xref: /openbmc/linux/fs/btrfs/inode.c (revision 33ac9dba)
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 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 	/*
398 	 * skip compression for a small file range(<=blocksize) that
399 	 * isn't an inline extent, since it dosen't save disk space at all.
400 	 */
401 	if ((end - start + 1) <= blocksize &&
402 	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
403 		goto cleanup_and_bail_uncompressed;
404 
405 	actual_end = min_t(u64, isize, end + 1);
406 again:
407 	will_compress = 0;
408 	nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
409 	nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
410 
411 	/*
412 	 * we don't want to send crud past the end of i_size through
413 	 * compression, that's just a waste of CPU time.  So, if the
414 	 * end of the file is before the start of our current
415 	 * requested range of bytes, we bail out to the uncompressed
416 	 * cleanup code that can deal with all of this.
417 	 *
418 	 * It isn't really the fastest way to fix things, but this is a
419 	 * very uncommon corner.
420 	 */
421 	if (actual_end <= start)
422 		goto cleanup_and_bail_uncompressed;
423 
424 	total_compressed = actual_end - start;
425 
426 	/* we want to make sure that amount of ram required to uncompress
427 	 * an extent is reasonable, so we limit the total size in ram
428 	 * of a compressed extent to 128k.  This is a crucial number
429 	 * because it also controls how easily we can spread reads across
430 	 * cpus for decompression.
431 	 *
432 	 * We also want to make sure the amount of IO required to do
433 	 * a random read is reasonably small, so we limit the size of
434 	 * a compressed extent to 128k.
435 	 */
436 	total_compressed = min(total_compressed, max_uncompressed);
437 	num_bytes = ALIGN(end - start + 1, blocksize);
438 	num_bytes = max(blocksize,  num_bytes);
439 	total_in = 0;
440 	ret = 0;
441 
442 	/*
443 	 * we do compression for mount -o compress and when the
444 	 * inode has not been flagged as nocompress.  This flag can
445 	 * change at any time if we discover bad compression ratios.
446 	 */
447 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
448 	    (btrfs_test_opt(root, COMPRESS) ||
449 	     (BTRFS_I(inode)->force_compress) ||
450 	     (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
451 		WARN_ON(pages);
452 		pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
453 		if (!pages) {
454 			/* just bail out to the uncompressed code */
455 			goto cont;
456 		}
457 
458 		if (BTRFS_I(inode)->force_compress)
459 			compress_type = BTRFS_I(inode)->force_compress;
460 
461 		/*
462 		 * we need to call clear_page_dirty_for_io on each
463 		 * page in the range.  Otherwise applications with the file
464 		 * mmap'd can wander in and change the page contents while
465 		 * we are compressing them.
466 		 *
467 		 * If the compression fails for any reason, we set the pages
468 		 * dirty again later on.
469 		 */
470 		extent_range_clear_dirty_for_io(inode, start, end);
471 		redirty = 1;
472 		ret = btrfs_compress_pages(compress_type,
473 					   inode->i_mapping, start,
474 					   total_compressed, pages,
475 					   nr_pages, &nr_pages_ret,
476 					   &total_in,
477 					   &total_compressed,
478 					   max_compressed);
479 
480 		if (!ret) {
481 			unsigned long offset = total_compressed &
482 				(PAGE_CACHE_SIZE - 1);
483 			struct page *page = pages[nr_pages_ret - 1];
484 			char *kaddr;
485 
486 			/* zero the tail end of the last page, we might be
487 			 * sending it down to disk
488 			 */
489 			if (offset) {
490 				kaddr = kmap_atomic(page);
491 				memset(kaddr + offset, 0,
492 				       PAGE_CACHE_SIZE - offset);
493 				kunmap_atomic(kaddr);
494 			}
495 			will_compress = 1;
496 		}
497 	}
498 cont:
499 	if (start == 0) {
500 		/* lets try to make an inline extent */
501 		if (ret || total_in < (actual_end - start)) {
502 			/* we didn't compress the entire range, try
503 			 * to make an uncompressed inline extent.
504 			 */
505 			ret = cow_file_range_inline(root, inode, start, end,
506 						    0, 0, NULL);
507 		} else {
508 			/* try making a compressed inline extent */
509 			ret = cow_file_range_inline(root, inode, start, end,
510 						    total_compressed,
511 						    compress_type, pages);
512 		}
513 		if (ret <= 0) {
514 			unsigned long clear_flags = EXTENT_DELALLOC |
515 				EXTENT_DEFRAG;
516 			clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
517 
518 			/*
519 			 * inline extent creation worked or returned error,
520 			 * we don't need to create any more async work items.
521 			 * Unlock and free up our temp pages.
522 			 */
523 			extent_clear_unlock_delalloc(inode, start, end, NULL,
524 						     clear_flags, PAGE_UNLOCK |
525 						     PAGE_CLEAR_DIRTY |
526 						     PAGE_SET_WRITEBACK |
527 						     PAGE_END_WRITEBACK);
528 			goto free_pages_out;
529 		}
530 	}
531 
532 	if (will_compress) {
533 		/*
534 		 * we aren't doing an inline extent round the compressed size
535 		 * up to a block size boundary so the allocator does sane
536 		 * things
537 		 */
538 		total_compressed = ALIGN(total_compressed, blocksize);
539 
540 		/*
541 		 * one last check to make sure the compression is really a
542 		 * win, compare the page count read with the blocks on disk
543 		 */
544 		total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
545 		if (total_compressed >= total_in) {
546 			will_compress = 0;
547 		} else {
548 			num_bytes = total_in;
549 		}
550 	}
551 	if (!will_compress && pages) {
552 		/*
553 		 * the compression code ran but failed to make things smaller,
554 		 * free any pages it allocated and our page pointer array
555 		 */
556 		for (i = 0; i < nr_pages_ret; i++) {
557 			WARN_ON(pages[i]->mapping);
558 			page_cache_release(pages[i]);
559 		}
560 		kfree(pages);
561 		pages = NULL;
562 		total_compressed = 0;
563 		nr_pages_ret = 0;
564 
565 		/* flag the file so we don't compress in the future */
566 		if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
567 		    !(BTRFS_I(inode)->force_compress)) {
568 			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
569 		}
570 	}
571 	if (will_compress) {
572 		*num_added += 1;
573 
574 		/* the async work queues will take care of doing actual
575 		 * allocation on disk for these compressed pages,
576 		 * and will submit them to the elevator.
577 		 */
578 		add_async_extent(async_cow, start, num_bytes,
579 				 total_compressed, pages, nr_pages_ret,
580 				 compress_type);
581 
582 		if (start + num_bytes < end) {
583 			start += num_bytes;
584 			pages = NULL;
585 			cond_resched();
586 			goto again;
587 		}
588 	} else {
589 cleanup_and_bail_uncompressed:
590 		/*
591 		 * No compression, but we still need to write the pages in
592 		 * the file we've been given so far.  redirty the locked
593 		 * page if it corresponds to our extent and set things up
594 		 * for the async work queue to run cow_file_range to do
595 		 * the normal delalloc dance
596 		 */
597 		if (page_offset(locked_page) >= start &&
598 		    page_offset(locked_page) <= end) {
599 			__set_page_dirty_nobuffers(locked_page);
600 			/* unlocked later on in the async handlers */
601 		}
602 		if (redirty)
603 			extent_range_redirty_for_io(inode, start, end);
604 		add_async_extent(async_cow, start, end - start + 1,
605 				 0, NULL, 0, BTRFS_COMPRESS_NONE);
606 		*num_added += 1;
607 	}
608 
609 out:
610 	return ret;
611 
612 free_pages_out:
613 	for (i = 0; i < nr_pages_ret; i++) {
614 		WARN_ON(pages[i]->mapping);
615 		page_cache_release(pages[i]);
616 	}
617 	kfree(pages);
618 
619 	goto out;
620 }
621 
622 /*
623  * phase two of compressed writeback.  This is the ordered portion
624  * of the code, which only gets called in the order the work was
625  * queued.  We walk all the async extents created by compress_file_range
626  * and send them down to the disk.
627  */
628 static noinline int submit_compressed_extents(struct inode *inode,
629 					      struct async_cow *async_cow)
630 {
631 	struct async_extent *async_extent;
632 	u64 alloc_hint = 0;
633 	struct btrfs_key ins;
634 	struct extent_map *em;
635 	struct btrfs_root *root = BTRFS_I(inode)->root;
636 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
637 	struct extent_io_tree *io_tree;
638 	int ret = 0;
639 
640 	if (list_empty(&async_cow->extents))
641 		return 0;
642 
643 again:
644 	while (!list_empty(&async_cow->extents)) {
645 		async_extent = list_entry(async_cow->extents.next,
646 					  struct async_extent, list);
647 		list_del(&async_extent->list);
648 
649 		io_tree = &BTRFS_I(inode)->io_tree;
650 
651 retry:
652 		/* did the compression code fall back to uncompressed IO? */
653 		if (!async_extent->pages) {
654 			int page_started = 0;
655 			unsigned long nr_written = 0;
656 
657 			lock_extent(io_tree, async_extent->start,
658 					 async_extent->start +
659 					 async_extent->ram_size - 1);
660 
661 			/* allocate blocks */
662 			ret = cow_file_range(inode, async_cow->locked_page,
663 					     async_extent->start,
664 					     async_extent->start +
665 					     async_extent->ram_size - 1,
666 					     &page_started, &nr_written, 0);
667 
668 			/* JDM XXX */
669 
670 			/*
671 			 * if page_started, cow_file_range inserted an
672 			 * inline extent and took care of all the unlocking
673 			 * and IO for us.  Otherwise, we need to submit
674 			 * all those pages down to the drive.
675 			 */
676 			if (!page_started && !ret)
677 				extent_write_locked_range(io_tree,
678 						  inode, async_extent->start,
679 						  async_extent->start +
680 						  async_extent->ram_size - 1,
681 						  btrfs_get_extent,
682 						  WB_SYNC_ALL);
683 			else if (ret)
684 				unlock_page(async_cow->locked_page);
685 			kfree(async_extent);
686 			cond_resched();
687 			continue;
688 		}
689 
690 		lock_extent(io_tree, async_extent->start,
691 			    async_extent->start + async_extent->ram_size - 1);
692 
693 		ret = btrfs_reserve_extent(root,
694 					   async_extent->compressed_size,
695 					   async_extent->compressed_size,
696 					   0, alloc_hint, &ins, 1, 1);
697 		if (ret) {
698 			int i;
699 
700 			for (i = 0; i < async_extent->nr_pages; i++) {
701 				WARN_ON(async_extent->pages[i]->mapping);
702 				page_cache_release(async_extent->pages[i]);
703 			}
704 			kfree(async_extent->pages);
705 			async_extent->nr_pages = 0;
706 			async_extent->pages = NULL;
707 
708 			if (ret == -ENOSPC) {
709 				unlock_extent(io_tree, async_extent->start,
710 					      async_extent->start +
711 					      async_extent->ram_size - 1);
712 
713 				/*
714 				 * we need to redirty the pages if we decide to
715 				 * fallback to uncompressed IO, otherwise we
716 				 * will not submit these pages down to lower
717 				 * layers.
718 				 */
719 				extent_range_redirty_for_io(inode,
720 						async_extent->start,
721 						async_extent->start +
722 						async_extent->ram_size - 1);
723 
724 				goto retry;
725 			}
726 			goto out_free;
727 		}
728 
729 		/*
730 		 * here we're doing allocation and writeback of the
731 		 * compressed pages
732 		 */
733 		btrfs_drop_extent_cache(inode, async_extent->start,
734 					async_extent->start +
735 					async_extent->ram_size - 1, 0);
736 
737 		em = alloc_extent_map();
738 		if (!em) {
739 			ret = -ENOMEM;
740 			goto out_free_reserve;
741 		}
742 		em->start = async_extent->start;
743 		em->len = async_extent->ram_size;
744 		em->orig_start = em->start;
745 		em->mod_start = em->start;
746 		em->mod_len = em->len;
747 
748 		em->block_start = ins.objectid;
749 		em->block_len = ins.offset;
750 		em->orig_block_len = ins.offset;
751 		em->ram_bytes = async_extent->ram_size;
752 		em->bdev = root->fs_info->fs_devices->latest_bdev;
753 		em->compress_type = async_extent->compress_type;
754 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
755 		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
756 		em->generation = -1;
757 
758 		while (1) {
759 			write_lock(&em_tree->lock);
760 			ret = add_extent_mapping(em_tree, em, 1);
761 			write_unlock(&em_tree->lock);
762 			if (ret != -EEXIST) {
763 				free_extent_map(em);
764 				break;
765 			}
766 			btrfs_drop_extent_cache(inode, async_extent->start,
767 						async_extent->start +
768 						async_extent->ram_size - 1, 0);
769 		}
770 
771 		if (ret)
772 			goto out_free_reserve;
773 
774 		ret = btrfs_add_ordered_extent_compress(inode,
775 						async_extent->start,
776 						ins.objectid,
777 						async_extent->ram_size,
778 						ins.offset,
779 						BTRFS_ORDERED_COMPRESSED,
780 						async_extent->compress_type);
781 		if (ret)
782 			goto out_free_reserve;
783 
784 		/*
785 		 * clear dirty, set writeback and unlock the pages.
786 		 */
787 		extent_clear_unlock_delalloc(inode, async_extent->start,
788 				async_extent->start +
789 				async_extent->ram_size - 1,
790 				NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
791 				PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
792 				PAGE_SET_WRITEBACK);
793 		ret = btrfs_submit_compressed_write(inode,
794 				    async_extent->start,
795 				    async_extent->ram_size,
796 				    ins.objectid,
797 				    ins.offset, async_extent->pages,
798 				    async_extent->nr_pages);
799 		alloc_hint = ins.objectid + ins.offset;
800 		kfree(async_extent);
801 		if (ret)
802 			goto out;
803 		cond_resched();
804 	}
805 	ret = 0;
806 out:
807 	return ret;
808 out_free_reserve:
809 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
810 out_free:
811 	extent_clear_unlock_delalloc(inode, async_extent->start,
812 				     async_extent->start +
813 				     async_extent->ram_size - 1,
814 				     NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
815 				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
816 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
817 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
818 	kfree(async_extent);
819 	goto again;
820 }
821 
822 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
823 				      u64 num_bytes)
824 {
825 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826 	struct extent_map *em;
827 	u64 alloc_hint = 0;
828 
829 	read_lock(&em_tree->lock);
830 	em = search_extent_mapping(em_tree, start, num_bytes);
831 	if (em) {
832 		/*
833 		 * if block start isn't an actual block number then find the
834 		 * first block in this inode and use that as a hint.  If that
835 		 * block is also bogus then just don't worry about it.
836 		 */
837 		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
838 			free_extent_map(em);
839 			em = search_extent_mapping(em_tree, 0, 0);
840 			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
841 				alloc_hint = em->block_start;
842 			if (em)
843 				free_extent_map(em);
844 		} else {
845 			alloc_hint = em->block_start;
846 			free_extent_map(em);
847 		}
848 	}
849 	read_unlock(&em_tree->lock);
850 
851 	return alloc_hint;
852 }
853 
854 /*
855  * when extent_io.c finds a delayed allocation range in the file,
856  * the call backs end up in this code.  The basic idea is to
857  * allocate extents on disk for the range, and create ordered data structs
858  * in ram to track those extents.
859  *
860  * locked_page is the page that writepage had locked already.  We use
861  * it to make sure we don't do extra locks or unlocks.
862  *
863  * *page_started is set to one if we unlock locked_page and do everything
864  * required to start IO on it.  It may be clean and already done with
865  * IO when we return.
866  */
867 static noinline int cow_file_range(struct inode *inode,
868 				   struct page *locked_page,
869 				   u64 start, u64 end, int *page_started,
870 				   unsigned long *nr_written,
871 				   int unlock)
872 {
873 	struct btrfs_root *root = BTRFS_I(inode)->root;
874 	u64 alloc_hint = 0;
875 	u64 num_bytes;
876 	unsigned long ram_size;
877 	u64 disk_num_bytes;
878 	u64 cur_alloc_size;
879 	u64 blocksize = root->sectorsize;
880 	struct btrfs_key ins;
881 	struct extent_map *em;
882 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 	int ret = 0;
884 
885 	if (btrfs_is_free_space_inode(inode)) {
886 		WARN_ON_ONCE(1);
887 		ret = -EINVAL;
888 		goto out_unlock;
889 	}
890 
891 	num_bytes = ALIGN(end - start + 1, blocksize);
892 	num_bytes = max(blocksize,  num_bytes);
893 	disk_num_bytes = num_bytes;
894 
895 	/* if this is a small write inside eof, kick off defrag */
896 	if (num_bytes < 64 * 1024 &&
897 	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
898 		btrfs_add_inode_defrag(NULL, inode);
899 
900 	if (start == 0) {
901 		/* lets try to make an inline extent */
902 		ret = cow_file_range_inline(root, inode, start, end, 0, 0,
903 					    NULL);
904 		if (ret == 0) {
905 			extent_clear_unlock_delalloc(inode, start, end, NULL,
906 				     EXTENT_LOCKED | EXTENT_DELALLOC |
907 				     EXTENT_DEFRAG, PAGE_UNLOCK |
908 				     PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
909 				     PAGE_END_WRITEBACK);
910 
911 			*nr_written = *nr_written +
912 			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
913 			*page_started = 1;
914 			goto out;
915 		} else if (ret < 0) {
916 			goto out_unlock;
917 		}
918 	}
919 
920 	BUG_ON(disk_num_bytes >
921 	       btrfs_super_total_bytes(root->fs_info->super_copy));
922 
923 	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
924 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
925 
926 	while (disk_num_bytes > 0) {
927 		unsigned long op;
928 
929 		cur_alloc_size = disk_num_bytes;
930 		ret = btrfs_reserve_extent(root, cur_alloc_size,
931 					   root->sectorsize, 0, alloc_hint,
932 					   &ins, 1, 1);
933 		if (ret < 0)
934 			goto out_unlock;
935 
936 		em = alloc_extent_map();
937 		if (!em) {
938 			ret = -ENOMEM;
939 			goto out_reserve;
940 		}
941 		em->start = start;
942 		em->orig_start = em->start;
943 		ram_size = ins.offset;
944 		em->len = ins.offset;
945 		em->mod_start = em->start;
946 		em->mod_len = em->len;
947 
948 		em->block_start = ins.objectid;
949 		em->block_len = ins.offset;
950 		em->orig_block_len = ins.offset;
951 		em->ram_bytes = ram_size;
952 		em->bdev = root->fs_info->fs_devices->latest_bdev;
953 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
954 		em->generation = -1;
955 
956 		while (1) {
957 			write_lock(&em_tree->lock);
958 			ret = add_extent_mapping(em_tree, em, 1);
959 			write_unlock(&em_tree->lock);
960 			if (ret != -EEXIST) {
961 				free_extent_map(em);
962 				break;
963 			}
964 			btrfs_drop_extent_cache(inode, start,
965 						start + ram_size - 1, 0);
966 		}
967 		if (ret)
968 			goto out_reserve;
969 
970 		cur_alloc_size = ins.offset;
971 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
972 					       ram_size, cur_alloc_size, 0);
973 		if (ret)
974 			goto out_reserve;
975 
976 		if (root->root_key.objectid ==
977 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
978 			ret = btrfs_reloc_clone_csums(inode, start,
979 						      cur_alloc_size);
980 			if (ret)
981 				goto out_reserve;
982 		}
983 
984 		if (disk_num_bytes < cur_alloc_size)
985 			break;
986 
987 		/* we're not doing compressed IO, don't unlock the first
988 		 * page (which the caller expects to stay locked), don't
989 		 * clear any dirty bits and don't set any writeback bits
990 		 *
991 		 * Do set the Private2 bit so we know this page was properly
992 		 * setup for writepage
993 		 */
994 		op = unlock ? PAGE_UNLOCK : 0;
995 		op |= PAGE_SET_PRIVATE2;
996 
997 		extent_clear_unlock_delalloc(inode, start,
998 					     start + ram_size - 1, locked_page,
999 					     EXTENT_LOCKED | EXTENT_DELALLOC,
1000 					     op);
1001 		disk_num_bytes -= cur_alloc_size;
1002 		num_bytes -= cur_alloc_size;
1003 		alloc_hint = ins.objectid + ins.offset;
1004 		start += cur_alloc_size;
1005 	}
1006 out:
1007 	return ret;
1008 
1009 out_reserve:
1010 	btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1011 out_unlock:
1012 	extent_clear_unlock_delalloc(inode, start, end, locked_page,
1013 				     EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1014 				     EXTENT_DELALLOC | EXTENT_DEFRAG,
1015 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1016 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1017 	goto out;
1018 }
1019 
1020 /*
1021  * work queue call back to started compression on a file and pages
1022  */
1023 static noinline void async_cow_start(struct btrfs_work *work)
1024 {
1025 	struct async_cow *async_cow;
1026 	int num_added = 0;
1027 	async_cow = container_of(work, struct async_cow, work);
1028 
1029 	compress_file_range(async_cow->inode, async_cow->locked_page,
1030 			    async_cow->start, async_cow->end, async_cow,
1031 			    &num_added);
1032 	if (num_added == 0) {
1033 		btrfs_add_delayed_iput(async_cow->inode);
1034 		async_cow->inode = NULL;
1035 	}
1036 }
1037 
1038 /*
1039  * work queue call back to submit previously compressed pages
1040  */
1041 static noinline void async_cow_submit(struct btrfs_work *work)
1042 {
1043 	struct async_cow *async_cow;
1044 	struct btrfs_root *root;
1045 	unsigned long nr_pages;
1046 
1047 	async_cow = container_of(work, struct async_cow, work);
1048 
1049 	root = async_cow->root;
1050 	nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1051 		PAGE_CACHE_SHIFT;
1052 
1053 	if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1054 	    5 * 1024 * 1024 &&
1055 	    waitqueue_active(&root->fs_info->async_submit_wait))
1056 		wake_up(&root->fs_info->async_submit_wait);
1057 
1058 	if (async_cow->inode)
1059 		submit_compressed_extents(async_cow->inode, async_cow);
1060 }
1061 
1062 static noinline void async_cow_free(struct btrfs_work *work)
1063 {
1064 	struct async_cow *async_cow;
1065 	async_cow = container_of(work, struct async_cow, work);
1066 	if (async_cow->inode)
1067 		btrfs_add_delayed_iput(async_cow->inode);
1068 	kfree(async_cow);
1069 }
1070 
1071 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1072 				u64 start, u64 end, int *page_started,
1073 				unsigned long *nr_written)
1074 {
1075 	struct async_cow *async_cow;
1076 	struct btrfs_root *root = BTRFS_I(inode)->root;
1077 	unsigned long nr_pages;
1078 	u64 cur_end;
1079 	int limit = 10 * 1024 * 1024;
1080 
1081 	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1082 			 1, 0, NULL, GFP_NOFS);
1083 	while (start < end) {
1084 		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1085 		BUG_ON(!async_cow); /* -ENOMEM */
1086 		async_cow->inode = igrab(inode);
1087 		async_cow->root = root;
1088 		async_cow->locked_page = locked_page;
1089 		async_cow->start = start;
1090 
1091 		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1092 			cur_end = end;
1093 		else
1094 			cur_end = min(end, start + 512 * 1024 - 1);
1095 
1096 		async_cow->end = cur_end;
1097 		INIT_LIST_HEAD(&async_cow->extents);
1098 
1099 		btrfs_init_work(&async_cow->work, async_cow_start,
1100 				async_cow_submit, async_cow_free);
1101 
1102 		nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1103 			PAGE_CACHE_SHIFT;
1104 		atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1105 
1106 		btrfs_queue_work(root->fs_info->delalloc_workers,
1107 				 &async_cow->work);
1108 
1109 		if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1110 			wait_event(root->fs_info->async_submit_wait,
1111 			   (atomic_read(&root->fs_info->async_delalloc_pages) <
1112 			    limit));
1113 		}
1114 
1115 		while (atomic_read(&root->fs_info->async_submit_draining) &&
1116 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
1117 			wait_event(root->fs_info->async_submit_wait,
1118 			  (atomic_read(&root->fs_info->async_delalloc_pages) ==
1119 			   0));
1120 		}
1121 
1122 		*nr_written += nr_pages;
1123 		start = cur_end + 1;
1124 	}
1125 	*page_started = 1;
1126 	return 0;
1127 }
1128 
1129 static noinline int csum_exist_in_range(struct btrfs_root *root,
1130 					u64 bytenr, u64 num_bytes)
1131 {
1132 	int ret;
1133 	struct btrfs_ordered_sum *sums;
1134 	LIST_HEAD(list);
1135 
1136 	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1137 				       bytenr + num_bytes - 1, &list, 0);
1138 	if (ret == 0 && list_empty(&list))
1139 		return 0;
1140 
1141 	while (!list_empty(&list)) {
1142 		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1143 		list_del(&sums->list);
1144 		kfree(sums);
1145 	}
1146 	return 1;
1147 }
1148 
1149 /*
1150  * when nowcow writeback call back.  This checks for snapshots or COW copies
1151  * of the extents that exist in the file, and COWs the file as required.
1152  *
1153  * If no cow copies or snapshots exist, we write directly to the existing
1154  * blocks on disk
1155  */
1156 static noinline int run_delalloc_nocow(struct inode *inode,
1157 				       struct page *locked_page,
1158 			      u64 start, u64 end, int *page_started, int force,
1159 			      unsigned long *nr_written)
1160 {
1161 	struct btrfs_root *root = BTRFS_I(inode)->root;
1162 	struct btrfs_trans_handle *trans;
1163 	struct extent_buffer *leaf;
1164 	struct btrfs_path *path;
1165 	struct btrfs_file_extent_item *fi;
1166 	struct btrfs_key found_key;
1167 	u64 cow_start;
1168 	u64 cur_offset;
1169 	u64 extent_end;
1170 	u64 extent_offset;
1171 	u64 disk_bytenr;
1172 	u64 num_bytes;
1173 	u64 disk_num_bytes;
1174 	u64 ram_bytes;
1175 	int extent_type;
1176 	int ret, err;
1177 	int type;
1178 	int nocow;
1179 	int check_prev = 1;
1180 	bool nolock;
1181 	u64 ino = btrfs_ino(inode);
1182 
1183 	path = btrfs_alloc_path();
1184 	if (!path) {
1185 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1186 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1187 					     EXTENT_DO_ACCOUNTING |
1188 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1189 					     PAGE_CLEAR_DIRTY |
1190 					     PAGE_SET_WRITEBACK |
1191 					     PAGE_END_WRITEBACK);
1192 		return -ENOMEM;
1193 	}
1194 
1195 	nolock = btrfs_is_free_space_inode(inode);
1196 
1197 	if (nolock)
1198 		trans = btrfs_join_transaction_nolock(root);
1199 	else
1200 		trans = btrfs_join_transaction(root);
1201 
1202 	if (IS_ERR(trans)) {
1203 		extent_clear_unlock_delalloc(inode, start, end, locked_page,
1204 					     EXTENT_LOCKED | EXTENT_DELALLOC |
1205 					     EXTENT_DO_ACCOUNTING |
1206 					     EXTENT_DEFRAG, PAGE_UNLOCK |
1207 					     PAGE_CLEAR_DIRTY |
1208 					     PAGE_SET_WRITEBACK |
1209 					     PAGE_END_WRITEBACK);
1210 		btrfs_free_path(path);
1211 		return PTR_ERR(trans);
1212 	}
1213 
1214 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1215 
1216 	cow_start = (u64)-1;
1217 	cur_offset = start;
1218 	while (1) {
1219 		ret = btrfs_lookup_file_extent(trans, root, path, ino,
1220 					       cur_offset, 0);
1221 		if (ret < 0)
1222 			goto error;
1223 		if (ret > 0 && path->slots[0] > 0 && check_prev) {
1224 			leaf = path->nodes[0];
1225 			btrfs_item_key_to_cpu(leaf, &found_key,
1226 					      path->slots[0] - 1);
1227 			if (found_key.objectid == ino &&
1228 			    found_key.type == BTRFS_EXTENT_DATA_KEY)
1229 				path->slots[0]--;
1230 		}
1231 		check_prev = 0;
1232 next_slot:
1233 		leaf = path->nodes[0];
1234 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1235 			ret = btrfs_next_leaf(root, path);
1236 			if (ret < 0)
1237 				goto error;
1238 			if (ret > 0)
1239 				break;
1240 			leaf = path->nodes[0];
1241 		}
1242 
1243 		nocow = 0;
1244 		disk_bytenr = 0;
1245 		num_bytes = 0;
1246 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1247 
1248 		if (found_key.objectid > ino ||
1249 		    found_key.type > BTRFS_EXTENT_DATA_KEY ||
1250 		    found_key.offset > end)
1251 			break;
1252 
1253 		if (found_key.offset > cur_offset) {
1254 			extent_end = found_key.offset;
1255 			extent_type = 0;
1256 			goto out_check;
1257 		}
1258 
1259 		fi = btrfs_item_ptr(leaf, path->slots[0],
1260 				    struct btrfs_file_extent_item);
1261 		extent_type = btrfs_file_extent_type(leaf, fi);
1262 
1263 		ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1264 		if (extent_type == BTRFS_FILE_EXTENT_REG ||
1265 		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1266 			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1267 			extent_offset = btrfs_file_extent_offset(leaf, fi);
1268 			extent_end = found_key.offset +
1269 				btrfs_file_extent_num_bytes(leaf, fi);
1270 			disk_num_bytes =
1271 				btrfs_file_extent_disk_num_bytes(leaf, fi);
1272 			if (extent_end <= start) {
1273 				path->slots[0]++;
1274 				goto next_slot;
1275 			}
1276 			if (disk_bytenr == 0)
1277 				goto out_check;
1278 			if (btrfs_file_extent_compression(leaf, fi) ||
1279 			    btrfs_file_extent_encryption(leaf, fi) ||
1280 			    btrfs_file_extent_other_encoding(leaf, fi))
1281 				goto out_check;
1282 			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1283 				goto out_check;
1284 			if (btrfs_extent_readonly(root, disk_bytenr))
1285 				goto out_check;
1286 			if (btrfs_cross_ref_exist(trans, root, ino,
1287 						  found_key.offset -
1288 						  extent_offset, disk_bytenr))
1289 				goto out_check;
1290 			disk_bytenr += extent_offset;
1291 			disk_bytenr += cur_offset - found_key.offset;
1292 			num_bytes = min(end + 1, extent_end) - cur_offset;
1293 			/*
1294 			 * if there are pending snapshots for this root,
1295 			 * we fall into common COW way.
1296 			 */
1297 			if (!nolock) {
1298 				err = btrfs_start_nocow_write(root);
1299 				if (!err)
1300 					goto out_check;
1301 			}
1302 			/*
1303 			 * force cow if csum exists in the range.
1304 			 * this ensure that csum for a given extent are
1305 			 * either valid or do not exist.
1306 			 */
1307 			if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1308 				goto out_check;
1309 			nocow = 1;
1310 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1311 			extent_end = found_key.offset +
1312 				btrfs_file_extent_inline_len(leaf,
1313 						     path->slots[0], fi);
1314 			extent_end = ALIGN(extent_end, root->sectorsize);
1315 		} else {
1316 			BUG_ON(1);
1317 		}
1318 out_check:
1319 		if (extent_end <= start) {
1320 			path->slots[0]++;
1321 			if (!nolock && nocow)
1322 				btrfs_end_nocow_write(root);
1323 			goto next_slot;
1324 		}
1325 		if (!nocow) {
1326 			if (cow_start == (u64)-1)
1327 				cow_start = cur_offset;
1328 			cur_offset = extent_end;
1329 			if (cur_offset > end)
1330 				break;
1331 			path->slots[0]++;
1332 			goto next_slot;
1333 		}
1334 
1335 		btrfs_release_path(path);
1336 		if (cow_start != (u64)-1) {
1337 			ret = cow_file_range(inode, locked_page,
1338 					     cow_start, found_key.offset - 1,
1339 					     page_started, nr_written, 1);
1340 			if (ret) {
1341 				if (!nolock && nocow)
1342 					btrfs_end_nocow_write(root);
1343 				goto error;
1344 			}
1345 			cow_start = (u64)-1;
1346 		}
1347 
1348 		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1349 			struct extent_map *em;
1350 			struct extent_map_tree *em_tree;
1351 			em_tree = &BTRFS_I(inode)->extent_tree;
1352 			em = alloc_extent_map();
1353 			BUG_ON(!em); /* -ENOMEM */
1354 			em->start = cur_offset;
1355 			em->orig_start = found_key.offset - extent_offset;
1356 			em->len = num_bytes;
1357 			em->block_len = num_bytes;
1358 			em->block_start = disk_bytenr;
1359 			em->orig_block_len = disk_num_bytes;
1360 			em->ram_bytes = ram_bytes;
1361 			em->bdev = root->fs_info->fs_devices->latest_bdev;
1362 			em->mod_start = em->start;
1363 			em->mod_len = em->len;
1364 			set_bit(EXTENT_FLAG_PINNED, &em->flags);
1365 			set_bit(EXTENT_FLAG_FILLING, &em->flags);
1366 			em->generation = -1;
1367 			while (1) {
1368 				write_lock(&em_tree->lock);
1369 				ret = add_extent_mapping(em_tree, em, 1);
1370 				write_unlock(&em_tree->lock);
1371 				if (ret != -EEXIST) {
1372 					free_extent_map(em);
1373 					break;
1374 				}
1375 				btrfs_drop_extent_cache(inode, em->start,
1376 						em->start + em->len - 1, 0);
1377 			}
1378 			type = BTRFS_ORDERED_PREALLOC;
1379 		} else {
1380 			type = BTRFS_ORDERED_NOCOW;
1381 		}
1382 
1383 		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1384 					       num_bytes, num_bytes, type);
1385 		BUG_ON(ret); /* -ENOMEM */
1386 
1387 		if (root->root_key.objectid ==
1388 		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
1389 			ret = btrfs_reloc_clone_csums(inode, cur_offset,
1390 						      num_bytes);
1391 			if (ret) {
1392 				if (!nolock && nocow)
1393 					btrfs_end_nocow_write(root);
1394 				goto error;
1395 			}
1396 		}
1397 
1398 		extent_clear_unlock_delalloc(inode, cur_offset,
1399 					     cur_offset + num_bytes - 1,
1400 					     locked_page, EXTENT_LOCKED |
1401 					     EXTENT_DELALLOC, PAGE_UNLOCK |
1402 					     PAGE_SET_PRIVATE2);
1403 		if (!nolock && nocow)
1404 			btrfs_end_nocow_write(root);
1405 		cur_offset = extent_end;
1406 		if (cur_offset > end)
1407 			break;
1408 	}
1409 	btrfs_release_path(path);
1410 
1411 	if (cur_offset <= end && cow_start == (u64)-1) {
1412 		cow_start = cur_offset;
1413 		cur_offset = end;
1414 	}
1415 
1416 	if (cow_start != (u64)-1) {
1417 		ret = cow_file_range(inode, locked_page, cow_start, end,
1418 				     page_started, nr_written, 1);
1419 		if (ret)
1420 			goto error;
1421 	}
1422 
1423 error:
1424 	err = btrfs_end_transaction(trans, root);
1425 	if (!ret)
1426 		ret = err;
1427 
1428 	if (ret && cur_offset < end)
1429 		extent_clear_unlock_delalloc(inode, cur_offset, end,
1430 					     locked_page, EXTENT_LOCKED |
1431 					     EXTENT_DELALLOC | EXTENT_DEFRAG |
1432 					     EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1433 					     PAGE_CLEAR_DIRTY |
1434 					     PAGE_SET_WRITEBACK |
1435 					     PAGE_END_WRITEBACK);
1436 	btrfs_free_path(path);
1437 	return ret;
1438 }
1439 
1440 /*
1441  * extent_io.c call back to do delayed allocation processing
1442  */
1443 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1444 			      u64 start, u64 end, int *page_started,
1445 			      unsigned long *nr_written)
1446 {
1447 	int ret;
1448 	struct btrfs_root *root = BTRFS_I(inode)->root;
1449 
1450 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1451 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1452 					 page_started, 1, nr_written);
1453 	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1454 		ret = run_delalloc_nocow(inode, locked_page, start, end,
1455 					 page_started, 0, nr_written);
1456 	} else if (!btrfs_test_opt(root, COMPRESS) &&
1457 		   !(BTRFS_I(inode)->force_compress) &&
1458 		   !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1459 		ret = cow_file_range(inode, locked_page, start, end,
1460 				      page_started, nr_written, 1);
1461 	} else {
1462 		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1463 			&BTRFS_I(inode)->runtime_flags);
1464 		ret = cow_file_range_async(inode, locked_page, start, end,
1465 					   page_started, nr_written);
1466 	}
1467 	return ret;
1468 }
1469 
1470 static void btrfs_split_extent_hook(struct inode *inode,
1471 				    struct extent_state *orig, u64 split)
1472 {
1473 	/* not delalloc, ignore it */
1474 	if (!(orig->state & EXTENT_DELALLOC))
1475 		return;
1476 
1477 	spin_lock(&BTRFS_I(inode)->lock);
1478 	BTRFS_I(inode)->outstanding_extents++;
1479 	spin_unlock(&BTRFS_I(inode)->lock);
1480 }
1481 
1482 /*
1483  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1484  * extents so we can keep track of new extents that are just merged onto old
1485  * extents, such as when we are doing sequential writes, so we can properly
1486  * account for the metadata space we'll need.
1487  */
1488 static void btrfs_merge_extent_hook(struct inode *inode,
1489 				    struct extent_state *new,
1490 				    struct extent_state *other)
1491 {
1492 	/* not delalloc, ignore it */
1493 	if (!(other->state & EXTENT_DELALLOC))
1494 		return;
1495 
1496 	spin_lock(&BTRFS_I(inode)->lock);
1497 	BTRFS_I(inode)->outstanding_extents--;
1498 	spin_unlock(&BTRFS_I(inode)->lock);
1499 }
1500 
1501 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1502 				      struct inode *inode)
1503 {
1504 	spin_lock(&root->delalloc_lock);
1505 	if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1506 		list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1507 			      &root->delalloc_inodes);
1508 		set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1509 			&BTRFS_I(inode)->runtime_flags);
1510 		root->nr_delalloc_inodes++;
1511 		if (root->nr_delalloc_inodes == 1) {
1512 			spin_lock(&root->fs_info->delalloc_root_lock);
1513 			BUG_ON(!list_empty(&root->delalloc_root));
1514 			list_add_tail(&root->delalloc_root,
1515 				      &root->fs_info->delalloc_roots);
1516 			spin_unlock(&root->fs_info->delalloc_root_lock);
1517 		}
1518 	}
1519 	spin_unlock(&root->delalloc_lock);
1520 }
1521 
1522 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1523 				     struct inode *inode)
1524 {
1525 	spin_lock(&root->delalloc_lock);
1526 	if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527 		list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1528 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1529 			  &BTRFS_I(inode)->runtime_flags);
1530 		root->nr_delalloc_inodes--;
1531 		if (!root->nr_delalloc_inodes) {
1532 			spin_lock(&root->fs_info->delalloc_root_lock);
1533 			BUG_ON(list_empty(&root->delalloc_root));
1534 			list_del_init(&root->delalloc_root);
1535 			spin_unlock(&root->fs_info->delalloc_root_lock);
1536 		}
1537 	}
1538 	spin_unlock(&root->delalloc_lock);
1539 }
1540 
1541 /*
1542  * extent_io.c set_bit_hook, used to track delayed allocation
1543  * bytes in this file, and to maintain the list of inodes that
1544  * have pending delalloc work to be done.
1545  */
1546 static void btrfs_set_bit_hook(struct inode *inode,
1547 			       struct extent_state *state, unsigned long *bits)
1548 {
1549 
1550 	/*
1551 	 * set_bit and clear bit hooks normally require _irqsave/restore
1552 	 * but in this case, we are only testing for the DELALLOC
1553 	 * bit, which is only set or cleared with irqs on
1554 	 */
1555 	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1556 		struct btrfs_root *root = BTRFS_I(inode)->root;
1557 		u64 len = state->end + 1 - state->start;
1558 		bool do_list = !btrfs_is_free_space_inode(inode);
1559 
1560 		if (*bits & EXTENT_FIRST_DELALLOC) {
1561 			*bits &= ~EXTENT_FIRST_DELALLOC;
1562 		} else {
1563 			spin_lock(&BTRFS_I(inode)->lock);
1564 			BTRFS_I(inode)->outstanding_extents++;
1565 			spin_unlock(&BTRFS_I(inode)->lock);
1566 		}
1567 
1568 		__percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1569 				     root->fs_info->delalloc_batch);
1570 		spin_lock(&BTRFS_I(inode)->lock);
1571 		BTRFS_I(inode)->delalloc_bytes += len;
1572 		if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1573 					 &BTRFS_I(inode)->runtime_flags))
1574 			btrfs_add_delalloc_inodes(root, inode);
1575 		spin_unlock(&BTRFS_I(inode)->lock);
1576 	}
1577 }
1578 
1579 /*
1580  * extent_io.c clear_bit_hook, see set_bit_hook for why
1581  */
1582 static void btrfs_clear_bit_hook(struct inode *inode,
1583 				 struct extent_state *state,
1584 				 unsigned long *bits)
1585 {
1586 	/*
1587 	 * set_bit and clear bit hooks normally require _irqsave/restore
1588 	 * but in this case, we are only testing for the DELALLOC
1589 	 * bit, which is only set or cleared with irqs on
1590 	 */
1591 	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1592 		struct btrfs_root *root = BTRFS_I(inode)->root;
1593 		u64 len = state->end + 1 - state->start;
1594 		bool do_list = !btrfs_is_free_space_inode(inode);
1595 
1596 		if (*bits & EXTENT_FIRST_DELALLOC) {
1597 			*bits &= ~EXTENT_FIRST_DELALLOC;
1598 		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1599 			spin_lock(&BTRFS_I(inode)->lock);
1600 			BTRFS_I(inode)->outstanding_extents--;
1601 			spin_unlock(&BTRFS_I(inode)->lock);
1602 		}
1603 
1604 		/*
1605 		 * We don't reserve metadata space for space cache inodes so we
1606 		 * don't need to call dellalloc_release_metadata if there is an
1607 		 * error.
1608 		 */
1609 		if (*bits & EXTENT_DO_ACCOUNTING &&
1610 		    root != root->fs_info->tree_root)
1611 			btrfs_delalloc_release_metadata(inode, len);
1612 
1613 		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1614 		    && do_list && !(state->state & EXTENT_NORESERVE))
1615 			btrfs_free_reserved_data_space(inode, len);
1616 
1617 		__percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1618 				     root->fs_info->delalloc_batch);
1619 		spin_lock(&BTRFS_I(inode)->lock);
1620 		BTRFS_I(inode)->delalloc_bytes -= len;
1621 		if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1622 		    test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1623 			     &BTRFS_I(inode)->runtime_flags))
1624 			btrfs_del_delalloc_inode(root, inode);
1625 		spin_unlock(&BTRFS_I(inode)->lock);
1626 	}
1627 }
1628 
1629 /*
1630  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631  * we don't create bios that span stripes or chunks
1632  */
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634 			 size_t size, struct bio *bio,
1635 			 unsigned long bio_flags)
1636 {
1637 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1639 	u64 length = 0;
1640 	u64 map_length;
1641 	int ret;
1642 
1643 	if (bio_flags & EXTENT_BIO_COMPRESSED)
1644 		return 0;
1645 
1646 	length = bio->bi_iter.bi_size;
1647 	map_length = length;
1648 	ret = btrfs_map_block(root->fs_info, rw, logical,
1649 			      &map_length, NULL, 0);
1650 	/* Will always return 0 with map_multi == NULL */
1651 	BUG_ON(ret < 0);
1652 	if (map_length < length + size)
1653 		return 1;
1654 	return 0;
1655 }
1656 
1657 /*
1658  * in order to insert checksums into the metadata in large chunks,
1659  * we wait until bio submission time.   All the pages in the bio are
1660  * checksummed and sums are attached onto the ordered extent record.
1661  *
1662  * At IO completion time the cums attached on the ordered extent record
1663  * are inserted into the btree
1664  */
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666 				    struct bio *bio, int mirror_num,
1667 				    unsigned long bio_flags,
1668 				    u64 bio_offset)
1669 {
1670 	struct btrfs_root *root = BTRFS_I(inode)->root;
1671 	int ret = 0;
1672 
1673 	ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674 	BUG_ON(ret); /* -ENOMEM */
1675 	return 0;
1676 }
1677 
1678 /*
1679  * in order to insert checksums into the metadata in large chunks,
1680  * we wait until bio submission time.   All the pages in the bio are
1681  * checksummed and sums are attached onto the ordered extent record.
1682  *
1683  * At IO completion time the cums attached on the ordered extent record
1684  * are inserted into the btree
1685  */
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687 			  int mirror_num, unsigned long bio_flags,
1688 			  u64 bio_offset)
1689 {
1690 	struct btrfs_root *root = BTRFS_I(inode)->root;
1691 	int ret;
1692 
1693 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1694 	if (ret)
1695 		bio_endio(bio, ret);
1696 	return ret;
1697 }
1698 
1699 /*
1700  * extent_io.c submission hook. This does the right thing for csum calculation
1701  * on write, or reading the csums from the tree before a read
1702  */
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704 			  int mirror_num, unsigned long bio_flags,
1705 			  u64 bio_offset)
1706 {
1707 	struct btrfs_root *root = BTRFS_I(inode)->root;
1708 	int ret = 0;
1709 	int skip_sum;
1710 	int metadata = 0;
1711 	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1712 
1713 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1714 
1715 	if (btrfs_is_free_space_inode(inode))
1716 		metadata = 2;
1717 
1718 	if (!(rw & REQ_WRITE)) {
1719 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1720 		if (ret)
1721 			goto out;
1722 
1723 		if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724 			ret = btrfs_submit_compressed_read(inode, bio,
1725 							   mirror_num,
1726 							   bio_flags);
1727 			goto out;
1728 		} else if (!skip_sum) {
1729 			ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1730 			if (ret)
1731 				goto out;
1732 		}
1733 		goto mapit;
1734 	} else if (async && !skip_sum) {
1735 		/* csum items have already been cloned */
1736 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1737 			goto mapit;
1738 		/* we're doing a write, do the async checksumming */
1739 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740 				   inode, rw, bio, mirror_num,
1741 				   bio_flags, bio_offset,
1742 				   __btrfs_submit_bio_start,
1743 				   __btrfs_submit_bio_done);
1744 		goto out;
1745 	} else if (!skip_sum) {
1746 		ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1747 		if (ret)
1748 			goto out;
1749 	}
1750 
1751 mapit:
1752 	ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1753 
1754 out:
1755 	if (ret < 0)
1756 		bio_endio(bio, ret);
1757 	return ret;
1758 }
1759 
1760 /*
1761  * given a list of ordered sums record them in the inode.  This happens
1762  * at IO completion time based on sums calculated at bio submission time.
1763  */
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765 			     struct inode *inode, u64 file_offset,
1766 			     struct list_head *list)
1767 {
1768 	struct btrfs_ordered_sum *sum;
1769 
1770 	list_for_each_entry(sum, list, list) {
1771 		trans->adding_csums = 1;
1772 		btrfs_csum_file_blocks(trans,
1773 		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774 		trans->adding_csums = 0;
1775 	}
1776 	return 0;
1777 }
1778 
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780 			      struct extent_state **cached_state)
1781 {
1782 	WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783 	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784 				   cached_state, GFP_NOFS);
1785 }
1786 
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1789 	struct page *page;
1790 	struct btrfs_work work;
1791 };
1792 
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1794 {
1795 	struct btrfs_writepage_fixup *fixup;
1796 	struct btrfs_ordered_extent *ordered;
1797 	struct extent_state *cached_state = NULL;
1798 	struct page *page;
1799 	struct inode *inode;
1800 	u64 page_start;
1801 	u64 page_end;
1802 	int ret;
1803 
1804 	fixup = container_of(work, struct btrfs_writepage_fixup, work);
1805 	page = fixup->page;
1806 again:
1807 	lock_page(page);
1808 	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809 		ClearPageChecked(page);
1810 		goto out_page;
1811 	}
1812 
1813 	inode = page->mapping->host;
1814 	page_start = page_offset(page);
1815 	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1816 
1817 	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1818 			 &cached_state);
1819 
1820 	/* already ordered? We're done */
1821 	if (PagePrivate2(page))
1822 		goto out;
1823 
1824 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
1825 	if (ordered) {
1826 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827 				     page_end, &cached_state, GFP_NOFS);
1828 		unlock_page(page);
1829 		btrfs_start_ordered_extent(inode, ordered, 1);
1830 		btrfs_put_ordered_extent(ordered);
1831 		goto again;
1832 	}
1833 
1834 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1835 	if (ret) {
1836 		mapping_set_error(page->mapping, ret);
1837 		end_extent_writepage(page, ret, page_start, page_end);
1838 		ClearPageChecked(page);
1839 		goto out;
1840 	 }
1841 
1842 	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843 	ClearPageChecked(page);
1844 	set_page_dirty(page);
1845 out:
1846 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847 			     &cached_state, GFP_NOFS);
1848 out_page:
1849 	unlock_page(page);
1850 	page_cache_release(page);
1851 	kfree(fixup);
1852 }
1853 
1854 /*
1855  * There are a few paths in the higher layers of the kernel that directly
1856  * set the page dirty bit without asking the filesystem if it is a
1857  * good idea.  This causes problems because we want to make sure COW
1858  * properly happens and the data=ordered rules are followed.
1859  *
1860  * In our case any range that doesn't have the ORDERED bit set
1861  * hasn't been properly setup for IO.  We kick off an async process
1862  * to fix it up.  The async helper will wait for ordered extents, set
1863  * the delalloc bit and make it safe to write the page.
1864  */
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1866 {
1867 	struct inode *inode = page->mapping->host;
1868 	struct btrfs_writepage_fixup *fixup;
1869 	struct btrfs_root *root = BTRFS_I(inode)->root;
1870 
1871 	/* this page is properly in the ordered list */
1872 	if (TestClearPagePrivate2(page))
1873 		return 0;
1874 
1875 	if (PageChecked(page))
1876 		return -EAGAIN;
1877 
1878 	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1879 	if (!fixup)
1880 		return -EAGAIN;
1881 
1882 	SetPageChecked(page);
1883 	page_cache_get(page);
1884 	btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1885 	fixup->page = page;
1886 	btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1887 	return -EBUSY;
1888 }
1889 
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891 				       struct inode *inode, u64 file_pos,
1892 				       u64 disk_bytenr, u64 disk_num_bytes,
1893 				       u64 num_bytes, u64 ram_bytes,
1894 				       u8 compression, u8 encryption,
1895 				       u16 other_encoding, int extent_type)
1896 {
1897 	struct btrfs_root *root = BTRFS_I(inode)->root;
1898 	struct btrfs_file_extent_item *fi;
1899 	struct btrfs_path *path;
1900 	struct extent_buffer *leaf;
1901 	struct btrfs_key ins;
1902 	int extent_inserted = 0;
1903 	int ret;
1904 
1905 	path = btrfs_alloc_path();
1906 	if (!path)
1907 		return -ENOMEM;
1908 
1909 	/*
1910 	 * we may be replacing one extent in the tree with another.
1911 	 * The new extent is pinned in the extent map, and we don't want
1912 	 * to drop it from the cache until it is completely in the btree.
1913 	 *
1914 	 * So, tell btrfs_drop_extents to leave this extent in the cache.
1915 	 * the caller is expected to unpin it and allow it to be merged
1916 	 * with the others.
1917 	 */
1918 	ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1919 				   file_pos + num_bytes, NULL, 0,
1920 				   1, sizeof(*fi), &extent_inserted);
1921 	if (ret)
1922 		goto out;
1923 
1924 	if (!extent_inserted) {
1925 		ins.objectid = btrfs_ino(inode);
1926 		ins.offset = file_pos;
1927 		ins.type = BTRFS_EXTENT_DATA_KEY;
1928 
1929 		path->leave_spinning = 1;
1930 		ret = btrfs_insert_empty_item(trans, root, path, &ins,
1931 					      sizeof(*fi));
1932 		if (ret)
1933 			goto out;
1934 	}
1935 	leaf = path->nodes[0];
1936 	fi = btrfs_item_ptr(leaf, path->slots[0],
1937 			    struct btrfs_file_extent_item);
1938 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1939 	btrfs_set_file_extent_type(leaf, fi, extent_type);
1940 	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1941 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1942 	btrfs_set_file_extent_offset(leaf, fi, 0);
1943 	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1944 	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1945 	btrfs_set_file_extent_compression(leaf, fi, compression);
1946 	btrfs_set_file_extent_encryption(leaf, fi, encryption);
1947 	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1948 
1949 	btrfs_mark_buffer_dirty(leaf);
1950 	btrfs_release_path(path);
1951 
1952 	inode_add_bytes(inode, num_bytes);
1953 
1954 	ins.objectid = disk_bytenr;
1955 	ins.offset = disk_num_bytes;
1956 	ins.type = BTRFS_EXTENT_ITEM_KEY;
1957 	ret = btrfs_alloc_reserved_file_extent(trans, root,
1958 					root->root_key.objectid,
1959 					btrfs_ino(inode), file_pos, &ins);
1960 out:
1961 	btrfs_free_path(path);
1962 
1963 	return ret;
1964 }
1965 
1966 /* snapshot-aware defrag */
1967 struct sa_defrag_extent_backref {
1968 	struct rb_node node;
1969 	struct old_sa_defrag_extent *old;
1970 	u64 root_id;
1971 	u64 inum;
1972 	u64 file_pos;
1973 	u64 extent_offset;
1974 	u64 num_bytes;
1975 	u64 generation;
1976 };
1977 
1978 struct old_sa_defrag_extent {
1979 	struct list_head list;
1980 	struct new_sa_defrag_extent *new;
1981 
1982 	u64 extent_offset;
1983 	u64 bytenr;
1984 	u64 offset;
1985 	u64 len;
1986 	int count;
1987 };
1988 
1989 struct new_sa_defrag_extent {
1990 	struct rb_root root;
1991 	struct list_head head;
1992 	struct btrfs_path *path;
1993 	struct inode *inode;
1994 	u64 file_pos;
1995 	u64 len;
1996 	u64 bytenr;
1997 	u64 disk_len;
1998 	u8 compress_type;
1999 };
2000 
2001 static int backref_comp(struct sa_defrag_extent_backref *b1,
2002 			struct sa_defrag_extent_backref *b2)
2003 {
2004 	if (b1->root_id < b2->root_id)
2005 		return -1;
2006 	else if (b1->root_id > b2->root_id)
2007 		return 1;
2008 
2009 	if (b1->inum < b2->inum)
2010 		return -1;
2011 	else if (b1->inum > b2->inum)
2012 		return 1;
2013 
2014 	if (b1->file_pos < b2->file_pos)
2015 		return -1;
2016 	else if (b1->file_pos > b2->file_pos)
2017 		return 1;
2018 
2019 	/*
2020 	 * [------------------------------] ===> (a range of space)
2021 	 *     |<--->|   |<---->| =============> (fs/file tree A)
2022 	 * |<---------------------------->| ===> (fs/file tree B)
2023 	 *
2024 	 * A range of space can refer to two file extents in one tree while
2025 	 * refer to only one file extent in another tree.
2026 	 *
2027 	 * So we may process a disk offset more than one time(two extents in A)
2028 	 * and locate at the same extent(one extent in B), then insert two same
2029 	 * backrefs(both refer to the extent in B).
2030 	 */
2031 	return 0;
2032 }
2033 
2034 static void backref_insert(struct rb_root *root,
2035 			   struct sa_defrag_extent_backref *backref)
2036 {
2037 	struct rb_node **p = &root->rb_node;
2038 	struct rb_node *parent = NULL;
2039 	struct sa_defrag_extent_backref *entry;
2040 	int ret;
2041 
2042 	while (*p) {
2043 		parent = *p;
2044 		entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2045 
2046 		ret = backref_comp(backref, entry);
2047 		if (ret < 0)
2048 			p = &(*p)->rb_left;
2049 		else
2050 			p = &(*p)->rb_right;
2051 	}
2052 
2053 	rb_link_node(&backref->node, parent, p);
2054 	rb_insert_color(&backref->node, root);
2055 }
2056 
2057 /*
2058  * Note the backref might has changed, and in this case we just return 0.
2059  */
2060 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2061 				       void *ctx)
2062 {
2063 	struct btrfs_file_extent_item *extent;
2064 	struct btrfs_fs_info *fs_info;
2065 	struct old_sa_defrag_extent *old = ctx;
2066 	struct new_sa_defrag_extent *new = old->new;
2067 	struct btrfs_path *path = new->path;
2068 	struct btrfs_key key;
2069 	struct btrfs_root *root;
2070 	struct sa_defrag_extent_backref *backref;
2071 	struct extent_buffer *leaf;
2072 	struct inode *inode = new->inode;
2073 	int slot;
2074 	int ret;
2075 	u64 extent_offset;
2076 	u64 num_bytes;
2077 
2078 	if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2079 	    inum == btrfs_ino(inode))
2080 		return 0;
2081 
2082 	key.objectid = root_id;
2083 	key.type = BTRFS_ROOT_ITEM_KEY;
2084 	key.offset = (u64)-1;
2085 
2086 	fs_info = BTRFS_I(inode)->root->fs_info;
2087 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2088 	if (IS_ERR(root)) {
2089 		if (PTR_ERR(root) == -ENOENT)
2090 			return 0;
2091 		WARN_ON(1);
2092 		pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2093 			 inum, offset, root_id);
2094 		return PTR_ERR(root);
2095 	}
2096 
2097 	key.objectid = inum;
2098 	key.type = BTRFS_EXTENT_DATA_KEY;
2099 	if (offset > (u64)-1 << 32)
2100 		key.offset = 0;
2101 	else
2102 		key.offset = offset;
2103 
2104 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2105 	if (WARN_ON(ret < 0))
2106 		return ret;
2107 	ret = 0;
2108 
2109 	while (1) {
2110 		cond_resched();
2111 
2112 		leaf = path->nodes[0];
2113 		slot = path->slots[0];
2114 
2115 		if (slot >= btrfs_header_nritems(leaf)) {
2116 			ret = btrfs_next_leaf(root, path);
2117 			if (ret < 0) {
2118 				goto out;
2119 			} else if (ret > 0) {
2120 				ret = 0;
2121 				goto out;
2122 			}
2123 			continue;
2124 		}
2125 
2126 		path->slots[0]++;
2127 
2128 		btrfs_item_key_to_cpu(leaf, &key, slot);
2129 
2130 		if (key.objectid > inum)
2131 			goto out;
2132 
2133 		if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2134 			continue;
2135 
2136 		extent = btrfs_item_ptr(leaf, slot,
2137 					struct btrfs_file_extent_item);
2138 
2139 		if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2140 			continue;
2141 
2142 		/*
2143 		 * 'offset' refers to the exact key.offset,
2144 		 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2145 		 * (key.offset - extent_offset).
2146 		 */
2147 		if (key.offset != offset)
2148 			continue;
2149 
2150 		extent_offset = btrfs_file_extent_offset(leaf, extent);
2151 		num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2152 
2153 		if (extent_offset >= old->extent_offset + old->offset +
2154 		    old->len || extent_offset + num_bytes <=
2155 		    old->extent_offset + old->offset)
2156 			continue;
2157 		break;
2158 	}
2159 
2160 	backref = kmalloc(sizeof(*backref), GFP_NOFS);
2161 	if (!backref) {
2162 		ret = -ENOENT;
2163 		goto out;
2164 	}
2165 
2166 	backref->root_id = root_id;
2167 	backref->inum = inum;
2168 	backref->file_pos = offset;
2169 	backref->num_bytes = num_bytes;
2170 	backref->extent_offset = extent_offset;
2171 	backref->generation = btrfs_file_extent_generation(leaf, extent);
2172 	backref->old = old;
2173 	backref_insert(&new->root, backref);
2174 	old->count++;
2175 out:
2176 	btrfs_release_path(path);
2177 	WARN_ON(ret);
2178 	return ret;
2179 }
2180 
2181 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2182 				   struct new_sa_defrag_extent *new)
2183 {
2184 	struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2185 	struct old_sa_defrag_extent *old, *tmp;
2186 	int ret;
2187 
2188 	new->path = path;
2189 
2190 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2191 		ret = iterate_inodes_from_logical(old->bytenr +
2192 						  old->extent_offset, fs_info,
2193 						  path, record_one_backref,
2194 						  old);
2195 		if (ret < 0 && ret != -ENOENT)
2196 			return false;
2197 
2198 		/* no backref to be processed for this extent */
2199 		if (!old->count) {
2200 			list_del(&old->list);
2201 			kfree(old);
2202 		}
2203 	}
2204 
2205 	if (list_empty(&new->head))
2206 		return false;
2207 
2208 	return true;
2209 }
2210 
2211 static int relink_is_mergable(struct extent_buffer *leaf,
2212 			      struct btrfs_file_extent_item *fi,
2213 			      struct new_sa_defrag_extent *new)
2214 {
2215 	if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2216 		return 0;
2217 
2218 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2219 		return 0;
2220 
2221 	if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2222 		return 0;
2223 
2224 	if (btrfs_file_extent_encryption(leaf, fi) ||
2225 	    btrfs_file_extent_other_encoding(leaf, fi))
2226 		return 0;
2227 
2228 	return 1;
2229 }
2230 
2231 /*
2232  * Note the backref might has changed, and in this case we just return 0.
2233  */
2234 static noinline int relink_extent_backref(struct btrfs_path *path,
2235 				 struct sa_defrag_extent_backref *prev,
2236 				 struct sa_defrag_extent_backref *backref)
2237 {
2238 	struct btrfs_file_extent_item *extent;
2239 	struct btrfs_file_extent_item *item;
2240 	struct btrfs_ordered_extent *ordered;
2241 	struct btrfs_trans_handle *trans;
2242 	struct btrfs_fs_info *fs_info;
2243 	struct btrfs_root *root;
2244 	struct btrfs_key key;
2245 	struct extent_buffer *leaf;
2246 	struct old_sa_defrag_extent *old = backref->old;
2247 	struct new_sa_defrag_extent *new = old->new;
2248 	struct inode *src_inode = new->inode;
2249 	struct inode *inode;
2250 	struct extent_state *cached = NULL;
2251 	int ret = 0;
2252 	u64 start;
2253 	u64 len;
2254 	u64 lock_start;
2255 	u64 lock_end;
2256 	bool merge = false;
2257 	int index;
2258 
2259 	if (prev && prev->root_id == backref->root_id &&
2260 	    prev->inum == backref->inum &&
2261 	    prev->file_pos + prev->num_bytes == backref->file_pos)
2262 		merge = true;
2263 
2264 	/* step 1: get root */
2265 	key.objectid = backref->root_id;
2266 	key.type = BTRFS_ROOT_ITEM_KEY;
2267 	key.offset = (u64)-1;
2268 
2269 	fs_info = BTRFS_I(src_inode)->root->fs_info;
2270 	index = srcu_read_lock(&fs_info->subvol_srcu);
2271 
2272 	root = btrfs_read_fs_root_no_name(fs_info, &key);
2273 	if (IS_ERR(root)) {
2274 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2275 		if (PTR_ERR(root) == -ENOENT)
2276 			return 0;
2277 		return PTR_ERR(root);
2278 	}
2279 
2280 	if (btrfs_root_readonly(root)) {
2281 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2282 		return 0;
2283 	}
2284 
2285 	/* step 2: get inode */
2286 	key.objectid = backref->inum;
2287 	key.type = BTRFS_INODE_ITEM_KEY;
2288 	key.offset = 0;
2289 
2290 	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2291 	if (IS_ERR(inode)) {
2292 		srcu_read_unlock(&fs_info->subvol_srcu, index);
2293 		return 0;
2294 	}
2295 
2296 	srcu_read_unlock(&fs_info->subvol_srcu, index);
2297 
2298 	/* step 3: relink backref */
2299 	lock_start = backref->file_pos;
2300 	lock_end = backref->file_pos + backref->num_bytes - 1;
2301 	lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2302 			 0, &cached);
2303 
2304 	ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2305 	if (ordered) {
2306 		btrfs_put_ordered_extent(ordered);
2307 		goto out_unlock;
2308 	}
2309 
2310 	trans = btrfs_join_transaction(root);
2311 	if (IS_ERR(trans)) {
2312 		ret = PTR_ERR(trans);
2313 		goto out_unlock;
2314 	}
2315 
2316 	key.objectid = backref->inum;
2317 	key.type = BTRFS_EXTENT_DATA_KEY;
2318 	key.offset = backref->file_pos;
2319 
2320 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2321 	if (ret < 0) {
2322 		goto out_free_path;
2323 	} else if (ret > 0) {
2324 		ret = 0;
2325 		goto out_free_path;
2326 	}
2327 
2328 	extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2329 				struct btrfs_file_extent_item);
2330 
2331 	if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2332 	    backref->generation)
2333 		goto out_free_path;
2334 
2335 	btrfs_release_path(path);
2336 
2337 	start = backref->file_pos;
2338 	if (backref->extent_offset < old->extent_offset + old->offset)
2339 		start += old->extent_offset + old->offset -
2340 			 backref->extent_offset;
2341 
2342 	len = min(backref->extent_offset + backref->num_bytes,
2343 		  old->extent_offset + old->offset + old->len);
2344 	len -= max(backref->extent_offset, old->extent_offset + old->offset);
2345 
2346 	ret = btrfs_drop_extents(trans, root, inode, start,
2347 				 start + len, 1);
2348 	if (ret)
2349 		goto out_free_path;
2350 again:
2351 	key.objectid = btrfs_ino(inode);
2352 	key.type = BTRFS_EXTENT_DATA_KEY;
2353 	key.offset = start;
2354 
2355 	path->leave_spinning = 1;
2356 	if (merge) {
2357 		struct btrfs_file_extent_item *fi;
2358 		u64 extent_len;
2359 		struct btrfs_key found_key;
2360 
2361 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2362 		if (ret < 0)
2363 			goto out_free_path;
2364 
2365 		path->slots[0]--;
2366 		leaf = path->nodes[0];
2367 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2368 
2369 		fi = btrfs_item_ptr(leaf, path->slots[0],
2370 				    struct btrfs_file_extent_item);
2371 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2372 
2373 		if (extent_len + found_key.offset == start &&
2374 		    relink_is_mergable(leaf, fi, new)) {
2375 			btrfs_set_file_extent_num_bytes(leaf, fi,
2376 							extent_len + len);
2377 			btrfs_mark_buffer_dirty(leaf);
2378 			inode_add_bytes(inode, len);
2379 
2380 			ret = 1;
2381 			goto out_free_path;
2382 		} else {
2383 			merge = false;
2384 			btrfs_release_path(path);
2385 			goto again;
2386 		}
2387 	}
2388 
2389 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2390 					sizeof(*extent));
2391 	if (ret) {
2392 		btrfs_abort_transaction(trans, root, ret);
2393 		goto out_free_path;
2394 	}
2395 
2396 	leaf = path->nodes[0];
2397 	item = btrfs_item_ptr(leaf, path->slots[0],
2398 				struct btrfs_file_extent_item);
2399 	btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2400 	btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2401 	btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2402 	btrfs_set_file_extent_num_bytes(leaf, item, len);
2403 	btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2404 	btrfs_set_file_extent_generation(leaf, item, trans->transid);
2405 	btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2406 	btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2407 	btrfs_set_file_extent_encryption(leaf, item, 0);
2408 	btrfs_set_file_extent_other_encoding(leaf, item, 0);
2409 
2410 	btrfs_mark_buffer_dirty(leaf);
2411 	inode_add_bytes(inode, len);
2412 	btrfs_release_path(path);
2413 
2414 	ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2415 			new->disk_len, 0,
2416 			backref->root_id, backref->inum,
2417 			new->file_pos, 0);	/* start - extent_offset */
2418 	if (ret) {
2419 		btrfs_abort_transaction(trans, root, ret);
2420 		goto out_free_path;
2421 	}
2422 
2423 	ret = 1;
2424 out_free_path:
2425 	btrfs_release_path(path);
2426 	path->leave_spinning = 0;
2427 	btrfs_end_transaction(trans, root);
2428 out_unlock:
2429 	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2430 			     &cached, GFP_NOFS);
2431 	iput(inode);
2432 	return ret;
2433 }
2434 
2435 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2436 {
2437 	struct old_sa_defrag_extent *old, *tmp;
2438 
2439 	if (!new)
2440 		return;
2441 
2442 	list_for_each_entry_safe(old, tmp, &new->head, list) {
2443 		list_del(&old->list);
2444 		kfree(old);
2445 	}
2446 	kfree(new);
2447 }
2448 
2449 static void relink_file_extents(struct new_sa_defrag_extent *new)
2450 {
2451 	struct btrfs_path *path;
2452 	struct sa_defrag_extent_backref *backref;
2453 	struct sa_defrag_extent_backref *prev = NULL;
2454 	struct inode *inode;
2455 	struct btrfs_root *root;
2456 	struct rb_node *node;
2457 	int ret;
2458 
2459 	inode = new->inode;
2460 	root = BTRFS_I(inode)->root;
2461 
2462 	path = btrfs_alloc_path();
2463 	if (!path)
2464 		return;
2465 
2466 	if (!record_extent_backrefs(path, new)) {
2467 		btrfs_free_path(path);
2468 		goto out;
2469 	}
2470 	btrfs_release_path(path);
2471 
2472 	while (1) {
2473 		node = rb_first(&new->root);
2474 		if (!node)
2475 			break;
2476 		rb_erase(node, &new->root);
2477 
2478 		backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2479 
2480 		ret = relink_extent_backref(path, prev, backref);
2481 		WARN_ON(ret < 0);
2482 
2483 		kfree(prev);
2484 
2485 		if (ret == 1)
2486 			prev = backref;
2487 		else
2488 			prev = NULL;
2489 		cond_resched();
2490 	}
2491 	kfree(prev);
2492 
2493 	btrfs_free_path(path);
2494 out:
2495 	free_sa_defrag_extent(new);
2496 
2497 	atomic_dec(&root->fs_info->defrag_running);
2498 	wake_up(&root->fs_info->transaction_wait);
2499 }
2500 
2501 static struct new_sa_defrag_extent *
2502 record_old_file_extents(struct inode *inode,
2503 			struct btrfs_ordered_extent *ordered)
2504 {
2505 	struct btrfs_root *root = BTRFS_I(inode)->root;
2506 	struct btrfs_path *path;
2507 	struct btrfs_key key;
2508 	struct old_sa_defrag_extent *old;
2509 	struct new_sa_defrag_extent *new;
2510 	int ret;
2511 
2512 	new = kmalloc(sizeof(*new), GFP_NOFS);
2513 	if (!new)
2514 		return NULL;
2515 
2516 	new->inode = inode;
2517 	new->file_pos = ordered->file_offset;
2518 	new->len = ordered->len;
2519 	new->bytenr = ordered->start;
2520 	new->disk_len = ordered->disk_len;
2521 	new->compress_type = ordered->compress_type;
2522 	new->root = RB_ROOT;
2523 	INIT_LIST_HEAD(&new->head);
2524 
2525 	path = btrfs_alloc_path();
2526 	if (!path)
2527 		goto out_kfree;
2528 
2529 	key.objectid = btrfs_ino(inode);
2530 	key.type = BTRFS_EXTENT_DATA_KEY;
2531 	key.offset = new->file_pos;
2532 
2533 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2534 	if (ret < 0)
2535 		goto out_free_path;
2536 	if (ret > 0 && path->slots[0] > 0)
2537 		path->slots[0]--;
2538 
2539 	/* find out all the old extents for the file range */
2540 	while (1) {
2541 		struct btrfs_file_extent_item *extent;
2542 		struct extent_buffer *l;
2543 		int slot;
2544 		u64 num_bytes;
2545 		u64 offset;
2546 		u64 end;
2547 		u64 disk_bytenr;
2548 		u64 extent_offset;
2549 
2550 		l = path->nodes[0];
2551 		slot = path->slots[0];
2552 
2553 		if (slot >= btrfs_header_nritems(l)) {
2554 			ret = btrfs_next_leaf(root, path);
2555 			if (ret < 0)
2556 				goto out_free_path;
2557 			else if (ret > 0)
2558 				break;
2559 			continue;
2560 		}
2561 
2562 		btrfs_item_key_to_cpu(l, &key, slot);
2563 
2564 		if (key.objectid != btrfs_ino(inode))
2565 			break;
2566 		if (key.type != BTRFS_EXTENT_DATA_KEY)
2567 			break;
2568 		if (key.offset >= new->file_pos + new->len)
2569 			break;
2570 
2571 		extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2572 
2573 		num_bytes = btrfs_file_extent_num_bytes(l, extent);
2574 		if (key.offset + num_bytes < new->file_pos)
2575 			goto next;
2576 
2577 		disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2578 		if (!disk_bytenr)
2579 			goto next;
2580 
2581 		extent_offset = btrfs_file_extent_offset(l, extent);
2582 
2583 		old = kmalloc(sizeof(*old), GFP_NOFS);
2584 		if (!old)
2585 			goto out_free_path;
2586 
2587 		offset = max(new->file_pos, key.offset);
2588 		end = min(new->file_pos + new->len, key.offset + num_bytes);
2589 
2590 		old->bytenr = disk_bytenr;
2591 		old->extent_offset = extent_offset;
2592 		old->offset = offset - key.offset;
2593 		old->len = end - offset;
2594 		old->new = new;
2595 		old->count = 0;
2596 		list_add_tail(&old->list, &new->head);
2597 next:
2598 		path->slots[0]++;
2599 		cond_resched();
2600 	}
2601 
2602 	btrfs_free_path(path);
2603 	atomic_inc(&root->fs_info->defrag_running);
2604 
2605 	return new;
2606 
2607 out_free_path:
2608 	btrfs_free_path(path);
2609 out_kfree:
2610 	free_sa_defrag_extent(new);
2611 	return NULL;
2612 }
2613 
2614 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2615 					 u64 start, u64 len)
2616 {
2617 	struct btrfs_block_group_cache *cache;
2618 
2619 	cache = btrfs_lookup_block_group(root->fs_info, start);
2620 	ASSERT(cache);
2621 
2622 	spin_lock(&cache->lock);
2623 	cache->delalloc_bytes -= len;
2624 	spin_unlock(&cache->lock);
2625 
2626 	btrfs_put_block_group(cache);
2627 }
2628 
2629 /* as ordered data IO finishes, this gets called so we can finish
2630  * an ordered extent if the range of bytes in the file it covers are
2631  * fully written.
2632  */
2633 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2634 {
2635 	struct inode *inode = ordered_extent->inode;
2636 	struct btrfs_root *root = BTRFS_I(inode)->root;
2637 	struct btrfs_trans_handle *trans = NULL;
2638 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2639 	struct extent_state *cached_state = NULL;
2640 	struct new_sa_defrag_extent *new = NULL;
2641 	int compress_type = 0;
2642 	int ret = 0;
2643 	u64 logical_len = ordered_extent->len;
2644 	bool nolock;
2645 	bool truncated = false;
2646 
2647 	nolock = btrfs_is_free_space_inode(inode);
2648 
2649 	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2650 		ret = -EIO;
2651 		goto out;
2652 	}
2653 
2654 	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2655 		truncated = true;
2656 		logical_len = ordered_extent->truncated_len;
2657 		/* Truncated the entire extent, don't bother adding */
2658 		if (!logical_len)
2659 			goto out;
2660 	}
2661 
2662 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2663 		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2664 		btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2665 		if (nolock)
2666 			trans = btrfs_join_transaction_nolock(root);
2667 		else
2668 			trans = btrfs_join_transaction(root);
2669 		if (IS_ERR(trans)) {
2670 			ret = PTR_ERR(trans);
2671 			trans = NULL;
2672 			goto out;
2673 		}
2674 		trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2675 		ret = btrfs_update_inode_fallback(trans, root, inode);
2676 		if (ret) /* -ENOMEM or corruption */
2677 			btrfs_abort_transaction(trans, root, ret);
2678 		goto out;
2679 	}
2680 
2681 	lock_extent_bits(io_tree, ordered_extent->file_offset,
2682 			 ordered_extent->file_offset + ordered_extent->len - 1,
2683 			 0, &cached_state);
2684 
2685 	ret = test_range_bit(io_tree, ordered_extent->file_offset,
2686 			ordered_extent->file_offset + ordered_extent->len - 1,
2687 			EXTENT_DEFRAG, 1, cached_state);
2688 	if (ret) {
2689 		u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2690 		if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2691 			/* the inode is shared */
2692 			new = record_old_file_extents(inode, ordered_extent);
2693 
2694 		clear_extent_bit(io_tree, ordered_extent->file_offset,
2695 			ordered_extent->file_offset + ordered_extent->len - 1,
2696 			EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2697 	}
2698 
2699 	if (nolock)
2700 		trans = btrfs_join_transaction_nolock(root);
2701 	else
2702 		trans = btrfs_join_transaction(root);
2703 	if (IS_ERR(trans)) {
2704 		ret = PTR_ERR(trans);
2705 		trans = NULL;
2706 		goto out_unlock;
2707 	}
2708 
2709 	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2710 
2711 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2712 		compress_type = ordered_extent->compress_type;
2713 	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2714 		BUG_ON(compress_type);
2715 		ret = btrfs_mark_extent_written(trans, inode,
2716 						ordered_extent->file_offset,
2717 						ordered_extent->file_offset +
2718 						logical_len);
2719 	} else {
2720 		BUG_ON(root == root->fs_info->tree_root);
2721 		ret = insert_reserved_file_extent(trans, inode,
2722 						ordered_extent->file_offset,
2723 						ordered_extent->start,
2724 						ordered_extent->disk_len,
2725 						logical_len, logical_len,
2726 						compress_type, 0, 0,
2727 						BTRFS_FILE_EXTENT_REG);
2728 		if (!ret)
2729 			btrfs_release_delalloc_bytes(root,
2730 						     ordered_extent->start,
2731 						     ordered_extent->disk_len);
2732 	}
2733 	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2734 			   ordered_extent->file_offset, ordered_extent->len,
2735 			   trans->transid);
2736 	if (ret < 0) {
2737 		btrfs_abort_transaction(trans, root, ret);
2738 		goto out_unlock;
2739 	}
2740 
2741 	add_pending_csums(trans, inode, ordered_extent->file_offset,
2742 			  &ordered_extent->list);
2743 
2744 	btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2745 	ret = btrfs_update_inode_fallback(trans, root, inode);
2746 	if (ret) { /* -ENOMEM or corruption */
2747 		btrfs_abort_transaction(trans, root, ret);
2748 		goto out_unlock;
2749 	}
2750 	ret = 0;
2751 out_unlock:
2752 	unlock_extent_cached(io_tree, ordered_extent->file_offset,
2753 			     ordered_extent->file_offset +
2754 			     ordered_extent->len - 1, &cached_state, GFP_NOFS);
2755 out:
2756 	if (root != root->fs_info->tree_root)
2757 		btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2758 	if (trans)
2759 		btrfs_end_transaction(trans, root);
2760 
2761 	if (ret || truncated) {
2762 		u64 start, end;
2763 
2764 		if (truncated)
2765 			start = ordered_extent->file_offset + logical_len;
2766 		else
2767 			start = ordered_extent->file_offset;
2768 		end = ordered_extent->file_offset + ordered_extent->len - 1;
2769 		clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2770 
2771 		/* Drop the cache for the part of the extent we didn't write. */
2772 		btrfs_drop_extent_cache(inode, start, end, 0);
2773 
2774 		/*
2775 		 * If the ordered extent had an IOERR or something else went
2776 		 * wrong we need to return the space for this ordered extent
2777 		 * back to the allocator.  We only free the extent in the
2778 		 * truncated case if we didn't write out the extent at all.
2779 		 */
2780 		if ((ret || !logical_len) &&
2781 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2782 		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2783 			btrfs_free_reserved_extent(root, ordered_extent->start,
2784 						   ordered_extent->disk_len, 1);
2785 	}
2786 
2787 
2788 	/*
2789 	 * This needs to be done to make sure anybody waiting knows we are done
2790 	 * updating everything for this ordered extent.
2791 	 */
2792 	btrfs_remove_ordered_extent(inode, ordered_extent);
2793 
2794 	/* for snapshot-aware defrag */
2795 	if (new) {
2796 		if (ret) {
2797 			free_sa_defrag_extent(new);
2798 			atomic_dec(&root->fs_info->defrag_running);
2799 		} else {
2800 			relink_file_extents(new);
2801 		}
2802 	}
2803 
2804 	/* once for us */
2805 	btrfs_put_ordered_extent(ordered_extent);
2806 	/* once for the tree */
2807 	btrfs_put_ordered_extent(ordered_extent);
2808 
2809 	return ret;
2810 }
2811 
2812 static void finish_ordered_fn(struct btrfs_work *work)
2813 {
2814 	struct btrfs_ordered_extent *ordered_extent;
2815 	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2816 	btrfs_finish_ordered_io(ordered_extent);
2817 }
2818 
2819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2820 				struct extent_state *state, int uptodate)
2821 {
2822 	struct inode *inode = page->mapping->host;
2823 	struct btrfs_root *root = BTRFS_I(inode)->root;
2824 	struct btrfs_ordered_extent *ordered_extent = NULL;
2825 	struct btrfs_workqueue *workers;
2826 
2827 	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2828 
2829 	ClearPagePrivate2(page);
2830 	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2831 					    end - start + 1, uptodate))
2832 		return 0;
2833 
2834 	btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2835 
2836 	if (btrfs_is_free_space_inode(inode))
2837 		workers = root->fs_info->endio_freespace_worker;
2838 	else
2839 		workers = root->fs_info->endio_write_workers;
2840 	btrfs_queue_work(workers, &ordered_extent->work);
2841 
2842 	return 0;
2843 }
2844 
2845 /*
2846  * when reads are done, we need to check csums to verify the data is correct
2847  * if there's a match, we allow the bio to finish.  If not, the code in
2848  * extent_io.c will try to find good copies for us.
2849  */
2850 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2851 				      u64 phy_offset, struct page *page,
2852 				      u64 start, u64 end, int mirror)
2853 {
2854 	size_t offset = start - page_offset(page);
2855 	struct inode *inode = page->mapping->host;
2856 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2857 	char *kaddr;
2858 	struct btrfs_root *root = BTRFS_I(inode)->root;
2859 	u32 csum_expected;
2860 	u32 csum = ~(u32)0;
2861 	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2862 	                              DEFAULT_RATELIMIT_BURST);
2863 
2864 	if (PageChecked(page)) {
2865 		ClearPageChecked(page);
2866 		goto good;
2867 	}
2868 
2869 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2870 		goto good;
2871 
2872 	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2873 	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2874 		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2875 				  GFP_NOFS);
2876 		return 0;
2877 	}
2878 
2879 	phy_offset >>= inode->i_sb->s_blocksize_bits;
2880 	csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2881 
2882 	kaddr = kmap_atomic(page);
2883 	csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2884 	btrfs_csum_final(csum, (char *)&csum);
2885 	if (csum != csum_expected)
2886 		goto zeroit;
2887 
2888 	kunmap_atomic(kaddr);
2889 good:
2890 	return 0;
2891 
2892 zeroit:
2893 	if (__ratelimit(&_rs))
2894 		btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2895 			btrfs_ino(page->mapping->host), start, csum, csum_expected);
2896 	memset(kaddr + offset, 1, end - start + 1);
2897 	flush_dcache_page(page);
2898 	kunmap_atomic(kaddr);
2899 	if (csum_expected == 0)
2900 		return 0;
2901 	return -EIO;
2902 }
2903 
2904 struct delayed_iput {
2905 	struct list_head list;
2906 	struct inode *inode;
2907 };
2908 
2909 /* JDM: If this is fs-wide, why can't we add a pointer to
2910  * btrfs_inode instead and avoid the allocation? */
2911 void btrfs_add_delayed_iput(struct inode *inode)
2912 {
2913 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2914 	struct delayed_iput *delayed;
2915 
2916 	if (atomic_add_unless(&inode->i_count, -1, 1))
2917 		return;
2918 
2919 	delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2920 	delayed->inode = inode;
2921 
2922 	spin_lock(&fs_info->delayed_iput_lock);
2923 	list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2924 	spin_unlock(&fs_info->delayed_iput_lock);
2925 }
2926 
2927 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2928 {
2929 	LIST_HEAD(list);
2930 	struct btrfs_fs_info *fs_info = root->fs_info;
2931 	struct delayed_iput *delayed;
2932 	int empty;
2933 
2934 	spin_lock(&fs_info->delayed_iput_lock);
2935 	empty = list_empty(&fs_info->delayed_iputs);
2936 	spin_unlock(&fs_info->delayed_iput_lock);
2937 	if (empty)
2938 		return;
2939 
2940 	spin_lock(&fs_info->delayed_iput_lock);
2941 	list_splice_init(&fs_info->delayed_iputs, &list);
2942 	spin_unlock(&fs_info->delayed_iput_lock);
2943 
2944 	while (!list_empty(&list)) {
2945 		delayed = list_entry(list.next, struct delayed_iput, list);
2946 		list_del(&delayed->list);
2947 		iput(delayed->inode);
2948 		kfree(delayed);
2949 	}
2950 }
2951 
2952 /*
2953  * This is called in transaction commit time. If there are no orphan
2954  * files in the subvolume, it removes orphan item and frees block_rsv
2955  * structure.
2956  */
2957 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2958 			      struct btrfs_root *root)
2959 {
2960 	struct btrfs_block_rsv *block_rsv;
2961 	int ret;
2962 
2963 	if (atomic_read(&root->orphan_inodes) ||
2964 	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2965 		return;
2966 
2967 	spin_lock(&root->orphan_lock);
2968 	if (atomic_read(&root->orphan_inodes)) {
2969 		spin_unlock(&root->orphan_lock);
2970 		return;
2971 	}
2972 
2973 	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2974 		spin_unlock(&root->orphan_lock);
2975 		return;
2976 	}
2977 
2978 	block_rsv = root->orphan_block_rsv;
2979 	root->orphan_block_rsv = NULL;
2980 	spin_unlock(&root->orphan_lock);
2981 
2982 	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2983 	    btrfs_root_refs(&root->root_item) > 0) {
2984 		ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2985 					    root->root_key.objectid);
2986 		if (ret)
2987 			btrfs_abort_transaction(trans, root, ret);
2988 		else
2989 			clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
2990 				  &root->state);
2991 	}
2992 
2993 	if (block_rsv) {
2994 		WARN_ON(block_rsv->size > 0);
2995 		btrfs_free_block_rsv(root, block_rsv);
2996 	}
2997 }
2998 
2999 /*
3000  * This creates an orphan entry for the given inode in case something goes
3001  * wrong in the middle of an unlink/truncate.
3002  *
3003  * NOTE: caller of this function should reserve 5 units of metadata for
3004  *	 this function.
3005  */
3006 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3007 {
3008 	struct btrfs_root *root = BTRFS_I(inode)->root;
3009 	struct btrfs_block_rsv *block_rsv = NULL;
3010 	int reserve = 0;
3011 	int insert = 0;
3012 	int ret;
3013 
3014 	if (!root->orphan_block_rsv) {
3015 		block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3016 		if (!block_rsv)
3017 			return -ENOMEM;
3018 	}
3019 
3020 	spin_lock(&root->orphan_lock);
3021 	if (!root->orphan_block_rsv) {
3022 		root->orphan_block_rsv = block_rsv;
3023 	} else if (block_rsv) {
3024 		btrfs_free_block_rsv(root, block_rsv);
3025 		block_rsv = NULL;
3026 	}
3027 
3028 	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3029 			      &BTRFS_I(inode)->runtime_flags)) {
3030 #if 0
3031 		/*
3032 		 * For proper ENOSPC handling, we should do orphan
3033 		 * cleanup when mounting. But this introduces backward
3034 		 * compatibility issue.
3035 		 */
3036 		if (!xchg(&root->orphan_item_inserted, 1))
3037 			insert = 2;
3038 		else
3039 			insert = 1;
3040 #endif
3041 		insert = 1;
3042 		atomic_inc(&root->orphan_inodes);
3043 	}
3044 
3045 	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3046 			      &BTRFS_I(inode)->runtime_flags))
3047 		reserve = 1;
3048 	spin_unlock(&root->orphan_lock);
3049 
3050 	/* grab metadata reservation from transaction handle */
3051 	if (reserve) {
3052 		ret = btrfs_orphan_reserve_metadata(trans, inode);
3053 		BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3054 	}
3055 
3056 	/* insert an orphan item to track this unlinked/truncated file */
3057 	if (insert >= 1) {
3058 		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3059 		if (ret) {
3060 			atomic_dec(&root->orphan_inodes);
3061 			if (reserve) {
3062 				clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3063 					  &BTRFS_I(inode)->runtime_flags);
3064 				btrfs_orphan_release_metadata(inode);
3065 			}
3066 			if (ret != -EEXIST) {
3067 				clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3068 					  &BTRFS_I(inode)->runtime_flags);
3069 				btrfs_abort_transaction(trans, root, ret);
3070 				return ret;
3071 			}
3072 		}
3073 		ret = 0;
3074 	}
3075 
3076 	/* insert an orphan item to track subvolume contains orphan files */
3077 	if (insert >= 2) {
3078 		ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3079 					       root->root_key.objectid);
3080 		if (ret && ret != -EEXIST) {
3081 			btrfs_abort_transaction(trans, root, ret);
3082 			return ret;
3083 		}
3084 	}
3085 	return 0;
3086 }
3087 
3088 /*
3089  * We have done the truncate/delete so we can go ahead and remove the orphan
3090  * item for this particular inode.
3091  */
3092 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3093 			    struct inode *inode)
3094 {
3095 	struct btrfs_root *root = BTRFS_I(inode)->root;
3096 	int delete_item = 0;
3097 	int release_rsv = 0;
3098 	int ret = 0;
3099 
3100 	spin_lock(&root->orphan_lock);
3101 	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3102 			       &BTRFS_I(inode)->runtime_flags))
3103 		delete_item = 1;
3104 
3105 	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3106 			       &BTRFS_I(inode)->runtime_flags))
3107 		release_rsv = 1;
3108 	spin_unlock(&root->orphan_lock);
3109 
3110 	if (delete_item) {
3111 		atomic_dec(&root->orphan_inodes);
3112 		if (trans)
3113 			ret = btrfs_del_orphan_item(trans, root,
3114 						    btrfs_ino(inode));
3115 	}
3116 
3117 	if (release_rsv)
3118 		btrfs_orphan_release_metadata(inode);
3119 
3120 	return ret;
3121 }
3122 
3123 /*
3124  * this cleans up any orphans that may be left on the list from the last use
3125  * of this root.
3126  */
3127 int btrfs_orphan_cleanup(struct btrfs_root *root)
3128 {
3129 	struct btrfs_path *path;
3130 	struct extent_buffer *leaf;
3131 	struct btrfs_key key, found_key;
3132 	struct btrfs_trans_handle *trans;
3133 	struct inode *inode;
3134 	u64 last_objectid = 0;
3135 	int ret = 0, nr_unlink = 0, nr_truncate = 0;
3136 
3137 	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3138 		return 0;
3139 
3140 	path = btrfs_alloc_path();
3141 	if (!path) {
3142 		ret = -ENOMEM;
3143 		goto out;
3144 	}
3145 	path->reada = -1;
3146 
3147 	key.objectid = BTRFS_ORPHAN_OBJECTID;
3148 	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3149 	key.offset = (u64)-1;
3150 
3151 	while (1) {
3152 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3153 		if (ret < 0)
3154 			goto out;
3155 
3156 		/*
3157 		 * if ret == 0 means we found what we were searching for, which
3158 		 * is weird, but possible, so only screw with path if we didn't
3159 		 * find the key and see if we have stuff that matches
3160 		 */
3161 		if (ret > 0) {
3162 			ret = 0;
3163 			if (path->slots[0] == 0)
3164 				break;
3165 			path->slots[0]--;
3166 		}
3167 
3168 		/* pull out the item */
3169 		leaf = path->nodes[0];
3170 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3171 
3172 		/* make sure the item matches what we want */
3173 		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3174 			break;
3175 		if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3176 			break;
3177 
3178 		/* release the path since we're done with it */
3179 		btrfs_release_path(path);
3180 
3181 		/*
3182 		 * this is where we are basically btrfs_lookup, without the
3183 		 * crossing root thing.  we store the inode number in the
3184 		 * offset of the orphan item.
3185 		 */
3186 
3187 		if (found_key.offset == last_objectid) {
3188 			btrfs_err(root->fs_info,
3189 				"Error removing orphan entry, stopping orphan cleanup");
3190 			ret = -EINVAL;
3191 			goto out;
3192 		}
3193 
3194 		last_objectid = found_key.offset;
3195 
3196 		found_key.objectid = found_key.offset;
3197 		found_key.type = BTRFS_INODE_ITEM_KEY;
3198 		found_key.offset = 0;
3199 		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3200 		ret = PTR_ERR_OR_ZERO(inode);
3201 		if (ret && ret != -ESTALE)
3202 			goto out;
3203 
3204 		if (ret == -ESTALE && root == root->fs_info->tree_root) {
3205 			struct btrfs_root *dead_root;
3206 			struct btrfs_fs_info *fs_info = root->fs_info;
3207 			int is_dead_root = 0;
3208 
3209 			/*
3210 			 * this is an orphan in the tree root. Currently these
3211 			 * could come from 2 sources:
3212 			 *  a) a snapshot deletion in progress
3213 			 *  b) a free space cache inode
3214 			 * We need to distinguish those two, as the snapshot
3215 			 * orphan must not get deleted.
3216 			 * find_dead_roots already ran before us, so if this
3217 			 * is a snapshot deletion, we should find the root
3218 			 * in the dead_roots list
3219 			 */
3220 			spin_lock(&fs_info->trans_lock);
3221 			list_for_each_entry(dead_root, &fs_info->dead_roots,
3222 					    root_list) {
3223 				if (dead_root->root_key.objectid ==
3224 				    found_key.objectid) {
3225 					is_dead_root = 1;
3226 					break;
3227 				}
3228 			}
3229 			spin_unlock(&fs_info->trans_lock);
3230 			if (is_dead_root) {
3231 				/* prevent this orphan from being found again */
3232 				key.offset = found_key.objectid - 1;
3233 				continue;
3234 			}
3235 		}
3236 		/*
3237 		 * Inode is already gone but the orphan item is still there,
3238 		 * kill the orphan item.
3239 		 */
3240 		if (ret == -ESTALE) {
3241 			trans = btrfs_start_transaction(root, 1);
3242 			if (IS_ERR(trans)) {
3243 				ret = PTR_ERR(trans);
3244 				goto out;
3245 			}
3246 			btrfs_debug(root->fs_info, "auto deleting %Lu",
3247 				found_key.objectid);
3248 			ret = btrfs_del_orphan_item(trans, root,
3249 						    found_key.objectid);
3250 			btrfs_end_transaction(trans, root);
3251 			if (ret)
3252 				goto out;
3253 			continue;
3254 		}
3255 
3256 		/*
3257 		 * add this inode to the orphan list so btrfs_orphan_del does
3258 		 * the proper thing when we hit it
3259 		 */
3260 		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3261 			&BTRFS_I(inode)->runtime_flags);
3262 		atomic_inc(&root->orphan_inodes);
3263 
3264 		/* if we have links, this was a truncate, lets do that */
3265 		if (inode->i_nlink) {
3266 			if (WARN_ON(!S_ISREG(inode->i_mode))) {
3267 				iput(inode);
3268 				continue;
3269 			}
3270 			nr_truncate++;
3271 
3272 			/* 1 for the orphan item deletion. */
3273 			trans = btrfs_start_transaction(root, 1);
3274 			if (IS_ERR(trans)) {
3275 				iput(inode);
3276 				ret = PTR_ERR(trans);
3277 				goto out;
3278 			}
3279 			ret = btrfs_orphan_add(trans, inode);
3280 			btrfs_end_transaction(trans, root);
3281 			if (ret) {
3282 				iput(inode);
3283 				goto out;
3284 			}
3285 
3286 			ret = btrfs_truncate(inode);
3287 			if (ret)
3288 				btrfs_orphan_del(NULL, inode);
3289 		} else {
3290 			nr_unlink++;
3291 		}
3292 
3293 		/* this will do delete_inode and everything for us */
3294 		iput(inode);
3295 		if (ret)
3296 			goto out;
3297 	}
3298 	/* release the path since we're done with it */
3299 	btrfs_release_path(path);
3300 
3301 	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3302 
3303 	if (root->orphan_block_rsv)
3304 		btrfs_block_rsv_release(root, root->orphan_block_rsv,
3305 					(u64)-1);
3306 
3307 	if (root->orphan_block_rsv ||
3308 	    test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3309 		trans = btrfs_join_transaction(root);
3310 		if (!IS_ERR(trans))
3311 			btrfs_end_transaction(trans, root);
3312 	}
3313 
3314 	if (nr_unlink)
3315 		btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3316 	if (nr_truncate)
3317 		btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3318 
3319 out:
3320 	if (ret)
3321 		btrfs_crit(root->fs_info,
3322 			"could not do orphan cleanup %d", ret);
3323 	btrfs_free_path(path);
3324 	return ret;
3325 }
3326 
3327 /*
3328  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3329  * don't find any xattrs, we know there can't be any acls.
3330  *
3331  * slot is the slot the inode is in, objectid is the objectid of the inode
3332  */
3333 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3334 					  int slot, u64 objectid,
3335 					  int *first_xattr_slot)
3336 {
3337 	u32 nritems = btrfs_header_nritems(leaf);
3338 	struct btrfs_key found_key;
3339 	static u64 xattr_access = 0;
3340 	static u64 xattr_default = 0;
3341 	int scanned = 0;
3342 
3343 	if (!xattr_access) {
3344 		xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3345 					strlen(POSIX_ACL_XATTR_ACCESS));
3346 		xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3347 					strlen(POSIX_ACL_XATTR_DEFAULT));
3348 	}
3349 
3350 	slot++;
3351 	*first_xattr_slot = -1;
3352 	while (slot < nritems) {
3353 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3354 
3355 		/* we found a different objectid, there must not be acls */
3356 		if (found_key.objectid != objectid)
3357 			return 0;
3358 
3359 		/* we found an xattr, assume we've got an acl */
3360 		if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3361 			if (*first_xattr_slot == -1)
3362 				*first_xattr_slot = slot;
3363 			if (found_key.offset == xattr_access ||
3364 			    found_key.offset == xattr_default)
3365 				return 1;
3366 		}
3367 
3368 		/*
3369 		 * we found a key greater than an xattr key, there can't
3370 		 * be any acls later on
3371 		 */
3372 		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3373 			return 0;
3374 
3375 		slot++;
3376 		scanned++;
3377 
3378 		/*
3379 		 * it goes inode, inode backrefs, xattrs, extents,
3380 		 * so if there are a ton of hard links to an inode there can
3381 		 * be a lot of backrefs.  Don't waste time searching too hard,
3382 		 * this is just an optimization
3383 		 */
3384 		if (scanned >= 8)
3385 			break;
3386 	}
3387 	/* we hit the end of the leaf before we found an xattr or
3388 	 * something larger than an xattr.  We have to assume the inode
3389 	 * has acls
3390 	 */
3391 	if (*first_xattr_slot == -1)
3392 		*first_xattr_slot = slot;
3393 	return 1;
3394 }
3395 
3396 /*
3397  * read an inode from the btree into the in-memory inode
3398  */
3399 static void btrfs_read_locked_inode(struct inode *inode)
3400 {
3401 	struct btrfs_path *path;
3402 	struct extent_buffer *leaf;
3403 	struct btrfs_inode_item *inode_item;
3404 	struct btrfs_timespec *tspec;
3405 	struct btrfs_root *root = BTRFS_I(inode)->root;
3406 	struct btrfs_key location;
3407 	unsigned long ptr;
3408 	int maybe_acls;
3409 	u32 rdev;
3410 	int ret;
3411 	bool filled = false;
3412 	int first_xattr_slot;
3413 
3414 	ret = btrfs_fill_inode(inode, &rdev);
3415 	if (!ret)
3416 		filled = true;
3417 
3418 	path = btrfs_alloc_path();
3419 	if (!path)
3420 		goto make_bad;
3421 
3422 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3423 
3424 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3425 	if (ret)
3426 		goto make_bad;
3427 
3428 	leaf = path->nodes[0];
3429 
3430 	if (filled)
3431 		goto cache_index;
3432 
3433 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3434 				    struct btrfs_inode_item);
3435 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3436 	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3437 	i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3438 	i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3439 	btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3440 
3441 	tspec = btrfs_inode_atime(inode_item);
3442 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3443 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3444 
3445 	tspec = btrfs_inode_mtime(inode_item);
3446 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3447 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3448 
3449 	tspec = btrfs_inode_ctime(inode_item);
3450 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3451 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3452 
3453 	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3454 	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3455 	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3456 
3457 	/*
3458 	 * If we were modified in the current generation and evicted from memory
3459 	 * and then re-read we need to do a full sync since we don't have any
3460 	 * idea about which extents were modified before we were evicted from
3461 	 * cache.
3462 	 */
3463 	if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3464 		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3465 			&BTRFS_I(inode)->runtime_flags);
3466 
3467 	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3468 	inode->i_generation = BTRFS_I(inode)->generation;
3469 	inode->i_rdev = 0;
3470 	rdev = btrfs_inode_rdev(leaf, inode_item);
3471 
3472 	BTRFS_I(inode)->index_cnt = (u64)-1;
3473 	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3474 
3475 cache_index:
3476 	path->slots[0]++;
3477 	if (inode->i_nlink != 1 ||
3478 	    path->slots[0] >= btrfs_header_nritems(leaf))
3479 		goto cache_acl;
3480 
3481 	btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3482 	if (location.objectid != btrfs_ino(inode))
3483 		goto cache_acl;
3484 
3485 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3486 	if (location.type == BTRFS_INODE_REF_KEY) {
3487 		struct btrfs_inode_ref *ref;
3488 
3489 		ref = (struct btrfs_inode_ref *)ptr;
3490 		BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3491 	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3492 		struct btrfs_inode_extref *extref;
3493 
3494 		extref = (struct btrfs_inode_extref *)ptr;
3495 		BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3496 								     extref);
3497 	}
3498 cache_acl:
3499 	/*
3500 	 * try to precache a NULL acl entry for files that don't have
3501 	 * any xattrs or acls
3502 	 */
3503 	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3504 					   btrfs_ino(inode), &first_xattr_slot);
3505 	if (first_xattr_slot != -1) {
3506 		path->slots[0] = first_xattr_slot;
3507 		ret = btrfs_load_inode_props(inode, path);
3508 		if (ret)
3509 			btrfs_err(root->fs_info,
3510 				  "error loading props for ino %llu (root %llu): %d",
3511 				  btrfs_ino(inode),
3512 				  root->root_key.objectid, ret);
3513 	}
3514 	btrfs_free_path(path);
3515 
3516 	if (!maybe_acls)
3517 		cache_no_acl(inode);
3518 
3519 	switch (inode->i_mode & S_IFMT) {
3520 	case S_IFREG:
3521 		inode->i_mapping->a_ops = &btrfs_aops;
3522 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3523 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3524 		inode->i_fop = &btrfs_file_operations;
3525 		inode->i_op = &btrfs_file_inode_operations;
3526 		break;
3527 	case S_IFDIR:
3528 		inode->i_fop = &btrfs_dir_file_operations;
3529 		if (root == root->fs_info->tree_root)
3530 			inode->i_op = &btrfs_dir_ro_inode_operations;
3531 		else
3532 			inode->i_op = &btrfs_dir_inode_operations;
3533 		break;
3534 	case S_IFLNK:
3535 		inode->i_op = &btrfs_symlink_inode_operations;
3536 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
3537 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3538 		break;
3539 	default:
3540 		inode->i_op = &btrfs_special_inode_operations;
3541 		init_special_inode(inode, inode->i_mode, rdev);
3542 		break;
3543 	}
3544 
3545 	btrfs_update_iflags(inode);
3546 	return;
3547 
3548 make_bad:
3549 	btrfs_free_path(path);
3550 	make_bad_inode(inode);
3551 }
3552 
3553 /*
3554  * given a leaf and an inode, copy the inode fields into the leaf
3555  */
3556 static void fill_inode_item(struct btrfs_trans_handle *trans,
3557 			    struct extent_buffer *leaf,
3558 			    struct btrfs_inode_item *item,
3559 			    struct inode *inode)
3560 {
3561 	struct btrfs_map_token token;
3562 
3563 	btrfs_init_map_token(&token);
3564 
3565 	btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3566 	btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3567 	btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3568 				   &token);
3569 	btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3570 	btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3571 
3572 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3573 				     inode->i_atime.tv_sec, &token);
3574 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3575 				      inode->i_atime.tv_nsec, &token);
3576 
3577 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3578 				     inode->i_mtime.tv_sec, &token);
3579 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3580 				      inode->i_mtime.tv_nsec, &token);
3581 
3582 	btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3583 				     inode->i_ctime.tv_sec, &token);
3584 	btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3585 				      inode->i_ctime.tv_nsec, &token);
3586 
3587 	btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3588 				     &token);
3589 	btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3590 					 &token);
3591 	btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3592 	btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3593 	btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3594 	btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3595 	btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3596 }
3597 
3598 /*
3599  * copy everything in the in-memory inode into the btree.
3600  */
3601 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3602 				struct btrfs_root *root, struct inode *inode)
3603 {
3604 	struct btrfs_inode_item *inode_item;
3605 	struct btrfs_path *path;
3606 	struct extent_buffer *leaf;
3607 	int ret;
3608 
3609 	path = btrfs_alloc_path();
3610 	if (!path)
3611 		return -ENOMEM;
3612 
3613 	path->leave_spinning = 1;
3614 	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3615 				 1);
3616 	if (ret) {
3617 		if (ret > 0)
3618 			ret = -ENOENT;
3619 		goto failed;
3620 	}
3621 
3622 	leaf = path->nodes[0];
3623 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3624 				    struct btrfs_inode_item);
3625 
3626 	fill_inode_item(trans, leaf, inode_item, inode);
3627 	btrfs_mark_buffer_dirty(leaf);
3628 	btrfs_set_inode_last_trans(trans, inode);
3629 	ret = 0;
3630 failed:
3631 	btrfs_free_path(path);
3632 	return ret;
3633 }
3634 
3635 /*
3636  * copy everything in the in-memory inode into the btree.
3637  */
3638 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3639 				struct btrfs_root *root, struct inode *inode)
3640 {
3641 	int ret;
3642 
3643 	/*
3644 	 * If the inode is a free space inode, we can deadlock during commit
3645 	 * if we put it into the delayed code.
3646 	 *
3647 	 * The data relocation inode should also be directly updated
3648 	 * without delay
3649 	 */
3650 	if (!btrfs_is_free_space_inode(inode)
3651 	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3652 		btrfs_update_root_times(trans, root);
3653 
3654 		ret = btrfs_delayed_update_inode(trans, root, inode);
3655 		if (!ret)
3656 			btrfs_set_inode_last_trans(trans, inode);
3657 		return ret;
3658 	}
3659 
3660 	return btrfs_update_inode_item(trans, root, inode);
3661 }
3662 
3663 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3664 					 struct btrfs_root *root,
3665 					 struct inode *inode)
3666 {
3667 	int ret;
3668 
3669 	ret = btrfs_update_inode(trans, root, inode);
3670 	if (ret == -ENOSPC)
3671 		return btrfs_update_inode_item(trans, root, inode);
3672 	return ret;
3673 }
3674 
3675 /*
3676  * unlink helper that gets used here in inode.c and in the tree logging
3677  * recovery code.  It remove a link in a directory with a given name, and
3678  * also drops the back refs in the inode to the directory
3679  */
3680 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3681 				struct btrfs_root *root,
3682 				struct inode *dir, struct inode *inode,
3683 				const char *name, int name_len)
3684 {
3685 	struct btrfs_path *path;
3686 	int ret = 0;
3687 	struct extent_buffer *leaf;
3688 	struct btrfs_dir_item *di;
3689 	struct btrfs_key key;
3690 	u64 index;
3691 	u64 ino = btrfs_ino(inode);
3692 	u64 dir_ino = btrfs_ino(dir);
3693 
3694 	path = btrfs_alloc_path();
3695 	if (!path) {
3696 		ret = -ENOMEM;
3697 		goto out;
3698 	}
3699 
3700 	path->leave_spinning = 1;
3701 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3702 				    name, name_len, -1);
3703 	if (IS_ERR(di)) {
3704 		ret = PTR_ERR(di);
3705 		goto err;
3706 	}
3707 	if (!di) {
3708 		ret = -ENOENT;
3709 		goto err;
3710 	}
3711 	leaf = path->nodes[0];
3712 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3713 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3714 	if (ret)
3715 		goto err;
3716 	btrfs_release_path(path);
3717 
3718 	/*
3719 	 * If we don't have dir index, we have to get it by looking up
3720 	 * the inode ref, since we get the inode ref, remove it directly,
3721 	 * it is unnecessary to do delayed deletion.
3722 	 *
3723 	 * But if we have dir index, needn't search inode ref to get it.
3724 	 * Since the inode ref is close to the inode item, it is better
3725 	 * that we delay to delete it, and just do this deletion when
3726 	 * we update the inode item.
3727 	 */
3728 	if (BTRFS_I(inode)->dir_index) {
3729 		ret = btrfs_delayed_delete_inode_ref(inode);
3730 		if (!ret) {
3731 			index = BTRFS_I(inode)->dir_index;
3732 			goto skip_backref;
3733 		}
3734 	}
3735 
3736 	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3737 				  dir_ino, &index);
3738 	if (ret) {
3739 		btrfs_info(root->fs_info,
3740 			"failed to delete reference to %.*s, inode %llu parent %llu",
3741 			name_len, name, ino, dir_ino);
3742 		btrfs_abort_transaction(trans, root, ret);
3743 		goto err;
3744 	}
3745 skip_backref:
3746 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3747 	if (ret) {
3748 		btrfs_abort_transaction(trans, root, ret);
3749 		goto err;
3750 	}
3751 
3752 	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3753 					 inode, dir_ino);
3754 	if (ret != 0 && ret != -ENOENT) {
3755 		btrfs_abort_transaction(trans, root, ret);
3756 		goto err;
3757 	}
3758 
3759 	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3760 					   dir, index);
3761 	if (ret == -ENOENT)
3762 		ret = 0;
3763 	else if (ret)
3764 		btrfs_abort_transaction(trans, root, ret);
3765 err:
3766 	btrfs_free_path(path);
3767 	if (ret)
3768 		goto out;
3769 
3770 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3771 	inode_inc_iversion(inode);
3772 	inode_inc_iversion(dir);
3773 	inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3774 	ret = btrfs_update_inode(trans, root, dir);
3775 out:
3776 	return ret;
3777 }
3778 
3779 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3780 		       struct btrfs_root *root,
3781 		       struct inode *dir, struct inode *inode,
3782 		       const char *name, int name_len)
3783 {
3784 	int ret;
3785 	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3786 	if (!ret) {
3787 		drop_nlink(inode);
3788 		ret = btrfs_update_inode(trans, root, inode);
3789 	}
3790 	return ret;
3791 }
3792 
3793 /*
3794  * helper to start transaction for unlink and rmdir.
3795  *
3796  * unlink and rmdir are special in btrfs, they do not always free space, so
3797  * if we cannot make our reservations the normal way try and see if there is
3798  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3799  * allow the unlink to occur.
3800  */
3801 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3802 {
3803 	struct btrfs_trans_handle *trans;
3804 	struct btrfs_root *root = BTRFS_I(dir)->root;
3805 	int ret;
3806 
3807 	/*
3808 	 * 1 for the possible orphan item
3809 	 * 1 for the dir item
3810 	 * 1 for the dir index
3811 	 * 1 for the inode ref
3812 	 * 1 for the inode
3813 	 */
3814 	trans = btrfs_start_transaction(root, 5);
3815 	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3816 		return trans;
3817 
3818 	if (PTR_ERR(trans) == -ENOSPC) {
3819 		u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3820 
3821 		trans = btrfs_start_transaction(root, 0);
3822 		if (IS_ERR(trans))
3823 			return trans;
3824 		ret = btrfs_cond_migrate_bytes(root->fs_info,
3825 					       &root->fs_info->trans_block_rsv,
3826 					       num_bytes, 5);
3827 		if (ret) {
3828 			btrfs_end_transaction(trans, root);
3829 			return ERR_PTR(ret);
3830 		}
3831 		trans->block_rsv = &root->fs_info->trans_block_rsv;
3832 		trans->bytes_reserved = num_bytes;
3833 	}
3834 	return trans;
3835 }
3836 
3837 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3838 {
3839 	struct btrfs_root *root = BTRFS_I(dir)->root;
3840 	struct btrfs_trans_handle *trans;
3841 	struct inode *inode = dentry->d_inode;
3842 	int ret;
3843 
3844 	trans = __unlink_start_trans(dir);
3845 	if (IS_ERR(trans))
3846 		return PTR_ERR(trans);
3847 
3848 	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3849 
3850 	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3851 				 dentry->d_name.name, dentry->d_name.len);
3852 	if (ret)
3853 		goto out;
3854 
3855 	if (inode->i_nlink == 0) {
3856 		ret = btrfs_orphan_add(trans, inode);
3857 		if (ret)
3858 			goto out;
3859 	}
3860 
3861 out:
3862 	btrfs_end_transaction(trans, root);
3863 	btrfs_btree_balance_dirty(root);
3864 	return ret;
3865 }
3866 
3867 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3868 			struct btrfs_root *root,
3869 			struct inode *dir, u64 objectid,
3870 			const char *name, int name_len)
3871 {
3872 	struct btrfs_path *path;
3873 	struct extent_buffer *leaf;
3874 	struct btrfs_dir_item *di;
3875 	struct btrfs_key key;
3876 	u64 index;
3877 	int ret;
3878 	u64 dir_ino = btrfs_ino(dir);
3879 
3880 	path = btrfs_alloc_path();
3881 	if (!path)
3882 		return -ENOMEM;
3883 
3884 	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3885 				   name, name_len, -1);
3886 	if (IS_ERR_OR_NULL(di)) {
3887 		if (!di)
3888 			ret = -ENOENT;
3889 		else
3890 			ret = PTR_ERR(di);
3891 		goto out;
3892 	}
3893 
3894 	leaf = path->nodes[0];
3895 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
3896 	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3897 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3898 	if (ret) {
3899 		btrfs_abort_transaction(trans, root, ret);
3900 		goto out;
3901 	}
3902 	btrfs_release_path(path);
3903 
3904 	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3905 				 objectid, root->root_key.objectid,
3906 				 dir_ino, &index, name, name_len);
3907 	if (ret < 0) {
3908 		if (ret != -ENOENT) {
3909 			btrfs_abort_transaction(trans, root, ret);
3910 			goto out;
3911 		}
3912 		di = btrfs_search_dir_index_item(root, path, dir_ino,
3913 						 name, name_len);
3914 		if (IS_ERR_OR_NULL(di)) {
3915 			if (!di)
3916 				ret = -ENOENT;
3917 			else
3918 				ret = PTR_ERR(di);
3919 			btrfs_abort_transaction(trans, root, ret);
3920 			goto out;
3921 		}
3922 
3923 		leaf = path->nodes[0];
3924 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3925 		btrfs_release_path(path);
3926 		index = key.offset;
3927 	}
3928 	btrfs_release_path(path);
3929 
3930 	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3931 	if (ret) {
3932 		btrfs_abort_transaction(trans, root, ret);
3933 		goto out;
3934 	}
3935 
3936 	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3937 	inode_inc_iversion(dir);
3938 	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3939 	ret = btrfs_update_inode_fallback(trans, root, dir);
3940 	if (ret)
3941 		btrfs_abort_transaction(trans, root, ret);
3942 out:
3943 	btrfs_free_path(path);
3944 	return ret;
3945 }
3946 
3947 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3948 {
3949 	struct inode *inode = dentry->d_inode;
3950 	int err = 0;
3951 	struct btrfs_root *root = BTRFS_I(dir)->root;
3952 	struct btrfs_trans_handle *trans;
3953 
3954 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3955 		return -ENOTEMPTY;
3956 	if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3957 		return -EPERM;
3958 
3959 	trans = __unlink_start_trans(dir);
3960 	if (IS_ERR(trans))
3961 		return PTR_ERR(trans);
3962 
3963 	if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3964 		err = btrfs_unlink_subvol(trans, root, dir,
3965 					  BTRFS_I(inode)->location.objectid,
3966 					  dentry->d_name.name,
3967 					  dentry->d_name.len);
3968 		goto out;
3969 	}
3970 
3971 	err = btrfs_orphan_add(trans, inode);
3972 	if (err)
3973 		goto out;
3974 
3975 	/* now the directory is empty */
3976 	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3977 				 dentry->d_name.name, dentry->d_name.len);
3978 	if (!err)
3979 		btrfs_i_size_write(inode, 0);
3980 out:
3981 	btrfs_end_transaction(trans, root);
3982 	btrfs_btree_balance_dirty(root);
3983 
3984 	return err;
3985 }
3986 
3987 /*
3988  * this can truncate away extent items, csum items and directory items.
3989  * It starts at a high offset and removes keys until it can't find
3990  * any higher than new_size
3991  *
3992  * csum items that cross the new i_size are truncated to the new size
3993  * as well.
3994  *
3995  * min_type is the minimum key type to truncate down to.  If set to 0, this
3996  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3997  */
3998 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3999 			       struct btrfs_root *root,
4000 			       struct inode *inode,
4001 			       u64 new_size, u32 min_type)
4002 {
4003 	struct btrfs_path *path;
4004 	struct extent_buffer *leaf;
4005 	struct btrfs_file_extent_item *fi;
4006 	struct btrfs_key key;
4007 	struct btrfs_key found_key;
4008 	u64 extent_start = 0;
4009 	u64 extent_num_bytes = 0;
4010 	u64 extent_offset = 0;
4011 	u64 item_end = 0;
4012 	u64 last_size = (u64)-1;
4013 	u32 found_type = (u8)-1;
4014 	int found_extent;
4015 	int del_item;
4016 	int pending_del_nr = 0;
4017 	int pending_del_slot = 0;
4018 	int extent_type = -1;
4019 	int ret;
4020 	int err = 0;
4021 	u64 ino = btrfs_ino(inode);
4022 
4023 	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4024 
4025 	path = btrfs_alloc_path();
4026 	if (!path)
4027 		return -ENOMEM;
4028 	path->reada = -1;
4029 
4030 	/*
4031 	 * We want to drop from the next block forward in case this new size is
4032 	 * not block aligned since we will be keeping the last block of the
4033 	 * extent just the way it is.
4034 	 */
4035 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4036 	    root == root->fs_info->tree_root)
4037 		btrfs_drop_extent_cache(inode, ALIGN(new_size,
4038 					root->sectorsize), (u64)-1, 0);
4039 
4040 	/*
4041 	 * This function is also used to drop the items in the log tree before
4042 	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4043 	 * it is used to drop the loged items. So we shouldn't kill the delayed
4044 	 * items.
4045 	 */
4046 	if (min_type == 0 && root == BTRFS_I(inode)->root)
4047 		btrfs_kill_delayed_inode_items(inode);
4048 
4049 	key.objectid = ino;
4050 	key.offset = (u64)-1;
4051 	key.type = (u8)-1;
4052 
4053 search_again:
4054 	path->leave_spinning = 1;
4055 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4056 	if (ret < 0) {
4057 		err = ret;
4058 		goto out;
4059 	}
4060 
4061 	if (ret > 0) {
4062 		/* there are no items in the tree for us to truncate, we're
4063 		 * done
4064 		 */
4065 		if (path->slots[0] == 0)
4066 			goto out;
4067 		path->slots[0]--;
4068 	}
4069 
4070 	while (1) {
4071 		fi = NULL;
4072 		leaf = path->nodes[0];
4073 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4074 		found_type = btrfs_key_type(&found_key);
4075 
4076 		if (found_key.objectid != ino)
4077 			break;
4078 
4079 		if (found_type < min_type)
4080 			break;
4081 
4082 		item_end = found_key.offset;
4083 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
4084 			fi = btrfs_item_ptr(leaf, path->slots[0],
4085 					    struct btrfs_file_extent_item);
4086 			extent_type = btrfs_file_extent_type(leaf, fi);
4087 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4088 				item_end +=
4089 				    btrfs_file_extent_num_bytes(leaf, fi);
4090 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4091 				item_end += btrfs_file_extent_inline_len(leaf,
4092 							 path->slots[0], fi);
4093 			}
4094 			item_end--;
4095 		}
4096 		if (found_type > min_type) {
4097 			del_item = 1;
4098 		} else {
4099 			if (item_end < new_size)
4100 				break;
4101 			if (found_key.offset >= new_size)
4102 				del_item = 1;
4103 			else
4104 				del_item = 0;
4105 		}
4106 		found_extent = 0;
4107 		/* FIXME, shrink the extent if the ref count is only 1 */
4108 		if (found_type != BTRFS_EXTENT_DATA_KEY)
4109 			goto delete;
4110 
4111 		if (del_item)
4112 			last_size = found_key.offset;
4113 		else
4114 			last_size = new_size;
4115 
4116 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4117 			u64 num_dec;
4118 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4119 			if (!del_item) {
4120 				u64 orig_num_bytes =
4121 					btrfs_file_extent_num_bytes(leaf, fi);
4122 				extent_num_bytes = ALIGN(new_size -
4123 						found_key.offset,
4124 						root->sectorsize);
4125 				btrfs_set_file_extent_num_bytes(leaf, fi,
4126 							 extent_num_bytes);
4127 				num_dec = (orig_num_bytes -
4128 					   extent_num_bytes);
4129 				if (test_bit(BTRFS_ROOT_REF_COWS,
4130 					     &root->state) &&
4131 				    extent_start != 0)
4132 					inode_sub_bytes(inode, num_dec);
4133 				btrfs_mark_buffer_dirty(leaf);
4134 			} else {
4135 				extent_num_bytes =
4136 					btrfs_file_extent_disk_num_bytes(leaf,
4137 									 fi);
4138 				extent_offset = found_key.offset -
4139 					btrfs_file_extent_offset(leaf, fi);
4140 
4141 				/* FIXME blocksize != 4096 */
4142 				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4143 				if (extent_start != 0) {
4144 					found_extent = 1;
4145 					if (test_bit(BTRFS_ROOT_REF_COWS,
4146 						     &root->state))
4147 						inode_sub_bytes(inode, num_dec);
4148 				}
4149 			}
4150 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4151 			/*
4152 			 * we can't truncate inline items that have had
4153 			 * special encodings
4154 			 */
4155 			if (!del_item &&
4156 			    btrfs_file_extent_compression(leaf, fi) == 0 &&
4157 			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
4158 			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4159 				u32 size = new_size - found_key.offset;
4160 
4161 				if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4162 					inode_sub_bytes(inode, item_end + 1 -
4163 							new_size);
4164 
4165 				/*
4166 				 * update the ram bytes to properly reflect
4167 				 * the new size of our item
4168 				 */
4169 				btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4170 				size =
4171 				    btrfs_file_extent_calc_inline_size(size);
4172 				btrfs_truncate_item(root, path, size, 1);
4173 			} else if (test_bit(BTRFS_ROOT_REF_COWS,
4174 					    &root->state)) {
4175 				inode_sub_bytes(inode, item_end + 1 -
4176 						found_key.offset);
4177 			}
4178 		}
4179 delete:
4180 		if (del_item) {
4181 			if (!pending_del_nr) {
4182 				/* no pending yet, add ourselves */
4183 				pending_del_slot = path->slots[0];
4184 				pending_del_nr = 1;
4185 			} else if (pending_del_nr &&
4186 				   path->slots[0] + 1 == pending_del_slot) {
4187 				/* hop on the pending chunk */
4188 				pending_del_nr++;
4189 				pending_del_slot = path->slots[0];
4190 			} else {
4191 				BUG();
4192 			}
4193 		} else {
4194 			break;
4195 		}
4196 		if (found_extent &&
4197 		    (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4198 		     root == root->fs_info->tree_root)) {
4199 			btrfs_set_path_blocking(path);
4200 			ret = btrfs_free_extent(trans, root, extent_start,
4201 						extent_num_bytes, 0,
4202 						btrfs_header_owner(leaf),
4203 						ino, extent_offset, 0);
4204 			BUG_ON(ret);
4205 		}
4206 
4207 		if (found_type == BTRFS_INODE_ITEM_KEY)
4208 			break;
4209 
4210 		if (path->slots[0] == 0 ||
4211 		    path->slots[0] != pending_del_slot) {
4212 			if (pending_del_nr) {
4213 				ret = btrfs_del_items(trans, root, path,
4214 						pending_del_slot,
4215 						pending_del_nr);
4216 				if (ret) {
4217 					btrfs_abort_transaction(trans,
4218 								root, ret);
4219 					goto error;
4220 				}
4221 				pending_del_nr = 0;
4222 			}
4223 			btrfs_release_path(path);
4224 			goto search_again;
4225 		} else {
4226 			path->slots[0]--;
4227 		}
4228 	}
4229 out:
4230 	if (pending_del_nr) {
4231 		ret = btrfs_del_items(trans, root, path, pending_del_slot,
4232 				      pending_del_nr);
4233 		if (ret)
4234 			btrfs_abort_transaction(trans, root, ret);
4235 	}
4236 error:
4237 	if (last_size != (u64)-1)
4238 		btrfs_ordered_update_i_size(inode, last_size, NULL);
4239 	btrfs_free_path(path);
4240 	return err;
4241 }
4242 
4243 /*
4244  * btrfs_truncate_page - read, zero a chunk and write a page
4245  * @inode - inode that we're zeroing
4246  * @from - the offset to start zeroing
4247  * @len - the length to zero, 0 to zero the entire range respective to the
4248  *	offset
4249  * @front - zero up to the offset instead of from the offset on
4250  *
4251  * This will find the page for the "from" offset and cow the page and zero the
4252  * part we want to zero.  This is used with truncate and hole punching.
4253  */
4254 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4255 			int front)
4256 {
4257 	struct address_space *mapping = inode->i_mapping;
4258 	struct btrfs_root *root = BTRFS_I(inode)->root;
4259 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4260 	struct btrfs_ordered_extent *ordered;
4261 	struct extent_state *cached_state = NULL;
4262 	char *kaddr;
4263 	u32 blocksize = root->sectorsize;
4264 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
4265 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
4266 	struct page *page;
4267 	gfp_t mask = btrfs_alloc_write_mask(mapping);
4268 	int ret = 0;
4269 	u64 page_start;
4270 	u64 page_end;
4271 
4272 	if ((offset & (blocksize - 1)) == 0 &&
4273 	    (!len || ((len & (blocksize - 1)) == 0)))
4274 		goto out;
4275 	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4276 	if (ret)
4277 		goto out;
4278 
4279 again:
4280 	page = find_or_create_page(mapping, index, mask);
4281 	if (!page) {
4282 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4283 		ret = -ENOMEM;
4284 		goto out;
4285 	}
4286 
4287 	page_start = page_offset(page);
4288 	page_end = page_start + PAGE_CACHE_SIZE - 1;
4289 
4290 	if (!PageUptodate(page)) {
4291 		ret = btrfs_readpage(NULL, page);
4292 		lock_page(page);
4293 		if (page->mapping != mapping) {
4294 			unlock_page(page);
4295 			page_cache_release(page);
4296 			goto again;
4297 		}
4298 		if (!PageUptodate(page)) {
4299 			ret = -EIO;
4300 			goto out_unlock;
4301 		}
4302 	}
4303 	wait_on_page_writeback(page);
4304 
4305 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4306 	set_page_extent_mapped(page);
4307 
4308 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
4309 	if (ordered) {
4310 		unlock_extent_cached(io_tree, page_start, page_end,
4311 				     &cached_state, GFP_NOFS);
4312 		unlock_page(page);
4313 		page_cache_release(page);
4314 		btrfs_start_ordered_extent(inode, ordered, 1);
4315 		btrfs_put_ordered_extent(ordered);
4316 		goto again;
4317 	}
4318 
4319 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4320 			  EXTENT_DIRTY | EXTENT_DELALLOC |
4321 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4322 			  0, 0, &cached_state, GFP_NOFS);
4323 
4324 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4325 					&cached_state);
4326 	if (ret) {
4327 		unlock_extent_cached(io_tree, page_start, page_end,
4328 				     &cached_state, GFP_NOFS);
4329 		goto out_unlock;
4330 	}
4331 
4332 	if (offset != PAGE_CACHE_SIZE) {
4333 		if (!len)
4334 			len = PAGE_CACHE_SIZE - offset;
4335 		kaddr = kmap(page);
4336 		if (front)
4337 			memset(kaddr, 0, offset);
4338 		else
4339 			memset(kaddr + offset, 0, len);
4340 		flush_dcache_page(page);
4341 		kunmap(page);
4342 	}
4343 	ClearPageChecked(page);
4344 	set_page_dirty(page);
4345 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4346 			     GFP_NOFS);
4347 
4348 out_unlock:
4349 	if (ret)
4350 		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4351 	unlock_page(page);
4352 	page_cache_release(page);
4353 out:
4354 	return ret;
4355 }
4356 
4357 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4358 			     u64 offset, u64 len)
4359 {
4360 	struct btrfs_trans_handle *trans;
4361 	int ret;
4362 
4363 	/*
4364 	 * Still need to make sure the inode looks like it's been updated so
4365 	 * that any holes get logged if we fsync.
4366 	 */
4367 	if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4368 		BTRFS_I(inode)->last_trans = root->fs_info->generation;
4369 		BTRFS_I(inode)->last_sub_trans = root->log_transid;
4370 		BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4371 		return 0;
4372 	}
4373 
4374 	/*
4375 	 * 1 - for the one we're dropping
4376 	 * 1 - for the one we're adding
4377 	 * 1 - for updating the inode.
4378 	 */
4379 	trans = btrfs_start_transaction(root, 3);
4380 	if (IS_ERR(trans))
4381 		return PTR_ERR(trans);
4382 
4383 	ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4384 	if (ret) {
4385 		btrfs_abort_transaction(trans, root, ret);
4386 		btrfs_end_transaction(trans, root);
4387 		return ret;
4388 	}
4389 
4390 	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4391 				       0, 0, len, 0, len, 0, 0, 0);
4392 	if (ret)
4393 		btrfs_abort_transaction(trans, root, ret);
4394 	else
4395 		btrfs_update_inode(trans, root, inode);
4396 	btrfs_end_transaction(trans, root);
4397 	return ret;
4398 }
4399 
4400 /*
4401  * This function puts in dummy file extents for the area we're creating a hole
4402  * for.  So if we are truncating this file to a larger size we need to insert
4403  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4404  * the range between oldsize and size
4405  */
4406 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4407 {
4408 	struct btrfs_root *root = BTRFS_I(inode)->root;
4409 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4410 	struct extent_map *em = NULL;
4411 	struct extent_state *cached_state = NULL;
4412 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4413 	u64 hole_start = ALIGN(oldsize, root->sectorsize);
4414 	u64 block_end = ALIGN(size, root->sectorsize);
4415 	u64 last_byte;
4416 	u64 cur_offset;
4417 	u64 hole_size;
4418 	int err = 0;
4419 
4420 	/*
4421 	 * If our size started in the middle of a page we need to zero out the
4422 	 * rest of the page before we expand the i_size, otherwise we could
4423 	 * expose stale data.
4424 	 */
4425 	err = btrfs_truncate_page(inode, oldsize, 0, 0);
4426 	if (err)
4427 		return err;
4428 
4429 	if (size <= hole_start)
4430 		return 0;
4431 
4432 	while (1) {
4433 		struct btrfs_ordered_extent *ordered;
4434 
4435 		lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4436 				 &cached_state);
4437 		ordered = btrfs_lookup_ordered_range(inode, hole_start,
4438 						     block_end - hole_start);
4439 		if (!ordered)
4440 			break;
4441 		unlock_extent_cached(io_tree, hole_start, block_end - 1,
4442 				     &cached_state, GFP_NOFS);
4443 		btrfs_start_ordered_extent(inode, ordered, 1);
4444 		btrfs_put_ordered_extent(ordered);
4445 	}
4446 
4447 	cur_offset = hole_start;
4448 	while (1) {
4449 		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4450 				block_end - cur_offset, 0);
4451 		if (IS_ERR(em)) {
4452 			err = PTR_ERR(em);
4453 			em = NULL;
4454 			break;
4455 		}
4456 		last_byte = min(extent_map_end(em), block_end);
4457 		last_byte = ALIGN(last_byte , root->sectorsize);
4458 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4459 			struct extent_map *hole_em;
4460 			hole_size = last_byte - cur_offset;
4461 
4462 			err = maybe_insert_hole(root, inode, cur_offset,
4463 						hole_size);
4464 			if (err)
4465 				break;
4466 			btrfs_drop_extent_cache(inode, cur_offset,
4467 						cur_offset + hole_size - 1, 0);
4468 			hole_em = alloc_extent_map();
4469 			if (!hole_em) {
4470 				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4471 					&BTRFS_I(inode)->runtime_flags);
4472 				goto next;
4473 			}
4474 			hole_em->start = cur_offset;
4475 			hole_em->len = hole_size;
4476 			hole_em->orig_start = cur_offset;
4477 
4478 			hole_em->block_start = EXTENT_MAP_HOLE;
4479 			hole_em->block_len = 0;
4480 			hole_em->orig_block_len = 0;
4481 			hole_em->ram_bytes = hole_size;
4482 			hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4483 			hole_em->compress_type = BTRFS_COMPRESS_NONE;
4484 			hole_em->generation = root->fs_info->generation;
4485 
4486 			while (1) {
4487 				write_lock(&em_tree->lock);
4488 				err = add_extent_mapping(em_tree, hole_em, 1);
4489 				write_unlock(&em_tree->lock);
4490 				if (err != -EEXIST)
4491 					break;
4492 				btrfs_drop_extent_cache(inode, cur_offset,
4493 							cur_offset +
4494 							hole_size - 1, 0);
4495 			}
4496 			free_extent_map(hole_em);
4497 		}
4498 next:
4499 		free_extent_map(em);
4500 		em = NULL;
4501 		cur_offset = last_byte;
4502 		if (cur_offset >= block_end)
4503 			break;
4504 	}
4505 	free_extent_map(em);
4506 	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4507 			     GFP_NOFS);
4508 	return err;
4509 }
4510 
4511 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4512 {
4513 	struct btrfs_root *root = BTRFS_I(inode)->root;
4514 	struct btrfs_trans_handle *trans;
4515 	loff_t oldsize = i_size_read(inode);
4516 	loff_t newsize = attr->ia_size;
4517 	int mask = attr->ia_valid;
4518 	int ret;
4519 
4520 	/*
4521 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4522 	 * special case where we need to update the times despite not having
4523 	 * these flags set.  For all other operations the VFS set these flags
4524 	 * explicitly if it wants a timestamp update.
4525 	 */
4526 	if (newsize != oldsize) {
4527 		inode_inc_iversion(inode);
4528 		if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4529 			inode->i_ctime = inode->i_mtime =
4530 				current_fs_time(inode->i_sb);
4531 	}
4532 
4533 	if (newsize > oldsize) {
4534 		truncate_pagecache(inode, newsize);
4535 		ret = btrfs_cont_expand(inode, oldsize, newsize);
4536 		if (ret)
4537 			return ret;
4538 
4539 		trans = btrfs_start_transaction(root, 1);
4540 		if (IS_ERR(trans))
4541 			return PTR_ERR(trans);
4542 
4543 		i_size_write(inode, newsize);
4544 		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4545 		ret = btrfs_update_inode(trans, root, inode);
4546 		btrfs_end_transaction(trans, root);
4547 	} else {
4548 
4549 		/*
4550 		 * We're truncating a file that used to have good data down to
4551 		 * zero. Make sure it gets into the ordered flush list so that
4552 		 * any new writes get down to disk quickly.
4553 		 */
4554 		if (newsize == 0)
4555 			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4556 				&BTRFS_I(inode)->runtime_flags);
4557 
4558 		/*
4559 		 * 1 for the orphan item we're going to add
4560 		 * 1 for the orphan item deletion.
4561 		 */
4562 		trans = btrfs_start_transaction(root, 2);
4563 		if (IS_ERR(trans))
4564 			return PTR_ERR(trans);
4565 
4566 		/*
4567 		 * We need to do this in case we fail at _any_ point during the
4568 		 * actual truncate.  Once we do the truncate_setsize we could
4569 		 * invalidate pages which forces any outstanding ordered io to
4570 		 * be instantly completed which will give us extents that need
4571 		 * to be truncated.  If we fail to get an orphan inode down we
4572 		 * could have left over extents that were never meant to live,
4573 		 * so we need to garuntee from this point on that everything
4574 		 * will be consistent.
4575 		 */
4576 		ret = btrfs_orphan_add(trans, inode);
4577 		btrfs_end_transaction(trans, root);
4578 		if (ret)
4579 			return ret;
4580 
4581 		/* we don't support swapfiles, so vmtruncate shouldn't fail */
4582 		truncate_setsize(inode, newsize);
4583 
4584 		/* Disable nonlocked read DIO to avoid the end less truncate */
4585 		btrfs_inode_block_unlocked_dio(inode);
4586 		inode_dio_wait(inode);
4587 		btrfs_inode_resume_unlocked_dio(inode);
4588 
4589 		ret = btrfs_truncate(inode);
4590 		if (ret && inode->i_nlink) {
4591 			int err;
4592 
4593 			/*
4594 			 * failed to truncate, disk_i_size is only adjusted down
4595 			 * as we remove extents, so it should represent the true
4596 			 * size of the inode, so reset the in memory size and
4597 			 * delete our orphan entry.
4598 			 */
4599 			trans = btrfs_join_transaction(root);
4600 			if (IS_ERR(trans)) {
4601 				btrfs_orphan_del(NULL, inode);
4602 				return ret;
4603 			}
4604 			i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4605 			err = btrfs_orphan_del(trans, inode);
4606 			if (err)
4607 				btrfs_abort_transaction(trans, root, err);
4608 			btrfs_end_transaction(trans, root);
4609 		}
4610 	}
4611 
4612 	return ret;
4613 }
4614 
4615 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4616 {
4617 	struct inode *inode = dentry->d_inode;
4618 	struct btrfs_root *root = BTRFS_I(inode)->root;
4619 	int err;
4620 
4621 	if (btrfs_root_readonly(root))
4622 		return -EROFS;
4623 
4624 	err = inode_change_ok(inode, attr);
4625 	if (err)
4626 		return err;
4627 
4628 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4629 		err = btrfs_setsize(inode, attr);
4630 		if (err)
4631 			return err;
4632 	}
4633 
4634 	if (attr->ia_valid) {
4635 		setattr_copy(inode, attr);
4636 		inode_inc_iversion(inode);
4637 		err = btrfs_dirty_inode(inode);
4638 
4639 		if (!err && attr->ia_valid & ATTR_MODE)
4640 			err = posix_acl_chmod(inode, inode->i_mode);
4641 	}
4642 
4643 	return err;
4644 }
4645 
4646 /*
4647  * While truncating the inode pages during eviction, we get the VFS calling
4648  * btrfs_invalidatepage() against each page of the inode. This is slow because
4649  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4650  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4651  * extent_state structures over and over, wasting lots of time.
4652  *
4653  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4654  * those expensive operations on a per page basis and do only the ordered io
4655  * finishing, while we release here the extent_map and extent_state structures,
4656  * without the excessive merging and splitting.
4657  */
4658 static void evict_inode_truncate_pages(struct inode *inode)
4659 {
4660 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4661 	struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4662 	struct rb_node *node;
4663 
4664 	ASSERT(inode->i_state & I_FREEING);
4665 	truncate_inode_pages_final(&inode->i_data);
4666 
4667 	write_lock(&map_tree->lock);
4668 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
4669 		struct extent_map *em;
4670 
4671 		node = rb_first(&map_tree->map);
4672 		em = rb_entry(node, struct extent_map, rb_node);
4673 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4674 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4675 		remove_extent_mapping(map_tree, em);
4676 		free_extent_map(em);
4677 	}
4678 	write_unlock(&map_tree->lock);
4679 
4680 	spin_lock(&io_tree->lock);
4681 	while (!RB_EMPTY_ROOT(&io_tree->state)) {
4682 		struct extent_state *state;
4683 		struct extent_state *cached_state = NULL;
4684 
4685 		node = rb_first(&io_tree->state);
4686 		state = rb_entry(node, struct extent_state, rb_node);
4687 		atomic_inc(&state->refs);
4688 		spin_unlock(&io_tree->lock);
4689 
4690 		lock_extent_bits(io_tree, state->start, state->end,
4691 				 0, &cached_state);
4692 		clear_extent_bit(io_tree, state->start, state->end,
4693 				 EXTENT_LOCKED | EXTENT_DIRTY |
4694 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4695 				 EXTENT_DEFRAG, 1, 1,
4696 				 &cached_state, GFP_NOFS);
4697 		free_extent_state(state);
4698 
4699 		spin_lock(&io_tree->lock);
4700 	}
4701 	spin_unlock(&io_tree->lock);
4702 }
4703 
4704 void btrfs_evict_inode(struct inode *inode)
4705 {
4706 	struct btrfs_trans_handle *trans;
4707 	struct btrfs_root *root = BTRFS_I(inode)->root;
4708 	struct btrfs_block_rsv *rsv, *global_rsv;
4709 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4710 	int ret;
4711 
4712 	trace_btrfs_inode_evict(inode);
4713 
4714 	evict_inode_truncate_pages(inode);
4715 
4716 	if (inode->i_nlink &&
4717 	    ((btrfs_root_refs(&root->root_item) != 0 &&
4718 	      root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4719 	     btrfs_is_free_space_inode(inode)))
4720 		goto no_delete;
4721 
4722 	if (is_bad_inode(inode)) {
4723 		btrfs_orphan_del(NULL, inode);
4724 		goto no_delete;
4725 	}
4726 	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4727 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
4728 
4729 	if (root->fs_info->log_root_recovering) {
4730 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4731 				 &BTRFS_I(inode)->runtime_flags));
4732 		goto no_delete;
4733 	}
4734 
4735 	if (inode->i_nlink > 0) {
4736 		BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4737 		       root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4738 		goto no_delete;
4739 	}
4740 
4741 	ret = btrfs_commit_inode_delayed_inode(inode);
4742 	if (ret) {
4743 		btrfs_orphan_del(NULL, inode);
4744 		goto no_delete;
4745 	}
4746 
4747 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4748 	if (!rsv) {
4749 		btrfs_orphan_del(NULL, inode);
4750 		goto no_delete;
4751 	}
4752 	rsv->size = min_size;
4753 	rsv->failfast = 1;
4754 	global_rsv = &root->fs_info->global_block_rsv;
4755 
4756 	btrfs_i_size_write(inode, 0);
4757 
4758 	/*
4759 	 * This is a bit simpler than btrfs_truncate since we've already
4760 	 * reserved our space for our orphan item in the unlink, so we just
4761 	 * need to reserve some slack space in case we add bytes and update
4762 	 * inode item when doing the truncate.
4763 	 */
4764 	while (1) {
4765 		ret = btrfs_block_rsv_refill(root, rsv, min_size,
4766 					     BTRFS_RESERVE_FLUSH_LIMIT);
4767 
4768 		/*
4769 		 * Try and steal from the global reserve since we will
4770 		 * likely not use this space anyway, we want to try as
4771 		 * hard as possible to get this to work.
4772 		 */
4773 		if (ret)
4774 			ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4775 
4776 		if (ret) {
4777 			btrfs_warn(root->fs_info,
4778 				"Could not get space for a delete, will truncate on mount %d",
4779 				ret);
4780 			btrfs_orphan_del(NULL, inode);
4781 			btrfs_free_block_rsv(root, rsv);
4782 			goto no_delete;
4783 		}
4784 
4785 		trans = btrfs_join_transaction(root);
4786 		if (IS_ERR(trans)) {
4787 			btrfs_orphan_del(NULL, inode);
4788 			btrfs_free_block_rsv(root, rsv);
4789 			goto no_delete;
4790 		}
4791 
4792 		trans->block_rsv = rsv;
4793 
4794 		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4795 		if (ret != -ENOSPC)
4796 			break;
4797 
4798 		trans->block_rsv = &root->fs_info->trans_block_rsv;
4799 		btrfs_end_transaction(trans, root);
4800 		trans = NULL;
4801 		btrfs_btree_balance_dirty(root);
4802 	}
4803 
4804 	btrfs_free_block_rsv(root, rsv);
4805 
4806 	/*
4807 	 * Errors here aren't a big deal, it just means we leave orphan items
4808 	 * in the tree.  They will be cleaned up on the next mount.
4809 	 */
4810 	if (ret == 0) {
4811 		trans->block_rsv = root->orphan_block_rsv;
4812 		btrfs_orphan_del(trans, inode);
4813 	} else {
4814 		btrfs_orphan_del(NULL, inode);
4815 	}
4816 
4817 	trans->block_rsv = &root->fs_info->trans_block_rsv;
4818 	if (!(root == root->fs_info->tree_root ||
4819 	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4820 		btrfs_return_ino(root, btrfs_ino(inode));
4821 
4822 	btrfs_end_transaction(trans, root);
4823 	btrfs_btree_balance_dirty(root);
4824 no_delete:
4825 	btrfs_remove_delayed_node(inode);
4826 	clear_inode(inode);
4827 	return;
4828 }
4829 
4830 /*
4831  * this returns the key found in the dir entry in the location pointer.
4832  * If no dir entries were found, location->objectid is 0.
4833  */
4834 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4835 			       struct btrfs_key *location)
4836 {
4837 	const char *name = dentry->d_name.name;
4838 	int namelen = dentry->d_name.len;
4839 	struct btrfs_dir_item *di;
4840 	struct btrfs_path *path;
4841 	struct btrfs_root *root = BTRFS_I(dir)->root;
4842 	int ret = 0;
4843 
4844 	path = btrfs_alloc_path();
4845 	if (!path)
4846 		return -ENOMEM;
4847 
4848 	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4849 				    namelen, 0);
4850 	if (IS_ERR(di))
4851 		ret = PTR_ERR(di);
4852 
4853 	if (IS_ERR_OR_NULL(di))
4854 		goto out_err;
4855 
4856 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4857 out:
4858 	btrfs_free_path(path);
4859 	return ret;
4860 out_err:
4861 	location->objectid = 0;
4862 	goto out;
4863 }
4864 
4865 /*
4866  * when we hit a tree root in a directory, the btrfs part of the inode
4867  * needs to be changed to reflect the root directory of the tree root.  This
4868  * is kind of like crossing a mount point.
4869  */
4870 static int fixup_tree_root_location(struct btrfs_root *root,
4871 				    struct inode *dir,
4872 				    struct dentry *dentry,
4873 				    struct btrfs_key *location,
4874 				    struct btrfs_root **sub_root)
4875 {
4876 	struct btrfs_path *path;
4877 	struct btrfs_root *new_root;
4878 	struct btrfs_root_ref *ref;
4879 	struct extent_buffer *leaf;
4880 	int ret;
4881 	int err = 0;
4882 
4883 	path = btrfs_alloc_path();
4884 	if (!path) {
4885 		err = -ENOMEM;
4886 		goto out;
4887 	}
4888 
4889 	err = -ENOENT;
4890 	ret = btrfs_find_item(root->fs_info->tree_root, path,
4891 				BTRFS_I(dir)->root->root_key.objectid,
4892 				location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4893 	if (ret) {
4894 		if (ret < 0)
4895 			err = ret;
4896 		goto out;
4897 	}
4898 
4899 	leaf = path->nodes[0];
4900 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4901 	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4902 	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4903 		goto out;
4904 
4905 	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4906 				   (unsigned long)(ref + 1),
4907 				   dentry->d_name.len);
4908 	if (ret)
4909 		goto out;
4910 
4911 	btrfs_release_path(path);
4912 
4913 	new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4914 	if (IS_ERR(new_root)) {
4915 		err = PTR_ERR(new_root);
4916 		goto out;
4917 	}
4918 
4919 	*sub_root = new_root;
4920 	location->objectid = btrfs_root_dirid(&new_root->root_item);
4921 	location->type = BTRFS_INODE_ITEM_KEY;
4922 	location->offset = 0;
4923 	err = 0;
4924 out:
4925 	btrfs_free_path(path);
4926 	return err;
4927 }
4928 
4929 static void inode_tree_add(struct inode *inode)
4930 {
4931 	struct btrfs_root *root = BTRFS_I(inode)->root;
4932 	struct btrfs_inode *entry;
4933 	struct rb_node **p;
4934 	struct rb_node *parent;
4935 	struct rb_node *new = &BTRFS_I(inode)->rb_node;
4936 	u64 ino = btrfs_ino(inode);
4937 
4938 	if (inode_unhashed(inode))
4939 		return;
4940 	parent = NULL;
4941 	spin_lock(&root->inode_lock);
4942 	p = &root->inode_tree.rb_node;
4943 	while (*p) {
4944 		parent = *p;
4945 		entry = rb_entry(parent, struct btrfs_inode, rb_node);
4946 
4947 		if (ino < btrfs_ino(&entry->vfs_inode))
4948 			p = &parent->rb_left;
4949 		else if (ino > btrfs_ino(&entry->vfs_inode))
4950 			p = &parent->rb_right;
4951 		else {
4952 			WARN_ON(!(entry->vfs_inode.i_state &
4953 				  (I_WILL_FREE | I_FREEING)));
4954 			rb_replace_node(parent, new, &root->inode_tree);
4955 			RB_CLEAR_NODE(parent);
4956 			spin_unlock(&root->inode_lock);
4957 			return;
4958 		}
4959 	}
4960 	rb_link_node(new, parent, p);
4961 	rb_insert_color(new, &root->inode_tree);
4962 	spin_unlock(&root->inode_lock);
4963 }
4964 
4965 static void inode_tree_del(struct inode *inode)
4966 {
4967 	struct btrfs_root *root = BTRFS_I(inode)->root;
4968 	int empty = 0;
4969 
4970 	spin_lock(&root->inode_lock);
4971 	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4972 		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4973 		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4974 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4975 	}
4976 	spin_unlock(&root->inode_lock);
4977 
4978 	if (empty && btrfs_root_refs(&root->root_item) == 0) {
4979 		synchronize_srcu(&root->fs_info->subvol_srcu);
4980 		spin_lock(&root->inode_lock);
4981 		empty = RB_EMPTY_ROOT(&root->inode_tree);
4982 		spin_unlock(&root->inode_lock);
4983 		if (empty)
4984 			btrfs_add_dead_root(root);
4985 	}
4986 }
4987 
4988 void btrfs_invalidate_inodes(struct btrfs_root *root)
4989 {
4990 	struct rb_node *node;
4991 	struct rb_node *prev;
4992 	struct btrfs_inode *entry;
4993 	struct inode *inode;
4994 	u64 objectid = 0;
4995 
4996 	if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4997 		WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4998 
4999 	spin_lock(&root->inode_lock);
5000 again:
5001 	node = root->inode_tree.rb_node;
5002 	prev = NULL;
5003 	while (node) {
5004 		prev = node;
5005 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5006 
5007 		if (objectid < btrfs_ino(&entry->vfs_inode))
5008 			node = node->rb_left;
5009 		else if (objectid > btrfs_ino(&entry->vfs_inode))
5010 			node = node->rb_right;
5011 		else
5012 			break;
5013 	}
5014 	if (!node) {
5015 		while (prev) {
5016 			entry = rb_entry(prev, struct btrfs_inode, rb_node);
5017 			if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5018 				node = prev;
5019 				break;
5020 			}
5021 			prev = rb_next(prev);
5022 		}
5023 	}
5024 	while (node) {
5025 		entry = rb_entry(node, struct btrfs_inode, rb_node);
5026 		objectid = btrfs_ino(&entry->vfs_inode) + 1;
5027 		inode = igrab(&entry->vfs_inode);
5028 		if (inode) {
5029 			spin_unlock(&root->inode_lock);
5030 			if (atomic_read(&inode->i_count) > 1)
5031 				d_prune_aliases(inode);
5032 			/*
5033 			 * btrfs_drop_inode will have it removed from
5034 			 * the inode cache when its usage count
5035 			 * hits zero.
5036 			 */
5037 			iput(inode);
5038 			cond_resched();
5039 			spin_lock(&root->inode_lock);
5040 			goto again;
5041 		}
5042 
5043 		if (cond_resched_lock(&root->inode_lock))
5044 			goto again;
5045 
5046 		node = rb_next(node);
5047 	}
5048 	spin_unlock(&root->inode_lock);
5049 }
5050 
5051 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5052 {
5053 	struct btrfs_iget_args *args = p;
5054 	inode->i_ino = args->location->objectid;
5055 	memcpy(&BTRFS_I(inode)->location, args->location,
5056 	       sizeof(*args->location));
5057 	BTRFS_I(inode)->root = args->root;
5058 	return 0;
5059 }
5060 
5061 static int btrfs_find_actor(struct inode *inode, void *opaque)
5062 {
5063 	struct btrfs_iget_args *args = opaque;
5064 	return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5065 		args->root == BTRFS_I(inode)->root;
5066 }
5067 
5068 static struct inode *btrfs_iget_locked(struct super_block *s,
5069 				       struct btrfs_key *location,
5070 				       struct btrfs_root *root)
5071 {
5072 	struct inode *inode;
5073 	struct btrfs_iget_args args;
5074 	unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5075 
5076 	args.location = location;
5077 	args.root = root;
5078 
5079 	inode = iget5_locked(s, hashval, btrfs_find_actor,
5080 			     btrfs_init_locked_inode,
5081 			     (void *)&args);
5082 	return inode;
5083 }
5084 
5085 /* Get an inode object given its location and corresponding root.
5086  * Returns in *is_new if the inode was read from disk
5087  */
5088 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5089 			 struct btrfs_root *root, int *new)
5090 {
5091 	struct inode *inode;
5092 
5093 	inode = btrfs_iget_locked(s, location, root);
5094 	if (!inode)
5095 		return ERR_PTR(-ENOMEM);
5096 
5097 	if (inode->i_state & I_NEW) {
5098 		btrfs_read_locked_inode(inode);
5099 		if (!is_bad_inode(inode)) {
5100 			inode_tree_add(inode);
5101 			unlock_new_inode(inode);
5102 			if (new)
5103 				*new = 1;
5104 		} else {
5105 			unlock_new_inode(inode);
5106 			iput(inode);
5107 			inode = ERR_PTR(-ESTALE);
5108 		}
5109 	}
5110 
5111 	return inode;
5112 }
5113 
5114 static struct inode *new_simple_dir(struct super_block *s,
5115 				    struct btrfs_key *key,
5116 				    struct btrfs_root *root)
5117 {
5118 	struct inode *inode = new_inode(s);
5119 
5120 	if (!inode)
5121 		return ERR_PTR(-ENOMEM);
5122 
5123 	BTRFS_I(inode)->root = root;
5124 	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5125 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5126 
5127 	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5128 	inode->i_op = &btrfs_dir_ro_inode_operations;
5129 	inode->i_fop = &simple_dir_operations;
5130 	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5131 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5132 
5133 	return inode;
5134 }
5135 
5136 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5137 {
5138 	struct inode *inode;
5139 	struct btrfs_root *root = BTRFS_I(dir)->root;
5140 	struct btrfs_root *sub_root = root;
5141 	struct btrfs_key location;
5142 	int index;
5143 	int ret = 0;
5144 
5145 	if (dentry->d_name.len > BTRFS_NAME_LEN)
5146 		return ERR_PTR(-ENAMETOOLONG);
5147 
5148 	ret = btrfs_inode_by_name(dir, dentry, &location);
5149 	if (ret < 0)
5150 		return ERR_PTR(ret);
5151 
5152 	if (location.objectid == 0)
5153 		return ERR_PTR(-ENOENT);
5154 
5155 	if (location.type == BTRFS_INODE_ITEM_KEY) {
5156 		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5157 		return inode;
5158 	}
5159 
5160 	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5161 
5162 	index = srcu_read_lock(&root->fs_info->subvol_srcu);
5163 	ret = fixup_tree_root_location(root, dir, dentry,
5164 				       &location, &sub_root);
5165 	if (ret < 0) {
5166 		if (ret != -ENOENT)
5167 			inode = ERR_PTR(ret);
5168 		else
5169 			inode = new_simple_dir(dir->i_sb, &location, sub_root);
5170 	} else {
5171 		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5172 	}
5173 	srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5174 
5175 	if (!IS_ERR(inode) && root != sub_root) {
5176 		down_read(&root->fs_info->cleanup_work_sem);
5177 		if (!(inode->i_sb->s_flags & MS_RDONLY))
5178 			ret = btrfs_orphan_cleanup(sub_root);
5179 		up_read(&root->fs_info->cleanup_work_sem);
5180 		if (ret) {
5181 			iput(inode);
5182 			inode = ERR_PTR(ret);
5183 		}
5184 	}
5185 
5186 	return inode;
5187 }
5188 
5189 static int btrfs_dentry_delete(const struct dentry *dentry)
5190 {
5191 	struct btrfs_root *root;
5192 	struct inode *inode = dentry->d_inode;
5193 
5194 	if (!inode && !IS_ROOT(dentry))
5195 		inode = dentry->d_parent->d_inode;
5196 
5197 	if (inode) {
5198 		root = BTRFS_I(inode)->root;
5199 		if (btrfs_root_refs(&root->root_item) == 0)
5200 			return 1;
5201 
5202 		if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5203 			return 1;
5204 	}
5205 	return 0;
5206 }
5207 
5208 static void btrfs_dentry_release(struct dentry *dentry)
5209 {
5210 	kfree(dentry->d_fsdata);
5211 }
5212 
5213 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5214 				   unsigned int flags)
5215 {
5216 	struct inode *inode;
5217 
5218 	inode = btrfs_lookup_dentry(dir, dentry);
5219 	if (IS_ERR(inode)) {
5220 		if (PTR_ERR(inode) == -ENOENT)
5221 			inode = NULL;
5222 		else
5223 			return ERR_CAST(inode);
5224 	}
5225 
5226 	return d_materialise_unique(dentry, inode);
5227 }
5228 
5229 unsigned char btrfs_filetype_table[] = {
5230 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5231 };
5232 
5233 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5234 {
5235 	struct inode *inode = file_inode(file);
5236 	struct btrfs_root *root = BTRFS_I(inode)->root;
5237 	struct btrfs_item *item;
5238 	struct btrfs_dir_item *di;
5239 	struct btrfs_key key;
5240 	struct btrfs_key found_key;
5241 	struct btrfs_path *path;
5242 	struct list_head ins_list;
5243 	struct list_head del_list;
5244 	int ret;
5245 	struct extent_buffer *leaf;
5246 	int slot;
5247 	unsigned char d_type;
5248 	int over = 0;
5249 	u32 di_cur;
5250 	u32 di_total;
5251 	u32 di_len;
5252 	int key_type = BTRFS_DIR_INDEX_KEY;
5253 	char tmp_name[32];
5254 	char *name_ptr;
5255 	int name_len;
5256 	int is_curr = 0;	/* ctx->pos points to the current index? */
5257 
5258 	/* FIXME, use a real flag for deciding about the key type */
5259 	if (root->fs_info->tree_root == root)
5260 		key_type = BTRFS_DIR_ITEM_KEY;
5261 
5262 	if (!dir_emit_dots(file, ctx))
5263 		return 0;
5264 
5265 	path = btrfs_alloc_path();
5266 	if (!path)
5267 		return -ENOMEM;
5268 
5269 	path->reada = 1;
5270 
5271 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5272 		INIT_LIST_HEAD(&ins_list);
5273 		INIT_LIST_HEAD(&del_list);
5274 		btrfs_get_delayed_items(inode, &ins_list, &del_list);
5275 	}
5276 
5277 	btrfs_set_key_type(&key, key_type);
5278 	key.offset = ctx->pos;
5279 	key.objectid = btrfs_ino(inode);
5280 
5281 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5282 	if (ret < 0)
5283 		goto err;
5284 
5285 	while (1) {
5286 		leaf = path->nodes[0];
5287 		slot = path->slots[0];
5288 		if (slot >= btrfs_header_nritems(leaf)) {
5289 			ret = btrfs_next_leaf(root, path);
5290 			if (ret < 0)
5291 				goto err;
5292 			else if (ret > 0)
5293 				break;
5294 			continue;
5295 		}
5296 
5297 		item = btrfs_item_nr(slot);
5298 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
5299 
5300 		if (found_key.objectid != key.objectid)
5301 			break;
5302 		if (btrfs_key_type(&found_key) != key_type)
5303 			break;
5304 		if (found_key.offset < ctx->pos)
5305 			goto next;
5306 		if (key_type == BTRFS_DIR_INDEX_KEY &&
5307 		    btrfs_should_delete_dir_index(&del_list,
5308 						  found_key.offset))
5309 			goto next;
5310 
5311 		ctx->pos = found_key.offset;
5312 		is_curr = 1;
5313 
5314 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5315 		di_cur = 0;
5316 		di_total = btrfs_item_size(leaf, item);
5317 
5318 		while (di_cur < di_total) {
5319 			struct btrfs_key location;
5320 
5321 			if (verify_dir_item(root, leaf, di))
5322 				break;
5323 
5324 			name_len = btrfs_dir_name_len(leaf, di);
5325 			if (name_len <= sizeof(tmp_name)) {
5326 				name_ptr = tmp_name;
5327 			} else {
5328 				name_ptr = kmalloc(name_len, GFP_NOFS);
5329 				if (!name_ptr) {
5330 					ret = -ENOMEM;
5331 					goto err;
5332 				}
5333 			}
5334 			read_extent_buffer(leaf, name_ptr,
5335 					   (unsigned long)(di + 1), name_len);
5336 
5337 			d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5338 			btrfs_dir_item_key_to_cpu(leaf, di, &location);
5339 
5340 
5341 			/* is this a reference to our own snapshot? If so
5342 			 * skip it.
5343 			 *
5344 			 * In contrast to old kernels, we insert the snapshot's
5345 			 * dir item and dir index after it has been created, so
5346 			 * we won't find a reference to our own snapshot. We
5347 			 * still keep the following code for backward
5348 			 * compatibility.
5349 			 */
5350 			if (location.type == BTRFS_ROOT_ITEM_KEY &&
5351 			    location.objectid == root->root_key.objectid) {
5352 				over = 0;
5353 				goto skip;
5354 			}
5355 			over = !dir_emit(ctx, name_ptr, name_len,
5356 				       location.objectid, d_type);
5357 
5358 skip:
5359 			if (name_ptr != tmp_name)
5360 				kfree(name_ptr);
5361 
5362 			if (over)
5363 				goto nopos;
5364 			di_len = btrfs_dir_name_len(leaf, di) +
5365 				 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5366 			di_cur += di_len;
5367 			di = (struct btrfs_dir_item *)((char *)di + di_len);
5368 		}
5369 next:
5370 		path->slots[0]++;
5371 	}
5372 
5373 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5374 		if (is_curr)
5375 			ctx->pos++;
5376 		ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5377 		if (ret)
5378 			goto nopos;
5379 	}
5380 
5381 	/* Reached end of directory/root. Bump pos past the last item. */
5382 	ctx->pos++;
5383 
5384 	/*
5385 	 * Stop new entries from being returned after we return the last
5386 	 * entry.
5387 	 *
5388 	 * New directory entries are assigned a strictly increasing
5389 	 * offset.  This means that new entries created during readdir
5390 	 * are *guaranteed* to be seen in the future by that readdir.
5391 	 * This has broken buggy programs which operate on names as
5392 	 * they're returned by readdir.  Until we re-use freed offsets
5393 	 * we have this hack to stop new entries from being returned
5394 	 * under the assumption that they'll never reach this huge
5395 	 * offset.
5396 	 *
5397 	 * This is being careful not to overflow 32bit loff_t unless the
5398 	 * last entry requires it because doing so has broken 32bit apps
5399 	 * in the past.
5400 	 */
5401 	if (key_type == BTRFS_DIR_INDEX_KEY) {
5402 		if (ctx->pos >= INT_MAX)
5403 			ctx->pos = LLONG_MAX;
5404 		else
5405 			ctx->pos = INT_MAX;
5406 	}
5407 nopos:
5408 	ret = 0;
5409 err:
5410 	if (key_type == BTRFS_DIR_INDEX_KEY)
5411 		btrfs_put_delayed_items(&ins_list, &del_list);
5412 	btrfs_free_path(path);
5413 	return ret;
5414 }
5415 
5416 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5417 {
5418 	struct btrfs_root *root = BTRFS_I(inode)->root;
5419 	struct btrfs_trans_handle *trans;
5420 	int ret = 0;
5421 	bool nolock = false;
5422 
5423 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5424 		return 0;
5425 
5426 	if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5427 		nolock = true;
5428 
5429 	if (wbc->sync_mode == WB_SYNC_ALL) {
5430 		if (nolock)
5431 			trans = btrfs_join_transaction_nolock(root);
5432 		else
5433 			trans = btrfs_join_transaction(root);
5434 		if (IS_ERR(trans))
5435 			return PTR_ERR(trans);
5436 		ret = btrfs_commit_transaction(trans, root);
5437 	}
5438 	return ret;
5439 }
5440 
5441 /*
5442  * This is somewhat expensive, updating the tree every time the
5443  * inode changes.  But, it is most likely to find the inode in cache.
5444  * FIXME, needs more benchmarking...there are no reasons other than performance
5445  * to keep or drop this code.
5446  */
5447 static int btrfs_dirty_inode(struct inode *inode)
5448 {
5449 	struct btrfs_root *root = BTRFS_I(inode)->root;
5450 	struct btrfs_trans_handle *trans;
5451 	int ret;
5452 
5453 	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5454 		return 0;
5455 
5456 	trans = btrfs_join_transaction(root);
5457 	if (IS_ERR(trans))
5458 		return PTR_ERR(trans);
5459 
5460 	ret = btrfs_update_inode(trans, root, inode);
5461 	if (ret && ret == -ENOSPC) {
5462 		/* whoops, lets try again with the full transaction */
5463 		btrfs_end_transaction(trans, root);
5464 		trans = btrfs_start_transaction(root, 1);
5465 		if (IS_ERR(trans))
5466 			return PTR_ERR(trans);
5467 
5468 		ret = btrfs_update_inode(trans, root, inode);
5469 	}
5470 	btrfs_end_transaction(trans, root);
5471 	if (BTRFS_I(inode)->delayed_node)
5472 		btrfs_balance_delayed_items(root);
5473 
5474 	return ret;
5475 }
5476 
5477 /*
5478  * This is a copy of file_update_time.  We need this so we can return error on
5479  * ENOSPC for updating the inode in the case of file write and mmap writes.
5480  */
5481 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5482 			     int flags)
5483 {
5484 	struct btrfs_root *root = BTRFS_I(inode)->root;
5485 
5486 	if (btrfs_root_readonly(root))
5487 		return -EROFS;
5488 
5489 	if (flags & S_VERSION)
5490 		inode_inc_iversion(inode);
5491 	if (flags & S_CTIME)
5492 		inode->i_ctime = *now;
5493 	if (flags & S_MTIME)
5494 		inode->i_mtime = *now;
5495 	if (flags & S_ATIME)
5496 		inode->i_atime = *now;
5497 	return btrfs_dirty_inode(inode);
5498 }
5499 
5500 /*
5501  * find the highest existing sequence number in a directory
5502  * and then set the in-memory index_cnt variable to reflect
5503  * free sequence numbers
5504  */
5505 static int btrfs_set_inode_index_count(struct inode *inode)
5506 {
5507 	struct btrfs_root *root = BTRFS_I(inode)->root;
5508 	struct btrfs_key key, found_key;
5509 	struct btrfs_path *path;
5510 	struct extent_buffer *leaf;
5511 	int ret;
5512 
5513 	key.objectid = btrfs_ino(inode);
5514 	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5515 	key.offset = (u64)-1;
5516 
5517 	path = btrfs_alloc_path();
5518 	if (!path)
5519 		return -ENOMEM;
5520 
5521 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5522 	if (ret < 0)
5523 		goto out;
5524 	/* FIXME: we should be able to handle this */
5525 	if (ret == 0)
5526 		goto out;
5527 	ret = 0;
5528 
5529 	/*
5530 	 * MAGIC NUMBER EXPLANATION:
5531 	 * since we search a directory based on f_pos we have to start at 2
5532 	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5533 	 * else has to start at 2
5534 	 */
5535 	if (path->slots[0] == 0) {
5536 		BTRFS_I(inode)->index_cnt = 2;
5537 		goto out;
5538 	}
5539 
5540 	path->slots[0]--;
5541 
5542 	leaf = path->nodes[0];
5543 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5544 
5545 	if (found_key.objectid != btrfs_ino(inode) ||
5546 	    btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5547 		BTRFS_I(inode)->index_cnt = 2;
5548 		goto out;
5549 	}
5550 
5551 	BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5552 out:
5553 	btrfs_free_path(path);
5554 	return ret;
5555 }
5556 
5557 /*
5558  * helper to find a free sequence number in a given directory.  This current
5559  * code is very simple, later versions will do smarter things in the btree
5560  */
5561 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5562 {
5563 	int ret = 0;
5564 
5565 	if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5566 		ret = btrfs_inode_delayed_dir_index_count(dir);
5567 		if (ret) {
5568 			ret = btrfs_set_inode_index_count(dir);
5569 			if (ret)
5570 				return ret;
5571 		}
5572 	}
5573 
5574 	*index = BTRFS_I(dir)->index_cnt;
5575 	BTRFS_I(dir)->index_cnt++;
5576 
5577 	return ret;
5578 }
5579 
5580 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5581 				     struct btrfs_root *root,
5582 				     struct inode *dir,
5583 				     const char *name, int name_len,
5584 				     u64 ref_objectid, u64 objectid,
5585 				     umode_t mode, u64 *index)
5586 {
5587 	struct inode *inode;
5588 	struct btrfs_inode_item *inode_item;
5589 	struct btrfs_key *location;
5590 	struct btrfs_path *path;
5591 	struct btrfs_inode_ref *ref;
5592 	struct btrfs_key key[2];
5593 	u32 sizes[2];
5594 	int nitems = name ? 2 : 1;
5595 	unsigned long ptr;
5596 	int ret;
5597 
5598 	path = btrfs_alloc_path();
5599 	if (!path)
5600 		return ERR_PTR(-ENOMEM);
5601 
5602 	inode = new_inode(root->fs_info->sb);
5603 	if (!inode) {
5604 		btrfs_free_path(path);
5605 		return ERR_PTR(-ENOMEM);
5606 	}
5607 
5608 	/*
5609 	 * we have to initialize this early, so we can reclaim the inode
5610 	 * number if we fail afterwards in this function.
5611 	 */
5612 	inode->i_ino = objectid;
5613 
5614 	if (dir && name) {
5615 		trace_btrfs_inode_request(dir);
5616 
5617 		ret = btrfs_set_inode_index(dir, index);
5618 		if (ret) {
5619 			btrfs_free_path(path);
5620 			iput(inode);
5621 			return ERR_PTR(ret);
5622 		}
5623 	} else if (dir) {
5624 		*index = 0;
5625 	}
5626 	/*
5627 	 * index_cnt is ignored for everything but a dir,
5628 	 * btrfs_get_inode_index_count has an explanation for the magic
5629 	 * number
5630 	 */
5631 	BTRFS_I(inode)->index_cnt = 2;
5632 	BTRFS_I(inode)->dir_index = *index;
5633 	BTRFS_I(inode)->root = root;
5634 	BTRFS_I(inode)->generation = trans->transid;
5635 	inode->i_generation = BTRFS_I(inode)->generation;
5636 
5637 	/*
5638 	 * We could have gotten an inode number from somebody who was fsynced
5639 	 * and then removed in this same transaction, so let's just set full
5640 	 * sync since it will be a full sync anyway and this will blow away the
5641 	 * old info in the log.
5642 	 */
5643 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5644 
5645 	key[0].objectid = objectid;
5646 	btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5647 	key[0].offset = 0;
5648 
5649 	sizes[0] = sizeof(struct btrfs_inode_item);
5650 
5651 	if (name) {
5652 		/*
5653 		 * Start new inodes with an inode_ref. This is slightly more
5654 		 * efficient for small numbers of hard links since they will
5655 		 * be packed into one item. Extended refs will kick in if we
5656 		 * add more hard links than can fit in the ref item.
5657 		 */
5658 		key[1].objectid = objectid;
5659 		btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5660 		key[1].offset = ref_objectid;
5661 
5662 		sizes[1] = name_len + sizeof(*ref);
5663 	}
5664 
5665 	path->leave_spinning = 1;
5666 	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5667 	if (ret != 0)
5668 		goto fail;
5669 
5670 	inode_init_owner(inode, dir, mode);
5671 	inode_set_bytes(inode, 0);
5672 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5673 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5674 				  struct btrfs_inode_item);
5675 	memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5676 			     sizeof(*inode_item));
5677 	fill_inode_item(trans, path->nodes[0], inode_item, inode);
5678 
5679 	if (name) {
5680 		ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5681 				     struct btrfs_inode_ref);
5682 		btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5683 		btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5684 		ptr = (unsigned long)(ref + 1);
5685 		write_extent_buffer(path->nodes[0], name, ptr, name_len);
5686 	}
5687 
5688 	btrfs_mark_buffer_dirty(path->nodes[0]);
5689 	btrfs_free_path(path);
5690 
5691 	location = &BTRFS_I(inode)->location;
5692 	location->objectid = objectid;
5693 	location->offset = 0;
5694 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5695 
5696 	btrfs_inherit_iflags(inode, dir);
5697 
5698 	if (S_ISREG(mode)) {
5699 		if (btrfs_test_opt(root, NODATASUM))
5700 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5701 		if (btrfs_test_opt(root, NODATACOW))
5702 			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5703 				BTRFS_INODE_NODATASUM;
5704 	}
5705 
5706 	btrfs_insert_inode_hash(inode);
5707 	inode_tree_add(inode);
5708 
5709 	trace_btrfs_inode_new(inode);
5710 	btrfs_set_inode_last_trans(trans, inode);
5711 
5712 	btrfs_update_root_times(trans, root);
5713 
5714 	ret = btrfs_inode_inherit_props(trans, inode, dir);
5715 	if (ret)
5716 		btrfs_err(root->fs_info,
5717 			  "error inheriting props for ino %llu (root %llu): %d",
5718 			  btrfs_ino(inode), root->root_key.objectid, ret);
5719 
5720 	return inode;
5721 fail:
5722 	if (dir && name)
5723 		BTRFS_I(dir)->index_cnt--;
5724 	btrfs_free_path(path);
5725 	iput(inode);
5726 	return ERR_PTR(ret);
5727 }
5728 
5729 static inline u8 btrfs_inode_type(struct inode *inode)
5730 {
5731 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5732 }
5733 
5734 /*
5735  * utility function to add 'inode' into 'parent_inode' with
5736  * a give name and a given sequence number.
5737  * if 'add_backref' is true, also insert a backref from the
5738  * inode to the parent directory.
5739  */
5740 int btrfs_add_link(struct btrfs_trans_handle *trans,
5741 		   struct inode *parent_inode, struct inode *inode,
5742 		   const char *name, int name_len, int add_backref, u64 index)
5743 {
5744 	int ret = 0;
5745 	struct btrfs_key key;
5746 	struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5747 	u64 ino = btrfs_ino(inode);
5748 	u64 parent_ino = btrfs_ino(parent_inode);
5749 
5750 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5751 		memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5752 	} else {
5753 		key.objectid = ino;
5754 		btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5755 		key.offset = 0;
5756 	}
5757 
5758 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5759 		ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5760 					 key.objectid, root->root_key.objectid,
5761 					 parent_ino, index, name, name_len);
5762 	} else if (add_backref) {
5763 		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5764 					     parent_ino, index);
5765 	}
5766 
5767 	/* Nothing to clean up yet */
5768 	if (ret)
5769 		return ret;
5770 
5771 	ret = btrfs_insert_dir_item(trans, root, name, name_len,
5772 				    parent_inode, &key,
5773 				    btrfs_inode_type(inode), index);
5774 	if (ret == -EEXIST || ret == -EOVERFLOW)
5775 		goto fail_dir_item;
5776 	else if (ret) {
5777 		btrfs_abort_transaction(trans, root, ret);
5778 		return ret;
5779 	}
5780 
5781 	btrfs_i_size_write(parent_inode, parent_inode->i_size +
5782 			   name_len * 2);
5783 	inode_inc_iversion(parent_inode);
5784 	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5785 	ret = btrfs_update_inode(trans, root, parent_inode);
5786 	if (ret)
5787 		btrfs_abort_transaction(trans, root, ret);
5788 	return ret;
5789 
5790 fail_dir_item:
5791 	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5792 		u64 local_index;
5793 		int err;
5794 		err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5795 				 key.objectid, root->root_key.objectid,
5796 				 parent_ino, &local_index, name, name_len);
5797 
5798 	} else if (add_backref) {
5799 		u64 local_index;
5800 		int err;
5801 
5802 		err = btrfs_del_inode_ref(trans, root, name, name_len,
5803 					  ino, parent_ino, &local_index);
5804 	}
5805 	return ret;
5806 }
5807 
5808 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5809 			    struct inode *dir, struct dentry *dentry,
5810 			    struct inode *inode, int backref, u64 index)
5811 {
5812 	int err = btrfs_add_link(trans, dir, inode,
5813 				 dentry->d_name.name, dentry->d_name.len,
5814 				 backref, index);
5815 	if (err > 0)
5816 		err = -EEXIST;
5817 	return err;
5818 }
5819 
5820 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5821 			umode_t mode, dev_t rdev)
5822 {
5823 	struct btrfs_trans_handle *trans;
5824 	struct btrfs_root *root = BTRFS_I(dir)->root;
5825 	struct inode *inode = NULL;
5826 	int err;
5827 	int drop_inode = 0;
5828 	u64 objectid;
5829 	u64 index = 0;
5830 
5831 	if (!new_valid_dev(rdev))
5832 		return -EINVAL;
5833 
5834 	/*
5835 	 * 2 for inode item and ref
5836 	 * 2 for dir items
5837 	 * 1 for xattr if selinux is on
5838 	 */
5839 	trans = btrfs_start_transaction(root, 5);
5840 	if (IS_ERR(trans))
5841 		return PTR_ERR(trans);
5842 
5843 	err = btrfs_find_free_ino(root, &objectid);
5844 	if (err)
5845 		goto out_unlock;
5846 
5847 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5848 				dentry->d_name.len, btrfs_ino(dir), objectid,
5849 				mode, &index);
5850 	if (IS_ERR(inode)) {
5851 		err = PTR_ERR(inode);
5852 		goto out_unlock;
5853 	}
5854 
5855 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5856 	if (err) {
5857 		drop_inode = 1;
5858 		goto out_unlock;
5859 	}
5860 
5861 	/*
5862 	* If the active LSM wants to access the inode during
5863 	* d_instantiate it needs these. Smack checks to see
5864 	* if the filesystem supports xattrs by looking at the
5865 	* ops vector.
5866 	*/
5867 
5868 	inode->i_op = &btrfs_special_inode_operations;
5869 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5870 	if (err)
5871 		drop_inode = 1;
5872 	else {
5873 		init_special_inode(inode, inode->i_mode, rdev);
5874 		btrfs_update_inode(trans, root, inode);
5875 		d_instantiate(dentry, inode);
5876 	}
5877 out_unlock:
5878 	btrfs_end_transaction(trans, root);
5879 	btrfs_balance_delayed_items(root);
5880 	btrfs_btree_balance_dirty(root);
5881 	if (drop_inode) {
5882 		inode_dec_link_count(inode);
5883 		iput(inode);
5884 	}
5885 	return err;
5886 }
5887 
5888 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5889 			umode_t mode, bool excl)
5890 {
5891 	struct btrfs_trans_handle *trans;
5892 	struct btrfs_root *root = BTRFS_I(dir)->root;
5893 	struct inode *inode = NULL;
5894 	int drop_inode_on_err = 0;
5895 	int err;
5896 	u64 objectid;
5897 	u64 index = 0;
5898 
5899 	/*
5900 	 * 2 for inode item and ref
5901 	 * 2 for dir items
5902 	 * 1 for xattr if selinux is on
5903 	 */
5904 	trans = btrfs_start_transaction(root, 5);
5905 	if (IS_ERR(trans))
5906 		return PTR_ERR(trans);
5907 
5908 	err = btrfs_find_free_ino(root, &objectid);
5909 	if (err)
5910 		goto out_unlock;
5911 
5912 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5913 				dentry->d_name.len, btrfs_ino(dir), objectid,
5914 				mode, &index);
5915 	if (IS_ERR(inode)) {
5916 		err = PTR_ERR(inode);
5917 		goto out_unlock;
5918 	}
5919 	drop_inode_on_err = 1;
5920 
5921 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5922 	if (err)
5923 		goto out_unlock;
5924 
5925 	err = btrfs_update_inode(trans, root, inode);
5926 	if (err)
5927 		goto out_unlock;
5928 
5929 	/*
5930 	* If the active LSM wants to access the inode during
5931 	* d_instantiate it needs these. Smack checks to see
5932 	* if the filesystem supports xattrs by looking at the
5933 	* ops vector.
5934 	*/
5935 	inode->i_fop = &btrfs_file_operations;
5936 	inode->i_op = &btrfs_file_inode_operations;
5937 
5938 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5939 	if (err)
5940 		goto out_unlock;
5941 
5942 	inode->i_mapping->a_ops = &btrfs_aops;
5943 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5944 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5945 	d_instantiate(dentry, inode);
5946 
5947 out_unlock:
5948 	btrfs_end_transaction(trans, root);
5949 	if (err && drop_inode_on_err) {
5950 		inode_dec_link_count(inode);
5951 		iput(inode);
5952 	}
5953 	btrfs_balance_delayed_items(root);
5954 	btrfs_btree_balance_dirty(root);
5955 	return err;
5956 }
5957 
5958 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5959 		      struct dentry *dentry)
5960 {
5961 	struct btrfs_trans_handle *trans;
5962 	struct btrfs_root *root = BTRFS_I(dir)->root;
5963 	struct inode *inode = old_dentry->d_inode;
5964 	u64 index;
5965 	int err;
5966 	int drop_inode = 0;
5967 
5968 	/* do not allow sys_link's with other subvols of the same device */
5969 	if (root->objectid != BTRFS_I(inode)->root->objectid)
5970 		return -EXDEV;
5971 
5972 	if (inode->i_nlink >= BTRFS_LINK_MAX)
5973 		return -EMLINK;
5974 
5975 	err = btrfs_set_inode_index(dir, &index);
5976 	if (err)
5977 		goto fail;
5978 
5979 	/*
5980 	 * 2 items for inode and inode ref
5981 	 * 2 items for dir items
5982 	 * 1 item for parent inode
5983 	 */
5984 	trans = btrfs_start_transaction(root, 5);
5985 	if (IS_ERR(trans)) {
5986 		err = PTR_ERR(trans);
5987 		goto fail;
5988 	}
5989 
5990 	/* There are several dir indexes for this inode, clear the cache. */
5991 	BTRFS_I(inode)->dir_index = 0ULL;
5992 	inc_nlink(inode);
5993 	inode_inc_iversion(inode);
5994 	inode->i_ctime = CURRENT_TIME;
5995 	ihold(inode);
5996 	set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5997 
5998 	err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5999 
6000 	if (err) {
6001 		drop_inode = 1;
6002 	} else {
6003 		struct dentry *parent = dentry->d_parent;
6004 		err = btrfs_update_inode(trans, root, inode);
6005 		if (err)
6006 			goto fail;
6007 		if (inode->i_nlink == 1) {
6008 			/*
6009 			 * If new hard link count is 1, it's a file created
6010 			 * with open(2) O_TMPFILE flag.
6011 			 */
6012 			err = btrfs_orphan_del(trans, inode);
6013 			if (err)
6014 				goto fail;
6015 		}
6016 		d_instantiate(dentry, inode);
6017 		btrfs_log_new_name(trans, inode, NULL, parent);
6018 	}
6019 
6020 	btrfs_end_transaction(trans, root);
6021 	btrfs_balance_delayed_items(root);
6022 fail:
6023 	if (drop_inode) {
6024 		inode_dec_link_count(inode);
6025 		iput(inode);
6026 	}
6027 	btrfs_btree_balance_dirty(root);
6028 	return err;
6029 }
6030 
6031 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6032 {
6033 	struct inode *inode = NULL;
6034 	struct btrfs_trans_handle *trans;
6035 	struct btrfs_root *root = BTRFS_I(dir)->root;
6036 	int err = 0;
6037 	int drop_on_err = 0;
6038 	u64 objectid = 0;
6039 	u64 index = 0;
6040 
6041 	/*
6042 	 * 2 items for inode and ref
6043 	 * 2 items for dir items
6044 	 * 1 for xattr if selinux is on
6045 	 */
6046 	trans = btrfs_start_transaction(root, 5);
6047 	if (IS_ERR(trans))
6048 		return PTR_ERR(trans);
6049 
6050 	err = btrfs_find_free_ino(root, &objectid);
6051 	if (err)
6052 		goto out_fail;
6053 
6054 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6055 				dentry->d_name.len, btrfs_ino(dir), objectid,
6056 				S_IFDIR | mode, &index);
6057 	if (IS_ERR(inode)) {
6058 		err = PTR_ERR(inode);
6059 		goto out_fail;
6060 	}
6061 
6062 	drop_on_err = 1;
6063 
6064 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6065 	if (err)
6066 		goto out_fail;
6067 
6068 	inode->i_op = &btrfs_dir_inode_operations;
6069 	inode->i_fop = &btrfs_dir_file_operations;
6070 
6071 	btrfs_i_size_write(inode, 0);
6072 	err = btrfs_update_inode(trans, root, inode);
6073 	if (err)
6074 		goto out_fail;
6075 
6076 	err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6077 			     dentry->d_name.len, 0, index);
6078 	if (err)
6079 		goto out_fail;
6080 
6081 	d_instantiate(dentry, inode);
6082 	drop_on_err = 0;
6083 
6084 out_fail:
6085 	btrfs_end_transaction(trans, root);
6086 	if (drop_on_err)
6087 		iput(inode);
6088 	btrfs_balance_delayed_items(root);
6089 	btrfs_btree_balance_dirty(root);
6090 	return err;
6091 }
6092 
6093 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6094  * and an extent that you want to insert, deal with overlap and insert
6095  * the new extent into the tree.
6096  */
6097 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6098 				struct extent_map *existing,
6099 				struct extent_map *em,
6100 				u64 map_start, u64 map_len)
6101 {
6102 	u64 start_diff;
6103 
6104 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6105 	start_diff = map_start - em->start;
6106 	em->start = map_start;
6107 	em->len = map_len;
6108 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6109 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6110 		em->block_start += start_diff;
6111 		em->block_len -= start_diff;
6112 	}
6113 	return add_extent_mapping(em_tree, em, 0);
6114 }
6115 
6116 static noinline int uncompress_inline(struct btrfs_path *path,
6117 				      struct inode *inode, struct page *page,
6118 				      size_t pg_offset, u64 extent_offset,
6119 				      struct btrfs_file_extent_item *item)
6120 {
6121 	int ret;
6122 	struct extent_buffer *leaf = path->nodes[0];
6123 	char *tmp;
6124 	size_t max_size;
6125 	unsigned long inline_size;
6126 	unsigned long ptr;
6127 	int compress_type;
6128 
6129 	WARN_ON(pg_offset != 0);
6130 	compress_type = btrfs_file_extent_compression(leaf, item);
6131 	max_size = btrfs_file_extent_ram_bytes(leaf, item);
6132 	inline_size = btrfs_file_extent_inline_item_len(leaf,
6133 					btrfs_item_nr(path->slots[0]));
6134 	tmp = kmalloc(inline_size, GFP_NOFS);
6135 	if (!tmp)
6136 		return -ENOMEM;
6137 	ptr = btrfs_file_extent_inline_start(item);
6138 
6139 	read_extent_buffer(leaf, tmp, ptr, inline_size);
6140 
6141 	max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6142 	ret = btrfs_decompress(compress_type, tmp, page,
6143 			       extent_offset, inline_size, max_size);
6144 	kfree(tmp);
6145 	return ret;
6146 }
6147 
6148 /*
6149  * a bit scary, this does extent mapping from logical file offset to the disk.
6150  * the ugly parts come from merging extents from the disk with the in-ram
6151  * representation.  This gets more complex because of the data=ordered code,
6152  * where the in-ram extents might be locked pending data=ordered completion.
6153  *
6154  * This also copies inline extents directly into the page.
6155  */
6156 
6157 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6158 				    size_t pg_offset, u64 start, u64 len,
6159 				    int create)
6160 {
6161 	int ret;
6162 	int err = 0;
6163 	u64 extent_start = 0;
6164 	u64 extent_end = 0;
6165 	u64 objectid = btrfs_ino(inode);
6166 	u32 found_type;
6167 	struct btrfs_path *path = NULL;
6168 	struct btrfs_root *root = BTRFS_I(inode)->root;
6169 	struct btrfs_file_extent_item *item;
6170 	struct extent_buffer *leaf;
6171 	struct btrfs_key found_key;
6172 	struct extent_map *em = NULL;
6173 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6174 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6175 	struct btrfs_trans_handle *trans = NULL;
6176 	const bool new_inline = !page || create;
6177 
6178 again:
6179 	read_lock(&em_tree->lock);
6180 	em = lookup_extent_mapping(em_tree, start, len);
6181 	if (em)
6182 		em->bdev = root->fs_info->fs_devices->latest_bdev;
6183 	read_unlock(&em_tree->lock);
6184 
6185 	if (em) {
6186 		if (em->start > start || em->start + em->len <= start)
6187 			free_extent_map(em);
6188 		else if (em->block_start == EXTENT_MAP_INLINE && page)
6189 			free_extent_map(em);
6190 		else
6191 			goto out;
6192 	}
6193 	em = alloc_extent_map();
6194 	if (!em) {
6195 		err = -ENOMEM;
6196 		goto out;
6197 	}
6198 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6199 	em->start = EXTENT_MAP_HOLE;
6200 	em->orig_start = EXTENT_MAP_HOLE;
6201 	em->len = (u64)-1;
6202 	em->block_len = (u64)-1;
6203 
6204 	if (!path) {
6205 		path = btrfs_alloc_path();
6206 		if (!path) {
6207 			err = -ENOMEM;
6208 			goto out;
6209 		}
6210 		/*
6211 		 * Chances are we'll be called again, so go ahead and do
6212 		 * readahead
6213 		 */
6214 		path->reada = 1;
6215 	}
6216 
6217 	ret = btrfs_lookup_file_extent(trans, root, path,
6218 				       objectid, start, trans != NULL);
6219 	if (ret < 0) {
6220 		err = ret;
6221 		goto out;
6222 	}
6223 
6224 	if (ret != 0) {
6225 		if (path->slots[0] == 0)
6226 			goto not_found;
6227 		path->slots[0]--;
6228 	}
6229 
6230 	leaf = path->nodes[0];
6231 	item = btrfs_item_ptr(leaf, path->slots[0],
6232 			      struct btrfs_file_extent_item);
6233 	/* are we inside the extent that was found? */
6234 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6235 	found_type = btrfs_key_type(&found_key);
6236 	if (found_key.objectid != objectid ||
6237 	    found_type != BTRFS_EXTENT_DATA_KEY) {
6238 		/*
6239 		 * If we backup past the first extent we want to move forward
6240 		 * and see if there is an extent in front of us, otherwise we'll
6241 		 * say there is a hole for our whole search range which can
6242 		 * cause problems.
6243 		 */
6244 		extent_end = start;
6245 		goto next;
6246 	}
6247 
6248 	found_type = btrfs_file_extent_type(leaf, item);
6249 	extent_start = found_key.offset;
6250 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6251 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6252 		extent_end = extent_start +
6253 		       btrfs_file_extent_num_bytes(leaf, item);
6254 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6255 		size_t size;
6256 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6257 		extent_end = ALIGN(extent_start + size, root->sectorsize);
6258 	}
6259 next:
6260 	if (start >= extent_end) {
6261 		path->slots[0]++;
6262 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6263 			ret = btrfs_next_leaf(root, path);
6264 			if (ret < 0) {
6265 				err = ret;
6266 				goto out;
6267 			}
6268 			if (ret > 0)
6269 				goto not_found;
6270 			leaf = path->nodes[0];
6271 		}
6272 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6273 		if (found_key.objectid != objectid ||
6274 		    found_key.type != BTRFS_EXTENT_DATA_KEY)
6275 			goto not_found;
6276 		if (start + len <= found_key.offset)
6277 			goto not_found;
6278 		em->start = start;
6279 		em->orig_start = start;
6280 		em->len = found_key.offset - start;
6281 		goto not_found_em;
6282 	}
6283 
6284 	btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6285 
6286 	if (found_type == BTRFS_FILE_EXTENT_REG ||
6287 	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6288 		goto insert;
6289 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6290 		unsigned long ptr;
6291 		char *map;
6292 		size_t size;
6293 		size_t extent_offset;
6294 		size_t copy_size;
6295 
6296 		if (new_inline)
6297 			goto out;
6298 
6299 		size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6300 		extent_offset = page_offset(page) + pg_offset - extent_start;
6301 		copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6302 				size - extent_offset);
6303 		em->start = extent_start + extent_offset;
6304 		em->len = ALIGN(copy_size, root->sectorsize);
6305 		em->orig_block_len = em->len;
6306 		em->orig_start = em->start;
6307 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6308 		if (create == 0 && !PageUptodate(page)) {
6309 			if (btrfs_file_extent_compression(leaf, item) !=
6310 			    BTRFS_COMPRESS_NONE) {
6311 				ret = uncompress_inline(path, inode, page,
6312 							pg_offset,
6313 							extent_offset, item);
6314 				if (ret) {
6315 					err = ret;
6316 					goto out;
6317 				}
6318 			} else {
6319 				map = kmap(page);
6320 				read_extent_buffer(leaf, map + pg_offset, ptr,
6321 						   copy_size);
6322 				if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6323 					memset(map + pg_offset + copy_size, 0,
6324 					       PAGE_CACHE_SIZE - pg_offset -
6325 					       copy_size);
6326 				}
6327 				kunmap(page);
6328 			}
6329 			flush_dcache_page(page);
6330 		} else if (create && PageUptodate(page)) {
6331 			BUG();
6332 			if (!trans) {
6333 				kunmap(page);
6334 				free_extent_map(em);
6335 				em = NULL;
6336 
6337 				btrfs_release_path(path);
6338 				trans = btrfs_join_transaction(root);
6339 
6340 				if (IS_ERR(trans))
6341 					return ERR_CAST(trans);
6342 				goto again;
6343 			}
6344 			map = kmap(page);
6345 			write_extent_buffer(leaf, map + pg_offset, ptr,
6346 					    copy_size);
6347 			kunmap(page);
6348 			btrfs_mark_buffer_dirty(leaf);
6349 		}
6350 		set_extent_uptodate(io_tree, em->start,
6351 				    extent_map_end(em) - 1, NULL, GFP_NOFS);
6352 		goto insert;
6353 	}
6354 not_found:
6355 	em->start = start;
6356 	em->orig_start = start;
6357 	em->len = len;
6358 not_found_em:
6359 	em->block_start = EXTENT_MAP_HOLE;
6360 	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6361 insert:
6362 	btrfs_release_path(path);
6363 	if (em->start > start || extent_map_end(em) <= start) {
6364 		btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6365 			em->start, em->len, start, len);
6366 		err = -EIO;
6367 		goto out;
6368 	}
6369 
6370 	err = 0;
6371 	write_lock(&em_tree->lock);
6372 	ret = add_extent_mapping(em_tree, em, 0);
6373 	/* it is possible that someone inserted the extent into the tree
6374 	 * while we had the lock dropped.  It is also possible that
6375 	 * an overlapping map exists in the tree
6376 	 */
6377 	if (ret == -EEXIST) {
6378 		struct extent_map *existing;
6379 
6380 		ret = 0;
6381 
6382 		existing = lookup_extent_mapping(em_tree, start, len);
6383 		if (existing && (existing->start > start ||
6384 		    existing->start + existing->len <= start)) {
6385 			free_extent_map(existing);
6386 			existing = NULL;
6387 		}
6388 		if (!existing) {
6389 			existing = lookup_extent_mapping(em_tree, em->start,
6390 							 em->len);
6391 			if (existing) {
6392 				err = merge_extent_mapping(em_tree, existing,
6393 							   em, start,
6394 							   root->sectorsize);
6395 				free_extent_map(existing);
6396 				if (err) {
6397 					free_extent_map(em);
6398 					em = NULL;
6399 				}
6400 			} else {
6401 				err = -EIO;
6402 				free_extent_map(em);
6403 				em = NULL;
6404 			}
6405 		} else {
6406 			free_extent_map(em);
6407 			em = existing;
6408 			err = 0;
6409 		}
6410 	}
6411 	write_unlock(&em_tree->lock);
6412 out:
6413 
6414 	trace_btrfs_get_extent(root, em);
6415 
6416 	if (path)
6417 		btrfs_free_path(path);
6418 	if (trans) {
6419 		ret = btrfs_end_transaction(trans, root);
6420 		if (!err)
6421 			err = ret;
6422 	}
6423 	if (err) {
6424 		free_extent_map(em);
6425 		return ERR_PTR(err);
6426 	}
6427 	BUG_ON(!em); /* Error is always set */
6428 	return em;
6429 }
6430 
6431 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6432 					   size_t pg_offset, u64 start, u64 len,
6433 					   int create)
6434 {
6435 	struct extent_map *em;
6436 	struct extent_map *hole_em = NULL;
6437 	u64 range_start = start;
6438 	u64 end;
6439 	u64 found;
6440 	u64 found_end;
6441 	int err = 0;
6442 
6443 	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6444 	if (IS_ERR(em))
6445 		return em;
6446 	if (em) {
6447 		/*
6448 		 * if our em maps to
6449 		 * -  a hole or
6450 		 * -  a pre-alloc extent,
6451 		 * there might actually be delalloc bytes behind it.
6452 		 */
6453 		if (em->block_start != EXTENT_MAP_HOLE &&
6454 		    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6455 			return em;
6456 		else
6457 			hole_em = em;
6458 	}
6459 
6460 	/* check to see if we've wrapped (len == -1 or similar) */
6461 	end = start + len;
6462 	if (end < start)
6463 		end = (u64)-1;
6464 	else
6465 		end -= 1;
6466 
6467 	em = NULL;
6468 
6469 	/* ok, we didn't find anything, lets look for delalloc */
6470 	found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6471 				 end, len, EXTENT_DELALLOC, 1);
6472 	found_end = range_start + found;
6473 	if (found_end < range_start)
6474 		found_end = (u64)-1;
6475 
6476 	/*
6477 	 * we didn't find anything useful, return
6478 	 * the original results from get_extent()
6479 	 */
6480 	if (range_start > end || found_end <= start) {
6481 		em = hole_em;
6482 		hole_em = NULL;
6483 		goto out;
6484 	}
6485 
6486 	/* adjust the range_start to make sure it doesn't
6487 	 * go backwards from the start they passed in
6488 	 */
6489 	range_start = max(start, range_start);
6490 	found = found_end - range_start;
6491 
6492 	if (found > 0) {
6493 		u64 hole_start = start;
6494 		u64 hole_len = len;
6495 
6496 		em = alloc_extent_map();
6497 		if (!em) {
6498 			err = -ENOMEM;
6499 			goto out;
6500 		}
6501 		/*
6502 		 * when btrfs_get_extent can't find anything it
6503 		 * returns one huge hole
6504 		 *
6505 		 * make sure what it found really fits our range, and
6506 		 * adjust to make sure it is based on the start from
6507 		 * the caller
6508 		 */
6509 		if (hole_em) {
6510 			u64 calc_end = extent_map_end(hole_em);
6511 
6512 			if (calc_end <= start || (hole_em->start > end)) {
6513 				free_extent_map(hole_em);
6514 				hole_em = NULL;
6515 			} else {
6516 				hole_start = max(hole_em->start, start);
6517 				hole_len = calc_end - hole_start;
6518 			}
6519 		}
6520 		em->bdev = NULL;
6521 		if (hole_em && range_start > hole_start) {
6522 			/* our hole starts before our delalloc, so we
6523 			 * have to return just the parts of the hole
6524 			 * that go until  the delalloc starts
6525 			 */
6526 			em->len = min(hole_len,
6527 				      range_start - hole_start);
6528 			em->start = hole_start;
6529 			em->orig_start = hole_start;
6530 			/*
6531 			 * don't adjust block start at all,
6532 			 * it is fixed at EXTENT_MAP_HOLE
6533 			 */
6534 			em->block_start = hole_em->block_start;
6535 			em->block_len = hole_len;
6536 			if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6537 				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6538 		} else {
6539 			em->start = range_start;
6540 			em->len = found;
6541 			em->orig_start = range_start;
6542 			em->block_start = EXTENT_MAP_DELALLOC;
6543 			em->block_len = found;
6544 		}
6545 	} else if (hole_em) {
6546 		return hole_em;
6547 	}
6548 out:
6549 
6550 	free_extent_map(hole_em);
6551 	if (err) {
6552 		free_extent_map(em);
6553 		return ERR_PTR(err);
6554 	}
6555 	return em;
6556 }
6557 
6558 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6559 						  u64 start, u64 len)
6560 {
6561 	struct btrfs_root *root = BTRFS_I(inode)->root;
6562 	struct extent_map *em;
6563 	struct btrfs_key ins;
6564 	u64 alloc_hint;
6565 	int ret;
6566 
6567 	alloc_hint = get_extent_allocation_hint(inode, start, len);
6568 	ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6569 				   alloc_hint, &ins, 1, 1);
6570 	if (ret)
6571 		return ERR_PTR(ret);
6572 
6573 	em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6574 			      ins.offset, ins.offset, ins.offset, 0);
6575 	if (IS_ERR(em)) {
6576 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6577 		return em;
6578 	}
6579 
6580 	ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6581 					   ins.offset, ins.offset, 0);
6582 	if (ret) {
6583 		btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6584 		free_extent_map(em);
6585 		return ERR_PTR(ret);
6586 	}
6587 
6588 	return em;
6589 }
6590 
6591 /*
6592  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6593  * block must be cow'd
6594  */
6595 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6596 			      u64 *orig_start, u64 *orig_block_len,
6597 			      u64 *ram_bytes)
6598 {
6599 	struct btrfs_trans_handle *trans;
6600 	struct btrfs_path *path;
6601 	int ret;
6602 	struct extent_buffer *leaf;
6603 	struct btrfs_root *root = BTRFS_I(inode)->root;
6604 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6605 	struct btrfs_file_extent_item *fi;
6606 	struct btrfs_key key;
6607 	u64 disk_bytenr;
6608 	u64 backref_offset;
6609 	u64 extent_end;
6610 	u64 num_bytes;
6611 	int slot;
6612 	int found_type;
6613 	bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6614 
6615 	path = btrfs_alloc_path();
6616 	if (!path)
6617 		return -ENOMEM;
6618 
6619 	ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6620 				       offset, 0);
6621 	if (ret < 0)
6622 		goto out;
6623 
6624 	slot = path->slots[0];
6625 	if (ret == 1) {
6626 		if (slot == 0) {
6627 			/* can't find the item, must cow */
6628 			ret = 0;
6629 			goto out;
6630 		}
6631 		slot--;
6632 	}
6633 	ret = 0;
6634 	leaf = path->nodes[0];
6635 	btrfs_item_key_to_cpu(leaf, &key, slot);
6636 	if (key.objectid != btrfs_ino(inode) ||
6637 	    key.type != BTRFS_EXTENT_DATA_KEY) {
6638 		/* not our file or wrong item type, must cow */
6639 		goto out;
6640 	}
6641 
6642 	if (key.offset > offset) {
6643 		/* Wrong offset, must cow */
6644 		goto out;
6645 	}
6646 
6647 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6648 	found_type = btrfs_file_extent_type(leaf, fi);
6649 	if (found_type != BTRFS_FILE_EXTENT_REG &&
6650 	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6651 		/* not a regular extent, must cow */
6652 		goto out;
6653 	}
6654 
6655 	if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6656 		goto out;
6657 
6658 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6659 	if (extent_end <= offset)
6660 		goto out;
6661 
6662 	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6663 	if (disk_bytenr == 0)
6664 		goto out;
6665 
6666 	if (btrfs_file_extent_compression(leaf, fi) ||
6667 	    btrfs_file_extent_encryption(leaf, fi) ||
6668 	    btrfs_file_extent_other_encoding(leaf, fi))
6669 		goto out;
6670 
6671 	backref_offset = btrfs_file_extent_offset(leaf, fi);
6672 
6673 	if (orig_start) {
6674 		*orig_start = key.offset - backref_offset;
6675 		*orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6676 		*ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6677 	}
6678 
6679 	if (btrfs_extent_readonly(root, disk_bytenr))
6680 		goto out;
6681 
6682 	num_bytes = min(offset + *len, extent_end) - offset;
6683 	if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6684 		u64 range_end;
6685 
6686 		range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6687 		ret = test_range_bit(io_tree, offset, range_end,
6688 				     EXTENT_DELALLOC, 0, NULL);
6689 		if (ret) {
6690 			ret = -EAGAIN;
6691 			goto out;
6692 		}
6693 	}
6694 
6695 	btrfs_release_path(path);
6696 
6697 	/*
6698 	 * look for other files referencing this extent, if we
6699 	 * find any we must cow
6700 	 */
6701 	trans = btrfs_join_transaction(root);
6702 	if (IS_ERR(trans)) {
6703 		ret = 0;
6704 		goto out;
6705 	}
6706 
6707 	ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6708 				    key.offset - backref_offset, disk_bytenr);
6709 	btrfs_end_transaction(trans, root);
6710 	if (ret) {
6711 		ret = 0;
6712 		goto out;
6713 	}
6714 
6715 	/*
6716 	 * adjust disk_bytenr and num_bytes to cover just the bytes
6717 	 * in this extent we are about to write.  If there
6718 	 * are any csums in that range we have to cow in order
6719 	 * to keep the csums correct
6720 	 */
6721 	disk_bytenr += backref_offset;
6722 	disk_bytenr += offset - key.offset;
6723 	if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6724 				goto out;
6725 	/*
6726 	 * all of the above have passed, it is safe to overwrite this extent
6727 	 * without cow
6728 	 */
6729 	*len = num_bytes;
6730 	ret = 1;
6731 out:
6732 	btrfs_free_path(path);
6733 	return ret;
6734 }
6735 
6736 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6737 {
6738 	struct radix_tree_root *root = &inode->i_mapping->page_tree;
6739 	int found = false;
6740 	void **pagep = NULL;
6741 	struct page *page = NULL;
6742 	int start_idx;
6743 	int end_idx;
6744 
6745 	start_idx = start >> PAGE_CACHE_SHIFT;
6746 
6747 	/*
6748 	 * end is the last byte in the last page.  end == start is legal
6749 	 */
6750 	end_idx = end >> PAGE_CACHE_SHIFT;
6751 
6752 	rcu_read_lock();
6753 
6754 	/* Most of the code in this while loop is lifted from
6755 	 * find_get_page.  It's been modified to begin searching from a
6756 	 * page and return just the first page found in that range.  If the
6757 	 * found idx is less than or equal to the end idx then we know that
6758 	 * a page exists.  If no pages are found or if those pages are
6759 	 * outside of the range then we're fine (yay!) */
6760 	while (page == NULL &&
6761 	       radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6762 		page = radix_tree_deref_slot(pagep);
6763 		if (unlikely(!page))
6764 			break;
6765 
6766 		if (radix_tree_exception(page)) {
6767 			if (radix_tree_deref_retry(page)) {
6768 				page = NULL;
6769 				continue;
6770 			}
6771 			/*
6772 			 * Otherwise, shmem/tmpfs must be storing a swap entry
6773 			 * here as an exceptional entry: so return it without
6774 			 * attempting to raise page count.
6775 			 */
6776 			page = NULL;
6777 			break; /* TODO: Is this relevant for this use case? */
6778 		}
6779 
6780 		if (!page_cache_get_speculative(page)) {
6781 			page = NULL;
6782 			continue;
6783 		}
6784 
6785 		/*
6786 		 * Has the page moved?
6787 		 * This is part of the lockless pagecache protocol. See
6788 		 * include/linux/pagemap.h for details.
6789 		 */
6790 		if (unlikely(page != *pagep)) {
6791 			page_cache_release(page);
6792 			page = NULL;
6793 		}
6794 	}
6795 
6796 	if (page) {
6797 		if (page->index <= end_idx)
6798 			found = true;
6799 		page_cache_release(page);
6800 	}
6801 
6802 	rcu_read_unlock();
6803 	return found;
6804 }
6805 
6806 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6807 			      struct extent_state **cached_state, int writing)
6808 {
6809 	struct btrfs_ordered_extent *ordered;
6810 	int ret = 0;
6811 
6812 	while (1) {
6813 		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6814 				 0, cached_state);
6815 		/*
6816 		 * We're concerned with the entire range that we're going to be
6817 		 * doing DIO to, so we need to make sure theres no ordered
6818 		 * extents in this range.
6819 		 */
6820 		ordered = btrfs_lookup_ordered_range(inode, lockstart,
6821 						     lockend - lockstart + 1);
6822 
6823 		/*
6824 		 * We need to make sure there are no buffered pages in this
6825 		 * range either, we could have raced between the invalidate in
6826 		 * generic_file_direct_write and locking the extent.  The
6827 		 * invalidate needs to happen so that reads after a write do not
6828 		 * get stale data.
6829 		 */
6830 		if (!ordered &&
6831 		    (!writing ||
6832 		     !btrfs_page_exists_in_range(inode, lockstart, lockend)))
6833 			break;
6834 
6835 		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6836 				     cached_state, GFP_NOFS);
6837 
6838 		if (ordered) {
6839 			btrfs_start_ordered_extent(inode, ordered, 1);
6840 			btrfs_put_ordered_extent(ordered);
6841 		} else {
6842 			/* Screw you mmap */
6843 			ret = filemap_write_and_wait_range(inode->i_mapping,
6844 							   lockstart,
6845 							   lockend);
6846 			if (ret)
6847 				break;
6848 
6849 			/*
6850 			 * If we found a page that couldn't be invalidated just
6851 			 * fall back to buffered.
6852 			 */
6853 			ret = invalidate_inode_pages2_range(inode->i_mapping,
6854 					lockstart >> PAGE_CACHE_SHIFT,
6855 					lockend >> PAGE_CACHE_SHIFT);
6856 			if (ret)
6857 				break;
6858 		}
6859 
6860 		cond_resched();
6861 	}
6862 
6863 	return ret;
6864 }
6865 
6866 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6867 					   u64 len, u64 orig_start,
6868 					   u64 block_start, u64 block_len,
6869 					   u64 orig_block_len, u64 ram_bytes,
6870 					   int type)
6871 {
6872 	struct extent_map_tree *em_tree;
6873 	struct extent_map *em;
6874 	struct btrfs_root *root = BTRFS_I(inode)->root;
6875 	int ret;
6876 
6877 	em_tree = &BTRFS_I(inode)->extent_tree;
6878 	em = alloc_extent_map();
6879 	if (!em)
6880 		return ERR_PTR(-ENOMEM);
6881 
6882 	em->start = start;
6883 	em->orig_start = orig_start;
6884 	em->mod_start = start;
6885 	em->mod_len = len;
6886 	em->len = len;
6887 	em->block_len = block_len;
6888 	em->block_start = block_start;
6889 	em->bdev = root->fs_info->fs_devices->latest_bdev;
6890 	em->orig_block_len = orig_block_len;
6891 	em->ram_bytes = ram_bytes;
6892 	em->generation = -1;
6893 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
6894 	if (type == BTRFS_ORDERED_PREALLOC)
6895 		set_bit(EXTENT_FLAG_FILLING, &em->flags);
6896 
6897 	do {
6898 		btrfs_drop_extent_cache(inode, em->start,
6899 				em->start + em->len - 1, 0);
6900 		write_lock(&em_tree->lock);
6901 		ret = add_extent_mapping(em_tree, em, 1);
6902 		write_unlock(&em_tree->lock);
6903 	} while (ret == -EEXIST);
6904 
6905 	if (ret) {
6906 		free_extent_map(em);
6907 		return ERR_PTR(ret);
6908 	}
6909 
6910 	return em;
6911 }
6912 
6913 
6914 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6915 				   struct buffer_head *bh_result, int create)
6916 {
6917 	struct extent_map *em;
6918 	struct btrfs_root *root = BTRFS_I(inode)->root;
6919 	struct extent_state *cached_state = NULL;
6920 	u64 start = iblock << inode->i_blkbits;
6921 	u64 lockstart, lockend;
6922 	u64 len = bh_result->b_size;
6923 	int unlock_bits = EXTENT_LOCKED;
6924 	int ret = 0;
6925 
6926 	if (create)
6927 		unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6928 	else
6929 		len = min_t(u64, len, root->sectorsize);
6930 
6931 	lockstart = start;
6932 	lockend = start + len - 1;
6933 
6934 	/*
6935 	 * If this errors out it's because we couldn't invalidate pagecache for
6936 	 * this range and we need to fallback to buffered.
6937 	 */
6938 	if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6939 		return -ENOTBLK;
6940 
6941 	em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6942 	if (IS_ERR(em)) {
6943 		ret = PTR_ERR(em);
6944 		goto unlock_err;
6945 	}
6946 
6947 	/*
6948 	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6949 	 * io.  INLINE is special, and we could probably kludge it in here, but
6950 	 * it's still buffered so for safety lets just fall back to the generic
6951 	 * buffered path.
6952 	 *
6953 	 * For COMPRESSED we _have_ to read the entire extent in so we can
6954 	 * decompress it, so there will be buffering required no matter what we
6955 	 * do, so go ahead and fallback to buffered.
6956 	 *
6957 	 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6958 	 * to buffered IO.  Don't blame me, this is the price we pay for using
6959 	 * the generic code.
6960 	 */
6961 	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6962 	    em->block_start == EXTENT_MAP_INLINE) {
6963 		free_extent_map(em);
6964 		ret = -ENOTBLK;
6965 		goto unlock_err;
6966 	}
6967 
6968 	/* Just a good old fashioned hole, return */
6969 	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6970 			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6971 		free_extent_map(em);
6972 		goto unlock_err;
6973 	}
6974 
6975 	/*
6976 	 * We don't allocate a new extent in the following cases
6977 	 *
6978 	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6979 	 * existing extent.
6980 	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
6981 	 * just use the extent.
6982 	 *
6983 	 */
6984 	if (!create) {
6985 		len = min(len, em->len - (start - em->start));
6986 		lockstart = start + len;
6987 		goto unlock;
6988 	}
6989 
6990 	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6991 	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6992 	     em->block_start != EXTENT_MAP_HOLE)) {
6993 		int type;
6994 		int ret;
6995 		u64 block_start, orig_start, orig_block_len, ram_bytes;
6996 
6997 		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6998 			type = BTRFS_ORDERED_PREALLOC;
6999 		else
7000 			type = BTRFS_ORDERED_NOCOW;
7001 		len = min(len, em->len - (start - em->start));
7002 		block_start = em->block_start + (start - em->start);
7003 
7004 		if (can_nocow_extent(inode, start, &len, &orig_start,
7005 				     &orig_block_len, &ram_bytes) == 1) {
7006 			if (type == BTRFS_ORDERED_PREALLOC) {
7007 				free_extent_map(em);
7008 				em = create_pinned_em(inode, start, len,
7009 						       orig_start,
7010 						       block_start, len,
7011 						       orig_block_len,
7012 						       ram_bytes, type);
7013 				if (IS_ERR(em))
7014 					goto unlock_err;
7015 			}
7016 
7017 			ret = btrfs_add_ordered_extent_dio(inode, start,
7018 					   block_start, len, len, type);
7019 			if (ret) {
7020 				free_extent_map(em);
7021 				goto unlock_err;
7022 			}
7023 			goto unlock;
7024 		}
7025 	}
7026 
7027 	/*
7028 	 * this will cow the extent, reset the len in case we changed
7029 	 * it above
7030 	 */
7031 	len = bh_result->b_size;
7032 	free_extent_map(em);
7033 	em = btrfs_new_extent_direct(inode, start, len);
7034 	if (IS_ERR(em)) {
7035 		ret = PTR_ERR(em);
7036 		goto unlock_err;
7037 	}
7038 	len = min(len, em->len - (start - em->start));
7039 unlock:
7040 	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7041 		inode->i_blkbits;
7042 	bh_result->b_size = len;
7043 	bh_result->b_bdev = em->bdev;
7044 	set_buffer_mapped(bh_result);
7045 	if (create) {
7046 		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7047 			set_buffer_new(bh_result);
7048 
7049 		/*
7050 		 * Need to update the i_size under the extent lock so buffered
7051 		 * readers will get the updated i_size when we unlock.
7052 		 */
7053 		if (start + len > i_size_read(inode))
7054 			i_size_write(inode, start + len);
7055 
7056 		spin_lock(&BTRFS_I(inode)->lock);
7057 		BTRFS_I(inode)->outstanding_extents++;
7058 		spin_unlock(&BTRFS_I(inode)->lock);
7059 
7060 		ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7061 				     lockstart + len - 1, EXTENT_DELALLOC, NULL,
7062 				     &cached_state, GFP_NOFS);
7063 		BUG_ON(ret);
7064 	}
7065 
7066 	/*
7067 	 * In the case of write we need to clear and unlock the entire range,
7068 	 * in the case of read we need to unlock only the end area that we
7069 	 * aren't using if there is any left over space.
7070 	 */
7071 	if (lockstart < lockend) {
7072 		clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7073 				 lockend, unlock_bits, 1, 0,
7074 				 &cached_state, GFP_NOFS);
7075 	} else {
7076 		free_extent_state(cached_state);
7077 	}
7078 
7079 	free_extent_map(em);
7080 
7081 	return 0;
7082 
7083 unlock_err:
7084 	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7085 			 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7086 	return ret;
7087 }
7088 
7089 static void btrfs_endio_direct_read(struct bio *bio, int err)
7090 {
7091 	struct btrfs_dio_private *dip = bio->bi_private;
7092 	struct bio_vec *bvec;
7093 	struct inode *inode = dip->inode;
7094 	struct btrfs_root *root = BTRFS_I(inode)->root;
7095 	struct bio *dio_bio;
7096 	u32 *csums = (u32 *)dip->csum;
7097 	u64 start;
7098 	int i;
7099 
7100 	start = dip->logical_offset;
7101 	bio_for_each_segment_all(bvec, bio, i) {
7102 		if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7103 			struct page *page = bvec->bv_page;
7104 			char *kaddr;
7105 			u32 csum = ~(u32)0;
7106 			unsigned long flags;
7107 
7108 			local_irq_save(flags);
7109 			kaddr = kmap_atomic(page);
7110 			csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7111 					       csum, bvec->bv_len);
7112 			btrfs_csum_final(csum, (char *)&csum);
7113 			kunmap_atomic(kaddr);
7114 			local_irq_restore(flags);
7115 
7116 			flush_dcache_page(bvec->bv_page);
7117 			if (csum != csums[i]) {
7118 				btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7119 					  btrfs_ino(inode), start, csum,
7120 					  csums[i]);
7121 				err = -EIO;
7122 			}
7123 		}
7124 
7125 		start += bvec->bv_len;
7126 	}
7127 
7128 	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7129 		      dip->logical_offset + dip->bytes - 1);
7130 	dio_bio = dip->dio_bio;
7131 
7132 	kfree(dip);
7133 
7134 	/* If we had a csum failure make sure to clear the uptodate flag */
7135 	if (err)
7136 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7137 	dio_end_io(dio_bio, err);
7138 	bio_put(bio);
7139 }
7140 
7141 static void btrfs_endio_direct_write(struct bio *bio, int err)
7142 {
7143 	struct btrfs_dio_private *dip = bio->bi_private;
7144 	struct inode *inode = dip->inode;
7145 	struct btrfs_root *root = BTRFS_I(inode)->root;
7146 	struct btrfs_ordered_extent *ordered = NULL;
7147 	u64 ordered_offset = dip->logical_offset;
7148 	u64 ordered_bytes = dip->bytes;
7149 	struct bio *dio_bio;
7150 	int ret;
7151 
7152 	if (err)
7153 		goto out_done;
7154 again:
7155 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7156 						   &ordered_offset,
7157 						   ordered_bytes, !err);
7158 	if (!ret)
7159 		goto out_test;
7160 
7161 	btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7162 	btrfs_queue_work(root->fs_info->endio_write_workers,
7163 			 &ordered->work);
7164 out_test:
7165 	/*
7166 	 * our bio might span multiple ordered extents.  If we haven't
7167 	 * completed the accounting for the whole dio, go back and try again
7168 	 */
7169 	if (ordered_offset < dip->logical_offset + dip->bytes) {
7170 		ordered_bytes = dip->logical_offset + dip->bytes -
7171 			ordered_offset;
7172 		ordered = NULL;
7173 		goto again;
7174 	}
7175 out_done:
7176 	dio_bio = dip->dio_bio;
7177 
7178 	kfree(dip);
7179 
7180 	/* If we had an error make sure to clear the uptodate flag */
7181 	if (err)
7182 		clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7183 	dio_end_io(dio_bio, err);
7184 	bio_put(bio);
7185 }
7186 
7187 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7188 				    struct bio *bio, int mirror_num,
7189 				    unsigned long bio_flags, u64 offset)
7190 {
7191 	int ret;
7192 	struct btrfs_root *root = BTRFS_I(inode)->root;
7193 	ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7194 	BUG_ON(ret); /* -ENOMEM */
7195 	return 0;
7196 }
7197 
7198 static void btrfs_end_dio_bio(struct bio *bio, int err)
7199 {
7200 	struct btrfs_dio_private *dip = bio->bi_private;
7201 
7202 	if (err) {
7203 		btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7204 			  "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7205 		      btrfs_ino(dip->inode), bio->bi_rw,
7206 		      (unsigned long long)bio->bi_iter.bi_sector,
7207 		      bio->bi_iter.bi_size, err);
7208 		dip->errors = 1;
7209 
7210 		/*
7211 		 * before atomic variable goto zero, we must make sure
7212 		 * dip->errors is perceived to be set.
7213 		 */
7214 		smp_mb__before_atomic();
7215 	}
7216 
7217 	/* if there are more bios still pending for this dio, just exit */
7218 	if (!atomic_dec_and_test(&dip->pending_bios))
7219 		goto out;
7220 
7221 	if (dip->errors) {
7222 		bio_io_error(dip->orig_bio);
7223 	} else {
7224 		set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7225 		bio_endio(dip->orig_bio, 0);
7226 	}
7227 out:
7228 	bio_put(bio);
7229 }
7230 
7231 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7232 				       u64 first_sector, gfp_t gfp_flags)
7233 {
7234 	int nr_vecs = bio_get_nr_vecs(bdev);
7235 	return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7236 }
7237 
7238 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7239 					 int rw, u64 file_offset, int skip_sum,
7240 					 int async_submit)
7241 {
7242 	struct btrfs_dio_private *dip = bio->bi_private;
7243 	int write = rw & REQ_WRITE;
7244 	struct btrfs_root *root = BTRFS_I(inode)->root;
7245 	int ret;
7246 
7247 	if (async_submit)
7248 		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7249 
7250 	bio_get(bio);
7251 
7252 	if (!write) {
7253 		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7254 		if (ret)
7255 			goto err;
7256 	}
7257 
7258 	if (skip_sum)
7259 		goto map;
7260 
7261 	if (write && async_submit) {
7262 		ret = btrfs_wq_submit_bio(root->fs_info,
7263 				   inode, rw, bio, 0, 0,
7264 				   file_offset,
7265 				   __btrfs_submit_bio_start_direct_io,
7266 				   __btrfs_submit_bio_done);
7267 		goto err;
7268 	} else if (write) {
7269 		/*
7270 		 * If we aren't doing async submit, calculate the csum of the
7271 		 * bio now.
7272 		 */
7273 		ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7274 		if (ret)
7275 			goto err;
7276 	} else if (!skip_sum) {
7277 		ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7278 						file_offset);
7279 		if (ret)
7280 			goto err;
7281 	}
7282 
7283 map:
7284 	ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7285 err:
7286 	bio_put(bio);
7287 	return ret;
7288 }
7289 
7290 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7291 				    int skip_sum)
7292 {
7293 	struct inode *inode = dip->inode;
7294 	struct btrfs_root *root = BTRFS_I(inode)->root;
7295 	struct bio *bio;
7296 	struct bio *orig_bio = dip->orig_bio;
7297 	struct bio_vec *bvec = orig_bio->bi_io_vec;
7298 	u64 start_sector = orig_bio->bi_iter.bi_sector;
7299 	u64 file_offset = dip->logical_offset;
7300 	u64 submit_len = 0;
7301 	u64 map_length;
7302 	int nr_pages = 0;
7303 	int ret = 0;
7304 	int async_submit = 0;
7305 
7306 	map_length = orig_bio->bi_iter.bi_size;
7307 	ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7308 			      &map_length, NULL, 0);
7309 	if (ret) {
7310 		bio_put(orig_bio);
7311 		return -EIO;
7312 	}
7313 
7314 	if (map_length >= orig_bio->bi_iter.bi_size) {
7315 		bio = orig_bio;
7316 		goto submit;
7317 	}
7318 
7319 	/* async crcs make it difficult to collect full stripe writes. */
7320 	if (btrfs_get_alloc_profile(root, 1) &
7321 	    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7322 		async_submit = 0;
7323 	else
7324 		async_submit = 1;
7325 
7326 	bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7327 	if (!bio)
7328 		return -ENOMEM;
7329 	bio->bi_private = dip;
7330 	bio->bi_end_io = btrfs_end_dio_bio;
7331 	atomic_inc(&dip->pending_bios);
7332 
7333 	while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7334 		if (unlikely(map_length < submit_len + bvec->bv_len ||
7335 		    bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7336 				 bvec->bv_offset) < bvec->bv_len)) {
7337 			/*
7338 			 * inc the count before we submit the bio so
7339 			 * we know the end IO handler won't happen before
7340 			 * we inc the count. Otherwise, the dip might get freed
7341 			 * before we're done setting it up
7342 			 */
7343 			atomic_inc(&dip->pending_bios);
7344 			ret = __btrfs_submit_dio_bio(bio, inode, rw,
7345 						     file_offset, skip_sum,
7346 						     async_submit);
7347 			if (ret) {
7348 				bio_put(bio);
7349 				atomic_dec(&dip->pending_bios);
7350 				goto out_err;
7351 			}
7352 
7353 			start_sector += submit_len >> 9;
7354 			file_offset += submit_len;
7355 
7356 			submit_len = 0;
7357 			nr_pages = 0;
7358 
7359 			bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7360 						  start_sector, GFP_NOFS);
7361 			if (!bio)
7362 				goto out_err;
7363 			bio->bi_private = dip;
7364 			bio->bi_end_io = btrfs_end_dio_bio;
7365 
7366 			map_length = orig_bio->bi_iter.bi_size;
7367 			ret = btrfs_map_block(root->fs_info, rw,
7368 					      start_sector << 9,
7369 					      &map_length, NULL, 0);
7370 			if (ret) {
7371 				bio_put(bio);
7372 				goto out_err;
7373 			}
7374 		} else {
7375 			submit_len += bvec->bv_len;
7376 			nr_pages++;
7377 			bvec++;
7378 		}
7379 	}
7380 
7381 submit:
7382 	ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7383 				     async_submit);
7384 	if (!ret)
7385 		return 0;
7386 
7387 	bio_put(bio);
7388 out_err:
7389 	dip->errors = 1;
7390 	/*
7391 	 * before atomic variable goto zero, we must
7392 	 * make sure dip->errors is perceived to be set.
7393 	 */
7394 	smp_mb__before_atomic();
7395 	if (atomic_dec_and_test(&dip->pending_bios))
7396 		bio_io_error(dip->orig_bio);
7397 
7398 	/* bio_end_io() will handle error, so we needn't return it */
7399 	return 0;
7400 }
7401 
7402 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7403 				struct inode *inode, loff_t file_offset)
7404 {
7405 	struct btrfs_root *root = BTRFS_I(inode)->root;
7406 	struct btrfs_dio_private *dip;
7407 	struct bio *io_bio;
7408 	int skip_sum;
7409 	int sum_len;
7410 	int write = rw & REQ_WRITE;
7411 	int ret = 0;
7412 	u16 csum_size;
7413 
7414 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7415 
7416 	io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7417 	if (!io_bio) {
7418 		ret = -ENOMEM;
7419 		goto free_ordered;
7420 	}
7421 
7422 	if (!skip_sum && !write) {
7423 		csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7424 		sum_len = dio_bio->bi_iter.bi_size >>
7425 			inode->i_sb->s_blocksize_bits;
7426 		sum_len *= csum_size;
7427 	} else {
7428 		sum_len = 0;
7429 	}
7430 
7431 	dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7432 	if (!dip) {
7433 		ret = -ENOMEM;
7434 		goto free_io_bio;
7435 	}
7436 
7437 	dip->private = dio_bio->bi_private;
7438 	dip->inode = inode;
7439 	dip->logical_offset = file_offset;
7440 	dip->bytes = dio_bio->bi_iter.bi_size;
7441 	dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7442 	io_bio->bi_private = dip;
7443 	dip->errors = 0;
7444 	dip->orig_bio = io_bio;
7445 	dip->dio_bio = dio_bio;
7446 	atomic_set(&dip->pending_bios, 0);
7447 
7448 	if (write)
7449 		io_bio->bi_end_io = btrfs_endio_direct_write;
7450 	else
7451 		io_bio->bi_end_io = btrfs_endio_direct_read;
7452 
7453 	ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7454 	if (!ret)
7455 		return;
7456 
7457 free_io_bio:
7458 	bio_put(io_bio);
7459 
7460 free_ordered:
7461 	/*
7462 	 * If this is a write, we need to clean up the reserved space and kill
7463 	 * the ordered extent.
7464 	 */
7465 	if (write) {
7466 		struct btrfs_ordered_extent *ordered;
7467 		ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7468 		if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7469 		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7470 			btrfs_free_reserved_extent(root, ordered->start,
7471 						   ordered->disk_len, 1);
7472 		btrfs_put_ordered_extent(ordered);
7473 		btrfs_put_ordered_extent(ordered);
7474 	}
7475 	bio_endio(dio_bio, ret);
7476 }
7477 
7478 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7479 			const struct iov_iter *iter, loff_t offset)
7480 {
7481 	int seg;
7482 	int i;
7483 	unsigned blocksize_mask = root->sectorsize - 1;
7484 	ssize_t retval = -EINVAL;
7485 
7486 	if (offset & blocksize_mask)
7487 		goto out;
7488 
7489 	if (iov_iter_alignment(iter) & blocksize_mask)
7490 		goto out;
7491 
7492 	/* If this is a write we don't need to check anymore */
7493 	if (rw & WRITE)
7494 		return 0;
7495 	/*
7496 	 * Check to make sure we don't have duplicate iov_base's in this
7497 	 * iovec, if so return EINVAL, otherwise we'll get csum errors
7498 	 * when reading back.
7499 	 */
7500 	for (seg = 0; seg < iter->nr_segs; seg++) {
7501 		for (i = seg + 1; i < iter->nr_segs; i++) {
7502 			if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7503 				goto out;
7504 		}
7505 	}
7506 	retval = 0;
7507 out:
7508 	return retval;
7509 }
7510 
7511 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7512 			struct iov_iter *iter, loff_t offset)
7513 {
7514 	struct file *file = iocb->ki_filp;
7515 	struct inode *inode = file->f_mapping->host;
7516 	size_t count = 0;
7517 	int flags = 0;
7518 	bool wakeup = true;
7519 	bool relock = false;
7520 	ssize_t ret;
7521 
7522 	if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
7523 		return 0;
7524 
7525 	atomic_inc(&inode->i_dio_count);
7526 	smp_mb__after_atomic();
7527 
7528 	/*
7529 	 * The generic stuff only does filemap_write_and_wait_range, which
7530 	 * isn't enough if we've written compressed pages to this area, so
7531 	 * we need to flush the dirty pages again to make absolutely sure
7532 	 * that any outstanding dirty pages are on disk.
7533 	 */
7534 	count = iov_iter_count(iter);
7535 	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7536 		     &BTRFS_I(inode)->runtime_flags))
7537 		filemap_fdatawrite_range(inode->i_mapping, offset, count);
7538 
7539 	if (rw & WRITE) {
7540 		/*
7541 		 * If the write DIO is beyond the EOF, we need update
7542 		 * the isize, but it is protected by i_mutex. So we can
7543 		 * not unlock the i_mutex at this case.
7544 		 */
7545 		if (offset + count <= inode->i_size) {
7546 			mutex_unlock(&inode->i_mutex);
7547 			relock = true;
7548 		}
7549 		ret = btrfs_delalloc_reserve_space(inode, count);
7550 		if (ret)
7551 			goto out;
7552 	} else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7553 				     &BTRFS_I(inode)->runtime_flags))) {
7554 		inode_dio_done(inode);
7555 		flags = DIO_LOCKING | DIO_SKIP_HOLES;
7556 		wakeup = false;
7557 	}
7558 
7559 	ret = __blockdev_direct_IO(rw, iocb, inode,
7560 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7561 			iter, offset, btrfs_get_blocks_direct, NULL,
7562 			btrfs_submit_direct, flags);
7563 	if (rw & WRITE) {
7564 		if (ret < 0 && ret != -EIOCBQUEUED)
7565 			btrfs_delalloc_release_space(inode, count);
7566 		else if (ret >= 0 && (size_t)ret < count)
7567 			btrfs_delalloc_release_space(inode,
7568 						     count - (size_t)ret);
7569 		else
7570 			btrfs_delalloc_release_metadata(inode, 0);
7571 	}
7572 out:
7573 	if (wakeup)
7574 		inode_dio_done(inode);
7575 	if (relock)
7576 		mutex_lock(&inode->i_mutex);
7577 
7578 	return ret;
7579 }
7580 
7581 #define BTRFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
7582 
7583 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7584 		__u64 start, __u64 len)
7585 {
7586 	int	ret;
7587 
7588 	ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7589 	if (ret)
7590 		return ret;
7591 
7592 	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7593 }
7594 
7595 int btrfs_readpage(struct file *file, struct page *page)
7596 {
7597 	struct extent_io_tree *tree;
7598 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7599 	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7600 }
7601 
7602 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7603 {
7604 	struct extent_io_tree *tree;
7605 
7606 
7607 	if (current->flags & PF_MEMALLOC) {
7608 		redirty_page_for_writepage(wbc, page);
7609 		unlock_page(page);
7610 		return 0;
7611 	}
7612 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7613 	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7614 }
7615 
7616 static int btrfs_writepages(struct address_space *mapping,
7617 			    struct writeback_control *wbc)
7618 {
7619 	struct extent_io_tree *tree;
7620 
7621 	tree = &BTRFS_I(mapping->host)->io_tree;
7622 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7623 }
7624 
7625 static int
7626 btrfs_readpages(struct file *file, struct address_space *mapping,
7627 		struct list_head *pages, unsigned nr_pages)
7628 {
7629 	struct extent_io_tree *tree;
7630 	tree = &BTRFS_I(mapping->host)->io_tree;
7631 	return extent_readpages(tree, mapping, pages, nr_pages,
7632 				btrfs_get_extent);
7633 }
7634 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7635 {
7636 	struct extent_io_tree *tree;
7637 	struct extent_map_tree *map;
7638 	int ret;
7639 
7640 	tree = &BTRFS_I(page->mapping->host)->io_tree;
7641 	map = &BTRFS_I(page->mapping->host)->extent_tree;
7642 	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7643 	if (ret == 1) {
7644 		ClearPagePrivate(page);
7645 		set_page_private(page, 0);
7646 		page_cache_release(page);
7647 	}
7648 	return ret;
7649 }
7650 
7651 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7652 {
7653 	if (PageWriteback(page) || PageDirty(page))
7654 		return 0;
7655 	return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7656 }
7657 
7658 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7659 				 unsigned int length)
7660 {
7661 	struct inode *inode = page->mapping->host;
7662 	struct extent_io_tree *tree;
7663 	struct btrfs_ordered_extent *ordered;
7664 	struct extent_state *cached_state = NULL;
7665 	u64 page_start = page_offset(page);
7666 	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7667 	int inode_evicting = inode->i_state & I_FREEING;
7668 
7669 	/*
7670 	 * we have the page locked, so new writeback can't start,
7671 	 * and the dirty bit won't be cleared while we are here.
7672 	 *
7673 	 * Wait for IO on this page so that we can safely clear
7674 	 * the PagePrivate2 bit and do ordered accounting
7675 	 */
7676 	wait_on_page_writeback(page);
7677 
7678 	tree = &BTRFS_I(inode)->io_tree;
7679 	if (offset) {
7680 		btrfs_releasepage(page, GFP_NOFS);
7681 		return;
7682 	}
7683 
7684 	if (!inode_evicting)
7685 		lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7686 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7687 	if (ordered) {
7688 		/*
7689 		 * IO on this page will never be started, so we need
7690 		 * to account for any ordered extents now
7691 		 */
7692 		if (!inode_evicting)
7693 			clear_extent_bit(tree, page_start, page_end,
7694 					 EXTENT_DIRTY | EXTENT_DELALLOC |
7695 					 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7696 					 EXTENT_DEFRAG, 1, 0, &cached_state,
7697 					 GFP_NOFS);
7698 		/*
7699 		 * whoever cleared the private bit is responsible
7700 		 * for the finish_ordered_io
7701 		 */
7702 		if (TestClearPagePrivate2(page)) {
7703 			struct btrfs_ordered_inode_tree *tree;
7704 			u64 new_len;
7705 
7706 			tree = &BTRFS_I(inode)->ordered_tree;
7707 
7708 			spin_lock_irq(&tree->lock);
7709 			set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7710 			new_len = page_start - ordered->file_offset;
7711 			if (new_len < ordered->truncated_len)
7712 				ordered->truncated_len = new_len;
7713 			spin_unlock_irq(&tree->lock);
7714 
7715 			if (btrfs_dec_test_ordered_pending(inode, &ordered,
7716 							   page_start,
7717 							   PAGE_CACHE_SIZE, 1))
7718 				btrfs_finish_ordered_io(ordered);
7719 		}
7720 		btrfs_put_ordered_extent(ordered);
7721 		if (!inode_evicting) {
7722 			cached_state = NULL;
7723 			lock_extent_bits(tree, page_start, page_end, 0,
7724 					 &cached_state);
7725 		}
7726 	}
7727 
7728 	if (!inode_evicting) {
7729 		clear_extent_bit(tree, page_start, page_end,
7730 				 EXTENT_LOCKED | EXTENT_DIRTY |
7731 				 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7732 				 EXTENT_DEFRAG, 1, 1,
7733 				 &cached_state, GFP_NOFS);
7734 
7735 		__btrfs_releasepage(page, GFP_NOFS);
7736 	}
7737 
7738 	ClearPageChecked(page);
7739 	if (PagePrivate(page)) {
7740 		ClearPagePrivate(page);
7741 		set_page_private(page, 0);
7742 		page_cache_release(page);
7743 	}
7744 }
7745 
7746 /*
7747  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7748  * called from a page fault handler when a page is first dirtied. Hence we must
7749  * be careful to check for EOF conditions here. We set the page up correctly
7750  * for a written page which means we get ENOSPC checking when writing into
7751  * holes and correct delalloc and unwritten extent mapping on filesystems that
7752  * support these features.
7753  *
7754  * We are not allowed to take the i_mutex here so we have to play games to
7755  * protect against truncate races as the page could now be beyond EOF.  Because
7756  * vmtruncate() writes the inode size before removing pages, once we have the
7757  * page lock we can determine safely if the page is beyond EOF. If it is not
7758  * beyond EOF, then the page is guaranteed safe against truncation until we
7759  * unlock the page.
7760  */
7761 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7762 {
7763 	struct page *page = vmf->page;
7764 	struct inode *inode = file_inode(vma->vm_file);
7765 	struct btrfs_root *root = BTRFS_I(inode)->root;
7766 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7767 	struct btrfs_ordered_extent *ordered;
7768 	struct extent_state *cached_state = NULL;
7769 	char *kaddr;
7770 	unsigned long zero_start;
7771 	loff_t size;
7772 	int ret;
7773 	int reserved = 0;
7774 	u64 page_start;
7775 	u64 page_end;
7776 
7777 	sb_start_pagefault(inode->i_sb);
7778 	ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7779 	if (!ret) {
7780 		ret = file_update_time(vma->vm_file);
7781 		reserved = 1;
7782 	}
7783 	if (ret) {
7784 		if (ret == -ENOMEM)
7785 			ret = VM_FAULT_OOM;
7786 		else /* -ENOSPC, -EIO, etc */
7787 			ret = VM_FAULT_SIGBUS;
7788 		if (reserved)
7789 			goto out;
7790 		goto out_noreserve;
7791 	}
7792 
7793 	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7794 again:
7795 	lock_page(page);
7796 	size = i_size_read(inode);
7797 	page_start = page_offset(page);
7798 	page_end = page_start + PAGE_CACHE_SIZE - 1;
7799 
7800 	if ((page->mapping != inode->i_mapping) ||
7801 	    (page_start >= size)) {
7802 		/* page got truncated out from underneath us */
7803 		goto out_unlock;
7804 	}
7805 	wait_on_page_writeback(page);
7806 
7807 	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7808 	set_page_extent_mapped(page);
7809 
7810 	/*
7811 	 * we can't set the delalloc bits if there are pending ordered
7812 	 * extents.  Drop our locks and wait for them to finish
7813 	 */
7814 	ordered = btrfs_lookup_ordered_extent(inode, page_start);
7815 	if (ordered) {
7816 		unlock_extent_cached(io_tree, page_start, page_end,
7817 				     &cached_state, GFP_NOFS);
7818 		unlock_page(page);
7819 		btrfs_start_ordered_extent(inode, ordered, 1);
7820 		btrfs_put_ordered_extent(ordered);
7821 		goto again;
7822 	}
7823 
7824 	/*
7825 	 * XXX - page_mkwrite gets called every time the page is dirtied, even
7826 	 * if it was already dirty, so for space accounting reasons we need to
7827 	 * clear any delalloc bits for the range we are fixing to save.  There
7828 	 * is probably a better way to do this, but for now keep consistent with
7829 	 * prepare_pages in the normal write path.
7830 	 */
7831 	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7832 			  EXTENT_DIRTY | EXTENT_DELALLOC |
7833 			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7834 			  0, 0, &cached_state, GFP_NOFS);
7835 
7836 	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7837 					&cached_state);
7838 	if (ret) {
7839 		unlock_extent_cached(io_tree, page_start, page_end,
7840 				     &cached_state, GFP_NOFS);
7841 		ret = VM_FAULT_SIGBUS;
7842 		goto out_unlock;
7843 	}
7844 	ret = 0;
7845 
7846 	/* page is wholly or partially inside EOF */
7847 	if (page_start + PAGE_CACHE_SIZE > size)
7848 		zero_start = size & ~PAGE_CACHE_MASK;
7849 	else
7850 		zero_start = PAGE_CACHE_SIZE;
7851 
7852 	if (zero_start != PAGE_CACHE_SIZE) {
7853 		kaddr = kmap(page);
7854 		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7855 		flush_dcache_page(page);
7856 		kunmap(page);
7857 	}
7858 	ClearPageChecked(page);
7859 	set_page_dirty(page);
7860 	SetPageUptodate(page);
7861 
7862 	BTRFS_I(inode)->last_trans = root->fs_info->generation;
7863 	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7864 	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7865 
7866 	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7867 
7868 out_unlock:
7869 	if (!ret) {
7870 		sb_end_pagefault(inode->i_sb);
7871 		return VM_FAULT_LOCKED;
7872 	}
7873 	unlock_page(page);
7874 out:
7875 	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7876 out_noreserve:
7877 	sb_end_pagefault(inode->i_sb);
7878 	return ret;
7879 }
7880 
7881 static int btrfs_truncate(struct inode *inode)
7882 {
7883 	struct btrfs_root *root = BTRFS_I(inode)->root;
7884 	struct btrfs_block_rsv *rsv;
7885 	int ret = 0;
7886 	int err = 0;
7887 	struct btrfs_trans_handle *trans;
7888 	u64 mask = root->sectorsize - 1;
7889 	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7890 
7891 	ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7892 				       (u64)-1);
7893 	if (ret)
7894 		return ret;
7895 
7896 	/*
7897 	 * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7898 	 * 3 things going on here
7899 	 *
7900 	 * 1) We need to reserve space for our orphan item and the space to
7901 	 * delete our orphan item.  Lord knows we don't want to have a dangling
7902 	 * orphan item because we didn't reserve space to remove it.
7903 	 *
7904 	 * 2) We need to reserve space to update our inode.
7905 	 *
7906 	 * 3) We need to have something to cache all the space that is going to
7907 	 * be free'd up by the truncate operation, but also have some slack
7908 	 * space reserved in case it uses space during the truncate (thank you
7909 	 * very much snapshotting).
7910 	 *
7911 	 * And we need these to all be seperate.  The fact is we can use alot of
7912 	 * space doing the truncate, and we have no earthly idea how much space
7913 	 * we will use, so we need the truncate reservation to be seperate so it
7914 	 * doesn't end up using space reserved for updating the inode or
7915 	 * removing the orphan item.  We also need to be able to stop the
7916 	 * transaction and start a new one, which means we need to be able to
7917 	 * update the inode several times, and we have no idea of knowing how
7918 	 * many times that will be, so we can't just reserve 1 item for the
7919 	 * entirety of the opration, so that has to be done seperately as well.
7920 	 * Then there is the orphan item, which does indeed need to be held on
7921 	 * to for the whole operation, and we need nobody to touch this reserved
7922 	 * space except the orphan code.
7923 	 *
7924 	 * So that leaves us with
7925 	 *
7926 	 * 1) root->orphan_block_rsv - for the orphan deletion.
7927 	 * 2) rsv - for the truncate reservation, which we will steal from the
7928 	 * transaction reservation.
7929 	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7930 	 * updating the inode.
7931 	 */
7932 	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7933 	if (!rsv)
7934 		return -ENOMEM;
7935 	rsv->size = min_size;
7936 	rsv->failfast = 1;
7937 
7938 	/*
7939 	 * 1 for the truncate slack space
7940 	 * 1 for updating the inode.
7941 	 */
7942 	trans = btrfs_start_transaction(root, 2);
7943 	if (IS_ERR(trans)) {
7944 		err = PTR_ERR(trans);
7945 		goto out;
7946 	}
7947 
7948 	/* Migrate the slack space for the truncate to our reserve */
7949 	ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7950 				      min_size);
7951 	BUG_ON(ret);
7952 
7953 	/*
7954 	 * So if we truncate and then write and fsync we normally would just
7955 	 * write the extents that changed, which is a problem if we need to
7956 	 * first truncate that entire inode.  So set this flag so we write out
7957 	 * all of the extents in the inode to the sync log so we're completely
7958 	 * safe.
7959 	 */
7960 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7961 	trans->block_rsv = rsv;
7962 
7963 	while (1) {
7964 		ret = btrfs_truncate_inode_items(trans, root, inode,
7965 						 inode->i_size,
7966 						 BTRFS_EXTENT_DATA_KEY);
7967 		if (ret != -ENOSPC) {
7968 			err = ret;
7969 			break;
7970 		}
7971 
7972 		trans->block_rsv = &root->fs_info->trans_block_rsv;
7973 		ret = btrfs_update_inode(trans, root, inode);
7974 		if (ret) {
7975 			err = ret;
7976 			break;
7977 		}
7978 
7979 		btrfs_end_transaction(trans, root);
7980 		btrfs_btree_balance_dirty(root);
7981 
7982 		trans = btrfs_start_transaction(root, 2);
7983 		if (IS_ERR(trans)) {
7984 			ret = err = PTR_ERR(trans);
7985 			trans = NULL;
7986 			break;
7987 		}
7988 
7989 		ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7990 					      rsv, min_size);
7991 		BUG_ON(ret);	/* shouldn't happen */
7992 		trans->block_rsv = rsv;
7993 	}
7994 
7995 	if (ret == 0 && inode->i_nlink > 0) {
7996 		trans->block_rsv = root->orphan_block_rsv;
7997 		ret = btrfs_orphan_del(trans, inode);
7998 		if (ret)
7999 			err = ret;
8000 	}
8001 
8002 	if (trans) {
8003 		trans->block_rsv = &root->fs_info->trans_block_rsv;
8004 		ret = btrfs_update_inode(trans, root, inode);
8005 		if (ret && !err)
8006 			err = ret;
8007 
8008 		ret = btrfs_end_transaction(trans, root);
8009 		btrfs_btree_balance_dirty(root);
8010 	}
8011 
8012 out:
8013 	btrfs_free_block_rsv(root, rsv);
8014 
8015 	if (ret && !err)
8016 		err = ret;
8017 
8018 	return err;
8019 }
8020 
8021 /*
8022  * create a new subvolume directory/inode (helper for the ioctl).
8023  */
8024 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8025 			     struct btrfs_root *new_root,
8026 			     struct btrfs_root *parent_root,
8027 			     u64 new_dirid)
8028 {
8029 	struct inode *inode;
8030 	int err;
8031 	u64 index = 0;
8032 
8033 	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8034 				new_dirid, new_dirid,
8035 				S_IFDIR | (~current_umask() & S_IRWXUGO),
8036 				&index);
8037 	if (IS_ERR(inode))
8038 		return PTR_ERR(inode);
8039 	inode->i_op = &btrfs_dir_inode_operations;
8040 	inode->i_fop = &btrfs_dir_file_operations;
8041 
8042 	set_nlink(inode, 1);
8043 	btrfs_i_size_write(inode, 0);
8044 
8045 	err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8046 	if (err)
8047 		btrfs_err(new_root->fs_info,
8048 			  "error inheriting subvolume %llu properties: %d",
8049 			  new_root->root_key.objectid, err);
8050 
8051 	err = btrfs_update_inode(trans, new_root, inode);
8052 
8053 	iput(inode);
8054 	return err;
8055 }
8056 
8057 struct inode *btrfs_alloc_inode(struct super_block *sb)
8058 {
8059 	struct btrfs_inode *ei;
8060 	struct inode *inode;
8061 
8062 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8063 	if (!ei)
8064 		return NULL;
8065 
8066 	ei->root = NULL;
8067 	ei->generation = 0;
8068 	ei->last_trans = 0;
8069 	ei->last_sub_trans = 0;
8070 	ei->logged_trans = 0;
8071 	ei->delalloc_bytes = 0;
8072 	ei->disk_i_size = 0;
8073 	ei->flags = 0;
8074 	ei->csum_bytes = 0;
8075 	ei->index_cnt = (u64)-1;
8076 	ei->dir_index = 0;
8077 	ei->last_unlink_trans = 0;
8078 	ei->last_log_commit = 0;
8079 
8080 	spin_lock_init(&ei->lock);
8081 	ei->outstanding_extents = 0;
8082 	ei->reserved_extents = 0;
8083 
8084 	ei->runtime_flags = 0;
8085 	ei->force_compress = BTRFS_COMPRESS_NONE;
8086 
8087 	ei->delayed_node = NULL;
8088 
8089 	inode = &ei->vfs_inode;
8090 	extent_map_tree_init(&ei->extent_tree);
8091 	extent_io_tree_init(&ei->io_tree, &inode->i_data);
8092 	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8093 	ei->io_tree.track_uptodate = 1;
8094 	ei->io_failure_tree.track_uptodate = 1;
8095 	atomic_set(&ei->sync_writers, 0);
8096 	mutex_init(&ei->log_mutex);
8097 	mutex_init(&ei->delalloc_mutex);
8098 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8099 	INIT_LIST_HEAD(&ei->delalloc_inodes);
8100 	RB_CLEAR_NODE(&ei->rb_node);
8101 
8102 	return inode;
8103 }
8104 
8105 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8106 void btrfs_test_destroy_inode(struct inode *inode)
8107 {
8108 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8109 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8110 }
8111 #endif
8112 
8113 static void btrfs_i_callback(struct rcu_head *head)
8114 {
8115 	struct inode *inode = container_of(head, struct inode, i_rcu);
8116 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8117 }
8118 
8119 void btrfs_destroy_inode(struct inode *inode)
8120 {
8121 	struct btrfs_ordered_extent *ordered;
8122 	struct btrfs_root *root = BTRFS_I(inode)->root;
8123 
8124 	WARN_ON(!hlist_empty(&inode->i_dentry));
8125 	WARN_ON(inode->i_data.nrpages);
8126 	WARN_ON(BTRFS_I(inode)->outstanding_extents);
8127 	WARN_ON(BTRFS_I(inode)->reserved_extents);
8128 	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8129 	WARN_ON(BTRFS_I(inode)->csum_bytes);
8130 
8131 	/*
8132 	 * This can happen where we create an inode, but somebody else also
8133 	 * created the same inode and we need to destroy the one we already
8134 	 * created.
8135 	 */
8136 	if (!root)
8137 		goto free;
8138 
8139 	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8140 		     &BTRFS_I(inode)->runtime_flags)) {
8141 		btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8142 			btrfs_ino(inode));
8143 		atomic_dec(&root->orphan_inodes);
8144 	}
8145 
8146 	while (1) {
8147 		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8148 		if (!ordered)
8149 			break;
8150 		else {
8151 			btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8152 				ordered->file_offset, ordered->len);
8153 			btrfs_remove_ordered_extent(inode, ordered);
8154 			btrfs_put_ordered_extent(ordered);
8155 			btrfs_put_ordered_extent(ordered);
8156 		}
8157 	}
8158 	inode_tree_del(inode);
8159 	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8160 free:
8161 	call_rcu(&inode->i_rcu, btrfs_i_callback);
8162 }
8163 
8164 int btrfs_drop_inode(struct inode *inode)
8165 {
8166 	struct btrfs_root *root = BTRFS_I(inode)->root;
8167 
8168 	if (root == NULL)
8169 		return 1;
8170 
8171 	/* the snap/subvol tree is on deleting */
8172 	if (btrfs_root_refs(&root->root_item) == 0)
8173 		return 1;
8174 	else
8175 		return generic_drop_inode(inode);
8176 }
8177 
8178 static void init_once(void *foo)
8179 {
8180 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8181 
8182 	inode_init_once(&ei->vfs_inode);
8183 }
8184 
8185 void btrfs_destroy_cachep(void)
8186 {
8187 	/*
8188 	 * Make sure all delayed rcu free inodes are flushed before we
8189 	 * destroy cache.
8190 	 */
8191 	rcu_barrier();
8192 	if (btrfs_inode_cachep)
8193 		kmem_cache_destroy(btrfs_inode_cachep);
8194 	if (btrfs_trans_handle_cachep)
8195 		kmem_cache_destroy(btrfs_trans_handle_cachep);
8196 	if (btrfs_transaction_cachep)
8197 		kmem_cache_destroy(btrfs_transaction_cachep);
8198 	if (btrfs_path_cachep)
8199 		kmem_cache_destroy(btrfs_path_cachep);
8200 	if (btrfs_free_space_cachep)
8201 		kmem_cache_destroy(btrfs_free_space_cachep);
8202 	if (btrfs_delalloc_work_cachep)
8203 		kmem_cache_destroy(btrfs_delalloc_work_cachep);
8204 }
8205 
8206 int btrfs_init_cachep(void)
8207 {
8208 	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8209 			sizeof(struct btrfs_inode), 0,
8210 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8211 	if (!btrfs_inode_cachep)
8212 		goto fail;
8213 
8214 	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8215 			sizeof(struct btrfs_trans_handle), 0,
8216 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8217 	if (!btrfs_trans_handle_cachep)
8218 		goto fail;
8219 
8220 	btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8221 			sizeof(struct btrfs_transaction), 0,
8222 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8223 	if (!btrfs_transaction_cachep)
8224 		goto fail;
8225 
8226 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
8227 			sizeof(struct btrfs_path), 0,
8228 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8229 	if (!btrfs_path_cachep)
8230 		goto fail;
8231 
8232 	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8233 			sizeof(struct btrfs_free_space), 0,
8234 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8235 	if (!btrfs_free_space_cachep)
8236 		goto fail;
8237 
8238 	btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8239 			sizeof(struct btrfs_delalloc_work), 0,
8240 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8241 			NULL);
8242 	if (!btrfs_delalloc_work_cachep)
8243 		goto fail;
8244 
8245 	return 0;
8246 fail:
8247 	btrfs_destroy_cachep();
8248 	return -ENOMEM;
8249 }
8250 
8251 static int btrfs_getattr(struct vfsmount *mnt,
8252 			 struct dentry *dentry, struct kstat *stat)
8253 {
8254 	u64 delalloc_bytes;
8255 	struct inode *inode = dentry->d_inode;
8256 	u32 blocksize = inode->i_sb->s_blocksize;
8257 
8258 	generic_fillattr(inode, stat);
8259 	stat->dev = BTRFS_I(inode)->root->anon_dev;
8260 	stat->blksize = PAGE_CACHE_SIZE;
8261 
8262 	spin_lock(&BTRFS_I(inode)->lock);
8263 	delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8264 	spin_unlock(&BTRFS_I(inode)->lock);
8265 	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8266 			ALIGN(delalloc_bytes, blocksize)) >> 9;
8267 	return 0;
8268 }
8269 
8270 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8271 			   struct inode *new_dir, struct dentry *new_dentry)
8272 {
8273 	struct btrfs_trans_handle *trans;
8274 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
8275 	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8276 	struct inode *new_inode = new_dentry->d_inode;
8277 	struct inode *old_inode = old_dentry->d_inode;
8278 	struct timespec ctime = CURRENT_TIME;
8279 	u64 index = 0;
8280 	u64 root_objectid;
8281 	int ret;
8282 	u64 old_ino = btrfs_ino(old_inode);
8283 
8284 	if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8285 		return -EPERM;
8286 
8287 	/* we only allow rename subvolume link between subvolumes */
8288 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8289 		return -EXDEV;
8290 
8291 	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8292 	    (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8293 		return -ENOTEMPTY;
8294 
8295 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
8296 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8297 		return -ENOTEMPTY;
8298 
8299 
8300 	/* check for collisions, even if the  name isn't there */
8301 	ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8302 			     new_dentry->d_name.name,
8303 			     new_dentry->d_name.len);
8304 
8305 	if (ret) {
8306 		if (ret == -EEXIST) {
8307 			/* we shouldn't get
8308 			 * eexist without a new_inode */
8309 			if (WARN_ON(!new_inode)) {
8310 				return ret;
8311 			}
8312 		} else {
8313 			/* maybe -EOVERFLOW */
8314 			return ret;
8315 		}
8316 	}
8317 	ret = 0;
8318 
8319 	/*
8320 	 * we're using rename to replace one file with another.  Start IO on it
8321 	 * now so  we don't add too much work to the end of the transaction
8322 	 */
8323 	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8324 		filemap_flush(old_inode->i_mapping);
8325 
8326 	/* close the racy window with snapshot create/destroy ioctl */
8327 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8328 		down_read(&root->fs_info->subvol_sem);
8329 	/*
8330 	 * We want to reserve the absolute worst case amount of items.  So if
8331 	 * both inodes are subvols and we need to unlink them then that would
8332 	 * require 4 item modifications, but if they are both normal inodes it
8333 	 * would require 5 item modifications, so we'll assume their normal
8334 	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8335 	 * should cover the worst case number of items we'll modify.
8336 	 */
8337 	trans = btrfs_start_transaction(root, 11);
8338 	if (IS_ERR(trans)) {
8339                 ret = PTR_ERR(trans);
8340                 goto out_notrans;
8341         }
8342 
8343 	if (dest != root)
8344 		btrfs_record_root_in_trans(trans, dest);
8345 
8346 	ret = btrfs_set_inode_index(new_dir, &index);
8347 	if (ret)
8348 		goto out_fail;
8349 
8350 	BTRFS_I(old_inode)->dir_index = 0ULL;
8351 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8352 		/* force full log commit if subvolume involved. */
8353 		btrfs_set_log_full_commit(root->fs_info, trans);
8354 	} else {
8355 		ret = btrfs_insert_inode_ref(trans, dest,
8356 					     new_dentry->d_name.name,
8357 					     new_dentry->d_name.len,
8358 					     old_ino,
8359 					     btrfs_ino(new_dir), index);
8360 		if (ret)
8361 			goto out_fail;
8362 		/*
8363 		 * this is an ugly little race, but the rename is required
8364 		 * to make sure that if we crash, the inode is either at the
8365 		 * old name or the new one.  pinning the log transaction lets
8366 		 * us make sure we don't allow a log commit to come in after
8367 		 * we unlink the name but before we add the new name back in.
8368 		 */
8369 		btrfs_pin_log_trans(root);
8370 	}
8371 
8372 	inode_inc_iversion(old_dir);
8373 	inode_inc_iversion(new_dir);
8374 	inode_inc_iversion(old_inode);
8375 	old_dir->i_ctime = old_dir->i_mtime = ctime;
8376 	new_dir->i_ctime = new_dir->i_mtime = ctime;
8377 	old_inode->i_ctime = ctime;
8378 
8379 	if (old_dentry->d_parent != new_dentry->d_parent)
8380 		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8381 
8382 	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8383 		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8384 		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8385 					old_dentry->d_name.name,
8386 					old_dentry->d_name.len);
8387 	} else {
8388 		ret = __btrfs_unlink_inode(trans, root, old_dir,
8389 					old_dentry->d_inode,
8390 					old_dentry->d_name.name,
8391 					old_dentry->d_name.len);
8392 		if (!ret)
8393 			ret = btrfs_update_inode(trans, root, old_inode);
8394 	}
8395 	if (ret) {
8396 		btrfs_abort_transaction(trans, root, ret);
8397 		goto out_fail;
8398 	}
8399 
8400 	if (new_inode) {
8401 		inode_inc_iversion(new_inode);
8402 		new_inode->i_ctime = CURRENT_TIME;
8403 		if (unlikely(btrfs_ino(new_inode) ==
8404 			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8405 			root_objectid = BTRFS_I(new_inode)->location.objectid;
8406 			ret = btrfs_unlink_subvol(trans, dest, new_dir,
8407 						root_objectid,
8408 						new_dentry->d_name.name,
8409 						new_dentry->d_name.len);
8410 			BUG_ON(new_inode->i_nlink == 0);
8411 		} else {
8412 			ret = btrfs_unlink_inode(trans, dest, new_dir,
8413 						 new_dentry->d_inode,
8414 						 new_dentry->d_name.name,
8415 						 new_dentry->d_name.len);
8416 		}
8417 		if (!ret && new_inode->i_nlink == 0)
8418 			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8419 		if (ret) {
8420 			btrfs_abort_transaction(trans, root, ret);
8421 			goto out_fail;
8422 		}
8423 	}
8424 
8425 	ret = btrfs_add_link(trans, new_dir, old_inode,
8426 			     new_dentry->d_name.name,
8427 			     new_dentry->d_name.len, 0, index);
8428 	if (ret) {
8429 		btrfs_abort_transaction(trans, root, ret);
8430 		goto out_fail;
8431 	}
8432 
8433 	if (old_inode->i_nlink == 1)
8434 		BTRFS_I(old_inode)->dir_index = index;
8435 
8436 	if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8437 		struct dentry *parent = new_dentry->d_parent;
8438 		btrfs_log_new_name(trans, old_inode, old_dir, parent);
8439 		btrfs_end_log_trans(root);
8440 	}
8441 out_fail:
8442 	btrfs_end_transaction(trans, root);
8443 out_notrans:
8444 	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8445 		up_read(&root->fs_info->subvol_sem);
8446 
8447 	return ret;
8448 }
8449 
8450 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8451 			 struct inode *new_dir, struct dentry *new_dentry,
8452 			 unsigned int flags)
8453 {
8454 	if (flags & ~RENAME_NOREPLACE)
8455 		return -EINVAL;
8456 
8457 	return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8458 }
8459 
8460 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8461 {
8462 	struct btrfs_delalloc_work *delalloc_work;
8463 	struct inode *inode;
8464 
8465 	delalloc_work = container_of(work, struct btrfs_delalloc_work,
8466 				     work);
8467 	inode = delalloc_work->inode;
8468 	if (delalloc_work->wait) {
8469 		btrfs_wait_ordered_range(inode, 0, (u64)-1);
8470 	} else {
8471 		filemap_flush(inode->i_mapping);
8472 		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8473 			     &BTRFS_I(inode)->runtime_flags))
8474 			filemap_flush(inode->i_mapping);
8475 	}
8476 
8477 	if (delalloc_work->delay_iput)
8478 		btrfs_add_delayed_iput(inode);
8479 	else
8480 		iput(inode);
8481 	complete(&delalloc_work->completion);
8482 }
8483 
8484 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8485 						    int wait, int delay_iput)
8486 {
8487 	struct btrfs_delalloc_work *work;
8488 
8489 	work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8490 	if (!work)
8491 		return NULL;
8492 
8493 	init_completion(&work->completion);
8494 	INIT_LIST_HEAD(&work->list);
8495 	work->inode = inode;
8496 	work->wait = wait;
8497 	work->delay_iput = delay_iput;
8498 	btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8499 
8500 	return work;
8501 }
8502 
8503 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8504 {
8505 	wait_for_completion(&work->completion);
8506 	kmem_cache_free(btrfs_delalloc_work_cachep, work);
8507 }
8508 
8509 /*
8510  * some fairly slow code that needs optimization. This walks the list
8511  * of all the inodes with pending delalloc and forces them to disk.
8512  */
8513 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8514 				   int nr)
8515 {
8516 	struct btrfs_inode *binode;
8517 	struct inode *inode;
8518 	struct btrfs_delalloc_work *work, *next;
8519 	struct list_head works;
8520 	struct list_head splice;
8521 	int ret = 0;
8522 
8523 	INIT_LIST_HEAD(&works);
8524 	INIT_LIST_HEAD(&splice);
8525 
8526 	mutex_lock(&root->delalloc_mutex);
8527 	spin_lock(&root->delalloc_lock);
8528 	list_splice_init(&root->delalloc_inodes, &splice);
8529 	while (!list_empty(&splice)) {
8530 		binode = list_entry(splice.next, struct btrfs_inode,
8531 				    delalloc_inodes);
8532 
8533 		list_move_tail(&binode->delalloc_inodes,
8534 			       &root->delalloc_inodes);
8535 		inode = igrab(&binode->vfs_inode);
8536 		if (!inode) {
8537 			cond_resched_lock(&root->delalloc_lock);
8538 			continue;
8539 		}
8540 		spin_unlock(&root->delalloc_lock);
8541 
8542 		work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8543 		if (unlikely(!work)) {
8544 			if (delay_iput)
8545 				btrfs_add_delayed_iput(inode);
8546 			else
8547 				iput(inode);
8548 			ret = -ENOMEM;
8549 			goto out;
8550 		}
8551 		list_add_tail(&work->list, &works);
8552 		btrfs_queue_work(root->fs_info->flush_workers,
8553 				 &work->work);
8554 		ret++;
8555 		if (nr != -1 && ret >= nr)
8556 			goto out;
8557 		cond_resched();
8558 		spin_lock(&root->delalloc_lock);
8559 	}
8560 	spin_unlock(&root->delalloc_lock);
8561 
8562 out:
8563 	list_for_each_entry_safe(work, next, &works, list) {
8564 		list_del_init(&work->list);
8565 		btrfs_wait_and_free_delalloc_work(work);
8566 	}
8567 
8568 	if (!list_empty_careful(&splice)) {
8569 		spin_lock(&root->delalloc_lock);
8570 		list_splice_tail(&splice, &root->delalloc_inodes);
8571 		spin_unlock(&root->delalloc_lock);
8572 	}
8573 	mutex_unlock(&root->delalloc_mutex);
8574 	return ret;
8575 }
8576 
8577 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8578 {
8579 	int ret;
8580 
8581 	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8582 		return -EROFS;
8583 
8584 	ret = __start_delalloc_inodes(root, delay_iput, -1);
8585 	if (ret > 0)
8586 		ret = 0;
8587 	/*
8588 	 * the filemap_flush will queue IO into the worker threads, but
8589 	 * we have to make sure the IO is actually started and that
8590 	 * ordered extents get created before we return
8591 	 */
8592 	atomic_inc(&root->fs_info->async_submit_draining);
8593 	while (atomic_read(&root->fs_info->nr_async_submits) ||
8594 	      atomic_read(&root->fs_info->async_delalloc_pages)) {
8595 		wait_event(root->fs_info->async_submit_wait,
8596 		   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8597 		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8598 	}
8599 	atomic_dec(&root->fs_info->async_submit_draining);
8600 	return ret;
8601 }
8602 
8603 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8604 			       int nr)
8605 {
8606 	struct btrfs_root *root;
8607 	struct list_head splice;
8608 	int ret;
8609 
8610 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8611 		return -EROFS;
8612 
8613 	INIT_LIST_HEAD(&splice);
8614 
8615 	mutex_lock(&fs_info->delalloc_root_mutex);
8616 	spin_lock(&fs_info->delalloc_root_lock);
8617 	list_splice_init(&fs_info->delalloc_roots, &splice);
8618 	while (!list_empty(&splice) && nr) {
8619 		root = list_first_entry(&splice, struct btrfs_root,
8620 					delalloc_root);
8621 		root = btrfs_grab_fs_root(root);
8622 		BUG_ON(!root);
8623 		list_move_tail(&root->delalloc_root,
8624 			       &fs_info->delalloc_roots);
8625 		spin_unlock(&fs_info->delalloc_root_lock);
8626 
8627 		ret = __start_delalloc_inodes(root, delay_iput, nr);
8628 		btrfs_put_fs_root(root);
8629 		if (ret < 0)
8630 			goto out;
8631 
8632 		if (nr != -1) {
8633 			nr -= ret;
8634 			WARN_ON(nr < 0);
8635 		}
8636 		spin_lock(&fs_info->delalloc_root_lock);
8637 	}
8638 	spin_unlock(&fs_info->delalloc_root_lock);
8639 
8640 	ret = 0;
8641 	atomic_inc(&fs_info->async_submit_draining);
8642 	while (atomic_read(&fs_info->nr_async_submits) ||
8643 	      atomic_read(&fs_info->async_delalloc_pages)) {
8644 		wait_event(fs_info->async_submit_wait,
8645 		   (atomic_read(&fs_info->nr_async_submits) == 0 &&
8646 		    atomic_read(&fs_info->async_delalloc_pages) == 0));
8647 	}
8648 	atomic_dec(&fs_info->async_submit_draining);
8649 out:
8650 	if (!list_empty_careful(&splice)) {
8651 		spin_lock(&fs_info->delalloc_root_lock);
8652 		list_splice_tail(&splice, &fs_info->delalloc_roots);
8653 		spin_unlock(&fs_info->delalloc_root_lock);
8654 	}
8655 	mutex_unlock(&fs_info->delalloc_root_mutex);
8656 	return ret;
8657 }
8658 
8659 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8660 			 const char *symname)
8661 {
8662 	struct btrfs_trans_handle *trans;
8663 	struct btrfs_root *root = BTRFS_I(dir)->root;
8664 	struct btrfs_path *path;
8665 	struct btrfs_key key;
8666 	struct inode *inode = NULL;
8667 	int err;
8668 	int drop_inode = 0;
8669 	u64 objectid;
8670 	u64 index = 0;
8671 	int name_len;
8672 	int datasize;
8673 	unsigned long ptr;
8674 	struct btrfs_file_extent_item *ei;
8675 	struct extent_buffer *leaf;
8676 
8677 	name_len = strlen(symname);
8678 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8679 		return -ENAMETOOLONG;
8680 
8681 	/*
8682 	 * 2 items for inode item and ref
8683 	 * 2 items for dir items
8684 	 * 1 item for xattr if selinux is on
8685 	 */
8686 	trans = btrfs_start_transaction(root, 5);
8687 	if (IS_ERR(trans))
8688 		return PTR_ERR(trans);
8689 
8690 	err = btrfs_find_free_ino(root, &objectid);
8691 	if (err)
8692 		goto out_unlock;
8693 
8694 	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8695 				dentry->d_name.len, btrfs_ino(dir), objectid,
8696 				S_IFLNK|S_IRWXUGO, &index);
8697 	if (IS_ERR(inode)) {
8698 		err = PTR_ERR(inode);
8699 		goto out_unlock;
8700 	}
8701 
8702 	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8703 	if (err) {
8704 		drop_inode = 1;
8705 		goto out_unlock;
8706 	}
8707 
8708 	/*
8709 	* If the active LSM wants to access the inode during
8710 	* d_instantiate it needs these. Smack checks to see
8711 	* if the filesystem supports xattrs by looking at the
8712 	* ops vector.
8713 	*/
8714 	inode->i_fop = &btrfs_file_operations;
8715 	inode->i_op = &btrfs_file_inode_operations;
8716 
8717 	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8718 	if (err)
8719 		drop_inode = 1;
8720 	else {
8721 		inode->i_mapping->a_ops = &btrfs_aops;
8722 		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8723 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8724 	}
8725 	if (drop_inode)
8726 		goto out_unlock;
8727 
8728 	path = btrfs_alloc_path();
8729 	if (!path) {
8730 		err = -ENOMEM;
8731 		drop_inode = 1;
8732 		goto out_unlock;
8733 	}
8734 	key.objectid = btrfs_ino(inode);
8735 	key.offset = 0;
8736 	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8737 	datasize = btrfs_file_extent_calc_inline_size(name_len);
8738 	err = btrfs_insert_empty_item(trans, root, path, &key,
8739 				      datasize);
8740 	if (err) {
8741 		drop_inode = 1;
8742 		btrfs_free_path(path);
8743 		goto out_unlock;
8744 	}
8745 	leaf = path->nodes[0];
8746 	ei = btrfs_item_ptr(leaf, path->slots[0],
8747 			    struct btrfs_file_extent_item);
8748 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8749 	btrfs_set_file_extent_type(leaf, ei,
8750 				   BTRFS_FILE_EXTENT_INLINE);
8751 	btrfs_set_file_extent_encryption(leaf, ei, 0);
8752 	btrfs_set_file_extent_compression(leaf, ei, 0);
8753 	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8754 	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8755 
8756 	ptr = btrfs_file_extent_inline_start(ei);
8757 	write_extent_buffer(leaf, symname, ptr, name_len);
8758 	btrfs_mark_buffer_dirty(leaf);
8759 	btrfs_free_path(path);
8760 
8761 	inode->i_op = &btrfs_symlink_inode_operations;
8762 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
8763 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8764 	inode_set_bytes(inode, name_len);
8765 	btrfs_i_size_write(inode, name_len);
8766 	err = btrfs_update_inode(trans, root, inode);
8767 	if (err)
8768 		drop_inode = 1;
8769 
8770 out_unlock:
8771 	if (!err)
8772 		d_instantiate(dentry, inode);
8773 	btrfs_end_transaction(trans, root);
8774 	if (drop_inode) {
8775 		inode_dec_link_count(inode);
8776 		iput(inode);
8777 	}
8778 	btrfs_btree_balance_dirty(root);
8779 	return err;
8780 }
8781 
8782 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8783 				       u64 start, u64 num_bytes, u64 min_size,
8784 				       loff_t actual_len, u64 *alloc_hint,
8785 				       struct btrfs_trans_handle *trans)
8786 {
8787 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8788 	struct extent_map *em;
8789 	struct btrfs_root *root = BTRFS_I(inode)->root;
8790 	struct btrfs_key ins;
8791 	u64 cur_offset = start;
8792 	u64 i_size;
8793 	u64 cur_bytes;
8794 	int ret = 0;
8795 	bool own_trans = true;
8796 
8797 	if (trans)
8798 		own_trans = false;
8799 	while (num_bytes > 0) {
8800 		if (own_trans) {
8801 			trans = btrfs_start_transaction(root, 3);
8802 			if (IS_ERR(trans)) {
8803 				ret = PTR_ERR(trans);
8804 				break;
8805 			}
8806 		}
8807 
8808 		cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8809 		cur_bytes = max(cur_bytes, min_size);
8810 		ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8811 					   *alloc_hint, &ins, 1, 0);
8812 		if (ret) {
8813 			if (own_trans)
8814 				btrfs_end_transaction(trans, root);
8815 			break;
8816 		}
8817 
8818 		ret = insert_reserved_file_extent(trans, inode,
8819 						  cur_offset, ins.objectid,
8820 						  ins.offset, ins.offset,
8821 						  ins.offset, 0, 0, 0,
8822 						  BTRFS_FILE_EXTENT_PREALLOC);
8823 		if (ret) {
8824 			btrfs_free_reserved_extent(root, ins.objectid,
8825 						   ins.offset, 0);
8826 			btrfs_abort_transaction(trans, root, ret);
8827 			if (own_trans)
8828 				btrfs_end_transaction(trans, root);
8829 			break;
8830 		}
8831 		btrfs_drop_extent_cache(inode, cur_offset,
8832 					cur_offset + ins.offset -1, 0);
8833 
8834 		em = alloc_extent_map();
8835 		if (!em) {
8836 			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8837 				&BTRFS_I(inode)->runtime_flags);
8838 			goto next;
8839 		}
8840 
8841 		em->start = cur_offset;
8842 		em->orig_start = cur_offset;
8843 		em->len = ins.offset;
8844 		em->block_start = ins.objectid;
8845 		em->block_len = ins.offset;
8846 		em->orig_block_len = ins.offset;
8847 		em->ram_bytes = ins.offset;
8848 		em->bdev = root->fs_info->fs_devices->latest_bdev;
8849 		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8850 		em->generation = trans->transid;
8851 
8852 		while (1) {
8853 			write_lock(&em_tree->lock);
8854 			ret = add_extent_mapping(em_tree, em, 1);
8855 			write_unlock(&em_tree->lock);
8856 			if (ret != -EEXIST)
8857 				break;
8858 			btrfs_drop_extent_cache(inode, cur_offset,
8859 						cur_offset + ins.offset - 1,
8860 						0);
8861 		}
8862 		free_extent_map(em);
8863 next:
8864 		num_bytes -= ins.offset;
8865 		cur_offset += ins.offset;
8866 		*alloc_hint = ins.objectid + ins.offset;
8867 
8868 		inode_inc_iversion(inode);
8869 		inode->i_ctime = CURRENT_TIME;
8870 		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8871 		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8872 		    (actual_len > inode->i_size) &&
8873 		    (cur_offset > inode->i_size)) {
8874 			if (cur_offset > actual_len)
8875 				i_size = actual_len;
8876 			else
8877 				i_size = cur_offset;
8878 			i_size_write(inode, i_size);
8879 			btrfs_ordered_update_i_size(inode, i_size, NULL);
8880 		}
8881 
8882 		ret = btrfs_update_inode(trans, root, inode);
8883 
8884 		if (ret) {
8885 			btrfs_abort_transaction(trans, root, ret);
8886 			if (own_trans)
8887 				btrfs_end_transaction(trans, root);
8888 			break;
8889 		}
8890 
8891 		if (own_trans)
8892 			btrfs_end_transaction(trans, root);
8893 	}
8894 	return ret;
8895 }
8896 
8897 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8898 			      u64 start, u64 num_bytes, u64 min_size,
8899 			      loff_t actual_len, u64 *alloc_hint)
8900 {
8901 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8902 					   min_size, actual_len, alloc_hint,
8903 					   NULL);
8904 }
8905 
8906 int btrfs_prealloc_file_range_trans(struct inode *inode,
8907 				    struct btrfs_trans_handle *trans, int mode,
8908 				    u64 start, u64 num_bytes, u64 min_size,
8909 				    loff_t actual_len, u64 *alloc_hint)
8910 {
8911 	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8912 					   min_size, actual_len, alloc_hint, trans);
8913 }
8914 
8915 static int btrfs_set_page_dirty(struct page *page)
8916 {
8917 	return __set_page_dirty_nobuffers(page);
8918 }
8919 
8920 static int btrfs_permission(struct inode *inode, int mask)
8921 {
8922 	struct btrfs_root *root = BTRFS_I(inode)->root;
8923 	umode_t mode = inode->i_mode;
8924 
8925 	if (mask & MAY_WRITE &&
8926 	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8927 		if (btrfs_root_readonly(root))
8928 			return -EROFS;
8929 		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8930 			return -EACCES;
8931 	}
8932 	return generic_permission(inode, mask);
8933 }
8934 
8935 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8936 {
8937 	struct btrfs_trans_handle *trans;
8938 	struct btrfs_root *root = BTRFS_I(dir)->root;
8939 	struct inode *inode = NULL;
8940 	u64 objectid;
8941 	u64 index;
8942 	int ret = 0;
8943 
8944 	/*
8945 	 * 5 units required for adding orphan entry
8946 	 */
8947 	trans = btrfs_start_transaction(root, 5);
8948 	if (IS_ERR(trans))
8949 		return PTR_ERR(trans);
8950 
8951 	ret = btrfs_find_free_ino(root, &objectid);
8952 	if (ret)
8953 		goto out;
8954 
8955 	inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8956 				btrfs_ino(dir), objectid, mode, &index);
8957 	if (IS_ERR(inode)) {
8958 		ret = PTR_ERR(inode);
8959 		inode = NULL;
8960 		goto out;
8961 	}
8962 
8963 	ret = btrfs_init_inode_security(trans, inode, dir, NULL);
8964 	if (ret)
8965 		goto out;
8966 
8967 	ret = btrfs_update_inode(trans, root, inode);
8968 	if (ret)
8969 		goto out;
8970 
8971 	inode->i_fop = &btrfs_file_operations;
8972 	inode->i_op = &btrfs_file_inode_operations;
8973 
8974 	inode->i_mapping->a_ops = &btrfs_aops;
8975 	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8976 	BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8977 
8978 	ret = btrfs_orphan_add(trans, inode);
8979 	if (ret)
8980 		goto out;
8981 
8982 	d_tmpfile(dentry, inode);
8983 	mark_inode_dirty(inode);
8984 
8985 out:
8986 	btrfs_end_transaction(trans, root);
8987 	if (ret)
8988 		iput(inode);
8989 	btrfs_balance_delayed_items(root);
8990 	btrfs_btree_balance_dirty(root);
8991 
8992 	return ret;
8993 }
8994 
8995 static const struct inode_operations btrfs_dir_inode_operations = {
8996 	.getattr	= btrfs_getattr,
8997 	.lookup		= btrfs_lookup,
8998 	.create		= btrfs_create,
8999 	.unlink		= btrfs_unlink,
9000 	.link		= btrfs_link,
9001 	.mkdir		= btrfs_mkdir,
9002 	.rmdir		= btrfs_rmdir,
9003 	.rename2	= btrfs_rename2,
9004 	.symlink	= btrfs_symlink,
9005 	.setattr	= btrfs_setattr,
9006 	.mknod		= btrfs_mknod,
9007 	.setxattr	= btrfs_setxattr,
9008 	.getxattr	= btrfs_getxattr,
9009 	.listxattr	= btrfs_listxattr,
9010 	.removexattr	= btrfs_removexattr,
9011 	.permission	= btrfs_permission,
9012 	.get_acl	= btrfs_get_acl,
9013 	.set_acl	= btrfs_set_acl,
9014 	.update_time	= btrfs_update_time,
9015 	.tmpfile        = btrfs_tmpfile,
9016 };
9017 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9018 	.lookup		= btrfs_lookup,
9019 	.permission	= btrfs_permission,
9020 	.get_acl	= btrfs_get_acl,
9021 	.set_acl	= btrfs_set_acl,
9022 	.update_time	= btrfs_update_time,
9023 };
9024 
9025 static const struct file_operations btrfs_dir_file_operations = {
9026 	.llseek		= generic_file_llseek,
9027 	.read		= generic_read_dir,
9028 	.iterate	= btrfs_real_readdir,
9029 	.unlocked_ioctl	= btrfs_ioctl,
9030 #ifdef CONFIG_COMPAT
9031 	.compat_ioctl	= btrfs_ioctl,
9032 #endif
9033 	.release        = btrfs_release_file,
9034 	.fsync		= btrfs_sync_file,
9035 };
9036 
9037 static struct extent_io_ops btrfs_extent_io_ops = {
9038 	.fill_delalloc = run_delalloc_range,
9039 	.submit_bio_hook = btrfs_submit_bio_hook,
9040 	.merge_bio_hook = btrfs_merge_bio_hook,
9041 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
9042 	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
9043 	.writepage_start_hook = btrfs_writepage_start_hook,
9044 	.set_bit_hook = btrfs_set_bit_hook,
9045 	.clear_bit_hook = btrfs_clear_bit_hook,
9046 	.merge_extent_hook = btrfs_merge_extent_hook,
9047 	.split_extent_hook = btrfs_split_extent_hook,
9048 };
9049 
9050 /*
9051  * btrfs doesn't support the bmap operation because swapfiles
9052  * use bmap to make a mapping of extents in the file.  They assume
9053  * these extents won't change over the life of the file and they
9054  * use the bmap result to do IO directly to the drive.
9055  *
9056  * the btrfs bmap call would return logical addresses that aren't
9057  * suitable for IO and they also will change frequently as COW
9058  * operations happen.  So, swapfile + btrfs == corruption.
9059  *
9060  * For now we're avoiding this by dropping bmap.
9061  */
9062 static const struct address_space_operations btrfs_aops = {
9063 	.readpage	= btrfs_readpage,
9064 	.writepage	= btrfs_writepage,
9065 	.writepages	= btrfs_writepages,
9066 	.readpages	= btrfs_readpages,
9067 	.direct_IO	= btrfs_direct_IO,
9068 	.invalidatepage = btrfs_invalidatepage,
9069 	.releasepage	= btrfs_releasepage,
9070 	.set_page_dirty	= btrfs_set_page_dirty,
9071 	.error_remove_page = generic_error_remove_page,
9072 };
9073 
9074 static const struct address_space_operations btrfs_symlink_aops = {
9075 	.readpage	= btrfs_readpage,
9076 	.writepage	= btrfs_writepage,
9077 	.invalidatepage = btrfs_invalidatepage,
9078 	.releasepage	= btrfs_releasepage,
9079 };
9080 
9081 static const struct inode_operations btrfs_file_inode_operations = {
9082 	.getattr	= btrfs_getattr,
9083 	.setattr	= btrfs_setattr,
9084 	.setxattr	= btrfs_setxattr,
9085 	.getxattr	= btrfs_getxattr,
9086 	.listxattr      = btrfs_listxattr,
9087 	.removexattr	= btrfs_removexattr,
9088 	.permission	= btrfs_permission,
9089 	.fiemap		= btrfs_fiemap,
9090 	.get_acl	= btrfs_get_acl,
9091 	.set_acl	= btrfs_set_acl,
9092 	.update_time	= btrfs_update_time,
9093 };
9094 static const struct inode_operations btrfs_special_inode_operations = {
9095 	.getattr	= btrfs_getattr,
9096 	.setattr	= btrfs_setattr,
9097 	.permission	= btrfs_permission,
9098 	.setxattr	= btrfs_setxattr,
9099 	.getxattr	= btrfs_getxattr,
9100 	.listxattr	= btrfs_listxattr,
9101 	.removexattr	= btrfs_removexattr,
9102 	.get_acl	= btrfs_get_acl,
9103 	.set_acl	= btrfs_set_acl,
9104 	.update_time	= btrfs_update_time,
9105 };
9106 static const struct inode_operations btrfs_symlink_inode_operations = {
9107 	.readlink	= generic_readlink,
9108 	.follow_link	= page_follow_link_light,
9109 	.put_link	= page_put_link,
9110 	.getattr	= btrfs_getattr,
9111 	.setattr	= btrfs_setattr,
9112 	.permission	= btrfs_permission,
9113 	.setxattr	= btrfs_setxattr,
9114 	.getxattr	= btrfs_getxattr,
9115 	.listxattr	= btrfs_listxattr,
9116 	.removexattr	= btrfs_removexattr,
9117 	.update_time	= btrfs_update_time,
9118 };
9119 
9120 const struct dentry_operations btrfs_dentry_operations = {
9121 	.d_delete	= btrfs_dentry_delete,
9122 	.d_release	= btrfs_dentry_release,
9123 };
9124