~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/file.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 
 19 #include <linux/fs.h>
 20 #include <linux/pagemap.h>
 21 #include <linux/highmem.h>
 22 #include <linux/time.h>
 23 #include <linux/init.h>
 24 #include <linux/string.h>
 25 #include <linux/backing-dev.h>
 26 #include <linux/mpage.h>
 27 #include <linux/aio.h>
 28 #include <linux/falloc.h>
 29 #include <linux/swap.h>
 30 #include <linux/writeback.h>
 31 #include <linux/statfs.h>
 32 #include <linux/compat.h>
 33 #include <linux/slab.h>
 34 #include <linux/btrfs.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 "tree-log.h"
 41 #include "locking.h"
 42 #include "volumes.h"
 43 #include "qgroup.h"
 44 
 45 static struct kmem_cache *btrfs_inode_defrag_cachep;
 46 /*
 47  * when auto defrag is enabled we
 48  * queue up these defrag structs to remember which
 49  * inodes need defragging passes
 50  */
 51 struct inode_defrag {
 52         struct rb_node rb_node;
 53         /* objectid */
 54         u64 ino;
 55         /*
 56          * transid where the defrag was added, we search for
 57          * extents newer than this
 58          */
 59         u64 transid;
 60 
 61         /* root objectid */
 62         u64 root;
 63 
 64         /* last offset we were able to defrag */
 65         u64 last_offset;
 66 
 67         /* if we've wrapped around back to zero once already */
 68         int cycled;
 69 };
 70 
 71 static int __compare_inode_defrag(struct inode_defrag *defrag1,
 72                                   struct inode_defrag *defrag2)
 73 {
 74         if (defrag1->root > defrag2->root)
 75                 return 1;
 76         else if (defrag1->root < defrag2->root)
 77                 return -1;
 78         else if (defrag1->ino > defrag2->ino)
 79                 return 1;
 80         else if (defrag1->ino < defrag2->ino)
 81                 return -1;
 82         else
 83                 return 0;
 84 }
 85 
 86 /* pop a record for an inode into the defrag tree.  The lock
 87  * must be held already
 88  *
 89  * If you're inserting a record for an older transid than an
 90  * existing record, the transid already in the tree is lowered
 91  *
 92  * If an existing record is found the defrag item you
 93  * pass in is freed
 94  */
 95 static int __btrfs_add_inode_defrag(struct inode *inode,
 96                                     struct inode_defrag *defrag)
 97 {
 98         struct btrfs_root *root = BTRFS_I(inode)->root;
 99         struct inode_defrag *entry;
100         struct rb_node **p;
101         struct rb_node *parent = NULL;
102         int ret;
103 
104         p = &root->fs_info->defrag_inodes.rb_node;
105         while (*p) {
106                 parent = *p;
107                 entry = rb_entry(parent, struct inode_defrag, rb_node);
108 
109                 ret = __compare_inode_defrag(defrag, entry);
110                 if (ret < 0)
111                         p = &parent->rb_left;
112                 else if (ret > 0)
113                         p = &parent->rb_right;
114                 else {
115                         /* if we're reinserting an entry for
116                          * an old defrag run, make sure to
117                          * lower the transid of our existing record
118                          */
119                         if (defrag->transid < entry->transid)
120                                 entry->transid = defrag->transid;
121                         if (defrag->last_offset > entry->last_offset)
122                                 entry->last_offset = defrag->last_offset;
123                         return -EEXIST;
124                 }
125         }
126         set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
127         rb_link_node(&defrag->rb_node, parent, p);
128         rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
129         return 0;
130 }
131 
132 static inline int __need_auto_defrag(struct btrfs_root *root)
133 {
134         if (!btrfs_test_opt(root, AUTO_DEFRAG))
135                 return 0;
136 
137         if (btrfs_fs_closing(root->fs_info))
138                 return 0;
139 
140         return 1;
141 }
142 
143 /*
144  * insert a defrag record for this inode if auto defrag is
145  * enabled
146  */
147 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
148                            struct inode *inode)
149 {
150         struct btrfs_root *root = BTRFS_I(inode)->root;
151         struct inode_defrag *defrag;
152         u64 transid;
153         int ret;
154 
155         if (!__need_auto_defrag(root))
156                 return 0;
157 
158         if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
159                 return 0;
160 
161         if (trans)
162                 transid = trans->transid;
163         else
164                 transid = BTRFS_I(inode)->root->last_trans;
165 
166         defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
167         if (!defrag)
168                 return -ENOMEM;
169 
170         defrag->ino = btrfs_ino(inode);
171         defrag->transid = transid;
172         defrag->root = root->root_key.objectid;
173 
174         spin_lock(&root->fs_info->defrag_inodes_lock);
175         if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
176                 /*
177                  * If we set IN_DEFRAG flag and evict the inode from memory,
178                  * and then re-read this inode, this new inode doesn't have
179                  * IN_DEFRAG flag. At the case, we may find the existed defrag.
180                  */
181                 ret = __btrfs_add_inode_defrag(inode, defrag);
182                 if (ret)
183                         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
184         } else {
185                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186         }
187         spin_unlock(&root->fs_info->defrag_inodes_lock);
188         return 0;
189 }
190 
191 /*
192  * Requeue the defrag object. If there is a defrag object that points to
193  * the same inode in the tree, we will merge them together (by
194  * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195  */
196 static void btrfs_requeue_inode_defrag(struct inode *inode,
197                                        struct inode_defrag *defrag)
198 {
199         struct btrfs_root *root = BTRFS_I(inode)->root;
200         int ret;
201 
202         if (!__need_auto_defrag(root))
203                 goto out;
204 
205         /*
206          * Here we don't check the IN_DEFRAG flag, because we need merge
207          * them together.
208          */
209         spin_lock(&root->fs_info->defrag_inodes_lock);
210         ret = __btrfs_add_inode_defrag(inode, defrag);
211         spin_unlock(&root->fs_info->defrag_inodes_lock);
212         if (ret)
213                 goto out;
214         return;
215 out:
216         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
217 }
218 
219 /*
220  * pick the defragable inode that we want, if it doesn't exist, we will get
221  * the next one.
222  */
223 static struct inode_defrag *
224 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
225 {
226         struct inode_defrag *entry = NULL;
227         struct inode_defrag tmp;
228         struct rb_node *p;
229         struct rb_node *parent = NULL;
230         int ret;
231 
232         tmp.ino = ino;
233         tmp.root = root;
234 
235         spin_lock(&fs_info->defrag_inodes_lock);
236         p = fs_info->defrag_inodes.rb_node;
237         while (p) {
238                 parent = p;
239                 entry = rb_entry(parent, struct inode_defrag, rb_node);
240 
241                 ret = __compare_inode_defrag(&tmp, entry);
242                 if (ret < 0)
243                         p = parent->rb_left;
244                 else if (ret > 0)
245                         p = parent->rb_right;
246                 else
247                         goto out;
248         }
249 
250         if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
251                 parent = rb_next(parent);
252                 if (parent)
253                         entry = rb_entry(parent, struct inode_defrag, rb_node);
254                 else
255                         entry = NULL;
256         }
257 out:
258         if (entry)
259                 rb_erase(parent, &fs_info->defrag_inodes);
260         spin_unlock(&fs_info->defrag_inodes_lock);
261         return entry;
262 }
263 
264 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
265 {
266         struct inode_defrag *defrag;
267         struct rb_node *node;
268 
269         spin_lock(&fs_info->defrag_inodes_lock);
270         node = rb_first(&fs_info->defrag_inodes);
271         while (node) {
272                 rb_erase(node, &fs_info->defrag_inodes);
273                 defrag = rb_entry(node, struct inode_defrag, rb_node);
274                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
275 
276                 if (need_resched()) {
277                         spin_unlock(&fs_info->defrag_inodes_lock);
278                         cond_resched();
279                         spin_lock(&fs_info->defrag_inodes_lock);
280                 }
281 
282                 node = rb_first(&fs_info->defrag_inodes);
283         }
284         spin_unlock(&fs_info->defrag_inodes_lock);
285 }
286 
287 #define BTRFS_DEFRAG_BATCH      1024
288 
289 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
290                                     struct inode_defrag *defrag)
291 {
292         struct btrfs_root *inode_root;
293         struct inode *inode;
294         struct btrfs_key key;
295         struct btrfs_ioctl_defrag_range_args range;
296         int num_defrag;
297         int index;
298         int ret;
299 
300         /* get the inode */
301         key.objectid = defrag->root;
302         key.type = BTRFS_ROOT_ITEM_KEY;
303         key.offset = (u64)-1;
304 
305         index = srcu_read_lock(&fs_info->subvol_srcu);
306 
307         inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
308         if (IS_ERR(inode_root)) {
309                 ret = PTR_ERR(inode_root);
310                 goto cleanup;
311         }
312 
313         key.objectid = defrag->ino;
314         key.type = BTRFS_INODE_ITEM_KEY;
315         key.offset = 0;
316         inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
317         if (IS_ERR(inode)) {
318                 ret = PTR_ERR(inode);
319                 goto cleanup;
320         }
321         srcu_read_unlock(&fs_info->subvol_srcu, index);
322 
323         /* do a chunk of defrag */
324         clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
325         memset(&range, 0, sizeof(range));
326         range.len = (u64)-1;
327         range.start = defrag->last_offset;
328 
329         sb_start_write(fs_info->sb);
330         num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
331                                        BTRFS_DEFRAG_BATCH);
332         sb_end_write(fs_info->sb);
333         /*
334          * if we filled the whole defrag batch, there
335          * must be more work to do.  Queue this defrag
336          * again
337          */
338         if (num_defrag == BTRFS_DEFRAG_BATCH) {
339                 defrag->last_offset = range.start;
340                 btrfs_requeue_inode_defrag(inode, defrag);
341         } else if (defrag->last_offset && !defrag->cycled) {
342                 /*
343                  * we didn't fill our defrag batch, but
344                  * we didn't start at zero.  Make sure we loop
345                  * around to the start of the file.
346                  */
347                 defrag->last_offset = 0;
348                 defrag->cycled = 1;
349                 btrfs_requeue_inode_defrag(inode, defrag);
350         } else {
351                 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
352         }
353 
354         iput(inode);
355         return 0;
356 cleanup:
357         srcu_read_unlock(&fs_info->subvol_srcu, index);
358         kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
359         return ret;
360 }
361 
362 /*
363  * run through the list of inodes in the FS that need
364  * defragging
365  */
366 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
367 {
368         struct inode_defrag *defrag;
369         u64 first_ino = 0;
370         u64 root_objectid = 0;
371 
372         atomic_inc(&fs_info->defrag_running);
373         while (1) {
374                 /* Pause the auto defragger. */
375                 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
376                              &fs_info->fs_state))
377                         break;
378 
379                 if (!__need_auto_defrag(fs_info->tree_root))
380                         break;
381 
382                 /* find an inode to defrag */
383                 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
384                                                  first_ino);
385                 if (!defrag) {
386                         if (root_objectid || first_ino) {
387                                 root_objectid = 0;
388                                 first_ino = 0;
389                                 continue;
390                         } else {
391                                 break;
392                         }
393                 }
394 
395                 first_ino = defrag->ino + 1;
396                 root_objectid = defrag->root;
397 
398                 __btrfs_run_defrag_inode(fs_info, defrag);
399         }
400         atomic_dec(&fs_info->defrag_running);
401 
402         /*
403          * during unmount, we use the transaction_wait queue to
404          * wait for the defragger to stop
405          */
406         wake_up(&fs_info->transaction_wait);
407         return 0;
408 }
409 
410 /* simple helper to fault in pages and copy.  This should go away
411  * and be replaced with calls into generic code.
412  */
413 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
414                                          size_t write_bytes,
415                                          struct page **prepared_pages,
416                                          struct iov_iter *i)
417 {
418         size_t copied = 0;
419         size_t total_copied = 0;
420         int pg = 0;
421         int offset = pos & (PAGE_CACHE_SIZE - 1);
422 
423         while (write_bytes > 0) {
424                 size_t count = min_t(size_t,
425                                      PAGE_CACHE_SIZE - offset, write_bytes);
426                 struct page *page = prepared_pages[pg];
427                 /*
428                  * Copy data from userspace to the current page
429                  */
430                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
431 
432                 /* Flush processor's dcache for this page */
433                 flush_dcache_page(page);
434 
435                 /*
436                  * if we get a partial write, we can end up with
437                  * partially up to date pages.  These add
438                  * a lot of complexity, so make sure they don't
439                  * happen by forcing this copy to be retried.
440                  *
441                  * The rest of the btrfs_file_write code will fall
442                  * back to page at a time copies after we return 0.
443                  */
444                 if (!PageUptodate(page) && copied < count)
445                         copied = 0;
446 
447                 iov_iter_advance(i, copied);
448                 write_bytes -= copied;
449                 total_copied += copied;
450 
451                 /* Return to btrfs_file_write_iter to fault page */
452                 if (unlikely(copied == 0))
453                         break;
454 
455                 if (copied < PAGE_CACHE_SIZE - offset) {
456                         offset += copied;
457                 } else {
458                         pg++;
459                         offset = 0;
460                 }
461         }
462         return total_copied;
463 }
464 
465 /*
466  * unlocks pages after btrfs_file_write is done with them
467  */
468 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
469 {
470         size_t i;
471         for (i = 0; i < num_pages; i++) {
472                 /* page checked is some magic around finding pages that
473                  * have been modified without going through btrfs_set_page_dirty
474                  * clear it here. There should be no need to mark the pages
475                  * accessed as prepare_pages should have marked them accessed
476                  * in prepare_pages via find_or_create_page()
477                  */
478                 ClearPageChecked(pages[i]);
479                 unlock_page(pages[i]);
480                 page_cache_release(pages[i]);
481         }
482 }
483 
484 /*
485  * after copy_from_user, pages need to be dirtied and we need to make
486  * sure holes are created between the current EOF and the start of
487  * any next extents (if required).
488  *
489  * this also makes the decision about creating an inline extent vs
490  * doing real data extents, marking pages dirty and delalloc as required.
491  */
492 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
493                              struct page **pages, size_t num_pages,
494                              loff_t pos, size_t write_bytes,
495                              struct extent_state **cached)
496 {
497         int err = 0;
498         int i;
499         u64 num_bytes;
500         u64 start_pos;
501         u64 end_of_last_block;
502         u64 end_pos = pos + write_bytes;
503         loff_t isize = i_size_read(inode);
504 
505         start_pos = pos & ~((u64)root->sectorsize - 1);
506         num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
507 
508         end_of_last_block = start_pos + num_bytes - 1;
509         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
510                                         cached);
511         if (err)
512                 return err;
513 
514         for (i = 0; i < num_pages; i++) {
515                 struct page *p = pages[i];
516                 SetPageUptodate(p);
517                 ClearPageChecked(p);
518                 set_page_dirty(p);
519         }
520 
521         /*
522          * we've only changed i_size in ram, and we haven't updated
523          * the disk i_size.  There is no need to log the inode
524          * at this time.
525          */
526         if (end_pos > isize)
527                 i_size_write(inode, end_pos);
528         return 0;
529 }
530 
531 /*
532  * this drops all the extents in the cache that intersect the range
533  * [start, end].  Existing extents are split as required.
534  */
535 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
536                              int skip_pinned)
537 {
538         struct extent_map *em;
539         struct extent_map *split = NULL;
540         struct extent_map *split2 = NULL;
541         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
542         u64 len = end - start + 1;
543         u64 gen;
544         int ret;
545         int testend = 1;
546         unsigned long flags;
547         int compressed = 0;
548         bool modified;
549 
550         WARN_ON(end < start);
551         if (end == (u64)-1) {
552                 len = (u64)-1;
553                 testend = 0;
554         }
555         while (1) {
556                 int no_splits = 0;
557 
558                 modified = false;
559                 if (!split)
560                         split = alloc_extent_map();
561                 if (!split2)
562                         split2 = alloc_extent_map();
563                 if (!split || !split2)
564                         no_splits = 1;
565 
566                 write_lock(&em_tree->lock);
567                 em = lookup_extent_mapping(em_tree, start, len);
568                 if (!em) {
569                         write_unlock(&em_tree->lock);
570                         break;
571                 }
572                 flags = em->flags;
573                 gen = em->generation;
574                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
575                         if (testend && em->start + em->len >= start + len) {
576                                 free_extent_map(em);
577                                 write_unlock(&em_tree->lock);
578                                 break;
579                         }
580                         start = em->start + em->len;
581                         if (testend)
582                                 len = start + len - (em->start + em->len);
583                         free_extent_map(em);
584                         write_unlock(&em_tree->lock);
585                         continue;
586                 }
587                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
588                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
589                 clear_bit(EXTENT_FLAG_LOGGING, &flags);
590                 modified = !list_empty(&em->list);
591                 if (no_splits)
592                         goto next;
593 
594                 if (em->start < start) {
595                         split->start = em->start;
596                         split->len = start - em->start;
597 
598                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
599                                 split->orig_start = em->orig_start;
600                                 split->block_start = em->block_start;
601 
602                                 if (compressed)
603                                         split->block_len = em->block_len;
604                                 else
605                                         split->block_len = split->len;
606                                 split->orig_block_len = max(split->block_len,
607                                                 em->orig_block_len);
608                                 split->ram_bytes = em->ram_bytes;
609                         } else {
610                                 split->orig_start = split->start;
611                                 split->block_len = 0;
612                                 split->block_start = em->block_start;
613                                 split->orig_block_len = 0;
614                                 split->ram_bytes = split->len;
615                         }
616 
617                         split->generation = gen;
618                         split->bdev = em->bdev;
619                         split->flags = flags;
620                         split->compress_type = em->compress_type;
621                         replace_extent_mapping(em_tree, em, split, modified);
622                         free_extent_map(split);
623                         split = split2;
624                         split2 = NULL;
625                 }
626                 if (testend && em->start + em->len > start + len) {
627                         u64 diff = start + len - em->start;
628 
629                         split->start = start + len;
630                         split->len = em->start + em->len - (start + len);
631                         split->bdev = em->bdev;
632                         split->flags = flags;
633                         split->compress_type = em->compress_type;
634                         split->generation = gen;
635 
636                         if (em->block_start < EXTENT_MAP_LAST_BYTE) {
637                                 split->orig_block_len = max(em->block_len,
638                                                     em->orig_block_len);
639 
640                                 split->ram_bytes = em->ram_bytes;
641                                 if (compressed) {
642                                         split->block_len = em->block_len;
643                                         split->block_start = em->block_start;
644                                         split->orig_start = em->orig_start;
645                                 } else {
646                                         split->block_len = split->len;
647                                         split->block_start = em->block_start
648                                                 + diff;
649                                         split->orig_start = em->orig_start;
650                                 }
651                         } else {
652                                 split->ram_bytes = split->len;
653                                 split->orig_start = split->start;
654                                 split->block_len = 0;
655                                 split->block_start = em->block_start;
656                                 split->orig_block_len = 0;
657                         }
658 
659                         if (extent_map_in_tree(em)) {
660                                 replace_extent_mapping(em_tree, em, split,
661                                                        modified);
662                         } else {
663                                 ret = add_extent_mapping(em_tree, split,
664                                                          modified);
665                                 ASSERT(ret == 0); /* Logic error */
666                         }
667                         free_extent_map(split);
668                         split = NULL;
669                 }
670 next:
671                 if (extent_map_in_tree(em))
672                         remove_extent_mapping(em_tree, em);
673                 write_unlock(&em_tree->lock);
674 
675                 /* once for us */
676                 free_extent_map(em);
677                 /* once for the tree*/
678                 free_extent_map(em);
679         }
680         if (split)
681                 free_extent_map(split);
682         if (split2)
683                 free_extent_map(split2);
684 }
685 
686 /*
687  * this is very complex, but the basic idea is to drop all extents
688  * in the range start - end.  hint_block is filled in with a block number
689  * that would be a good hint to the block allocator for this file.
690  *
691  * If an extent intersects the range but is not entirely inside the range
692  * it is either truncated or split.  Anything entirely inside the range
693  * is deleted from the tree.
694  */
695 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
696                          struct btrfs_root *root, struct inode *inode,
697                          struct btrfs_path *path, u64 start, u64 end,
698                          u64 *drop_end, int drop_cache,
699                          int replace_extent,
700                          u32 extent_item_size,
701                          int *key_inserted)
702 {
703         struct extent_buffer *leaf;
704         struct btrfs_file_extent_item *fi;
705         struct btrfs_key key;
706         struct btrfs_key new_key;
707         u64 ino = btrfs_ino(inode);
708         u64 search_start = start;
709         u64 disk_bytenr = 0;
710         u64 num_bytes = 0;
711         u64 extent_offset = 0;
712         u64 extent_end = 0;
713         int del_nr = 0;
714         int del_slot = 0;
715         int extent_type;
716         int recow;
717         int ret;
718         int modify_tree = -1;
719         int update_refs;
720         int found = 0;
721         int leafs_visited = 0;
722 
723         if (drop_cache)
724                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
725 
726         if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
727                 modify_tree = 0;
728 
729         update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
730                        root == root->fs_info->tree_root);
731         while (1) {
732                 recow = 0;
733                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
734                                                search_start, modify_tree);
735                 if (ret < 0)
736                         break;
737                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
738                         leaf = path->nodes[0];
739                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
740                         if (key.objectid == ino &&
741                             key.type == BTRFS_EXTENT_DATA_KEY)
742                                 path->slots[0]--;
743                 }
744                 ret = 0;
745                 leafs_visited++;
746 next_slot:
747                 leaf = path->nodes[0];
748                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
749                         BUG_ON(del_nr > 0);
750                         ret = btrfs_next_leaf(root, path);
751                         if (ret < 0)
752                                 break;
753                         if (ret > 0) {
754                                 ret = 0;
755                                 break;
756                         }
757                         leafs_visited++;
758                         leaf = path->nodes[0];
759                         recow = 1;
760                 }
761 
762                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
763                 if (key.objectid > ino ||
764                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
765                         break;
766 
767                 fi = btrfs_item_ptr(leaf, path->slots[0],
768                                     struct btrfs_file_extent_item);
769                 extent_type = btrfs_file_extent_type(leaf, fi);
770 
771                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
772                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
773                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
774                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
775                         extent_offset = btrfs_file_extent_offset(leaf, fi);
776                         extent_end = key.offset +
777                                 btrfs_file_extent_num_bytes(leaf, fi);
778                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
779                         extent_end = key.offset +
780                                 btrfs_file_extent_inline_len(leaf,
781                                                      path->slots[0], fi);
782                 } else {
783                         WARN_ON(1);
784                         extent_end = search_start;
785                 }
786 
787                 /*
788                  * Don't skip extent items representing 0 byte lengths. They
789                  * used to be created (bug) if while punching holes we hit
790                  * -ENOSPC condition. So if we find one here, just ensure we
791                  * delete it, otherwise we would insert a new file extent item
792                  * with the same key (offset) as that 0 bytes length file
793                  * extent item in the call to setup_items_for_insert() later
794                  * in this function.
795                  */
796                 if (extent_end == key.offset && extent_end >= search_start)
797                         goto delete_extent_item;
798 
799                 if (extent_end <= search_start) {
800                         path->slots[0]++;
801                         goto next_slot;
802                 }
803 
804                 found = 1;
805                 search_start = max(key.offset, start);
806                 if (recow || !modify_tree) {
807                         modify_tree = -1;
808                         btrfs_release_path(path);
809                         continue;
810                 }
811 
812                 /*
813                  *     | - range to drop - |
814                  *  | -------- extent -------- |
815                  */
816                 if (start > key.offset && end < extent_end) {
817                         BUG_ON(del_nr > 0);
818                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
819                                 ret = -EOPNOTSUPP;
820                                 break;
821                         }
822 
823                         memcpy(&new_key, &key, sizeof(new_key));
824                         new_key.offset = start;
825                         ret = btrfs_duplicate_item(trans, root, path,
826                                                    &new_key);
827                         if (ret == -EAGAIN) {
828                                 btrfs_release_path(path);
829                                 continue;
830                         }
831                         if (ret < 0)
832                                 break;
833 
834                         leaf = path->nodes[0];
835                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
836                                             struct btrfs_file_extent_item);
837                         btrfs_set_file_extent_num_bytes(leaf, fi,
838                                                         start - key.offset);
839 
840                         fi = btrfs_item_ptr(leaf, path->slots[0],
841                                             struct btrfs_file_extent_item);
842 
843                         extent_offset += start - key.offset;
844                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
845                         btrfs_set_file_extent_num_bytes(leaf, fi,
846                                                         extent_end - start);
847                         btrfs_mark_buffer_dirty(leaf);
848 
849                         if (update_refs && disk_bytenr > 0) {
850                                 ret = btrfs_inc_extent_ref(trans, root,
851                                                 disk_bytenr, num_bytes, 0,
852                                                 root->root_key.objectid,
853                                                 new_key.objectid,
854                                                 start - extent_offset, 1);
855                                 BUG_ON(ret); /* -ENOMEM */
856                         }
857                         key.offset = start;
858                 }
859                 /*
860                  *  | ---- range to drop ----- |
861                  *      | -------- extent -------- |
862                  */
863                 if (start <= key.offset && end < extent_end) {
864                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
865                                 ret = -EOPNOTSUPP;
866                                 break;
867                         }
868 
869                         memcpy(&new_key, &key, sizeof(new_key));
870                         new_key.offset = end;
871                         btrfs_set_item_key_safe(root, path, &new_key);
872 
873                         extent_offset += end - key.offset;
874                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
875                         btrfs_set_file_extent_num_bytes(leaf, fi,
876                                                         extent_end - end);
877                         btrfs_mark_buffer_dirty(leaf);
878                         if (update_refs && disk_bytenr > 0)
879                                 inode_sub_bytes(inode, end - key.offset);
880                         break;
881                 }
882 
883                 search_start = extent_end;
884                 /*
885                  *       | ---- range to drop ----- |
886                  *  | -------- extent -------- |
887                  */
888                 if (start > key.offset && end >= extent_end) {
889                         BUG_ON(del_nr > 0);
890                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
891                                 ret = -EOPNOTSUPP;
892                                 break;
893                         }
894 
895                         btrfs_set_file_extent_num_bytes(leaf, fi,
896                                                         start - key.offset);
897                         btrfs_mark_buffer_dirty(leaf);
898                         if (update_refs && disk_bytenr > 0)
899                                 inode_sub_bytes(inode, extent_end - start);
900                         if (end == extent_end)
901                                 break;
902 
903                         path->slots[0]++;
904                         goto next_slot;
905                 }
906 
907                 /*
908                  *  | ---- range to drop ----- |
909                  *    | ------ extent ------ |
910                  */
911                 if (start <= key.offset && end >= extent_end) {
912 delete_extent_item:
913                         if (del_nr == 0) {
914                                 del_slot = path->slots[0];
915                                 del_nr = 1;
916                         } else {
917                                 BUG_ON(del_slot + del_nr != path->slots[0]);
918                                 del_nr++;
919                         }
920 
921                         if (update_refs &&
922                             extent_type == BTRFS_FILE_EXTENT_INLINE) {
923                                 inode_sub_bytes(inode,
924                                                 extent_end - key.offset);
925                                 extent_end = ALIGN(extent_end,
926                                                    root->sectorsize);
927                         } else if (update_refs && disk_bytenr > 0) {
928                                 ret = btrfs_free_extent(trans, root,
929                                                 disk_bytenr, num_bytes, 0,
930                                                 root->root_key.objectid,
931                                                 key.objectid, key.offset -
932                                                 extent_offset, 0);
933                                 BUG_ON(ret); /* -ENOMEM */
934                                 inode_sub_bytes(inode,
935                                                 extent_end - key.offset);
936                         }
937 
938                         if (end == extent_end)
939                                 break;
940 
941                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
942                                 path->slots[0]++;
943                                 goto next_slot;
944                         }
945 
946                         ret = btrfs_del_items(trans, root, path, del_slot,
947                                               del_nr);
948                         if (ret) {
949                                 btrfs_abort_transaction(trans, root, ret);
950                                 break;
951                         }
952 
953                         del_nr = 0;
954                         del_slot = 0;
955 
956                         btrfs_release_path(path);
957                         continue;
958                 }
959 
960                 BUG_ON(1);
961         }
962 
963         if (!ret && del_nr > 0) {
964                 /*
965                  * Set path->slots[0] to first slot, so that after the delete
966                  * if items are move off from our leaf to its immediate left or
967                  * right neighbor leafs, we end up with a correct and adjusted
968                  * path->slots[0] for our insertion (if replace_extent != 0).
969                  */
970                 path->slots[0] = del_slot;
971                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
972                 if (ret)
973                         btrfs_abort_transaction(trans, root, ret);
974         }
975 
976         leaf = path->nodes[0];
977         /*
978          * If btrfs_del_items() was called, it might have deleted a leaf, in
979          * which case it unlocked our path, so check path->locks[0] matches a
980          * write lock.
981          */
982         if (!ret && replace_extent && leafs_visited == 1 &&
983             (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
984              path->locks[0] == BTRFS_WRITE_LOCK) &&
985             btrfs_leaf_free_space(root, leaf) >=
986             sizeof(struct btrfs_item) + extent_item_size) {
987 
988                 key.objectid = ino;
989                 key.type = BTRFS_EXTENT_DATA_KEY;
990                 key.offset = start;
991                 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
992                         struct btrfs_key slot_key;
993 
994                         btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
995                         if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
996                                 path->slots[0]++;
997                 }
998                 setup_items_for_insert(root, path, &key,
999                                        &extent_item_size,
1000                                        extent_item_size,
1001                                        sizeof(struct btrfs_item) +
1002                                        extent_item_size, 1);
1003                 *key_inserted = 1;
1004         }
1005 
1006         if (!replace_extent || !(*key_inserted))
1007                 btrfs_release_path(path);
1008         if (drop_end)
1009                 *drop_end = found ? min(end, extent_end) : end;
1010         return ret;
1011 }
1012 
1013 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1014                        struct btrfs_root *root, struct inode *inode, u64 start,
1015                        u64 end, int drop_cache)
1016 {
1017         struct btrfs_path *path;
1018         int ret;
1019 
1020         path = btrfs_alloc_path();
1021         if (!path)
1022                 return -ENOMEM;
1023         ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1024                                    drop_cache, 0, 0, NULL);
1025         btrfs_free_path(path);
1026         return ret;
1027 }
1028 
1029 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1030                             u64 objectid, u64 bytenr, u64 orig_offset,
1031                             u64 *start, u64 *end)
1032 {
1033         struct btrfs_file_extent_item *fi;
1034         struct btrfs_key key;
1035         u64 extent_end;
1036 
1037         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1038                 return 0;
1039 
1040         btrfs_item_key_to_cpu(leaf, &key, slot);
1041         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1042                 return 0;
1043 
1044         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1045         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1046             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1047             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1048             btrfs_file_extent_compression(leaf, fi) ||
1049             btrfs_file_extent_encryption(leaf, fi) ||
1050             btrfs_file_extent_other_encoding(leaf, fi))
1051                 return 0;
1052 
1053         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1054         if ((*start && *start != key.offset) || (*end && *end != extent_end))
1055                 return 0;
1056 
1057         *start = key.offset;
1058         *end = extent_end;
1059         return 1;
1060 }
1061 
1062 /*
1063  * Mark extent in the range start - end as written.
1064  *
1065  * This changes extent type from 'pre-allocated' to 'regular'. If only
1066  * part of extent is marked as written, the extent will be split into
1067  * two or three.
1068  */
1069 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1070                               struct inode *inode, u64 start, u64 end)
1071 {
1072         struct btrfs_root *root = BTRFS_I(inode)->root;
1073         struct extent_buffer *leaf;
1074         struct btrfs_path *path;
1075         struct btrfs_file_extent_item *fi;
1076         struct btrfs_key key;
1077         struct btrfs_key new_key;
1078         u64 bytenr;
1079         u64 num_bytes;
1080         u64 extent_end;
1081         u64 orig_offset;
1082         u64 other_start;
1083         u64 other_end;
1084         u64 split;
1085         int del_nr = 0;
1086         int del_slot = 0;
1087         int recow;
1088         int ret;
1089         u64 ino = btrfs_ino(inode);
1090 
1091         path = btrfs_alloc_path();
1092         if (!path)
1093                 return -ENOMEM;
1094 again:
1095         recow = 0;
1096         split = start;
1097         key.objectid = ino;
1098         key.type = BTRFS_EXTENT_DATA_KEY;
1099         key.offset = split;
1100 
1101         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1102         if (ret < 0)
1103                 goto out;
1104         if (ret > 0 && path->slots[0] > 0)
1105                 path->slots[0]--;
1106 
1107         leaf = path->nodes[0];
1108         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1109         BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1110         fi = btrfs_item_ptr(leaf, path->slots[0],
1111                             struct btrfs_file_extent_item);
1112         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1113                BTRFS_FILE_EXTENT_PREALLOC);
1114         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1115         BUG_ON(key.offset > start || extent_end < end);
1116 
1117         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1118         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1119         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1120         memcpy(&new_key, &key, sizeof(new_key));
1121 
1122         if (start == key.offset && end < extent_end) {
1123                 other_start = 0;
1124                 other_end = start;
1125                 if (extent_mergeable(leaf, path->slots[0] - 1,
1126                                      ino, bytenr, orig_offset,
1127                                      &other_start, &other_end)) {
1128                         new_key.offset = end;
1129                         btrfs_set_item_key_safe(root, path, &new_key);
1130                         fi = btrfs_item_ptr(leaf, path->slots[0],
1131                                             struct btrfs_file_extent_item);
1132                         btrfs_set_file_extent_generation(leaf, fi,
1133                                                          trans->transid);
1134                         btrfs_set_file_extent_num_bytes(leaf, fi,
1135                                                         extent_end - end);
1136                         btrfs_set_file_extent_offset(leaf, fi,
1137                                                      end - orig_offset);
1138                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1139                                             struct btrfs_file_extent_item);
1140                         btrfs_set_file_extent_generation(leaf, fi,
1141                                                          trans->transid);
1142                         btrfs_set_file_extent_num_bytes(leaf, fi,
1143                                                         end - other_start);
1144                         btrfs_mark_buffer_dirty(leaf);
1145                         goto out;
1146                 }
1147         }
1148 
1149         if (start > key.offset && end == extent_end) {
1150                 other_start = end;
1151                 other_end = 0;
1152                 if (extent_mergeable(leaf, path->slots[0] + 1,
1153                                      ino, bytenr, orig_offset,
1154                                      &other_start, &other_end)) {
1155                         fi = btrfs_item_ptr(leaf, path->slots[0],
1156                                             struct btrfs_file_extent_item);
1157                         btrfs_set_file_extent_num_bytes(leaf, fi,
1158                                                         start - key.offset);
1159                         btrfs_set_file_extent_generation(leaf, fi,
1160                                                          trans->transid);
1161                         path->slots[0]++;
1162                         new_key.offset = start;
1163                         btrfs_set_item_key_safe(root, path, &new_key);
1164 
1165                         fi = btrfs_item_ptr(leaf, path->slots[0],
1166                                             struct btrfs_file_extent_item);
1167                         btrfs_set_file_extent_generation(leaf, fi,
1168                                                          trans->transid);
1169                         btrfs_set_file_extent_num_bytes(leaf, fi,
1170                                                         other_end - start);
1171                         btrfs_set_file_extent_offset(leaf, fi,
1172                                                      start - orig_offset);
1173                         btrfs_mark_buffer_dirty(leaf);
1174                         goto out;
1175                 }
1176         }
1177 
1178         while (start > key.offset || end < extent_end) {
1179                 if (key.offset == start)
1180                         split = end;
1181 
1182                 new_key.offset = split;
1183                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1184                 if (ret == -EAGAIN) {
1185                         btrfs_release_path(path);
1186                         goto again;
1187                 }
1188                 if (ret < 0) {
1189                         btrfs_abort_transaction(trans, root, ret);
1190                         goto out;
1191                 }
1192 
1193                 leaf = path->nodes[0];
1194                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1195                                     struct btrfs_file_extent_item);
1196                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1197                 btrfs_set_file_extent_num_bytes(leaf, fi,
1198                                                 split - key.offset);
1199 
1200                 fi = btrfs_item_ptr(leaf, path->slots[0],
1201                                     struct btrfs_file_extent_item);
1202 
1203                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1204                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1205                 btrfs_set_file_extent_num_bytes(leaf, fi,
1206                                                 extent_end - split);
1207                 btrfs_mark_buffer_dirty(leaf);
1208 
1209                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1210                                            root->root_key.objectid,
1211                                            ino, orig_offset, 1);
1212                 BUG_ON(ret); /* -ENOMEM */
1213 
1214                 if (split == start) {
1215                         key.offset = start;
1216                 } else {
1217                         BUG_ON(start != key.offset);
1218                         path->slots[0]--;
1219                         extent_end = end;
1220                 }
1221                 recow = 1;
1222         }
1223 
1224         other_start = end;
1225         other_end = 0;
1226         if (extent_mergeable(leaf, path->slots[0] + 1,
1227                              ino, bytenr, orig_offset,
1228                              &other_start, &other_end)) {
1229                 if (recow) {
1230                         btrfs_release_path(path);
1231                         goto again;
1232                 }
1233                 extent_end = other_end;
1234                 del_slot = path->slots[0] + 1;
1235                 del_nr++;
1236                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1237                                         0, root->root_key.objectid,
1238                                         ino, orig_offset, 0);
1239                 BUG_ON(ret); /* -ENOMEM */
1240         }
1241         other_start = 0;
1242         other_end = start;
1243         if (extent_mergeable(leaf, path->slots[0] - 1,
1244                              ino, bytenr, orig_offset,
1245                              &other_start, &other_end)) {
1246                 if (recow) {
1247                         btrfs_release_path(path);
1248                         goto again;
1249                 }
1250                 key.offset = other_start;
1251                 del_slot = path->slots[0];
1252                 del_nr++;
1253                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1254                                         0, root->root_key.objectid,
1255                                         ino, orig_offset, 0);
1256                 BUG_ON(ret); /* -ENOMEM */
1257         }
1258         if (del_nr == 0) {
1259                 fi = btrfs_item_ptr(leaf, path->slots[0],
1260                            struct btrfs_file_extent_item);
1261                 btrfs_set_file_extent_type(leaf, fi,
1262                                            BTRFS_FILE_EXTENT_REG);
1263                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1264                 btrfs_mark_buffer_dirty(leaf);
1265         } else {
1266                 fi = btrfs_item_ptr(leaf, del_slot - 1,
1267                            struct btrfs_file_extent_item);
1268                 btrfs_set_file_extent_type(leaf, fi,
1269                                            BTRFS_FILE_EXTENT_REG);
1270                 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1271                 btrfs_set_file_extent_num_bytes(leaf, fi,
1272                                                 extent_end - key.offset);
1273                 btrfs_mark_buffer_dirty(leaf);
1274 
1275                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1276                 if (ret < 0) {
1277                         btrfs_abort_transaction(trans, root, ret);
1278                         goto out;
1279                 }
1280         }
1281 out:
1282         btrfs_free_path(path);
1283         return 0;
1284 }
1285 
1286 /*
1287  * on error we return an unlocked page and the error value
1288  * on success we return a locked page and 0
1289  */
1290 static int prepare_uptodate_page(struct page *page, u64 pos,
1291                                  bool force_uptodate)
1292 {
1293         int ret = 0;
1294 
1295         if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1296             !PageUptodate(page)) {
1297                 ret = btrfs_readpage(NULL, page);
1298                 if (ret)
1299                         return ret;
1300                 lock_page(page);
1301                 if (!PageUptodate(page)) {
1302                         unlock_page(page);
1303                         return -EIO;
1304                 }
1305         }
1306         return 0;
1307 }
1308 
1309 /*
1310  * this just gets pages into the page cache and locks them down.
1311  */
1312 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1313                                   size_t num_pages, loff_t pos,
1314                                   size_t write_bytes, bool force_uptodate)
1315 {
1316         int i;
1317         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1318         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1319         int err = 0;
1320         int faili;
1321 
1322         for (i = 0; i < num_pages; i++) {
1323                 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1324                                                mask | __GFP_WRITE);
1325                 if (!pages[i]) {
1326                         faili = i - 1;
1327                         err = -ENOMEM;
1328                         goto fail;
1329                 }
1330 
1331                 if (i == 0)
1332                         err = prepare_uptodate_page(pages[i], pos,
1333                                                     force_uptodate);
1334                 if (i == num_pages - 1)
1335                         err = prepare_uptodate_page(pages[i],
1336                                                     pos + write_bytes, false);
1337                 if (err) {
1338                         page_cache_release(pages[i]);
1339                         faili = i - 1;
1340                         goto fail;
1341                 }
1342                 wait_on_page_writeback(pages[i]);
1343         }
1344 
1345         return 0;
1346 fail:
1347         while (faili >= 0) {
1348                 unlock_page(pages[faili]);
1349                 page_cache_release(pages[faili]);
1350                 faili--;
1351         }
1352         return err;
1353 
1354 }
1355 
1356 /*
1357  * This function locks the extent and properly waits for data=ordered extents
1358  * to finish before allowing the pages to be modified if need.
1359  *
1360  * The return value:
1361  * 1 - the extent is locked
1362  * 0 - the extent is not locked, and everything is OK
1363  * -EAGAIN - need re-prepare the pages
1364  * the other < 0 number - Something wrong happens
1365  */
1366 static noinline int
1367 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1368                                 size_t num_pages, loff_t pos,
1369                                 u64 *lockstart, u64 *lockend,
1370                                 struct extent_state **cached_state)
1371 {
1372         u64 start_pos;
1373         u64 last_pos;
1374         int i;
1375         int ret = 0;
1376 
1377         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1378         last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1379 
1380         if (start_pos < inode->i_size) {
1381                 struct btrfs_ordered_extent *ordered;
1382                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1383                                  start_pos, last_pos, 0, cached_state);
1384                 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1385                                                      last_pos - start_pos + 1);
1386                 if (ordered &&
1387                     ordered->file_offset + ordered->len > start_pos &&
1388                     ordered->file_offset <= last_pos) {
1389                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1390                                              start_pos, last_pos,
1391                                              cached_state, GFP_NOFS);
1392                         for (i = 0; i < num_pages; i++) {
1393                                 unlock_page(pages[i]);
1394                                 page_cache_release(pages[i]);
1395                         }
1396                         btrfs_start_ordered_extent(inode, ordered, 1);
1397                         btrfs_put_ordered_extent(ordered);
1398                         return -EAGAIN;
1399                 }
1400                 if (ordered)
1401                         btrfs_put_ordered_extent(ordered);
1402 
1403                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1404                                   last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1405                                   EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1406                                   0, 0, cached_state, GFP_NOFS);
1407                 *lockstart = start_pos;
1408                 *lockend = last_pos;
1409                 ret = 1;
1410         }
1411 
1412         for (i = 0; i < num_pages; i++) {
1413                 if (clear_page_dirty_for_io(pages[i]))
1414                         account_page_redirty(pages[i]);
1415                 set_page_extent_mapped(pages[i]);
1416                 WARN_ON(!PageLocked(pages[i]));
1417         }
1418 
1419         return ret;
1420 }
1421 
1422 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1423                                     size_t *write_bytes)
1424 {
1425         struct btrfs_root *root = BTRFS_I(inode)->root;
1426         struct btrfs_ordered_extent *ordered;
1427         u64 lockstart, lockend;
1428         u64 num_bytes;
1429         int ret;
1430 
1431         ret = btrfs_start_write_no_snapshoting(root);
1432         if (!ret)
1433                 return -ENOSPC;
1434 
1435         lockstart = round_down(pos, root->sectorsize);
1436         lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1437 
1438         while (1) {
1439                 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1440                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1441                                                      lockend - lockstart + 1);
1442                 if (!ordered) {
1443                         break;
1444                 }
1445                 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1446                 btrfs_start_ordered_extent(inode, ordered, 1);
1447                 btrfs_put_ordered_extent(ordered);
1448         }
1449 
1450         num_bytes = lockend - lockstart + 1;
1451         ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1452         if (ret <= 0) {
1453                 ret = 0;
1454                 btrfs_end_write_no_snapshoting(root);
1455         } else {
1456                 *write_bytes = min_t(size_t, *write_bytes ,
1457                                      num_bytes - pos + lockstart);
1458         }
1459 
1460         unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1461 
1462         return ret;
1463 }
1464 
1465 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1466                                                struct iov_iter *i,
1467                                                loff_t pos)
1468 {
1469         struct inode *inode = file_inode(file);
1470         struct btrfs_root *root = BTRFS_I(inode)->root;
1471         struct page **pages = NULL;
1472         struct extent_state *cached_state = NULL;
1473         u64 release_bytes = 0;
1474         u64 lockstart;
1475         u64 lockend;
1476         unsigned long first_index;
1477         size_t num_written = 0;
1478         int nrptrs;
1479         int ret = 0;
1480         bool only_release_metadata = false;
1481         bool force_page_uptodate = false;
1482         bool need_unlock;
1483 
1484         nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE),
1485                         PAGE_CACHE_SIZE / (sizeof(struct page *)));
1486         nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1487         nrptrs = max(nrptrs, 8);
1488         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1489         if (!pages)
1490                 return -ENOMEM;
1491 
1492         first_index = pos >> PAGE_CACHE_SHIFT;
1493 
1494         while (iov_iter_count(i) > 0) {
1495                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1496                 size_t write_bytes = min(iov_iter_count(i),
1497                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1498                                          offset);
1499                 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1500                                                 PAGE_CACHE_SIZE);
1501                 size_t reserve_bytes;
1502                 size_t dirty_pages;
1503                 size_t copied;
1504 
1505                 WARN_ON(num_pages > nrptrs);
1506 
1507                 /*
1508                  * Fault pages before locking them in prepare_pages
1509                  * to avoid recursive lock
1510                  */
1511                 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1512                         ret = -EFAULT;
1513                         break;
1514                 }
1515 
1516                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1517                 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1518                 if (ret == -ENOSPC &&
1519                     (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1520                                               BTRFS_INODE_PREALLOC))) {
1521                         ret = check_can_nocow(inode, pos, &write_bytes);
1522                         if (ret > 0) {
1523                                 only_release_metadata = true;
1524                                 /*
1525                                  * our prealloc extent may be smaller than
1526                                  * write_bytes, so scale down.
1527                                  */
1528                                 num_pages = DIV_ROUND_UP(write_bytes + offset,
1529                                                          PAGE_CACHE_SIZE);
1530                                 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1531                                 ret = 0;
1532                         } else {
1533                                 ret = -ENOSPC;
1534                         }
1535                 }
1536 
1537                 if (ret)
1538                         break;
1539 
1540                 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1541                 if (ret) {
1542                         if (!only_release_metadata)
1543                                 btrfs_free_reserved_data_space(inode,
1544                                                                reserve_bytes);
1545                         else
1546                                 btrfs_end_write_no_snapshoting(root);
1547                         break;
1548                 }
1549 
1550                 release_bytes = reserve_bytes;
1551                 need_unlock = false;
1552 again:
1553                 /*
1554                  * This is going to setup the pages array with the number of
1555                  * pages we want, so we don't really need to worry about the
1556                  * contents of pages from loop to loop
1557                  */
1558                 ret = prepare_pages(inode, pages, num_pages,
1559                                     pos, write_bytes,
1560                                     force_page_uptodate);
1561                 if (ret)
1562                         break;
1563 
1564                 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1565                                                       pos, &lockstart, &lockend,
1566                                                       &cached_state);
1567                 if (ret < 0) {
1568                         if (ret == -EAGAIN)
1569                                 goto again;
1570                         break;
1571                 } else if (ret > 0) {
1572                         need_unlock = true;
1573                         ret = 0;
1574                 }
1575 
1576                 copied = btrfs_copy_from_user(pos, num_pages,
1577                                            write_bytes, pages, i);
1578 
1579                 /*
1580                  * if we have trouble faulting in the pages, fall
1581                  * back to one page at a time
1582                  */
1583                 if (copied < write_bytes)
1584                         nrptrs = 1;
1585 
1586                 if (copied == 0) {
1587                         force_page_uptodate = true;
1588                         dirty_pages = 0;
1589                 } else {
1590                         force_page_uptodate = false;
1591                         dirty_pages = DIV_ROUND_UP(copied + offset,
1592                                                    PAGE_CACHE_SIZE);
1593                 }
1594 
1595                 /*
1596                  * If we had a short copy we need to release the excess delaloc
1597                  * bytes we reserved.  We need to increment outstanding_extents
1598                  * because btrfs_delalloc_release_space will decrement it, but
1599                  * we still have an outstanding extent for the chunk we actually
1600                  * managed to copy.
1601                  */
1602                 if (num_pages > dirty_pages) {
1603                         release_bytes = (num_pages - dirty_pages) <<
1604                                 PAGE_CACHE_SHIFT;
1605                         if (copied > 0) {
1606                                 spin_lock(&BTRFS_I(inode)->lock);
1607                                 BTRFS_I(inode)->outstanding_extents++;
1608                                 spin_unlock(&BTRFS_I(inode)->lock);
1609                         }
1610                         if (only_release_metadata)
1611                                 btrfs_delalloc_release_metadata(inode,
1612                                                                 release_bytes);
1613                         else
1614                                 btrfs_delalloc_release_space(inode,
1615                                                              release_bytes);
1616                 }
1617 
1618                 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1619 
1620                 if (copied > 0)
1621                         ret = btrfs_dirty_pages(root, inode, pages,
1622                                                 dirty_pages, pos, copied,
1623                                                 NULL);
1624                 if (need_unlock)
1625                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1626                                              lockstart, lockend, &cached_state,
1627                                              GFP_NOFS);
1628                 if (ret) {
1629                         btrfs_drop_pages(pages, num_pages);
1630                         break;
1631                 }
1632 
1633                 release_bytes = 0;
1634                 if (only_release_metadata)
1635                         btrfs_end_write_no_snapshoting(root);
1636 
1637                 if (only_release_metadata && copied > 0) {
1638                         u64 lockstart = round_down(pos, root->sectorsize);
1639                         u64 lockend = lockstart +
1640                                 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1641 
1642                         set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1643                                        lockend, EXTENT_NORESERVE, NULL,
1644                                        NULL, GFP_NOFS);
1645                         only_release_metadata = false;
1646                 }
1647 
1648                 btrfs_drop_pages(pages, num_pages);
1649 
1650                 cond_resched();
1651 
1652                 balance_dirty_pages_ratelimited(inode->i_mapping);
1653                 if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1)
1654                         btrfs_btree_balance_dirty(root);
1655 
1656                 pos += copied;
1657                 num_written += copied;
1658         }
1659 
1660         kfree(pages);
1661 
1662         if (release_bytes) {
1663                 if (only_release_metadata) {
1664                         btrfs_end_write_no_snapshoting(root);
1665                         btrfs_delalloc_release_metadata(inode, release_bytes);
1666                 } else {
1667                         btrfs_delalloc_release_space(inode, release_bytes);
1668                 }
1669         }
1670 
1671         return num_written ? num_written : ret;
1672 }
1673 
1674 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1675                                     struct iov_iter *from,
1676                                     loff_t pos)
1677 {
1678         struct file *file = iocb->ki_filp;
1679         struct inode *inode = file_inode(file);
1680         ssize_t written;
1681         ssize_t written_buffered;
1682         loff_t endbyte;
1683         int err;
1684 
1685         written = generic_file_direct_write(iocb, from, pos);
1686 
1687         if (written < 0 || !iov_iter_count(from))
1688                 return written;
1689 
1690         pos += written;
1691         written_buffered = __btrfs_buffered_write(file, from, pos);
1692         if (written_buffered < 0) {
1693                 err = written_buffered;
1694                 goto out;
1695         }
1696         /*
1697          * Ensure all data is persisted. We want the next direct IO read to be
1698          * able to read what was just written.
1699          */
1700         endbyte = pos + written_buffered - 1;
1701         err = btrfs_fdatawrite_range(inode, pos, endbyte);
1702         if (err)
1703                 goto out;
1704         err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1705         if (err)
1706                 goto out;
1707         written += written_buffered;
1708         iocb->ki_pos = pos + written_buffered;
1709         invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1710                                  endbyte >> PAGE_CACHE_SHIFT);
1711 out:
1712         return written ? written : err;
1713 }
1714 
1715 static void update_time_for_write(struct inode *inode)
1716 {
1717         struct timespec now;
1718 
1719         if (IS_NOCMTIME(inode))
1720                 return;
1721 
1722         now = current_fs_time(inode->i_sb);
1723         if (!timespec_equal(&inode->i_mtime, &now))
1724                 inode->i_mtime = now;
1725 
1726         if (!timespec_equal(&inode->i_ctime, &now))
1727                 inode->i_ctime = now;
1728 
1729         if (IS_I_VERSION(inode))
1730                 inode_inc_iversion(inode);
1731 }
1732 
1733 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1734                                     struct iov_iter *from)
1735 {
1736         struct file *file = iocb->ki_filp;
1737         struct inode *inode = file_inode(file);
1738         struct btrfs_root *root = BTRFS_I(inode)->root;
1739         u64 start_pos;
1740         u64 end_pos;
1741         ssize_t num_written = 0;
1742         ssize_t err = 0;
1743         size_t count = iov_iter_count(from);
1744         bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1745         loff_t pos = iocb->ki_pos;
1746 
1747         mutex_lock(&inode->i_mutex);
1748 
1749         current->backing_dev_info = inode_to_bdi(inode);
1750         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1751         if (err) {
1752                 mutex_unlock(&inode->i_mutex);
1753                 goto out;
1754         }
1755 
1756         if (count == 0) {
1757                 mutex_unlock(&inode->i_mutex);
1758                 goto out;
1759         }
1760 
1761         iov_iter_truncate(from, count);
1762 
1763         err = file_remove_suid(file);
1764         if (err) {
1765                 mutex_unlock(&inode->i_mutex);
1766                 goto out;
1767         }
1768 
1769         /*
1770          * If BTRFS flips readonly due to some impossible error
1771          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1772          * although we have opened a file as writable, we have
1773          * to stop this write operation to ensure FS consistency.
1774          */
1775         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1776                 mutex_unlock(&inode->i_mutex);
1777                 err = -EROFS;
1778                 goto out;
1779         }
1780 
1781         /*
1782          * We reserve space for updating the inode when we reserve space for the
1783          * extent we are going to write, so we will enospc out there.  We don't
1784          * need to start yet another transaction to update the inode as we will
1785          * update the inode when we finish writing whatever data we write.
1786          */
1787         update_time_for_write(inode);
1788 
1789         start_pos = round_down(pos, root->sectorsize);
1790         if (start_pos > i_size_read(inode)) {
1791                 /* Expand hole size to cover write data, preventing empty gap */
1792                 end_pos = round_up(pos + count, root->sectorsize);
1793                 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1794                 if (err) {
1795                         mutex_unlock(&inode->i_mutex);
1796                         goto out;
1797                 }
1798         }
1799 
1800         if (sync)
1801                 atomic_inc(&BTRFS_I(inode)->sync_writers);
1802 
1803         if (file->f_flags & O_DIRECT) {
1804                 num_written = __btrfs_direct_write(iocb, from, pos);
1805         } else {
1806                 num_written = __btrfs_buffered_write(file, from, pos);
1807                 if (num_written > 0)
1808                         iocb->ki_pos = pos + num_written;
1809         }
1810 
1811         mutex_unlock(&inode->i_mutex);
1812 
1813         /*
1814          * We also have to set last_sub_trans to the current log transid,
1815          * otherwise subsequent syncs to a file that's been synced in this
1816          * transaction will appear to have already occured.
1817          */
1818         BTRFS_I(inode)->last_sub_trans = root->log_transid;
1819         if (num_written > 0) {
1820                 err = generic_write_sync(file, pos, num_written);
1821                 if (err < 0)
1822                         num_written = err;
1823         }
1824 
1825         if (sync)
1826                 atomic_dec(&BTRFS_I(inode)->sync_writers);
1827 out:
1828         current->backing_dev_info = NULL;
1829         return num_written ? num_written : err;
1830 }
1831 
1832 int btrfs_release_file(struct inode *inode, struct file *filp)
1833 {
1834         if (filp->private_data)
1835                 btrfs_ioctl_trans_end(filp);
1836         /*
1837          * ordered_data_close is set by settattr when we are about to truncate
1838          * a file from a non-zero size to a zero size.  This tries to
1839          * flush down new bytes that may have been written if the
1840          * application were using truncate to replace a file in place.
1841          */
1842         if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1843                                &BTRFS_I(inode)->runtime_flags))
1844                         filemap_flush(inode->i_mapping);
1845         return 0;
1846 }
1847 
1848 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1849 {
1850         int ret;
1851 
1852         atomic_inc(&BTRFS_I(inode)->sync_writers);
1853         ret = btrfs_fdatawrite_range(inode, start, end);
1854         atomic_dec(&BTRFS_I(inode)->sync_writers);
1855 
1856         return ret;
1857 }
1858 
1859 /*
1860  * fsync call for both files and directories.  This logs the inode into
1861  * the tree log instead of forcing full commits whenever possible.
1862  *
1863  * It needs to call filemap_fdatawait so that all ordered extent updates are
1864  * in the metadata btree are up to date for copying to the log.
1865  *
1866  * It drops the inode mutex before doing the tree log commit.  This is an
1867  * important optimization for directories because holding the mutex prevents
1868  * new operations on the dir while we write to disk.
1869  */
1870 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1871 {
1872         struct dentry *dentry = file->f_path.dentry;
1873         struct inode *inode = dentry->d_inode;
1874         struct btrfs_root *root = BTRFS_I(inode)->root;
1875         struct btrfs_trans_handle *trans;
1876         struct btrfs_log_ctx ctx;
1877         int ret = 0;
1878         bool full_sync = 0;
1879 
1880         trace_btrfs_sync_file(file, datasync);
1881 
1882         /*
1883          * We write the dirty pages in the range and wait until they complete
1884          * out of the ->i_mutex. If so, we can flush the dirty pages by
1885          * multi-task, and make the performance up.  See
1886          * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1887          */
1888         ret = start_ordered_ops(inode, start, end);
1889         if (ret)
1890                 return ret;
1891 
1892         mutex_lock(&inode->i_mutex);
1893         atomic_inc(&root->log_batch);
1894         full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1895                              &BTRFS_I(inode)->runtime_flags);
1896         /*
1897          * We might have have had more pages made dirty after calling
1898          * start_ordered_ops and before acquiring the inode's i_mutex.
1899          */
1900         if (full_sync) {
1901                 /*
1902                  * For a full sync, we need to make sure any ordered operations
1903                  * start and finish before we start logging the inode, so that
1904                  * all extents are persisted and the respective file extent
1905                  * items are in the fs/subvol btree.
1906                  */
1907                 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1908         } else {
1909                 /*
1910                  * Start any new ordered operations before starting to log the
1911                  * inode. We will wait for them to finish in btrfs_sync_log().
1912                  *
1913                  * Right before acquiring the inode's mutex, we might have new
1914                  * writes dirtying pages, which won't immediately start the
1915                  * respective ordered operations - that is done through the
1916                  * fill_delalloc callbacks invoked from the writepage and
1917                  * writepages address space operations. So make sure we start
1918                  * all ordered operations before starting to log our inode. Not
1919                  * doing this means that while logging the inode, writeback
1920                  * could start and invoke writepage/writepages, which would call
1921                  * the fill_delalloc callbacks (cow_file_range,
1922                  * submit_compressed_extents). These callbacks add first an
1923                  * extent map to the modified list of extents and then create
1924                  * the respective ordered operation, which means in
1925                  * tree-log.c:btrfs_log_inode() we might capture all existing
1926                  * ordered operations (with btrfs_get_logged_extents()) before
1927                  * the fill_delalloc callback adds its ordered operation, and by
1928                  * the time we visit the modified list of extent maps (with
1929                  * btrfs_log_changed_extents()), we see and process the extent
1930                  * map they created. We then use the extent map to construct a
1931                  * file extent item for logging without waiting for the
1932                  * respective ordered operation to finish - this file extent
1933                  * item points to a disk location that might not have yet been
1934                  * written to, containing random data - so after a crash a log
1935                  * replay will make our inode have file extent items that point
1936                  * to disk locations containing invalid data, as we returned
1937                  * success to userspace without waiting for the respective
1938                  * ordered operation to finish, because it wasn't captured by
1939                  * btrfs_get_logged_extents().
1940                  */
1941                 ret = start_ordered_ops(inode, start, end);
1942         }
1943         if (ret) {
1944                 mutex_unlock(&inode->i_mutex);
1945                 goto out;
1946         }
1947         atomic_inc(&root->log_batch);
1948 
1949         /*
1950          * If the last transaction that changed this file was before the current
1951          * transaction and we have the full sync flag set in our inode, we can
1952          * bail out now without any syncing.
1953          *
1954          * Note that we can't bail out if the full sync flag isn't set. This is
1955          * because when the full sync flag is set we start all ordered extents
1956          * and wait for them to fully complete - when they complete they update
1957          * the inode's last_trans field through:
1958          *
1959          *     btrfs_finish_ordered_io() ->
1960          *         btrfs_update_inode_fallback() ->
1961          *             btrfs_update_inode() ->
1962          *                 btrfs_set_inode_last_trans()
1963          *
1964          * So we are sure that last_trans is up to date and can do this check to
1965          * bail out safely. For the fast path, when the full sync flag is not
1966          * set in our inode, we can not do it because we start only our ordered
1967          * extents and don't wait for them to complete (that is when
1968          * btrfs_finish_ordered_io runs), so here at this point their last_trans
1969          * value might be less than or equals to fs_info->last_trans_committed,
1970          * and setting a speculative last_trans for an inode when a buffered
1971          * write is made (such as fs_info->generation + 1 for example) would not
1972          * be reliable since after setting the value and before fsync is called
1973          * any number of transactions can start and commit (transaction kthread
1974          * commits the current transaction periodically), and a transaction
1975          * commit does not start nor waits for ordered extents to complete.
1976          */
1977         smp_mb();
1978         if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1979             (full_sync && BTRFS_I(inode)->last_trans <=
1980              root->fs_info->last_trans_committed)) {
1981                 /*
1982                  * We'v had everything committed since the last time we were
1983                  * modified so clear this flag in case it was set for whatever
1984                  * reason, it's no longer relevant.
1985                  */
1986                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1987                           &BTRFS_I(inode)->runtime_flags);
1988                 mutex_unlock(&inode->i_mutex);
1989                 goto out;
1990         }
1991 
1992         /*
1993          * ok we haven't committed the transaction yet, lets do a commit
1994          */
1995         if (file->private_data)
1996                 btrfs_ioctl_trans_end(file);
1997 
1998         /*
1999          * We use start here because we will need to wait on the IO to complete
2000          * in btrfs_sync_log, which could require joining a transaction (for
2001          * example checking cross references in the nocow path).  If we use join
2002          * here we could get into a situation where we're waiting on IO to
2003          * happen that is blocked on a transaction trying to commit.  With start
2004          * we inc the extwriter counter, so we wait for all extwriters to exit
2005          * before we start blocking join'ers.  This comment is to keep somebody
2006          * from thinking they are super smart and changing this to
2007          * btrfs_join_transaction *cough*Josef*cough*.
2008          */
2009         trans = btrfs_start_transaction(root, 0);
2010         if (IS_ERR(trans)) {
2011                 ret = PTR_ERR(trans);
2012                 mutex_unlock(&inode->i_mutex);
2013                 goto out;
2014         }
2015         trans->sync = true;
2016 
2017         btrfs_init_log_ctx(&ctx);
2018 
2019         ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2020         if (ret < 0) {
2021                 /* Fallthrough and commit/free transaction. */
2022                 ret = 1;
2023         }
2024 
2025         /* we've logged all the items and now have a consistent
2026          * version of the file in the log.  It is possible that
2027          * someone will come in and modify the file, but that's
2028          * fine because the log is consistent on disk, and we
2029          * have references to all of the file's extents
2030          *
2031          * It is possible that someone will come in and log the
2032          * file again, but that will end up using the synchronization
2033          * inside btrfs_sync_log to keep things safe.
2034          */
2035         mutex_unlock(&inode->i_mutex);
2036 
2037         /*
2038          * If any of the ordered extents had an error, just return it to user
2039          * space, so that the application knows some writes didn't succeed and
2040          * can take proper action (retry for e.g.). Blindly committing the
2041          * transaction in this case, would fool userspace that everything was
2042          * successful. And we also want to make sure our log doesn't contain
2043          * file extent items pointing to extents that weren't fully written to -
2044          * just like in the non fast fsync path, where we check for the ordered
2045          * operation's error flag before writing to the log tree and return -EIO
2046          * if any of them had this flag set (btrfs_wait_ordered_range) -
2047          * therefore we need to check for errors in the ordered operations,
2048          * which are indicated by ctx.io_err.
2049          */
2050         if (ctx.io_err) {
2051                 btrfs_end_transaction(trans, root);
2052                 ret = ctx.io_err;
2053                 goto out;
2054         }
2055 
2056         if (ret != BTRFS_NO_LOG_SYNC) {
2057                 if (!ret) {
2058                         ret = btrfs_sync_log(trans, root, &ctx);
2059                         if (!ret) {
2060                                 ret = btrfs_end_transaction(trans, root);
2061                                 goto out;
2062                         }
2063                 }
2064                 if (!full_sync) {
2065                         ret = btrfs_wait_ordered_range(inode, start,
2066                                                        end - start + 1);
2067                         if (ret) {
2068                                 btrfs_end_transaction(trans, root);
2069                                 goto out;
2070                         }
2071                 }
2072                 ret = btrfs_commit_transaction(trans, root);
2073         } else {
2074                 ret = btrfs_end_transaction(trans, root);
2075         }
2076 out:
2077         return ret > 0 ? -EIO : ret;
2078 }
2079 
2080 static const struct vm_operations_struct btrfs_file_vm_ops = {
2081         .fault          = filemap_fault,
2082         .map_pages      = filemap_map_pages,
2083         .page_mkwrite   = btrfs_page_mkwrite,
2084 };
2085 
2086 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
2087 {
2088         struct address_space *mapping = filp->f_mapping;
2089 
2090         if (!mapping->a_ops->readpage)
2091                 return -ENOEXEC;
2092 
2093         file_accessed(filp);
2094         vma->vm_ops = &btrfs_file_vm_ops;
2095 
2096         return 0;
2097 }
2098 
2099 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2100                           int slot, u64 start, u64 end)
2101 {
2102         struct btrfs_file_extent_item *fi;
2103         struct btrfs_key key;
2104 
2105         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2106                 return 0;
2107 
2108         btrfs_item_key_to_cpu(leaf, &key, slot);
2109         if (key.objectid != btrfs_ino(inode) ||
2110             key.type != BTRFS_EXTENT_DATA_KEY)
2111                 return 0;
2112 
2113         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2114 
2115         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2116                 return 0;
2117 
2118         if (btrfs_file_extent_disk_bytenr(leaf, fi))
2119                 return 0;
2120 
2121         if (key.offset == end)
2122                 return 1;
2123         if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2124                 return 1;
2125         return 0;
2126 }
2127 
2128 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2129                       struct btrfs_path *path, u64 offset, u64 end)
2130 {
2131         struct btrfs_root *root = BTRFS_I(inode)->root;
2132         struct extent_buffer *leaf;
2133         struct btrfs_file_extent_item *fi;
2134         struct extent_map *hole_em;
2135         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2136         struct btrfs_key key;
2137         int ret;
2138 
2139         if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2140                 goto out;
2141 
2142         key.objectid = btrfs_ino(inode);
2143         key.type = BTRFS_EXTENT_DATA_KEY;
2144         key.offset = offset;
2145 
2146         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2147         if (ret < 0)
2148                 return ret;
2149         BUG_ON(!ret);
2150 
2151         leaf = path->nodes[0];
2152         if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2153                 u64 num_bytes;
2154 
2155                 path->slots[0]--;
2156                 fi = btrfs_item_ptr(leaf, path->slots[0],
2157                                     struct btrfs_file_extent_item);
2158                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2159                         end - offset;
2160                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2161                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2162                 btrfs_set_file_extent_offset(leaf, fi, 0);
2163                 btrfs_mark_buffer_dirty(leaf);
2164                 goto out;
2165         }
2166 
2167         if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2168                 u64 num_bytes;
2169 
2170                 key.offset = offset;
2171                 btrfs_set_item_key_safe(root, path, &key);
2172                 fi = btrfs_item_ptr(leaf, path->slots[0],
2173                                     struct btrfs_file_extent_item);
2174                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2175                         offset;
2176                 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2177                 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2178                 btrfs_set_file_extent_offset(leaf, fi, 0);
2179                 btrfs_mark_buffer_dirty(leaf);
2180                 goto out;
2181         }
2182         btrfs_release_path(path);
2183 
2184         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2185                                        0, 0, end - offset, 0, end - offset,
2186                                        0, 0, 0);
2187         if (ret)
2188                 return ret;
2189 
2190 out:
2191         btrfs_release_path(path);
2192 
2193         hole_em = alloc_extent_map();
2194         if (!hole_em) {
2195                 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2196                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2197                         &BTRFS_I(inode)->runtime_flags);
2198         } else {
2199                 hole_em->start = offset;
2200                 hole_em->len = end - offset;
2201                 hole_em->ram_bytes = hole_em->len;
2202                 hole_em->orig_start = offset;
2203 
2204                 hole_em->block_start = EXTENT_MAP_HOLE;
2205                 hole_em->block_len = 0;
2206                 hole_em->orig_block_len = 0;
2207                 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2208                 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2209                 hole_em->generation = trans->transid;
2210 
2211                 do {
2212                         btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2213                         write_lock(&em_tree->lock);
2214                         ret = add_extent_mapping(em_tree, hole_em, 1);
2215                         write_unlock(&em_tree->lock);
2216                 } while (ret == -EEXIST);
2217                 free_extent_map(hole_em);
2218                 if (ret)
2219                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2220                                 &BTRFS_I(inode)->runtime_flags);
2221         }
2222 
2223         return 0;
2224 }
2225 
2226 /*
2227  * Find a hole extent on given inode and change start/len to the end of hole
2228  * extent.(hole/vacuum extent whose em->start <= start &&
2229  *         em->start + em->len > start)
2230  * When a hole extent is found, return 1 and modify start/len.
2231  */
2232 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2233 {
2234         struct extent_map *em;
2235         int ret = 0;
2236 
2237         em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
2238         if (IS_ERR_OR_NULL(em)) {
2239                 if (!em)
2240                         ret = -ENOMEM;
2241                 else
2242                         ret = PTR_ERR(em);
2243                 return ret;
2244         }
2245 
2246         /* Hole or vacuum extent(only exists in no-hole mode) */
2247         if (em->block_start == EXTENT_MAP_HOLE) {
2248                 ret = 1;
2249                 *len = em->start + em->len > *start + *len ?
2250                        0 : *start + *len - em->start - em->len;
2251                 *start = em->start + em->len;
2252         }
2253         free_extent_map(em);
2254         return ret;
2255 }
2256 
2257 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2258 {
2259         struct btrfs_root *root = BTRFS_I(inode)->root;
2260         struct extent_state *cached_state = NULL;
2261         struct btrfs_path *path;
2262         struct btrfs_block_rsv *rsv;
2263         struct btrfs_trans_handle *trans;
2264         u64 lockstart;
2265         u64 lockend;
2266         u64 tail_start;
2267         u64 tail_len;
2268         u64 orig_start = offset;
2269         u64 cur_offset;
2270         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2271         u64 drop_end;
2272         int ret = 0;
2273         int err = 0;
2274         int rsv_count;
2275         bool same_page;
2276         bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2277         u64 ino_size;
2278         bool truncated_page = false;
2279         bool updated_inode = false;
2280 
2281         ret = btrfs_wait_ordered_range(inode, offset, len);
2282         if (ret)
2283                 return ret;
2284 
2285         mutex_lock(&inode->i_mutex);
2286         ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2287         ret = find_first_non_hole(inode, &offset, &len);
2288         if (ret < 0)
2289                 goto out_only_mutex;
2290         if (ret && !len) {
2291                 /* Already in a large hole */
2292                 ret = 0;
2293                 goto out_only_mutex;
2294         }
2295 
2296         lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2297         lockend = round_down(offset + len,
2298                              BTRFS_I(inode)->root->sectorsize) - 1;
2299         same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2300                     ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2301 
2302         /*
2303          * We needn't truncate any page which is beyond the end of the file
2304          * because we are sure there is no data there.
2305          */
2306         /*
2307          * Only do this if we are in the same page and we aren't doing the
2308          * entire page.
2309          */
2310         if (same_page && len < PAGE_CACHE_SIZE) {
2311                 if (offset < ino_size) {
2312                         truncated_page = true;
2313                         ret = btrfs_truncate_page(inode, offset, len, 0);
2314                 } else {
2315                         ret = 0;
2316                 }
2317                 goto out_only_mutex;
2318         }
2319 
2320         /* zero back part of the first page */
2321         if (offset < ino_size) {
2322                 truncated_page = true;
2323                 ret = btrfs_truncate_page(inode, offset, 0, 0);
2324                 if (ret) {
2325                         mutex_unlock(&inode->i_mutex);
2326                         return ret;
2327                 }
2328         }
2329 
2330         /* Check the aligned pages after the first unaligned page,
2331          * if offset != orig_start, which means the first unaligned page
2332          * including serveral following pages are already in holes,
2333          * the extra check can be skipped */
2334         if (offset == orig_start) {
2335                 /* after truncate page, check hole again */
2336                 len = offset + len - lockstart;
2337                 offset = lockstart;
2338                 ret = find_first_non_hole(inode, &offset, &len);
2339                 if (ret < 0)
2340                         goto out_only_mutex;
2341                 if (ret && !len) {
2342                         ret = 0;
2343                         goto out_only_mutex;
2344                 }
2345                 lockstart = offset;
2346         }
2347 
2348         /* Check the tail unaligned part is in a hole */
2349         tail_start = lockend + 1;
2350         tail_len = offset + len - tail_start;
2351         if (tail_len) {
2352                 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2353                 if (unlikely(ret < 0))
2354                         goto out_only_mutex;
2355                 if (!ret) {
2356                         /* zero the front end of the last page */
2357                         if (tail_start + tail_len < ino_size) {
2358                                 truncated_page = true;
2359                                 ret = btrfs_truncate_page(inode,
2360                                                 tail_start + tail_len, 0, 1);
2361                                 if (ret)
2362                                         goto out_only_mutex;
2363                         }
2364                 }
2365         }
2366 
2367         if (lockend < lockstart) {
2368                 ret = 0;
2369                 goto out_only_mutex;
2370         }
2371 
2372         while (1) {
2373                 struct btrfs_ordered_extent *ordered;
2374 
2375                 truncate_pagecache_range(inode, lockstart, lockend);
2376 
2377                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2378                                  0, &cached_state);
2379                 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2380 
2381                 /*
2382                  * We need to make sure we have no ordered extents in this range
2383                  * and nobody raced in and read a page in this range, if we did
2384                  * we need to try again.
2385                  */
2386                 if ((!ordered ||
2387                     (ordered->file_offset + ordered->len <= lockstart ||
2388                      ordered->file_offset > lockend)) &&
2389                      !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2390                         if (ordered)
2391                                 btrfs_put_ordered_extent(ordered);
2392                         break;
2393                 }
2394                 if (ordered)
2395                         btrfs_put_ordered_extent(ordered);
2396                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2397                                      lockend, &cached_state, GFP_NOFS);
2398                 ret = btrfs_wait_ordered_range(inode, lockstart,
2399                                                lockend - lockstart + 1);
2400                 if (ret) {
2401                         mutex_unlock(&inode->i_mutex);
2402                         return ret;
2403                 }
2404         }
2405 
2406         path = btrfs_alloc_path();
2407         if (!path) {
2408                 ret = -ENOMEM;
2409                 goto out;
2410         }
2411 
2412         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2413         if (!rsv) {
2414                 ret = -ENOMEM;
2415                 goto out_free;
2416         }
2417         rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2418         rsv->failfast = 1;
2419 
2420         /*
2421          * 1 - update the inode
2422          * 1 - removing the extents in the range
2423          * 1 - adding the hole extent if no_holes isn't set
2424          */
2425         rsv_count = no_holes ? 2 : 3;
2426         trans = btrfs_start_transaction(root, rsv_count);
2427         if (IS_ERR(trans)) {
2428                 err = PTR_ERR(trans);
2429                 goto out_free;
2430         }
2431 
2432         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2433                                       min_size);
2434         BUG_ON(ret);
2435         trans->block_rsv = rsv;
2436 
2437         cur_offset = lockstart;
2438         len = lockend - cur_offset;
2439         while (cur_offset < lockend) {
2440                 ret = __btrfs_drop_extents(trans, root, inode, path,
2441                                            cur_offset, lockend + 1,
2442                                            &drop_end, 1, 0, 0, NULL);
2443                 if (ret != -ENOSPC)
2444                         break;
2445 
2446                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2447 
2448                 if (cur_offset < ino_size) {
2449                         ret = fill_holes(trans, inode, path, cur_offset,
2450                                          drop_end);
2451                         if (ret) {
2452                                 err = ret;
2453                                 break;
2454                         }
2455                 }
2456 
2457                 cur_offset = drop_end;
2458 
2459                 ret = btrfs_update_inode(trans, root, inode);
2460                 if (ret) {
2461                         err = ret;
2462                         break;
2463                 }
2464 
2465                 btrfs_end_transaction(trans, root);
2466                 btrfs_btree_balance_dirty(root);
2467 
2468                 trans = btrfs_start_transaction(root, rsv_count);
2469                 if (IS_ERR(trans)) {
2470                         ret = PTR_ERR(trans);
2471                         trans = NULL;
2472                         break;
2473                 }
2474 
2475                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2476                                               rsv, min_size);
2477                 BUG_ON(ret);    /* shouldn't happen */
2478                 trans->block_rsv = rsv;
2479 
2480                 ret = find_first_non_hole(inode, &cur_offset, &len);
2481                 if (unlikely(ret < 0))
2482                         break;
2483                 if (ret && !len) {
2484                         ret = 0;
2485                         break;
2486                 }
2487         }
2488 
2489         if (ret) {
2490                 err = ret;
2491                 goto out_trans;
2492         }
2493 
2494         trans->block_rsv = &root->fs_info->trans_block_rsv;
2495         /*
2496          * Don't insert file hole extent item if it's for a range beyond eof
2497          * (because it's useless) or if it represents a 0 bytes range (when
2498          * cur_offset == drop_end).
2499          */
2500         if (cur_offset < ino_size && cur_offset < drop_end) {
2501                 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2502                 if (ret) {
2503                         err = ret;
2504                         goto out_trans;
2505                 }
2506         }
2507 
2508 out_trans:
2509         if (!trans)
2510                 goto out_free;
2511 
2512         inode_inc_iversion(inode);
2513         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2514 
2515         trans->block_rsv = &root->fs_info->trans_block_rsv;
2516         ret = btrfs_update_inode(trans, root, inode);
2517         updated_inode = true;
2518         btrfs_end_transaction(trans, root);
2519         btrfs_btree_balance_dirty(root);
2520 out_free:
2521         btrfs_free_path(path);
2522         btrfs_free_block_rsv(root, rsv);
2523 out:
2524         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2525                              &cached_state, GFP_NOFS);
2526 out_only_mutex:
2527         if (!updated_inode && truncated_page && !ret && !err) {
2528                 /*
2529                  * If we only end up zeroing part of a page, we still need to
2530                  * update the inode item, so that all the time fields are
2531                  * updated as well as the necessary btrfs inode in memory fields
2532                  * for detecting, at fsync time, if the inode isn't yet in the
2533                  * log tree or it's there but not up to date.
2534                  */
2535                 trans = btrfs_start_transaction(root, 1);
2536                 if (IS_ERR(trans)) {
2537                         err = PTR_ERR(trans);
2538                 } else {
2539                         err = btrfs_update_inode(trans, root, inode);
2540                         ret = btrfs_end_transaction(trans, root);
2541                 }
2542         }
2543         mutex_unlock(&inode->i_mutex);
2544         if (ret && !err)
2545                 err = ret;
2546         return err;
2547 }
2548 
2549 static long btrfs_fallocate(struct file *file, int mode,
2550                             loff_t offset, loff_t len)
2551 {
2552         struct inode *inode = file_inode(file);
2553         struct extent_state *cached_state = NULL;
2554         struct btrfs_root *root = BTRFS_I(inode)->root;
2555         u64 cur_offset;
2556         u64 last_byte;
2557         u64 alloc_start;
2558         u64 alloc_end;
2559         u64 alloc_hint = 0;
2560         u64 locked_end;
2561         struct extent_map *em;
2562         int blocksize = BTRFS_I(inode)->root->sectorsize;
2563         int ret;
2564 
2565         alloc_start = round_down(offset, blocksize);
2566         alloc_end = round_up(offset + len, blocksize);
2567 
2568         /* Make sure we aren't being give some crap mode */
2569         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2570                 return -EOPNOTSUPP;
2571 
2572         if (mode & FALLOC_FL_PUNCH_HOLE)
2573                 return btrfs_punch_hole(inode, offset, len);
2574 
2575         /*
2576          * Make sure we have enough space before we do the
2577          * allocation.
2578          */
2579         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2580         if (ret)
2581                 return ret;
2582         if (root->fs_info->quota_enabled) {
2583                 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2584                 if (ret)
2585                         goto out_reserve_fail;
2586         }
2587 
2588         mutex_lock(&inode->i_mutex);
2589         ret = inode_newsize_ok(inode, alloc_end);
2590         if (ret)
2591                 goto out;
2592 
2593         if (alloc_start > inode->i_size) {
2594                 ret = btrfs_cont_expand(inode, i_size_read(inode),
2595                                         alloc_start);
2596                 if (ret)
2597                         goto out;
2598         } else {
2599                 /*
2600                  * If we are fallocating from the end of the file onward we
2601                  * need to zero out the end of the page if i_size lands in the
2602                  * middle of a page.
2603                  */
2604                 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2605                 if (ret)
2606                         goto out;
2607         }
2608 
2609         /*
2610          * wait for ordered IO before we have any locks.  We'll loop again
2611          * below with the locks held.
2612          */
2613         ret = btrfs_wait_ordered_range(inode, alloc_start,
2614                                        alloc_end - alloc_start);
2615         if (ret)
2616                 goto out;
2617 
2618         locked_end = alloc_end - 1;
2619         while (1) {
2620                 struct btrfs_ordered_extent *ordered;
2621 
2622                 /* the extent lock is ordered inside the running
2623                  * transaction
2624                  */
2625                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2626                                  locked_end, 0, &cached_state);
2627                 ordered = btrfs_lookup_first_ordered_extent(inode,
2628                                                             alloc_end - 1);
2629                 if (ordered &&
2630                     ordered->file_offset + ordered->len > alloc_start &&
2631                     ordered->file_offset < alloc_end) {
2632                         btrfs_put_ordered_extent(ordered);
2633                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2634                                              alloc_start, locked_end,
2635                                              &cached_state, GFP_NOFS);
2636                         /*
2637                          * we can't wait on the range with the transaction
2638                          * running or with the extent lock held
2639                          */
2640                         ret = btrfs_wait_ordered_range(inode, alloc_start,
2641                                                        alloc_end - alloc_start);
2642                         if (ret)
2643                                 goto out;
2644                 } else {
2645                         if (ordered)
2646                                 btrfs_put_ordered_extent(ordered);
2647                         break;
2648                 }
2649         }
2650 
2651         cur_offset = alloc_start;
2652         while (1) {
2653                 u64 actual_end;
2654 
2655                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2656                                       alloc_end - cur_offset, 0);
2657                 if (IS_ERR_OR_NULL(em)) {
2658                         if (!em)
2659                                 ret = -ENOMEM;
2660                         else
2661                                 ret = PTR_ERR(em);
2662                         break;
2663                 }
2664                 last_byte = min(extent_map_end(em), alloc_end);
2665                 actual_end = min_t(u64, extent_map_end(em), offset + len);
2666                 last_byte = ALIGN(last_byte, blocksize);
2667 
2668                 if (em->block_start == EXTENT_MAP_HOLE ||
2669                     (cur_offset >= inode->i_size &&
2670                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2671                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2672                                                         last_byte - cur_offset,
2673                                                         1 << inode->i_blkbits,
2674                                                         offset + len,
2675                                                         &alloc_hint);
2676 
2677                         if (ret < 0) {
2678                                 free_extent_map(em);
2679                                 break;
2680                         }
2681                 } else if (actual_end > inode->i_size &&
2682                            !(mode & FALLOC_FL_KEEP_SIZE)) {
2683                         /*
2684                          * We didn't need to allocate any more space, but we
2685                          * still extended the size of the file so we need to
2686                          * update i_size.
2687                          */
2688                         inode->i_ctime = CURRENT_TIME;
2689                         i_size_write(inode, actual_end);
2690                         btrfs_ordered_update_i_size(inode, actual_end, NULL);
2691                 }
2692                 free_extent_map(em);
2693 
2694                 cur_offset = last_byte;
2695                 if (cur_offset >= alloc_end) {
2696                         ret = 0;
2697                         break;
2698                 }
2699         }
2700         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2701                              &cached_state, GFP_NOFS);
2702 out:
2703         mutex_unlock(&inode->i_mutex);
2704         if (root->fs_info->quota_enabled)
2705                 btrfs_qgroup_free(root, alloc_end - alloc_start);
2706 out_reserve_fail:
2707         /* Let go of our reservation. */
2708         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2709         return ret;
2710 }
2711 
2712 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2713 {
2714         struct btrfs_root *root = BTRFS_I(inode)->root;
2715         struct extent_map *em = NULL;
2716         struct extent_state *cached_state = NULL;
2717         u64 lockstart;
2718         u64 lockend;
2719         u64 start;
2720         u64 len;
2721         int ret = 0;
2722 
2723         if (inode->i_size == 0)
2724                 return -ENXIO;
2725 
2726         /*
2727          * *offset can be negative, in this case we start finding DATA/HOLE from
2728          * the very start of the file.
2729          */
2730         start = max_t(loff_t, 0, *offset);
2731 
2732         lockstart = round_down(start, root->sectorsize);
2733         lockend = round_up(i_size_read(inode), root->sectorsize);
2734         if (lockend <= lockstart)
2735                 lockend = lockstart + root->sectorsize;
2736         lockend--;
2737         len = lockend - lockstart + 1;
2738 
2739         lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2740                          &cached_state);
2741 
2742         while (start < inode->i_size) {
2743                 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2744                 if (IS_ERR(em)) {
2745                         ret = PTR_ERR(em);
2746                         em = NULL;
2747                         break;
2748                 }
2749 
2750                 if (whence == SEEK_HOLE &&
2751                     (em->block_start == EXTENT_MAP_HOLE ||
2752                      test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2753                         break;
2754                 else if (whence == SEEK_DATA &&
2755                            (em->block_start != EXTENT_MAP_HOLE &&
2756                             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2757                         break;
2758 
2759                 start = em->start + em->len;
2760                 free_extent_map(em);
2761                 em = NULL;
2762                 cond_resched();
2763         }
2764         free_extent_map(em);
2765         if (!ret) {
2766                 if (whence == SEEK_DATA && start >= inode->i_size)
2767                         ret = -ENXIO;
2768                 else
2769                         *offset = min_t(loff_t, start, inode->i_size);
2770         }
2771         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2772                              &cached_state, GFP_NOFS);
2773         return ret;
2774 }
2775 
2776 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2777 {
2778         struct inode *inode = file->f_mapping->host;
2779         int ret;
2780 
2781         mutex_lock(&inode->i_mutex);
2782         switch (whence) {
2783         case SEEK_END:
2784         case SEEK_CUR:
2785                 offset = generic_file_llseek(file, offset, whence);
2786                 goto out;
2787         case SEEK_DATA:
2788         case SEEK_HOLE:
2789                 if (offset >= i_size_read(inode)) {
2790                         mutex_unlock(&inode->i_mutex);
2791                         return -ENXIO;
2792                 }
2793 
2794                 ret = find_desired_extent(inode, &offset, whence);
2795                 if (ret) {
2796                         mutex_unlock(&inode->i_mutex);
2797                         return ret;
2798                 }
2799         }
2800 
2801         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2802 out:
2803         mutex_unlock(&inode->i_mutex);
2804         return offset;
2805 }
2806 
2807 const struct file_operations btrfs_file_operations = {
2808         .llseek         = btrfs_file_llseek,
2809         .read           = new_sync_read,
2810         .write          = new_sync_write,
2811         .read_iter      = generic_file_read_iter,
2812         .splice_read    = generic_file_splice_read,
2813         .write_iter     = btrfs_file_write_iter,
2814         .mmap           = btrfs_file_mmap,
2815         .open           = generic_file_open,
2816         .release        = btrfs_release_file,
2817         .fsync          = btrfs_sync_file,
2818         .fallocate      = btrfs_fallocate,
2819         .unlocked_ioctl = btrfs_ioctl,
2820 #ifdef CONFIG_COMPAT
2821         .compat_ioctl   = btrfs_ioctl,
2822 #endif
2823 };
2824 
2825 void btrfs_auto_defrag_exit(void)
2826 {
2827         if (btrfs_inode_defrag_cachep)
2828                 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2829 }
2830 
2831 int btrfs_auto_defrag_init(void)
2832 {
2833         btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2834                                         sizeof(struct inode_defrag), 0,
2835                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2836                                         NULL);
2837         if (!btrfs_inode_defrag_cachep)
2838                 return -ENOMEM;
2839 
2840         return 0;
2841 }
2842 
2843 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
2844 {
2845         int ret;
2846 
2847         /*
2848          * So with compression we will find and lock a dirty page and clear the
2849          * first one as dirty, setup an async extent, and immediately return
2850          * with the entire range locked but with nobody actually marked with
2851          * writeback.  So we can't just filemap_write_and_wait_range() and
2852          * expect it to work since it will just kick off a thread to do the
2853          * actual work.  So we need to call filemap_fdatawrite_range _again_
2854          * since it will wait on the page lock, which won't be unlocked until
2855          * after the pages have been marked as writeback and so we're good to go
2856          * from there.  We have to do this otherwise we'll miss the ordered
2857          * extents and that results in badness.  Please Josef, do not think you
2858          * know better and pull this out at some point in the future, it is
2859          * right and you are wrong.
2860          */
2861         ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
2862         if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
2863                              &BTRFS_I(inode)->runtime_flags))
2864                 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
2865 
2866         return ret;
2867 }
2868 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp