && root != sub_root) {
5743 		down_read(&fs_info->cleanup_work_sem);
5744 		if (!sb_rdonly(inode->i_sb))
5745 			ret = btrfs_orphan_cleanup(sub_root);
5746 		up_read(&fs_info->cleanup_work_sem);
5747 		if (ret) {
5748 			iput(inode);
5749 			inode = ERR_PTR(ret);
5750 		}
5751 	}
5752 
5753 	return inode;
5754 }
5755 
5756 static int btrfs_dentry_delete(const struct dentry *dentry)
5757 {
5758 	struct btrfs_root *root;
5759 	struct inode *inode = d_inode(dentry);
5760 
5761 	if (!inode && !IS_ROOT(dentry))
5762 		inode = d_inode(dentry->d_parent);
5763 
5764 	if (inode) {
5765 		root = BTRFS_I(inode)->root;
5766 		if (btrfs_root_refs(&root->root_item) == 0)
5767 			return 1;
5768 
5769 		if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5770 			return 1;
5771 	}
5772 	return 0;
5773 }
5774 
5775 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5776 				   unsigned int flags)
5777 {
5778 	struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5779 
5780 	if (inode == ERR_PTR(-ENOENT))
5781 		inode = NULL;
5782 	return d_splice_alias(inode, dentry);
5783 }
5784 
5785 unsigned char btrfs_filetype_table[] = {
5786 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5787 };
5788 
5789 /*
5790  * All this infrastructure exists because dir_emit can fault, and we are holding
5791  * the tree lock when doing readdir.  For now just allocate a buffer and copy
5792  * our information into that, and then dir_emit from the buffer.  This is
5793  * similar to what NFS does, only we don't keep the buffer around in pagecache
5794  * because I'm afraid I'll mess that up.  Long term we need to make filldir do
5795  * copy_to_user_inatomic so we don't have to worry about page faulting under the
5796  * tree lock.
5797  */
5798 static int btrfs_opendir(struct inode *inode, struct file *file)
5799 {
5800 	struct btrfs_file_private *private;
5801 
5802 	private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5803 	if (!private)
5804 		return -ENOMEM;
5805 	private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5806 	if (!private->filldir_buf) {
5807 		kfree(private);
5808 		return -ENOMEM;
5809 	}
5810 	file->private_data = private;
5811 	return 0;
5812 }
5813 
5814 struct dir_entry {
5815 	u64 ino;
5816 	u64 offset;
5817 	unsigned type;
5818 	int name_len;
5819 };
5820 
5821 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5822 {
5823 	while (entries--) {
5824 		struct dir_entry *entry = addr;
5825 		char *name = (char *)(entry + 1);
5826 
5827 		ctx->pos = get_unaligned(&entry->offset);
5828 		if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5829 					 get_unaligned(&entry->ino),
5830 					 get_unaligned(&entry->type)))
5831 			return 1;
5832 		addr += sizeof(struct dir_entry) +
5833 			get_unaligned(&entry->name_len);
5834 		ctx->pos++;
5835 	}
5836 	return 0;
5837 }
5838 
5839 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5840 {
5841 	struct inode *inode = file_inode(file);
5842 	struct btrfs_root *root = BTRFS_I(inode)->root;
5843 	struct btrfs_file_private *private = file->private_data;
5844 	struct btrfs_dir_item *di;
5845 	struct btrfs_key key;
5846 	struct btrfs_key found_key;
5847 	struct btrfs_path *path;
5848 	void *addr;
5849 	struct list_head ins_list;
5850 	struct list_head del_list;
5851 	int ret;
5852 	struct extent_buffer *leaf;
5853 	int slot;
5854 	char *name_ptr;
5855 	int name_len;
5856 	int entries = 0;
5857 	int total_len = 0;
5858 	bool put = false;
5859 	struct btrfs_key location;
5860 
5861 	if (!dir_emit_dots(file, ctx))
5862 		return 0;
5863 
5864 	path = btrfs_alloc_path();
5865 	if (!path)
5866 		return -ENOMEM;
5867 
5868 	addr = private->filldir_buf;
5869 	path->reada = READA_FORWARD;
5870 
5871 	INIT_LIST_HEAD(&ins_list);
5872 	INIT_LIST_HEAD(&del_list);
5873 	put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5874 
5875 again:
5876 	key.type = BTRFS_DIR_INDEX_KEY;
5877 	key.offset = ctx->pos;
5878 	key.objectid = btrfs_ino(BTRFS_I(inode));
5879 
5880 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5881 	if (ret < 0)
5882 		goto err;
5883 
5884 	while (1) {
5885 		struct dir_entry *entry;
5886 
5887 		leaf = path->nodes[0];
5888 		slot = path->slots[0];
5889 		if (slot >= btrfs_header_nritems(leaf)) {
5890 			ret = btrfs_next_leaf(root, path);
5891 			if (ret < 0)
5892 				goto err;
5893 			else if (ret > 0)
5894 				break;
5895 			continue;
5896 		}
5897 
5898 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5899 
5900 		if (found_key.objectid != key.objectid)
5901 			break;
5902 		if (found_key.type != BTRFS_DIR_INDEX_KEY)
5903 			break;
5904 		if (found_key.offset < ctx->pos)
5905 			goto next;
5906 		if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5907 			goto next;
5908 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5909 		name_len = btrfs_dir_name_len(leaf, di);
5910 		if ((total_len + sizeof(struct dir_entry) + name_len) >=
5911 		    PAGE_SIZE) {
5912 			btrfs_release_path(path);
5913 			ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5914 			if (ret)
5915 				goto nopos;
5916 			addr = private->filldir_buf;
5917 			entries = 0;
5918 			total_len = 0;
5919 			goto again;
5920 		}
5921 
5922 		entry = addr;
5923 		put_unaligned(name_len, &entry->name_len);
5924 		name_ptr = (char *)(entry + 1);
5925 		read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5926 				   name_len);
5927 		put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5928 				&entry->type);
5929 		btrfs_dir_item_key_to_cpu(leaf, di, &location);
5930 		put_unaligned(location.objectid, &entry->ino);
5931 		put_unaligned(found_key.offset, &entry->offset);
5932 		entries++;
5933 		addr += sizeof(struct dir_entry) + name_len;
5934 		total_len += sizeof(struct dir_entry) + name_len;
5935 next:
5936 		path->slots[0]++;
5937 	}
5938 	btrfs_release_path(path);
5939 
5940 	ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5941 	if (ret)
5942 		goto nopos;
5943 
5944 	ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5945 	if (ret)
5946 		goto nopos;
5947 
5948 	/*
5949 	 * Stop new entries from being returned after we return the last
5950 	 * entry.
5951 	 *
5952 	 * New directory entries are assigned a strictly increasing
5953 	 * offset.  This means that new entries created during readdir
5954 	 * are *guaranteed* to be seen in the future by that readdir.
5955 	 * This has broken buggy programs which operate on names as
5956 	 * they're returned by readdir.  Until we re-use freed offsets
5957 	 * we have this hack to stop new entries from being returned
5958 	 * under the assumption that they'll never reach this huge
5959 	 * offset.
5960 	 *
5961 	 * This is being careful not to overflow 32bit loff_t unless the
5962 	 * last entry requires it because doing so has broken 32bit apps
5963 	 * in the past.
5964 	 */
5965 	if (ctx->pos >= INT_MAX)
5966 		ctx->pos = LLONG_MAX;
5967 	else
5968 		ctx->pos = INT_MAX;
5969 nopos:
5970 	ret = 0;
5971 err:
5972 	if (put)
5973 		btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5974 	btrfs_free_path(path);
5975 	return ret;
5976 }
5977 
5978 /*
5979  * This is somewhat expensive, updating the tree every time the
5980  * inode changes.  But, it is most likely to find the inode in cache.
5981  * FIXME, needs more benchmarking...there are no reasons other than performance
5982  * to keep or drop this code.
5983  */
5984 static int btrfs_dirty_inode(struct inode *inode)
5985 {
5986 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5987 	struct btrfs_root *root = BTRFS_I(inode)->root;
5988 	struct btrfs_trans_handle *trans;
5989 	int ret;
5990 
5991 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5992 		return 0;
5993 
5994 	trans = btrfs_join_transaction(root);
5995 	if (IS_ERR(trans))
5996 		return PTR_ERR(trans);
5997 
5998 	ret = btrfs_update_inode(trans, root, inode);
5999 	if (ret && ret == -ENOSPC) {
6000 		/* whoops, lets try again with the full transaction */
6001 		btrfs_end_transaction(trans);
6002 		trans = btrfs_start_transaction(root, 1);
6003 		if (IS_ERR(trans))
6004 			return PTR_ERR(trans);
6005 
6006 		ret = btrfs_update_inode(trans, root, inode);
6007 	}
6008 	btrfs_end_transaction(trans);
6009 	if (BTRFS_I(inode)->delayed_node)
6010 		btrfs_balance_delayed_items(fs_info);
6011 
6012 	return ret;
6013 }
6014 
6015 /*
6016  * This is a copy of file_update_time.  We need this so we can return error on
6017  * ENOSPC for updating the inode in the case of file write and mmap writes.
6018  */
6019 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6020 			     int flags)
6021 {
6022 	struct btrfs_root *root = BTRFS_I(inode)->root;
6023 	bool dirty = flags & ~S_VERSION;
6024 
6025 	if (btrfs_root_readonly(root))
6026 		return -EROFS;
6027 
6028 	if (flags & S_VERSION)
6029 		dirty |= inode_maybe_inc_iversion(inode, dirty);
6030 	if (flags & S_CTIME)
6031 		inode->i_ctime = *now;
6032 	if (flags & S_MTIME)
6033 		inode->i_mtime = *now;
6034 	if (flags & S_ATIME)
6035 		inode->i_atime = *now;
6036 	return dirty ? btrfs_dirty_inode(inode) : 0;
6037 }
6038 
6039 /*
6040  * find the highest existing sequence number in a directory
6041  * and then set the in-memory index_cnt variable to reflect
6042  * free sequence numbers
6043  */
6044 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6045 {
6046 	struct btrfs_root *root = inode->root;
6047 	struct btrfs_key key, found_key;
6048 	struct btrfs_path *path;
6049 	struct extent_buffer *leaf;
6050 	int ret;
6051 
6052 	key.objectid = btrfs_ino(inode);
6053 	key.type = BTRFS_DIR_INDEX_KEY;
6054 	key.offset = (u64)-1;
6055 
6056 	path = btrfs_alloc_path();
6057 	if (!path)
6058 		return -ENOMEM;
6059 
6060 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6061 	if (ret < 0)
6062 		goto out;
6063 	/* FIXME: we should be able to handle this */
6064 	if (ret == 0)
6065 		goto out;
6066 	ret = 0;
6067 
6068 	/*
6069 	 * MAGIC NUMBER EXPLANATION:
6070 	 * since we search a directory based on f_pos we have to start at 2
6071 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6072 	 * else has to start at 2
6073 	 */
6074 	if (path->slots[0] == 0) {
6075 		inode->index_cnt = 2;
6076 		goto out;
6077 	}
6078 
6079 	path->slots[0]--;
6080 
6081 	leaf = path->nodes[0];
6082 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6083 
6084 	if (found_key.objectid != btrfs_ino(inode) ||
6085 	    found_key.type != BTRFS_DIR_INDEX_KEY) {
6086 		inode->index_cnt = 2;
6087 		goto out;
6088 	}
6089 
6090 	inode->index_cnt = found_key.offset + 1;
6091 out:
6092 	btrfs_free_path(path);
6093 	return ret;
6094 }
6095 
6096 /*
6097  * helper to find a free sequence number in a given directory.  This current
6098  * code is very simple, later versions will do smarter things in the btree
6099  */
6100 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6101 {
6102 	int ret = 0;
6103 
6104 	if (dir->index_cnt == (u64)-1) {
6105 		ret = btrfs_inode_delayed_dir_index_count(dir);
6106 		if (ret) {
6107 			ret = btrfs_set_inode_index_count(dir);
6108 			if (ret)
6109 				return ret;
6110 		}
6111 	}
6112 
6113 	*index = dir->index_cnt;
6114 	dir->index_cnt++;
6115 
6116 	return ret;
6117 }
6118 
6119 static int btrfs_insert_inode_locked(struct inode *inode)
6120 {
6121 	struct btrfs_iget_args args;
6122 	args.location = &BTRFS_I(inode)->location;
6123 	args.root = BTRFS_I(inode)->root;
6124 
6125 	return insert_inode_locked4(inode,
6126 		   btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6127 		   btrfs_find_actor, &args);
6128 }
6129 
6130 /*
6131  * Inherit flags from the parent inode.
6132  *
6133  * Currently only the compression flags and the cow flags are inherited.
6134  */
6135 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6136 {
6137 	unsigned int flags;
6138 
6139 	if (!dir)
6140 		return;
6141 
6142 	flags = BTRFS_I(dir)->flags;
6143 
6144 	if (flags & BTRFS_INODE_NOCOMPRESS) {
6145 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6146 		BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6147 	} else if (flags & BTRFS_INODE_COMPRESS) {
6148 		BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6149 		BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6150 	}
6151 
6152 	if (flags & BTRFS_INODE_NODATACOW) {
6153 		BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6154 		if (S_ISREG(inode->i_mode))
6155 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6156 	}
6157 
6158 	btrfs_sync_inode_flags_to_i_flags(inode);
6159 }
6160 
6161 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6162 				     struct btrfs_root *root,
6163 				     struct inode *dir,
6164 				     const char *name, int name_len,
6165 				     u64 ref_objectid, u64 objectid,
6166 				     umode_t mode, u64 *index)
6167 {
6168 	struct btrfs_fs_info *fs_info = root->fs_info;
6169 	struct inode *inode;
6170 	struct btrfs_inode_item *inode_item;
6171 	struct btrfs_key *location;
6172 	struct btrfs_path *path;
6173 	struct btrfs_inode_ref *ref;
6174 	struct btrfs_key key[2];
6175 	u32 sizes[2];
6176 	int nitems = name ? 2 : 1;
6177 	unsigned long ptr;
6178 	int ret;
6179 
6180 	path = btrfs_alloc_path();
6181 	if (!path)
6182 		return ERR_PTR(-ENOMEM);
6183 
6184 	inode = new_inode(fs_info->sb);
6185 	if (!inode) {
6186 		btrfs_free_path(path);
6187 		return ERR_PTR(-ENOMEM);
6188 	}
6189 
6190 	/*
6191 	 * O_TMPFILE, set link count to 0, so that after this point,
6192 	 * we fill in an inode item with the correct link count.
6193 	 */
6194 	if (!name)
6195 		set_nlink(inode, 0);
6196 
6197 	/*
6198 	 * we have to initialize this early, so we can reclaim the inode
6199 	 * number if we fail afterwards in this function.
6200 	 */
6201 	inode->i_ino = objectid;
6202 
6203 	if (dir && name) {
6204 		trace_btrfs_inode_request(dir);
6205 
6206 		ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6207 		if (ret) {
6208 			btrfs_free_path(path);
6209 			iput(inode);
6210 			return ERR_PTR(ret);
6211 		}
6212 	} else if (dir) {
6213 		*index = 0;
6214 	}
6215 	/*
6216 	 * index_cnt is ignored for everything but a dir,
6217 	 * btrfs_set_inode_index_count has an explanation for the magic
6218 	 * number
6219 	 */
6220 	BTRFS_I(inode)->index_cnt = 2;
6221 	BTRFS_I(inode)->dir_index = *index;
6222 	BTRFS_I(inode)->root = root;
6223 	BTRFS_I(inode)->generation = trans->transid;
6224 	inode->i_generation = BTRFS_I(inode)->generation;
6225 
6226 	/*
6227 	 * We could have gotten an inode number from somebody who was fsynced
6228 	 * and then removed in this same transaction, so let's just set full
6229 	 * sync since it will be a full sync anyway and this will blow away the
6230 	 * old info in the log.
6231 	 */
6232 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6233 
6234 	key[0].objectid = objectid;
6235 	key[0].type = BTRFS_INODE_ITEM_KEY;
6236 	key[0].offset = 0;
6237 
6238 	sizes[0] = sizeof(struct btrfs_inode_item);
6239 
6240 	if (name) {
6241 		/*
6242 		 * Start new inodes with an inode_ref. This is slightly more
6243 		 * efficient for small numbers of hard links since they will
6244 		 * be packed into one item. Extended refs will kick in if we
6245 		 * add more hard links than can fit in the ref item.
6246 		 */
6247 		key[1].objectid = objectid;
6248 		key[1].type = BTRFS_INODE_REF_KEY;
6249 		key[1].offset = ref_objectid;
6250 
6251 		sizes[1] = name_len + sizeof(*ref);
6252 	}
6253 
6254 	location = &BTRFS_I(inode)->location;
6255 	location->objectid = objectid;
6256 	location->offset = 0;
6257 	location->type = BTRFS_INODE_ITEM_KEY;
6258 
6259 	ret = btrfs_insert_inode_locked(inode);
6260 	if (ret < 0) {
6261 		iput(inode);
6262 		goto fail;
6263 	}
6264 
6265 	path->leave_spinning = 1;
6266 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6267 	if (ret != 0)
6268 		goto fail_unlock;
6269 
6270 	inode_init_owner(inode, dir, mode);
6271 	inode_set_bytes(inode, 0);
6272 
6273 	inode->i_mtime = current_time(inode);
6274 	inode->i_atime = inode->i_mtime;
6275 	inode->i_ctime = inode->i_mtime;
6276 	BTRFS_I(inode)->i_otime = inode->i_mtime;
6277 
6278 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6279 				  struct btrfs_inode_item);
6280 	memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6281 			     sizeof(*inode_item));
6282 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
6283 
6284 	if (name) {
6285 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6286 				     struct btrfs_inode_ref);
6287 		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6288 		btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6289 		ptr = (unsigned long)(ref + 1);
6290 		write_extent_buffer(path->nodes[0], name, ptr, name_len);
6291 	}
6292 
6293 	btrfs_mark_buffer_dirty(path->nodes[0]);
6294 	btrfs_free_path(path);
6295 
6296 	btrfs_inherit_iflags(inode, dir);
6297 
6298 	if (S_ISREG(mode)) {
6299 		if (btrfs_test_opt(fs_info, NODATASUM))
6300 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6301 		if (btrfs_test_opt(fs_info, NODATACOW))
6302 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6303 				BTRFS_INODE_NODATASUM;
6304 	}
6305 
6306 	inode_tree_add(inode);
6307 
6308 	trace_btrfs_inode_new(inode);
6309 	btrfs_set_inode_last_trans(trans, inode);
6310 
6311 	btrfs_update_root_times(trans, root);
6312 
6313 	ret = btrfs_inode_inherit_props(trans, inode, dir);
6314 	if (ret)
6315 		btrfs_err(fs_info,
6316 			  "error inheriting props for ino %llu (root %llu): %d",
6317 			btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6318 
6319 	return inode;
6320 
6321 fail_unlock:
6322 	discard_new_inode(inode);
6323 fail:
6324 	if (dir && name)
6325 		BTRFS_I(dir)->index_cnt--;
6326 	btrfs_free_path(path);
6327 	return ERR_PTR(ret);
6328 }
6329 
6330 static inline u8 btrfs_inode_type(struct inode *inode)
6331 {
6332 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6333 }
6334 
6335 /*
6336  * utility function to add 'inode' into 'parent_inode' with
6337  * a give name and a given sequence number.
6338  * if 'add_backref' is true, also insert a backref from the
6339  * inode to the parent directory.
6340  */
6341 int btrfs_add_link(struct btrfs_trans_handle *trans,
6342 		   struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6343 		   const char *name, int name_len, int add_backref, u64 index)
6344 {
6345 	int ret = 0;
6346 	struct btrfs_key key;
6347 	struct btrfs_root *root = parent_inode->root;
6348 	u64 ino = btrfs_ino(inode);
6349 	u64 parent_ino = btrfs_ino(parent_inode);
6350 
6351 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6352 		memcpy(&key, &inode->root->root_key, sizeof(key));
6353 	} else {
6354 		key.objectid = ino;
6355 		key.type = BTRFS_INODE_ITEM_KEY;
6356 		key.offset = 0;
6357 	}
6358 
6359 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6360 		ret = btrfs_add_root_ref(trans, key.objectid,
6361 					 root->root_key.objectid, parent_ino,
6362 					 index, name, name_len);
6363 	} else if (add_backref) {
6364 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6365 					     parent_ino, index);
6366 	}
6367 
6368 	/* Nothing to clean up yet */
6369 	if (ret)
6370 		return ret;
6371 
6372 	ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6373 				    btrfs_inode_type(&inode->vfs_inode), index);
6374 	if (ret == -EEXIST || ret == -EOVERFLOW)
6375 		goto fail_dir_item;
6376 	else if (ret) {
6377 		btrfs_abort_transaction(trans, ret);
6378 		return ret;
6379 	}
6380 
6381 	btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6382 			   name_len * 2);
6383 	inode_inc_iversion(&parent_inode->vfs_inode);
6384 	parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6385 		current_time(&parent_inode->vfs_inode);
6386 	ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6387 	if (ret)
6388 		btrfs_abort_transaction(trans, ret);
6389 	return ret;
6390 
6391 fail_dir_item:
6392 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6393 		u64 local_index;
6394 		int err;
6395 		err = btrfs_del_root_ref(trans, key.objectid,
6396 					 root->root_key.objectid, parent_ino,
6397 					 &local_index, name, name_len);
6398 
6399 	} else if (add_backref) {
6400 		u64 local_index;
6401 		int err;
6402 
6403 		err = btrfs_del_inode_ref(trans, root, name, name_len,
6404 					  ino, parent_ino, &local_index);
6405 	}
6406 	return ret;
6407 }
6408 
6409 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6410 			    struct btrfs_inode *dir, struct dentry *dentry,
6411 			    struct btrfs_inode *inode, int backref, u64 index)
6412 {
6413 	int err = btrfs_add_link(trans, dir, inode,
6414 				 dentry->d_name.name, dentry->d_name.len,
6415 				 backref, index);
6416 	if (err > 0)
6417 		err = -EEXIST;
6418 	return err;
6419 }
6420 
6421 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6422 			umode_t mode, dev_t rdev)
6423 {
6424 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6425 	struct btrfs_trans_handle *trans;
6426 	struct btrfs_root *root = BTRFS_I(dir)->root;
6427 	struct inode *inode = NULL;
6428 	int err;
6429 	u64 objectid;
6430 	u64 index = 0;
6431 
6432 	/*
6433 	 * 2 for inode item and ref
6434 	 * 2 for dir items
6435 	 * 1 for xattr if selinux is on
6436 	 */
6437 	trans = btrfs_start_transaction(root, 5);
6438 	if (IS_ERR(trans))
6439 		return PTR_ERR(trans);
6440 
6441 	err = btrfs_find_free_ino(root, &objectid);
6442 	if (err)
6443 		goto out_unlock;
6444 
6445 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6446 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6447 			mode, &index);
6448 	if (IS_ERR(inode)) {
6449 		err = PTR_ERR(inode);
6450 		inode = NULL;
6451 		goto out_unlock;
6452 	}
6453 
6454 	/*
6455 	* If the active LSM wants to access the inode during
6456 	* d_instantiate it needs these. Smack checks to see
6457 	* if the filesystem supports xattrs by looking at the
6458 	* ops vector.
6459 	*/
6460 	inode->i_op = &btrfs_special_inode_operations;
6461 	init_special_inode(inode, inode->i_mode, rdev);
6462 
6463 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6464 	if (err)
6465 		goto out_unlock;
6466 
6467 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6468 			0, index);
6469 	if (err)
6470 		goto out_unlock;
6471 
6472 	btrfs_update_inode(trans, root, inode);
6473 	d_instantiate_new(dentry, inode);
6474 
6475 out_unlock:
6476 	btrfs_end_transaction(trans);
6477 	btrfs_btree_balance_dirty(fs_info);
6478 	if (err && inode) {
6479 		inode_dec_link_count(inode);
6480 		discard_new_inode(inode);
6481 	}
6482 	return err;
6483 }
6484 
6485 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6486 			umode_t mode, bool excl)
6487 {
6488 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6489 	struct btrfs_trans_handle *trans;
6490 	struct btrfs_root *root = BTRFS_I(dir)->root;
6491 	struct inode *inode = NULL;
6492 	int err;
6493 	u64 objectid;
6494 	u64 index = 0;
6495 
6496 	/*
6497 	 * 2 for inode item and ref
6498 	 * 2 for dir items
6499 	 * 1 for xattr if selinux is on
6500 	 */
6501 	trans = btrfs_start_transaction(root, 5);
6502 	if (IS_ERR(trans))
6503 		return PTR_ERR(trans);
6504 
6505 	err = btrfs_find_free_ino(root, &objectid);
6506 	if (err)
6507 		goto out_unlock;
6508 
6509 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6510 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6511 			mode, &index);
6512 	if (IS_ERR(inode)) {
6513 		err = PTR_ERR(inode);
6514 		inode = NULL;
6515 		goto out_unlock;
6516 	}
6517 	/*
6518 	* If the active LSM wants to access the inode during
6519 	* d_instantiate it needs these. Smack checks to see
6520 	* if the filesystem supports xattrs by looking at the
6521 	* ops vector.
6522 	*/
6523 	inode->i_fop = &btrfs_file_operations;
6524 	inode->i_op = &btrfs_file_inode_operations;
6525 	inode->i_mapping->a_ops = &btrfs_aops;
6526 
6527 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6528 	if (err)
6529 		goto out_unlock;
6530 
6531 	err = btrfs_update_inode(trans, root, inode);
6532 	if (err)
6533 		goto out_unlock;
6534 
6535 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6536 			0, index);
6537 	if (err)
6538 		goto out_unlock;
6539 
6540 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6541 	d_instantiate_new(dentry, inode);
6542 
6543 out_unlock:
6544 	btrfs_end_transaction(trans);
6545 	if (err && inode) {
6546 		inode_dec_link_count(inode);
6547 		discard_new_inode(inode);
6548 	}
6549 	btrfs_btree_balance_dirty(fs_info);
6550 	return err;
6551 }
6552 
6553 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6554 		      struct dentry *dentry)
6555 {
6556 	struct btrfs_trans_handle *trans = NULL;
6557 	struct btrfs_root *root = BTRFS_I(dir)->root;
6558 	struct inode *inode = d_inode(old_dentry);
6559 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6560 	u64 index;
6561 	int err;
6562 	int drop_inode = 0;
6563 
6564 	/* do not allow sys_link's with other subvols of the same device */
6565 	if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6566 		return -EXDEV;
6567 
6568 	if (inode->i_nlink >= BTRFS_LINK_MAX)
6569 		return -EMLINK;
6570 
6571 	err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6572 	if (err)
6573 		goto fail;
6574 
6575 	/*
6576 	 * 2 items for inode and inode ref
6577 	 * 2 items for dir items
6578 	 * 1 item for parent inode
6579 	 * 1 item for orphan item deletion if O_TMPFILE
6580 	 */
6581 	trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6582 	if (IS_ERR(trans)) {
6583 		err = PTR_ERR(trans);
6584 		trans = NULL;
6585 		goto fail;
6586 	}
6587 
6588 	/* There are several dir indexes for this inode, clear the cache. */
6589 	BTRFS_I(inode)->dir_index = 0ULL;
6590 	inc_nlink(inode);
6591 	inode_inc_iversion(inode);
6592 	inode->i_ctime = current_time(inode);
6593 	ihold(inode);
6594 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6595 
6596 	err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6597 			1, index);
6598 
6599 	if (err) {
6600 		drop_inode = 1;
6601 	} else {
6602 		struct dentry *parent = dentry->d_parent;
6603 		int ret;
6604 
6605 		err = btrfs_update_inode(trans, root, inode);
6606 		if (err)
6607 			goto fail;
6608 		if (inode->i_nlink == 1) {
6609 			/*
6610 			 * If new hard link count is 1, it's a file created
6611 			 * with open(2) O_TMPFILE flag.
6612 			 */
6613 			err = btrfs_orphan_del(trans, BTRFS_I(inode));
6614 			if (err)
6615 				goto fail;
6616 		}
6617 		d_instantiate(dentry, inode);
6618 		ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6619 					 true, NULL);
6620 		if (ret == BTRFS_NEED_TRANS_COMMIT) {
6621 			err = btrfs_commit_transaction(trans);
6622 			trans = NULL;
6623 		}
6624 	}
6625 
6626 fail:
6627 	if (trans)
6628 		btrfs_end_transaction(trans);
6629 	if (drop_inode) {
6630 		inode_dec_link_count(inode);
6631 		iput(inode);
6632 	}
6633 	btrfs_btree_balance_dirty(fs_info);
6634 	return err;
6635 }
6636 
6637 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6638 {
6639 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6640 	struct inode *inode = NULL;
6641 	struct btrfs_trans_handle *trans;
6642 	struct btrfs_root *root = BTRFS_I(dir)->root;
6643 	int err = 0;
6644 	int drop_on_err = 0;
6645 	u64 objectid = 0;
6646 	u64 index = 0;
6647 
6648 	/*
6649 	 * 2 items for inode and ref
6650 	 * 2 items for dir items
6651 	 * 1 for xattr if selinux is on
6652 	 */
6653 	trans = btrfs_start_transaction(root, 5);
6654 	if (IS_ERR(trans))
6655 		return PTR_ERR(trans);
6656 
6657 	err = btrfs_find_free_ino(root, &objectid);
6658 	if (err)
6659 		goto out_fail;
6660 
6661 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6662 			dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6663 			S_IFDIR | mode, &index);
6664 	if (IS_ERR(inode)) {
6665 		err = PTR_ERR(inode);
6666 		inode = NULL;
6667 		goto out_fail;
6668 	}
6669 
6670 	drop_on_err = 1;
6671 	/* these must be set before we unlock the inode */
6672 	inode->i_op = &btrfs_dir_inode_operations;
6673 	inode->i_fop = &btrfs_dir_file_operations;
6674 
6675 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6676 	if (err)
6677 		goto out_fail;
6678 
6679 	btrfs_i_size_write(BTRFS_I(inode), 0);
6680 	err = btrfs_update_inode(trans, root, inode);
6681 	if (err)
6682 		goto out_fail;
6683 
6684 	err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6685 			dentry->d_name.name,
6686 			dentry->d_name.len, 0, index);
6687 	if (err)
6688 		goto out_fail;
6689 
6690 	d_instantiate_new(dentry, inode);
6691 	drop_on_err = 0;
6692 
6693 out_fail:
6694 	btrfs_end_transaction(trans);
6695 	if (err && inode) {
6696 		inode_dec_link_count(inode);
6697 		discard_new_inode(inode);
6698 	}
6699 	btrfs_btree_balance_dirty(fs_info);
6700 	return err;
6701 }
6702 
6703 static noinline int uncompress_inline(struct btrfs_path *path,
6704 				      struct page *page,
6705 				      size_t pg_offset, u64 extent_offset,
6706 				      struct btrfs_file_extent_item *item)
6707 {
6708 	int ret;
6709 	struct extent_buffer *leaf = path->nodes[0];
6710 	char *tmp;
6711 	size_t max_size;
6712 	unsigned long inline_size;
6713 	unsigned long ptr;
6714 	int compress_type;
6715 
6716 	WARN_ON(pg_offset != 0);
6717 	compress_type = btrfs_file_extent_compression(leaf, item);
6718 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6719 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6720 					btrfs_item_nr(path->slots[0]));
6721 	tmp = kmalloc(inline_size, GFP_NOFS);
6722 	if (!tmp)
6723 		return -ENOMEM;
6724 	ptr = btrfs_file_extent_inline_start(item);
6725 
6726 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6727 
6728 	max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6729 	ret = btrfs_decompress(compress_type, tmp, page,
6730 			       extent_offset, inline_size, max_size);
6731 
6732 	/*
6733 	 * decompression code contains a memset to fill in any space between the end
6734 	 * of the uncompressed data and the end of max_size in case the decompressed
6735 	 * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6736 	 * the end of an inline extent and the beginning of the next block, so we
6737 	 * cover that region here.
6738 	 */
6739 
6740 	if (max_size + pg_offset < PAGE_SIZE) {
6741 		char *map = kmap(page);
6742 		memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6743 		kunmap(page);
6744 	}
6745 	kfree(tmp);
6746 	return ret;
6747 }
6748 
6749 /*
6750  * a bit scary, this does extent mapping from logical file offset to the disk.
6751  * the ugly parts come from merging extents from the disk with the in-ram
6752  * representation.  This gets more complex because of the data=ordered code,
6753  * where the in-ram extents might be locked pending data=ordered completion.
6754  *
6755  * This also copies inline extents directly into the page.
6756  */
6757 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6758 				    struct page *page,
6759 				    size_t pg_offset, u64 start, u64 len,
6760 				    int create)
6761 {
6762 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
6763 	int ret;
6764 	int err = 0;
6765 	u64 extent_start = 0;
6766 	u64 extent_end = 0;
6767 	u64 objectid = btrfs_ino(inode);
6768 	u32 found_type;
6769 	struct btrfs_path *path = NULL;
6770 	struct btrfs_root *root = inode->root;
6771 	struct btrfs_file_extent_item *item;
6772 	struct extent_buffer *leaf;
6773 	struct btrfs_key found_key;
6774 	struct extent_map *em = NULL;
6775 	struct extent_map_tree *em_tree = &inode->extent_tree;
6776 	struct extent_io_tree *io_tree = &inode->io_tree;
6777 	const bool new_inline = !page || create;
6778 
6779 	read_lock(&em_tree->lock);
6780 	em = lookup_extent_mapping(em_tree, start, len);
6781 	if (em)
6782 		em->bdev = fs_info->fs_devices->latest_bdev;
6783 	read_unlock(&em_tree->lock);
6784 
6785 	if (em) {
6786 		if (em->start > start || em->start + em->len <= start)
6787 			free_extent_map(em);
6788 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6789 			free_extent_map(em);
6790 		else
6791 			goto out;
6792 	}
6793 	em = alloc_extent_map();
6794 	if (!em) {
6795 		err = -ENOMEM;
6796 		goto out;
6797 	}
6798 	em->bdev = fs_info->fs_devices->latest_bdev;
6799 	em->start = EXTENT_MAP_HOLE;
6800 	em->orig_start = EXTENT_MAP_HOLE;
6801 	em->len = (u64)-1;
6802 	em->block_len = (u64)-1;
6803 
6804 	path = btrfs_alloc_path();
6805 	if (!path) {
6806 		err = -ENOMEM;
6807 		goto out;
6808 	}
6809 
6810 	/* Chances are we'll be called again, so go ahead and do readahead */
6811 	path->reada = READA_FORWARD;
6812 
6813 	/*
6814 	 * Unless we're going to uncompress the inline extent, no sleep would
6815 	 * happen.
6816 	 */
6817 	path->leave_spinning = 1;
6818 
6819 	ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6820 	if (ret < 0) {
6821 		err = ret;
6822 		goto out;
6823 	}
6824 
6825 	if (ret != 0) {
6826 		if (path->slots[0] == 0)
6827 			goto not_found;
6828 		path->slots[0]--;
6829 	}
6830 
6831 	leaf = path->nodes[0];
6832 	item = btrfs_item_ptr(leaf, path->slots[0],
6833 			      struct btrfs_file_extent_item);
6834 	/* are we inside the extent that was found? */
6835 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6836 	found_type = found_key.type;
6837 	if (found_key.objectid != objectid ||
6838 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6839 		/*
6840 		 * If we backup past the first extent we want to move forward
6841 		 * and see if there is an extent in front of us, otherwise we'll
6842 		 * say there is a hole for our whole search range which can
6843 		 * cause problems.
6844 		 */
6845 		extent_end = start;
6846 		goto next;
6847 	}
6848 
6849 	found_type = btrfs_file_extent_type(leaf, item);
6850 	extent_start = found_key.offset;
6851 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6852 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6853 		extent_end = extent_start +
6854 		       btrfs_file_extent_num_bytes(leaf, item);
6855 
6856 		trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6857 						       extent_start);
6858 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6859 		size_t size;
6860 
6861 		size = btrfs_file_extent_ram_bytes(leaf, item);
6862 		extent_end = ALIGN(extent_start + size,
6863 				   fs_info->sectorsize);
6864 
6865 		trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6866 						      path->slots[0],
6867 						      extent_start);
6868 	}
6869 next:
6870 	if (start >= extent_end) {
6871 		path->slots[0]++;
6872 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6873 			ret = btrfs_next_leaf(root, path);
6874 			if (ret < 0) {
6875 				err = ret;
6876 				goto out;
6877 			}
6878 			if (ret > 0)
6879 				goto not_found;
6880 			leaf = path->nodes[0];
6881 		}
6882 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6883 		if (found_key.objectid != objectid ||
6884 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6885 			goto not_found;
6886 		if (start + len <= found_key.offset)
6887 			goto not_found;
6888 		if (start > found_key.offset)
6889 			goto next;
6890 		em->start = start;
6891 		em->orig_start = start;
6892 		em->len = found_key.offset - start;
6893 		goto not_found_em;
6894 	}
6895 
6896 	btrfs_extent_item_to_extent_map(inode, path, item,
6897 			new_inline, em);
6898 
6899 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6900 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6901 		goto insert;
6902 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6903 		unsigned long ptr;
6904 		char *map;
6905 		size_t size;
6906 		size_t extent_offset;
6907 		size_t copy_size;
6908 
6909 		if (new_inline)
6910 			goto out;
6911 
6912 		size = btrfs_file_extent_ram_bytes(leaf, item);
6913 		extent_offset = page_offset(page) + pg_offset - extent_start;
6914 		copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6915 				  size - extent_offset);
6916 		em->start = extent_start + extent_offset;
6917 		em->len = ALIGN(copy_size, fs_info->sectorsize);
6918 		em->orig_block_len = em->len;
6919 		em->orig_start = em->start;
6920 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6921 
6922 		btrfs_set_path_blocking(path);
6923 		if (!PageUptodate(page)) {
6924 			if (btrfs_file_extent_compression(leaf, item) !=
6925 			    BTRFS_COMPRESS_NONE) {
6926 				ret = uncompress_inline(path, page, pg_offset,
6927 							extent_offset, item);
6928 				if (ret) {
6929 					err = ret;
6930 					goto out;
6931 				}
6932 			} else {
6933 				map = kmap(page);
6934 				read_extent_buffer(leaf, map + pg_offset, ptr,
6935 						   copy_size);
6936 				if (pg_offset + copy_size < PAGE_SIZE) {
6937 					memset(map + pg_offset + copy_size, 0,
6938 					       PAGE_SIZE - pg_offset -
6939 					       copy_size);
6940 				}
6941 				kunmap(page);
6942 			}
6943 			flush_dcache_page(page);
6944 		}
6945 		set_extent_uptodate(io_tree, em->start,
6946 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
6947 		goto insert;
6948 	}
6949 not_found:
6950 	em->start = start;
6951 	em->orig_start = start;
6952 	em->len = len;
6953 not_found_em:
6954 	em->block_start = EXTENT_MAP_HOLE;
6955 insert:
6956 	btrfs_release_path(path);
6957 	if (em->start > start || extent_map_end(em) <= start) {
6958 		btrfs_err(fs_info,
6959 			  "bad extent! em: [%llu %llu] passed [%llu %llu]",
6960 			  em->start, em->len, start, len);
6961 		err = -EIO;
6962 		goto out;
6963 	}
6964 
6965 	err = 0;
6966 	write_lock(&em_tree->lock);
6967 	err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6968 	write_unlock(&em_tree->lock);
6969 out:
6970 	btrfs_free_path(path);
6971 
6972 	trace_btrfs_get_extent(root, inode, em);
6973 
6974 	if (err) {
6975 		free_extent_map(em);
6976 		return ERR_PTR(err);
6977 	}
6978 	BUG_ON(!em); /* Error is always set */
6979 	return em;
6980 }
6981 
6982 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6983 		struct page *page,
6984 		size_t pg_offset, u64 start, u64 len,
6985 		int create)
6986 {
6987 	struct extent_map *em;
6988 	struct extent_map *hole_em = NULL;
6989 	u64 range_start = start;
6990 	u64 end;
6991 	u64 found;
6992 	u64 found_end;
6993 	int err = 0;
6994 
6995 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6996 	if (IS_ERR(em))
6997 		return em;
6998 	/*
6999 	 * If our em maps to:
7000 	 * - a hole or
7001 	 * - a pre-alloc extent,
7002 	 * there might actually be delalloc bytes behind it.
7003 	 */
7004 	if (em->block_start != EXTENT_MAP_HOLE &&
7005 	    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7006 		return em;
7007 	else
7008 		hole_em = em;
7009 
7010 	/* check to see if we've wrapped (len == -1 or similar) */
7011 	end = start + len;
7012 	if (end < start)
7013 		end = (u64)-1;
7014 	else
7015 		end -= 1;
7016 
7017 	em = NULL;
7018 
7019 	/* ok, we didn't find anything, lets look for delalloc */
7020 	found = count_range_bits(&inode->io_tree, &range_start,
7021 				 end, len, EXTENT_DELALLOC, 1);
7022 	found_end = range_start + found;
7023 	if (found_end < range_start)
7024 		found_end = (u64)-1;
7025 
7026 	/*
7027 	 * we didn't find anything useful, return
7028 	 * the original results from get_extent()
7029 	 */
7030 	if (range_start > end || found_end <= start) {
7031 		em = hole_em;
7032 		hole_em = NULL;
7033 		goto out;
7034 	}
7035 
7036 	/* adjust the range_start to make sure it doesn't
7037 	 * go backwards from the start they passed in
7038 	 */
7039 	range_start = max(start, range_start);
7040 	found = found_end - range_start;
7041 
7042 	if (found > 0) {
7043 		u64 hole_start = start;
7044 		u64 hole_len = len;
7045 
7046 		em = alloc_extent_map();
7047 		if (!em) {
7048 			err = -ENOMEM;
7049 			goto out;
7050 		}
7051 		/*
7052 		 * when btrfs_get_extent can't find anything it
7053 		 * returns one huge hole
7054 		 *
7055 		 * make sure what it found really fits our range, and
7056 		 * adjust to make sure it is based on the start from
7057 		 * the caller
7058 		 */
7059 		if (hole_em) {
7060 			u64 calc_end = extent_map_end(hole_em);
7061 
7062 			if (calc_end <= start || (hole_em->start > end)) {
7063 				free_extent_map(hole_em);
7064 				hole_em = NULL;
7065 			} else {
7066 				hole_start = max(hole_em->start, start);
7067 				hole_len = calc_end - hole_start;
7068 			}
7069 		}
7070 		em->bdev = NULL;
7071 		if (hole_em && range_start > hole_start) {
7072 			/* our hole starts before our delalloc, so we
7073 			 * have to return just the parts of the hole
7074 			 * that go until  the delalloc starts
7075 			 */
7076 			em->len = min(hole_len,
7077 				      range_start - hole_start);
7078 			em->start = hole_start;
7079 			em->orig_start = hole_start;
7080 			/*
7081 			 * don't adjust block start at all,
7082 			 * it is fixed at EXTENT_MAP_HOLE
7083 			 */
7084 			em->block_start = hole_em->block_start;
7085 			em->block_len = hole_len;
7086 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7087 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7088 		} else {
7089 			em->start = range_start;
7090 			em->len = found;
7091 			em->orig_start = range_start;
7092 			em->block_start = EXTENT_MAP_DELALLOC;
7093 			em->block_len = found;
7094 		}
7095 	} else {
7096 		return hole_em;
7097 	}
7098 out:
7099 
7100 	free_extent_map(hole_em);
7101 	if (err) {
7102 		free_extent_map(em);
7103 		return ERR_PTR(err);
7104 	}
7105 	return em;
7106 }
7107 
7108 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7109 						  const u64 start,
7110 						  const u64 len,
7111 						  const u64 orig_start,
7112 						  const u64 block_start,
7113 						  const u64 block_len,
7114 						  const u64 orig_block_len,
7115 						  const u64 ram_bytes,
7116 						  const int type)
7117 {
7118 	struct extent_map *em = NULL;
7119 	int ret;
7120 
7121 	if (type != BTRFS_ORDERED_NOCOW) {
7122 		em = create_io_em(inode, start, len, orig_start,
7123 				  block_start, block_len, orig_block_len,
7124 				  ram_bytes,
7125 				  BTRFS_COMPRESS_NONE, /* compress_type */
7126 				  type);
7127 		if (IS_ERR(em))
7128 			goto out;
7129 	}
7130 	ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7131 					   len, block_len, type);
7132 	if (ret) {
7133 		if (em) {
7134 			free_extent_map(em);
7135 			btrfs_drop_extent_cache(BTRFS_I(inode), start,
7136 						start + len - 1, 0);
7137 		}
7138 		em = ERR_PTR(ret);
7139 	}
7140  out:
7141 
7142 	return em;
7143 }
7144 
7145 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7146 						  u64 start, u64 len)
7147 {
7148 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7149 	struct btrfs_root *root = BTRFS_I(inode)->root;
7150 	struct extent_map *em;
7151 	struct btrfs_key ins;
7152 	u64 alloc_hint;
7153 	int ret;
7154 
7155 	alloc_hint = get_extent_allocation_hint(inode, start, len);
7156 	ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7157 				   0, alloc_hint, &ins, 1, 1);
7158 	if (ret)
7159 		return ERR_PTR(ret);
7160 
7161 	em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7162 				     ins.objectid, ins.offset, ins.offset,
7163 				     ins.offset, BTRFS_ORDERED_REGULAR);
7164 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7165 	if (IS_ERR(em))
7166 		btrfs_free_reserved_extent(fs_info, ins.objectid,
7167 					   ins.offset, 1);
7168 
7169 	return em;
7170 }
7171 
7172 /*
7173  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7174  * block must be cow'd
7175  */
7176 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7177 			      u64 *orig_start, u64 *orig_block_len,
7178 			      u64 *ram_bytes)
7179 {
7180 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7181 	struct btrfs_path *path;
7182 	int ret;
7183 	struct extent_buffer *leaf;
7184 	struct btrfs_root *root = BTRFS_I(inode)->root;
7185 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7186 	struct btrfs_file_extent_item *fi;
7187 	struct btrfs_key key;
7188 	u64 disk_bytenr;
7189 	u64 backref_offset;
7190 	u64 extent_end;
7191 	u64 num_bytes;
7192 	int slot;
7193 	int found_type;
7194 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7195 
7196 	path = btrfs_alloc_path();
7197 	if (!path)
7198 		return -ENOMEM;
7199 
7200 	ret = btrfs_lookup_file_extent(NULL, root, path,
7201 			btrfs_ino(BTRFS_I(inode)), offset, 0);
7202 	if (ret < 0)
7203 		goto out;
7204 
7205 	slot = path->slots[0];
7206 	if (ret == 1) {
7207 		if (slot == 0) {
7208 			/* can't find the item, must cow */
7209 			ret = 0;
7210 			goto out;
7211 		}
7212 		slot--;
7213 	}
7214 	ret = 0;
7215 	leaf = path->nodes[0];
7216 	btrfs_item_key_to_cpu(leaf, &key, slot);
7217 	if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7218 	    key.type != BTRFS_EXTENT_DATA_KEY) {
7219 		/* not our file or wrong item type, must cow */
7220 		goto out;
7221 	}
7222 
7223 	if (key.offset > offset) {
7224 		/* Wrong offset, must cow */
7225 		goto out;
7226 	}
7227 
7228 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7229 	found_type = btrfs_file_extent_type(leaf, fi);
7230 	if (found_type != BTRFS_FILE_EXTENT_REG &&
7231 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7232 		/* not a regular extent, must cow */
7233 		goto out;
7234 	}
7235 
7236 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7237 		goto out;
7238 
7239 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7240 	if (extent_end <= offset)
7241 		goto out;
7242 
7243 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7244 	if (disk_bytenr == 0)
7245 		goto out;
7246 
7247 	if (btrfs_file_extent_compression(leaf, fi) ||
7248 	    btrfs_file_extent_encryption(leaf, fi) ||
7249 	    btrfs_file_extent_other_encoding(leaf, fi))
7250 		goto out;
7251 
7252 	/*
7253 	 * Do the same check as in btrfs_cross_ref_exist but without the
7254 	 * unnecessary search.
7255 	 */
7256 	if (btrfs_file_extent_generation(leaf, fi) <=
7257 	    btrfs_root_last_snapshot(&root->root_item))
7258 		goto out;
7259 
7260 	backref_offset = btrfs_file_extent_offset(leaf, fi);
7261 
7262 	if (orig_start) {
7263 		*orig_start = key.offset - backref_offset;
7264 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7265 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7266 	}
7267 
7268 	if (btrfs_extent_readonly(fs_info, disk_bytenr))
7269 		goto out;
7270 
7271 	num_bytes = min(offset + *len, extent_end) - offset;
7272 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7273 		u64 range_end;
7274 
7275 		range_end = round_up(offset + num_bytes,
7276 				     root->fs_info->sectorsize) - 1;
7277 		ret = test_range_bit(io_tree, offset, range_end,
7278 				     EXTENT_DELALLOC, 0, NULL);
7279 		if (ret) {
7280 			ret = -EAGAIN;
7281 			goto out;
7282 		}
7283 	}
7284 
7285 	btrfs_release_path(path);
7286 
7287 	/*
7288 	 * look for other files referencing this extent, if we
7289 	 * find any we must cow
7290 	 */
7291 
7292 	ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7293 				    key.offset - backref_offset, disk_bytenr);
7294 	if (ret) {
7295 		ret = 0;
7296 		goto out;
7297 	}
7298 
7299 	/*
7300 	 * adjust disk_bytenr and num_bytes to cover just the bytes
7301 	 * in this extent we are about to write.  If there
7302 	 * are any csums in that range we have to cow in order
7303 	 * to keep the csums correct
7304 	 */
7305 	disk_bytenr += backref_offset;
7306 	disk_bytenr += offset - key.offset;
7307 	if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7308 		goto out;
7309 	/*
7310 	 * all of the above have passed, it is safe to overwrite this extent
7311 	 * without cow
7312 	 */
7313 	*len = num_bytes;
7314 	ret = 1;
7315 out:
7316 	btrfs_free_path(path);
7317 	return ret;
7318 }
7319 
7320 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7321 			      struct extent_state **cached_state, int writing)
7322 {
7323 	struct btrfs_ordered_extent *ordered;
7324 	int ret = 0;
7325 
7326 	while (1) {
7327 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7328 				 cached_state);
7329 		/*
7330 		 * We're concerned with the entire range that we're going to be
7331 		 * doing DIO to, so we need to make sure there's no ordered
7332 		 * extents in this range.
7333 		 */
7334 		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7335 						     lockend - lockstart + 1);
7336 
7337 		/*
7338 		 * We need to make sure there are no buffered pages in this
7339 		 * range either, we could have raced between the invalidate in
7340 		 * generic_file_direct_write and locking the extent.  The
7341 		 * invalidate needs to happen so that reads after a write do not
7342 		 * get stale data.
7343 		 */
7344 		if (!ordered &&
7345 		    (!writing || !filemap_range_has_page(inode->i_mapping,
7346 							 lockstart, lockend)))
7347 			break;
7348 
7349 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7350 				     cached_state);
7351 
7352 		if (ordered) {
7353 			/*
7354 			 * If we are doing a DIO read and the ordered extent we
7355 			 * found is for a buffered write, we can not wait for it
7356 			 * to complete and retry, because if we do so we can
7357 			 * deadlock with concurrent buffered writes on page
7358 			 * locks. This happens only if our DIO read covers more
7359 			 * than one extent map, if at this point has already
7360 			 * created an ordered extent for a previous extent map
7361 			 * and locked its range in the inode's io tree, and a
7362 			 * concurrent write against that previous extent map's
7363 			 * range and this range started (we unlock the ranges
7364 			 * in the io tree only when the bios complete and
7365 			 * buffered writes always lock pages before attempting
7366 			 * to lock range in the io tree).
7367 			 */
7368 			if (writing ||
7369 			    test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7370 				btrfs_start_ordered_extent(inode, ordered, 1);
7371 			else
7372 				ret = -ENOTBLK;
7373 			btrfs_put_ordered_extent(ordered);
7374 		} else {
7375 			/*
7376 			 * We could trigger writeback for this range (and wait
7377 			 * for it to complete) and then invalidate the pages for
7378 			 * this range (through invalidate_inode_pages2_range()),
7379 			 * but that can lead us to a deadlock with a concurrent
7380 			 * call to readpages() (a buffered read or a defrag call
7381 			 * triggered a readahead) on a page lock due to an
7382 			 * ordered dio extent we created before but did not have
7383 			 * yet a corresponding bio submitted (whence it can not
7384 			 * complete), which makes readpages() wait for that
7385 			 * ordered extent to complete while holding a lock on
7386 			 * that page.
7387 			 */
7388 			ret = -ENOTBLK;
7389 		}
7390 
7391 		if (ret)
7392 			break;
7393 
7394 		cond_resched();
7395 	}
7396 
7397 	return ret;
7398 }
7399 
7400 /* The callers of this must take lock_extent() */
7401 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7402 				       u64 orig_start, u64 block_start,
7403 				       u64 block_len, u64 orig_block_len,
7404 				       u64 ram_bytes, int compress_type,
7405 				       int type)
7406 {
7407 	struct extent_map_tree *em_tree;
7408 	struct extent_map *em;
7409 	struct btrfs_root *root = BTRFS_I(inode)->root;
7410 	int ret;
7411 
7412 	ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7413 	       type == BTRFS_ORDERED_COMPRESSED ||
7414 	       type == BTRFS_ORDERED_NOCOW ||
7415 	       type == BTRFS_ORDERED_REGULAR);
7416 
7417 	em_tree = &BTRFS_I(inode)->extent_tree;
7418 	em = alloc_extent_map();
7419 	if (!em)
7420 		return ERR_PTR(-ENOMEM);
7421 
7422 	em->start = start;
7423 	em->orig_start = orig_start;
7424 	em->len = len;
7425 	em->block_len = block_len;
7426 	em->block_start = block_start;
7427 	em->bdev = root->fs_info->fs_devices->latest_bdev;
7428 	em->orig_block_len = orig_block_len;
7429 	em->ram_bytes = ram_bytes;
7430 	em->generation = -1;
7431 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
7432 	if (type == BTRFS_ORDERED_PREALLOC) {
7433 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
7434 	} else if (type == BTRFS_ORDERED_COMPRESSED) {
7435 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7436 		em->compress_type = compress_type;
7437 	}
7438 
7439 	do {
7440 		btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7441 				em->start + em->len - 1, 0);
7442 		write_lock(&em_tree->lock);
7443 		ret = add_extent_mapping(em_tree, em, 1);
7444 		write_unlock(&em_tree->lock);
7445 		/*
7446 		 * The caller has taken lock_extent(), who could race with us
7447 		 * to add em?
7448 		 */
7449 	} while (ret == -EEXIST);
7450 
7451 	if (ret) {
7452 		free_extent_map(em);
7453 		return ERR_PTR(ret);
7454 	}
7455 
7456 	/* em got 2 refs now, callers needs to do free_extent_map once. */
7457 	return em;
7458 }
7459 
7460 
7461 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7462 					struct buffer_head *bh_result,
7463 					struct inode *inode,
7464 					u64 start, u64 len)
7465 {
7466 	if (em->block_start == EXTENT_MAP_HOLE ||
7467 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7468 		return -ENOENT;
7469 
7470 	len = min(len, em->len - (start - em->start));
7471 
7472 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7473 		inode->i_blkbits;
7474 	bh_result->b_size = len;
7475 	bh_result->b_bdev = em->bdev;
7476 	set_buffer_mapped(bh_result);
7477 
7478 	return 0;
7479 }
7480 
7481 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7482 					 struct buffer_head *bh_result,
7483 					 struct inode *inode,
7484 					 struct btrfs_dio_data *dio_data,
7485 					 u64 start, u64 len)
7486 {
7487 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7488 	struct extent_map *em = *map;
7489 	int ret = 0;
7490 
7491 	/*
7492 	 * We don't allocate a new extent in the following cases
7493 	 *
7494 	 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7495 	 * existing extent.
7496 	 * 2) The extent is marked as PREALLOC. We're good to go here and can
7497 	 * just use the extent.
7498 	 *
7499 	 */
7500 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7501 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7502 	     em->block_start != EXTENT_MAP_HOLE)) {
7503 		int type;
7504 		u64 block_start, orig_start, orig_block_len, ram_bytes;
7505 
7506 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7507 			type = BTRFS_ORDERED_PREALLOC;
7508 		else
7509 			type = BTRFS_ORDERED_NOCOW;
7510 		len = min(len, em->len - (start - em->start));
7511 		block_start = em->block_start + (start - em->start);
7512 
7513 		if (can_nocow_extent(inode, start, &len, &orig_start,
7514 				     &orig_block_len, &ram_bytes) == 1 &&
7515 		    btrfs_inc_nocow_writers(fs_info, block_start)) {
7516 			struct extent_map *em2;
7517 
7518 			em2 = btrfs_create_dio_extent(inode, start, len,
7519 						      orig_start, block_start,
7520 						      len, orig_block_len,
7521 						      ram_bytes, type);
7522 			btrfs_dec_nocow_writers(fs_info, block_start);
7523 			if (type == BTRFS_ORDERED_PREALLOC) {
7524 				free_extent_map(em);
7525 				*map = em = em2;
7526 			}
7527 
7528 			if (em2 && IS_ERR(em2)) {
7529 				ret = PTR_ERR(em2);
7530 				goto out;
7531 			}
7532 			/*
7533 			 * For inode marked NODATACOW or extent marked PREALLOC,
7534 			 * use the existing or preallocated extent, so does not
7535 			 * need to adjust btrfs_space_info's bytes_may_use.
7536 			 */
7537 			btrfs_free_reserved_data_space_noquota(inode, start,
7538 							       len);
7539 			goto skip_cow;
7540 		}
7541 	}
7542 
7543 	/* this will cow the extent */
7544 	len = bh_result->b_size;
7545 	free_extent_map(em);
7546 	*map = em = btrfs_new_extent_direct(inode, start, len);
7547 	if (IS_ERR(em)) {
7548 		ret = PTR_ERR(em);
7549 		goto out;
7550 	}
7551 
7552 	len = min(len, em->len - (start - em->start));
7553 
7554 skip_cow:
7555 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7556 		inode->i_blkbits;
7557 	bh_result->b_size = len;
7558 	bh_result->b_bdev = em->bdev;
7559 	set_buffer_mapped(bh_result);
7560 
7561 	if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7562 		set_buffer_new(bh_result);
7563 
7564 	/*
7565 	 * Need to update the i_size under the extent lock so buffered
7566 	 * readers will get the updated i_size when we unlock.
7567 	 */
7568 	if (!dio_data->overwrite && start + len > i_size_read(inode))
7569 		i_size_write(inode, start + len);
7570 
7571 	WARN_ON(dio_data->reserve < len);
7572 	dio_data->reserve -= len;
7573 	dio_data->unsubmitted_oe_range_end = start + len;
7574 	current->journal_info = dio_data;
7575 out:
7576 	return ret;
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 	if (create) {
7648 		ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7649 						    dio_data, start, len);
7650 		if (ret < 0)
7651 			goto unlock_err;
7652 
7653 		/* clear and unlock the entire range */
7654 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7655 				 unlock_bits, 1, 0, &cached_state);
7656 	} else {
7657 		ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7658 						   start, len);
7659 		/* Can be negative only if we read from a hole */
7660 		if (ret < 0) {
7661 			ret = 0;
7662 			free_extent_map(em);
7663 			goto unlock_err;
7664 		}
7665 		/*
7666 		 * We need to unlock only the end area that we aren't using.
7667 		 * The rest is going to be unlocked by the endio routine.
7668 		 */
7669 		lockstart = start + bh_result->b_size;
7670 		if (lockstart < lockend) {
7671 			clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7672 					 lockend, unlock_bits, 1, 0,
7673 					 &cached_state);
7674 		} else {
7675 			free_extent_state(cached_state);
7676 		}
7677 	}
7678 
7679 	free_extent_map(em);
7680 
7681 	return 0;
7682 
7683 unlock_err:
7684 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7685 			 unlock_bits, 1, 0, &cached_state);
7686 err:
7687 	if (dio_data)
7688 		current->journal_info = dio_data;
7689 	return ret;
7690 }
7691 
7692 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7693 						 struct bio *bio,
7694 						 int mirror_num)
7695 {
7696 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7697 	blk_status_t ret;
7698 
7699 	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7700 
7701 	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7702 	if (ret)
7703 		return ret;
7704 
7705 	ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7706 
7707 	return ret;
7708 }
7709 
7710 static int btrfs_check_dio_repairable(struct inode *inode,
7711 				      struct bio *failed_bio,
7712 				      struct io_failure_record *failrec,
7713 				      int failed_mirror)
7714 {
7715 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7716 	int num_copies;
7717 
7718 	num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7719 	if (num_copies == 1) {
7720 		/*
7721 		 * we only have a single copy of the data, so don't bother with
7722 		 * all the retry and error correction code that follows. no
7723 		 * matter what the error is, it is very likely to persist.
7724 		 */
7725 		btrfs_debug(fs_info,
7726 			"Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7727 			num_copies, failrec->this_mirror, failed_mirror);
7728 		return 0;
7729 	}
7730 
7731 	failrec->failed_mirror = failed_mirror;
7732 	failrec->this_mirror++;
7733 	if (failrec->this_mirror == failed_mirror)
7734 		failrec->this_mirror++;
7735 
7736 	if (failrec->this_mirror > num_copies) {
7737 		btrfs_debug(fs_info,
7738 			"Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7739 			num_copies, failrec->this_mirror, failed_mirror);
7740 		return 0;
7741 	}
7742 
7743 	return 1;
7744 }
7745 
7746 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7747 				   struct page *page, unsigned int pgoff,
7748 				   u64 start, u64 end, int failed_mirror,
7749 				   bio_end_io_t *repair_endio, void *repair_arg)
7750 {
7751 	struct io_failure_record *failrec;
7752 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7753 	struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7754 	struct bio *bio;
7755 	int isector;
7756 	unsigned int read_mode = 0;
7757 	int segs;
7758 	int ret;
7759 	blk_status_t status;
7760 	struct bio_vec bvec;
7761 
7762 	BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7763 
7764 	ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7765 	if (ret)
7766 		return errno_to_blk_status(ret);
7767 
7768 	ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7769 					 failed_mirror);
7770 	if (!ret) {
7771 		free_io_failure(failure_tree, io_tree, failrec);
7772 		return BLK_STS_IOERR;
7773 	}
7774 
7775 	segs = bio_segments(failed_bio);
7776 	bio_get_first_bvec(failed_bio, &bvec);
7777 	if (segs > 1 ||
7778 	    (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7779 		read_mode |= REQ_FAILFAST_DEV;
7780 
7781 	isector = start - btrfs_io_bio(failed_bio)->logical;
7782 	isector >>= inode->i_sb->s_blocksize_bits;
7783 	bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7784 				pgoff, isector, repair_endio, repair_arg);
7785 	bio->bi_opf = REQ_OP_READ | read_mode;
7786 
7787 	btrfs_debug(BTRFS_I(inode)->root->fs_info,
7788 		    "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7789 		    read_mode, failrec->this_mirror, failrec->in_validation);
7790 
7791 	status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7792 	if (status) {
7793 		free_io_failure(failure_tree, io_tree, failrec);
7794 		bio_put(bio);
7795 	}
7796 
7797 	return status;
7798 }
7799 
7800 struct btrfs_retry_complete {
7801 	struct completion done;
7802 	struct inode *inode;
7803 	u64 start;
7804 	int uptodate;
7805 };
7806 
7807 static void btrfs_retry_endio_nocsum(struct bio *bio)
7808 {
7809 	struct btrfs_retry_complete *done = bio->bi_private;
7810 	struct inode *inode = done->inode;
7811 	struct bio_vec *bvec;
7812 	struct extent_io_tree *io_tree, *failure_tree;
7813 	int i;
7814 
7815 	if (bio->bi_status)
7816 		goto end;
7817 
7818 	ASSERT(bio->bi_vcnt == 1);
7819 	io_tree = &BTRFS_I(inode)->io_tree;
7820 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
7821 	ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7822 
7823 	done->uptodate = 1;
7824 	ASSERT(!bio_flagged(bio, BIO_CLONED));
7825 	bio_for_each_segment_all(bvec, bio, i)
7826 		clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7827 				 io_tree, done->start, bvec->bv_page,
7828 				 btrfs_ino(BTRFS_I(inode)), 0);
7829 end:
7830 	complete(&done->done);
7831 	bio_put(bio);
7832 }
7833 
7834 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7835 						struct btrfs_io_bio *io_bio)
7836 {
7837 	struct btrfs_fs_info *fs_info;
7838 	struct bio_vec bvec;
7839 	struct bvec_iter iter;
7840 	struct btrfs_retry_complete done;
7841 	u64 start;
7842 	unsigned int pgoff;
7843 	u32 sectorsize;
7844 	int nr_sectors;
7845 	blk_status_t ret;
7846 	blk_status_t err = BLK_STS_OK;
7847 
7848 	fs_info = BTRFS_I(inode)->root->fs_info;
7849 	sectorsize = fs_info->sectorsize;
7850 
7851 	start = io_bio->logical;
7852 	done.inode = inode;
7853 	io_bio->bio.bi_iter = io_bio->iter;
7854 
7855 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
7856 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7857 		pgoff = bvec.bv_offset;
7858 
7859 next_block_or_try_again:
7860 		done.uptodate = 0;
7861 		done.start = start;
7862 		init_completion(&done.done);
7863 
7864 		ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7865 				pgoff, start, start + sectorsize - 1,
7866 				io_bio->mirror_num,
7867 				btrfs_retry_endio_nocsum, &done);
7868 		if (ret) {
7869 			err = ret;
7870 			goto next;
7871 		}
7872 
7873 		wait_for_completion_io(&done.done);
7874 
7875 		if (!done.uptodate) {
7876 			/* We might have another mirror, so try again */
7877 			goto next_block_or_try_again;
7878 		}
7879 
7880 next:
7881 		start += sectorsize;
7882 
7883 		nr_sectors--;
7884 		if (nr_sectors) {
7885 			pgoff += sectorsize;
7886 			ASSERT(pgoff < PAGE_SIZE);
7887 			goto next_block_or_try_again;
7888 		}
7889 	}
7890 
7891 	return err;
7892 }
7893 
7894 static void btrfs_retry_endio(struct bio *bio)
7895 {
7896 	struct btrfs_retry_complete *done = bio->bi_private;
7897 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7898 	struct extent_io_tree *io_tree, *failure_tree;
7899 	struct inode *inode = done->inode;
7900 	struct bio_vec *bvec;
7901 	int uptodate;
7902 	int ret;
7903 	int i;
7904 
7905 	if (bio->bi_status)
7906 		goto end;
7907 
7908 	uptodate = 1;
7909 
7910 	ASSERT(bio->bi_vcnt == 1);
7911 	ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7912 
7913 	io_tree = &BTRFS_I(inode)->io_tree;
7914 	failure_tree = &BTRFS_I(inode)->io_failure_tree;
7915 
7916 	ASSERT(!bio_flagged(bio, BIO_CLONED));
7917 	bio_for_each_segment_all(bvec, bio, i) {
7918 		ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7919 					     bvec->bv_offset, done->start,
7920 					     bvec->bv_len);
7921 		if (!ret)
7922 			clean_io_failure(BTRFS_I(inode)->root->fs_info,
7923 					 failure_tree, io_tree, done->start,
7924 					 bvec->bv_page,
7925 					 btrfs_ino(BTRFS_I(inode)),
7926 					 bvec->bv_offset);
7927 		else
7928 			uptodate = 0;
7929 	}
7930 
7931 	done->uptodate = uptodate;
7932 end:
7933 	complete(&done->done);
7934 	bio_put(bio);
7935 }
7936 
7937 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7938 		struct btrfs_io_bio *io_bio, blk_status_t err)
7939 {
7940 	struct btrfs_fs_info *fs_info;
7941 	struct bio_vec bvec;
7942 	struct bvec_iter iter;
7943 	struct btrfs_retry_complete done;
7944 	u64 start;
7945 	u64 offset = 0;
7946 	u32 sectorsize;
7947 	int nr_sectors;
7948 	unsigned int pgoff;
7949 	int csum_pos;
7950 	bool uptodate = (err == 0);
7951 	int ret;
7952 	blk_status_t status;
7953 
7954 	fs_info = BTRFS_I(inode)->root->fs_info;
7955 	sectorsize = fs_info->sectorsize;
7956 
7957 	err = BLK_STS_OK;
7958 	start = io_bio->logical;
7959 	done.inode = inode;
7960 	io_bio->bio.bi_iter = io_bio->iter;
7961 
7962 	bio_for_each_segment(bvec, &io_bio->bio, iter) {
7963 		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7964 
7965 		pgoff = bvec.bv_offset;
7966 next_block:
7967 		if (uptodate) {
7968 			csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7969 			ret = __readpage_endio_check(inode, io_bio, csum_pos,
7970 					bvec.bv_page, pgoff, start, sectorsize);
7971 			if (likely(!ret))
7972 				goto next;
7973 		}
7974 try_again:
7975 		done.uptodate = 0;
7976 		done.start = start;
7977 		init_completion(&done.done);
7978 
7979 		status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7980 					pgoff, start, start + sectorsize - 1,
7981 					io_bio->mirror_num, btrfs_retry_endio,
7982 					&done);
7983 		if (status) {
7984 			err = status;
7985 			goto next;
7986 		}
7987 
7988 		wait_for_completion_io(&done.done);
7989 
7990 		if (!done.uptodate) {
7991 			/* We might have another mirror, so try again */
7992 			goto try_again;
7993 		}
7994 next:
7995 		offset += sectorsize;
7996 		start += sectorsize;
7997 
7998 		ASSERT(nr_sectors);
7999 
8000 		nr_sectors--;
8001 		if (nr_sectors) {
8002 			pgoff += sectorsize;
8003 			ASSERT(pgoff < PAGE_SIZE);
8004 			goto next_block;
8005 		}
8006 	}
8007 
8008 	return err;
8009 }
8010 
8011 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8012 		struct btrfs_io_bio *io_bio, blk_status_t err)
8013 {
8014 	bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8015 
8016 	if (skip_csum) {
8017 		if (unlikely(err))
8018 			return __btrfs_correct_data_nocsum(inode, io_bio);
8019 		else
8020 			return BLK_STS_OK;
8021 	} else {
8022 		return __btrfs_subio_endio_read(inode, io_bio, err);
8023 	}
8024 }
8025 
8026 static void btrfs_endio_direct_read(struct bio *bio)
8027 {
8028 	struct btrfs_dio_private *dip = bio->bi_private;
8029 	struct inode *inode = dip->inode;
8030 	struct bio *dio_bio;
8031 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8032 	blk_status_t err = bio->bi_status;
8033 
8034 	if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8035 		err = btrfs_subio_endio_read(inode, io_bio, err);
8036 
8037 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8038 		      dip->logical_offset + dip->bytes - 1);
8039 	dio_bio = dip->dio_bio;
8040 
8041 	kfree(dip);
8042 
8043 	dio_bio->bi_status = err;
8044 	dio_end_io(dio_bio);
8045 
8046 	if (io_bio->end_io)
8047 		io_bio->end_io(io_bio, blk_status_to_errno(err));
8048 	bio_put(bio);
8049 }
8050 
8051 static void __endio_write_update_ordered(struct inode *inode,
8052 					 const u64 offset, const u64 bytes,
8053 					 const bool uptodate)
8054 {
8055 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8056 	struct btrfs_ordered_extent *ordered = NULL;
8057 	struct btrfs_workqueue *wq;
8058 	btrfs_work_func_t func;
8059 	u64 ordered_offset = offset;
8060 	u64 ordered_bytes = bytes;
8061 	u64 last_offset;
8062 
8063 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8064 		wq = fs_info->endio_freespace_worker;
8065 		func = btrfs_freespace_write_helper;
8066 	} else {
8067 		wq = fs_info->endio_write_workers;
8068 		func = btrfs_endio_write_helper;
8069 	}
8070 
8071 	while (ordered_offset < offset + bytes) {
8072 		last_offset = ordered_offset;
8073 		if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8074 							   &ordered_offset,
8075 							   ordered_bytes,
8076 							   uptodate)) {
8077 			btrfs_init_work(&ordered->work, func,
8078 					finish_ordered_fn,
8079 					NULL, NULL);
8080 			btrfs_queue_work(wq, &ordered->work);
8081 		}
8082 		/*
8083 		 * If btrfs_dec_test_ordered_pending does not find any ordered
8084 		 * extent in the range, we can exit.
8085 		 */
8086 		if (ordered_offset == last_offset)
8087 			return;
8088 		/*
8089 		 * Our bio might span multiple ordered extents. In this case
8090 		 * we keep goin until we have accounted the whole dio.
8091 		 */
8092 		if (ordered_offset < offset + bytes) {
8093 			ordered_bytes = offset + bytes - ordered_offset;
8094 			ordered = NULL;
8095 		}
8096 	}
8097 }
8098 
8099 static void btrfs_endio_direct_write(struct bio *bio)
8100 {
8101 	struct btrfs_dio_private *dip = bio->bi_private;
8102 	struct bio *dio_bio = dip->dio_bio;
8103 
8104 	__endio_write_update_ordered(dip->inode, dip->logical_offset,
8105 				     dip->bytes, !bio->bi_status);
8106 
8107 	kfree(dip);
8108 
8109 	dio_bio->bi_status = bio->bi_status;
8110 	dio_end_io(dio_bio);
8111 	bio_put(bio);
8112 }
8113 
8114 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8115 				    struct bio *bio, u64 offset)
8116 {
8117 	struct inode *inode = private_data;
8118 	blk_status_t ret;
8119 	ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8120 	BUG_ON(ret); /* -ENOMEM */
8121 	return 0;
8122 }
8123 
8124 static void btrfs_end_dio_bio(struct bio *bio)
8125 {
8126 	struct btrfs_dio_private *dip = bio->bi_private;
8127 	blk_status_t err = bio->bi_status;
8128 
8129 	if (err)
8130 		btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8131 			   "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8132 			   btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8133 			   bio->bi_opf,
8134 			   (unsigned long long)bio->bi_iter.bi_sector,
8135 			   bio->bi_iter.bi_size, err);
8136 
8137 	if (dip->subio_endio)
8138 		err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8139 
8140 	if (err) {
8141 		/*
8142 		 * We want to perceive the errors flag being set before
8143 		 * decrementing the reference count. We don't need a barrier
8144 		 * since atomic operations with a return value are fully
8145 		 * ordered as per atomic_t.txt
8146 		 */
8147 		dip->errors = 1;
8148 	}
8149 
8150 	/* if there are more bios still pending for this dio, just exit */
8151 	if (!atomic_dec_and_test(&dip->pending_bios))
8152 		goto out;
8153 
8154 	if (dip->errors) {
8155 		bio_io_error(dip->orig_bio);
8156 	} else {
8157 		dip->dio_bio->bi_status = BLK_STS_OK;
8158 		bio_endio(dip->orig_bio);
8159 	}
8160 out:
8161 	bio_put(bio);
8162 }
8163 
8164 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8165 						 struct btrfs_dio_private *dip,
8166 						 struct bio *bio,
8167 						 u64 file_offset)
8168 {
8169 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8170 	struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8171 	blk_status_t ret;
8172 
8173 	/*
8174 	 * We load all the csum data we need when we submit
8175 	 * the first bio to reduce the csum tree search and
8176 	 * contention.
8177 	 */
8178 	if (dip->logical_offset == file_offset) {
8179 		ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8180 						file_offset);
8181 		if (ret)
8182 			return ret;
8183 	}
8184 
8185 	if (bio == dip->orig_bio)
8186 		return 0;
8187 
8188 	file_offset -= dip->logical_offset;
8189 	file_offset >>= inode->i_sb->s_blocksize_bits;
8190 	io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8191 
8192 	return 0;
8193 }
8194 
8195 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8196 		struct inode *inode, u64 file_offset, int async_submit)
8197 {
8198 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8199 	struct btrfs_dio_private *dip = bio->bi_private;
8200 	bool write = bio_op(bio) == REQ_OP_WRITE;
8201 	blk_status_t ret;
8202 
8203 	/* Check btrfs_submit_bio_hook() for rules about async submit. */
8204 	if (async_submit)
8205 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8206 
8207 	if (!write) {
8208 		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8209 		if (ret)
8210 			goto err;
8211 	}
8212 
8213 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8214 		goto map;
8215 
8216 	if (write && async_submit) {
8217 		ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8218 					  file_offset, inode,
8219 					  btrfs_submit_bio_start_direct_io);
8220 		goto err;
8221 	} else if (write) {
8222 		/*
8223 		 * If we aren't doing async submit, calculate the csum of the
8224 		 * bio now.
8225 		 */
8226 		ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8227 		if (ret)
8228 			goto err;
8229 	} else {
8230 		ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8231 						     file_offset);
8232 		if (ret)
8233 			goto err;
8234 	}
8235 map:
8236 	ret = btrfs_map_bio(fs_info, bio, 0, 0);
8237 err:
8238 	return ret;
8239 }
8240 
8241 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8242 {
8243 	struct inode *inode = dip->inode;
8244 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8245 	struct bio *bio;
8246 	struct bio *orig_bio = dip->orig_bio;
8247 	u64 start_sector = orig_bio->bi_iter.bi_sector;
8248 	u64 file_offset = dip->logical_offset;
8249 	u64 map_length;
8250 	int async_submit = 0;
8251 	u64 submit_len;
8252 	int clone_offset = 0;
8253 	int clone_len;
8254 	int ret;
8255 	blk_status_t status;
8256 
8257 	map_length = orig_bio->bi_iter.bi_size;
8258 	submit_len = map_length;
8259 	ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8260 			      &map_length, NULL, 0);
8261 	if (ret)
8262 		return -EIO;
8263 
8264 	if (map_length >= submit_len) {
8265 		bio = orig_bio;
8266 		dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8267 		goto submit;
8268 	}
8269 
8270 	/* async crcs make it difficult to collect full stripe writes. */
8271 	if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8272 		async_submit = 0;
8273 	else
8274 		async_submit = 1;
8275 
8276 	/* bio split */
8277 	ASSERT(map_length <= INT_MAX);
8278 	atomic_inc(&dip->pending_bios);
8279 	do {
8280 		clone_len = min_t(int, submit_len, map_length);
8281 
8282 		/*
8283 		 * This will never fail as it's passing GPF_NOFS and
8284 		 * the allocation is backed by btrfs_bioset.
8285 		 */
8286 		bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8287 					      clone_len);
8288 		bio->bi_private = dip;
8289 		bio->bi_end_io = btrfs_end_dio_bio;
8290 		btrfs_io_bio(bio)->logical = file_offset;
8291 
8292 		ASSERT(submit_len >= clone_len);
8293 		submit_len -= clone_len;
8294 		if (submit_len == 0)
8295 			break;
8296 
8297 		/*
8298 		 * Increase the count before we submit the bio so we know
8299 		 * the end IO handler won't happen before we increase the
8300 		 * count. Otherwise, the dip might get freed before we're
8301 		 * done setting it up.
8302 		 */
8303 		atomic_inc(&dip->pending_bios);
8304 
8305 		status = btrfs_submit_dio_bio(bio, inode, file_offset,
8306 						async_submit);
8307 		if (status) {
8308 			bio_put(bio);
8309 			atomic_dec(&dip->pending_bios);
8310 			goto out_err;
8311 		}
8312 
8313 		clone_offset += clone_len;
8314 		start_sector += clone_len >> 9;
8315 		file_offset += clone_len;
8316 
8317 		map_length = submit_len;
8318 		ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8319 				      start_sector << 9, &map_length, NULL, 0);
8320 		if (ret)
8321 			goto out_err;
8322 	} while (submit_len > 0);
8323 
8324 submit:
8325 	status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8326 	if (!status)
8327 		return 0;
8328 
8329 	bio_put(bio);
8330 out_err:
8331 	dip->errors = 1;
8332 	/*
8333 	 * Before atomic variable goto zero, we must  make sure dip->errors is
8334 	 * perceived to be set. This ordering is ensured by the fact that an
8335 	 * atomic operations with a return value are fully ordered as per
8336 	 * atomic_t.txt
8337 	 */
8338 	if (atomic_dec_and_test(&dip->pending_bios))
8339 		bio_io_error(dip->orig_bio);
8340 
8341 	/* bio_end_io() will handle error, so we needn't return it */
8342 	return 0;
8343 }
8344 
8345 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8346 				loff_t file_offset)
8347 {
8348 	struct btrfs_dio_private *dip = NULL;
8349 	struct bio *bio = NULL;
8350 	struct btrfs_io_bio *io_bio;
8351 	bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8352 	int ret = 0;
8353 
8354 	bio = btrfs_bio_clone(dio_bio);
8355 
8356 	dip = kzalloc(sizeof(*dip), GFP_NOFS);
8357 	if (!dip) {
8358 		ret = -ENOMEM;
8359 		goto free_ordered;
8360 	}
8361 
8362 	dip->private = dio_bio->bi_private;
8363 	dip->inode = inode;
8364 	dip->logical_offset = file_offset;
8365 	dip->bytes = dio_bio->bi_iter.bi_size;
8366 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8367 	bio->bi_private = dip;
8368 	dip->orig_bio = bio;
8369 	dip->dio_bio = dio_bio;
8370 	atomic_set(&dip->pending_bios, 0);
8371 	io_bio = btrfs_io_bio(bio);
8372 	io_bio->logical = file_offset;
8373 
8374 	if (write) {
8375 		bio->bi_end_io = btrfs_endio_direct_write;
8376 	} else {
8377 		bio->bi_end_io = btrfs_endio_direct_read;
8378 		dip->subio_endio = btrfs_subio_endio_read;
8379 	}
8380 
8381 	/*
8382 	 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8383 	 * even if we fail to submit a bio, because in such case we do the
8384 	 * corresponding error handling below and it must not be done a second
8385 	 * time by btrfs_direct_IO().
8386 	 */
8387 	if (write) {
8388 		struct btrfs_dio_data *dio_data = current->journal_info;
8389 
8390 		dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8391 			dip->bytes;
8392 		dio_data->unsubmitted_oe_range_start =
8393 			dio_data->unsubmitted_oe_range_end;
8394 	}
8395 
8396 	ret = btrfs_submit_direct_hook(dip);
8397 	if (!ret)
8398 		return;
8399 
8400 	if (io_bio->end_io)
8401 		io_bio->end_io(io_bio, ret);
8402 
8403 free_ordered:
8404 	/*
8405 	 * If we arrived here it means either we failed to submit the dip
8406 	 * or we either failed to clone the dio_bio or failed to allocate the
8407 	 * dip. If we cloned the dio_bio and allocated the dip, we can just
8408 	 * call bio_endio against our io_bio so that we get proper resource
8409 	 * cleanup if we fail to submit the dip, otherwise, we must do the
8410 	 * same as btrfs_endio_direct_[write|read] because we can't call these
8411 	 * callbacks - they require an allocated dip and a clone of dio_bio.
8412 	 */
8413 	if (bio && dip) {
8414 		bio_io_error(bio);
8415 		/*
8416 		 * The end io callbacks free our dip, do the final put on bio
8417 		 * and all the cleanup and final put for dio_bio (through
8418 		 * dio_end_io()).
8419 		 */
8420 		dip = NULL;
8421 		bio = NULL;
8422 	} else {
8423 		if (write)
8424 			__endio_write_update_ordered(inode,
8425 						file_offset,
8426 						dio_bio->bi_iter.bi_size,
8427 						false);
8428 		else
8429 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8430 			      file_offset + dio_bio->bi_iter.bi_size - 1);
8431 
8432 		dio_bio->bi_status = BLK_STS_IOERR;
8433 		/*
8434 		 * Releases and cleans up our dio_bio, no need to bio_put()
8435 		 * nor bio_endio()/bio_io_error() against dio_bio.
8436 		 */
8437 		dio_end_io(dio_bio);
8438 	}
8439 	if (bio)
8440 		bio_put(bio);
8441 	kfree(dip);
8442 }
8443 
8444 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8445 			       const struct iov_iter *iter, loff_t offset)
8446 {
8447 	int seg;
8448 	int i;
8449 	unsigned int blocksize_mask = fs_info->sectorsize - 1;
8450 	ssize_t retval = -EINVAL;
8451 
8452 	if (offset & blocksize_mask)
8453 		goto out;
8454 
8455 	if (iov_iter_alignment(iter) & blocksize_mask)
8456 		goto out;
8457 
8458 	/* If this is a write we don't need to check anymore */
8459 	if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8460 		return 0;
8461 	/*
8462 	 * Check to make sure we don't have duplicate iov_base's in this
8463 	 * iovec, if so return EINVAL, otherwise we'll get csum errors
8464 	 * when reading back.
8465 	 */
8466 	for (seg = 0; seg < iter->nr_segs; seg++) {
8467 		for (i = seg + 1; i < iter->nr_segs; i++) {
8468 			if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8469 				goto out;
8470 		}
8471 	}
8472 	retval = 0;
8473 out:
8474 	return retval;
8475 }
8476 
8477 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8478 {
8479 	struct file *file = iocb->ki_filp;
8480 	struct inode *inode = file->f_mapping->host;
8481 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8482 	struct btrfs_dio_data dio_data = { 0 };
8483 	struct extent_changeset *data_reserved = NULL;
8484 	loff_t offset = iocb->ki_pos;
8485 	size_t count = 0;
8486 	int flags = 0;
8487 	bool wakeup = true;
8488 	bool relock = false;
8489 	ssize_t ret;
8490 
8491 	if (check_direct_IO(fs_info, iter, offset))
8492 		return 0;
8493 
8494 	inode_dio_begin(inode);
8495 
8496 	/*
8497 	 * The generic stuff only does filemap_write_and_wait_range, which
8498 	 * isn't enough if we've written compressed pages to this area, so
8499 	 * we need to flush the dirty pages again to make absolutely sure
8500 	 * that any outstanding dirty pages are on disk.
8501 	 */
8502 	count = iov_iter_count(iter);
8503 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8504 		     &BTRFS_I(inode)->runtime_flags))
8505 		filemap_fdatawrite_range(inode->i_mapping, offset,
8506 					 offset + count - 1);
8507 
8508 	if (iov_iter_rw(iter) == WRITE) {
8509 		/*
8510 		 * If the write DIO is beyond the EOF, we need update
8511 		 * the isize, but it is protected by i_mutex. So we can
8512 		 * not unlock the i_mutex at this case.
8513 		 */
8514 		if (offset + count <= inode->i_size) {
8515 			dio_data.overwrite = 1;
8516 			inode_unlock(inode);
8517 			relock = true;
8518 		} else if (iocb->ki_flags & IOCB_NOWAIT) {
8519 			ret = -EAGAIN;
8520 			goto out;
8521 		}
8522 		ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8523 						   offset, count);
8524 		if (ret)
8525 			goto out;
8526 
8527 		/*
8528 		 * We need to know how many extents we reserved so that we can
8529 		 * do the accounting properly if we go over the number we
8530 		 * originally calculated.  Abuse current->journal_info for this.
8531 		 */
8532 		dio_data.reserve = round_up(count,
8533 					    fs_info->sectorsize);
8534 		dio_data.unsubmitted_oe_range_start = (u64)offset;
8535 		dio_data.unsubmitted_oe_range_end = (u64)offset;
8536 		current->journal_info = &dio_data;
8537 		down_read(&BTRFS_I(inode)->dio_sem);
8538 	} else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8539 				     &BTRFS_I(inode)->runtime_flags)) {
8540 		inode_dio_end(inode);
8541 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
8542 		wakeup = false;
8543 	}
8544 
8545 	ret = __blockdev_direct_IO(iocb, inode,
8546 				   fs_info->fs_devices->latest_bdev,
8547 				   iter, btrfs_get_blocks_direct, NULL,
8548 				   btrfs_submit_direct, flags);
8549 	if (iov_iter_rw(iter) == WRITE) {
8550 		up_read(&BTRFS_I(inode)->dio_sem);
8551 		current->journal_info = NULL;
8552 		if (ret < 0 && ret != -EIOCBQUEUED) {
8553 			if (dio_data.reserve)
8554 				btrfs_delalloc_release_space(inode, data_reserved,
8555 					offset, dio_data.reserve, true);
8556 			/*
8557 			 * On error we might have left some ordered extents
8558 			 * without submitting corresponding bios for them, so
8559 			 * cleanup them up to avoid other tasks getting them
8560 			 * and waiting for them to complete forever.
8561 			 */
8562 			if (dio_data.unsubmitted_oe_range_start <
8563 			    dio_data.unsubmitted_oe_range_end)
8564 				__endio_write_update_ordered(inode,
8565 					dio_data.unsubmitted_oe_range_start,
8566 					dio_data.unsubmitted_oe_range_end -
8567 					dio_data.unsubmitted_oe_range_start,
8568 					false);
8569 		} else if (ret >= 0 && (size_t)ret < count)
8570 			btrfs_delalloc_release_space(inode, data_reserved,
8571 					offset, count - (size_t)ret, true);
8572 		btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8573 	}
8574 out:
8575 	if (wakeup)
8576 		inode_dio_end(inode);
8577 	if (relock)
8578 		inode_lock(inode);
8579 
8580 	extent_changeset_free(data_reserved);
8581 	return ret;
8582 }
8583 
8584 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
8585 
8586 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8587 		__u64 start, __u64 len)
8588 {
8589 	int	ret;
8590 
8591 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8592 	if (ret)
8593 		return ret;
8594 
8595 	return extent_fiemap(inode, fieinfo, start, len);
8596 }
8597 
8598 int btrfs_readpage(struct file *file, struct page *page)
8599 {
8600 	struct extent_io_tree *tree;
8601 	tree = &BTRFS_I(page->mapping->host)->io_tree;
8602 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8603 }
8604 
8605 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8606 {
8607 	struct inode *inode = page->mapping->host;
8608 	int ret;
8609 
8610 	if (current->flags & PF_MEMALLOC) {
8611 		redirty_page_for_writepage(wbc, page);
8612 		unlock_page(page);
8613 		return 0;
8614 	}
8615 
8616 	/*
8617 	 * If we are under memory pressure we will call this directly from the
8618 	 * VM, we need to make sure we have the inode referenced for the ordered
8619 	 * extent.  If not just return like we didn't do anything.
8620 	 */
8621 	if (!igrab(inode)) {
8622 		redirty_page_for_writepage(wbc, page);
8623 		return AOP_WRITEPAGE_ACTIVATE;
8624 	}
8625 	ret = extent_write_full_page(page, wbc);
8626 	btrfs_add_delayed_iput(inode);
8627 	return ret;
8628 }
8629 
8630 static int btrfs_writepages(struct address_space *mapping,
8631 			    struct writeback_control *wbc)
8632 {
8633 	return extent_writepages(mapping, wbc);
8634 }
8635 
8636 static int
8637 btrfs_readpages(struct file *file, struct address_space *mapping,
8638 		struct list_head *pages, unsigned nr_pages)
8639 {
8640 	return extent_readpages(mapping, pages, nr_pages);
8641 }
8642 
8643 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8644 {
8645 	int ret = try_release_extent_mapping(page, gfp_flags);
8646 	if (ret == 1) {
8647 		ClearPagePrivate(page);
8648 		set_page_private(page, 0);
8649 		put_page(page);
8650 	}
8651 	return ret;
8652 }
8653 
8654 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8655 {
8656 	if (PageWriteback(page) || PageDirty(page))
8657 		return 0;
8658 	return __btrfs_releasepage(page, gfp_flags);
8659 }
8660 
8661 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8662 				 unsigned int length)
8663 {
8664 	struct inode *inode = page->mapping->host;
8665 	struct extent_io_tree *tree;
8666 	struct btrfs_ordered_extent *ordered;
8667 	struct extent_state *cached_state = NULL;
8668 	u64 page_start = page_offset(page);
8669 	u64 page_end = page_start + PAGE_SIZE - 1;
8670 	u64 start;
8671 	u64 end;
8672 	int inode_evicting = inode->i_state & I_FREEING;
8673 
8674 	/*
8675 	 * we have the page locked, so new writeback can't start,
8676 	 * and the dirty bit won't be cleared while we are here.
8677 	 *
8678 	 * Wait for IO on this page so that we can safely clear
8679 	 * the PagePrivate2 bit and do ordered accounting
8680 	 */
8681 	wait_on_page_writeback(page);
8682 
8683 	tree = &BTRFS_I(inode)->io_tree;
8684 	if (offset) {
8685 		btrfs_releasepage(page, GFP_NOFS);
8686 		return;
8687 	}
8688 
8689 	if (!inode_evicting)
8690 		lock_extent_bits(tree, page_start, page_end, &cached_state);
8691 again:
8692 	start = page_start;
8693 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8694 					page_end - start + 1);
8695 	if (ordered) {
8696 		end = min(page_end, ordered->file_offset + ordered->len - 1);
8697 		/*
8698 		 * IO on this page will never be started, so we need
8699 		 * to account for any ordered extents now
8700 		 */
8701 		if (!inode_evicting)
8702 			clear_extent_bit(tree, start, end,
8703 					 EXTENT_DIRTY | EXTENT_DELALLOC |
8704 					 EXTENT_DELALLOC_NEW |
8705 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8706 					 EXTENT_DEFRAG, 1, 0, &cached_state);
8707 		/*
8708 		 * whoever cleared the private bit is responsible
8709 		 * for the finish_ordered_io
8710 		 */
8711 		if (TestClearPagePrivate2(page)) {
8712 			struct btrfs_ordered_inode_tree *tree;
8713 			u64 new_len;
8714 
8715 			tree = &BTRFS_I(inode)->ordered_tree;
8716 
8717 			spin_lock_irq(&tree->lock);
8718 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8719 			new_len = start - ordered->file_offset;
8720 			if (new_len < ordered->truncated_len)
8721 				ordered->truncated_len = new_len;
8722 			spin_unlock_irq(&tree->lock);
8723 
8724 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
8725 							   start,
8726 							   end - start + 1, 1))
8727 				btrfs_finish_ordered_io(ordered);
8728 		}
8729 		btrfs_put_ordered_extent(ordered);
8730 		if (!inode_evicting) {
8731 			cached_state = NULL;
8732 			lock_extent_bits(tree, start, end,
8733 					 &cached_state);
8734 		}
8735 
8736 		start = end + 1;
8737 		if (start < page_end)
8738 			goto again;
8739 	}
8740 
8741 	/*
8742 	 * Qgroup reserved space handler
8743 	 * Page here will be either
8744 	 * 1) Already written to disk
8745 	 *    In this case, its reserved space is released from data rsv map
8746 	 *    and will be freed by delayed_ref handler finally.
8747 	 *    So even we call qgroup_free_data(), it won't decrease reserved
8748 	 *    space.
8749 	 * 2) Not written to disk
8750 	 *    This means the reserved space should be freed here. However,
8751 	 *    if a truncate invalidates the page (by clearing PageDirty)
8752 	 *    and the page is accounted for while allocating extent
8753 	 *    in btrfs_check_data_free_space() we let delayed_ref to
8754 	 *    free the entire extent.
8755 	 */
8756 	if (PageDirty(page))
8757 		btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8758 	if (!inode_evicting) {
8759 		clear_extent_bit(tree, page_start, page_end,
8760 				 EXTENT_LOCKED | EXTENT_DIRTY |
8761 				 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8762 				 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8763 				 &cached_state);
8764 
8765 		__btrfs_releasepage(page, GFP_NOFS);
8766 	}
8767 
8768 	ClearPageChecked(page);
8769 	if (PagePrivate(page)) {
8770 		ClearPagePrivate(page);
8771 		set_page_private(page, 0);
8772 		put_page(page);
8773 	}
8774 }
8775 
8776 /*
8777  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8778  * called from a page fault handler when a page is first dirtied. Hence we must
8779  * be careful to check for EOF conditions here. We set the page up correctly
8780  * for a written page which means we get ENOSPC checking when writing into
8781  * holes and correct delalloc and unwritten extent mapping on filesystems that
8782  * support these features.
8783  *
8784  * We are not allowed to take the i_mutex here so we have to play games to
8785  * protect against truncate races as the page could now be beyond EOF.  Because
8786  * truncate_setsize() writes the inode size before removing pages, once we have
8787  * the page lock we can determine safely if the page is beyond EOF. If it is not
8788  * beyond EOF, then the page is guaranteed safe against truncation until we
8789  * unlock the page.
8790  */
8791 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8792 {
8793 	struct page *page = vmf->page;
8794 	struct inode *inode = file_inode(vmf->vma->vm_file);
8795 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8796 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8797 	struct btrfs_ordered_extent *ordered;
8798 	struct extent_state *cached_state = NULL;
8799 	struct extent_changeset *data_reserved = NULL;
8800 	char *kaddr;
8801 	unsigned long zero_start;
8802 	loff_t size;
8803 	vm_fault_t ret;
8804 	int ret2;
8805 	int reserved = 0;
8806 	u64 reserved_space;
8807 	u64 page_start;
8808 	u64 page_end;
8809 	u64 end;
8810 
8811 	reserved_space = PAGE_SIZE;
8812 
8813 	sb_start_pagefault(inode->i_sb);
8814 	page_start = page_offset(page);
8815 	page_end = page_start + PAGE_SIZE - 1;
8816 	end = page_end;
8817 
8818 	/*
8819 	 * Reserving delalloc space after obtaining the page lock can lead to
8820 	 * deadlock. For example, if a dirty page is locked by this function
8821 	 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8822 	 * dirty page write out, then the btrfs_writepage() function could
8823 	 * end up waiting indefinitely to get a lock on the page currently
8824 	 * being processed by btrfs_page_mkwrite() function.
8825 	 */
8826 	ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8827 					   reserved_space);
8828 	if (!ret2) {
8829 		ret2 = file_update_time(vmf->vma->vm_file);
8830 		reserved = 1;
8831 	}
8832 	if (ret2) {
8833 		ret = vmf_error(ret2);
8834 		if (reserved)
8835 			goto out;
8836 		goto out_noreserve;
8837 	}
8838 
8839 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8840 again:
8841 	lock_page(page);
8842 	size = i_size_read(inode);
8843 
8844 	if ((page->mapping != inode->i_mapping) ||
8845 	    (page_start >= size)) {
8846 		/* page got truncated out from underneath us */
8847 		goto out_unlock;
8848 	}
8849 	wait_on_page_writeback(page);
8850 
8851 	lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8852 	set_page_extent_mapped(page);
8853 
8854 	/*
8855 	 * we can't set the delalloc bits if there are pending ordered
8856 	 * extents.  Drop our locks and wait for them to finish
8857 	 */
8858 	ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8859 			PAGE_SIZE);
8860 	if (ordered) {
8861 		unlock_extent_cached(io_tree, page_start, page_end,
8862 				     &cached_state);
8863 		unlock_page(page);
8864 		btrfs_start_ordered_extent(inode, ordered, 1);
8865 		btrfs_put_ordered_extent(ordered);
8866 		goto again;
8867 	}
8868 
8869 	if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8870 		reserved_space = round_up(size - page_start,
8871 					  fs_info->sectorsize);
8872 		if (reserved_space < PAGE_SIZE) {
8873 			end = page_start + reserved_space - 1;
8874 			btrfs_delalloc_release_space(inode, data_reserved,
8875 					page_start, PAGE_SIZE - reserved_space,
8876 					true);
8877 		}
8878 	}
8879 
8880 	/*
8881 	 * page_mkwrite gets called when the page is firstly dirtied after it's
8882 	 * faulted in, but write(2) could also dirty a page and set delalloc
8883 	 * bits, thus in this case for space account reason, we still need to
8884 	 * clear any delalloc bits within this page range since we have to
8885 	 * reserve data&meta space before lock_page() (see above comments).
8886 	 */
8887 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8888 			  EXTENT_DIRTY | EXTENT_DELALLOC |
8889 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8890 			  0, 0, &cached_state);
8891 
8892 	ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8893 					&cached_state, 0);
8894 	if (ret2) {
8895 		unlock_extent_cached(io_tree, page_start, page_end,
8896 				     &cached_state);
8897 		ret = VM_FAULT_SIGBUS;
8898 		goto out_unlock;
8899 	}
8900 	ret2 = 0;
8901 
8902 	/* page is wholly or partially inside EOF */
8903 	if (page_start + PAGE_SIZE > size)
8904 		zero_start = size & ~PAGE_MASK;
8905 	else
8906 		zero_start = PAGE_SIZE;
8907 
8908 	if (zero_start != PAGE_SIZE) {
8909 		kaddr = kmap(page);
8910 		memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8911 		flush_dcache_page(page);
8912 		kunmap(page);
8913 	}
8914 	ClearPageChecked(page);
8915 	set_page_dirty(page);
8916 	SetPageUptodate(page);
8917 
8918 	BTRFS_I(inode)->last_trans = fs_info->generation;
8919 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8920 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8921 
8922 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8923 
8924 	if (!ret2) {
8925 		btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8926 		sb_end_pagefault(inode->i_sb);
8927 		extent_changeset_free(data_reserved);
8928 		return VM_FAULT_LOCKED;
8929 	}
8930 
8931 out_unlock:
8932 	unlock_page(page);
8933 out:
8934 	btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8935 	btrfs_delalloc_release_space(inode, data_reserved, page_start,
8936 				     reserved_space, (ret != 0));
8937 out_noreserve:
8938 	sb_end_pagefault(inode->i_sb);
8939 	extent_changeset_free(data_reserved);
8940 	return ret;
8941 }
8942 
8943 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8944 {
8945 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8946 	struct btrfs_root *root = BTRFS_I(inode)->root;
8947 	struct btrfs_block_rsv *rsv;
8948 	int ret;
8949 	struct btrfs_trans_handle *trans;
8950 	u64 mask = fs_info->sectorsize - 1;
8951 	u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8952 
8953 	if (!skip_writeback) {
8954 		ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8955 					       (u64)-1);
8956 		if (ret)
8957 			return ret;
8958 	}
8959 
8960 	/*
8961 	 * Yes ladies and gentlemen, this is indeed ugly.  We have a couple of
8962 	 * things going on here:
8963 	 *
8964 	 * 1) We need to reserve space to update our inode.
8965 	 *
8966 	 * 2) We need to have something to cache all the space that is going to
8967 	 * be free'd up by the truncate operation, but also have some slack
8968 	 * space reserved in case it uses space during the truncate (thank you
8969 	 * very much snapshotting).
8970 	 *
8971 	 * And we need these to be separate.  The fact is we can use a lot of
8972 	 * space doing the truncate, and we have no earthly idea how much space
8973 	 * we will use, so we need the truncate reservation to be separate so it
8974 	 * doesn't end up using space reserved for updating the inode.  We also
8975 	 * need to be able to stop the transaction and start a new one, which
8976 	 * means we need to be able to update the inode several times, and we
8977 	 * have no idea of knowing how many times that will be, so we can't just
8978 	 * reserve 1 item for the entirety of the operation, so that has to be
8979 	 * done separately as well.
8980 	 *
8981 	 * So that leaves us with
8982 	 *
8983 	 * 1) rsv - for the truncate reservation, which we will steal from the
8984 	 * transaction reservation.
8985 	 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8986 	 * updating the inode.
8987 	 */
8988 	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8989 	if (!rsv)
8990 		return -ENOMEM;
8991 	rsv->size = min_size;
8992 	rsv->failfast = 1;
8993 
8994 	/*
8995 	 * 1 for the truncate slack space
8996 	 * 1 for updating the inode.
8997 	 */
8998 	trans = btrfs_start_transaction(root, 2);
8999 	if (IS_ERR(trans)) {
9000 		ret = PTR_ERR(trans);
9001 		goto out;
9002 	}
9003 
9004 	/* Migrate the slack space for the truncate to our reserve */
9005 	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9006 				      min_size, false);
9007 	BUG_ON(ret);
9008 
9009 	/*
9010 	 * So if we truncate and then write and fsync we normally would just
9011 	 * write the extents that changed, which is a problem if we need to
9012 	 * first truncate that entire inode.  So set this flag so we write out
9013 	 * all of the extents in the inode to the sync log so we're completely
9014 	 * safe.
9015 	 */
9016 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9017 	trans->block_rsv = rsv;
9018 
9019 	while (1) {
9020 		ret = btrfs_truncate_inode_items(trans, root, inode,
9021 						 inode->i_size,
9022 						 BTRFS_EXTENT_DATA_KEY);
9023 		trans->block_rsv = &fs_info->trans_block_rsv;
9024 		if (ret != -ENOSPC && ret != -EAGAIN)
9025 			break;
9026 
9027 		ret = btrfs_update_inode(trans, root, inode);
9028 		if (ret)
9029 			break;
9030 
9031 		btrfs_end_transaction(trans);
9032 		btrfs_btree_balance_dirty(fs_info);
9033 
9034 		trans = btrfs_start_transaction(root, 2);
9035 		if (IS_ERR(trans)) {
9036 			ret = PTR_ERR(trans);
9037 			trans = NULL;
9038 			break;
9039 		}
9040 
9041 		btrfs_block_rsv_release(fs_info, rsv, -1);
9042 		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9043 					      rsv, min_size, false);
9044 		BUG_ON(ret);	/* shouldn't happen */
9045 		trans->block_rsv = rsv;
9046 	}
9047 
9048 	/*
9049 	 * We can't call btrfs_truncate_block inside a trans handle as we could
9050 	 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9051 	 * we've truncated everything except the last little bit, and can do
9052 	 * btrfs_truncate_block and then update the disk_i_size.
9053 	 */
9054 	if (ret == NEED_TRUNCATE_BLOCK) {
9055 		btrfs_end_transaction(trans);
9056 		btrfs_btree_balance_dirty(fs_info);
9057 
9058 		ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9059 		if (ret)
9060 			goto out;
9061 		trans = btrfs_start_transaction(root, 1);
9062 		if (IS_ERR(trans)) {
9063 			ret = PTR_ERR(trans);
9064 			goto out;
9065 		}
9066 		btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9067 	}
9068 
9069 	if (trans) {
9070 		int ret2;
9071 
9072 		trans->block_rsv = &fs_info->trans_block_rsv;
9073 		ret2 = btrfs_update_inode(trans, root, inode);
9074 		if (ret2 && !ret)
9075 			ret = ret2;
9076 
9077 		ret2 = btrfs_end_transaction(trans);
9078 		if (ret2 && !ret)
9079 			ret = ret2;
9080 		btrfs_btree_balance_dirty(fs_info);
9081 	}
9082 out:
9083 	btrfs_free_block_rsv(fs_info, rsv);
9084 
9085 	return ret;
9086 }
9087 
9088 /*
9089  * create a new subvolume directory/inode (helper for the ioctl).
9090  */
9091 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9092 			     struct btrfs_root *new_root,
9093 			     struct btrfs_root *parent_root,
9094 			     u64 new_dirid)
9095 {
9096 	struct inode *inode;
9097 	int err;
9098 	u64 index = 0;
9099 
9100 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9101 				new_dirid, new_dirid,
9102 				S_IFDIR | (~current_umask() & S_IRWXUGO),
9103 				&index);
9104 	if (IS_ERR(inode))
9105 		return PTR_ERR(inode);
9106 	inode->i_op = &btrfs_dir_inode_operations;
9107 	inode->i_fop = &btrfs_dir_file_operations;
9108 
9109 	set_nlink(inode, 1);
9110 	btrfs_i_size_write(BTRFS_I(inode), 0);
9111 	unlock_new_inode(inode);
9112 
9113 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9114 	if (err)
9115 		btrfs_err(new_root->fs_info,
9116 			  "error inheriting subvolume %llu properties: %d",
9117 			  new_root->root_key.objectid, err);
9118 
9119 	err = btrfs_update_inode(trans, new_root, inode);
9120 
9121 	iput(inode);
9122 	return err;
9123 }
9124 
9125 struct inode *btrfs_alloc_inode(struct super_block *sb)
9126 {
9127 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9128 	struct btrfs_inode *ei;
9129 	struct inode *inode;
9130 
9131 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9132 	if (!ei)
9133 		return NULL;
9134 
9135 	ei->root = NULL;
9136 	ei->generation = 0;
9137 	ei->last_trans = 0;
9138 	ei->last_sub_trans = 0;
9139 	ei->logged_trans = 0;
9140 	ei->delalloc_bytes = 0;
9141 	ei->new_delalloc_bytes = 0;
9142 	ei->defrag_bytes = 0;
9143 	ei->disk_i_size = 0;
9144 	ei->flags = 0;
9145 	ei->csum_bytes = 0;
9146 	ei->index_cnt = (u64)-1;
9147 	ei->dir_index = 0;
9148 	ei->last_unlink_trans = 0;
9149 	ei->last_log_commit = 0;
9150 
9151 	spin_lock_init(&ei->lock);
9152 	ei->outstanding_extents = 0;
9153 	if (sb->s_magic != BTRFS_TEST_MAGIC)
9154 		btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9155 					      BTRFS_BLOCK_RSV_DELALLOC);
9156 	ei->runtime_flags = 0;
9157 	ei->prop_compress = BTRFS_COMPRESS_NONE;
9158 	ei->defrag_compress = BTRFS_COMPRESS_NONE;
9159 
9160 	ei->delayed_node = NULL;
9161 
9162 	ei->i_otime.tv_sec = 0;
9163 	ei->i_otime.tv_nsec = 0;
9164 
9165 	inode = &ei->vfs_inode;
9166 	extent_map_tree_init(&ei->extent_tree);
9167 	extent_io_tree_init(&ei->io_tree, inode);
9168 	extent_io_tree_init(&ei->io_failure_tree, inode);
9169 	ei->io_tree.track_uptodate = 1;
9170 	ei->io_failure_tree.track_uptodate = 1;
9171 	atomic_set(&ei->sync_writers, 0);
9172 	mutex_init(&ei->log_mutex);
9173 	mutex_init(&ei->delalloc_mutex);
9174 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9175 	INIT_LIST_HEAD(&ei->delalloc_inodes);
9176 	INIT_LIST_HEAD(&ei->delayed_iput);
9177 	RB_CLEAR_NODE(&ei->rb_node);
9178 	init_rwsem(&ei->dio_sem);
9179 
9180 	return inode;
9181 }
9182 
9183 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9184 void btrfs_test_destroy_inode(struct inode *inode)
9185 {
9186 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9187 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9188 }
9189 #endif
9190 
9191 static void btrfs_i_callback(struct rcu_head *head)
9192 {
9193 	struct inode *inode = container_of(head, struct inode, i_rcu);
9194 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9195 }
9196 
9197 void btrfs_destroy_inode(struct inode *inode)
9198 {
9199 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9200 	struct btrfs_ordered_extent *ordered;
9201 	struct btrfs_root *root = BTRFS_I(inode)->root;
9202 
9203 	WARN_ON(!hlist_empty(&inode->i_dentry));
9204 	WARN_ON(inode->i_data.nrpages);
9205 	WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9206 	WARN_ON(BTRFS_I(inode)->block_rsv.size);
9207 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
9208 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9209 	WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9210 	WARN_ON(BTRFS_I(inode)->csum_bytes);
9211 	WARN_ON(BTRFS_I(inode)->defrag_bytes);
9212 
9213 	/*
9214 	 * This can happen where we create an inode, but somebody else also
9215 	 * created the same inode and we need to destroy the one we already
9216 	 * created.
9217 	 */
9218 	if (!root)
9219 		goto free;
9220 
9221 	while (1) {
9222 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9223 		if (!ordered)
9224 			break;
9225 		else {
9226 			btrfs_err(fs_info,
9227 				  "found ordered extent %llu %llu on inode cleanup",
9228 				  ordered->file_offset, ordered->len);
9229 			btrfs_remove_ordered_extent(inode, ordered);
9230 			btrfs_put_ordered_extent(ordered);
9231 			btrfs_put_ordered_extent(ordered);
9232 		}
9233 	}
9234 	btrfs_qgroup_check_reserved_leak(inode);
9235 	inode_tree_del(inode);
9236 	btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9237 free:
9238 	call_rcu(&inode->i_rcu, btrfs_i_callback);
9239 }
9240 
9241 int btrfs_drop_inode(struct inode *inode)
9242 {
9243 	struct btrfs_root *root = BTRFS_I(inode)->root;
9244 
9245 	if (root == NULL)
9246 		return 1;
9247 
9248 	/* the snap/subvol tree is on deleting */
9249 	if (btrfs_root_refs(&root->root_item) == 0)
9250 		return 1;
9251 	else
9252 		return generic_drop_inode(inode);
9253 }
9254 
9255 static void init_once(void *foo)
9256 {
9257 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9258 
9259 	inode_init_once(&ei->vfs_inode);
9260 }
9261 
9262 void __cold btrfs_destroy_cachep(void)
9263 {
9264 	/*
9265 	 * Make sure all delayed rcu free inodes are flushed before we
9266 	 * destroy cache.
9267 	 */
9268 	rcu_barrier();
9269 	kmem_cache_destroy(btrfs_inode_cachep);
9270 	kmem_cache_destroy(btrfs_trans_handle_cachep);
9271 	kmem_cache_destroy(btrfs_path_cachep);
9272 	kmem_cache_destroy(btrfs_free_space_cachep);
9273 }
9274 
9275 int __init btrfs_init_cachep(void)
9276 {
9277 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9278 			sizeof(struct btrfs_inode), 0,
9279 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9280 			init_once);
9281 	if (!btrfs_inode_cachep)
9282 		goto fail;
9283 
9284 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9285 			sizeof(struct btrfs_trans_handle), 0,
9286 			SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9287 	if (!btrfs_trans_handle_cachep)
9288 		goto fail;
9289 
9290 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
9291 			sizeof(struct btrfs_path), 0,
9292 			SLAB_MEM_SPREAD, NULL);
9293 	if (!btrfs_path_cachep)
9294 		goto fail;
9295 
9296 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9297 			sizeof(struct btrfs_free_space), 0,
9298 			SLAB_MEM_SPREAD, NULL);
9299 	if (!btrfs_free_space_cachep)
9300 		goto fail;
9301 
9302 	return 0;
9303 fail:
9304 	btrfs_destroy_cachep();
9305 	return -ENOMEM;
9306 }
9307 
9308 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9309 			 u32 request_mask, unsigned int flags)
9310 {
9311 	u64 delalloc_bytes;
9312 	struct inode *inode = d_inode(path->dentry);
9313 	u32 blocksize = inode->i_sb->s_blocksize;
9314 	u32 bi_flags = BTRFS_I(inode)->flags;
9315 
9316 	stat->result_mask |= STATX_BTIME;
9317 	stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9318 	stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9319 	if (bi_flags & BTRFS_INODE_APPEND)
9320 		stat->attributes |= STATX_ATTR_APPEND;
9321 	if (bi_flags & BTRFS_INODE_COMPRESS)
9322 		stat->attributes |= STATX_ATTR_COMPRESSED;
9323 	if (bi_flags & BTRFS_INODE_IMMUTABLE)
9324 		stat->attributes |= STATX_ATTR_IMMUTABLE;
9325 	if (bi_flags & BTRFS_INODE_NODUMP)
9326 		stat->attributes |= STATX_ATTR_NODUMP;
9327 
9328 	stat->attributes_mask |= (STATX_ATTR_APPEND |
9329 				  STATX_ATTR_COMPRESSED |
9330 				  STATX_ATTR_IMMUTABLE |
9331 				  STATX_ATTR_NODUMP);
9332 
9333 	generic_fillattr(inode, stat);
9334 	stat->dev = BTRFS_I(inode)->root->anon_dev;
9335 
9336 	spin_lock(&BTRFS_I(inode)->lock);
9337 	delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9338 	spin_unlock(&BTRFS_I(inode)->lock);
9339 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9340 			ALIGN(delalloc_bytes, blocksize)) >> 9;
9341 	return 0;
9342 }
9343 
9344 static int btrfs_rename_exchange(struct inode *old_dir,
9345 			      struct dentry *old_dentry,
9346 			      struct inode *new_dir,
9347 			      struct dentry *new_dentry)
9348 {
9349 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9350 	struct btrfs_trans_handle *trans;
9351 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9352 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9353 	struct inode *new_inode = new_dentry->d_inode;
9354 	struct inode *old_inode = old_dentry->d_inode;
9355 	struct timespec64 ctime = current_time(old_inode);
9356 	struct dentry *parent;
9357 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9358 	u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9359 	u64 old_idx = 0;
9360 	u64 new_idx = 0;
9361 	u64 root_objectid;
9362 	int ret;
9363 	bool root_log_pinned = false;
9364 	bool dest_log_pinned = false;
9365 	struct btrfs_log_ctx ctx_root;
9366 	struct btrfs_log_ctx ctx_dest;
9367 	bool sync_log_root = false;
9368 	bool sync_log_dest = false;
9369 	bool commit_transaction = false;
9370 
9371 	/* we only allow rename subvolume link between subvolumes */
9372 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9373 		return -EXDEV;
9374 
9375 	btrfs_init_log_ctx(&ctx_root, old_inode);
9376 	btrfs_init_log_ctx(&ctx_dest, new_inode);
9377 
9378 	/* close the race window with snapshot create/destroy ioctl */
9379 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9380 		down_read(&fs_info->subvol_sem);
9381 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9382 		down_read(&fs_info->subvol_sem);
9383 
9384 	/*
9385 	 * We want to reserve the absolute worst case amount of items.  So if
9386 	 * both inodes are subvols and we need to unlink them then that would
9387 	 * require 4 item modifications, but if they are both normal inodes it
9388 	 * would require 5 item modifications, so we'll assume their normal
9389 	 * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9390 	 * should cover the worst case number of items we'll modify.
9391 	 */
9392 	trans = btrfs_start_transaction(root, 12);
9393 	if (IS_ERR(trans)) {
9394 		ret = PTR_ERR(trans);
9395 		goto out_notrans;
9396 	}
9397 
9398 	/*
9399 	 * We need to find a free sequence number both in the source and
9400 	 * in the destination directory for the exchange.
9401 	 */
9402 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9403 	if (ret)
9404 		goto out_fail;
9405 	ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9406 	if (ret)
9407 		goto out_fail;
9408 
9409 	BTRFS_I(old_inode)->dir_index = 0ULL;
9410 	BTRFS_I(new_inode)->dir_index = 0ULL;
9411 
9412 	/* Reference for the source. */
9413 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9414 		/* force full log commit if subvolume involved. */
9415 		btrfs_set_log_full_commit(fs_info, trans);
9416 	} else {
9417 		btrfs_pin_log_trans(root);
9418 		root_log_pinned = true;
9419 		ret = btrfs_insert_inode_ref(trans, dest,
9420 					     new_dentry->d_name.name,
9421 					     new_dentry->d_name.len,
9422 					     old_ino,
9423 					     btrfs_ino(BTRFS_I(new_dir)),
9424 					     old_idx);
9425 		if (ret)
9426 			goto out_fail;
9427 	}
9428 
9429 	/* And now for the dest. */
9430 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9431 		/* force full log commit if subvolume involved. */
9432 		btrfs_set_log_full_commit(fs_info, trans);
9433 	} else {
9434 		btrfs_pin_log_trans(dest);
9435 		dest_log_pinned = true;
9436 		ret = btrfs_insert_inode_ref(trans, root,
9437 					     old_dentry->d_name.name,
9438 					     old_dentry->d_name.len,
9439 					     new_ino,
9440 					     btrfs_ino(BTRFS_I(old_dir)),
9441 					     new_idx);
9442 		if (ret)
9443 			goto out_fail;
9444 	}
9445 
9446 	/* Update inode version and ctime/mtime. */
9447 	inode_inc_iversion(old_dir);
9448 	inode_inc_iversion(new_dir);
9449 	inode_inc_iversion(old_inode);
9450 	inode_inc_iversion(new_inode);
9451 	old_dir->i_ctime = old_dir->i_mtime = ctime;
9452 	new_dir->i_ctime = new_dir->i_mtime = ctime;
9453 	old_inode->i_ctime = ctime;
9454 	new_inode->i_ctime = ctime;
9455 
9456 	if (old_dentry->d_parent != new_dentry->d_parent) {
9457 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9458 				BTRFS_I(old_inode), 1);
9459 		btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9460 				BTRFS_I(new_inode), 1);
9461 	}
9462 
9463 	/* src is a subvolume */
9464 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9465 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9466 		ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9467 					  old_dentry->d_name.name,
9468 					  old_dentry->d_name.len);
9469 	} else { /* src is an inode */
9470 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9471 					   BTRFS_I(old_dentry->d_inode),
9472 					   old_dentry->d_name.name,
9473 					   old_dentry->d_name.len);
9474 		if (!ret)
9475 			ret = btrfs_update_inode(trans, root, old_inode);
9476 	}
9477 	if (ret) {
9478 		btrfs_abort_transaction(trans, ret);
9479 		goto out_fail;
9480 	}
9481 
9482 	/* dest is a subvolume */
9483 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9484 		root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9485 		ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9486 					  new_dentry->d_name.name,
9487 					  new_dentry->d_name.len);
9488 	} else { /* dest is an inode */
9489 		ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9490 					   BTRFS_I(new_dentry->d_inode),
9491 					   new_dentry->d_name.name,
9492 					   new_dentry->d_name.len);
9493 		if (!ret)
9494 			ret = btrfs_update_inode(trans, dest, new_inode);
9495 	}
9496 	if (ret) {
9497 		btrfs_abort_transaction(trans, ret);
9498 		goto out_fail;
9499 	}
9500 
9501 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9502 			     new_dentry->d_name.name,
9503 			     new_dentry->d_name.len, 0, old_idx);
9504 	if (ret) {
9505 		btrfs_abort_transaction(trans, ret);
9506 		goto out_fail;
9507 	}
9508 
9509 	ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9510 			     old_dentry->d_name.name,
9511 			     old_dentry->d_name.len, 0, new_idx);
9512 	if (ret) {
9513 		btrfs_abort_transaction(trans, ret);
9514 		goto out_fail;
9515 	}
9516 
9517 	if (old_inode->i_nlink == 1)
9518 		BTRFS_I(old_inode)->dir_index = old_idx;
9519 	if (new_inode->i_nlink == 1)
9520 		BTRFS_I(new_inode)->dir_index = new_idx;
9521 
9522 	if (root_log_pinned) {
9523 		parent = new_dentry->d_parent;
9524 		ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9525 					 BTRFS_I(old_dir), parent,
9526 					 false, &ctx_root);
9527 		if (ret == BTRFS_NEED_LOG_SYNC)
9528 			sync_log_root = true;
9529 		else if (ret == BTRFS_NEED_TRANS_COMMIT)
9530 			commit_transaction = true;
9531 		ret = 0;
9532 		btrfs_end_log_trans(root);
9533 		root_log_pinned = false;
9534 	}
9535 	if (dest_log_pinned) {
9536 		if (!commit_transaction) {
9537 			parent = old_dentry->d_parent;
9538 			ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9539 						 BTRFS_I(new_dir), parent,
9540 						 false, &ctx_dest);
9541 			if (ret == BTRFS_NEED_LOG_SYNC)
9542 				sync_log_dest = true;
9543 			else if (ret == BTRFS_NEED_TRANS_COMMIT)
9544 				commit_transaction = true;
9545 			ret = 0;
9546 		}
9547 		btrfs_end_log_trans(dest);
9548 		dest_log_pinned = false;
9549 	}
9550 out_fail:
9551 	/*
9552 	 * If we have pinned a log and an error happened, we unpin tasks
9553 	 * trying to sync the log and force them to fallback to a transaction
9554 	 * commit if the log currently contains any of the inodes involved in
9555 	 * this rename operation (to ensure we do not persist a log with an
9556 	 * inconsistent state for any of these inodes or leading to any
9557 	 * inconsistencies when replayed). If the transaction was aborted, the
9558 	 * abortion reason is propagated to userspace when attempting to commit
9559 	 * the transaction. If the log does not contain any of these inodes, we
9560 	 * allow the tasks to sync it.
9561 	 */
9562 	if (ret && (root_log_pinned || dest_log_pinned)) {
9563 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9564 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9565 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9566 		    (new_inode &&
9567 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9568 			btrfs_set_log_full_commit(fs_info, trans);
9569 
9570 		if (root_log_pinned) {
9571 			btrfs_end_log_trans(root);
9572 			root_log_pinned = false;
9573 		}
9574 		if (dest_log_pinned) {
9575 			btrfs_end_log_trans(dest);
9576 			dest_log_pinned = false;
9577 		}
9578 	}
9579 	if (!ret && sync_log_root && !commit_transaction) {
9580 		ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9581 				     &ctx_root);
9582 		if (ret)
9583 			commit_transaction = true;
9584 	}
9585 	if (!ret && sync_log_dest && !commit_transaction) {
9586 		ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9587 				     &ctx_dest);
9588 		if (ret)
9589 			commit_transaction = true;
9590 	}
9591 	if (commit_transaction) {
9592 		ret = btrfs_commit_transaction(trans);
9593 	} else {
9594 		int ret2;
9595 
9596 		ret2 = btrfs_end_transaction(trans);
9597 		ret = ret ? ret : ret2;
9598 	}
9599 out_notrans:
9600 	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9601 		up_read(&fs_info->subvol_sem);
9602 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9603 		up_read(&fs_info->subvol_sem);
9604 
9605 	return ret;
9606 }
9607 
9608 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9609 				     struct btrfs_root *root,
9610 				     struct inode *dir,
9611 				     struct dentry *dentry)
9612 {
9613 	int ret;
9614 	struct inode *inode;
9615 	u64 objectid;
9616 	u64 index;
9617 
9618 	ret = btrfs_find_free_ino(root, &objectid);
9619 	if (ret)
9620 		return ret;
9621 
9622 	inode = btrfs_new_inode(trans, root, dir,
9623 				dentry->d_name.name,
9624 				dentry->d_name.len,
9625 				btrfs_ino(BTRFS_I(dir)),
9626 				objectid,
9627 				S_IFCHR | WHITEOUT_MODE,
9628 				&index);
9629 
9630 	if (IS_ERR(inode)) {
9631 		ret = PTR_ERR(inode);
9632 		return ret;
9633 	}
9634 
9635 	inode->i_op = &btrfs_special_inode_operations;
9636 	init_special_inode(inode, inode->i_mode,
9637 		WHITEOUT_DEV);
9638 
9639 	ret = btrfs_init_inode_security(trans, inode, dir,
9640 				&dentry->d_name);
9641 	if (ret)
9642 		goto out;
9643 
9644 	ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9645 				BTRFS_I(inode), 0, index);
9646 	if (ret)
9647 		goto out;
9648 
9649 	ret = btrfs_update_inode(trans, root, inode);
9650 out:
9651 	unlock_new_inode(inode);
9652 	if (ret)
9653 		inode_dec_link_count(inode);
9654 	iput(inode);
9655 
9656 	return ret;
9657 }
9658 
9659 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9660 			   struct inode *new_dir, struct dentry *new_dentry,
9661 			   unsigned int flags)
9662 {
9663 	struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9664 	struct btrfs_trans_handle *trans;
9665 	unsigned int trans_num_items;
9666 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
9667 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9668 	struct inode *new_inode = d_inode(new_dentry);
9669 	struct inode *old_inode = d_inode(old_dentry);
9670 	u64 index = 0;
9671 	u64 root_objectid;
9672 	int ret;
9673 	u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9674 	bool log_pinned = false;
9675 	struct btrfs_log_ctx ctx;
9676 	bool sync_log = false;
9677 	bool commit_transaction = false;
9678 
9679 	if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9680 		return -EPERM;
9681 
9682 	/* we only allow rename subvolume link between subvolumes */
9683 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9684 		return -EXDEV;
9685 
9686 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9687 	    (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9688 		return -ENOTEMPTY;
9689 
9690 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
9691 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9692 		return -ENOTEMPTY;
9693 
9694 
9695 	/* check for collisions, even if the  name isn't there */
9696 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9697 			     new_dentry->d_name.name,
9698 			     new_dentry->d_name.len);
9699 
9700 	if (ret) {
9701 		if (ret == -EEXIST) {
9702 			/* we shouldn't get
9703 			 * eexist without a new_inode */
9704 			if (WARN_ON(!new_inode)) {
9705 				return ret;
9706 			}
9707 		} else {
9708 			/* maybe -EOVERFLOW */
9709 			return ret;
9710 		}
9711 	}
9712 	ret = 0;
9713 
9714 	/*
9715 	 * we're using rename to replace one file with another.  Start IO on it
9716 	 * now so  we don't add too much work to the end of the transaction
9717 	 */
9718 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9719 		filemap_flush(old_inode->i_mapping);
9720 
9721 	/* close the racy window with snapshot create/destroy ioctl */
9722 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9723 		down_read(&fs_info->subvol_sem);
9724 	/*
9725 	 * We want to reserve the absolute worst case amount of items.  So if
9726 	 * both inodes are subvols and we need to unlink them then that would
9727 	 * require 4 item modifications, but if they are both normal inodes it
9728 	 * would require 5 item modifications, so we'll assume they are normal
9729 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9730 	 * should cover the worst case number of items we'll modify.
9731 	 * If our rename has the whiteout flag, we need more 5 units for the
9732 	 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9733 	 * when selinux is enabled).
9734 	 */
9735 	trans_num_items = 11;
9736 	if (flags & RENAME_WHITEOUT)
9737 		trans_num_items += 5;
9738 	trans = btrfs_start_transaction(root, trans_num_items);
9739 	if (IS_ERR(trans)) {
9740 		ret = PTR_ERR(trans);
9741 		goto out_notrans;
9742 	}
9743 
9744 	if (dest != root)
9745 		btrfs_record_root_in_trans(trans, dest);
9746 
9747 	ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9748 	if (ret)
9749 		goto out_fail;
9750 
9751 	BTRFS_I(old_inode)->dir_index = 0ULL;
9752 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9753 		/* force full log commit if subvolume involved. */
9754 		btrfs_set_log_full_commit(fs_info, trans);
9755 	} else {
9756 		btrfs_pin_log_trans(root);
9757 		log_pinned = true;
9758 		ret = btrfs_insert_inode_ref(trans, dest,
9759 					     new_dentry->d_name.name,
9760 					     new_dentry->d_name.len,
9761 					     old_ino,
9762 					     btrfs_ino(BTRFS_I(new_dir)), index);
9763 		if (ret)
9764 			goto out_fail;
9765 	}
9766 
9767 	inode_inc_iversion(old_dir);
9768 	inode_inc_iversion(new_dir);
9769 	inode_inc_iversion(old_inode);
9770 	old_dir->i_ctime = old_dir->i_mtime =
9771 	new_dir->i_ctime = new_dir->i_mtime =
9772 	old_inode->i_ctime = current_time(old_dir);
9773 
9774 	if (old_dentry->d_parent != new_dentry->d_parent)
9775 		btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9776 				BTRFS_I(old_inode), 1);
9777 
9778 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9779 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9780 		ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9781 					old_dentry->d_name.name,
9782 					old_dentry->d_name.len);
9783 	} else {
9784 		ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9785 					BTRFS_I(d_inode(old_dentry)),
9786 					old_dentry->d_name.name,
9787 					old_dentry->d_name.len);
9788 		if (!ret)
9789 			ret = btrfs_update_inode(trans, root, old_inode);
9790 	}
9791 	if (ret) {
9792 		btrfs_abort_transaction(trans, ret);
9793 		goto out_fail;
9794 	}
9795 
9796 	if (new_inode) {
9797 		inode_inc_iversion(new_inode);
9798 		new_inode->i_ctime = current_time(new_inode);
9799 		if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9800 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9801 			root_objectid = BTRFS_I(new_inode)->location.objectid;
9802 			ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9803 						new_dentry->d_name.name,
9804 						new_dentry->d_name.len);
9805 			BUG_ON(new_inode->i_nlink == 0);
9806 		} else {
9807 			ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9808 						 BTRFS_I(d_inode(new_dentry)),
9809 						 new_dentry->d_name.name,
9810 						 new_dentry->d_name.len);
9811 		}
9812 		if (!ret && new_inode->i_nlink == 0)
9813 			ret = btrfs_orphan_add(trans,
9814 					BTRFS_I(d_inode(new_dentry)));
9815 		if (ret) {
9816 			btrfs_abort_transaction(trans, ret);
9817 			goto out_fail;
9818 		}
9819 	}
9820 
9821 	ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9822 			     new_dentry->d_name.name,
9823 			     new_dentry->d_name.len, 0, index);
9824 	if (ret) {
9825 		btrfs_abort_transaction(trans, ret);
9826 		goto out_fail;
9827 	}
9828 
9829 	if (old_inode->i_nlink == 1)
9830 		BTRFS_I(old_inode)->dir_index = index;
9831 
9832 	if (log_pinned) {
9833 		struct dentry *parent = new_dentry->d_parent;
9834 
9835 		btrfs_init_log_ctx(&ctx, old_inode);
9836 		ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9837 					 BTRFS_I(old_dir), parent,
9838 					 false, &ctx);
9839 		if (ret == BTRFS_NEED_LOG_SYNC)
9840 			sync_log = true;
9841 		else if (ret == BTRFS_NEED_TRANS_COMMIT)
9842 			commit_transaction = true;
9843 		ret = 0;
9844 		btrfs_end_log_trans(root);
9845 		log_pinned = false;
9846 	}
9847 
9848 	if (flags & RENAME_WHITEOUT) {
9849 		ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9850 						old_dentry);
9851 
9852 		if (ret) {
9853 			btrfs_abort_transaction(trans, ret);
9854 			goto out_fail;
9855 		}
9856 	}
9857 out_fail:
9858 	/*
9859 	 * If we have pinned the log and an error happened, we unpin tasks
9860 	 * trying to sync the log and force them to fallback to a transaction
9861 	 * commit if the log currently contains any of the inodes involved in
9862 	 * this rename operation (to ensure we do not persist a log with an
9863 	 * inconsistent state for any of these inodes or leading to any
9864 	 * inconsistencies when replayed). If the transaction was aborted, the
9865 	 * abortion reason is propagated to userspace when attempting to commit
9866 	 * the transaction. If the log does not contain any of these inodes, we
9867 	 * allow the tasks to sync it.
9868 	 */
9869 	if (ret && log_pinned) {
9870 		if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9871 		    btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9872 		    btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9873 		    (new_inode &&
9874 		     btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9875 			btrfs_set_log_full_commit(fs_info, trans);
9876 
9877 		btrfs_end_log_trans(root);
9878 		log_pinned = false;
9879 	}
9880 	if (!ret && sync_log) {
9881 		ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9882 		if (ret)
9883 			commit_transaction = true;
9884 	}
9885 	if (commit_transaction) {
9886 		ret = btrfs_commit_transaction(trans);
9887 	} else {
9888 		int ret2;
9889 
9890 		ret2 = btrfs_end_transaction(trans);
9891 		ret = ret ? ret : ret2;
9892 	}
9893 out_notrans:
9894 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9895 		up_read(&fs_info->subvol_sem);
9896 
9897 	return ret;
9898 }
9899 
9900 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9901 			 struct inode *new_dir, struct dentry *new_dentry,
9902 			 unsigned int flags)
9903 {
9904 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9905 		return -EINVAL;
9906 
9907 	if (flags & RENAME_EXCHANGE)
9908 		return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9909 					  new_dentry);
9910 
9911 	return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9912 }
9913 
9914 struct btrfs_delalloc_work {
9915 	struct inode *inode;
9916 	struct completion completion;
9917 	struct list_head list;
9918 	struct btrfs_work work;
9919 };
9920 
9921 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9922 {
9923 	struct btrfs_delalloc_work *delalloc_work;
9924 	struct inode *inode;
9925 
9926 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
9927 				     work);
9928 	inode = delalloc_work->inode;
9929 	filemap_flush(inode->i_mapping);
9930 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9931 				&BTRFS_I(inode)->runtime_flags))
9932 		filemap_flush(inode->i_mapping);
9933 
9934 	iput(inode);
9935 	complete(&delalloc_work->completion);
9936 }
9937 
9938 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9939 {
9940 	struct btrfs_delalloc_work *work;
9941 
9942 	work = kmalloc(sizeof(*work), GFP_NOFS);
9943 	if (!work)
9944 		return NULL;
9945 
9946 	init_completion(&work->completion);
9947 	INIT_LIST_HEAD(&work->list);
9948 	work->inode = inode;
9949 	WARN_ON_ONCE(!inode);
9950 	btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9951 			btrfs_run_delalloc_work, NULL, NULL);
9952 
9953 	return work;
9954 }
9955 
9956 /*
9957  * some fairly slow code that needs optimization. This walks the list
9958  * of all the inodes with pending delalloc and forces them to disk.
9959  */
9960 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9961 {
9962 	struct btrfs_inode *binode;
9963 	struct inode *inode;
9964 	struct btrfs_delalloc_work *work, *next;
9965 	struct list_head works;
9966 	struct list_head splice;
9967 	int ret = 0;
9968 
9969 	INIT_LIST_HEAD(&works);
9970 	INIT_LIST_HEAD(&splice);
9971 
9972 	mutex_lock(&root->delalloc_mutex);
9973 	spin_lock(&root->delalloc_lock);
9974 	list_splice_init(&root->delalloc_inodes, &splice);
9975 	while (!list_empty(&splice)) {
9976 		binode = list_entry(splice.next, struct btrfs_inode,
9977 				    delalloc_inodes);
9978 
9979 		list_move_tail(&binode->delalloc_inodes,
9980 			       &root->delalloc_inodes);
9981 		inode = igrab(&binode->vfs_inode);
9982 		if (!inode) {
9983 			cond_resched_lock(&root->delalloc_lock);
9984 			continue;
9985 		}
9986 		spin_unlock(&root->delalloc_lock);
9987 
9988 		work = btrfs_alloc_delalloc_work(inode);
9989 		if (!work) {
9990 			iput(inode);
9991 			ret = -ENOMEM;
9992 			goto out;
9993 		}
9994 		list_add_tail(&work->list, &works);
9995 		btrfs_queue_work(root->fs_info->flush_workers,
9996 				 &work->work);
9997 		ret++;
9998 		if (nr != -1 && ret >= nr)
9999 			goto out;
10000 		cond_resched();
10001 		spin_lock(&root->delalloc_lock);
10002 	}
10003 	spin_unlock(&root->delalloc_lock);
10004 
10005 out:
10006 	list_for_each_entry_safe(work, next, &works, list) {
10007 		list_del_init(&work->list);
10008 		wait_for_completion(&work->completion);
10009 		kfree(work);
10010 	}
10011 
10012 	if (!list_empty(&splice)) {
10013 		spin_lock(&root->delalloc_lock);
10014 		list_splice_tail(&splice, &root->delalloc_inodes);
10015 		spin_unlock(&root->delalloc_lock);
10016 	}
10017 	mutex_unlock(&root->delalloc_mutex);
10018 	return ret;
10019 }
10020 
10021 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10022 {
10023 	struct btrfs_fs_info *fs_info = root->fs_info;
10024 	int ret;
10025 
10026 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10027 		return -EROFS;
10028 
10029 	ret = start_delalloc_inodes(root, -1);
10030 	if (ret > 0)
10031 		ret = 0;
10032 	return ret;
10033 }
10034 
10035 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10036 {
10037 	struct btrfs_root *root;
10038 	struct list_head splice;
10039 	int ret;
10040 
10041 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10042 		return -EROFS;
10043 
10044 	INIT_LIST_HEAD(&splice);
10045 
10046 	mutex_lock(&fs_info->delalloc_root_mutex);
10047 	spin_lock(&fs_info->delalloc_root_lock);
10048 	list_splice_init(&fs_info->delalloc_roots, &splice);
10049 	while (!list_empty(&splice) && nr) {
10050 		root = list_first_entry(&splice, struct btrfs_root,
10051 					delalloc_root);
10052 		root = btrfs_grab_fs_root(root);
10053 		BUG_ON(!root);
10054 		list_move_tail(&root->delalloc_root,
10055 			       &fs_info->delalloc_roots);
10056 		spin_unlock(&fs_info->delalloc_root_lock);
10057 
10058 		ret = start_delalloc_inodes(root, nr);
10059 		btrfs_put_fs_root(root);
10060 		if (ret < 0)
10061 			goto out;
10062 
10063 		if (nr != -1) {
10064 			nr -= ret;
10065 			WARN_ON(nr < 0);
10066 		}
10067 		spin_lock(&fs_info->delalloc_root_lock);
10068 	}
10069 	spin_unlock(&fs_info->delalloc_root_lock);
10070 
10071 	ret = 0;
10072 out:
10073 	if (!list_empty(&splice)) {
10074 		spin_lock(&fs_info->delalloc_root_lock);
10075 		list_splice_tail(&splice, &fs_info->delalloc_roots);
10076 		spin_unlock(&fs_info->delalloc_root_lock);
10077 	}
10078 	mutex_unlock(&fs_info->delalloc_root_mutex);
10079 	return ret;
10080 }
10081 
10082 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10083 			 const char *symname)
10084 {
10085 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10086 	struct btrfs_trans_handle *trans;
10087 	struct btrfs_root *root = BTRFS_I(dir)->root;
10088 	struct btrfs_path *path;
10089 	struct btrfs_key key;
10090 	struct inode *inode = NULL;
10091 	int err;
10092 	u64 objectid;
10093 	u64 index = 0;
10094 	int name_len;
10095 	int datasize;
10096 	unsigned long ptr;
10097 	struct btrfs_file_extent_item *ei;
10098 	struct extent_buffer *leaf;
10099 
10100 	name_len = strlen(symname);
10101 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10102 		return -ENAMETOOLONG;
10103 
10104 	/*
10105 	 * 2 items for inode item and ref
10106 	 * 2 items for dir items
10107 	 * 1 item for updating parent inode item
10108 	 * 1 item for the inline extent item
10109 	 * 1 item for xattr if selinux is on
10110 	 */
10111 	trans = btrfs_start_transaction(root, 7);
10112 	if (IS_ERR(trans))
10113 		return PTR_ERR(trans);
10114 
10115 	err = btrfs_find_free_ino(root, &objectid);
10116 	if (err)
10117 		goto out_unlock;
10118 
10119 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10120 				dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10121 				objectid, S_IFLNK|S_IRWXUGO, &index);
10122 	if (IS_ERR(inode)) {
10123 		err = PTR_ERR(inode);
10124 		inode = NULL;
10125 		goto out_unlock;
10126 	}
10127 
10128 	/*
10129 	* If the active LSM wants to access the inode during
10130 	* d_instantiate it needs these. Smack checks to see
10131 	* if the filesystem supports xattrs by looking at the
10132 	* ops vector.
10133 	*/
10134 	inode->i_fop = &btrfs_file_operations;
10135 	inode->i_op = &btrfs_file_inode_operations;
10136 	inode->i_mapping->a_ops = &btrfs_aops;
10137 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10138 
10139 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10140 	if (err)
10141 		goto out_unlock;
10142 
10143 	path = btrfs_alloc_path();
10144 	if (!path) {
10145 		err = -ENOMEM;
10146 		goto out_unlock;
10147 	}
10148 	key.objectid = btrfs_ino(BTRFS_I(inode));
10149 	key.offset = 0;
10150 	key.type = BTRFS_EXTENT_DATA_KEY;
10151 	datasize = btrfs_file_extent_calc_inline_size(name_len);
10152 	err = btrfs_insert_empty_item(trans, root, path, &key,
10153 				      datasize);
10154 	if (err) {
10155 		btrfs_free_path(path);
10156 		goto out_unlock;
10157 	}
10158 	leaf = path->nodes[0];
10159 	ei = btrfs_item_ptr(leaf, path->slots[0],
10160 			    struct btrfs_file_extent_item);
10161 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10162 	btrfs_set_file_extent_type(leaf, ei,
10163 				   BTRFS_FILE_EXTENT_INLINE);
10164 	btrfs_set_file_extent_encryption(leaf, ei, 0);
10165 	btrfs_set_file_extent_compression(leaf, ei, 0);
10166 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10167 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10168 
10169 	ptr = btrfs_file_extent_inline_start(ei);
10170 	write_extent_buffer(leaf, symname, ptr, name_len);
10171 	btrfs_mark_buffer_dirty(leaf);
10172 	btrfs_free_path(path);
10173 
10174 	inode->i_op = &btrfs_symlink_inode_operations;
10175 	inode_nohighmem(inode);
10176 	inode->i_mapping->a_ops = &btrfs_aops;
10177 	inode_set_bytes(inode, name_len);
10178 	btrfs_i_size_write(BTRFS_I(inode), name_len);
10179 	err = btrfs_update_inode(trans, root, inode);
10180 	/*
10181 	 * Last step, add directory indexes for our symlink inode. This is the
10182 	 * last step to avoid extra cleanup of these indexes if an error happens
10183 	 * elsewhere above.
10184 	 */
10185 	if (!err)
10186 		err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10187 				BTRFS_I(inode), 0, index);
10188 	if (err)
10189 		goto out_unlock;
10190 
10191 	d_instantiate_new(dentry, inode);
10192 
10193 out_unlock:
10194 	btrfs_end_transaction(trans);
10195 	if (err && inode) {
10196 		inode_dec_link_count(inode);
10197 		discard_new_inode(inode);
10198 	}
10199 	btrfs_btree_balance_dirty(fs_info);
10200 	return err;
10201 }
10202 
10203 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10204 				       u64 start, u64 num_bytes, u64 min_size,
10205 				       loff_t actual_len, u64 *alloc_hint,
10206 				       struct btrfs_trans_handle *trans)
10207 {
10208 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10209 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10210 	struct extent_map *em;
10211 	struct btrfs_root *root = BTRFS_I(inode)->root;
10212 	struct btrfs_key ins;
10213 	u64 cur_offset = start;
10214 	u64 i_size;
10215 	u64 cur_bytes;
10216 	u64 last_alloc = (u64)-1;
10217 	int ret = 0;
10218 	bool own_trans = true;
10219 	u64 end = start + num_bytes - 1;
10220 
10221 	if (trans)
10222 		own_trans = false;
10223 	while (num_bytes > 0) {
10224 		if (own_trans) {
10225 			trans = btrfs_start_transaction(root, 3);
10226 			if (IS_ERR(trans)) {
10227 				ret = PTR_ERR(trans);
10228 				break;
10229 			}
10230 		}
10231 
10232 		cur_bytes = min_t(u64, num_bytes, SZ_256M);
10233 		cur_bytes = max(cur_bytes, min_size);
10234 		/*
10235 		 * If we are severely fragmented we could end up with really
10236 		 * small allocations, so if the allocator is returning small
10237 		 * chunks lets make its job easier by only searching for those
10238 		 * sized chunks.
10239 		 */
10240 		cur_bytes = min(cur_bytes, last_alloc);
10241 		ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10242 				min_size, 0, *alloc_hint, &ins, 1, 0);
10243 		if (ret) {
10244 			if (own_trans)
10245 				btrfs_end_transaction(trans);
10246 			break;
10247 		}
10248 		btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10249 
10250 		last_alloc = ins.offset;
10251 		ret = insert_reserved_file_extent(trans, inode,
10252 						  cur_offset, ins.objectid,
10253 						  ins.offset, ins.offset,
10254 						  ins.offset, 0, 0, 0,
10255 						  BTRFS_FILE_EXTENT_PREALLOC);
10256 		if (ret) {
10257 			btrfs_free_reserved_extent(fs_info, ins.objectid,
10258 						   ins.offset, 0);
10259 			btrfs_abort_transaction(trans, ret);
10260 			if (own_trans)
10261 				btrfs_end_transaction(trans);
10262 			break;
10263 		}
10264 
10265 		btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10266 					cur_offset + ins.offset -1, 0);
10267 
10268 		em = alloc_extent_map();
10269 		if (!em) {
10270 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10271 				&BTRFS_I(inode)->runtime_flags);
10272 			goto next;
10273 		}
10274 
10275 		em->start = cur_offset;
10276 		em->orig_start = cur_offset;
10277 		em->len = ins.offset;
10278 		em->block_start = ins.objectid;
10279 		em->block_len = ins.offset;
10280 		em->orig_block_len = ins.offset;
10281 		em->ram_bytes = ins.offset;
10282 		em->bdev = fs_info->fs_devices->latest_bdev;
10283 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10284 		em->generation = trans->transid;
10285 
10286 		while (1) {
10287 			write_lock(&em_tree->lock);
10288 			ret = add_extent_mapping(em_tree, em, 1);
10289 			write_unlock(&em_tree->lock);
10290 			if (ret != -EEXIST)
10291 				break;
10292 			btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10293 						cur_offset + ins.offset - 1,
10294 						0);
10295 		}
10296 		free_extent_map(em);
10297 next:
10298 		num_bytes -= ins.offset;
10299 		cur_offset += ins.offset;
10300 		*alloc_hint = ins.objectid + ins.offset;
10301 
10302 		inode_inc_iversion(inode);
10303 		inode->i_ctime = current_time(inode);
10304 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10305 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10306 		    (actual_len > inode->i_size) &&
10307 		    (cur_offset > inode->i_size)) {
10308 			if (cur_offset > actual_len)
10309 				i_size = actual_len;
10310 			else
10311 				i_size = cur_offset;
10312 			i_size_write(inode, i_size);
10313 			btrfs_ordered_update_i_size(inode, i_size, NULL);
10314 		}
10315 
10316 		ret = btrfs_update_inode(trans, root, inode);
10317 
10318 		if (ret) {
10319 			btrfs_abort_transaction(trans, ret);
10320 			if (own_trans)
10321 				btrfs_end_transaction(trans);
10322 			break;
10323 		}
10324 
10325 		if (own_trans)
10326 			btrfs_end_transaction(trans);
10327 	}
10328 	if (cur_offset < end)
10329 		btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10330 			end - cur_offset + 1);
10331 	return ret;
10332 }
10333 
10334 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10335 			      u64 start, u64 num_bytes, u64 min_size,
10336 			      loff_t actual_len, u64 *alloc_hint)
10337 {
10338 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10339 					   min_size, actual_len, alloc_hint,
10340 					   NULL);
10341 }
10342 
10343 int btrfs_prealloc_file_range_trans(struct inode *inode,
10344 				    struct btrfs_trans_handle *trans, int mode,
10345 				    u64 start, u64 num_bytes, u64 min_size,
10346 				    loff_t actual_len, u64 *alloc_hint)
10347 {
10348 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10349 					   min_size, actual_len, alloc_hint, trans);
10350 }
10351 
10352 static int btrfs_set_page_dirty(struct page *page)
10353 {
10354 	return __set_page_dirty_nobuffers(page);
10355 }
10356 
10357 static int btrfs_permission(struct inode *inode, int mask)
10358 {
10359 	struct btrfs_root *root = BTRFS_I(inode)->root;
10360 	umode_t mode = inode->i_mode;
10361 
10362 	if (mask & MAY_WRITE &&
10363 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10364 		if (btrfs_root_readonly(root))
10365 			return -EROFS;
10366 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10367 			return -EACCES;
10368 	}
10369 	return generic_permission(inode, mask);
10370 }
10371 
10372 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10373 {
10374 	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10375 	struct btrfs_trans_handle *trans;
10376 	struct btrfs_root *root = BTRFS_I(dir)->root;
10377 	struct inode *inode = NULL;
10378 	u64 objectid;
10379 	u64 index;
10380 	int ret = 0;
10381 
10382 	/*
10383 	 * 5 units required for adding orphan entry
10384 	 */
10385 	trans = btrfs_start_transaction(root, 5);
10386 	if (IS_ERR(trans))
10387 		return PTR_ERR(trans);
10388 
10389 	ret = btrfs_find_free_ino(root, &objectid);
10390 	if (ret)
10391 		goto out;
10392 
10393 	inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10394 			btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10395 	if (IS_ERR(inode)) {
10396 		ret = PTR_ERR(inode);
10397 		inode = NULL;
10398 		goto out;
10399 	}
10400 
10401 	inode->i_fop = &btrfs_file_operations;
10402 	inode->i_op = &btrfs_file_inode_operations;
10403 
10404 	inode->i_mapping->a_ops = &btrfs_aops;
10405 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10406 
10407 	ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10408 	if (ret)
10409 		goto out;
10410 
10411 	ret = btrfs_update_inode(trans, root, inode);
10412 	if (ret)
10413 		goto out;
10414 	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10415 	if (ret)
10416 		goto out;
10417 
10418 	/*
10419 	 * We set number of links to 0 in btrfs_new_inode(), and here we set
10420 	 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10421 	 * through:
10422 	 *
10423 	 *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10424 	 */
10425 	set_nlink(inode, 1);
10426 	d_tmpfile(dentry, inode);
10427 	unlock_new_inode(inode);
10428 	mark_inode_dirty(inode);
10429 out:
10430 	btrfs_end_transaction(trans);
10431 	if (ret && inode)
10432 		discard_new_inode(inode);
10433 	btrfs_btree_balance_dirty(fs_info);
10434 	return ret;
10435 }
10436 
10437 __attribute__((const))
10438 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10439 {
10440 	return -EAGAIN;
10441 }
10442 
10443 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10444 					u64 start, u64 end)
10445 {
10446 	struct inode *inode = private_data;
10447 	u64 isize;
10448 
10449 	isize = i_size_read(inode);
10450 	if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10451 		btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10452 		    "%s: ino %llu isize %llu odd range [%llu,%llu]",
10453 			caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10454 	}
10455 }
10456 
10457 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10458 {
10459 	struct inode *inode = tree->private_data;
10460 	unsigned long index = start >> PAGE_SHIFT;
10461 	unsigned long end_index = end >> PAGE_SHIFT;
10462 	struct page *page;
10463 
10464 	while (index <= end_index) {
10465 		page = find_get_page(inode->i_mapping, index);
10466 		ASSERT(page); /* Pages should be in the extent_io_tree */
10467 		set_page_writeback(page);
10468 		put_page(page);
10469 		index++;
10470 	}
10471 }
10472 
10473 static const struct inode_operations btrfs_dir_inode_operations = {
10474 	.getattr	= btrfs_getattr,
10475 	.lookup		= btrfs_lookup,
10476 	.create		= btrfs_create,
10477 	.unlink		= btrfs_unlink,
10478 	.link		= btrfs_link,
10479 	.mkdir		= btrfs_mkdir,
10480 	.rmdir		= btrfs_rmdir,
10481 	.rename		= btrfs_rename2,
10482 	.symlink	= btrfs_symlink,
10483 	.setattr	= btrfs_setattr,
10484 	.mknod		= btrfs_mknod,
10485 	.listxattr	= btrfs_listxattr,
10486 	.permission	= btrfs_permission,
10487 	.get_acl	= btrfs_get_acl,
10488 	.set_acl	= btrfs_set_acl,
10489 	.update_time	= btrfs_update_time,
10490 	.tmpfile        = btrfs_tmpfile,
10491 };
10492 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10493 	.lookup		= btrfs_lookup,
10494 	.permission	= btrfs_permission,
10495 	.update_time	= btrfs_update_time,
10496 };
10497 
10498 static const struct file_operations btrfs_dir_file_operations = {
10499 	.llseek		= generic_file_llseek,
10500 	.read		= generic_read_dir,
10501 	.iterate_shared	= btrfs_real_readdir,
10502 	.open		= btrfs_opendir,
10503 	.unlocked_ioctl	= btrfs_ioctl,
10504 #ifdef CONFIG_COMPAT
10505 	.compat_ioctl	= btrfs_compat_ioctl,
10506 #endif
10507 	.release        = btrfs_release_file,
10508 	.fsync		= btrfs_sync_file,
10509 };
10510 
10511 static const struct extent_io_ops btrfs_extent_io_ops = {
10512 	/* mandatory callbacks */
10513 	.submit_bio_hook = btrfs_submit_bio_hook,
10514 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
10515 	.readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10516 
10517 	/* optional callbacks */
10518 	.fill_delalloc = run_delalloc_range,
10519 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
10520 	.writepage_start_hook = btrfs_writepage_start_hook,
10521 	.set_bit_hook = btrfs_set_bit_hook,
10522 	.clear_bit_hook = btrfs_clear_bit_hook,
10523 	.merge_extent_hook = btrfs_merge_extent_hook,
10524 	.split_extent_hook = btrfs_split_extent_hook,
10525 	.check_extent_io_range = btrfs_check_extent_io_range,
10526 };
10527 
10528 /*
10529  * btrfs doesn't support the bmap operation because swapfiles
10530  * use bmap to make a mapping of extents in the file.  They assume
10531  * these extents won't change over the life of the file and they
10532  * use the bmap result to do IO directly to the drive.
10533  *
10534  * the btrfs bmap call would return logical addresses that aren't
10535  * suitable for IO and they also will change frequently as COW
10536  * operations happen.  So, swapfile + btrfs == corruption.
10537  *
10538  * For now we're avoiding this by dropping bmap.
10539  */
10540 static const struct address_space_operations btrfs_aops = {
10541 	.readpage	= btrfs_readpage,
10542 	.writepage	= btrfs_writepage,
10543 	.writepages	= btrfs_writepages,
10544 	.readpages	= btrfs_readpages,
10545 	.direct_IO	= btrfs_direct_IO,
10546 	.invalidatepage = btrfs_invalidatepage,
10547 	.releasepage	= btrfs_releasepage,
10548 	.set_page_dirty	= btrfs_set_page_dirty,
10549 	.error_remove_page = generic_error_remove_page,
10550 };
10551 
10552 static const struct inode_operations btrfs_file_inode_operations = {
10553 	.getattr	= btrfs_getattr,
10554 	.setattr	= btrfs_setattr,
10555 	.listxattr      = btrfs_listxattr,
10556 	.permission	= btrfs_permission,
10557 	.fiemap		= btrfs_fiemap,
10558 	.get_acl	= btrfs_get_acl,
10559 	.set_acl	= btrfs_set_acl,
10560 	.update_time	= btrfs_update_time,
10561 };
10562 static const struct inode_operations btrfs_special_inode_operations = {
10563 	.getattr	= btrfs_getattr,
10564 	.setattr	= btrfs_setattr,
10565 	.permission	= btrfs_permission,
10566 	.listxattr	= btrfs_listxattr,
10567 	.get_acl	= btrfs_get_acl,
10568 	.set_acl	= btrfs_set_acl,
10569 	.update_time	= btrfs_update_time,
10570 };
10571 static const struct inode_operations btrfs_symlink_inode_operations = {
10572 	.get_link	= page_get_link,
10573 	.getattr	= btrfs_getattr,
10574 	.setattr	= btrfs_setattr,
10575 	.permission	= btrfs_permission,
10576 	.listxattr	= btrfs_listxattr,
10577 	.update_time	= btrfs_update_time,
10578 };
10579 
10580 const struct dentry_operations btrfs_dentry_operations = {
10581 	.d_delete	= btrfs_dentry_delete,
10582 };
10583