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

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

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ 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/kernel.h>
 20 #include <linux/bio.h>
 21 #include <linux/buffer_head.h>
 22 #include <linux/file.h>
 23 #include <linux/fs.h>
 24 #include <linux/pagemap.h>
 25 #include <linux/highmem.h>
 26 #include <linux/time.h>
 27 #include <linux/init.h>
 28 #include <linux/string.h>
 29 #include <linux/backing-dev.h>
 30 #include <linux/mpage.h>
 31 #include <linux/swap.h>
 32 #include <linux/writeback.h>
 33 #include <linux/statfs.h>
 34 #include <linux/compat.h>
 35 #include <linux/bit_spinlock.h>
 36 #include <linux/xattr.h>
 37 #include <linux/posix_acl.h>
 38 #include <linux/falloc.h>
 39 #include <linux/slab.h>
 40 #include <linux/ratelimit.h>
 41 #include <linux/mount.h>
 42 #include <linux/btrfs.h>
 43 #include <linux/blkdev.h>
 44 #include <linux/posix_acl_xattr.h>
 45 #include <linux/uio.h>
 46 #include "ctree.h"
 47 #include "disk-io.h"
 48 #include "transaction.h"
 49 #include "btrfs_inode.h"
 50 #include "print-tree.h"
 51 #include "ordered-data.h"
 52 #include "xattr.h"
 53 #include "tree-log.h"
 54 #include "volumes.h"
 55 #include "compression.h"
 56 #include "locking.h"
 57 #include "free-space-cache.h"
 58 #include "inode-map.h"
 59 #include "backref.h"
 60 #include "hash.h"
 61 #include "props.h"
 62 #include "qgroup.h"
 63 
 64 struct btrfs_iget_args {
 65         struct btrfs_key *location;
 66         struct btrfs_root *root;
 67 };
 68 
 69 struct btrfs_dio_data {
 70         u64 outstanding_extents;
 71         u64 reserve;
 72         u64 unsubmitted_oe_range_start;
 73         u64 unsubmitted_oe_range_end;
 74 };
 75 
 76 static const struct inode_operations btrfs_dir_inode_operations;
 77 static const struct inode_operations btrfs_symlink_inode_operations;
 78 static const struct inode_operations btrfs_dir_ro_inode_operations;
 79 static const struct inode_operations btrfs_special_inode_operations;
 80 static const struct inode_operations btrfs_file_inode_operations;
 81 static const struct address_space_operations btrfs_aops;
 82 static const struct address_space_operations btrfs_symlink_aops;
 83 static const struct file_operations btrfs_dir_file_operations;
 84 static const struct extent_io_ops btrfs_extent_io_ops;
 85 
 86 static struct kmem_cache *btrfs_inode_cachep;
 87 struct kmem_cache *btrfs_trans_handle_cachep;
 88 struct kmem_cache *btrfs_transaction_cachep;
 89 struct kmem_cache *btrfs_path_cachep;
 90 struct kmem_cache *btrfs_free_space_cachep;
 91 
 92 #define S_SHIFT 12
 93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
 94         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
 95         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
 96         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
 97         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
 98         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
 99         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
100         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
101 };
102 
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107                                    struct page *locked_page,
108                                    u64 start, u64 end, int *page_started,
109                                    unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111                                            u64 len, u64 orig_start,
112                                            u64 block_start, u64 block_len,
113                                            u64 orig_block_len, u64 ram_bytes,
114                                            int type);
115 
116 static int btrfs_dirty_inode(struct inode *inode);
117 
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
120 {
121         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
122 }
123 #endif
124 
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126                                      struct inode *inode,  struct inode *dir,
127                                      const struct qstr *qstr)
128 {
129         int err;
130 
131         err = btrfs_init_acl(trans, inode, dir);
132         if (!err)
133                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
134         return err;
135 }
136 
137 /*
138  * this does all the hard work for inserting an inline extent into
139  * the btree.  The caller should have done a btrfs_drop_extents so that
140  * no overlapping inline items exist in the btree
141  */
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143                                 struct btrfs_path *path, int extent_inserted,
144                                 struct btrfs_root *root, struct inode *inode,
145                                 u64 start, size_t size, size_t compressed_size,
146                                 int compress_type,
147                                 struct page **compressed_pages)
148 {
149         struct extent_buffer *leaf;
150         struct page *page = NULL;
151         char *kaddr;
152         unsigned long ptr;
153         struct btrfs_file_extent_item *ei;
154         int err = 0;
155         int ret;
156         size_t cur_size = size;
157         unsigned long offset;
158 
159         if (compressed_size && compressed_pages)
160                 cur_size = compressed_size;
161 
162         inode_add_bytes(inode, size);
163 
164         if (!extent_inserted) {
165                 struct btrfs_key key;
166                 size_t datasize;
167 
168                 key.objectid = btrfs_ino(inode);
169                 key.offset = start;
170                 key.type = BTRFS_EXTENT_DATA_KEY;
171 
172                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173                 path->leave_spinning = 1;
174                 ret = btrfs_insert_empty_item(trans, root, path, &key,
175                                               datasize);
176                 if (ret) {
177                         err = ret;
178                         goto fail;
179                 }
180         }
181         leaf = path->nodes[0];
182         ei = btrfs_item_ptr(leaf, path->slots[0],
183                             struct btrfs_file_extent_item);
184         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186         btrfs_set_file_extent_encryption(leaf, ei, 0);
187         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189         ptr = btrfs_file_extent_inline_start(ei);
190 
191         if (compress_type != BTRFS_COMPRESS_NONE) {
192                 struct page *cpage;
193                 int i = 0;
194                 while (compressed_size > 0) {
195                         cpage = compressed_pages[i];
196                         cur_size = min_t(unsigned long, compressed_size,
197                                        PAGE_CACHE_SIZE);
198 
199                         kaddr = kmap_atomic(cpage);
200                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
201                         kunmap_atomic(kaddr);
202 
203                         i++;
204                         ptr += cur_size;
205                         compressed_size -= cur_size;
206                 }
207                 btrfs_set_file_extent_compression(leaf, ei,
208                                                   compress_type);
209         } else {
210                 page = find_get_page(inode->i_mapping,
211                                      start >> PAGE_CACHE_SHIFT);
212                 btrfs_set_file_extent_compression(leaf, ei, 0);
213                 kaddr = kmap_atomic(page);
214                 offset = start & (PAGE_CACHE_SIZE - 1);
215                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216                 kunmap_atomic(kaddr);
217                 page_cache_release(page);
218         }
219         btrfs_mark_buffer_dirty(leaf);
220         btrfs_release_path(path);
221 
222         /*
223          * we're an inline extent, so nobody can
224          * extend the file past i_size without locking
225          * a page we already have locked.
226          *
227          * We must do any isize and inode updates
228          * before we unlock the pages.  Otherwise we
229          * could end up racing with unlink.
230          */
231         BTRFS_I(inode)->disk_i_size = inode->i_size;
232         ret = btrfs_update_inode(trans, root, inode);
233 
234         return ret;
235 fail:
236         return err;
237 }
238 
239 
240 /*
241  * conditionally insert an inline extent into the file.  This
242  * does the checks required to make sure the data is small enough
243  * to fit as an inline extent.
244  */
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246                                           struct inode *inode, u64 start,
247                                           u64 end, size_t compressed_size,
248                                           int compress_type,
249                                           struct page **compressed_pages)
250 {
251         struct btrfs_trans_handle *trans;
252         u64 isize = i_size_read(inode);
253         u64 actual_end = min(end + 1, isize);
254         u64 inline_len = actual_end - start;
255         u64 aligned_end = ALIGN(end, root->sectorsize);
256         u64 data_len = inline_len;
257         int ret;
258         struct btrfs_path *path;
259         int extent_inserted = 0;
260         u32 extent_item_size;
261 
262         if (compressed_size)
263                 data_len = compressed_size;
264 
265         if (start > 0 ||
266             actual_end > PAGE_CACHE_SIZE ||
267             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
268             (!compressed_size &&
269             (actual_end & (root->sectorsize - 1)) == 0) ||
270             end + 1 < isize ||
271             data_len > root->fs_info->max_inline) {
272                 return 1;
273         }
274 
275         path = btrfs_alloc_path();
276         if (!path)
277                 return -ENOMEM;
278 
279         trans = btrfs_join_transaction(root);
280         if (IS_ERR(trans)) {
281                 btrfs_free_path(path);
282                 return PTR_ERR(trans);
283         }
284         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
285 
286         if (compressed_size && compressed_pages)
287                 extent_item_size = btrfs_file_extent_calc_inline_size(
288                    compressed_size);
289         else
290                 extent_item_size = btrfs_file_extent_calc_inline_size(
291                     inline_len);
292 
293         ret = __btrfs_drop_extents(trans, root, inode, path,
294                                    start, aligned_end, NULL,
295                                    1, 1, extent_item_size, &extent_inserted);
296         if (ret) {
297                 btrfs_abort_transaction(trans, root, ret);
298                 goto out;
299         }
300 
301         if (isize > actual_end)
302                 inline_len = min_t(u64, isize, actual_end);
303         ret = insert_inline_extent(trans, path, extent_inserted,
304                                    root, inode, start,
305                                    inline_len, compressed_size,
306                                    compress_type, compressed_pages);
307         if (ret && ret != -ENOSPC) {
308                 btrfs_abort_transaction(trans, root, ret);
309                 goto out;
310         } else if (ret == -ENOSPC) {
311                 ret = 1;
312                 goto out;
313         }
314 
315         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316         btrfs_delalloc_release_metadata(inode, end + 1 - start);
317         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
318 out:
319         /*
320          * Don't forget to free the reserved space, as for inlined extent
321          * it won't count as data extent, free them directly here.
322          * And at reserve time, it's always aligned to page size, so
323          * just free one page here.
324          */
325         btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
326         btrfs_free_path(path);
327         btrfs_end_transaction(trans, root);
328         return ret;
329 }
330 
331 struct async_extent {
332         u64 start;
333         u64 ram_size;
334         u64 compressed_size;
335         struct page **pages;
336         unsigned long nr_pages;
337         int compress_type;
338         struct list_head list;
339 };
340 
341 struct async_cow {
342         struct inode *inode;
343         struct btrfs_root *root;
344         struct page *locked_page;
345         u64 start;
346         u64 end;
347         struct list_head extents;
348         struct btrfs_work work;
349 };
350 
351 static noinline int add_async_extent(struct async_cow *cow,
352                                      u64 start, u64 ram_size,
353                                      u64 compressed_size,
354                                      struct page **pages,
355                                      unsigned long nr_pages,
356                                      int compress_type)
357 {
358         struct async_extent *async_extent;
359 
360         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361         BUG_ON(!async_extent); /* -ENOMEM */
362         async_extent->start = start;
363         async_extent->ram_size = ram_size;
364         async_extent->compressed_size = compressed_size;
365         async_extent->pages = pages;
366         async_extent->nr_pages = nr_pages;
367         async_extent->compress_type = compress_type;
368         list_add_tail(&async_extent->list, &cow->extents);
369         return 0;
370 }
371 
372 static inline int inode_need_compress(struct inode *inode)
373 {
374         struct btrfs_root *root = BTRFS_I(inode)->root;
375 
376         /* force compress */
377         if (btrfs_test_opt(root, FORCE_COMPRESS))
378                 return 1;
379         /* bad compression ratios */
380         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
381                 return 0;
382         if (btrfs_test_opt(root, COMPRESS) ||
383             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384             BTRFS_I(inode)->force_compress)
385                 return 1;
386         return 0;
387 }
388 
389 /*
390  * we create compressed extents in two phases.  The first
391  * phase compresses a range of pages that have already been
392  * locked (both pages and state bits are locked).
393  *
394  * This is done inside an ordered work queue, and the compression
395  * is spread across many cpus.  The actual IO submission is step
396  * two, and the ordered work queue takes care of making sure that
397  * happens in the same order things were put onto the queue by
398  * writepages and friends.
399  *
400  * If this code finds it can't get good compression, it puts an
401  * entry onto the work queue to write the uncompressed bytes.  This
402  * makes sure that both compressed inodes and uncompressed inodes
403  * are written in the same order that the flusher thread sent them
404  * down.
405  */
406 static noinline void compress_file_range(struct inode *inode,
407                                         struct page *locked_page,
408                                         u64 start, u64 end,
409                                         struct async_cow *async_cow,
410                                         int *num_added)
411 {
412         struct btrfs_root *root = BTRFS_I(inode)->root;
413         u64 num_bytes;
414         u64 blocksize = root->sectorsize;
415         u64 actual_end;
416         u64 isize = i_size_read(inode);
417         int ret = 0;
418         struct page **pages = NULL;
419         unsigned long nr_pages;
420         unsigned long nr_pages_ret = 0;
421         unsigned long total_compressed = 0;
422         unsigned long total_in = 0;
423         unsigned long max_compressed = SZ_128K;
424         unsigned long max_uncompressed = SZ_128K;
425         int i;
426         int will_compress;
427         int compress_type = root->fs_info->compress_type;
428         int redirty = 0;
429 
430         /* if this is a small write inside eof, kick off a defrag */
431         if ((end - start + 1) < SZ_16K &&
432             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433                 btrfs_add_inode_defrag(NULL, inode);
434 
435         actual_end = min_t(u64, isize, end + 1);
436 again:
437         will_compress = 0;
438         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
439         nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_SIZE);
440 
441         /*
442          * we don't want to send crud past the end of i_size through
443          * compression, that's just a waste of CPU time.  So, if the
444          * end of the file is before the start of our current
445          * requested range of bytes, we bail out to the uncompressed
446          * cleanup code that can deal with all of this.
447          *
448          * It isn't really the fastest way to fix things, but this is a
449          * very uncommon corner.
450          */
451         if (actual_end <= start)
452                 goto cleanup_and_bail_uncompressed;
453 
454         total_compressed = actual_end - start;
455 
456         /*
457          * skip compression for a small file range(<=blocksize) that
458          * isn't an inline extent, since it dosen't save disk space at all.
459          */
460         if (total_compressed <= blocksize &&
461            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462                 goto cleanup_and_bail_uncompressed;
463 
464         /* we want to make sure that amount of ram required to uncompress
465          * an extent is reasonable, so we limit the total size in ram
466          * of a compressed extent to 128k.  This is a crucial number
467          * because it also controls how easily we can spread reads across
468          * cpus for decompression.
469          *
470          * We also want to make sure the amount of IO required to do
471          * a random read is reasonably small, so we limit the size of
472          * a compressed extent to 128k.
473          */
474         total_compressed = min(total_compressed, max_uncompressed);
475         num_bytes = ALIGN(end - start + 1, blocksize);
476         num_bytes = max(blocksize,  num_bytes);
477         total_in = 0;
478         ret = 0;
479 
480         /*
481          * we do compression for mount -o compress and when the
482          * inode has not been flagged as nocompress.  This flag can
483          * change at any time if we discover bad compression ratios.
484          */
485         if (inode_need_compress(inode)) {
486                 WARN_ON(pages);
487                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
488                 if (!pages) {
489                         /* just bail out to the uncompressed code */
490                         goto cont;
491                 }
492 
493                 if (BTRFS_I(inode)->force_compress)
494                         compress_type = BTRFS_I(inode)->force_compress;
495 
496                 /*
497                  * we need to call clear_page_dirty_for_io on each
498                  * page in the range.  Otherwise applications with the file
499                  * mmap'd can wander in and change the page contents while
500                  * we are compressing them.
501                  *
502                  * If the compression fails for any reason, we set the pages
503                  * dirty again later on.
504                  */
505                 extent_range_clear_dirty_for_io(inode, start, end);
506                 redirty = 1;
507                 ret = btrfs_compress_pages(compress_type,
508                                            inode->i_mapping, start,
509                                            total_compressed, pages,
510                                            nr_pages, &nr_pages_ret,
511                                            &total_in,
512                                            &total_compressed,
513                                            max_compressed);
514 
515                 if (!ret) {
516                         unsigned long offset = total_compressed &
517                                 (PAGE_CACHE_SIZE - 1);
518                         struct page *page = pages[nr_pages_ret - 1];
519                         char *kaddr;
520 
521                         /* zero the tail end of the last page, we might be
522                          * sending it down to disk
523                          */
524                         if (offset) {
525                                 kaddr = kmap_atomic(page);
526                                 memset(kaddr + offset, 0,
527                                        PAGE_CACHE_SIZE - offset);
528                                 kunmap_atomic(kaddr);
529                         }
530                         will_compress = 1;
531                 }
532         }
533 cont:
534         if (start == 0) {
535                 /* lets try to make an inline extent */
536                 if (ret || total_in < (actual_end - start)) {
537                         /* we didn't compress the entire range, try
538                          * to make an uncompressed inline extent.
539                          */
540                         ret = cow_file_range_inline(root, inode, start, end,
541                                                     0, 0, NULL);
542                 } else {
543                         /* try making a compressed inline extent */
544                         ret = cow_file_range_inline(root, inode, start, end,
545                                                     total_compressed,
546                                                     compress_type, pages);
547                 }
548                 if (ret <= 0) {
549                         unsigned long clear_flags = EXTENT_DELALLOC |
550                                 EXTENT_DEFRAG;
551                         unsigned long page_error_op;
552 
553                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
555 
556                         /*
557                          * inline extent creation worked or returned error,
558                          * we don't need to create any more async work items.
559                          * Unlock and free up our temp pages.
560                          */
561                         extent_clear_unlock_delalloc(inode, start, end, NULL,
562                                                      clear_flags, PAGE_UNLOCK |
563                                                      PAGE_CLEAR_DIRTY |
564                                                      PAGE_SET_WRITEBACK |
565                                                      page_error_op |
566                                                      PAGE_END_WRITEBACK);
567                         goto free_pages_out;
568                 }
569         }
570 
571         if (will_compress) {
572                 /*
573                  * we aren't doing an inline extent round the compressed size
574                  * up to a block size boundary so the allocator does sane
575                  * things
576                  */
577                 total_compressed = ALIGN(total_compressed, blocksize);
578 
579                 /*
580                  * one last check to make sure the compression is really a
581                  * win, compare the page count read with the blocks on disk
582                  */
583                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
584                 if (total_compressed >= total_in) {
585                         will_compress = 0;
586                 } else {
587                         num_bytes = total_in;
588                 }
589         }
590         if (!will_compress && pages) {
591                 /*
592                  * the compression code ran but failed to make things smaller,
593                  * free any pages it allocated and our page pointer array
594                  */
595                 for (i = 0; i < nr_pages_ret; i++) {
596                         WARN_ON(pages[i]->mapping);
597                         page_cache_release(pages[i]);
598                 }
599                 kfree(pages);
600                 pages = NULL;
601                 total_compressed = 0;
602                 nr_pages_ret = 0;
603 
604                 /* flag the file so we don't compress in the future */
605                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606                     !(BTRFS_I(inode)->force_compress)) {
607                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
608                 }
609         }
610         if (will_compress) {
611                 *num_added += 1;
612 
613                 /* the async work queues will take care of doing actual
614                  * allocation on disk for these compressed pages,
615                  * and will submit them to the elevator.
616                  */
617                 add_async_extent(async_cow, start, num_bytes,
618                                  total_compressed, pages, nr_pages_ret,
619                                  compress_type);
620 
621                 if (start + num_bytes < end) {
622                         start += num_bytes;
623                         pages = NULL;
624                         cond_resched();
625                         goto again;
626                 }
627         } else {
628 cleanup_and_bail_uncompressed:
629                 /*
630                  * No compression, but we still need to write the pages in
631                  * the file we've been given so far.  redirty the locked
632                  * page if it corresponds to our extent and set things up
633                  * for the async work queue to run cow_file_range to do
634                  * the normal delalloc dance
635                  */
636                 if (page_offset(locked_page) >= start &&
637                     page_offset(locked_page) <= end) {
638                         __set_page_dirty_nobuffers(locked_page);
639                         /* unlocked later on in the async handlers */
640                 }
641                 if (redirty)
642                         extent_range_redirty_for_io(inode, start, end);
643                 add_async_extent(async_cow, start, end - start + 1,
644                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
645                 *num_added += 1;
646         }
647 
648         return;
649 
650 free_pages_out:
651         for (i = 0; i < nr_pages_ret; i++) {
652                 WARN_ON(pages[i]->mapping);
653                 page_cache_release(pages[i]);
654         }
655         kfree(pages);
656 }
657 
658 static void free_async_extent_pages(struct async_extent *async_extent)
659 {
660         int i;
661 
662         if (!async_extent->pages)
663                 return;
664 
665         for (i = 0; i < async_extent->nr_pages; i++) {
666                 WARN_ON(async_extent->pages[i]->mapping);
667                 page_cache_release(async_extent->pages[i]);
668         }
669         kfree(async_extent->pages);
670         async_extent->nr_pages = 0;
671         async_extent->pages = NULL;
672 }
673 
674 /*
675  * phase two of compressed writeback.  This is the ordered portion
676  * of the code, which only gets called in the order the work was
677  * queued.  We walk all the async extents created by compress_file_range
678  * and send them down to the disk.
679  */
680 static noinline void submit_compressed_extents(struct inode *inode,
681                                               struct async_cow *async_cow)
682 {
683         struct async_extent *async_extent;
684         u64 alloc_hint = 0;
685         struct btrfs_key ins;
686         struct extent_map *em;
687         struct btrfs_root *root = BTRFS_I(inode)->root;
688         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689         struct extent_io_tree *io_tree;
690         int ret = 0;
691 
692 again:
693         while (!list_empty(&async_cow->extents)) {
694                 async_extent = list_entry(async_cow->extents.next,
695                                           struct async_extent, list);
696                 list_del(&async_extent->list);
697 
698                 io_tree = &BTRFS_I(inode)->io_tree;
699 
700 retry:
701                 /* did the compression code fall back to uncompressed IO? */
702                 if (!async_extent->pages) {
703                         int page_started = 0;
704                         unsigned long nr_written = 0;
705 
706                         lock_extent(io_tree, async_extent->start,
707                                          async_extent->start +
708                                          async_extent->ram_size - 1);
709 
710                         /* allocate blocks */
711                         ret = cow_file_range(inode, async_cow->locked_page,
712                                              async_extent->start,
713                                              async_extent->start +
714                                              async_extent->ram_size - 1,
715                                              &page_started, &nr_written, 0);
716 
717                         /* JDM XXX */
718 
719                         /*
720                          * if page_started, cow_file_range inserted an
721                          * inline extent and took care of all the unlocking
722                          * and IO for us.  Otherwise, we need to submit
723                          * all those pages down to the drive.
724                          */
725                         if (!page_started && !ret)
726                                 extent_write_locked_range(io_tree,
727                                                   inode, async_extent->start,
728                                                   async_extent->start +
729                                                   async_extent->ram_size - 1,
730                                                   btrfs_get_extent,
731                                                   WB_SYNC_ALL);
732                         else if (ret)
733                                 unlock_page(async_cow->locked_page);
734                         kfree(async_extent);
735                         cond_resched();
736                         continue;
737                 }
738 
739                 lock_extent(io_tree, async_extent->start,
740                             async_extent->start + async_extent->ram_size - 1);
741 
742                 ret = btrfs_reserve_extent(root,
743                                            async_extent->compressed_size,
744                                            async_extent->compressed_size,
745                                            0, alloc_hint, &ins, 1, 1);
746                 if (ret) {
747                         free_async_extent_pages(async_extent);
748 
749                         if (ret == -ENOSPC) {
750                                 unlock_extent(io_tree, async_extent->start,
751                                               async_extent->start +
752                                               async_extent->ram_size - 1);
753 
754                                 /*
755                                  * we need to redirty the pages if we decide to
756                                  * fallback to uncompressed IO, otherwise we
757                                  * will not submit these pages down to lower
758                                  * layers.
759                                  */
760                                 extent_range_redirty_for_io(inode,
761                                                 async_extent->start,
762                                                 async_extent->start +
763                                                 async_extent->ram_size - 1);
764 
765                                 goto retry;
766                         }
767                         goto out_free;
768                 }
769                 /*
770                  * here we're doing allocation and writeback of the
771                  * compressed pages
772                  */
773                 btrfs_drop_extent_cache(inode, async_extent->start,
774                                         async_extent->start +
775                                         async_extent->ram_size - 1, 0);
776 
777                 em = alloc_extent_map();
778                 if (!em) {
779                         ret = -ENOMEM;
780                         goto out_free_reserve;
781                 }
782                 em->start = async_extent->start;
783                 em->len = async_extent->ram_size;
784                 em->orig_start = em->start;
785                 em->mod_start = em->start;
786                 em->mod_len = em->len;
787 
788                 em->block_start = ins.objectid;
789                 em->block_len = ins.offset;
790                 em->orig_block_len = ins.offset;
791                 em->ram_bytes = async_extent->ram_size;
792                 em->bdev = root->fs_info->fs_devices->latest_bdev;
793                 em->compress_type = async_extent->compress_type;
794                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
796                 em->generation = -1;
797 
798                 while (1) {
799                         write_lock(&em_tree->lock);
800                         ret = add_extent_mapping(em_tree, em, 1);
801                         write_unlock(&em_tree->lock);
802                         if (ret != -EEXIST) {
803                                 free_extent_map(em);
804                                 break;
805                         }
806                         btrfs_drop_extent_cache(inode, async_extent->start,
807                                                 async_extent->start +
808                                                 async_extent->ram_size - 1, 0);
809                 }
810 
811                 if (ret)
812                         goto out_free_reserve;
813 
814                 ret = btrfs_add_ordered_extent_compress(inode,
815                                                 async_extent->start,
816                                                 ins.objectid,
817                                                 async_extent->ram_size,
818                                                 ins.offset,
819                                                 BTRFS_ORDERED_COMPRESSED,
820                                                 async_extent->compress_type);
821                 if (ret) {
822                         btrfs_drop_extent_cache(inode, async_extent->start,
823                                                 async_extent->start +
824                                                 async_extent->ram_size - 1, 0);
825                         goto out_free_reserve;
826                 }
827 
828                 /*
829                  * clear dirty, set writeback and unlock the pages.
830                  */
831                 extent_clear_unlock_delalloc(inode, async_extent->start,
832                                 async_extent->start +
833                                 async_extent->ram_size - 1,
834                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
836                                 PAGE_SET_WRITEBACK);
837                 ret = btrfs_submit_compressed_write(inode,
838                                     async_extent->start,
839                                     async_extent->ram_size,
840                                     ins.objectid,
841                                     ins.offset, async_extent->pages,
842                                     async_extent->nr_pages);
843                 if (ret) {
844                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845                         struct page *p = async_extent->pages[0];
846                         const u64 start = async_extent->start;
847                         const u64 end = start + async_extent->ram_size - 1;
848 
849                         p->mapping = inode->i_mapping;
850                         tree->ops->writepage_end_io_hook(p, start, end,
851                                                          NULL, 0);
852                         p->mapping = NULL;
853                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
854                                                      PAGE_END_WRITEBACK |
855                                                      PAGE_SET_ERROR);
856                         free_async_extent_pages(async_extent);
857                 }
858                 alloc_hint = ins.objectid + ins.offset;
859                 kfree(async_extent);
860                 cond_resched();
861         }
862         return;
863 out_free_reserve:
864         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
865 out_free:
866         extent_clear_unlock_delalloc(inode, async_extent->start,
867                                      async_extent->start +
868                                      async_extent->ram_size - 1,
869                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
873                                      PAGE_SET_ERROR);
874         free_async_extent_pages(async_extent);
875         kfree(async_extent);
876         goto again;
877 }
878 
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
880                                       u64 num_bytes)
881 {
882         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883         struct extent_map *em;
884         u64 alloc_hint = 0;
885 
886         read_lock(&em_tree->lock);
887         em = search_extent_mapping(em_tree, start, num_bytes);
888         if (em) {
889                 /*
890                  * if block start isn't an actual block number then find the
891                  * first block in this inode and use that as a hint.  If that
892                  * block is also bogus then just don't worry about it.
893                  */
894                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
895                         free_extent_map(em);
896                         em = search_extent_mapping(em_tree, 0, 0);
897                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898                                 alloc_hint = em->block_start;
899                         if (em)
900                                 free_extent_map(em);
901                 } else {
902                         alloc_hint = em->block_start;
903                         free_extent_map(em);
904                 }
905         }
906         read_unlock(&em_tree->lock);
907 
908         return alloc_hint;
909 }
910 
911 /*
912  * when extent_io.c finds a delayed allocation range in the file,
913  * the call backs end up in this code.  The basic idea is to
914  * allocate extents on disk for the range, and create ordered data structs
915  * in ram to track those extents.
916  *
917  * locked_page is the page that writepage had locked already.  We use
918  * it to make sure we don't do extra locks or unlocks.
919  *
920  * *page_started is set to one if we unlock locked_page and do everything
921  * required to start IO on it.  It may be clean and already done with
922  * IO when we return.
923  */
924 static noinline int cow_file_range(struct inode *inode,
925                                    struct page *locked_page,
926                                    u64 start, u64 end, int *page_started,
927                                    unsigned long *nr_written,
928                                    int unlock)
929 {
930         struct btrfs_root *root = BTRFS_I(inode)->root;
931         u64 alloc_hint = 0;
932         u64 num_bytes;
933         unsigned long ram_size;
934         u64 disk_num_bytes;
935         u64 cur_alloc_size;
936         u64 blocksize = root->sectorsize;
937         struct btrfs_key ins;
938         struct extent_map *em;
939         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
940         int ret = 0;
941 
942         if (btrfs_is_free_space_inode(inode)) {
943                 WARN_ON_ONCE(1);
944                 ret = -EINVAL;
945                 goto out_unlock;
946         }
947 
948         num_bytes = ALIGN(end - start + 1, blocksize);
949         num_bytes = max(blocksize,  num_bytes);
950         disk_num_bytes = num_bytes;
951 
952         /* if this is a small write inside eof, kick off defrag */
953         if (num_bytes < SZ_64K &&
954             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955                 btrfs_add_inode_defrag(NULL, inode);
956 
957         if (start == 0) {
958                 /* lets try to make an inline extent */
959                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
960                                             NULL);
961                 if (ret == 0) {
962                         extent_clear_unlock_delalloc(inode, start, end, NULL,
963                                      EXTENT_LOCKED | EXTENT_DELALLOC |
964                                      EXTENT_DEFRAG, PAGE_UNLOCK |
965                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
966                                      PAGE_END_WRITEBACK);
967 
968                         *nr_written = *nr_written +
969                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
970                         *page_started = 1;
971                         goto out;
972                 } else if (ret < 0) {
973                         goto out_unlock;
974                 }
975         }
976 
977         BUG_ON(disk_num_bytes >
978                btrfs_super_total_bytes(root->fs_info->super_copy));
979 
980         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
982 
983         while (disk_num_bytes > 0) {
984                 unsigned long op;
985 
986                 cur_alloc_size = disk_num_bytes;
987                 ret = btrfs_reserve_extent(root, cur_alloc_size,
988                                            root->sectorsize, 0, alloc_hint,
989                                            &ins, 1, 1);
990                 if (ret < 0)
991                         goto out_unlock;
992 
993                 em = alloc_extent_map();
994                 if (!em) {
995                         ret = -ENOMEM;
996                         goto out_reserve;
997                 }
998                 em->start = start;
999                 em->orig_start = em->start;
1000                 ram_size = ins.offset;
1001                 em->len = ins.offset;
1002                 em->mod_start = em->start;
1003                 em->mod_len = em->len;
1004 
1005                 em->block_start = ins.objectid;
1006                 em->block_len = ins.offset;
1007                 em->orig_block_len = ins.offset;
1008                 em->ram_bytes = ram_size;
1009                 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011                 em->generation = -1;
1012 
1013                 while (1) {
1014                         write_lock(&em_tree->lock);
1015                         ret = add_extent_mapping(em_tree, em, 1);
1016                         write_unlock(&em_tree->lock);
1017                         if (ret != -EEXIST) {
1018                                 free_extent_map(em);
1019                                 break;
1020                         }
1021                         btrfs_drop_extent_cache(inode, start,
1022                                                 start + ram_size - 1, 0);
1023                 }
1024                 if (ret)
1025                         goto out_reserve;
1026 
1027                 cur_alloc_size = ins.offset;
1028                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029                                                ram_size, cur_alloc_size, 0);
1030                 if (ret)
1031                         goto out_drop_extent_cache;
1032 
1033                 if (root->root_key.objectid ==
1034                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035                         ret = btrfs_reloc_clone_csums(inode, start,
1036                                                       cur_alloc_size);
1037                         if (ret)
1038                                 goto out_drop_extent_cache;
1039                 }
1040 
1041                 if (disk_num_bytes < cur_alloc_size)
1042                         break;
1043 
1044                 /* we're not doing compressed IO, don't unlock the first
1045                  * page (which the caller expects to stay locked), don't
1046                  * clear any dirty bits and don't set any writeback bits
1047                  *
1048                  * Do set the Private2 bit so we know this page was properly
1049                  * setup for writepage
1050                  */
1051                 op = unlock ? PAGE_UNLOCK : 0;
1052                 op |= PAGE_SET_PRIVATE2;
1053 
1054                 extent_clear_unlock_delalloc(inode, start,
1055                                              start + ram_size - 1, locked_page,
1056                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1057                                              op);
1058                 disk_num_bytes -= cur_alloc_size;
1059                 num_bytes -= cur_alloc_size;
1060                 alloc_hint = ins.objectid + ins.offset;
1061                 start += cur_alloc_size;
1062         }
1063 out:
1064         return ret;
1065 
1066 out_drop_extent_cache:
1067         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1068 out_reserve:
1069         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1070 out_unlock:
1071         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1074                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1076         goto out;
1077 }
1078 
1079 /*
1080  * work queue call back to started compression on a file and pages
1081  */
1082 static noinline void async_cow_start(struct btrfs_work *work)
1083 {
1084         struct async_cow *async_cow;
1085         int num_added = 0;
1086         async_cow = container_of(work, struct async_cow, work);
1087 
1088         compress_file_range(async_cow->inode, async_cow->locked_page,
1089                             async_cow->start, async_cow->end, async_cow,
1090                             &num_added);
1091         if (num_added == 0) {
1092                 btrfs_add_delayed_iput(async_cow->inode);
1093                 async_cow->inode = NULL;
1094         }
1095 }
1096 
1097 /*
1098  * work queue call back to submit previously compressed pages
1099  */
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1101 {
1102         struct async_cow *async_cow;
1103         struct btrfs_root *root;
1104         unsigned long nr_pages;
1105 
1106         async_cow = container_of(work, struct async_cow, work);
1107 
1108         root = async_cow->root;
1109         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1110                 PAGE_CACHE_SHIFT;
1111 
1112         /*
1113          * atomic_sub_return implies a barrier for waitqueue_active
1114          */
1115         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1116             5 * SZ_1M &&
1117             waitqueue_active(&root->fs_info->async_submit_wait))
1118                 wake_up(&root->fs_info->async_submit_wait);
1119 
1120         if (async_cow->inode)
1121                 submit_compressed_extents(async_cow->inode, async_cow);
1122 }
1123 
1124 static noinline void async_cow_free(struct btrfs_work *work)
1125 {
1126         struct async_cow *async_cow;
1127         async_cow = container_of(work, struct async_cow, work);
1128         if (async_cow->inode)
1129                 btrfs_add_delayed_iput(async_cow->inode);
1130         kfree(async_cow);
1131 }
1132 
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134                                 u64 start, u64 end, int *page_started,
1135                                 unsigned long *nr_written)
1136 {
1137         struct async_cow *async_cow;
1138         struct btrfs_root *root = BTRFS_I(inode)->root;
1139         unsigned long nr_pages;
1140         u64 cur_end;
1141         int limit = 10 * SZ_1M;
1142 
1143         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144                          1, 0, NULL, GFP_NOFS);
1145         while (start < end) {
1146                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147                 BUG_ON(!async_cow); /* -ENOMEM */
1148                 async_cow->inode = igrab(inode);
1149                 async_cow->root = root;
1150                 async_cow->locked_page = locked_page;
1151                 async_cow->start = start;
1152 
1153                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154                     !btrfs_test_opt(root, FORCE_COMPRESS))
1155                         cur_end = end;
1156                 else
1157                         cur_end = min(end, start + SZ_512K - 1);
1158 
1159                 async_cow->end = cur_end;
1160                 INIT_LIST_HEAD(&async_cow->extents);
1161 
1162                 btrfs_init_work(&async_cow->work,
1163                                 btrfs_delalloc_helper,
1164                                 async_cow_start, async_cow_submit,
1165                                 async_cow_free);
1166 
1167                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1168                         PAGE_CACHE_SHIFT;
1169                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1170 
1171                 btrfs_queue_work(root->fs_info->delalloc_workers,
1172                                  &async_cow->work);
1173 
1174                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175                         wait_event(root->fs_info->async_submit_wait,
1176                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1177                             limit));
1178                 }
1179 
1180                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1182                         wait_event(root->fs_info->async_submit_wait,
1183                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1184                            0));
1185                 }
1186 
1187                 *nr_written += nr_pages;
1188                 start = cur_end + 1;
1189         }
1190         *page_started = 1;
1191         return 0;
1192 }
1193 
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195                                         u64 bytenr, u64 num_bytes)
1196 {
1197         int ret;
1198         struct btrfs_ordered_sum *sums;
1199         LIST_HEAD(list);
1200 
1201         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202                                        bytenr + num_bytes - 1, &list, 0);
1203         if (ret == 0 && list_empty(&list))
1204                 return 0;
1205 
1206         while (!list_empty(&list)) {
1207                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208                 list_del(&sums->list);
1209                 kfree(sums);
1210         }
1211         return 1;
1212 }
1213 
1214 /*
1215  * when nowcow writeback call back.  This checks for snapshots or COW copies
1216  * of the extents that exist in the file, and COWs the file as required.
1217  *
1218  * If no cow copies or snapshots exist, we write directly to the existing
1219  * blocks on disk
1220  */
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222                                        struct page *locked_page,
1223                               u64 start, u64 end, int *page_started, int force,
1224                               unsigned long *nr_written)
1225 {
1226         struct btrfs_root *root = BTRFS_I(inode)->root;
1227         struct btrfs_trans_handle *trans;
1228         struct extent_buffer *leaf;
1229         struct btrfs_path *path;
1230         struct btrfs_file_extent_item *fi;
1231         struct btrfs_key found_key;
1232         u64 cow_start;
1233         u64 cur_offset;
1234         u64 extent_end;
1235         u64 extent_offset;
1236         u64 disk_bytenr;
1237         u64 num_bytes;
1238         u64 disk_num_bytes;
1239         u64 ram_bytes;
1240         int extent_type;
1241         int ret, err;
1242         int type;
1243         int nocow;
1244         int check_prev = 1;
1245         bool nolock;
1246         u64 ino = btrfs_ino(inode);
1247 
1248         path = btrfs_alloc_path();
1249         if (!path) {
1250                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1252                                              EXTENT_DO_ACCOUNTING |
1253                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1254                                              PAGE_CLEAR_DIRTY |
1255                                              PAGE_SET_WRITEBACK |
1256                                              PAGE_END_WRITEBACK);
1257                 return -ENOMEM;
1258         }
1259 
1260         nolock = btrfs_is_free_space_inode(inode);
1261 
1262         if (nolock)
1263                 trans = btrfs_join_transaction_nolock(root);
1264         else
1265                 trans = btrfs_join_transaction(root);
1266 
1267         if (IS_ERR(trans)) {
1268                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1270                                              EXTENT_DO_ACCOUNTING |
1271                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1272                                              PAGE_CLEAR_DIRTY |
1273                                              PAGE_SET_WRITEBACK |
1274                                              PAGE_END_WRITEBACK);
1275                 btrfs_free_path(path);
1276                 return PTR_ERR(trans);
1277         }
1278 
1279         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1280 
1281         cow_start = (u64)-1;
1282         cur_offset = start;
1283         while (1) {
1284                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1285                                                cur_offset, 0);
1286                 if (ret < 0)
1287                         goto error;
1288                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289                         leaf = path->nodes[0];
1290                         btrfs_item_key_to_cpu(leaf, &found_key,
1291                                               path->slots[0] - 1);
1292                         if (found_key.objectid == ino &&
1293                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1294                                 path->slots[0]--;
1295                 }
1296                 check_prev = 0;
1297 next_slot:
1298                 leaf = path->nodes[0];
1299                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300                         ret = btrfs_next_leaf(root, path);
1301                         if (ret < 0)
1302                                 goto error;
1303                         if (ret > 0)
1304                                 break;
1305                         leaf = path->nodes[0];
1306                 }
1307 
1308                 nocow = 0;
1309                 disk_bytenr = 0;
1310                 num_bytes = 0;
1311                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1312 
1313                 if (found_key.objectid > ino)
1314                         break;
1315                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1317                         path->slots[0]++;
1318                         goto next_slot;
1319                 }
1320                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321                     found_key.offset > end)
1322                         break;
1323 
1324                 if (found_key.offset > cur_offset) {
1325                         extent_end = found_key.offset;
1326                         extent_type = 0;
1327                         goto out_check;
1328                 }
1329 
1330                 fi = btrfs_item_ptr(leaf, path->slots[0],
1331                                     struct btrfs_file_extent_item);
1332                 extent_type = btrfs_file_extent_type(leaf, fi);
1333 
1334                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1339                         extent_end = found_key.offset +
1340                                 btrfs_file_extent_num_bytes(leaf, fi);
1341                         disk_num_bytes =
1342                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343                         if (extent_end <= start) {
1344                                 path->slots[0]++;
1345                                 goto next_slot;
1346                         }
1347                         if (disk_bytenr == 0)
1348                                 goto out_check;
1349                         if (btrfs_file_extent_compression(leaf, fi) ||
1350                             btrfs_file_extent_encryption(leaf, fi) ||
1351                             btrfs_file_extent_other_encoding(leaf, fi))
1352                                 goto out_check;
1353                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1354                                 goto out_check;
1355                         if (btrfs_extent_readonly(root, disk_bytenr))
1356                                 goto out_check;
1357                         if (btrfs_cross_ref_exist(trans, root, ino,
1358                                                   found_key.offset -
1359                                                   extent_offset, disk_bytenr))
1360                                 goto out_check;
1361                         disk_bytenr += extent_offset;
1362                         disk_bytenr += cur_offset - found_key.offset;
1363                         num_bytes = min(end + 1, extent_end) - cur_offset;
1364                         /*
1365                          * if there are pending snapshots for this root,
1366                          * we fall into common COW way.
1367                          */
1368                         if (!nolock) {
1369                                 err = btrfs_start_write_no_snapshoting(root);
1370                                 if (!err)
1371                                         goto out_check;
1372                         }
1373                         /*
1374                          * force cow if csum exists in the range.
1375                          * this ensure that csum for a given extent are
1376                          * either valid or do not exist.
1377                          */
1378                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1379                                 goto out_check;
1380                         nocow = 1;
1381                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382                         extent_end = found_key.offset +
1383                                 btrfs_file_extent_inline_len(leaf,
1384                                                      path->slots[0], fi);
1385                         extent_end = ALIGN(extent_end, root->sectorsize);
1386                 } else {
1387                         BUG_ON(1);
1388                 }
1389 out_check:
1390                 if (extent_end <= start) {
1391                         path->slots[0]++;
1392                         if (!nolock && nocow)
1393                                 btrfs_end_write_no_snapshoting(root);
1394                         goto next_slot;
1395                 }
1396                 if (!nocow) {
1397                         if (cow_start == (u64)-1)
1398                                 cow_start = cur_offset;
1399                         cur_offset = extent_end;
1400                         if (cur_offset > end)
1401                                 break;
1402                         path->slots[0]++;
1403                         goto next_slot;
1404                 }
1405 
1406                 btrfs_release_path(path);
1407                 if (cow_start != (u64)-1) {
1408                         ret = cow_file_range(inode, locked_page,
1409                                              cow_start, found_key.offset - 1,
1410                                              page_started, nr_written, 1);
1411                         if (ret) {
1412                                 if (!nolock && nocow)
1413                                         btrfs_end_write_no_snapshoting(root);
1414                                 goto error;
1415                         }
1416                         cow_start = (u64)-1;
1417                 }
1418 
1419                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420                         struct extent_map *em;
1421                         struct extent_map_tree *em_tree;
1422                         em_tree = &BTRFS_I(inode)->extent_tree;
1423                         em = alloc_extent_map();
1424                         BUG_ON(!em); /* -ENOMEM */
1425                         em->start = cur_offset;
1426                         em->orig_start = found_key.offset - extent_offset;
1427                         em->len = num_bytes;
1428                         em->block_len = num_bytes;
1429                         em->block_start = disk_bytenr;
1430                         em->orig_block_len = disk_num_bytes;
1431                         em->ram_bytes = ram_bytes;
1432                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1433                         em->mod_start = em->start;
1434                         em->mod_len = em->len;
1435                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437                         em->generation = -1;
1438                         while (1) {
1439                                 write_lock(&em_tree->lock);
1440                                 ret = add_extent_mapping(em_tree, em, 1);
1441                                 write_unlock(&em_tree->lock);
1442                                 if (ret != -EEXIST) {
1443                                         free_extent_map(em);
1444                                         break;
1445                                 }
1446                                 btrfs_drop_extent_cache(inode, em->start,
1447                                                 em->start + em->len - 1, 0);
1448                         }
1449                         type = BTRFS_ORDERED_PREALLOC;
1450                 } else {
1451                         type = BTRFS_ORDERED_NOCOW;
1452                 }
1453 
1454                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455                                                num_bytes, num_bytes, type);
1456                 BUG_ON(ret); /* -ENOMEM */
1457 
1458                 if (root->root_key.objectid ==
1459                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1461                                                       num_bytes);
1462                         if (ret) {
1463                                 if (!nolock && nocow)
1464                                         btrfs_end_write_no_snapshoting(root);
1465                                 goto error;
1466                         }
1467                 }
1468 
1469                 extent_clear_unlock_delalloc(inode, cur_offset,
1470                                              cur_offset + num_bytes - 1,
1471                                              locked_page, EXTENT_LOCKED |
1472                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1473                                              PAGE_SET_PRIVATE2);
1474                 if (!nolock && nocow)
1475                         btrfs_end_write_no_snapshoting(root);
1476                 cur_offset = extent_end;
1477                 if (cur_offset > end)
1478                         break;
1479         }
1480         btrfs_release_path(path);
1481 
1482         if (cur_offset <= end && cow_start == (u64)-1) {
1483                 cow_start = cur_offset;
1484                 cur_offset = end;
1485         }
1486 
1487         if (cow_start != (u64)-1) {
1488                 ret = cow_file_range(inode, locked_page, cow_start, end,
1489                                      page_started, nr_written, 1);
1490                 if (ret)
1491                         goto error;
1492         }
1493 
1494 error:
1495         err = btrfs_end_transaction(trans, root);
1496         if (!ret)
1497                 ret = err;
1498 
1499         if (ret && cur_offset < end)
1500                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501                                              locked_page, EXTENT_LOCKED |
1502                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1503                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1504                                              PAGE_CLEAR_DIRTY |
1505                                              PAGE_SET_WRITEBACK |
1506                                              PAGE_END_WRITEBACK);
1507         btrfs_free_path(path);
1508         return ret;
1509 }
1510 
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1512 {
1513 
1514         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1516                 return 0;
1517 
1518         /*
1519          * @defrag_bytes is a hint value, no spinlock held here,
1520          * if is not zero, it means the file is defragging.
1521          * Force cow if given extent needs to be defragged.
1522          */
1523         if (BTRFS_I(inode)->defrag_bytes &&
1524             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525                            EXTENT_DEFRAG, 0, NULL))
1526                 return 1;
1527 
1528         return 0;
1529 }
1530 
1531 /*
1532  * extent_io.c call back to do delayed allocation processing
1533  */
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535                               u64 start, u64 end, int *page_started,
1536                               unsigned long *nr_written)
1537 {
1538         int ret;
1539         int force_cow = need_force_cow(inode, start, end);
1540 
1541         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543                                          page_started, 1, nr_written);
1544         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546                                          page_started, 0, nr_written);
1547         } else if (!inode_need_compress(inode)) {
1548                 ret = cow_file_range(inode, locked_page, start, end,
1549                                       page_started, nr_written, 1);
1550         } else {
1551                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552                         &BTRFS_I(inode)->runtime_flags);
1553                 ret = cow_file_range_async(inode, locked_page, start, end,
1554                                            page_started, nr_written);
1555         }
1556         return ret;
1557 }
1558 
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560                                     struct extent_state *orig, u64 split)
1561 {
1562         u64 size;
1563 
1564         /* not delalloc, ignore it */
1565         if (!(orig->state & EXTENT_DELALLOC))
1566                 return;
1567 
1568         size = orig->end - orig->start + 1;
1569         if (size > BTRFS_MAX_EXTENT_SIZE) {
1570                 u64 num_extents;
1571                 u64 new_size;
1572 
1573                 /*
1574                  * See the explanation in btrfs_merge_extent_hook, the same
1575                  * applies here, just in reverse.
1576                  */
1577                 new_size = orig->end - split + 1;
1578                 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579                                         BTRFS_MAX_EXTENT_SIZE);
1580                 new_size = split - orig->start;
1581                 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582                                         BTRFS_MAX_EXTENT_SIZE);
1583                 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584                               BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1585                         return;
1586         }
1587 
1588         spin_lock(&BTRFS_I(inode)->lock);
1589         BTRFS_I(inode)->outstanding_extents++;
1590         spin_unlock(&BTRFS_I(inode)->lock);
1591 }
1592 
1593 /*
1594  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595  * extents so we can keep track of new extents that are just merged onto old
1596  * extents, such as when we are doing sequential writes, so we can properly
1597  * account for the metadata space we'll need.
1598  */
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600                                     struct extent_state *new,
1601                                     struct extent_state *other)
1602 {
1603         u64 new_size, old_size;
1604         u64 num_extents;
1605 
1606         /* not delalloc, ignore it */
1607         if (!(other->state & EXTENT_DELALLOC))
1608                 return;
1609 
1610         if (new->start > other->start)
1611                 new_size = new->end - other->start + 1;
1612         else
1613                 new_size = other->end - new->start + 1;
1614 
1615         /* we're not bigger than the max, unreserve the space and go */
1616         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617                 spin_lock(&BTRFS_I(inode)->lock);
1618                 BTRFS_I(inode)->outstanding_extents--;
1619                 spin_unlock(&BTRFS_I(inode)->lock);
1620                 return;
1621         }
1622 
1623         /*
1624          * We have to add up either side to figure out how many extents were
1625          * accounted for before we merged into one big extent.  If the number of
1626          * extents we accounted for is <= the amount we need for the new range
1627          * then we can return, otherwise drop.  Think of it like this
1628          *
1629          * [ 4k][MAX_SIZE]
1630          *
1631          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632          * need 2 outstanding extents, on one side we have 1 and the other side
1633          * we have 1 so they are == and we can return.  But in this case
1634          *
1635          * [MAX_SIZE+4k][MAX_SIZE+4k]
1636          *
1637          * Each range on their own accounts for 2 extents, but merged together
1638          * they are only 3 extents worth of accounting, so we need to drop in
1639          * this case.
1640          */
1641         old_size = other->end - other->start + 1;
1642         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643                                 BTRFS_MAX_EXTENT_SIZE);
1644         old_size = new->end - new->start + 1;
1645         num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646                                  BTRFS_MAX_EXTENT_SIZE);
1647 
1648         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649                       BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1650                 return;
1651 
1652         spin_lock(&BTRFS_I(inode)->lock);
1653         BTRFS_I(inode)->outstanding_extents--;
1654         spin_unlock(&BTRFS_I(inode)->lock);
1655 }
1656 
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658                                       struct inode *inode)
1659 {
1660         spin_lock(&root->delalloc_lock);
1661         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663                               &root->delalloc_inodes);
1664                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665                         &BTRFS_I(inode)->runtime_flags);
1666                 root->nr_delalloc_inodes++;
1667                 if (root->nr_delalloc_inodes == 1) {
1668                         spin_lock(&root->fs_info->delalloc_root_lock);
1669                         BUG_ON(!list_empty(&root->delalloc_root));
1670                         list_add_tail(&root->delalloc_root,
1671                                       &root->fs_info->delalloc_roots);
1672                         spin_unlock(&root->fs_info->delalloc_root_lock);
1673                 }
1674         }
1675         spin_unlock(&root->delalloc_lock);
1676 }
1677 
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679                                      struct inode *inode)
1680 {
1681         spin_lock(&root->delalloc_lock);
1682         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685                           &BTRFS_I(inode)->runtime_flags);
1686                 root->nr_delalloc_inodes--;
1687                 if (!root->nr_delalloc_inodes) {
1688                         spin_lock(&root->fs_info->delalloc_root_lock);
1689                         BUG_ON(list_empty(&root->delalloc_root));
1690                         list_del_init(&root->delalloc_root);
1691                         spin_unlock(&root->fs_info->delalloc_root_lock);
1692                 }
1693         }
1694         spin_unlock(&root->delalloc_lock);
1695 }
1696 
1697 /*
1698  * extent_io.c set_bit_hook, used to track delayed allocation
1699  * bytes in this file, and to maintain the list of inodes that
1700  * have pending delalloc work to be done.
1701  */
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703                                struct extent_state *state, unsigned *bits)
1704 {
1705 
1706         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1707                 WARN_ON(1);
1708         /*
1709          * set_bit and clear bit hooks normally require _irqsave/restore
1710          * but in this case, we are only testing for the DELALLOC
1711          * bit, which is only set or cleared with irqs on
1712          */
1713         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714                 struct btrfs_root *root = BTRFS_I(inode)->root;
1715                 u64 len = state->end + 1 - state->start;
1716                 bool do_list = !btrfs_is_free_space_inode(inode);
1717 
1718                 if (*bits & EXTENT_FIRST_DELALLOC) {
1719                         *bits &= ~EXTENT_FIRST_DELALLOC;
1720                 } else {
1721                         spin_lock(&BTRFS_I(inode)->lock);
1722                         BTRFS_I(inode)->outstanding_extents++;
1723                         spin_unlock(&BTRFS_I(inode)->lock);
1724                 }
1725 
1726                 /* For sanity tests */
1727                 if (btrfs_test_is_dummy_root(root))
1728                         return;
1729 
1730                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731                                      root->fs_info->delalloc_batch);
1732                 spin_lock(&BTRFS_I(inode)->lock);
1733                 BTRFS_I(inode)->delalloc_bytes += len;
1734                 if (*bits & EXTENT_DEFRAG)
1735                         BTRFS_I(inode)->defrag_bytes += len;
1736                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737                                          &BTRFS_I(inode)->runtime_flags))
1738                         btrfs_add_delalloc_inodes(root, inode);
1739                 spin_unlock(&BTRFS_I(inode)->lock);
1740         }
1741 }
1742 
1743 /*
1744  * extent_io.c clear_bit_hook, see set_bit_hook for why
1745  */
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747                                  struct extent_state *state,
1748                                  unsigned *bits)
1749 {
1750         u64 len = state->end + 1 - state->start;
1751         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752                                     BTRFS_MAX_EXTENT_SIZE);
1753 
1754         spin_lock(&BTRFS_I(inode)->lock);
1755         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756                 BTRFS_I(inode)->defrag_bytes -= len;
1757         spin_unlock(&BTRFS_I(inode)->lock);
1758 
1759         /*
1760          * set_bit and clear bit hooks normally require _irqsave/restore
1761          * but in this case, we are only testing for the DELALLOC
1762          * bit, which is only set or cleared with irqs on
1763          */
1764         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765                 struct btrfs_root *root = BTRFS_I(inode)->root;
1766                 bool do_list = !btrfs_is_free_space_inode(inode);
1767 
1768                 if (*bits & EXTENT_FIRST_DELALLOC) {
1769                         *bits &= ~EXTENT_FIRST_DELALLOC;
1770                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771                         spin_lock(&BTRFS_I(inode)->lock);
1772                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1773                         spin_unlock(&BTRFS_I(inode)->lock);
1774                 }
1775 
1776                 /*
1777                  * We don't reserve metadata space for space cache inodes so we
1778                  * don't need to call dellalloc_release_metadata if there is an
1779                  * error.
1780                  */
1781                 if (*bits & EXTENT_DO_ACCOUNTING &&
1782                     root != root->fs_info->tree_root)
1783                         btrfs_delalloc_release_metadata(inode, len);
1784 
1785                 /* For sanity tests. */
1786                 if (btrfs_test_is_dummy_root(root))
1787                         return;
1788 
1789                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790                     && do_list && !(state->state & EXTENT_NORESERVE))
1791                         btrfs_free_reserved_data_space_noquota(inode,
1792                                         state->start, len);
1793 
1794                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795                                      root->fs_info->delalloc_batch);
1796                 spin_lock(&BTRFS_I(inode)->lock);
1797                 BTRFS_I(inode)->delalloc_bytes -= len;
1798                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800                              &BTRFS_I(inode)->runtime_flags))
1801                         btrfs_del_delalloc_inode(root, inode);
1802                 spin_unlock(&BTRFS_I(inode)->lock);
1803         }
1804 }
1805 
1806 /*
1807  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808  * we don't create bios that span stripes or chunks
1809  */
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811                          size_t size, struct bio *bio,
1812                          unsigned long bio_flags)
1813 {
1814         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1816         u64 length = 0;
1817         u64 map_length;
1818         int ret;
1819 
1820         if (bio_flags & EXTENT_BIO_COMPRESSED)
1821                 return 0;
1822 
1823         length = bio->bi_iter.bi_size;
1824         map_length = length;
1825         ret = btrfs_map_block(root->fs_info, rw, logical,
1826                               &map_length, NULL, 0);
1827         /* Will always return 0 with map_multi == NULL */
1828         BUG_ON(ret < 0);
1829         if (map_length < length + size)
1830                 return 1;
1831         return 0;
1832 }
1833 
1834 /*
1835  * in order to insert checksums into the metadata in large chunks,
1836  * we wait until bio submission time.   All the pages in the bio are
1837  * checksummed and sums are attached onto the ordered extent record.
1838  *
1839  * At IO completion time the cums attached on the ordered extent record
1840  * are inserted into the btree
1841  */
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843                                     struct bio *bio, int mirror_num,
1844                                     unsigned long bio_flags,
1845                                     u64 bio_offset)
1846 {
1847         struct btrfs_root *root = BTRFS_I(inode)->root;
1848         int ret = 0;
1849 
1850         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851         BUG_ON(ret); /* -ENOMEM */
1852         return 0;
1853 }
1854 
1855 /*
1856  * in order to insert checksums into the metadata in large chunks,
1857  * we wait until bio submission time.   All the pages in the bio are
1858  * checksummed and sums are attached onto the ordered extent record.
1859  *
1860  * At IO completion time the cums attached on the ordered extent record
1861  * are inserted into the btree
1862  */
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864                           int mirror_num, unsigned long bio_flags,
1865                           u64 bio_offset)
1866 {
1867         struct btrfs_root *root = BTRFS_I(inode)->root;
1868         int ret;
1869 
1870         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1871         if (ret) {
1872                 bio->bi_error = ret;
1873                 bio_endio(bio);
1874         }
1875         return ret;
1876 }
1877 
1878 /*
1879  * extent_io.c submission hook. This does the right thing for csum calculation
1880  * on write, or reading the csums from the tree before a read
1881  */
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883                           int mirror_num, unsigned long bio_flags,
1884                           u64 bio_offset)
1885 {
1886         struct btrfs_root *root = BTRFS_I(inode)->root;
1887         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1888         int ret = 0;
1889         int skip_sum;
1890         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1891 
1892         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1893 
1894         if (btrfs_is_free_space_inode(inode))
1895                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1896 
1897         if (!(rw & REQ_WRITE)) {
1898                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1899                 if (ret)
1900                         goto out;
1901 
1902                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903                         ret = btrfs_submit_compressed_read(inode, bio,
1904                                                            mirror_num,
1905                                                            bio_flags);
1906                         goto out;
1907                 } else if (!skip_sum) {
1908                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1909                         if (ret)
1910                                 goto out;
1911                 }
1912                 goto mapit;
1913         } else if (async && !skip_sum) {
1914                 /* csum items have already been cloned */
1915                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1916                         goto mapit;
1917                 /* we're doing a write, do the async checksumming */
1918                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919                                    inode, rw, bio, mirror_num,
1920                                    bio_flags, bio_offset,
1921                                    __btrfs_submit_bio_start,
1922                                    __btrfs_submit_bio_done);
1923                 goto out;
1924         } else if (!skip_sum) {
1925                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1926                 if (ret)
1927                         goto out;
1928         }
1929 
1930 mapit:
1931         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1932 
1933 out:
1934         if (ret < 0) {
1935                 bio->bi_error = ret;
1936                 bio_endio(bio);
1937         }
1938         return ret;
1939 }
1940 
1941 /*
1942  * given a list of ordered sums record them in the inode.  This happens
1943  * at IO completion time based on sums calculated at bio submission time.
1944  */
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946                              struct inode *inode, u64 file_offset,
1947                              struct list_head *list)
1948 {
1949         struct btrfs_ordered_sum *sum;
1950 
1951         list_for_each_entry(sum, list, list) {
1952                 trans->adding_csums = 1;
1953                 btrfs_csum_file_blocks(trans,
1954                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955                 trans->adding_csums = 0;
1956         }
1957         return 0;
1958 }
1959 
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961                               struct extent_state **cached_state)
1962 {
1963         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1964         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965                                    cached_state, GFP_NOFS);
1966 }
1967 
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1970         struct page *page;
1971         struct btrfs_work work;
1972 };
1973 
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1975 {
1976         struct btrfs_writepage_fixup *fixup;
1977         struct btrfs_ordered_extent *ordered;
1978         struct extent_state *cached_state = NULL;
1979         struct page *page;
1980         struct inode *inode;
1981         u64 page_start;
1982         u64 page_end;
1983         int ret;
1984 
1985         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1986         page = fixup->page;
1987 again:
1988         lock_page(page);
1989         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990                 ClearPageChecked(page);
1991                 goto out_page;
1992         }
1993 
1994         inode = page->mapping->host;
1995         page_start = page_offset(page);
1996         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1997 
1998         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1999                          &cached_state);
2000 
2001         /* already ordered? We're done */
2002         if (PagePrivate2(page))
2003                 goto out;
2004 
2005         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2006         if (ordered) {
2007                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2008                                      page_end, &cached_state, GFP_NOFS);
2009                 unlock_page(page);
2010                 btrfs_start_ordered_extent(inode, ordered, 1);
2011                 btrfs_put_ordered_extent(ordered);
2012                 goto again;
2013         }
2014 
2015         ret = btrfs_delalloc_reserve_space(inode, page_start,
2016                                            PAGE_CACHE_SIZE);
2017         if (ret) {
2018                 mapping_set_error(page->mapping, ret);
2019                 end_extent_writepage(page, ret, page_start, page_end);
2020                 ClearPageChecked(page);
2021                 goto out;
2022          }
2023 
2024         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2025         ClearPageChecked(page);
2026         set_page_dirty(page);
2027 out:
2028         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2029                              &cached_state, GFP_NOFS);
2030 out_page:
2031         unlock_page(page);
2032         page_cache_release(page);
2033         kfree(fixup);
2034 }
2035 
2036 /*
2037  * There are a few paths in the higher layers of the kernel that directly
2038  * set the page dirty bit without asking the filesystem if it is a
2039  * good idea.  This causes problems because we want to make sure COW
2040  * properly happens and the data=ordered rules are followed.
2041  *
2042  * In our case any range that doesn't have the ORDERED bit set
2043  * hasn't been properly setup for IO.  We kick off an async process
2044  * to fix it up.  The async helper will wait for ordered extents, set
2045  * the delalloc bit and make it safe to write the page.
2046  */
2047 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2048 {
2049         struct inode *inode = page->mapping->host;
2050         struct btrfs_writepage_fixup *fixup;
2051         struct btrfs_root *root = BTRFS_I(inode)->root;
2052 
2053         /* this page is properly in the ordered list */
2054         if (TestClearPagePrivate2(page))
2055                 return 0;
2056 
2057         if (PageChecked(page))
2058                 return -EAGAIN;
2059 
2060         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2061         if (!fixup)
2062                 return -EAGAIN;
2063 
2064         SetPageChecked(page);
2065         page_cache_get(page);
2066         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2067                         btrfs_writepage_fixup_worker, NULL, NULL);
2068         fixup->page = page;
2069         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2070         return -EBUSY;
2071 }
2072 
2073 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2074                                        struct inode *inode, u64 file_pos,
2075                                        u64 disk_bytenr, u64 disk_num_bytes,
2076                                        u64 num_bytes, u64 ram_bytes,
2077                                        u8 compression, u8 encryption,
2078                                        u16 other_encoding, int extent_type)
2079 {
2080         struct btrfs_root *root = BTRFS_I(inode)->root;
2081         struct btrfs_file_extent_item *fi;
2082         struct btrfs_path *path;
2083         struct extent_buffer *leaf;
2084         struct btrfs_key ins;
2085         int extent_inserted = 0;
2086         int ret;
2087 
2088         path = btrfs_alloc_path();
2089         if (!path)
2090                 return -ENOMEM;
2091 
2092         /*
2093          * we may be replacing one extent in the tree with another.
2094          * The new extent is pinned in the extent map, and we don't want
2095          * to drop it from the cache until it is completely in the btree.
2096          *
2097          * So, tell btrfs_drop_extents to leave this extent in the cache.
2098          * the caller is expected to unpin it and allow it to be merged
2099          * with the others.
2100          */
2101         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2102                                    file_pos + num_bytes, NULL, 0,
2103                                    1, sizeof(*fi), &extent_inserted);
2104         if (ret)
2105                 goto out;
2106 
2107         if (!extent_inserted) {
2108                 ins.objectid = btrfs_ino(inode);
2109                 ins.offset = file_pos;
2110                 ins.type = BTRFS_EXTENT_DATA_KEY;
2111 
2112                 path->leave_spinning = 1;
2113                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2114                                               sizeof(*fi));
2115                 if (ret)
2116                         goto out;
2117         }
2118         leaf = path->nodes[0];
2119         fi = btrfs_item_ptr(leaf, path->slots[0],
2120                             struct btrfs_file_extent_item);
2121         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2122         btrfs_set_file_extent_type(leaf, fi, extent_type);
2123         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2124         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2125         btrfs_set_file_extent_offset(leaf, fi, 0);
2126         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2127         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2128         btrfs_set_file_extent_compression(leaf, fi, compression);
2129         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2130         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2131 
2132         btrfs_mark_buffer_dirty(leaf);
2133         btrfs_release_path(path);
2134 
2135         inode_add_bytes(inode, num_bytes);
2136 
2137         ins.objectid = disk_bytenr;
2138         ins.offset = disk_num_bytes;
2139         ins.type = BTRFS_EXTENT_ITEM_KEY;
2140         ret = btrfs_alloc_reserved_file_extent(trans, root,
2141                                         root->root_key.objectid,
2142                                         btrfs_ino(inode), file_pos,
2143                                         ram_bytes, &ins);
2144         /*
2145          * Release the reserved range from inode dirty range map, as it is
2146          * already moved into delayed_ref_head
2147          */
2148         btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2149 out:
2150         btrfs_free_path(path);
2151 
2152         return ret;
2153 }
2154 
2155 /* snapshot-aware defrag */
2156 struct sa_defrag_extent_backref {
2157         struct rb_node node;
2158         struct old_sa_defrag_extent *old;
2159         u64 root_id;
2160         u64 inum;
2161         u64 file_pos;
2162         u64 extent_offset;
2163         u64 num_bytes;
2164         u64 generation;
2165 };
2166 
2167 struct old_sa_defrag_extent {
2168         struct list_head list;
2169         struct new_sa_defrag_extent *new;
2170 
2171         u64 extent_offset;
2172         u64 bytenr;
2173         u64 offset;
2174         u64 len;
2175         int count;
2176 };
2177 
2178 struct new_sa_defrag_extent {
2179         struct rb_root root;
2180         struct list_head head;
2181         struct btrfs_path *path;
2182         struct inode *inode;
2183         u64 file_pos;
2184         u64 len;
2185         u64 bytenr;
2186         u64 disk_len;
2187         u8 compress_type;
2188 };
2189 
2190 static int backref_comp(struct sa_defrag_extent_backref *b1,
2191                         struct sa_defrag_extent_backref *b2)
2192 {
2193         if (b1->root_id < b2->root_id)
2194                 return -1;
2195         else if (b1->root_id > b2->root_id)
2196                 return 1;
2197 
2198         if (b1->inum < b2->inum)
2199                 return -1;
2200         else if (b1->inum > b2->inum)
2201                 return 1;
2202 
2203         if (b1->file_pos < b2->file_pos)
2204                 return -1;
2205         else if (b1->file_pos > b2->file_pos)
2206                 return 1;
2207 
2208         /*
2209          * [------------------------------] ===> (a range of space)
2210          *     |<--->|   |<---->| =============> (fs/file tree A)
2211          * |<---------------------------->| ===> (fs/file tree B)
2212          *
2213          * A range of space can refer to two file extents in one tree while
2214          * refer to only one file extent in another tree.
2215          *
2216          * So we may process a disk offset more than one time(two extents in A)
2217          * and locate at the same extent(one extent in B), then insert two same
2218          * backrefs(both refer to the extent in B).
2219          */
2220         return 0;
2221 }
2222 
2223 static void backref_insert(struct rb_root *root,
2224                            struct sa_defrag_extent_backref *backref)
2225 {
2226         struct rb_node **p = &root->rb_node;
2227         struct rb_node *parent = NULL;
2228         struct sa_defrag_extent_backref *entry;
2229         int ret;
2230 
2231         while (*p) {
2232                 parent = *p;
2233                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2234 
2235                 ret = backref_comp(backref, entry);
2236                 if (ret < 0)
2237                         p = &(*p)->rb_left;
2238                 else
2239                         p = &(*p)->rb_right;
2240         }
2241 
2242         rb_link_node(&backref->node, parent, p);
2243         rb_insert_color(&backref->node, root);
2244 }
2245 
2246 /*
2247  * Note the backref might has changed, and in this case we just return 0.
2248  */
2249 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2250                                        void *ctx)
2251 {
2252         struct btrfs_file_extent_item *extent;
2253         struct btrfs_fs_info *fs_info;
2254         struct old_sa_defrag_extent *old = ctx;
2255         struct new_sa_defrag_extent *new = old->new;
2256         struct btrfs_path *path = new->path;
2257         struct btrfs_key key;
2258         struct btrfs_root *root;
2259         struct sa_defrag_extent_backref *backref;
2260         struct extent_buffer *leaf;
2261         struct inode *inode = new->inode;
2262         int slot;
2263         int ret;
2264         u64 extent_offset;
2265         u64 num_bytes;
2266 
2267         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2268             inum == btrfs_ino(inode))
2269                 return 0;
2270 
2271         key.objectid = root_id;
2272         key.type = BTRFS_ROOT_ITEM_KEY;
2273         key.offset = (u64)-1;
2274 
2275         fs_info = BTRFS_I(inode)->root->fs_info;
2276         root = btrfs_read_fs_root_no_name(fs_info, &key);
2277         if (IS_ERR(root)) {
2278                 if (PTR_ERR(root) == -ENOENT)
2279                         return 0;
2280                 WARN_ON(1);
2281                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282                          inum, offset, root_id);
2283                 return PTR_ERR(root);
2284         }
2285 
2286         key.objectid = inum;
2287         key.type = BTRFS_EXTENT_DATA_KEY;
2288         if (offset > (u64)-1 << 32)
2289                 key.offset = 0;
2290         else
2291                 key.offset = offset;
2292 
2293         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2294         if (WARN_ON(ret < 0))
2295                 return ret;
2296         ret = 0;
2297 
2298         while (1) {
2299                 cond_resched();
2300 
2301                 leaf = path->nodes[0];
2302                 slot = path->slots[0];
2303 
2304                 if (slot >= btrfs_header_nritems(leaf)) {
2305                         ret = btrfs_next_leaf(root, path);
2306                         if (ret < 0) {
2307                                 goto out;
2308                         } else if (ret > 0) {
2309                                 ret = 0;
2310                                 goto out;
2311                         }
2312                         continue;
2313                 }
2314 
2315                 path->slots[0]++;
2316 
2317                 btrfs_item_key_to_cpu(leaf, &key, slot);
2318 
2319                 if (key.objectid > inum)
2320                         goto out;
2321 
2322                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2323                         continue;
2324 
2325                 extent = btrfs_item_ptr(leaf, slot,
2326                                         struct btrfs_file_extent_item);
2327 
2328                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2329                         continue;
2330 
2331                 /*
2332                  * 'offset' refers to the exact key.offset,
2333                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334                  * (key.offset - extent_offset).
2335                  */
2336                 if (key.offset != offset)
2337                         continue;
2338 
2339                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2340                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2341 
2342                 if (extent_offset >= old->extent_offset + old->offset +
2343                     old->len || extent_offset + num_bytes <=
2344                     old->extent_offset + old->offset)
2345                         continue;
2346                 break;
2347         }
2348 
2349         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2350         if (!backref) {
2351                 ret = -ENOENT;
2352                 goto out;
2353         }
2354 
2355         backref->root_id = root_id;
2356         backref->inum = inum;
2357         backref->file_pos = offset;
2358         backref->num_bytes = num_bytes;
2359         backref->extent_offset = extent_offset;
2360         backref->generation = btrfs_file_extent_generation(leaf, extent);
2361         backref->old = old;
2362         backref_insert(&new->root, backref);
2363         old->count++;
2364 out:
2365         btrfs_release_path(path);
2366         WARN_ON(ret);
2367         return ret;
2368 }
2369 
2370 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2371                                    struct new_sa_defrag_extent *new)
2372 {
2373         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2374         struct old_sa_defrag_extent *old, *tmp;
2375         int ret;
2376 
2377         new->path = path;
2378 
2379         list_for_each_entry_safe(old, tmp, &new->head, list) {
2380                 ret = iterate_inodes_from_logical(old->bytenr +
2381                                                   old->extent_offset, fs_info,
2382                                                   path, record_one_backref,
2383                                                   old);
2384                 if (ret < 0 && ret != -ENOENT)
2385                         return false;
2386 
2387                 /* no backref to be processed for this extent */
2388                 if (!old->count) {
2389                         list_del(&old->list);
2390                         kfree(old);
2391                 }
2392         }
2393 
2394         if (list_empty(&new->head))
2395                 return false;
2396 
2397         return true;
2398 }
2399 
2400 static int relink_is_mergable(struct extent_buffer *leaf,
2401                               struct btrfs_file_extent_item *fi,
2402                               struct new_sa_defrag_extent *new)
2403 {
2404         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2405                 return 0;
2406 
2407         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2408                 return 0;
2409 
2410         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2411                 return 0;
2412 
2413         if (btrfs_file_extent_encryption(leaf, fi) ||
2414             btrfs_file_extent_other_encoding(leaf, fi))
2415                 return 0;
2416 
2417         return 1;
2418 }
2419 
2420 /*
2421  * Note the backref might has changed, and in this case we just return 0.
2422  */
2423 static noinline int relink_extent_backref(struct btrfs_path *path,
2424                                  struct sa_defrag_extent_backref *prev,
2425                                  struct sa_defrag_extent_backref *backref)
2426 {
2427         struct btrfs_file_extent_item *extent;
2428         struct btrfs_file_extent_item *item;
2429         struct btrfs_ordered_extent *ordered;
2430         struct btrfs_trans_handle *trans;
2431         struct btrfs_fs_info *fs_info;
2432         struct btrfs_root *root;
2433         struct btrfs_key key;
2434         struct extent_buffer *leaf;
2435         struct old_sa_defrag_extent *old = backref->old;
2436         struct new_sa_defrag_extent *new = old->new;
2437         struct inode *src_inode = new->inode;
2438         struct inode *inode;
2439         struct extent_state *cached = NULL;
2440         int ret = 0;
2441         u64 start;
2442         u64 len;
2443         u64 lock_start;
2444         u64 lock_end;
2445         bool merge = false;
2446         int index;
2447 
2448         if (prev && prev->root_id == backref->root_id &&
2449             prev->inum == backref->inum &&
2450             prev->file_pos + prev->num_bytes == backref->file_pos)
2451                 merge = true;
2452 
2453         /* step 1: get root */
2454         key.objectid = backref->root_id;
2455         key.type = BTRFS_ROOT_ITEM_KEY;
2456         key.offset = (u64)-1;
2457 
2458         fs_info = BTRFS_I(src_inode)->root->fs_info;
2459         index = srcu_read_lock(&fs_info->subvol_srcu);
2460 
2461         root = btrfs_read_fs_root_no_name(fs_info, &key);
2462         if (IS_ERR(root)) {
2463                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2464                 if (PTR_ERR(root) == -ENOENT)
2465                         return 0;
2466                 return PTR_ERR(root);
2467         }
2468 
2469         if (btrfs_root_readonly(root)) {
2470                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2471                 return 0;
2472         }
2473 
2474         /* step 2: get inode */
2475         key.objectid = backref->inum;
2476         key.type = BTRFS_INODE_ITEM_KEY;
2477         key.offset = 0;
2478 
2479         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2480         if (IS_ERR(inode)) {
2481                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2482                 return 0;
2483         }
2484 
2485         srcu_read_unlock(&fs_info->subvol_srcu, index);
2486 
2487         /* step 3: relink backref */
2488         lock_start = backref->file_pos;
2489         lock_end = backref->file_pos + backref->num_bytes - 1;
2490         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2491                          &cached);
2492 
2493         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2494         if (ordered) {
2495                 btrfs_put_ordered_extent(ordered);
2496                 goto out_unlock;
2497         }
2498 
2499         trans = btrfs_join_transaction(root);
2500         if (IS_ERR(trans)) {
2501                 ret = PTR_ERR(trans);
2502                 goto out_unlock;
2503         }
2504 
2505         key.objectid = backref->inum;
2506         key.type = BTRFS_EXTENT_DATA_KEY;
2507         key.offset = backref->file_pos;
2508 
2509         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510         if (ret < 0) {
2511                 goto out_free_path;
2512         } else if (ret > 0) {
2513                 ret = 0;
2514                 goto out_free_path;
2515         }
2516 
2517         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2518                                 struct btrfs_file_extent_item);
2519 
2520         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2521             backref->generation)
2522                 goto out_free_path;
2523 
2524         btrfs_release_path(path);
2525 
2526         start = backref->file_pos;
2527         if (backref->extent_offset < old->extent_offset + old->offset)
2528                 start += old->extent_offset + old->offset -
2529                          backref->extent_offset;
2530 
2531         len = min(backref->extent_offset + backref->num_bytes,
2532                   old->extent_offset + old->offset + old->len);
2533         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2534 
2535         ret = btrfs_drop_extents(trans, root, inode, start,
2536                                  start + len, 1);
2537         if (ret)
2538                 goto out_free_path;
2539 again:
2540         key.objectid = btrfs_ino(inode);
2541         key.type = BTRFS_EXTENT_DATA_KEY;
2542         key.offset = start;
2543 
2544         path->leave_spinning = 1;
2545         if (merge) {
2546                 struct btrfs_file_extent_item *fi;
2547                 u64 extent_len;
2548                 struct btrfs_key found_key;
2549 
2550                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2551                 if (ret < 0)
2552                         goto out_free_path;
2553 
2554                 path->slots[0]--;
2555                 leaf = path->nodes[0];
2556                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2557 
2558                 fi = btrfs_item_ptr(leaf, path->slots[0],
2559                                     struct btrfs_file_extent_item);
2560                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2561 
2562                 if (extent_len + found_key.offset == start &&
2563                     relink_is_mergable(leaf, fi, new)) {
2564                         btrfs_set_file_extent_num_bytes(leaf, fi,
2565                                                         extent_len + len);
2566                         btrfs_mark_buffer_dirty(leaf);
2567                         inode_add_bytes(inode, len);
2568 
2569                         ret = 1;
2570                         goto out_free_path;
2571                 } else {
2572                         merge = false;
2573                         btrfs_release_path(path);
2574                         goto again;
2575                 }
2576         }
2577 
2578         ret = btrfs_insert_empty_item(trans, root, path, &key,
2579                                         sizeof(*extent));
2580         if (ret) {
2581                 btrfs_abort_transaction(trans, root, ret);
2582                 goto out_free_path;
2583         }
2584 
2585         leaf = path->nodes[0];
2586         item = btrfs_item_ptr(leaf, path->slots[0],
2587                                 struct btrfs_file_extent_item);
2588         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2589         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2590         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2591         btrfs_set_file_extent_num_bytes(leaf, item, len);
2592         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2593         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2594         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2595         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2596         btrfs_set_file_extent_encryption(leaf, item, 0);
2597         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2598 
2599         btrfs_mark_buffer_dirty(leaf);
2600         inode_add_bytes(inode, len);
2601         btrfs_release_path(path);
2602 
2603         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2604                         new->disk_len, 0,
2605                         backref->root_id, backref->inum,
2606                         new->file_pos); /* start - extent_offset */
2607         if (ret) {
2608                 btrfs_abort_transaction(trans, root, ret);
2609                 goto out_free_path;
2610         }
2611 
2612         ret = 1;
2613 out_free_path:
2614         btrfs_release_path(path);
2615         path->leave_spinning = 0;
2616         btrfs_end_transaction(trans, root);
2617 out_unlock:
2618         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2619                              &cached, GFP_NOFS);
2620         iput(inode);
2621         return ret;
2622 }
2623 
2624 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2625 {
2626         struct old_sa_defrag_extent *old, *tmp;
2627 
2628         if (!new)
2629                 return;
2630 
2631         list_for_each_entry_safe(old, tmp, &new->head, list) {
2632                 kfree(old);
2633         }
2634         kfree(new);
2635 }
2636 
2637 static void relink_file_extents(struct new_sa_defrag_extent *new)
2638 {
2639         struct btrfs_path *path;
2640         struct sa_defrag_extent_backref *backref;
2641         struct sa_defrag_extent_backref *prev = NULL;
2642         struct inode *inode;
2643         struct btrfs_root *root;
2644         struct rb_node *node;
2645         int ret;
2646 
2647         inode = new->inode;
2648         root = BTRFS_I(inode)->root;
2649 
2650         path = btrfs_alloc_path();
2651         if (!path)
2652                 return;
2653 
2654         if (!record_extent_backrefs(path, new)) {
2655                 btrfs_free_path(path);
2656                 goto out;
2657         }
2658         btrfs_release_path(path);
2659 
2660         while (1) {
2661                 node = rb_first(&new->root);
2662                 if (!node)
2663                         break;
2664                 rb_erase(node, &new->root);
2665 
2666                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2667 
2668                 ret = relink_extent_backref(path, prev, backref);
2669                 WARN_ON(ret < 0);
2670 
2671                 kfree(prev);
2672 
2673                 if (ret == 1)
2674                         prev = backref;
2675                 else
2676                         prev = NULL;
2677                 cond_resched();
2678         }
2679         kfree(prev);
2680 
2681         btrfs_free_path(path);
2682 out:
2683         free_sa_defrag_extent(new);
2684 
2685         atomic_dec(&root->fs_info->defrag_running);
2686         wake_up(&root->fs_info->transaction_wait);
2687 }
2688 
2689 static struct new_sa_defrag_extent *
2690 record_old_file_extents(struct inode *inode,
2691                         struct btrfs_ordered_extent *ordered)
2692 {
2693         struct btrfs_root *root = BTRFS_I(inode)->root;
2694         struct btrfs_path *path;
2695         struct btrfs_key key;
2696         struct old_sa_defrag_extent *old;
2697         struct new_sa_defrag_extent *new;
2698         int ret;
2699 
2700         new = kmalloc(sizeof(*new), GFP_NOFS);
2701         if (!new)
2702                 return NULL;
2703 
2704         new->inode = inode;
2705         new->file_pos = ordered->file_offset;
2706         new->len = ordered->len;
2707         new->bytenr = ordered->start;
2708         new->disk_len = ordered->disk_len;
2709         new->compress_type = ordered->compress_type;
2710         new->root = RB_ROOT;
2711         INIT_LIST_HEAD(&new->head);
2712 
2713         path = btrfs_alloc_path();
2714         if (!path)
2715                 goto out_kfree;
2716 
2717         key.objectid = btrfs_ino(inode);
2718         key.type = BTRFS_EXTENT_DATA_KEY;
2719         key.offset = new->file_pos;
2720 
2721         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2722         if (ret < 0)
2723                 goto out_free_path;
2724         if (ret > 0 && path->slots[0] > 0)
2725                 path->slots[0]--;
2726 
2727         /* find out all the old extents for the file range */
2728         while (1) {
2729                 struct btrfs_file_extent_item *extent;
2730                 struct extent_buffer *l;
2731                 int slot;
2732                 u64 num_bytes;
2733                 u64 offset;
2734                 u64 end;
2735                 u64 disk_bytenr;
2736                 u64 extent_offset;
2737 
2738                 l = path->nodes[0];
2739                 slot = path->slots[0];
2740 
2741                 if (slot >= btrfs_header_nritems(l)) {
2742                         ret = btrfs_next_leaf(root, path);
2743                         if (ret < 0)
2744                                 goto out_free_path;
2745                         else if (ret > 0)
2746                                 break;
2747                         continue;
2748                 }
2749 
2750                 btrfs_item_key_to_cpu(l, &key, slot);
2751 
2752                 if (key.objectid != btrfs_ino(inode))
2753                         break;
2754                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2755                         break;
2756                 if (key.offset >= new->file_pos + new->len)
2757                         break;
2758 
2759                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2760 
2761                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2762                 if (key.offset + num_bytes < new->file_pos)
2763                         goto next;
2764 
2765                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2766                 if (!disk_bytenr)
2767                         goto next;
2768 
2769                 extent_offset = btrfs_file_extent_offset(l, extent);
2770 
2771                 old = kmalloc(sizeof(*old), GFP_NOFS);
2772                 if (!old)
2773                         goto out_free_path;
2774 
2775                 offset = max(new->file_pos, key.offset);
2776                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2777 
2778                 old->bytenr = disk_bytenr;
2779                 old->extent_offset = extent_offset;
2780                 old->offset = offset - key.offset;
2781                 old->len = end - offset;
2782                 old->new = new;
2783                 old->count = 0;
2784                 list_add_tail(&old->list, &new->head);
2785 next:
2786                 path->slots[0]++;
2787                 cond_resched();
2788         }
2789 
2790         btrfs_free_path(path);
2791         atomic_inc(&root->fs_info->defrag_running);
2792 
2793         return new;
2794 
2795 out_free_path:
2796         btrfs_free_path(path);
2797 out_kfree:
2798         free_sa_defrag_extent(new);
2799         return NULL;
2800 }
2801 
2802 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2803                                          u64 start, u64 len)
2804 {
2805         struct btrfs_block_group_cache *cache;
2806 
2807         cache = btrfs_lookup_block_group(root->fs_info, start);
2808         ASSERT(cache);
2809 
2810         spin_lock(&cache->lock);
2811         cache->delalloc_bytes -= len;
2812         spin_unlock(&cache->lock);
2813 
2814         btrfs_put_block_group(cache);
2815 }
2816 
2817 /* as ordered data IO finishes, this gets called so we can finish
2818  * an ordered extent if the range of bytes in the file it covers are
2819  * fully written.
2820  */
2821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2822 {
2823         struct inode *inode = ordered_extent->inode;
2824         struct btrfs_root *root = BTRFS_I(inode)->root;
2825         struct btrfs_trans_handle *trans = NULL;
2826         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2827         struct extent_state *cached_state = NULL;
2828         struct new_sa_defrag_extent *new = NULL;
2829         int compress_type = 0;
2830         int ret = 0;
2831         u64 logical_len = ordered_extent->len;
2832         bool nolock;
2833         bool truncated = false;
2834 
2835         nolock = btrfs_is_free_space_inode(inode);
2836 
2837         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2838                 ret = -EIO;
2839                 goto out;
2840         }
2841 
2842         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2843                                      ordered_extent->file_offset +
2844                                      ordered_extent->len - 1);
2845 
2846         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2847                 truncated = true;
2848                 logical_len = ordered_extent->truncated_len;
2849                 /* Truncated the entire extent, don't bother adding */
2850                 if (!logical_len)
2851                         goto out;
2852         }
2853 
2854         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2855                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2856 
2857                 /*
2858                  * For mwrite(mmap + memset to write) case, we still reserve
2859                  * space for NOCOW range.
2860                  * As NOCOW won't cause a new delayed ref, just free the space
2861                  */
2862                 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2863                                        ordered_extent->len);
2864                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2865                 if (nolock)
2866                         trans = btrfs_join_transaction_nolock(root);
2867                 else
2868                         trans = btrfs_join_transaction(root);
2869                 if (IS_ERR(trans)) {
2870                         ret = PTR_ERR(trans);
2871                         trans = NULL;
2872                         goto out;
2873                 }
2874                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2875                 ret = btrfs_update_inode_fallback(trans, root, inode);
2876                 if (ret) /* -ENOMEM or corruption */
2877                         btrfs_abort_transaction(trans, root, ret);
2878                 goto out;
2879         }
2880 
2881         lock_extent_bits(io_tree, ordered_extent->file_offset,
2882                          ordered_extent->file_offset + ordered_extent->len - 1,
2883                          &cached_state);
2884 
2885         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2886                         ordered_extent->file_offset + ordered_extent->len - 1,
2887                         EXTENT_DEFRAG, 1, cached_state);
2888         if (ret) {
2889                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2890                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2891                         /* the inode is shared */
2892                         new = record_old_file_extents(inode, ordered_extent);
2893 
2894                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2895                         ordered_extent->file_offset + ordered_extent->len - 1,
2896                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2897         }
2898 
2899         if (nolock)
2900                 trans = btrfs_join_transaction_nolock(root);
2901         else
2902                 trans = btrfs_join_transaction(root);
2903         if (IS_ERR(trans)) {
2904                 ret = PTR_ERR(trans);
2905                 trans = NULL;
2906                 goto out_unlock;
2907         }
2908 
2909         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2910 
2911         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2912                 compress_type = ordered_extent->compress_type;
2913         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2914                 BUG_ON(compress_type);
2915                 ret = btrfs_mark_extent_written(trans, inode,
2916                                                 ordered_extent->file_offset,
2917                                                 ordered_extent->file_offset +
2918                                                 logical_len);
2919         } else {
2920                 BUG_ON(root == root->fs_info->tree_root);
2921                 ret = insert_reserved_file_extent(trans, inode,
2922                                                 ordered_extent->file_offset,
2923                                                 ordered_extent->start,
2924                                                 ordered_extent->disk_len,
2925                                                 logical_len, logical_len,
2926                                                 compress_type, 0, 0,
2927                                                 BTRFS_FILE_EXTENT_REG);
2928                 if (!ret)
2929                         btrfs_release_delalloc_bytes(root,
2930                                                      ordered_extent->start,
2931                                                      ordered_extent->disk_len);
2932         }
2933         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2934                            ordered_extent->file_offset, ordered_extent->len,
2935                            trans->transid);
2936         if (ret < 0) {
2937                 btrfs_abort_transaction(trans, root, ret);
2938                 goto out_unlock;
2939         }
2940 
2941         add_pending_csums(trans, inode, ordered_extent->file_offset,
2942                           &ordered_extent->list);
2943 
2944         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2945         ret = btrfs_update_inode_fallback(trans, root, inode);
2946         if (ret) { /* -ENOMEM or corruption */
2947                 btrfs_abort_transaction(trans, root, ret);
2948                 goto out_unlock;
2949         }
2950         ret = 0;
2951 out_unlock:
2952         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2953                              ordered_extent->file_offset +
2954                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2955 out:
2956         if (root != root->fs_info->tree_root)
2957                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2958         if (trans)
2959                 btrfs_end_transaction(trans, root);
2960 
2961         if (ret || truncated) {
2962                 u64 start, end;
2963 
2964                 if (truncated)
2965                         start = ordered_extent->file_offset + logical_len;
2966                 else
2967                         start = ordered_extent->file_offset;
2968                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2969                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2970 
2971                 /* Drop the cache for the part of the extent we didn't write. */
2972                 btrfs_drop_extent_cache(inode, start, end, 0);
2973 
2974                 /*
2975                  * If the ordered extent had an IOERR or something else went
2976                  * wrong we need to return the space for this ordered extent
2977                  * back to the allocator.  We only free the extent in the
2978                  * truncated case if we didn't write out the extent at all.
2979                  */
2980                 if ((ret || !logical_len) &&
2981                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2982                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2983                         btrfs_free_reserved_extent(root, ordered_extent->start,
2984                                                    ordered_extent->disk_len, 1);
2985         }
2986 
2987 
2988         /*
2989          * This needs to be done to make sure anybody waiting knows we are done
2990          * updating everything for this ordered extent.
2991          */
2992         btrfs_remove_ordered_extent(inode, ordered_extent);
2993 
2994         /* for snapshot-aware defrag */
2995         if (new) {
2996                 if (ret) {
2997                         free_sa_defrag_extent(new);
2998                         atomic_dec(&root->fs_info->defrag_running);
2999                 } else {
3000                         relink_file_extents(new);
3001                 }
3002         }
3003 
3004         /* once for us */
3005         btrfs_put_ordered_extent(ordered_extent);
3006         /* once for the tree */
3007         btrfs_put_ordered_extent(ordered_extent);
3008 
3009         return ret;
3010 }
3011 
3012 static void finish_ordered_fn(struct btrfs_work *work)
3013 {
3014         struct btrfs_ordered_extent *ordered_extent;
3015         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3016         btrfs_finish_ordered_io(ordered_extent);
3017 }
3018 
3019 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3020                                 struct extent_state *state, int uptodate)
3021 {
3022         struct inode *inode = page->mapping->host;
3023         struct btrfs_root *root = BTRFS_I(inode)->root;
3024         struct btrfs_ordered_extent *ordered_extent = NULL;
3025         struct btrfs_workqueue *wq;
3026         btrfs_work_func_t func;
3027 
3028         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3029 
3030         ClearPagePrivate2(page);
3031         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3032                                             end - start + 1, uptodate))
3033                 return 0;
3034 
3035         if (btrfs_is_free_space_inode(inode)) {
3036                 wq = root->fs_info->endio_freespace_worker;
3037                 func = btrfs_freespace_write_helper;
3038         } else {
3039                 wq = root->fs_info->endio_write_workers;
3040                 func = btrfs_endio_write_helper;
3041         }
3042 
3043         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3044                         NULL);
3045         btrfs_queue_work(wq, &ordered_extent->work);
3046 
3047         return 0;
3048 }
3049 
3050 static int __readpage_endio_check(struct inode *inode,
3051                                   struct btrfs_io_bio *io_bio,
3052                                   int icsum, struct page *page,
3053                                   int pgoff, u64 start, size_t len)
3054 {
3055         char *kaddr;
3056         u32 csum_expected;
3057         u32 csum = ~(u32)0;
3058 
3059         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3060 
3061         kaddr = kmap_atomic(page);
3062         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3063         btrfs_csum_final(csum, (char *)&csum);
3064         if (csum != csum_expected)
3065                 goto zeroit;
3066 
3067         kunmap_atomic(kaddr);
3068         return 0;
3069 zeroit:
3070         btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3071                 "csum failed ino %llu off %llu csum %u expected csum %u",
3072                            btrfs_ino(inode), start, csum, csum_expected);
3073         memset(kaddr + pgoff, 1, len);
3074         flush_dcache_page(page);
3075         kunmap_atomic(kaddr);
3076         if (csum_expected == 0)
3077                 return 0;
3078         return -EIO;
3079 }
3080 
3081 /*
3082  * when reads are done, we need to check csums to verify the data is correct
3083  * if there's a match, we allow the bio to finish.  If not, the code in
3084  * extent_io.c will try to find good copies for us.
3085  */
3086 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3087                                       u64 phy_offset, struct page *page,
3088                                       u64 start, u64 end, int mirror)
3089 {
3090         size_t offset = start - page_offset(page);
3091         struct inode *inode = page->mapping->host;
3092         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3093         struct btrfs_root *root = BTRFS_I(inode)->root;
3094 
3095         if (PageChecked(page)) {
3096                 ClearPageChecked(page);
3097                 return 0;
3098         }
3099 
3100         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3101                 return 0;
3102 
3103         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3104             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3105                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3106                                   GFP_NOFS);
3107                 return 0;
3108         }
3109 
3110         phy_offset >>= inode->i_sb->s_blocksize_bits;
3111         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112                                       start, (size_t)(end - start + 1));
3113 }
3114 
3115 void btrfs_add_delayed_iput(struct inode *inode)
3116 {
3117         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3118         struct btrfs_inode *binode = BTRFS_I(inode);
3119 
3120         if (atomic_add_unless(&inode->i_count, -1, 1))
3121                 return;
3122 
3123         spin_lock(&fs_info->delayed_iput_lock);
3124         if (binode->delayed_iput_count == 0) {
3125                 ASSERT(list_empty(&binode->delayed_iput));
3126                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3127         } else {
3128                 binode->delayed_iput_count++;
3129         }
3130         spin_unlock(&fs_info->delayed_iput_lock);
3131 }
3132 
3133 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3134 {
3135         struct btrfs_fs_info *fs_info = root->fs_info;
3136 
3137         spin_lock(&fs_info->delayed_iput_lock);
3138         while (!list_empty(&fs_info->delayed_iputs)) {
3139                 struct btrfs_inode *inode;
3140 
3141                 inode = list_first_entry(&fs_info->delayed_iputs,
3142                                 struct btrfs_inode, delayed_iput);
3143                 if (inode->delayed_iput_count) {
3144                         inode->delayed_iput_count--;
3145                         list_move_tail(&inode->delayed_iput,
3146                                         &fs_info->delayed_iputs);
3147                 } else {
3148                         list_del_init(&inode->delayed_iput);
3149                 }
3150                 spin_unlock(&fs_info->delayed_iput_lock);
3151                 iput(&inode->vfs_inode);
3152                 spin_lock(&fs_info->delayed_iput_lock);
3153         }
3154         spin_unlock(&fs_info->delayed_iput_lock);
3155 }
3156 
3157 /*
3158  * This is called in transaction commit time. If there are no orphan
3159  * files in the subvolume, it removes orphan item and frees block_rsv
3160  * structure.
3161  */
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163                               struct btrfs_root *root)
3164 {
3165         struct btrfs_block_rsv *block_rsv;
3166         int ret;
3167 
3168         if (atomic_read(&root->orphan_inodes) ||
3169             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170                 return;
3171 
3172         spin_lock(&root->orphan_lock);
3173         if (atomic_read(&root->orphan_inodes)) {
3174                 spin_unlock(&root->orphan_lock);
3175                 return;
3176         }
3177 
3178         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179                 spin_unlock(&root->orphan_lock);
3180                 return;
3181         }
3182 
3183         block_rsv = root->orphan_block_rsv;
3184         root->orphan_block_rsv = NULL;
3185         spin_unlock(&root->orphan_lock);
3186 
3187         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3188             btrfs_root_refs(&root->root_item) > 0) {
3189                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190                                             root->root_key.objectid);
3191                 if (ret)
3192                         btrfs_abort_transaction(trans, root, ret);
3193                 else
3194                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3195                                   &root->state);
3196         }
3197 
3198         if (block_rsv) {
3199                 WARN_ON(block_rsv->size > 0);
3200                 btrfs_free_block_rsv(root, block_rsv);
3201         }
3202 }
3203 
3204 /*
3205  * This creates an orphan entry for the given inode in case something goes
3206  * wrong in the middle of an unlink/truncate.
3207  *
3208  * NOTE: caller of this function should reserve 5 units of metadata for
3209  *       this function.
3210  */
3211 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3212 {
3213         struct btrfs_root *root = BTRFS_I(inode)->root;
3214         struct btrfs_block_rsv *block_rsv = NULL;
3215         int reserve = 0;
3216         int insert = 0;
3217         int ret;
3218 
3219         if (!root->orphan_block_rsv) {
3220                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3221                 if (!block_rsv)
3222                         return -ENOMEM;
3223         }
3224 
3225         spin_lock(&root->orphan_lock);
3226         if (!root->orphan_block_rsv) {
3227                 root->orphan_block_rsv = block_rsv;
3228         } else if (block_rsv) {
3229                 btrfs_free_block_rsv(root, block_rsv);
3230                 block_rsv = NULL;
3231         }
3232 
3233         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234                               &BTRFS_I(inode)->runtime_flags)) {
3235 #if 0
3236                 /*
3237                  * For proper ENOSPC handling, we should do orphan
3238                  * cleanup when mounting. But this introduces backward
3239                  * compatibility issue.
3240                  */
3241                 if (!xchg(&root->orphan_item_inserted, 1))
3242                         insert = 2;
3243                 else
3244                         insert = 1;
3245 #endif
3246                 insert = 1;
3247                 atomic_inc(&root->orphan_inodes);
3248         }
3249 
3250         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251                               &BTRFS_I(inode)->runtime_flags))
3252                 reserve = 1;
3253         spin_unlock(&root->orphan_lock);
3254 
3255         /* grab metadata reservation from transaction handle */
3256         if (reserve) {
3257                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3258                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3259         }
3260 
3261         /* insert an orphan item to track this unlinked/truncated file */
3262         if (insert >= 1) {
3263                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3264                 if (ret) {
3265                         atomic_dec(&root->orphan_inodes);
3266                         if (reserve) {
3267                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268                                           &BTRFS_I(inode)->runtime_flags);
3269                                 btrfs_orphan_release_metadata(inode);
3270                         }
3271                         if (ret != -EEXIST) {
3272                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273                                           &BTRFS_I(inode)->runtime_flags);
3274                                 btrfs_abort_transaction(trans, root, ret);
3275                                 return ret;
3276                         }
3277                 }
3278                 ret = 0;
3279         }
3280 
3281         /* insert an orphan item to track subvolume contains orphan files */
3282         if (insert >= 2) {
3283                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284                                                root->root_key.objectid);
3285                 if (ret && ret != -EEXIST) {
3286                         btrfs_abort_transaction(trans, root, ret);
3287                         return ret;
3288                 }
3289         }
3290         return 0;
3291 }
3292 
3293 /*
3294  * We have done the truncate/delete so we can go ahead and remove the orphan
3295  * item for this particular inode.
3296  */
3297 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298                             struct inode *inode)
3299 {
3300         struct btrfs_root *root = BTRFS_I(inode)->root;
3301         int delete_item = 0;
3302         int release_rsv = 0;
3303         int ret = 0;
3304 
3305         spin_lock(&root->orphan_lock);
3306         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307                                &BTRFS_I(inode)->runtime_flags))
3308                 delete_item = 1;
3309 
3310         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311                                &BTRFS_I(inode)->runtime_flags))
3312                 release_rsv = 1;
3313         spin_unlock(&root->orphan_lock);
3314 
3315         if (delete_item) {
3316                 atomic_dec(&root->orphan_inodes);
3317                 if (trans)
3318                         ret = btrfs_del_orphan_item(trans, root,
3319                                                     btrfs_ino(inode));
3320         }
3321 
3322         if (release_rsv)
3323                 btrfs_orphan_release_metadata(inode);
3324 
3325         return ret;
3326 }
3327 
3328 /*
3329  * this cleans up any orphans that may be left on the list from the last use
3330  * of this root.
3331  */
3332 int btrfs_orphan_cleanup(struct btrfs_root *root)
3333 {
3334         struct btrfs_path *path;
3335         struct extent_buffer *leaf;
3336         struct btrfs_key key, found_key;
3337         struct btrfs_trans_handle *trans;
3338         struct inode *inode;
3339         u64 last_objectid = 0;
3340         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3341 
3342         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3343                 return 0;
3344 
3345         path = btrfs_alloc_path();
3346         if (!path) {
3347                 ret = -ENOMEM;
3348                 goto out;
3349         }
3350         path->reada = READA_BACK;
3351 
3352         key.objectid = BTRFS_ORPHAN_OBJECTID;
3353         key.type = BTRFS_ORPHAN_ITEM_KEY;
3354         key.offset = (u64)-1;
3355 
3356         while (1) {
3357                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3358                 if (ret < 0)
3359                         goto out;
3360 
3361                 /*
3362                  * if ret == 0 means we found what we were searching for, which
3363                  * is weird, but possible, so only screw with path if we didn't
3364                  * find the key and see if we have stuff that matches
3365                  */
3366                 if (ret > 0) {
3367                         ret = 0;
3368                         if (path->slots[0] == 0)
3369                                 break;
3370                         path->slots[0]--;
3371                 }
3372 
3373                 /* pull out the item */
3374                 leaf = path->nodes[0];
3375                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376 
3377                 /* make sure the item matches what we want */
3378                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3379                         break;
3380                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3381                         break;
3382 
3383                 /* release the path since we're done with it */
3384                 btrfs_release_path(path);
3385 
3386                 /*
3387                  * this is where we are basically btrfs_lookup, without the
3388                  * crossing root thing.  we store the inode number in the
3389                  * offset of the orphan item.
3390                  */
3391 
3392                 if (found_key.offset == last_objectid) {
3393                         btrfs_err(root->fs_info,
3394                                 "Error removing orphan entry, stopping orphan cleanup");
3395                         ret = -EINVAL;
3396                         goto out;
3397                 }
3398 
3399                 last_objectid = found_key.offset;
3400 
3401                 found_key.objectid = found_key.offset;
3402                 found_key.type = BTRFS_INODE_ITEM_KEY;
3403                 found_key.offset = 0;
3404                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3405                 ret = PTR_ERR_OR_ZERO(inode);
3406                 if (ret && ret != -ESTALE)
3407                         goto out;
3408 
3409                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410                         struct btrfs_root *dead_root;
3411                         struct btrfs_fs_info *fs_info = root->fs_info;
3412                         int is_dead_root = 0;
3413 
3414                         /*
3415                          * this is an orphan in the tree root. Currently these
3416                          * could come from 2 sources:
3417                          *  a) a snapshot deletion in progress
3418                          *  b) a free space cache inode
3419                          * We need to distinguish those two, as the snapshot
3420                          * orphan must not get deleted.
3421                          * find_dead_roots already ran before us, so if this
3422                          * is a snapshot deletion, we should find the root
3423                          * in the dead_roots list
3424                          */
3425                         spin_lock(&fs_info->trans_lock);
3426                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3427                                             root_list) {
3428                                 if (dead_root->root_key.objectid ==
3429                                     found_key.objectid) {
3430                                         is_dead_root = 1;
3431                                         break;
3432                                 }
3433                         }
3434                         spin_unlock(&fs_info->trans_lock);
3435                         if (is_dead_root) {
3436                                 /* prevent this orphan from being found again */
3437                                 key.offset = found_key.objectid - 1;
3438                                 continue;
3439                         }
3440                 }
3441                 /*
3442                  * Inode is already gone but the orphan item is still there,
3443                  * kill the orphan item.
3444                  */
3445                 if (ret == -ESTALE) {
3446                         trans = btrfs_start_transaction(root, 1);
3447                         if (IS_ERR(trans)) {
3448                                 ret = PTR_ERR(trans);
3449                                 goto out;
3450                         }
3451                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3452                                 found_key.objectid);
3453                         ret = btrfs_del_orphan_item(trans, root,
3454                                                     found_key.objectid);
3455                         btrfs_end_transaction(trans, root);
3456                         if (ret)
3457                                 goto out;
3458                         continue;
3459                 }
3460 
3461                 /*
3462                  * add this inode to the orphan list so btrfs_orphan_del does
3463                  * the proper thing when we hit it
3464                  */
3465                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466                         &BTRFS_I(inode)->runtime_flags);
3467                 atomic_inc(&root->orphan_inodes);
3468 
3469                 /* if we have links, this was a truncate, lets do that */
3470                 if (inode->i_nlink) {
3471                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3472                                 iput(inode);
3473                                 continue;
3474                         }
3475                         nr_truncate++;
3476 
3477                         /* 1 for the orphan item deletion. */
3478                         trans = btrfs_start_transaction(root, 1);
3479                         if (IS_ERR(trans)) {
3480                                 iput(inode);
3481                                 ret = PTR_ERR(trans);
3482                                 goto out;
3483                         }
3484                         ret = btrfs_orphan_add(trans, inode);
3485                         btrfs_end_transaction(trans, root);
3486                         if (ret) {
3487                                 iput(inode);
3488                                 goto out;
3489                         }
3490 
3491                         ret = btrfs_truncate(inode);
3492                         if (ret)
3493                                 btrfs_orphan_del(NULL, inode);
3494                 } else {
3495                         nr_unlink++;
3496                 }
3497 
3498                 /* this will do delete_inode and everything for us */
3499                 iput(inode);
3500                 if (ret)
3501                         goto out;
3502         }
3503         /* release the path since we're done with it */
3504         btrfs_release_path(path);
3505 
3506         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3507 
3508         if (root->orphan_block_rsv)
3509                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510                                         (u64)-1);
3511 
3512         if (root->orphan_block_rsv ||
3513             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3514                 trans = btrfs_join_transaction(root);
3515                 if (!IS_ERR(trans))
3516                         btrfs_end_transaction(trans, root);
3517         }
3518 
3519         if (nr_unlink)
3520                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3521         if (nr_truncate)
3522                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3523 
3524 out:
3525         if (ret)
3526                 btrfs_err(root->fs_info,
3527                         "could not do orphan cleanup %d", ret);
3528         btrfs_free_path(path);
3529         return ret;
3530 }
3531 
3532 /*
3533  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3534  * don't find any xattrs, we know there can't be any acls.
3535  *
3536  * slot is the slot the inode is in, objectid is the objectid of the inode
3537  */
3538 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3539                                           int slot, u64 objectid,
3540                                           int *first_xattr_slot)
3541 {
3542         u32 nritems = btrfs_header_nritems(leaf);
3543         struct btrfs_key found_key;
3544         static u64 xattr_access = 0;
3545         static u64 xattr_default = 0;
3546         int scanned = 0;
3547 
3548         if (!xattr_access) {
3549                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3550                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3551                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3552                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3553         }
3554 
3555         slot++;
3556         *first_xattr_slot = -1;
3557         while (slot < nritems) {
3558                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3559 
3560                 /* we found a different objectid, there must not be acls */
3561                 if (found_key.objectid != objectid)
3562                         return 0;
3563 
3564                 /* we found an xattr, assume we've got an acl */
3565                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3566                         if (*first_xattr_slot == -1)
3567                                 *first_xattr_slot = slot;
3568                         if (found_key.offset == xattr_access ||
3569                             found_key.offset == xattr_default)
3570                                 return 1;
3571                 }
3572 
3573                 /*
3574                  * we found a key greater than an xattr key, there can't
3575                  * be any acls later on
3576                  */
3577                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3578                         return 0;
3579 
3580                 slot++;
3581                 scanned++;
3582 
3583                 /*
3584                  * it goes inode, inode backrefs, xattrs, extents,
3585                  * so if there are a ton of hard links to an inode there can
3586                  * be a lot of backrefs.  Don't waste time searching too hard,
3587                  * this is just an optimization
3588                  */
3589                 if (scanned >= 8)
3590                         break;
3591         }
3592         /* we hit the end of the leaf before we found an xattr or
3593          * something larger than an xattr.  We have to assume the inode
3594          * has acls
3595          */
3596         if (*first_xattr_slot == -1)
3597                 *first_xattr_slot = slot;
3598         return 1;
3599 }
3600 
3601 /*
3602  * read an inode from the btree into the in-memory inode
3603  */
3604 static void btrfs_read_locked_inode(struct inode *inode)
3605 {
3606         struct btrfs_path *path;
3607         struct extent_buffer *leaf;
3608         struct btrfs_inode_item *inode_item;
3609         struct btrfs_root *root = BTRFS_I(inode)->root;
3610         struct btrfs_key location;
3611         unsigned long ptr;
3612         int maybe_acls;
3613         u32 rdev;
3614         int ret;
3615         bool filled = false;
3616         int first_xattr_slot;
3617 
3618         ret = btrfs_fill_inode(inode, &rdev);
3619         if (!ret)
3620                 filled = true;
3621 
3622         path = btrfs_alloc_path();
3623         if (!path)
3624                 goto make_bad;
3625 
3626         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3627 
3628         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3629         if (ret)
3630                 goto make_bad;
3631 
3632         leaf = path->nodes[0];
3633 
3634         if (filled)
3635                 goto cache_index;
3636 
3637         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638                                     struct btrfs_inode_item);
3639         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3640         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3641         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3643         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3644 
3645         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3647 
3648         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3650 
3651         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3653 
3654         BTRFS_I(inode)->i_otime.tv_sec =
3655                 btrfs_timespec_sec(leaf, &inode_item->otime);
3656         BTRFS_I(inode)->i_otime.tv_nsec =
3657                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3658 
3659         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3660         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3661         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3662 
3663         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664         inode->i_generation = BTRFS_I(inode)->generation;
3665         inode->i_rdev = 0;
3666         rdev = btrfs_inode_rdev(leaf, inode_item);
3667 
3668         BTRFS_I(inode)->index_cnt = (u64)-1;
3669         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3670 
3671 cache_index:
3672         /*
3673          * If we were modified in the current generation and evicted from memory
3674          * and then re-read we need to do a full sync since we don't have any
3675          * idea about which extents were modified before we were evicted from
3676          * cache.
3677          *
3678          * This is required for both inode re-read from disk and delayed inode
3679          * in delayed_nodes_tree.
3680          */
3681         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683                         &BTRFS_I(inode)->runtime_flags);
3684 
3685         /*
3686          * We don't persist the id of the transaction where an unlink operation
3687          * against the inode was last made. So here we assume the inode might
3688          * have been evicted, and therefore the exact value of last_unlink_trans
3689          * lost, and set it to last_trans to avoid metadata inconsistencies
3690          * between the inode and its parent if the inode is fsync'ed and the log
3691          * replayed. For example, in the scenario:
3692          *
3693          * touch mydir/foo
3694          * ln mydir/foo mydir/bar
3695          * sync
3696          * unlink mydir/bar
3697          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3698          * xfs_io -c fsync mydir/foo
3699          * <power failure>
3700          * mount fs, triggers fsync log replay
3701          *
3702          * We must make sure that when we fsync our inode foo we also log its
3703          * parent inode, otherwise after log replay the parent still has the
3704          * dentry with the "bar" name but our inode foo has a link count of 1
3705          * and doesn't have an inode ref with the name "bar" anymore.
3706          *
3707          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708          * but it guarantees correctness at the expense of ocassional full
3709          * transaction commits on fsync if our inode is a directory, or if our
3710          * inode is not a directory, logging its parent unnecessarily.
3711          */
3712         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713 
3714         path->slots[0]++;
3715         if (inode->i_nlink != 1 ||
3716             path->slots[0] >= btrfs_header_nritems(leaf))
3717                 goto cache_acl;
3718 
3719         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720         if (location.objectid != btrfs_ino(inode))
3721                 goto cache_acl;
3722 
3723         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724         if (location.type == BTRFS_INODE_REF_KEY) {
3725                 struct btrfs_inode_ref *ref;
3726 
3727                 ref = (struct btrfs_inode_ref *)ptr;
3728                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730                 struct btrfs_inode_extref *extref;
3731 
3732                 extref = (struct btrfs_inode_extref *)ptr;
3733                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734                                                                      extref);
3735         }
3736 cache_acl:
3737         /*
3738          * try to precache a NULL acl entry for files that don't have
3739          * any xattrs or acls
3740          */
3741         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3742                                            btrfs_ino(inode), &first_xattr_slot);
3743         if (first_xattr_slot != -1) {
3744                 path->slots[0] = first_xattr_slot;
3745                 ret = btrfs_load_inode_props(inode, path);
3746                 if (ret)
3747                         btrfs_err(root->fs_info,
3748                                   "error loading props for ino %llu (root %llu): %d",
3749                                   btrfs_ino(inode),
3750                                   root->root_key.objectid, ret);
3751         }
3752         btrfs_free_path(path);
3753 
3754         if (!maybe_acls)
3755                 cache_no_acl(inode);
3756 
3757         switch (inode->i_mode & S_IFMT) {
3758         case S_IFREG:
3759                 inode->i_mapping->a_ops = &btrfs_aops;
3760                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3761                 inode->i_fop = &btrfs_file_operations;
3762                 inode->i_op = &btrfs_file_inode_operations;
3763                 break;
3764         case S_IFDIR:
3765                 inode->i_fop = &btrfs_dir_file_operations;
3766                 if (root == root->fs_info->tree_root)
3767                         inode->i_op = &btrfs_dir_ro_inode_operations;
3768                 else
3769                         inode->i_op = &btrfs_dir_inode_operations;
3770                 break;
3771         case S_IFLNK:
3772                 inode->i_op = &btrfs_symlink_inode_operations;
3773                 inode_nohighmem(inode);
3774                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3775                 break;
3776         default:
3777                 inode->i_op = &btrfs_special_inode_operations;
3778                 init_special_inode(inode, inode->i_mode, rdev);
3779                 break;
3780         }
3781 
3782         btrfs_update_iflags(inode);
3783         return;
3784 
3785 make_bad:
3786         btrfs_free_path(path);
3787         make_bad_inode(inode);
3788 }
3789 
3790 /*
3791  * given a leaf and an inode, copy the inode fields into the leaf
3792  */
3793 static void fill_inode_item(struct btrfs_trans_handle *trans,
3794                             struct extent_buffer *leaf,
3795                             struct btrfs_inode_item *item,
3796                             struct inode *inode)
3797 {
3798         struct btrfs_map_token token;
3799 
3800         btrfs_init_map_token(&token);
3801 
3802         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3803         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3804         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805                                    &token);
3806         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3807         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3808 
3809         btrfs_set_token_timespec_sec(leaf, &item->atime,
3810                                      inode->i_atime.tv_sec, &token);
3811         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3812                                       inode->i_atime.tv_nsec, &token);
3813 
3814         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3815                                      inode->i_mtime.tv_sec, &token);
3816         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3817                                       inode->i_mtime.tv_nsec, &token);
3818 
3819         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3820                                      inode->i_ctime.tv_sec, &token);
3821         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3822                                       inode->i_ctime.tv_nsec, &token);
3823 
3824         btrfs_set_token_timespec_sec(leaf, &item->otime,
3825                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3826         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3827                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828 
3829         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830                                      &token);
3831         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832                                          &token);
3833         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3834         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3835         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3836         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3837         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3838 }
3839 
3840 /*
3841  * copy everything in the in-memory inode into the btree.
3842  */
3843 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3844                                 struct btrfs_root *root, struct inode *inode)
3845 {
3846         struct btrfs_inode_item *inode_item;
3847         struct btrfs_path *path;
3848         struct extent_buffer *leaf;
3849         int ret;
3850 
3851         path = btrfs_alloc_path();
3852         if (!path)
3853                 return -ENOMEM;
3854 
3855         path->leave_spinning = 1;
3856         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3857                                  1);
3858         if (ret) {
3859                 if (ret > 0)
3860                         ret = -ENOENT;
3861                 goto failed;
3862         }
3863 
3864         leaf = path->nodes[0];
3865         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3866                                     struct btrfs_inode_item);
3867 
3868         fill_inode_item(trans, leaf, inode_item, inode);
3869         btrfs_mark_buffer_dirty(leaf);
3870         btrfs_set_inode_last_trans(trans, inode);
3871         ret = 0;
3872 failed:
3873         btrfs_free_path(path);
3874         return ret;
3875 }
3876 
3877 /*
3878  * copy everything in the in-memory inode into the btree.
3879  */
3880 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3881                                 struct btrfs_root *root, struct inode *inode)
3882 {
3883         int ret;
3884 
3885         /*
3886          * If the inode is a free space inode, we can deadlock during commit
3887          * if we put it into the delayed code.
3888          *
3889          * The data relocation inode should also be directly updated
3890          * without delay
3891          */
3892         if (!btrfs_is_free_space_inode(inode)
3893             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3894             && !root->fs_info->log_root_recovering) {
3895                 btrfs_update_root_times(trans, root);
3896 
3897                 ret = btrfs_delayed_update_inode(trans, root, inode);
3898                 if (!ret)
3899                         btrfs_set_inode_last_trans(trans, inode);
3900                 return ret;
3901         }
3902 
3903         return btrfs_update_inode_item(trans, root, inode);
3904 }
3905 
3906 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3907                                          struct btrfs_root *root,
3908                                          struct inode *inode)
3909 {
3910         int ret;
3911 
3912         ret = btrfs_update_inode(trans, root, inode);
3913         if (ret == -ENOSPC)
3914                 return btrfs_update_inode_item(trans, root, inode);
3915         return ret;
3916 }
3917 
3918 /*
3919  * unlink helper that gets used here in inode.c and in the tree logging
3920  * recovery code.  It remove a link in a directory with a given name, and
3921  * also drops the back refs in the inode to the directory
3922  */
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3924                                 struct btrfs_root *root,
3925                                 struct inode *dir, struct inode *inode,
3926                                 const char *name, int name_len)
3927 {
3928         struct btrfs_path *path;
3929         int ret = 0;
3930         struct extent_buffer *leaf;
3931         struct btrfs_dir_item *di;
3932         struct btrfs_key key;
3933         u64 index;
3934         u64 ino = btrfs_ino(inode);
3935         u64 dir_ino = btrfs_ino(dir);
3936 
3937         path = btrfs_alloc_path();
3938         if (!path) {
3939                 ret = -ENOMEM;
3940                 goto out;
3941         }
3942 
3943         path->leave_spinning = 1;
3944         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3945                                     name, name_len, -1);
3946         if (IS_ERR(di)) {
3947                 ret = PTR_ERR(di);
3948                 goto err;
3949         }
3950         if (!di) {
3951                 ret = -ENOENT;
3952                 goto err;
3953         }
3954         leaf = path->nodes[0];
3955         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3956         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3957         if (ret)
3958                 goto err;
3959         btrfs_release_path(path);
3960 
3961         /*
3962          * If we don't have dir index, we have to get it by looking up
3963          * the inode ref, since we get the inode ref, remove it directly,
3964          * it is unnecessary to do delayed deletion.
3965          *
3966          * But if we have dir index, needn't search inode ref to get it.
3967          * Since the inode ref is close to the inode item, it is better
3968          * that we delay to delete it, and just do this deletion when
3969          * we update the inode item.
3970          */
3971         if (BTRFS_I(inode)->dir_index) {
3972                 ret = btrfs_delayed_delete_inode_ref(inode);
3973                 if (!ret) {
3974                         index = BTRFS_I(inode)->dir_index;
3975                         goto skip_backref;
3976                 }
3977         }
3978 
3979         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3980                                   dir_ino, &index);
3981         if (ret) {
3982                 btrfs_info(root->fs_info,
3983                         "failed to delete reference to %.*s, inode %llu parent %llu",
3984                         name_len, name, ino, dir_ino);
3985                 btrfs_abort_transaction(trans, root, ret);
3986                 goto err;
3987         }
3988 skip_backref:
3989         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3990         if (ret) {
3991                 btrfs_abort_transaction(trans, root, ret);
3992                 goto err;
3993         }
3994 
3995         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3996                                          inode, dir_ino);
3997         if (ret != 0 && ret != -ENOENT) {
3998                 btrfs_abort_transaction(trans, root, ret);
3999                 goto err;
4000         }
4001 
4002         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4003                                            dir, index);
4004         if (ret == -ENOENT)
4005                 ret = 0;
4006         else if (ret)
4007                 btrfs_abort_transaction(trans, root, ret);
4008 err:
4009         btrfs_free_path(path);
4010         if (ret)
4011                 goto out;
4012 
4013         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4014         inode_inc_iversion(inode);
4015         inode_inc_iversion(dir);
4016         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4017         ret = btrfs_update_inode(trans, root, dir);
4018 out:
4019         return ret;
4020 }
4021 
4022 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4023                        struct btrfs_root *root,
4024                        struct inode *dir, struct inode *inode,
4025                        const char *name, int name_len)
4026 {
4027         int ret;
4028         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4029         if (!ret) {
4030                 drop_nlink(inode);
4031                 ret = btrfs_update_inode(trans, root, inode);
4032         }
4033         return ret;
4034 }
4035 
4036 /*
4037  * helper to start transaction for unlink and rmdir.
4038  *
4039  * unlink and rmdir are special in btrfs, they do not always free space, so
4040  * if we cannot make our reservations the normal way try and see if there is
4041  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4042  * allow the unlink to occur.
4043  */
4044 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4045 {
4046         struct btrfs_root *root = BTRFS_I(dir)->root;
4047 
4048         /*
4049          * 1 for the possible orphan item
4050          * 1 for the dir item
4051          * 1 for the dir index
4052          * 1 for the inode ref
4053          * 1 for the inode
4054          */
4055         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4056 }
4057 
4058 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4059 {
4060         struct btrfs_root *root = BTRFS_I(dir)->root;
4061         struct btrfs_trans_handle *trans;
4062         struct inode *inode = d_inode(dentry);
4063         int ret;
4064 
4065         trans = __unlink_start_trans(dir);
4066         if (IS_ERR(trans))
4067                 return PTR_ERR(trans);
4068 
4069         btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4070 
4071         ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4072                                  dentry->d_name.name, dentry->d_name.len);
4073         if (ret)
4074                 goto out;
4075 
4076         if (inode->i_nlink == 0) {
4077                 ret = btrfs_orphan_add(trans, inode);
4078                 if (ret)
4079                         goto out;
4080         }
4081 
4082 out:
4083         btrfs_end_transaction(trans, root);
4084         btrfs_btree_balance_dirty(root);
4085         return ret;
4086 }
4087 
4088 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4089                         struct btrfs_root *root,
4090                         struct inode *dir, u64 objectid,
4091                         const char *name, int name_len)
4092 {
4093         struct btrfs_path *path;
4094         struct extent_buffer *leaf;
4095         struct btrfs_dir_item *di;
4096         struct btrfs_key key;
4097         u64 index;
4098         int ret;
4099         u64 dir_ino = btrfs_ino(dir);
4100 
4101         path = btrfs_alloc_path();
4102         if (!path)
4103                 return -ENOMEM;
4104 
4105         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4106                                    name, name_len, -1);
4107         if (IS_ERR_OR_NULL(di)) {
4108                 if (!di)
4109                         ret = -ENOENT;
4110                 else
4111                         ret = PTR_ERR(di);
4112                 goto out;
4113         }
4114 
4115         leaf = path->nodes[0];
4116         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4117         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4118         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119         if (ret) {
4120                 btrfs_abort_transaction(trans, root, ret);
4121                 goto out;
4122         }
4123         btrfs_release_path(path);
4124 
4125         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4126                                  objectid, root->root_key.objectid,
4127                                  dir_ino, &index, name, name_len);
4128         if (ret < 0) {
4129                 if (ret != -ENOENT) {
4130                         btrfs_abort_transaction(trans, root, ret);
4131                         goto out;
4132                 }
4133                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4134                                                  name, name_len);
4135                 if (IS_ERR_OR_NULL(di)) {
4136                         if (!di)
4137                                 ret = -ENOENT;
4138                         else
4139                                 ret = PTR_ERR(di);
4140                         btrfs_abort_transaction(trans, root, ret);
4141                         goto out;
4142                 }
4143 
4144                 leaf = path->nodes[0];
4145                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4146                 btrfs_release_path(path);
4147                 index = key.offset;
4148         }
4149         btrfs_release_path(path);
4150 
4151         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4152         if (ret) {
4153                 btrfs_abort_transaction(trans, root, ret);
4154                 goto out;
4155         }
4156 
4157         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4158         inode_inc_iversion(dir);
4159         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4160         ret = btrfs_update_inode_fallback(trans, root, dir);
4161         if (ret)
4162                 btrfs_abort_transaction(trans, root, ret);
4163 out:
4164         btrfs_free_path(path);
4165         return ret;
4166 }
4167 
4168 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4169 {
4170         struct inode *inode = d_inode(dentry);
4171         int err = 0;
4172         struct btrfs_root *root = BTRFS_I(dir)->root;
4173         struct btrfs_trans_handle *trans;
4174 
4175         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4176                 return -ENOTEMPTY;
4177         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4178                 return -EPERM;
4179 
4180         trans = __unlink_start_trans(dir);
4181         if (IS_ERR(trans))
4182                 return PTR_ERR(trans);
4183 
4184         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4185                 err = btrfs_unlink_subvol(trans, root, dir,
4186                                           BTRFS_I(inode)->location.objectid,
4187                                           dentry->d_name.name,
4188                                           dentry->d_name.len);
4189                 goto out;
4190         }
4191 
4192         err = btrfs_orphan_add(trans, inode);
4193         if (err)
4194                 goto out;
4195 
4196         /* now the directory is empty */
4197         err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4198                                  dentry->d_name.name, dentry->d_name.len);
4199         if (!err)
4200                 btrfs_i_size_write(inode, 0);
4201 out:
4202         btrfs_end_transaction(trans, root);
4203         btrfs_btree_balance_dirty(root);
4204 
4205         return err;
4206 }
4207 
4208 static int truncate_space_check(struct btrfs_trans_handle *trans,
4209                                 struct btrfs_root *root,
4210                                 u64 bytes_deleted)
4211 {
4212         int ret;
4213 
4214         /*
4215          * This is only used to apply pressure to the enospc system, we don't
4216          * intend to use this reservation at all.
4217          */
4218         bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4219         bytes_deleted *= root->nodesize;
4220         ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4221                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4222         if (!ret) {
4223                 trace_btrfs_space_reservation(root->fs_info, "transaction",
4224                                               trans->transid,
4225                                               bytes_deleted, 1);
4226                 trans->bytes_reserved += bytes_deleted;
4227         }
4228         return ret;
4229 
4230 }
4231 
4232 static int truncate_inline_extent(struct inode *inode,
4233                                   struct btrfs_path *path,
4234                                   struct btrfs_key *found_key,
4235                                   const u64 item_end,
4236                                   const u64 new_size)
4237 {
4238         struct extent_buffer *leaf = path->nodes[0];
4239         int slot = path->slots[0];
4240         struct btrfs_file_extent_item *fi;
4241         u32 size = (u32)(new_size - found_key->offset);
4242         struct btrfs_root *root = BTRFS_I(inode)->root;
4243 
4244         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4245 
4246         if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4247                 loff_t offset = new_size;
4248                 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4249 
4250                 /*
4251                  * Zero out the remaining of the last page of our inline extent,
4252                  * instead of directly truncating our inline extent here - that
4253                  * would be much more complex (decompressing all the data, then
4254                  * compressing the truncated data, which might be bigger than
4255                  * the size of the inline extent, resize the extent, etc).
4256                  * We release the path because to get the page we might need to
4257                  * read the extent item from disk (data not in the page cache).
4258                  */
4259                 btrfs_release_path(path);
4260                 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4261         }
4262 
4263         btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4264         size = btrfs_file_extent_calc_inline_size(size);
4265         btrfs_truncate_item(root, path, size, 1);
4266 
4267         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4268                 inode_sub_bytes(inode, item_end + 1 - new_size);
4269 
4270         return 0;
4271 }
4272 
4273 /*
4274  * this can truncate away extent items, csum items and directory items.
4275  * It starts at a high offset and removes keys until it can't find
4276  * any higher than new_size
4277  *
4278  * csum items that cross the new i_size are truncated to the new size
4279  * as well.
4280  *
4281  * min_type is the minimum key type to truncate down to.  If set to 0, this
4282  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4283  */
4284 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4285                                struct btrfs_root *root,
4286                                struct inode *inode,
4287                                u64 new_size, u32 min_type)
4288 {
4289         struct btrfs_path *path;
4290         struct extent_buffer *leaf;
4291         struct btrfs_file_extent_item *fi;
4292         struct btrfs_key key;
4293         struct btrfs_key found_key;
4294         u64 extent_start = 0;
4295         u64 extent_num_bytes = 0;
4296         u64 extent_offset = 0;
4297         u64 item_end = 0;
4298         u64 last_size = new_size;
4299         u32 found_type = (u8)-1;
4300         int found_extent;
4301         int del_item;
4302         int pending_del_nr = 0;
4303         int pending_del_slot = 0;
4304         int extent_type = -1;
4305         int ret;
4306         int err = 0;
4307         u64 ino = btrfs_ino(inode);
4308         u64 bytes_deleted = 0;
4309         bool be_nice = 0;
4310         bool should_throttle = 0;
4311         bool should_end = 0;
4312 
4313         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4314 
4315         /*
4316          * for non-free space inodes and ref cows, we want to back off from
4317          * time to time
4318          */
4319         if (!btrfs_is_free_space_inode(inode) &&
4320             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4321                 be_nice = 1;
4322 
4323         path = btrfs_alloc_path();
4324         if (!path)
4325                 return -ENOMEM;
4326         path->reada = READA_BACK;
4327 
4328         /*
4329          * We want to drop from the next block forward in case this new size is
4330          * not block aligned since we will be keeping the last block of the
4331          * extent just the way it is.
4332          */
4333         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4334             root == root->fs_info->tree_root)
4335                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4336                                         root->sectorsize), (u64)-1, 0);
4337 
4338         /*
4339          * This function is also used to drop the items in the log tree before
4340          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4341          * it is used to drop the loged items. So we shouldn't kill the delayed
4342          * items.
4343          */
4344         if (min_type == 0 && root == BTRFS_I(inode)->root)
4345                 btrfs_kill_delayed_inode_items(inode);
4346 
4347         key.objectid = ino;
4348         key.offset = (u64)-1;
4349         key.type = (u8)-1;
4350 
4351 search_again:
4352         /*
4353          * with a 16K leaf size and 128MB extents, you can actually queue
4354          * up a huge file in a single leaf.  Most of the time that
4355          * bytes_deleted is > 0, it will be huge by the time we get here
4356          */
4357         if (be_nice && bytes_deleted > SZ_32M) {
4358                 if (btrfs_should_end_transaction(trans, root)) {
4359                         err = -EAGAIN;
4360                         goto error;
4361                 }
4362         }
4363 
4364 
4365         path->leave_spinning = 1;
4366         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4367         if (ret < 0) {
4368                 err = ret;
4369                 goto out;
4370         }
4371 
4372         if (ret > 0) {
4373                 /* there are no items in the tree for us to truncate, we're
4374                  * done
4375                  */
4376                 if (path->slots[0] == 0)
4377                         goto out;
4378                 path->slots[0]--;
4379         }
4380 
4381         while (1) {
4382                 fi = NULL;
4383                 leaf = path->nodes[0];
4384                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4385                 found_type = found_key.type;
4386 
4387                 if (found_key.objectid != ino)
4388                         break;
4389 
4390                 if (found_type < min_type)
4391                         break;
4392 
4393                 item_end = found_key.offset;
4394                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4395                         fi = btrfs_item_ptr(leaf, path->slots[0],
4396                                             struct btrfs_file_extent_item);
4397                         extent_type = btrfs_file_extent_type(leaf, fi);
4398                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4399                                 item_end +=
4400                                     btrfs_file_extent_num_bytes(leaf, fi);
4401                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4402                                 item_end += btrfs_file_extent_inline_len(leaf,
4403                                                          path->slots[0], fi);
4404                         }
4405                         item_end--;
4406                 }
4407                 if (found_type > min_type) {
4408                         del_item = 1;
4409                 } else {
4410                         if (item_end < new_size)
4411                                 break;
4412                         if (found_key.offset >= new_size)
4413                                 del_item = 1;
4414                         else
4415                                 del_item = 0;
4416                 }
4417                 found_extent = 0;
4418                 /* FIXME, shrink the extent if the ref count is only 1 */
4419                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4420                         goto delete;
4421 
4422                 if (del_item)
4423                         last_size = found_key.offset;
4424                 else
4425                         last_size = new_size;
4426 
4427                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4428                         u64 num_dec;
4429                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4430                         if (!del_item) {
4431                                 u64 orig_num_bytes =
4432                                         btrfs_file_extent_num_bytes(leaf, fi);
4433                                 extent_num_bytes = ALIGN(new_size -
4434                                                 found_key.offset,
4435                                                 root->sectorsize);
4436                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4437                                                          extent_num_bytes);
4438                                 num_dec = (orig_num_bytes -
4439                                            extent_num_bytes);
4440                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4441                                              &root->state) &&
4442                                     extent_start != 0)
4443                                         inode_sub_bytes(inode, num_dec);
4444                                 btrfs_mark_buffer_dirty(leaf);
4445                         } else {
4446                                 extent_num_bytes =
4447                                         btrfs_file_extent_disk_num_bytes(leaf,
4448                                                                          fi);
4449                                 extent_offset = found_key.offset -
4450                                         btrfs_file_extent_offset(leaf, fi);
4451 
4452                                 /* FIXME blocksize != 4096 */
4453                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4454                                 if (extent_start != 0) {
4455                                         found_extent = 1;
4456                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4457                                                      &root->state))
4458                                                 inode_sub_bytes(inode, num_dec);
4459                                 }
4460                         }
4461                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4462                         /*
4463                          * we can't truncate inline items that have had
4464                          * special encodings
4465                          */
4466                         if (!del_item &&
4467                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4468                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4469 
4470                                 /*
4471                                  * Need to release path in order to truncate a
4472                                  * compressed extent. So delete any accumulated
4473                                  * extent items so far.
4474                                  */
4475                                 if (btrfs_file_extent_compression(leaf, fi) !=
4476                                     BTRFS_COMPRESS_NONE && pending_del_nr) {
4477                                         err = btrfs_del_items(trans, root, path,
4478                                                               pending_del_slot,
4479                                                               pending_del_nr);
4480                                         if (err) {
4481                                                 btrfs_abort_transaction(trans,
4482                                                                         root,
4483                                                                         err);
4484                                                 goto error;
4485                                         }
4486                                         pending_del_nr = 0;
4487                                 }
4488 
4489                                 err = truncate_inline_extent(inode, path,
4490                                                              &found_key,
4491                                                              item_end,
4492                                                              new_size);
4493                                 if (err) {
4494                                         btrfs_abort_transaction(trans,
4495                                                                 root, err);
4496                                         goto error;
4497                                 }
4498                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4499                                             &root->state)) {
4500                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4501                         }
4502                 }
4503 delete:
4504                 if (del_item) {
4505                         if (!pending_del_nr) {
4506                                 /* no pending yet, add ourselves */
4507                                 pending_del_slot = path->slots[0];
4508                                 pending_del_nr = 1;
4509                         } else if (pending_del_nr &&
4510                                    path->slots[0] + 1 == pending_del_slot) {
4511                                 /* hop on the pending chunk */
4512                                 pending_del_nr++;
4513                                 pending_del_slot = path->slots[0];
4514                         } else {
4515                                 BUG();
4516                         }
4517                 } else {
4518                         break;
4519                 }
4520                 should_throttle = 0;
4521 
4522                 if (found_extent &&
4523                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4524                      root == root->fs_info->tree_root)) {
4525                         btrfs_set_path_blocking(path);
4526                         bytes_deleted += extent_num_bytes;
4527                         ret = btrfs_free_extent(trans, root, extent_start,
4528                                                 extent_num_bytes, 0,
4529                                                 btrfs_header_owner(leaf),
4530                                                 ino, extent_offset);
4531                         BUG_ON(ret);
4532                         if (btrfs_should_throttle_delayed_refs(trans, root))
4533                                 btrfs_async_run_delayed_refs(root,
4534                                         trans->delayed_ref_updates * 2, 0);
4535                         if (be_nice) {
4536                                 if (truncate_space_check(trans, root,
4537                                                          extent_num_bytes)) {
4538                                         should_end = 1;
4539                                 }
4540                                 if (btrfs_should_throttle_delayed_refs(trans,
4541                                                                        root)) {
4542                                         should_throttle = 1;
4543                                 }
4544                         }
4545                 }
4546 
4547                 if (found_type == BTRFS_INODE_ITEM_KEY)
4548                         break;
4549 
4550                 if (path->slots[0] == 0 ||
4551                     path->slots[0] != pending_del_slot ||
4552                     should_throttle || should_end) {
4553                         if (pending_del_nr) {
4554                                 ret = btrfs_del_items(trans, root, path,
4555                                                 pending_del_slot,
4556                                                 pending_del_nr);
4557                                 if (ret) {
4558                                         btrfs_abort_transaction(trans,
4559                                                                 root, ret);
4560                                         goto error;
4561                                 }
4562                                 pending_del_nr = 0;
4563                         }
4564                         btrfs_release_path(path);
4565                         if (should_throttle) {
4566                                 unsigned long updates = trans->delayed_ref_updates;
4567                                 if (updates) {
4568                                         trans->delayed_ref_updates = 0;
4569                                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4570                                         if (ret && !err)
4571                                                 err = ret;
4572                                 }
4573                         }
4574                         /*
4575                          * if we failed to refill our space rsv, bail out
4576                          * and let the transaction restart
4577                          */
4578                         if (should_end) {
4579                                 err = -EAGAIN;
4580                                 goto error;
4581                         }
4582                         goto search_again;
4583                 } else {
4584                         path->slots[0]--;
4585                 }
4586         }
4587 out:
4588         if (pending_del_nr) {
4589                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4590                                       pending_del_nr);
4591                 if (ret)
4592                         btrfs_abort_transaction(trans, root, ret);
4593         }
4594 error:
4595         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4596                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4597 
4598         btrfs_free_path(path);
4599 
4600         if (be_nice && bytes_deleted > SZ_32M) {
4601                 unsigned long updates = trans->delayed_ref_updates;
4602                 if (updates) {
4603                         trans->delayed_ref_updates = 0;
4604                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4605                         if (ret && !err)
4606                                 err = ret;
4607                 }
4608         }
4609         return err;
4610 }
4611 
4612 /*
4613  * btrfs_truncate_page - read, zero a chunk and write a page
4614  * @inode - inode that we're zeroing
4615  * @from - the offset to start zeroing
4616  * @len - the length to zero, 0 to zero the entire range respective to the
4617  *      offset
4618  * @front - zero up to the offset instead of from the offset on
4619  *
4620  * This will find the page for the "from" offset and cow the page and zero the
4621  * part we want to zero.  This is used with truncate and hole punching.
4622  */
4623 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4624                         int front)
4625 {
4626         struct address_space *mapping = inode->i_mapping;
4627         struct btrfs_root *root = BTRFS_I(inode)->root;
4628         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4629         struct btrfs_ordered_extent *ordered;
4630         struct extent_state *cached_state = NULL;
4631         char *kaddr;
4632         u32 blocksize = root->sectorsize;
4633         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4634         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4635         struct page *page;
4636         gfp_t mask = btrfs_alloc_write_mask(mapping);
4637         int ret = 0;
4638         u64 page_start;
4639         u64 page_end;
4640 
4641         if ((offset & (blocksize - 1)) == 0 &&
4642             (!len || ((len & (blocksize - 1)) == 0)))
4643                 goto out;
4644         ret = btrfs_delalloc_reserve_space(inode,
4645                         round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4646         if (ret)
4647                 goto out;
4648 
4649 again:
4650         page = find_or_create_page(mapping, index, mask);
4651         if (!page) {
4652                 btrfs_delalloc_release_space(inode,
4653                                 round_down(from, PAGE_CACHE_SIZE),
4654                                 PAGE_CACHE_SIZE);
4655                 ret = -ENOMEM;
4656                 goto out;
4657         }
4658 
4659         page_start = page_offset(page);
4660         page_end = page_start + PAGE_CACHE_SIZE - 1;
4661 
4662         if (!PageUptodate(page)) {
4663                 ret = btrfs_readpage(NULL, page);
4664                 lock_page(page);
4665                 if (page->mapping != mapping) {
4666                         unlock_page(page);
4667                         page_cache_release(page);
4668                         goto again;
4669                 }
4670                 if (!PageUptodate(page)) {
4671                         ret = -EIO;
4672                         goto out_unlock;
4673                 }
4674         }
4675         wait_on_page_writeback(page);
4676 
4677         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
4678         set_page_extent_mapped(page);
4679 
4680         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4681         if (ordered) {
4682                 unlock_extent_cached(io_tree, page_start, page_end,
4683                                      &cached_state, GFP_NOFS);
4684                 unlock_page(page);
4685                 page_cache_release(page);
4686                 btrfs_start_ordered_extent(inode, ordered, 1);
4687                 btrfs_put_ordered_extent(ordered);
4688                 goto again;
4689         }
4690 
4691         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4692                           EXTENT_DIRTY | EXTENT_DELALLOC |
4693                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4694                           0, 0, &cached_state, GFP_NOFS);
4695 
4696         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4697                                         &cached_state);
4698         if (ret) {
4699                 unlock_extent_cached(io_tree, page_start, page_end,
4700                                      &cached_state, GFP_NOFS);
4701                 goto out_unlock;
4702         }
4703 
4704         if (offset != PAGE_CACHE_SIZE) {
4705                 if (!len)
4706                         len = PAGE_CACHE_SIZE - offset;
4707                 kaddr = kmap(page);
4708                 if (front)
4709                         memset(kaddr, 0, offset);
4710                 else
4711                         memset(kaddr + offset, 0, len);
4712                 flush_dcache_page(page);
4713                 kunmap(page);
4714         }
4715         ClearPageChecked(page);
4716         set_page_dirty(page);
4717         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4718                              GFP_NOFS);
4719 
4720 out_unlock:
4721         if (ret)
4722                 btrfs_delalloc_release_space(inode, page_start,
4723                                              PAGE_CACHE_SIZE);
4724         unlock_page(page);
4725         page_cache_release(page);
4726 out:
4727         return ret;
4728 }
4729 
4730 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4731                              u64 offset, u64 len)
4732 {
4733         struct btrfs_trans_handle *trans;
4734         int ret;
4735 
4736         /*
4737          * Still need to make sure the inode looks like it's been updated so
4738          * that any holes get logged if we fsync.
4739          */
4740         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4741                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4742                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4743                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4744                 return 0;
4745         }
4746 
4747         /*
4748          * 1 - for the one we're dropping
4749          * 1 - for the one we're adding
4750          * 1 - for updating the inode.
4751          */
4752         trans = btrfs_start_transaction(root, 3);
4753         if (IS_ERR(trans))
4754                 return PTR_ERR(trans);
4755 
4756         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4757         if (ret) {
4758                 btrfs_abort_transaction(trans, root, ret);
4759                 btrfs_end_transaction(trans, root);
4760                 return ret;
4761         }
4762 
4763         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4764                                        0, 0, len, 0, len, 0, 0, 0);
4765         if (ret)
4766                 btrfs_abort_transaction(trans, root, ret);
4767         else
4768                 btrfs_update_inode(trans, root, inode);
4769         btrfs_end_transaction(trans, root);
4770         return ret;
4771 }
4772 
4773 /*
4774  * This function puts in dummy file extents for the area we're creating a hole
4775  * for.  So if we are truncating this file to a larger size we need to insert
4776  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4777  * the range between oldsize and size
4778  */
4779 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4780 {
4781         struct btrfs_root *root = BTRFS_I(inode)->root;
4782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4783         struct extent_map *em = NULL;
4784         struct extent_state *cached_state = NULL;
4785         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4786         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4787         u64 block_end = ALIGN(size, root->sectorsize);
4788         u64 last_byte;
4789         u64 cur_offset;
4790         u64 hole_size;
4791         int err = 0;
4792 
4793         /*
4794          * If our size started in the middle of a page we need to zero out the
4795          * rest of the page before we expand the i_size, otherwise we could
4796          * expose stale data.
4797          */
4798         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4799         if (err)
4800                 return err;
4801 
4802         if (size <= hole_start)
4803                 return 0;
4804 
4805         while (1) {
4806                 struct btrfs_ordered_extent *ordered;
4807 
4808                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4809                                  &cached_state);
4810                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4811                                                      block_end - hole_start);
4812                 if (!ordered)
4813                         break;
4814                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4815                                      &cached_state, GFP_NOFS);
4816                 btrfs_start_ordered_extent(inode, ordered, 1);
4817                 btrfs_put_ordered_extent(ordered);
4818         }
4819 
4820         cur_offset = hole_start;
4821         while (1) {
4822                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4823                                 block_end - cur_offset, 0);
4824                 if (IS_ERR(em)) {
4825                         err = PTR_ERR(em);
4826                         em = NULL;
4827                         break;
4828                 }
4829                 last_byte = min(extent_map_end(em), block_end);
4830                 last_byte = ALIGN(last_byte , root->sectorsize);
4831                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4832                         struct extent_map *hole_em;
4833                         hole_size = last_byte - cur_offset;
4834 
4835                         err = maybe_insert_hole(root, inode, cur_offset,
4836                                                 hole_size);
4837                         if (err)
4838                                 break;
4839                         btrfs_drop_extent_cache(inode, cur_offset,
4840                                                 cur_offset + hole_size - 1, 0);
4841                         hole_em = alloc_extent_map();
4842                         if (!hole_em) {
4843                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4844                                         &BTRFS_I(inode)->runtime_flags);
4845                                 goto next;
4846                         }
4847                         hole_em->start = cur_offset;
4848                         hole_em->len = hole_size;
4849                         hole_em->orig_start = cur_offset;
4850 
4851                         hole_em->block_start = EXTENT_MAP_HOLE;
4852                         hole_em->block_len = 0;
4853                         hole_em->orig_block_len = 0;
4854                         hole_em->ram_bytes = hole_size;
4855                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4856                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4857                         hole_em->generation = root->fs_info->generation;
4858 
4859                         while (1) {
4860                                 write_lock(&em_tree->lock);
4861                                 err = add_extent_mapping(em_tree, hole_em, 1);
4862                                 write_unlock(&em_tree->lock);
4863                                 if (err != -EEXIST)
4864                                         break;
4865                                 btrfs_drop_extent_cache(inode, cur_offset,
4866                                                         cur_offset +
4867                                                         hole_size - 1, 0);
4868                         }
4869                         free_extent_map(hole_em);
4870                 }
4871 next:
4872                 free_extent_map(em);
4873                 em = NULL;
4874                 cur_offset = last_byte;
4875                 if (cur_offset >= block_end)
4876                         break;
4877         }
4878         free_extent_map(em);
4879         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4880                              GFP_NOFS);
4881         return err;
4882 }
4883 
4884 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4885 {
4886         struct btrfs_root *root = BTRFS_I(inode)->root;
4887         struct btrfs_trans_handle *trans;
4888         loff_t oldsize = i_size_read(inode);
4889         loff_t newsize = attr->ia_size;
4890         int mask = attr->ia_valid;
4891         int ret;
4892 
4893         /*
4894          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4895          * special case where we need to update the times despite not having
4896          * these flags set.  For all other operations the VFS set these flags
4897          * explicitly if it wants a timestamp update.
4898          */
4899         if (newsize != oldsize) {
4900                 inode_inc_iversion(inode);
4901                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4902                         inode->i_ctime = inode->i_mtime =
4903                                 current_fs_time(inode->i_sb);
4904         }
4905 
4906         if (newsize > oldsize) {
4907                 truncate_pagecache(inode, newsize);
4908                 /*
4909                  * Don't do an expanding truncate while snapshoting is ongoing.
4910                  * This is to ensure the snapshot captures a fully consistent
4911                  * state of this file - if the snapshot captures this expanding
4912                  * truncation, it must capture all writes that happened before
4913                  * this truncation.
4914                  */
4915                 btrfs_wait_for_snapshot_creation(root);
4916                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4917                 if (ret) {
4918                         btrfs_end_write_no_snapshoting(root);
4919                         return ret;
4920                 }
4921 
4922                 trans = btrfs_start_transaction(root, 1);
4923                 if (IS_ERR(trans)) {
4924                         btrfs_end_write_no_snapshoting(root);
4925                         return PTR_ERR(trans);
4926                 }
4927 
4928                 i_size_write(inode, newsize);
4929                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4930                 ret = btrfs_update_inode(trans, root, inode);
4931                 btrfs_end_write_no_snapshoting(root);
4932                 btrfs_end_transaction(trans, root);
4933         } else {
4934 
4935                 /*
4936                  * We're truncating a file that used to have good data down to
4937                  * zero. Make sure it gets into the ordered flush list so that
4938                  * any new writes get down to disk quickly.
4939                  */
4940                 if (newsize == 0)
4941                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4942                                 &BTRFS_I(inode)->runtime_flags);
4943 
4944                 /*
4945                  * 1 for the orphan item we're going to add
4946                  * 1 for the orphan item deletion.
4947                  */
4948                 trans = btrfs_start_transaction(root, 2);
4949                 if (IS_ERR(trans))
4950                         return PTR_ERR(trans);
4951 
4952                 /*
4953                  * We need to do this in case we fail at _any_ point during the
4954                  * actual truncate.  Once we do the truncate_setsize we could
4955                  * invalidate pages which forces any outstanding ordered io to
4956                  * be instantly completed which will give us extents that need
4957                  * to be truncated.  If we fail to get an orphan inode down we
4958                  * could have left over extents that were never meant to live,
4959                  * so we need to garuntee from this point on that everything
4960                  * will be consistent.
4961                  */
4962                 ret = btrfs_orphan_add(trans, inode);
4963                 btrfs_end_transaction(trans, root);
4964                 if (ret)
4965                         return ret;
4966 
4967                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4968                 truncate_setsize(inode, newsize);
4969 
4970                 /* Disable nonlocked read DIO to avoid the end less truncate */
4971                 btrfs_inode_block_unlocked_dio(inode);
4972                 inode_dio_wait(inode);
4973                 btrfs_inode_resume_unlocked_dio(inode);
4974 
4975                 ret = btrfs_truncate(inode);
4976                 if (ret && inode->i_nlink) {
4977                         int err;
4978 
4979                         /*
4980                          * failed to truncate, disk_i_size is only adjusted down
4981                          * as we remove extents, so it should represent the true
4982                          * size of the inode, so reset the in memory size and
4983                          * delete our orphan entry.
4984                          */
4985                         trans = btrfs_join_transaction(root);
4986                         if (IS_ERR(trans)) {
4987                                 btrfs_orphan_del(NULL, inode);
4988                                 return ret;
4989                         }
4990                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4991                         err = btrfs_orphan_del(trans, inode);
4992                         if (err)
4993                                 btrfs_abort_transaction(trans, root, err);
4994                         btrfs_end_transaction(trans, root);
4995                 }
4996         }
4997 
4998         return ret;
4999 }
5000 
5001 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5002 {
5003         struct inode *inode = d_inode(dentry);
5004         struct btrfs_root *root = BTRFS_I(inode)->root;
5005         int err;
5006 
5007         if (btrfs_root_readonly(root))
5008                 return -EROFS;
5009 
5010         err = inode_change_ok(inode, attr);
5011         if (err)
5012                 return err;
5013 
5014         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5015                 err = btrfs_setsize(inode, attr);
5016                 if (err)
5017                         return err;
5018         }
5019 
5020         if (attr->ia_valid) {
5021                 setattr_copy(inode, attr);
5022                 inode_inc_iversion(inode);
5023                 err = btrfs_dirty_inode(inode);
5024 
5025                 if (!err && attr->ia_valid & ATTR_MODE)
5026                         err = posix_acl_chmod(inode, inode->i_mode);
5027         }
5028 
5029         return err;
5030 }
5031 
5032 /*
5033  * While truncating the inode pages during eviction, we get the VFS calling
5034  * btrfs_invalidatepage() against each page of the inode. This is slow because
5035  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5036  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5037  * extent_state structures over and over, wasting lots of time.
5038  *
5039  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5040  * those expensive operations on a per page basis and do only the ordered io
5041  * finishing, while we release here the extent_map and extent_state structures,
5042  * without the excessive merging and splitting.
5043  */
5044 static void evict_inode_truncate_pages(struct inode *inode)
5045 {
5046         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5047         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5048         struct rb_node *node;
5049 
5050         ASSERT(inode->i_state & I_FREEING);
5051         truncate_inode_pages_final(&inode->i_data);
5052 
5053         write_lock(&map_tree->lock);
5054         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5055                 struct extent_map *em;
5056 
5057                 node = rb_first(&map_tree->map);
5058                 em = rb_entry(node, struct extent_map, rb_node);
5059                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5060                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5061                 remove_extent_mapping(map_tree, em);
5062                 free_extent_map(em);
5063                 if (need_resched()) {
5064                         write_unlock(&map_tree->lock);
5065                         cond_resched();
5066                         write_lock(&map_tree->lock);
5067                 }
5068         }
5069         write_unlock(&map_tree->lock);
5070 
5071         /*
5072          * Keep looping until we have no more ranges in the io tree.
5073          * We can have ongoing bios started by readpages (called from readahead)
5074          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5075          * still in progress (unlocked the pages in the bio but did not yet
5076          * unlocked the ranges in the io tree). Therefore this means some
5077          * ranges can still be locked and eviction started because before
5078          * submitting those bios, which are executed by a separate task (work
5079          * queue kthread), inode references (inode->i_count) were not taken
5080          * (which would be dropped in the end io callback of each bio).
5081          * Therefore here we effectively end up waiting for those bios and
5082          * anyone else holding locked ranges without having bumped the inode's
5083          * reference count - if we don't do it, when they access the inode's
5084          * io_tree to unlock a range it may be too late, leading to an
5085          * use-after-free issue.
5086          */
5087         spin_lock(&io_tree->lock);
5088         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5089                 struct extent_state *state;
5090                 struct extent_state *cached_state = NULL;
5091                 u64 start;
5092                 u64 end;
5093 
5094                 node = rb_first(&io_tree->state);
5095                 state = rb_entry(node, struct extent_state, rb_node);
5096                 start = state->start;
5097                 end = state->end;
5098                 spin_unlock(&io_tree->lock);
5099 
5100                 lock_extent_bits(io_tree, start, end, &cached_state);
5101 
5102                 /*
5103                  * If still has DELALLOC flag, the extent didn't reach disk,
5104                  * and its reserved space won't be freed by delayed_ref.
5105                  * So we need to free its reserved space here.
5106                  * (Refer to comment in btrfs_invalidatepage, case 2)
5107                  *
5108                  * Note, end is the bytenr of last byte, so we need + 1 here.
5109                  */
5110                 if (state->state & EXTENT_DELALLOC)
5111                         btrfs_qgroup_free_data(inode, start, end - start + 1);
5112 
5113                 clear_extent_bit(io_tree, start, end,
5114                                  EXTENT_LOCKED | EXTENT_DIRTY |
5115                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5116                                  EXTENT_DEFRAG, 1, 1,
5117                                  &cached_state, GFP_NOFS);
5118 
5119                 cond_resched();
5120                 spin_lock(&io_tree->lock);
5121         }
5122         spin_unlock(&io_tree->lock);
5123 }
5124 
5125 void btrfs_evict_inode(struct inode *inode)
5126 {
5127         struct btrfs_trans_handle *trans;
5128         struct btrfs_root *root = BTRFS_I(inode)->root;
5129         struct btrfs_block_rsv *rsv, *global_rsv;
5130         int steal_from_global = 0;
5131         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5132         int ret;
5133 
5134         trace_btrfs_inode_evict(inode);
5135 
5136         evict_inode_truncate_pages(inode);
5137 
5138         if (inode->i_nlink &&
5139             ((btrfs_root_refs(&root->root_item) != 0 &&
5140               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5141              btrfs_is_free_space_inode(inode)))
5142                 goto no_delete;
5143 
5144         if (is_bad_inode(inode)) {
5145                 btrfs_orphan_del(NULL, inode);
5146                 goto no_delete;
5147         }
5148         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5149         if (!special_file(inode->i_mode))
5150                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5151 
5152         btrfs_free_io_failure_record(inode, 0, (u64)-1);
5153 
5154         if (root->fs_info->log_root_recovering) {
5155                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5156                                  &BTRFS_I(inode)->runtime_flags));
5157                 goto no_delete;
5158         }
5159 
5160         if (inode->i_nlink > 0) {
5161                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5162                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5163                 goto no_delete;
5164         }
5165 
5166         ret = btrfs_commit_inode_delayed_inode(inode);
5167         if (ret) {
5168                 btrfs_orphan_del(NULL, inode);
5169                 goto no_delete;
5170         }
5171 
5172         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5173         if (!rsv) {
5174                 btrfs_orphan_del(NULL, inode);
5175                 goto no_delete;
5176         }
5177         rsv->size = min_size;
5178         rsv->failfast = 1;
5179         global_rsv = &root->fs_info->global_block_rsv;
5180 
5181         btrfs_i_size_write(inode, 0);
5182 
5183         /*
5184          * This is a bit simpler than btrfs_truncate since we've already
5185          * reserved our space for our orphan item in the unlink, so we just
5186          * need to reserve some slack space in case we add bytes and update
5187          * inode item when doing the truncate.
5188          */
5189         while (1) {
5190                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5191                                              BTRFS_RESERVE_FLUSH_LIMIT);
5192 
5193                 /*
5194                  * Try and steal from the global reserve since we will
5195                  * likely not use this space anyway, we want to try as
5196                  * hard as possible to get this to work.
5197                  */
5198                 if (ret)
5199                         steal_from_global++;
5200                 else
5201                         steal_from_global = 0;
5202                 ret = 0;
5203 
5204                 /*
5205                  * steal_from_global == 0: we reserved stuff, hooray!
5206                  * steal_from_global == 1: we didn't reserve stuff, boo!
5207                  * steal_from_global == 2: we've committed, still not a lot of
5208                  * room but maybe we'll have room in the global reserve this
5209                  * time.
5210                  * steal_from_global == 3: abandon all hope!
5211                  */
5212                 if (steal_from_global > 2) {
5213                         btrfs_warn(root->fs_info,
5214                                 "Could not get space for a delete, will truncate on mount %d",
5215                                 ret);
5216                         btrfs_orphan_del(NULL, inode);
5217                         btrfs_free_block_rsv(root, rsv);
5218                         goto no_delete;
5219                 }
5220 
5221                 trans = btrfs_join_transaction(root);
5222                 if (IS_ERR(trans)) {
5223                         btrfs_orphan_del(NULL, inode);
5224                         btrfs_free_block_rsv(root, rsv);
5225                         goto no_delete;
5226                 }
5227 
5228                 /*
5229                  * We can't just steal from the global reserve, we need tomake
5230                  * sure there is room to do it, if not we need to commit and try
5231                  * again.
5232                  */
5233                 if (steal_from_global) {
5234                         if (!btrfs_check_space_for_delayed_refs(trans, root))
5235                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5236                                                               min_size);
5237                         else
5238                                 ret = -ENOSPC;
5239                 }
5240 
5241                 /*
5242                  * Couldn't steal from the global reserve, we have too much
5243                  * pending stuff built up, commit the transaction and try it
5244                  * again.
5245                  */
5246                 if (ret) {
5247                         ret = btrfs_commit_transaction(trans, root);
5248                         if (ret) {
5249                                 btrfs_orphan_del(NULL, inode);
5250                                 btrfs_free_block_rsv(root, rsv);
5251                                 goto no_delete;
5252                         }
5253                         continue;
5254                 } else {
5255                         steal_from_global = 0;
5256                 }
5257 
5258                 trans->block_rsv = rsv;
5259 
5260                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5261                 if (ret != -ENOSPC && ret != -EAGAIN)
5262                         break;
5263 
5264                 trans->block_rsv = &root->fs_info->trans_block_rsv;
5265                 btrfs_end_transaction(trans, root);
5266                 trans = NULL;
5267                 btrfs_btree_balance_dirty(root);
5268         }
5269 
5270         btrfs_free_block_rsv(root, rsv);
5271 
5272         /*
5273          * Errors here aren't a big deal, it just means we leave orphan items
5274          * in the tree.  They will be cleaned up on the next mount.
5275          */
5276         if (ret == 0) {
5277                 trans->block_rsv = root->orphan_block_rsv;
5278                 btrfs_orphan_del(trans, inode);
5279         } else {
5280                 btrfs_orphan_del(NULL, inode);
5281         }
5282 
5283         trans->block_rsv = &root->fs_info->trans_block_rsv;
5284         if (!(root == root->fs_info->tree_root ||
5285               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5286                 btrfs_return_ino(root, btrfs_ino(inode));
5287 
5288         btrfs_end_transaction(trans, root);
5289         btrfs_btree_balance_dirty(root);
5290 no_delete:
5291         btrfs_remove_delayed_node(inode);
5292         clear_inode(inode);
5293 }
5294 
5295 /*
5296  * this returns the key found in the dir entry in the location pointer.
5297  * If no dir entries were found, location->objectid is 0.
5298  */
5299 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5300                                struct btrfs_key *location)
5301 {
5302         const char *name = dentry->d_name.name;
5303         int namelen = dentry->d_name.len;
5304         struct btrfs_dir_item *di;
5305         struct btrfs_path *path;
5306         struct btrfs_root *root = BTRFS_I(dir)->root;
5307         int ret = 0;
5308 
5309         path = btrfs_alloc_path();
5310         if (!path)
5311                 return -ENOMEM;
5312 
5313         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5314                                     namelen, 0);
5315         if (IS_ERR(di))
5316                 ret = PTR_ERR(di);
5317 
5318         if (IS_ERR_OR_NULL(di))
5319                 goto out_err;
5320 
5321         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5322 out:
5323         btrfs_free_path(path);
5324         return ret;
5325 out_err:
5326         location->objectid = 0;
5327         goto out;
5328 }
5329 
5330 /*
5331  * when we hit a tree root in a directory, the btrfs part of the inode
5332  * needs to be changed to reflect the root directory of the tree root.  This
5333  * is kind of like crossing a mount point.
5334  */
5335 static int fixup_tree_root_location(struct btrfs_root *root,
5336                                     struct inode *dir,
5337                                     struct dentry *dentry,
5338                                     struct btrfs_key *location,
5339                                     struct btrfs_root **sub_root)
5340 {
5341         struct btrfs_path *path;
5342         struct btrfs_root *new_root;
5343         struct btrfs_root_ref *ref;
5344         struct extent_buffer *leaf;
5345         struct btrfs_key key;
5346         int ret;
5347         int err = 0;
5348 
5349         path = btrfs_alloc_path();
5350         if (!path) {
5351                 err = -ENOMEM;
5352                 goto out;
5353         }
5354 
5355         err = -ENOENT;
5356         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5357         key.type = BTRFS_ROOT_REF_KEY;
5358         key.offset = location->objectid;
5359 
5360         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5361                                 0, 0);
5362         if (ret) {
5363                 if (ret < 0)
5364                         err = ret;
5365                 goto out;
5366         }
5367 
5368         leaf = path->nodes[0];
5369         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5370         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5371             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5372                 goto out;
5373 
5374         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5375                                    (unsigned long)(ref + 1),
5376                                    dentry->d_name.len);
5377         if (ret)
5378                 goto out;
5379 
5380         btrfs_release_path(path);
5381 
5382         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5383         if (IS_ERR(new_root)) {
5384                 err = PTR_ERR(new_root);
5385                 goto out;
5386         }
5387 
5388         *sub_root = new_root;
5389         location->objectid = btrfs_root_dirid(&new_root->root_item);
5390         location->type = BTRFS_INODE_ITEM_KEY;
5391         location->offset = 0;
5392         err = 0;
5393 out:
5394         btrfs_free_path(path);
5395         return err;
5396 }
5397 
5398 static void inode_tree_add(struct inode *inode)
5399 {
5400         struct btrfs_root *root = BTRFS_I(inode)->root;
5401         struct btrfs_inode *entry;
5402         struct rb_node **p;
5403         struct rb_node *parent;
5404         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5405         u64 ino = btrfs_ino(inode);
5406 
5407         if (inode_unhashed(inode))
5408                 return;
5409         parent = NULL;
5410         spin_lock(&root->inode_lock);
5411         p = &root->inode_tree.rb_node;
5412         while (*p) {
5413                 parent = *p;
5414                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5415 
5416                 if (ino < btrfs_ino(&entry->vfs_inode))
5417                         p = &parent->rb_left;
5418                 else if (ino > btrfs_ino(&entry->vfs_inode))
5419                         p = &parent->rb_right;
5420                 else {
5421                         WARN_ON(!(entry->vfs_inode.i_state &
5422                                   (I_WILL_FREE | I_FREEING)));
5423                         rb_replace_node(parent, new, &root->inode_tree);
5424                         RB_CLEAR_NODE(parent);
5425                         spin_unlock(&root->inode_lock);
5426                         return;
5427                 }
5428         }
5429         rb_link_node(new, parent, p);
5430         rb_insert_color(new, &root->inode_tree);
5431         spin_unlock(&root->inode_lock);
5432 }
5433 
5434 static void inode_tree_del(struct inode *inode)
5435 {
5436         struct btrfs_root *root = BTRFS_I(inode)->root;
5437         int empty = 0;
5438 
5439         spin_lock(&root->inode_lock);
5440         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5441                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5442                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5443                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5444         }
5445         spin_unlock(&root->inode_lock);
5446 
5447         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5448                 synchronize_srcu(&root->fs_info->subvol_srcu);
5449                 spin_lock(&root->inode_lock);
5450                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5451                 spin_unlock(&root->inode_lock);
5452                 if (empty)
5453                         btrfs_add_dead_root(root);
5454         }
5455 }
5456 
5457 void btrfs_invalidate_inodes(struct btrfs_root *root)
5458 {
5459         struct rb_node *node;
5460         struct rb_node *prev;
5461         struct btrfs_inode *entry;
5462         struct inode *inode;
5463         u64 objectid = 0;
5464 
5465         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5466                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5467 
5468         spin_lock(&root->inode_lock);
5469 again:
5470         node = root->inode_tree.rb_node;
5471         prev = NULL;
5472         while (node) {
5473                 prev = node;
5474                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5475 
5476                 if (objectid < btrfs_ino(&entry->vfs_inode))
5477                         node = node->rb_left;
5478                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5479                         node = node->rb_right;
5480                 else
5481                         break;
5482         }
5483         if (!node) {
5484                 while (prev) {
5485                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5486                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5487                                 node = prev;
5488                                 break;
5489                         }
5490                         prev = rb_next(prev);
5491                 }
5492         }
5493         while (node) {
5494                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5495                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5496                 inode = igrab(&entry->vfs_inode);
5497                 if (inode) {
5498                         spin_unlock(&root->inode_lock);
5499                         if (atomic_read(&inode->i_count) > 1)
5500                                 d_prune_aliases(inode);
5501                         /*
5502                          * btrfs_drop_inode will have it removed from
5503                          * the inode cache when its usage count
5504                          * hits zero.
5505                          */
5506                         iput(inode);
5507                         cond_resched();
5508                         spin_lock(&root->inode_lock);
5509                         goto again;
5510                 }
5511 
5512                 if (cond_resched_lock(&root->inode_lock))
5513                         goto again;
5514 
5515                 node = rb_next(node);
5516         }
5517         spin_unlock(&root->inode_lock);
5518 }
5519 
5520 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5521 {
5522         struct btrfs_iget_args *args = p;
5523         inode->i_ino = args->location->objectid;
5524         memcpy(&BTRFS_I(inode)->location, args->location,
5525                sizeof(*args->location));
5526         BTRFS_I(inode)->root = args->root;
5527         return 0;
5528 }
5529 
5530 static int btrfs_find_actor(struct inode *inode, void *opaque)
5531 {
5532         struct btrfs_iget_args *args = opaque;
5533         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5534                 args->root == BTRFS_I(inode)->root;
5535 }
5536 
5537 static struct inode *btrfs_iget_locked(struct super_block *s,
5538                                        struct btrfs_key *location,
5539                                        struct btrfs_root *root)
5540 {
5541         struct inode *inode;
5542         struct btrfs_iget_args args;
5543         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5544 
5545         args.location = location;
5546         args.root = root;
5547 
5548         inode = iget5_locked(s, hashval, btrfs_find_actor,
5549                              btrfs_init_locked_inode,
5550                              (void *)&args);
5551         return inode;
5552 }
5553 
5554 /* Get an inode object given its location and corresponding root.
5555  * Returns in *is_new if the inode was read from disk
5556  */
5557 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5558                          struct btrfs_root *root, int *new)
5559 {
5560         struct inode *inode;
5561 
5562         inode = btrfs_iget_locked(s, location, root);
5563         if (!inode)
5564                 return ERR_PTR(-ENOMEM);
5565 
5566         if (inode->i_state & I_NEW) {
5567                 btrfs_read_locked_inode(inode);
5568                 if (!is_bad_inode(inode)) {
5569                         inode_tree_add(inode);
5570                         unlock_new_inode(inode);
5571                         if (new)
5572                                 *new = 1;
5573                 } else {
5574                         unlock_new_inode(inode);
5575                         iput(inode);
5576                         inode = ERR_PTR(-ESTALE);
5577                 }
5578         }
5579 
5580         return inode;
5581 }
5582 
5583 static struct inode *new_simple_dir(struct super_block *s,
5584                                     struct btrfs_key *key,
5585                                     struct btrfs_root *root)
5586 {
5587         struct inode *inode = new_inode(s);
5588 
5589         if (!inode)
5590                 return ERR_PTR(-ENOMEM);
5591 
5592         BTRFS_I(inode)->root = root;
5593         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5594         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5595 
5596         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5597         inode->i_op = &btrfs_dir_ro_inode_operations;
5598         inode->i_fop = &simple_dir_operations;
5599         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5600         inode->i_mtime = CURRENT_TIME;
5601         inode->i_atime = inode->i_mtime;
5602         inode->i_ctime = inode->i_mtime;
5603         BTRFS_I(inode)->i_otime = inode->i_mtime;
5604 
5605         return inode;
5606 }
5607 
5608 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5609 {
5610         struct inode *inode;
5611         struct btrfs_root *root = BTRFS_I(dir)->root;
5612         struct btrfs_root *sub_root = root;
5613         struct btrfs_key location;
5614         int index;
5615         int ret = 0;
5616 
5617         if (dentry->d_name.len > BTRFS_NAME_LEN)
5618                 return ERR_PTR(-ENAMETOOLONG);
5619 
5620         ret = btrfs_inode_by_name(dir, dentry, &location);
5621         if (ret < 0)
5622                 return ERR_PTR(ret);
5623 
5624         if (location.objectid == 0)
5625                 return ERR_PTR(-ENOENT);
5626 
5627         if (location.type == BTRFS_INODE_ITEM_KEY) {
5628                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5629                 return inode;
5630         }
5631 
5632         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5633 
5634         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5635         ret = fixup_tree_root_location(root, dir, dentry,
5636                                        &location, &sub_root);
5637         if (ret < 0) {
5638                 if (ret != -ENOENT)
5639                         inode = ERR_PTR(ret);
5640                 else
5641                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5642         } else {
5643                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5644         }
5645         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5646 
5647         if (!IS_ERR(inode) && root != sub_root) {
5648                 down_read(&root->fs_info->cleanup_work_sem);
5649                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5650                         ret = btrfs_orphan_cleanup(sub_root);
5651                 up_read(&root->fs_info->cleanup_work_sem);
5652                 if (ret) {
5653                         iput(inode);
5654                         inode = ERR_PTR(ret);
5655                 }
5656         }
5657 
5658         return inode;
5659 }
5660 
5661 static int btrfs_dentry_delete(const struct dentry *dentry)
5662 {
5663         struct btrfs_root *root;
5664         struct inode *inode = d_inode(dentry);
5665 
5666         if (!inode && !IS_ROOT(dentry))
5667                 inode = d_inode(dentry->d_parent);
5668 
5669         if (inode) {
5670                 root = BTRFS_I(inode)->root;
5671                 if (btrfs_root_refs(&root->root_item) == 0)
5672                         return 1;
5673 
5674                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5675                         return 1;
5676         }
5677         return 0;
5678 }
5679 
5680 static void btrfs_dentry_release(struct dentry *dentry)
5681 {
5682         kfree(dentry->d_fsdata);
5683 }
5684 
5685 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5686                                    unsigned int flags)
5687 {
5688         struct inode *inode;
5689 
5690         inode = btrfs_lookup_dentry(dir, dentry);
5691         if (IS_ERR(inode)) {
5692                 if (PTR_ERR(inode) == -ENOENT)
5693                         inode = NULL;
5694                 else
5695                         return ERR_CAST(inode);
5696         }
5697 
5698         return d_splice_alias(inode, dentry);
5699 }
5700 
5701 unsigned char btrfs_filetype_table[] = {
5702         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5703 };
5704 
5705 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5706 {
5707         struct inode *inode = file_inode(file);
5708         struct btrfs_root *root = BTRFS_I(inode)->root;
5709         struct btrfs_item *item;
5710         struct btrfs_dir_item *di;
5711         struct btrfs_key key;
5712         struct btrfs_key found_key;
5713         struct btrfs_path *path;
5714         struct list_head ins_list;
5715         struct list_head del_list;
5716         int ret;
5717         struct extent_buffer *leaf;
5718         int slot;
5719         unsigned char d_type;
5720         int over = 0;
5721         u32 di_cur;
5722         u32 di_total;
5723         u32 di_len;
5724         int key_type = BTRFS_DIR_INDEX_KEY;
5725         char tmp_name[32];
5726         char *name_ptr;
5727         int name_len;
5728         int is_curr = 0;        /* ctx->pos points to the current index? */
5729         bool emitted;
5730 
5731         /* FIXME, use a real flag for deciding about the key type */
5732         if (root->fs_info->tree_root == root)
5733                 key_type = BTRFS_DIR_ITEM_KEY;
5734 
5735         if (!dir_emit_dots(file, ctx))
5736                 return 0;
5737 
5738         path = btrfs_alloc_path();
5739         if (!path)
5740                 return -ENOMEM;
5741 
5742         path->reada = READA_FORWARD;
5743 
5744         if (key_type == BTRFS_DIR_INDEX_KEY) {
5745                 INIT_LIST_HEAD(&ins_list);
5746                 INIT_LIST_HEAD(&del_list);
5747                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5748         }
5749 
5750         key.type = key_type;
5751         key.offset = ctx->pos;
5752         key.objectid = btrfs_ino(inode);
5753 
5754         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5755         if (ret < 0)
5756                 goto err;
5757 
5758         emitted = false;
5759         while (1) {
5760                 leaf = path->nodes[0];
5761                 slot = path->slots[0];
5762                 if (slot >= btrfs_header_nritems(leaf)) {
5763                         ret = btrfs_next_leaf(root, path);
5764                         if (ret < 0)
5765                                 goto err;
5766                         else if (ret > 0)
5767                                 break;
5768                         continue;
5769                 }
5770 
5771                 item = btrfs_item_nr(slot);
5772                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5773 
5774                 if (found_key.objectid != key.objectid)
5775                         break;
5776                 if (found_key.type != key_type)
5777                         break;
5778                 if (found_key.offset < ctx->pos)
5779                         goto next;
5780                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5781                     btrfs_should_delete_dir_index(&del_list,
5782                                                   found_key.offset))
5783                         goto next;
5784 
5785                 ctx->pos = found_key.offset;
5786                 is_curr = 1;
5787 
5788                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5789                 di_cur = 0;
5790                 di_total = btrfs_item_size(leaf, item);
5791 
5792                 while (di_cur < di_total) {
5793                         struct btrfs_key location;
5794 
5795                         if (verify_dir_item(root, leaf, di))
5796                                 break;
5797 
5798                         name_len = btrfs_dir_name_len(leaf, di);
5799                         if (name_len <= sizeof(tmp_name)) {
5800                                 name_ptr = tmp_name;
5801                         } else {
5802                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5803                                 if (!name_ptr) {
5804                                         ret = -ENOMEM;
5805