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