xref: /linux/fs/btrfs/inode.c (revision c98be0c96db00e9b6b02d31e0fa7590c54cdaaac)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 
63 struct btrfs_iget_args {
64 	struct btrfs_key *location;
65 	struct btrfs_root *root;
66 };
67 
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
77 
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
84 
85 #define S_SHIFT 12
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
88 	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
89 	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
90 	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
91 	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
92 	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
93 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
94 };
95 
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 				   struct page *locked_page,
101 				   u64 start, u64 end, int *page_started,
102 				   unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 					   u64 len, u64 orig_start,
105 					   u64 block_start, u64 block_len,
106 					   u64 orig_block_len, u64 ram_bytes,
107 					   int type);
108 
109 static int btrfs_dirty_inode(struct inode *inode);
110 
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 				     struct inode *inode,  struct inode *dir,
113 				     const struct qstr *qstr)
114 {
115 	int err;
116 
117 	err = btrfs_init_acl(trans, inode, dir);
118 	if (!err)
119 		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
120 	return err;
121 }
122 
123 /*
124  * this does all the hard work for inserting an inline extent into
125  * the btree.  The caller should have done a btrfs_drop_extents so that
126  * no overlapping inline items exist in the btree
127  */
128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
129 				struct btrfs_path *path, int extent_inserted,
130 				struct btrfs_root *root, struct inode *inode,
131 				u64 start, size_t size, size_t compressed_size,
132 				int compress_type,
133 				struct page **compressed_pages)
134 {
135 	struct extent_buffer *leaf;
136 	struct page *page = NULL;
137 	char *kaddr;
138 	unsigned long ptr;
139 	struct btrfs_file_extent_item *ei;
140 	int err = 0;
141 	int ret;
142 	size_t cur_size = size;
143 	unsigned long offset;
144 
145 	if (compressed_size && compressed_pages)
146 		cur_size = compressed_size;
147 
148 	inode_add_bytes(inode, size);
149 
150 	if (!extent_inserted) {
151 		struct btrfs_key key;
152 		size_t datasize;
153 
154 		key.objectid = btrfs_ino(inode);
155 		key.offset = start;
156 		btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
157 
158 		datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 		path->leave_spinning = 1;
160 		ret = btrfs_insert_empty_item(trans, root, path, &key,
161 					      datasize);
162 		if (ret) {
163 			err = ret;
164 			goto fail;
165 		}
166 	}
167 	leaf = path->nodes[0];
168 	ei = btrfs_item_ptr(leaf, path->slots[0],
169 			    struct btrfs_file_extent_item);
170 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 	btrfs_set_file_extent_encryption(leaf, ei, 0);
173 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 	ptr = btrfs_file_extent_inline_start(ei);
176 
177 	if (compress_type != BTRFS_COMPRESS_NONE) {
178 		struct page *cpage;
179 		int i = 0;
180 		while (compressed_size > 0) {
181 			cpage = compressed_pages[i];
182 			cur_size = min_t(unsigned long, compressed_size,
183 				       PAGE_CACHE_SIZE);
184 
185 			kaddr = kmap_atomic(cpage);
186 			write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 			kunmap_atomic(kaddr);
188 
189 			i++;
190 			ptr += cur_size;
191 			compressed_size -= cur_size;
192 		}
193 		btrfs_set_file_extent_compression(leaf, ei,
194 						  compress_type);
195 	} else {
196 		page = find_get_page(inode->i_mapping,
197 				     start >> PAGE_CACHE_SHIFT);
198 		btrfs_set_file_extent_compression(leaf, ei, 0);
199 		kaddr = kmap_atomic(page);
200 		offset = start & (PAGE_CACHE_SIZE - 1);
201 		write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 		kunmap_atomic(kaddr);
203 		page_cache_release(page);
204 	}
205 	btrfs_mark_buffer_dirty(leaf);
206 	btrfs_release_path(path);
207 
208 	/*
209 	 * we're an inline extent, so nobody can
210 	 * extend the file past i_size without locking
211 	 * a page we already have locked.
212 	 *
213 	 * We must do any isize and inode updates
214 	 * before we unlock the pages.  Otherwise we
215 	 * could end up racing with unlink.
216 	 */
217 	BTRFS_I(inode)->disk_i_size = inode->i_size;
218 	ret = btrfs_update_inode(trans, root, inode);
219 
220 	return ret;
221 fail:
222 	return err;
223 }
224 
225 
226 /*
227  * conditionally insert an inline extent into the file.  This
228  * does the checks required to make sure the data is small enough
229  * to fit as an inline extent.
230  */
231 static noinline int cow_file_range_inline(struct btrfs_root *root,
232 					  struct inode *inode, u64 start,
233 					  u64 end, size_t compressed_size,
234 					  int compress_type,
235 					  struct page **compressed_pages)
236 {
237 	struct btrfs_trans_handle *trans;
238 	u64 isize = i_size_read(inode);
239 	u64 actual_end = min(end + 1, isize);
240 	u64 inline_len = actual_end - start;
241 	u64 aligned_end = ALIGN(end, root->sectorsize);
242 	u64 data_len = inline_len;
243 	int ret;
244 	struct btrfs_path *path;
245 	int extent_inserted = 0;
246 	u32 extent_item_size;
247 
248 	if (compressed_size)
249 		data_len = compressed_size;
250 
251 	if (start > 0 ||
252 	    actual_end >= PAGE_CACHE_SIZE ||
253 	    data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
254 	    (!compressed_size &&
255 	    (actual_end & (root->sectorsize - 1)) == 0) ||
256 	    end + 1 < isize ||
257 	    data_len > root->fs_info->max_inline) {
258 		return 1;
259 	}
260 
261 	path = btrfs_alloc_path();
262 	if (!path)
263 		return -ENOMEM;
264 
265 	trans = btrfs_join_transaction(root);
266 	if (IS_ERR(trans)) {
267 		btrfs_free_path(path);
268 		return PTR_ERR(trans);
269 	}
270 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
271 
272 	if (compressed_size && compressed_pages)
273 		extent_item_size = btrfs_file_extent_calc_inline_size(
274 		   compressed_size);
275 	else
276 		extent_item_size = btrfs_file_extent_calc_inline_size(
277 		    inline_len);
278 
279 	ret = __btrfs_drop_extents(trans, root, inode, path,
280 				   start, aligned_end, NULL,
281 				   1, 1, extent_item_size, &extent_inserted);
282 	if (ret) {
283 		btrfs_abort_transaction(trans, root, ret);
284 		goto out;
285 	}
286 
287 	if (isize > actual_end)
288 		inline_len = min_t(u64, isize, actual_end);
289 	ret = insert_inline_extent(trans, path, extent_inserted,
290 				   root, inode, start,
291 				   inline_len, compressed_size,
292 				   compress_type, compressed_pages);
293 	if (ret && ret != -ENOSPC) {
294 		btrfs_abort_transaction(trans, root, ret);
295 		goto out;
296 	} else if (ret == -ENOSPC) {
297 		ret = 1;
298 		goto out;
299 	}
300 
301 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
302 	btrfs_delalloc_release_metadata(inode, end + 1 - start);
303 	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
304 out:
305 	btrfs_free_path(path);
306 	btrfs_end_transaction(trans, root);
307 	return ret;
308 }
309 
310 struct async_extent {
311 	u64 start;
312 	u64 ram_size;
313 	u64 compressed_size;
314 	struct page **pages;
315 	unsigned long nr_pages;
316 	int compress_type;
317 	struct list_head list;
318 };
319 
320 struct async_cow {
321 	struct inode *inode;
322 	struct btrfs_root *root;
323 	struct page *locked_page;
324 	u64 start;
325 	u64 end;
326 	struct list_head extents;
327 	struct btrfs_work work;
328 };
329 
330 static noinline int add_async_extent(struct async_cow *cow,
331 				     u64 start, u64 ram_size,
332 				     u64 compressed_size,
333 				     struct page **pages,
334 				     unsigned long nr_pages,
335 				     int compress_type)
336 {
337 	struct async_extent *async_extent;
338 
339 	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
340 	BUG_ON(!async_extent); /* -ENOMEM */
341 	async_extent->start = start;
342 	async_extent->ram_size = ram_size;
343 	async_extent->compressed_size = compressed_size;
344 	async_extent->pages = pages;
345 	async_extent->nr_pages = nr_pages;
346 	async_extent->compress_type = compress_type;
347 	list_add_tail(&async_extent->list, &cow->extents);
348 	return 0;
349 }
350 
351 /*
352  * we create compressed extents in two phases.  The first
353  * phase compresses a range of pages that have already been
354  * locked (both pages and state bits are locked).
355  *
356  * This is done inside an ordered work queue, and the compression
357  * is spread across many cpus.  The actual IO submission is step
358  * two, and the ordered work queue takes care of making sure that
359  * happens in the same order things were put onto the queue by
360  * writepages and friends.
361  *
362  * If this code finds it can't get good compression, it puts an
363  * entry onto the work queue to write the uncompressed bytes.  This
364  * makes sure that both compressed inodes and uncompressed inodes
365  * are written in the same order that the flusher thread sent them
366  * down.
367  */
368 static noinline int compress_file_range(struct inode *inode,
369 					struct page *locked_page,
370 					u64 start, u64 end,
371 					struct async_cow *async_cow,
372 					int *num_added)
373 {
374 	struct btrfs_root *root = BTRFS_I(inode)->root;
375 	u64 num_bytes;
376 	u64 blocksize = root->sectorsize;
377 	u64 actual_end;
378 	u64 isize = i_size_read(inode);
379 	int ret = 0;
380 	struct page **pages = NULL;
381 	unsigned long nr_pages;
382 	unsigned long nr_pages_ret = 0;
383 	unsigned long total_compressed = 0;
384 	unsigned long total_in = 0;
385 	unsigned long max_compressed = 128 * 1024;
386 	unsigned long max_uncompressed = 128 * 1024;
387 	int i;
388 	int will_compress;
389 	int compress_type = root->fs_info->compress_type;
390 	int redirty = 0;
391 
392 	/* if this is a small write inside eof, kick off a defrag */
393 	if ((end - start + 1) < 16 * 1024 &&
394 	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
395 		btrfs_add_inode_defrag(NULL, inode);
396 
397 	actual_end = min_t(u64, isize, end + 1);
398 again:
399 	will_compress = 0;
400 	nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
401 	nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
402 
403 	/*
404 	 * we don't want to send crud past the end of i_size through
405 	 * compression, that's just a waste of CPU time.  So, if the
406 	 * end of the file is before the start of our current
407 	 * requested range of bytes, we bail out to the uncompressed
408 	 * cleanup code that can deal with all of this.
409 	 *
410 	 * It isn't really the fastest way to fix things, but this is a
411 	 * very uncommon corner.
412 	 */
413 	if (actual_end <= start)
414 		goto cleanup_and_bail_uncompressed;
415 
416 	total_compressed = actual_end - start;
417 
418 	/* we want to make sure that amount of ram required to uncompress
419 	 * an extent is reasonable, so we limit the total size in ram
420 	 * of a compressed extent to 128k.  This is a crucial number
421 	 * because it also controls how easily we can spread reads across
422 	 * cpus for decompression.
423 	 *
424 	 * We also want to make sure the amount of IO required to do
425 	 * a random read is reasonably small, so we limit the size of
426 	 * a compressed extent to 128k.
427 	 */
428 	total_compressed = min(total_compressed, max_uncompressed);
429 	num_bytes = ALIGN(end - start + 1, blocksize);
430 	num_bytes = max(blocksize,  num_bytes);
431 	total_in = 0;
432 	ret = 0;
433 
434 	/*
435 	 * we do compression for mount -o compress and when the
436 	 * inode has not been flagged as nocompress.  This flag can
437 	 * change at any time if we discover bad compression ratios.
438 	 */
439 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
440 	    (btrfs_test_opt(root, COMPRESS) ||
441 	     (BTRFS_I(inode)->force_compress) ||
442 	     (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
443 		WARN_ON(pages);
444 		pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
445 		if (!pages) {
446 			/* just bail out to the uncompressed code */
447 			goto cont;
448 		}
449 
450 		if (BTRFS_I(inode)->force_compress)
451 			compress_type = BTRFS_I(inode)->force_compress;
452 
453 		/*
454 		 * we need to call clear_page_dirty_for_io on each
455 		 * page in the range.  Otherwise applications with the file
456 		 * mmap'd can wander in and change the page contents while
457 		 * we are compressing them.
458 		 *
459 		 * If the compression fails for any reason, we set the pages
460 		 * dirty again later on.
461 		 */
462 		extent_range_clear_dirty_for_io(inode, start, end);
463 		redirty = 1;
464 		ret = btrfs_compress_pages(compress_type,
465 					   inode->i_mapping, start,
466 					   total_compressed, pages,
467 					   nr_pages, &nr_pages_ret,
468 					   &total_in,
469 					   &total_compressed,
470 					   max_compressed);
471 
472 		if (!ret) {
473 			unsigned long offset = total_compressed &
474 				(PAGE_CACHE_SIZE - 1);
475 			struct page *page = pages[nr_pages_ret - 1];
476 			char *kaddr;
477 
478 			/* zero the tail end of the last page, we might be
479 			 * sending it down to disk
480 			 */
481 			if (offset) {
482 				kaddr = kmap_atomic(page);
483 				memset(kaddr + offset, 0,
484 				       PAGE_CACHE_SIZE - offset);
485 				kunmap_atomic(kaddr);
486 			}
487 			will_compress = 1;
488 		}
489 	}
490 cont:
491 	if (start == 0) {
492 		/* lets try to make an inline extent */
493 		if (ret || total_in < (actual_end - start)) {
494 			/* we didn't compress the entire range, try
495 			 * to make an uncompressed inline extent.
496 			 */
497 			ret = cow_file_range_inline(root, inode, start, end,
498 						    0, 0, NULL);
499 		} else {
500 			/* try making a compressed inline extent */
501 			ret = cow_file_range_inline(root, inode, start, end,
502 						    total_compressed,
503 						    compress_type, pages);
504 		}
505 		if (ret <= 0) {
506 			unsigned long clear_flags = EXTENT_DELALLOC |
507 				EXTENT_DEFRAG;
508 			clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
509 
510 			/*
511 			 * inline extent creation worked or returned error,
512 			 * we don't need to create any more async work items.
513 			 * Unlock and free up our temp pages.
514 			 */
515 			extent_clear_unlock_delalloc(inode, start, end, NULL,
516 						     clear_flags, PAGE_UNLOCK |
517 						     PAGE_CLEAR_DIRTY |
518 						     PAGE_SET_WRITEBACK |
519 						     PAGE_END_WRITEBACK);
520 			goto free_pages_out;
521 		}
522 	}
523 
524 	if (will_compress) {
525 		/*
526 		 * we aren't doing an inline extent round the compressed size
527 		 * up to a block size boundary so the allocator does sane
528 		 * things
529 		 */
530 		total_compressed = ALIGN(total_compressed, blocksize);
531 
532 		/*
533 		 * one last check to make sure the compression is really a
534 		 * win, compare the page count read with the blocks on disk
535 		 */
536 		total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
537 		if (total_compressed >= total_in) {
538 			will_compress = 0;
539 		} else {
540 			num_bytes = total_in;
541 		}
542 	}
543 	if (!will_compress && pages) {
544 		/*
545 		 * the compression code ran but failed to make things smaller,
546 		 * free any pages it allocated and our page pointer array
547 		 */
548 		for (i = 0; i < nr_pages_ret; i++) {
549 			WARN_ON(pages[i]->mapping);
550 			page_cache_release(pages[i]);
551 		}
552 		kfree(pages);
553 		pages = NULL;
554 		total_compressed = 0;
555 		nr_pages_ret = 0;
556 
557 		/* flag the file so we don't compress in the future */
558 		if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
559 		    !(BTRFS_I(inode)->force_compress)) {
560 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
561 		}
562 	}
563 	if (will_compress) {
564 		*num_added += 1;
565 
566 		/* the async work queues will take care of doing actual
567 		 * allocation on disk for these compressed pages,
568 		 * and will submit them to the elevator.
569 		 */
570 		add_async_extent(async_cow, start, num_bytes,
571 				 total_compressed, pages, nr_pages_ret,
572 				 compress_type);
573 
574 		if (start + num_bytes < end) {
575 			start += num_bytes;
576 			pages = NULL;
577 			cond_resched();
578 			goto again;
579 		}
580 	} else {
581 cleanup_and_bail_uncompressed:
582 		/*
583 		 * No compression, but we still need to write the pages in
584 		 * the file we've been given so far.  redirty the locked
585 		 * page if it corresponds to our extent and set things up
586 		 * for the async work queue to run cow_file_range to do
587 		 * the normal delalloc dance
588 		 */
589 		if (page_offset(locked_page) >= start &&
590 		    page_offset(locked_page) <= end) {
591 			__set_page_dirty_nobuffers(locked_page);
592 			/* unlocked later on in the async handlers */
593 		}
594 		if (redirty)
595 			extent_range_redirty_for_io(inode, start, end);
596 		add_async_extent(async_cow, start, end - start + 1,
597 				 0, NULL, 0, BTRFS_COMPRESS_NONE);
598 		*num_added += 1;
599 	}
600 
601 out:
602 	return ret;
603 
604 free_pages_out:
605 	for (i = 0; i < nr_pages_ret; i++) {
606 		WARN_ON(pages[i]->mapping);
607 		page_cache_release(pages[i]);
608 	}
609 	kfree(pages);
610 
611 	goto out;
612 }
613 
614 /*
615  * phase two of compressed writeback.  This is the ordered portion
616  * of the code, which only gets called in the order the work was
617  * queued.  We walk all the async extents created by compress_file_range
618  * and send them down to the disk.
619  */
620 static noinline int submit_compressed_extents(struct inode *inode,
621 					      struct async_cow *async_cow)
622 {
623 	struct async_extent *async_extent;
624 	u64 alloc_hint = 0;
625 	struct btrfs_key ins;
626 	struct extent_map *em;
627 	struct btrfs_root *root = BTRFS_I(inode)->root;
628 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
629 	struct extent_io_tree *io_tree;
630 	int ret = 0;
631 
632 	if (list_empty(&async_cow->extents))
633 		return 0;
634 
635 again:
636 	while (!list_empty(&async_cow->extents)) {
637 		async_extent = list_entry(async_cow->extents.next,
638 					  struct async_extent, list);
639 		list_del(&async_extent->list);
640 
641 		io_tree = &BTRFS_I(inode)->io_tree;
642 
643 retry:
644 		/* did the compression code fall back to uncompressed IO? */
645 		if (!async_extent->pages) {
646 			int page_started = 0;
647 			unsigned long nr_written = 0;
648 
649 			lock_extent(io_tree, async_extent->start,
650 					 async_extent->start +
651 					 async_extent->ram_size - 1);
652 
653 			/* allocate blocks */
654 			ret = cow_file_range(inode, async_cow->locked_page,
655 					     async_extent->start,
656 					     async_extent->start +
657 					     async_extent->ram_size - 1,
658 					     &page_started, &nr_written, 0);
659 
660 			/* JDM XXX */
661 
662 			/*
663 			 * if page_started, cow_file_range inserted an
664 			 * inline extent and took care of all the unlocking
665 			 * and IO for us.  Otherwise, we need to submit
666 			 * all those pages down to the drive.
667 			 */
668 			if (!page_started && !ret)
669 				extent_write_locked_range(io_tree,
670 						  inode, async_extent->start,
671 						  async_extent->start +
672 						  async_extent->ram_size - 1,
673 						  btrfs_get_extent,
674 						  WB_SYNC_ALL);
675 			else if (ret)
676 				unlock_page(async_cow->locked_page);
677 			kfree(async_extent);
678 			cond_resched();
679 			continue;
680 		}
681 
682 		lock_extent(io_tree, async_extent->start,
683 			    async_extent->start + async_extent->ram_size - 1);
684 
685 		ret = btrfs_reserve_extent(root,
686 					   async_extent->compressed_size,
687 					   async_extent->compressed_size,
688 					   0, alloc_hint, &ins, 1);
689 		if (ret) {
690 			int i;
691 
692 			for (i = 0; i < async_extent->nr_pages; i++) {
693 				WARN_ON(async_extent->pages[i]->mapping);
694 				page_cache_release(async_extent->pages[i]);
695 			}
696 			kfree(async_extent->pages);
697 			async_extent->nr_pages = 0;
698 			async_extent->pages = NULL;
699 
700 			if (ret == -ENOSPC) {
701 				unlock_extent(io_tree, async_extent->start,
702 					      async_extent->start +
703 					      async_extent->ram_size - 1);
704 				goto retry;
705 			}
706 			goto out_free;
707 		}
708 
709 		/*
710 		 * here we're doing allocation and writeback of the
711 		 * compressed pages
712 		 */
713 		btrfs_drop_extent_cache(inode, async_extent->start,
714 					async_extent->start +
715 					async_extent->ram_size - 1, 0);
716 
717 		em = alloc_extent_map();
718 		if (!em) {
719 			ret = -ENOMEM;
720 			goto out_free_reserve;
721 		}
722 		em->start = async_extent->start;
723 		em->len = async_extent->ram_size;
724 		em->orig_start = em->start;
725 		em->mod_start = em->start;
726 		em->mod_len = em->len;
727 
728 		em->block_start = ins.objectid;
729 		em->block_len = ins.offset;
730 		em->orig_block_len = ins.offset;
731 		em->ram_bytes = async_extent->ram_size;
732 		em->bdev = root->fs_info->fs_devices->latest_bdev;
733 		em->compress_type = async_extent->compress_type;
734 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
735 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
736 		em->generation = -1;
737 
738 		while (1) {
739 			write_lock(&em_tree->lock);
740 			ret = add_extent_mapping(em_tree, em, 1);
741 			write_unlock(&em_tree->lock);
742 			if (ret != -EEXIST) {
743 				free_extent_map(em);
744 				break;
745 			}
746 			btrfs_drop_extent_cache(inode, async_extent->start,
747 						async_extent->start +
748 						async_extent->ram_size - 1, 0);
749 		}
750 
751 		if (ret)
752 			goto out_free_reserve;
753 
754 		ret = btrfs_add_ordered_extent_compress(inode,
755 						async_extent->start,
756 						ins.objectid,
757 						async_extent->ram_size,
758 						ins.offset,
759 						BTRFS_ORDERED_COMPRESSED,
760 						async_extent->compress_type);
761 		if (ret)
762 			goto out_free_reserve;
763 
764 		/*
765 		 * clear dirty, set writeback and unlock the pages.
766 		 */
767 		extent_clear_unlock_delalloc(inode, async_extent->start,
768 				async_extent->start +
769 				async_extent->ram_size - 1,
770 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
771 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
772 				PAGE_SET_WRITEBACK);
773 		ret = btrfs_submit_compressed_write(inode,
774 				    async_extent->start,
775 				    async_extent->ram_size,
776 				    ins.objectid,
777 				    ins.offset, async_extent->pages,
778 				    async_extent->nr_pages);
779 		alloc_hint = ins.objectid + ins.offset;
780 		kfree(async_extent);
781 		if (ret)
782 			goto out;
783 		cond_resched();
784 	}
785 	ret = 0;
786 out:
787 	return ret;
788 out_free_reserve:
789 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
790 out_free:
791 	extent_clear_unlock_delalloc(inode, async_extent->start,
792 				     async_extent->start +
793 				     async_extent->ram_size - 1,
794 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
795 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
796 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
797 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
798 	kfree(async_extent);
799 	goto again;
800 }
801 
802 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
803 				      u64 num_bytes)
804 {
805 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
806 	struct extent_map *em;
807 	u64 alloc_hint = 0;
808 
809 	read_lock(&em_tree->lock);
810 	em = search_extent_mapping(em_tree, start, num_bytes);
811 	if (em) {
812 		/*
813 		 * if block start isn't an actual block number then find the
814 		 * first block in this inode and use that as a hint.  If that
815 		 * block is also bogus then just don't worry about it.
816 		 */
817 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
818 			free_extent_map(em);
819 			em = search_extent_mapping(em_tree, 0, 0);
820 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
821 				alloc_hint = em->block_start;
822 			if (em)
823 				free_extent_map(em);
824 		} else {
825 			alloc_hint = em->block_start;
826 			free_extent_map(em);
827 		}
828 	}
829 	read_unlock(&em_tree->lock);
830 
831 	return alloc_hint;
832 }
833 
834 /*
835  * when extent_io.c finds a delayed allocation range in the file,
836  * the call backs end up in this code.  The basic idea is to
837  * allocate extents on disk for the range, and create ordered data structs
838  * in ram to track those extents.
839  *
840  * locked_page is the page that writepage had locked already.  We use
841  * it to make sure we don't do extra locks or unlocks.
842  *
843  * *page_started is set to one if we unlock locked_page and do everything
844  * required to start IO on it.  It may be clean and already done with
845  * IO when we return.
846  */
847 static noinline int cow_file_range(struct inode *inode,
848 				   struct page *locked_page,
849 				   u64 start, u64 end, int *page_started,
850 				   unsigned long *nr_written,
851 				   int unlock)
852 {
853 	struct btrfs_root *root = BTRFS_I(inode)->root;
854 	u64 alloc_hint = 0;
855 	u64 num_bytes;
856 	unsigned long ram_size;
857 	u64 disk_num_bytes;
858 	u64 cur_alloc_size;
859 	u64 blocksize = root->sectorsize;
860 	struct btrfs_key ins;
861 	struct extent_map *em;
862 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 	int ret = 0;
864 
865 	if (btrfs_is_free_space_inode(inode)) {
866 		WARN_ON_ONCE(1);
867 		return -EINVAL;
868 	}
869 
870 	num_bytes = ALIGN(end - start + 1, blocksize);
871 	num_bytes = max(blocksize,  num_bytes);
872 	disk_num_bytes = num_bytes;
873 
874 	/* if this is a small write inside eof, kick off defrag */
875 	if (num_bytes < 64 * 1024 &&
876 	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
877 		btrfs_add_inode_defrag(NULL, inode);
878 
879 	if (start == 0) {
880 		/* lets try to make an inline extent */
881 		ret = cow_file_range_inline(root, inode, start, end, 0, 0,
882 					    NULL);
883 		if (ret == 0) {
884 			extent_clear_unlock_delalloc(inode, start, end, NULL,
885 				     EXTENT_LOCKED | EXTENT_DELALLOC |
886 				     EXTENT_DEFRAG, PAGE_UNLOCK |
887 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
888 				     PAGE_END_WRITEBACK);
889 
890 			*nr_written = *nr_written +
891 			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
892 			*page_started = 1;
893 			goto out;
894 		} else if (ret < 0) {
895 			goto out_unlock;
896 		}
897 	}
898 
899 	BUG_ON(disk_num_bytes >
900 	       btrfs_super_total_bytes(root->fs_info->super_copy));
901 
902 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
903 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
904 
905 	while (disk_num_bytes > 0) {
906 		unsigned long op;
907 
908 		cur_alloc_size = disk_num_bytes;
909 		ret = btrfs_reserve_extent(root, cur_alloc_size,
910 					   root->sectorsize, 0, alloc_hint,
911 					   &ins, 1);
912 		if (ret < 0)
913 			goto out_unlock;
914 
915 		em = alloc_extent_map();
916 		if (!em) {
917 			ret = -ENOMEM;
918 			goto out_reserve;
919 		}
920 		em->start = start;
921 		em->orig_start = em->start;
922 		ram_size = ins.offset;
923 		em->len = ins.offset;
924 		em->mod_start = em->start;
925 		em->mod_len = em->len;
926 
927 		em->block_start = ins.objectid;
928 		em->block_len = ins.offset;
929 		em->orig_block_len = ins.offset;
930 		em->ram_bytes = ram_size;
931 		em->bdev = root->fs_info->fs_devices->latest_bdev;
932 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
933 		em->generation = -1;
934 
935 		while (1) {
936 			write_lock(&em_tree->lock);
937 			ret = add_extent_mapping(em_tree, em, 1);
938 			write_unlock(&em_tree->lock);
939 			if (ret != -EEXIST) {
940 				free_extent_map(em);
941 				break;
942 			}
943 			btrfs_drop_extent_cache(inode, start,
944 						start + ram_size - 1, 0);
945 		}
946 		if (ret)
947 			goto out_reserve;
948 
949 		cur_alloc_size = ins.offset;
950 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
951 					       ram_size, cur_alloc_size, 0);
952 		if (ret)
953 			goto out_reserve;
954 
955 		if (root->root_key.objectid ==
956 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
957 			ret = btrfs_reloc_clone_csums(inode, start,
958 						      cur_alloc_size);
959 			if (ret)
960 				goto out_reserve;
961 		}
962 
963 		if (disk_num_bytes < cur_alloc_size)
964 			break;
965 
966 		/* we're not doing compressed IO, don't unlock the first
967 		 * page (which the caller expects to stay locked), don't
968 		 * clear any dirty bits and don't set any writeback bits
969 		 *
970 		 * Do set the Private2 bit so we know this page was properly
971 		 * setup for writepage
972 		 */
973 		op = unlock ? PAGE_UNLOCK : 0;
974 		op |= PAGE_SET_PRIVATE2;
975 
976 		extent_clear_unlock_delalloc(inode, start,
977 					     start + ram_size - 1, locked_page,
978 					     EXTENT_LOCKED | EXTENT_DELALLOC,
979 					     op);
980 		disk_num_bytes -= cur_alloc_size;
981 		num_bytes -= cur_alloc_size;
982 		alloc_hint = ins.objectid + ins.offset;
983 		start += cur_alloc_size;
984 	}
985 out:
986 	return ret;
987 
988 out_reserve:
989 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
990 out_unlock:
991 	extent_clear_unlock_delalloc(inode, start, end, locked_page,
992 				     EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
993 				     EXTENT_DELALLOC | EXTENT_DEFRAG,
994 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
995 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
996 	goto out;
997 }
998 
999 /*
1000  * work queue call back to started compression on a file and pages
1001  */
1002 static noinline void async_cow_start(struct btrfs_work *work)
1003 {
1004 	struct async_cow *async_cow;
1005 	int num_added = 0;
1006 	async_cow = container_of(work, struct async_cow, work);
1007 
1008 	compress_file_range(async_cow->inode, async_cow->locked_page,
1009 			    async_cow->start, async_cow->end, async_cow,
1010 			    &num_added);
1011 	if (num_added == 0) {
1012 		btrfs_add_delayed_iput(async_cow->inode);
1013 		async_cow->inode = NULL;
1014 	}
1015 }
1016 
1017 /*
1018  * work queue call back to submit previously compressed pages
1019  */
1020 static noinline void async_cow_submit(struct btrfs_work *work)
1021 {
1022 	struct async_cow *async_cow;
1023 	struct btrfs_root *root;
1024 	unsigned long nr_pages;
1025 
1026 	async_cow = container_of(work, struct async_cow, work);
1027 
1028 	root = async_cow->root;
1029 	nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1030 		PAGE_CACHE_SHIFT;
1031 
1032 	if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1033 	    5 * 1024 * 1024 &&
1034 	    waitqueue_active(&root->fs_info->async_submit_wait))
1035 		wake_up(&root->fs_info->async_submit_wait);
1036 
1037 	if (async_cow->inode)
1038 		submit_compressed_extents(async_cow->inode, async_cow);
1039 }
1040 
1041 static noinline void async_cow_free(struct btrfs_work *work)
1042 {
1043 	struct async_cow *async_cow;
1044 	async_cow = container_of(work, struct async_cow, work);
1045 	if (async_cow->inode)
1046 		btrfs_add_delayed_iput(async_cow->inode);
1047 	kfree(async_cow);
1048 }
1049 
1050 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1051 				u64 start, u64 end, int *page_started,
1052 				unsigned long *nr_written)
1053 {
1054 	struct async_cow *async_cow;
1055 	struct btrfs_root *root = BTRFS_I(inode)->root;
1056 	unsigned long nr_pages;
1057 	u64 cur_end;
1058 	int limit = 10 * 1024 * 1024;
1059 
1060 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1061 			 1, 0, NULL, GFP_NOFS);
1062 	while (start < end) {
1063 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1064 		BUG_ON(!async_cow); /* -ENOMEM */
1065 		async_cow->inode = igrab(inode);
1066 		async_cow->root = root;
1067 		async_cow->locked_page = locked_page;
1068 		async_cow->start = start;
1069 
1070 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1071 			cur_end = end;
1072 		else
1073 			cur_end = min(end, start + 512 * 1024 - 1);
1074 
1075 		async_cow->end = cur_end;
1076 		INIT_LIST_HEAD(&async_cow->extents);
1077 
1078 		async_cow->work.func = async_cow_start;
1079 		async_cow->work.ordered_func = async_cow_submit;
1080 		async_cow->work.ordered_free = async_cow_free;
1081 		async_cow->work.flags = 0;
1082 
1083 		nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1084 			PAGE_CACHE_SHIFT;
1085 		atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1086 
1087 		btrfs_queue_worker(&root->fs_info->delalloc_workers,
1088 				   &async_cow->work);
1089 
1090 		if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1091 			wait_event(root->fs_info->async_submit_wait,
1092 			   (atomic_read(&root->fs_info->async_delalloc_pages) <
1093 			    limit));
1094 		}
1095 
1096 		while (atomic_read(&root->fs_info->async_submit_draining) &&
1097 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
1098 			wait_event(root->fs_info->async_submit_wait,
1099 			  (atomic_read(&root->fs_info->async_delalloc_pages) ==
1100 			   0));
1101 		}
1102 
1103 		*nr_written += nr_pages;
1104 		start = cur_end + 1;
1105 	}
1106 	*page_started = 1;
1107 	return 0;
1108 }
1109 
1110 static noinline int csum_exist_in_range(struct btrfs_root *root,
1111 					u64 bytenr, u64 num_bytes)
1112 {
1113 	int ret;
1114 	struct btrfs_ordered_sum *sums;
1115 	LIST_HEAD(list);
1116 
1117 	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1118 				       bytenr + num_bytes - 1, &list, 0);
1119 	if (ret == 0 && list_empty(&list))
1120 		return 0;
1121 
1122 	while (!list_empty(&list)) {
1123 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1124 		list_del(&sums->list);
1125 		kfree(sums);
1126 	}
1127 	return 1;
1128 }
1129 
1130 /*
1131  * when nowcow writeback call back.  This checks for snapshots or COW copies
1132  * of the extents that exist in the file, and COWs the file as required.
1133  *
1134  * If no cow copies or snapshots exist, we write directly to the existing
1135  * blocks on disk
1136  */
1137 static noinline int run_delalloc_nocow(struct inode *inode,
1138 				       struct page *locked_page,
1139 			      u64 start, u64 end, int *page_started, int force,
1140 			      unsigned long *nr_written)
1141 {
1142 	struct btrfs_root *root = BTRFS_I(inode)->root;
1143 	struct btrfs_trans_handle *trans;
1144 	struct extent_buffer *leaf;
1145 	struct btrfs_path *path;
1146 	struct btrfs_file_extent_item *fi;
1147 	struct btrfs_key found_key;
1148 	u64 cow_start;
1149 	u64 cur_offset;
1150 	u64 extent_end;
1151 	u64 extent_offset;
1152 	u64 disk_bytenr;
1153 	u64 num_bytes;
1154 	u64 disk_num_bytes;
1155 	u64 ram_bytes;
1156 	int extent_type;
1157 	int ret, err;
1158 	int type;
1159 	int nocow;
1160 	int check_prev = 1;
1161 	bool nolock;
1162 	u64 ino = btrfs_ino(inode);
1163 
1164 	path = btrfs_alloc_path();
1165 	if (!path) {
1166 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1167 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1168 					     EXTENT_DO_ACCOUNTING |
1169 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1170 					     PAGE_CLEAR_DIRTY |
1171 					     PAGE_SET_WRITEBACK |
1172 					     PAGE_END_WRITEBACK);
1173 		return -ENOMEM;
1174 	}
1175 
1176 	nolock = btrfs_is_free_space_inode(inode);
1177 
1178 	if (nolock)
1179 		trans = btrfs_join_transaction_nolock(root);
1180 	else
1181 		trans = btrfs_join_transaction(root);
1182 
1183 	if (IS_ERR(trans)) {
1184 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1185 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1186 					     EXTENT_DO_ACCOUNTING |
1187 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1188 					     PAGE_CLEAR_DIRTY |
1189 					     PAGE_SET_WRITEBACK |
1190 					     PAGE_END_WRITEBACK);
1191 		btrfs_free_path(path);
1192 		return PTR_ERR(trans);
1193 	}
1194 
1195 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1196 
1197 	cow_start = (u64)-1;
1198 	cur_offset = start;
1199 	while (1) {
1200 		ret = btrfs_lookup_file_extent(trans, root, path, ino,
1201 					       cur_offset, 0);
1202 		if (ret < 0)
1203 			goto error;
1204 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1205 			leaf = path->nodes[0];
1206 			btrfs_item_key_to_cpu(leaf, &found_key,
1207 					      path->slots[0] - 1);
1208 			if (found_key.objectid == ino &&
1209 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1210 				path->slots[0]--;
1211 		}
1212 		check_prev = 0;
1213 next_slot:
1214 		leaf = path->nodes[0];
1215 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1216 			ret = btrfs_next_leaf(root, path);
1217 			if (ret < 0)
1218 				goto error;
1219 			if (ret > 0)
1220 				break;
1221 			leaf = path->nodes[0];
1222 		}
1223 
1224 		nocow = 0;
1225 		disk_bytenr = 0;
1226 		num_bytes = 0;
1227 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1228 
1229 		if (found_key.objectid > ino ||
1230 		    found_key.type > BTRFS_EXTENT_DATA_KEY ||
1231 		    found_key.offset > end)
1232 			break;
1233 
1234 		if (found_key.offset > cur_offset) {
1235 			extent_end = found_key.offset;
1236 			extent_type = 0;
1237 			goto out_check;
1238 		}
1239 
1240 		fi = btrfs_item_ptr(leaf, path->slots[0],
1241 				    struct btrfs_file_extent_item);
1242 		extent_type = btrfs_file_extent_type(leaf, fi);
1243 
1244 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1245 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1246 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1247 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1248 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1249 			extent_end = found_key.offset +
1250 				btrfs_file_extent_num_bytes(leaf, fi);
1251 			disk_num_bytes =
1252 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1253 			if (extent_end <= start) {
1254 				path->slots[0]++;
1255 				goto next_slot;
1256 			}
1257 			if (disk_bytenr == 0)
1258 				goto out_check;
1259 			if (btrfs_file_extent_compression(leaf, fi) ||
1260 			    btrfs_file_extent_encryption(leaf, fi) ||
1261 			    btrfs_file_extent_other_encoding(leaf, fi))
1262 				goto out_check;
1263 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1264 				goto out_check;
1265 			if (btrfs_extent_readonly(root, disk_bytenr))
1266 				goto out_check;
1267 			if (btrfs_cross_ref_exist(trans, root, ino,
1268 						  found_key.offset -
1269 						  extent_offset, disk_bytenr))
1270 				goto out_check;
1271 			disk_bytenr += extent_offset;
1272 			disk_bytenr += cur_offset - found_key.offset;
1273 			num_bytes = min(end + 1, extent_end) - cur_offset;
1274 			/*
1275 			 * force cow if csum exists in the range.
1276 			 * this ensure that csum for a given extent are
1277 			 * either valid or do not exist.
1278 			 */
1279 			if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1280 				goto out_check;
1281 			nocow = 1;
1282 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1283 			extent_end = found_key.offset +
1284 				btrfs_file_extent_inline_len(leaf,
1285 						     path->slots[0], fi);
1286 			extent_end = ALIGN(extent_end, root->sectorsize);
1287 		} else {
1288 			BUG_ON(1);
1289 		}
1290 out_check:
1291 		if (extent_end <= start) {
1292 			path->slots[0]++;
1293 			goto next_slot;
1294 		}
1295 		if (!nocow) {
1296 			if (cow_start == (u64)-1)
1297 				cow_start = cur_offset;
1298 			cur_offset = extent_end;
1299 			if (cur_offset > end)
1300 				break;
1301 			path->slots[0]++;
1302 			goto next_slot;
1303 		}
1304 
1305 		btrfs_release_path(path);
1306 		if (cow_start != (u64)-1) {
1307 			ret = cow_file_range(inode, locked_page,
1308 					     cow_start, found_key.offset - 1,
1309 					     page_started, nr_written, 1);
1310 			if (ret)
1311 				goto error;
1312 			cow_start = (u64)-1;
1313 		}
1314 
1315 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1316 			struct extent_map *em;
1317 			struct extent_map_tree *em_tree;
1318 			em_tree = &BTRFS_I(inode)->extent_tree;
1319 			em = alloc_extent_map();
1320 			BUG_ON(!em); /* -ENOMEM */
1321 			em->start = cur_offset;
1322 			em->orig_start = found_key.offset - extent_offset;
1323 			em->len = num_bytes;
1324 			em->block_len = num_bytes;
1325 			em->block_start = disk_bytenr;
1326 			em->orig_block_len = disk_num_bytes;
1327 			em->ram_bytes = ram_bytes;
1328 			em->bdev = root->fs_info->fs_devices->latest_bdev;
1329 			em->mod_start = em->start;
1330 			em->mod_len = em->len;
1331 			set_bit(EXTENT_FLAG_PINNED, &em->flags);
1332 			set_bit(EXTENT_FLAG_FILLING, &em->flags);
1333 			em->generation = -1;
1334 			while (1) {
1335 				write_lock(&em_tree->lock);
1336 				ret = add_extent_mapping(em_tree, em, 1);
1337 				write_unlock(&em_tree->lock);
1338 				if (ret != -EEXIST) {
1339 					free_extent_map(em);
1340 					break;
1341 				}
1342 				btrfs_drop_extent_cache(inode, em->start,
1343 						em->start + em->len - 1, 0);
1344 			}
1345 			type = BTRFS_ORDERED_PREALLOC;
1346 		} else {
1347 			type = BTRFS_ORDERED_NOCOW;
1348 		}
1349 
1350 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1351 					       num_bytes, num_bytes, type);
1352 		BUG_ON(ret); /* -ENOMEM */
1353 
1354 		if (root->root_key.objectid ==
1355 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1356 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1357 						      num_bytes);
1358 			if (ret)
1359 				goto error;
1360 		}
1361 
1362 		extent_clear_unlock_delalloc(inode, cur_offset,
1363 					     cur_offset + num_bytes - 1,
1364 					     locked_page, EXTENT_LOCKED |
1365 					     EXTENT_DELALLOC, PAGE_UNLOCK |
1366 					     PAGE_SET_PRIVATE2);
1367 		cur_offset = extent_end;
1368 		if (cur_offset > end)
1369 			break;
1370 	}
1371 	btrfs_release_path(path);
1372 
1373 	if (cur_offset <= end && cow_start == (u64)-1) {
1374 		cow_start = cur_offset;
1375 		cur_offset = end;
1376 	}
1377 
1378 	if (cow_start != (u64)-1) {
1379 		ret = cow_file_range(inode, locked_page, cow_start, end,
1380 				     page_started, nr_written, 1);
1381 		if (ret)
1382 			goto error;
1383 	}
1384 
1385 error:
1386 	err = btrfs_end_transaction(trans, root);
1387 	if (!ret)
1388 		ret = err;
1389 
1390 	if (ret && cur_offset < end)
1391 		extent_clear_unlock_delalloc(inode, cur_offset, end,
1392 					     locked_page, EXTENT_LOCKED |
1393 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1394 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1395 					     PAGE_CLEAR_DIRTY |
1396 					     PAGE_SET_WRITEBACK |
1397 					     PAGE_END_WRITEBACK);
1398 	btrfs_free_path(path);
1399 	return ret;
1400 }
1401 
1402 /*
1403  * extent_io.c call back to do delayed allocation processing
1404  */
1405 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1406 			      u64 start, u64 end, int *page_started,
1407 			      unsigned long *nr_written)
1408 {
1409 	int ret;
1410 	struct btrfs_root *root = BTRFS_I(inode)->root;
1411 
1412 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1413 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1414 					 page_started, 1, nr_written);
1415 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1416 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1417 					 page_started, 0, nr_written);
1418 	} else if (!btrfs_test_opt(root, COMPRESS) &&
1419 		   !(BTRFS_I(inode)->force_compress) &&
1420 		   !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1421 		ret = cow_file_range(inode, locked_page, start, end,
1422 				      page_started, nr_written, 1);
1423 	} else {
1424 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1425 			&BTRFS_I(inode)->runtime_flags);
1426 		ret = cow_file_range_async(inode, locked_page, start, end,
1427 					   page_started, nr_written);
1428 	}
1429 	return ret;
1430 }
1431 
1432 static void btrfs_split_extent_hook(struct inode *inode,
1433 				    struct extent_state *orig, u64 split)
1434 {
1435 	/* not delalloc, ignore it */
1436 	if (!(orig->state & EXTENT_DELALLOC))
1437 		return;
1438 
1439 	spin_lock(&BTRFS_I(inode)->lock);
1440 	BTRFS_I(inode)->outstanding_extents++;
1441 	spin_unlock(&BTRFS_I(inode)->lock);
1442 }
1443 
1444 /*
1445  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1446  * extents so we can keep track of new extents that are just merged onto old
1447  * extents, such as when we are doing sequential writes, so we can properly
1448  * account for the metadata space we'll need.
1449  */
1450 static void btrfs_merge_extent_hook(struct inode *inode,
1451 				    struct extent_state *new,
1452 				    struct extent_state *other)
1453 {
1454 	/* not delalloc, ignore it */
1455 	if (!(other->state & EXTENT_DELALLOC))
1456 		return;
1457 
1458 	spin_lock(&BTRFS_I(inode)->lock);
1459 	BTRFS_I(inode)->outstanding_extents--;
1460 	spin_unlock(&BTRFS_I(inode)->lock);
1461 }
1462 
1463 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1464 				      struct inode *inode)
1465 {
1466 	spin_lock(&root->delalloc_lock);
1467 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1468 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1469 			      &root->delalloc_inodes);
1470 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471 			&BTRFS_I(inode)->runtime_flags);
1472 		root->nr_delalloc_inodes++;
1473 		if (root->nr_delalloc_inodes == 1) {
1474 			spin_lock(&root->fs_info->delalloc_root_lock);
1475 			BUG_ON(!list_empty(&root->delalloc_root));
1476 			list_add_tail(&root->delalloc_root,
1477 				      &root->fs_info->delalloc_roots);
1478 			spin_unlock(&root->fs_info->delalloc_root_lock);
1479 		}
1480 	}
1481 	spin_unlock(&root->delalloc_lock);
1482 }
1483 
1484 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1485 				     struct inode *inode)
1486 {
1487 	spin_lock(&root->delalloc_lock);
1488 	if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1489 		list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1490 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1491 			  &BTRFS_I(inode)->runtime_flags);
1492 		root->nr_delalloc_inodes--;
1493 		if (!root->nr_delalloc_inodes) {
1494 			spin_lock(&root->fs_info->delalloc_root_lock);
1495 			BUG_ON(list_empty(&root->delalloc_root));
1496 			list_del_init(&root->delalloc_root);
1497 			spin_unlock(&root->fs_info->delalloc_root_lock);
1498 		}
1499 	}
1500 	spin_unlock(&root->delalloc_lock);
1501 }
1502 
1503 /*
1504  * extent_io.c set_bit_hook, used to track delayed allocation
1505  * bytes in this file, and to maintain the list of inodes that
1506  * have pending delalloc work to be done.
1507  */
1508 static void btrfs_set_bit_hook(struct inode *inode,
1509 			       struct extent_state *state, unsigned long *bits)
1510 {
1511 
1512 	/*
1513 	 * set_bit and clear bit hooks normally require _irqsave/restore
1514 	 * but in this case, we are only testing for the DELALLOC
1515 	 * bit, which is only set or cleared with irqs on
1516 	 */
1517 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1518 		struct btrfs_root *root = BTRFS_I(inode)->root;
1519 		u64 len = state->end + 1 - state->start;
1520 		bool do_list = !btrfs_is_free_space_inode(inode);
1521 
1522 		if (*bits & EXTENT_FIRST_DELALLOC) {
1523 			*bits &= ~EXTENT_FIRST_DELALLOC;
1524 		} else {
1525 			spin_lock(&BTRFS_I(inode)->lock);
1526 			BTRFS_I(inode)->outstanding_extents++;
1527 			spin_unlock(&BTRFS_I(inode)->lock);
1528 		}
1529 
1530 		__percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1531 				     root->fs_info->delalloc_batch);
1532 		spin_lock(&BTRFS_I(inode)->lock);
1533 		BTRFS_I(inode)->delalloc_bytes += len;
1534 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1535 					 &BTRFS_I(inode)->runtime_flags))
1536 			btrfs_add_delalloc_inodes(root, inode);
1537 		spin_unlock(&BTRFS_I(inode)->lock);
1538 	}
1539 }
1540 
1541 /*
1542  * extent_io.c clear_bit_hook, see set_bit_hook for why
1543  */
1544 static void btrfs_clear_bit_hook(struct inode *inode,
1545 				 struct extent_state *state,
1546 				 unsigned long *bits)
1547 {
1548 	/*
1549 	 * set_bit and clear bit hooks normally require _irqsave/restore
1550 	 * but in this case, we are only testing for the DELALLOC
1551 	 * bit, which is only set or cleared with irqs on
1552 	 */
1553 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1554 		struct btrfs_root *root = BTRFS_I(inode)->root;
1555 		u64 len = state->end + 1 - state->start;
1556 		bool do_list = !btrfs_is_free_space_inode(inode);
1557 
1558 		if (*bits & EXTENT_FIRST_DELALLOC) {
1559 			*bits &= ~EXTENT_FIRST_DELALLOC;
1560 		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1561 			spin_lock(&BTRFS_I(inode)->lock);
1562 			BTRFS_I(inode)->outstanding_extents--;
1563 			spin_unlock(&BTRFS_I(inode)->lock);
1564 		}
1565 
1566 		/*
1567 		 * We don't reserve metadata space for space cache inodes so we
1568 		 * don't need to call dellalloc_release_metadata if there is an
1569 		 * error.
1570 		 */
1571 		if (*bits & EXTENT_DO_ACCOUNTING &&
1572 		    root != root->fs_info->tree_root)
1573 			btrfs_delalloc_release_metadata(inode, len);
1574 
1575 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1576 		    && do_list && !(state->state & EXTENT_NORESERVE))
1577 			btrfs_free_reserved_data_space(inode, len);
1578 
1579 		__percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1580 				     root->fs_info->delalloc_batch);
1581 		spin_lock(&BTRFS_I(inode)->lock);
1582 		BTRFS_I(inode)->delalloc_bytes -= len;
1583 		if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1584 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1585 			     &BTRFS_I(inode)->runtime_flags))
1586 			btrfs_del_delalloc_inode(root, inode);
1587 		spin_unlock(&BTRFS_I(inode)->lock);
1588 	}
1589 }
1590 
1591 /*
1592  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1593  * we don't create bios that span stripes or chunks
1594  */
1595 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1596 			 size_t size, struct bio *bio,
1597 			 unsigned long bio_flags)
1598 {
1599 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1600 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1601 	u64 length = 0;
1602 	u64 map_length;
1603 	int ret;
1604 
1605 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1606 		return 0;
1607 
1608 	length = bio->bi_iter.bi_size;
1609 	map_length = length;
1610 	ret = btrfs_map_block(root->fs_info, rw, logical,
1611 			      &map_length, NULL, 0);
1612 	/* Will always return 0 with map_multi == NULL */
1613 	BUG_ON(ret < 0);
1614 	if (map_length < length + size)
1615 		return 1;
1616 	return 0;
1617 }
1618 
1619 /*
1620  * in order to insert checksums into the metadata in large chunks,
1621  * we wait until bio submission time.   All the pages in the bio are
1622  * checksummed and sums are attached onto the ordered extent record.
1623  *
1624  * At IO completion time the cums attached on the ordered extent record
1625  * are inserted into the btree
1626  */
1627 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1628 				    struct bio *bio, int mirror_num,
1629 				    unsigned long bio_flags,
1630 				    u64 bio_offset)
1631 {
1632 	struct btrfs_root *root = BTRFS_I(inode)->root;
1633 	int ret = 0;
1634 
1635 	ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1636 	BUG_ON(ret); /* -ENOMEM */
1637 	return 0;
1638 }
1639 
1640 /*
1641  * in order to insert checksums into the metadata in large chunks,
1642  * we wait until bio submission time.   All the pages in the bio are
1643  * checksummed and sums are attached onto the ordered extent record.
1644  *
1645  * At IO completion time the cums attached on the ordered extent record
1646  * are inserted into the btree
1647  */
1648 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1649 			  int mirror_num, unsigned long bio_flags,
1650 			  u64 bio_offset)
1651 {
1652 	struct btrfs_root *root = BTRFS_I(inode)->root;
1653 	int ret;
1654 
1655 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1656 	if (ret)
1657 		bio_endio(bio, ret);
1658 	return ret;
1659 }
1660 
1661 /*
1662  * extent_io.c submission hook. This does the right thing for csum calculation
1663  * on write, or reading the csums from the tree before a read
1664  */
1665 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1666 			  int mirror_num, unsigned long bio_flags,
1667 			  u64 bio_offset)
1668 {
1669 	struct btrfs_root *root = BTRFS_I(inode)->root;
1670 	int ret = 0;
1671 	int skip_sum;
1672 	int metadata = 0;
1673 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1674 
1675 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1676 
1677 	if (btrfs_is_free_space_inode(inode))
1678 		metadata = 2;
1679 
1680 	if (!(rw & REQ_WRITE)) {
1681 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1682 		if (ret)
1683 			goto out;
1684 
1685 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1686 			ret = btrfs_submit_compressed_read(inode, bio,
1687 							   mirror_num,
1688 							   bio_flags);
1689 			goto out;
1690 		} else if (!skip_sum) {
1691 			ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1692 			if (ret)
1693 				goto out;
1694 		}
1695 		goto mapit;
1696 	} else if (async && !skip_sum) {
1697 		/* csum items have already been cloned */
1698 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1699 			goto mapit;
1700 		/* we're doing a write, do the async checksumming */
1701 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1702 				   inode, rw, bio, mirror_num,
1703 				   bio_flags, bio_offset,
1704 				   __btrfs_submit_bio_start,
1705 				   __btrfs_submit_bio_done);
1706 		goto out;
1707 	} else if (!skip_sum) {
1708 		ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1709 		if (ret)
1710 			goto out;
1711 	}
1712 
1713 mapit:
1714 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1715 
1716 out:
1717 	if (ret < 0)
1718 		bio_endio(bio, ret);
1719 	return ret;
1720 }
1721 
1722 /*
1723  * given a list of ordered sums record them in the inode.  This happens
1724  * at IO completion time based on sums calculated at bio submission time.
1725  */
1726 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1727 			     struct inode *inode, u64 file_offset,
1728 			     struct list_head *list)
1729 {
1730 	struct btrfs_ordered_sum *sum;
1731 
1732 	list_for_each_entry(sum, list, list) {
1733 		trans->adding_csums = 1;
1734 		btrfs_csum_file_blocks(trans,
1735 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
1736 		trans->adding_csums = 0;
1737 	}
1738 	return 0;
1739 }
1740 
1741 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1742 			      struct extent_state **cached_state)
1743 {
1744 	WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1745 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1746 				   cached_state, GFP_NOFS);
1747 }
1748 
1749 /* see btrfs_writepage_start_hook for details on why this is required */
1750 struct btrfs_writepage_fixup {
1751 	struct page *page;
1752 	struct btrfs_work work;
1753 };
1754 
1755 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1756 {
1757 	struct btrfs_writepage_fixup *fixup;
1758 	struct btrfs_ordered_extent *ordered;
1759 	struct extent_state *cached_state = NULL;
1760 	struct page *page;
1761 	struct inode *inode;
1762 	u64 page_start;
1763 	u64 page_end;
1764 	int ret;
1765 
1766 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
1767 	page = fixup->page;
1768 again:
1769 	lock_page(page);
1770 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1771 		ClearPageChecked(page);
1772 		goto out_page;
1773 	}
1774 
1775 	inode = page->mapping->host;
1776 	page_start = page_offset(page);
1777 	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1778 
1779 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1780 			 &cached_state);
1781 
1782 	/* already ordered? We're done */
1783 	if (PagePrivate2(page))
1784 		goto out;
1785 
1786 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
1787 	if (ordered) {
1788 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1789 				     page_end, &cached_state, GFP_NOFS);
1790 		unlock_page(page);
1791 		btrfs_start_ordered_extent(inode, ordered, 1);
1792 		btrfs_put_ordered_extent(ordered);
1793 		goto again;
1794 	}
1795 
1796 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1797 	if (ret) {
1798 		mapping_set_error(page->mapping, ret);
1799 		end_extent_writepage(page, ret, page_start, page_end);
1800 		ClearPageChecked(page);
1801 		goto out;
1802 	 }
1803 
1804 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1805 	ClearPageChecked(page);
1806 	set_page_dirty(page);
1807 out:
1808 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1809 			     &cached_state, GFP_NOFS);
1810 out_page:
1811 	unlock_page(page);
1812 	page_cache_release(page);
1813 	kfree(fixup);
1814 }
1815 
1816 /*
1817  * There are a few paths in the higher layers of the kernel that directly
1818  * set the page dirty bit without asking the filesystem if it is a
1819  * good idea.  This causes problems because we want to make sure COW
1820  * properly happens and the data=ordered rules are followed.
1821  *
1822  * In our case any range that doesn't have the ORDERED bit set
1823  * hasn't been properly setup for IO.  We kick off an async process
1824  * to fix it up.  The async helper will wait for ordered extents, set
1825  * the delalloc bit and make it safe to write the page.
1826  */
1827 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1828 {
1829 	struct inode *inode = page->mapping->host;
1830 	struct btrfs_writepage_fixup *fixup;
1831 	struct btrfs_root *root = BTRFS_I(inode)->root;
1832 
1833 	/* this page is properly in the ordered list */
1834 	if (TestClearPagePrivate2(page))
1835 		return 0;
1836 
1837 	if (PageChecked(page))
1838 		return -EAGAIN;
1839 
1840 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1841 	if (!fixup)
1842 		return -EAGAIN;
1843 
1844 	SetPageChecked(page);
1845 	page_cache_get(page);
1846 	fixup->work.func = btrfs_writepage_fixup_worker;
1847 	fixup->page = page;
1848 	btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1849 	return -EBUSY;
1850 }
1851 
1852 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1853 				       struct inode *inode, u64 file_pos,
1854 				       u64 disk_bytenr, u64 disk_num_bytes,
1855 				       u64 num_bytes, u64 ram_bytes,
1856 				       u8 compression, u8 encryption,
1857 				       u16 other_encoding, int extent_type)
1858 {
1859 	struct btrfs_root *root = BTRFS_I(inode)->root;
1860 	struct btrfs_file_extent_item *fi;
1861 	struct btrfs_path *path;
1862 	struct extent_buffer *leaf;
1863 	struct btrfs_key ins;
1864 	int extent_inserted = 0;
1865 	int ret;
1866 
1867 	path = btrfs_alloc_path();
1868 	if (!path)
1869 		return -ENOMEM;
1870 
1871 	/*
1872 	 * we may be replacing one extent in the tree with another.
1873 	 * The new extent is pinned in the extent map, and we don't want
1874 	 * to drop it from the cache until it is completely in the btree.
1875 	 *
1876 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
1877 	 * the caller is expected to unpin it and allow it to be merged
1878 	 * with the others.
1879 	 */
1880 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1881 				   file_pos + num_bytes, NULL, 0,
1882 				   1, sizeof(*fi), &extent_inserted);
1883 	if (ret)
1884 		goto out;
1885 
1886 	if (!extent_inserted) {
1887 		ins.objectid = btrfs_ino(inode);
1888 		ins.offset = file_pos;
1889 		ins.type = BTRFS_EXTENT_DATA_KEY;
1890 
1891 		path->leave_spinning = 1;
1892 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
1893 					      sizeof(*fi));
1894 		if (ret)
1895 			goto out;
1896 	}
1897 	leaf = path->nodes[0];
1898 	fi = btrfs_item_ptr(leaf, path->slots[0],
1899 			    struct btrfs_file_extent_item);
1900 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1901 	btrfs_set_file_extent_type(leaf, fi, extent_type);
1902 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1903 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1904 	btrfs_set_file_extent_offset(leaf, fi, 0);
1905 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1906 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1907 	btrfs_set_file_extent_compression(leaf, fi, compression);
1908 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
1909 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1910 
1911 	btrfs_mark_buffer_dirty(leaf);
1912 	btrfs_release_path(path);
1913 
1914 	inode_add_bytes(inode, num_bytes);
1915 
1916 	ins.objectid = disk_bytenr;
1917 	ins.offset = disk_num_bytes;
1918 	ins.type = BTRFS_EXTENT_ITEM_KEY;
1919 	ret = btrfs_alloc_reserved_file_extent(trans, root,
1920 					root->root_key.objectid,
1921 					btrfs_ino(inode), file_pos, &ins);
1922 out:
1923 	btrfs_free_path(path);
1924 
1925 	return ret;
1926 }
1927 
1928 /* snapshot-aware defrag */
1929 struct sa_defrag_extent_backref {
1930 	struct rb_node node;
1931 	struct old_sa_defrag_extent *old;
1932 	u64 root_id;
1933 	u64 inum;
1934 	u64 file_pos;
1935 	u64 extent_offset;
1936 	u64 num_bytes;
1937 	u64 generation;
1938 };
1939 
1940 struct old_sa_defrag_extent {
1941 	struct list_head list;
1942 	struct new_sa_defrag_extent *new;
1943 
1944 	u64 extent_offset;
1945 	u64 bytenr;
1946 	u64 offset;
1947 	u64 len;
1948 	int count;
1949 };
1950 
1951 struct new_sa_defrag_extent {
1952 	struct rb_root root;
1953 	struct list_head head;
1954 	struct btrfs_path *path;
1955 	struct inode *inode;
1956 	u64 file_pos;
1957 	u64 len;
1958 	u64 bytenr;
1959 	u64 disk_len;
1960 	u8 compress_type;
1961 };
1962 
1963 static int backref_comp(struct sa_defrag_extent_backref *b1,
1964 			struct sa_defrag_extent_backref *b2)
1965 {
1966 	if (b1->root_id < b2->root_id)
1967 		return -1;
1968 	else if (b1->root_id > b2->root_id)
1969 		return 1;
1970 
1971 	if (b1->inum < b2->inum)
1972 		return -1;
1973 	else if (b1->inum > b2->inum)
1974 		return 1;
1975 
1976 	if (b1->file_pos < b2->file_pos)
1977 		return -1;
1978 	else if (b1->file_pos > b2->file_pos)
1979 		return 1;
1980 
1981 	/*
1982 	 * [------------------------------] ===> (a range of space)
1983 	 *     |<--->|   |<---->| =============> (fs/file tree A)
1984 	 * |<---------------------------->| ===> (fs/file tree B)
1985 	 *
1986 	 * A range of space can refer to two file extents in one tree while
1987 	 * refer to only one file extent in another tree.
1988 	 *
1989 	 * So we may process a disk offset more than one time(two extents in A)
1990 	 * and locate at the same extent(one extent in B), then insert two same
1991 	 * backrefs(both refer to the extent in B).
1992 	 */
1993 	return 0;
1994 }
1995 
1996 static void backref_insert(struct rb_root *root,
1997 			   struct sa_defrag_extent_backref *backref)
1998 {
1999 	struct rb_node **p = &root->rb_node;
2000 	struct rb_node *parent = NULL;
2001 	struct sa_defrag_extent_backref *entry;
2002 	int ret;
2003 
2004 	while (*p) {
2005 		parent = *p;
2006 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2007 
2008 		ret = backref_comp(backref, entry);
2009 		if (ret < 0)
2010 			p = &(*p)->rb_left;
2011 		else
2012 			p = &(*p)->rb_right;
2013 	}
2014 
2015 	rb_link_node(&backref->node, parent, p);
2016 	rb_insert_color(&backref->node, root);
2017 }
2018 
2019 /*
2020  * Note the backref might has changed, and in this case we just return 0.
2021  */
2022 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2023 				       void *ctx)
2024 {
2025 	struct btrfs_file_extent_item *extent;
2026 	struct btrfs_fs_info *fs_info;
2027 	struct old_sa_defrag_extent *old = ctx;
2028 	struct new_sa_defrag_extent *new = old->new;
2029 	struct btrfs_path *path = new->path;
2030 	struct btrfs_key key;
2031 	struct btrfs_root *root;
2032 	struct sa_defrag_extent_backref *backref;
2033 	struct extent_buffer *leaf;
2034 	struct inode *inode = new->inode;
2035 	int slot;
2036 	int ret;
2037 	u64 extent_offset;
2038 	u64 num_bytes;
2039 
2040 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2041 	    inum == btrfs_ino(inode))
2042 		return 0;
2043 
2044 	key.objectid = root_id;
2045 	key.type = BTRFS_ROOT_ITEM_KEY;
2046 	key.offset = (u64)-1;
2047 
2048 	fs_info = BTRFS_I(inode)->root->fs_info;
2049 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2050 	if (IS_ERR(root)) {
2051 		if (PTR_ERR(root) == -ENOENT)
2052 			return 0;
2053 		WARN_ON(1);
2054 		pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2055 			 inum, offset, root_id);
2056 		return PTR_ERR(root);
2057 	}
2058 
2059 	key.objectid = inum;
2060 	key.type = BTRFS_EXTENT_DATA_KEY;
2061 	if (offset > (u64)-1 << 32)
2062 		key.offset = 0;
2063 	else
2064 		key.offset = offset;
2065 
2066 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2067 	if (WARN_ON(ret < 0))
2068 		return ret;
2069 	ret = 0;
2070 
2071 	while (1) {
2072 		cond_resched();
2073 
2074 		leaf = path->nodes[0];
2075 		slot = path->slots[0];
2076 
2077 		if (slot >= btrfs_header_nritems(leaf)) {
2078 			ret = btrfs_next_leaf(root, path);
2079 			if (ret < 0) {
2080 				goto out;
2081 			} else if (ret > 0) {
2082 				ret = 0;
2083 				goto out;
2084 			}
2085 			continue;
2086 		}
2087 
2088 		path->slots[0]++;
2089 
2090 		btrfs_item_key_to_cpu(leaf, &key, slot);
2091 
2092 		if (key.objectid > inum)
2093 			goto out;
2094 
2095 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2096 			continue;
2097 
2098 		extent = btrfs_item_ptr(leaf, slot,
2099 					struct btrfs_file_extent_item);
2100 
2101 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2102 			continue;
2103 
2104 		/*
2105 		 * 'offset' refers to the exact key.offset,
2106 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2107 		 * (key.offset - extent_offset).
2108 		 */
2109 		if (key.offset != offset)
2110 			continue;
2111 
2112 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2113 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2114 
2115 		if (extent_offset >= old->extent_offset + old->offset +
2116 		    old->len || extent_offset + num_bytes <=
2117 		    old->extent_offset + old->offset)
2118 			continue;
2119 		break;
2120 	}
2121 
2122 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2123 	if (!backref) {
2124 		ret = -ENOENT;
2125 		goto out;
2126 	}
2127 
2128 	backref->root_id = root_id;
2129 	backref->inum = inum;
2130 	backref->file_pos = offset;
2131 	backref->num_bytes = num_bytes;
2132 	backref->extent_offset = extent_offset;
2133 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2134 	backref->old = old;
2135 	backref_insert(&new->root, backref);
2136 	old->count++;
2137 out:
2138 	btrfs_release_path(path);
2139 	WARN_ON(ret);
2140 	return ret;
2141 }
2142 
2143 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2144 				   struct new_sa_defrag_extent *new)
2145 {
2146 	struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2147 	struct old_sa_defrag_extent *old, *tmp;
2148 	int ret;
2149 
2150 	new->path = path;
2151 
2152 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2153 		ret = iterate_inodes_from_logical(old->bytenr +
2154 						  old->extent_offset, fs_info,
2155 						  path, record_one_backref,
2156 						  old);
2157 		if (ret < 0 && ret != -ENOENT)
2158 			return false;
2159 
2160 		/* no backref to be processed for this extent */
2161 		if (!old->count) {
2162 			list_del(&old->list);
2163 			kfree(old);
2164 		}
2165 	}
2166 
2167 	if (list_empty(&new->head))
2168 		return false;
2169 
2170 	return true;
2171 }
2172 
2173 static int relink_is_mergable(struct extent_buffer *leaf,
2174 			      struct btrfs_file_extent_item *fi,
2175 			      struct new_sa_defrag_extent *new)
2176 {
2177 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2178 		return 0;
2179 
2180 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2181 		return 0;
2182 
2183 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2184 		return 0;
2185 
2186 	if (btrfs_file_extent_encryption(leaf, fi) ||
2187 	    btrfs_file_extent_other_encoding(leaf, fi))
2188 		return 0;
2189 
2190 	return 1;
2191 }
2192 
2193 /*
2194  * Note the backref might has changed, and in this case we just return 0.
2195  */
2196 static noinline int relink_extent_backref(struct btrfs_path *path,
2197 				 struct sa_defrag_extent_backref *prev,
2198 				 struct sa_defrag_extent_backref *backref)
2199 {
2200 	struct btrfs_file_extent_item *extent;
2201 	struct btrfs_file_extent_item *item;
2202 	struct btrfs_ordered_extent *ordered;
2203 	struct btrfs_trans_handle *trans;
2204 	struct btrfs_fs_info *fs_info;
2205 	struct btrfs_root *root;
2206 	struct btrfs_key key;
2207 	struct extent_buffer *leaf;
2208 	struct old_sa_defrag_extent *old = backref->old;
2209 	struct new_sa_defrag_extent *new = old->new;
2210 	struct inode *src_inode = new->inode;
2211 	struct inode *inode;
2212 	struct extent_state *cached = NULL;
2213 	int ret = 0;
2214 	u64 start;
2215 	u64 len;
2216 	u64 lock_start;
2217 	u64 lock_end;
2218 	bool merge = false;
2219 	int index;
2220 
2221 	if (prev && prev->root_id == backref->root_id &&
2222 	    prev->inum == backref->inum &&
2223 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2224 		merge = true;
2225 
2226 	/* step 1: get root */
2227 	key.objectid = backref->root_id;
2228 	key.type = BTRFS_ROOT_ITEM_KEY;
2229 	key.offset = (u64)-1;
2230 
2231 	fs_info = BTRFS_I(src_inode)->root->fs_info;
2232 	index = srcu_read_lock(&fs_info->subvol_srcu);
2233 
2234 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2235 	if (IS_ERR(root)) {
2236 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2237 		if (PTR_ERR(root) == -ENOENT)
2238 			return 0;
2239 		return PTR_ERR(root);
2240 	}
2241 
2242 	/* step 2: get inode */
2243 	key.objectid = backref->inum;
2244 	key.type = BTRFS_INODE_ITEM_KEY;
2245 	key.offset = 0;
2246 
2247 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2248 	if (IS_ERR(inode)) {
2249 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2250 		return 0;
2251 	}
2252 
2253 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2254 
2255 	/* step 3: relink backref */
2256 	lock_start = backref->file_pos;
2257 	lock_end = backref->file_pos + backref->num_bytes - 1;
2258 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2259 			 0, &cached);
2260 
2261 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2262 	if (ordered) {
2263 		btrfs_put_ordered_extent(ordered);
2264 		goto out_unlock;
2265 	}
2266 
2267 	trans = btrfs_join_transaction(root);
2268 	if (IS_ERR(trans)) {
2269 		ret = PTR_ERR(trans);
2270 		goto out_unlock;
2271 	}
2272 
2273 	key.objectid = backref->inum;
2274 	key.type = BTRFS_EXTENT_DATA_KEY;
2275 	key.offset = backref->file_pos;
2276 
2277 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2278 	if (ret < 0) {
2279 		goto out_free_path;
2280 	} else if (ret > 0) {
2281 		ret = 0;
2282 		goto out_free_path;
2283 	}
2284 
2285 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2286 				struct btrfs_file_extent_item);
2287 
2288 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2289 	    backref->generation)
2290 		goto out_free_path;
2291 
2292 	btrfs_release_path(path);
2293 
2294 	start = backref->file_pos;
2295 	if (backref->extent_offset < old->extent_offset + old->offset)
2296 		start += old->extent_offset + old->offset -
2297 			 backref->extent_offset;
2298 
2299 	len = min(backref->extent_offset + backref->num_bytes,
2300 		  old->extent_offset + old->offset + old->len);
2301 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2302 
2303 	ret = btrfs_drop_extents(trans, root, inode, start,
2304 				 start + len, 1);
2305 	if (ret)
2306 		goto out_free_path;
2307 again:
2308 	key.objectid = btrfs_ino(inode);
2309 	key.type = BTRFS_EXTENT_DATA_KEY;
2310 	key.offset = start;
2311 
2312 	path->leave_spinning = 1;
2313 	if (merge) {
2314 		struct btrfs_file_extent_item *fi;
2315 		u64 extent_len;
2316 		struct btrfs_key found_key;
2317 
2318 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2319 		if (ret < 0)
2320 			goto out_free_path;
2321 
2322 		path->slots[0]--;
2323 		leaf = path->nodes[0];
2324 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2325 
2326 		fi = btrfs_item_ptr(leaf, path->slots[0],
2327 				    struct btrfs_file_extent_item);
2328 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2329 
2330 		if (extent_len + found_key.offset == start &&
2331 		    relink_is_mergable(leaf, fi, new)) {
2332 			btrfs_set_file_extent_num_bytes(leaf, fi,
2333 							extent_len + len);
2334 			btrfs_mark_buffer_dirty(leaf);
2335 			inode_add_bytes(inode, len);
2336 
2337 			ret = 1;
2338 			goto out_free_path;
2339 		} else {
2340 			merge = false;
2341 			btrfs_release_path(path);
2342 			goto again;
2343 		}
2344 	}
2345 
2346 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2347 					sizeof(*extent));
2348 	if (ret) {
2349 		btrfs_abort_transaction(trans, root, ret);
2350 		goto out_free_path;
2351 	}
2352 
2353 	leaf = path->nodes[0];
2354 	item = btrfs_item_ptr(leaf, path->slots[0],
2355 				struct btrfs_file_extent_item);
2356 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2357 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2358 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2359 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2360 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2361 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2362 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2363 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2364 	btrfs_set_file_extent_encryption(leaf, item, 0);
2365 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2366 
2367 	btrfs_mark_buffer_dirty(leaf);
2368 	inode_add_bytes(inode, len);
2369 	btrfs_release_path(path);
2370 
2371 	ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2372 			new->disk_len, 0,
2373 			backref->root_id, backref->inum,
2374 			new->file_pos, 0);	/* start - extent_offset */
2375 	if (ret) {
2376 		btrfs_abort_transaction(trans, root, ret);
2377 		goto out_free_path;
2378 	}
2379 
2380 	ret = 1;
2381 out_free_path:
2382 	btrfs_release_path(path);
2383 	path->leave_spinning = 0;
2384 	btrfs_end_transaction(trans, root);
2385 out_unlock:
2386 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2387 			     &cached, GFP_NOFS);
2388 	iput(inode);
2389 	return ret;
2390 }
2391 
2392 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2393 {
2394 	struct old_sa_defrag_extent *old, *tmp;
2395 
2396 	if (!new)
2397 		return;
2398 
2399 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2400 		list_del(&old->list);
2401 		kfree(old);
2402 	}
2403 	kfree(new);
2404 }
2405 
2406 static void relink_file_extents(struct new_sa_defrag_extent *new)
2407 {
2408 	struct btrfs_path *path;
2409 	struct sa_defrag_extent_backref *backref;
2410 	struct sa_defrag_extent_backref *prev = NULL;
2411 	struct inode *inode;
2412 	struct btrfs_root *root;
2413 	struct rb_node *node;
2414 	int ret;
2415 
2416 	inode = new->inode;
2417 	root = BTRFS_I(inode)->root;
2418 
2419 	path = btrfs_alloc_path();
2420 	if (!path)
2421 		return;
2422 
2423 	if (!record_extent_backrefs(path, new)) {
2424 		btrfs_free_path(path);
2425 		goto out;
2426 	}
2427 	btrfs_release_path(path);
2428 
2429 	while (1) {
2430 		node = rb_first(&new->root);
2431 		if (!node)
2432 			break;
2433 		rb_erase(node, &new->root);
2434 
2435 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2436 
2437 		ret = relink_extent_backref(path, prev, backref);
2438 		WARN_ON(ret < 0);
2439 
2440 		kfree(prev);
2441 
2442 		if (ret == 1)
2443 			prev = backref;
2444 		else
2445 			prev = NULL;
2446 		cond_resched();
2447 	}
2448 	kfree(prev);
2449 
2450 	btrfs_free_path(path);
2451 out:
2452 	free_sa_defrag_extent(new);
2453 
2454 	atomic_dec(&root->fs_info->defrag_running);
2455 	wake_up(&root->fs_info->transaction_wait);
2456 }
2457 
2458 static struct new_sa_defrag_extent *
2459 record_old_file_extents(struct inode *inode,
2460 			struct btrfs_ordered_extent *ordered)
2461 {
2462 	struct btrfs_root *root = BTRFS_I(inode)->root;
2463 	struct btrfs_path *path;
2464 	struct btrfs_key key;
2465 	struct old_sa_defrag_extent *old;
2466 	struct new_sa_defrag_extent *new;
2467 	int ret;
2468 
2469 	new = kmalloc(sizeof(*new), GFP_NOFS);
2470 	if (!new)
2471 		return NULL;
2472 
2473 	new->inode = inode;
2474 	new->file_pos = ordered->file_offset;
2475 	new->len = ordered->len;
2476 	new->bytenr = ordered->start;
2477 	new->disk_len = ordered->disk_len;
2478 	new->compress_type = ordered->compress_type;
2479 	new->root = RB_ROOT;
2480 	INIT_LIST_HEAD(&new->head);
2481 
2482 	path = btrfs_alloc_path();
2483 	if (!path)
2484 		goto out_kfree;
2485 
2486 	key.objectid = btrfs_ino(inode);
2487 	key.type = BTRFS_EXTENT_DATA_KEY;
2488 	key.offset = new->file_pos;
2489 
2490 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2491 	if (ret < 0)
2492 		goto out_free_path;
2493 	if (ret > 0 && path->slots[0] > 0)
2494 		path->slots[0]--;
2495 
2496 	/* find out all the old extents for the file range */
2497 	while (1) {
2498 		struct btrfs_file_extent_item *extent;
2499 		struct extent_buffer *l;
2500 		int slot;
2501 		u64 num_bytes;
2502 		u64 offset;
2503 		u64 end;
2504 		u64 disk_bytenr;
2505 		u64 extent_offset;
2506 
2507 		l = path->nodes[0];
2508 		slot = path->slots[0];
2509 
2510 		if (slot >= btrfs_header_nritems(l)) {
2511 			ret = btrfs_next_leaf(root, path);
2512 			if (ret < 0)
2513 				goto out_free_path;
2514 			else if (ret > 0)
2515 				break;
2516 			continue;
2517 		}
2518 
2519 		btrfs_item_key_to_cpu(l, &key, slot);
2520 
2521 		if (key.objectid != btrfs_ino(inode))
2522 			break;
2523 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2524 			break;
2525 		if (key.offset >= new->file_pos + new->len)
2526 			break;
2527 
2528 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2529 
2530 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2531 		if (key.offset + num_bytes < new->file_pos)
2532 			goto next;
2533 
2534 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2535 		if (!disk_bytenr)
2536 			goto next;
2537 
2538 		extent_offset = btrfs_file_extent_offset(l, extent);
2539 
2540 		old = kmalloc(sizeof(*old), GFP_NOFS);
2541 		if (!old)
2542 			goto out_free_path;
2543 
2544 		offset = max(new->file_pos, key.offset);
2545 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2546 
2547 		old->bytenr = disk_bytenr;
2548 		old->extent_offset = extent_offset;
2549 		old->offset = offset - key.offset;
2550 		old->len = end - offset;
2551 		old->new = new;
2552 		old->count = 0;
2553 		list_add_tail(&old->list, &new->head);
2554 next:
2555 		path->slots[0]++;
2556 		cond_resched();
2557 	}
2558 
2559 	btrfs_free_path(path);
2560 	atomic_inc(&root->fs_info->defrag_running);
2561 
2562 	return new;
2563 
2564 out_free_path:
2565 	btrfs_free_path(path);
2566 out_kfree:
2567 	free_sa_defrag_extent(new);
2568 	return NULL;
2569 }
2570 
2571 /* as ordered data IO finishes, this gets called so we can finish
2572  * an ordered extent if the range of bytes in the file it covers are
2573  * fully written.
2574  */
2575 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2576 {
2577 	struct inode *inode = ordered_extent->inode;
2578 	struct btrfs_root *root = BTRFS_I(inode)->root;
2579 	struct btrfs_trans_handle *trans = NULL;
2580 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2581 	struct extent_state *cached_state = NULL;
2582 	struct new_sa_defrag_extent *new = NULL;
2583 	int compress_type = 0;
2584 	int ret = 0;
2585 	u64 logical_len = ordered_extent->len;
2586 	bool nolock;
2587 	bool truncated = false;
2588 
2589 	nolock = btrfs_is_free_space_inode(inode);
2590 
2591 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2592 		ret = -EIO;
2593 		goto out;
2594 	}
2595 
2596 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2597 		truncated = true;
2598 		logical_len = ordered_extent->truncated_len;
2599 		/* Truncated the entire extent, don't bother adding */
2600 		if (!logical_len)
2601 			goto out;
2602 	}
2603 
2604 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2605 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2606 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2607 		if (nolock)
2608 			trans = btrfs_join_transaction_nolock(root);
2609 		else
2610 			trans = btrfs_join_transaction(root);
2611 		if (IS_ERR(trans)) {
2612 			ret = PTR_ERR(trans);
2613 			trans = NULL;
2614 			goto out;
2615 		}
2616 		trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2617 		ret = btrfs_update_inode_fallback(trans, root, inode);
2618 		if (ret) /* -ENOMEM or corruption */
2619 			btrfs_abort_transaction(trans, root, ret);
2620 		goto out;
2621 	}
2622 
2623 	lock_extent_bits(io_tree, ordered_extent->file_offset,
2624 			 ordered_extent->file_offset + ordered_extent->len - 1,
2625 			 0, &cached_state);
2626 
2627 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
2628 			ordered_extent->file_offset + ordered_extent->len - 1,
2629 			EXTENT_DEFRAG, 1, cached_state);
2630 	if (ret) {
2631 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2632 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2633 			/* the inode is shared */
2634 			new = record_old_file_extents(inode, ordered_extent);
2635 
2636 		clear_extent_bit(io_tree, ordered_extent->file_offset,
2637 			ordered_extent->file_offset + ordered_extent->len - 1,
2638 			EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2639 	}
2640 
2641 	if (nolock)
2642 		trans = btrfs_join_transaction_nolock(root);
2643 	else
2644 		trans = btrfs_join_transaction(root);
2645 	if (IS_ERR(trans)) {
2646 		ret = PTR_ERR(trans);
2647 		trans = NULL;
2648 		goto out_unlock;
2649 	}
2650 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2651 
2652 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2653 		compress_type = ordered_extent->compress_type;
2654 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2655 		BUG_ON(compress_type);
2656 		ret = btrfs_mark_extent_written(trans, inode,
2657 						ordered_extent->file_offset,
2658 						ordered_extent->file_offset +
2659 						logical_len);
2660 	} else {
2661 		BUG_ON(root == root->fs_info->tree_root);
2662 		ret = insert_reserved_file_extent(trans, inode,
2663 						ordered_extent->file_offset,
2664 						ordered_extent->start,
2665 						ordered_extent->disk_len,
2666 						logical_len, logical_len,
2667 						compress_type, 0, 0,
2668 						BTRFS_FILE_EXTENT_REG);
2669 	}
2670 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2671 			   ordered_extent->file_offset, ordered_extent->len,
2672 			   trans->transid);
2673 	if (ret < 0) {
2674 		btrfs_abort_transaction(trans, root, ret);
2675 		goto out_unlock;
2676 	}
2677 
2678 	add_pending_csums(trans, inode, ordered_extent->file_offset,
2679 			  &ordered_extent->list);
2680 
2681 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2682 	ret = btrfs_update_inode_fallback(trans, root, inode);
2683 	if (ret) { /* -ENOMEM or corruption */
2684 		btrfs_abort_transaction(trans, root, ret);
2685 		goto out_unlock;
2686 	}
2687 	ret = 0;
2688 out_unlock:
2689 	unlock_extent_cached(io_tree, ordered_extent->file_offset,
2690 			     ordered_extent->file_offset +
2691 			     ordered_extent->len - 1, &cached_state, GFP_NOFS);
2692 out:
2693 	if (root != root->fs_info->tree_root)
2694 		btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2695 	if (trans)
2696 		btrfs_end_transaction(trans, root);
2697 
2698 	if (ret || truncated) {
2699 		u64 start, end;
2700 
2701 		if (truncated)
2702 			start = ordered_extent->file_offset + logical_len;
2703 		else
2704 			start = ordered_extent->file_offset;
2705 		end = ordered_extent->file_offset + ordered_extent->len - 1;
2706 		clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2707 
2708 		/* Drop the cache for the part of the extent we didn't write. */
2709 		btrfs_drop_extent_cache(inode, start, end, 0);
2710 
2711 		/*
2712 		 * If the ordered extent had an IOERR or something else went
2713 		 * wrong we need to return the space for this ordered extent
2714 		 * back to the allocator.  We only free the extent in the
2715 		 * truncated case if we didn't write out the extent at all.
2716 		 */
2717 		if ((ret || !logical_len) &&
2718 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2719 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2720 			btrfs_free_reserved_extent(root, ordered_extent->start,
2721 						   ordered_extent->disk_len);
2722 	}
2723 
2724 
2725 	/*
2726 	 * This needs to be done to make sure anybody waiting knows we are done
2727 	 * updating everything for this ordered extent.
2728 	 */
2729 	btrfs_remove_ordered_extent(inode, ordered_extent);
2730 
2731 	/* for snapshot-aware defrag */
2732 	if (new) {
2733 		if (ret) {
2734 			free_sa_defrag_extent(new);
2735 			atomic_dec(&root->fs_info->defrag_running);
2736 		} else {
2737 			relink_file_extents(new);
2738 		}
2739 	}
2740 
2741 	/* once for us */
2742 	btrfs_put_ordered_extent(ordered_extent);
2743 	/* once for the tree */
2744 	btrfs_put_ordered_extent(ordered_extent);
2745 
2746 	return ret;
2747 }
2748 
2749 static void finish_ordered_fn(struct btrfs_work *work)
2750 {
2751 	struct btrfs_ordered_extent *ordered_extent;
2752 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2753 	btrfs_finish_ordered_io(ordered_extent);
2754 }
2755 
2756 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2757 				struct extent_state *state, int uptodate)
2758 {
2759 	struct inode *inode = page->mapping->host;
2760 	struct btrfs_root *root = BTRFS_I(inode)->root;
2761 	struct btrfs_ordered_extent *ordered_extent = NULL;
2762 	struct btrfs_workers *workers;
2763 
2764 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2765 
2766 	ClearPagePrivate2(page);
2767 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2768 					    end - start + 1, uptodate))
2769 		return 0;
2770 
2771 	ordered_extent->work.func = finish_ordered_fn;
2772 	ordered_extent->work.flags = 0;
2773 
2774 	if (btrfs_is_free_space_inode(inode))
2775 		workers = &root->fs_info->endio_freespace_worker;
2776 	else
2777 		workers = &root->fs_info->endio_write_workers;
2778 	btrfs_queue_worker(workers, &ordered_extent->work);
2779 
2780 	return 0;
2781 }
2782 
2783 /*
2784  * when reads are done, we need to check csums to verify the data is correct
2785  * if there's a match, we allow the bio to finish.  If not, the code in
2786  * extent_io.c will try to find good copies for us.
2787  */
2788 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2789 				      u64 phy_offset, struct page *page,
2790 				      u64 start, u64 end, int mirror)
2791 {
2792 	size_t offset = start - page_offset(page);
2793 	struct inode *inode = page->mapping->host;
2794 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2795 	char *kaddr;
2796 	struct btrfs_root *root = BTRFS_I(inode)->root;
2797 	u32 csum_expected;
2798 	u32 csum = ~(u32)0;
2799 	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2800 	                              DEFAULT_RATELIMIT_BURST);
2801 
2802 	if (PageChecked(page)) {
2803 		ClearPageChecked(page);
2804 		goto good;
2805 	}
2806 
2807 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2808 		goto good;
2809 
2810 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2811 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2812 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2813 				  GFP_NOFS);
2814 		return 0;
2815 	}
2816 
2817 	phy_offset >>= inode->i_sb->s_blocksize_bits;
2818 	csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2819 
2820 	kaddr = kmap_atomic(page);
2821 	csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2822 	btrfs_csum_final(csum, (char *)&csum);
2823 	if (csum != csum_expected)
2824 		goto zeroit;
2825 
2826 	kunmap_atomic(kaddr);
2827 good:
2828 	return 0;
2829 
2830 zeroit:
2831 	if (__ratelimit(&_rs))
2832 		btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2833 			btrfs_ino(page->mapping->host), start, csum, csum_expected);
2834 	memset(kaddr + offset, 1, end - start + 1);
2835 	flush_dcache_page(page);
2836 	kunmap_atomic(kaddr);
2837 	if (csum_expected == 0)
2838 		return 0;
2839 	return -EIO;
2840 }
2841 
2842 struct delayed_iput {
2843 	struct list_head list;
2844 	struct inode *inode;
2845 };
2846 
2847 /* JDM: If this is fs-wide, why can't we add a pointer to
2848  * btrfs_inode instead and avoid the allocation? */
2849 void btrfs_add_delayed_iput(struct inode *inode)
2850 {
2851 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2852 	struct delayed_iput *delayed;
2853 
2854 	if (atomic_add_unless(&inode->i_count, -1, 1))
2855 		return;
2856 
2857 	delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2858 	delayed->inode = inode;
2859 
2860 	spin_lock(&fs_info->delayed_iput_lock);
2861 	list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2862 	spin_unlock(&fs_info->delayed_iput_lock);
2863 }
2864 
2865 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2866 {
2867 	LIST_HEAD(list);
2868 	struct btrfs_fs_info *fs_info = root->fs_info;
2869 	struct delayed_iput *delayed;
2870 	int empty;
2871 
2872 	spin_lock(&fs_info->delayed_iput_lock);
2873 	empty = list_empty(&fs_info->delayed_iputs);
2874 	spin_unlock(&fs_info->delayed_iput_lock);
2875 	if (empty)
2876 		return;
2877 
2878 	spin_lock(&fs_info->delayed_iput_lock);
2879 	list_splice_init(&fs_info->delayed_iputs, &list);
2880 	spin_unlock(&fs_info->delayed_iput_lock);
2881 
2882 	while (!list_empty(&list)) {
2883 		delayed = list_entry(list.next, struct delayed_iput, list);
2884 		list_del(&delayed->list);
2885 		iput(delayed->inode);
2886 		kfree(delayed);
2887 	}
2888 }
2889 
2890 /*
2891  * This is called in transaction commit time. If there are no orphan
2892  * files in the subvolume, it removes orphan item and frees block_rsv
2893  * structure.
2894  */
2895 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2896 			      struct btrfs_root *root)
2897 {
2898 	struct btrfs_block_rsv *block_rsv;
2899 	int ret;
2900 
2901 	if (atomic_read(&root->orphan_inodes) ||
2902 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2903 		return;
2904 
2905 	spin_lock(&root->orphan_lock);
2906 	if (atomic_read(&root->orphan_inodes)) {
2907 		spin_unlock(&root->orphan_lock);
2908 		return;
2909 	}
2910 
2911 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2912 		spin_unlock(&root->orphan_lock);
2913 		return;
2914 	}
2915 
2916 	block_rsv = root->orphan_block_rsv;
2917 	root->orphan_block_rsv = NULL;
2918 	spin_unlock(&root->orphan_lock);
2919 
2920 	if (root->orphan_item_inserted &&
2921 	    btrfs_root_refs(&root->root_item) > 0) {
2922 		ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2923 					    root->root_key.objectid);
2924 		if (ret)
2925 			btrfs_abort_transaction(trans, root, ret);
2926 		else
2927 			root->orphan_item_inserted = 0;
2928 	}
2929 
2930 	if (block_rsv) {
2931 		WARN_ON(block_rsv->size > 0);
2932 		btrfs_free_block_rsv(root, block_rsv);
2933 	}
2934 }
2935 
2936 /*
2937  * This creates an orphan entry for the given inode in case something goes
2938  * wrong in the middle of an unlink/truncate.
2939  *
2940  * NOTE: caller of this function should reserve 5 units of metadata for
2941  *	 this function.
2942  */
2943 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2944 {
2945 	struct btrfs_root *root = BTRFS_I(inode)->root;
2946 	struct btrfs_block_rsv *block_rsv = NULL;
2947 	int reserve = 0;
2948 	int insert = 0;
2949 	int ret;
2950 
2951 	if (!root->orphan_block_rsv) {
2952 		block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2953 		if (!block_rsv)
2954 			return -ENOMEM;
2955 	}
2956 
2957 	spin_lock(&root->orphan_lock);
2958 	if (!root->orphan_block_rsv) {
2959 		root->orphan_block_rsv = block_rsv;
2960 	} else if (block_rsv) {
2961 		btrfs_free_block_rsv(root, block_rsv);
2962 		block_rsv = NULL;
2963 	}
2964 
2965 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2966 			      &BTRFS_I(inode)->runtime_flags)) {
2967 #if 0
2968 		/*
2969 		 * For proper ENOSPC handling, we should do orphan
2970 		 * cleanup when mounting. But this introduces backward
2971 		 * compatibility issue.
2972 		 */
2973 		if (!xchg(&root->orphan_item_inserted, 1))
2974 			insert = 2;
2975 		else
2976 			insert = 1;
2977 #endif
2978 		insert = 1;
2979 		atomic_inc(&root->orphan_inodes);
2980 	}
2981 
2982 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2983 			      &BTRFS_I(inode)->runtime_flags))
2984 		reserve = 1;
2985 	spin_unlock(&root->orphan_lock);
2986 
2987 	/* grab metadata reservation from transaction handle */
2988 	if (reserve) {
2989 		ret = btrfs_orphan_reserve_metadata(trans, inode);
2990 		BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2991 	}
2992 
2993 	/* insert an orphan item to track this unlinked/truncated file */
2994 	if (insert >= 1) {
2995 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2996 		if (ret) {
2997 			atomic_dec(&root->orphan_inodes);
2998 			if (reserve) {
2999 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3000 					  &BTRFS_I(inode)->runtime_flags);
3001 				btrfs_orphan_release_metadata(inode);
3002 			}
3003 			if (ret != -EEXIST) {
3004 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3005 					  &BTRFS_I(inode)->runtime_flags);
3006 				btrfs_abort_transaction(trans, root, ret);
3007 				return ret;
3008 			}
3009 		}
3010 		ret = 0;
3011 	}
3012 
3013 	/* insert an orphan item to track subvolume contains orphan files */
3014 	if (insert >= 2) {
3015 		ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3016 					       root->root_key.objectid);
3017 		if (ret && ret != -EEXIST) {
3018 			btrfs_abort_transaction(trans, root, ret);
3019 			return ret;
3020 		}
3021 	}
3022 	return 0;
3023 }
3024 
3025 /*
3026  * We have done the truncate/delete so we can go ahead and remove the orphan
3027  * item for this particular inode.
3028  */
3029 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3030 			    struct inode *inode)
3031 {
3032 	struct btrfs_root *root = BTRFS_I(inode)->root;
3033 	int delete_item = 0;
3034 	int release_rsv = 0;
3035 	int ret = 0;
3036 
3037 	spin_lock(&root->orphan_lock);
3038 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3039 			       &BTRFS_I(inode)->runtime_flags))
3040 		delete_item = 1;
3041 
3042 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3043 			       &BTRFS_I(inode)->runtime_flags))
3044 		release_rsv = 1;
3045 	spin_unlock(&root->orphan_lock);
3046 
3047 	if (delete_item) {
3048 		atomic_dec(&root->orphan_inodes);
3049 		if (trans)
3050 			ret = btrfs_del_orphan_item(trans, root,
3051 						    btrfs_ino(inode));
3052 	}
3053 
3054 	if (release_rsv)
3055 		btrfs_orphan_release_metadata(inode);
3056 
3057 	return ret;
3058 }
3059 
3060 /*
3061  * this cleans up any orphans that may be left on the list from the last use
3062  * of this root.
3063  */
3064 int btrfs_orphan_cleanup(struct btrfs_root *root)
3065 {
3066 	struct btrfs_path *path;
3067 	struct extent_buffer *leaf;
3068 	struct btrfs_key key, found_key;
3069 	struct btrfs_trans_handle *trans;
3070 	struct inode *inode;
3071 	u64 last_objectid = 0;
3072 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3073 
3074 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3075 		return 0;
3076 
3077 	path = btrfs_alloc_path();
3078 	if (!path) {
3079 		ret = -ENOMEM;
3080 		goto out;
3081 	}
3082 	path->reada = -1;
3083 
3084 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3085 	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3086 	key.offset = (u64)-1;
3087 
3088 	while (1) {
3089 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3090 		if (ret < 0)
3091 			goto out;
3092 
3093 		/*
3094 		 * if ret == 0 means we found what we were searching for, which
3095 		 * is weird, but possible, so only screw with path if we didn't
3096 		 * find the key and see if we have stuff that matches
3097 		 */
3098 		if (ret > 0) {
3099 			ret = 0;
3100 			if (path->slots[0] == 0)
3101 				break;
3102 			path->slots[0]--;
3103 		}
3104 
3105 		/* pull out the item */
3106 		leaf = path->nodes[0];
3107 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3108 
3109 		/* make sure the item matches what we want */
3110 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3111 			break;
3112 		if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3113 			break;
3114 
3115 		/* release the path since we're done with it */
3116 		btrfs_release_path(path);
3117 
3118 		/*
3119 		 * this is where we are basically btrfs_lookup, without the
3120 		 * crossing root thing.  we store the inode number in the
3121 		 * offset of the orphan item.
3122 		 */
3123 
3124 		if (found_key.offset == last_objectid) {
3125 			btrfs_err(root->fs_info,
3126 				"Error removing orphan entry, stopping orphan cleanup");
3127 			ret = -EINVAL;
3128 			goto out;
3129 		}
3130 
3131 		last_objectid = found_key.offset;
3132 
3133 		found_key.objectid = found_key.offset;
3134 		found_key.type = BTRFS_INODE_ITEM_KEY;
3135 		found_key.offset = 0;
3136 		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3137 		ret = PTR_ERR_OR_ZERO(inode);
3138 		if (ret && ret != -ESTALE)
3139 			goto out;
3140 
3141 		if (ret == -ESTALE && root == root->fs_info->tree_root) {
3142 			struct btrfs_root *dead_root;
3143 			struct btrfs_fs_info *fs_info = root->fs_info;
3144 			int is_dead_root = 0;
3145 
3146 			/*
3147 			 * this is an orphan in the tree root. Currently these
3148 			 * could come from 2 sources:
3149 			 *  a) a snapshot deletion in progress
3150 			 *  b) a free space cache inode
3151 			 * We need to distinguish those two, as the snapshot
3152 			 * orphan must not get deleted.
3153 			 * find_dead_roots already ran before us, so if this
3154 			 * is a snapshot deletion, we should find the root
3155 			 * in the dead_roots list
3156 			 */
3157 			spin_lock(&fs_info->trans_lock);
3158 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3159 					    root_list) {
3160 				if (dead_root->root_key.objectid ==
3161 				    found_key.objectid) {
3162 					is_dead_root = 1;
3163 					break;
3164 				}
3165 			}
3166 			spin_unlock(&fs_info->trans_lock);
3167 			if (is_dead_root) {
3168 				/* prevent this orphan from being found again */
3169 				key.offset = found_key.objectid - 1;
3170 				continue;
3171 			}
3172 		}
3173 		/*
3174 		 * Inode is already gone but the orphan item is still there,
3175 		 * kill the orphan item.
3176 		 */
3177 		if (ret == -ESTALE) {
3178 			trans = btrfs_start_transaction(root, 1);
3179 			if (IS_ERR(trans)) {
3180 				ret = PTR_ERR(trans);
3181 				goto out;
3182 			}
3183 			btrfs_debug(root->fs_info, "auto deleting %Lu",
3184 				found_key.objectid);
3185 			ret = btrfs_del_orphan_item(trans, root,
3186 						    found_key.objectid);
3187 			btrfs_end_transaction(trans, root);
3188 			if (ret)
3189 				goto out;
3190 			continue;
3191 		}
3192 
3193 		/*
3194 		 * add this inode to the orphan list so btrfs_orphan_del does
3195 		 * the proper thing when we hit it
3196 		 */
3197 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3198 			&BTRFS_I(inode)->runtime_flags);
3199 		atomic_inc(&root->orphan_inodes);
3200 
3201 		/* if we have links, this was a truncate, lets do that */
3202 		if (inode->i_nlink) {
3203 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3204 				iput(inode);
3205 				continue;
3206 			}
3207 			nr_truncate++;
3208 
3209 			/* 1 for the orphan item deletion. */
3210 			trans = btrfs_start_transaction(root, 1);
3211 			if (IS_ERR(trans)) {
3212 				iput(inode);
3213 				ret = PTR_ERR(trans);
3214 				goto out;
3215 			}
3216 			ret = btrfs_orphan_add(trans, inode);
3217 			btrfs_end_transaction(trans, root);
3218 			if (ret) {
3219 				iput(inode);
3220 				goto out;
3221 			}
3222 
3223 			ret = btrfs_truncate(inode);
3224 			if (ret)
3225 				btrfs_orphan_del(NULL, inode);
3226 		} else {
3227 			nr_unlink++;
3228 		}
3229 
3230 		/* this will do delete_inode and everything for us */
3231 		iput(inode);
3232 		if (ret)
3233 			goto out;
3234 	}
3235 	/* release the path since we're done with it */
3236 	btrfs_release_path(path);
3237 
3238 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3239 
3240 	if (root->orphan_block_rsv)
3241 		btrfs_block_rsv_release(root, root->orphan_block_rsv,
3242 					(u64)-1);
3243 
3244 	if (root->orphan_block_rsv || root->orphan_item_inserted) {
3245 		trans = btrfs_join_transaction(root);
3246 		if (!IS_ERR(trans))
3247 			btrfs_end_transaction(trans, root);
3248 	}
3249 
3250 	if (nr_unlink)
3251 		btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3252 	if (nr_truncate)
3253 		btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3254 
3255 out:
3256 	if (ret)
3257 		btrfs_crit(root->fs_info,
3258 			"could not do orphan cleanup %d", ret);
3259 	btrfs_free_path(path);
3260 	return ret;
3261 }
3262 
3263 /*
3264  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3265  * don't find any xattrs, we know there can't be any acls.
3266  *
3267  * slot is the slot the inode is in, objectid is the objectid of the inode
3268  */
3269 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3270 					  int slot, u64 objectid,
3271 					  int *first_xattr_slot)
3272 {
3273 	u32 nritems = btrfs_header_nritems(leaf);
3274 	struct btrfs_key found_key;
3275 	static u64 xattr_access = 0;
3276 	static u64 xattr_default = 0;
3277 	int scanned = 0;
3278 
3279 	if (!xattr_access) {
3280 		xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3281 					strlen(POSIX_ACL_XATTR_ACCESS));
3282 		xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3283 					strlen(POSIX_ACL_XATTR_DEFAULT));
3284 	}
3285 
3286 	slot++;
3287 	*first_xattr_slot = -1;
3288 	while (slot < nritems) {
3289 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3290 
3291 		/* we found a different objectid, there must not be acls */
3292 		if (found_key.objectid != objectid)
3293 			return 0;
3294 
3295 		/* we found an xattr, assume we've got an acl */
3296 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3297 			if (*first_xattr_slot == -1)
3298 				*first_xattr_slot = slot;
3299 			if (found_key.offset == xattr_access ||
3300 			    found_key.offset == xattr_default)
3301 				return 1;
3302 		}
3303 
3304 		/*
3305 		 * we found a key greater than an xattr key, there can't
3306 		 * be any acls later on
3307 		 */
3308 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3309 			return 0;
3310 
3311 		slot++;
3312 		scanned++;
3313 
3314 		/*
3315 		 * it goes inode, inode backrefs, xattrs, extents,
3316 		 * so if there are a ton of hard links to an inode there can
3317 		 * be a lot of backrefs.  Don't waste time searching too hard,
3318 		 * this is just an optimization
3319 		 */
3320 		if (scanned >= 8)
3321 			break;
3322 	}
3323 	/* we hit the end of the leaf before we found an xattr or
3324 	 * something larger than an xattr.  We have to assume the inode
3325 	 * has acls
3326 	 */
3327 	if (*first_xattr_slot == -1)
3328 		*first_xattr_slot = slot;
3329 	return 1;
3330 }
3331 
3332 /*
3333  * read an inode from the btree into the in-memory inode
3334  */
3335 static void btrfs_read_locked_inode(struct inode *inode)
3336 {
3337 	struct btrfs_path *path;
3338 	struct extent_buffer *leaf;
3339 	struct btrfs_inode_item *inode_item;
3340 	struct btrfs_timespec *tspec;
3341 	struct btrfs_root *root = BTRFS_I(inode)->root;
3342 	struct btrfs_key location;
3343 	unsigned long ptr;
3344 	int maybe_acls;
3345 	u32 rdev;
3346 	int ret;
3347 	bool filled = false;
3348 	int first_xattr_slot;
3349 
3350 	ret = btrfs_fill_inode(inode, &rdev);
3351 	if (!ret)
3352 		filled = true;
3353 
3354 	path = btrfs_alloc_path();
3355 	if (!path)
3356 		goto make_bad;
3357 
3358 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3359 
3360 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3361 	if (ret)
3362 		goto make_bad;
3363 
3364 	leaf = path->nodes[0];
3365 
3366 	if (filled)
3367 		goto cache_index;
3368 
3369 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3370 				    struct btrfs_inode_item);
3371 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3372 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3373 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3374 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3375 	btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3376 
3377 	tspec = btrfs_inode_atime(inode_item);
3378 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3379 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3380 
3381 	tspec = btrfs_inode_mtime(inode_item);
3382 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3383 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3384 
3385 	tspec = btrfs_inode_ctime(inode_item);
3386 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3387 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3388 
3389 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3390 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3391 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3392 
3393 	/*
3394 	 * If we were modified in the current generation and evicted from memory
3395 	 * and then re-read we need to do a full sync since we don't have any
3396 	 * idea about which extents were modified before we were evicted from
3397 	 * cache.
3398 	 */
3399 	if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3400 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3401 			&BTRFS_I(inode)->runtime_flags);
3402 
3403 	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3404 	inode->i_generation = BTRFS_I(inode)->generation;
3405 	inode->i_rdev = 0;
3406 	rdev = btrfs_inode_rdev(leaf, inode_item);
3407 
3408 	BTRFS_I(inode)->index_cnt = (u64)-1;
3409 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3410 
3411 cache_index:
3412 	path->slots[0]++;
3413 	if (inode->i_nlink != 1 ||
3414 	    path->slots[0] >= btrfs_header_nritems(leaf))
3415 		goto cache_acl;
3416 
3417 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3418 	if (location.objectid != btrfs_ino(inode))
3419 		goto cache_acl;
3420 
3421 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3422 	if (location.type == BTRFS_INODE_REF_KEY) {
3423 		struct btrfs_inode_ref *ref;
3424 
3425 		ref = (struct btrfs_inode_ref *)ptr;
3426 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3427 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3428 		struct btrfs_inode_extref *extref;
3429 
3430 		extref = (struct btrfs_inode_extref *)ptr;
3431 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3432 								     extref);
3433 	}
3434 cache_acl:
3435 	/*
3436 	 * try to precache a NULL acl entry for files that don't have
3437 	 * any xattrs or acls
3438 	 */
3439 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3440 					   btrfs_ino(inode), &first_xattr_slot);
3441 	if (first_xattr_slot != -1) {
3442 		path->slots[0] = first_xattr_slot;
3443 		ret = btrfs_load_inode_props(inode, path);
3444 		if (ret)
3445 			btrfs_err(root->fs_info,
3446 				  "error loading props for ino %llu (root %llu): %d\n",
3447 				  btrfs_ino(inode),
3448 				  root->root_key.objectid, ret);
3449 	}
3450 	btrfs_free_path(path);
3451 
3452 	if (!maybe_acls)
3453 		cache_no_acl(inode);
3454 
3455 	switch (inode->i_mode & S_IFMT) {
3456 	case S_IFREG:
3457 		inode->i_mapping->a_ops = &btrfs_aops;
3458 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3459 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3460 		inode->i_fop = &btrfs_file_operations;
3461 		inode->i_op = &btrfs_file_inode_operations;
3462 		break;
3463 	case S_IFDIR:
3464 		inode->i_fop = &btrfs_dir_file_operations;
3465 		if (root == root->fs_info->tree_root)
3466 			inode->i_op = &btrfs_dir_ro_inode_operations;
3467 		else
3468 			inode->i_op = &btrfs_dir_inode_operations;
3469 		break;
3470 	case S_IFLNK:
3471 		inode->i_op = &btrfs_symlink_inode_operations;
3472 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3473 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3474 		break;
3475 	default:
3476 		inode->i_op = &btrfs_special_inode_operations;
3477 		init_special_inode(inode, inode->i_mode, rdev);
3478 		break;
3479 	}
3480 
3481 	btrfs_update_iflags(inode);
3482 	return;
3483 
3484 make_bad:
3485 	btrfs_free_path(path);
3486 	make_bad_inode(inode);
3487 }
3488 
3489 /*
3490  * given a leaf and an inode, copy the inode fields into the leaf
3491  */
3492 static void fill_inode_item(struct btrfs_trans_handle *trans,
3493 			    struct extent_buffer *leaf,
3494 			    struct btrfs_inode_item *item,
3495 			    struct inode *inode)
3496 {
3497 	struct btrfs_map_token token;
3498 
3499 	btrfs_init_map_token(&token);
3500 
3501 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3502 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3503 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3504 				   &token);
3505 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3506 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3507 
3508 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3509 				     inode->i_atime.tv_sec, &token);
3510 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3511 				      inode->i_atime.tv_nsec, &token);
3512 
3513 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3514 				     inode->i_mtime.tv_sec, &token);
3515 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3516 				      inode->i_mtime.tv_nsec, &token);
3517 
3518 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3519 				     inode->i_ctime.tv_sec, &token);
3520 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3521 				      inode->i_ctime.tv_nsec, &token);
3522 
3523 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3524 				     &token);
3525 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3526 					 &token);
3527 	btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3528 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3529 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3530 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3531 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3532 }
3533 
3534 /*
3535  * copy everything in the in-memory inode into the btree.
3536  */
3537 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3538 				struct btrfs_root *root, struct inode *inode)
3539 {
3540 	struct btrfs_inode_item *inode_item;
3541 	struct btrfs_path *path;
3542 	struct extent_buffer *leaf;
3543 	int ret;
3544 
3545 	path = btrfs_alloc_path();
3546 	if (!path)
3547 		return -ENOMEM;
3548 
3549 	path->leave_spinning = 1;
3550 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3551 				 1);
3552 	if (ret) {
3553 		if (ret > 0)
3554 			ret = -ENOENT;
3555 		goto failed;
3556 	}
3557 
3558 	leaf = path->nodes[0];
3559 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3560 				    struct btrfs_inode_item);
3561 
3562 	fill_inode_item(trans, leaf, inode_item, inode);
3563 	btrfs_mark_buffer_dirty(leaf);
3564 	btrfs_set_inode_last_trans(trans, inode);
3565 	ret = 0;
3566 failed:
3567 	btrfs_free_path(path);
3568 	return ret;
3569 }
3570 
3571 /*
3572  * copy everything in the in-memory inode into the btree.
3573  */
3574 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3575 				struct btrfs_root *root, struct inode *inode)
3576 {
3577 	int ret;
3578 
3579 	/*
3580 	 * If the inode is a free space inode, we can deadlock during commit
3581 	 * if we put it into the delayed code.
3582 	 *
3583 	 * The data relocation inode should also be directly updated
3584 	 * without delay
3585 	 */
3586 	if (!btrfs_is_free_space_inode(inode)
3587 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3588 		btrfs_update_root_times(trans, root);
3589 
3590 		ret = btrfs_delayed_update_inode(trans, root, inode);
3591 		if (!ret)
3592 			btrfs_set_inode_last_trans(trans, inode);
3593 		return ret;
3594 	}
3595 
3596 	return btrfs_update_inode_item(trans, root, inode);
3597 }
3598 
3599 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3600 					 struct btrfs_root *root,
3601 					 struct inode *inode)
3602 {
3603 	int ret;
3604 
3605 	ret = btrfs_update_inode(trans, root, inode);
3606 	if (ret == -ENOSPC)
3607 		return btrfs_update_inode_item(trans, root, inode);
3608 	return ret;
3609 }
3610 
3611 /*
3612  * unlink helper that gets used here in inode.c and in the tree logging
3613  * recovery code.  It remove a link in a directory with a given name, and
3614  * also drops the back refs in the inode to the directory
3615  */
3616 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3617 				struct btrfs_root *root,
3618 				struct inode *dir, struct inode *inode,
3619 				const char *name, int name_len)
3620 {
3621 	struct btrfs_path *path;
3622 	int ret = 0;
3623 	struct extent_buffer *leaf;
3624 	struct btrfs_dir_item *di;
3625 	struct btrfs_key key;
3626 	u64 index;
3627 	u64 ino = btrfs_ino(inode);
3628 	u64 dir_ino = btrfs_ino(dir);
3629 
3630 	path = btrfs_alloc_path();
3631 	if (!path) {
3632 		ret = -ENOMEM;
3633 		goto out;
3634 	}
3635 
3636 	path->leave_spinning = 1;
3637 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3638 				    name, name_len, -1);
3639 	if (IS_ERR(di)) {
3640 		ret = PTR_ERR(di);
3641 		goto err;
3642 	}
3643 	if (!di) {
3644 		ret = -ENOENT;
3645 		goto err;
3646 	}
3647 	leaf = path->nodes[0];
3648 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3649 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3650 	if (ret)
3651 		goto err;
3652 	btrfs_release_path(path);
3653 
3654 	/*
3655 	 * If we don't have dir index, we have to get it by looking up
3656 	 * the inode ref, since we get the inode ref, remove it directly,
3657 	 * it is unnecessary to do delayed deletion.
3658 	 *
3659 	 * But if we have dir index, needn't search inode ref to get it.
3660 	 * Since the inode ref is close to the inode item, it is better
3661 	 * that we delay to delete it, and just do this deletion when
3662 	 * we update the inode item.
3663 	 */
3664 	if (BTRFS_I(inode)->dir_index) {
3665 		ret = btrfs_delayed_delete_inode_ref(inode);
3666 		if (!ret) {
3667 			index = BTRFS_I(inode)->dir_index;
3668 			goto skip_backref;
3669 		}
3670 	}
3671 
3672 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3673 				  dir_ino, &index);
3674 	if (ret) {
3675 		btrfs_info(root->fs_info,
3676 			"failed to delete reference to %.*s, inode %llu parent %llu",
3677 			name_len, name, ino, dir_ino);
3678 		btrfs_abort_transaction(trans, root, ret);
3679 		goto err;
3680 	}
3681 skip_backref:
3682 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3683 	if (ret) {
3684 		btrfs_abort_transaction(trans, root, ret);
3685 		goto err;
3686 	}
3687 
3688 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3689 					 inode, dir_ino);
3690 	if (ret != 0 && ret != -ENOENT) {
3691 		btrfs_abort_transaction(trans, root, ret);
3692 		goto err;
3693 	}
3694 
3695 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3696 					   dir, index);
3697 	if (ret == -ENOENT)
3698 		ret = 0;
3699 	else if (ret)
3700 		btrfs_abort_transaction(trans, root, ret);
3701 err:
3702 	btrfs_free_path(path);
3703 	if (ret)
3704 		goto out;
3705 
3706 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3707 	inode_inc_iversion(inode);
3708 	inode_inc_iversion(dir);
3709 	inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3710 	ret = btrfs_update_inode(trans, root, dir);
3711 out:
3712 	return ret;
3713 }
3714 
3715 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3716 		       struct btrfs_root *root,
3717 		       struct inode *dir, struct inode *inode,
3718 		       const char *name, int name_len)
3719 {
3720 	int ret;
3721 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3722 	if (!ret) {
3723 		drop_nlink(inode);
3724 		ret = btrfs_update_inode(trans, root, inode);
3725 	}
3726 	return ret;
3727 }
3728 
3729 /*
3730  * helper to start transaction for unlink and rmdir.
3731  *
3732  * unlink and rmdir are special in btrfs, they do not always free space, so
3733  * if we cannot make our reservations the normal way try and see if there is
3734  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3735  * allow the unlink to occur.
3736  */
3737 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3738 {
3739 	struct btrfs_trans_handle *trans;
3740 	struct btrfs_root *root = BTRFS_I(dir)->root;
3741 	int ret;
3742 
3743 	/*
3744 	 * 1 for the possible orphan item
3745 	 * 1 for the dir item
3746 	 * 1 for the dir index
3747 	 * 1 for the inode ref
3748 	 * 1 for the inode
3749 	 */
3750 	trans = btrfs_start_transaction(root, 5);
3751 	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3752 		return trans;
3753 
3754 	if (PTR_ERR(trans) == -ENOSPC) {
3755 		u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3756 
3757 		trans = btrfs_start_transaction(root, 0);
3758 		if (IS_ERR(trans))
3759 			return trans;
3760 		ret = btrfs_cond_migrate_bytes(root->fs_info,
3761 					       &root->fs_info->trans_block_rsv,
3762 					       num_bytes, 5);
3763 		if (ret) {
3764 			btrfs_end_transaction(trans, root);
3765 			return ERR_PTR(ret);
3766 		}
3767 		trans->block_rsv = &root->fs_info->trans_block_rsv;
3768 		trans->bytes_reserved = num_bytes;
3769 	}
3770 	return trans;
3771 }
3772 
3773 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3774 {
3775 	struct btrfs_root *root = BTRFS_I(dir)->root;
3776 	struct btrfs_trans_handle *trans;
3777 	struct inode *inode = dentry->d_inode;
3778 	int ret;
3779 
3780 	trans = __unlink_start_trans(dir);
3781 	if (IS_ERR(trans))
3782 		return PTR_ERR(trans);
3783 
3784 	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3785 
3786 	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3787 				 dentry->d_name.name, dentry->d_name.len);
3788 	if (ret)
3789 		goto out;
3790 
3791 	if (inode->i_nlink == 0) {
3792 		ret = btrfs_orphan_add(trans, inode);
3793 		if (ret)
3794 			goto out;
3795 	}
3796 
3797 out:
3798 	btrfs_end_transaction(trans, root);
3799 	btrfs_btree_balance_dirty(root);
3800 	return ret;
3801 }
3802 
3803 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3804 			struct btrfs_root *root,
3805 			struct inode *dir, u64 objectid,
3806 			const char *name, int name_len)
3807 {
3808 	struct btrfs_path *path;
3809 	struct extent_buffer *leaf;
3810 	struct btrfs_dir_item *di;
3811 	struct btrfs_key key;
3812 	u64 index;
3813 	int ret;
3814 	u64 dir_ino = btrfs_ino(dir);
3815 
3816 	path = btrfs_alloc_path();
3817 	if (!path)
3818 		return -ENOMEM;
3819 
3820 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3821 				   name, name_len, -1);
3822 	if (IS_ERR_OR_NULL(di)) {
3823 		if (!di)
3824 			ret = -ENOENT;
3825 		else
3826 			ret = PTR_ERR(di);
3827 		goto out;
3828 	}
3829 
3830 	leaf = path->nodes[0];
3831 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3832 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3833 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3834 	if (ret) {
3835 		btrfs_abort_transaction(trans, root, ret);
3836 		goto out;
3837 	}
3838 	btrfs_release_path(path);
3839 
3840 	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3841 				 objectid, root->root_key.objectid,
3842 				 dir_ino, &index, name, name_len);
3843 	if (ret < 0) {
3844 		if (ret != -ENOENT) {
3845 			btrfs_abort_transaction(trans, root, ret);
3846 			goto out;
3847 		}
3848 		di = btrfs_search_dir_index_item(root, path, dir_ino,
3849 						 name, name_len);
3850 		if (IS_ERR_OR_NULL(di)) {
3851 			if (!di)
3852 				ret = -ENOENT;
3853 			else
3854 				ret = PTR_ERR(di);
3855 			btrfs_abort_transaction(trans, root, ret);
3856 			goto out;
3857 		}
3858 
3859 		leaf = path->nodes[0];
3860 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3861 		btrfs_release_path(path);
3862 		index = key.offset;
3863 	}
3864 	btrfs_release_path(path);
3865 
3866 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3867 	if (ret) {
3868 		btrfs_abort_transaction(trans, root, ret);
3869 		goto out;
3870 	}
3871 
3872 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3873 	inode_inc_iversion(dir);
3874 	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3875 	ret = btrfs_update_inode_fallback(trans, root, dir);
3876 	if (ret)
3877 		btrfs_abort_transaction(trans, root, ret);
3878 out:
3879 	btrfs_free_path(path);
3880 	return ret;
3881 }
3882 
3883 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3884 {
3885 	struct inode *inode = dentry->d_inode;
3886 	int err = 0;
3887 	struct btrfs_root *root = BTRFS_I(dir)->root;
3888 	struct btrfs_trans_handle *trans;
3889 
3890 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3891 		return -ENOTEMPTY;
3892 	if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3893 		return -EPERM;
3894 
3895 	trans = __unlink_start_trans(dir);
3896 	if (IS_ERR(trans))
3897 		return PTR_ERR(trans);
3898 
3899 	if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3900 		err = btrfs_unlink_subvol(trans, root, dir,
3901 					  BTRFS_I(inode)->location.objectid,
3902 					  dentry->d_name.name,
3903 					  dentry->d_name.len);
3904 		goto out;
3905 	}
3906 
3907 	err = btrfs_orphan_add(trans, inode);
3908 	if (err)
3909 		goto out;
3910 
3911 	/* now the directory is empty */
3912 	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3913 				 dentry->d_name.name, dentry->d_name.len);
3914 	if (!err)
3915 		btrfs_i_size_write(inode, 0);
3916 out:
3917 	btrfs_end_transaction(trans, root);
3918 	btrfs_btree_balance_dirty(root);
3919 
3920 	return err;
3921 }
3922 
3923 /*
3924  * this can truncate away extent items, csum items and directory items.
3925  * It starts at a high offset and removes keys until it can't find
3926  * any higher than new_size
3927  *
3928  * csum items that cross the new i_size are truncated to the new size
3929  * as well.
3930  *
3931  * min_type is the minimum key type to truncate down to.  If set to 0, this
3932  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3933  */
3934 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3935 			       struct btrfs_root *root,
3936 			       struct inode *inode,
3937 			       u64 new_size, u32 min_type)
3938 {
3939 	struct btrfs_path *path;
3940 	struct extent_buffer *leaf;
3941 	struct btrfs_file_extent_item *fi;
3942 	struct btrfs_key key;
3943 	struct btrfs_key found_key;
3944 	u64 extent_start = 0;
3945 	u64 extent_num_bytes = 0;
3946 	u64 extent_offset = 0;
3947 	u64 item_end = 0;
3948 	u64 last_size = (u64)-1;
3949 	u32 found_type = (u8)-1;
3950 	int found_extent;
3951 	int del_item;
3952 	int pending_del_nr = 0;
3953 	int pending_del_slot = 0;
3954 	int extent_type = -1;
3955 	int ret;
3956 	int err = 0;
3957 	u64 ino = btrfs_ino(inode);
3958 
3959 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3960 
3961 	path = btrfs_alloc_path();
3962 	if (!path)
3963 		return -ENOMEM;
3964 	path->reada = -1;
3965 
3966 	/*
3967 	 * We want to drop from the next block forward in case this new size is
3968 	 * not block aligned since we will be keeping the last block of the
3969 	 * extent just the way it is.
3970 	 */
3971 	if (root->ref_cows || root == root->fs_info->tree_root)
3972 		btrfs_drop_extent_cache(inode, ALIGN(new_size,
3973 					root->sectorsize), (u64)-1, 0);
3974 
3975 	/*
3976 	 * This function is also used to drop the items in the log tree before
3977 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3978 	 * it is used to drop the loged items. So we shouldn't kill the delayed
3979 	 * items.
3980 	 */
3981 	if (min_type == 0 && root == BTRFS_I(inode)->root)
3982 		btrfs_kill_delayed_inode_items(inode);
3983 
3984 	key.objectid = ino;
3985 	key.offset = (u64)-1;
3986 	key.type = (u8)-1;
3987 
3988 search_again:
3989 	path->leave_spinning = 1;
3990 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3991 	if (ret < 0) {
3992 		err = ret;
3993 		goto out;
3994 	}
3995 
3996 	if (ret > 0) {
3997 		/* there are no items in the tree for us to truncate, we're
3998 		 * done
3999 		 */
4000 		if (path->slots[0] == 0)
4001 			goto out;
4002 		path->slots[0]--;
4003 	}
4004 
4005 	while (1) {
4006 		fi = NULL;
4007 		leaf = path->nodes[0];
4008 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4009 		found_type = btrfs_key_type(&found_key);
4010 
4011 		if (found_key.objectid != ino)
4012 			break;
4013 
4014 		if (found_type < min_type)
4015 			break;
4016 
4017 		item_end = found_key.offset;
4018 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4019 			fi = btrfs_item_ptr(leaf, path->slots[0],
4020 					    struct btrfs_file_extent_item);
4021 			extent_type = btrfs_file_extent_type(leaf, fi);
4022 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4023 				item_end +=
4024 				    btrfs_file_extent_num_bytes(leaf, fi);
4025 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4026 				item_end += btrfs_file_extent_inline_len(leaf,
4027 							 path->slots[0], fi);
4028 			}
4029 			item_end--;
4030 		}
4031 		if (found_type > min_type) {
4032 			del_item = 1;
4033 		} else {
4034 			if (item_end < new_size)
4035 				break;
4036 			if (found_key.offset >= new_size)
4037 				del_item = 1;
4038 			else
4039 				del_item = 0;
4040 		}
4041 		found_extent = 0;
4042 		/* FIXME, shrink the extent if the ref count is only 1 */
4043 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4044 			goto delete;
4045 
4046 		if (del_item)
4047 			last_size = found_key.offset;
4048 		else
4049 			last_size = new_size;
4050 
4051 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4052 			u64 num_dec;
4053 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4054 			if (!del_item) {
4055 				u64 orig_num_bytes =
4056 					btrfs_file_extent_num_bytes(leaf, fi);
4057 				extent_num_bytes = ALIGN(new_size -
4058 						found_key.offset,
4059 						root->sectorsize);
4060 				btrfs_set_file_extent_num_bytes(leaf, fi,
4061 							 extent_num_bytes);
4062 				num_dec = (orig_num_bytes -
4063 					   extent_num_bytes);
4064 				if (root->ref_cows && extent_start != 0)
4065 					inode_sub_bytes(inode, num_dec);
4066 				btrfs_mark_buffer_dirty(leaf);
4067 			} else {
4068 				extent_num_bytes =
4069 					btrfs_file_extent_disk_num_bytes(leaf,
4070 									 fi);
4071 				extent_offset = found_key.offset -
4072 					btrfs_file_extent_offset(leaf, fi);
4073 
4074 				/* FIXME blocksize != 4096 */
4075 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4076 				if (extent_start != 0) {
4077 					found_extent = 1;
4078 					if (root->ref_cows)
4079 						inode_sub_bytes(inode, num_dec);
4080 				}
4081 			}
4082 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4083 			/*
4084 			 * we can't truncate inline items that have had
4085 			 * special encodings
4086 			 */
4087 			if (!del_item &&
4088 			    btrfs_file_extent_compression(leaf, fi) == 0 &&
4089 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4090 			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4091 				u32 size = new_size - found_key.offset;
4092 
4093 				if (root->ref_cows) {
4094 					inode_sub_bytes(inode, item_end + 1 -
4095 							new_size);
4096 				}
4097 
4098 				/*
4099 				 * update the ram bytes to properly reflect
4100 				 * the new size of our item
4101 				 */
4102 				btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4103 				size =
4104 				    btrfs_file_extent_calc_inline_size(size);
4105 				btrfs_truncate_item(root, path, size, 1);
4106 			} else if (root->ref_cows) {
4107 				inode_sub_bytes(inode, item_end + 1 -
4108 						found_key.offset);
4109 			}
4110 		}
4111 delete:
4112 		if (del_item) {
4113 			if (!pending_del_nr) {
4114 				/* no pending yet, add ourselves */
4115 				pending_del_slot = path->slots[0];
4116 				pending_del_nr = 1;
4117 			} else if (pending_del_nr &&
4118 				   path->slots[0] + 1 == pending_del_slot) {
4119 				/* hop on the pending chunk */
4120 				pending_del_nr++;
4121 				pending_del_slot = path->slots[0];
4122 			} else {
4123 				BUG();
4124 			}
4125 		} else {
4126 			break;
4127 		}
4128 		if (found_extent && (root->ref_cows ||
4129 				     root == root->fs_info->tree_root)) {
4130 			btrfs_set_path_blocking(path);
4131 			ret = btrfs_free_extent(trans, root, extent_start,
4132 						extent_num_bytes, 0,
4133 						btrfs_header_owner(leaf),
4134 						ino, extent_offset, 0);
4135 			BUG_ON(ret);
4136 		}
4137 
4138 		if (found_type == BTRFS_INODE_ITEM_KEY)
4139 			break;
4140 
4141 		if (path->slots[0] == 0 ||
4142 		    path->slots[0] != pending_del_slot) {
4143 			if (pending_del_nr) {
4144 				ret = btrfs_del_items(trans, root, path,
4145 						pending_del_slot,
4146 						pending_del_nr);
4147 				if (ret) {
4148 					btrfs_abort_transaction(trans,
4149 								root, ret);
4150 					goto error;
4151 				}
4152 				pending_del_nr = 0;
4153 			}
4154 			btrfs_release_path(path);
4155 			goto search_again;
4156 		} else {
4157 			path->slots[0]--;
4158 		}
4159 	}
4160 out:
4161 	if (pending_del_nr) {
4162 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4163 				      pending_del_nr);
4164 		if (ret)
4165 			btrfs_abort_transaction(trans, root, ret);
4166 	}
4167 error:
4168 	if (last_size != (u64)-1)
4169 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4170 	btrfs_free_path(path);
4171 	return err;
4172 }
4173 
4174 /*
4175  * btrfs_truncate_page - read, zero a chunk and write a page
4176  * @inode - inode that we're zeroing
4177  * @from - the offset to start zeroing
4178  * @len - the length to zero, 0 to zero the entire range respective to the
4179  *	offset
4180  * @front - zero up to the offset instead of from the offset on
4181  *
4182  * This will find the page for the "from" offset and cow the page and zero the
4183  * part we want to zero.  This is used with truncate and hole punching.
4184  */
4185 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4186 			int front)
4187 {
4188 	struct address_space *mapping = inode->i_mapping;
4189 	struct btrfs_root *root = BTRFS_I(inode)->root;
4190 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4191 	struct btrfs_ordered_extent *ordered;
4192 	struct extent_state *cached_state = NULL;
4193 	char *kaddr;
4194 	u32 blocksize = root->sectorsize;
4195 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
4196 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
4197 	struct page *page;
4198 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4199 	int ret = 0;
4200 	u64 page_start;
4201 	u64 page_end;
4202 
4203 	if ((offset & (blocksize - 1)) == 0 &&
4204 	    (!len || ((len & (blocksize - 1)) == 0)))
4205 		goto out;
4206 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4207 	if (ret)
4208 		goto out;
4209 
4210 again:
4211 	page = find_or_create_page(mapping, index, mask);
4212 	if (!page) {
4213 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4214 		ret = -ENOMEM;
4215 		goto out;
4216 	}
4217 
4218 	page_start = page_offset(page);
4219 	page_end = page_start + PAGE_CACHE_SIZE - 1;
4220 
4221 	if (!PageUptodate(page)) {
4222 		ret = btrfs_readpage(NULL, page);
4223 		lock_page(page);
4224 		if (page->mapping != mapping) {
4225 			unlock_page(page);
4226 			page_cache_release(page);
4227 			goto again;
4228 		}
4229 		if (!PageUptodate(page)) {
4230 			ret = -EIO;
4231 			goto out_unlock;
4232 		}
4233 	}
4234 	wait_on_page_writeback(page);
4235 
4236 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4237 	set_page_extent_mapped(page);
4238 
4239 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
4240 	if (ordered) {
4241 		unlock_extent_cached(io_tree, page_start, page_end,
4242 				     &cached_state, GFP_NOFS);
4243 		unlock_page(page);
4244 		page_cache_release(page);
4245 		btrfs_start_ordered_extent(inode, ordered, 1);
4246 		btrfs_put_ordered_extent(ordered);
4247 		goto again;
4248 	}
4249 
4250 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4251 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4252 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4253 			  0, 0, &cached_state, GFP_NOFS);
4254 
4255 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4256 					&cached_state);
4257 	if (ret) {
4258 		unlock_extent_cached(io_tree, page_start, page_end,
4259 				     &cached_state, GFP_NOFS);
4260 		goto out_unlock;
4261 	}
4262 
4263 	if (offset != PAGE_CACHE_SIZE) {
4264 		if (!len)
4265 			len = PAGE_CACHE_SIZE - offset;
4266 		kaddr = kmap(page);
4267 		if (front)
4268 			memset(kaddr, 0, offset);
4269 		else
4270 			memset(kaddr + offset, 0, len);
4271 		flush_dcache_page(page);
4272 		kunmap(page);
4273 	}
4274 	ClearPageChecked(page);
4275 	set_page_dirty(page);
4276 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4277 			     GFP_NOFS);
4278 
4279 out_unlock:
4280 	if (ret)
4281 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4282 	unlock_page(page);
4283 	page_cache_release(page);
4284 out:
4285 	return ret;
4286 }
4287 
4288 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4289 			     u64 offset, u64 len)
4290 {
4291 	struct btrfs_trans_handle *trans;
4292 	int ret;
4293 
4294 	/*
4295 	 * Still need to make sure the inode looks like it's been updated so
4296 	 * that any holes get logged if we fsync.
4297 	 */
4298 	if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4299 		BTRFS_I(inode)->last_trans = root->fs_info->generation;
4300 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4301 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4302 		return 0;
4303 	}
4304 
4305 	/*
4306 	 * 1 - for the one we're dropping
4307 	 * 1 - for the one we're adding
4308 	 * 1 - for updating the inode.
4309 	 */
4310 	trans = btrfs_start_transaction(root, 3);
4311 	if (IS_ERR(trans))
4312 		return PTR_ERR(trans);
4313 
4314 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4315 	if (ret) {
4316 		btrfs_abort_transaction(trans, root, ret);
4317 		btrfs_end_transaction(trans, root);
4318 		return ret;
4319 	}
4320 
4321 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4322 				       0, 0, len, 0, len, 0, 0, 0);
4323 	if (ret)
4324 		btrfs_abort_transaction(trans, root, ret);
4325 	else
4326 		btrfs_update_inode(trans, root, inode);
4327 	btrfs_end_transaction(trans, root);
4328 	return ret;
4329 }
4330 
4331 /*
4332  * This function puts in dummy file extents for the area we're creating a hole
4333  * for.  So if we are truncating this file to a larger size we need to insert
4334  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4335  * the range between oldsize and size
4336  */
4337 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4338 {
4339 	struct btrfs_root *root = BTRFS_I(inode)->root;
4340 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4341 	struct extent_map *em = NULL;
4342 	struct extent_state *cached_state = NULL;
4343 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4344 	u64 hole_start = ALIGN(oldsize, root->sectorsize);
4345 	u64 block_end = ALIGN(size, root->sectorsize);
4346 	u64 last_byte;
4347 	u64 cur_offset;
4348 	u64 hole_size;
4349 	int err = 0;
4350 
4351 	/*
4352 	 * If our size started in the middle of a page we need to zero out the
4353 	 * rest of the page before we expand the i_size, otherwise we could
4354 	 * expose stale data.
4355 	 */
4356 	err = btrfs_truncate_page(inode, oldsize, 0, 0);
4357 	if (err)
4358 		return err;
4359 
4360 	if (size <= hole_start)
4361 		return 0;
4362 
4363 	while (1) {
4364 		struct btrfs_ordered_extent *ordered;
4365 
4366 		lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4367 				 &cached_state);
4368 		ordered = btrfs_lookup_ordered_range(inode, hole_start,
4369 						     block_end - hole_start);
4370 		if (!ordered)
4371 			break;
4372 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4373 				     &cached_state, GFP_NOFS);
4374 		btrfs_start_ordered_extent(inode, ordered, 1);
4375 		btrfs_put_ordered_extent(ordered);
4376 	}
4377 
4378 	cur_offset = hole_start;
4379 	while (1) {
4380 		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4381 				block_end - cur_offset, 0);
4382 		if (IS_ERR(em)) {
4383 			err = PTR_ERR(em);
4384 			em = NULL;
4385 			break;
4386 		}
4387 		last_byte = min(extent_map_end(em), block_end);
4388 		last_byte = ALIGN(last_byte , root->sectorsize);
4389 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4390 			struct extent_map *hole_em;
4391 			hole_size = last_byte - cur_offset;
4392 
4393 			err = maybe_insert_hole(root, inode, cur_offset,
4394 						hole_size);
4395 			if (err)
4396 				break;
4397 			btrfs_drop_extent_cache(inode, cur_offset,
4398 						cur_offset + hole_size - 1, 0);
4399 			hole_em = alloc_extent_map();
4400 			if (!hole_em) {
4401 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4402 					&BTRFS_I(inode)->runtime_flags);
4403 				goto next;
4404 			}
4405 			hole_em->start = cur_offset;
4406 			hole_em->len = hole_size;
4407 			hole_em->orig_start = cur_offset;
4408 
4409 			hole_em->block_start = EXTENT_MAP_HOLE;
4410 			hole_em->block_len = 0;
4411 			hole_em->orig_block_len = 0;
4412 			hole_em->ram_bytes = hole_size;
4413 			hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4414 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
4415 			hole_em->generation = root->fs_info->generation;
4416 
4417 			while (1) {
4418 				write_lock(&em_tree->lock);
4419 				err = add_extent_mapping(em_tree, hole_em, 1);
4420 				write_unlock(&em_tree->lock);
4421 				if (err != -EEXIST)
4422 					break;
4423 				btrfs_drop_extent_cache(inode, cur_offset,
4424 							cur_offset +
4425 							hole_size - 1, 0);
4426 			}
4427 			free_extent_map(hole_em);
4428 		}
4429 next:
4430 		free_extent_map(em);
4431 		em = NULL;
4432 		cur_offset = last_byte;
4433 		if (cur_offset >= block_end)
4434 			break;
4435 	}
4436 	free_extent_map(em);
4437 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4438 			     GFP_NOFS);
4439 	return err;
4440 }
4441 
4442 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4443 {
4444 	struct btrfs_root *root = BTRFS_I(inode)->root;
4445 	struct btrfs_trans_handle *trans;
4446 	loff_t oldsize = i_size_read(inode);
4447 	loff_t newsize = attr->ia_size;
4448 	int mask = attr->ia_valid;
4449 	int ret;
4450 
4451 	/*
4452 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4453 	 * special case where we need to update the times despite not having
4454 	 * these flags set.  For all other operations the VFS set these flags
4455 	 * explicitly if it wants a timestamp update.
4456 	 */
4457 	if (newsize != oldsize) {
4458 		inode_inc_iversion(inode);
4459 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4460 			inode->i_ctime = inode->i_mtime =
4461 				current_fs_time(inode->i_sb);
4462 	}
4463 
4464 	if (newsize > oldsize) {
4465 		truncate_pagecache(inode, newsize);
4466 		ret = btrfs_cont_expand(inode, oldsize, newsize);
4467 		if (ret)
4468 			return ret;
4469 
4470 		trans = btrfs_start_transaction(root, 1);
4471 		if (IS_ERR(trans))
4472 			return PTR_ERR(trans);
4473 
4474 		i_size_write(inode, newsize);
4475 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4476 		ret = btrfs_update_inode(trans, root, inode);
4477 		btrfs_end_transaction(trans, root);
4478 	} else {
4479 
4480 		/*
4481 		 * We're truncating a file that used to have good data down to
4482 		 * zero. Make sure it gets into the ordered flush list so that
4483 		 * any new writes get down to disk quickly.
4484 		 */
4485 		if (newsize == 0)
4486 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4487 				&BTRFS_I(inode)->runtime_flags);
4488 
4489 		/*
4490 		 * 1 for the orphan item we're going to add
4491 		 * 1 for the orphan item deletion.
4492 		 */
4493 		trans = btrfs_start_transaction(root, 2);
4494 		if (IS_ERR(trans))
4495 			return PTR_ERR(trans);
4496 
4497 		/*
4498 		 * We need to do this in case we fail at _any_ point during the
4499 		 * actual truncate.  Once we do the truncate_setsize we could
4500 		 * invalidate pages which forces any outstanding ordered io to
4501 		 * be instantly completed which will give us extents that need
4502 		 * to be truncated.  If we fail to get an orphan inode down we
4503 		 * could have left over extents that were never meant to live,
4504 		 * so we need to garuntee from this point on that everything
4505 		 * will be consistent.
4506 		 */
4507 		ret = btrfs_orphan_add(trans, inode);
4508 		btrfs_end_transaction(trans, root);
4509 		if (ret)
4510 			return ret;
4511 
4512 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
4513 		truncate_setsize(inode, newsize);
4514 
4515 		/* Disable nonlocked read DIO to avoid the end less truncate */
4516 		btrfs_inode_block_unlocked_dio(inode);
4517 		inode_dio_wait(inode);
4518 		btrfs_inode_resume_unlocked_dio(inode);
4519 
4520 		ret = btrfs_truncate(inode);
4521 		if (ret && inode->i_nlink) {
4522 			int err;
4523 
4524 			/*
4525 			 * failed to truncate, disk_i_size is only adjusted down
4526 			 * as we remove extents, so it should represent the true
4527 			 * size of the inode, so reset the in memory size and
4528 			 * delete our orphan entry.
4529 			 */
4530 			trans = btrfs_join_transaction(root);
4531 			if (IS_ERR(trans)) {
4532 				btrfs_orphan_del(NULL, inode);
4533 				return ret;
4534 			}
4535 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4536 			err = btrfs_orphan_del(trans, inode);
4537 			if (err)
4538 				btrfs_abort_transaction(trans, root, err);
4539 			btrfs_end_transaction(trans, root);
4540 		}
4541 	}
4542 
4543 	return ret;
4544 }
4545 
4546 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4547 {
4548 	struct inode *inode = dentry->d_inode;
4549 	struct btrfs_root *root = BTRFS_I(inode)->root;
4550 	int err;
4551 
4552 	if (btrfs_root_readonly(root))
4553 		return -EROFS;
4554 
4555 	err = inode_change_ok(inode, attr);
4556 	if (err)
4557 		return err;
4558 
4559 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4560 		err = btrfs_setsize(inode, attr);
4561 		if (err)
4562 			return err;
4563 	}
4564 
4565 	if (attr->ia_valid) {
4566 		setattr_copy(inode, attr);
4567 		inode_inc_iversion(inode);
4568 		err = btrfs_dirty_inode(inode);
4569 
4570 		if (!err && attr->ia_valid & ATTR_MODE)
4571 			err = posix_acl_chmod(inode, inode->i_mode);
4572 	}
4573 
4574 	return err;
4575 }
4576 
4577 /*
4578  * While truncating the inode pages during eviction, we get the VFS calling
4579  * btrfs_invalidatepage() against each page of the inode. This is slow because
4580  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4581  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4582  * extent_state structures over and over, wasting lots of time.
4583  *
4584  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4585  * those expensive operations on a per page basis and do only the ordered io
4586  * finishing, while we release here the extent_map and extent_state structures,
4587  * without the excessive merging and splitting.
4588  */
4589 static void evict_inode_truncate_pages(struct inode *inode)
4590 {
4591 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4592 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4593 	struct rb_node *node;
4594 
4595 	ASSERT(inode->i_state & I_FREEING);
4596 	truncate_inode_pages(&inode->i_data, 0);
4597 
4598 	write_lock(&map_tree->lock);
4599 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
4600 		struct extent_map *em;
4601 
4602 		node = rb_first(&map_tree->map);
4603 		em = rb_entry(node, struct extent_map, rb_node);
4604 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4605 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4606 		remove_extent_mapping(map_tree, em);
4607 		free_extent_map(em);
4608 	}
4609 	write_unlock(&map_tree->lock);
4610 
4611 	spin_lock(&io_tree->lock);
4612 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
4613 		struct extent_state *state;
4614 		struct extent_state *cached_state = NULL;
4615 
4616 		node = rb_first(&io_tree->state);
4617 		state = rb_entry(node, struct extent_state, rb_node);
4618 		atomic_inc(&state->refs);
4619 		spin_unlock(&io_tree->lock);
4620 
4621 		lock_extent_bits(io_tree, state->start, state->end,
4622 				 0, &cached_state);
4623 		clear_extent_bit(io_tree, state->start, state->end,
4624 				 EXTENT_LOCKED | EXTENT_DIRTY |
4625 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4626 				 EXTENT_DEFRAG, 1, 1,
4627 				 &cached_state, GFP_NOFS);
4628 		free_extent_state(state);
4629 
4630 		spin_lock(&io_tree->lock);
4631 	}
4632 	spin_unlock(&io_tree->lock);
4633 }
4634 
4635 void btrfs_evict_inode(struct inode *inode)
4636 {
4637 	struct btrfs_trans_handle *trans;
4638 	struct btrfs_root *root = BTRFS_I(inode)->root;
4639 	struct btrfs_block_rsv *rsv, *global_rsv;
4640 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4641 	int ret;
4642 
4643 	trace_btrfs_inode_evict(inode);
4644 
4645 	evict_inode_truncate_pages(inode);
4646 
4647 	if (inode->i_nlink &&
4648 	    ((btrfs_root_refs(&root->root_item) != 0 &&
4649 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4650 	     btrfs_is_free_space_inode(inode)))
4651 		goto no_delete;
4652 
4653 	if (is_bad_inode(inode)) {
4654 		btrfs_orphan_del(NULL, inode);
4655 		goto no_delete;
4656 	}
4657 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4658 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
4659 
4660 	if (root->fs_info->log_root_recovering) {
4661 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4662 				 &BTRFS_I(inode)->runtime_flags));
4663 		goto no_delete;
4664 	}
4665 
4666 	if (inode->i_nlink > 0) {
4667 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4668 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4669 		goto no_delete;
4670 	}
4671 
4672 	ret = btrfs_commit_inode_delayed_inode(inode);
4673 	if (ret) {
4674 		btrfs_orphan_del(NULL, inode);
4675 		goto no_delete;
4676 	}
4677 
4678 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4679 	if (!rsv) {
4680 		btrfs_orphan_del(NULL, inode);
4681 		goto no_delete;
4682 	}
4683 	rsv->size = min_size;
4684 	rsv->failfast = 1;
4685 	global_rsv = &root->fs_info->global_block_rsv;
4686 
4687 	btrfs_i_size_write(inode, 0);
4688 
4689 	/*
4690 	 * This is a bit simpler than btrfs_truncate since we've already
4691 	 * reserved our space for our orphan item in the unlink, so we just
4692 	 * need to reserve some slack space in case we add bytes and update
4693 	 * inode item when doing the truncate.
4694 	 */
4695 	while (1) {
4696 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
4697 					     BTRFS_RESERVE_FLUSH_LIMIT);
4698 
4699 		/*
4700 		 * Try and steal from the global reserve since we will
4701 		 * likely not use this space anyway, we want to try as
4702 		 * hard as possible to get this to work.
4703 		 */
4704 		if (ret)
4705 			ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4706 
4707 		if (ret) {
4708 			btrfs_warn(root->fs_info,
4709 				"Could not get space for a delete, will truncate on mount %d",
4710 				ret);
4711 			btrfs_orphan_del(NULL, inode);
4712 			btrfs_free_block_rsv(root, rsv);
4713 			goto no_delete;
4714 		}
4715 
4716 		trans = btrfs_join_transaction(root);
4717 		if (IS_ERR(trans)) {
4718 			btrfs_orphan_del(NULL, inode);
4719 			btrfs_free_block_rsv(root, rsv);
4720 			goto no_delete;
4721 		}
4722 
4723 		trans->block_rsv = rsv;
4724 
4725 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4726 		if (ret != -ENOSPC)
4727 			break;
4728 
4729 		trans->block_rsv = &root->fs_info->trans_block_rsv;
4730 		btrfs_end_transaction(trans, root);
4731 		trans = NULL;
4732 		btrfs_btree_balance_dirty(root);
4733 	}
4734 
4735 	btrfs_free_block_rsv(root, rsv);
4736 
4737 	/*
4738 	 * Errors here aren't a big deal, it just means we leave orphan items
4739 	 * in the tree.  They will be cleaned up on the next mount.
4740 	 */
4741 	if (ret == 0) {
4742 		trans->block_rsv = root->orphan_block_rsv;
4743 		btrfs_orphan_del(trans, inode);
4744 	} else {
4745 		btrfs_orphan_del(NULL, inode);
4746 	}
4747 
4748 	trans->block_rsv = &root->fs_info->trans_block_rsv;
4749 	if (!(root == root->fs_info->tree_root ||
4750 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4751 		btrfs_return_ino(root, btrfs_ino(inode));
4752 
4753 	btrfs_end_transaction(trans, root);
4754 	btrfs_btree_balance_dirty(root);
4755 no_delete:
4756 	btrfs_remove_delayed_node(inode);
4757 	clear_inode(inode);
4758 	return;
4759 }
4760 
4761 /*
4762  * this returns the key found in the dir entry in the location pointer.
4763  * If no dir entries were found, location->objectid is 0.
4764  */
4765 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4766 			       struct btrfs_key *location)
4767 {
4768 	const char *name = dentry->d_name.name;
4769 	int namelen = dentry->d_name.len;
4770 	struct btrfs_dir_item *di;
4771 	struct btrfs_path *path;
4772 	struct btrfs_root *root = BTRFS_I(dir)->root;
4773 	int ret = 0;
4774 
4775 	path = btrfs_alloc_path();
4776 	if (!path)
4777 		return -ENOMEM;
4778 
4779 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4780 				    namelen, 0);
4781 	if (IS_ERR(di))
4782 		ret = PTR_ERR(di);
4783 
4784 	if (IS_ERR_OR_NULL(di))
4785 		goto out_err;
4786 
4787 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4788 out:
4789 	btrfs_free_path(path);
4790 	return ret;
4791 out_err:
4792 	location->objectid = 0;
4793 	goto out;
4794 }
4795 
4796 /*
4797  * when we hit a tree root in a directory, the btrfs part of the inode
4798  * needs to be changed to reflect the root directory of the tree root.  This
4799  * is kind of like crossing a mount point.
4800  */
4801 static int fixup_tree_root_location(struct btrfs_root *root,
4802 				    struct inode *dir,
4803 				    struct dentry *dentry,
4804 				    struct btrfs_key *location,
4805 				    struct btrfs_root **sub_root)
4806 {
4807 	struct btrfs_path *path;
4808 	struct btrfs_root *new_root;
4809 	struct btrfs_root_ref *ref;
4810 	struct extent_buffer *leaf;
4811 	int ret;
4812 	int err = 0;
4813 
4814 	path = btrfs_alloc_path();
4815 	if (!path) {
4816 		err = -ENOMEM;
4817 		goto out;
4818 	}
4819 
4820 	err = -ENOENT;
4821 	ret = btrfs_find_item(root->fs_info->tree_root, path,
4822 				BTRFS_I(dir)->root->root_key.objectid,
4823 				location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4824 	if (ret) {
4825 		if (ret < 0)
4826 			err = ret;
4827 		goto out;
4828 	}
4829 
4830 	leaf = path->nodes[0];
4831 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4832 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4833 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4834 		goto out;
4835 
4836 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4837 				   (unsigned long)(ref + 1),
4838 				   dentry->d_name.len);
4839 	if (ret)
4840 		goto out;
4841 
4842 	btrfs_release_path(path);
4843 
4844 	new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4845 	if (IS_ERR(new_root)) {
4846 		err = PTR_ERR(new_root);
4847 		goto out;
4848 	}
4849 
4850 	*sub_root = new_root;
4851 	location->objectid = btrfs_root_dirid(&new_root->root_item);
4852 	location->type = BTRFS_INODE_ITEM_KEY;
4853 	location->offset = 0;
4854 	err = 0;
4855 out:
4856 	btrfs_free_path(path);
4857 	return err;
4858 }
4859 
4860 static void inode_tree_add(struct inode *inode)
4861 {
4862 	struct btrfs_root *root = BTRFS_I(inode)->root;
4863 	struct btrfs_inode *entry;
4864 	struct rb_node **p;
4865 	struct rb_node *parent;
4866 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
4867 	u64 ino = btrfs_ino(inode);
4868 
4869 	if (inode_unhashed(inode))
4870 		return;
4871 	parent = NULL;
4872 	spin_lock(&root->inode_lock);
4873 	p = &root->inode_tree.rb_node;
4874 	while (*p) {
4875 		parent = *p;
4876 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
4877 
4878 		if (ino < btrfs_ino(&entry->vfs_inode))
4879 			p = &parent->rb_left;
4880 		else if (ino > btrfs_ino(&entry->vfs_inode))
4881 			p = &parent->rb_right;
4882 		else {
4883 			WARN_ON(!(entry->vfs_inode.i_state &
4884 				  (I_WILL_FREE | I_FREEING)));
4885 			rb_replace_node(parent, new, &root->inode_tree);
4886 			RB_CLEAR_NODE(parent);
4887 			spin_unlock(&root->inode_lock);
4888 			return;
4889 		}
4890 	}
4891 	rb_link_node(new, parent, p);
4892 	rb_insert_color(new, &root->inode_tree);
4893 	spin_unlock(&root->inode_lock);
4894 }
4895 
4896 static void inode_tree_del(struct inode *inode)
4897 {
4898 	struct btrfs_root *root = BTRFS_I(inode)->root;
4899 	int empty = 0;
4900 
4901 	spin_lock(&root->inode_lock);
4902 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4903 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4904 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4905 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4906 	}
4907 	spin_unlock(&root->inode_lock);
4908 
4909 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
4910 		synchronize_srcu(&root->fs_info->subvol_srcu);
4911 		spin_lock(&root->inode_lock);
4912 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4913 		spin_unlock(&root->inode_lock);
4914 		if (empty)
4915 			btrfs_add_dead_root(root);
4916 	}
4917 }
4918 
4919 void btrfs_invalidate_inodes(struct btrfs_root *root)
4920 {
4921 	struct rb_node *node;
4922 	struct rb_node *prev;
4923 	struct btrfs_inode *entry;
4924 	struct inode *inode;
4925 	u64 objectid = 0;
4926 
4927 	WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4928 
4929 	spin_lock(&root->inode_lock);
4930 again:
4931 	node = root->inode_tree.rb_node;
4932 	prev = NULL;
4933 	while (node) {
4934 		prev = node;
4935 		entry = rb_entry(node, struct btrfs_inode, rb_node);
4936 
4937 		if (objectid < btrfs_ino(&entry->vfs_inode))
4938 			node = node->rb_left;
4939 		else if (objectid > btrfs_ino(&entry->vfs_inode))
4940 			node = node->rb_right;
4941 		else
4942 			break;
4943 	}
4944 	if (!node) {
4945 		while (prev) {
4946 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
4947 			if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4948 				node = prev;
4949 				break;
4950 			}
4951 			prev = rb_next(prev);
4952 		}
4953 	}
4954 	while (node) {
4955 		entry = rb_entry(node, struct btrfs_inode, rb_node);
4956 		objectid = btrfs_ino(&entry->vfs_inode) + 1;
4957 		inode = igrab(&entry->vfs_inode);
4958 		if (inode) {
4959 			spin_unlock(&root->inode_lock);
4960 			if (atomic_read(&inode->i_count) > 1)
4961 				d_prune_aliases(inode);
4962 			/*
4963 			 * btrfs_drop_inode will have it removed from
4964 			 * the inode cache when its usage count
4965 			 * hits zero.
4966 			 */
4967 			iput(inode);
4968 			cond_resched();
4969 			spin_lock(&root->inode_lock);
4970 			goto again;
4971 		}
4972 
4973 		if (cond_resched_lock(&root->inode_lock))
4974 			goto again;
4975 
4976 		node = rb_next(node);
4977 	}
4978 	spin_unlock(&root->inode_lock);
4979 }
4980 
4981 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4982 {
4983 	struct btrfs_iget_args *args = p;
4984 	inode->i_ino = args->location->objectid;
4985 	memcpy(&BTRFS_I(inode)->location, args->location,
4986 	       sizeof(*args->location));
4987 	BTRFS_I(inode)->root = args->root;
4988 	return 0;
4989 }
4990 
4991 static int btrfs_find_actor(struct inode *inode, void *opaque)
4992 {
4993 	struct btrfs_iget_args *args = opaque;
4994 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
4995 		args->root == BTRFS_I(inode)->root;
4996 }
4997 
4998 static struct inode *btrfs_iget_locked(struct super_block *s,
4999 				       struct btrfs_key *location,
5000 				       struct btrfs_root *root)
5001 {
5002 	struct inode *inode;
5003 	struct btrfs_iget_args args;
5004 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5005 
5006 	args.location = location;
5007 	args.root = root;
5008 
5009 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5010 			     btrfs_init_locked_inode,
5011 			     (void *)&args);
5012 	return inode;
5013 }
5014 
5015 /* Get an inode object given its location and corresponding root.
5016  * Returns in *is_new if the inode was read from disk
5017  */
5018 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5019 			 struct btrfs_root *root, int *new)
5020 {
5021 	struct inode *inode;
5022 
5023 	inode = btrfs_iget_locked(s, location, root);
5024 	if (!inode)
5025 		return ERR_PTR(-ENOMEM);
5026 
5027 	if (inode->i_state & I_NEW) {
5028 		btrfs_read_locked_inode(inode);
5029 		if (!is_bad_inode(inode)) {
5030 			inode_tree_add(inode);
5031 			unlock_new_inode(inode);
5032 			if (new)
5033 				*new = 1;
5034 		} else {
5035 			unlock_new_inode(inode);
5036 			iput(inode);
5037 			inode = ERR_PTR(-ESTALE);
5038 		}
5039 	}
5040 
5041 	return inode;
5042 }
5043 
5044 static struct inode *new_simple_dir(struct super_block *s,
5045 				    struct btrfs_key *key,
5046 				    struct btrfs_root *root)
5047 {
5048 	struct inode *inode = new_inode(s);
5049 
5050 	if (!inode)
5051 		return ERR_PTR(-ENOMEM);
5052 
5053 	BTRFS_I(inode)->root = root;
5054 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5055 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5056 
5057 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5058 	inode->i_op = &btrfs_dir_ro_inode_operations;
5059 	inode->i_fop = &simple_dir_operations;
5060 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5061 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5062 
5063 	return inode;
5064 }
5065 
5066 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5067 {
5068 	struct inode *inode;
5069 	struct btrfs_root *root = BTRFS_I(dir)->root;
5070 	struct btrfs_root *sub_root = root;
5071 	struct btrfs_key location;
5072 	int index;
5073 	int ret = 0;
5074 
5075 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5076 		return ERR_PTR(-ENAMETOOLONG);
5077 
5078 	ret = btrfs_inode_by_name(dir, dentry, &location);
5079 	if (ret < 0)
5080 		return ERR_PTR(ret);
5081 
5082 	if (location.objectid == 0)
5083 		return ERR_PTR(-ENOENT);
5084 
5085 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5086 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5087 		return inode;
5088 	}
5089 
5090 	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5091 
5092 	index = srcu_read_lock(&root->fs_info->subvol_srcu);
5093 	ret = fixup_tree_root_location(root, dir, dentry,
5094 				       &location, &sub_root);
5095 	if (ret < 0) {
5096 		if (ret != -ENOENT)
5097 			inode = ERR_PTR(ret);
5098 		else
5099 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5100 	} else {
5101 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5102 	}
5103 	srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5104 
5105 	if (!IS_ERR(inode) && root != sub_root) {
5106 		down_read(&root->fs_info->cleanup_work_sem);
5107 		if (!(inode->i_sb->s_flags & MS_RDONLY))
5108 			ret = btrfs_orphan_cleanup(sub_root);
5109 		up_read(&root->fs_info->cleanup_work_sem);
5110 		if (ret) {
5111 			iput(inode);
5112 			inode = ERR_PTR(ret);
5113 		}
5114 	}
5115 
5116 	return inode;
5117 }
5118 
5119 static int btrfs_dentry_delete(const struct dentry *dentry)
5120 {
5121 	struct btrfs_root *root;
5122 	struct inode *inode = dentry->d_inode;
5123 
5124 	if (!inode && !IS_ROOT(dentry))
5125 		inode = dentry->d_parent->d_inode;
5126 
5127 	if (inode) {
5128 		root = BTRFS_I(inode)->root;
5129 		if (btrfs_root_refs(&root->root_item) == 0)
5130 			return 1;
5131 
5132 		if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5133 			return 1;
5134 	}
5135 	return 0;
5136 }
5137 
5138 static void btrfs_dentry_release(struct dentry *dentry)
5139 {
5140 	if (dentry->d_fsdata)
5141 		kfree(dentry->d_fsdata);
5142 }
5143 
5144 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5145 				   unsigned int flags)
5146 {
5147 	struct inode *inode;
5148 
5149 	inode = btrfs_lookup_dentry(dir, dentry);
5150 	if (IS_ERR(inode)) {
5151 		if (PTR_ERR(inode) == -ENOENT)
5152 			inode = NULL;
5153 		else
5154 			return ERR_CAST(inode);
5155 	}
5156 
5157 	return d_materialise_unique(dentry, inode);
5158 }
5159 
5160 unsigned char btrfs_filetype_table[] = {
5161 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5162 };
5163 
5164 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5165 {
5166 	struct inode *inode = file_inode(file);
5167 	struct btrfs_root *root = BTRFS_I(inode)->root;
5168 	struct btrfs_item *item;
5169 	struct btrfs_dir_item *di;
5170 	struct btrfs_key key;
5171 	struct btrfs_key found_key;
5172 	struct btrfs_path *path;
5173 	struct list_head ins_list;
5174 	struct list_head del_list;
5175 	int ret;
5176 	struct extent_buffer *leaf;
5177 	int slot;
5178 	unsigned char d_type;
5179 	int over = 0;
5180 	u32 di_cur;
5181 	u32 di_total;
5182 	u32 di_len;
5183 	int key_type = BTRFS_DIR_INDEX_KEY;
5184 	char tmp_name[32];
5185 	char *name_ptr;
5186 	int name_len;
5187 	int is_curr = 0;	/* ctx->pos points to the current index? */
5188 
5189 	/* FIXME, use a real flag for deciding about the key type */
5190 	if (root->fs_info->tree_root == root)
5191 		key_type = BTRFS_DIR_ITEM_KEY;
5192 
5193 	if (!dir_emit_dots(file, ctx))
5194 		return 0;
5195 
5196 	path = btrfs_alloc_path();
5197 	if (!path)
5198 		return -ENOMEM;
5199 
5200 	path->reada = 1;
5201 
5202 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5203 		INIT_LIST_HEAD(&ins_list);
5204 		INIT_LIST_HEAD(&del_list);
5205 		btrfs_get_delayed_items(inode, &ins_list, &del_list);
5206 	}
5207 
5208 	btrfs_set_key_type(&key, key_type);
5209 	key.offset = ctx->pos;
5210 	key.objectid = btrfs_ino(inode);
5211 
5212 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5213 	if (ret < 0)
5214 		goto err;
5215 
5216 	while (1) {
5217 		leaf = path->nodes[0];
5218 		slot = path->slots[0];
5219 		if (slot >= btrfs_header_nritems(leaf)) {
5220 			ret = btrfs_next_leaf(root, path);
5221 			if (ret < 0)
5222 				goto err;
5223 			else if (ret > 0)
5224 				break;
5225 			continue;
5226 		}
5227 
5228 		item = btrfs_item_nr(slot);
5229 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5230 
5231 		if (found_key.objectid != key.objectid)
5232 			break;
5233 		if (btrfs_key_type(&found_key) != key_type)
5234 			break;
5235 		if (found_key.offset < ctx->pos)
5236 			goto next;
5237 		if (key_type == BTRFS_DIR_INDEX_KEY &&
5238 		    btrfs_should_delete_dir_index(&del_list,
5239 						  found_key.offset))
5240 			goto next;
5241 
5242 		ctx->pos = found_key.offset;
5243 		is_curr = 1;
5244 
5245 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5246 		di_cur = 0;
5247 		di_total = btrfs_item_size(leaf, item);
5248 
5249 		while (di_cur < di_total) {
5250 			struct btrfs_key location;
5251 
5252 			if (verify_dir_item(root, leaf, di))
5253 				break;
5254 
5255 			name_len = btrfs_dir_name_len(leaf, di);
5256 			if (name_len <= sizeof(tmp_name)) {
5257 				name_ptr = tmp_name;
5258 			} else {
5259 				name_ptr = kmalloc(name_len, GFP_NOFS);
5260 				if (!name_ptr) {
5261 					ret = -ENOMEM;
5262 					goto err;
5263 				}
5264 			}
5265 			read_extent_buffer(leaf, name_ptr,
5266 					   (unsigned long)(di + 1), name_len);
5267 
5268 			d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5269 			btrfs_dir_item_key_to_cpu(leaf, di, &location);
5270 
5271 
5272 			/* is this a reference to our own snapshot? If so
5273 			 * skip it.
5274 			 *
5275 			 * In contrast to old kernels, we insert the snapshot's
5276 			 * dir item and dir index after it has been created, so
5277 			 * we won't find a reference to our own snapshot. We
5278 			 * still keep the following code for backward
5279 			 * compatibility.
5280 			 */
5281 			if (location.type == BTRFS_ROOT_ITEM_KEY &&
5282 			    location.objectid == root->root_key.objectid) {
5283 				over = 0;
5284 				goto skip;
5285 			}
5286 			over = !dir_emit(ctx, name_ptr, name_len,
5287 				       location.objectid, d_type);
5288 
5289 skip:
5290 			if (name_ptr != tmp_name)
5291 				kfree(name_ptr);
5292 
5293 			if (over)
5294 				goto nopos;
5295 			di_len = btrfs_dir_name_len(leaf, di) +
5296 				 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5297 			di_cur += di_len;
5298 			di = (struct btrfs_dir_item *)((char *)di + di_len);
5299 		}
5300 next:
5301 		path->slots[0]++;
5302 	}
5303 
5304 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5305 		if (is_curr)
5306 			ctx->pos++;
5307 		ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5308 		if (ret)
5309 			goto nopos;
5310 	}
5311 
5312 	/* Reached end of directory/root. Bump pos past the last item. */
5313 	ctx->pos++;
5314 
5315 	/*
5316 	 * Stop new entries from being returned after we return the last
5317 	 * entry.
5318 	 *
5319 	 * New directory entries are assigned a strictly increasing
5320 	 * offset.  This means that new entries created during readdir
5321 	 * are *guaranteed* to be seen in the future by that readdir.
5322 	 * This has broken buggy programs which operate on names as
5323 	 * they're returned by readdir.  Until we re-use freed offsets
5324 	 * we have this hack to stop new entries from being returned
5325 	 * under the assumption that they'll never reach this huge
5326 	 * offset.
5327 	 *
5328 	 * This is being careful not to overflow 32bit loff_t unless the
5329 	 * last entry requires it because doing so has broken 32bit apps
5330 	 * in the past.
5331 	 */
5332 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5333 		if (ctx->pos >= INT_MAX)
5334 			ctx->pos = LLONG_MAX;
5335 		else
5336 			ctx->pos = INT_MAX;
5337 	}
5338 nopos:
5339 	ret = 0;
5340 err:
5341 	if (key_type == BTRFS_DIR_INDEX_KEY)
5342 		btrfs_put_delayed_items(&ins_list, &del_list);
5343 	btrfs_free_path(path);
5344 	return ret;
5345 }
5346 
5347 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5348 {
5349 	struct btrfs_root *root = BTRFS_I(inode)->root;
5350 	struct btrfs_trans_handle *trans;
5351 	int ret = 0;
5352 	bool nolock = false;
5353 
5354 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5355 		return 0;
5356 
5357 	if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5358 		nolock = true;
5359 
5360 	if (wbc->sync_mode == WB_SYNC_ALL) {
5361 		if (nolock)
5362 			trans = btrfs_join_transaction_nolock(root);
5363 		else
5364 			trans = btrfs_join_transaction(root);
5365 		if (IS_ERR(trans))
5366 			return PTR_ERR(trans);
5367 		ret = btrfs_commit_transaction(trans, root);
5368 	}
5369 	return ret;
5370 }
5371 
5372 /*
5373  * This is somewhat expensive, updating the tree every time the
5374  * inode changes.  But, it is most likely to find the inode in cache.
5375  * FIXME, needs more benchmarking...there are no reasons other than performance
5376  * to keep or drop this code.
5377  */
5378 static int btrfs_dirty_inode(struct inode *inode)
5379 {
5380 	struct btrfs_root *root = BTRFS_I(inode)->root;
5381 	struct btrfs_trans_handle *trans;
5382 	int ret;
5383 
5384 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5385 		return 0;
5386 
5387 	trans = btrfs_join_transaction(root);
5388 	if (IS_ERR(trans))
5389 		return PTR_ERR(trans);
5390 
5391 	ret = btrfs_update_inode(trans, root, inode);
5392 	if (ret && ret == -ENOSPC) {
5393 		/* whoops, lets try again with the full transaction */
5394 		btrfs_end_transaction(trans, root);
5395 		trans = btrfs_start_transaction(root, 1);
5396 		if (IS_ERR(trans))
5397 			return PTR_ERR(trans);
5398 
5399 		ret = btrfs_update_inode(trans, root, inode);
5400 	}
5401 	btrfs_end_transaction(trans, root);
5402 	if (BTRFS_I(inode)->delayed_node)
5403 		btrfs_balance_delayed_items(root);
5404 
5405 	return ret;
5406 }
5407 
5408 /*
5409  * This is a copy of file_update_time.  We need this so we can return error on
5410  * ENOSPC for updating the inode in the case of file write and mmap writes.
5411  */
5412 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5413 			     int flags)
5414 {
5415 	struct btrfs_root *root = BTRFS_I(inode)->root;
5416 
5417 	if (btrfs_root_readonly(root))
5418 		return -EROFS;
5419 
5420 	if (flags & S_VERSION)
5421 		inode_inc_iversion(inode);
5422 	if (flags & S_CTIME)
5423 		inode->i_ctime = *now;
5424 	if (flags & S_MTIME)
5425 		inode->i_mtime = *now;
5426 	if (flags & S_ATIME)
5427 		inode->i_atime = *now;
5428 	return btrfs_dirty_inode(inode);
5429 }
5430 
5431 /*
5432  * find the highest existing sequence number in a directory
5433  * and then set the in-memory index_cnt variable to reflect
5434  * free sequence numbers
5435  */
5436 static int btrfs_set_inode_index_count(struct inode *inode)
5437 {
5438 	struct btrfs_root *root = BTRFS_I(inode)->root;
5439 	struct btrfs_key key, found_key;
5440 	struct btrfs_path *path;
5441 	struct extent_buffer *leaf;
5442 	int ret;
5443 
5444 	key.objectid = btrfs_ino(inode);
5445 	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5446 	key.offset = (u64)-1;
5447 
5448 	path = btrfs_alloc_path();
5449 	if (!path)
5450 		return -ENOMEM;
5451 
5452 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5453 	if (ret < 0)
5454 		goto out;
5455 	/* FIXME: we should be able to handle this */
5456 	if (ret == 0)
5457 		goto out;
5458 	ret = 0;
5459 
5460 	/*
5461 	 * MAGIC NUMBER EXPLANATION:
5462 	 * since we search a directory based on f_pos we have to start at 2
5463 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5464 	 * else has to start at 2
5465 	 */
5466 	if (path->slots[0] == 0) {
5467 		BTRFS_I(inode)->index_cnt = 2;
5468 		goto out;
5469 	}
5470 
5471 	path->slots[0]--;
5472 
5473 	leaf = path->nodes[0];
5474 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5475 
5476 	if (found_key.objectid != btrfs_ino(inode) ||
5477 	    btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5478 		BTRFS_I(inode)->index_cnt = 2;
5479 		goto out;
5480 	}
5481 
5482 	BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5483 out:
5484 	btrfs_free_path(path);
5485 	return ret;
5486 }
5487 
5488 /*
5489  * helper to find a free sequence number in a given directory.  This current
5490  * code is very simple, later versions will do smarter things in the btree
5491  */
5492 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5493 {
5494 	int ret = 0;
5495 
5496 	if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5497 		ret = btrfs_inode_delayed_dir_index_count(dir);
5498 		if (ret) {
5499 			ret = btrfs_set_inode_index_count(dir);
5500 			if (ret)
5501 				return ret;
5502 		}
5503 	}
5504 
5505 	*index = BTRFS_I(dir)->index_cnt;
5506 	BTRFS_I(dir)->index_cnt++;
5507 
5508 	return ret;
5509 }
5510 
5511 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5512 				     struct btrfs_root *root,
5513 				     struct inode *dir,
5514 				     const char *name, int name_len,
5515 				     u64 ref_objectid, u64 objectid,
5516 				     umode_t mode, u64 *index)
5517 {
5518 	struct inode *inode;
5519 	struct btrfs_inode_item *inode_item;
5520 	struct btrfs_key *location;
5521 	struct btrfs_path *path;
5522 	struct btrfs_inode_ref *ref;
5523 	struct btrfs_key key[2];
5524 	u32 sizes[2];
5525 	unsigned long ptr;
5526 	int ret;
5527 
5528 	path = btrfs_alloc_path();
5529 	if (!path)
5530 		return ERR_PTR(-ENOMEM);
5531 
5532 	inode = new_inode(root->fs_info->sb);
5533 	if (!inode) {
5534 		btrfs_free_path(path);
5535 		return ERR_PTR(-ENOMEM);
5536 	}
5537 
5538 	/*
5539 	 * we have to initialize this early, so we can reclaim the inode
5540 	 * number if we fail afterwards in this function.
5541 	 */
5542 	inode->i_ino = objectid;
5543 
5544 	if (dir) {
5545 		trace_btrfs_inode_request(dir);
5546 
5547 		ret = btrfs_set_inode_index(dir, index);
5548 		if (ret) {
5549 			btrfs_free_path(path);
5550 			iput(inode);
5551 			return ERR_PTR(ret);
5552 		}
5553 	}
5554 	/*
5555 	 * index_cnt is ignored for everything but a dir,
5556 	 * btrfs_get_inode_index_count has an explanation for the magic
5557 	 * number
5558 	 */
5559 	BTRFS_I(inode)->index_cnt = 2;
5560 	BTRFS_I(inode)->dir_index = *index;
5561 	BTRFS_I(inode)->root = root;
5562 	BTRFS_I(inode)->generation = trans->transid;
5563 	inode->i_generation = BTRFS_I(inode)->generation;
5564 
5565 	/*
5566 	 * We could have gotten an inode number from somebody who was fsynced
5567 	 * and then removed in this same transaction, so let's just set full
5568 	 * sync since it will be a full sync anyway and this will blow away the
5569 	 * old info in the log.
5570 	 */
5571 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5572 
5573 	key[0].objectid = objectid;
5574 	btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5575 	key[0].offset = 0;
5576 
5577 	/*
5578 	 * Start new inodes with an inode_ref. This is slightly more
5579 	 * efficient for small numbers of hard links since they will
5580 	 * be packed into one item. Extended refs will kick in if we
5581 	 * add more hard links than can fit in the ref item.
5582 	 */
5583 	key[1].objectid = objectid;
5584 	btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5585 	key[1].offset = ref_objectid;
5586 
5587 	sizes[0] = sizeof(struct btrfs_inode_item);
5588 	sizes[1] = name_len + sizeof(*ref);
5589 
5590 	path->leave_spinning = 1;
5591 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5592 	if (ret != 0)
5593 		goto fail;
5594 
5595 	inode_init_owner(inode, dir, mode);
5596 	inode_set_bytes(inode, 0);
5597 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5598 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5599 				  struct btrfs_inode_item);
5600 	memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5601 			     sizeof(*inode_item));
5602 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
5603 
5604 	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5605 			     struct btrfs_inode_ref);
5606 	btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5607 	btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5608 	ptr = (unsigned long)(ref + 1);
5609 	write_extent_buffer(path->nodes[0], name, ptr, name_len);
5610 
5611 	btrfs_mark_buffer_dirty(path->nodes[0]);
5612 	btrfs_free_path(path);
5613 
5614 	location = &BTRFS_I(inode)->location;
5615 	location->objectid = objectid;
5616 	location->offset = 0;
5617 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5618 
5619 	btrfs_inherit_iflags(inode, dir);
5620 
5621 	if (S_ISREG(mode)) {
5622 		if (btrfs_test_opt(root, NODATASUM))
5623 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5624 		if (btrfs_test_opt(root, NODATACOW))
5625 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5626 				BTRFS_INODE_NODATASUM;
5627 	}
5628 
5629 	btrfs_insert_inode_hash(inode);
5630 	inode_tree_add(inode);
5631 
5632 	trace_btrfs_inode_new(inode);
5633 	btrfs_set_inode_last_trans(trans, inode);
5634 
5635 	btrfs_update_root_times(trans, root);
5636 
5637 	ret = btrfs_inode_inherit_props(trans, inode, dir);
5638 	if (ret)
5639 		btrfs_err(root->fs_info,
5640 			  "error inheriting props for ino %llu (root %llu): %d",
5641 			  btrfs_ino(inode), root->root_key.objectid, ret);
5642 
5643 	return inode;
5644 fail:
5645 	if (dir)
5646 		BTRFS_I(dir)->index_cnt--;
5647 	btrfs_free_path(path);
5648 	iput(inode);
5649 	return ERR_PTR(ret);
5650 }
5651 
5652 static inline u8 btrfs_inode_type(struct inode *inode)
5653 {
5654 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5655 }
5656 
5657 /*
5658  * utility function to add 'inode' into 'parent_inode' with
5659  * a give name and a given sequence number.
5660  * if 'add_backref' is true, also insert a backref from the
5661  * inode to the parent directory.
5662  */
5663 int btrfs_add_link(struct btrfs_trans_handle *trans,
5664 		   struct inode *parent_inode, struct inode *inode,
5665 		   const char *name, int name_len, int add_backref, u64 index)
5666 {
5667 	int ret = 0;
5668 	struct btrfs_key key;
5669 	struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5670 	u64 ino = btrfs_ino(inode);
5671 	u64 parent_ino = btrfs_ino(parent_inode);
5672 
5673 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5674 		memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5675 	} else {
5676 		key.objectid = ino;
5677 		btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5678 		key.offset = 0;
5679 	}
5680 
5681 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5682 		ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5683 					 key.objectid, root->root_key.objectid,
5684 					 parent_ino, index, name, name_len);
5685 	} else if (add_backref) {
5686 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5687 					     parent_ino, index);
5688 	}
5689 
5690 	/* Nothing to clean up yet */
5691 	if (ret)
5692 		return ret;
5693 
5694 	ret = btrfs_insert_dir_item(trans, root, name, name_len,
5695 				    parent_inode, &key,
5696 				    btrfs_inode_type(inode), index);
5697 	if (ret == -EEXIST || ret == -EOVERFLOW)
5698 		goto fail_dir_item;
5699 	else if (ret) {
5700 		btrfs_abort_transaction(trans, root, ret);
5701 		return ret;
5702 	}
5703 
5704 	btrfs_i_size_write(parent_inode, parent_inode->i_size +
5705 			   name_len * 2);
5706 	inode_inc_iversion(parent_inode);
5707 	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5708 	ret = btrfs_update_inode(trans, root, parent_inode);
5709 	if (ret)
5710 		btrfs_abort_transaction(trans, root, ret);
5711 	return ret;
5712 
5713 fail_dir_item:
5714 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5715 		u64 local_index;
5716 		int err;
5717 		err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5718 				 key.objectid, root->root_key.objectid,
5719 				 parent_ino, &local_index, name, name_len);
5720 
5721 	} else if (add_backref) {
5722 		u64 local_index;
5723 		int err;
5724 
5725 		err = btrfs_del_inode_ref(trans, root, name, name_len,
5726 					  ino, parent_ino, &local_index);
5727 	}
5728 	return ret;
5729 }
5730 
5731 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5732 			    struct inode *dir, struct dentry *dentry,
5733 			    struct inode *inode, int backref, u64 index)
5734 {
5735 	int err = btrfs_add_link(trans, dir, inode,
5736 				 dentry->d_name.name, dentry->d_name.len,
5737 				 backref, index);
5738 	if (err > 0)
5739 		err = -EEXIST;
5740 	return err;
5741 }
5742 
5743 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5744 			umode_t mode, dev_t rdev)
5745 {
5746 	struct btrfs_trans_handle *trans;
5747 	struct btrfs_root *root = BTRFS_I(dir)->root;
5748 	struct inode *inode = NULL;
5749 	int err;
5750 	int drop_inode = 0;
5751 	u64 objectid;
5752 	u64 index = 0;
5753 
5754 	if (!new_valid_dev(rdev))
5755 		return -EINVAL;
5756 
5757 	/*
5758 	 * 2 for inode item and ref
5759 	 * 2 for dir items
5760 	 * 1 for xattr if selinux is on
5761 	 */
5762 	trans = btrfs_start_transaction(root, 5);
5763 	if (IS_ERR(trans))
5764 		return PTR_ERR(trans);
5765 
5766 	err = btrfs_find_free_ino(root, &objectid);
5767 	if (err)
5768 		goto out_unlock;
5769 
5770 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5771 				dentry->d_name.len, btrfs_ino(dir), objectid,
5772 				mode, &index);
5773 	if (IS_ERR(inode)) {
5774 		err = PTR_ERR(inode);
5775 		goto out_unlock;
5776 	}
5777 
5778 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5779 	if (err) {
5780 		drop_inode = 1;
5781 		goto out_unlock;
5782 	}
5783 
5784 	/*
5785 	* If the active LSM wants to access the inode during
5786 	* d_instantiate it needs these. Smack checks to see
5787 	* if the filesystem supports xattrs by looking at the
5788 	* ops vector.
5789 	*/
5790 
5791 	inode->i_op = &btrfs_special_inode_operations;
5792 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5793 	if (err)
5794 		drop_inode = 1;
5795 	else {
5796 		init_special_inode(inode, inode->i_mode, rdev);
5797 		btrfs_update_inode(trans, root, inode);
5798 		d_instantiate(dentry, inode);
5799 	}
5800 out_unlock:
5801 	btrfs_end_transaction(trans, root);
5802 	btrfs_btree_balance_dirty(root);
5803 	if (drop_inode) {
5804 		inode_dec_link_count(inode);
5805 		iput(inode);
5806 	}
5807 	return err;
5808 }
5809 
5810 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5811 			umode_t mode, bool excl)
5812 {
5813 	struct btrfs_trans_handle *trans;
5814 	struct btrfs_root *root = BTRFS_I(dir)->root;
5815 	struct inode *inode = NULL;
5816 	int drop_inode_on_err = 0;
5817 	int err;
5818 	u64 objectid;
5819 	u64 index = 0;
5820 
5821 	/*
5822 	 * 2 for inode item and ref
5823 	 * 2 for dir items
5824 	 * 1 for xattr if selinux is on
5825 	 */
5826 	trans = btrfs_start_transaction(root, 5);
5827 	if (IS_ERR(trans))
5828 		return PTR_ERR(trans);
5829 
5830 	err = btrfs_find_free_ino(root, &objectid);
5831 	if (err)
5832 		goto out_unlock;
5833 
5834 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5835 				dentry->d_name.len, btrfs_ino(dir), objectid,
5836 				mode, &index);
5837 	if (IS_ERR(inode)) {
5838 		err = PTR_ERR(inode);
5839 		goto out_unlock;
5840 	}
5841 	drop_inode_on_err = 1;
5842 
5843 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5844 	if (err)
5845 		goto out_unlock;
5846 
5847 	err = btrfs_update_inode(trans, root, inode);
5848 	if (err)
5849 		goto out_unlock;
5850 
5851 	/*
5852 	* If the active LSM wants to access the inode during
5853 	* d_instantiate it needs these. Smack checks to see
5854 	* if the filesystem supports xattrs by looking at the
5855 	* ops vector.
5856 	*/
5857 	inode->i_fop = &btrfs_file_operations;
5858 	inode->i_op = &btrfs_file_inode_operations;
5859 
5860 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5861 	if (err)
5862 		goto out_unlock;
5863 
5864 	inode->i_mapping->a_ops = &btrfs_aops;
5865 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5866 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5867 	d_instantiate(dentry, inode);
5868 
5869 out_unlock:
5870 	btrfs_end_transaction(trans, root);
5871 	if (err && drop_inode_on_err) {
5872 		inode_dec_link_count(inode);
5873 		iput(inode);
5874 	}
5875 	btrfs_btree_balance_dirty(root);
5876 	return err;
5877 }
5878 
5879 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5880 		      struct dentry *dentry)
5881 {
5882 	struct btrfs_trans_handle *trans;
5883 	struct btrfs_root *root = BTRFS_I(dir)->root;
5884 	struct inode *inode = old_dentry->d_inode;
5885 	u64 index;
5886 	int err;
5887 	int drop_inode = 0;
5888 
5889 	/* do not allow sys_link's with other subvols of the same device */
5890 	if (root->objectid != BTRFS_I(inode)->root->objectid)
5891 		return -EXDEV;
5892 
5893 	if (inode->i_nlink >= BTRFS_LINK_MAX)
5894 		return -EMLINK;
5895 
5896 	err = btrfs_set_inode_index(dir, &index);
5897 	if (err)
5898 		goto fail;
5899 
5900 	/*
5901 	 * 2 items for inode and inode ref
5902 	 * 2 items for dir items
5903 	 * 1 item for parent inode
5904 	 */
5905 	trans = btrfs_start_transaction(root, 5);
5906 	if (IS_ERR(trans)) {
5907 		err = PTR_ERR(trans);
5908 		goto fail;
5909 	}
5910 
5911 	/* There are several dir indexes for this inode, clear the cache. */
5912 	BTRFS_I(inode)->dir_index = 0ULL;
5913 	inc_nlink(inode);
5914 	inode_inc_iversion(inode);
5915 	inode->i_ctime = CURRENT_TIME;
5916 	ihold(inode);
5917 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5918 
5919 	err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5920 
5921 	if (err) {
5922 		drop_inode = 1;
5923 	} else {
5924 		struct dentry *parent = dentry->d_parent;
5925 		err = btrfs_update_inode(trans, root, inode);
5926 		if (err)
5927 			goto fail;
5928 		d_instantiate(dentry, inode);
5929 		btrfs_log_new_name(trans, inode, NULL, parent);
5930 	}
5931 
5932 	btrfs_end_transaction(trans, root);
5933 fail:
5934 	if (drop_inode) {
5935 		inode_dec_link_count(inode);
5936 		iput(inode);
5937 	}
5938 	btrfs_btree_balance_dirty(root);
5939 	return err;
5940 }
5941 
5942 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5943 {
5944 	struct inode *inode = NULL;
5945 	struct btrfs_trans_handle *trans;
5946 	struct btrfs_root *root = BTRFS_I(dir)->root;
5947 	int err = 0;
5948 	int drop_on_err = 0;
5949 	u64 objectid = 0;
5950 	u64 index = 0;
5951 
5952 	/*
5953 	 * 2 items for inode and ref
5954 	 * 2 items for dir items
5955 	 * 1 for xattr if selinux is on
5956 	 */
5957 	trans = btrfs_start_transaction(root, 5);
5958 	if (IS_ERR(trans))
5959 		return PTR_ERR(trans);
5960 
5961 	err = btrfs_find_free_ino(root, &objectid);
5962 	if (err)
5963 		goto out_fail;
5964 
5965 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5966 				dentry->d_name.len, btrfs_ino(dir), objectid,
5967 				S_IFDIR | mode, &index);
5968 	if (IS_ERR(inode)) {
5969 		err = PTR_ERR(inode);
5970 		goto out_fail;
5971 	}
5972 
5973 	drop_on_err = 1;
5974 
5975 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5976 	if (err)
5977 		goto out_fail;
5978 
5979 	inode->i_op = &btrfs_dir_inode_operations;
5980 	inode->i_fop = &btrfs_dir_file_operations;
5981 
5982 	btrfs_i_size_write(inode, 0);
5983 	err = btrfs_update_inode(trans, root, inode);
5984 	if (err)
5985 		goto out_fail;
5986 
5987 	err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5988 			     dentry->d_name.len, 0, index);
5989 	if (err)
5990 		goto out_fail;
5991 
5992 	d_instantiate(dentry, inode);
5993 	drop_on_err = 0;
5994 
5995 out_fail:
5996 	btrfs_end_transaction(trans, root);
5997 	if (drop_on_err)
5998 		iput(inode);
5999 	btrfs_btree_balance_dirty(root);
6000 	return err;
6001 }
6002 
6003 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6004  * and an extent that you want to insert, deal with overlap and insert
6005  * the new extent into the tree.
6006  */
6007 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6008 				struct extent_map *existing,
6009 				struct extent_map *em,
6010 				u64 map_start, u64 map_len)
6011 {
6012 	u64 start_diff;
6013 
6014 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6015 	start_diff = map_start - em->start;
6016 	em->start = map_start;
6017 	em->len = map_len;
6018 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6019 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6020 		em->block_start += start_diff;
6021 		em->block_len -= start_diff;
6022 	}
6023 	return add_extent_mapping(em_tree, em, 0);
6024 }
6025 
6026 static noinline int uncompress_inline(struct btrfs_path *path,
6027 				      struct inode *inode, struct page *page,
6028 				      size_t pg_offset, u64 extent_offset,
6029 				      struct btrfs_file_extent_item *item)
6030 {
6031 	int ret;
6032 	struct extent_buffer *leaf = path->nodes[0];
6033 	char *tmp;
6034 	size_t max_size;
6035 	unsigned long inline_size;
6036 	unsigned long ptr;
6037 	int compress_type;
6038 
6039 	WARN_ON(pg_offset != 0);
6040 	compress_type = btrfs_file_extent_compression(leaf, item);
6041 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6042 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6043 					btrfs_item_nr(path->slots[0]));
6044 	tmp = kmalloc(inline_size, GFP_NOFS);
6045 	if (!tmp)
6046 		return -ENOMEM;
6047 	ptr = btrfs_file_extent_inline_start(item);
6048 
6049 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6050 
6051 	max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6052 	ret = btrfs_decompress(compress_type, tmp, page,
6053 			       extent_offset, inline_size, max_size);
6054 	if (ret) {
6055 		char *kaddr = kmap_atomic(page);
6056 		unsigned long copy_size = min_t(u64,
6057 				  PAGE_CACHE_SIZE - pg_offset,
6058 				  max_size - extent_offset);
6059 		memset(kaddr + pg_offset, 0, copy_size);
6060 		kunmap_atomic(kaddr);
6061 	}
6062 	kfree(tmp);
6063 	return 0;
6064 }
6065 
6066 /*
6067  * a bit scary, this does extent mapping from logical file offset to the disk.
6068  * the ugly parts come from merging extents from the disk with the in-ram
6069  * representation.  This gets more complex because of the data=ordered code,
6070  * where the in-ram extents might be locked pending data=ordered completion.
6071  *
6072  * This also copies inline extents directly into the page.
6073  */
6074 
6075 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6076 				    size_t pg_offset, u64 start, u64 len,
6077 				    int create)
6078 {
6079 	int ret;
6080 	int err = 0;
6081 	u64 bytenr;
6082 	u64 extent_start = 0;
6083 	u64 extent_end = 0;
6084 	u64 objectid = btrfs_ino(inode);
6085 	u32 found_type;
6086 	struct btrfs_path *path = NULL;
6087 	struct btrfs_root *root = BTRFS_I(inode)->root;
6088 	struct btrfs_file_extent_item *item;
6089 	struct extent_buffer *leaf;
6090 	struct btrfs_key found_key;
6091 	struct extent_map *em = NULL;
6092 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6093 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6094 	struct btrfs_trans_handle *trans = NULL;
6095 	int compress_type;
6096 
6097 again:
6098 	read_lock(&em_tree->lock);
6099 	em = lookup_extent_mapping(em_tree, start, len);
6100 	if (em)
6101 		em->bdev = root->fs_info->fs_devices->latest_bdev;
6102 	read_unlock(&em_tree->lock);
6103 
6104 	if (em) {
6105 		if (em->start > start || em->start + em->len <= start)
6106 			free_extent_map(em);
6107 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6108 			free_extent_map(em);
6109 		else
6110 			goto out;
6111 	}
6112 	em = alloc_extent_map();
6113 	if (!em) {
6114 		err = -ENOMEM;
6115 		goto out;
6116 	}
6117 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6118 	em->start = EXTENT_MAP_HOLE;
6119 	em->orig_start = EXTENT_MAP_HOLE;
6120 	em->len = (u64)-1;
6121 	em->block_len = (u64)-1;
6122 
6123 	if (!path) {
6124 		path = btrfs_alloc_path();
6125 		if (!path) {
6126 			err = -ENOMEM;
6127 			goto out;
6128 		}
6129 		/*
6130 		 * Chances are we'll be called again, so go ahead and do
6131 		 * readahead
6132 		 */
6133 		path->reada = 1;
6134 	}
6135 
6136 	ret = btrfs_lookup_file_extent(trans, root, path,
6137 				       objectid, start, trans != NULL);
6138 	if (ret < 0) {
6139 		err = ret;
6140 		goto out;
6141 	}
6142 
6143 	if (ret != 0) {
6144 		if (path->slots[0] == 0)
6145 			goto not_found;
6146 		path->slots[0]--;
6147 	}
6148 
6149 	leaf = path->nodes[0];
6150 	item = btrfs_item_ptr(leaf, path->slots[0],
6151 			      struct btrfs_file_extent_item);
6152 	/* are we inside the extent that was found? */
6153 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6154 	found_type = btrfs_key_type(&found_key);
6155 	if (found_key.objectid != objectid ||
6156 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6157 		/*
6158 		 * If we backup past the first extent we want to move forward
6159 		 * and see if there is an extent in front of us, otherwise we'll
6160 		 * say there is a hole for our whole search range which can
6161 		 * cause problems.
6162 		 */
6163 		extent_end = start;
6164 		goto next;
6165 	}
6166 
6167 	found_type = btrfs_file_extent_type(leaf, item);
6168 	extent_start = found_key.offset;
6169 	compress_type = btrfs_file_extent_compression(leaf, item);
6170 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6171 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6172 		extent_end = extent_start +
6173 		       btrfs_file_extent_num_bytes(leaf, item);
6174 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6175 		size_t size;
6176 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6177 		extent_end = ALIGN(extent_start + size, root->sectorsize);
6178 	}
6179 next:
6180 	if (start >= extent_end) {
6181 		path->slots[0]++;
6182 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6183 			ret = btrfs_next_leaf(root, path);
6184 			if (ret < 0) {
6185 				err = ret;
6186 				goto out;
6187 			}
6188 			if (ret > 0)
6189 				goto not_found;
6190 			leaf = path->nodes[0];
6191 		}
6192 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6193 		if (found_key.objectid != objectid ||
6194 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6195 			goto not_found;
6196 		if (start + len <= found_key.offset)
6197 			goto not_found;
6198 		em->start = start;
6199 		em->orig_start = start;
6200 		em->len = found_key.offset - start;
6201 		goto not_found_em;
6202 	}
6203 
6204 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6205 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6206 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6207 		em->start = extent_start;
6208 		em->len = extent_end - extent_start;
6209 		em->orig_start = extent_start -
6210 				 btrfs_file_extent_offset(leaf, item);
6211 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6212 								      item);
6213 		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6214 		if (bytenr == 0) {
6215 			em->block_start = EXTENT_MAP_HOLE;
6216 			goto insert;
6217 		}
6218 		if (compress_type != BTRFS_COMPRESS_NONE) {
6219 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6220 			em->compress_type = compress_type;
6221 			em->block_start = bytenr;
6222 			em->block_len = em->orig_block_len;
6223 		} else {
6224 			bytenr += btrfs_file_extent_offset(leaf, item);
6225 			em->block_start = bytenr;
6226 			em->block_len = em->len;
6227 			if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6228 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6229 		}
6230 		goto insert;
6231 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6232 		unsigned long ptr;
6233 		char *map;
6234 		size_t size;
6235 		size_t extent_offset;
6236 		size_t copy_size;
6237 
6238 		em->block_start = EXTENT_MAP_INLINE;
6239 		if (!page || create) {
6240 			em->start = extent_start;
6241 			em->len = extent_end - extent_start;
6242 			goto out;
6243 		}
6244 
6245 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6246 		extent_offset = page_offset(page) + pg_offset - extent_start;
6247 		copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6248 				size - extent_offset);
6249 		em->start = extent_start + extent_offset;
6250 		em->len = ALIGN(copy_size, root->sectorsize);
6251 		em->orig_block_len = em->len;
6252 		em->orig_start = em->start;
6253 		if (compress_type) {
6254 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6255 			em->compress_type = compress_type;
6256 		}
6257 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6258 		if (create == 0 && !PageUptodate(page)) {
6259 			if (btrfs_file_extent_compression(leaf, item) !=
6260 			    BTRFS_COMPRESS_NONE) {
6261 				ret = uncompress_inline(path, inode, page,
6262 							pg_offset,
6263 							extent_offset, item);
6264 				BUG_ON(ret); /* -ENOMEM */
6265 			} else {
6266 				map = kmap(page);
6267 				read_extent_buffer(leaf, map + pg_offset, ptr,
6268 						   copy_size);
6269 				if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6270 					memset(map + pg_offset + copy_size, 0,
6271 					       PAGE_CACHE_SIZE - pg_offset -
6272 					       copy_size);
6273 				}
6274 				kunmap(page);
6275 			}
6276 			flush_dcache_page(page);
6277 		} else if (create && PageUptodate(page)) {
6278 			BUG();
6279 			if (!trans) {
6280 				kunmap(page);
6281 				free_extent_map(em);
6282 				em = NULL;
6283 
6284 				btrfs_release_path(path);
6285 				trans = btrfs_join_transaction(root);
6286 
6287 				if (IS_ERR(trans))
6288 					return ERR_CAST(trans);
6289 				goto again;
6290 			}
6291 			map = kmap(page);
6292 			write_extent_buffer(leaf, map + pg_offset, ptr,
6293 					    copy_size);
6294 			kunmap(page);
6295 			btrfs_mark_buffer_dirty(leaf);
6296 		}
6297 		set_extent_uptodate(io_tree, em->start,
6298 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
6299 		goto insert;
6300 	} else {
6301 		WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6302 	}
6303 not_found:
6304 	em->start = start;
6305 	em->orig_start = start;
6306 	em->len = len;
6307 not_found_em:
6308 	em->block_start = EXTENT_MAP_HOLE;
6309 	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6310 insert:
6311 	btrfs_release_path(path);
6312 	if (em->start > start || extent_map_end(em) <= start) {
6313 		btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6314 			em->start, em->len, start, len);
6315 		err = -EIO;
6316 		goto out;
6317 	}
6318 
6319 	err = 0;
6320 	write_lock(&em_tree->lock);
6321 	ret = add_extent_mapping(em_tree, em, 0);
6322 	/* it is possible that someone inserted the extent into the tree
6323 	 * while we had the lock dropped.  It is also possible that
6324 	 * an overlapping map exists in the tree
6325 	 */
6326 	if (ret == -EEXIST) {
6327 		struct extent_map *existing;
6328 
6329 		ret = 0;
6330 
6331 		existing = lookup_extent_mapping(em_tree, start, len);
6332 		if (existing && (existing->start > start ||
6333 		    existing->start + existing->len <= start)) {
6334 			free_extent_map(existing);
6335 			existing = NULL;
6336 		}
6337 		if (!existing) {
6338 			existing = lookup_extent_mapping(em_tree, em->start,
6339 							 em->len);
6340 			if (existing) {
6341 				err = merge_extent_mapping(em_tree, existing,
6342 							   em, start,
6343 							   root->sectorsize);
6344 				free_extent_map(existing);
6345 				if (err) {
6346 					free_extent_map(em);
6347 					em = NULL;
6348 				}
6349 			} else {
6350 				err = -EIO;
6351 				free_extent_map(em);
6352 				em = NULL;
6353 			}
6354 		} else {
6355 			free_extent_map(em);
6356 			em = existing;
6357 			err = 0;
6358 		}
6359 	}
6360 	write_unlock(&em_tree->lock);
6361 out:
6362 
6363 	trace_btrfs_get_extent(root, em);
6364 
6365 	if (path)
6366 		btrfs_free_path(path);
6367 	if (trans) {
6368 		ret = btrfs_end_transaction(trans, root);
6369 		if (!err)
6370 			err = ret;
6371 	}
6372 	if (err) {
6373 		free_extent_map(em);
6374 		return ERR_PTR(err);
6375 	}
6376 	BUG_ON(!em); /* Error is always set */
6377 	return em;
6378 }
6379 
6380 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6381 					   size_t pg_offset, u64 start, u64 len,
6382 					   int create)
6383 {
6384 	struct extent_map *em;
6385 	struct extent_map *hole_em = NULL;
6386 	u64 range_start = start;
6387 	u64 end;
6388 	u64 found;
6389 	u64 found_end;
6390 	int err = 0;
6391 
6392 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6393 	if (IS_ERR(em))
6394 		return em;
6395 	if (em) {
6396 		/*
6397 		 * if our em maps to
6398 		 * -  a hole or
6399 		 * -  a pre-alloc extent,
6400 		 * there might actually be delalloc bytes behind it.
6401 		 */
6402 		if (em->block_start != EXTENT_MAP_HOLE &&
6403 		    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6404 			return em;
6405 		else
6406 			hole_em = em;
6407 	}
6408 
6409 	/* check to see if we've wrapped (len == -1 or similar) */
6410 	end = start + len;
6411 	if (end < start)
6412 		end = (u64)-1;
6413 	else
6414 		end -= 1;
6415 
6416 	em = NULL;
6417 
6418 	/* ok, we didn't find anything, lets look for delalloc */
6419 	found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6420 				 end, len, EXTENT_DELALLOC, 1);
6421 	found_end = range_start + found;
6422 	if (found_end < range_start)
6423 		found_end = (u64)-1;
6424 
6425 	/*
6426 	 * we didn't find anything useful, return
6427 	 * the original results from get_extent()
6428 	 */
6429 	if (range_start > end || found_end <= start) {
6430 		em = hole_em;
6431 		hole_em = NULL;
6432 		goto out;
6433 	}
6434 
6435 	/* adjust the range_start to make sure it doesn't
6436 	 * go backwards from the start they passed in
6437 	 */
6438 	range_start = max(start, range_start);
6439 	found = found_end - range_start;
6440 
6441 	if (found > 0) {
6442 		u64 hole_start = start;
6443 		u64 hole_len = len;
6444 
6445 		em = alloc_extent_map();
6446 		if (!em) {
6447 			err = -ENOMEM;
6448 			goto out;
6449 		}
6450 		/*
6451 		 * when btrfs_get_extent can't find anything it
6452 		 * returns one huge hole
6453 		 *
6454 		 * make sure what it found really fits our range, and
6455 		 * adjust to make sure it is based on the start from
6456 		 * the caller
6457 		 */
6458 		if (hole_em) {
6459 			u64 calc_end = extent_map_end(hole_em);
6460 
6461 			if (calc_end <= start || (hole_em->start > end)) {
6462 				free_extent_map(hole_em);
6463 				hole_em = NULL;
6464 			} else {
6465 				hole_start = max(hole_em->start, start);
6466 				hole_len = calc_end - hole_start;
6467 			}
6468 		}
6469 		em->bdev = NULL;
6470 		if (hole_em && range_start > hole_start) {
6471 			/* our hole starts before our delalloc, so we
6472 			 * have to return just the parts of the hole
6473 			 * that go until  the delalloc starts
6474 			 */
6475 			em->len = min(hole_len,
6476 				      range_start - hole_start);
6477 			em->start = hole_start;
6478 			em->orig_start = hole_start;
6479 			/*
6480 			 * don't adjust block start at all,
6481 			 * it is fixed at EXTENT_MAP_HOLE
6482 			 */
6483 			em->block_start = hole_em->block_start;
6484 			em->block_len = hole_len;
6485 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6486 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6487 		} else {
6488 			em->start = range_start;
6489 			em->len = found;
6490 			em->orig_start = range_start;
6491 			em->block_start = EXTENT_MAP_DELALLOC;
6492 			em->block_len = found;
6493 		}
6494 	} else if (hole_em) {
6495 		return hole_em;
6496 	}
6497 out:
6498 
6499 	free_extent_map(hole_em);
6500 	if (err) {
6501 		free_extent_map(em);
6502 		return ERR_PTR(err);
6503 	}
6504 	return em;
6505 }
6506 
6507 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6508 						  u64 start, u64 len)
6509 {
6510 	struct btrfs_root *root = BTRFS_I(inode)->root;
6511 	struct extent_map *em;
6512 	struct btrfs_key ins;
6513 	u64 alloc_hint;
6514 	int ret;
6515 
6516 	alloc_hint = get_extent_allocation_hint(inode, start, len);
6517 	ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6518 				   alloc_hint, &ins, 1);
6519 	if (ret)
6520 		return ERR_PTR(ret);
6521 
6522 	em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6523 			      ins.offset, ins.offset, ins.offset, 0);
6524 	if (IS_ERR(em)) {
6525 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6526 		return em;
6527 	}
6528 
6529 	ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6530 					   ins.offset, ins.offset, 0);
6531 	if (ret) {
6532 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6533 		free_extent_map(em);
6534 		return ERR_PTR(ret);
6535 	}
6536 
6537 	return em;
6538 }
6539 
6540 /*
6541  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6542  * block must be cow'd
6543  */
6544 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6545 			      u64 *orig_start, u64 *orig_block_len,
6546 			      u64 *ram_bytes)
6547 {
6548 	struct btrfs_trans_handle *trans;
6549 	struct btrfs_path *path;
6550 	int ret;
6551 	struct extent_buffer *leaf;
6552 	struct btrfs_root *root = BTRFS_I(inode)->root;
6553 	struct btrfs_file_extent_item *fi;
6554 	struct btrfs_key key;
6555 	u64 disk_bytenr;
6556 	u64 backref_offset;
6557 	u64 extent_end;
6558 	u64 num_bytes;
6559 	int slot;
6560 	int found_type;
6561 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6562 
6563 	path = btrfs_alloc_path();
6564 	if (!path)
6565 		return -ENOMEM;
6566 
6567 	ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6568 				       offset, 0);
6569 	if (ret < 0)
6570 		goto out;
6571 
6572 	slot = path->slots[0];
6573 	if (ret == 1) {
6574 		if (slot == 0) {
6575 			/* can't find the item, must cow */
6576 			ret = 0;
6577 			goto out;
6578 		}
6579 		slot--;
6580 	}
6581 	ret = 0;
6582 	leaf = path->nodes[0];
6583 	btrfs_item_key_to_cpu(leaf, &key, slot);
6584 	if (key.objectid != btrfs_ino(inode) ||
6585 	    key.type != BTRFS_EXTENT_DATA_KEY) {
6586 		/* not our file or wrong item type, must cow */
6587 		goto out;
6588 	}
6589 
6590 	if (key.offset > offset) {
6591 		/* Wrong offset, must cow */
6592 		goto out;
6593 	}
6594 
6595 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6596 	found_type = btrfs_file_extent_type(leaf, fi);
6597 	if (found_type != BTRFS_FILE_EXTENT_REG &&
6598 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6599 		/* not a regular extent, must cow */
6600 		goto out;
6601 	}
6602 
6603 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6604 		goto out;
6605 
6606 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6607 	if (extent_end <= offset)
6608 		goto out;
6609 
6610 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6611 	if (disk_bytenr == 0)
6612 		goto out;
6613 
6614 	if (btrfs_file_extent_compression(leaf, fi) ||
6615 	    btrfs_file_extent_encryption(leaf, fi) ||
6616 	    btrfs_file_extent_other_encoding(leaf, fi))
6617 		goto out;
6618 
6619 	backref_offset = btrfs_file_extent_offset(leaf, fi);
6620 
6621 	if (orig_start) {
6622 		*orig_start = key.offset - backref_offset;
6623 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6624 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6625 	}
6626 
6627 	if (btrfs_extent_readonly(root, disk_bytenr))
6628 		goto out;
6629 	btrfs_release_path(path);
6630 
6631 	/*
6632 	 * look for other files referencing this extent, if we
6633 	 * find any we must cow
6634 	 */
6635 	trans = btrfs_join_transaction(root);
6636 	if (IS_ERR(trans)) {
6637 		ret = 0;
6638 		goto out;
6639 	}
6640 
6641 	ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6642 				    key.offset - backref_offset, disk_bytenr);
6643 	btrfs_end_transaction(trans, root);
6644 	if (ret) {
6645 		ret = 0;
6646 		goto out;
6647 	}
6648 
6649 	/*
6650 	 * adjust disk_bytenr and num_bytes to cover just the bytes
6651 	 * in this extent we are about to write.  If there
6652 	 * are any csums in that range we have to cow in order
6653 	 * to keep the csums correct
6654 	 */
6655 	disk_bytenr += backref_offset;
6656 	disk_bytenr += offset - key.offset;
6657 	num_bytes = min(offset + *len, extent_end) - offset;
6658 	if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6659 				goto out;
6660 	/*
6661 	 * all of the above have passed, it is safe to overwrite this extent
6662 	 * without cow
6663 	 */
6664 	*len = num_bytes;
6665 	ret = 1;
6666 out:
6667 	btrfs_free_path(path);
6668 	return ret;
6669 }
6670 
6671 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6672 			      struct extent_state **cached_state, int writing)
6673 {
6674 	struct btrfs_ordered_extent *ordered;
6675 	int ret = 0;
6676 
6677 	while (1) {
6678 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6679 				 0, cached_state);
6680 		/*
6681 		 * We're concerned with the entire range that we're going to be
6682 		 * doing DIO to, so we need to make sure theres no ordered
6683 		 * extents in this range.
6684 		 */
6685 		ordered = btrfs_lookup_ordered_range(inode, lockstart,
6686 						     lockend - lockstart + 1);
6687 
6688 		/*
6689 		 * We need to make sure there are no buffered pages in this
6690 		 * range either, we could have raced between the invalidate in
6691 		 * generic_file_direct_write and locking the extent.  The
6692 		 * invalidate needs to happen so that reads after a write do not
6693 		 * get stale data.
6694 		 */
6695 		if (!ordered && (!writing ||
6696 		    !test_range_bit(&BTRFS_I(inode)->io_tree,
6697 				    lockstart, lockend, EXTENT_UPTODATE, 0,
6698 				    *cached_state)))
6699 			break;
6700 
6701 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6702 				     cached_state, GFP_NOFS);
6703 
6704 		if (ordered) {
6705 			btrfs_start_ordered_extent(inode, ordered, 1);
6706 			btrfs_put_ordered_extent(ordered);
6707 		} else {
6708 			/* Screw you mmap */
6709 			ret = filemap_write_and_wait_range(inode->i_mapping,
6710 							   lockstart,
6711 							   lockend);
6712 			if (ret)
6713 				break;
6714 
6715 			/*
6716 			 * If we found a page that couldn't be invalidated just
6717 			 * fall back to buffered.
6718 			 */
6719 			ret = invalidate_inode_pages2_range(inode->i_mapping,
6720 					lockstart >> PAGE_CACHE_SHIFT,
6721 					lockend >> PAGE_CACHE_SHIFT);
6722 			if (ret)
6723 				break;
6724 		}
6725 
6726 		cond_resched();
6727 	}
6728 
6729 	return ret;
6730 }
6731 
6732 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6733 					   u64 len, u64 orig_start,
6734 					   u64 block_start, u64 block_len,
6735 					   u64 orig_block_len, u64 ram_bytes,
6736 					   int type)
6737 {
6738 	struct extent_map_tree *em_tree;
6739 	struct extent_map *em;
6740 	struct btrfs_root *root = BTRFS_I(inode)->root;
6741 	int ret;
6742 
6743 	em_tree = &BTRFS_I(inode)->extent_tree;
6744 	em = alloc_extent_map();
6745 	if (!em)
6746 		return ERR_PTR(-ENOMEM);
6747 
6748 	em->start = start;
6749 	em->orig_start = orig_start;
6750 	em->mod_start = start;
6751 	em->mod_len = len;
6752 	em->len = len;
6753 	em->block_len = block_len;
6754 	em->block_start = block_start;
6755 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6756 	em->orig_block_len = orig_block_len;
6757 	em->ram_bytes = ram_bytes;
6758 	em->generation = -1;
6759 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
6760 	if (type == BTRFS_ORDERED_PREALLOC)
6761 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
6762 
6763 	do {
6764 		btrfs_drop_extent_cache(inode, em->start,
6765 				em->start + em->len - 1, 0);
6766 		write_lock(&em_tree->lock);
6767 		ret = add_extent_mapping(em_tree, em, 1);
6768 		write_unlock(&em_tree->lock);
6769 	} while (ret == -EEXIST);
6770 
6771 	if (ret) {
6772 		free_extent_map(em);
6773 		return ERR_PTR(ret);
6774 	}
6775 
6776 	return em;
6777 }
6778 
6779 
6780 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6781 				   struct buffer_head *bh_result, int create)
6782 {
6783 	struct extent_map *em;
6784 	struct btrfs_root *root = BTRFS_I(inode)->root;
6785 	struct extent_state *cached_state = NULL;
6786 	u64 start = iblock << inode->i_blkbits;
6787 	u64 lockstart, lockend;
6788 	u64 len = bh_result->b_size;
6789 	int unlock_bits = EXTENT_LOCKED;
6790 	int ret = 0;
6791 
6792 	if (create)
6793 		unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6794 	else
6795 		len = min_t(u64, len, root->sectorsize);
6796 
6797 	lockstart = start;
6798 	lockend = start + len - 1;
6799 
6800 	/*
6801 	 * If this errors out it's because we couldn't invalidate pagecache for
6802 	 * this range and we need to fallback to buffered.
6803 	 */
6804 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6805 		return -ENOTBLK;
6806 
6807 	em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6808 	if (IS_ERR(em)) {
6809 		ret = PTR_ERR(em);
6810 		goto unlock_err;
6811 	}
6812 
6813 	/*
6814 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6815 	 * io.  INLINE is special, and we could probably kludge it in here, but
6816 	 * it's still buffered so for safety lets just fall back to the generic
6817 	 * buffered path.
6818 	 *
6819 	 * For COMPRESSED we _have_ to read the entire extent in so we can
6820 	 * decompress it, so there will be buffering required no matter what we
6821 	 * do, so go ahead and fallback to buffered.
6822 	 *
6823 	 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6824 	 * to buffered IO.  Don't blame me, this is the price we pay for using
6825 	 * the generic code.
6826 	 */
6827 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6828 	    em->block_start == EXTENT_MAP_INLINE) {
6829 		free_extent_map(em);
6830 		ret = -ENOTBLK;
6831 		goto unlock_err;
6832 	}
6833 
6834 	/* Just a good old fashioned hole, return */
6835 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6836 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6837 		free_extent_map(em);
6838 		goto unlock_err;
6839 	}
6840 
6841 	/*
6842 	 * We don't allocate a new extent in the following cases
6843 	 *
6844 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6845 	 * existing extent.
6846 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
6847 	 * just use the extent.
6848 	 *
6849 	 */
6850 	if (!create) {
6851 		len = min(len, em->len - (start - em->start));
6852 		lockstart = start + len;
6853 		goto unlock;
6854 	}
6855 
6856 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6857 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6858 	     em->block_start != EXTENT_MAP_HOLE)) {
6859 		int type;
6860 		int ret;
6861 		u64 block_start, orig_start, orig_block_len, ram_bytes;
6862 
6863 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6864 			type = BTRFS_ORDERED_PREALLOC;
6865 		else
6866 			type = BTRFS_ORDERED_NOCOW;
6867 		len = min(len, em->len - (start - em->start));
6868 		block_start = em->block_start + (start - em->start);
6869 
6870 		if (can_nocow_extent(inode, start, &len, &orig_start,
6871 				     &orig_block_len, &ram_bytes) == 1) {
6872 			if (type == BTRFS_ORDERED_PREALLOC) {
6873 				free_extent_map(em);
6874 				em = create_pinned_em(inode, start, len,
6875 						       orig_start,
6876 						       block_start, len,
6877 						       orig_block_len,
6878 						       ram_bytes, type);
6879 				if (IS_ERR(em))
6880 					goto unlock_err;
6881 			}
6882 
6883 			ret = btrfs_add_ordered_extent_dio(inode, start,
6884 					   block_start, len, len, type);
6885 			if (ret) {
6886 				free_extent_map(em);
6887 				goto unlock_err;
6888 			}
6889 			goto unlock;
6890 		}
6891 	}
6892 
6893 	/*
6894 	 * this will cow the extent, reset the len in case we changed
6895 	 * it above
6896 	 */
6897 	len = bh_result->b_size;
6898 	free_extent_map(em);
6899 	em = btrfs_new_extent_direct(inode, start, len);
6900 	if (IS_ERR(em)) {
6901 		ret = PTR_ERR(em);
6902 		goto unlock_err;
6903 	}
6904 	len = min(len, em->len - (start - em->start));
6905 unlock:
6906 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6907 		inode->i_blkbits;
6908 	bh_result->b_size = len;
6909 	bh_result->b_bdev = em->bdev;
6910 	set_buffer_mapped(bh_result);
6911 	if (create) {
6912 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6913 			set_buffer_new(bh_result);
6914 
6915 		/*
6916 		 * Need to update the i_size under the extent lock so buffered
6917 		 * readers will get the updated i_size when we unlock.
6918 		 */
6919 		if (start + len > i_size_read(inode))
6920 			i_size_write(inode, start + len);
6921 
6922 		spin_lock(&BTRFS_I(inode)->lock);
6923 		BTRFS_I(inode)->outstanding_extents++;
6924 		spin_unlock(&BTRFS_I(inode)->lock);
6925 
6926 		ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6927 				     lockstart + len - 1, EXTENT_DELALLOC, NULL,
6928 				     &cached_state, GFP_NOFS);
6929 		BUG_ON(ret);
6930 	}
6931 
6932 	/*
6933 	 * In the case of write we need to clear and unlock the entire range,
6934 	 * in the case of read we need to unlock only the end area that we
6935 	 * aren't using if there is any left over space.
6936 	 */
6937 	if (lockstart < lockend) {
6938 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6939 				 lockend, unlock_bits, 1, 0,
6940 				 &cached_state, GFP_NOFS);
6941 	} else {
6942 		free_extent_state(cached_state);
6943 	}
6944 
6945 	free_extent_map(em);
6946 
6947 	return 0;
6948 
6949 unlock_err:
6950 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6951 			 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6952 	return ret;
6953 }
6954 
6955 static void btrfs_endio_direct_read(struct bio *bio, int err)
6956 {
6957 	struct btrfs_dio_private *dip = bio->bi_private;
6958 	struct bio_vec *bvec;
6959 	struct inode *inode = dip->inode;
6960 	struct btrfs_root *root = BTRFS_I(inode)->root;
6961 	struct bio *dio_bio;
6962 	u32 *csums = (u32 *)dip->csum;
6963 	u64 start;
6964 	int i;
6965 
6966 	start = dip->logical_offset;
6967 	bio_for_each_segment_all(bvec, bio, i) {
6968 		if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6969 			struct page *page = bvec->bv_page;
6970 			char *kaddr;
6971 			u32 csum = ~(u32)0;
6972 			unsigned long flags;
6973 
6974 			local_irq_save(flags);
6975 			kaddr = kmap_atomic(page);
6976 			csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6977 					       csum, bvec->bv_len);
6978 			btrfs_csum_final(csum, (char *)&csum);
6979 			kunmap_atomic(kaddr);
6980 			local_irq_restore(flags);
6981 
6982 			flush_dcache_page(bvec->bv_page);
6983 			if (csum != csums[i]) {
6984 				btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6985 					  btrfs_ino(inode), start, csum,
6986 					  csums[i]);
6987 				err = -EIO;
6988 			}
6989 		}
6990 
6991 		start += bvec->bv_len;
6992 	}
6993 
6994 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6995 		      dip->logical_offset + dip->bytes - 1);
6996 	dio_bio = dip->dio_bio;
6997 
6998 	kfree(dip);
6999 
7000 	/* If we had a csum failure make sure to clear the uptodate flag */
7001 	if (err)
7002 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7003 	dio_end_io(dio_bio, err);
7004 	bio_put(bio);
7005 }
7006 
7007 static void btrfs_endio_direct_write(struct bio *bio, int err)
7008 {
7009 	struct btrfs_dio_private *dip = bio->bi_private;
7010 	struct inode *inode = dip->inode;
7011 	struct btrfs_root *root = BTRFS_I(inode)->root;
7012 	struct btrfs_ordered_extent *ordered = NULL;
7013 	u64 ordered_offset = dip->logical_offset;
7014 	u64 ordered_bytes = dip->bytes;
7015 	struct bio *dio_bio;
7016 	int ret;
7017 
7018 	if (err)
7019 		goto out_done;
7020 again:
7021 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7022 						   &ordered_offset,
7023 						   ordered_bytes, !err);
7024 	if (!ret)
7025 		goto out_test;
7026 
7027 	ordered->work.func = finish_ordered_fn;
7028 	ordered->work.flags = 0;
7029 	btrfs_queue_worker(&root->fs_info->endio_write_workers,
7030 			   &ordered->work);
7031 out_test:
7032 	/*
7033 	 * our bio might span multiple ordered extents.  If we haven't
7034 	 * completed the accounting for the whole dio, go back and try again
7035 	 */
7036 	if (ordered_offset < dip->logical_offset + dip->bytes) {
7037 		ordered_bytes = dip->logical_offset + dip->bytes -
7038 			ordered_offset;
7039 		ordered = NULL;
7040 		goto again;
7041 	}
7042 out_done:
7043 	dio_bio = dip->dio_bio;
7044 
7045 	kfree(dip);
7046 
7047 	/* If we had an error make sure to clear the uptodate flag */
7048 	if (err)
7049 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7050 	dio_end_io(dio_bio, err);
7051 	bio_put(bio);
7052 }
7053 
7054 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7055 				    struct bio *bio, int mirror_num,
7056 				    unsigned long bio_flags, u64 offset)
7057 {
7058 	int ret;
7059 	struct btrfs_root *root = BTRFS_I(inode)->root;
7060 	ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7061 	BUG_ON(ret); /* -ENOMEM */
7062 	return 0;
7063 }
7064 
7065 static void btrfs_end_dio_bio(struct bio *bio, int err)
7066 {
7067 	struct btrfs_dio_private *dip = bio->bi_private;
7068 
7069 	if (err) {
7070 		btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7071 			  "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7072 		      btrfs_ino(dip->inode), bio->bi_rw,
7073 		      (unsigned long long)bio->bi_iter.bi_sector,
7074 		      bio->bi_iter.bi_size, err);
7075 		dip->errors = 1;
7076 
7077 		/*
7078 		 * before atomic variable goto zero, we must make sure
7079 		 * dip->errors is perceived to be set.
7080 		 */
7081 		smp_mb__before_atomic_dec();
7082 	}
7083 
7084 	/* if there are more bios still pending for this dio, just exit */
7085 	if (!atomic_dec_and_test(&dip->pending_bios))
7086 		goto out;
7087 
7088 	if (dip->errors) {
7089 		bio_io_error(dip->orig_bio);
7090 	} else {
7091 		set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7092 		bio_endio(dip->orig_bio, 0);
7093 	}
7094 out:
7095 	bio_put(bio);
7096 }
7097 
7098 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7099 				       u64 first_sector, gfp_t gfp_flags)
7100 {
7101 	int nr_vecs = bio_get_nr_vecs(bdev);
7102 	return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7103 }
7104 
7105 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7106 					 int rw, u64 file_offset, int skip_sum,
7107 					 int async_submit)
7108 {
7109 	struct btrfs_dio_private *dip = bio->bi_private;
7110 	int write = rw & REQ_WRITE;
7111 	struct btrfs_root *root = BTRFS_I(inode)->root;
7112 	int ret;
7113 
7114 	if (async_submit)
7115 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7116 
7117 	bio_get(bio);
7118 
7119 	if (!write) {
7120 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7121 		if (ret)
7122 			goto err;
7123 	}
7124 
7125 	if (skip_sum)
7126 		goto map;
7127 
7128 	if (write && async_submit) {
7129 		ret = btrfs_wq_submit_bio(root->fs_info,
7130 				   inode, rw, bio, 0, 0,
7131 				   file_offset,
7132 				   __btrfs_submit_bio_start_direct_io,
7133 				   __btrfs_submit_bio_done);
7134 		goto err;
7135 	} else if (write) {
7136 		/*
7137 		 * If we aren't doing async submit, calculate the csum of the
7138 		 * bio now.
7139 		 */
7140 		ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7141 		if (ret)
7142 			goto err;
7143 	} else if (!skip_sum) {
7144 		ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7145 						file_offset);
7146 		if (ret)
7147 			goto err;
7148 	}
7149 
7150 map:
7151 	ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7152 err:
7153 	bio_put(bio);
7154 	return ret;
7155 }
7156 
7157 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7158 				    int skip_sum)
7159 {
7160 	struct inode *inode = dip->inode;
7161 	struct btrfs_root *root = BTRFS_I(inode)->root;
7162 	struct bio *bio;
7163 	struct bio *orig_bio = dip->orig_bio;
7164 	struct bio_vec *bvec = orig_bio->bi_io_vec;
7165 	u64 start_sector = orig_bio->bi_iter.bi_sector;
7166 	u64 file_offset = dip->logical_offset;
7167 	u64 submit_len = 0;
7168 	u64 map_length;
7169 	int nr_pages = 0;
7170 	int ret = 0;
7171 	int async_submit = 0;
7172 
7173 	map_length = orig_bio->bi_iter.bi_size;
7174 	ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7175 			      &map_length, NULL, 0);
7176 	if (ret) {
7177 		bio_put(orig_bio);
7178 		return -EIO;
7179 	}
7180 
7181 	if (map_length >= orig_bio->bi_iter.bi_size) {
7182 		bio = orig_bio;
7183 		goto submit;
7184 	}
7185 
7186 	/* async crcs make it difficult to collect full stripe writes. */
7187 	if (btrfs_get_alloc_profile(root, 1) &
7188 	    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7189 		async_submit = 0;
7190 	else
7191 		async_submit = 1;
7192 
7193 	bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7194 	if (!bio)
7195 		return -ENOMEM;
7196 	bio->bi_private = dip;
7197 	bio->bi_end_io = btrfs_end_dio_bio;
7198 	atomic_inc(&dip->pending_bios);
7199 
7200 	while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7201 		if (unlikely(map_length < submit_len + bvec->bv_len ||
7202 		    bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7203 				 bvec->bv_offset) < bvec->bv_len)) {
7204 			/*
7205 			 * inc the count before we submit the bio so
7206 			 * we know the end IO handler won't happen before
7207 			 * we inc the count. Otherwise, the dip might get freed
7208 			 * before we're done setting it up
7209 			 */
7210 			atomic_inc(&dip->pending_bios);
7211 			ret = __btrfs_submit_dio_bio(bio, inode, rw,
7212 						     file_offset, skip_sum,
7213 						     async_submit);
7214 			if (ret) {
7215 				bio_put(bio);
7216 				atomic_dec(&dip->pending_bios);
7217 				goto out_err;
7218 			}
7219 
7220 			start_sector += submit_len >> 9;
7221 			file_offset += submit_len;
7222 
7223 			submit_len = 0;
7224 			nr_pages = 0;
7225 
7226 			bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7227 						  start_sector, GFP_NOFS);
7228 			if (!bio)
7229 				goto out_err;
7230 			bio->bi_private = dip;
7231 			bio->bi_end_io = btrfs_end_dio_bio;
7232 
7233 			map_length = orig_bio->bi_iter.bi_size;
7234 			ret = btrfs_map_block(root->fs_info, rw,
7235 					      start_sector << 9,
7236 					      &map_length, NULL, 0);
7237 			if (ret) {
7238 				bio_put(bio);
7239 				goto out_err;
7240 			}
7241 		} else {
7242 			submit_len += bvec->bv_len;
7243 			nr_pages++;
7244 			bvec++;
7245 		}
7246 	}
7247 
7248 submit:
7249 	ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7250 				     async_submit);
7251 	if (!ret)
7252 		return 0;
7253 
7254 	bio_put(bio);
7255 out_err:
7256 	dip->errors = 1;
7257 	/*
7258 	 * before atomic variable goto zero, we must
7259 	 * make sure dip->errors is perceived to be set.
7260 	 */
7261 	smp_mb__before_atomic_dec();
7262 	if (atomic_dec_and_test(&dip->pending_bios))
7263 		bio_io_error(dip->orig_bio);
7264 
7265 	/* bio_end_io() will handle error, so we needn't return it */
7266 	return 0;
7267 }
7268 
7269 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7270 				struct inode *inode, loff_t file_offset)
7271 {
7272 	struct btrfs_root *root = BTRFS_I(inode)->root;
7273 	struct btrfs_dio_private *dip;
7274 	struct bio *io_bio;
7275 	int skip_sum;
7276 	int sum_len;
7277 	int write = rw & REQ_WRITE;
7278 	int ret = 0;
7279 	u16 csum_size;
7280 
7281 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7282 
7283 	io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7284 	if (!io_bio) {
7285 		ret = -ENOMEM;
7286 		goto free_ordered;
7287 	}
7288 
7289 	if (!skip_sum && !write) {
7290 		csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7291 		sum_len = dio_bio->bi_iter.bi_size >>
7292 			inode->i_sb->s_blocksize_bits;
7293 		sum_len *= csum_size;
7294 	} else {
7295 		sum_len = 0;
7296 	}
7297 
7298 	dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7299 	if (!dip) {
7300 		ret = -ENOMEM;
7301 		goto free_io_bio;
7302 	}
7303 
7304 	dip->private = dio_bio->bi_private;
7305 	dip->inode = inode;
7306 	dip->logical_offset = file_offset;
7307 	dip->bytes = dio_bio->bi_iter.bi_size;
7308 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7309 	io_bio->bi_private = dip;
7310 	dip->errors = 0;
7311 	dip->orig_bio = io_bio;
7312 	dip->dio_bio = dio_bio;
7313 	atomic_set(&dip->pending_bios, 0);
7314 
7315 	if (write)
7316 		io_bio->bi_end_io = btrfs_endio_direct_write;
7317 	else
7318 		io_bio->bi_end_io = btrfs_endio_direct_read;
7319 
7320 	ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7321 	if (!ret)
7322 		return;
7323 
7324 free_io_bio:
7325 	bio_put(io_bio);
7326 
7327 free_ordered:
7328 	/*
7329 	 * If this is a write, we need to clean up the reserved space and kill
7330 	 * the ordered extent.
7331 	 */
7332 	if (write) {
7333 		struct btrfs_ordered_extent *ordered;
7334 		ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7335 		if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7336 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7337 			btrfs_free_reserved_extent(root, ordered->start,
7338 						   ordered->disk_len);
7339 		btrfs_put_ordered_extent(ordered);
7340 		btrfs_put_ordered_extent(ordered);
7341 	}
7342 	bio_endio(dio_bio, ret);
7343 }
7344 
7345 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7346 			const struct iovec *iov, loff_t offset,
7347 			unsigned long nr_segs)
7348 {
7349 	int seg;
7350 	int i;
7351 	size_t size;
7352 	unsigned long addr;
7353 	unsigned blocksize_mask = root->sectorsize - 1;
7354 	ssize_t retval = -EINVAL;
7355 	loff_t end = offset;
7356 
7357 	if (offset & blocksize_mask)
7358 		goto out;
7359 
7360 	/* Check the memory alignment.  Blocks cannot straddle pages */
7361 	for (seg = 0; seg < nr_segs; seg++) {
7362 		addr = (unsigned long)iov[seg].iov_base;
7363 		size = iov[seg].iov_len;
7364 		end += size;
7365 		if ((addr & blocksize_mask) || (size & blocksize_mask))
7366 			goto out;
7367 
7368 		/* If this is a write we don't need to check anymore */
7369 		if (rw & WRITE)
7370 			continue;
7371 
7372 		/*
7373 		 * Check to make sure we don't have duplicate iov_base's in this
7374 		 * iovec, if so return EINVAL, otherwise we'll get csum errors
7375 		 * when reading back.
7376 		 */
7377 		for (i = seg + 1; i < nr_segs; i++) {
7378 			if (iov[seg].iov_base == iov[i].iov_base)
7379 				goto out;
7380 		}
7381 	}
7382 	retval = 0;
7383 out:
7384 	return retval;
7385 }
7386 
7387 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7388 			const struct iovec *iov, loff_t offset,
7389 			unsigned long nr_segs)
7390 {
7391 	struct file *file = iocb->ki_filp;
7392 	struct inode *inode = file->f_mapping->host;
7393 	size_t count = 0;
7394 	int flags = 0;
7395 	bool wakeup = true;
7396 	bool relock = false;
7397 	ssize_t ret;
7398 
7399 	if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7400 			    offset, nr_segs))
7401 		return 0;
7402 
7403 	atomic_inc(&inode->i_dio_count);
7404 	smp_mb__after_atomic_inc();
7405 
7406 	/*
7407 	 * The generic stuff only does filemap_write_and_wait_range, which isn't
7408 	 * enough if we've written compressed pages to this area, so we need to
7409 	 * call btrfs_wait_ordered_range to make absolutely sure that any
7410 	 * outstanding dirty pages are on disk.
7411 	 */
7412 	count = iov_length(iov, nr_segs);
7413 	ret = btrfs_wait_ordered_range(inode, offset, count);
7414 	if (ret)
7415 		return ret;
7416 
7417 	if (rw & WRITE) {
7418 		/*
7419 		 * If the write DIO is beyond the EOF, we need update
7420 		 * the isize, but it is protected by i_mutex. So we can
7421 		 * not unlock the i_mutex at this case.
7422 		 */
7423 		if (offset + count <= inode->i_size) {
7424 			mutex_unlock(&inode->i_mutex);
7425 			relock = true;
7426 		}
7427 		ret = btrfs_delalloc_reserve_space(inode, count);
7428 		if (ret)
7429 			goto out;
7430 	} else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7431 				     &BTRFS_I(inode)->runtime_flags))) {
7432 		inode_dio_done(inode);
7433 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
7434 		wakeup = false;
7435 	}
7436 
7437 	ret = __blockdev_direct_IO(rw, iocb, inode,
7438 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7439 			iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7440 			btrfs_submit_direct, flags);
7441 	if (rw & WRITE) {
7442 		if (ret < 0 && ret != -EIOCBQUEUED)
7443 			btrfs_delalloc_release_space(inode, count);
7444 		else if (ret >= 0 && (size_t)ret < count)
7445 			btrfs_delalloc_release_space(inode,
7446 						     count - (size_t)ret);
7447 		else
7448 			btrfs_delalloc_release_metadata(inode, 0);
7449 	}
7450 out:
7451 	if (wakeup)
7452 		inode_dio_done(inode);
7453 	if (relock)
7454 		mutex_lock(&inode->i_mutex);
7455 
7456 	return ret;
7457 }
7458 
7459 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
7460 
7461 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7462 		__u64 start, __u64 len)
7463 {
7464 	int	ret;
7465 
7466 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7467 	if (ret)
7468 		return ret;
7469 
7470 	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7471 }
7472 
7473 int btrfs_readpage(struct file *file, struct page *page)
7474 {
7475 	struct extent_io_tree *tree;
7476 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7477 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7478 }
7479 
7480 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7481 {
7482 	struct extent_io_tree *tree;
7483 
7484 
7485 	if (current->flags & PF_MEMALLOC) {
7486 		redirty_page_for_writepage(wbc, page);
7487 		unlock_page(page);
7488 		return 0;
7489 	}
7490 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7491 	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7492 }
7493 
7494 static int btrfs_writepages(struct address_space *mapping,
7495 			    struct writeback_control *wbc)
7496 {
7497 	struct extent_io_tree *tree;
7498 
7499 	tree = &BTRFS_I(mapping->host)->io_tree;
7500 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7501 }
7502 
7503 static int
7504 btrfs_readpages(struct file *file, struct address_space *mapping,
7505 		struct list_head *pages, unsigned nr_pages)
7506 {
7507 	struct extent_io_tree *tree;
7508 	tree = &BTRFS_I(mapping->host)->io_tree;
7509 	return extent_readpages(tree, mapping, pages, nr_pages,
7510 				btrfs_get_extent);
7511 }
7512 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7513 {
7514 	struct extent_io_tree *tree;
7515 	struct extent_map_tree *map;
7516 	int ret;
7517 
7518 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7519 	map = &BTRFS_I(page->mapping->host)->extent_tree;
7520 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7521 	if (ret == 1) {
7522 		ClearPagePrivate(page);
7523 		set_page_private(page, 0);
7524 		page_cache_release(page);
7525 	}
7526 	return ret;
7527 }
7528 
7529 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7530 {
7531 	if (PageWriteback(page) || PageDirty(page))
7532 		return 0;
7533 	return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7534 }
7535 
7536 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7537 				 unsigned int length)
7538 {
7539 	struct inode *inode = page->mapping->host;
7540 	struct extent_io_tree *tree;
7541 	struct btrfs_ordered_extent *ordered;
7542 	struct extent_state *cached_state = NULL;
7543 	u64 page_start = page_offset(page);
7544 	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7545 	int inode_evicting = inode->i_state & I_FREEING;
7546 
7547 	/*
7548 	 * we have the page locked, so new writeback can't start,
7549 	 * and the dirty bit won't be cleared while we are here.
7550 	 *
7551 	 * Wait for IO on this page so that we can safely clear
7552 	 * the PagePrivate2 bit and do ordered accounting
7553 	 */
7554 	wait_on_page_writeback(page);
7555 
7556 	tree = &BTRFS_I(inode)->io_tree;
7557 	if (offset) {
7558 		btrfs_releasepage(page, GFP_NOFS);
7559 		return;
7560 	}
7561 
7562 	if (!inode_evicting)
7563 		lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7564 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7565 	if (ordered) {
7566 		/*
7567 		 * IO on this page will never be started, so we need
7568 		 * to account for any ordered extents now
7569 		 */
7570 		if (!inode_evicting)
7571 			clear_extent_bit(tree, page_start, page_end,
7572 					 EXTENT_DIRTY | EXTENT_DELALLOC |
7573 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7574 					 EXTENT_DEFRAG, 1, 0, &cached_state,
7575 					 GFP_NOFS);
7576 		/*
7577 		 * whoever cleared the private bit is responsible
7578 		 * for the finish_ordered_io
7579 		 */
7580 		if (TestClearPagePrivate2(page)) {
7581 			struct btrfs_ordered_inode_tree *tree;
7582 			u64 new_len;
7583 
7584 			tree = &BTRFS_I(inode)->ordered_tree;
7585 
7586 			spin_lock_irq(&tree->lock);
7587 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7588 			new_len = page_start - ordered->file_offset;
7589 			if (new_len < ordered->truncated_len)
7590 				ordered->truncated_len = new_len;
7591 			spin_unlock_irq(&tree->lock);
7592 
7593 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
7594 							   page_start,
7595 							   PAGE_CACHE_SIZE, 1))
7596 				btrfs_finish_ordered_io(ordered);
7597 		}
7598 		btrfs_put_ordered_extent(ordered);
7599 		if (!inode_evicting) {
7600 			cached_state = NULL;
7601 			lock_extent_bits(tree, page_start, page_end, 0,
7602 					 &cached_state);
7603 		}
7604 	}
7605 
7606 	if (!inode_evicting) {
7607 		clear_extent_bit(tree, page_start, page_end,
7608 				 EXTENT_LOCKED | EXTENT_DIRTY |
7609 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7610 				 EXTENT_DEFRAG, 1, 1,
7611 				 &cached_state, GFP_NOFS);
7612 
7613 		__btrfs_releasepage(page, GFP_NOFS);
7614 	}
7615 
7616 	ClearPageChecked(page);
7617 	if (PagePrivate(page)) {
7618 		ClearPagePrivate(page);
7619 		set_page_private(page, 0);
7620 		page_cache_release(page);
7621 	}
7622 }
7623 
7624 /*
7625  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7626  * called from a page fault handler when a page is first dirtied. Hence we must
7627  * be careful to check for EOF conditions here. We set the page up correctly
7628  * for a written page which means we get ENOSPC checking when writing into
7629  * holes and correct delalloc and unwritten extent mapping on filesystems that
7630  * support these features.
7631  *
7632  * We are not allowed to take the i_mutex here so we have to play games to
7633  * protect against truncate races as the page could now be beyond EOF.  Because
7634  * vmtruncate() writes the inode size before removing pages, once we have the
7635  * page lock we can determine safely if the page is beyond EOF. If it is not
7636  * beyond EOF, then the page is guaranteed safe against truncation until we
7637  * unlock the page.
7638  */
7639 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7640 {
7641 	struct page *page = vmf->page;
7642 	struct inode *inode = file_inode(vma->vm_file);
7643 	struct btrfs_root *root = BTRFS_I(inode)->root;
7644 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7645 	struct btrfs_ordered_extent *ordered;
7646 	struct extent_state *cached_state = NULL;
7647 	char *kaddr;
7648 	unsigned long zero_start;
7649 	loff_t size;
7650 	int ret;
7651 	int reserved = 0;
7652 	u64 page_start;
7653 	u64 page_end;
7654 
7655 	sb_start_pagefault(inode->i_sb);
7656 	ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7657 	if (!ret) {
7658 		ret = file_update_time(vma->vm_file);
7659 		reserved = 1;
7660 	}
7661 	if (ret) {
7662 		if (ret == -ENOMEM)
7663 			ret = VM_FAULT_OOM;
7664 		else /* -ENOSPC, -EIO, etc */
7665 			ret = VM_FAULT_SIGBUS;
7666 		if (reserved)
7667 			goto out;
7668 		goto out_noreserve;
7669 	}
7670 
7671 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7672 again:
7673 	lock_page(page);
7674 	size = i_size_read(inode);
7675 	page_start = page_offset(page);
7676 	page_end = page_start + PAGE_CACHE_SIZE - 1;
7677 
7678 	if ((page->mapping != inode->i_mapping) ||
7679 	    (page_start >= size)) {
7680 		/* page got truncated out from underneath us */
7681 		goto out_unlock;
7682 	}
7683 	wait_on_page_writeback(page);
7684 
7685 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7686 	set_page_extent_mapped(page);
7687 
7688 	/*
7689 	 * we can't set the delalloc bits if there are pending ordered
7690 	 * extents.  Drop our locks and wait for them to finish
7691 	 */
7692 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7693 	if (ordered) {
7694 		unlock_extent_cached(io_tree, page_start, page_end,
7695 				     &cached_state, GFP_NOFS);
7696 		unlock_page(page);
7697 		btrfs_start_ordered_extent(inode, ordered, 1);
7698 		btrfs_put_ordered_extent(ordered);
7699 		goto again;
7700 	}
7701 
7702 	/*
7703 	 * XXX - page_mkwrite gets called every time the page is dirtied, even
7704 	 * if it was already dirty, so for space accounting reasons we need to
7705 	 * clear any delalloc bits for the range we are fixing to save.  There
7706 	 * is probably a better way to do this, but for now keep consistent with
7707 	 * prepare_pages in the normal write path.
7708 	 */
7709 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7710 			  EXTENT_DIRTY | EXTENT_DELALLOC |
7711 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7712 			  0, 0, &cached_state, GFP_NOFS);
7713 
7714 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7715 					&cached_state);
7716 	if (ret) {
7717 		unlock_extent_cached(io_tree, page_start, page_end,
7718 				     &cached_state, GFP_NOFS);
7719 		ret = VM_FAULT_SIGBUS;
7720 		goto out_unlock;
7721 	}
7722 	ret = 0;
7723 
7724 	/* page is wholly or partially inside EOF */
7725 	if (page_start + PAGE_CACHE_SIZE > size)
7726 		zero_start = size & ~PAGE_CACHE_MASK;
7727 	else
7728 		zero_start = PAGE_CACHE_SIZE;
7729 
7730 	if (zero_start != PAGE_CACHE_SIZE) {
7731 		kaddr = kmap(page);
7732 		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7733 		flush_dcache_page(page);
7734 		kunmap(page);
7735 	}
7736 	ClearPageChecked(page);
7737 	set_page_dirty(page);
7738 	SetPageUptodate(page);
7739 
7740 	BTRFS_I(inode)->last_trans = root->fs_info->generation;
7741 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7742 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7743 
7744 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7745 
7746 out_unlock:
7747 	if (!ret) {
7748 		sb_end_pagefault(inode->i_sb);
7749 		return VM_FAULT_LOCKED;
7750 	}
7751 	unlock_page(page);
7752 out:
7753 	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7754 out_noreserve:
7755 	sb_end_pagefault(inode->i_sb);
7756 	return ret;
7757 }
7758 
7759 static int btrfs_truncate(struct inode *inode)
7760 {
7761 	struct btrfs_root *root = BTRFS_I(inode)->root;
7762 	struct btrfs_block_rsv *rsv;
7763 	int ret = 0;
7764 	int err = 0;
7765 	struct btrfs_trans_handle *trans;
7766 	u64 mask = root->sectorsize - 1;
7767 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7768 
7769 	ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7770 				       (u64)-1);
7771 	if (ret)
7772 		return ret;
7773 
7774 	/*
7775 	 * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7776 	 * 3 things going on here
7777 	 *
7778 	 * 1) We need to reserve space for our orphan item and the space to
7779 	 * delete our orphan item.  Lord knows we don't want to have a dangling
7780 	 * orphan item because we didn't reserve space to remove it.
7781 	 *
7782 	 * 2) We need to reserve space to update our inode.
7783 	 *
7784 	 * 3) We need to have something to cache all the space that is going to
7785 	 * be free'd up by the truncate operation, but also have some slack
7786 	 * space reserved in case it uses space during the truncate (thank you
7787 	 * very much snapshotting).
7788 	 *
7789 	 * And we need these to all be seperate.  The fact is we can use alot of
7790 	 * space doing the truncate, and we have no earthly idea how much space
7791 	 * we will use, so we need the truncate reservation to be seperate so it
7792 	 * doesn't end up using space reserved for updating the inode or
7793 	 * removing the orphan item.  We also need to be able to stop the
7794 	 * transaction and start a new one, which means we need to be able to
7795 	 * update the inode several times, and we have no idea of knowing how
7796 	 * many times that will be, so we can't just reserve 1 item for the
7797 	 * entirety of the opration, so that has to be done seperately as well.
7798 	 * Then there is the orphan item, which does indeed need to be held on
7799 	 * to for the whole operation, and we need nobody to touch this reserved
7800 	 * space except the orphan code.
7801 	 *
7802 	 * So that leaves us with
7803 	 *
7804 	 * 1) root->orphan_block_rsv - for the orphan deletion.
7805 	 * 2) rsv - for the truncate reservation, which we will steal from the
7806 	 * transaction reservation.
7807 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7808 	 * updating the inode.
7809 	 */
7810 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7811 	if (!rsv)
7812 		return -ENOMEM;
7813 	rsv->size = min_size;
7814 	rsv->failfast = 1;
7815 
7816 	/*
7817 	 * 1 for the truncate slack space
7818 	 * 1 for updating the inode.
7819 	 */
7820 	trans = btrfs_start_transaction(root, 2);
7821 	if (IS_ERR(trans)) {
7822 		err = PTR_ERR(trans);
7823 		goto out;
7824 	}
7825 
7826 	/* Migrate the slack space for the truncate to our reserve */
7827 	ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7828 				      min_size);
7829 	BUG_ON(ret);
7830 
7831 	/*
7832 	 * setattr is responsible for setting the ordered_data_close flag,
7833 	 * but that is only tested during the last file release.  That
7834 	 * could happen well after the next commit, leaving a great big
7835 	 * window where new writes may get lost if someone chooses to write
7836 	 * to this file after truncating to zero
7837 	 *
7838 	 * The inode doesn't have any dirty data here, and so if we commit
7839 	 * this is a noop.  If someone immediately starts writing to the inode
7840 	 * it is very likely we'll catch some of their writes in this
7841 	 * transaction, and the commit will find this file on the ordered
7842 	 * data list with good things to send down.
7843 	 *
7844 	 * This is a best effort solution, there is still a window where
7845 	 * using truncate to replace the contents of the file will
7846 	 * end up with a zero length file after a crash.
7847 	 */
7848 	if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7849 					   &BTRFS_I(inode)->runtime_flags))
7850 		btrfs_add_ordered_operation(trans, root, inode);
7851 
7852 	/*
7853 	 * So if we truncate and then write and fsync we normally would just
7854 	 * write the extents that changed, which is a problem if we need to
7855 	 * first truncate that entire inode.  So set this flag so we write out
7856 	 * all of the extents in the inode to the sync log so we're completely
7857 	 * safe.
7858 	 */
7859 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7860 	trans->block_rsv = rsv;
7861 
7862 	while (1) {
7863 		ret = btrfs_truncate_inode_items(trans, root, inode,
7864 						 inode->i_size,
7865 						 BTRFS_EXTENT_DATA_KEY);
7866 		if (ret != -ENOSPC) {
7867 			err = ret;
7868 			break;
7869 		}
7870 
7871 		trans->block_rsv = &root->fs_info->trans_block_rsv;
7872 		ret = btrfs_update_inode(trans, root, inode);
7873 		if (ret) {
7874 			err = ret;
7875 			break;
7876 		}
7877 
7878 		btrfs_end_transaction(trans, root);
7879 		btrfs_btree_balance_dirty(root);
7880 
7881 		trans = btrfs_start_transaction(root, 2);
7882 		if (IS_ERR(trans)) {
7883 			ret = err = PTR_ERR(trans);
7884 			trans = NULL;
7885 			break;
7886 		}
7887 
7888 		ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7889 					      rsv, min_size);
7890 		BUG_ON(ret);	/* shouldn't happen */
7891 		trans->block_rsv = rsv;
7892 	}
7893 
7894 	if (ret == 0 && inode->i_nlink > 0) {
7895 		trans->block_rsv = root->orphan_block_rsv;
7896 		ret = btrfs_orphan_del(trans, inode);
7897 		if (ret)
7898 			err = ret;
7899 	}
7900 
7901 	if (trans) {
7902 		trans->block_rsv = &root->fs_info->trans_block_rsv;
7903 		ret = btrfs_update_inode(trans, root, inode);
7904 		if (ret && !err)
7905 			err = ret;
7906 
7907 		ret = btrfs_end_transaction(trans, root);
7908 		btrfs_btree_balance_dirty(root);
7909 	}
7910 
7911 out:
7912 	btrfs_free_block_rsv(root, rsv);
7913 
7914 	if (ret && !err)
7915 		err = ret;
7916 
7917 	return err;
7918 }
7919 
7920 /*
7921  * create a new subvolume directory/inode (helper for the ioctl).
7922  */
7923 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7924 			     struct btrfs_root *new_root,
7925 			     struct btrfs_root *parent_root,
7926 			     u64 new_dirid)
7927 {
7928 	struct inode *inode;
7929 	int err;
7930 	u64 index = 0;
7931 
7932 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7933 				new_dirid, new_dirid,
7934 				S_IFDIR | (~current_umask() & S_IRWXUGO),
7935 				&index);
7936 	if (IS_ERR(inode))
7937 		return PTR_ERR(inode);
7938 	inode->i_op = &btrfs_dir_inode_operations;
7939 	inode->i_fop = &btrfs_dir_file_operations;
7940 
7941 	set_nlink(inode, 1);
7942 	btrfs_i_size_write(inode, 0);
7943 
7944 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7945 	if (err)
7946 		btrfs_err(new_root->fs_info,
7947 			  "error inheriting subvolume %llu properties: %d\n",
7948 			  new_root->root_key.objectid, err);
7949 
7950 	err = btrfs_update_inode(trans, new_root, inode);
7951 
7952 	iput(inode);
7953 	return err;
7954 }
7955 
7956 struct inode *btrfs_alloc_inode(struct super_block *sb)
7957 {
7958 	struct btrfs_inode *ei;
7959 	struct inode *inode;
7960 
7961 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7962 	if (!ei)
7963 		return NULL;
7964 
7965 	ei->root = NULL;
7966 	ei->generation = 0;
7967 	ei->last_trans = 0;
7968 	ei->last_sub_trans = 0;
7969 	ei->logged_trans = 0;
7970 	ei->delalloc_bytes = 0;
7971 	ei->disk_i_size = 0;
7972 	ei->flags = 0;
7973 	ei->csum_bytes = 0;
7974 	ei->index_cnt = (u64)-1;
7975 	ei->dir_index = 0;
7976 	ei->last_unlink_trans = 0;
7977 	ei->last_log_commit = 0;
7978 
7979 	spin_lock_init(&ei->lock);
7980 	ei->outstanding_extents = 0;
7981 	ei->reserved_extents = 0;
7982 
7983 	ei->runtime_flags = 0;
7984 	ei->force_compress = BTRFS_COMPRESS_NONE;
7985 
7986 	ei->delayed_node = NULL;
7987 
7988 	inode = &ei->vfs_inode;
7989 	extent_map_tree_init(&ei->extent_tree);
7990 	extent_io_tree_init(&ei->io_tree, &inode->i_data);
7991 	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7992 	ei->io_tree.track_uptodate = 1;
7993 	ei->io_failure_tree.track_uptodate = 1;
7994 	atomic_set(&ei->sync_writers, 0);
7995 	mutex_init(&ei->log_mutex);
7996 	mutex_init(&ei->delalloc_mutex);
7997 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7998 	INIT_LIST_HEAD(&ei->delalloc_inodes);
7999 	INIT_LIST_HEAD(&ei->ordered_operations);
8000 	RB_CLEAR_NODE(&ei->rb_node);
8001 
8002 	return inode;
8003 }
8004 
8005 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8006 void btrfs_test_destroy_inode(struct inode *inode)
8007 {
8008 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8009 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8010 }
8011 #endif
8012 
8013 static void btrfs_i_callback(struct rcu_head *head)
8014 {
8015 	struct inode *inode = container_of(head, struct inode, i_rcu);
8016 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8017 }
8018 
8019 void btrfs_destroy_inode(struct inode *inode)
8020 {
8021 	struct btrfs_ordered_extent *ordered;
8022 	struct btrfs_root *root = BTRFS_I(inode)->root;
8023 
8024 	WARN_ON(!hlist_empty(&inode->i_dentry));
8025 	WARN_ON(inode->i_data.nrpages);
8026 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
8027 	WARN_ON(BTRFS_I(inode)->reserved_extents);
8028 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8029 	WARN_ON(BTRFS_I(inode)->csum_bytes);
8030 
8031 	/*
8032 	 * This can happen where we create an inode, but somebody else also
8033 	 * created the same inode and we need to destroy the one we already
8034 	 * created.
8035 	 */
8036 	if (!root)
8037 		goto free;
8038 
8039 	/*
8040 	 * Make sure we're properly removed from the ordered operation
8041 	 * lists.
8042 	 */
8043 	smp_mb();
8044 	if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8045 		spin_lock(&root->fs_info->ordered_root_lock);
8046 		list_del_init(&BTRFS_I(inode)->ordered_operations);
8047 		spin_unlock(&root->fs_info->ordered_root_lock);
8048 	}
8049 
8050 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8051 		     &BTRFS_I(inode)->runtime_flags)) {
8052 		btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8053 			btrfs_ino(inode));
8054 		atomic_dec(&root->orphan_inodes);
8055 	}
8056 
8057 	while (1) {
8058 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8059 		if (!ordered)
8060 			break;
8061 		else {
8062 			btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8063 				ordered->file_offset, ordered->len);
8064 			btrfs_remove_ordered_extent(inode, ordered);
8065 			btrfs_put_ordered_extent(ordered);
8066 			btrfs_put_ordered_extent(ordered);
8067 		}
8068 	}
8069 	inode_tree_del(inode);
8070 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8071 free:
8072 	call_rcu(&inode->i_rcu, btrfs_i_callback);
8073 }
8074 
8075 int btrfs_drop_inode(struct inode *inode)
8076 {
8077 	struct btrfs_root *root = BTRFS_I(inode)->root;
8078 
8079 	if (root == NULL)
8080 		return 1;
8081 
8082 	/* the snap/subvol tree is on deleting */
8083 	if (btrfs_root_refs(&root->root_item) == 0)
8084 		return 1;
8085 	else
8086 		return generic_drop_inode(inode);
8087 }
8088 
8089 static void init_once(void *foo)
8090 {
8091 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8092 
8093 	inode_init_once(&ei->vfs_inode);
8094 }
8095 
8096 void btrfs_destroy_cachep(void)
8097 {
8098 	/*
8099 	 * Make sure all delayed rcu free inodes are flushed before we
8100 	 * destroy cache.
8101 	 */
8102 	rcu_barrier();
8103 	if (btrfs_inode_cachep)
8104 		kmem_cache_destroy(btrfs_inode_cachep);
8105 	if (btrfs_trans_handle_cachep)
8106 		kmem_cache_destroy(btrfs_trans_handle_cachep);
8107 	if (btrfs_transaction_cachep)
8108 		kmem_cache_destroy(btrfs_transaction_cachep);
8109 	if (btrfs_path_cachep)
8110 		kmem_cache_destroy(btrfs_path_cachep);
8111 	if (btrfs_free_space_cachep)
8112 		kmem_cache_destroy(btrfs_free_space_cachep);
8113 	if (btrfs_delalloc_work_cachep)
8114 		kmem_cache_destroy(btrfs_delalloc_work_cachep);
8115 }
8116 
8117 int btrfs_init_cachep(void)
8118 {
8119 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8120 			sizeof(struct btrfs_inode), 0,
8121 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8122 	if (!btrfs_inode_cachep)
8123 		goto fail;
8124 
8125 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8126 			sizeof(struct btrfs_trans_handle), 0,
8127 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8128 	if (!btrfs_trans_handle_cachep)
8129 		goto fail;
8130 
8131 	btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8132 			sizeof(struct btrfs_transaction), 0,
8133 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8134 	if (!btrfs_transaction_cachep)
8135 		goto fail;
8136 
8137 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
8138 			sizeof(struct btrfs_path), 0,
8139 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8140 	if (!btrfs_path_cachep)
8141 		goto fail;
8142 
8143 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8144 			sizeof(struct btrfs_free_space), 0,
8145 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8146 	if (!btrfs_free_space_cachep)
8147 		goto fail;
8148 
8149 	btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8150 			sizeof(struct btrfs_delalloc_work), 0,
8151 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8152 			NULL);
8153 	if (!btrfs_delalloc_work_cachep)
8154 		goto fail;
8155 
8156 	return 0;
8157 fail:
8158 	btrfs_destroy_cachep();
8159 	return -ENOMEM;
8160 }
8161 
8162 static int btrfs_getattr(struct vfsmount *mnt,
8163 			 struct dentry *dentry, struct kstat *stat)
8164 {
8165 	u64 delalloc_bytes;
8166 	struct inode *inode = dentry->d_inode;
8167 	u32 blocksize = inode->i_sb->s_blocksize;
8168 
8169 	generic_fillattr(inode, stat);
8170 	stat->dev = BTRFS_I(inode)->root->anon_dev;
8171 	stat->blksize = PAGE_CACHE_SIZE;
8172 
8173 	spin_lock(&BTRFS_I(inode)->lock);
8174 	delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8175 	spin_unlock(&BTRFS_I(inode)->lock);
8176 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8177 			ALIGN(delalloc_bytes, blocksize)) >> 9;
8178 	return 0;
8179 }
8180 
8181 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8182 			   struct inode *new_dir, struct dentry *new_dentry)
8183 {
8184 	struct btrfs_trans_handle *trans;
8185 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
8186 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8187 	struct inode *new_inode = new_dentry->d_inode;
8188 	struct inode *old_inode = old_dentry->d_inode;
8189 	struct timespec ctime = CURRENT_TIME;
8190 	u64 index = 0;
8191 	u64 root_objectid;
8192 	int ret;
8193 	u64 old_ino = btrfs_ino(old_inode);
8194 
8195 	if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8196 		return -EPERM;
8197 
8198 	/* we only allow rename subvolume link between subvolumes */
8199 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8200 		return -EXDEV;
8201 
8202 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8203 	    (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8204 		return -ENOTEMPTY;
8205 
8206 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
8207 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8208 		return -ENOTEMPTY;
8209 
8210 
8211 	/* check for collisions, even if the  name isn't there */
8212 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8213 			     new_dentry->d_name.name,
8214 			     new_dentry->d_name.len);
8215 
8216 	if (ret) {
8217 		if (ret == -EEXIST) {
8218 			/* we shouldn't get
8219 			 * eexist without a new_inode */
8220 			if (WARN_ON(!new_inode)) {
8221 				return ret;
8222 			}
8223 		} else {
8224 			/* maybe -EOVERFLOW */
8225 			return ret;
8226 		}
8227 	}
8228 	ret = 0;
8229 
8230 	/*
8231 	 * we're using rename to replace one file with another.
8232 	 * and the replacement file is large.  Start IO on it now so
8233 	 * we don't add too much work to the end of the transaction
8234 	 */
8235 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8236 	    old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8237 		filemap_flush(old_inode->i_mapping);
8238 
8239 	/* close the racy window with snapshot create/destroy ioctl */
8240 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8241 		down_read(&root->fs_info->subvol_sem);
8242 	/*
8243 	 * We want to reserve the absolute worst case amount of items.  So if
8244 	 * both inodes are subvols and we need to unlink them then that would
8245 	 * require 4 item modifications, but if they are both normal inodes it
8246 	 * would require 5 item modifications, so we'll assume their normal
8247 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8248 	 * should cover the worst case number of items we'll modify.
8249 	 */
8250 	trans = btrfs_start_transaction(root, 11);
8251 	if (IS_ERR(trans)) {
8252                 ret = PTR_ERR(trans);
8253                 goto out_notrans;
8254         }
8255 
8256 	if (dest != root)
8257 		btrfs_record_root_in_trans(trans, dest);
8258 
8259 	ret = btrfs_set_inode_index(new_dir, &index);
8260 	if (ret)
8261 		goto out_fail;
8262 
8263 	BTRFS_I(old_inode)->dir_index = 0ULL;
8264 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8265 		/* force full log commit if subvolume involved. */
8266 		root->fs_info->last_trans_log_full_commit = trans->transid;
8267 	} else {
8268 		ret = btrfs_insert_inode_ref(trans, dest,
8269 					     new_dentry->d_name.name,
8270 					     new_dentry->d_name.len,
8271 					     old_ino,
8272 					     btrfs_ino(new_dir), index);
8273 		if (ret)
8274 			goto out_fail;
8275 		/*
8276 		 * this is an ugly little race, but the rename is required
8277 		 * to make sure that if we crash, the inode is either at the
8278 		 * old name or the new one.  pinning the log transaction lets
8279 		 * us make sure we don't allow a log commit to come in after
8280 		 * we unlink the name but before we add the new name back in.
8281 		 */
8282 		btrfs_pin_log_trans(root);
8283 	}
8284 	/*
8285 	 * make sure the inode gets flushed if it is replacing
8286 	 * something.
8287 	 */
8288 	if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8289 		btrfs_add_ordered_operation(trans, root, old_inode);
8290 
8291 	inode_inc_iversion(old_dir);
8292 	inode_inc_iversion(new_dir);
8293 	inode_inc_iversion(old_inode);
8294 	old_dir->i_ctime = old_dir->i_mtime = ctime;
8295 	new_dir->i_ctime = new_dir->i_mtime = ctime;
8296 	old_inode->i_ctime = ctime;
8297 
8298 	if (old_dentry->d_parent != new_dentry->d_parent)
8299 		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8300 
8301 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8302 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8303 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8304 					old_dentry->d_name.name,
8305 					old_dentry->d_name.len);
8306 	} else {
8307 		ret = __btrfs_unlink_inode(trans, root, old_dir,
8308 					old_dentry->d_inode,
8309 					old_dentry->d_name.name,
8310 					old_dentry->d_name.len);
8311 		if (!ret)
8312 			ret = btrfs_update_inode(trans, root, old_inode);
8313 	}
8314 	if (ret) {
8315 		btrfs_abort_transaction(trans, root, ret);
8316 		goto out_fail;
8317 	}
8318 
8319 	if (new_inode) {
8320 		inode_inc_iversion(new_inode);
8321 		new_inode->i_ctime = CURRENT_TIME;
8322 		if (unlikely(btrfs_ino(new_inode) ==
8323 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8324 			root_objectid = BTRFS_I(new_inode)->location.objectid;
8325 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
8326 						root_objectid,
8327 						new_dentry->d_name.name,
8328 						new_dentry->d_name.len);
8329 			BUG_ON(new_inode->i_nlink == 0);
8330 		} else {
8331 			ret = btrfs_unlink_inode(trans, dest, new_dir,
8332 						 new_dentry->d_inode,
8333 						 new_dentry->d_name.name,
8334 						 new_dentry->d_name.len);
8335 		}
8336 		if (!ret && new_inode->i_nlink == 0)
8337 			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8338 		if (ret) {
8339 			btrfs_abort_transaction(trans, root, ret);
8340 			goto out_fail;
8341 		}
8342 	}
8343 
8344 	ret = btrfs_add_link(trans, new_dir, old_inode,
8345 			     new_dentry->d_name.name,
8346 			     new_dentry->d_name.len, 0, index);
8347 	if (ret) {
8348 		btrfs_abort_transaction(trans, root, ret);
8349 		goto out_fail;
8350 	}
8351 
8352 	if (old_inode->i_nlink == 1)
8353 		BTRFS_I(old_inode)->dir_index = index;
8354 
8355 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8356 		struct dentry *parent = new_dentry->d_parent;
8357 		btrfs_log_new_name(trans, old_inode, old_dir, parent);
8358 		btrfs_end_log_trans(root);
8359 	}
8360 out_fail:
8361 	btrfs_end_transaction(trans, root);
8362 out_notrans:
8363 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8364 		up_read(&root->fs_info->subvol_sem);
8365 
8366 	return ret;
8367 }
8368 
8369 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8370 {
8371 	struct btrfs_delalloc_work *delalloc_work;
8372 	struct inode *inode;
8373 
8374 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
8375 				     work);
8376 	inode = delalloc_work->inode;
8377 	if (delalloc_work->wait) {
8378 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
8379 	} else {
8380 		filemap_flush(inode->i_mapping);
8381 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8382 			     &BTRFS_I(inode)->runtime_flags))
8383 			filemap_flush(inode->i_mapping);
8384 	}
8385 
8386 	if (delalloc_work->delay_iput)
8387 		btrfs_add_delayed_iput(inode);
8388 	else
8389 		iput(inode);
8390 	complete(&delalloc_work->completion);
8391 }
8392 
8393 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8394 						    int wait, int delay_iput)
8395 {
8396 	struct btrfs_delalloc_work *work;
8397 
8398 	work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8399 	if (!work)
8400 		return NULL;
8401 
8402 	init_completion(&work->completion);
8403 	INIT_LIST_HEAD(&work->list);
8404 	work->inode = inode;
8405 	work->wait = wait;
8406 	work->delay_iput = delay_iput;
8407 	work->work.func = btrfs_run_delalloc_work;
8408 
8409 	return work;
8410 }
8411 
8412 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8413 {
8414 	wait_for_completion(&work->completion);
8415 	kmem_cache_free(btrfs_delalloc_work_cachep, work);
8416 }
8417 
8418 /*
8419  * some fairly slow code that needs optimization. This walks the list
8420  * of all the inodes with pending delalloc and forces them to disk.
8421  */
8422 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8423 {
8424 	struct btrfs_inode *binode;
8425 	struct inode *inode;
8426 	struct btrfs_delalloc_work *work, *next;
8427 	struct list_head works;
8428 	struct list_head splice;
8429 	int ret = 0;
8430 
8431 	INIT_LIST_HEAD(&works);
8432 	INIT_LIST_HEAD(&splice);
8433 
8434 	spin_lock(&root->delalloc_lock);
8435 	list_splice_init(&root->delalloc_inodes, &splice);
8436 	while (!list_empty(&splice)) {
8437 		binode = list_entry(splice.next, struct btrfs_inode,
8438 				    delalloc_inodes);
8439 
8440 		list_move_tail(&binode->delalloc_inodes,
8441 			       &root->delalloc_inodes);
8442 		inode = igrab(&binode->vfs_inode);
8443 		if (!inode) {
8444 			cond_resched_lock(&root->delalloc_lock);
8445 			continue;
8446 		}
8447 		spin_unlock(&root->delalloc_lock);
8448 
8449 		work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8450 		if (unlikely(!work)) {
8451 			if (delay_iput)
8452 				btrfs_add_delayed_iput(inode);
8453 			else
8454 				iput(inode);
8455 			ret = -ENOMEM;
8456 			goto out;
8457 		}
8458 		list_add_tail(&work->list, &works);
8459 		btrfs_queue_worker(&root->fs_info->flush_workers,
8460 				   &work->work);
8461 
8462 		cond_resched();
8463 		spin_lock(&root->delalloc_lock);
8464 	}
8465 	spin_unlock(&root->delalloc_lock);
8466 
8467 	list_for_each_entry_safe(work, next, &works, list) {
8468 		list_del_init(&work->list);
8469 		btrfs_wait_and_free_delalloc_work(work);
8470 	}
8471 	return 0;
8472 out:
8473 	list_for_each_entry_safe(work, next, &works, list) {
8474 		list_del_init(&work->list);
8475 		btrfs_wait_and_free_delalloc_work(work);
8476 	}
8477 
8478 	if (!list_empty_careful(&splice)) {
8479 		spin_lock(&root->delalloc_lock);
8480 		list_splice_tail(&splice, &root->delalloc_inodes);
8481 		spin_unlock(&root->delalloc_lock);
8482 	}
8483 	return ret;
8484 }
8485 
8486 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8487 {
8488 	int ret;
8489 
8490 	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8491 		return -EROFS;
8492 
8493 	ret = __start_delalloc_inodes(root, delay_iput);
8494 	/*
8495 	 * the filemap_flush will queue IO into the worker threads, but
8496 	 * we have to make sure the IO is actually started and that
8497 	 * ordered extents get created before we return
8498 	 */
8499 	atomic_inc(&root->fs_info->async_submit_draining);
8500 	while (atomic_read(&root->fs_info->nr_async_submits) ||
8501 	      atomic_read(&root->fs_info->async_delalloc_pages)) {
8502 		wait_event(root->fs_info->async_submit_wait,
8503 		   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8504 		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8505 	}
8506 	atomic_dec(&root->fs_info->async_submit_draining);
8507 	return ret;
8508 }
8509 
8510 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8511 {
8512 	struct btrfs_root *root;
8513 	struct list_head splice;
8514 	int ret;
8515 
8516 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8517 		return -EROFS;
8518 
8519 	INIT_LIST_HEAD(&splice);
8520 
8521 	spin_lock(&fs_info->delalloc_root_lock);
8522 	list_splice_init(&fs_info->delalloc_roots, &splice);
8523 	while (!list_empty(&splice)) {
8524 		root = list_first_entry(&splice, struct btrfs_root,
8525 					delalloc_root);
8526 		root = btrfs_grab_fs_root(root);
8527 		BUG_ON(!root);
8528 		list_move_tail(&root->delalloc_root,
8529 			       &fs_info->delalloc_roots);
8530 		spin_unlock(&fs_info->delalloc_root_lock);
8531 
8532 		ret = __start_delalloc_inodes(root, delay_iput);
8533 		btrfs_put_fs_root(root);
8534 		if (ret)
8535 			goto out;
8536 
8537 		spin_lock(&fs_info->delalloc_root_lock);
8538 	}
8539 	spin_unlock(&fs_info->delalloc_root_lock);
8540 
8541 	atomic_inc(&fs_info->async_submit_draining);
8542 	while (atomic_read(&fs_info->nr_async_submits) ||
8543 	      atomic_read(&fs_info->async_delalloc_pages)) {
8544 		wait_event(fs_info->async_submit_wait,
8545 		   (atomic_read(&fs_info->nr_async_submits) == 0 &&
8546 		    atomic_read(&fs_info->async_delalloc_pages) == 0));
8547 	}
8548 	atomic_dec(&fs_info->async_submit_draining);
8549 	return 0;
8550 out:
8551 	if (!list_empty_careful(&splice)) {
8552 		spin_lock(&fs_info->delalloc_root_lock);
8553 		list_splice_tail(&splice, &fs_info->delalloc_roots);
8554 		spin_unlock(&fs_info->delalloc_root_lock);
8555 	}
8556 	return ret;
8557 }
8558 
8559 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8560 			 const char *symname)
8561 {
8562 	struct btrfs_trans_handle *trans;
8563 	struct btrfs_root *root = BTRFS_I(dir)->root;
8564 	struct btrfs_path *path;
8565 	struct btrfs_key key;
8566 	struct inode *inode = NULL;
8567 	int err;
8568 	int drop_inode = 0;
8569 	u64 objectid;
8570 	u64 index = 0;
8571 	int name_len;
8572 	int datasize;
8573 	unsigned long ptr;
8574 	struct btrfs_file_extent_item *ei;
8575 	struct extent_buffer *leaf;
8576 
8577 	name_len = strlen(symname);
8578 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8579 		return -ENAMETOOLONG;
8580 
8581 	/*
8582 	 * 2 items for inode item and ref
8583 	 * 2 items for dir items
8584 	 * 1 item for xattr if selinux is on
8585 	 */
8586 	trans = btrfs_start_transaction(root, 5);
8587 	if (IS_ERR(trans))
8588 		return PTR_ERR(trans);
8589 
8590 	err = btrfs_find_free_ino(root, &objectid);
8591 	if (err)
8592 		goto out_unlock;
8593 
8594 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8595 				dentry->d_name.len, btrfs_ino(dir), objectid,
8596 				S_IFLNK|S_IRWXUGO, &index);
8597 	if (IS_ERR(inode)) {
8598 		err = PTR_ERR(inode);
8599 		goto out_unlock;
8600 	}
8601 
8602 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8603 	if (err) {
8604 		drop_inode = 1;
8605 		goto out_unlock;
8606 	}
8607 
8608 	/*
8609 	* If the active LSM wants to access the inode during
8610 	* d_instantiate it needs these. Smack checks to see
8611 	* if the filesystem supports xattrs by looking at the
8612 	* ops vector.
8613 	*/
8614 	inode->i_fop = &btrfs_file_operations;
8615 	inode->i_op = &btrfs_file_inode_operations;
8616 
8617 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8618 	if (err)
8619 		drop_inode = 1;
8620 	else {
8621 		inode->i_mapping->a_ops = &btrfs_aops;
8622 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8623 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8624 	}
8625 	if (drop_inode)
8626 		goto out_unlock;
8627 
8628 	path = btrfs_alloc_path();
8629 	if (!path) {
8630 		err = -ENOMEM;
8631 		drop_inode = 1;
8632 		goto out_unlock;
8633 	}
8634 	key.objectid = btrfs_ino(inode);
8635 	key.offset = 0;
8636 	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8637 	datasize = btrfs_file_extent_calc_inline_size(name_len);
8638 	err = btrfs_insert_empty_item(trans, root, path, &key,
8639 				      datasize);
8640 	if (err) {
8641 		drop_inode = 1;
8642 		btrfs_free_path(path);
8643 		goto out_unlock;
8644 	}
8645 	leaf = path->nodes[0];
8646 	ei = btrfs_item_ptr(leaf, path->slots[0],
8647 			    struct btrfs_file_extent_item);
8648 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8649 	btrfs_set_file_extent_type(leaf, ei,
8650 				   BTRFS_FILE_EXTENT_INLINE);
8651 	btrfs_set_file_extent_encryption(leaf, ei, 0);
8652 	btrfs_set_file_extent_compression(leaf, ei, 0);
8653 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8654 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8655 
8656 	ptr = btrfs_file_extent_inline_start(ei);
8657 	write_extent_buffer(leaf, symname, ptr, name_len);
8658 	btrfs_mark_buffer_dirty(leaf);
8659 	btrfs_free_path(path);
8660 
8661 	inode->i_op = &btrfs_symlink_inode_operations;
8662 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
8663 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8664 	inode_set_bytes(inode, name_len);
8665 	btrfs_i_size_write(inode, name_len);
8666 	err = btrfs_update_inode(trans, root, inode);
8667 	if (err)
8668 		drop_inode = 1;
8669 
8670 out_unlock:
8671 	if (!err)
8672 		d_instantiate(dentry, inode);
8673 	btrfs_end_transaction(trans, root);
8674 	if (drop_inode) {
8675 		inode_dec_link_count(inode);
8676 		iput(inode);
8677 	}
8678 	btrfs_btree_balance_dirty(root);
8679 	return err;
8680 }
8681 
8682 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8683 				       u64 start, u64 num_bytes, u64 min_size,
8684 				       loff_t actual_len, u64 *alloc_hint,
8685 				       struct btrfs_trans_handle *trans)
8686 {
8687 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8688 	struct extent_map *em;
8689 	struct btrfs_root *root = BTRFS_I(inode)->root;
8690 	struct btrfs_key ins;
8691 	u64 cur_offset = start;
8692 	u64 i_size;
8693 	u64 cur_bytes;
8694 	int ret = 0;
8695 	bool own_trans = true;
8696 
8697 	if (trans)
8698 		own_trans = false;
8699 	while (num_bytes > 0) {
8700 		if (own_trans) {
8701 			trans = btrfs_start_transaction(root, 3);
8702 			if (IS_ERR(trans)) {
8703 				ret = PTR_ERR(trans);
8704 				break;
8705 			}
8706 		}
8707 
8708 		cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8709 		cur_bytes = max(cur_bytes, min_size);
8710 		ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8711 					   *alloc_hint, &ins, 1);
8712 		if (ret) {
8713 			if (own_trans)
8714 				btrfs_end_transaction(trans, root);
8715 			break;
8716 		}
8717 
8718 		ret = insert_reserved_file_extent(trans, inode,
8719 						  cur_offset, ins.objectid,
8720 						  ins.offset, ins.offset,
8721 						  ins.offset, 0, 0, 0,
8722 						  BTRFS_FILE_EXTENT_PREALLOC);
8723 		if (ret) {
8724 			btrfs_free_reserved_extent(root, ins.objectid,
8725 						   ins.offset);
8726 			btrfs_abort_transaction(trans, root, ret);
8727 			if (own_trans)
8728 				btrfs_end_transaction(trans, root);
8729 			break;
8730 		}
8731 		btrfs_drop_extent_cache(inode, cur_offset,
8732 					cur_offset + ins.offset -1, 0);
8733 
8734 		em = alloc_extent_map();
8735 		if (!em) {
8736 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8737 				&BTRFS_I(inode)->runtime_flags);
8738 			goto next;
8739 		}
8740 
8741 		em->start = cur_offset;
8742 		em->orig_start = cur_offset;
8743 		em->len = ins.offset;
8744 		em->block_start = ins.objectid;
8745 		em->block_len = ins.offset;
8746 		em->orig_block_len = ins.offset;
8747 		em->ram_bytes = ins.offset;
8748 		em->bdev = root->fs_info->fs_devices->latest_bdev;
8749 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8750 		em->generation = trans->transid;
8751 
8752 		while (1) {
8753 			write_lock(&em_tree->lock);
8754 			ret = add_extent_mapping(em_tree, em, 1);
8755 			write_unlock(&em_tree->lock);
8756 			if (ret != -EEXIST)
8757 				break;
8758 			btrfs_drop_extent_cache(inode, cur_offset,
8759 						cur_offset + ins.offset - 1,
8760 						0);
8761 		}
8762 		free_extent_map(em);
8763 next:
8764 		num_bytes -= ins.offset;
8765 		cur_offset += ins.offset;
8766 		*alloc_hint = ins.objectid + ins.offset;
8767 
8768 		inode_inc_iversion(inode);
8769 		inode->i_ctime = CURRENT_TIME;
8770 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8771 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8772 		    (actual_len > inode->i_size) &&
8773 		    (cur_offset > inode->i_size)) {
8774 			if (cur_offset > actual_len)
8775 				i_size = actual_len;
8776 			else
8777 				i_size = cur_offset;
8778 			i_size_write(inode, i_size);
8779 			btrfs_ordered_update_i_size(inode, i_size, NULL);
8780 		}
8781 
8782 		ret = btrfs_update_inode(trans, root, inode);
8783 
8784 		if (ret) {
8785 			btrfs_abort_transaction(trans, root, ret);
8786 			if (own_trans)
8787 				btrfs_end_transaction(trans, root);
8788 			break;
8789 		}
8790 
8791 		if (own_trans)
8792 			btrfs_end_transaction(trans, root);
8793 	}
8794 	return ret;
8795 }
8796 
8797 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8798 			      u64 start, u64 num_bytes, u64 min_size,
8799 			      loff_t actual_len, u64 *alloc_hint)
8800 {
8801 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8802 					   min_size, actual_len, alloc_hint,
8803 					   NULL);
8804 }
8805 
8806 int btrfs_prealloc_file_range_trans(struct inode *inode,
8807 				    struct btrfs_trans_handle *trans, int mode,
8808 				    u64 start, u64 num_bytes, u64 min_size,
8809 				    loff_t actual_len, u64 *alloc_hint)
8810 {
8811 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8812 					   min_size, actual_len, alloc_hint, trans);
8813 }
8814 
8815 static int btrfs_set_page_dirty(struct page *page)
8816 {
8817 	return __set_page_dirty_nobuffers(page);
8818 }
8819 
8820 static int btrfs_permission(struct inode *inode, int mask)
8821 {
8822 	struct btrfs_root *root = BTRFS_I(inode)->root;
8823 	umode_t mode = inode->i_mode;
8824 
8825 	if (mask & MAY_WRITE &&
8826 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8827 		if (btrfs_root_readonly(root))
8828 			return -EROFS;
8829 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8830 			return -EACCES;
8831 	}
8832 	return generic_permission(inode, mask);
8833 }
8834 
8835 static const struct inode_operations btrfs_dir_inode_operations = {
8836 	.getattr	= btrfs_getattr,
8837 	.lookup		= btrfs_lookup,
8838 	.create		= btrfs_create,
8839 	.unlink		= btrfs_unlink,
8840 	.link		= btrfs_link,
8841 	.mkdir		= btrfs_mkdir,
8842 	.rmdir		= btrfs_rmdir,
8843 	.rename		= btrfs_rename,
8844 	.symlink	= btrfs_symlink,
8845 	.setattr	= btrfs_setattr,
8846 	.mknod		= btrfs_mknod,
8847 	.setxattr	= btrfs_setxattr,
8848 	.getxattr	= btrfs_getxattr,
8849 	.listxattr	= btrfs_listxattr,
8850 	.removexattr	= btrfs_removexattr,
8851 	.permission	= btrfs_permission,
8852 	.get_acl	= btrfs_get_acl,
8853 	.set_acl	= btrfs_set_acl,
8854 	.update_time	= btrfs_update_time,
8855 };
8856 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8857 	.lookup		= btrfs_lookup,
8858 	.permission	= btrfs_permission,
8859 	.get_acl	= btrfs_get_acl,
8860 	.set_acl	= btrfs_set_acl,
8861 	.update_time	= btrfs_update_time,
8862 };
8863 
8864 static const struct file_operations btrfs_dir_file_operations = {
8865 	.llseek		= generic_file_llseek,
8866 	.read		= generic_read_dir,
8867 	.iterate	= btrfs_real_readdir,
8868 	.unlocked_ioctl	= btrfs_ioctl,
8869 #ifdef CONFIG_COMPAT
8870 	.compat_ioctl	= btrfs_ioctl,
8871 #endif
8872 	.release        = btrfs_release_file,
8873 	.fsync		= btrfs_sync_file,
8874 };
8875 
8876 static struct extent_io_ops btrfs_extent_io_ops = {
8877 	.fill_delalloc = run_delalloc_range,
8878 	.submit_bio_hook = btrfs_submit_bio_hook,
8879 	.merge_bio_hook = btrfs_merge_bio_hook,
8880 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
8881 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
8882 	.writepage_start_hook = btrfs_writepage_start_hook,
8883 	.set_bit_hook = btrfs_set_bit_hook,
8884 	.clear_bit_hook = btrfs_clear_bit_hook,
8885 	.merge_extent_hook = btrfs_merge_extent_hook,
8886 	.split_extent_hook = btrfs_split_extent_hook,
8887 };
8888 
8889 /*
8890  * btrfs doesn't support the bmap operation because swapfiles
8891  * use bmap to make a mapping of extents in the file.  They assume
8892  * these extents won't change over the life of the file and they
8893  * use the bmap result to do IO directly to the drive.
8894  *
8895  * the btrfs bmap call would return logical addresses that aren't
8896  * suitable for IO and they also will change frequently as COW
8897  * operations happen.  So, swapfile + btrfs == corruption.
8898  *
8899  * For now we're avoiding this by dropping bmap.
8900  */
8901 static const struct address_space_operations btrfs_aops = {
8902 	.readpage	= btrfs_readpage,
8903 	.writepage	= btrfs_writepage,
8904 	.writepages	= btrfs_writepages,
8905 	.readpages	= btrfs_readpages,
8906 	.direct_IO	= btrfs_direct_IO,
8907 	.invalidatepage = btrfs_invalidatepage,
8908 	.releasepage	= btrfs_releasepage,
8909 	.set_page_dirty	= btrfs_set_page_dirty,
8910 	.error_remove_page = generic_error_remove_page,
8911 };
8912 
8913 static const struct address_space_operations btrfs_symlink_aops = {
8914 	.readpage	= btrfs_readpage,
8915 	.writepage	= btrfs_writepage,
8916 	.invalidatepage = btrfs_invalidatepage,
8917 	.releasepage	= btrfs_releasepage,
8918 };
8919 
8920 static const struct inode_operations btrfs_file_inode_operations = {
8921 	.getattr	= btrfs_getattr,
8922 	.setattr	= btrfs_setattr,
8923 	.setxattr	= btrfs_setxattr,
8924 	.getxattr	= btrfs_getxattr,
8925 	.listxattr      = btrfs_listxattr,
8926 	.removexattr	= btrfs_removexattr,
8927 	.permission	= btrfs_permission,
8928 	.fiemap		= btrfs_fiemap,
8929 	.get_acl	= btrfs_get_acl,
8930 	.set_acl	= btrfs_set_acl,
8931 	.update_time	= btrfs_update_time,
8932 };
8933 static const struct inode_operations btrfs_special_inode_operations = {
8934 	.getattr	= btrfs_getattr,
8935 	.setattr	= btrfs_setattr,
8936 	.permission	= btrfs_permission,
8937 	.setxattr	= btrfs_setxattr,
8938 	.getxattr	= btrfs_getxattr,
8939 	.listxattr	= btrfs_listxattr,
8940 	.removexattr	= btrfs_removexattr,
8941 	.get_acl	= btrfs_get_acl,
8942 	.set_acl	= btrfs_set_acl,
8943 	.update_time	= btrfs_update_time,
8944 };
8945 static const struct inode_operations btrfs_symlink_inode_operations = {
8946 	.readlink	= generic_readlink,
8947 	.follow_link	= page_follow_link_light,
8948 	.put_link	= page_put_link,
8949 	.getattr	= btrfs_getattr,
8950 	.setattr	= btrfs_setattr,
8951 	.permission	= btrfs_permission,
8952 	.setxattr	= btrfs_setxattr,
8953 	.getxattr	= btrfs_getxattr,
8954 	.listxattr	= btrfs_listxattr,
8955 	.removexattr	= btrfs_removexattr,
8956 	.update_time	= btrfs_update_time,
8957 };
8958 
8959 const struct dentry_operations btrfs_dentry_operations = {
8960 	.d_delete	= btrfs_dentry_delete,
8961 	.d_release	= btrfs_dentry_release,
8962 };
8963