xref: /openbmc/linux/fs/btrfs/inode.c (revision 41bd9ca459a007cc5588563bb08de9677c8d23fd)
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 		ret = -EINVAL;
868 		goto out_unlock;
869 	}
870 
871 	num_bytes = ALIGN(end - start + 1, blocksize);
872 	num_bytes = max(blocksize,  num_bytes);
873 	disk_num_bytes = num_bytes;
874 
875 	/* if this is a small write inside eof, kick off defrag */
876 	if (num_bytes < 64 * 1024 &&
877 	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
878 		btrfs_add_inode_defrag(NULL, inode);
879 
880 	if (start == 0) {
881 		/* lets try to make an inline extent */
882 		ret = cow_file_range_inline(root, inode, start, end, 0, 0,
883 					    NULL);
884 		if (ret == 0) {
885 			extent_clear_unlock_delalloc(inode, start, end, NULL,
886 				     EXTENT_LOCKED | EXTENT_DELALLOC |
887 				     EXTENT_DEFRAG, PAGE_UNLOCK |
888 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
889 				     PAGE_END_WRITEBACK);
890 
891 			*nr_written = *nr_written +
892 			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
893 			*page_started = 1;
894 			goto out;
895 		} else if (ret < 0) {
896 			goto out_unlock;
897 		}
898 	}
899 
900 	BUG_ON(disk_num_bytes >
901 	       btrfs_super_total_bytes(root->fs_info->super_copy));
902 
903 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
904 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
905 
906 	while (disk_num_bytes > 0) {
907 		unsigned long op;
908 
909 		cur_alloc_size = disk_num_bytes;
910 		ret = btrfs_reserve_extent(root, cur_alloc_size,
911 					   root->sectorsize, 0, alloc_hint,
912 					   &ins, 1);
913 		if (ret < 0)
914 			goto out_unlock;
915 
916 		em = alloc_extent_map();
917 		if (!em) {
918 			ret = -ENOMEM;
919 			goto out_reserve;
920 		}
921 		em->start = start;
922 		em->orig_start = em->start;
923 		ram_size = ins.offset;
924 		em->len = ins.offset;
925 		em->mod_start = em->start;
926 		em->mod_len = em->len;
927 
928 		em->block_start = ins.objectid;
929 		em->block_len = ins.offset;
930 		em->orig_block_len = ins.offset;
931 		em->ram_bytes = ram_size;
932 		em->bdev = root->fs_info->fs_devices->latest_bdev;
933 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
934 		em->generation = -1;
935 
936 		while (1) {
937 			write_lock(&em_tree->lock);
938 			ret = add_extent_mapping(em_tree, em, 1);
939 			write_unlock(&em_tree->lock);
940 			if (ret != -EEXIST) {
941 				free_extent_map(em);
942 				break;
943 			}
944 			btrfs_drop_extent_cache(inode, start,
945 						start + ram_size - 1, 0);
946 		}
947 		if (ret)
948 			goto out_reserve;
949 
950 		cur_alloc_size = ins.offset;
951 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
952 					       ram_size, cur_alloc_size, 0);
953 		if (ret)
954 			goto out_reserve;
955 
956 		if (root->root_key.objectid ==
957 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
958 			ret = btrfs_reloc_clone_csums(inode, start,
959 						      cur_alloc_size);
960 			if (ret)
961 				goto out_reserve;
962 		}
963 
964 		if (disk_num_bytes < cur_alloc_size)
965 			break;
966 
967 		/* we're not doing compressed IO, don't unlock the first
968 		 * page (which the caller expects to stay locked), don't
969 		 * clear any dirty bits and don't set any writeback bits
970 		 *
971 		 * Do set the Private2 bit so we know this page was properly
972 		 * setup for writepage
973 		 */
974 		op = unlock ? PAGE_UNLOCK : 0;
975 		op |= PAGE_SET_PRIVATE2;
976 
977 		extent_clear_unlock_delalloc(inode, start,
978 					     start + ram_size - 1, locked_page,
979 					     EXTENT_LOCKED | EXTENT_DELALLOC,
980 					     op);
981 		disk_num_bytes -= cur_alloc_size;
982 		num_bytes -= cur_alloc_size;
983 		alloc_hint = ins.objectid + ins.offset;
984 		start += cur_alloc_size;
985 	}
986 out:
987 	return ret;
988 
989 out_reserve:
990 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
991 out_unlock:
992 	extent_clear_unlock_delalloc(inode, start, end, locked_page,
993 				     EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
994 				     EXTENT_DELALLOC | EXTENT_DEFRAG,
995 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
996 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
997 	goto out;
998 }
999 
1000 /*
1001  * work queue call back to started compression on a file and pages
1002  */
1003 static noinline void async_cow_start(struct btrfs_work *work)
1004 {
1005 	struct async_cow *async_cow;
1006 	int num_added = 0;
1007 	async_cow = container_of(work, struct async_cow, work);
1008 
1009 	compress_file_range(async_cow->inode, async_cow->locked_page,
1010 			    async_cow->start, async_cow->end, async_cow,
1011 			    &num_added);
1012 	if (num_added == 0) {
1013 		btrfs_add_delayed_iput(async_cow->inode);
1014 		async_cow->inode = NULL;
1015 	}
1016 }
1017 
1018 /*
1019  * work queue call back to submit previously compressed pages
1020  */
1021 static noinline void async_cow_submit(struct btrfs_work *work)
1022 {
1023 	struct async_cow *async_cow;
1024 	struct btrfs_root *root;
1025 	unsigned long nr_pages;
1026 
1027 	async_cow = container_of(work, struct async_cow, work);
1028 
1029 	root = async_cow->root;
1030 	nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1031 		PAGE_CACHE_SHIFT;
1032 
1033 	if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1034 	    5 * 1024 * 1024 &&
1035 	    waitqueue_active(&root->fs_info->async_submit_wait))
1036 		wake_up(&root->fs_info->async_submit_wait);
1037 
1038 	if (async_cow->inode)
1039 		submit_compressed_extents(async_cow->inode, async_cow);
1040 }
1041 
1042 static noinline void async_cow_free(struct btrfs_work *work)
1043 {
1044 	struct async_cow *async_cow;
1045 	async_cow = container_of(work, struct async_cow, work);
1046 	if (async_cow->inode)
1047 		btrfs_add_delayed_iput(async_cow->inode);
1048 	kfree(async_cow);
1049 }
1050 
1051 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1052 				u64 start, u64 end, int *page_started,
1053 				unsigned long *nr_written)
1054 {
1055 	struct async_cow *async_cow;
1056 	struct btrfs_root *root = BTRFS_I(inode)->root;
1057 	unsigned long nr_pages;
1058 	u64 cur_end;
1059 	int limit = 10 * 1024 * 1024;
1060 
1061 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1062 			 1, 0, NULL, GFP_NOFS);
1063 	while (start < end) {
1064 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1065 		BUG_ON(!async_cow); /* -ENOMEM */
1066 		async_cow->inode = igrab(inode);
1067 		async_cow->root = root;
1068 		async_cow->locked_page = locked_page;
1069 		async_cow->start = start;
1070 
1071 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1072 			cur_end = end;
1073 		else
1074 			cur_end = min(end, start + 512 * 1024 - 1);
1075 
1076 		async_cow->end = cur_end;
1077 		INIT_LIST_HEAD(&async_cow->extents);
1078 
1079 		btrfs_init_work(&async_cow->work, async_cow_start,
1080 				async_cow_submit, async_cow_free);
1081 
1082 		nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1083 			PAGE_CACHE_SHIFT;
1084 		atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1085 
1086 		btrfs_queue_work(root->fs_info->delalloc_workers,
1087 				 &async_cow->work);
1088 
1089 		if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1090 			wait_event(root->fs_info->async_submit_wait,
1091 			   (atomic_read(&root->fs_info->async_delalloc_pages) <
1092 			    limit));
1093 		}
1094 
1095 		while (atomic_read(&root->fs_info->async_submit_draining) &&
1096 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
1097 			wait_event(root->fs_info->async_submit_wait,
1098 			  (atomic_read(&root->fs_info->async_delalloc_pages) ==
1099 			   0));
1100 		}
1101 
1102 		*nr_written += nr_pages;
1103 		start = cur_end + 1;
1104 	}
1105 	*page_started = 1;
1106 	return 0;
1107 }
1108 
1109 static noinline int csum_exist_in_range(struct btrfs_root *root,
1110 					u64 bytenr, u64 num_bytes)
1111 {
1112 	int ret;
1113 	struct btrfs_ordered_sum *sums;
1114 	LIST_HEAD(list);
1115 
1116 	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1117 				       bytenr + num_bytes - 1, &list, 0);
1118 	if (ret == 0 && list_empty(&list))
1119 		return 0;
1120 
1121 	while (!list_empty(&list)) {
1122 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1123 		list_del(&sums->list);
1124 		kfree(sums);
1125 	}
1126 	return 1;
1127 }
1128 
1129 /*
1130  * when nowcow writeback call back.  This checks for snapshots or COW copies
1131  * of the extents that exist in the file, and COWs the file as required.
1132  *
1133  * If no cow copies or snapshots exist, we write directly to the existing
1134  * blocks on disk
1135  */
1136 static noinline int run_delalloc_nocow(struct inode *inode,
1137 				       struct page *locked_page,
1138 			      u64 start, u64 end, int *page_started, int force,
1139 			      unsigned long *nr_written)
1140 {
1141 	struct btrfs_root *root = BTRFS_I(inode)->root;
1142 	struct btrfs_trans_handle *trans;
1143 	struct extent_buffer *leaf;
1144 	struct btrfs_path *path;
1145 	struct btrfs_file_extent_item *fi;
1146 	struct btrfs_key found_key;
1147 	u64 cow_start;
1148 	u64 cur_offset;
1149 	u64 extent_end;
1150 	u64 extent_offset;
1151 	u64 disk_bytenr;
1152 	u64 num_bytes;
1153 	u64 disk_num_bytes;
1154 	u64 ram_bytes;
1155 	int extent_type;
1156 	int ret, err;
1157 	int type;
1158 	int nocow;
1159 	int check_prev = 1;
1160 	bool nolock;
1161 	u64 ino = btrfs_ino(inode);
1162 
1163 	path = btrfs_alloc_path();
1164 	if (!path) {
1165 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1167 					     EXTENT_DO_ACCOUNTING |
1168 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1169 					     PAGE_CLEAR_DIRTY |
1170 					     PAGE_SET_WRITEBACK |
1171 					     PAGE_END_WRITEBACK);
1172 		return -ENOMEM;
1173 	}
1174 
1175 	nolock = btrfs_is_free_space_inode(inode);
1176 
1177 	if (nolock)
1178 		trans = btrfs_join_transaction_nolock(root);
1179 	else
1180 		trans = btrfs_join_transaction(root);
1181 
1182 	if (IS_ERR(trans)) {
1183 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1184 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1185 					     EXTENT_DO_ACCOUNTING |
1186 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1187 					     PAGE_CLEAR_DIRTY |
1188 					     PAGE_SET_WRITEBACK |
1189 					     PAGE_END_WRITEBACK);
1190 		btrfs_free_path(path);
1191 		return PTR_ERR(trans);
1192 	}
1193 
1194 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1195 
1196 	cow_start = (u64)-1;
1197 	cur_offset = start;
1198 	while (1) {
1199 		ret = btrfs_lookup_file_extent(trans, root, path, ino,
1200 					       cur_offset, 0);
1201 		if (ret < 0)
1202 			goto error;
1203 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1204 			leaf = path->nodes[0];
1205 			btrfs_item_key_to_cpu(leaf, &found_key,
1206 					      path->slots[0] - 1);
1207 			if (found_key.objectid == ino &&
1208 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1209 				path->slots[0]--;
1210 		}
1211 		check_prev = 0;
1212 next_slot:
1213 		leaf = path->nodes[0];
1214 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1215 			ret = btrfs_next_leaf(root, path);
1216 			if (ret < 0)
1217 				goto error;
1218 			if (ret > 0)
1219 				break;
1220 			leaf = path->nodes[0];
1221 		}
1222 
1223 		nocow = 0;
1224 		disk_bytenr = 0;
1225 		num_bytes = 0;
1226 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1227 
1228 		if (found_key.objectid > ino ||
1229 		    found_key.type > BTRFS_EXTENT_DATA_KEY ||
1230 		    found_key.offset > end)
1231 			break;
1232 
1233 		if (found_key.offset > cur_offset) {
1234 			extent_end = found_key.offset;
1235 			extent_type = 0;
1236 			goto out_check;
1237 		}
1238 
1239 		fi = btrfs_item_ptr(leaf, path->slots[0],
1240 				    struct btrfs_file_extent_item);
1241 		extent_type = btrfs_file_extent_type(leaf, fi);
1242 
1243 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1244 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1245 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1246 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1247 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1248 			extent_end = found_key.offset +
1249 				btrfs_file_extent_num_bytes(leaf, fi);
1250 			disk_num_bytes =
1251 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1252 			if (extent_end <= start) {
1253 				path->slots[0]++;
1254 				goto next_slot;
1255 			}
1256 			if (disk_bytenr == 0)
1257 				goto out_check;
1258 			if (btrfs_file_extent_compression(leaf, fi) ||
1259 			    btrfs_file_extent_encryption(leaf, fi) ||
1260 			    btrfs_file_extent_other_encoding(leaf, fi))
1261 				goto out_check;
1262 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1263 				goto out_check;
1264 			if (btrfs_extent_readonly(root, disk_bytenr))
1265 				goto out_check;
1266 			if (btrfs_cross_ref_exist(trans, root, ino,
1267 						  found_key.offset -
1268 						  extent_offset, disk_bytenr))
1269 				goto out_check;
1270 			disk_bytenr += extent_offset;
1271 			disk_bytenr += cur_offset - found_key.offset;
1272 			num_bytes = min(end + 1, extent_end) - cur_offset;
1273 			/*
1274 			 * force cow if csum exists in the range.
1275 			 * this ensure that csum for a given extent are
1276 			 * either valid or do not exist.
1277 			 */
1278 			if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1279 				goto out_check;
1280 			nocow = 1;
1281 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1282 			extent_end = found_key.offset +
1283 				btrfs_file_extent_inline_len(leaf,
1284 						     path->slots[0], fi);
1285 			extent_end = ALIGN(extent_end, root->sectorsize);
1286 		} else {
1287 			BUG_ON(1);
1288 		}
1289 out_check:
1290 		if (extent_end <= start) {
1291 			path->slots[0]++;
1292 			goto next_slot;
1293 		}
1294 		if (!nocow) {
1295 			if (cow_start == (u64)-1)
1296 				cow_start = cur_offset;
1297 			cur_offset = extent_end;
1298 			if (cur_offset > end)
1299 				break;
1300 			path->slots[0]++;
1301 			goto next_slot;
1302 		}
1303 
1304 		btrfs_release_path(path);
1305 		if (cow_start != (u64)-1) {
1306 			ret = cow_file_range(inode, locked_page,
1307 					     cow_start, found_key.offset - 1,
1308 					     page_started, nr_written, 1);
1309 			if (ret)
1310 				goto error;
1311 			cow_start = (u64)-1;
1312 		}
1313 
1314 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315 			struct extent_map *em;
1316 			struct extent_map_tree *em_tree;
1317 			em_tree = &BTRFS_I(inode)->extent_tree;
1318 			em = alloc_extent_map();
1319 			BUG_ON(!em); /* -ENOMEM */
1320 			em->start = cur_offset;
1321 			em->orig_start = found_key.offset - extent_offset;
1322 			em->len = num_bytes;
1323 			em->block_len = num_bytes;
1324 			em->block_start = disk_bytenr;
1325 			em->orig_block_len = disk_num_bytes;
1326 			em->ram_bytes = ram_bytes;
1327 			em->bdev = root->fs_info->fs_devices->latest_bdev;
1328 			em->mod_start = em->start;
1329 			em->mod_len = em->len;
1330 			set_bit(EXTENT_FLAG_PINNED, &em->flags);
1331 			set_bit(EXTENT_FLAG_FILLING, &em->flags);
1332 			em->generation = -1;
1333 			while (1) {
1334 				write_lock(&em_tree->lock);
1335 				ret = add_extent_mapping(em_tree, em, 1);
1336 				write_unlock(&em_tree->lock);
1337 				if (ret != -EEXIST) {
1338 					free_extent_map(em);
1339 					break;
1340 				}
1341 				btrfs_drop_extent_cache(inode, em->start,
1342 						em->start + em->len - 1, 0);
1343 			}
1344 			type = BTRFS_ORDERED_PREALLOC;
1345 		} else {
1346 			type = BTRFS_ORDERED_NOCOW;
1347 		}
1348 
1349 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1350 					       num_bytes, num_bytes, type);
1351 		BUG_ON(ret); /* -ENOMEM */
1352 
1353 		if (root->root_key.objectid ==
1354 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1355 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1356 						      num_bytes);
1357 			if (ret)
1358 				goto error;
1359 		}
1360 
1361 		extent_clear_unlock_delalloc(inode, cur_offset,
1362 					     cur_offset + num_bytes - 1,
1363 					     locked_page, EXTENT_LOCKED |
1364 					     EXTENT_DELALLOC, PAGE_UNLOCK |
1365 					     PAGE_SET_PRIVATE2);
1366 		cur_offset = extent_end;
1367 		if (cur_offset > end)
1368 			break;
1369 	}
1370 	btrfs_release_path(path);
1371 
1372 	if (cur_offset <= end && cow_start == (u64)-1) {
1373 		cow_start = cur_offset;
1374 		cur_offset = end;
1375 	}
1376 
1377 	if (cow_start != (u64)-1) {
1378 		ret = cow_file_range(inode, locked_page, cow_start, end,
1379 				     page_started, nr_written, 1);
1380 		if (ret)
1381 			goto error;
1382 	}
1383 
1384 error:
1385 	err = btrfs_end_transaction(trans, root);
1386 	if (!ret)
1387 		ret = err;
1388 
1389 	if (ret && cur_offset < end)
1390 		extent_clear_unlock_delalloc(inode, cur_offset, end,
1391 					     locked_page, EXTENT_LOCKED |
1392 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1393 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1394 					     PAGE_CLEAR_DIRTY |
1395 					     PAGE_SET_WRITEBACK |
1396 					     PAGE_END_WRITEBACK);
1397 	btrfs_free_path(path);
1398 	return ret;
1399 }
1400 
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405 			      u64 start, u64 end, int *page_started,
1406 			      unsigned long *nr_written)
1407 {
1408 	int ret;
1409 	struct btrfs_root *root = BTRFS_I(inode)->root;
1410 
1411 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1413 					 page_started, 1, nr_written);
1414 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1416 					 page_started, 0, nr_written);
1417 	} else if (!btrfs_test_opt(root, COMPRESS) &&
1418 		   !(BTRFS_I(inode)->force_compress) &&
1419 		   !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420 		ret = cow_file_range(inode, locked_page, start, end,
1421 				      page_started, nr_written, 1);
1422 	} else {
1423 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424 			&BTRFS_I(inode)->runtime_flags);
1425 		ret = cow_file_range_async(inode, locked_page, start, end,
1426 					   page_started, nr_written);
1427 	}
1428 	return ret;
1429 }
1430 
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432 				    struct extent_state *orig, u64 split)
1433 {
1434 	/* not delalloc, ignore it */
1435 	if (!(orig->state & EXTENT_DELALLOC))
1436 		return;
1437 
1438 	spin_lock(&BTRFS_I(inode)->lock);
1439 	BTRFS_I(inode)->outstanding_extents++;
1440 	spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442 
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450 				    struct extent_state *new,
1451 				    struct extent_state *other)
1452 {
1453 	/* not delalloc, ignore it */
1454 	if (!(other->state & EXTENT_DELALLOC))
1455 		return;
1456 
1457 	spin_lock(&BTRFS_I(inode)->lock);
1458 	BTRFS_I(inode)->outstanding_extents--;
1459 	spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461 
1462 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1463 				      struct inode *inode)
1464 {
1465 	spin_lock(&root->delalloc_lock);
1466 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1467 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1468 			      &root->delalloc_inodes);
1469 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1470 			&BTRFS_I(inode)->runtime_flags);
1471 		root->nr_delalloc_inodes++;
1472 		if (root->nr_delalloc_inodes == 1) {
1473 			spin_lock(&root->fs_info->delalloc_root_lock);
1474 			BUG_ON(!list_empty(&root->delalloc_root));
1475 			list_add_tail(&root->delalloc_root,
1476 				      &root->fs_info->delalloc_roots);
1477 			spin_unlock(&root->fs_info->delalloc_root_lock);
1478 		}
1479 	}
1480 	spin_unlock(&root->delalloc_lock);
1481 }
1482 
1483 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1484 				     struct inode *inode)
1485 {
1486 	spin_lock(&root->delalloc_lock);
1487 	if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1488 		list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1489 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1490 			  &BTRFS_I(inode)->runtime_flags);
1491 		root->nr_delalloc_inodes--;
1492 		if (!root->nr_delalloc_inodes) {
1493 			spin_lock(&root->fs_info->delalloc_root_lock);
1494 			BUG_ON(list_empty(&root->delalloc_root));
1495 			list_del_init(&root->delalloc_root);
1496 			spin_unlock(&root->fs_info->delalloc_root_lock);
1497 		}
1498 	}
1499 	spin_unlock(&root->delalloc_lock);
1500 }
1501 
1502 /*
1503  * extent_io.c set_bit_hook, used to track delayed allocation
1504  * bytes in this file, and to maintain the list of inodes that
1505  * have pending delalloc work to be done.
1506  */
1507 static void btrfs_set_bit_hook(struct inode *inode,
1508 			       struct extent_state *state, unsigned long *bits)
1509 {
1510 
1511 	/*
1512 	 * set_bit and clear bit hooks normally require _irqsave/restore
1513 	 * but in this case, we are only testing for the DELALLOC
1514 	 * bit, which is only set or cleared with irqs on
1515 	 */
1516 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1517 		struct btrfs_root *root = BTRFS_I(inode)->root;
1518 		u64 len = state->end + 1 - state->start;
1519 		bool do_list = !btrfs_is_free_space_inode(inode);
1520 
1521 		if (*bits & EXTENT_FIRST_DELALLOC) {
1522 			*bits &= ~EXTENT_FIRST_DELALLOC;
1523 		} else {
1524 			spin_lock(&BTRFS_I(inode)->lock);
1525 			BTRFS_I(inode)->outstanding_extents++;
1526 			spin_unlock(&BTRFS_I(inode)->lock);
1527 		}
1528 
1529 		__percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1530 				     root->fs_info->delalloc_batch);
1531 		spin_lock(&BTRFS_I(inode)->lock);
1532 		BTRFS_I(inode)->delalloc_bytes += len;
1533 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1534 					 &BTRFS_I(inode)->runtime_flags))
1535 			btrfs_add_delalloc_inodes(root, inode);
1536 		spin_unlock(&BTRFS_I(inode)->lock);
1537 	}
1538 }
1539 
1540 /*
1541  * extent_io.c clear_bit_hook, see set_bit_hook for why
1542  */
1543 static void btrfs_clear_bit_hook(struct inode *inode,
1544 				 struct extent_state *state,
1545 				 unsigned long *bits)
1546 {
1547 	/*
1548 	 * set_bit and clear bit hooks normally require _irqsave/restore
1549 	 * but in this case, we are only testing for the DELALLOC
1550 	 * bit, which is only set or cleared with irqs on
1551 	 */
1552 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1553 		struct btrfs_root *root = BTRFS_I(inode)->root;
1554 		u64 len = state->end + 1 - state->start;
1555 		bool do_list = !btrfs_is_free_space_inode(inode);
1556 
1557 		if (*bits & EXTENT_FIRST_DELALLOC) {
1558 			*bits &= ~EXTENT_FIRST_DELALLOC;
1559 		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1560 			spin_lock(&BTRFS_I(inode)->lock);
1561 			BTRFS_I(inode)->outstanding_extents--;
1562 			spin_unlock(&BTRFS_I(inode)->lock);
1563 		}
1564 
1565 		/*
1566 		 * We don't reserve metadata space for space cache inodes so we
1567 		 * don't need to call dellalloc_release_metadata if there is an
1568 		 * error.
1569 		 */
1570 		if (*bits & EXTENT_DO_ACCOUNTING &&
1571 		    root != root->fs_info->tree_root)
1572 			btrfs_delalloc_release_metadata(inode, len);
1573 
1574 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1575 		    && do_list && !(state->state & EXTENT_NORESERVE))
1576 			btrfs_free_reserved_data_space(inode, len);
1577 
1578 		__percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1579 				     root->fs_info->delalloc_batch);
1580 		spin_lock(&BTRFS_I(inode)->lock);
1581 		BTRFS_I(inode)->delalloc_bytes -= len;
1582 		if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1583 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1584 			     &BTRFS_I(inode)->runtime_flags))
1585 			btrfs_del_delalloc_inode(root, inode);
1586 		spin_unlock(&BTRFS_I(inode)->lock);
1587 	}
1588 }
1589 
1590 /*
1591  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1592  * we don't create bios that span stripes or chunks
1593  */
1594 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1595 			 size_t size, struct bio *bio,
1596 			 unsigned long bio_flags)
1597 {
1598 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1599 	u64 logical = (u64)bio->bi_sector << 9;
1600 	u64 length = 0;
1601 	u64 map_length;
1602 	int ret;
1603 
1604 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1605 		return 0;
1606 
1607 	length = bio->bi_size;
1608 	map_length = length;
1609 	ret = btrfs_map_block(root->fs_info, rw, logical,
1610 			      &map_length, NULL, 0);
1611 	/* Will always return 0 with map_multi == NULL */
1612 	BUG_ON(ret < 0);
1613 	if (map_length < length + size)
1614 		return 1;
1615 	return 0;
1616 }
1617 
1618 /*
1619  * in order to insert checksums into the metadata in large chunks,
1620  * we wait until bio submission time.   All the pages in the bio are
1621  * checksummed and sums are attached onto the ordered extent record.
1622  *
1623  * At IO completion time the cums attached on the ordered extent record
1624  * are inserted into the btree
1625  */
1626 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1627 				    struct bio *bio, int mirror_num,
1628 				    unsigned long bio_flags,
1629 				    u64 bio_offset)
1630 {
1631 	struct btrfs_root *root = BTRFS_I(inode)->root;
1632 	int ret = 0;
1633 
1634 	ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1635 	BUG_ON(ret); /* -ENOMEM */
1636 	return 0;
1637 }
1638 
1639 /*
1640  * in order to insert checksums into the metadata in large chunks,
1641  * we wait until bio submission time.   All the pages in the bio are
1642  * checksummed and sums are attached onto the ordered extent record.
1643  *
1644  * At IO completion time the cums attached on the ordered extent record
1645  * are inserted into the btree
1646  */
1647 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1648 			  int mirror_num, unsigned long bio_flags,
1649 			  u64 bio_offset)
1650 {
1651 	struct btrfs_root *root = BTRFS_I(inode)->root;
1652 	int ret;
1653 
1654 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1655 	if (ret)
1656 		bio_endio(bio, ret);
1657 	return ret;
1658 }
1659 
1660 /*
1661  * extent_io.c submission hook. This does the right thing for csum calculation
1662  * on write, or reading the csums from the tree before a read
1663  */
1664 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1665 			  int mirror_num, unsigned long bio_flags,
1666 			  u64 bio_offset)
1667 {
1668 	struct btrfs_root *root = BTRFS_I(inode)->root;
1669 	int ret = 0;
1670 	int skip_sum;
1671 	int metadata = 0;
1672 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1673 
1674 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1675 
1676 	if (btrfs_is_free_space_inode(inode))
1677 		metadata = 2;
1678 
1679 	if (!(rw & REQ_WRITE)) {
1680 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1681 		if (ret)
1682 			goto out;
1683 
1684 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1685 			ret = btrfs_submit_compressed_read(inode, bio,
1686 							   mirror_num,
1687 							   bio_flags);
1688 			goto out;
1689 		} else if (!skip_sum) {
1690 			ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1691 			if (ret)
1692 				goto out;
1693 		}
1694 		goto mapit;
1695 	} else if (async && !skip_sum) {
1696 		/* csum items have already been cloned */
1697 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1698 			goto mapit;
1699 		/* we're doing a write, do the async checksumming */
1700 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1701 				   inode, rw, bio, mirror_num,
1702 				   bio_flags, bio_offset,
1703 				   __btrfs_submit_bio_start,
1704 				   __btrfs_submit_bio_done);
1705 		goto out;
1706 	} else if (!skip_sum) {
1707 		ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1708 		if (ret)
1709 			goto out;
1710 	}
1711 
1712 mapit:
1713 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1714 
1715 out:
1716 	if (ret < 0)
1717 		bio_endio(bio, ret);
1718 	return ret;
1719 }
1720 
1721 /*
1722  * given a list of ordered sums record them in the inode.  This happens
1723  * at IO completion time based on sums calculated at bio submission time.
1724  */
1725 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1726 			     struct inode *inode, u64 file_offset,
1727 			     struct list_head *list)
1728 {
1729 	struct btrfs_ordered_sum *sum;
1730 
1731 	list_for_each_entry(sum, list, list) {
1732 		trans->adding_csums = 1;
1733 		btrfs_csum_file_blocks(trans,
1734 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
1735 		trans->adding_csums = 0;
1736 	}
1737 	return 0;
1738 }
1739 
1740 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1741 			      struct extent_state **cached_state)
1742 {
1743 	WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1744 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1745 				   cached_state, GFP_NOFS);
1746 }
1747 
1748 /* see btrfs_writepage_start_hook for details on why this is required */
1749 struct btrfs_writepage_fixup {
1750 	struct page *page;
1751 	struct btrfs_work work;
1752 };
1753 
1754 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1755 {
1756 	struct btrfs_writepage_fixup *fixup;
1757 	struct btrfs_ordered_extent *ordered;
1758 	struct extent_state *cached_state = NULL;
1759 	struct page *page;
1760 	struct inode *inode;
1761 	u64 page_start;
1762 	u64 page_end;
1763 	int ret;
1764 
1765 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
1766 	page = fixup->page;
1767 again:
1768 	lock_page(page);
1769 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1770 		ClearPageChecked(page);
1771 		goto out_page;
1772 	}
1773 
1774 	inode = page->mapping->host;
1775 	page_start = page_offset(page);
1776 	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1777 
1778 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1779 			 &cached_state);
1780 
1781 	/* already ordered? We're done */
1782 	if (PagePrivate2(page))
1783 		goto out;
1784 
1785 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
1786 	if (ordered) {
1787 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1788 				     page_end, &cached_state, GFP_NOFS);
1789 		unlock_page(page);
1790 		btrfs_start_ordered_extent(inode, ordered, 1);
1791 		btrfs_put_ordered_extent(ordered);
1792 		goto again;
1793 	}
1794 
1795 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1796 	if (ret) {
1797 		mapping_set_error(page->mapping, ret);
1798 		end_extent_writepage(page, ret, page_start, page_end);
1799 		ClearPageChecked(page);
1800 		goto out;
1801 	 }
1802 
1803 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1804 	ClearPageChecked(page);
1805 	set_page_dirty(page);
1806 out:
1807 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1808 			     &cached_state, GFP_NOFS);
1809 out_page:
1810 	unlock_page(page);
1811 	page_cache_release(page);
1812 	kfree(fixup);
1813 }
1814 
1815 /*
1816  * There are a few paths in the higher layers of the kernel that directly
1817  * set the page dirty bit without asking the filesystem if it is a
1818  * good idea.  This causes problems because we want to make sure COW
1819  * properly happens and the data=ordered rules are followed.
1820  *
1821  * In our case any range that doesn't have the ORDERED bit set
1822  * hasn't been properly setup for IO.  We kick off an async process
1823  * to fix it up.  The async helper will wait for ordered extents, set
1824  * the delalloc bit and make it safe to write the page.
1825  */
1826 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1827 {
1828 	struct inode *inode = page->mapping->host;
1829 	struct btrfs_writepage_fixup *fixup;
1830 	struct btrfs_root *root = BTRFS_I(inode)->root;
1831 
1832 	/* this page is properly in the ordered list */
1833 	if (TestClearPagePrivate2(page))
1834 		return 0;
1835 
1836 	if (PageChecked(page))
1837 		return -EAGAIN;
1838 
1839 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1840 	if (!fixup)
1841 		return -EAGAIN;
1842 
1843 	SetPageChecked(page);
1844 	page_cache_get(page);
1845 	btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1846 	fixup->page = page;
1847 	btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1848 	return -EBUSY;
1849 }
1850 
1851 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1852 				       struct inode *inode, u64 file_pos,
1853 				       u64 disk_bytenr, u64 disk_num_bytes,
1854 				       u64 num_bytes, u64 ram_bytes,
1855 				       u8 compression, u8 encryption,
1856 				       u16 other_encoding, int extent_type)
1857 {
1858 	struct btrfs_root *root = BTRFS_I(inode)->root;
1859 	struct btrfs_file_extent_item *fi;
1860 	struct btrfs_path *path;
1861 	struct extent_buffer *leaf;
1862 	struct btrfs_key ins;
1863 	int extent_inserted = 0;
1864 	int ret;
1865 
1866 	path = btrfs_alloc_path();
1867 	if (!path)
1868 		return -ENOMEM;
1869 
1870 	/*
1871 	 * we may be replacing one extent in the tree with another.
1872 	 * The new extent is pinned in the extent map, and we don't want
1873 	 * to drop it from the cache until it is completely in the btree.
1874 	 *
1875 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
1876 	 * the caller is expected to unpin it and allow it to be merged
1877 	 * with the others.
1878 	 */
1879 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1880 				   file_pos + num_bytes, NULL, 0,
1881 				   1, sizeof(*fi), &extent_inserted);
1882 	if (ret)
1883 		goto out;
1884 
1885 	if (!extent_inserted) {
1886 		ins.objectid = btrfs_ino(inode);
1887 		ins.offset = file_pos;
1888 		ins.type = BTRFS_EXTENT_DATA_KEY;
1889 
1890 		path->leave_spinning = 1;
1891 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
1892 					      sizeof(*fi));
1893 		if (ret)
1894 			goto out;
1895 	}
1896 	leaf = path->nodes[0];
1897 	fi = btrfs_item_ptr(leaf, path->slots[0],
1898 			    struct btrfs_file_extent_item);
1899 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1900 	btrfs_set_file_extent_type(leaf, fi, extent_type);
1901 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1902 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1903 	btrfs_set_file_extent_offset(leaf, fi, 0);
1904 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1905 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1906 	btrfs_set_file_extent_compression(leaf, fi, compression);
1907 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
1908 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1909 
1910 	btrfs_mark_buffer_dirty(leaf);
1911 	btrfs_release_path(path);
1912 
1913 	inode_add_bytes(inode, num_bytes);
1914 
1915 	ins.objectid = disk_bytenr;
1916 	ins.offset = disk_num_bytes;
1917 	ins.type = BTRFS_EXTENT_ITEM_KEY;
1918 	ret = btrfs_alloc_reserved_file_extent(trans, root,
1919 					root->root_key.objectid,
1920 					btrfs_ino(inode), file_pos, &ins);
1921 out:
1922 	btrfs_free_path(path);
1923 
1924 	return ret;
1925 }
1926 
1927 /* snapshot-aware defrag */
1928 struct sa_defrag_extent_backref {
1929 	struct rb_node node;
1930 	struct old_sa_defrag_extent *old;
1931 	u64 root_id;
1932 	u64 inum;
1933 	u64 file_pos;
1934 	u64 extent_offset;
1935 	u64 num_bytes;
1936 	u64 generation;
1937 };
1938 
1939 struct old_sa_defrag_extent {
1940 	struct list_head list;
1941 	struct new_sa_defrag_extent *new;
1942 
1943 	u64 extent_offset;
1944 	u64 bytenr;
1945 	u64 offset;
1946 	u64 len;
1947 	int count;
1948 };
1949 
1950 struct new_sa_defrag_extent {
1951 	struct rb_root root;
1952 	struct list_head head;
1953 	struct btrfs_path *path;
1954 	struct inode *inode;
1955 	u64 file_pos;
1956 	u64 len;
1957 	u64 bytenr;
1958 	u64 disk_len;
1959 	u8 compress_type;
1960 };
1961 
1962 static int backref_comp(struct sa_defrag_extent_backref *b1,
1963 			struct sa_defrag_extent_backref *b2)
1964 {
1965 	if (b1->root_id < b2->root_id)
1966 		return -1;
1967 	else if (b1->root_id > b2->root_id)
1968 		return 1;
1969 
1970 	if (b1->inum < b2->inum)
1971 		return -1;
1972 	else if (b1->inum > b2->inum)
1973 		return 1;
1974 
1975 	if (b1->file_pos < b2->file_pos)
1976 		return -1;
1977 	else if (b1->file_pos > b2->file_pos)
1978 		return 1;
1979 
1980 	/*
1981 	 * [------------------------------] ===> (a range of space)
1982 	 *     |<--->|   |<---->| =============> (fs/file tree A)
1983 	 * |<---------------------------->| ===> (fs/file tree B)
1984 	 *
1985 	 * A range of space can refer to two file extents in one tree while
1986 	 * refer to only one file extent in another tree.
1987 	 *
1988 	 * So we may process a disk offset more than one time(two extents in A)
1989 	 * and locate at the same extent(one extent in B), then insert two same
1990 	 * backrefs(both refer to the extent in B).
1991 	 */
1992 	return 0;
1993 }
1994 
1995 static void backref_insert(struct rb_root *root,
1996 			   struct sa_defrag_extent_backref *backref)
1997 {
1998 	struct rb_node **p = &root->rb_node;
1999 	struct rb_node *parent = NULL;
2000 	struct sa_defrag_extent_backref *entry;
2001 	int ret;
2002 
2003 	while (*p) {
2004 		parent = *p;
2005 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2006 
2007 		ret = backref_comp(backref, entry);
2008 		if (ret < 0)
2009 			p = &(*p)->rb_left;
2010 		else
2011 			p = &(*p)->rb_right;
2012 	}
2013 
2014 	rb_link_node(&backref->node, parent, p);
2015 	rb_insert_color(&backref->node, root);
2016 }
2017 
2018 /*
2019  * Note the backref might has changed, and in this case we just return 0.
2020  */
2021 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2022 				       void *ctx)
2023 {
2024 	struct btrfs_file_extent_item *extent;
2025 	struct btrfs_fs_info *fs_info;
2026 	struct old_sa_defrag_extent *old = ctx;
2027 	struct new_sa_defrag_extent *new = old->new;
2028 	struct btrfs_path *path = new->path;
2029 	struct btrfs_key key;
2030 	struct btrfs_root *root;
2031 	struct sa_defrag_extent_backref *backref;
2032 	struct extent_buffer *leaf;
2033 	struct inode *inode = new->inode;
2034 	int slot;
2035 	int ret;
2036 	u64 extent_offset;
2037 	u64 num_bytes;
2038 
2039 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2040 	    inum == btrfs_ino(inode))
2041 		return 0;
2042 
2043 	key.objectid = root_id;
2044 	key.type = BTRFS_ROOT_ITEM_KEY;
2045 	key.offset = (u64)-1;
2046 
2047 	fs_info = BTRFS_I(inode)->root->fs_info;
2048 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2049 	if (IS_ERR(root)) {
2050 		if (PTR_ERR(root) == -ENOENT)
2051 			return 0;
2052 		WARN_ON(1);
2053 		pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2054 			 inum, offset, root_id);
2055 		return PTR_ERR(root);
2056 	}
2057 
2058 	key.objectid = inum;
2059 	key.type = BTRFS_EXTENT_DATA_KEY;
2060 	if (offset > (u64)-1 << 32)
2061 		key.offset = 0;
2062 	else
2063 		key.offset = offset;
2064 
2065 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066 	if (WARN_ON(ret < 0))
2067 		return ret;
2068 	ret = 0;
2069 
2070 	while (1) {
2071 		cond_resched();
2072 
2073 		leaf = path->nodes[0];
2074 		slot = path->slots[0];
2075 
2076 		if (slot >= btrfs_header_nritems(leaf)) {
2077 			ret = btrfs_next_leaf(root, path);
2078 			if (ret < 0) {
2079 				goto out;
2080 			} else if (ret > 0) {
2081 				ret = 0;
2082 				goto out;
2083 			}
2084 			continue;
2085 		}
2086 
2087 		path->slots[0]++;
2088 
2089 		btrfs_item_key_to_cpu(leaf, &key, slot);
2090 
2091 		if (key.objectid > inum)
2092 			goto out;
2093 
2094 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2095 			continue;
2096 
2097 		extent = btrfs_item_ptr(leaf, slot,
2098 					struct btrfs_file_extent_item);
2099 
2100 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2101 			continue;
2102 
2103 		/*
2104 		 * 'offset' refers to the exact key.offset,
2105 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2106 		 * (key.offset - extent_offset).
2107 		 */
2108 		if (key.offset != offset)
2109 			continue;
2110 
2111 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2112 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2113 
2114 		if (extent_offset >= old->extent_offset + old->offset +
2115 		    old->len || extent_offset + num_bytes <=
2116 		    old->extent_offset + old->offset)
2117 			continue;
2118 		break;
2119 	}
2120 
2121 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2122 	if (!backref) {
2123 		ret = -ENOENT;
2124 		goto out;
2125 	}
2126 
2127 	backref->root_id = root_id;
2128 	backref->inum = inum;
2129 	backref->file_pos = offset;
2130 	backref->num_bytes = num_bytes;
2131 	backref->extent_offset = extent_offset;
2132 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2133 	backref->old = old;
2134 	backref_insert(&new->root, backref);
2135 	old->count++;
2136 out:
2137 	btrfs_release_path(path);
2138 	WARN_ON(ret);
2139 	return ret;
2140 }
2141 
2142 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2143 				   struct new_sa_defrag_extent *new)
2144 {
2145 	struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2146 	struct old_sa_defrag_extent *old, *tmp;
2147 	int ret;
2148 
2149 	new->path = path;
2150 
2151 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2152 		ret = iterate_inodes_from_logical(old->bytenr +
2153 						  old->extent_offset, fs_info,
2154 						  path, record_one_backref,
2155 						  old);
2156 		if (ret < 0 && ret != -ENOENT)
2157 			return false;
2158 
2159 		/* no backref to be processed for this extent */
2160 		if (!old->count) {
2161 			list_del(&old->list);
2162 			kfree(old);
2163 		}
2164 	}
2165 
2166 	if (list_empty(&new->head))
2167 		return false;
2168 
2169 	return true;
2170 }
2171 
2172 static int relink_is_mergable(struct extent_buffer *leaf,
2173 			      struct btrfs_file_extent_item *fi,
2174 			      struct new_sa_defrag_extent *new)
2175 {
2176 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2177 		return 0;
2178 
2179 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2180 		return 0;
2181 
2182 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2183 		return 0;
2184 
2185 	if (btrfs_file_extent_encryption(leaf, fi) ||
2186 	    btrfs_file_extent_other_encoding(leaf, fi))
2187 		return 0;
2188 
2189 	return 1;
2190 }
2191 
2192 /*
2193  * Note the backref might has changed, and in this case we just return 0.
2194  */
2195 static noinline int relink_extent_backref(struct btrfs_path *path,
2196 				 struct sa_defrag_extent_backref *prev,
2197 				 struct sa_defrag_extent_backref *backref)
2198 {
2199 	struct btrfs_file_extent_item *extent;
2200 	struct btrfs_file_extent_item *item;
2201 	struct btrfs_ordered_extent *ordered;
2202 	struct btrfs_trans_handle *trans;
2203 	struct btrfs_fs_info *fs_info;
2204 	struct btrfs_root *root;
2205 	struct btrfs_key key;
2206 	struct extent_buffer *leaf;
2207 	struct old_sa_defrag_extent *old = backref->old;
2208 	struct new_sa_defrag_extent *new = old->new;
2209 	struct inode *src_inode = new->inode;
2210 	struct inode *inode;
2211 	struct extent_state *cached = NULL;
2212 	int ret = 0;
2213 	u64 start;
2214 	u64 len;
2215 	u64 lock_start;
2216 	u64 lock_end;
2217 	bool merge = false;
2218 	int index;
2219 
2220 	if (prev && prev->root_id == backref->root_id &&
2221 	    prev->inum == backref->inum &&
2222 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2223 		merge = true;
2224 
2225 	/* step 1: get root */
2226 	key.objectid = backref->root_id;
2227 	key.type = BTRFS_ROOT_ITEM_KEY;
2228 	key.offset = (u64)-1;
2229 
2230 	fs_info = BTRFS_I(src_inode)->root->fs_info;
2231 	index = srcu_read_lock(&fs_info->subvol_srcu);
2232 
2233 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2234 	if (IS_ERR(root)) {
2235 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2236 		if (PTR_ERR(root) == -ENOENT)
2237 			return 0;
2238 		return PTR_ERR(root);
2239 	}
2240 
2241 	if (btrfs_root_readonly(root)) {
2242 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2243 		return 0;
2244 	}
2245 
2246 	/* step 2: get inode */
2247 	key.objectid = backref->inum;
2248 	key.type = BTRFS_INODE_ITEM_KEY;
2249 	key.offset = 0;
2250 
2251 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2252 	if (IS_ERR(inode)) {
2253 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2254 		return 0;
2255 	}
2256 
2257 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2258 
2259 	/* step 3: relink backref */
2260 	lock_start = backref->file_pos;
2261 	lock_end = backref->file_pos + backref->num_bytes - 1;
2262 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2263 			 0, &cached);
2264 
2265 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2266 	if (ordered) {
2267 		btrfs_put_ordered_extent(ordered);
2268 		goto out_unlock;
2269 	}
2270 
2271 	trans = btrfs_join_transaction(root);
2272 	if (IS_ERR(trans)) {
2273 		ret = PTR_ERR(trans);
2274 		goto out_unlock;
2275 	}
2276 
2277 	key.objectid = backref->inum;
2278 	key.type = BTRFS_EXTENT_DATA_KEY;
2279 	key.offset = backref->file_pos;
2280 
2281 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2282 	if (ret < 0) {
2283 		goto out_free_path;
2284 	} else if (ret > 0) {
2285 		ret = 0;
2286 		goto out_free_path;
2287 	}
2288 
2289 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2290 				struct btrfs_file_extent_item);
2291 
2292 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2293 	    backref->generation)
2294 		goto out_free_path;
2295 
2296 	btrfs_release_path(path);
2297 
2298 	start = backref->file_pos;
2299 	if (backref->extent_offset < old->extent_offset + old->offset)
2300 		start += old->extent_offset + old->offset -
2301 			 backref->extent_offset;
2302 
2303 	len = min(backref->extent_offset + backref->num_bytes,
2304 		  old->extent_offset + old->offset + old->len);
2305 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2306 
2307 	ret = btrfs_drop_extents(trans, root, inode, start,
2308 				 start + len, 1);
2309 	if (ret)
2310 		goto out_free_path;
2311 again:
2312 	key.objectid = btrfs_ino(inode);
2313 	key.type = BTRFS_EXTENT_DATA_KEY;
2314 	key.offset = start;
2315 
2316 	path->leave_spinning = 1;
2317 	if (merge) {
2318 		struct btrfs_file_extent_item *fi;
2319 		u64 extent_len;
2320 		struct btrfs_key found_key;
2321 
2322 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2323 		if (ret < 0)
2324 			goto out_free_path;
2325 
2326 		path->slots[0]--;
2327 		leaf = path->nodes[0];
2328 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2329 
2330 		fi = btrfs_item_ptr(leaf, path->slots[0],
2331 				    struct btrfs_file_extent_item);
2332 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2333 
2334 		if (extent_len + found_key.offset == start &&
2335 		    relink_is_mergable(leaf, fi, new)) {
2336 			btrfs_set_file_extent_num_bytes(leaf, fi,
2337 							extent_len + len);
2338 			btrfs_mark_buffer_dirty(leaf);
2339 			inode_add_bytes(inode, len);
2340 
2341 			ret = 1;
2342 			goto out_free_path;
2343 		} else {
2344 			merge = false;
2345 			btrfs_release_path(path);
2346 			goto again;
2347 		}
2348 	}
2349 
2350 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2351 					sizeof(*extent));
2352 	if (ret) {
2353 		btrfs_abort_transaction(trans, root, ret);
2354 		goto out_free_path;
2355 	}
2356 
2357 	leaf = path->nodes[0];
2358 	item = btrfs_item_ptr(leaf, path->slots[0],
2359 				struct btrfs_file_extent_item);
2360 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2361 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2362 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2363 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2364 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2365 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2366 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2367 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2368 	btrfs_set_file_extent_encryption(leaf, item, 0);
2369 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2370 
2371 	btrfs_mark_buffer_dirty(leaf);
2372 	inode_add_bytes(inode, len);
2373 	btrfs_release_path(path);
2374 
2375 	ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2376 			new->disk_len, 0,
2377 			backref->root_id, backref->inum,
2378 			new->file_pos, 0);	/* start - extent_offset */
2379 	if (ret) {
2380 		btrfs_abort_transaction(trans, root, ret);
2381 		goto out_free_path;
2382 	}
2383 
2384 	ret = 1;
2385 out_free_path:
2386 	btrfs_release_path(path);
2387 	path->leave_spinning = 0;
2388 	btrfs_end_transaction(trans, root);
2389 out_unlock:
2390 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2391 			     &cached, GFP_NOFS);
2392 	iput(inode);
2393 	return ret;
2394 }
2395 
2396 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2397 {
2398 	struct old_sa_defrag_extent *old, *tmp;
2399 
2400 	if (!new)
2401 		return;
2402 
2403 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2404 		list_del(&old->list);
2405 		kfree(old);
2406 	}
2407 	kfree(new);
2408 }
2409 
2410 static void relink_file_extents(struct new_sa_defrag_extent *new)
2411 {
2412 	struct btrfs_path *path;
2413 	struct sa_defrag_extent_backref *backref;
2414 	struct sa_defrag_extent_backref *prev = NULL;
2415 	struct inode *inode;
2416 	struct btrfs_root *root;
2417 	struct rb_node *node;
2418 	int ret;
2419 
2420 	inode = new->inode;
2421 	root = BTRFS_I(inode)->root;
2422 
2423 	path = btrfs_alloc_path();
2424 	if (!path)
2425 		return;
2426 
2427 	if (!record_extent_backrefs(path, new)) {
2428 		btrfs_free_path(path);
2429 		goto out;
2430 	}
2431 	btrfs_release_path(path);
2432 
2433 	while (1) {
2434 		node = rb_first(&new->root);
2435 		if (!node)
2436 			break;
2437 		rb_erase(node, &new->root);
2438 
2439 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2440 
2441 		ret = relink_extent_backref(path, prev, backref);
2442 		WARN_ON(ret < 0);
2443 
2444 		kfree(prev);
2445 
2446 		if (ret == 1)
2447 			prev = backref;
2448 		else
2449 			prev = NULL;
2450 		cond_resched();
2451 	}
2452 	kfree(prev);
2453 
2454 	btrfs_free_path(path);
2455 out:
2456 	free_sa_defrag_extent(new);
2457 
2458 	atomic_dec(&root->fs_info->defrag_running);
2459 	wake_up(&root->fs_info->transaction_wait);
2460 }
2461 
2462 static struct new_sa_defrag_extent *
2463 record_old_file_extents(struct inode *inode,
2464 			struct btrfs_ordered_extent *ordered)
2465 {
2466 	struct btrfs_root *root = BTRFS_I(inode)->root;
2467 	struct btrfs_path *path;
2468 	struct btrfs_key key;
2469 	struct old_sa_defrag_extent *old;
2470 	struct new_sa_defrag_extent *new;
2471 	int ret;
2472 
2473 	new = kmalloc(sizeof(*new), GFP_NOFS);
2474 	if (!new)
2475 		return NULL;
2476 
2477 	new->inode = inode;
2478 	new->file_pos = ordered->file_offset;
2479 	new->len = ordered->len;
2480 	new->bytenr = ordered->start;
2481 	new->disk_len = ordered->disk_len;
2482 	new->compress_type = ordered->compress_type;
2483 	new->root = RB_ROOT;
2484 	INIT_LIST_HEAD(&new->head);
2485 
2486 	path = btrfs_alloc_path();
2487 	if (!path)
2488 		goto out_kfree;
2489 
2490 	key.objectid = btrfs_ino(inode);
2491 	key.type = BTRFS_EXTENT_DATA_KEY;
2492 	key.offset = new->file_pos;
2493 
2494 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2495 	if (ret < 0)
2496 		goto out_free_path;
2497 	if (ret > 0 && path->slots[0] > 0)
2498 		path->slots[0]--;
2499 
2500 	/* find out all the old extents for the file range */
2501 	while (1) {
2502 		struct btrfs_file_extent_item *extent;
2503 		struct extent_buffer *l;
2504 		int slot;
2505 		u64 num_bytes;
2506 		u64 offset;
2507 		u64 end;
2508 		u64 disk_bytenr;
2509 		u64 extent_offset;
2510 
2511 		l = path->nodes[0];
2512 		slot = path->slots[0];
2513 
2514 		if (slot >= btrfs_header_nritems(l)) {
2515 			ret = btrfs_next_leaf(root, path);
2516 			if (ret < 0)
2517 				goto out_free_path;
2518 			else if (ret > 0)
2519 				break;
2520 			continue;
2521 		}
2522 
2523 		btrfs_item_key_to_cpu(l, &key, slot);
2524 
2525 		if (key.objectid != btrfs_ino(inode))
2526 			break;
2527 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2528 			break;
2529 		if (key.offset >= new->file_pos + new->len)
2530 			break;
2531 
2532 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2533 
2534 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2535 		if (key.offset + num_bytes < new->file_pos)
2536 			goto next;
2537 
2538 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2539 		if (!disk_bytenr)
2540 			goto next;
2541 
2542 		extent_offset = btrfs_file_extent_offset(l, extent);
2543 
2544 		old = kmalloc(sizeof(*old), GFP_NOFS);
2545 		if (!old)
2546 			goto out_free_path;
2547 
2548 		offset = max(new->file_pos, key.offset);
2549 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2550 
2551 		old->bytenr = disk_bytenr;
2552 		old->extent_offset = extent_offset;
2553 		old->offset = offset - key.offset;
2554 		old->len = end - offset;
2555 		old->new = new;
2556 		old->count = 0;
2557 		list_add_tail(&old->list, &new->head);
2558 next:
2559 		path->slots[0]++;
2560 		cond_resched();
2561 	}
2562 
2563 	btrfs_free_path(path);
2564 	atomic_inc(&root->fs_info->defrag_running);
2565 
2566 	return new;
2567 
2568 out_free_path:
2569 	btrfs_free_path(path);
2570 out_kfree:
2571 	free_sa_defrag_extent(new);
2572 	return NULL;
2573 }
2574 
2575 /* as ordered data IO finishes, this gets called so we can finish
2576  * an ordered extent if the range of bytes in the file it covers are
2577  * fully written.
2578  */
2579 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2580 {
2581 	struct inode *inode = ordered_extent->inode;
2582 	struct btrfs_root *root = BTRFS_I(inode)->root;
2583 	struct btrfs_trans_handle *trans = NULL;
2584 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2585 	struct extent_state *cached_state = NULL;
2586 	struct new_sa_defrag_extent *new = NULL;
2587 	int compress_type = 0;
2588 	int ret = 0;
2589 	u64 logical_len = ordered_extent->len;
2590 	bool nolock;
2591 	bool truncated = false;
2592 
2593 	nolock = btrfs_is_free_space_inode(inode);
2594 
2595 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2596 		ret = -EIO;
2597 		goto out;
2598 	}
2599 
2600 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2601 		truncated = true;
2602 		logical_len = ordered_extent->truncated_len;
2603 		/* Truncated the entire extent, don't bother adding */
2604 		if (!logical_len)
2605 			goto out;
2606 	}
2607 
2608 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2609 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2610 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2611 		if (nolock)
2612 			trans = btrfs_join_transaction_nolock(root);
2613 		else
2614 			trans = btrfs_join_transaction(root);
2615 		if (IS_ERR(trans)) {
2616 			ret = PTR_ERR(trans);
2617 			trans = NULL;
2618 			goto out;
2619 		}
2620 		trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2621 		ret = btrfs_update_inode_fallback(trans, root, inode);
2622 		if (ret) /* -ENOMEM or corruption */
2623 			btrfs_abort_transaction(trans, root, ret);
2624 		goto out;
2625 	}
2626 
2627 	lock_extent_bits(io_tree, ordered_extent->file_offset,
2628 			 ordered_extent->file_offset + ordered_extent->len - 1,
2629 			 0, &cached_state);
2630 
2631 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
2632 			ordered_extent->file_offset + ordered_extent->len - 1,
2633 			EXTENT_DEFRAG, 1, cached_state);
2634 	if (ret) {
2635 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2636 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2637 			/* the inode is shared */
2638 			new = record_old_file_extents(inode, ordered_extent);
2639 
2640 		clear_extent_bit(io_tree, ordered_extent->file_offset,
2641 			ordered_extent->file_offset + ordered_extent->len - 1,
2642 			EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2643 	}
2644 
2645 	if (nolock)
2646 		trans = btrfs_join_transaction_nolock(root);
2647 	else
2648 		trans = btrfs_join_transaction(root);
2649 	if (IS_ERR(trans)) {
2650 		ret = PTR_ERR(trans);
2651 		trans = NULL;
2652 		goto out_unlock;
2653 	}
2654 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2655 
2656 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2657 		compress_type = ordered_extent->compress_type;
2658 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2659 		BUG_ON(compress_type);
2660 		ret = btrfs_mark_extent_written(trans, inode,
2661 						ordered_extent->file_offset,
2662 						ordered_extent->file_offset +
2663 						logical_len);
2664 	} else {
2665 		BUG_ON(root == root->fs_info->tree_root);
2666 		ret = insert_reserved_file_extent(trans, inode,
2667 						ordered_extent->file_offset,
2668 						ordered_extent->start,
2669 						ordered_extent->disk_len,
2670 						logical_len, logical_len,
2671 						compress_type, 0, 0,
2672 						BTRFS_FILE_EXTENT_REG);
2673 	}
2674 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2675 			   ordered_extent->file_offset, ordered_extent->len,
2676 			   trans->transid);
2677 	if (ret < 0) {
2678 		btrfs_abort_transaction(trans, root, ret);
2679 		goto out_unlock;
2680 	}
2681 
2682 	add_pending_csums(trans, inode, ordered_extent->file_offset,
2683 			  &ordered_extent->list);
2684 
2685 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2686 	ret = btrfs_update_inode_fallback(trans, root, inode);
2687 	if (ret) { /* -ENOMEM or corruption */
2688 		btrfs_abort_transaction(trans, root, ret);
2689 		goto out_unlock;
2690 	}
2691 	ret = 0;
2692 out_unlock:
2693 	unlock_extent_cached(io_tree, ordered_extent->file_offset,
2694 			     ordered_extent->file_offset +
2695 			     ordered_extent->len - 1, &cached_state, GFP_NOFS);
2696 out:
2697 	if (root != root->fs_info->tree_root)
2698 		btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2699 	if (trans)
2700 		btrfs_end_transaction(trans, root);
2701 
2702 	if (ret || truncated) {
2703 		u64 start, end;
2704 
2705 		if (truncated)
2706 			start = ordered_extent->file_offset + logical_len;
2707 		else
2708 			start = ordered_extent->file_offset;
2709 		end = ordered_extent->file_offset + ordered_extent->len - 1;
2710 		clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2711 
2712 		/* Drop the cache for the part of the extent we didn't write. */
2713 		btrfs_drop_extent_cache(inode, start, end, 0);
2714 
2715 		/*
2716 		 * If the ordered extent had an IOERR or something else went
2717 		 * wrong we need to return the space for this ordered extent
2718 		 * back to the allocator.  We only free the extent in the
2719 		 * truncated case if we didn't write out the extent at all.
2720 		 */
2721 		if ((ret || !logical_len) &&
2722 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2723 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2724 			btrfs_free_reserved_extent(root, ordered_extent->start,
2725 						   ordered_extent->disk_len);
2726 	}
2727 
2728 
2729 	/*
2730 	 * This needs to be done to make sure anybody waiting knows we are done
2731 	 * updating everything for this ordered extent.
2732 	 */
2733 	btrfs_remove_ordered_extent(inode, ordered_extent);
2734 
2735 	/* for snapshot-aware defrag */
2736 	if (new) {
2737 		if (ret) {
2738 			free_sa_defrag_extent(new);
2739 			atomic_dec(&root->fs_info->defrag_running);
2740 		} else {
2741 			relink_file_extents(new);
2742 		}
2743 	}
2744 
2745 	/* once for us */
2746 	btrfs_put_ordered_extent(ordered_extent);
2747 	/* once for the tree */
2748 	btrfs_put_ordered_extent(ordered_extent);
2749 
2750 	return ret;
2751 }
2752 
2753 static void finish_ordered_fn(struct btrfs_work *work)
2754 {
2755 	struct btrfs_ordered_extent *ordered_extent;
2756 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2757 	btrfs_finish_ordered_io(ordered_extent);
2758 }
2759 
2760 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2761 				struct extent_state *state, int uptodate)
2762 {
2763 	struct inode *inode = page->mapping->host;
2764 	struct btrfs_root *root = BTRFS_I(inode)->root;
2765 	struct btrfs_ordered_extent *ordered_extent = NULL;
2766 	struct btrfs_workqueue *workers;
2767 
2768 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2769 
2770 	ClearPagePrivate2(page);
2771 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2772 					    end - start + 1, uptodate))
2773 		return 0;
2774 
2775 	btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2776 
2777 	if (btrfs_is_free_space_inode(inode))
2778 		workers = root->fs_info->endio_freespace_worker;
2779 	else
2780 		workers = root->fs_info->endio_write_workers;
2781 	btrfs_queue_work(workers, &ordered_extent->work);
2782 
2783 	return 0;
2784 }
2785 
2786 /*
2787  * when reads are done, we need to check csums to verify the data is correct
2788  * if there's a match, we allow the bio to finish.  If not, the code in
2789  * extent_io.c will try to find good copies for us.
2790  */
2791 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2792 				      u64 phy_offset, struct page *page,
2793 				      u64 start, u64 end, int mirror)
2794 {
2795 	size_t offset = start - page_offset(page);
2796 	struct inode *inode = page->mapping->host;
2797 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2798 	char *kaddr;
2799 	struct btrfs_root *root = BTRFS_I(inode)->root;
2800 	u32 csum_expected;
2801 	u32 csum = ~(u32)0;
2802 	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2803 	                              DEFAULT_RATELIMIT_BURST);
2804 
2805 	if (PageChecked(page)) {
2806 		ClearPageChecked(page);
2807 		goto good;
2808 	}
2809 
2810 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2811 		goto good;
2812 
2813 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2814 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2815 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2816 				  GFP_NOFS);
2817 		return 0;
2818 	}
2819 
2820 	phy_offset >>= inode->i_sb->s_blocksize_bits;
2821 	csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2822 
2823 	kaddr = kmap_atomic(page);
2824 	csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2825 	btrfs_csum_final(csum, (char *)&csum);
2826 	if (csum != csum_expected)
2827 		goto zeroit;
2828 
2829 	kunmap_atomic(kaddr);
2830 good:
2831 	return 0;
2832 
2833 zeroit:
2834 	if (__ratelimit(&_rs))
2835 		btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2836 			btrfs_ino(page->mapping->host), start, csum, csum_expected);
2837 	memset(kaddr + offset, 1, end - start + 1);
2838 	flush_dcache_page(page);
2839 	kunmap_atomic(kaddr);
2840 	if (csum_expected == 0)
2841 		return 0;
2842 	return -EIO;
2843 }
2844 
2845 struct delayed_iput {
2846 	struct list_head list;
2847 	struct inode *inode;
2848 };
2849 
2850 /* JDM: If this is fs-wide, why can't we add a pointer to
2851  * btrfs_inode instead and avoid the allocation? */
2852 void btrfs_add_delayed_iput(struct inode *inode)
2853 {
2854 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2855 	struct delayed_iput *delayed;
2856 
2857 	if (atomic_add_unless(&inode->i_count, -1, 1))
2858 		return;
2859 
2860 	delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2861 	delayed->inode = inode;
2862 
2863 	spin_lock(&fs_info->delayed_iput_lock);
2864 	list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2865 	spin_unlock(&fs_info->delayed_iput_lock);
2866 }
2867 
2868 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2869 {
2870 	LIST_HEAD(list);
2871 	struct btrfs_fs_info *fs_info = root->fs_info;
2872 	struct delayed_iput *delayed;
2873 	int empty;
2874 
2875 	spin_lock(&fs_info->delayed_iput_lock);
2876 	empty = list_empty(&fs_info->delayed_iputs);
2877 	spin_unlock(&fs_info->delayed_iput_lock);
2878 	if (empty)
2879 		return;
2880 
2881 	spin_lock(&fs_info->delayed_iput_lock);
2882 	list_splice_init(&fs_info->delayed_iputs, &list);
2883 	spin_unlock(&fs_info->delayed_iput_lock);
2884 
2885 	while (!list_empty(&list)) {
2886 		delayed = list_entry(list.next, struct delayed_iput, list);
2887 		list_del(&delayed->list);
2888 		iput(delayed->inode);
2889 		kfree(delayed);
2890 	}
2891 }
2892 
2893 /*
2894  * This is called in transaction commit time. If there are no orphan
2895  * files in the subvolume, it removes orphan item and frees block_rsv
2896  * structure.
2897  */
2898 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2899 			      struct btrfs_root *root)
2900 {
2901 	struct btrfs_block_rsv *block_rsv;
2902 	int ret;
2903 
2904 	if (atomic_read(&root->orphan_inodes) ||
2905 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2906 		return;
2907 
2908 	spin_lock(&root->orphan_lock);
2909 	if (atomic_read(&root->orphan_inodes)) {
2910 		spin_unlock(&root->orphan_lock);
2911 		return;
2912 	}
2913 
2914 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2915 		spin_unlock(&root->orphan_lock);
2916 		return;
2917 	}
2918 
2919 	block_rsv = root->orphan_block_rsv;
2920 	root->orphan_block_rsv = NULL;
2921 	spin_unlock(&root->orphan_lock);
2922 
2923 	if (root->orphan_item_inserted &&
2924 	    btrfs_root_refs(&root->root_item) > 0) {
2925 		ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2926 					    root->root_key.objectid);
2927 		if (ret)
2928 			btrfs_abort_transaction(trans, root, ret);
2929 		else
2930 			root->orphan_item_inserted = 0;
2931 	}
2932 
2933 	if (block_rsv) {
2934 		WARN_ON(block_rsv->size > 0);
2935 		btrfs_free_block_rsv(root, block_rsv);
2936 	}
2937 }
2938 
2939 /*
2940  * This creates an orphan entry for the given inode in case something goes
2941  * wrong in the middle of an unlink/truncate.
2942  *
2943  * NOTE: caller of this function should reserve 5 units of metadata for
2944  *	 this function.
2945  */
2946 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2947 {
2948 	struct btrfs_root *root = BTRFS_I(inode)->root;
2949 	struct btrfs_block_rsv *block_rsv = NULL;
2950 	int reserve = 0;
2951 	int insert = 0;
2952 	int ret;
2953 
2954 	if (!root->orphan_block_rsv) {
2955 		block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2956 		if (!block_rsv)
2957 			return -ENOMEM;
2958 	}
2959 
2960 	spin_lock(&root->orphan_lock);
2961 	if (!root->orphan_block_rsv) {
2962 		root->orphan_block_rsv = block_rsv;
2963 	} else if (block_rsv) {
2964 		btrfs_free_block_rsv(root, block_rsv);
2965 		block_rsv = NULL;
2966 	}
2967 
2968 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2969 			      &BTRFS_I(inode)->runtime_flags)) {
2970 #if 0
2971 		/*
2972 		 * For proper ENOSPC handling, we should do orphan
2973 		 * cleanup when mounting. But this introduces backward
2974 		 * compatibility issue.
2975 		 */
2976 		if (!xchg(&root->orphan_item_inserted, 1))
2977 			insert = 2;
2978 		else
2979 			insert = 1;
2980 #endif
2981 		insert = 1;
2982 		atomic_inc(&root->orphan_inodes);
2983 	}
2984 
2985 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2986 			      &BTRFS_I(inode)->runtime_flags))
2987 		reserve = 1;
2988 	spin_unlock(&root->orphan_lock);
2989 
2990 	/* grab metadata reservation from transaction handle */
2991 	if (reserve) {
2992 		ret = btrfs_orphan_reserve_metadata(trans, inode);
2993 		BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2994 	}
2995 
2996 	/* insert an orphan item to track this unlinked/truncated file */
2997 	if (insert >= 1) {
2998 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2999 		if (ret) {
3000 			atomic_dec(&root->orphan_inodes);
3001 			if (reserve) {
3002 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3003 					  &BTRFS_I(inode)->runtime_flags);
3004 				btrfs_orphan_release_metadata(inode);
3005 			}
3006 			if (ret != -EEXIST) {
3007 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3008 					  &BTRFS_I(inode)->runtime_flags);
3009 				btrfs_abort_transaction(trans, root, ret);
3010 				return ret;
3011 			}
3012 		}
3013 		ret = 0;
3014 	}
3015 
3016 	/* insert an orphan item to track subvolume contains orphan files */
3017 	if (insert >= 2) {
3018 		ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3019 					       root->root_key.objectid);
3020 		if (ret && ret != -EEXIST) {
3021 			btrfs_abort_transaction(trans, root, ret);
3022 			return ret;
3023 		}
3024 	}
3025 	return 0;
3026 }
3027 
3028 /*
3029  * We have done the truncate/delete so we can go ahead and remove the orphan
3030  * item for this particular inode.
3031  */
3032 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3033 			    struct inode *inode)
3034 {
3035 	struct btrfs_root *root = BTRFS_I(inode)->root;
3036 	int delete_item = 0;
3037 	int release_rsv = 0;
3038 	int ret = 0;
3039 
3040 	spin_lock(&root->orphan_lock);
3041 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3042 			       &BTRFS_I(inode)->runtime_flags))
3043 		delete_item = 1;
3044 
3045 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3046 			       &BTRFS_I(inode)->runtime_flags))
3047 		release_rsv = 1;
3048 	spin_unlock(&root->orphan_lock);
3049 
3050 	if (delete_item) {
3051 		atomic_dec(&root->orphan_inodes);
3052 		if (trans)
3053 			ret = btrfs_del_orphan_item(trans, root,
3054 						    btrfs_ino(inode));
3055 	}
3056 
3057 	if (release_rsv)
3058 		btrfs_orphan_release_metadata(inode);
3059 
3060 	return ret;
3061 }
3062 
3063 /*
3064  * this cleans up any orphans that may be left on the list from the last use
3065  * of this root.
3066  */
3067 int btrfs_orphan_cleanup(struct btrfs_root *root)
3068 {
3069 	struct btrfs_path *path;
3070 	struct extent_buffer *leaf;
3071 	struct btrfs_key key, found_key;
3072 	struct btrfs_trans_handle *trans;
3073 	struct inode *inode;
3074 	u64 last_objectid = 0;
3075 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3076 
3077 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3078 		return 0;
3079 
3080 	path = btrfs_alloc_path();
3081 	if (!path) {
3082 		ret = -ENOMEM;
3083 		goto out;
3084 	}
3085 	path->reada = -1;
3086 
3087 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3088 	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3089 	key.offset = (u64)-1;
3090 
3091 	while (1) {
3092 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3093 		if (ret < 0)
3094 			goto out;
3095 
3096 		/*
3097 		 * if ret == 0 means we found what we were searching for, which
3098 		 * is weird, but possible, so only screw with path if we didn't
3099 		 * find the key and see if we have stuff that matches
3100 		 */
3101 		if (ret > 0) {
3102 			ret = 0;
3103 			if (path->slots[0] == 0)
3104 				break;
3105 			path->slots[0]--;
3106 		}
3107 
3108 		/* pull out the item */
3109 		leaf = path->nodes[0];
3110 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3111 
3112 		/* make sure the item matches what we want */
3113 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3114 			break;
3115 		if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3116 			break;
3117 
3118 		/* release the path since we're done with it */
3119 		btrfs_release_path(path);
3120 
3121 		/*
3122 		 * this is where we are basically btrfs_lookup, without the
3123 		 * crossing root thing.  we store the inode number in the
3124 		 * offset of the orphan item.
3125 		 */
3126 
3127 		if (found_key.offset == last_objectid) {
3128 			btrfs_err(root->fs_info,
3129 				"Error removing orphan entry, stopping orphan cleanup");
3130 			ret = -EINVAL;
3131 			goto out;
3132 		}
3133 
3134 		last_objectid = found_key.offset;
3135 
3136 		found_key.objectid = found_key.offset;
3137 		found_key.type = BTRFS_INODE_ITEM_KEY;
3138 		found_key.offset = 0;
3139 		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3140 		ret = PTR_ERR_OR_ZERO(inode);
3141 		if (ret && ret != -ESTALE)
3142 			goto out;
3143 
3144 		if (ret == -ESTALE && root == root->fs_info->tree_root) {
3145 			struct btrfs_root *dead_root;
3146 			struct btrfs_fs_info *fs_info = root->fs_info;
3147 			int is_dead_root = 0;
3148 
3149 			/*
3150 			 * this is an orphan in the tree root. Currently these
3151 			 * could come from 2 sources:
3152 			 *  a) a snapshot deletion in progress
3153 			 *  b) a free space cache inode
3154 			 * We need to distinguish those two, as the snapshot
3155 			 * orphan must not get deleted.
3156 			 * find_dead_roots already ran before us, so if this
3157 			 * is a snapshot deletion, we should find the root
3158 			 * in the dead_roots list
3159 			 */
3160 			spin_lock(&fs_info->trans_lock);
3161 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3162 					    root_list) {
3163 				if (dead_root->root_key.objectid ==
3164 				    found_key.objectid) {
3165 					is_dead_root = 1;
3166 					break;
3167 				}
3168 			}
3169 			spin_unlock(&fs_info->trans_lock);
3170 			if (is_dead_root) {
3171 				/* prevent this orphan from being found again */
3172 				key.offset = found_key.objectid - 1;
3173 				continue;
3174 			}
3175 		}
3176 		/*
3177 		 * Inode is already gone but the orphan item is still there,
3178 		 * kill the orphan item.
3179 		 */
3180 		if (ret == -ESTALE) {
3181 			trans = btrfs_start_transaction(root, 1);
3182 			if (IS_ERR(trans)) {
3183 				ret = PTR_ERR(trans);
3184 				goto out;
3185 			}
3186 			btrfs_debug(root->fs_info, "auto deleting %Lu",
3187 				found_key.objectid);
3188 			ret = btrfs_del_orphan_item(trans, root,
3189 						    found_key.objectid);
3190 			btrfs_end_transaction(trans, root);
3191 			if (ret)
3192 				goto out;
3193 			continue;
3194 		}
3195 
3196 		/*
3197 		 * add this inode to the orphan list so btrfs_orphan_del does
3198 		 * the proper thing when we hit it
3199 		 */
3200 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3201 			&BTRFS_I(inode)->runtime_flags);
3202 		atomic_inc(&root->orphan_inodes);
3203 
3204 		/* if we have links, this was a truncate, lets do that */
3205 		if (inode->i_nlink) {
3206 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3207 				iput(inode);
3208 				continue;
3209 			}
3210 			nr_truncate++;
3211 
3212 			/* 1 for the orphan item deletion. */
3213 			trans = btrfs_start_transaction(root, 1);
3214 			if (IS_ERR(trans)) {
3215 				iput(inode);
3216 				ret = PTR_ERR(trans);
3217 				goto out;
3218 			}
3219 			ret = btrfs_orphan_add(trans, inode);
3220 			btrfs_end_transaction(trans, root);
3221 			if (ret) {
3222 				iput(inode);
3223 				goto out;
3224 			}
3225 
3226 			ret = btrfs_truncate(inode);
3227 			if (ret)
3228 				btrfs_orphan_del(NULL, inode);
3229 		} else {
3230 			nr_unlink++;
3231 		}
3232 
3233 		/* this will do delete_inode and everything for us */
3234 		iput(inode);
3235 		if (ret)
3236 			goto out;
3237 	}
3238 	/* release the path since we're done with it */
3239 	btrfs_release_path(path);
3240 
3241 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3242 
3243 	if (root->orphan_block_rsv)
3244 		btrfs_block_rsv_release(root, root->orphan_block_rsv,
3245 					(u64)-1);
3246 
3247 	if (root->orphan_block_rsv || root->orphan_item_inserted) {
3248 		trans = btrfs_join_transaction(root);
3249 		if (!IS_ERR(trans))
3250 			btrfs_end_transaction(trans, root);
3251 	}
3252 
3253 	if (nr_unlink)
3254 		btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3255 	if (nr_truncate)
3256 		btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3257 
3258 out:
3259 	if (ret)
3260 		btrfs_crit(root->fs_info,
3261 			"could not do orphan cleanup %d", ret);
3262 	btrfs_free_path(path);
3263 	return ret;
3264 }
3265 
3266 /*
3267  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3268  * don't find any xattrs, we know there can't be any acls.
3269  *
3270  * slot is the slot the inode is in, objectid is the objectid of the inode
3271  */
3272 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3273 					  int slot, u64 objectid,
3274 					  int *first_xattr_slot)
3275 {
3276 	u32 nritems = btrfs_header_nritems(leaf);
3277 	struct btrfs_key found_key;
3278 	static u64 xattr_access = 0;
3279 	static u64 xattr_default = 0;
3280 	int scanned = 0;
3281 
3282 	if (!xattr_access) {
3283 		xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3284 					strlen(POSIX_ACL_XATTR_ACCESS));
3285 		xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3286 					strlen(POSIX_ACL_XATTR_DEFAULT));
3287 	}
3288 
3289 	slot++;
3290 	*first_xattr_slot = -1;
3291 	while (slot < nritems) {
3292 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3293 
3294 		/* we found a different objectid, there must not be acls */
3295 		if (found_key.objectid != objectid)
3296 			return 0;
3297 
3298 		/* we found an xattr, assume we've got an acl */
3299 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3300 			if (*first_xattr_slot == -1)
3301 				*first_xattr_slot = slot;
3302 			if (found_key.offset == xattr_access ||
3303 			    found_key.offset == xattr_default)
3304 				return 1;
3305 		}
3306 
3307 		/*
3308 		 * we found a key greater than an xattr key, there can't
3309 		 * be any acls later on
3310 		 */
3311 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3312 			return 0;
3313 
3314 		slot++;
3315 		scanned++;
3316 
3317 		/*
3318 		 * it goes inode, inode backrefs, xattrs, extents,
3319 		 * so if there are a ton of hard links to an inode there can
3320 		 * be a lot of backrefs.  Don't waste time searching too hard,
3321 		 * this is just an optimization
3322 		 */
3323 		if (scanned >= 8)
3324 			break;
3325 	}
3326 	/* we hit the end of the leaf before we found an xattr or
3327 	 * something larger than an xattr.  We have to assume the inode
3328 	 * has acls
3329 	 */
3330 	if (*first_xattr_slot == -1)
3331 		*first_xattr_slot = slot;
3332 	return 1;
3333 }
3334 
3335 /*
3336  * read an inode from the btree into the in-memory inode
3337  */
3338 static void btrfs_read_locked_inode(struct inode *inode)
3339 {
3340 	struct btrfs_path *path;
3341 	struct extent_buffer *leaf;
3342 	struct btrfs_inode_item *inode_item;
3343 	struct btrfs_timespec *tspec;
3344 	struct btrfs_root *root = BTRFS_I(inode)->root;
3345 	struct btrfs_key location;
3346 	unsigned long ptr;
3347 	int maybe_acls;
3348 	u32 rdev;
3349 	int ret;
3350 	bool filled = false;
3351 	int first_xattr_slot;
3352 
3353 	ret = btrfs_fill_inode(inode, &rdev);
3354 	if (!ret)
3355 		filled = true;
3356 
3357 	path = btrfs_alloc_path();
3358 	if (!path)
3359 		goto make_bad;
3360 
3361 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3362 
3363 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3364 	if (ret)
3365 		goto make_bad;
3366 
3367 	leaf = path->nodes[0];
3368 
3369 	if (filled)
3370 		goto cache_index;
3371 
3372 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3373 				    struct btrfs_inode_item);
3374 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3375 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3376 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3377 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3378 	btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3379 
3380 	tspec = btrfs_inode_atime(inode_item);
3381 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383 
3384 	tspec = btrfs_inode_mtime(inode_item);
3385 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387 
3388 	tspec = btrfs_inode_ctime(inode_item);
3389 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3390 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391 
3392 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3393 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3394 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3395 
3396 	/*
3397 	 * If we were modified in the current generation and evicted from memory
3398 	 * and then re-read we need to do a full sync since we don't have any
3399 	 * idea about which extents were modified before we were evicted from
3400 	 * cache.
3401 	 */
3402 	if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3403 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3404 			&BTRFS_I(inode)->runtime_flags);
3405 
3406 	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3407 	inode->i_generation = BTRFS_I(inode)->generation;
3408 	inode->i_rdev = 0;
3409 	rdev = btrfs_inode_rdev(leaf, inode_item);
3410 
3411 	BTRFS_I(inode)->index_cnt = (u64)-1;
3412 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3413 
3414 cache_index:
3415 	path->slots[0]++;
3416 	if (inode->i_nlink != 1 ||
3417 	    path->slots[0] >= btrfs_header_nritems(leaf))
3418 		goto cache_acl;
3419 
3420 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3421 	if (location.objectid != btrfs_ino(inode))
3422 		goto cache_acl;
3423 
3424 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3425 	if (location.type == BTRFS_INODE_REF_KEY) {
3426 		struct btrfs_inode_ref *ref;
3427 
3428 		ref = (struct btrfs_inode_ref *)ptr;
3429 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3430 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3431 		struct btrfs_inode_extref *extref;
3432 
3433 		extref = (struct btrfs_inode_extref *)ptr;
3434 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3435 								     extref);
3436 	}
3437 cache_acl:
3438 	/*
3439 	 * try to precache a NULL acl entry for files that don't have
3440 	 * any xattrs or acls
3441 	 */
3442 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3443 					   btrfs_ino(inode), &first_xattr_slot);
3444 	if (first_xattr_slot != -1) {
3445 		path->slots[0] = first_xattr_slot;
3446 		ret = btrfs_load_inode_props(inode, path);
3447 		if (ret)
3448 			btrfs_err(root->fs_info,
3449 				  "error loading props for ino %llu (root %llu): %d\n",
3450 				  btrfs_ino(inode),
3451 				  root->root_key.objectid, ret);
3452 	}
3453 	btrfs_free_path(path);
3454 
3455 	if (!maybe_acls)
3456 		cache_no_acl(inode);
3457 
3458 	switch (inode->i_mode & S_IFMT) {
3459 	case S_IFREG:
3460 		inode->i_mapping->a_ops = &btrfs_aops;
3461 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3462 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3463 		inode->i_fop = &btrfs_file_operations;
3464 		inode->i_op = &btrfs_file_inode_operations;
3465 		break;
3466 	case S_IFDIR:
3467 		inode->i_fop = &btrfs_dir_file_operations;
3468 		if (root == root->fs_info->tree_root)
3469 			inode->i_op = &btrfs_dir_ro_inode_operations;
3470 		else
3471 			inode->i_op = &btrfs_dir_inode_operations;
3472 		break;
3473 	case S_IFLNK:
3474 		inode->i_op = &btrfs_symlink_inode_operations;
3475 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3476 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3477 		break;
3478 	default:
3479 		inode->i_op = &btrfs_special_inode_operations;
3480 		init_special_inode(inode, inode->i_mode, rdev);
3481 		break;
3482 	}
3483 
3484 	btrfs_update_iflags(inode);
3485 	return;
3486 
3487 make_bad:
3488 	btrfs_free_path(path);
3489 	make_bad_inode(inode);
3490 }
3491 
3492 /*
3493  * given a leaf and an inode, copy the inode fields into the leaf
3494  */
3495 static void fill_inode_item(struct btrfs_trans_handle *trans,
3496 			    struct extent_buffer *leaf,
3497 			    struct btrfs_inode_item *item,
3498 			    struct inode *inode)
3499 {
3500 	struct btrfs_map_token token;
3501 
3502 	btrfs_init_map_token(&token);
3503 
3504 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3505 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3506 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3507 				   &token);
3508 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3509 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3510 
3511 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3512 				     inode->i_atime.tv_sec, &token);
3513 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3514 				      inode->i_atime.tv_nsec, &token);
3515 
3516 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3517 				     inode->i_mtime.tv_sec, &token);
3518 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3519 				      inode->i_mtime.tv_nsec, &token);
3520 
3521 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3522 				     inode->i_ctime.tv_sec, &token);
3523 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3524 				      inode->i_ctime.tv_nsec, &token);
3525 
3526 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3527 				     &token);
3528 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3529 					 &token);
3530 	btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3531 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3532 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3533 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3534 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3535 }
3536 
3537 /*
3538  * copy everything in the in-memory inode into the btree.
3539  */
3540 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3541 				struct btrfs_root *root, struct inode *inode)
3542 {
3543 	struct btrfs_inode_item *inode_item;
3544 	struct btrfs_path *path;
3545 	struct extent_buffer *leaf;
3546 	int ret;
3547 
3548 	path = btrfs_alloc_path();
3549 	if (!path)
3550 		return -ENOMEM;
3551 
3552 	path->leave_spinning = 1;
3553 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3554 				 1);
3555 	if (ret) {
3556 		if (ret > 0)
3557 			ret = -ENOENT;
3558 		goto failed;
3559 	}
3560 
3561 	leaf = path->nodes[0];
3562 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3563 				    struct btrfs_inode_item);
3564 
3565 	fill_inode_item(trans, leaf, inode_item, inode);
3566 	btrfs_mark_buffer_dirty(leaf);
3567 	btrfs_set_inode_last_trans(trans, inode);
3568 	ret = 0;
3569 failed:
3570 	btrfs_free_path(path);
3571 	return ret;
3572 }
3573 
3574 /*
3575  * copy everything in the in-memory inode into the btree.
3576  */
3577 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3578 				struct btrfs_root *root, struct inode *inode)
3579 {
3580 	int ret;
3581 
3582 	/*
3583 	 * If the inode is a free space inode, we can deadlock during commit
3584 	 * if we put it into the delayed code.
3585 	 *
3586 	 * The data relocation inode should also be directly updated
3587 	 * without delay
3588 	 */
3589 	if (!btrfs_is_free_space_inode(inode)
3590 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3591 		btrfs_update_root_times(trans, root);
3592 
3593 		ret = btrfs_delayed_update_inode(trans, root, inode);
3594 		if (!ret)
3595 			btrfs_set_inode_last_trans(trans, inode);
3596 		return ret;
3597 	}
3598 
3599 	return btrfs_update_inode_item(trans, root, inode);
3600 }
3601 
3602 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3603 					 struct btrfs_root *root,
3604 					 struct inode *inode)
3605 {
3606 	int ret;
3607 
3608 	ret = btrfs_update_inode(trans, root, inode);
3609 	if (ret == -ENOSPC)
3610 		return btrfs_update_inode_item(trans, root, inode);
3611 	return ret;
3612 }
3613 
3614 /*
3615  * unlink helper that gets used here in inode.c and in the tree logging
3616  * recovery code.  It remove a link in a directory with a given name, and
3617  * also drops the back refs in the inode to the directory
3618  */
3619 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3620 				struct btrfs_root *root,
3621 				struct inode *dir, struct inode *inode,
3622 				const char *name, int name_len)
3623 {
3624 	struct btrfs_path *path;
3625 	int ret = 0;
3626 	struct extent_buffer *leaf;
3627 	struct btrfs_dir_item *di;
3628 	struct btrfs_key key;
3629 	u64 index;
3630 	u64 ino = btrfs_ino(inode);
3631 	u64 dir_ino = btrfs_ino(dir);
3632 
3633 	path = btrfs_alloc_path();
3634 	if (!path) {
3635 		ret = -ENOMEM;
3636 		goto out;
3637 	}
3638 
3639 	path->leave_spinning = 1;
3640 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3641 				    name, name_len, -1);
3642 	if (IS_ERR(di)) {
3643 		ret = PTR_ERR(di);
3644 		goto err;
3645 	}
3646 	if (!di) {
3647 		ret = -ENOENT;
3648 		goto err;
3649 	}
3650 	leaf = path->nodes[0];
3651 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3652 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3653 	if (ret)
3654 		goto err;
3655 	btrfs_release_path(path);
3656 
3657 	/*
3658 	 * If we don't have dir index, we have to get it by looking up
3659 	 * the inode ref, since we get the inode ref, remove it directly,
3660 	 * it is unnecessary to do delayed deletion.
3661 	 *
3662 	 * But if we have dir index, needn't search inode ref to get it.
3663 	 * Since the inode ref is close to the inode item, it is better
3664 	 * that we delay to delete it, and just do this deletion when
3665 	 * we update the inode item.
3666 	 */
3667 	if (BTRFS_I(inode)->dir_index) {
3668 		ret = btrfs_delayed_delete_inode_ref(inode);
3669 		if (!ret) {
3670 			index = BTRFS_I(inode)->dir_index;
3671 			goto skip_backref;
3672 		}
3673 	}
3674 
3675 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3676 				  dir_ino, &index);
3677 	if (ret) {
3678 		btrfs_info(root->fs_info,
3679 			"failed to delete reference to %.*s, inode %llu parent %llu",
3680 			name_len, name, ino, dir_ino);
3681 		btrfs_abort_transaction(trans, root, ret);
3682 		goto err;
3683 	}
3684 skip_backref:
3685 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3686 	if (ret) {
3687 		btrfs_abort_transaction(trans, root, ret);
3688 		goto err;
3689 	}
3690 
3691 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3692 					 inode, dir_ino);
3693 	if (ret != 0 && ret != -ENOENT) {
3694 		btrfs_abort_transaction(trans, root, ret);
3695 		goto err;
3696 	}
3697 
3698 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3699 					   dir, index);
3700 	if (ret == -ENOENT)
3701 		ret = 0;
3702 	else if (ret)
3703 		btrfs_abort_transaction(trans, root, ret);
3704 err:
3705 	btrfs_free_path(path);
3706 	if (ret)
3707 		goto out;
3708 
3709 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3710 	inode_inc_iversion(inode);
3711 	inode_inc_iversion(dir);
3712 	inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3713 	ret = btrfs_update_inode(trans, root, dir);
3714 out:
3715 	return ret;
3716 }
3717 
3718 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3719 		       struct btrfs_root *root,
3720 		       struct inode *dir, struct inode *inode,
3721 		       const char *name, int name_len)
3722 {
3723 	int ret;
3724 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3725 	if (!ret) {
3726 		drop_nlink(inode);
3727 		ret = btrfs_update_inode(trans, root, inode);
3728 	}
3729 	return ret;
3730 }
3731 
3732 /*
3733  * helper to start transaction for unlink and rmdir.
3734  *
3735  * unlink and rmdir are special in btrfs, they do not always free space, so
3736  * if we cannot make our reservations the normal way try and see if there is
3737  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3738  * allow the unlink to occur.
3739  */
3740 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3741 {
3742 	struct btrfs_trans_handle *trans;
3743 	struct btrfs_root *root = BTRFS_I(dir)->root;
3744 	int ret;
3745 
3746 	/*
3747 	 * 1 for the possible orphan item
3748 	 * 1 for the dir item
3749 	 * 1 for the dir index
3750 	 * 1 for the inode ref
3751 	 * 1 for the inode
3752 	 */
3753 	trans = btrfs_start_transaction(root, 5);
3754 	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3755 		return trans;
3756 
3757 	if (PTR_ERR(trans) == -ENOSPC) {
3758 		u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3759 
3760 		trans = btrfs_start_transaction(root, 0);
3761 		if (IS_ERR(trans))
3762 			return trans;
3763 		ret = btrfs_cond_migrate_bytes(root->fs_info,
3764 					       &root->fs_info->trans_block_rsv,
3765 					       num_bytes, 5);
3766 		if (ret) {
3767 			btrfs_end_transaction(trans, root);
3768 			return ERR_PTR(ret);
3769 		}
3770 		trans->block_rsv = &root->fs_info->trans_block_rsv;
3771 		trans->bytes_reserved = num_bytes;
3772 	}
3773 	return trans;
3774 }
3775 
3776 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3777 {
3778 	struct btrfs_root *root = BTRFS_I(dir)->root;
3779 	struct btrfs_trans_handle *trans;
3780 	struct inode *inode = dentry->d_inode;
3781 	int ret;
3782 
3783 	trans = __unlink_start_trans(dir);
3784 	if (IS_ERR(trans))
3785 		return PTR_ERR(trans);
3786 
3787 	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3788 
3789 	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3790 				 dentry->d_name.name, dentry->d_name.len);
3791 	if (ret)
3792 		goto out;
3793 
3794 	if (inode->i_nlink == 0) {
3795 		ret = btrfs_orphan_add(trans, inode);
3796 		if (ret)
3797 			goto out;
3798 	}
3799 
3800 out:
3801 	btrfs_end_transaction(trans, root);
3802 	btrfs_btree_balance_dirty(root);
3803 	return ret;
3804 }
3805 
3806 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3807 			struct btrfs_root *root,
3808 			struct inode *dir, u64 objectid,
3809 			const char *name, int name_len)
3810 {
3811 	struct btrfs_path *path;
3812 	struct extent_buffer *leaf;
3813 	struct btrfs_dir_item *di;
3814 	struct btrfs_key key;
3815 	u64 index;
3816 	int ret;
3817 	u64 dir_ino = btrfs_ino(dir);
3818 
3819 	path = btrfs_alloc_path();
3820 	if (!path)
3821 		return -ENOMEM;
3822 
3823 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3824 				   name, name_len, -1);
3825 	if (IS_ERR_OR_NULL(di)) {
3826 		if (!di)
3827 			ret = -ENOENT;
3828 		else
3829 			ret = PTR_ERR(di);
3830 		goto out;
3831 	}
3832 
3833 	leaf = path->nodes[0];
3834 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3835 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3836 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3837 	if (ret) {
3838 		btrfs_abort_transaction(trans, root, ret);
3839 		goto out;
3840 	}
3841 	btrfs_release_path(path);
3842 
3843 	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3844 				 objectid, root->root_key.objectid,
3845 				 dir_ino, &index, name, name_len);
3846 	if (ret < 0) {
3847 		if (ret != -ENOENT) {
3848 			btrfs_abort_transaction(trans, root, ret);
3849 			goto out;
3850 		}
3851 		di = btrfs_search_dir_index_item(root, path, dir_ino,
3852 						 name, name_len);
3853 		if (IS_ERR_OR_NULL(di)) {
3854 			if (!di)
3855 				ret = -ENOENT;
3856 			else
3857 				ret = PTR_ERR(di);
3858 			btrfs_abort_transaction(trans, root, ret);
3859 			goto out;
3860 		}
3861 
3862 		leaf = path->nodes[0];
3863 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3864 		btrfs_release_path(path);
3865 		index = key.offset;
3866 	}
3867 	btrfs_release_path(path);
3868 
3869 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3870 	if (ret) {
3871 		btrfs_abort_transaction(trans, root, ret);
3872 		goto out;
3873 	}
3874 
3875 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3876 	inode_inc_iversion(dir);
3877 	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3878 	ret = btrfs_update_inode_fallback(trans, root, dir);
3879 	if (ret)
3880 		btrfs_abort_transaction(trans, root, ret);
3881 out:
3882 	btrfs_free_path(path);
3883 	return ret;
3884 }
3885 
3886 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3887 {
3888 	struct inode *inode = dentry->d_inode;
3889 	int err = 0;
3890 	struct btrfs_root *root = BTRFS_I(dir)->root;
3891 	struct btrfs_trans_handle *trans;
3892 
3893 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3894 		return -ENOTEMPTY;
3895 	if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3896 		return -EPERM;
3897 
3898 	trans = __unlink_start_trans(dir);
3899 	if (IS_ERR(trans))
3900 		return PTR_ERR(trans);
3901 
3902 	if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3903 		err = btrfs_unlink_subvol(trans, root, dir,
3904 					  BTRFS_I(inode)->location.objectid,
3905 					  dentry->d_name.name,
3906 					  dentry->d_name.len);
3907 		goto out;
3908 	}
3909 
3910 	err = btrfs_orphan_add(trans, inode);
3911 	if (err)
3912 		goto out;
3913 
3914 	/* now the directory is empty */
3915 	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3916 				 dentry->d_name.name, dentry->d_name.len);
3917 	if (!err)
3918 		btrfs_i_size_write(inode, 0);
3919 out:
3920 	btrfs_end_transaction(trans, root);
3921 	btrfs_btree_balance_dirty(root);
3922 
3923 	return err;
3924 }
3925 
3926 /*
3927  * this can truncate away extent items, csum items and directory items.
3928  * It starts at a high offset and removes keys until it can't find
3929  * any higher than new_size
3930  *
3931  * csum items that cross the new i_size are truncated to the new size
3932  * as well.
3933  *
3934  * min_type is the minimum key type to truncate down to.  If set to 0, this
3935  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3936  */
3937 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3938 			       struct btrfs_root *root,
3939 			       struct inode *inode,
3940 			       u64 new_size, u32 min_type)
3941 {
3942 	struct btrfs_path *path;
3943 	struct extent_buffer *leaf;
3944 	struct btrfs_file_extent_item *fi;
3945 	struct btrfs_key key;
3946 	struct btrfs_key found_key;
3947 	u64 extent_start = 0;
3948 	u64 extent_num_bytes = 0;
3949 	u64 extent_offset = 0;
3950 	u64 item_end = 0;
3951 	u64 last_size = (u64)-1;
3952 	u32 found_type = (u8)-1;
3953 	int found_extent;
3954 	int del_item;
3955 	int pending_del_nr = 0;
3956 	int pending_del_slot = 0;
3957 	int extent_type = -1;
3958 	int ret;
3959 	int err = 0;
3960 	u64 ino = btrfs_ino(inode);
3961 
3962 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3963 
3964 	path = btrfs_alloc_path();
3965 	if (!path)
3966 		return -ENOMEM;
3967 	path->reada = -1;
3968 
3969 	/*
3970 	 * We want to drop from the next block forward in case this new size is
3971 	 * not block aligned since we will be keeping the last block of the
3972 	 * extent just the way it is.
3973 	 */
3974 	if (root->ref_cows || root == root->fs_info->tree_root)
3975 		btrfs_drop_extent_cache(inode, ALIGN(new_size,
3976 					root->sectorsize), (u64)-1, 0);
3977 
3978 	/*
3979 	 * This function is also used to drop the items in the log tree before
3980 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3981 	 * it is used to drop the loged items. So we shouldn't kill the delayed
3982 	 * items.
3983 	 */
3984 	if (min_type == 0 && root == BTRFS_I(inode)->root)
3985 		btrfs_kill_delayed_inode_items(inode);
3986 
3987 	key.objectid = ino;
3988 	key.offset = (u64)-1;
3989 	key.type = (u8)-1;
3990 
3991 search_again:
3992 	path->leave_spinning = 1;
3993 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3994 	if (ret < 0) {
3995 		err = ret;
3996 		goto out;
3997 	}
3998 
3999 	if (ret > 0) {
4000 		/* there are no items in the tree for us to truncate, we're
4001 		 * done
4002 		 */
4003 		if (path->slots[0] == 0)
4004 			goto out;
4005 		path->slots[0]--;
4006 	}
4007 
4008 	while (1) {
4009 		fi = NULL;
4010 		leaf = path->nodes[0];
4011 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4012 		found_type = btrfs_key_type(&found_key);
4013 
4014 		if (found_key.objectid != ino)
4015 			break;
4016 
4017 		if (found_type < min_type)
4018 			break;
4019 
4020 		item_end = found_key.offset;
4021 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4022 			fi = btrfs_item_ptr(leaf, path->slots[0],
4023 					    struct btrfs_file_extent_item);
4024 			extent_type = btrfs_file_extent_type(leaf, fi);
4025 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4026 				item_end +=
4027 				    btrfs_file_extent_num_bytes(leaf, fi);
4028 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4029 				item_end += btrfs_file_extent_inline_len(leaf,
4030 							 path->slots[0], fi);
4031 			}
4032 			item_end--;
4033 		}
4034 		if (found_type > min_type) {
4035 			del_item = 1;
4036 		} else {
4037 			if (item_end < new_size)
4038 				break;
4039 			if (found_key.offset >= new_size)
4040 				del_item = 1;
4041 			else
4042 				del_item = 0;
4043 		}
4044 		found_extent = 0;
4045 		/* FIXME, shrink the extent if the ref count is only 1 */
4046 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4047 			goto delete;
4048 
4049 		if (del_item)
4050 			last_size = found_key.offset;
4051 		else
4052 			last_size = new_size;
4053 
4054 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4055 			u64 num_dec;
4056 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4057 			if (!del_item) {
4058 				u64 orig_num_bytes =
4059 					btrfs_file_extent_num_bytes(leaf, fi);
4060 				extent_num_bytes = ALIGN(new_size -
4061 						found_key.offset,
4062 						root->sectorsize);
4063 				btrfs_set_file_extent_num_bytes(leaf, fi,
4064 							 extent_num_bytes);
4065 				num_dec = (orig_num_bytes -
4066 					   extent_num_bytes);
4067 				if (root->ref_cows && extent_start != 0)
4068 					inode_sub_bytes(inode, num_dec);
4069 				btrfs_mark_buffer_dirty(leaf);
4070 			} else {
4071 				extent_num_bytes =
4072 					btrfs_file_extent_disk_num_bytes(leaf,
4073 									 fi);
4074 				extent_offset = found_key.offset -
4075 					btrfs_file_extent_offset(leaf, fi);
4076 
4077 				/* FIXME blocksize != 4096 */
4078 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4079 				if (extent_start != 0) {
4080 					found_extent = 1;
4081 					if (root->ref_cows)
4082 						inode_sub_bytes(inode, num_dec);
4083 				}
4084 			}
4085 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4086 			/*
4087 			 * we can't truncate inline items that have had
4088 			 * special encodings
4089 			 */
4090 			if (!del_item &&
4091 			    btrfs_file_extent_compression(leaf, fi) == 0 &&
4092 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4093 			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4094 				u32 size = new_size - found_key.offset;
4095 
4096 				if (root->ref_cows) {
4097 					inode_sub_bytes(inode, item_end + 1 -
4098 							new_size);
4099 				}
4100 
4101 				/*
4102 				 * update the ram bytes to properly reflect
4103 				 * the new size of our item
4104 				 */
4105 				btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4106 				size =
4107 				    btrfs_file_extent_calc_inline_size(size);
4108 				btrfs_truncate_item(root, path, size, 1);
4109 			} else if (root->ref_cows) {
4110 				inode_sub_bytes(inode, item_end + 1 -
4111 						found_key.offset);
4112 			}
4113 		}
4114 delete:
4115 		if (del_item) {
4116 			if (!pending_del_nr) {
4117 				/* no pending yet, add ourselves */
4118 				pending_del_slot = path->slots[0];
4119 				pending_del_nr = 1;
4120 			} else if (pending_del_nr &&
4121 				   path->slots[0] + 1 == pending_del_slot) {
4122 				/* hop on the pending chunk */
4123 				pending_del_nr++;
4124 				pending_del_slot = path->slots[0];
4125 			} else {
4126 				BUG();
4127 			}
4128 		} else {
4129 			break;
4130 		}
4131 		if (found_extent && (root->ref_cows ||
4132 				     root == root->fs_info->tree_root)) {
4133 			btrfs_set_path_blocking(path);
4134 			ret = btrfs_free_extent(trans, root, extent_start,
4135 						extent_num_bytes, 0,
4136 						btrfs_header_owner(leaf),
4137 						ino, extent_offset, 0);
4138 			BUG_ON(ret);
4139 		}
4140 
4141 		if (found_type == BTRFS_INODE_ITEM_KEY)
4142 			break;
4143 
4144 		if (path->slots[0] == 0 ||
4145 		    path->slots[0] != pending_del_slot) {
4146 			if (pending_del_nr) {
4147 				ret = btrfs_del_items(trans, root, path,
4148 						pending_del_slot,
4149 						pending_del_nr);
4150 				if (ret) {
4151 					btrfs_abort_transaction(trans,
4152 								root, ret);
4153 					goto error;
4154 				}
4155 				pending_del_nr = 0;
4156 			}
4157 			btrfs_release_path(path);
4158 			goto search_again;
4159 		} else {
4160 			path->slots[0]--;
4161 		}
4162 	}
4163 out:
4164 	if (pending_del_nr) {
4165 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4166 				      pending_del_nr);
4167 		if (ret)
4168 			btrfs_abort_transaction(trans, root, ret);
4169 	}
4170 error:
4171 	if (last_size != (u64)-1)
4172 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4173 	btrfs_free_path(path);
4174 	return err;
4175 }
4176 
4177 /*
4178  * btrfs_truncate_page - read, zero a chunk and write a page
4179  * @inode - inode that we're zeroing
4180  * @from - the offset to start zeroing
4181  * @len - the length to zero, 0 to zero the entire range respective to the
4182  *	offset
4183  * @front - zero up to the offset instead of from the offset on
4184  *
4185  * This will find the page for the "from" offset and cow the page and zero the
4186  * part we want to zero.  This is used with truncate and hole punching.
4187  */
4188 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4189 			int front)
4190 {
4191 	struct address_space *mapping = inode->i_mapping;
4192 	struct btrfs_root *root = BTRFS_I(inode)->root;
4193 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4194 	struct btrfs_ordered_extent *ordered;
4195 	struct extent_state *cached_state = NULL;
4196 	char *kaddr;
4197 	u32 blocksize = root->sectorsize;
4198 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
4199 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
4200 	struct page *page;
4201 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4202 	int ret = 0;
4203 	u64 page_start;
4204 	u64 page_end;
4205 
4206 	if ((offset & (blocksize - 1)) == 0 &&
4207 	    (!len || ((len & (blocksize - 1)) == 0)))
4208 		goto out;
4209 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4210 	if (ret)
4211 		goto out;
4212 
4213 again:
4214 	page = find_or_create_page(mapping, index, mask);
4215 	if (!page) {
4216 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4217 		ret = -ENOMEM;
4218 		goto out;
4219 	}
4220 
4221 	page_start = page_offset(page);
4222 	page_end = page_start + PAGE_CACHE_SIZE - 1;
4223 
4224 	if (!PageUptodate(page)) {
4225 		ret = btrfs_readpage(NULL, page);
4226 		lock_page(page);
4227 		if (page->mapping != mapping) {
4228 			unlock_page(page);
4229 			page_cache_release(page);
4230 			goto again;
4231 		}
4232 		if (!PageUptodate(page)) {
4233 			ret = -EIO;
4234 			goto out_unlock;
4235 		}
4236 	}
4237 	wait_on_page_writeback(page);
4238 
4239 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4240 	set_page_extent_mapped(page);
4241 
4242 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
4243 	if (ordered) {
4244 		unlock_extent_cached(io_tree, page_start, page_end,
4245 				     &cached_state, GFP_NOFS);
4246 		unlock_page(page);
4247 		page_cache_release(page);
4248 		btrfs_start_ordered_extent(inode, ordered, 1);
4249 		btrfs_put_ordered_extent(ordered);
4250 		goto again;
4251 	}
4252 
4253 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4254 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4255 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4256 			  0, 0, &cached_state, GFP_NOFS);
4257 
4258 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4259 					&cached_state);
4260 	if (ret) {
4261 		unlock_extent_cached(io_tree, page_start, page_end,
4262 				     &cached_state, GFP_NOFS);
4263 		goto out_unlock;
4264 	}
4265 
4266 	if (offset != PAGE_CACHE_SIZE) {
4267 		if (!len)
4268 			len = PAGE_CACHE_SIZE - offset;
4269 		kaddr = kmap(page);
4270 		if (front)
4271 			memset(kaddr, 0, offset);
4272 		else
4273 			memset(kaddr + offset, 0, len);
4274 		flush_dcache_page(page);
4275 		kunmap(page);
4276 	}
4277 	ClearPageChecked(page);
4278 	set_page_dirty(page);
4279 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4280 			     GFP_NOFS);
4281 
4282 out_unlock:
4283 	if (ret)
4284 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4285 	unlock_page(page);
4286 	page_cache_release(page);
4287 out:
4288 	return ret;
4289 }
4290 
4291 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4292 			     u64 offset, u64 len)
4293 {
4294 	struct btrfs_trans_handle *trans;
4295 	int ret;
4296 
4297 	/*
4298 	 * Still need to make sure the inode looks like it's been updated so
4299 	 * that any holes get logged if we fsync.
4300 	 */
4301 	if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4302 		BTRFS_I(inode)->last_trans = root->fs_info->generation;
4303 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4304 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4305 		return 0;
4306 	}
4307 
4308 	/*
4309 	 * 1 - for the one we're dropping
4310 	 * 1 - for the one we're adding
4311 	 * 1 - for updating the inode.
4312 	 */
4313 	trans = btrfs_start_transaction(root, 3);
4314 	if (IS_ERR(trans))
4315 		return PTR_ERR(trans);
4316 
4317 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4318 	if (ret) {
4319 		btrfs_abort_transaction(trans, root, ret);
4320 		btrfs_end_transaction(trans, root);
4321 		return ret;
4322 	}
4323 
4324 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4325 				       0, 0, len, 0, len, 0, 0, 0);
4326 	if (ret)
4327 		btrfs_abort_transaction(trans, root, ret);
4328 	else
4329 		btrfs_update_inode(trans, root, inode);
4330 	btrfs_end_transaction(trans, root);
4331 	return ret;
4332 }
4333 
4334 /*
4335  * This function puts in dummy file extents for the area we're creating a hole
4336  * for.  So if we are truncating this file to a larger size we need to insert
4337  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4338  * the range between oldsize and size
4339  */
4340 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4341 {
4342 	struct btrfs_root *root = BTRFS_I(inode)->root;
4343 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4344 	struct extent_map *em = NULL;
4345 	struct extent_state *cached_state = NULL;
4346 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4347 	u64 hole_start = ALIGN(oldsize, root->sectorsize);
4348 	u64 block_end = ALIGN(size, root->sectorsize);
4349 	u64 last_byte;
4350 	u64 cur_offset;
4351 	u64 hole_size;
4352 	int err = 0;
4353 
4354 	/*
4355 	 * If our size started in the middle of a page we need to zero out the
4356 	 * rest of the page before we expand the i_size, otherwise we could
4357 	 * expose stale data.
4358 	 */
4359 	err = btrfs_truncate_page(inode, oldsize, 0, 0);
4360 	if (err)
4361 		return err;
4362 
4363 	if (size <= hole_start)
4364 		return 0;
4365 
4366 	while (1) {
4367 		struct btrfs_ordered_extent *ordered;
4368 
4369 		lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4370 				 &cached_state);
4371 		ordered = btrfs_lookup_ordered_range(inode, hole_start,
4372 						     block_end - hole_start);
4373 		if (!ordered)
4374 			break;
4375 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4376 				     &cached_state, GFP_NOFS);
4377 		btrfs_start_ordered_extent(inode, ordered, 1);
4378 		btrfs_put_ordered_extent(ordered);
4379 	}
4380 
4381 	cur_offset = hole_start;
4382 	while (1) {
4383 		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4384 				block_end - cur_offset, 0);
4385 		if (IS_ERR(em)) {
4386 			err = PTR_ERR(em);
4387 			em = NULL;
4388 			break;
4389 		}
4390 		last_byte = min(extent_map_end(em), block_end);
4391 		last_byte = ALIGN(last_byte , root->sectorsize);
4392 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4393 			struct extent_map *hole_em;
4394 			hole_size = last_byte - cur_offset;
4395 
4396 			err = maybe_insert_hole(root, inode, cur_offset,
4397 						hole_size);
4398 			if (err)
4399 				break;
4400 			btrfs_drop_extent_cache(inode, cur_offset,
4401 						cur_offset + hole_size - 1, 0);
4402 			hole_em = alloc_extent_map();
4403 			if (!hole_em) {
4404 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4405 					&BTRFS_I(inode)->runtime_flags);
4406 				goto next;
4407 			}
4408 			hole_em->start = cur_offset;
4409 			hole_em->len = hole_size;
4410 			hole_em->orig_start = cur_offset;
4411 
4412 			hole_em->block_start = EXTENT_MAP_HOLE;
4413 			hole_em->block_len = 0;
4414 			hole_em->orig_block_len = 0;
4415 			hole_em->ram_bytes = hole_size;
4416 			hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4417 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
4418 			hole_em->generation = root->fs_info->generation;
4419 
4420 			while (1) {
4421 				write_lock(&em_tree->lock);
4422 				err = add_extent_mapping(em_tree, hole_em, 1);
4423 				write_unlock(&em_tree->lock);
4424 				if (err != -EEXIST)
4425 					break;
4426 				btrfs_drop_extent_cache(inode, cur_offset,
4427 							cur_offset +
4428 							hole_size - 1, 0);
4429 			}
4430 			free_extent_map(hole_em);
4431 		}
4432 next:
4433 		free_extent_map(em);
4434 		em = NULL;
4435 		cur_offset = last_byte;
4436 		if (cur_offset >= block_end)
4437 			break;
4438 	}
4439 	free_extent_map(em);
4440 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4441 			     GFP_NOFS);
4442 	return err;
4443 }
4444 
4445 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4446 {
4447 	struct btrfs_root *root = BTRFS_I(inode)->root;
4448 	struct btrfs_trans_handle *trans;
4449 	loff_t oldsize = i_size_read(inode);
4450 	loff_t newsize = attr->ia_size;
4451 	int mask = attr->ia_valid;
4452 	int ret;
4453 
4454 	/*
4455 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4456 	 * special case where we need to update the times despite not having
4457 	 * these flags set.  For all other operations the VFS set these flags
4458 	 * explicitly if it wants a timestamp update.
4459 	 */
4460 	if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4461 		inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4462 
4463 	if (newsize > oldsize) {
4464 		truncate_pagecache(inode, newsize);
4465 		ret = btrfs_cont_expand(inode, oldsize, newsize);
4466 		if (ret)
4467 			return ret;
4468 
4469 		trans = btrfs_start_transaction(root, 1);
4470 		if (IS_ERR(trans))
4471 			return PTR_ERR(trans);
4472 
4473 		i_size_write(inode, newsize);
4474 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4475 		ret = btrfs_update_inode(trans, root, inode);
4476 		btrfs_end_transaction(trans, root);
4477 	} else {
4478 
4479 		/*
4480 		 * We're truncating a file that used to have good data down to
4481 		 * zero. Make sure it gets into the ordered flush list so that
4482 		 * any new writes get down to disk quickly.
4483 		 */
4484 		if (newsize == 0)
4485 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4486 				&BTRFS_I(inode)->runtime_flags);
4487 
4488 		/*
4489 		 * 1 for the orphan item we're going to add
4490 		 * 1 for the orphan item deletion.
4491 		 */
4492 		trans = btrfs_start_transaction(root, 2);
4493 		if (IS_ERR(trans))
4494 			return PTR_ERR(trans);
4495 
4496 		/*
4497 		 * We need to do this in case we fail at _any_ point during the
4498 		 * actual truncate.  Once we do the truncate_setsize we could
4499 		 * invalidate pages which forces any outstanding ordered io to
4500 		 * be instantly completed which will give us extents that need
4501 		 * to be truncated.  If we fail to get an orphan inode down we
4502 		 * could have left over extents that were never meant to live,
4503 		 * so we need to garuntee from this point on that everything
4504 		 * will be consistent.
4505 		 */
4506 		ret = btrfs_orphan_add(trans, inode);
4507 		btrfs_end_transaction(trans, root);
4508 		if (ret)
4509 			return ret;
4510 
4511 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
4512 		truncate_setsize(inode, newsize);
4513 
4514 		/* Disable nonlocked read DIO to avoid the end less truncate */
4515 		btrfs_inode_block_unlocked_dio(inode);
4516 		inode_dio_wait(inode);
4517 		btrfs_inode_resume_unlocked_dio(inode);
4518 
4519 		ret = btrfs_truncate(inode);
4520 		if (ret && inode->i_nlink) {
4521 			int err;
4522 
4523 			/*
4524 			 * failed to truncate, disk_i_size is only adjusted down
4525 			 * as we remove extents, so it should represent the true
4526 			 * size of the inode, so reset the in memory size and
4527 			 * delete our orphan entry.
4528 			 */
4529 			trans = btrfs_join_transaction(root);
4530 			if (IS_ERR(trans)) {
4531 				btrfs_orphan_del(NULL, inode);
4532 				return ret;
4533 			}
4534 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4535 			err = btrfs_orphan_del(trans, inode);
4536 			if (err)
4537 				btrfs_abort_transaction(trans, root, err);
4538 			btrfs_end_transaction(trans, root);
4539 		}
4540 	}
4541 
4542 	return ret;
4543 }
4544 
4545 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4546 {
4547 	struct inode *inode = dentry->d_inode;
4548 	struct btrfs_root *root = BTRFS_I(inode)->root;
4549 	int err;
4550 
4551 	if (btrfs_root_readonly(root))
4552 		return -EROFS;
4553 
4554 	err = inode_change_ok(inode, attr);
4555 	if (err)
4556 		return err;
4557 
4558 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4559 		err = btrfs_setsize(inode, attr);
4560 		if (err)
4561 			return err;
4562 	}
4563 
4564 	if (attr->ia_valid) {
4565 		setattr_copy(inode, attr);
4566 		inode_inc_iversion(inode);
4567 		err = btrfs_dirty_inode(inode);
4568 
4569 		if (!err && attr->ia_valid & ATTR_MODE)
4570 			err = btrfs_acl_chmod(inode);
4571 	}
4572 
4573 	return err;
4574 }
4575 
4576 /*
4577  * While truncating the inode pages during eviction, we get the VFS calling
4578  * btrfs_invalidatepage() against each page of the inode. This is slow because
4579  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4580  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4581  * extent_state structures over and over, wasting lots of time.
4582  *
4583  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4584  * those expensive operations on a per page basis and do only the ordered io
4585  * finishing, while we release here the extent_map and extent_state structures,
4586  * without the excessive merging and splitting.
4587  */
4588 static void evict_inode_truncate_pages(struct inode *inode)
4589 {
4590 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4591 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4592 	struct rb_node *node;
4593 
4594 	ASSERT(inode->i_state & I_FREEING);
4595 	truncate_inode_pages(&inode->i_data, 0);
4596 
4597 	write_lock(&map_tree->lock);
4598 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
4599 		struct extent_map *em;
4600 
4601 		node = rb_first(&map_tree->map);
4602 		em = rb_entry(node, struct extent_map, rb_node);
4603 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4604 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4605 		remove_extent_mapping(map_tree, em);
4606 		free_extent_map(em);
4607 	}
4608 	write_unlock(&map_tree->lock);
4609 
4610 	spin_lock(&io_tree->lock);
4611 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
4612 		struct extent_state *state;
4613 		struct extent_state *cached_state = NULL;
4614 
4615 		node = rb_first(&io_tree->state);
4616 		state = rb_entry(node, struct extent_state, rb_node);
4617 		atomic_inc(&state->refs);
4618 		spin_unlock(&io_tree->lock);
4619 
4620 		lock_extent_bits(io_tree, state->start, state->end,
4621 				 0, &cached_state);
4622 		clear_extent_bit(io_tree, state->start, state->end,
4623 				 EXTENT_LOCKED | EXTENT_DIRTY |
4624 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4625 				 EXTENT_DEFRAG, 1, 1,
4626 				 &cached_state, GFP_NOFS);
4627 		free_extent_state(state);
4628 
4629 		spin_lock(&io_tree->lock);
4630 	}
4631 	spin_unlock(&io_tree->lock);
4632 }
4633 
4634 void btrfs_evict_inode(struct inode *inode)
4635 {
4636 	struct btrfs_trans_handle *trans;
4637 	struct btrfs_root *root = BTRFS_I(inode)->root;
4638 	struct btrfs_block_rsv *rsv, *global_rsv;
4639 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4640 	int ret;
4641 
4642 	trace_btrfs_inode_evict(inode);
4643 
4644 	evict_inode_truncate_pages(inode);
4645 
4646 	if (inode->i_nlink &&
4647 	    ((btrfs_root_refs(&root->root_item) != 0 &&
4648 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4649 	     btrfs_is_free_space_inode(inode)))
4650 		goto no_delete;
4651 
4652 	if (is_bad_inode(inode)) {
4653 		btrfs_orphan_del(NULL, inode);
4654 		goto no_delete;
4655 	}
4656 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4657 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
4658 
4659 	if (root->fs_info->log_root_recovering) {
4660 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4661 				 &BTRFS_I(inode)->runtime_flags));
4662 		goto no_delete;
4663 	}
4664 
4665 	if (inode->i_nlink > 0) {
4666 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4667 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4668 		goto no_delete;
4669 	}
4670 
4671 	ret = btrfs_commit_inode_delayed_inode(inode);
4672 	if (ret) {
4673 		btrfs_orphan_del(NULL, inode);
4674 		goto no_delete;
4675 	}
4676 
4677 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4678 	if (!rsv) {
4679 		btrfs_orphan_del(NULL, inode);
4680 		goto no_delete;
4681 	}
4682 	rsv->size = min_size;
4683 	rsv->failfast = 1;
4684 	global_rsv = &root->fs_info->global_block_rsv;
4685 
4686 	btrfs_i_size_write(inode, 0);
4687 
4688 	/*
4689 	 * This is a bit simpler than btrfs_truncate since we've already
4690 	 * reserved our space for our orphan item in the unlink, so we just
4691 	 * need to reserve some slack space in case we add bytes and update
4692 	 * inode item when doing the truncate.
4693 	 */
4694 	while (1) {
4695 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
4696 					     BTRFS_RESERVE_FLUSH_LIMIT);
4697 
4698 		/*
4699 		 * Try and steal from the global reserve since we will
4700 		 * likely not use this space anyway, we want to try as
4701 		 * hard as possible to get this to work.
4702 		 */
4703 		if (ret)
4704 			ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4705 
4706 		if (ret) {
4707 			btrfs_warn(root->fs_info,
4708 				"Could not get space for a delete, will truncate on mount %d",
4709 				ret);
4710 			btrfs_orphan_del(NULL, inode);
4711 			btrfs_free_block_rsv(root, rsv);
4712 			goto no_delete;
4713 		}
4714 
4715 		trans = btrfs_join_transaction(root);
4716 		if (IS_ERR(trans)) {
4717 			btrfs_orphan_del(NULL, inode);
4718 			btrfs_free_block_rsv(root, rsv);
4719 			goto no_delete;
4720 		}
4721 
4722 		trans->block_rsv = rsv;
4723 
4724 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4725 		if (ret != -ENOSPC)
4726 			break;
4727 
4728 		trans->block_rsv = &root->fs_info->trans_block_rsv;
4729 		btrfs_end_transaction(trans, root);
4730 		trans = NULL;
4731 		btrfs_btree_balance_dirty(root);
4732 	}
4733 
4734 	btrfs_free_block_rsv(root, rsv);
4735 
4736 	/*
4737 	 * Errors here aren't a big deal, it just means we leave orphan items
4738 	 * in the tree.  They will be cleaned up on the next mount.
4739 	 */
4740 	if (ret == 0) {
4741 		trans->block_rsv = root->orphan_block_rsv;
4742 		btrfs_orphan_del(trans, inode);
4743 	} else {
4744 		btrfs_orphan_del(NULL, inode);
4745 	}
4746 
4747 	trans->block_rsv = &root->fs_info->trans_block_rsv;
4748 	if (!(root == root->fs_info->tree_root ||
4749 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4750 		btrfs_return_ino(root, btrfs_ino(inode));
4751 
4752 	btrfs_end_transaction(trans, root);
4753 	btrfs_btree_balance_dirty(root);
4754 no_delete:
4755 	btrfs_remove_delayed_node(inode);
4756 	clear_inode(inode);
4757 	return;
4758 }
4759 
4760 /*
4761  * this returns the key found in the dir entry in the location pointer.
4762  * If no dir entries were found, location->objectid is 0.
4763  */
4764 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4765 			       struct btrfs_key *location)
4766 {
4767 	const char *name = dentry->d_name.name;
4768 	int namelen = dentry->d_name.len;
4769 	struct btrfs_dir_item *di;
4770 	struct btrfs_path *path;
4771 	struct btrfs_root *root = BTRFS_I(dir)->root;
4772 	int ret = 0;
4773 
4774 	path = btrfs_alloc_path();
4775 	if (!path)
4776 		return -ENOMEM;
4777 
4778 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4779 				    namelen, 0);
4780 	if (IS_ERR(di))
4781 		ret = PTR_ERR(di);
4782 
4783 	if (IS_ERR_OR_NULL(di))
4784 		goto out_err;
4785 
4786 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4787 out:
4788 	btrfs_free_path(path);
4789 	return ret;
4790 out_err:
4791 	location->objectid = 0;
4792 	goto out;
4793 }
4794 
4795 /*
4796  * when we hit a tree root in a directory, the btrfs part of the inode
4797  * needs to be changed to reflect the root directory of the tree root.  This
4798  * is kind of like crossing a mount point.
4799  */
4800 static int fixup_tree_root_location(struct btrfs_root *root,
4801 				    struct inode *dir,
4802 				    struct dentry *dentry,
4803 				    struct btrfs_key *location,
4804 				    struct btrfs_root **sub_root)
4805 {
4806 	struct btrfs_path *path;
4807 	struct btrfs_root *new_root;
4808 	struct btrfs_root_ref *ref;
4809 	struct extent_buffer *leaf;
4810 	int ret;
4811 	int err = 0;
4812 
4813 	path = btrfs_alloc_path();
4814 	if (!path) {
4815 		err = -ENOMEM;
4816 		goto out;
4817 	}
4818 
4819 	err = -ENOENT;
4820 	ret = btrfs_find_item(root->fs_info->tree_root, path,
4821 				BTRFS_I(dir)->root->root_key.objectid,
4822 				location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4823 	if (ret) {
4824 		if (ret < 0)
4825 			err = ret;
4826 		goto out;
4827 	}
4828 
4829 	leaf = path->nodes[0];
4830 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4831 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4832 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4833 		goto out;
4834 
4835 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4836 				   (unsigned long)(ref + 1),
4837 				   dentry->d_name.len);
4838 	if (ret)
4839 		goto out;
4840 
4841 	btrfs_release_path(path);
4842 
4843 	new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4844 	if (IS_ERR(new_root)) {
4845 		err = PTR_ERR(new_root);
4846 		goto out;
4847 	}
4848 
4849 	*sub_root = new_root;
4850 	location->objectid = btrfs_root_dirid(&new_root->root_item);
4851 	location->type = BTRFS_INODE_ITEM_KEY;
4852 	location->offset = 0;
4853 	err = 0;
4854 out:
4855 	btrfs_free_path(path);
4856 	return err;
4857 }
4858 
4859 static void inode_tree_add(struct inode *inode)
4860 {
4861 	struct btrfs_root *root = BTRFS_I(inode)->root;
4862 	struct btrfs_inode *entry;
4863 	struct rb_node **p;
4864 	struct rb_node *parent;
4865 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
4866 	u64 ino = btrfs_ino(inode);
4867 
4868 	if (inode_unhashed(inode))
4869 		return;
4870 	parent = NULL;
4871 	spin_lock(&root->inode_lock);
4872 	p = &root->inode_tree.rb_node;
4873 	while (*p) {
4874 		parent = *p;
4875 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
4876 
4877 		if (ino < btrfs_ino(&entry->vfs_inode))
4878 			p = &parent->rb_left;
4879 		else if (ino > btrfs_ino(&entry->vfs_inode))
4880 			p = &parent->rb_right;
4881 		else {
4882 			WARN_ON(!(entry->vfs_inode.i_state &
4883 				  (I_WILL_FREE | I_FREEING)));
4884 			rb_replace_node(parent, new, &root->inode_tree);
4885 			RB_CLEAR_NODE(parent);
4886 			spin_unlock(&root->inode_lock);
4887 			return;
4888 		}
4889 	}
4890 	rb_link_node(new, parent, p);
4891 	rb_insert_color(new, &root->inode_tree);
4892 	spin_unlock(&root->inode_lock);
4893 }
4894 
4895 static void inode_tree_del(struct inode *inode)
4896 {
4897 	struct btrfs_root *root = BTRFS_I(inode)->root;
4898 	int empty = 0;
4899 
4900 	spin_lock(&root->inode_lock);
4901 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4902 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4903 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4904 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4905 	}
4906 	spin_unlock(&root->inode_lock);
4907 
4908 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
4909 		synchronize_srcu(&root->fs_info->subvol_srcu);
4910 		spin_lock(&root->inode_lock);
4911 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4912 		spin_unlock(&root->inode_lock);
4913 		if (empty)
4914 			btrfs_add_dead_root(root);
4915 	}
4916 }
4917 
4918 void btrfs_invalidate_inodes(struct btrfs_root *root)
4919 {
4920 	struct rb_node *node;
4921 	struct rb_node *prev;
4922 	struct btrfs_inode *entry;
4923 	struct inode *inode;
4924 	u64 objectid = 0;
4925 
4926 	if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
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_balance_delayed_items(root);
5803 	btrfs_btree_balance_dirty(root);
5804 	if (drop_inode) {
5805 		inode_dec_link_count(inode);
5806 		iput(inode);
5807 	}
5808 	return err;
5809 }
5810 
5811 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5812 			umode_t mode, bool excl)
5813 {
5814 	struct btrfs_trans_handle *trans;
5815 	struct btrfs_root *root = BTRFS_I(dir)->root;
5816 	struct inode *inode = NULL;
5817 	int drop_inode_on_err = 0;
5818 	int err;
5819 	u64 objectid;
5820 	u64 index = 0;
5821 
5822 	/*
5823 	 * 2 for inode item and ref
5824 	 * 2 for dir items
5825 	 * 1 for xattr if selinux is on
5826 	 */
5827 	trans = btrfs_start_transaction(root, 5);
5828 	if (IS_ERR(trans))
5829 		return PTR_ERR(trans);
5830 
5831 	err = btrfs_find_free_ino(root, &objectid);
5832 	if (err)
5833 		goto out_unlock;
5834 
5835 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5836 				dentry->d_name.len, btrfs_ino(dir), objectid,
5837 				mode, &index);
5838 	if (IS_ERR(inode)) {
5839 		err = PTR_ERR(inode);
5840 		goto out_unlock;
5841 	}
5842 	drop_inode_on_err = 1;
5843 
5844 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5845 	if (err)
5846 		goto out_unlock;
5847 
5848 	err = btrfs_update_inode(trans, root, inode);
5849 	if (err)
5850 		goto out_unlock;
5851 
5852 	/*
5853 	* If the active LSM wants to access the inode during
5854 	* d_instantiate it needs these. Smack checks to see
5855 	* if the filesystem supports xattrs by looking at the
5856 	* ops vector.
5857 	*/
5858 	inode->i_fop = &btrfs_file_operations;
5859 	inode->i_op = &btrfs_file_inode_operations;
5860 
5861 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5862 	if (err)
5863 		goto out_unlock;
5864 
5865 	inode->i_mapping->a_ops = &btrfs_aops;
5866 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5867 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5868 	d_instantiate(dentry, inode);
5869 
5870 out_unlock:
5871 	btrfs_end_transaction(trans, root);
5872 	if (err && drop_inode_on_err) {
5873 		inode_dec_link_count(inode);
5874 		iput(inode);
5875 	}
5876 	btrfs_balance_delayed_items(root);
5877 	btrfs_btree_balance_dirty(root);
5878 	return err;
5879 }
5880 
5881 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5882 		      struct dentry *dentry)
5883 {
5884 	struct btrfs_trans_handle *trans;
5885 	struct btrfs_root *root = BTRFS_I(dir)->root;
5886 	struct inode *inode = old_dentry->d_inode;
5887 	u64 index;
5888 	int err;
5889 	int drop_inode = 0;
5890 
5891 	/* do not allow sys_link's with other subvols of the same device */
5892 	if (root->objectid != BTRFS_I(inode)->root->objectid)
5893 		return -EXDEV;
5894 
5895 	if (inode->i_nlink >= BTRFS_LINK_MAX)
5896 		return -EMLINK;
5897 
5898 	err = btrfs_set_inode_index(dir, &index);
5899 	if (err)
5900 		goto fail;
5901 
5902 	/*
5903 	 * 2 items for inode and inode ref
5904 	 * 2 items for dir items
5905 	 * 1 item for parent inode
5906 	 */
5907 	trans = btrfs_start_transaction(root, 5);
5908 	if (IS_ERR(trans)) {
5909 		err = PTR_ERR(trans);
5910 		goto fail;
5911 	}
5912 
5913 	/* There are several dir indexes for this inode, clear the cache. */
5914 	BTRFS_I(inode)->dir_index = 0ULL;
5915 	inc_nlink(inode);
5916 	inode_inc_iversion(inode);
5917 	inode->i_ctime = CURRENT_TIME;
5918 	ihold(inode);
5919 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5920 
5921 	err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5922 
5923 	if (err) {
5924 		drop_inode = 1;
5925 	} else {
5926 		struct dentry *parent = dentry->d_parent;
5927 		err = btrfs_update_inode(trans, root, inode);
5928 		if (err)
5929 			goto fail;
5930 		d_instantiate(dentry, inode);
5931 		btrfs_log_new_name(trans, inode, NULL, parent);
5932 	}
5933 
5934 	btrfs_end_transaction(trans, root);
5935 	btrfs_balance_delayed_items(root);
5936 fail:
5937 	if (drop_inode) {
5938 		inode_dec_link_count(inode);
5939 		iput(inode);
5940 	}
5941 	btrfs_btree_balance_dirty(root);
5942 	return err;
5943 }
5944 
5945 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5946 {
5947 	struct inode *inode = NULL;
5948 	struct btrfs_trans_handle *trans;
5949 	struct btrfs_root *root = BTRFS_I(dir)->root;
5950 	int err = 0;
5951 	int drop_on_err = 0;
5952 	u64 objectid = 0;
5953 	u64 index = 0;
5954 
5955 	/*
5956 	 * 2 items for inode and ref
5957 	 * 2 items for dir items
5958 	 * 1 for xattr if selinux is on
5959 	 */
5960 	trans = btrfs_start_transaction(root, 5);
5961 	if (IS_ERR(trans))
5962 		return PTR_ERR(trans);
5963 
5964 	err = btrfs_find_free_ino(root, &objectid);
5965 	if (err)
5966 		goto out_fail;
5967 
5968 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5969 				dentry->d_name.len, btrfs_ino(dir), objectid,
5970 				S_IFDIR | mode, &index);
5971 	if (IS_ERR(inode)) {
5972 		err = PTR_ERR(inode);
5973 		goto out_fail;
5974 	}
5975 
5976 	drop_on_err = 1;
5977 
5978 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5979 	if (err)
5980 		goto out_fail;
5981 
5982 	inode->i_op = &btrfs_dir_inode_operations;
5983 	inode->i_fop = &btrfs_dir_file_operations;
5984 
5985 	btrfs_i_size_write(inode, 0);
5986 	err = btrfs_update_inode(trans, root, inode);
5987 	if (err)
5988 		goto out_fail;
5989 
5990 	err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5991 			     dentry->d_name.len, 0, index);
5992 	if (err)
5993 		goto out_fail;
5994 
5995 	d_instantiate(dentry, inode);
5996 	drop_on_err = 0;
5997 
5998 out_fail:
5999 	btrfs_end_transaction(trans, root);
6000 	if (drop_on_err)
6001 		iput(inode);
6002 	btrfs_balance_delayed_items(root);
6003 	btrfs_btree_balance_dirty(root);
6004 	return err;
6005 }
6006 
6007 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6008  * and an extent that you want to insert, deal with overlap and insert
6009  * the new extent into the tree.
6010  */
6011 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6012 				struct extent_map *existing,
6013 				struct extent_map *em,
6014 				u64 map_start, u64 map_len)
6015 {
6016 	u64 start_diff;
6017 
6018 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6019 	start_diff = map_start - em->start;
6020 	em->start = map_start;
6021 	em->len = map_len;
6022 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6023 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6024 		em->block_start += start_diff;
6025 		em->block_len -= start_diff;
6026 	}
6027 	return add_extent_mapping(em_tree, em, 0);
6028 }
6029 
6030 static noinline int uncompress_inline(struct btrfs_path *path,
6031 				      struct inode *inode, struct page *page,
6032 				      size_t pg_offset, u64 extent_offset,
6033 				      struct btrfs_file_extent_item *item)
6034 {
6035 	int ret;
6036 	struct extent_buffer *leaf = path->nodes[0];
6037 	char *tmp;
6038 	size_t max_size;
6039 	unsigned long inline_size;
6040 	unsigned long ptr;
6041 	int compress_type;
6042 
6043 	WARN_ON(pg_offset != 0);
6044 	compress_type = btrfs_file_extent_compression(leaf, item);
6045 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6046 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6047 					btrfs_item_nr(path->slots[0]));
6048 	tmp = kmalloc(inline_size, GFP_NOFS);
6049 	if (!tmp)
6050 		return -ENOMEM;
6051 	ptr = btrfs_file_extent_inline_start(item);
6052 
6053 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6054 
6055 	max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6056 	ret = btrfs_decompress(compress_type, tmp, page,
6057 			       extent_offset, inline_size, max_size);
6058 	if (ret) {
6059 		char *kaddr = kmap_atomic(page);
6060 		unsigned long copy_size = min_t(u64,
6061 				  PAGE_CACHE_SIZE - pg_offset,
6062 				  max_size - extent_offset);
6063 		memset(kaddr + pg_offset, 0, copy_size);
6064 		kunmap_atomic(kaddr);
6065 	}
6066 	kfree(tmp);
6067 	return 0;
6068 }
6069 
6070 /*
6071  * a bit scary, this does extent mapping from logical file offset to the disk.
6072  * the ugly parts come from merging extents from the disk with the in-ram
6073  * representation.  This gets more complex because of the data=ordered code,
6074  * where the in-ram extents might be locked pending data=ordered completion.
6075  *
6076  * This also copies inline extents directly into the page.
6077  */
6078 
6079 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6080 				    size_t pg_offset, u64 start, u64 len,
6081 				    int create)
6082 {
6083 	int ret;
6084 	int err = 0;
6085 	u64 bytenr;
6086 	u64 extent_start = 0;
6087 	u64 extent_end = 0;
6088 	u64 objectid = btrfs_ino(inode);
6089 	u32 found_type;
6090 	struct btrfs_path *path = NULL;
6091 	struct btrfs_root *root = BTRFS_I(inode)->root;
6092 	struct btrfs_file_extent_item *item;
6093 	struct extent_buffer *leaf;
6094 	struct btrfs_key found_key;
6095 	struct extent_map *em = NULL;
6096 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6097 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6098 	struct btrfs_trans_handle *trans = NULL;
6099 	int compress_type;
6100 
6101 again:
6102 	read_lock(&em_tree->lock);
6103 	em = lookup_extent_mapping(em_tree, start, len);
6104 	if (em)
6105 		em->bdev = root->fs_info->fs_devices->latest_bdev;
6106 	read_unlock(&em_tree->lock);
6107 
6108 	if (em) {
6109 		if (em->start > start || em->start + em->len <= start)
6110 			free_extent_map(em);
6111 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6112 			free_extent_map(em);
6113 		else
6114 			goto out;
6115 	}
6116 	em = alloc_extent_map();
6117 	if (!em) {
6118 		err = -ENOMEM;
6119 		goto out;
6120 	}
6121 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6122 	em->start = EXTENT_MAP_HOLE;
6123 	em->orig_start = EXTENT_MAP_HOLE;
6124 	em->len = (u64)-1;
6125 	em->block_len = (u64)-1;
6126 
6127 	if (!path) {
6128 		path = btrfs_alloc_path();
6129 		if (!path) {
6130 			err = -ENOMEM;
6131 			goto out;
6132 		}
6133 		/*
6134 		 * Chances are we'll be called again, so go ahead and do
6135 		 * readahead
6136 		 */
6137 		path->reada = 1;
6138 	}
6139 
6140 	ret = btrfs_lookup_file_extent(trans, root, path,
6141 				       objectid, start, trans != NULL);
6142 	if (ret < 0) {
6143 		err = ret;
6144 		goto out;
6145 	}
6146 
6147 	if (ret != 0) {
6148 		if (path->slots[0] == 0)
6149 			goto not_found;
6150 		path->slots[0]--;
6151 	}
6152 
6153 	leaf = path->nodes[0];
6154 	item = btrfs_item_ptr(leaf, path->slots[0],
6155 			      struct btrfs_file_extent_item);
6156 	/* are we inside the extent that was found? */
6157 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6158 	found_type = btrfs_key_type(&found_key);
6159 	if (found_key.objectid != objectid ||
6160 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6161 		/*
6162 		 * If we backup past the first extent we want to move forward
6163 		 * and see if there is an extent in front of us, otherwise we'll
6164 		 * say there is a hole for our whole search range which can
6165 		 * cause problems.
6166 		 */
6167 		extent_end = start;
6168 		goto next;
6169 	}
6170 
6171 	found_type = btrfs_file_extent_type(leaf, item);
6172 	extent_start = found_key.offset;
6173 	compress_type = btrfs_file_extent_compression(leaf, item);
6174 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6175 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6176 		extent_end = extent_start +
6177 		       btrfs_file_extent_num_bytes(leaf, item);
6178 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6179 		size_t size;
6180 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6181 		extent_end = ALIGN(extent_start + size, root->sectorsize);
6182 	}
6183 next:
6184 	if (start >= extent_end) {
6185 		path->slots[0]++;
6186 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6187 			ret = btrfs_next_leaf(root, path);
6188 			if (ret < 0) {
6189 				err = ret;
6190 				goto out;
6191 			}
6192 			if (ret > 0)
6193 				goto not_found;
6194 			leaf = path->nodes[0];
6195 		}
6196 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6197 		if (found_key.objectid != objectid ||
6198 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6199 			goto not_found;
6200 		if (start + len <= found_key.offset)
6201 			goto not_found;
6202 		em->start = start;
6203 		em->orig_start = start;
6204 		em->len = found_key.offset - start;
6205 		goto not_found_em;
6206 	}
6207 
6208 	em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6209 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6210 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6211 		em->start = extent_start;
6212 		em->len = extent_end - extent_start;
6213 		em->orig_start = extent_start -
6214 				 btrfs_file_extent_offset(leaf, item);
6215 		em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6216 								      item);
6217 		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6218 		if (bytenr == 0) {
6219 			em->block_start = EXTENT_MAP_HOLE;
6220 			goto insert;
6221 		}
6222 		if (compress_type != BTRFS_COMPRESS_NONE) {
6223 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6224 			em->compress_type = compress_type;
6225 			em->block_start = bytenr;
6226 			em->block_len = em->orig_block_len;
6227 		} else {
6228 			bytenr += btrfs_file_extent_offset(leaf, item);
6229 			em->block_start = bytenr;
6230 			em->block_len = em->len;
6231 			if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6232 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6233 		}
6234 		goto insert;
6235 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6236 		unsigned long ptr;
6237 		char *map;
6238 		size_t size;
6239 		size_t extent_offset;
6240 		size_t copy_size;
6241 
6242 		em->block_start = EXTENT_MAP_INLINE;
6243 		if (!page || create) {
6244 			em->start = extent_start;
6245 			em->len = extent_end - extent_start;
6246 			goto out;
6247 		}
6248 
6249 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6250 		extent_offset = page_offset(page) + pg_offset - extent_start;
6251 		copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6252 				size - extent_offset);
6253 		em->start = extent_start + extent_offset;
6254 		em->len = ALIGN(copy_size, root->sectorsize);
6255 		em->orig_block_len = em->len;
6256 		em->orig_start = em->start;
6257 		if (compress_type) {
6258 			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6259 			em->compress_type = compress_type;
6260 		}
6261 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6262 		if (create == 0 && !PageUptodate(page)) {
6263 			if (btrfs_file_extent_compression(leaf, item) !=
6264 			    BTRFS_COMPRESS_NONE) {
6265 				ret = uncompress_inline(path, inode, page,
6266 							pg_offset,
6267 							extent_offset, item);
6268 				BUG_ON(ret); /* -ENOMEM */
6269 			} else {
6270 				map = kmap(page);
6271 				read_extent_buffer(leaf, map + pg_offset, ptr,
6272 						   copy_size);
6273 				if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6274 					memset(map + pg_offset + copy_size, 0,
6275 					       PAGE_CACHE_SIZE - pg_offset -
6276 					       copy_size);
6277 				}
6278 				kunmap(page);
6279 			}
6280 			flush_dcache_page(page);
6281 		} else if (create && PageUptodate(page)) {
6282 			BUG();
6283 			if (!trans) {
6284 				kunmap(page);
6285 				free_extent_map(em);
6286 				em = NULL;
6287 
6288 				btrfs_release_path(path);
6289 				trans = btrfs_join_transaction(root);
6290 
6291 				if (IS_ERR(trans))
6292 					return ERR_CAST(trans);
6293 				goto again;
6294 			}
6295 			map = kmap(page);
6296 			write_extent_buffer(leaf, map + pg_offset, ptr,
6297 					    copy_size);
6298 			kunmap(page);
6299 			btrfs_mark_buffer_dirty(leaf);
6300 		}
6301 		set_extent_uptodate(io_tree, em->start,
6302 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
6303 		goto insert;
6304 	} else {
6305 		WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6306 	}
6307 not_found:
6308 	em->start = start;
6309 	em->orig_start = start;
6310 	em->len = len;
6311 not_found_em:
6312 	em->block_start = EXTENT_MAP_HOLE;
6313 	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6314 insert:
6315 	btrfs_release_path(path);
6316 	if (em->start > start || extent_map_end(em) <= start) {
6317 		btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6318 			em->start, em->len, start, len);
6319 		err = -EIO;
6320 		goto out;
6321 	}
6322 
6323 	err = 0;
6324 	write_lock(&em_tree->lock);
6325 	ret = add_extent_mapping(em_tree, em, 0);
6326 	/* it is possible that someone inserted the extent into the tree
6327 	 * while we had the lock dropped.  It is also possible that
6328 	 * an overlapping map exists in the tree
6329 	 */
6330 	if (ret == -EEXIST) {
6331 		struct extent_map *existing;
6332 
6333 		ret = 0;
6334 
6335 		existing = lookup_extent_mapping(em_tree, start, len);
6336 		if (existing && (existing->start > start ||
6337 		    existing->start + existing->len <= start)) {
6338 			free_extent_map(existing);
6339 			existing = NULL;
6340 		}
6341 		if (!existing) {
6342 			existing = lookup_extent_mapping(em_tree, em->start,
6343 							 em->len);
6344 			if (existing) {
6345 				err = merge_extent_mapping(em_tree, existing,
6346 							   em, start,
6347 							   root->sectorsize);
6348 				free_extent_map(existing);
6349 				if (err) {
6350 					free_extent_map(em);
6351 					em = NULL;
6352 				}
6353 			} else {
6354 				err = -EIO;
6355 				free_extent_map(em);
6356 				em = NULL;
6357 			}
6358 		} else {
6359 			free_extent_map(em);
6360 			em = existing;
6361 			err = 0;
6362 		}
6363 	}
6364 	write_unlock(&em_tree->lock);
6365 out:
6366 
6367 	trace_btrfs_get_extent(root, em);
6368 
6369 	if (path)
6370 		btrfs_free_path(path);
6371 	if (trans) {
6372 		ret = btrfs_end_transaction(trans, root);
6373 		if (!err)
6374 			err = ret;
6375 	}
6376 	if (err) {
6377 		free_extent_map(em);
6378 		return ERR_PTR(err);
6379 	}
6380 	BUG_ON(!em); /* Error is always set */
6381 	return em;
6382 }
6383 
6384 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6385 					   size_t pg_offset, u64 start, u64 len,
6386 					   int create)
6387 {
6388 	struct extent_map *em;
6389 	struct extent_map *hole_em = NULL;
6390 	u64 range_start = start;
6391 	u64 end;
6392 	u64 found;
6393 	u64 found_end;
6394 	int err = 0;
6395 
6396 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6397 	if (IS_ERR(em))
6398 		return em;
6399 	if (em) {
6400 		/*
6401 		 * if our em maps to
6402 		 * -  a hole or
6403 		 * -  a pre-alloc extent,
6404 		 * there might actually be delalloc bytes behind it.
6405 		 */
6406 		if (em->block_start != EXTENT_MAP_HOLE &&
6407 		    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6408 			return em;
6409 		else
6410 			hole_em = em;
6411 	}
6412 
6413 	/* check to see if we've wrapped (len == -1 or similar) */
6414 	end = start + len;
6415 	if (end < start)
6416 		end = (u64)-1;
6417 	else
6418 		end -= 1;
6419 
6420 	em = NULL;
6421 
6422 	/* ok, we didn't find anything, lets look for delalloc */
6423 	found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6424 				 end, len, EXTENT_DELALLOC, 1);
6425 	found_end = range_start + found;
6426 	if (found_end < range_start)
6427 		found_end = (u64)-1;
6428 
6429 	/*
6430 	 * we didn't find anything useful, return
6431 	 * the original results from get_extent()
6432 	 */
6433 	if (range_start > end || found_end <= start) {
6434 		em = hole_em;
6435 		hole_em = NULL;
6436 		goto out;
6437 	}
6438 
6439 	/* adjust the range_start to make sure it doesn't
6440 	 * go backwards from the start they passed in
6441 	 */
6442 	range_start = max(start, range_start);
6443 	found = found_end - range_start;
6444 
6445 	if (found > 0) {
6446 		u64 hole_start = start;
6447 		u64 hole_len = len;
6448 
6449 		em = alloc_extent_map();
6450 		if (!em) {
6451 			err = -ENOMEM;
6452 			goto out;
6453 		}
6454 		/*
6455 		 * when btrfs_get_extent can't find anything it
6456 		 * returns one huge hole
6457 		 *
6458 		 * make sure what it found really fits our range, and
6459 		 * adjust to make sure it is based on the start from
6460 		 * the caller
6461 		 */
6462 		if (hole_em) {
6463 			u64 calc_end = extent_map_end(hole_em);
6464 
6465 			if (calc_end <= start || (hole_em->start > end)) {
6466 				free_extent_map(hole_em);
6467 				hole_em = NULL;
6468 			} else {
6469 				hole_start = max(hole_em->start, start);
6470 				hole_len = calc_end - hole_start;
6471 			}
6472 		}
6473 		em->bdev = NULL;
6474 		if (hole_em && range_start > hole_start) {
6475 			/* our hole starts before our delalloc, so we
6476 			 * have to return just the parts of the hole
6477 			 * that go until  the delalloc starts
6478 			 */
6479 			em->len = min(hole_len,
6480 				      range_start - hole_start);
6481 			em->start = hole_start;
6482 			em->orig_start = hole_start;
6483 			/*
6484 			 * don't adjust block start at all,
6485 			 * it is fixed at EXTENT_MAP_HOLE
6486 			 */
6487 			em->block_start = hole_em->block_start;
6488 			em->block_len = hole_len;
6489 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6490 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6491 		} else {
6492 			em->start = range_start;
6493 			em->len = found;
6494 			em->orig_start = range_start;
6495 			em->block_start = EXTENT_MAP_DELALLOC;
6496 			em->block_len = found;
6497 		}
6498 	} else if (hole_em) {
6499 		return hole_em;
6500 	}
6501 out:
6502 
6503 	free_extent_map(hole_em);
6504 	if (err) {
6505 		free_extent_map(em);
6506 		return ERR_PTR(err);
6507 	}
6508 	return em;
6509 }
6510 
6511 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6512 						  u64 start, u64 len)
6513 {
6514 	struct btrfs_root *root = BTRFS_I(inode)->root;
6515 	struct extent_map *em;
6516 	struct btrfs_key ins;
6517 	u64 alloc_hint;
6518 	int ret;
6519 
6520 	alloc_hint = get_extent_allocation_hint(inode, start, len);
6521 	ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6522 				   alloc_hint, &ins, 1);
6523 	if (ret)
6524 		return ERR_PTR(ret);
6525 
6526 	em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6527 			      ins.offset, ins.offset, ins.offset, 0);
6528 	if (IS_ERR(em)) {
6529 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6530 		return em;
6531 	}
6532 
6533 	ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6534 					   ins.offset, ins.offset, 0);
6535 	if (ret) {
6536 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6537 		free_extent_map(em);
6538 		return ERR_PTR(ret);
6539 	}
6540 
6541 	return em;
6542 }
6543 
6544 /*
6545  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6546  * block must be cow'd
6547  */
6548 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6549 			      u64 *orig_start, u64 *orig_block_len,
6550 			      u64 *ram_bytes)
6551 {
6552 	struct btrfs_trans_handle *trans;
6553 	struct btrfs_path *path;
6554 	int ret;
6555 	struct extent_buffer *leaf;
6556 	struct btrfs_root *root = BTRFS_I(inode)->root;
6557 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6558 	struct btrfs_file_extent_item *fi;
6559 	struct btrfs_key key;
6560 	u64 disk_bytenr;
6561 	u64 backref_offset;
6562 	u64 extent_end;
6563 	u64 num_bytes;
6564 	int slot;
6565 	int found_type;
6566 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6567 
6568 	path = btrfs_alloc_path();
6569 	if (!path)
6570 		return -ENOMEM;
6571 
6572 	ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6573 				       offset, 0);
6574 	if (ret < 0)
6575 		goto out;
6576 
6577 	slot = path->slots[0];
6578 	if (ret == 1) {
6579 		if (slot == 0) {
6580 			/* can't find the item, must cow */
6581 			ret = 0;
6582 			goto out;
6583 		}
6584 		slot--;
6585 	}
6586 	ret = 0;
6587 	leaf = path->nodes[0];
6588 	btrfs_item_key_to_cpu(leaf, &key, slot);
6589 	if (key.objectid != btrfs_ino(inode) ||
6590 	    key.type != BTRFS_EXTENT_DATA_KEY) {
6591 		/* not our file or wrong item type, must cow */
6592 		goto out;
6593 	}
6594 
6595 	if (key.offset > offset) {
6596 		/* Wrong offset, must cow */
6597 		goto out;
6598 	}
6599 
6600 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6601 	found_type = btrfs_file_extent_type(leaf, fi);
6602 	if (found_type != BTRFS_FILE_EXTENT_REG &&
6603 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6604 		/* not a regular extent, must cow */
6605 		goto out;
6606 	}
6607 
6608 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6609 		goto out;
6610 
6611 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6612 	if (extent_end <= offset)
6613 		goto out;
6614 
6615 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6616 	if (disk_bytenr == 0)
6617 		goto out;
6618 
6619 	if (btrfs_file_extent_compression(leaf, fi) ||
6620 	    btrfs_file_extent_encryption(leaf, fi) ||
6621 	    btrfs_file_extent_other_encoding(leaf, fi))
6622 		goto out;
6623 
6624 	backref_offset = btrfs_file_extent_offset(leaf, fi);
6625 
6626 	if (orig_start) {
6627 		*orig_start = key.offset - backref_offset;
6628 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6629 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6630 	}
6631 
6632 	if (btrfs_extent_readonly(root, disk_bytenr))
6633 		goto out;
6634 
6635 	num_bytes = min(offset + *len, extent_end) - offset;
6636 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6637 		u64 range_end;
6638 
6639 		range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6640 		ret = test_range_bit(io_tree, offset, range_end,
6641 				     EXTENT_DELALLOC, 0, NULL);
6642 		if (ret) {
6643 			ret = -EAGAIN;
6644 			goto out;
6645 		}
6646 	}
6647 
6648 	btrfs_release_path(path);
6649 
6650 	/*
6651 	 * look for other files referencing this extent, if we
6652 	 * find any we must cow
6653 	 */
6654 	trans = btrfs_join_transaction(root);
6655 	if (IS_ERR(trans)) {
6656 		ret = 0;
6657 		goto out;
6658 	}
6659 
6660 	ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6661 				    key.offset - backref_offset, disk_bytenr);
6662 	btrfs_end_transaction(trans, root);
6663 	if (ret) {
6664 		ret = 0;
6665 		goto out;
6666 	}
6667 
6668 	/*
6669 	 * adjust disk_bytenr and num_bytes to cover just the bytes
6670 	 * in this extent we are about to write.  If there
6671 	 * are any csums in that range we have to cow in order
6672 	 * to keep the csums correct
6673 	 */
6674 	disk_bytenr += backref_offset;
6675 	disk_bytenr += offset - key.offset;
6676 	if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6677 				goto out;
6678 	/*
6679 	 * all of the above have passed, it is safe to overwrite this extent
6680 	 * without cow
6681 	 */
6682 	*len = num_bytes;
6683 	ret = 1;
6684 out:
6685 	btrfs_free_path(path);
6686 	return ret;
6687 }
6688 
6689 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6690 			      struct extent_state **cached_state, int writing)
6691 {
6692 	struct btrfs_ordered_extent *ordered;
6693 	int ret = 0;
6694 
6695 	while (1) {
6696 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6697 				 0, cached_state);
6698 		/*
6699 		 * We're concerned with the entire range that we're going to be
6700 		 * doing DIO to, so we need to make sure theres no ordered
6701 		 * extents in this range.
6702 		 */
6703 		ordered = btrfs_lookup_ordered_range(inode, lockstart,
6704 						     lockend - lockstart + 1);
6705 
6706 		/*
6707 		 * We need to make sure there are no buffered pages in this
6708 		 * range either, we could have raced between the invalidate in
6709 		 * generic_file_direct_write and locking the extent.  The
6710 		 * invalidate needs to happen so that reads after a write do not
6711 		 * get stale data.
6712 		 */
6713 		if (!ordered && (!writing ||
6714 		    !test_range_bit(&BTRFS_I(inode)->io_tree,
6715 				    lockstart, lockend, EXTENT_UPTODATE, 0,
6716 				    *cached_state)))
6717 			break;
6718 
6719 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6720 				     cached_state, GFP_NOFS);
6721 
6722 		if (ordered) {
6723 			btrfs_start_ordered_extent(inode, ordered, 1);
6724 			btrfs_put_ordered_extent(ordered);
6725 		} else {
6726 			/* Screw you mmap */
6727 			ret = filemap_write_and_wait_range(inode->i_mapping,
6728 							   lockstart,
6729 							   lockend);
6730 			if (ret)
6731 				break;
6732 
6733 			/*
6734 			 * If we found a page that couldn't be invalidated just
6735 			 * fall back to buffered.
6736 			 */
6737 			ret = invalidate_inode_pages2_range(inode->i_mapping,
6738 					lockstart >> PAGE_CACHE_SHIFT,
6739 					lockend >> PAGE_CACHE_SHIFT);
6740 			if (ret)
6741 				break;
6742 		}
6743 
6744 		cond_resched();
6745 	}
6746 
6747 	return ret;
6748 }
6749 
6750 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6751 					   u64 len, u64 orig_start,
6752 					   u64 block_start, u64 block_len,
6753 					   u64 orig_block_len, u64 ram_bytes,
6754 					   int type)
6755 {
6756 	struct extent_map_tree *em_tree;
6757 	struct extent_map *em;
6758 	struct btrfs_root *root = BTRFS_I(inode)->root;
6759 	int ret;
6760 
6761 	em_tree = &BTRFS_I(inode)->extent_tree;
6762 	em = alloc_extent_map();
6763 	if (!em)
6764 		return ERR_PTR(-ENOMEM);
6765 
6766 	em->start = start;
6767 	em->orig_start = orig_start;
6768 	em->mod_start = start;
6769 	em->mod_len = len;
6770 	em->len = len;
6771 	em->block_len = block_len;
6772 	em->block_start = block_start;
6773 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6774 	em->orig_block_len = orig_block_len;
6775 	em->ram_bytes = ram_bytes;
6776 	em->generation = -1;
6777 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
6778 	if (type == BTRFS_ORDERED_PREALLOC)
6779 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
6780 
6781 	do {
6782 		btrfs_drop_extent_cache(inode, em->start,
6783 				em->start + em->len - 1, 0);
6784 		write_lock(&em_tree->lock);
6785 		ret = add_extent_mapping(em_tree, em, 1);
6786 		write_unlock(&em_tree->lock);
6787 	} while (ret == -EEXIST);
6788 
6789 	if (ret) {
6790 		free_extent_map(em);
6791 		return ERR_PTR(ret);
6792 	}
6793 
6794 	return em;
6795 }
6796 
6797 
6798 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6799 				   struct buffer_head *bh_result, int create)
6800 {
6801 	struct extent_map *em;
6802 	struct btrfs_root *root = BTRFS_I(inode)->root;
6803 	struct extent_state *cached_state = NULL;
6804 	u64 start = iblock << inode->i_blkbits;
6805 	u64 lockstart, lockend;
6806 	u64 len = bh_result->b_size;
6807 	int unlock_bits = EXTENT_LOCKED;
6808 	int ret = 0;
6809 
6810 	if (create)
6811 		unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6812 	else
6813 		len = min_t(u64, len, root->sectorsize);
6814 
6815 	lockstart = start;
6816 	lockend = start + len - 1;
6817 
6818 	/*
6819 	 * If this errors out it's because we couldn't invalidate pagecache for
6820 	 * this range and we need to fallback to buffered.
6821 	 */
6822 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6823 		return -ENOTBLK;
6824 
6825 	em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6826 	if (IS_ERR(em)) {
6827 		ret = PTR_ERR(em);
6828 		goto unlock_err;
6829 	}
6830 
6831 	/*
6832 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6833 	 * io.  INLINE is special, and we could probably kludge it in here, but
6834 	 * it's still buffered so for safety lets just fall back to the generic
6835 	 * buffered path.
6836 	 *
6837 	 * For COMPRESSED we _have_ to read the entire extent in so we can
6838 	 * decompress it, so there will be buffering required no matter what we
6839 	 * do, so go ahead and fallback to buffered.
6840 	 *
6841 	 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6842 	 * to buffered IO.  Don't blame me, this is the price we pay for using
6843 	 * the generic code.
6844 	 */
6845 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6846 	    em->block_start == EXTENT_MAP_INLINE) {
6847 		free_extent_map(em);
6848 		ret = -ENOTBLK;
6849 		goto unlock_err;
6850 	}
6851 
6852 	/* Just a good old fashioned hole, return */
6853 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6854 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6855 		free_extent_map(em);
6856 		goto unlock_err;
6857 	}
6858 
6859 	/*
6860 	 * We don't allocate a new extent in the following cases
6861 	 *
6862 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6863 	 * existing extent.
6864 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
6865 	 * just use the extent.
6866 	 *
6867 	 */
6868 	if (!create) {
6869 		len = min(len, em->len - (start - em->start));
6870 		lockstart = start + len;
6871 		goto unlock;
6872 	}
6873 
6874 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6875 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6876 	     em->block_start != EXTENT_MAP_HOLE)) {
6877 		int type;
6878 		int ret;
6879 		u64 block_start, orig_start, orig_block_len, ram_bytes;
6880 
6881 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6882 			type = BTRFS_ORDERED_PREALLOC;
6883 		else
6884 			type = BTRFS_ORDERED_NOCOW;
6885 		len = min(len, em->len - (start - em->start));
6886 		block_start = em->block_start + (start - em->start);
6887 
6888 		if (can_nocow_extent(inode, start, &len, &orig_start,
6889 				     &orig_block_len, &ram_bytes) == 1) {
6890 			if (type == BTRFS_ORDERED_PREALLOC) {
6891 				free_extent_map(em);
6892 				em = create_pinned_em(inode, start, len,
6893 						       orig_start,
6894 						       block_start, len,
6895 						       orig_block_len,
6896 						       ram_bytes, type);
6897 				if (IS_ERR(em))
6898 					goto unlock_err;
6899 			}
6900 
6901 			ret = btrfs_add_ordered_extent_dio(inode, start,
6902 					   block_start, len, len, type);
6903 			if (ret) {
6904 				free_extent_map(em);
6905 				goto unlock_err;
6906 			}
6907 			goto unlock;
6908 		}
6909 	}
6910 
6911 	/*
6912 	 * this will cow the extent, reset the len in case we changed
6913 	 * it above
6914 	 */
6915 	len = bh_result->b_size;
6916 	free_extent_map(em);
6917 	em = btrfs_new_extent_direct(inode, start, len);
6918 	if (IS_ERR(em)) {
6919 		ret = PTR_ERR(em);
6920 		goto unlock_err;
6921 	}
6922 	len = min(len, em->len - (start - em->start));
6923 unlock:
6924 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6925 		inode->i_blkbits;
6926 	bh_result->b_size = len;
6927 	bh_result->b_bdev = em->bdev;
6928 	set_buffer_mapped(bh_result);
6929 	if (create) {
6930 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6931 			set_buffer_new(bh_result);
6932 
6933 		/*
6934 		 * Need to update the i_size under the extent lock so buffered
6935 		 * readers will get the updated i_size when we unlock.
6936 		 */
6937 		if (start + len > i_size_read(inode))
6938 			i_size_write(inode, start + len);
6939 
6940 		spin_lock(&BTRFS_I(inode)->lock);
6941 		BTRFS_I(inode)->outstanding_extents++;
6942 		spin_unlock(&BTRFS_I(inode)->lock);
6943 
6944 		ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6945 				     lockstart + len - 1, EXTENT_DELALLOC, NULL,
6946 				     &cached_state, GFP_NOFS);
6947 		BUG_ON(ret);
6948 	}
6949 
6950 	/*
6951 	 * In the case of write we need to clear and unlock the entire range,
6952 	 * in the case of read we need to unlock only the end area that we
6953 	 * aren't using if there is any left over space.
6954 	 */
6955 	if (lockstart < lockend) {
6956 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6957 				 lockend, unlock_bits, 1, 0,
6958 				 &cached_state, GFP_NOFS);
6959 	} else {
6960 		free_extent_state(cached_state);
6961 	}
6962 
6963 	free_extent_map(em);
6964 
6965 	return 0;
6966 
6967 unlock_err:
6968 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6969 			 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6970 	return ret;
6971 }
6972 
6973 static void btrfs_endio_direct_read(struct bio *bio, int err)
6974 {
6975 	struct btrfs_dio_private *dip = bio->bi_private;
6976 	struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6977 	struct bio_vec *bvec = bio->bi_io_vec;
6978 	struct inode *inode = dip->inode;
6979 	struct btrfs_root *root = BTRFS_I(inode)->root;
6980 	struct bio *dio_bio;
6981 	u32 *csums = (u32 *)dip->csum;
6982 	int index = 0;
6983 	u64 start;
6984 
6985 	start = dip->logical_offset;
6986 	do {
6987 		if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6988 			struct page *page = bvec->bv_page;
6989 			char *kaddr;
6990 			u32 csum = ~(u32)0;
6991 			unsigned long flags;
6992 
6993 			local_irq_save(flags);
6994 			kaddr = kmap_atomic(page);
6995 			csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6996 					       csum, bvec->bv_len);
6997 			btrfs_csum_final(csum, (char *)&csum);
6998 			kunmap_atomic(kaddr);
6999 			local_irq_restore(flags);
7000 
7001 			flush_dcache_page(bvec->bv_page);
7002 			if (csum != csums[index]) {
7003 				btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7004 					  btrfs_ino(inode), start, csum,
7005 					  csums[index]);
7006 				err = -EIO;
7007 			}
7008 		}
7009 
7010 		start += bvec->bv_len;
7011 		bvec++;
7012 		index++;
7013 	} while (bvec <= bvec_end);
7014 
7015 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7016 		      dip->logical_offset + dip->bytes - 1);
7017 	dio_bio = dip->dio_bio;
7018 
7019 	kfree(dip);
7020 
7021 	/* If we had a csum failure make sure to clear the uptodate flag */
7022 	if (err)
7023 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7024 	dio_end_io(dio_bio, err);
7025 	bio_put(bio);
7026 }
7027 
7028 static void btrfs_endio_direct_write(struct bio *bio, int err)
7029 {
7030 	struct btrfs_dio_private *dip = bio->bi_private;
7031 	struct inode *inode = dip->inode;
7032 	struct btrfs_root *root = BTRFS_I(inode)->root;
7033 	struct btrfs_ordered_extent *ordered = NULL;
7034 	u64 ordered_offset = dip->logical_offset;
7035 	u64 ordered_bytes = dip->bytes;
7036 	struct bio *dio_bio;
7037 	int ret;
7038 
7039 	if (err)
7040 		goto out_done;
7041 again:
7042 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7043 						   &ordered_offset,
7044 						   ordered_bytes, !err);
7045 	if (!ret)
7046 		goto out_test;
7047 
7048 	btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7049 	btrfs_queue_work(root->fs_info->endio_write_workers,
7050 			 &ordered->work);
7051 out_test:
7052 	/*
7053 	 * our bio might span multiple ordered extents.  If we haven't
7054 	 * completed the accounting for the whole dio, go back and try again
7055 	 */
7056 	if (ordered_offset < dip->logical_offset + dip->bytes) {
7057 		ordered_bytes = dip->logical_offset + dip->bytes -
7058 			ordered_offset;
7059 		ordered = NULL;
7060 		goto again;
7061 	}
7062 out_done:
7063 	dio_bio = dip->dio_bio;
7064 
7065 	kfree(dip);
7066 
7067 	/* If we had an error make sure to clear the uptodate flag */
7068 	if (err)
7069 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7070 	dio_end_io(dio_bio, err);
7071 	bio_put(bio);
7072 }
7073 
7074 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7075 				    struct bio *bio, int mirror_num,
7076 				    unsigned long bio_flags, u64 offset)
7077 {
7078 	int ret;
7079 	struct btrfs_root *root = BTRFS_I(inode)->root;
7080 	ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7081 	BUG_ON(ret); /* -ENOMEM */
7082 	return 0;
7083 }
7084 
7085 static void btrfs_end_dio_bio(struct bio *bio, int err)
7086 {
7087 	struct btrfs_dio_private *dip = bio->bi_private;
7088 
7089 	if (err) {
7090 		btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7091 			  "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7092 		      btrfs_ino(dip->inode), bio->bi_rw,
7093 		      (unsigned long long)bio->bi_sector, bio->bi_size, err);
7094 		dip->errors = 1;
7095 
7096 		/*
7097 		 * before atomic variable goto zero, we must make sure
7098 		 * dip->errors is perceived to be set.
7099 		 */
7100 		smp_mb__before_atomic_dec();
7101 	}
7102 
7103 	/* if there are more bios still pending for this dio, just exit */
7104 	if (!atomic_dec_and_test(&dip->pending_bios))
7105 		goto out;
7106 
7107 	if (dip->errors) {
7108 		bio_io_error(dip->orig_bio);
7109 	} else {
7110 		set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7111 		bio_endio(dip->orig_bio, 0);
7112 	}
7113 out:
7114 	bio_put(bio);
7115 }
7116 
7117 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7118 				       u64 first_sector, gfp_t gfp_flags)
7119 {
7120 	int nr_vecs = bio_get_nr_vecs(bdev);
7121 	return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7122 }
7123 
7124 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7125 					 int rw, u64 file_offset, int skip_sum,
7126 					 int async_submit)
7127 {
7128 	struct btrfs_dio_private *dip = bio->bi_private;
7129 	int write = rw & REQ_WRITE;
7130 	struct btrfs_root *root = BTRFS_I(inode)->root;
7131 	int ret;
7132 
7133 	if (async_submit)
7134 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7135 
7136 	bio_get(bio);
7137 
7138 	if (!write) {
7139 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7140 		if (ret)
7141 			goto err;
7142 	}
7143 
7144 	if (skip_sum)
7145 		goto map;
7146 
7147 	if (write && async_submit) {
7148 		ret = btrfs_wq_submit_bio(root->fs_info,
7149 				   inode, rw, bio, 0, 0,
7150 				   file_offset,
7151 				   __btrfs_submit_bio_start_direct_io,
7152 				   __btrfs_submit_bio_done);
7153 		goto err;
7154 	} else if (write) {
7155 		/*
7156 		 * If we aren't doing async submit, calculate the csum of the
7157 		 * bio now.
7158 		 */
7159 		ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7160 		if (ret)
7161 			goto err;
7162 	} else if (!skip_sum) {
7163 		ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7164 						file_offset);
7165 		if (ret)
7166 			goto err;
7167 	}
7168 
7169 map:
7170 	ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7171 err:
7172 	bio_put(bio);
7173 	return ret;
7174 }
7175 
7176 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7177 				    int skip_sum)
7178 {
7179 	struct inode *inode = dip->inode;
7180 	struct btrfs_root *root = BTRFS_I(inode)->root;
7181 	struct bio *bio;
7182 	struct bio *orig_bio = dip->orig_bio;
7183 	struct bio_vec *bvec = orig_bio->bi_io_vec;
7184 	u64 start_sector = orig_bio->bi_sector;
7185 	u64 file_offset = dip->logical_offset;
7186 	u64 submit_len = 0;
7187 	u64 map_length;
7188 	int nr_pages = 0;
7189 	int ret = 0;
7190 	int async_submit = 0;
7191 
7192 	map_length = orig_bio->bi_size;
7193 	ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7194 			      &map_length, NULL, 0);
7195 	if (ret) {
7196 		bio_put(orig_bio);
7197 		return -EIO;
7198 	}
7199 
7200 	if (map_length >= orig_bio->bi_size) {
7201 		bio = orig_bio;
7202 		goto submit;
7203 	}
7204 
7205 	/* async crcs make it difficult to collect full stripe writes. */
7206 	if (btrfs_get_alloc_profile(root, 1) &
7207 	    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7208 		async_submit = 0;
7209 	else
7210 		async_submit = 1;
7211 
7212 	bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7213 	if (!bio)
7214 		return -ENOMEM;
7215 	bio->bi_private = dip;
7216 	bio->bi_end_io = btrfs_end_dio_bio;
7217 	atomic_inc(&dip->pending_bios);
7218 
7219 	while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7220 		if (unlikely(map_length < submit_len + bvec->bv_len ||
7221 		    bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7222 				 bvec->bv_offset) < bvec->bv_len)) {
7223 			/*
7224 			 * inc the count before we submit the bio so
7225 			 * we know the end IO handler won't happen before
7226 			 * we inc the count. Otherwise, the dip might get freed
7227 			 * before we're done setting it up
7228 			 */
7229 			atomic_inc(&dip->pending_bios);
7230 			ret = __btrfs_submit_dio_bio(bio, inode, rw,
7231 						     file_offset, skip_sum,
7232 						     async_submit);
7233 			if (ret) {
7234 				bio_put(bio);
7235 				atomic_dec(&dip->pending_bios);
7236 				goto out_err;
7237 			}
7238 
7239 			start_sector += submit_len >> 9;
7240 			file_offset += submit_len;
7241 
7242 			submit_len = 0;
7243 			nr_pages = 0;
7244 
7245 			bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7246 						  start_sector, GFP_NOFS);
7247 			if (!bio)
7248 				goto out_err;
7249 			bio->bi_private = dip;
7250 			bio->bi_end_io = btrfs_end_dio_bio;
7251 
7252 			map_length = orig_bio->bi_size;
7253 			ret = btrfs_map_block(root->fs_info, rw,
7254 					      start_sector << 9,
7255 					      &map_length, NULL, 0);
7256 			if (ret) {
7257 				bio_put(bio);
7258 				goto out_err;
7259 			}
7260 		} else {
7261 			submit_len += bvec->bv_len;
7262 			nr_pages++;
7263 			bvec++;
7264 		}
7265 	}
7266 
7267 submit:
7268 	ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7269 				     async_submit);
7270 	if (!ret)
7271 		return 0;
7272 
7273 	bio_put(bio);
7274 out_err:
7275 	dip->errors = 1;
7276 	/*
7277 	 * before atomic variable goto zero, we must
7278 	 * make sure dip->errors is perceived to be set.
7279 	 */
7280 	smp_mb__before_atomic_dec();
7281 	if (atomic_dec_and_test(&dip->pending_bios))
7282 		bio_io_error(dip->orig_bio);
7283 
7284 	/* bio_end_io() will handle error, so we needn't return it */
7285 	return 0;
7286 }
7287 
7288 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7289 				struct inode *inode, loff_t file_offset)
7290 {
7291 	struct btrfs_root *root = BTRFS_I(inode)->root;
7292 	struct btrfs_dio_private *dip;
7293 	struct bio *io_bio;
7294 	int skip_sum;
7295 	int sum_len;
7296 	int write = rw & REQ_WRITE;
7297 	int ret = 0;
7298 	u16 csum_size;
7299 
7300 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7301 
7302 	io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7303 	if (!io_bio) {
7304 		ret = -ENOMEM;
7305 		goto free_ordered;
7306 	}
7307 
7308 	if (!skip_sum && !write) {
7309 		csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7310 		sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7311 		sum_len *= csum_size;
7312 	} else {
7313 		sum_len = 0;
7314 	}
7315 
7316 	dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7317 	if (!dip) {
7318 		ret = -ENOMEM;
7319 		goto free_io_bio;
7320 	}
7321 
7322 	dip->private = dio_bio->bi_private;
7323 	dip->inode = inode;
7324 	dip->logical_offset = file_offset;
7325 	dip->bytes = dio_bio->bi_size;
7326 	dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7327 	io_bio->bi_private = dip;
7328 	dip->errors = 0;
7329 	dip->orig_bio = io_bio;
7330 	dip->dio_bio = dio_bio;
7331 	atomic_set(&dip->pending_bios, 0);
7332 
7333 	if (write)
7334 		io_bio->bi_end_io = btrfs_endio_direct_write;
7335 	else
7336 		io_bio->bi_end_io = btrfs_endio_direct_read;
7337 
7338 	ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7339 	if (!ret)
7340 		return;
7341 
7342 free_io_bio:
7343 	bio_put(io_bio);
7344 
7345 free_ordered:
7346 	/*
7347 	 * If this is a write, we need to clean up the reserved space and kill
7348 	 * the ordered extent.
7349 	 */
7350 	if (write) {
7351 		struct btrfs_ordered_extent *ordered;
7352 		ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7353 		if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7354 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7355 			btrfs_free_reserved_extent(root, ordered->start,
7356 						   ordered->disk_len);
7357 		btrfs_put_ordered_extent(ordered);
7358 		btrfs_put_ordered_extent(ordered);
7359 	}
7360 	bio_endio(dio_bio, ret);
7361 }
7362 
7363 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7364 			const struct iovec *iov, loff_t offset,
7365 			unsigned long nr_segs)
7366 {
7367 	int seg;
7368 	int i;
7369 	size_t size;
7370 	unsigned long addr;
7371 	unsigned blocksize_mask = root->sectorsize - 1;
7372 	ssize_t retval = -EINVAL;
7373 	loff_t end = offset;
7374 
7375 	if (offset & blocksize_mask)
7376 		goto out;
7377 
7378 	/* Check the memory alignment.  Blocks cannot straddle pages */
7379 	for (seg = 0; seg < nr_segs; seg++) {
7380 		addr = (unsigned long)iov[seg].iov_base;
7381 		size = iov[seg].iov_len;
7382 		end += size;
7383 		if ((addr & blocksize_mask) || (size & blocksize_mask))
7384 			goto out;
7385 
7386 		/* If this is a write we don't need to check anymore */
7387 		if (rw & WRITE)
7388 			continue;
7389 
7390 		/*
7391 		 * Check to make sure we don't have duplicate iov_base's in this
7392 		 * iovec, if so return EINVAL, otherwise we'll get csum errors
7393 		 * when reading back.
7394 		 */
7395 		for (i = seg + 1; i < nr_segs; i++) {
7396 			if (iov[seg].iov_base == iov[i].iov_base)
7397 				goto out;
7398 		}
7399 	}
7400 	retval = 0;
7401 out:
7402 	return retval;
7403 }
7404 
7405 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7406 			const struct iovec *iov, loff_t offset,
7407 			unsigned long nr_segs)
7408 {
7409 	struct file *file = iocb->ki_filp;
7410 	struct inode *inode = file->f_mapping->host;
7411 	size_t count = 0;
7412 	int flags = 0;
7413 	bool wakeup = true;
7414 	bool relock = false;
7415 	ssize_t ret;
7416 
7417 	if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7418 			    offset, nr_segs))
7419 		return 0;
7420 
7421 	atomic_inc(&inode->i_dio_count);
7422 	smp_mb__after_atomic_inc();
7423 
7424 	/*
7425 	 * The generic stuff only does filemap_write_and_wait_range, which
7426 	 * isn't enough if we've written compressed pages to this area, so
7427 	 * we need to flush the dirty pages again to make absolutely sure
7428 	 * that any outstanding dirty pages are on disk.
7429 	 */
7430 	count = iov_length(iov, nr_segs);
7431 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7432 		     &BTRFS_I(inode)->runtime_flags))
7433 		filemap_fdatawrite_range(inode->i_mapping, offset, count);
7434 
7435 	if (rw & WRITE) {
7436 		/*
7437 		 * If the write DIO is beyond the EOF, we need update
7438 		 * the isize, but it is protected by i_mutex. So we can
7439 		 * not unlock the i_mutex at this case.
7440 		 */
7441 		if (offset + count <= inode->i_size) {
7442 			mutex_unlock(&inode->i_mutex);
7443 			relock = true;
7444 		}
7445 		ret = btrfs_delalloc_reserve_space(inode, count);
7446 		if (ret)
7447 			goto out;
7448 	} else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7449 				     &BTRFS_I(inode)->runtime_flags))) {
7450 		inode_dio_done(inode);
7451 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
7452 		wakeup = false;
7453 	}
7454 
7455 	ret = __blockdev_direct_IO(rw, iocb, inode,
7456 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7457 			iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7458 			btrfs_submit_direct, flags);
7459 	if (rw & WRITE) {
7460 		if (ret < 0 && ret != -EIOCBQUEUED)
7461 			btrfs_delalloc_release_space(inode, count);
7462 		else if (ret >= 0 && (size_t)ret < count)
7463 			btrfs_delalloc_release_space(inode,
7464 						     count - (size_t)ret);
7465 		else
7466 			btrfs_delalloc_release_metadata(inode, 0);
7467 	}
7468 out:
7469 	if (wakeup)
7470 		inode_dio_done(inode);
7471 	if (relock)
7472 		mutex_lock(&inode->i_mutex);
7473 
7474 	return ret;
7475 }
7476 
7477 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
7478 
7479 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7480 		__u64 start, __u64 len)
7481 {
7482 	int	ret;
7483 
7484 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7485 	if (ret)
7486 		return ret;
7487 
7488 	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7489 }
7490 
7491 int btrfs_readpage(struct file *file, struct page *page)
7492 {
7493 	struct extent_io_tree *tree;
7494 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7495 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7496 }
7497 
7498 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7499 {
7500 	struct extent_io_tree *tree;
7501 
7502 
7503 	if (current->flags & PF_MEMALLOC) {
7504 		redirty_page_for_writepage(wbc, page);
7505 		unlock_page(page);
7506 		return 0;
7507 	}
7508 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7509 	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7510 }
7511 
7512 static int btrfs_writepages(struct address_space *mapping,
7513 			    struct writeback_control *wbc)
7514 {
7515 	struct extent_io_tree *tree;
7516 
7517 	tree = &BTRFS_I(mapping->host)->io_tree;
7518 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7519 }
7520 
7521 static int
7522 btrfs_readpages(struct file *file, struct address_space *mapping,
7523 		struct list_head *pages, unsigned nr_pages)
7524 {
7525 	struct extent_io_tree *tree;
7526 	tree = &BTRFS_I(mapping->host)->io_tree;
7527 	return extent_readpages(tree, mapping, pages, nr_pages,
7528 				btrfs_get_extent);
7529 }
7530 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7531 {
7532 	struct extent_io_tree *tree;
7533 	struct extent_map_tree *map;
7534 	int ret;
7535 
7536 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7537 	map = &BTRFS_I(page->mapping->host)->extent_tree;
7538 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7539 	if (ret == 1) {
7540 		ClearPagePrivate(page);
7541 		set_page_private(page, 0);
7542 		page_cache_release(page);
7543 	}
7544 	return ret;
7545 }
7546 
7547 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7548 {
7549 	if (PageWriteback(page) || PageDirty(page))
7550 		return 0;
7551 	return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7552 }
7553 
7554 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7555 				 unsigned int length)
7556 {
7557 	struct inode *inode = page->mapping->host;
7558 	struct extent_io_tree *tree;
7559 	struct btrfs_ordered_extent *ordered;
7560 	struct extent_state *cached_state = NULL;
7561 	u64 page_start = page_offset(page);
7562 	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7563 	int inode_evicting = inode->i_state & I_FREEING;
7564 
7565 	/*
7566 	 * we have the page locked, so new writeback can't start,
7567 	 * and the dirty bit won't be cleared while we are here.
7568 	 *
7569 	 * Wait for IO on this page so that we can safely clear
7570 	 * the PagePrivate2 bit and do ordered accounting
7571 	 */
7572 	wait_on_page_writeback(page);
7573 
7574 	tree = &BTRFS_I(inode)->io_tree;
7575 	if (offset) {
7576 		btrfs_releasepage(page, GFP_NOFS);
7577 		return;
7578 	}
7579 
7580 	if (!inode_evicting)
7581 		lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7582 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7583 	if (ordered) {
7584 		/*
7585 		 * IO on this page will never be started, so we need
7586 		 * to account for any ordered extents now
7587 		 */
7588 		if (!inode_evicting)
7589 			clear_extent_bit(tree, page_start, page_end,
7590 					 EXTENT_DIRTY | EXTENT_DELALLOC |
7591 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7592 					 EXTENT_DEFRAG, 1, 0, &cached_state,
7593 					 GFP_NOFS);
7594 		/*
7595 		 * whoever cleared the private bit is responsible
7596 		 * for the finish_ordered_io
7597 		 */
7598 		if (TestClearPagePrivate2(page)) {
7599 			struct btrfs_ordered_inode_tree *tree;
7600 			u64 new_len;
7601 
7602 			tree = &BTRFS_I(inode)->ordered_tree;
7603 
7604 			spin_lock_irq(&tree->lock);
7605 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7606 			new_len = page_start - ordered->file_offset;
7607 			if (new_len < ordered->truncated_len)
7608 				ordered->truncated_len = new_len;
7609 			spin_unlock_irq(&tree->lock);
7610 
7611 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
7612 							   page_start,
7613 							   PAGE_CACHE_SIZE, 1))
7614 				btrfs_finish_ordered_io(ordered);
7615 		}
7616 		btrfs_put_ordered_extent(ordered);
7617 		if (!inode_evicting) {
7618 			cached_state = NULL;
7619 			lock_extent_bits(tree, page_start, page_end, 0,
7620 					 &cached_state);
7621 		}
7622 	}
7623 
7624 	if (!inode_evicting) {
7625 		clear_extent_bit(tree, page_start, page_end,
7626 				 EXTENT_LOCKED | EXTENT_DIRTY |
7627 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7628 				 EXTENT_DEFRAG, 1, 1,
7629 				 &cached_state, GFP_NOFS);
7630 
7631 		__btrfs_releasepage(page, GFP_NOFS);
7632 	}
7633 
7634 	ClearPageChecked(page);
7635 	if (PagePrivate(page)) {
7636 		ClearPagePrivate(page);
7637 		set_page_private(page, 0);
7638 		page_cache_release(page);
7639 	}
7640 }
7641 
7642 /*
7643  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7644  * called from a page fault handler when a page is first dirtied. Hence we must
7645  * be careful to check for EOF conditions here. We set the page up correctly
7646  * for a written page which means we get ENOSPC checking when writing into
7647  * holes and correct delalloc and unwritten extent mapping on filesystems that
7648  * support these features.
7649  *
7650  * We are not allowed to take the i_mutex here so we have to play games to
7651  * protect against truncate races as the page could now be beyond EOF.  Because
7652  * vmtruncate() writes the inode size before removing pages, once we have the
7653  * page lock we can determine safely if the page is beyond EOF. If it is not
7654  * beyond EOF, then the page is guaranteed safe against truncation until we
7655  * unlock the page.
7656  */
7657 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7658 {
7659 	struct page *page = vmf->page;
7660 	struct inode *inode = file_inode(vma->vm_file);
7661 	struct btrfs_root *root = BTRFS_I(inode)->root;
7662 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7663 	struct btrfs_ordered_extent *ordered;
7664 	struct extent_state *cached_state = NULL;
7665 	char *kaddr;
7666 	unsigned long zero_start;
7667 	loff_t size;
7668 	int ret;
7669 	int reserved = 0;
7670 	u64 page_start;
7671 	u64 page_end;
7672 
7673 	sb_start_pagefault(inode->i_sb);
7674 	ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7675 	if (!ret) {
7676 		ret = file_update_time(vma->vm_file);
7677 		reserved = 1;
7678 	}
7679 	if (ret) {
7680 		if (ret == -ENOMEM)
7681 			ret = VM_FAULT_OOM;
7682 		else /* -ENOSPC, -EIO, etc */
7683 			ret = VM_FAULT_SIGBUS;
7684 		if (reserved)
7685 			goto out;
7686 		goto out_noreserve;
7687 	}
7688 
7689 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7690 again:
7691 	lock_page(page);
7692 	size = i_size_read(inode);
7693 	page_start = page_offset(page);
7694 	page_end = page_start + PAGE_CACHE_SIZE - 1;
7695 
7696 	if ((page->mapping != inode->i_mapping) ||
7697 	    (page_start >= size)) {
7698 		/* page got truncated out from underneath us */
7699 		goto out_unlock;
7700 	}
7701 	wait_on_page_writeback(page);
7702 
7703 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7704 	set_page_extent_mapped(page);
7705 
7706 	/*
7707 	 * we can't set the delalloc bits if there are pending ordered
7708 	 * extents.  Drop our locks and wait for them to finish
7709 	 */
7710 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7711 	if (ordered) {
7712 		unlock_extent_cached(io_tree, page_start, page_end,
7713 				     &cached_state, GFP_NOFS);
7714 		unlock_page(page);
7715 		btrfs_start_ordered_extent(inode, ordered, 1);
7716 		btrfs_put_ordered_extent(ordered);
7717 		goto again;
7718 	}
7719 
7720 	/*
7721 	 * XXX - page_mkwrite gets called every time the page is dirtied, even
7722 	 * if it was already dirty, so for space accounting reasons we need to
7723 	 * clear any delalloc bits for the range we are fixing to save.  There
7724 	 * is probably a better way to do this, but for now keep consistent with
7725 	 * prepare_pages in the normal write path.
7726 	 */
7727 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7728 			  EXTENT_DIRTY | EXTENT_DELALLOC |
7729 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7730 			  0, 0, &cached_state, GFP_NOFS);
7731 
7732 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7733 					&cached_state);
7734 	if (ret) {
7735 		unlock_extent_cached(io_tree, page_start, page_end,
7736 				     &cached_state, GFP_NOFS);
7737 		ret = VM_FAULT_SIGBUS;
7738 		goto out_unlock;
7739 	}
7740 	ret = 0;
7741 
7742 	/* page is wholly or partially inside EOF */
7743 	if (page_start + PAGE_CACHE_SIZE > size)
7744 		zero_start = size & ~PAGE_CACHE_MASK;
7745 	else
7746 		zero_start = PAGE_CACHE_SIZE;
7747 
7748 	if (zero_start != PAGE_CACHE_SIZE) {
7749 		kaddr = kmap(page);
7750 		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7751 		flush_dcache_page(page);
7752 		kunmap(page);
7753 	}
7754 	ClearPageChecked(page);
7755 	set_page_dirty(page);
7756 	SetPageUptodate(page);
7757 
7758 	BTRFS_I(inode)->last_trans = root->fs_info->generation;
7759 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7760 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7761 
7762 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7763 
7764 out_unlock:
7765 	if (!ret) {
7766 		sb_end_pagefault(inode->i_sb);
7767 		return VM_FAULT_LOCKED;
7768 	}
7769 	unlock_page(page);
7770 out:
7771 	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7772 out_noreserve:
7773 	sb_end_pagefault(inode->i_sb);
7774 	return ret;
7775 }
7776 
7777 static int btrfs_truncate(struct inode *inode)
7778 {
7779 	struct btrfs_root *root = BTRFS_I(inode)->root;
7780 	struct btrfs_block_rsv *rsv;
7781 	int ret = 0;
7782 	int err = 0;
7783 	struct btrfs_trans_handle *trans;
7784 	u64 mask = root->sectorsize - 1;
7785 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7786 
7787 	ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7788 				       (u64)-1);
7789 	if (ret)
7790 		return ret;
7791 
7792 	/*
7793 	 * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7794 	 * 3 things going on here
7795 	 *
7796 	 * 1) We need to reserve space for our orphan item and the space to
7797 	 * delete our orphan item.  Lord knows we don't want to have a dangling
7798 	 * orphan item because we didn't reserve space to remove it.
7799 	 *
7800 	 * 2) We need to reserve space to update our inode.
7801 	 *
7802 	 * 3) We need to have something to cache all the space that is going to
7803 	 * be free'd up by the truncate operation, but also have some slack
7804 	 * space reserved in case it uses space during the truncate (thank you
7805 	 * very much snapshotting).
7806 	 *
7807 	 * And we need these to all be seperate.  The fact is we can use alot of
7808 	 * space doing the truncate, and we have no earthly idea how much space
7809 	 * we will use, so we need the truncate reservation to be seperate so it
7810 	 * doesn't end up using space reserved for updating the inode or
7811 	 * removing the orphan item.  We also need to be able to stop the
7812 	 * transaction and start a new one, which means we need to be able to
7813 	 * update the inode several times, and we have no idea of knowing how
7814 	 * many times that will be, so we can't just reserve 1 item for the
7815 	 * entirety of the opration, so that has to be done seperately as well.
7816 	 * Then there is the orphan item, which does indeed need to be held on
7817 	 * to for the whole operation, and we need nobody to touch this reserved
7818 	 * space except the orphan code.
7819 	 *
7820 	 * So that leaves us with
7821 	 *
7822 	 * 1) root->orphan_block_rsv - for the orphan deletion.
7823 	 * 2) rsv - for the truncate reservation, which we will steal from the
7824 	 * transaction reservation.
7825 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7826 	 * updating the inode.
7827 	 */
7828 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7829 	if (!rsv)
7830 		return -ENOMEM;
7831 	rsv->size = min_size;
7832 	rsv->failfast = 1;
7833 
7834 	/*
7835 	 * 1 for the truncate slack space
7836 	 * 1 for updating the inode.
7837 	 */
7838 	trans = btrfs_start_transaction(root, 2);
7839 	if (IS_ERR(trans)) {
7840 		err = PTR_ERR(trans);
7841 		goto out;
7842 	}
7843 
7844 	/* Migrate the slack space for the truncate to our reserve */
7845 	ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7846 				      min_size);
7847 	BUG_ON(ret);
7848 
7849 	/*
7850 	 * setattr is responsible for setting the ordered_data_close flag,
7851 	 * but that is only tested during the last file release.  That
7852 	 * could happen well after the next commit, leaving a great big
7853 	 * window where new writes may get lost if someone chooses to write
7854 	 * to this file after truncating to zero
7855 	 *
7856 	 * The inode doesn't have any dirty data here, and so if we commit
7857 	 * this is a noop.  If someone immediately starts writing to the inode
7858 	 * it is very likely we'll catch some of their writes in this
7859 	 * transaction, and the commit will find this file on the ordered
7860 	 * data list with good things to send down.
7861 	 *
7862 	 * This is a best effort solution, there is still a window where
7863 	 * using truncate to replace the contents of the file will
7864 	 * end up with a zero length file after a crash.
7865 	 */
7866 	if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7867 					   &BTRFS_I(inode)->runtime_flags))
7868 		btrfs_add_ordered_operation(trans, root, inode);
7869 
7870 	/*
7871 	 * So if we truncate and then write and fsync we normally would just
7872 	 * write the extents that changed, which is a problem if we need to
7873 	 * first truncate that entire inode.  So set this flag so we write out
7874 	 * all of the extents in the inode to the sync log so we're completely
7875 	 * safe.
7876 	 */
7877 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7878 	trans->block_rsv = rsv;
7879 
7880 	while (1) {
7881 		ret = btrfs_truncate_inode_items(trans, root, inode,
7882 						 inode->i_size,
7883 						 BTRFS_EXTENT_DATA_KEY);
7884 		if (ret != -ENOSPC) {
7885 			err = ret;
7886 			break;
7887 		}
7888 
7889 		trans->block_rsv = &root->fs_info->trans_block_rsv;
7890 		ret = btrfs_update_inode(trans, root, inode);
7891 		if (ret) {
7892 			err = ret;
7893 			break;
7894 		}
7895 
7896 		btrfs_end_transaction(trans, root);
7897 		btrfs_btree_balance_dirty(root);
7898 
7899 		trans = btrfs_start_transaction(root, 2);
7900 		if (IS_ERR(trans)) {
7901 			ret = err = PTR_ERR(trans);
7902 			trans = NULL;
7903 			break;
7904 		}
7905 
7906 		ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7907 					      rsv, min_size);
7908 		BUG_ON(ret);	/* shouldn't happen */
7909 		trans->block_rsv = rsv;
7910 	}
7911 
7912 	if (ret == 0 && inode->i_nlink > 0) {
7913 		trans->block_rsv = root->orphan_block_rsv;
7914 		ret = btrfs_orphan_del(trans, inode);
7915 		if (ret)
7916 			err = ret;
7917 	}
7918 
7919 	if (trans) {
7920 		trans->block_rsv = &root->fs_info->trans_block_rsv;
7921 		ret = btrfs_update_inode(trans, root, inode);
7922 		if (ret && !err)
7923 			err = ret;
7924 
7925 		ret = btrfs_end_transaction(trans, root);
7926 		btrfs_btree_balance_dirty(root);
7927 	}
7928 
7929 out:
7930 	btrfs_free_block_rsv(root, rsv);
7931 
7932 	if (ret && !err)
7933 		err = ret;
7934 
7935 	return err;
7936 }
7937 
7938 /*
7939  * create a new subvolume directory/inode (helper for the ioctl).
7940  */
7941 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7942 			     struct btrfs_root *new_root,
7943 			     struct btrfs_root *parent_root,
7944 			     u64 new_dirid)
7945 {
7946 	struct inode *inode;
7947 	int err;
7948 	u64 index = 0;
7949 
7950 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7951 				new_dirid, new_dirid,
7952 				S_IFDIR | (~current_umask() & S_IRWXUGO),
7953 				&index);
7954 	if (IS_ERR(inode))
7955 		return PTR_ERR(inode);
7956 	inode->i_op = &btrfs_dir_inode_operations;
7957 	inode->i_fop = &btrfs_dir_file_operations;
7958 
7959 	set_nlink(inode, 1);
7960 	btrfs_i_size_write(inode, 0);
7961 
7962 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7963 	if (err)
7964 		btrfs_err(new_root->fs_info,
7965 			  "error inheriting subvolume %llu properties: %d\n",
7966 			  new_root->root_key.objectid, err);
7967 
7968 	err = btrfs_update_inode(trans, new_root, inode);
7969 
7970 	iput(inode);
7971 	return err;
7972 }
7973 
7974 struct inode *btrfs_alloc_inode(struct super_block *sb)
7975 {
7976 	struct btrfs_inode *ei;
7977 	struct inode *inode;
7978 
7979 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7980 	if (!ei)
7981 		return NULL;
7982 
7983 	ei->root = NULL;
7984 	ei->generation = 0;
7985 	ei->last_trans = 0;
7986 	ei->last_sub_trans = 0;
7987 	ei->logged_trans = 0;
7988 	ei->delalloc_bytes = 0;
7989 	ei->disk_i_size = 0;
7990 	ei->flags = 0;
7991 	ei->csum_bytes = 0;
7992 	ei->index_cnt = (u64)-1;
7993 	ei->dir_index = 0;
7994 	ei->last_unlink_trans = 0;
7995 	ei->last_log_commit = 0;
7996 
7997 	spin_lock_init(&ei->lock);
7998 	ei->outstanding_extents = 0;
7999 	ei->reserved_extents = 0;
8000 
8001 	ei->runtime_flags = 0;
8002 	ei->force_compress = BTRFS_COMPRESS_NONE;
8003 
8004 	ei->delayed_node = NULL;
8005 
8006 	inode = &ei->vfs_inode;
8007 	extent_map_tree_init(&ei->extent_tree);
8008 	extent_io_tree_init(&ei->io_tree, &inode->i_data);
8009 	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8010 	ei->io_tree.track_uptodate = 1;
8011 	ei->io_failure_tree.track_uptodate = 1;
8012 	atomic_set(&ei->sync_writers, 0);
8013 	mutex_init(&ei->log_mutex);
8014 	mutex_init(&ei->delalloc_mutex);
8015 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8016 	INIT_LIST_HEAD(&ei->delalloc_inodes);
8017 	INIT_LIST_HEAD(&ei->ordered_operations);
8018 	RB_CLEAR_NODE(&ei->rb_node);
8019 
8020 	return inode;
8021 }
8022 
8023 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8024 void btrfs_test_destroy_inode(struct inode *inode)
8025 {
8026 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8027 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8028 }
8029 #endif
8030 
8031 static void btrfs_i_callback(struct rcu_head *head)
8032 {
8033 	struct inode *inode = container_of(head, struct inode, i_rcu);
8034 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8035 }
8036 
8037 void btrfs_destroy_inode(struct inode *inode)
8038 {
8039 	struct btrfs_ordered_extent *ordered;
8040 	struct btrfs_root *root = BTRFS_I(inode)->root;
8041 
8042 	WARN_ON(!hlist_empty(&inode->i_dentry));
8043 	WARN_ON(inode->i_data.nrpages);
8044 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
8045 	WARN_ON(BTRFS_I(inode)->reserved_extents);
8046 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8047 	WARN_ON(BTRFS_I(inode)->csum_bytes);
8048 
8049 	/*
8050 	 * This can happen where we create an inode, but somebody else also
8051 	 * created the same inode and we need to destroy the one we already
8052 	 * created.
8053 	 */
8054 	if (!root)
8055 		goto free;
8056 
8057 	/*
8058 	 * Make sure we're properly removed from the ordered operation
8059 	 * lists.
8060 	 */
8061 	smp_mb();
8062 	if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8063 		spin_lock(&root->fs_info->ordered_root_lock);
8064 		list_del_init(&BTRFS_I(inode)->ordered_operations);
8065 		spin_unlock(&root->fs_info->ordered_root_lock);
8066 	}
8067 
8068 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8069 		     &BTRFS_I(inode)->runtime_flags)) {
8070 		btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8071 			btrfs_ino(inode));
8072 		atomic_dec(&root->orphan_inodes);
8073 	}
8074 
8075 	while (1) {
8076 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8077 		if (!ordered)
8078 			break;
8079 		else {
8080 			btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8081 				ordered->file_offset, ordered->len);
8082 			btrfs_remove_ordered_extent(inode, ordered);
8083 			btrfs_put_ordered_extent(ordered);
8084 			btrfs_put_ordered_extent(ordered);
8085 		}
8086 	}
8087 	inode_tree_del(inode);
8088 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8089 free:
8090 	call_rcu(&inode->i_rcu, btrfs_i_callback);
8091 }
8092 
8093 int btrfs_drop_inode(struct inode *inode)
8094 {
8095 	struct btrfs_root *root = BTRFS_I(inode)->root;
8096 
8097 	if (root == NULL)
8098 		return 1;
8099 
8100 	/* the snap/subvol tree is on deleting */
8101 	if (btrfs_root_refs(&root->root_item) == 0)
8102 		return 1;
8103 	else
8104 		return generic_drop_inode(inode);
8105 }
8106 
8107 static void init_once(void *foo)
8108 {
8109 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8110 
8111 	inode_init_once(&ei->vfs_inode);
8112 }
8113 
8114 void btrfs_destroy_cachep(void)
8115 {
8116 	/*
8117 	 * Make sure all delayed rcu free inodes are flushed before we
8118 	 * destroy cache.
8119 	 */
8120 	rcu_barrier();
8121 	if (btrfs_inode_cachep)
8122 		kmem_cache_destroy(btrfs_inode_cachep);
8123 	if (btrfs_trans_handle_cachep)
8124 		kmem_cache_destroy(btrfs_trans_handle_cachep);
8125 	if (btrfs_transaction_cachep)
8126 		kmem_cache_destroy(btrfs_transaction_cachep);
8127 	if (btrfs_path_cachep)
8128 		kmem_cache_destroy(btrfs_path_cachep);
8129 	if (btrfs_free_space_cachep)
8130 		kmem_cache_destroy(btrfs_free_space_cachep);
8131 	if (btrfs_delalloc_work_cachep)
8132 		kmem_cache_destroy(btrfs_delalloc_work_cachep);
8133 }
8134 
8135 int btrfs_init_cachep(void)
8136 {
8137 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8138 			sizeof(struct btrfs_inode), 0,
8139 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8140 	if (!btrfs_inode_cachep)
8141 		goto fail;
8142 
8143 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8144 			sizeof(struct btrfs_trans_handle), 0,
8145 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8146 	if (!btrfs_trans_handle_cachep)
8147 		goto fail;
8148 
8149 	btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8150 			sizeof(struct btrfs_transaction), 0,
8151 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8152 	if (!btrfs_transaction_cachep)
8153 		goto fail;
8154 
8155 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
8156 			sizeof(struct btrfs_path), 0,
8157 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8158 	if (!btrfs_path_cachep)
8159 		goto fail;
8160 
8161 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8162 			sizeof(struct btrfs_free_space), 0,
8163 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8164 	if (!btrfs_free_space_cachep)
8165 		goto fail;
8166 
8167 	btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8168 			sizeof(struct btrfs_delalloc_work), 0,
8169 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8170 			NULL);
8171 	if (!btrfs_delalloc_work_cachep)
8172 		goto fail;
8173 
8174 	return 0;
8175 fail:
8176 	btrfs_destroy_cachep();
8177 	return -ENOMEM;
8178 }
8179 
8180 static int btrfs_getattr(struct vfsmount *mnt,
8181 			 struct dentry *dentry, struct kstat *stat)
8182 {
8183 	u64 delalloc_bytes;
8184 	struct inode *inode = dentry->d_inode;
8185 	u32 blocksize = inode->i_sb->s_blocksize;
8186 
8187 	generic_fillattr(inode, stat);
8188 	stat->dev = BTRFS_I(inode)->root->anon_dev;
8189 	stat->blksize = PAGE_CACHE_SIZE;
8190 
8191 	spin_lock(&BTRFS_I(inode)->lock);
8192 	delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8193 	spin_unlock(&BTRFS_I(inode)->lock);
8194 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8195 			ALIGN(delalloc_bytes, blocksize)) >> 9;
8196 	return 0;
8197 }
8198 
8199 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8200 			   struct inode *new_dir, struct dentry *new_dentry)
8201 {
8202 	struct btrfs_trans_handle *trans;
8203 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
8204 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8205 	struct inode *new_inode = new_dentry->d_inode;
8206 	struct inode *old_inode = old_dentry->d_inode;
8207 	struct timespec ctime = CURRENT_TIME;
8208 	u64 index = 0;
8209 	u64 root_objectid;
8210 	int ret;
8211 	u64 old_ino = btrfs_ino(old_inode);
8212 
8213 	if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8214 		return -EPERM;
8215 
8216 	/* we only allow rename subvolume link between subvolumes */
8217 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8218 		return -EXDEV;
8219 
8220 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8221 	    (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8222 		return -ENOTEMPTY;
8223 
8224 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
8225 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8226 		return -ENOTEMPTY;
8227 
8228 
8229 	/* check for collisions, even if the  name isn't there */
8230 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8231 			     new_dentry->d_name.name,
8232 			     new_dentry->d_name.len);
8233 
8234 	if (ret) {
8235 		if (ret == -EEXIST) {
8236 			/* we shouldn't get
8237 			 * eexist without a new_inode */
8238 			if (WARN_ON(!new_inode)) {
8239 				return ret;
8240 			}
8241 		} else {
8242 			/* maybe -EOVERFLOW */
8243 			return ret;
8244 		}
8245 	}
8246 	ret = 0;
8247 
8248 	/*
8249 	 * we're using rename to replace one file with another.
8250 	 * and the replacement file is large.  Start IO on it now so
8251 	 * we don't add too much work to the end of the transaction
8252 	 */
8253 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8254 	    old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8255 		filemap_flush(old_inode->i_mapping);
8256 
8257 	/* close the racy window with snapshot create/destroy ioctl */
8258 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8259 		down_read(&root->fs_info->subvol_sem);
8260 	/*
8261 	 * We want to reserve the absolute worst case amount of items.  So if
8262 	 * both inodes are subvols and we need to unlink them then that would
8263 	 * require 4 item modifications, but if they are both normal inodes it
8264 	 * would require 5 item modifications, so we'll assume their normal
8265 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8266 	 * should cover the worst case number of items we'll modify.
8267 	 */
8268 	trans = btrfs_start_transaction(root, 11);
8269 	if (IS_ERR(trans)) {
8270                 ret = PTR_ERR(trans);
8271                 goto out_notrans;
8272         }
8273 
8274 	if (dest != root)
8275 		btrfs_record_root_in_trans(trans, dest);
8276 
8277 	ret = btrfs_set_inode_index(new_dir, &index);
8278 	if (ret)
8279 		goto out_fail;
8280 
8281 	BTRFS_I(old_inode)->dir_index = 0ULL;
8282 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8283 		/* force full log commit if subvolume involved. */
8284 		root->fs_info->last_trans_log_full_commit = trans->transid;
8285 	} else {
8286 		ret = btrfs_insert_inode_ref(trans, dest,
8287 					     new_dentry->d_name.name,
8288 					     new_dentry->d_name.len,
8289 					     old_ino,
8290 					     btrfs_ino(new_dir), index);
8291 		if (ret)
8292 			goto out_fail;
8293 		/*
8294 		 * this is an ugly little race, but the rename is required
8295 		 * to make sure that if we crash, the inode is either at the
8296 		 * old name or the new one.  pinning the log transaction lets
8297 		 * us make sure we don't allow a log commit to come in after
8298 		 * we unlink the name but before we add the new name back in.
8299 		 */
8300 		btrfs_pin_log_trans(root);
8301 	}
8302 	/*
8303 	 * make sure the inode gets flushed if it is replacing
8304 	 * something.
8305 	 */
8306 	if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8307 		btrfs_add_ordered_operation(trans, root, old_inode);
8308 
8309 	inode_inc_iversion(old_dir);
8310 	inode_inc_iversion(new_dir);
8311 	inode_inc_iversion(old_inode);
8312 	old_dir->i_ctime = old_dir->i_mtime = ctime;
8313 	new_dir->i_ctime = new_dir->i_mtime = ctime;
8314 	old_inode->i_ctime = ctime;
8315 
8316 	if (old_dentry->d_parent != new_dentry->d_parent)
8317 		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8318 
8319 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8320 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8321 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8322 					old_dentry->d_name.name,
8323 					old_dentry->d_name.len);
8324 	} else {
8325 		ret = __btrfs_unlink_inode(trans, root, old_dir,
8326 					old_dentry->d_inode,
8327 					old_dentry->d_name.name,
8328 					old_dentry->d_name.len);
8329 		if (!ret)
8330 			ret = btrfs_update_inode(trans, root, old_inode);
8331 	}
8332 	if (ret) {
8333 		btrfs_abort_transaction(trans, root, ret);
8334 		goto out_fail;
8335 	}
8336 
8337 	if (new_inode) {
8338 		inode_inc_iversion(new_inode);
8339 		new_inode->i_ctime = CURRENT_TIME;
8340 		if (unlikely(btrfs_ino(new_inode) ==
8341 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8342 			root_objectid = BTRFS_I(new_inode)->location.objectid;
8343 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
8344 						root_objectid,
8345 						new_dentry->d_name.name,
8346 						new_dentry->d_name.len);
8347 			BUG_ON(new_inode->i_nlink == 0);
8348 		} else {
8349 			ret = btrfs_unlink_inode(trans, dest, new_dir,
8350 						 new_dentry->d_inode,
8351 						 new_dentry->d_name.name,
8352 						 new_dentry->d_name.len);
8353 		}
8354 		if (!ret && new_inode->i_nlink == 0)
8355 			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8356 		if (ret) {
8357 			btrfs_abort_transaction(trans, root, ret);
8358 			goto out_fail;
8359 		}
8360 	}
8361 
8362 	ret = btrfs_add_link(trans, new_dir, old_inode,
8363 			     new_dentry->d_name.name,
8364 			     new_dentry->d_name.len, 0, index);
8365 	if (ret) {
8366 		btrfs_abort_transaction(trans, root, ret);
8367 		goto out_fail;
8368 	}
8369 
8370 	if (old_inode->i_nlink == 1)
8371 		BTRFS_I(old_inode)->dir_index = index;
8372 
8373 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8374 		struct dentry *parent = new_dentry->d_parent;
8375 		btrfs_log_new_name(trans, old_inode, old_dir, parent);
8376 		btrfs_end_log_trans(root);
8377 	}
8378 out_fail:
8379 	btrfs_end_transaction(trans, root);
8380 out_notrans:
8381 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8382 		up_read(&root->fs_info->subvol_sem);
8383 
8384 	return ret;
8385 }
8386 
8387 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8388 {
8389 	struct btrfs_delalloc_work *delalloc_work;
8390 	struct inode *inode;
8391 
8392 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
8393 				     work);
8394 	inode = delalloc_work->inode;
8395 	if (delalloc_work->wait) {
8396 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
8397 	} else {
8398 		filemap_flush(inode->i_mapping);
8399 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8400 			     &BTRFS_I(inode)->runtime_flags))
8401 			filemap_flush(inode->i_mapping);
8402 	}
8403 
8404 	if (delalloc_work->delay_iput)
8405 		btrfs_add_delayed_iput(inode);
8406 	else
8407 		iput(inode);
8408 	complete(&delalloc_work->completion);
8409 }
8410 
8411 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8412 						    int wait, int delay_iput)
8413 {
8414 	struct btrfs_delalloc_work *work;
8415 
8416 	work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8417 	if (!work)
8418 		return NULL;
8419 
8420 	init_completion(&work->completion);
8421 	INIT_LIST_HEAD(&work->list);
8422 	work->inode = inode;
8423 	work->wait = wait;
8424 	work->delay_iput = delay_iput;
8425 	btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8426 
8427 	return work;
8428 }
8429 
8430 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8431 {
8432 	wait_for_completion(&work->completion);
8433 	kmem_cache_free(btrfs_delalloc_work_cachep, work);
8434 }
8435 
8436 /*
8437  * some fairly slow code that needs optimization. This walks the list
8438  * of all the inodes with pending delalloc and forces them to disk.
8439  */
8440 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8441 {
8442 	struct btrfs_inode *binode;
8443 	struct inode *inode;
8444 	struct btrfs_delalloc_work *work, *next;
8445 	struct list_head works;
8446 	struct list_head splice;
8447 	int ret = 0;
8448 
8449 	INIT_LIST_HEAD(&works);
8450 	INIT_LIST_HEAD(&splice);
8451 
8452 	spin_lock(&root->delalloc_lock);
8453 	list_splice_init(&root->delalloc_inodes, &splice);
8454 	while (!list_empty(&splice)) {
8455 		binode = list_entry(splice.next, struct btrfs_inode,
8456 				    delalloc_inodes);
8457 
8458 		list_move_tail(&binode->delalloc_inodes,
8459 			       &root->delalloc_inodes);
8460 		inode = igrab(&binode->vfs_inode);
8461 		if (!inode) {
8462 			cond_resched_lock(&root->delalloc_lock);
8463 			continue;
8464 		}
8465 		spin_unlock(&root->delalloc_lock);
8466 
8467 		work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8468 		if (unlikely(!work)) {
8469 			if (delay_iput)
8470 				btrfs_add_delayed_iput(inode);
8471 			else
8472 				iput(inode);
8473 			ret = -ENOMEM;
8474 			goto out;
8475 		}
8476 		list_add_tail(&work->list, &works);
8477 		btrfs_queue_work(root->fs_info->flush_workers,
8478 				 &work->work);
8479 
8480 		cond_resched();
8481 		spin_lock(&root->delalloc_lock);
8482 	}
8483 	spin_unlock(&root->delalloc_lock);
8484 
8485 	list_for_each_entry_safe(work, next, &works, list) {
8486 		list_del_init(&work->list);
8487 		btrfs_wait_and_free_delalloc_work(work);
8488 	}
8489 	return 0;
8490 out:
8491 	list_for_each_entry_safe(work, next, &works, list) {
8492 		list_del_init(&work->list);
8493 		btrfs_wait_and_free_delalloc_work(work);
8494 	}
8495 
8496 	if (!list_empty_careful(&splice)) {
8497 		spin_lock(&root->delalloc_lock);
8498 		list_splice_tail(&splice, &root->delalloc_inodes);
8499 		spin_unlock(&root->delalloc_lock);
8500 	}
8501 	return ret;
8502 }
8503 
8504 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8505 {
8506 	int ret;
8507 
8508 	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8509 		return -EROFS;
8510 
8511 	ret = __start_delalloc_inodes(root, delay_iput);
8512 	/*
8513 	 * the filemap_flush will queue IO into the worker threads, but
8514 	 * we have to make sure the IO is actually started and that
8515 	 * ordered extents get created before we return
8516 	 */
8517 	atomic_inc(&root->fs_info->async_submit_draining);
8518 	while (atomic_read(&root->fs_info->nr_async_submits) ||
8519 	      atomic_read(&root->fs_info->async_delalloc_pages)) {
8520 		wait_event(root->fs_info->async_submit_wait,
8521 		   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8522 		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8523 	}
8524 	atomic_dec(&root->fs_info->async_submit_draining);
8525 	return ret;
8526 }
8527 
8528 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8529 {
8530 	struct btrfs_root *root;
8531 	struct list_head splice;
8532 	int ret;
8533 
8534 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8535 		return -EROFS;
8536 
8537 	INIT_LIST_HEAD(&splice);
8538 
8539 	spin_lock(&fs_info->delalloc_root_lock);
8540 	list_splice_init(&fs_info->delalloc_roots, &splice);
8541 	while (!list_empty(&splice)) {
8542 		root = list_first_entry(&splice, struct btrfs_root,
8543 					delalloc_root);
8544 		root = btrfs_grab_fs_root(root);
8545 		BUG_ON(!root);
8546 		list_move_tail(&root->delalloc_root,
8547 			       &fs_info->delalloc_roots);
8548 		spin_unlock(&fs_info->delalloc_root_lock);
8549 
8550 		ret = __start_delalloc_inodes(root, delay_iput);
8551 		btrfs_put_fs_root(root);
8552 		if (ret)
8553 			goto out;
8554 
8555 		spin_lock(&fs_info->delalloc_root_lock);
8556 	}
8557 	spin_unlock(&fs_info->delalloc_root_lock);
8558 
8559 	atomic_inc(&fs_info->async_submit_draining);
8560 	while (atomic_read(&fs_info->nr_async_submits) ||
8561 	      atomic_read(&fs_info->async_delalloc_pages)) {
8562 		wait_event(fs_info->async_submit_wait,
8563 		   (atomic_read(&fs_info->nr_async_submits) == 0 &&
8564 		    atomic_read(&fs_info->async_delalloc_pages) == 0));
8565 	}
8566 	atomic_dec(&fs_info->async_submit_draining);
8567 	return 0;
8568 out:
8569 	if (!list_empty_careful(&splice)) {
8570 		spin_lock(&fs_info->delalloc_root_lock);
8571 		list_splice_tail(&splice, &fs_info->delalloc_roots);
8572 		spin_unlock(&fs_info->delalloc_root_lock);
8573 	}
8574 	return ret;
8575 }
8576 
8577 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8578 			 const char *symname)
8579 {
8580 	struct btrfs_trans_handle *trans;
8581 	struct btrfs_root *root = BTRFS_I(dir)->root;
8582 	struct btrfs_path *path;
8583 	struct btrfs_key key;
8584 	struct inode *inode = NULL;
8585 	int err;
8586 	int drop_inode = 0;
8587 	u64 objectid;
8588 	u64 index = 0;
8589 	int name_len;
8590 	int datasize;
8591 	unsigned long ptr;
8592 	struct btrfs_file_extent_item *ei;
8593 	struct extent_buffer *leaf;
8594 
8595 	name_len = strlen(symname);
8596 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8597 		return -ENAMETOOLONG;
8598 
8599 	/*
8600 	 * 2 items for inode item and ref
8601 	 * 2 items for dir items
8602 	 * 1 item for xattr if selinux is on
8603 	 */
8604 	trans = btrfs_start_transaction(root, 5);
8605 	if (IS_ERR(trans))
8606 		return PTR_ERR(trans);
8607 
8608 	err = btrfs_find_free_ino(root, &objectid);
8609 	if (err)
8610 		goto out_unlock;
8611 
8612 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8613 				dentry->d_name.len, btrfs_ino(dir), objectid,
8614 				S_IFLNK|S_IRWXUGO, &index);
8615 	if (IS_ERR(inode)) {
8616 		err = PTR_ERR(inode);
8617 		goto out_unlock;
8618 	}
8619 
8620 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8621 	if (err) {
8622 		drop_inode = 1;
8623 		goto out_unlock;
8624 	}
8625 
8626 	/*
8627 	* If the active LSM wants to access the inode during
8628 	* d_instantiate it needs these. Smack checks to see
8629 	* if the filesystem supports xattrs by looking at the
8630 	* ops vector.
8631 	*/
8632 	inode->i_fop = &btrfs_file_operations;
8633 	inode->i_op = &btrfs_file_inode_operations;
8634 
8635 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8636 	if (err)
8637 		drop_inode = 1;
8638 	else {
8639 		inode->i_mapping->a_ops = &btrfs_aops;
8640 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8641 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8642 	}
8643 	if (drop_inode)
8644 		goto out_unlock;
8645 
8646 	path = btrfs_alloc_path();
8647 	if (!path) {
8648 		err = -ENOMEM;
8649 		drop_inode = 1;
8650 		goto out_unlock;
8651 	}
8652 	key.objectid = btrfs_ino(inode);
8653 	key.offset = 0;
8654 	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8655 	datasize = btrfs_file_extent_calc_inline_size(name_len);
8656 	err = btrfs_insert_empty_item(trans, root, path, &key,
8657 				      datasize);
8658 	if (err) {
8659 		drop_inode = 1;
8660 		btrfs_free_path(path);
8661 		goto out_unlock;
8662 	}
8663 	leaf = path->nodes[0];
8664 	ei = btrfs_item_ptr(leaf, path->slots[0],
8665 			    struct btrfs_file_extent_item);
8666 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8667 	btrfs_set_file_extent_type(leaf, ei,
8668 				   BTRFS_FILE_EXTENT_INLINE);
8669 	btrfs_set_file_extent_encryption(leaf, ei, 0);
8670 	btrfs_set_file_extent_compression(leaf, ei, 0);
8671 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8672 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8673 
8674 	ptr = btrfs_file_extent_inline_start(ei);
8675 	write_extent_buffer(leaf, symname, ptr, name_len);
8676 	btrfs_mark_buffer_dirty(leaf);
8677 	btrfs_free_path(path);
8678 
8679 	inode->i_op = &btrfs_symlink_inode_operations;
8680 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
8681 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8682 	inode_set_bytes(inode, name_len);
8683 	btrfs_i_size_write(inode, name_len);
8684 	err = btrfs_update_inode(trans, root, inode);
8685 	if (err)
8686 		drop_inode = 1;
8687 
8688 out_unlock:
8689 	if (!err)
8690 		d_instantiate(dentry, inode);
8691 	btrfs_end_transaction(trans, root);
8692 	if (drop_inode) {
8693 		inode_dec_link_count(inode);
8694 		iput(inode);
8695 	}
8696 	btrfs_btree_balance_dirty(root);
8697 	return err;
8698 }
8699 
8700 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8701 				       u64 start, u64 num_bytes, u64 min_size,
8702 				       loff_t actual_len, u64 *alloc_hint,
8703 				       struct btrfs_trans_handle *trans)
8704 {
8705 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8706 	struct extent_map *em;
8707 	struct btrfs_root *root = BTRFS_I(inode)->root;
8708 	struct btrfs_key ins;
8709 	u64 cur_offset = start;
8710 	u64 i_size;
8711 	u64 cur_bytes;
8712 	int ret = 0;
8713 	bool own_trans = true;
8714 
8715 	if (trans)
8716 		own_trans = false;
8717 	while (num_bytes > 0) {
8718 		if (own_trans) {
8719 			trans = btrfs_start_transaction(root, 3);
8720 			if (IS_ERR(trans)) {
8721 				ret = PTR_ERR(trans);
8722 				break;
8723 			}
8724 		}
8725 
8726 		cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8727 		cur_bytes = max(cur_bytes, min_size);
8728 		ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8729 					   *alloc_hint, &ins, 1);
8730 		if (ret) {
8731 			if (own_trans)
8732 				btrfs_end_transaction(trans, root);
8733 			break;
8734 		}
8735 
8736 		ret = insert_reserved_file_extent(trans, inode,
8737 						  cur_offset, ins.objectid,
8738 						  ins.offset, ins.offset,
8739 						  ins.offset, 0, 0, 0,
8740 						  BTRFS_FILE_EXTENT_PREALLOC);
8741 		if (ret) {
8742 			btrfs_free_reserved_extent(root, ins.objectid,
8743 						   ins.offset);
8744 			btrfs_abort_transaction(trans, root, ret);
8745 			if (own_trans)
8746 				btrfs_end_transaction(trans, root);
8747 			break;
8748 		}
8749 		btrfs_drop_extent_cache(inode, cur_offset,
8750 					cur_offset + ins.offset -1, 0);
8751 
8752 		em = alloc_extent_map();
8753 		if (!em) {
8754 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8755 				&BTRFS_I(inode)->runtime_flags);
8756 			goto next;
8757 		}
8758 
8759 		em->start = cur_offset;
8760 		em->orig_start = cur_offset;
8761 		em->len = ins.offset;
8762 		em->block_start = ins.objectid;
8763 		em->block_len = ins.offset;
8764 		em->orig_block_len = ins.offset;
8765 		em->ram_bytes = ins.offset;
8766 		em->bdev = root->fs_info->fs_devices->latest_bdev;
8767 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8768 		em->generation = trans->transid;
8769 
8770 		while (1) {
8771 			write_lock(&em_tree->lock);
8772 			ret = add_extent_mapping(em_tree, em, 1);
8773 			write_unlock(&em_tree->lock);
8774 			if (ret != -EEXIST)
8775 				break;
8776 			btrfs_drop_extent_cache(inode, cur_offset,
8777 						cur_offset + ins.offset - 1,
8778 						0);
8779 		}
8780 		free_extent_map(em);
8781 next:
8782 		num_bytes -= ins.offset;
8783 		cur_offset += ins.offset;
8784 		*alloc_hint = ins.objectid + ins.offset;
8785 
8786 		inode_inc_iversion(inode);
8787 		inode->i_ctime = CURRENT_TIME;
8788 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8789 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8790 		    (actual_len > inode->i_size) &&
8791 		    (cur_offset > inode->i_size)) {
8792 			if (cur_offset > actual_len)
8793 				i_size = actual_len;
8794 			else
8795 				i_size = cur_offset;
8796 			i_size_write(inode, i_size);
8797 			btrfs_ordered_update_i_size(inode, i_size, NULL);
8798 		}
8799 
8800 		ret = btrfs_update_inode(trans, root, inode);
8801 
8802 		if (ret) {
8803 			btrfs_abort_transaction(trans, root, ret);
8804 			if (own_trans)
8805 				btrfs_end_transaction(trans, root);
8806 			break;
8807 		}
8808 
8809 		if (own_trans)
8810 			btrfs_end_transaction(trans, root);
8811 	}
8812 	return ret;
8813 }
8814 
8815 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8816 			      u64 start, u64 num_bytes, u64 min_size,
8817 			      loff_t actual_len, u64 *alloc_hint)
8818 {
8819 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8820 					   min_size, actual_len, alloc_hint,
8821 					   NULL);
8822 }
8823 
8824 int btrfs_prealloc_file_range_trans(struct inode *inode,
8825 				    struct btrfs_trans_handle *trans, int mode,
8826 				    u64 start, u64 num_bytes, u64 min_size,
8827 				    loff_t actual_len, u64 *alloc_hint)
8828 {
8829 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8830 					   min_size, actual_len, alloc_hint, trans);
8831 }
8832 
8833 static int btrfs_set_page_dirty(struct page *page)
8834 {
8835 	return __set_page_dirty_nobuffers(page);
8836 }
8837 
8838 static int btrfs_permission(struct inode *inode, int mask)
8839 {
8840 	struct btrfs_root *root = BTRFS_I(inode)->root;
8841 	umode_t mode = inode->i_mode;
8842 
8843 	if (mask & MAY_WRITE &&
8844 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8845 		if (btrfs_root_readonly(root))
8846 			return -EROFS;
8847 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8848 			return -EACCES;
8849 	}
8850 	return generic_permission(inode, mask);
8851 }
8852 
8853 static const struct inode_operations btrfs_dir_inode_operations = {
8854 	.getattr	= btrfs_getattr,
8855 	.lookup		= btrfs_lookup,
8856 	.create		= btrfs_create,
8857 	.unlink		= btrfs_unlink,
8858 	.link		= btrfs_link,
8859 	.mkdir		= btrfs_mkdir,
8860 	.rmdir		= btrfs_rmdir,
8861 	.rename		= btrfs_rename,
8862 	.symlink	= btrfs_symlink,
8863 	.setattr	= btrfs_setattr,
8864 	.mknod		= btrfs_mknod,
8865 	.setxattr	= btrfs_setxattr,
8866 	.getxattr	= btrfs_getxattr,
8867 	.listxattr	= btrfs_listxattr,
8868 	.removexattr	= btrfs_removexattr,
8869 	.permission	= btrfs_permission,
8870 	.get_acl	= btrfs_get_acl,
8871 	.update_time	= btrfs_update_time,
8872 };
8873 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8874 	.lookup		= btrfs_lookup,
8875 	.permission	= btrfs_permission,
8876 	.get_acl	= btrfs_get_acl,
8877 	.update_time	= btrfs_update_time,
8878 };
8879 
8880 static const struct file_operations btrfs_dir_file_operations = {
8881 	.llseek		= generic_file_llseek,
8882 	.read		= generic_read_dir,
8883 	.iterate	= btrfs_real_readdir,
8884 	.unlocked_ioctl	= btrfs_ioctl,
8885 #ifdef CONFIG_COMPAT
8886 	.compat_ioctl	= btrfs_ioctl,
8887 #endif
8888 	.release        = btrfs_release_file,
8889 	.fsync		= btrfs_sync_file,
8890 };
8891 
8892 static struct extent_io_ops btrfs_extent_io_ops = {
8893 	.fill_delalloc = run_delalloc_range,
8894 	.submit_bio_hook = btrfs_submit_bio_hook,
8895 	.merge_bio_hook = btrfs_merge_bio_hook,
8896 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
8897 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
8898 	.writepage_start_hook = btrfs_writepage_start_hook,
8899 	.set_bit_hook = btrfs_set_bit_hook,
8900 	.clear_bit_hook = btrfs_clear_bit_hook,
8901 	.merge_extent_hook = btrfs_merge_extent_hook,
8902 	.split_extent_hook = btrfs_split_extent_hook,
8903 };
8904 
8905 /*
8906  * btrfs doesn't support the bmap operation because swapfiles
8907  * use bmap to make a mapping of extents in the file.  They assume
8908  * these extents won't change over the life of the file and they
8909  * use the bmap result to do IO directly to the drive.
8910  *
8911  * the btrfs bmap call would return logical addresses that aren't
8912  * suitable for IO and they also will change frequently as COW
8913  * operations happen.  So, swapfile + btrfs == corruption.
8914  *
8915  * For now we're avoiding this by dropping bmap.
8916  */
8917 static const struct address_space_operations btrfs_aops = {
8918 	.readpage	= btrfs_readpage,
8919 	.writepage	= btrfs_writepage,
8920 	.writepages	= btrfs_writepages,
8921 	.readpages	= btrfs_readpages,
8922 	.direct_IO	= btrfs_direct_IO,
8923 	.invalidatepage = btrfs_invalidatepage,
8924 	.releasepage	= btrfs_releasepage,
8925 	.set_page_dirty	= btrfs_set_page_dirty,
8926 	.error_remove_page = generic_error_remove_page,
8927 };
8928 
8929 static const struct address_space_operations btrfs_symlink_aops = {
8930 	.readpage	= btrfs_readpage,
8931 	.writepage	= btrfs_writepage,
8932 	.invalidatepage = btrfs_invalidatepage,
8933 	.releasepage	= btrfs_releasepage,
8934 };
8935 
8936 static const struct inode_operations btrfs_file_inode_operations = {
8937 	.getattr	= btrfs_getattr,
8938 	.setattr	= btrfs_setattr,
8939 	.setxattr	= btrfs_setxattr,
8940 	.getxattr	= btrfs_getxattr,
8941 	.listxattr      = btrfs_listxattr,
8942 	.removexattr	= btrfs_removexattr,
8943 	.permission	= btrfs_permission,
8944 	.fiemap		= btrfs_fiemap,
8945 	.get_acl	= btrfs_get_acl,
8946 	.update_time	= btrfs_update_time,
8947 };
8948 static const struct inode_operations btrfs_special_inode_operations = {
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 	.get_acl	= btrfs_get_acl,
8957 	.update_time	= btrfs_update_time,
8958 };
8959 static const struct inode_operations btrfs_symlink_inode_operations = {
8960 	.readlink	= generic_readlink,
8961 	.follow_link	= page_follow_link_light,
8962 	.put_link	= page_put_link,
8963 	.getattr	= btrfs_getattr,
8964 	.setattr	= btrfs_setattr,
8965 	.permission	= btrfs_permission,
8966 	.setxattr	= btrfs_setxattr,
8967 	.getxattr	= btrfs_getxattr,
8968 	.listxattr	= btrfs_listxattr,
8969 	.removexattr	= btrfs_removexattr,
8970 	.get_acl	= btrfs_get_acl,
8971 	.update_time	= btrfs_update_time,
8972 };
8973 
8974 const struct dentry_operations btrfs_dentry_operations = {
8975 	.d_delete	= btrfs_dentry_delete,
8976 	.d_release	= btrfs_dentry_release,
8977 };
8978