btrfs: rename btrfs_update_iflags to reflect which flags it touches
[sfrench/cifs-2.6.git] / fs / btrfs / inode.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
10 #include <linux/fs.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/mpage.h>
18 #include <linux/swap.h>
19 #include <linux/writeback.h>
20 #include <linux/compat.h>
21 #include <linux/bit_spinlock.h>
22 #include <linux/xattr.h>
23 #include <linux/posix_acl.h>
24 #include <linux/falloc.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/mount.h>
28 #include <linux/btrfs.h>
29 #include <linux/blkdev.h>
30 #include <linux/posix_acl_xattr.h>
31 #include <linux/uio.h>
32 #include <linux/magic.h>
33 #include <linux/iversion.h>
34 #include <asm/unaligned.h>
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
40 #include "ordered-data.h"
41 #include "xattr.h"
42 #include "tree-log.h"
43 #include "volumes.h"
44 #include "compression.h"
45 #include "locking.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "backref.h"
49 #include "props.h"
50 #include "qgroup.h"
51 #include "dedupe.h"
52
53 struct btrfs_iget_args {
54         struct btrfs_key *location;
55         struct btrfs_root *root;
56 };
57
58 struct btrfs_dio_data {
59         u64 reserve;
60         u64 unsubmitted_oe_range_start;
61         u64 unsubmitted_oe_range_end;
62         int overwrite;
63 };
64
65 static const struct inode_operations btrfs_dir_inode_operations;
66 static const struct inode_operations btrfs_symlink_inode_operations;
67 static const struct inode_operations btrfs_dir_ro_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct address_space_operations btrfs_symlink_aops;
72 static const struct file_operations btrfs_dir_file_operations;
73 static const struct extent_io_ops btrfs_extent_io_ops;
74
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79
80 #define S_SHIFT 12
81 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
89 };
90
91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
92 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95                                    struct page *locked_page,
96                                    u64 start, u64 end, u64 delalloc_end,
97                                    int *page_started, unsigned long *nr_written,
98                                    int unlock, struct btrfs_dedupe_hash *hash);
99 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
100                                        u64 orig_start, u64 block_start,
101                                        u64 block_len, u64 orig_block_len,
102                                        u64 ram_bytes, int compress_type,
103                                        int type);
104
105 static void __endio_write_update_ordered(struct inode *inode,
106                                          const u64 offset, const u64 bytes,
107                                          const bool uptodate);
108
109 /*
110  * Cleanup all submitted ordered extents in specified range to handle errors
111  * from the fill_dellaloc() callback.
112  *
113  * NOTE: caller must ensure that when an error happens, it can not call
114  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
115  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
116  * to be released, which we want to happen only when finishing the ordered
117  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
118  * fill_delalloc() callback already does proper cleanup for the first page of
119  * the range, that is, it invokes the callback writepage_end_io_hook() for the
120  * range of the first page.
121  */
122 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
123                                                  const u64 offset,
124                                                  const u64 bytes)
125 {
126         unsigned long index = offset >> PAGE_SHIFT;
127         unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
128         struct page *page;
129
130         while (index <= end_index) {
131                 page = find_get_page(inode->i_mapping, index);
132                 index++;
133                 if (!page)
134                         continue;
135                 ClearPagePrivate2(page);
136                 put_page(page);
137         }
138         return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139                                             bytes - PAGE_SIZE, false);
140 }
141
142 static int btrfs_dirty_inode(struct inode *inode);
143
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
146 {
147         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
148 }
149 #endif
150
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152                                      struct inode *inode,  struct inode *dir,
153                                      const struct qstr *qstr)
154 {
155         int err;
156
157         err = btrfs_init_acl(trans, inode, dir);
158         if (!err)
159                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160         return err;
161 }
162
163 /*
164  * this does all the hard work for inserting an inline extent into
165  * the btree.  The caller should have done a btrfs_drop_extents so that
166  * no overlapping inline items exist in the btree
167  */
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169                                 struct btrfs_path *path, int extent_inserted,
170                                 struct btrfs_root *root, struct inode *inode,
171                                 u64 start, size_t size, size_t compressed_size,
172                                 int compress_type,
173                                 struct page **compressed_pages)
174 {
175         struct extent_buffer *leaf;
176         struct page *page = NULL;
177         char *kaddr;
178         unsigned long ptr;
179         struct btrfs_file_extent_item *ei;
180         int ret;
181         size_t cur_size = size;
182         unsigned long offset;
183
184         if (compressed_size && compressed_pages)
185                 cur_size = compressed_size;
186
187         inode_add_bytes(inode, size);
188
189         if (!extent_inserted) {
190                 struct btrfs_key key;
191                 size_t datasize;
192
193                 key.objectid = btrfs_ino(BTRFS_I(inode));
194                 key.offset = start;
195                 key.type = BTRFS_EXTENT_DATA_KEY;
196
197                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198                 path->leave_spinning = 1;
199                 ret = btrfs_insert_empty_item(trans, root, path, &key,
200                                               datasize);
201                 if (ret)
202                         goto fail;
203         }
204         leaf = path->nodes[0];
205         ei = btrfs_item_ptr(leaf, path->slots[0],
206                             struct btrfs_file_extent_item);
207         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209         btrfs_set_file_extent_encryption(leaf, ei, 0);
210         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212         ptr = btrfs_file_extent_inline_start(ei);
213
214         if (compress_type != BTRFS_COMPRESS_NONE) {
215                 struct page *cpage;
216                 int i = 0;
217                 while (compressed_size > 0) {
218                         cpage = compressed_pages[i];
219                         cur_size = min_t(unsigned long, compressed_size,
220                                        PAGE_SIZE);
221
222                         kaddr = kmap_atomic(cpage);
223                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
224                         kunmap_atomic(kaddr);
225
226                         i++;
227                         ptr += cur_size;
228                         compressed_size -= cur_size;
229                 }
230                 btrfs_set_file_extent_compression(leaf, ei,
231                                                   compress_type);
232         } else {
233                 page = find_get_page(inode->i_mapping,
234                                      start >> PAGE_SHIFT);
235                 btrfs_set_file_extent_compression(leaf, ei, 0);
236                 kaddr = kmap_atomic(page);
237                 offset = start & (PAGE_SIZE - 1);
238                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239                 kunmap_atomic(kaddr);
240                 put_page(page);
241         }
242         btrfs_mark_buffer_dirty(leaf);
243         btrfs_release_path(path);
244
245         /*
246          * we're an inline extent, so nobody can
247          * extend the file past i_size without locking
248          * a page we already have locked.
249          *
250          * We must do any isize and inode updates
251          * before we unlock the pages.  Otherwise we
252          * could end up racing with unlink.
253          */
254         BTRFS_I(inode)->disk_i_size = inode->i_size;
255         ret = btrfs_update_inode(trans, root, inode);
256
257 fail:
258         return ret;
259 }
260
261
262 /*
263  * conditionally insert an inline extent into the file.  This
264  * does the checks required to make sure the data is small enough
265  * to fit as an inline extent.
266  */
267 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
268                                           u64 end, size_t compressed_size,
269                                           int compress_type,
270                                           struct page **compressed_pages)
271 {
272         struct btrfs_root *root = BTRFS_I(inode)->root;
273         struct btrfs_fs_info *fs_info = root->fs_info;
274         struct btrfs_trans_handle *trans;
275         u64 isize = i_size_read(inode);
276         u64 actual_end = min(end + 1, isize);
277         u64 inline_len = actual_end - start;
278         u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279         u64 data_len = inline_len;
280         int ret;
281         struct btrfs_path *path;
282         int extent_inserted = 0;
283         u32 extent_item_size;
284
285         if (compressed_size)
286                 data_len = compressed_size;
287
288         if (start > 0 ||
289             actual_end > fs_info->sectorsize ||
290             data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
291             (!compressed_size &&
292             (actual_end & (fs_info->sectorsize - 1)) == 0) ||
293             end + 1 < isize ||
294             data_len > fs_info->max_inline) {
295                 return 1;
296         }
297
298         path = btrfs_alloc_path();
299         if (!path)
300                 return -ENOMEM;
301
302         trans = btrfs_join_transaction(root);
303         if (IS_ERR(trans)) {
304                 btrfs_free_path(path);
305                 return PTR_ERR(trans);
306         }
307         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
308
309         if (compressed_size && compressed_pages)
310                 extent_item_size = btrfs_file_extent_calc_inline_size(
311                    compressed_size);
312         else
313                 extent_item_size = btrfs_file_extent_calc_inline_size(
314                     inline_len);
315
316         ret = __btrfs_drop_extents(trans, root, inode, path,
317                                    start, aligned_end, NULL,
318                                    1, 1, extent_item_size, &extent_inserted);
319         if (ret) {
320                 btrfs_abort_transaction(trans, ret);
321                 goto out;
322         }
323
324         if (isize > actual_end)
325                 inline_len = min_t(u64, isize, actual_end);
326         ret = insert_inline_extent(trans, path, extent_inserted,
327                                    root, inode, start,
328                                    inline_len, compressed_size,
329                                    compress_type, compressed_pages);
330         if (ret && ret != -ENOSPC) {
331                 btrfs_abort_transaction(trans, ret);
332                 goto out;
333         } else if (ret == -ENOSPC) {
334                 ret = 1;
335                 goto out;
336         }
337
338         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339         btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
340 out:
341         /*
342          * Don't forget to free the reserved space, as for inlined extent
343          * it won't count as data extent, free them directly here.
344          * And at reserve time, it's always aligned to page size, so
345          * just free one page here.
346          */
347         btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
348         btrfs_free_path(path);
349         btrfs_end_transaction(trans);
350         return ret;
351 }
352
353 struct async_extent {
354         u64 start;
355         u64 ram_size;
356         u64 compressed_size;
357         struct page **pages;
358         unsigned long nr_pages;
359         int compress_type;
360         struct list_head list;
361 };
362
363 struct async_cow {
364         struct inode *inode;
365         struct btrfs_root *root;
366         struct page *locked_page;
367         u64 start;
368         u64 end;
369         unsigned int write_flags;
370         struct list_head extents;
371         struct btrfs_work work;
372 };
373
374 static noinline int add_async_extent(struct async_cow *cow,
375                                      u64 start, u64 ram_size,
376                                      u64 compressed_size,
377                                      struct page **pages,
378                                      unsigned long nr_pages,
379                                      int compress_type)
380 {
381         struct async_extent *async_extent;
382
383         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384         BUG_ON(!async_extent); /* -ENOMEM */
385         async_extent->start = start;
386         async_extent->ram_size = ram_size;
387         async_extent->compressed_size = compressed_size;
388         async_extent->pages = pages;
389         async_extent->nr_pages = nr_pages;
390         async_extent->compress_type = compress_type;
391         list_add_tail(&async_extent->list, &cow->extents);
392         return 0;
393 }
394
395 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
396 {
397         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
398
399         /* force compress */
400         if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
401                 return 1;
402         /* defrag ioctl */
403         if (BTRFS_I(inode)->defrag_compress)
404                 return 1;
405         /* bad compression ratios */
406         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
407                 return 0;
408         if (btrfs_test_opt(fs_info, COMPRESS) ||
409             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
410             BTRFS_I(inode)->prop_compress)
411                 return btrfs_compress_heuristic(inode, start, end);
412         return 0;
413 }
414
415 static inline void inode_should_defrag(struct btrfs_inode *inode,
416                 u64 start, u64 end, u64 num_bytes, u64 small_write)
417 {
418         /* If this is a small write inside eof, kick off a defrag */
419         if (num_bytes < small_write &&
420             (start > 0 || end + 1 < inode->disk_i_size))
421                 btrfs_add_inode_defrag(NULL, inode);
422 }
423
424 /*
425  * we create compressed extents in two phases.  The first
426  * phase compresses a range of pages that have already been
427  * locked (both pages and state bits are locked).
428  *
429  * This is done inside an ordered work queue, and the compression
430  * is spread across many cpus.  The actual IO submission is step
431  * two, and the ordered work queue takes care of making sure that
432  * happens in the same order things were put onto the queue by
433  * writepages and friends.
434  *
435  * If this code finds it can't get good compression, it puts an
436  * entry onto the work queue to write the uncompressed bytes.  This
437  * makes sure that both compressed inodes and uncompressed inodes
438  * are written in the same order that the flusher thread sent them
439  * down.
440  */
441 static noinline void compress_file_range(struct inode *inode,
442                                         struct page *locked_page,
443                                         u64 start, u64 end,
444                                         struct async_cow *async_cow,
445                                         int *num_added)
446 {
447         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
448         u64 blocksize = fs_info->sectorsize;
449         u64 actual_end;
450         u64 isize = i_size_read(inode);
451         int ret = 0;
452         struct page **pages = NULL;
453         unsigned long nr_pages;
454         unsigned long total_compressed = 0;
455         unsigned long total_in = 0;
456         int i;
457         int will_compress;
458         int compress_type = fs_info->compress_type;
459         int redirty = 0;
460
461         inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
462                         SZ_16K);
463
464         actual_end = min_t(u64, isize, end + 1);
465 again:
466         will_compress = 0;
467         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
468         BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
469         nr_pages = min_t(unsigned long, nr_pages,
470                         BTRFS_MAX_COMPRESSED / PAGE_SIZE);
471
472         /*
473          * we don't want to send crud past the end of i_size through
474          * compression, that's just a waste of CPU time.  So, if the
475          * end of the file is before the start of our current
476          * requested range of bytes, we bail out to the uncompressed
477          * cleanup code that can deal with all of this.
478          *
479          * It isn't really the fastest way to fix things, but this is a
480          * very uncommon corner.
481          */
482         if (actual_end <= start)
483                 goto cleanup_and_bail_uncompressed;
484
485         total_compressed = actual_end - start;
486
487         /*
488          * skip compression for a small file range(<=blocksize) that
489          * isn't an inline extent, since it doesn't save disk space at all.
490          */
491         if (total_compressed <= blocksize &&
492            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
493                 goto cleanup_and_bail_uncompressed;
494
495         total_compressed = min_t(unsigned long, total_compressed,
496                         BTRFS_MAX_UNCOMPRESSED);
497         total_in = 0;
498         ret = 0;
499
500         /*
501          * we do compression for mount -o compress and when the
502          * inode has not been flagged as nocompress.  This flag can
503          * change at any time if we discover bad compression ratios.
504          */
505         if (inode_need_compress(inode, start, end)) {
506                 WARN_ON(pages);
507                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
508                 if (!pages) {
509                         /* just bail out to the uncompressed code */
510                         goto cont;
511                 }
512
513                 if (BTRFS_I(inode)->defrag_compress)
514                         compress_type = BTRFS_I(inode)->defrag_compress;
515                 else if (BTRFS_I(inode)->prop_compress)
516                         compress_type = BTRFS_I(inode)->prop_compress;
517
518                 /*
519                  * we need to call clear_page_dirty_for_io on each
520                  * page in the range.  Otherwise applications with the file
521                  * mmap'd can wander in and change the page contents while
522                  * we are compressing them.
523                  *
524                  * If the compression fails for any reason, we set the pages
525                  * dirty again later on.
526                  *
527                  * Note that the remaining part is redirtied, the start pointer
528                  * has moved, the end is the original one.
529                  */
530                 if (!redirty) {
531                         extent_range_clear_dirty_for_io(inode, start, end);
532                         redirty = 1;
533                 }
534
535                 /* Compression level is applied here and only here */
536                 ret = btrfs_compress_pages(
537                         compress_type | (fs_info->compress_level << 4),
538                                            inode->i_mapping, start,
539                                            pages,
540                                            &nr_pages,
541                                            &total_in,
542                                            &total_compressed);
543
544                 if (!ret) {
545                         unsigned long offset = total_compressed &
546                                 (PAGE_SIZE - 1);
547                         struct page *page = pages[nr_pages - 1];
548                         char *kaddr;
549
550                         /* zero the tail end of the last page, we might be
551                          * sending it down to disk
552                          */
553                         if (offset) {
554                                 kaddr = kmap_atomic(page);
555                                 memset(kaddr + offset, 0,
556                                        PAGE_SIZE - offset);
557                                 kunmap_atomic(kaddr);
558                         }
559                         will_compress = 1;
560                 }
561         }
562 cont:
563         if (start == 0) {
564                 /* lets try to make an inline extent */
565                 if (ret || total_in < actual_end) {
566                         /* we didn't compress the entire range, try
567                          * to make an uncompressed inline extent.
568                          */
569                         ret = cow_file_range_inline(inode, start, end, 0,
570                                                     BTRFS_COMPRESS_NONE, NULL);
571                 } else {
572                         /* try making a compressed inline extent */
573                         ret = cow_file_range_inline(inode, start, end,
574                                                     total_compressed,
575                                                     compress_type, pages);
576                 }
577                 if (ret <= 0) {
578                         unsigned long clear_flags = EXTENT_DELALLOC |
579                                 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
580                                 EXTENT_DO_ACCOUNTING;
581                         unsigned long page_error_op;
582
583                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
584
585                         /*
586                          * inline extent creation worked or returned error,
587                          * we don't need to create any more async work items.
588                          * Unlock and free up our temp pages.
589                          *
590                          * We use DO_ACCOUNTING here because we need the
591                          * delalloc_release_metadata to be done _after_ we drop
592                          * our outstanding extent for clearing delalloc for this
593                          * range.
594                          */
595                         extent_clear_unlock_delalloc(inode, start, end, end,
596                                                      NULL, clear_flags,
597                                                      PAGE_UNLOCK |
598                                                      PAGE_CLEAR_DIRTY |
599                                                      PAGE_SET_WRITEBACK |
600                                                      page_error_op |
601                                                      PAGE_END_WRITEBACK);
602                         goto free_pages_out;
603                 }
604         }
605
606         if (will_compress) {
607                 /*
608                  * we aren't doing an inline extent round the compressed size
609                  * up to a block size boundary so the allocator does sane
610                  * things
611                  */
612                 total_compressed = ALIGN(total_compressed, blocksize);
613
614                 /*
615                  * one last check to make sure the compression is really a
616                  * win, compare the page count read with the blocks on disk,
617                  * compression must free at least one sector size
618                  */
619                 total_in = ALIGN(total_in, PAGE_SIZE);
620                 if (total_compressed + blocksize <= total_in) {
621                         *num_added += 1;
622
623                         /*
624                          * The async work queues will take care of doing actual
625                          * allocation on disk for these compressed pages, and
626                          * will submit them to the elevator.
627                          */
628                         add_async_extent(async_cow, start, total_in,
629                                         total_compressed, pages, nr_pages,
630                                         compress_type);
631
632                         if (start + total_in < end) {
633                                 start += total_in;
634                                 pages = NULL;
635                                 cond_resched();
636                                 goto again;
637                         }
638                         return;
639                 }
640         }
641         if (pages) {
642                 /*
643                  * the compression code ran but failed to make things smaller,
644                  * free any pages it allocated and our page pointer array
645                  */
646                 for (i = 0; i < nr_pages; i++) {
647                         WARN_ON(pages[i]->mapping);
648                         put_page(pages[i]);
649                 }
650                 kfree(pages);
651                 pages = NULL;
652                 total_compressed = 0;
653                 nr_pages = 0;
654
655                 /* flag the file so we don't compress in the future */
656                 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
657                     !(BTRFS_I(inode)->prop_compress)) {
658                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
659                 }
660         }
661 cleanup_and_bail_uncompressed:
662         /*
663          * No compression, but we still need to write the pages in the file
664          * we've been given so far.  redirty the locked page if it corresponds
665          * to our extent and set things up for the async work queue to run
666          * cow_file_range to do the normal delalloc dance.
667          */
668         if (page_offset(locked_page) >= start &&
669             page_offset(locked_page) <= end)
670                 __set_page_dirty_nobuffers(locked_page);
671                 /* unlocked later on in the async handlers */
672
673         if (redirty)
674                 extent_range_redirty_for_io(inode, start, end);
675         add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
676                          BTRFS_COMPRESS_NONE);
677         *num_added += 1;
678
679         return;
680
681 free_pages_out:
682         for (i = 0; i < nr_pages; i++) {
683                 WARN_ON(pages[i]->mapping);
684                 put_page(pages[i]);
685         }
686         kfree(pages);
687 }
688
689 static void free_async_extent_pages(struct async_extent *async_extent)
690 {
691         int i;
692
693         if (!async_extent->pages)
694                 return;
695
696         for (i = 0; i < async_extent->nr_pages; i++) {
697                 WARN_ON(async_extent->pages[i]->mapping);
698                 put_page(async_extent->pages[i]);
699         }
700         kfree(async_extent->pages);
701         async_extent->nr_pages = 0;
702         async_extent->pages = NULL;
703 }
704
705 /*
706  * phase two of compressed writeback.  This is the ordered portion
707  * of the code, which only gets called in the order the work was
708  * queued.  We walk all the async extents created by compress_file_range
709  * and send them down to the disk.
710  */
711 static noinline void submit_compressed_extents(struct inode *inode,
712                                               struct async_cow *async_cow)
713 {
714         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
715         struct async_extent *async_extent;
716         u64 alloc_hint = 0;
717         struct btrfs_key ins;
718         struct extent_map *em;
719         struct btrfs_root *root = BTRFS_I(inode)->root;
720         struct extent_io_tree *io_tree;
721         int ret = 0;
722
723 again:
724         while (!list_empty(&async_cow->extents)) {
725                 async_extent = list_entry(async_cow->extents.next,
726                                           struct async_extent, list);
727                 list_del(&async_extent->list);
728
729                 io_tree = &BTRFS_I(inode)->io_tree;
730
731 retry:
732                 /* did the compression code fall back to uncompressed IO? */
733                 if (!async_extent->pages) {
734                         int page_started = 0;
735                         unsigned long nr_written = 0;
736
737                         lock_extent(io_tree, async_extent->start,
738                                          async_extent->start +
739                                          async_extent->ram_size - 1);
740
741                         /* allocate blocks */
742                         ret = cow_file_range(inode, async_cow->locked_page,
743                                              async_extent->start,
744                                              async_extent->start +
745                                              async_extent->ram_size - 1,
746                                              async_extent->start +
747                                              async_extent->ram_size - 1,
748                                              &page_started, &nr_written, 0,
749                                              NULL);
750
751                         /* JDM XXX */
752
753                         /*
754                          * if page_started, cow_file_range inserted an
755                          * inline extent and took care of all the unlocking
756                          * and IO for us.  Otherwise, we need to submit
757                          * all those pages down to the drive.
758                          */
759                         if (!page_started && !ret)
760                                 extent_write_locked_range(inode,
761                                                   async_extent->start,
762                                                   async_extent->start +
763                                                   async_extent->ram_size - 1,
764                                                   WB_SYNC_ALL);
765                         else if (ret)
766                                 unlock_page(async_cow->locked_page);
767                         kfree(async_extent);
768                         cond_resched();
769                         continue;
770                 }
771
772                 lock_extent(io_tree, async_extent->start,
773                             async_extent->start + async_extent->ram_size - 1);
774
775                 ret = btrfs_reserve_extent(root, async_extent->ram_size,
776                                            async_extent->compressed_size,
777                                            async_extent->compressed_size,
778                                            0, alloc_hint, &ins, 1, 1);
779                 if (ret) {
780                         free_async_extent_pages(async_extent);
781
782                         if (ret == -ENOSPC) {
783                                 unlock_extent(io_tree, async_extent->start,
784                                               async_extent->start +
785                                               async_extent->ram_size - 1);
786
787                                 /*
788                                  * we need to redirty the pages if we decide to
789                                  * fallback to uncompressed IO, otherwise we
790                                  * will not submit these pages down to lower
791                                  * layers.
792                                  */
793                                 extent_range_redirty_for_io(inode,
794                                                 async_extent->start,
795                                                 async_extent->start +
796                                                 async_extent->ram_size - 1);
797
798                                 goto retry;
799                         }
800                         goto out_free;
801                 }
802                 /*
803                  * here we're doing allocation and writeback of the
804                  * compressed pages
805                  */
806                 em = create_io_em(inode, async_extent->start,
807                                   async_extent->ram_size, /* len */
808                                   async_extent->start, /* orig_start */
809                                   ins.objectid, /* block_start */
810                                   ins.offset, /* block_len */
811                                   ins.offset, /* orig_block_len */
812                                   async_extent->ram_size, /* ram_bytes */
813                                   async_extent->compress_type,
814                                   BTRFS_ORDERED_COMPRESSED);
815                 if (IS_ERR(em))
816                         /* ret value is not necessary due to void function */
817                         goto out_free_reserve;
818                 free_extent_map(em);
819
820                 ret = btrfs_add_ordered_extent_compress(inode,
821                                                 async_extent->start,
822                                                 ins.objectid,
823                                                 async_extent->ram_size,
824                                                 ins.offset,
825                                                 BTRFS_ORDERED_COMPRESSED,
826                                                 async_extent->compress_type);
827                 if (ret) {
828                         btrfs_drop_extent_cache(BTRFS_I(inode),
829                                                 async_extent->start,
830                                                 async_extent->start +
831                                                 async_extent->ram_size - 1, 0);
832                         goto out_free_reserve;
833                 }
834                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
835
836                 /*
837                  * clear dirty, set writeback and unlock the pages.
838                  */
839                 extent_clear_unlock_delalloc(inode, async_extent->start,
840                                 async_extent->start +
841                                 async_extent->ram_size - 1,
842                                 async_extent->start +
843                                 async_extent->ram_size - 1,
844                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
845                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
846                                 PAGE_SET_WRITEBACK);
847                 if (btrfs_submit_compressed_write(inode,
848                                     async_extent->start,
849                                     async_extent->ram_size,
850                                     ins.objectid,
851                                     ins.offset, async_extent->pages,
852                                     async_extent->nr_pages,
853                                     async_cow->write_flags)) {
854                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
855                         struct page *p = async_extent->pages[0];
856                         const u64 start = async_extent->start;
857                         const u64 end = start + async_extent->ram_size - 1;
858
859                         p->mapping = inode->i_mapping;
860                         tree->ops->writepage_end_io_hook(p, start, end,
861                                                          NULL, 0);
862                         p->mapping = NULL;
863                         extent_clear_unlock_delalloc(inode, start, end, end,
864                                                      NULL, 0,
865                                                      PAGE_END_WRITEBACK |
866                                                      PAGE_SET_ERROR);
867                         free_async_extent_pages(async_extent);
868                 }
869                 alloc_hint = ins.objectid + ins.offset;
870                 kfree(async_extent);
871                 cond_resched();
872         }
873         return;
874 out_free_reserve:
875         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
876         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
877 out_free:
878         extent_clear_unlock_delalloc(inode, async_extent->start,
879                                      async_extent->start +
880                                      async_extent->ram_size - 1,
881                                      async_extent->start +
882                                      async_extent->ram_size - 1,
883                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
884                                      EXTENT_DELALLOC_NEW |
885                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
886                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
888                                      PAGE_SET_ERROR);
889         free_async_extent_pages(async_extent);
890         kfree(async_extent);
891         goto again;
892 }
893
894 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
895                                       u64 num_bytes)
896 {
897         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
898         struct extent_map *em;
899         u64 alloc_hint = 0;
900
901         read_lock(&em_tree->lock);
902         em = search_extent_mapping(em_tree, start, num_bytes);
903         if (em) {
904                 /*
905                  * if block start isn't an actual block number then find the
906                  * first block in this inode and use that as a hint.  If that
907                  * block is also bogus then just don't worry about it.
908                  */
909                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
910                         free_extent_map(em);
911                         em = search_extent_mapping(em_tree, 0, 0);
912                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
913                                 alloc_hint = em->block_start;
914                         if (em)
915                                 free_extent_map(em);
916                 } else {
917                         alloc_hint = em->block_start;
918                         free_extent_map(em);
919                 }
920         }
921         read_unlock(&em_tree->lock);
922
923         return alloc_hint;
924 }
925
926 /*
927  * when extent_io.c finds a delayed allocation range in the file,
928  * the call backs end up in this code.  The basic idea is to
929  * allocate extents on disk for the range, and create ordered data structs
930  * in ram to track those extents.
931  *
932  * locked_page is the page that writepage had locked already.  We use
933  * it to make sure we don't do extra locks or unlocks.
934  *
935  * *page_started is set to one if we unlock locked_page and do everything
936  * required to start IO on it.  It may be clean and already done with
937  * IO when we return.
938  */
939 static noinline int cow_file_range(struct inode *inode,
940                                    struct page *locked_page,
941                                    u64 start, u64 end, u64 delalloc_end,
942                                    int *page_started, unsigned long *nr_written,
943                                    int unlock, struct btrfs_dedupe_hash *hash)
944 {
945         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
946         struct btrfs_root *root = BTRFS_I(inode)->root;
947         u64 alloc_hint = 0;
948         u64 num_bytes;
949         unsigned long ram_size;
950         u64 cur_alloc_size = 0;
951         u64 blocksize = fs_info->sectorsize;
952         struct btrfs_key ins;
953         struct extent_map *em;
954         unsigned clear_bits;
955         unsigned long page_ops;
956         bool extent_reserved = false;
957         int ret = 0;
958
959         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
960                 WARN_ON_ONCE(1);
961                 ret = -EINVAL;
962                 goto out_unlock;
963         }
964
965         num_bytes = ALIGN(end - start + 1, blocksize);
966         num_bytes = max(blocksize,  num_bytes);
967         ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
968
969         inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
970
971         if (start == 0) {
972                 /* lets try to make an inline extent */
973                 ret = cow_file_range_inline(inode, start, end, 0,
974                                             BTRFS_COMPRESS_NONE, NULL);
975                 if (ret == 0) {
976                         /*
977                          * We use DO_ACCOUNTING here because we need the
978                          * delalloc_release_metadata to be run _after_ we drop
979                          * our outstanding extent for clearing delalloc for this
980                          * range.
981                          */
982                         extent_clear_unlock_delalloc(inode, start, end,
983                                      delalloc_end, NULL,
984                                      EXTENT_LOCKED | EXTENT_DELALLOC |
985                                      EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986                                      EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
987                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
988                                      PAGE_END_WRITEBACK);
989                         *nr_written = *nr_written +
990                              (end - start + PAGE_SIZE) / PAGE_SIZE;
991                         *page_started = 1;
992                         goto out;
993                 } else if (ret < 0) {
994                         goto out_unlock;
995                 }
996         }
997
998         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
999         btrfs_drop_extent_cache(BTRFS_I(inode), start,
1000                         start + num_bytes - 1, 0);
1001
1002         while (num_bytes > 0) {
1003                 cur_alloc_size = num_bytes;
1004                 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1005                                            fs_info->sectorsize, 0, alloc_hint,
1006                                            &ins, 1, 1);
1007                 if (ret < 0)
1008                         goto out_unlock;
1009                 cur_alloc_size = ins.offset;
1010                 extent_reserved = true;
1011
1012                 ram_size = ins.offset;
1013                 em = create_io_em(inode, start, ins.offset, /* len */
1014                                   start, /* orig_start */
1015                                   ins.objectid, /* block_start */
1016                                   ins.offset, /* block_len */
1017                                   ins.offset, /* orig_block_len */
1018                                   ram_size, /* ram_bytes */
1019                                   BTRFS_COMPRESS_NONE, /* compress_type */
1020                                   BTRFS_ORDERED_REGULAR /* type */);
1021                 if (IS_ERR(em))
1022                         goto out_reserve;
1023                 free_extent_map(em);
1024
1025                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1026                                                ram_size, cur_alloc_size, 0);
1027                 if (ret)
1028                         goto out_drop_extent_cache;
1029
1030                 if (root->root_key.objectid ==
1031                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1032                         ret = btrfs_reloc_clone_csums(inode, start,
1033                                                       cur_alloc_size);
1034                         /*
1035                          * Only drop cache here, and process as normal.
1036                          *
1037                          * We must not allow extent_clear_unlock_delalloc()
1038                          * at out_unlock label to free meta of this ordered
1039                          * extent, as its meta should be freed by
1040                          * btrfs_finish_ordered_io().
1041                          *
1042                          * So we must continue until @start is increased to
1043                          * skip current ordered extent.
1044                          */
1045                         if (ret)
1046                                 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1047                                                 start + ram_size - 1, 0);
1048                 }
1049
1050                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1051
1052                 /* we're not doing compressed IO, don't unlock the first
1053                  * page (which the caller expects to stay locked), don't
1054                  * clear any dirty bits and don't set any writeback bits
1055                  *
1056                  * Do set the Private2 bit so we know this page was properly
1057                  * setup for writepage
1058                  */
1059                 page_ops = unlock ? PAGE_UNLOCK : 0;
1060                 page_ops |= PAGE_SET_PRIVATE2;
1061
1062                 extent_clear_unlock_delalloc(inode, start,
1063                                              start + ram_size - 1,
1064                                              delalloc_end, locked_page,
1065                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1066                                              page_ops);
1067                 if (num_bytes < cur_alloc_size)
1068                         num_bytes = 0;
1069                 else
1070                         num_bytes -= cur_alloc_size;
1071                 alloc_hint = ins.objectid + ins.offset;
1072                 start += cur_alloc_size;
1073                 extent_reserved = false;
1074
1075                 /*
1076                  * btrfs_reloc_clone_csums() error, since start is increased
1077                  * extent_clear_unlock_delalloc() at out_unlock label won't
1078                  * free metadata of current ordered extent, we're OK to exit.
1079                  */
1080                 if (ret)
1081                         goto out_unlock;
1082         }
1083 out:
1084         return ret;
1085
1086 out_drop_extent_cache:
1087         btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1088 out_reserve:
1089         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1090         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1091 out_unlock:
1092         clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1093                 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1094         page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1095                 PAGE_END_WRITEBACK;
1096         /*
1097          * If we reserved an extent for our delalloc range (or a subrange) and
1098          * failed to create the respective ordered extent, then it means that
1099          * when we reserved the extent we decremented the extent's size from
1100          * the data space_info's bytes_may_use counter and incremented the
1101          * space_info's bytes_reserved counter by the same amount. We must make
1102          * sure extent_clear_unlock_delalloc() does not try to decrement again
1103          * the data space_info's bytes_may_use counter, therefore we do not pass
1104          * it the flag EXTENT_CLEAR_DATA_RESV.
1105          */
1106         if (extent_reserved) {
1107                 extent_clear_unlock_delalloc(inode, start,
1108                                              start + cur_alloc_size,
1109                                              start + cur_alloc_size,
1110                                              locked_page,
1111                                              clear_bits,
1112                                              page_ops);
1113                 start += cur_alloc_size;
1114                 if (start >= end)
1115                         goto out;
1116         }
1117         extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1118                                      locked_page,
1119                                      clear_bits | EXTENT_CLEAR_DATA_RESV,
1120                                      page_ops);
1121         goto out;
1122 }
1123
1124 /*
1125  * work queue call back to started compression on a file and pages
1126  */
1127 static noinline void async_cow_start(struct btrfs_work *work)
1128 {
1129         struct async_cow *async_cow;
1130         int num_added = 0;
1131         async_cow = container_of(work, struct async_cow, work);
1132
1133         compress_file_range(async_cow->inode, async_cow->locked_page,
1134                             async_cow->start, async_cow->end, async_cow,
1135                             &num_added);
1136         if (num_added == 0) {
1137                 btrfs_add_delayed_iput(async_cow->inode);
1138                 async_cow->inode = NULL;
1139         }
1140 }
1141
1142 /*
1143  * work queue call back to submit previously compressed pages
1144  */
1145 static noinline void async_cow_submit(struct btrfs_work *work)
1146 {
1147         struct btrfs_fs_info *fs_info;
1148         struct async_cow *async_cow;
1149         struct btrfs_root *root;
1150         unsigned long nr_pages;
1151
1152         async_cow = container_of(work, struct async_cow, work);
1153
1154         root = async_cow->root;
1155         fs_info = root->fs_info;
1156         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157                 PAGE_SHIFT;
1158
1159         /* atomic_sub_return implies a barrier */
1160         if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1161             5 * SZ_1M)
1162                 cond_wake_up_nomb(&fs_info->async_submit_wait);
1163
1164         if (async_cow->inode)
1165                 submit_compressed_extents(async_cow->inode, async_cow);
1166 }
1167
1168 static noinline void async_cow_free(struct btrfs_work *work)
1169 {
1170         struct async_cow *async_cow;
1171         async_cow = container_of(work, struct async_cow, work);
1172         if (async_cow->inode)
1173                 btrfs_add_delayed_iput(async_cow->inode);
1174         kfree(async_cow);
1175 }
1176
1177 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1178                                 u64 start, u64 end, int *page_started,
1179                                 unsigned long *nr_written,
1180                                 unsigned int write_flags)
1181 {
1182         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1183         struct async_cow *async_cow;
1184         struct btrfs_root *root = BTRFS_I(inode)->root;
1185         unsigned long nr_pages;
1186         u64 cur_end;
1187
1188         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1189                          1, 0, NULL);
1190         while (start < end) {
1191                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1192                 BUG_ON(!async_cow); /* -ENOMEM */
1193                 async_cow->inode = igrab(inode);
1194                 async_cow->root = root;
1195                 async_cow->locked_page = locked_page;
1196                 async_cow->start = start;
1197                 async_cow->write_flags = write_flags;
1198
1199                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1200                     !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1201                         cur_end = end;
1202                 else
1203                         cur_end = min(end, start + SZ_512K - 1);
1204
1205                 async_cow->end = cur_end;
1206                 INIT_LIST_HEAD(&async_cow->extents);
1207
1208                 btrfs_init_work(&async_cow->work,
1209                                 btrfs_delalloc_helper,
1210                                 async_cow_start, async_cow_submit,
1211                                 async_cow_free);
1212
1213                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1214                         PAGE_SHIFT;
1215                 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1216
1217                 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1218
1219                 *nr_written += nr_pages;
1220                 start = cur_end + 1;
1221         }
1222         *page_started = 1;
1223         return 0;
1224 }
1225
1226 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1227                                         u64 bytenr, u64 num_bytes)
1228 {
1229         int ret;
1230         struct btrfs_ordered_sum *sums;
1231         LIST_HEAD(list);
1232
1233         ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1234                                        bytenr + num_bytes - 1, &list, 0);
1235         if (ret == 0 && list_empty(&list))
1236                 return 0;
1237
1238         while (!list_empty(&list)) {
1239                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1240                 list_del(&sums->list);
1241                 kfree(sums);
1242         }
1243         if (ret < 0)
1244                 return ret;
1245         return 1;
1246 }
1247
1248 /*
1249  * when nowcow writeback call back.  This checks for snapshots or COW copies
1250  * of the extents that exist in the file, and COWs the file as required.
1251  *
1252  * If no cow copies or snapshots exist, we write directly to the existing
1253  * blocks on disk
1254  */
1255 static noinline int run_delalloc_nocow(struct inode *inode,
1256                                        struct page *locked_page,
1257                               u64 start, u64 end, int *page_started, int force,
1258                               unsigned long *nr_written)
1259 {
1260         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1261         struct btrfs_root *root = BTRFS_I(inode)->root;
1262         struct extent_buffer *leaf;
1263         struct btrfs_path *path;
1264         struct btrfs_file_extent_item *fi;
1265         struct btrfs_key found_key;
1266         struct extent_map *em;
1267         u64 cow_start;
1268         u64 cur_offset;
1269         u64 extent_end;
1270         u64 extent_offset;
1271         u64 disk_bytenr;
1272         u64 num_bytes;
1273         u64 disk_num_bytes;
1274         u64 ram_bytes;
1275         int extent_type;
1276         int ret, err;
1277         int type;
1278         int nocow;
1279         int check_prev = 1;
1280         bool nolock;
1281         u64 ino = btrfs_ino(BTRFS_I(inode));
1282
1283         path = btrfs_alloc_path();
1284         if (!path) {
1285                 extent_clear_unlock_delalloc(inode, start, end, end,
1286                                              locked_page,
1287                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1288                                              EXTENT_DO_ACCOUNTING |
1289                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1290                                              PAGE_CLEAR_DIRTY |
1291                                              PAGE_SET_WRITEBACK |
1292                                              PAGE_END_WRITEBACK);
1293                 return -ENOMEM;
1294         }
1295
1296         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1297
1298         cow_start = (u64)-1;
1299         cur_offset = start;
1300         while (1) {
1301                 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1302                                                cur_offset, 0);
1303                 if (ret < 0)
1304                         goto error;
1305                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1306                         leaf = path->nodes[0];
1307                         btrfs_item_key_to_cpu(leaf, &found_key,
1308                                               path->slots[0] - 1);
1309                         if (found_key.objectid == ino &&
1310                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1311                                 path->slots[0]--;
1312                 }
1313                 check_prev = 0;
1314 next_slot:
1315                 leaf = path->nodes[0];
1316                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1317                         ret = btrfs_next_leaf(root, path);
1318                         if (ret < 0) {
1319                                 if (cow_start != (u64)-1)
1320                                         cur_offset = cow_start;
1321                                 goto error;
1322                         }
1323                         if (ret > 0)
1324                                 break;
1325                         leaf = path->nodes[0];
1326                 }
1327
1328                 nocow = 0;
1329                 disk_bytenr = 0;
1330                 num_bytes = 0;
1331                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1332
1333                 if (found_key.objectid > ino)
1334                         break;
1335                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1336                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1337                         path->slots[0]++;
1338                         goto next_slot;
1339                 }
1340                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1341                     found_key.offset > end)
1342                         break;
1343
1344                 if (found_key.offset > cur_offset) {
1345                         extent_end = found_key.offset;
1346                         extent_type = 0;
1347                         goto out_check;
1348                 }
1349
1350                 fi = btrfs_item_ptr(leaf, path->slots[0],
1351                                     struct btrfs_file_extent_item);
1352                 extent_type = btrfs_file_extent_type(leaf, fi);
1353
1354                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1355                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1356                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1357                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1358                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1359                         extent_end = found_key.offset +
1360                                 btrfs_file_extent_num_bytes(leaf, fi);
1361                         disk_num_bytes =
1362                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1363                         if (extent_end <= start) {
1364                                 path->slots[0]++;
1365                                 goto next_slot;
1366                         }
1367                         if (disk_bytenr == 0)
1368                                 goto out_check;
1369                         if (btrfs_file_extent_compression(leaf, fi) ||
1370                             btrfs_file_extent_encryption(leaf, fi) ||
1371                             btrfs_file_extent_other_encoding(leaf, fi))
1372                                 goto out_check;
1373                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1374                                 goto out_check;
1375                         if (btrfs_extent_readonly(fs_info, disk_bytenr))
1376                                 goto out_check;
1377                         ret = btrfs_cross_ref_exist(root, ino,
1378                                                     found_key.offset -
1379                                                     extent_offset, disk_bytenr);
1380                         if (ret) {
1381                                 /*
1382                                  * ret could be -EIO if the above fails to read
1383                                  * metadata.
1384                                  */
1385                                 if (ret < 0) {
1386                                         if (cow_start != (u64)-1)
1387                                                 cur_offset = cow_start;
1388                                         goto error;
1389                                 }
1390
1391                                 WARN_ON_ONCE(nolock);
1392                                 goto out_check;
1393                         }
1394                         disk_bytenr += extent_offset;
1395                         disk_bytenr += cur_offset - found_key.offset;
1396                         num_bytes = min(end + 1, extent_end) - cur_offset;
1397                         /*
1398                          * if there are pending snapshots for this root,
1399                          * we fall into common COW way.
1400                          */
1401                         if (!nolock) {
1402                                 err = btrfs_start_write_no_snapshotting(root);
1403                                 if (!err)
1404                                         goto out_check;
1405                         }
1406                         /*
1407                          * force cow if csum exists in the range.
1408                          * this ensure that csum for a given extent are
1409                          * either valid or do not exist.
1410                          */
1411                         ret = csum_exist_in_range(fs_info, disk_bytenr,
1412                                                   num_bytes);
1413                         if (ret) {
1414                                 if (!nolock)
1415                                         btrfs_end_write_no_snapshotting(root);
1416
1417                                 /*
1418                                  * ret could be -EIO if the above fails to read
1419                                  * metadata.
1420                                  */
1421                                 if (ret < 0) {
1422                                         if (cow_start != (u64)-1)
1423                                                 cur_offset = cow_start;
1424                                         goto error;
1425                                 }
1426                                 WARN_ON_ONCE(nolock);
1427                                 goto out_check;
1428                         }
1429                         if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1430                                 if (!nolock)
1431                                         btrfs_end_write_no_snapshotting(root);
1432                                 goto out_check;
1433                         }
1434                         nocow = 1;
1435                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1436                         extent_end = found_key.offset +
1437                                 btrfs_file_extent_inline_len(leaf,
1438                                                      path->slots[0], fi);
1439                         extent_end = ALIGN(extent_end,
1440                                            fs_info->sectorsize);
1441                 } else {
1442                         BUG_ON(1);
1443                 }
1444 out_check:
1445                 if (extent_end <= start) {
1446                         path->slots[0]++;
1447                         if (!nolock && nocow)
1448                                 btrfs_end_write_no_snapshotting(root);
1449                         if (nocow)
1450                                 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1451                         goto next_slot;
1452                 }
1453                 if (!nocow) {
1454                         if (cow_start == (u64)-1)
1455                                 cow_start = cur_offset;
1456                         cur_offset = extent_end;
1457                         if (cur_offset > end)
1458                                 break;
1459                         path->slots[0]++;
1460                         goto next_slot;
1461                 }
1462
1463                 btrfs_release_path(path);
1464                 if (cow_start != (u64)-1) {
1465                         ret = cow_file_range(inode, locked_page,
1466                                              cow_start, found_key.offset - 1,
1467                                              end, page_started, nr_written, 1,
1468                                              NULL);
1469                         if (ret) {
1470                                 if (!nolock && nocow)
1471                                         btrfs_end_write_no_snapshotting(root);
1472                                 if (nocow)
1473                                         btrfs_dec_nocow_writers(fs_info,
1474                                                                 disk_bytenr);
1475                                 goto error;
1476                         }
1477                         cow_start = (u64)-1;
1478                 }
1479
1480                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1481                         u64 orig_start = found_key.offset - extent_offset;
1482
1483                         em = create_io_em(inode, cur_offset, num_bytes,
1484                                           orig_start,
1485                                           disk_bytenr, /* block_start */
1486                                           num_bytes, /* block_len */
1487                                           disk_num_bytes, /* orig_block_len */
1488                                           ram_bytes, BTRFS_COMPRESS_NONE,
1489                                           BTRFS_ORDERED_PREALLOC);
1490                         if (IS_ERR(em)) {
1491                                 if (!nolock && nocow)
1492                                         btrfs_end_write_no_snapshotting(root);
1493                                 if (nocow)
1494                                         btrfs_dec_nocow_writers(fs_info,
1495                                                                 disk_bytenr);
1496                                 ret = PTR_ERR(em);
1497                                 goto error;
1498                         }
1499                         free_extent_map(em);
1500                 }
1501
1502                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1503                         type = BTRFS_ORDERED_PREALLOC;
1504                 } else {
1505                         type = BTRFS_ORDERED_NOCOW;
1506                 }
1507
1508                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1509                                                num_bytes, num_bytes, type);
1510                 if (nocow)
1511                         btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1512                 BUG_ON(ret); /* -ENOMEM */
1513
1514                 if (root->root_key.objectid ==
1515                     BTRFS_DATA_RELOC_TREE_OBJECTID)
1516                         /*
1517                          * Error handled later, as we must prevent
1518                          * extent_clear_unlock_delalloc() in error handler
1519                          * from freeing metadata of created ordered extent.
1520                          */
1521                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1522                                                       num_bytes);
1523
1524                 extent_clear_unlock_delalloc(inode, cur_offset,
1525                                              cur_offset + num_bytes - 1, end,
1526                                              locked_page, EXTENT_LOCKED |
1527                                              EXTENT_DELALLOC |
1528                                              EXTENT_CLEAR_DATA_RESV,
1529                                              PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1530
1531                 if (!nolock && nocow)
1532                         btrfs_end_write_no_snapshotting(root);
1533                 cur_offset = extent_end;
1534
1535                 /*
1536                  * btrfs_reloc_clone_csums() error, now we're OK to call error
1537                  * handler, as metadata for created ordered extent will only
1538                  * be freed by btrfs_finish_ordered_io().
1539                  */
1540                 if (ret)
1541                         goto error;
1542                 if (cur_offset > end)
1543                         break;
1544         }
1545         btrfs_release_path(path);
1546
1547         if (cur_offset <= end && cow_start == (u64)-1) {
1548                 cow_start = cur_offset;
1549                 cur_offset = end;
1550         }
1551
1552         if (cow_start != (u64)-1) {
1553                 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1554                                      page_started, nr_written, 1, NULL);
1555                 if (ret)
1556                         goto error;
1557         }
1558
1559 error:
1560         if (ret && cur_offset < end)
1561                 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1562                                              locked_page, EXTENT_LOCKED |
1563                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1564                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1565                                              PAGE_CLEAR_DIRTY |
1566                                              PAGE_SET_WRITEBACK |
1567                                              PAGE_END_WRITEBACK);
1568         btrfs_free_path(path);
1569         return ret;
1570 }
1571
1572 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1573 {
1574
1575         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1576             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1577                 return 0;
1578
1579         /*
1580          * @defrag_bytes is a hint value, no spinlock held here,
1581          * if is not zero, it means the file is defragging.
1582          * Force cow if given extent needs to be defragged.
1583          */
1584         if (BTRFS_I(inode)->defrag_bytes &&
1585             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1586                            EXTENT_DEFRAG, 0, NULL))
1587                 return 1;
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * extent_io.c call back to do delayed allocation processing
1594  */
1595 static int run_delalloc_range(void *private_data, struct page *locked_page,
1596                               u64 start, u64 end, int *page_started,
1597                               unsigned long *nr_written,
1598                               struct writeback_control *wbc)
1599 {
1600         struct inode *inode = private_data;
1601         int ret;
1602         int force_cow = need_force_cow(inode, start, end);
1603         unsigned int write_flags = wbc_to_write_flags(wbc);
1604
1605         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1606                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1607                                          page_started, 1, nr_written);
1608         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1609                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1610                                          page_started, 0, nr_written);
1611         } else if (!inode_need_compress(inode, start, end)) {
1612                 ret = cow_file_range(inode, locked_page, start, end, end,
1613                                       page_started, nr_written, 1, NULL);
1614         } else {
1615                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1616                         &BTRFS_I(inode)->runtime_flags);
1617                 ret = cow_file_range_async(inode, locked_page, start, end,
1618                                            page_started, nr_written,
1619                                            write_flags);
1620         }
1621         if (ret)
1622                 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1623         return ret;
1624 }
1625
1626 static void btrfs_split_extent_hook(void *private_data,
1627                                     struct extent_state *orig, u64 split)
1628 {
1629         struct inode *inode = private_data;
1630         u64 size;
1631
1632         /* not delalloc, ignore it */
1633         if (!(orig->state & EXTENT_DELALLOC))
1634                 return;
1635
1636         size = orig->end - orig->start + 1;
1637         if (size > BTRFS_MAX_EXTENT_SIZE) {
1638                 u32 num_extents;
1639                 u64 new_size;
1640
1641                 /*
1642                  * See the explanation in btrfs_merge_extent_hook, the same
1643                  * applies here, just in reverse.
1644                  */
1645                 new_size = orig->end - split + 1;
1646                 num_extents = count_max_extents(new_size);
1647                 new_size = split - orig->start;
1648                 num_extents += count_max_extents(new_size);
1649                 if (count_max_extents(size) >= num_extents)
1650                         return;
1651         }
1652
1653         spin_lock(&BTRFS_I(inode)->lock);
1654         btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1655         spin_unlock(&BTRFS_I(inode)->lock);
1656 }
1657
1658 /*
1659  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1660  * extents so we can keep track of new extents that are just merged onto old
1661  * extents, such as when we are doing sequential writes, so we can properly
1662  * account for the metadata space we'll need.
1663  */
1664 static void btrfs_merge_extent_hook(void *private_data,
1665                                     struct extent_state *new,
1666                                     struct extent_state *other)
1667 {
1668         struct inode *inode = private_data;
1669         u64 new_size, old_size;
1670         u32 num_extents;
1671
1672         /* not delalloc, ignore it */
1673         if (!(other->state & EXTENT_DELALLOC))
1674                 return;
1675
1676         if (new->start > other->start)
1677                 new_size = new->end - other->start + 1;
1678         else
1679                 new_size = other->end - new->start + 1;
1680
1681         /* we're not bigger than the max, unreserve the space and go */
1682         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1683                 spin_lock(&BTRFS_I(inode)->lock);
1684                 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1685                 spin_unlock(&BTRFS_I(inode)->lock);
1686                 return;
1687         }
1688
1689         /*
1690          * We have to add up either side to figure out how many extents were
1691          * accounted for before we merged into one big extent.  If the number of
1692          * extents we accounted for is <= the amount we need for the new range
1693          * then we can return, otherwise drop.  Think of it like this
1694          *
1695          * [ 4k][MAX_SIZE]
1696          *
1697          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1698          * need 2 outstanding extents, on one side we have 1 and the other side
1699          * we have 1 so they are == and we can return.  But in this case
1700          *
1701          * [MAX_SIZE+4k][MAX_SIZE+4k]
1702          *
1703          * Each range on their own accounts for 2 extents, but merged together
1704          * they are only 3 extents worth of accounting, so we need to drop in
1705          * this case.
1706          */
1707         old_size = other->end - other->start + 1;
1708         num_extents = count_max_extents(old_size);
1709         old_size = new->end - new->start + 1;
1710         num_extents += count_max_extents(old_size);
1711         if (count_max_extents(new_size) >= num_extents)
1712                 return;
1713
1714         spin_lock(&BTRFS_I(inode)->lock);
1715         btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1716         spin_unlock(&BTRFS_I(inode)->lock);
1717 }
1718
1719 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1720                                       struct inode *inode)
1721 {
1722         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1723
1724         spin_lock(&root->delalloc_lock);
1725         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1726                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1727                               &root->delalloc_inodes);
1728                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1729                         &BTRFS_I(inode)->runtime_flags);
1730                 root->nr_delalloc_inodes++;
1731                 if (root->nr_delalloc_inodes == 1) {
1732                         spin_lock(&fs_info->delalloc_root_lock);
1733                         BUG_ON(!list_empty(&root->delalloc_root));
1734                         list_add_tail(&root->delalloc_root,
1735                                       &fs_info->delalloc_roots);
1736                         spin_unlock(&fs_info->delalloc_root_lock);
1737                 }
1738         }
1739         spin_unlock(&root->delalloc_lock);
1740 }
1741
1742
1743 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1744                                 struct btrfs_inode *inode)
1745 {
1746         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1747
1748         if (!list_empty(&inode->delalloc_inodes)) {
1749                 list_del_init(&inode->delalloc_inodes);
1750                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1751                           &inode->runtime_flags);
1752                 root->nr_delalloc_inodes--;
1753                 if (!root->nr_delalloc_inodes) {
1754                         ASSERT(list_empty(&root->delalloc_inodes));
1755                         spin_lock(&fs_info->delalloc_root_lock);
1756                         BUG_ON(list_empty(&root->delalloc_root));
1757                         list_del_init(&root->delalloc_root);
1758                         spin_unlock(&fs_info->delalloc_root_lock);
1759                 }
1760         }
1761 }
1762
1763 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1764                                      struct btrfs_inode *inode)
1765 {
1766         spin_lock(&root->delalloc_lock);
1767         __btrfs_del_delalloc_inode(root, inode);
1768         spin_unlock(&root->delalloc_lock);
1769 }
1770
1771 /*
1772  * extent_io.c set_bit_hook, used to track delayed allocation
1773  * bytes in this file, and to maintain the list of inodes that
1774  * have pending delalloc work to be done.
1775  */
1776 static void btrfs_set_bit_hook(void *private_data,
1777                                struct extent_state *state, unsigned *bits)
1778 {
1779         struct inode *inode = private_data;
1780
1781         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1782
1783         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1784                 WARN_ON(1);
1785         /*
1786          * set_bit and clear bit hooks normally require _irqsave/restore
1787          * but in this case, we are only testing for the DELALLOC
1788          * bit, which is only set or cleared with irqs on
1789          */
1790         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1791                 struct btrfs_root *root = BTRFS_I(inode)->root;
1792                 u64 len = state->end + 1 - state->start;
1793                 u32 num_extents = count_max_extents(len);
1794                 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1795
1796                 spin_lock(&BTRFS_I(inode)->lock);
1797                 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1798                 spin_unlock(&BTRFS_I(inode)->lock);
1799
1800                 /* For sanity tests */
1801                 if (btrfs_is_testing(fs_info))
1802                         return;
1803
1804                 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1805                                          fs_info->delalloc_batch);
1806                 spin_lock(&BTRFS_I(inode)->lock);
1807                 BTRFS_I(inode)->delalloc_bytes += len;
1808                 if (*bits & EXTENT_DEFRAG)
1809                         BTRFS_I(inode)->defrag_bytes += len;
1810                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1811                                          &BTRFS_I(inode)->runtime_flags))
1812                         btrfs_add_delalloc_inodes(root, inode);
1813                 spin_unlock(&BTRFS_I(inode)->lock);
1814         }
1815
1816         if (!(state->state & EXTENT_DELALLOC_NEW) &&
1817             (*bits & EXTENT_DELALLOC_NEW)) {
1818                 spin_lock(&BTRFS_I(inode)->lock);
1819                 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1820                         state->start;
1821                 spin_unlock(&BTRFS_I(inode)->lock);
1822         }
1823 }
1824
1825 /*
1826  * extent_io.c clear_bit_hook, see set_bit_hook for why
1827  */
1828 static void btrfs_clear_bit_hook(void *private_data,
1829                                  struct extent_state *state,
1830                                  unsigned *bits)
1831 {
1832         struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1833         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1834         u64 len = state->end + 1 - state->start;
1835         u32 num_extents = count_max_extents(len);
1836
1837         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1838                 spin_lock(&inode->lock);
1839                 inode->defrag_bytes -= len;
1840                 spin_unlock(&inode->lock);
1841         }
1842
1843         /*
1844          * set_bit and clear bit hooks normally require _irqsave/restore
1845          * but in this case, we are only testing for the DELALLOC
1846          * bit, which is only set or cleared with irqs on
1847          */
1848         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1849                 struct btrfs_root *root = inode->root;
1850                 bool do_list = !btrfs_is_free_space_inode(inode);
1851
1852                 spin_lock(&inode->lock);
1853                 btrfs_mod_outstanding_extents(inode, -num_extents);
1854                 spin_unlock(&inode->lock);
1855
1856                 /*
1857                  * We don't reserve metadata space for space cache inodes so we
1858                  * don't need to call dellalloc_release_metadata if there is an
1859                  * error.
1860                  */
1861                 if (*bits & EXTENT_CLEAR_META_RESV &&
1862                     root != fs_info->tree_root)
1863                         btrfs_delalloc_release_metadata(inode, len, false);
1864
1865                 /* For sanity tests. */
1866                 if (btrfs_is_testing(fs_info))
1867                         return;
1868
1869                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1870                     do_list && !(state->state & EXTENT_NORESERVE) &&
1871                     (*bits & EXTENT_CLEAR_DATA_RESV))
1872                         btrfs_free_reserved_data_space_noquota(
1873                                         &inode->vfs_inode,
1874                                         state->start, len);
1875
1876                 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1877                                          fs_info->delalloc_batch);
1878                 spin_lock(&inode->lock);
1879                 inode->delalloc_bytes -= len;
1880                 if (do_list && inode->delalloc_bytes == 0 &&
1881                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1882                                         &inode->runtime_flags))
1883                         btrfs_del_delalloc_inode(root, inode);
1884                 spin_unlock(&inode->lock);
1885         }
1886
1887         if ((state->state & EXTENT_DELALLOC_NEW) &&
1888             (*bits & EXTENT_DELALLOC_NEW)) {
1889                 spin_lock(&inode->lock);
1890                 ASSERT(inode->new_delalloc_bytes >= len);
1891                 inode->new_delalloc_bytes -= len;
1892                 spin_unlock(&inode->lock);
1893         }
1894 }
1895
1896 /*
1897  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1898  * we don't create bios that span stripes or chunks
1899  *
1900  * return 1 if page cannot be merged to bio
1901  * return 0 if page can be merged to bio
1902  * return error otherwise
1903  */
1904 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1905                          size_t size, struct bio *bio,
1906                          unsigned long bio_flags)
1907 {
1908         struct inode *inode = page->mapping->host;
1909         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1910         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1911         u64 length = 0;
1912         u64 map_length;
1913         int ret;
1914
1915         if (bio_flags & EXTENT_BIO_COMPRESSED)
1916                 return 0;
1917
1918         length = bio->bi_iter.bi_size;
1919         map_length = length;
1920         ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1921                               NULL, 0);
1922         if (ret < 0)
1923                 return ret;
1924         if (map_length < length + size)
1925                 return 1;
1926         return 0;
1927 }
1928
1929 /*
1930  * in order to insert checksums into the metadata in large chunks,
1931  * we wait until bio submission time.   All the pages in the bio are
1932  * checksummed and sums are attached onto the ordered extent record.
1933  *
1934  * At IO completion time the cums attached on the ordered extent record
1935  * are inserted into the btree
1936  */
1937 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1938                                     u64 bio_offset)
1939 {
1940         struct inode *inode = private_data;
1941         blk_status_t ret = 0;
1942
1943         ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1944         BUG_ON(ret); /* -ENOMEM */
1945         return 0;
1946 }
1947
1948 /*
1949  * in order to insert checksums into the metadata in large chunks,
1950  * we wait until bio submission time.   All the pages in the bio are
1951  * checksummed and sums are attached onto the ordered extent record.
1952  *
1953  * At IO completion time the cums attached on the ordered extent record
1954  * are inserted into the btree
1955  */
1956 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1957                           int mirror_num)
1958 {
1959         struct inode *inode = private_data;
1960         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1961         blk_status_t ret;
1962
1963         ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1964         if (ret) {
1965                 bio->bi_status = ret;
1966                 bio_endio(bio);
1967         }
1968         return ret;
1969 }
1970
1971 /*
1972  * extent_io.c submission hook. This does the right thing for csum calculation
1973  * on write, or reading the csums from the tree before a read.
1974  *
1975  * Rules about async/sync submit,
1976  * a) read:                             sync submit
1977  *
1978  * b) write without checksum:           sync submit
1979  *
1980  * c) write with checksum:
1981  *    c-1) if bio is issued by fsync:   sync submit
1982  *         (sync_writers != 0)
1983  *
1984  *    c-2) if root is reloc root:       sync submit
1985  *         (only in case of buffered IO)
1986  *
1987  *    c-3) otherwise:                   async submit
1988  */
1989 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1990                                  int mirror_num, unsigned long bio_flags,
1991                                  u64 bio_offset)
1992 {
1993         struct inode *inode = private_data;
1994         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1995         struct btrfs_root *root = BTRFS_I(inode)->root;
1996         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1997         blk_status_t ret = 0;
1998         int skip_sum;
1999         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2000
2001         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2002
2003         if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2004                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2005
2006         if (bio_op(bio) != REQ_OP_WRITE) {
2007                 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2008                 if (ret)
2009                         goto out;
2010
2011                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2012                         ret = btrfs_submit_compressed_read(inode, bio,
2013                                                            mirror_num,
2014                                                            bio_flags);
2015                         goto out;
2016                 } else if (!skip_sum) {
2017                         ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2018                         if (ret)
2019                                 goto out;
2020                 }
2021                 goto mapit;
2022         } else if (async && !skip_sum) {
2023                 /* csum items have already been cloned */
2024                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2025                         goto mapit;
2026                 /* we're doing a write, do the async checksumming */
2027                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2028                                           bio_offset, inode,
2029                                           btrfs_submit_bio_start,
2030                                           btrfs_submit_bio_done);
2031                 goto out;
2032         } else if (!skip_sum) {
2033                 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2034                 if (ret)
2035                         goto out;
2036         }
2037
2038 mapit:
2039         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2040
2041 out:
2042         if (ret) {
2043                 bio->bi_status = ret;
2044                 bio_endio(bio);
2045         }
2046         return ret;
2047 }
2048
2049 /*
2050  * given a list of ordered sums record them in the inode.  This happens
2051  * at IO completion time based on sums calculated at bio submission time.
2052  */
2053 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2054                              struct inode *inode, struct list_head *list)
2055 {
2056         struct btrfs_ordered_sum *sum;
2057         int ret;
2058
2059         list_for_each_entry(sum, list, list) {
2060                 trans->adding_csums = true;
2061                 ret = btrfs_csum_file_blocks(trans,
2062                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
2063                 trans->adding_csums = false;
2064                 if (ret)
2065                         return ret;
2066         }
2067         return 0;
2068 }
2069
2070 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2071                               unsigned int extra_bits,
2072                               struct extent_state **cached_state, int dedupe)
2073 {
2074         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2075         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2076                                    extra_bits, cached_state);
2077 }
2078
2079 /* see btrfs_writepage_start_hook for details on why this is required */
2080 struct btrfs_writepage_fixup {
2081         struct page *page;
2082         struct btrfs_work work;
2083 };
2084
2085 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2086 {
2087         struct btrfs_writepage_fixup *fixup;
2088         struct btrfs_ordered_extent *ordered;
2089         struct extent_state *cached_state = NULL;
2090         struct extent_changeset *data_reserved = NULL;
2091         struct page *page;
2092         struct inode *inode;
2093         u64 page_start;
2094         u64 page_end;
2095         int ret;
2096
2097         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2098         page = fixup->page;
2099 again:
2100         lock_page(page);
2101         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2102                 ClearPageChecked(page);
2103                 goto out_page;
2104         }
2105
2106         inode = page->mapping->host;
2107         page_start = page_offset(page);
2108         page_end = page_offset(page) + PAGE_SIZE - 1;
2109
2110         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2111                          &cached_state);
2112
2113         /* already ordered? We're done */
2114         if (PagePrivate2(page))
2115                 goto out;
2116
2117         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2118                                         PAGE_SIZE);
2119         if (ordered) {
2120                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2121                                      page_end, &cached_state);
2122                 unlock_page(page);
2123                 btrfs_start_ordered_extent(inode, ordered, 1);
2124                 btrfs_put_ordered_extent(ordered);
2125                 goto again;
2126         }
2127
2128         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2129                                            PAGE_SIZE);
2130         if (ret) {
2131                 mapping_set_error(page->mapping, ret);
2132                 end_extent_writepage(page, ret, page_start, page_end);
2133                 ClearPageChecked(page);
2134                 goto out;
2135          }
2136
2137         ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2138                                         &cached_state, 0);
2139         if (ret) {
2140                 mapping_set_error(page->mapping, ret);
2141                 end_extent_writepage(page, ret, page_start, page_end);
2142                 ClearPageChecked(page);
2143                 goto out;
2144         }
2145
2146         ClearPageChecked(page);
2147         set_page_dirty(page);
2148         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2149 out:
2150         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2151                              &cached_state);
2152 out_page:
2153         unlock_page(page);
2154         put_page(page);
2155         kfree(fixup);
2156         extent_changeset_free(data_reserved);
2157 }
2158
2159 /*
2160  * There are a few paths in the higher layers of the kernel that directly
2161  * set the page dirty bit without asking the filesystem if it is a
2162  * good idea.  This causes problems because we want to make sure COW
2163  * properly happens and the data=ordered rules are followed.
2164  *
2165  * In our case any range that doesn't have the ORDERED bit set
2166  * hasn't been properly setup for IO.  We kick off an async process
2167  * to fix it up.  The async helper will wait for ordered extents, set
2168  * the delalloc bit and make it safe to write the page.
2169  */
2170 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2171 {
2172         struct inode *inode = page->mapping->host;
2173         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2174         struct btrfs_writepage_fixup *fixup;
2175
2176         /* this page is properly in the ordered list */
2177         if (TestClearPagePrivate2(page))
2178                 return 0;
2179
2180         if (PageChecked(page))
2181                 return -EAGAIN;
2182
2183         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2184         if (!fixup)
2185                 return -EAGAIN;
2186
2187         SetPageChecked(page);
2188         get_page(page);
2189         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2190                         btrfs_writepage_fixup_worker, NULL, NULL);
2191         fixup->page = page;
2192         btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2193         return -EBUSY;
2194 }
2195
2196 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2197                                        struct inode *inode, u64 file_pos,
2198                                        u64 disk_bytenr, u64 disk_num_bytes,
2199                                        u64 num_bytes, u64 ram_bytes,
2200                                        u8 compression, u8 encryption,
2201                                        u16 other_encoding, int extent_type)
2202 {
2203         struct btrfs_root *root = BTRFS_I(inode)->root;
2204         struct btrfs_file_extent_item *fi;
2205         struct btrfs_path *path;
2206         struct extent_buffer *leaf;
2207         struct btrfs_key ins;
2208         u64 qg_released;
2209         int extent_inserted = 0;
2210         int ret;
2211
2212         path = btrfs_alloc_path();
2213         if (!path)
2214                 return -ENOMEM;
2215
2216         /*
2217          * we may be replacing one extent in the tree with another.
2218          * The new extent is pinned in the extent map, and we don't want
2219          * to drop it from the cache until it is completely in the btree.
2220          *
2221          * So, tell btrfs_drop_extents to leave this extent in the cache.
2222          * the caller is expected to unpin it and allow it to be merged
2223          * with the others.
2224          */
2225         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2226                                    file_pos + num_bytes, NULL, 0,
2227                                    1, sizeof(*fi), &extent_inserted);
2228         if (ret)
2229                 goto out;
2230
2231         if (!extent_inserted) {
2232                 ins.objectid = btrfs_ino(BTRFS_I(inode));
2233                 ins.offset = file_pos;
2234                 ins.type = BTRFS_EXTENT_DATA_KEY;
2235
2236                 path->leave_spinning = 1;
2237                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2238                                               sizeof(*fi));
2239                 if (ret)
2240                         goto out;
2241         }
2242         leaf = path->nodes[0];
2243         fi = btrfs_item_ptr(leaf, path->slots[0],
2244                             struct btrfs_file_extent_item);
2245         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2246         btrfs_set_file_extent_type(leaf, fi, extent_type);
2247         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2248         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2249         btrfs_set_file_extent_offset(leaf, fi, 0);
2250         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2251         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2252         btrfs_set_file_extent_compression(leaf, fi, compression);
2253         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2254         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2255
2256         btrfs_mark_buffer_dirty(leaf);
2257         btrfs_release_path(path);
2258
2259         inode_add_bytes(inode, num_bytes);
2260
2261         ins.objectid = disk_bytenr;
2262         ins.offset = disk_num_bytes;
2263         ins.type = BTRFS_EXTENT_ITEM_KEY;
2264
2265         /*
2266          * Release the reserved range from inode dirty range map, as it is
2267          * already moved into delayed_ref_head
2268          */
2269         ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2270         if (ret < 0)
2271                 goto out;
2272         qg_released = ret;
2273         ret = btrfs_alloc_reserved_file_extent(trans, root,
2274                                                btrfs_ino(BTRFS_I(inode)),
2275                                                file_pos, qg_released, &ins);
2276 out:
2277         btrfs_free_path(path);
2278
2279         return ret;
2280 }
2281
2282 /* snapshot-aware defrag */
2283 struct sa_defrag_extent_backref {
2284         struct rb_node node;
2285         struct old_sa_defrag_extent *old;
2286         u64 root_id;
2287         u64 inum;
2288         u64 file_pos;
2289         u64 extent_offset;
2290         u64 num_bytes;
2291         u64 generation;
2292 };
2293
2294 struct old_sa_defrag_extent {
2295         struct list_head list;
2296         struct new_sa_defrag_extent *new;
2297
2298         u64 extent_offset;
2299         u64 bytenr;
2300         u64 offset;
2301         u64 len;
2302         int count;
2303 };
2304
2305 struct new_sa_defrag_extent {
2306         struct rb_root root;
2307         struct list_head head;
2308         struct btrfs_path *path;
2309         struct inode *inode;
2310         u64 file_pos;
2311         u64 len;
2312         u64 bytenr;
2313         u64 disk_len;
2314         u8 compress_type;
2315 };
2316
2317 static int backref_comp(struct sa_defrag_extent_backref *b1,
2318                         struct sa_defrag_extent_backref *b2)
2319 {
2320         if (b1->root_id < b2->root_id)
2321                 return -1;
2322         else if (b1->root_id > b2->root_id)
2323                 return 1;
2324
2325         if (b1->inum < b2->inum)
2326                 return -1;
2327         else if (b1->inum > b2->inum)
2328                 return 1;
2329
2330         if (b1->file_pos < b2->file_pos)
2331                 return -1;
2332         else if (b1->file_pos > b2->file_pos)
2333                 return 1;
2334
2335         /*
2336          * [------------------------------] ===> (a range of space)
2337          *     |<--->|   |<---->| =============> (fs/file tree A)
2338          * |<---------------------------->| ===> (fs/file tree B)
2339          *
2340          * A range of space can refer to two file extents in one tree while
2341          * refer to only one file extent in another tree.
2342          *
2343          * So we may process a disk offset more than one time(two extents in A)
2344          * and locate at the same extent(one extent in B), then insert two same
2345          * backrefs(both refer to the extent in B).
2346          */
2347         return 0;
2348 }
2349
2350 static void backref_insert(struct rb_root *root,
2351                            struct sa_defrag_extent_backref *backref)
2352 {
2353         struct rb_node **p = &root->rb_node;
2354         struct rb_node *parent = NULL;
2355         struct sa_defrag_extent_backref *entry;
2356         int ret;
2357
2358         while (*p) {
2359                 parent = *p;
2360                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2361
2362                 ret = backref_comp(backref, entry);
2363                 if (ret < 0)
2364                         p = &(*p)->rb_left;
2365                 else
2366                         p = &(*p)->rb_right;
2367         }
2368
2369         rb_link_node(&backref->node, parent, p);
2370         rb_insert_color(&backref->node, root);
2371 }
2372
2373 /*
2374  * Note the backref might has changed, and in this case we just return 0.
2375  */
2376 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2377                                        void *ctx)
2378 {
2379         struct btrfs_file_extent_item *extent;
2380         struct old_sa_defrag_extent *old = ctx;
2381         struct new_sa_defrag_extent *new = old->new;
2382         struct btrfs_path *path = new->path;
2383         struct btrfs_key key;
2384         struct btrfs_root *root;
2385         struct sa_defrag_extent_backref *backref;
2386         struct extent_buffer *leaf;
2387         struct inode *inode = new->inode;
2388         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2389         int slot;
2390         int ret;
2391         u64 extent_offset;
2392         u64 num_bytes;
2393
2394         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2395             inum == btrfs_ino(BTRFS_I(inode)))
2396                 return 0;
2397
2398         key.objectid = root_id;
2399         key.type = BTRFS_ROOT_ITEM_KEY;
2400         key.offset = (u64)-1;
2401
2402         root = btrfs_read_fs_root_no_name(fs_info, &key);
2403         if (IS_ERR(root)) {
2404                 if (PTR_ERR(root) == -ENOENT)
2405                         return 0;
2406                 WARN_ON(1);
2407                 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2408                          inum, offset, root_id);
2409                 return PTR_ERR(root);
2410         }
2411
2412         key.objectid = inum;
2413         key.type = BTRFS_EXTENT_DATA_KEY;
2414         if (offset > (u64)-1 << 32)
2415                 key.offset = 0;
2416         else
2417                 key.offset = offset;
2418
2419         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2420         if (WARN_ON(ret < 0))
2421                 return ret;
2422         ret = 0;
2423
2424         while (1) {
2425                 cond_resched();
2426
2427                 leaf = path->nodes[0];
2428                 slot = path->slots[0];
2429
2430                 if (slot >= btrfs_header_nritems(leaf)) {
2431                         ret = btrfs_next_leaf(root, path);
2432                         if (ret < 0) {
2433                                 goto out;
2434                         } else if (ret > 0) {
2435                                 ret = 0;
2436                                 goto out;
2437                         }
2438                         continue;
2439                 }
2440
2441                 path->slots[0]++;
2442
2443                 btrfs_item_key_to_cpu(leaf, &key, slot);
2444
2445                 if (key.objectid > inum)
2446                         goto out;
2447
2448                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2449                         continue;
2450
2451                 extent = btrfs_item_ptr(leaf, slot,
2452                                         struct btrfs_file_extent_item);
2453
2454                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2455                         continue;
2456
2457                 /*
2458                  * 'offset' refers to the exact key.offset,
2459                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2460                  * (key.offset - extent_offset).
2461                  */
2462                 if (key.offset != offset)
2463                         continue;
2464
2465                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2466                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2467
2468                 if (extent_offset >= old->extent_offset + old->offset +
2469                     old->len || extent_offset + num_bytes <=
2470                     old->extent_offset + old->offset)
2471                         continue;
2472                 break;
2473         }
2474
2475         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2476         if (!backref) {
2477                 ret = -ENOENT;
2478                 goto out;
2479         }
2480
2481         backref->root_id = root_id;
2482         backref->inum = inum;
2483         backref->file_pos = offset;
2484         backref->num_bytes = num_bytes;
2485         backref->extent_offset = extent_offset;
2486         backref->generation = btrfs_file_extent_generation(leaf, extent);
2487         backref->old = old;
2488         backref_insert(&new->root, backref);
2489         old->count++;
2490 out:
2491         btrfs_release_path(path);
2492         WARN_ON(ret);
2493         return ret;
2494 }
2495
2496 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2497                                    struct new_sa_defrag_extent *new)
2498 {
2499         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2500         struct old_sa_defrag_extent *old, *tmp;
2501         int ret;
2502
2503         new->path = path;
2504
2505         list_for_each_entry_safe(old, tmp, &new->head, list) {
2506                 ret = iterate_inodes_from_logical(old->bytenr +
2507                                                   old->extent_offset, fs_info,
2508                                                   path, record_one_backref,
2509                                                   old, false);
2510                 if (ret < 0 && ret != -ENOENT)
2511                         return false;
2512
2513                 /* no backref to be processed for this extent */
2514                 if (!old->count) {
2515                         list_del(&old->list);
2516                         kfree(old);
2517                 }
2518         }
2519
2520         if (list_empty(&new->head))
2521                 return false;
2522
2523         return true;
2524 }
2525
2526 static int relink_is_mergable(struct extent_buffer *leaf,
2527                               struct btrfs_file_extent_item *fi,
2528                               struct new_sa_defrag_extent *new)
2529 {
2530         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2531                 return 0;
2532
2533         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2534                 return 0;
2535
2536         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2537                 return 0;
2538
2539         if (btrfs_file_extent_encryption(leaf, fi) ||
2540             btrfs_file_extent_other_encoding(leaf, fi))
2541                 return 0;
2542
2543         return 1;
2544 }
2545
2546 /*
2547  * Note the backref might has changed, and in this case we just return 0.
2548  */
2549 static noinline int relink_extent_backref(struct btrfs_path *path,
2550                                  struct sa_defrag_extent_backref *prev,
2551                                  struct sa_defrag_extent_backref *backref)
2552 {
2553         struct btrfs_file_extent_item *extent;
2554         struct btrfs_file_extent_item *item;
2555         struct btrfs_ordered_extent *ordered;
2556         struct btrfs_trans_handle *trans;
2557         struct btrfs_root *root;
2558         struct btrfs_key key;
2559         struct extent_buffer *leaf;
2560         struct old_sa_defrag_extent *old = backref->old;
2561         struct new_sa_defrag_extent *new = old->new;
2562         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2563         struct inode *inode;
2564         struct extent_state *cached = NULL;
2565         int ret = 0;
2566         u64 start;
2567         u64 len;
2568         u64 lock_start;
2569         u64 lock_end;
2570         bool merge = false;
2571         int index;
2572
2573         if (prev && prev->root_id == backref->root_id &&
2574             prev->inum == backref->inum &&
2575             prev->file_pos + prev->num_bytes == backref->file_pos)
2576                 merge = true;
2577
2578         /* step 1: get root */
2579         key.objectid = backref->root_id;
2580         key.type = BTRFS_ROOT_ITEM_KEY;
2581         key.offset = (u64)-1;
2582
2583         index = srcu_read_lock(&fs_info->subvol_srcu);
2584
2585         root = btrfs_read_fs_root_no_name(fs_info, &key);
2586         if (IS_ERR(root)) {
2587                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2588                 if (PTR_ERR(root) == -ENOENT)
2589                         return 0;
2590                 return PTR_ERR(root);
2591         }
2592
2593         if (btrfs_root_readonly(root)) {
2594                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2595                 return 0;
2596         }
2597
2598         /* step 2: get inode */
2599         key.objectid = backref->inum;
2600         key.type = BTRFS_INODE_ITEM_KEY;
2601         key.offset = 0;
2602
2603         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2604         if (IS_ERR(inode)) {
2605                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2606                 return 0;
2607         }
2608
2609         srcu_read_unlock(&fs_info->subvol_srcu, index);
2610
2611         /* step 3: relink backref */
2612         lock_start = backref->file_pos;
2613         lock_end = backref->file_pos + backref->num_bytes - 1;
2614         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2615                          &cached);
2616
2617         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2618         if (ordered) {
2619                 btrfs_put_ordered_extent(ordered);
2620                 goto out_unlock;
2621         }
2622
2623         trans = btrfs_join_transaction(root);
2624         if (IS_ERR(trans)) {
2625                 ret = PTR_ERR(trans);
2626                 goto out_unlock;
2627         }
2628
2629         key.objectid = backref->inum;
2630         key.type = BTRFS_EXTENT_DATA_KEY;
2631         key.offset = backref->file_pos;
2632
2633         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2634         if (ret < 0) {
2635                 goto out_free_path;
2636         } else if (ret > 0) {
2637                 ret = 0;
2638                 goto out_free_path;
2639         }
2640
2641         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2642                                 struct btrfs_file_extent_item);
2643
2644         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2645             backref->generation)
2646                 goto out_free_path;
2647
2648         btrfs_release_path(path);
2649
2650         start = backref->file_pos;
2651         if (backref->extent_offset < old->extent_offset + old->offset)
2652                 start += old->extent_offset + old->offset -
2653                          backref->extent_offset;
2654
2655         len = min(backref->extent_offset + backref->num_bytes,
2656                   old->extent_offset + old->offset + old->len);
2657         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2658
2659         ret = btrfs_drop_extents(trans, root, inode, start,
2660                                  start + len, 1);
2661         if (ret)
2662                 goto out_free_path;
2663 again:
2664         key.objectid = btrfs_ino(BTRFS_I(inode));
2665         key.type = BTRFS_EXTENT_DATA_KEY;
2666         key.offset = start;
2667
2668         path->leave_spinning = 1;
2669         if (merge) {
2670                 struct btrfs_file_extent_item *fi;
2671                 u64 extent_len;
2672                 struct btrfs_key found_key;
2673
2674                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2675                 if (ret < 0)
2676                         goto out_free_path;
2677
2678                 path->slots[0]--;
2679                 leaf = path->nodes[0];
2680                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2681
2682                 fi = btrfs_item_ptr(leaf, path->slots[0],
2683                                     struct btrfs_file_extent_item);
2684                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2685
2686                 if (extent_len + found_key.offset == start &&
2687                     relink_is_mergable(leaf, fi, new)) {
2688                         btrfs_set_file_extent_num_bytes(leaf, fi,
2689                                                         extent_len + len);
2690                         btrfs_mark_buffer_dirty(leaf);
2691                         inode_add_bytes(inode, len);
2692
2693                         ret = 1;
2694                         goto out_free_path;
2695                 } else {
2696                         merge = false;
2697                         btrfs_release_path(path);
2698                         goto again;
2699                 }
2700         }
2701
2702         ret = btrfs_insert_empty_item(trans, root, path, &key,
2703                                         sizeof(*extent));
2704         if (ret) {
2705                 btrfs_abort_transaction(trans, ret);
2706                 goto out_free_path;
2707         }
2708
2709         leaf = path->nodes[0];
2710         item = btrfs_item_ptr(leaf, path->slots[0],
2711                                 struct btrfs_file_extent_item);
2712         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2713         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2714         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2715         btrfs_set_file_extent_num_bytes(leaf, item, len);
2716         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2717         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2718         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2719         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2720         btrfs_set_file_extent_encryption(leaf, item, 0);
2721         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2722
2723         btrfs_mark_buffer_dirty(leaf);
2724         inode_add_bytes(inode, len);
2725         btrfs_release_path(path);
2726
2727         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2728                         new->disk_len, 0,
2729                         backref->root_id, backref->inum,
2730                         new->file_pos); /* start - extent_offset */
2731         if (ret) {
2732                 btrfs_abort_transaction(trans, ret);
2733                 goto out_free_path;
2734         }
2735
2736         ret = 1;
2737 out_free_path:
2738         btrfs_release_path(path);
2739         path->leave_spinning = 0;
2740         btrfs_end_transaction(trans);
2741 out_unlock:
2742         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2743                              &cached);
2744         iput(inode);
2745         return ret;
2746 }
2747
2748 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2749 {
2750         struct old_sa_defrag_extent *old, *tmp;
2751
2752         if (!new)
2753                 return;
2754
2755         list_for_each_entry_safe(old, tmp, &new->head, list) {
2756                 kfree(old);
2757         }
2758         kfree(new);
2759 }
2760
2761 static void relink_file_extents(struct new_sa_defrag_extent *new)
2762 {
2763         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2764         struct btrfs_path *path;
2765         struct sa_defrag_extent_backref *backref;
2766         struct sa_defrag_extent_backref *prev = NULL;
2767         struct inode *inode;
2768         struct rb_node *node;
2769         int ret;
2770
2771         inode = new->inode;
2772
2773         path = btrfs_alloc_path();
2774         if (!path)
2775                 return;
2776
2777         if (!record_extent_backrefs(path, new)) {
2778                 btrfs_free_path(path);
2779                 goto out;
2780         }
2781         btrfs_release_path(path);
2782
2783         while (1) {
2784                 node = rb_first(&new->root);
2785                 if (!node)
2786                         break;
2787                 rb_erase(node, &new->root);
2788
2789                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2790
2791                 ret = relink_extent_backref(path, prev, backref);
2792                 WARN_ON(ret < 0);
2793
2794                 kfree(prev);
2795
2796                 if (ret == 1)
2797                         prev = backref;
2798                 else
2799                         prev = NULL;
2800                 cond_resched();
2801         }
2802         kfree(prev);
2803
2804         btrfs_free_path(path);
2805 out:
2806         free_sa_defrag_extent(new);
2807
2808         atomic_dec(&fs_info->defrag_running);
2809         wake_up(&fs_info->transaction_wait);
2810 }
2811
2812 static struct new_sa_defrag_extent *
2813 record_old_file_extents(struct inode *inode,
2814                         struct btrfs_ordered_extent *ordered)
2815 {
2816         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2817         struct btrfs_root *root = BTRFS_I(inode)->root;
2818         struct btrfs_path *path;
2819         struct btrfs_key key;
2820         struct old_sa_defrag_extent *old;
2821         struct new_sa_defrag_extent *new;
2822         int ret;
2823
2824         new = kmalloc(sizeof(*new), GFP_NOFS);
2825         if (!new)
2826                 return NULL;
2827
2828         new->inode = inode;
2829         new->file_pos = ordered->file_offset;
2830         new->len = ordered->len;
2831         new->bytenr = ordered->start;
2832         new->disk_len = ordered->disk_len;
2833         new->compress_type = ordered->compress_type;
2834         new->root = RB_ROOT;
2835         INIT_LIST_HEAD(&new->head);
2836
2837         path = btrfs_alloc_path();
2838         if (!path)
2839                 goto out_kfree;
2840
2841         key.objectid = btrfs_ino(BTRFS_I(inode));
2842         key.type = BTRFS_EXTENT_DATA_KEY;
2843         key.offset = new->file_pos;
2844
2845         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2846         if (ret < 0)
2847                 goto out_free_path;
2848         if (ret > 0 && path->slots[0] > 0)
2849                 path->slots[0]--;
2850
2851         /* find out all the old extents for the file range */
2852         while (1) {
2853                 struct btrfs_file_extent_item *extent;
2854                 struct extent_buffer *l;
2855                 int slot;
2856                 u64 num_bytes;
2857                 u64 offset;
2858                 u64 end;
2859                 u64 disk_bytenr;
2860                 u64 extent_offset;
2861
2862                 l = path->nodes[0];
2863                 slot = path->slots[0];
2864
2865                 if (slot >= btrfs_header_nritems(l)) {
2866                         ret = btrfs_next_leaf(root, path);
2867                         if (ret < 0)
2868                                 goto out_free_path;
2869                         else if (ret > 0)
2870                                 break;
2871                         continue;
2872                 }
2873
2874                 btrfs_item_key_to_cpu(l, &key, slot);
2875
2876                 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2877                         break;
2878                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2879                         break;
2880                 if (key.offset >= new->file_pos + new->len)
2881                         break;
2882
2883                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2884
2885                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2886                 if (key.offset + num_bytes < new->file_pos)
2887                         goto next;
2888
2889                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2890                 if (!disk_bytenr)
2891                         goto next;
2892
2893                 extent_offset = btrfs_file_extent_offset(l, extent);
2894
2895                 old = kmalloc(sizeof(*old), GFP_NOFS);
2896                 if (!old)
2897                         goto out_free_path;
2898
2899                 offset = max(new->file_pos, key.offset);
2900                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2901
2902                 old->bytenr = disk_bytenr;
2903                 old->extent_offset = extent_offset;
2904                 old->offset = offset - key.offset;
2905                 old->len = end - offset;
2906                 old->new = new;
2907                 old->count = 0;
2908                 list_add_tail(&old->list, &new->head);
2909 next:
2910                 path->slots[0]++;
2911                 cond_resched();
2912         }
2913
2914         btrfs_free_path(path);
2915         atomic_inc(&fs_info->defrag_running);
2916
2917         return new;
2918
2919 out_free_path:
2920         btrfs_free_path(path);
2921 out_kfree:
2922         free_sa_defrag_extent(new);
2923         return NULL;
2924 }
2925
2926 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2927                                          u64 start, u64 len)
2928 {
2929         struct btrfs_block_group_cache *cache;
2930
2931         cache = btrfs_lookup_block_group(fs_info, start);
2932         ASSERT(cache);
2933
2934         spin_lock(&cache->lock);
2935         cache->delalloc_bytes -= len;
2936         spin_unlock(&cache->lock);
2937
2938         btrfs_put_block_group(cache);
2939 }
2940
2941 /* as ordered data IO finishes, this gets called so we can finish
2942  * an ordered extent if the range of bytes in the file it covers are
2943  * fully written.
2944  */
2945 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2946 {
2947         struct inode *inode = ordered_extent->inode;
2948         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2949         struct btrfs_root *root = BTRFS_I(inode)->root;
2950         struct btrfs_trans_handle *trans = NULL;
2951         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2952         struct extent_state *cached_state = NULL;
2953         struct new_sa_defrag_extent *new = NULL;
2954         int compress_type = 0;
2955         int ret = 0;
2956         u64 logical_len = ordered_extent->len;
2957         bool nolock;
2958         bool truncated = false;
2959         bool range_locked = false;
2960         bool clear_new_delalloc_bytes = false;
2961
2962         if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2963             !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2964             !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2965                 clear_new_delalloc_bytes = true;
2966
2967         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2968
2969         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2970                 ret = -EIO;
2971                 goto out;
2972         }
2973
2974         btrfs_free_io_failure_record(BTRFS_I(inode),
2975                         ordered_extent->file_offset,
2976                         ordered_extent->file_offset +
2977                         ordered_extent->len - 1);
2978
2979         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2980                 truncated = true;
2981                 logical_len = ordered_extent->truncated_len;
2982                 /* Truncated the entire extent, don't bother adding */
2983                 if (!logical_len)
2984                         goto out;
2985         }
2986
2987         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2988                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2989
2990                 /*
2991                  * For mwrite(mmap + memset to write) case, we still reserve
2992                  * space for NOCOW range.
2993                  * As NOCOW won't cause a new delayed ref, just free the space
2994                  */
2995                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2996                                        ordered_extent->len);
2997                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2998                 if (nolock)
2999                         trans = btrfs_join_transaction_nolock(root);
3000                 else
3001                         trans = btrfs_join_transaction(root);
3002                 if (IS_ERR(trans)) {
3003                         ret = PTR_ERR(trans);
3004                         trans = NULL;
3005                         goto out;
3006                 }
3007                 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3008                 ret = btrfs_update_inode_fallback(trans, root, inode);
3009                 if (ret) /* -ENOMEM or corruption */
3010                         btrfs_abort_transaction(trans, ret);
3011                 goto out;
3012         }
3013
3014         range_locked = true;
3015         lock_extent_bits(io_tree, ordered_extent->file_offset,
3016                          ordered_extent->file_offset + ordered_extent->len - 1,
3017                          &cached_state);
3018
3019         ret = test_range_bit(io_tree, ordered_extent->file_offset,
3020                         ordered_extent->file_offset + ordered_extent->len - 1,
3021                         EXTENT_DEFRAG, 0, cached_state);
3022         if (ret) {
3023                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3024                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3025                         /* the inode is shared */
3026                         new = record_old_file_extents(inode, ordered_extent);
3027
3028                 clear_extent_bit(io_tree, ordered_extent->file_offset,
3029                         ordered_extent->file_offset + ordered_extent->len - 1,
3030                         EXTENT_DEFRAG, 0, 0, &cached_state);
3031         }
3032
3033         if (nolock)
3034                 trans = btrfs_join_transaction_nolock(root);
3035         else
3036                 trans = btrfs_join_transaction(root);
3037         if (IS_ERR(trans)) {
3038                 ret = PTR_ERR(trans);
3039                 trans = NULL;
3040                 goto out;
3041         }
3042
3043         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3044
3045         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3046                 compress_type = ordered_extent->compress_type;
3047         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3048                 BUG_ON(compress_type);
3049                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3050                                        ordered_extent->len);
3051                 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3052                                                 ordered_extent->file_offset,
3053                                                 ordered_extent->file_offset +
3054                                                 logical_len);
3055         } else {
3056                 BUG_ON(root == fs_info->tree_root);
3057                 ret = insert_reserved_file_extent(trans, inode,
3058                                                 ordered_extent->file_offset,
3059                                                 ordered_extent->start,
3060                                                 ordered_extent->disk_len,
3061                                                 logical_len, logical_len,
3062                                                 compress_type, 0, 0,
3063                                                 BTRFS_FILE_EXTENT_REG);
3064                 if (!ret)
3065                         btrfs_release_delalloc_bytes(fs_info,
3066                                                      ordered_extent->start,
3067                &