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

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

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.3 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.30 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.107 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.164 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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_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_SHIFT);
212                 btrfs_set_file_extent_compression(leaf, ei, 0);
213                 kaddr = kmap_atomic(page);
214                 offset = start & (PAGE_SIZE - 1);
215                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216                 kunmap_atomic(kaddr);
217                 put_page(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 > root->sectorsize ||
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_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_SHIFT) - (start >> PAGE_SHIFT) + 1;
439         nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_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 doesn'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_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_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_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                         put_page(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                 put_page(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                 put_page(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                 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
828 
829                 /*
830                  * clear dirty, set writeback and unlock the pages.
831                  */
832                 extent_clear_unlock_delalloc(inode, async_extent->start,
833                                 async_extent->start +
834                                 async_extent->ram_size - 1,
835                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
836                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
837                                 PAGE_SET_WRITEBACK);
838                 ret = btrfs_submit_compressed_write(inode,
839                                     async_extent->start,
840                                     async_extent->ram_size,
841                                     ins.objectid,
842                                     ins.offset, async_extent->pages,
843                                     async_extent->nr_pages);
844                 if (ret) {
845                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846                         struct page *p = async_extent->pages[0];
847                         const u64 start = async_extent->start;
848                         const u64 end = start + async_extent->ram_size - 1;
849 
850                         p->mapping = inode->i_mapping;
851                         tree->ops->writepage_end_io_hook(p, start, end,
852                                                          NULL, 0);
853                         p->mapping = NULL;
854                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
855                                                      PAGE_END_WRITEBACK |
856                                                      PAGE_SET_ERROR);
857                         free_async_extent_pages(async_extent);
858                 }
859                 alloc_hint = ins.objectid + ins.offset;
860                 kfree(async_extent);
861                 cond_resched();
862         }
863         return;
864 out_free_reserve:
865         btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
866         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
867 out_free:
868         extent_clear_unlock_delalloc(inode, async_extent->start,
869                                      async_extent->start +
870                                      async_extent->ram_size - 1,
871                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
872                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
873                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
874                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
875                                      PAGE_SET_ERROR);
876         free_async_extent_pages(async_extent);
877         kfree(async_extent);
878         goto again;
879 }
880 
881 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882                                       u64 num_bytes)
883 {
884         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885         struct extent_map *em;
886         u64 alloc_hint = 0;
887 
888         read_lock(&em_tree->lock);
889         em = search_extent_mapping(em_tree, start, num_bytes);
890         if (em) {
891                 /*
892                  * if block start isn't an actual block number then find the
893                  * first block in this inode and use that as a hint.  If that
894                  * block is also bogus then just don't worry about it.
895                  */
896                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
897                         free_extent_map(em);
898                         em = search_extent_mapping(em_tree, 0, 0);
899                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
900                                 alloc_hint = em->block_start;
901                         if (em)
902                                 free_extent_map(em);
903                 } else {
904                         alloc_hint = em->block_start;
905                         free_extent_map(em);
906                 }
907         }
908         read_unlock(&em_tree->lock);
909 
910         return alloc_hint;
911 }
912 
913 /*
914  * when extent_io.c finds a delayed allocation range in the file,
915  * the call backs end up in this code.  The basic idea is to
916  * allocate extents on disk for the range, and create ordered data structs
917  * in ram to track those extents.
918  *
919  * locked_page is the page that writepage had locked already.  We use
920  * it to make sure we don't do extra locks or unlocks.
921  *
922  * *page_started is set to one if we unlock locked_page and do everything
923  * required to start IO on it.  It may be clean and already done with
924  * IO when we return.
925  */
926 static noinline int cow_file_range(struct inode *inode,
927                                    struct page *locked_page,
928                                    u64 start, u64 end, int *page_started,
929                                    unsigned long *nr_written,
930                                    int unlock)
931 {
932         struct btrfs_root *root = BTRFS_I(inode)->root;
933         u64 alloc_hint = 0;
934         u64 num_bytes;
935         unsigned long ram_size;
936         u64 disk_num_bytes;
937         u64 cur_alloc_size;
938         u64 blocksize = root->sectorsize;
939         struct btrfs_key ins;
940         struct extent_map *em;
941         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942         int ret = 0;
943 
944         if (btrfs_is_free_space_inode(inode)) {
945                 WARN_ON_ONCE(1);
946                 ret = -EINVAL;
947                 goto out_unlock;
948         }
949 
950         num_bytes = ALIGN(end - start + 1, blocksize);
951         num_bytes = max(blocksize,  num_bytes);
952         disk_num_bytes = num_bytes;
953 
954         /* if this is a small write inside eof, kick off defrag */
955         if (num_bytes < SZ_64K &&
956             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
957                 btrfs_add_inode_defrag(NULL, inode);
958 
959         if (start == 0) {
960                 /* lets try to make an inline extent */
961                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962                                             NULL);
963                 if (ret == 0) {
964                         extent_clear_unlock_delalloc(inode, start, end, NULL,
965                                      EXTENT_LOCKED | EXTENT_DELALLOC |
966                                      EXTENT_DEFRAG, PAGE_UNLOCK |
967                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968                                      PAGE_END_WRITEBACK);
969 
970                         *nr_written = *nr_written +
971                              (end - start + PAGE_SIZE) / PAGE_SIZE;
972                         *page_started = 1;
973                         goto out;
974                 } else if (ret < 0) {
975                         goto out_unlock;
976                 }
977         }
978 
979         BUG_ON(disk_num_bytes >
980                btrfs_super_total_bytes(root->fs_info->super_copy));
981 
982         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
983         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
984 
985         while (disk_num_bytes > 0) {
986                 unsigned long op;
987 
988                 cur_alloc_size = disk_num_bytes;
989                 ret = btrfs_reserve_extent(root, cur_alloc_size,
990                                            root->sectorsize, 0, alloc_hint,
991                                            &ins, 1, 1);
992                 if (ret < 0)
993                         goto out_unlock;
994 
995                 em = alloc_extent_map();
996                 if (!em) {
997                         ret = -ENOMEM;
998                         goto out_reserve;
999                 }
1000                 em->start = start;
1001                 em->orig_start = em->start;
1002                 ram_size = ins.offset;
1003                 em->len = ins.offset;
1004                 em->mod_start = em->start;
1005                 em->mod_len = em->len;
1006 
1007                 em->block_start = ins.objectid;
1008                 em->block_len = ins.offset;
1009                 em->orig_block_len = ins.offset;
1010                 em->ram_bytes = ram_size;
1011                 em->bdev = root->fs_info->fs_devices->latest_bdev;
1012                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1013                 em->generation = -1;
1014 
1015                 while (1) {
1016                         write_lock(&em_tree->lock);
1017                         ret = add_extent_mapping(em_tree, em, 1);
1018                         write_unlock(&em_tree->lock);
1019                         if (ret != -EEXIST) {
1020                                 free_extent_map(em);
1021                                 break;
1022                         }
1023                         btrfs_drop_extent_cache(inode, start,
1024                                                 start + ram_size - 1, 0);
1025                 }
1026                 if (ret)
1027                         goto out_reserve;
1028 
1029                 cur_alloc_size = ins.offset;
1030                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1031                                                ram_size, cur_alloc_size, 0);
1032                 if (ret)
1033                         goto out_drop_extent_cache;
1034 
1035                 if (root->root_key.objectid ==
1036                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1037                         ret = btrfs_reloc_clone_csums(inode, start,
1038                                                       cur_alloc_size);
1039                         if (ret)
1040                                 goto out_drop_extent_cache;
1041                 }
1042 
1043                 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1044 
1045                 if (disk_num_bytes < cur_alloc_size)
1046                         break;
1047 
1048                 /* we're not doing compressed IO, don't unlock the first
1049                  * page (which the caller expects to stay locked), don't
1050                  * clear any dirty bits and don't set any writeback bits
1051                  *
1052                  * Do set the Private2 bit so we know this page was properly
1053                  * setup for writepage
1054                  */
1055                 op = unlock ? PAGE_UNLOCK : 0;
1056                 op |= PAGE_SET_PRIVATE2;
1057 
1058                 extent_clear_unlock_delalloc(inode, start,
1059                                              start + ram_size - 1, locked_page,
1060                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1061                                              op);
1062                 disk_num_bytes -= cur_alloc_size;
1063                 num_bytes -= cur_alloc_size;
1064                 alloc_hint = ins.objectid + ins.offset;
1065                 start += cur_alloc_size;
1066         }
1067 out:
1068         return ret;
1069 
1070 out_drop_extent_cache:
1071         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1072 out_reserve:
1073         btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1074         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1075 out_unlock:
1076         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1077                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1078                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1079                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1080                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1081         goto out;
1082 }
1083 
1084 /*
1085  * work queue call back to started compression on a file and pages
1086  */
1087 static noinline void async_cow_start(struct btrfs_work *work)
1088 {
1089         struct async_cow *async_cow;
1090         int num_added = 0;
1091         async_cow = container_of(work, struct async_cow, work);
1092 
1093         compress_file_range(async_cow->inode, async_cow->locked_page,
1094                             async_cow->start, async_cow->end, async_cow,
1095                             &num_added);
1096         if (num_added == 0) {
1097                 btrfs_add_delayed_iput(async_cow->inode);
1098                 async_cow->inode = NULL;
1099         }
1100 }
1101 
1102 /*
1103  * work queue call back to submit previously compressed pages
1104  */
1105 static noinline void async_cow_submit(struct btrfs_work *work)
1106 {
1107         struct async_cow *async_cow;
1108         struct btrfs_root *root;
1109         unsigned long nr_pages;
1110 
1111         async_cow = container_of(work, struct async_cow, work);
1112 
1113         root = async_cow->root;
1114         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1115                 PAGE_SHIFT;
1116 
1117         /*
1118          * atomic_sub_return implies a barrier for waitqueue_active
1119          */
1120         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1121             5 * SZ_1M &&
1122             waitqueue_active(&root->fs_info->async_submit_wait))
1123                 wake_up(&root->fs_info->async_submit_wait);
1124 
1125         if (async_cow->inode)
1126                 submit_compressed_extents(async_cow->inode, async_cow);
1127 }
1128 
1129 static noinline void async_cow_free(struct btrfs_work *work)
1130 {
1131         struct async_cow *async_cow;
1132         async_cow = container_of(work, struct async_cow, work);
1133         if (async_cow->inode)
1134                 btrfs_add_delayed_iput(async_cow->inode);
1135         kfree(async_cow);
1136 }
1137 
1138 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1139                                 u64 start, u64 end, int *page_started,
1140                                 unsigned long *nr_written)
1141 {
1142         struct async_cow *async_cow;
1143         struct btrfs_root *root = BTRFS_I(inode)->root;
1144         unsigned long nr_pages;
1145         u64 cur_end;
1146         int limit = 10 * SZ_1M;
1147 
1148         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1149                          1, 0, NULL, GFP_NOFS);
1150         while (start < end) {
1151                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1152                 BUG_ON(!async_cow); /* -ENOMEM */
1153                 async_cow->inode = igrab(inode);
1154                 async_cow->root = root;
1155                 async_cow->locked_page = locked_page;
1156                 async_cow->start = start;
1157 
1158                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1159                     !btrfs_test_opt(root, FORCE_COMPRESS))
1160                         cur_end = end;
1161                 else
1162                         cur_end = min(end, start + SZ_512K - 1);
1163 
1164                 async_cow->end = cur_end;
1165                 INIT_LIST_HEAD(&async_cow->extents);
1166 
1167                 btrfs_init_work(&async_cow->work,
1168                                 btrfs_delalloc_helper,
1169                                 async_cow_start, async_cow_submit,
1170                                 async_cow_free);
1171 
1172                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1173                         PAGE_SHIFT;
1174                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1175 
1176                 btrfs_queue_work(root->fs_info->delalloc_workers,
1177                                  &async_cow->work);
1178 
1179                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1180                         wait_event(root->fs_info->async_submit_wait,
1181                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1182                             limit));
1183                 }
1184 
1185                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1186                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1187                         wait_event(root->fs_info->async_submit_wait,
1188                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1189                            0));
1190                 }
1191 
1192                 *nr_written += nr_pages;
1193                 start = cur_end + 1;
1194         }
1195         *page_started = 1;
1196         return 0;
1197 }
1198 
1199 static noinline int csum_exist_in_range(struct btrfs_root *root,
1200                                         u64 bytenr, u64 num_bytes)
1201 {
1202         int ret;
1203         struct btrfs_ordered_sum *sums;
1204         LIST_HEAD(list);
1205 
1206         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1207                                        bytenr + num_bytes - 1, &list, 0);
1208         if (ret == 0 && list_empty(&list))
1209                 return 0;
1210 
1211         while (!list_empty(&list)) {
1212                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1213                 list_del(&sums->list);
1214                 kfree(sums);
1215         }
1216         return 1;
1217 }
1218 
1219 /*
1220  * when nowcow writeback call back.  This checks for snapshots or COW copies
1221  * of the extents that exist in the file, and COWs the file as required.
1222  *
1223  * If no cow copies or snapshots exist, we write directly to the existing
1224  * blocks on disk
1225  */
1226 static noinline int run_delalloc_nocow(struct inode *inode,
1227                                        struct page *locked_page,
1228                               u64 start, u64 end, int *page_started, int force,
1229                               unsigned long *nr_written)
1230 {
1231         struct btrfs_root *root = BTRFS_I(inode)->root;
1232         struct btrfs_trans_handle *trans;
1233         struct extent_buffer *leaf;
1234         struct btrfs_path *path;
1235         struct btrfs_file_extent_item *fi;
1236         struct btrfs_key found_key;
1237         u64 cow_start;
1238         u64 cur_offset;
1239         u64 extent_end;
1240         u64 extent_offset;
1241         u64 disk_bytenr;
1242         u64 num_bytes;
1243         u64 disk_num_bytes;
1244         u64 ram_bytes;
1245         int extent_type;
1246         int ret, err;
1247         int type;
1248         int nocow;
1249         int check_prev = 1;
1250         bool nolock;
1251         u64 ino = btrfs_ino(inode);
1252 
1253         path = btrfs_alloc_path();
1254         if (!path) {
1255                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1256                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1257                                              EXTENT_DO_ACCOUNTING |
1258                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1259                                              PAGE_CLEAR_DIRTY |
1260                                              PAGE_SET_WRITEBACK |
1261                                              PAGE_END_WRITEBACK);
1262                 return -ENOMEM;
1263         }
1264 
1265         nolock = btrfs_is_free_space_inode(inode);
1266 
1267         if (nolock)
1268                 trans = btrfs_join_transaction_nolock(root);
1269         else
1270                 trans = btrfs_join_transaction(root);
1271 
1272         if (IS_ERR(trans)) {
1273                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1274                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1275                                              EXTENT_DO_ACCOUNTING |
1276                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1277                                              PAGE_CLEAR_DIRTY |
1278                                              PAGE_SET_WRITEBACK |
1279                                              PAGE_END_WRITEBACK);
1280                 btrfs_free_path(path);
1281                 return PTR_ERR(trans);
1282         }
1283 
1284         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1285 
1286         cow_start = (u64)-1;
1287         cur_offset = start;
1288         while (1) {
1289                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1290                                                cur_offset, 0);
1291                 if (ret < 0)
1292                         goto error;
1293                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1294                         leaf = path->nodes[0];
1295                         btrfs_item_key_to_cpu(leaf, &found_key,
1296                                               path->slots[0] - 1);
1297                         if (found_key.objectid == ino &&
1298                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1299                                 path->slots[0]--;
1300                 }
1301                 check_prev = 0;
1302 next_slot:
1303                 leaf = path->nodes[0];
1304                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1305                         ret = btrfs_next_leaf(root, path);
1306                         if (ret < 0)
1307                                 goto error;
1308                         if (ret > 0)
1309                                 break;
1310                         leaf = path->nodes[0];
1311                 }
1312 
1313                 nocow = 0;
1314                 disk_bytenr = 0;
1315                 num_bytes = 0;
1316                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1317 
1318                 if (found_key.objectid > ino)
1319                         break;
1320                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1321                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1322                         path->slots[0]++;
1323                         goto next_slot;
1324                 }
1325                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1326                     found_key.offset > end)
1327                         break;
1328 
1329                 if (found_key.offset > cur_offset) {
1330                         extent_end = found_key.offset;
1331                         extent_type = 0;
1332                         goto out_check;
1333                 }
1334 
1335                 fi = btrfs_item_ptr(leaf, path->slots[0],
1336                                     struct btrfs_file_extent_item);
1337                 extent_type = btrfs_file_extent_type(leaf, fi);
1338 
1339                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1340                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1341                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1342                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1343                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1344                         extent_end = found_key.offset +
1345                                 btrfs_file_extent_num_bytes(leaf, fi);
1346                         disk_num_bytes =
1347                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1348                         if (extent_end <= start) {
1349                                 path->slots[0]++;
1350                                 goto next_slot;
1351                         }
1352                         if (disk_bytenr == 0)
1353                                 goto out_check;
1354                         if (btrfs_file_extent_compression(leaf, fi) ||
1355                             btrfs_file_extent_encryption(leaf, fi) ||
1356                             btrfs_file_extent_other_encoding(leaf, fi))
1357                                 goto out_check;
1358                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1359                                 goto out_check;
1360                         if (btrfs_extent_readonly(root, disk_bytenr))
1361                                 goto out_check;
1362                         if (btrfs_cross_ref_exist(trans, root, ino,
1363                                                   found_key.offset -
1364                                                   extent_offset, disk_bytenr))
1365                                 goto out_check;
1366                         disk_bytenr += extent_offset;
1367                         disk_bytenr += cur_offset - found_key.offset;
1368                         num_bytes = min(end + 1, extent_end) - cur_offset;
1369                         /*
1370                          * if there are pending snapshots for this root,
1371                          * we fall into common COW way.
1372                          */
1373                         if (!nolock) {
1374                                 err = btrfs_start_write_no_snapshoting(root);
1375                                 if (!err)
1376                                         goto out_check;
1377                         }
1378                         /*
1379                          * force cow if csum exists in the range.
1380                          * this ensure that csum for a given extent are
1381                          * either valid or do not exist.
1382                          */
1383                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1384                                 goto out_check;
1385                         if (!btrfs_inc_nocow_writers(root->fs_info,
1386                                                      disk_bytenr))
1387                                 goto out_check;
1388                         nocow = 1;
1389                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1390                         extent_end = found_key.offset +
1391                                 btrfs_file_extent_inline_len(leaf,
1392                                                      path->slots[0], fi);
1393                         extent_end = ALIGN(extent_end, root->sectorsize);
1394                 } else {
1395                         BUG_ON(1);
1396                 }
1397 out_check:
1398                 if (extent_end <= start) {
1399                         path->slots[0]++;
1400                         if (!nolock && nocow)
1401                                 btrfs_end_write_no_snapshoting(root);
1402                         if (nocow)
1403                                 btrfs_dec_nocow_writers(root->fs_info,
1404                                                         disk_bytenr);
1405                         goto next_slot;
1406                 }
1407                 if (!nocow) {
1408                         if (cow_start == (u64)-1)
1409                                 cow_start = cur_offset;
1410                         cur_offset = extent_end;
1411                         if (cur_offset > end)
1412                                 break;
1413                         path->slots[0]++;
1414                         goto next_slot;
1415                 }
1416 
1417                 btrfs_release_path(path);
1418                 if (cow_start != (u64)-1) {
1419                         ret = cow_file_range(inode, locked_page,
1420                                              cow_start, found_key.offset - 1,
1421                                              page_started, nr_written, 1);
1422                         if (ret) {
1423                                 if (!nolock && nocow)
1424                                         btrfs_end_write_no_snapshoting(root);
1425                                 if (nocow)
1426                                         btrfs_dec_nocow_writers(root->fs_info,
1427                                                                 disk_bytenr);
1428                                 goto error;
1429                         }
1430                         cow_start = (u64)-1;
1431                 }
1432 
1433                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1434                         struct extent_map *em;
1435                         struct extent_map_tree *em_tree;
1436                         em_tree = &BTRFS_I(inode)->extent_tree;
1437                         em = alloc_extent_map();
1438                         BUG_ON(!em); /* -ENOMEM */
1439                         em->start = cur_offset;
1440                         em->orig_start = found_key.offset - extent_offset;
1441                         em->len = num_bytes;
1442                         em->block_len = num_bytes;
1443                         em->block_start = disk_bytenr;
1444                         em->orig_block_len = disk_num_bytes;
1445                         em->ram_bytes = ram_bytes;
1446                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1447                         em->mod_start = em->start;
1448                         em->mod_len = em->len;
1449                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1450                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1451                         em->generation = -1;
1452                         while (1) {
1453                                 write_lock(&em_tree->lock);
1454                                 ret = add_extent_mapping(em_tree, em, 1);
1455                                 write_unlock(&em_tree->lock);
1456                                 if (ret != -EEXIST) {
1457                                         free_extent_map(em);
1458                                         break;
1459                                 }
1460                                 btrfs_drop_extent_cache(inode, em->start,
1461                                                 em->start + em->len - 1, 0);
1462                         }
1463                         type = BTRFS_ORDERED_PREALLOC;
1464                 } else {
1465                         type = BTRFS_ORDERED_NOCOW;
1466                 }
1467 
1468                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1469                                                num_bytes, num_bytes, type);
1470                 if (nocow)
1471                         btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1472                 BUG_ON(ret); /* -ENOMEM */
1473 
1474                 if (root->root_key.objectid ==
1475                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1476                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1477                                                       num_bytes);
1478                         if (ret) {
1479                                 if (!nolock && nocow)
1480                                         btrfs_end_write_no_snapshoting(root);
1481                                 goto error;
1482                         }
1483                 }
1484 
1485                 extent_clear_unlock_delalloc(inode, cur_offset,
1486                                              cur_offset + num_bytes - 1,
1487                                              locked_page, EXTENT_LOCKED |
1488                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1489                                              PAGE_SET_PRIVATE2);
1490                 if (!nolock && nocow)
1491                         btrfs_end_write_no_snapshoting(root);
1492                 cur_offset = extent_end;
1493                 if (cur_offset > end)
1494                         break;
1495         }
1496         btrfs_release_path(path);
1497 
1498         if (cur_offset <= end && cow_start == (u64)-1) {
1499                 cow_start = cur_offset;
1500                 cur_offset = end;
1501         }
1502 
1503         if (cow_start != (u64)-1) {
1504                 ret = cow_file_range(inode, locked_page, cow_start, end,
1505                                      page_started, nr_written, 1);
1506                 if (ret)
1507                         goto error;
1508         }
1509 
1510 error:
1511         err = btrfs_end_transaction(trans, root);
1512         if (!ret)
1513                 ret = err;
1514 
1515         if (ret && cur_offset < end)
1516                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1517                                              locked_page, EXTENT_LOCKED |
1518                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1519                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1520                                              PAGE_CLEAR_DIRTY |
1521                                              PAGE_SET_WRITEBACK |
1522                                              PAGE_END_WRITEBACK);
1523         btrfs_free_path(path);
1524         return ret;
1525 }
1526 
1527 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1528 {
1529 
1530         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1531             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1532                 return 0;
1533 
1534         /*
1535          * @defrag_bytes is a hint value, no spinlock held here,
1536          * if is not zero, it means the file is defragging.
1537          * Force cow if given extent needs to be defragged.
1538          */
1539         if (BTRFS_I(inode)->defrag_bytes &&
1540             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1541                            EXTENT_DEFRAG, 0, NULL))
1542                 return 1;
1543 
1544         return 0;
1545 }
1546 
1547 /*
1548  * extent_io.c call back to do delayed allocation processing
1549  */
1550 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1551                               u64 start, u64 end, int *page_started,
1552                               unsigned long *nr_written)
1553 {
1554         int ret;
1555         int force_cow = need_force_cow(inode, start, end);
1556 
1557         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1558                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1559                                          page_started, 1, nr_written);
1560         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1561                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1562                                          page_started, 0, nr_written);
1563         } else if (!inode_need_compress(inode)) {
1564                 ret = cow_file_range(inode, locked_page, start, end,
1565                                       page_started, nr_written, 1);
1566         } else {
1567                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1568                         &BTRFS_I(inode)->runtime_flags);
1569                 ret = cow_file_range_async(inode, locked_page, start, end,
1570                                            page_started, nr_written);
1571         }
1572         return ret;
1573 }
1574 
1575 static void btrfs_split_extent_hook(struct inode *inode,
1576                                     struct extent_state *orig, u64 split)
1577 {
1578         u64 size;
1579 
1580         /* not delalloc, ignore it */
1581         if (!(orig->state & EXTENT_DELALLOC))
1582                 return;
1583 
1584         size = orig->end - orig->start + 1;
1585         if (size > BTRFS_MAX_EXTENT_SIZE) {
1586                 u64 num_extents;
1587                 u64 new_size;
1588 
1589                 /*
1590                  * See the explanation in btrfs_merge_extent_hook, the same
1591                  * applies here, just in reverse.
1592                  */
1593                 new_size = orig->end - split + 1;
1594                 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1595                                         BTRFS_MAX_EXTENT_SIZE);
1596                 new_size = split - orig->start;
1597                 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1598                                         BTRFS_MAX_EXTENT_SIZE);
1599                 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1600                               BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1601                         return;
1602         }
1603 
1604         spin_lock(&BTRFS_I(inode)->lock);
1605         BTRFS_I(inode)->outstanding_extents++;
1606         spin_unlock(&BTRFS_I(inode)->lock);
1607 }
1608 
1609 /*
1610  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1611  * extents so we can keep track of new extents that are just merged onto old
1612  * extents, such as when we are doing sequential writes, so we can properly
1613  * account for the metadata space we'll need.
1614  */
1615 static void btrfs_merge_extent_hook(struct inode *inode,
1616                                     struct extent_state *new,
1617                                     struct extent_state *other)
1618 {
1619         u64 new_size, old_size;
1620         u64 num_extents;
1621 
1622         /* not delalloc, ignore it */
1623         if (!(other->state & EXTENT_DELALLOC))
1624                 return;
1625 
1626         if (new->start > other->start)
1627                 new_size = new->end - other->start + 1;
1628         else
1629                 new_size = other->end - new->start + 1;
1630 
1631         /* we're not bigger than the max, unreserve the space and go */
1632         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1633                 spin_lock(&BTRFS_I(inode)->lock);
1634                 BTRFS_I(inode)->outstanding_extents--;
1635                 spin_unlock(&BTRFS_I(inode)->lock);
1636                 return;
1637         }
1638 
1639         /*
1640          * We have to add up either side to figure out how many extents were
1641          * accounted for before we merged into one big extent.  If the number of
1642          * extents we accounted for is <= the amount we need for the new range
1643          * then we can return, otherwise drop.  Think of it like this
1644          *
1645          * [ 4k][MAX_SIZE]
1646          *
1647          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1648          * need 2 outstanding extents, on one side we have 1 and the other side
1649          * we have 1 so they are == and we can return.  But in this case
1650          *
1651          * [MAX_SIZE+4k][MAX_SIZE+4k]
1652          *
1653          * Each range on their own accounts for 2 extents, but merged together
1654          * they are only 3 extents worth of accounting, so we need to drop in
1655          * this case.
1656          */
1657         old_size = other->end - other->start + 1;
1658         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1659                                 BTRFS_MAX_EXTENT_SIZE);
1660         old_size = new->end - new->start + 1;
1661         num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1662                                  BTRFS_MAX_EXTENT_SIZE);
1663 
1664         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1665                       BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1666                 return;
1667 
1668         spin_lock(&BTRFS_I(inode)->lock);
1669         BTRFS_I(inode)->outstanding_extents--;
1670         spin_unlock(&BTRFS_I(inode)->lock);
1671 }
1672 
1673 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1674                                       struct inode *inode)
1675 {
1676         spin_lock(&root->delalloc_lock);
1677         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1678                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1679                               &root->delalloc_inodes);
1680                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1681                         &BTRFS_I(inode)->runtime_flags);
1682                 root->nr_delalloc_inodes++;
1683                 if (root->nr_delalloc_inodes == 1) {
1684                         spin_lock(&root->fs_info->delalloc_root_lock);
1685                         BUG_ON(!list_empty(&root->delalloc_root));
1686                         list_add_tail(&root->delalloc_root,
1687                                       &root->fs_info->delalloc_roots);
1688                         spin_unlock(&root->fs_info->delalloc_root_lock);
1689                 }
1690         }
1691         spin_unlock(&root->delalloc_lock);
1692 }
1693 
1694 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1695                                      struct inode *inode)
1696 {
1697         spin_lock(&root->delalloc_lock);
1698         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1699                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1700                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701                           &BTRFS_I(inode)->runtime_flags);
1702                 root->nr_delalloc_inodes--;
1703                 if (!root->nr_delalloc_inodes) {
1704                         spin_lock(&root->fs_info->delalloc_root_lock);
1705                         BUG_ON(list_empty(&root->delalloc_root));
1706                         list_del_init(&root->delalloc_root);
1707                         spin_unlock(&root->fs_info->delalloc_root_lock);
1708                 }
1709         }
1710         spin_unlock(&root->delalloc_lock);
1711 }
1712 
1713 /*
1714  * extent_io.c set_bit_hook, used to track delayed allocation
1715  * bytes in this file, and to maintain the list of inodes that
1716  * have pending delalloc work to be done.
1717  */
1718 static void btrfs_set_bit_hook(struct inode *inode,
1719                                struct extent_state *state, unsigned *bits)
1720 {
1721 
1722         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1723                 WARN_ON(1);
1724         /*
1725          * set_bit and clear bit hooks normally require _irqsave/restore
1726          * but in this case, we are only testing for the DELALLOC
1727          * bit, which is only set or cleared with irqs on
1728          */
1729         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1730                 struct btrfs_root *root = BTRFS_I(inode)->root;
1731                 u64 len = state->end + 1 - state->start;
1732                 bool do_list = !btrfs_is_free_space_inode(inode);
1733 
1734                 if (*bits & EXTENT_FIRST_DELALLOC) {
1735                         *bits &= ~EXTENT_FIRST_DELALLOC;
1736                 } else {
1737                         spin_lock(&BTRFS_I(inode)->lock);
1738                         BTRFS_I(inode)->outstanding_extents++;
1739                         spin_unlock(&BTRFS_I(inode)->lock);
1740                 }
1741 
1742                 /* For sanity tests */
1743                 if (btrfs_test_is_dummy_root(root))
1744                         return;
1745 
1746                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1747                                      root->fs_info->delalloc_batch);
1748                 spin_lock(&BTRFS_I(inode)->lock);
1749                 BTRFS_I(inode)->delalloc_bytes += len;
1750                 if (*bits & EXTENT_DEFRAG)
1751                         BTRFS_I(inode)->defrag_bytes += len;
1752                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1753                                          &BTRFS_I(inode)->runtime_flags))
1754                         btrfs_add_delalloc_inodes(root, inode);
1755                 spin_unlock(&BTRFS_I(inode)->lock);
1756         }
1757 }
1758 
1759 /*
1760  * extent_io.c clear_bit_hook, see set_bit_hook for why
1761  */
1762 static void btrfs_clear_bit_hook(struct inode *inode,
1763                                  struct extent_state *state,
1764                                  unsigned *bits)
1765 {
1766         u64 len = state->end + 1 - state->start;
1767         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1768                                     BTRFS_MAX_EXTENT_SIZE);
1769 
1770         spin_lock(&BTRFS_I(inode)->lock);
1771         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1772                 BTRFS_I(inode)->defrag_bytes -= len;
1773         spin_unlock(&BTRFS_I(inode)->lock);
1774 
1775         /*
1776          * set_bit and clear bit hooks normally require _irqsave/restore
1777          * but in this case, we are only testing for the DELALLOC
1778          * bit, which is only set or cleared with irqs on
1779          */
1780         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1781                 struct btrfs_root *root = BTRFS_I(inode)->root;
1782                 bool do_list = !btrfs_is_free_space_inode(inode);
1783 
1784                 if (*bits & EXTENT_FIRST_DELALLOC) {
1785                         *bits &= ~EXTENT_FIRST_DELALLOC;
1786                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1787                         spin_lock(&BTRFS_I(inode)->lock);
1788                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1789                         spin_unlock(&BTRFS_I(inode)->lock);
1790                 }
1791 
1792                 /*
1793                  * We don't reserve metadata space for space cache inodes so we
1794                  * don't need to call dellalloc_release_metadata if there is an
1795                  * error.
1796                  */
1797                 if (*bits & EXTENT_DO_ACCOUNTING &&
1798                     root != root->fs_info->tree_root)
1799                         btrfs_delalloc_release_metadata(inode, len);
1800 
1801                 /* For sanity tests. */
1802                 if (btrfs_test_is_dummy_root(root))
1803                         return;
1804 
1805                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1806                     && do_list && !(state->state & EXTENT_NORESERVE))
1807                         btrfs_free_reserved_data_space_noquota(inode,
1808                                         state->start, len);
1809 
1810                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1811                                      root->fs_info->delalloc_batch);
1812                 spin_lock(&BTRFS_I(inode)->lock);
1813                 BTRFS_I(inode)->delalloc_bytes -= len;
1814                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1815                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1816                              &BTRFS_I(inode)->runtime_flags))
1817                         btrfs_del_delalloc_inode(root, inode);
1818                 spin_unlock(&BTRFS_I(inode)->lock);
1819         }
1820 }
1821 
1822 /*
1823  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1824  * we don't create bios that span stripes or chunks
1825  */
1826 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1827                          size_t size, struct bio *bio,
1828                          unsigned long bio_flags)
1829 {
1830         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1831         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1832         u64 length = 0;
1833         u64 map_length;
1834         int ret;
1835 
1836         if (bio_flags & EXTENT_BIO_COMPRESSED)
1837                 return 0;
1838 
1839         length = bio->bi_iter.bi_size;
1840         map_length = length;
1841         ret = btrfs_map_block(root->fs_info, rw, logical,
1842                               &map_length, NULL, 0);
1843         /* Will always return 0 with map_multi == NULL */
1844         BUG_ON(ret < 0);
1845         if (map_length < length + size)
1846                 return 1;
1847         return 0;
1848 }
1849 
1850 /*
1851  * in order to insert checksums into the metadata in large chunks,
1852  * we wait until bio submission time.   All the pages in the bio are
1853  * checksummed and sums are attached onto the ordered extent record.
1854  *
1855  * At IO completion time the cums attached on the ordered extent record
1856  * are inserted into the btree
1857  */
1858 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1859                                     struct bio *bio, int mirror_num,
1860                                     unsigned long bio_flags,
1861                                     u64 bio_offset)
1862 {
1863         struct btrfs_root *root = BTRFS_I(inode)->root;
1864         int ret = 0;
1865 
1866         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1867         BUG_ON(ret); /* -ENOMEM */
1868         return 0;
1869 }
1870 
1871 /*
1872  * in order to insert checksums into the metadata in large chunks,
1873  * we wait until bio submission time.   All the pages in the bio are
1874  * checksummed and sums are attached onto the ordered extent record.
1875  *
1876  * At IO completion time the cums attached on the ordered extent record
1877  * are inserted into the btree
1878  */
1879 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1880                           int mirror_num, unsigned long bio_flags,
1881                           u64 bio_offset)
1882 {
1883         struct btrfs_root *root = BTRFS_I(inode)->root;
1884         int ret;
1885 
1886         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1887         if (ret) {
1888                 bio->bi_error = ret;
1889                 bio_endio(bio);
1890         }
1891         return ret;
1892 }
1893 
1894 /*
1895  * extent_io.c submission hook. This does the right thing for csum calculation
1896  * on write, or reading the csums from the tree before a read
1897  */
1898 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1899                           int mirror_num, unsigned long bio_flags,
1900                           u64 bio_offset)
1901 {
1902         struct btrfs_root *root = BTRFS_I(inode)->root;
1903         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1904         int ret = 0;
1905         int skip_sum;
1906         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1907 
1908         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1909 
1910         if (btrfs_is_free_space_inode(inode))
1911                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1912 
1913         if (!(rw & REQ_WRITE)) {
1914                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1915                 if (ret)
1916                         goto out;
1917 
1918                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1919                         ret = btrfs_submit_compressed_read(inode, bio,
1920                                                            mirror_num,
1921                                                            bio_flags);
1922                         goto out;
1923                 } else if (!skip_sum) {
1924                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1925                         if (ret)
1926                                 goto out;
1927                 }
1928                 goto mapit;
1929         } else if (async && !skip_sum) {
1930                 /* csum items have already been cloned */
1931                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1932                         goto mapit;
1933                 /* we're doing a write, do the async checksumming */
1934                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1935                                    inode, rw, bio, mirror_num,
1936                                    bio_flags, bio_offset,
1937                                    __btrfs_submit_bio_start,
1938                                    __btrfs_submit_bio_done);
1939                 goto out;
1940         } else if (!skip_sum) {
1941                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1942                 if (ret)
1943                         goto out;
1944         }
1945 
1946 mapit:
1947         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1948 
1949 out:
1950         if (ret < 0) {
1951                 bio->bi_error = ret;
1952                 bio_endio(bio);
1953         }
1954         return ret;
1955 }
1956 
1957 /*
1958  * given a list of ordered sums record them in the inode.  This happens
1959  * at IO completion time based on sums calculated at bio submission time.
1960  */
1961 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1962                              struct inode *inode, u64 file_offset,
1963                              struct list_head *list)
1964 {
1965         struct btrfs_ordered_sum *sum;
1966 
1967         list_for_each_entry(sum, list, list) {
1968                 trans->adding_csums = 1;
1969                 btrfs_csum_file_blocks(trans,
1970                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1971                 trans->adding_csums = 0;
1972         }
1973         return 0;
1974 }
1975 
1976 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1977                               struct extent_state **cached_state)
1978 {
1979         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1980         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1981                                    cached_state);
1982 }
1983 
1984 /* see btrfs_writepage_start_hook for details on why this is required */
1985 struct btrfs_writepage_fixup {
1986         struct page *page;
1987         struct btrfs_work work;
1988 };
1989 
1990 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1991 {
1992         struct btrfs_writepage_fixup *fixup;
1993         struct btrfs_ordered_extent *ordered;
1994         struct extent_state *cached_state = NULL;
1995         struct page *page;
1996         struct inode *inode;
1997         u64 page_start;
1998         u64 page_end;
1999         int ret;
2000 
2001         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2002         page = fixup->page;
2003 again:
2004         lock_page(page);
2005         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2006                 ClearPageChecked(page);
2007                 goto out_page;
2008         }
2009 
2010         inode = page->mapping->host;
2011         page_start = page_offset(page);
2012         page_end = page_offset(page) + PAGE_SIZE - 1;
2013 
2014         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2015                          &cached_state);
2016 
2017         /* already ordered? We're done */
2018         if (PagePrivate2(page))
2019                 goto out;
2020 
2021         ordered = btrfs_lookup_ordered_range(inode, page_start,
2022                                         PAGE_SIZE);
2023         if (ordered) {
2024                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2025                                      page_end, &cached_state, GFP_NOFS);
2026                 unlock_page(page);
2027                 btrfs_start_ordered_extent(inode, ordered, 1);
2028                 btrfs_put_ordered_extent(ordered);
2029                 goto again;
2030         }
2031 
2032         ret = btrfs_delalloc_reserve_space(inode, page_start,
2033                                            PAGE_SIZE);
2034         if (ret) {
2035                 mapping_set_error(page->mapping, ret);
2036                 end_extent_writepage(page, ret, page_start, page_end);
2037                 ClearPageChecked(page);
2038                 goto out;
2039          }
2040 
2041         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2042         ClearPageChecked(page);
2043         set_page_dirty(page);
2044 out:
2045         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2046                              &cached_state, GFP_NOFS);
2047 out_page:
2048         unlock_page(page);
2049         put_page(page);
2050         kfree(fixup);
2051 }
2052 
2053 /*
2054  * There are a few paths in the higher layers of the kernel that directly
2055  * set the page dirty bit without asking the filesystem if it is a
2056  * good idea.  This causes problems because we want to make sure COW
2057  * properly happens and the data=ordered rules are followed.
2058  *
2059  * In our case any range that doesn't have the ORDERED bit set
2060  * hasn't been properly setup for IO.  We kick off an async process
2061  * to fix it up.  The async helper will wait for ordered extents, set
2062  * the delalloc bit and make it safe to write the page.
2063  */
2064 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2065 {
2066         struct inode *inode = page->mapping->host;
2067         struct btrfs_writepage_fixup *fixup;
2068         struct btrfs_root *root = BTRFS_I(inode)->root;
2069 
2070         /* this page is properly in the ordered list */
2071         if (TestClearPagePrivate2(page))
2072                 return 0;
2073 
2074         if (PageChecked(page))
2075                 return -EAGAIN;
2076 
2077         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2078         if (!fixup)
2079                 return -EAGAIN;
2080 
2081         SetPageChecked(page);
2082         get_page(page);
2083         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2084                         btrfs_writepage_fixup_worker, NULL, NULL);
2085         fixup->page = page;
2086         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2087         return -EBUSY;
2088 }
2089 
2090 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2091                                        struct inode *inode, u64 file_pos,
2092                                        u64 disk_bytenr, u64 disk_num_bytes,
2093                                        u64 num_bytes, u64 ram_bytes,
2094                                        u8 compression, u8 encryption,
2095                                        u16 other_encoding, int extent_type)
2096 {
2097         struct btrfs_root *root = BTRFS_I(inode)->root;
2098         struct btrfs_file_extent_item *fi;
2099         struct btrfs_path *path;
2100         struct extent_buffer *leaf;
2101         struct btrfs_key ins;
2102         int extent_inserted = 0;
2103         int ret;
2104 
2105         path = btrfs_alloc_path();
2106         if (!path)
2107                 return -ENOMEM;
2108 
2109         /*
2110          * we may be replacing one extent in the tree with another.
2111          * The new extent is pinned in the extent map, and we don't want
2112          * to drop it from the cache until it is completely in the btree.
2113          *
2114          * So, tell btrfs_drop_extents to leave this extent in the cache.
2115          * the caller is expected to unpin it and allow it to be merged
2116          * with the others.
2117          */
2118         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2119                                    file_pos + num_bytes, NULL, 0,
2120                                    1, sizeof(*fi), &extent_inserted);
2121         if (ret)
2122                 goto out;
2123 
2124         if (!extent_inserted) {
2125                 ins.objectid = btrfs_ino(inode);
2126                 ins.offset = file_pos;
2127                 ins.type = BTRFS_EXTENT_DATA_KEY;
2128 
2129                 path->leave_spinning = 1;
2130                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2131                                               sizeof(*fi));
2132                 if (ret)
2133                         goto out;
2134         }
2135         leaf = path->nodes[0];
2136         fi = btrfs_item_ptr(leaf, path->slots[0],
2137                             struct btrfs_file_extent_item);
2138         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2139         btrfs_set_file_extent_type(leaf, fi, extent_type);
2140         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2141         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2142         btrfs_set_file_extent_offset(leaf, fi, 0);
2143         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2144         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2145         btrfs_set_file_extent_compression(leaf, fi, compression);
2146         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2147         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2148 
2149         btrfs_mark_buffer_dirty(leaf);
2150         btrfs_release_path(path);
2151 
2152         inode_add_bytes(inode, num_bytes);
2153 
2154         ins.objectid = disk_bytenr;
2155         ins.offset = disk_num_bytes;
2156         ins.type = BTRFS_EXTENT_ITEM_KEY;
2157         ret = btrfs_alloc_reserved_file_extent(trans, root,
2158                                         root->root_key.objectid,
2159                                         btrfs_ino(inode), file_pos,
2160                                         ram_bytes, &ins);
2161         /*
2162          * Release the reserved range from inode dirty range map, as it is
2163          * already moved into delayed_ref_head
2164          */
2165         btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2166 out:
2167         btrfs_free_path(path);
2168 
2169         return ret;
2170 }
2171 
2172 /* snapshot-aware defrag */
2173 struct sa_defrag_extent_backref {
2174         struct rb_node node;
2175         struct old_sa_defrag_extent *old;
2176         u64 root_id;
2177         u64 inum;
2178         u64 file_pos;
2179         u64 extent_offset;
2180         u64 num_bytes;
2181         u64 generation;
2182 };
2183 
2184 struct old_sa_defrag_extent {
2185         struct list_head list;
2186         struct new_sa_defrag_extent *new;
2187 
2188         u64 extent_offset;
2189         u64 bytenr;
2190         u64 offset;
2191         u64 len;
2192         int count;
2193 };
2194 
2195 struct new_sa_defrag_extent {
2196         struct rb_root root;
2197         struct list_head head;
2198         struct btrfs_path *path;
2199         struct inode *inode;
2200         u64 file_pos;
2201         u64 len;
2202         u64 bytenr;
2203         u64 disk_len;
2204         u8 compress_type;
2205 };
2206 
2207 static int backref_comp(struct sa_defrag_extent_backref *b1,
2208                         struct sa_defrag_extent_backref *b2)
2209 {
2210         if (b1->root_id < b2->root_id)
2211                 return -1;
2212         else if (b1->root_id > b2->root_id)
2213                 return 1;
2214 
2215         if (b1->inum < b2->inum)
2216                 return -1;
2217         else if (b1->inum > b2->inum)
2218                 return 1;
2219 
2220         if (b1->file_pos < b2->file_pos)
2221                 return -1;
2222         else if (b1->file_pos > b2->file_pos)
2223                 return 1;
2224 
2225         /*
2226          * [------------------------------] ===> (a range of space)
2227          *     |<--->|   |<---->| =============> (fs/file tree A)
2228          * |<---------------------------->| ===> (fs/file tree B)
2229          *
2230          * A range of space can refer to two file extents in one tree while
2231          * refer to only one file extent in another tree.
2232          *
2233          * So we may process a disk offset more than one time(two extents in A)
2234          * and locate at the same extent(one extent in B), then insert two same
2235          * backrefs(both refer to the extent in B).
2236          */
2237         return 0;
2238 }
2239 
2240 static void backref_insert(struct rb_root *root,
2241                            struct sa_defrag_extent_backref *backref)
2242 {
2243         struct rb_node **p = &root->rb_node;
2244         struct rb_node *parent = NULL;
2245         struct sa_defrag_extent_backref *entry;
2246         int ret;
2247 
2248         while (*p) {
2249                 parent = *p;
2250                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2251 
2252                 ret = backref_comp(backref, entry);
2253                 if (ret < 0)
2254                         p = &(*p)->rb_left;
2255                 else
2256                         p = &(*p)->rb_right;
2257         }
2258 
2259         rb_link_node(&backref->node, parent, p);
2260         rb_insert_color(&backref->node, root);
2261 }
2262 
2263 /*
2264  * Note the backref might has changed, and in this case we just return 0.
2265  */
2266 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2267                                        void *ctx)
2268 {
2269         struct btrfs_file_extent_item *extent;
2270         struct btrfs_fs_info *fs_info;
2271         struct old_sa_defrag_extent *old = ctx;
2272         struct new_sa_defrag_extent *new = old->new;
2273         struct btrfs_path *path = new->path;
2274         struct btrfs_key key;
2275         struct btrfs_root *root;
2276         struct sa_defrag_extent_backref *backref;
2277         struct extent_buffer *leaf;
2278         struct inode *inode = new->inode;
2279         int slot;
2280         int ret;
2281         u64 extent_offset;
2282         u64 num_bytes;
2283 
2284         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2285             inum == btrfs_ino(inode))
2286                 return 0;
2287 
2288         key.objectid = root_id;
2289         key.type = BTRFS_ROOT_ITEM_KEY;
2290         key.offset = (u64)-1;
2291 
2292         fs_info = BTRFS_I(inode)->root->fs_info;
2293         root = btrfs_read_fs_root_no_name(fs_info, &key);
2294         if (IS_ERR(root)) {
2295                 if (PTR_ERR(root) == -ENOENT)
2296                         return 0;
2297                 WARN_ON(1);
2298                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2299                          inum, offset, root_id);
2300                 return PTR_ERR(root);
2301         }
2302 
2303         key.objectid = inum;
2304         key.type = BTRFS_EXTENT_DATA_KEY;
2305         if (offset > (u64)-1 << 32)
2306                 key.offset = 0;
2307         else
2308                 key.offset = offset;
2309 
2310         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311         if (WARN_ON(ret < 0))
2312                 return ret;
2313         ret = 0;
2314 
2315         while (1) {
2316                 cond_resched();
2317 
2318                 leaf = path->nodes[0];
2319                 slot = path->slots[0];
2320 
2321                 if (slot >= btrfs_header_nritems(leaf)) {
2322                         ret = btrfs_next_leaf(root, path);
2323                         if (ret < 0) {
2324                                 goto out;
2325                         } else if (ret > 0) {
2326                                 ret = 0;
2327                                 goto out;
2328                         }
2329                         continue;
2330                 }
2331 
2332                 path->slots[0]++;
2333 
2334                 btrfs_item_key_to_cpu(leaf, &key, slot);
2335 
2336                 if (key.objectid > inum)
2337                         goto out;
2338 
2339                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2340                         continue;
2341 
2342                 extent = btrfs_item_ptr(leaf, slot,
2343                                         struct btrfs_file_extent_item);
2344 
2345                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2346                         continue;
2347 
2348                 /*
2349                  * 'offset' refers to the exact key.offset,
2350                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2351                  * (key.offset - extent_offset).
2352                  */
2353                 if (key.offset != offset)
2354                         continue;
2355 
2356                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2357                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2358 
2359                 if (extent_offset >= old->extent_offset + old->offset +
2360                     old->len || extent_offset + num_bytes <=
2361                     old->extent_offset + old->offset)
2362                         continue;
2363                 break;
2364         }
2365 
2366         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2367         if (!backref) {
2368                 ret = -ENOENT;
2369                 goto out;
2370         }
2371 
2372         backref->root_id = root_id;
2373         backref->inum = inum;
2374         backref->file_pos = offset;
2375         backref->num_bytes = num_bytes;
2376         backref->extent_offset = extent_offset;
2377         backref->generation = btrfs_file_extent_generation(leaf, extent);
2378         backref->old = old;
2379         backref_insert(&new->root, backref);
2380         old->count++;
2381 out:
2382         btrfs_release_path(path);
2383         WARN_ON(ret);
2384         return ret;
2385 }
2386 
2387 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2388                                    struct new_sa_defrag_extent *new)
2389 {
2390         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2391         struct old_sa_defrag_extent *old, *tmp;
2392         int ret;
2393 
2394         new->path = path;
2395 
2396         list_for_each_entry_safe(old, tmp, &new->head, list) {
2397                 ret = iterate_inodes_from_logical(old->bytenr +
2398                                                   old->extent_offset, fs_info,
2399                                                   path, record_one_backref,
2400                                                   old);
2401                 if (ret < 0 && ret != -ENOENT)
2402                         return false;
2403 
2404                 /* no backref to be processed for this extent */
2405                 if (!old->count) {
2406                         list_del(&old->list);
2407                         kfree(old);
2408                 }
2409         }
2410 
2411         if (list_empty(&new->head))
2412                 return false;
2413 
2414         return true;
2415 }
2416 
2417 static int relink_is_mergable(struct extent_buffer *leaf,
2418                               struct btrfs_file_extent_item *fi,
2419                               struct new_sa_defrag_extent *new)
2420 {
2421         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2422                 return 0;
2423 
2424         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2425                 return 0;
2426 
2427         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2428                 return 0;
2429 
2430         if (btrfs_file_extent_encryption(leaf, fi) ||
2431             btrfs_file_extent_other_encoding(leaf, fi))
2432                 return 0;
2433 
2434         return 1;
2435 }
2436 
2437 /*
2438  * Note the backref might has changed, and in this case we just return 0.
2439  */
2440 static noinline int relink_extent_backref(struct btrfs_path *path,
2441                                  struct sa_defrag_extent_backref *prev,
2442                                  struct sa_defrag_extent_backref *backref)
2443 {
2444         struct btrfs_file_extent_item *extent;
2445         struct btrfs_file_extent_item *item;
2446         struct btrfs_ordered_extent *ordered;
2447         struct btrfs_trans_handle *trans;
2448         struct btrfs_fs_info *fs_info;
2449         struct btrfs_root *root;
2450         struct btrfs_key key;
2451         struct extent_buffer *leaf;
2452         struct old_sa_defrag_extent *old = backref->old;
2453         struct new_sa_defrag_extent *new = old->new;
2454         struct inode *src_inode = new->inode;
2455         struct inode *inode;
2456         struct extent_state *cached = NULL;
2457         int ret = 0;
2458         u64 start;
2459         u64 len;
2460         u64 lock_start;
2461         u64 lock_end;
2462         bool merge = false;
2463         int index;
2464 
2465         if (prev && prev->root_id == backref->root_id &&
2466             prev->inum == backref->inum &&
2467             prev->file_pos + prev->num_bytes == backref->file_pos)
2468                 merge = true;
2469 
2470         /* step 1: get root */
2471         key.objectid = backref->root_id;
2472         key.type = BTRFS_ROOT_ITEM_KEY;
2473         key.offset = (u64)-1;
2474 
2475         fs_info = BTRFS_I(src_inode)->root->fs_info;
2476         index = srcu_read_lock(&fs_info->subvol_srcu);
2477 
2478         root = btrfs_read_fs_root_no_name(fs_info, &key);
2479         if (IS_ERR(root)) {
2480                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2481                 if (PTR_ERR(root) == -ENOENT)
2482                         return 0;
2483                 return PTR_ERR(root);
2484         }
2485 
2486         if (btrfs_root_readonly(root)) {
2487                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2488                 return 0;
2489         }
2490 
2491         /* step 2: get inode */
2492         key.objectid = backref->inum;
2493         key.type = BTRFS_INODE_ITEM_KEY;
2494         key.offset = 0;
2495 
2496         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2497         if (IS_ERR(inode)) {
2498                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2499                 return 0;
2500         }
2501 
2502         srcu_read_unlock(&fs_info->subvol_srcu, index);
2503 
2504         /* step 3: relink backref */
2505         lock_start = backref->file_pos;
2506         lock_end = backref->file_pos + backref->num_bytes - 1;
2507         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2508                          &cached);
2509 
2510         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2511         if (ordered) {
2512                 btrfs_put_ordered_extent(ordered);
2513                 goto out_unlock;
2514         }
2515 
2516         trans = btrfs_join_transaction(root);
2517         if (IS_ERR(trans)) {
2518                 ret = PTR_ERR(trans);
2519                 goto out_unlock;
2520         }
2521 
2522         key.objectid = backref->inum;
2523         key.type = BTRFS_EXTENT_DATA_KEY;
2524         key.offset = backref->file_pos;
2525 
2526         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2527         if (ret < 0) {
2528                 goto out_free_path;
2529         } else if (ret > 0) {
2530                 ret = 0;
2531                 goto out_free_path;
2532         }
2533 
2534         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2535                                 struct btrfs_file_extent_item);
2536 
2537         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2538             backref->generation)
2539                 goto out_free_path;
2540 
2541         btrfs_release_path(path);
2542 
2543         start = backref->file_pos;
2544         if (backref->extent_offset < old->extent_offset + old->offset)
2545                 start += old->extent_offset + old->offset -
2546                          backref->extent_offset;
2547 
2548         len = min(backref->extent_offset + backref->num_bytes,
2549                   old->extent_offset + old->offset + old->len);
2550         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2551 
2552         ret = btrfs_drop_extents(trans, root, inode, start,
2553                                  start + len, 1);
2554         if (ret)
2555                 goto out_free_path;
2556 again:
2557         key.objectid = btrfs_ino(inode);
2558         key.type = BTRFS_EXTENT_DATA_KEY;
2559         key.offset = start;
2560 
2561         path->leave_spinning = 1;
2562         if (merge) {
2563                 struct btrfs_file_extent_item *fi;
2564                 u64 extent_len;
2565                 struct btrfs_key found_key;
2566 
2567                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2568                 if (ret < 0)
2569                         goto out_free_path;
2570 
2571                 path->slots[0]--;
2572                 leaf = path->nodes[0];
2573                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2574 
2575                 fi = btrfs_item_ptr(leaf, path->slots[0],
2576                                     struct btrfs_file_extent_item);
2577                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2578 
2579                 if (extent_len + found_key.offset == start &&
2580                     relink_is_mergable(leaf, fi, new)) {
2581                         btrfs_set_file_extent_num_bytes(leaf, fi,
2582                                                         extent_len + len);
2583                         btrfs_mark_buffer_dirty(leaf);
2584                         inode_add_bytes(inode, len);
2585 
2586                         ret = 1;
2587                         goto out_free_path;
2588                 } else {
2589                         merge = false;
2590                         btrfs_release_path(path);
2591                         goto again;
2592                 }
2593         }
2594 
2595         ret = btrfs_insert_empty_item(trans, root, path, &key,
2596                                         sizeof(*extent));
2597         if (ret) {
2598                 btrfs_abort_transaction(trans, root, ret);
2599                 goto out_free_path;
2600         }
2601 
2602         leaf = path->nodes[0];
2603         item = btrfs_item_ptr(leaf, path->slots[0],
2604                                 struct btrfs_file_extent_item);
2605         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2606         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2607         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2608         btrfs_set_file_extent_num_bytes(leaf, item, len);
2609         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2610         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2611         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2612         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2613         btrfs_set_file_extent_encryption(leaf, item, 0);
2614         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2615 
2616         btrfs_mark_buffer_dirty(leaf);
2617         inode_add_bytes(inode, len);
2618         btrfs_release_path(path);
2619 
2620         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2621                         new->disk_len, 0,
2622                         backref->root_id, backref->inum,
2623                         new->file_pos); /* start - extent_offset */
2624         if (ret) {
2625                 btrfs_abort_transaction(trans, root, ret);
2626                 goto out_free_path;
2627         }
2628 
2629         ret = 1;
2630 out_free_path:
2631         btrfs_release_path(path);
2632         path->leave_spinning = 0;
2633         btrfs_end_transaction(trans, root);
2634 out_unlock:
2635         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2636                              &cached, GFP_NOFS);
2637         iput(inode);
2638         return ret;
2639 }
2640 
2641 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2642 {
2643         struct old_sa_defrag_extent *old, *tmp;
2644 
2645         if (!new)
2646                 return;
2647 
2648         list_for_each_entry_safe(old, tmp, &new->head, list) {
2649                 kfree(old);
2650         }
2651         kfree(new);
2652 }
2653 
2654 static void relink_file_extents(struct new_sa_defrag_extent *new)
2655 {
2656         struct btrfs_path *path;
2657         struct sa_defrag_extent_backref *backref;
2658         struct sa_defrag_extent_backref *prev = NULL;
2659         struct inode *inode;
2660         struct btrfs_root *root;
2661         struct rb_node *node;
2662         int ret;
2663 
2664         inode = new->inode;
2665         root = BTRFS_I(inode)->root;
2666 
2667         path = btrfs_alloc_path();
2668         if (!path)
2669                 return;
2670 
2671         if (!record_extent_backrefs(path, new)) {
2672                 btrfs_free_path(path);
2673                 goto out;
2674         }
2675         btrfs_release_path(path);
2676 
2677         while (1) {
2678                 node = rb_first(&new->root);
2679                 if (!node)
2680                         break;
2681                 rb_erase(node, &new->root);
2682 
2683                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2684 
2685                 ret = relink_extent_backref(path, prev, backref);
2686                 WARN_ON(ret < 0);
2687 
2688                 kfree(prev);
2689 
2690                 if (ret == 1)
2691                         prev = backref;
2692                 else
2693                         prev = NULL;
2694                 cond_resched();
2695         }
2696         kfree(prev);
2697 
2698         btrfs_free_path(path);
2699 out:
2700         free_sa_defrag_extent(new);
2701 
2702         atomic_dec(&root->fs_info->defrag_running);
2703         wake_up(&root->fs_info->transaction_wait);
2704 }
2705 
2706 static struct new_sa_defrag_extent *
2707 record_old_file_extents(struct inode *inode,
2708                         struct btrfs_ordered_extent *ordered)
2709 {
2710         struct btrfs_root *root = BTRFS_I(inode)->root;
2711         struct btrfs_path *path;
2712         struct btrfs_key key;
2713         struct old_sa_defrag_extent *old;
2714         struct new_sa_defrag_extent *new;
2715         int ret;
2716 
2717         new = kmalloc(sizeof(*new), GFP_NOFS);
2718         if (!new)
2719                 return NULL;
2720 
2721         new->inode = inode;
2722         new->file_pos = ordered->file_offset;
2723         new->len = ordered->len;
2724         new->bytenr = ordered->start;
2725         new->disk_len = ordered->disk_len;
2726         new->compress_type = ordered->compress_type;
2727         new->root = RB_ROOT;
2728         INIT_LIST_HEAD(&new->head);
2729 
2730         path = btrfs_alloc_path();
2731         if (!path)
2732                 goto out_kfree;
2733 
2734         key.objectid = btrfs_ino(inode);
2735         key.type = BTRFS_EXTENT_DATA_KEY;
2736         key.offset = new->file_pos;
2737 
2738         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2739         if (ret < 0)
2740                 goto out_free_path;
2741         if (ret > 0 && path->slots[0] > 0)
2742                 path->slots[0]--;
2743 
2744         /* find out all the old extents for the file range */
2745         while (1) {
2746                 struct btrfs_file_extent_item *extent;
2747                 struct extent_buffer *l;
2748                 int slot;
2749                 u64 num_bytes;
2750                 u64 offset;
2751                 u64 end;
2752                 u64 disk_bytenr;
2753                 u64 extent_offset;
2754 
2755                 l = path->nodes[0];
2756                 slot = path->slots[0];
2757 
2758                 if (slot >= btrfs_header_nritems(l)) {
2759                         ret = btrfs_next_leaf(root, path);
2760                         if (ret < 0)
2761                                 goto out_free_path;
2762                         else if (ret > 0)
2763                                 break;
2764                         continue;
2765                 }
2766 
2767                 btrfs_item_key_to_cpu(l, &key, slot);
2768 
2769                 if (key.objectid != btrfs_ino(inode))
2770                         break;
2771                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2772                         break;
2773                 if (key.offset >= new->file_pos + new->len)
2774                         break;
2775 
2776                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2777 
2778                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2779                 if (key.offset + num_bytes < new->file_pos)
2780                         goto next;
2781 
2782                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2783                 if (!disk_bytenr)
2784                         goto next;
2785 
2786                 extent_offset = btrfs_file_extent_offset(l, extent);
2787 
2788                 old = kmalloc(sizeof(*old), GFP_NOFS);
2789                 if (!old)
2790                         goto out_free_path;
2791 
2792                 offset = max(new->file_pos, key.offset);
2793                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2794 
2795                 old->bytenr = disk_bytenr;
2796                 old->extent_offset = extent_offset;
2797                 old->offset = offset - key.offset;
2798                 old->len = end - offset;
2799                 old->new = new;
2800                 old->count = 0;
2801                 list_add_tail(&old->list, &new->head);
2802 next:
2803                 path->slots[0]++;
2804                 cond_resched();
2805         }
2806 
2807         btrfs_free_path(path);
2808         atomic_inc(&root->fs_info->defrag_running);
2809 
2810         return new;
2811 
2812 out_free_path:
2813         btrfs_free_path(path);
2814 out_kfree:
2815         free_sa_defrag_extent(new);
2816         return NULL;
2817 }
2818 
2819 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2820                                          u64 start, u64 len)
2821 {
2822         struct btrfs_block_group_cache *cache;
2823 
2824         cache = btrfs_lookup_block_group(root->fs_info, start);
2825         ASSERT(cache);
2826 
2827         spin_lock(&cache->lock);
2828         cache->delalloc_bytes -= len;
2829         spin_unlock(&cache->lock);
2830 
2831         btrfs_put_block_group(cache);
2832 }
2833 
2834 /* as ordered data IO finishes, this gets called so we can finish
2835  * an ordered extent if the range of bytes in the file it covers are
2836  * fully written.
2837  */
2838 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2839 {
2840         struct inode *inode = ordered_extent->inode;
2841         struct btrfs_root *root = BTRFS_I(inode)->root;
2842         struct btrfs_trans_handle *trans = NULL;
2843         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2844         struct extent_state *cached_state = NULL;
2845         struct new_sa_defrag_extent *new = NULL;
2846         int compress_type = 0;
2847         int ret = 0;
2848         u64 logical_len = ordered_extent->len;
2849         bool nolock;
2850         bool truncated = false;
2851 
2852         nolock = btrfs_is_free_space_inode(inode);
2853 
2854         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2855                 ret = -EIO;
2856                 goto out;
2857         }
2858 
2859         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2860                                      ordered_extent->file_offset +
2861                                      ordered_extent->len - 1);
2862 
2863         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2864                 truncated = true;
2865                 logical_len = ordered_extent->truncated_len;
2866                 /* Truncated the entire extent, don't bother adding */
2867                 if (!logical_len)
2868                         goto out;
2869         }
2870 
2871         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2872                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2873 
2874                 /*
2875                  * For mwrite(mmap + memset to write) case, we still reserve
2876                  * space for NOCOW range.
2877                  * As NOCOW won't cause a new delayed ref, just free the space
2878                  */
2879                 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2880                                        ordered_extent->len);
2881                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2882                 if (nolock)
2883                         trans = btrfs_join_transaction_nolock(root);
2884                 else
2885                         trans = btrfs_join_transaction(root);
2886                 if (IS_ERR(trans)) {
2887                         ret = PTR_ERR(trans);
2888                         trans = NULL;
2889                         goto out;
2890                 }
2891                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2892                 ret = btrfs_update_inode_fallback(trans, root, inode);
2893                 if (ret) /* -ENOMEM or corruption */
2894                         btrfs_abort_transaction(trans, root, ret);
2895                 goto out;
2896         }
2897 
2898         lock_extent_bits(io_tree, ordered_extent->file_offset,
2899                          ordered_extent->file_offset + ordered_extent->len - 1,
2900                          &cached_state);
2901 
2902         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2903                         ordered_extent->file_offset + ordered_extent->len - 1,
2904                         EXTENT_DEFRAG, 1, cached_state);
2905         if (ret) {
2906                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2907                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2908                         /* the inode is shared */
2909                         new = record_old_file_extents(inode, ordered_extent);
2910 
2911                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2912                         ordered_extent->file_offset + ordered_extent->len - 1,
2913                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2914         }
2915 
2916         if (nolock)
2917                 trans = btrfs_join_transaction_nolock(root);
2918         else
2919                 trans = btrfs_join_transaction(root);
2920         if (IS_ERR(trans)) {
2921                 ret = PTR_ERR(trans);
2922                 trans = NULL;
2923                 goto out_unlock;
2924         }
2925 
2926         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2927 
2928         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2929                 compress_type = ordered_extent->compress_type;
2930         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2931                 BUG_ON(compress_type);
2932                 ret = btrfs_mark_extent_written(trans, inode,
2933                                                 ordered_extent->file_offset,
2934                                                 ordered_extent->file_offset +
2935                                                 logical_len);
2936         } else {
2937                 BUG_ON(root == root->fs_info->tree_root);
2938                 ret = insert_reserved_file_extent(trans, inode,
2939                                                 ordered_extent->file_offset,
2940                                                 ordered_extent->start,
2941                                                 ordered_extent->disk_len,
2942                                                 logical_len, logical_len,
2943                                                 compress_type, 0, 0,
2944                                                 BTRFS_FILE_EXTENT_REG);
2945                 if (!ret)
2946                         btrfs_release_delalloc_bytes(root,
2947                                                      ordered_extent->start,
2948                                                      ordered_extent->disk_len);
2949         }
2950         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2951                            ordered_extent->file_offset, ordered_extent->len,
2952                            trans->transid);
2953         if (ret < 0) {
2954                 btrfs_abort_transaction(trans, root, ret);
2955                 goto out_unlock;
2956         }
2957 
2958         add_pending_csums(trans, inode, ordered_extent->file_offset,
2959                           &ordered_extent->list);
2960 
2961         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2962         ret = btrfs_update_inode_fallback(trans, root, inode);
2963         if (ret) { /* -ENOMEM or corruption */
2964                 btrfs_abort_transaction(trans, root, ret);
2965                 goto out_unlock;
2966         }
2967         ret = 0;
2968 out_unlock:
2969         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2970                              ordered_extent->file_offset +
2971                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2972 out:
2973         if (root != root->fs_info->tree_root)
2974                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2975         if (trans)
2976                 btrfs_end_transaction(trans, root);
2977 
2978         if (ret || truncated) {
2979                 u64 start, end;
2980 
2981                 if (truncated)
2982                         start = ordered_extent->file_offset + logical_len;
2983                 else
2984                         start = ordered_extent->file_offset;
2985                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2986                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2987 
2988                 /* Drop the cache for the part of the extent we didn't write. */
2989                 btrfs_drop_extent_cache(inode, start, end, 0);
2990 
2991                 /*
2992                  * If the ordered extent had an IOERR or something else went
2993                  * wrong we need to return the space for this ordered extent
2994                  * back to the allocator.  We only free the extent in the
2995                  * truncated case if we didn't write out the extent at all.
2996                  */
2997                 if ((ret || !logical_len) &&
2998                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2999                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3000                         btrfs_free_reserved_extent(root, ordered_extent->start,
3001                                                    ordered_extent->disk_len, 1);
3002         }
3003 
3004 
3005         /*
3006          * This needs to be done to make sure anybody waiting knows we are done
3007          * updating everything for this ordered extent.
3008          */
3009         btrfs_remove_ordered_extent(inode, ordered_extent);
3010 
3011         /* for snapshot-aware defrag */
3012         if (new) {
3013                 if (ret) {
3014                         free_sa_defrag_extent(new);
3015                         atomic_dec(&root->fs_info->defrag_running);
3016                 } else {
3017                         relink_file_extents(new);
3018                 }
3019         }
3020 
3021         /* once for us */
3022         btrfs_put_ordered_extent(ordered_extent);
3023         /* once for the tree */
3024         btrfs_put_ordered_extent(ordered_extent);
3025 
3026         return ret;
3027 }
3028 
3029 static void finish_ordered_fn(struct btrfs_work *work)
3030 {
3031         struct btrfs_ordered_extent *ordered_extent;
3032         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3033         btrfs_finish_ordered_io(ordered_extent);
3034 }
3035 
3036 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3037                                 struct extent_state *state, int uptodate)
3038 {
3039         struct inode *inode = page->mapping->host;
3040         struct btrfs_root *root = BTRFS_I(inode)->root;
3041         struct btrfs_ordered_extent *ordered_extent = NULL;
3042         struct btrfs_workqueue *wq;
3043         btrfs_work_func_t func;
3044 
3045         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3046 
3047         ClearPagePrivate2(page);
3048         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3049                                             end - start + 1, uptodate))
3050                 return 0;
3051 
3052         if (btrfs_is_free_space_inode(inode)) {
3053                 wq = root->fs_info->endio_freespace_worker;
3054                 func = btrfs_freespace_write_helper;
3055         } else {
3056                 wq = root->fs_info->endio_write_workers;
3057                 func = btrfs_endio_write_helper;
3058         }
3059 
3060         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3061                         NULL);
3062         btrfs_queue_work(wq, &ordered_extent->work);
3063 
3064         return 0;
3065 }
3066 
3067 static int __readpage_endio_check(struct inode *inode,
3068                                   struct btrfs_io_bio *io_bio,
3069                                   int icsum, struct page *page,
3070                                   int pgoff, u64 start, size_t len)
3071 {
3072         char *kaddr;
3073         u32 csum_expected;
3074         u32 csum = ~(u32)0;
3075 
3076         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3077 
3078         kaddr = kmap_atomic(page);
3079         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3080         btrfs_csum_final(csum, (char *)&csum);
3081         if (csum != csum_expected)
3082                 goto zeroit;
3083 
3084         kunmap_atomic(kaddr);
3085         return 0;
3086 zeroit:
3087         btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3088                 "csum failed ino %llu off %llu csum %u expected csum %u",
3089                            btrfs_ino(inode), start, csum, csum_expected);
3090         memset(kaddr + pgoff, 1, len);
3091         flush_dcache_page(page);
3092         kunmap_atomic(kaddr);
3093         if (csum_expected == 0)
3094                 return 0;
3095         return -EIO;
3096 }
3097 
3098 /*
3099  * when reads are done, we need to check csums to verify the data is correct
3100  * if there's a match, we allow the bio to finish.  If not, the code in
3101  * extent_io.c will try to find good copies for us.
3102  */
3103 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3104                                       u64 phy_offset, struct page *page,
3105                                       u64 start, u64 end, int mirror)
3106 {
3107         size_t offset = start - page_offset(page);
3108         struct inode *inode = page->mapping->host;
3109         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3110         struct btrfs_root *root = BTRFS_I(inode)->root;
3111 
3112         if (PageChecked(page)) {
3113                 ClearPageChecked(page);
3114                 return 0;
3115         }
3116 
3117         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3118                 return 0;
3119 
3120         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3121             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3122                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3123                 return 0;
3124         }
3125 
3126         phy_offset >>= inode->i_sb->s_blocksize_bits;
3127         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3128                                       start, (size_t)(end - start + 1));
3129 }
3130 
3131 void btrfs_add_delayed_iput(struct inode *inode)
3132 {
3133         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3134         struct btrfs_inode *binode = BTRFS_I(inode);
3135 
3136         if (atomic_add_unless(&inode->i_count, -1, 1))
3137                 return;
3138 
3139         spin_lock(&fs_info->delayed_iput_lock);
3140         if (binode->delayed_iput_count == 0) {
3141                 ASSERT(list_empty(&binode->delayed_iput));
3142                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3143         } else {
3144                 binode->delayed_iput_count++;
3145         }
3146         spin_unlock(&fs_info->delayed_iput_lock);
3147 }
3148 
3149 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3150 {
3151         struct btrfs_fs_info *fs_info = root->fs_info;
3152 
3153         spin_lock(&fs_info->delayed_iput_lock);
3154         while (!list_empty(&fs_info->delayed_iputs)) {
3155                 struct btrfs_inode *inode;
3156 
3157                 inode = list_first_entry(&fs_info->delayed_iputs,
3158                                 struct btrfs_inode, delayed_iput);
3159                 if (inode->delayed_iput_count) {
3160                         inode->delayed_iput_count--;
3161                         list_move_tail(&inode->delayed_iput,
3162                                         &fs_info->delayed_iputs);
3163                 } else {
3164                         list_del_init(&inode->delayed_iput);
3165                 }
3166                 spin_unlock(&fs_info->delayed_iput_lock);
3167                 iput(&inode->vfs_inode);
3168                 spin_lock(&fs_info->delayed_iput_lock);
3169         }
3170         spin_unlock(&fs_info->delayed_iput_lock);
3171 }
3172 
3173 /*
3174  * This is called in transaction commit time. If there are no orphan
3175  * files in the subvolume, it removes orphan item and frees block_rsv
3176  * structure.
3177  */
3178 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3179                               struct btrfs_root *root)
3180 {
3181         struct btrfs_block_rsv *block_rsv;
3182         int ret;
3183 
3184         if (atomic_read(&root->orphan_inodes) ||
3185             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3186                 return;
3187 
3188         spin_lock(&root->orphan_lock);
3189         if (atomic_read(&root->orphan_inodes)) {
3190                 spin_unlock(&root->orphan_lock);
3191                 return;
3192         }
3193 
3194         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3195                 spin_unlock(&root->orphan_lock);
3196                 return;
3197         }
3198 
3199         block_rsv = root->orphan_block_rsv;
3200         root->orphan_block_rsv = NULL;
3201         spin_unlock(&root->orphan_lock);
3202 
3203         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3204             btrfs_root_refs(&root->root_item) > 0) {
3205                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3206                                             root->root_key.objectid);
3207                 if (ret)
3208                         btrfs_abort_transaction(trans, root, ret);
3209                 else
3210                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3211                                   &root->state);
3212         }
3213 
3214         if (block_rsv) {
3215                 WARN_ON(block_rsv->size > 0);
3216                 btrfs_free_block_rsv(root, block_rsv);
3217         }
3218 }
3219 
3220 /*
3221  * This creates an orphan entry for the given inode in case something goes
3222  * wrong in the middle of an unlink/truncate.
3223  *
3224  * NOTE: caller of this function should reserve 5 units of metadata for
3225  *       this function.
3226  */
3227 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3228 {
3229         struct btrfs_root *root = BTRFS_I(inode)->root;
3230         struct btrfs_block_rsv *block_rsv = NULL;
3231         int reserve = 0;
3232         int insert = 0;
3233         int ret;
3234 
3235         if (!root->orphan_block_rsv) {
3236                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3237                 if (!block_rsv)
3238                         return -ENOMEM;
3239         }
3240 
3241         spin_lock(&root->orphan_lock);
3242         if (!root->orphan_block_rsv) {
3243                 root->orphan_block_rsv = block_rsv;
3244         } else if (block_rsv) {
3245                 btrfs_free_block_rsv(root, block_rsv);
3246                 block_rsv = NULL;
3247         }
3248 
3249         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3250                               &BTRFS_I(inode)->runtime_flags)) {
3251 #if 0
3252                 /*
3253                  * For proper ENOSPC handling, we should do orphan
3254                  * cleanup when mounting. But this introduces backward
3255                  * compatibility issue.
3256                  */
3257                 if (!xchg(&root->orphan_item_inserted, 1))
3258                         insert = 2;
3259                 else
3260                         insert = 1;
3261 #endif
3262                 insert = 1;
3263                 atomic_inc(&root->orphan_inodes);
3264         }
3265 
3266         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3267                               &BTRFS_I(inode)->runtime_flags))
3268                 reserve = 1;
3269         spin_unlock(&root->orphan_lock);
3270 
3271         /* grab metadata reservation from transaction handle */
3272         if (reserve) {
3273                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3274                 ASSERT(!ret);
3275                 if (ret) {
3276                         atomic_dec(&root->orphan_inodes);
3277                         clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3278                                   &BTRFS_I(inode)->runtime_flags);
3279                         if (insert)
3280                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3281                                           &BTRFS_I(inode)->runtime_flags);
3282                         return ret;
3283                 }
3284         }
3285 
3286         /* insert an orphan item to track this unlinked/truncated file */
3287         if (insert >= 1) {
3288                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3289                 if (ret) {
3290                         atomic_dec(&root->orphan_inodes);
3291                         if (reserve) {
3292                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3293                                           &BTRFS_I(inode)->runtime_flags);
3294                                 btrfs_orphan_release_metadata(inode);
3295                         }
3296                         if (ret != -EEXIST) {
3297                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3298                                           &BTRFS_I(inode)->runtime_flags);
3299                                 btrfs_abort_transaction(trans, root, ret);
3300                                 return ret;
3301                         }
3302                 }
3303                 ret = 0;
3304         }
3305 
3306         /* insert an orphan item to track subvolume contains orphan files */
3307         if (insert >= 2) {
3308                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3309                                                root->root_key.objectid);
3310                 if (ret && ret != -EEXIST) {
3311                         btrfs_abort_transaction(trans, root, ret);
3312                         return ret;
3313                 }
3314         }
3315         return 0;
3316 }
3317 
3318 /*
3319  * We have done the truncate/delete so we can go ahead and remove the orphan
3320  * item for this particular inode.
3321  */
3322 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3323                             struct inode *inode)
3324 {
3325         struct btrfs_root *root = BTRFS_I(inode)->root;
3326         int delete_item = 0;
3327         int release_rsv = 0;
3328         int ret = 0;
3329 
3330         spin_lock(&root->orphan_lock);
3331         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3332                                &BTRFS_I(inode)->runtime_flags))
3333                 delete_item = 1;
3334 
3335         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3336                                &BTRFS_I(inode)->runtime_flags))
3337                 release_rsv = 1;
3338         spin_unlock(&root->orphan_lock);
3339 
3340         if (delete_item) {
3341                 atomic_dec(&root->orphan_inodes);
3342                 if (trans)
3343                         ret = btrfs_del_orphan_item(trans, root,
3344                                                     btrfs_ino(inode));
3345         }
3346 
3347         if (release_rsv)
3348                 btrfs_orphan_release_metadata(inode);
3349 
3350         return ret;
3351 }
3352 
3353 /*
3354  * this cleans up any orphans that may be left on the list from the last use
3355  * of this root.
3356  */
3357 int btrfs_orphan_cleanup(struct btrfs_root *root)
3358 {
3359         struct btrfs_path *path;
3360         struct extent_buffer *leaf;
3361         struct btrfs_key key, found_key;
3362         struct btrfs_trans_handle *trans;
3363         struct inode *inode;
3364         u64 last_objectid = 0;
3365         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3366 
3367         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3368                 return 0;
3369 
3370         path = btrfs_alloc_path();
3371         if (!path) {
3372                 ret = -ENOMEM;
3373                 goto out;
3374         }
3375         path->reada = READA_BACK;
3376 
3377         key.objectid = BTRFS_ORPHAN_OBJECTID;
3378         key.type = BTRFS_ORPHAN_ITEM_KEY;
3379         key.offset = (u64)-1;
3380 
3381         while (1) {
3382                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3383                 if (ret < 0)
3384                         goto out;
3385 
3386                 /*
3387                  * if ret == 0 means we found what we were searching for, which
3388                  * is weird, but possible, so only screw with path if we didn't
3389                  * find the key and see if we have stuff that matches
3390                  */
3391                 if (ret > 0) {
3392                         ret = 0;
3393                         if (path->slots[0] == 0)
3394                                 break;
3395                         path->slots[0]--;
3396                 }
3397 
3398                 /* pull out the item */
3399                 leaf = path->nodes[0];
3400                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3401 
3402                 /* make sure the item matches what we want */
3403                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3404                         break;
3405                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3406                         break;
3407 
3408                 /* release the path since we're done with it */
3409                 btrfs_release_path(path);
3410 
3411                 /*
3412                  * this is where we are basically btrfs_lookup, without the
3413                  * crossing root thing.  we store the inode number in the
3414                  * offset of the orphan item.
3415                  */
3416 
3417                 if (found_key.offset == last_objectid) {
3418                         btrfs_err(root->fs_info,
3419                                 "Error removing orphan entry, stopping orphan cleanup");
3420                         ret = -EINVAL;
3421                         goto out;
3422                 }
3423 
3424                 last_objectid = found_key.offset;
3425 
3426                 found_key.objectid = found_key.offset;
3427                 found_key.type = BTRFS_INODE_ITEM_KEY;
3428                 found_key.offset = 0;
3429                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3430                 ret = PTR_ERR_OR_ZERO(inode);
3431                 if (ret && ret != -ESTALE)
3432                         goto out;
3433 
3434                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3435                         struct btrfs_root *dead_root;
3436                         struct btrfs_fs_info *fs_info = root->fs_info;
3437                         int is_dead_root = 0;
3438 
3439                         /*
3440                          * this is an orphan in the tree root. Currently these
3441                          * could come from 2 sources:
3442                          *  a) a snapshot deletion in progress
3443                          *  b) a free space cache inode
3444                          * We need to distinguish those two, as the snapshot
3445                          * orphan must not get deleted.
3446                          * find_dead_roots already ran before us, so if this
3447                          * is a snapshot deletion, we should find the root
3448                          * in the dead_roots list
3449                          */
3450                         spin_lock(&fs_info->trans_lock);
3451                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3452                                             root_list) {
3453                                 if (dead_root->root_key.objectid ==
3454                                     found_key.objectid) {
3455                                         is_dead_root = 1;
3456                                         break;
3457                                 }
3458                         }
3459                         spin_unlock(&fs_info->trans_lock);
3460                         if (is_dead_root) {
3461                                 /* prevent this orphan from being found again */
3462                                 key.offset = found_key.objectid - 1;
3463                                 continue;
3464                         }
3465                 }
3466                 /*
3467                  * Inode is already gone but the orphan item is still there,
3468                  * kill the orphan item.
3469                  */
3470                 if (ret == -ESTALE) {
3471                         trans = btrfs_start_transaction(root, 1);
3472                         if (IS_ERR(trans)) {
3473                                 ret = PTR_ERR(trans);
3474                                 goto out;
3475                         }
3476                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3477                                 found_key.objectid);
3478                         ret = btrfs_del_orphan_item(trans, root,
3479                                                     found_key.objectid);
3480                         btrfs_end_transaction(trans, root);
3481                         if (ret)
3482                                 goto out;
3483                         continue;
3484                 }
3485 
3486                 /*
3487                  * add this inode to the orphan list so btrfs_orphan_del does
3488                  * the proper thing when we hit it
3489                  */
3490                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3491                         &BTRFS_I(inode)->runtime_flags);
3492                 atomic_inc(&root->orphan_inodes);
3493 
3494                 /* if we have links, this was a truncate, lets do that */
3495                 if (inode->i_nlink) {
3496                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3497                                 iput(inode);
3498                                 continue;
3499                         }
3500                         nr_truncate++;
3501 
3502                         /* 1 for the orphan item deletion. */
3503                         trans = btrfs_start_transaction(root, 1);
3504                         if (IS_ERR(trans)) {
3505                                 iput(inode);
3506                                 ret = PTR_ERR(trans);
3507                                 goto out;
3508                         }
3509                         ret = btrfs_orphan_add(trans, inode);
3510                         btrfs_end_transaction(trans, root);
3511                         if (ret) {
3512                                 iput(inode);
3513                                 goto out;
3514                         }
3515 
3516                         ret = btrfs_truncate(inode);
3517                         if (ret)
3518                                 btrfs_orphan_del(NULL, inode);
3519                 } else {
3520                         nr_unlink++;
3521                 }
3522 
3523                 /* this will do delete_inode and everything for us */
3524                 iput(inode);
3525                 if (ret)
3526                         goto out;
3527         }
3528         /* release the path since we're done with it */
3529         btrfs_release_path(path);
3530 
3531         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3532 
3533         if (root->orphan_block_rsv)
3534                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3535                                         (u64)-1);
3536 
3537         if (root->orphan_block_rsv ||
3538             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3539                 trans = btrfs_join_transaction(root);
3540                 if (!IS_ERR(trans))
3541                         btrfs_end_transaction(trans, root);
3542         }
3543 
3544         if (nr_unlink)
3545                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3546         if (nr_truncate)
3547                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3548 
3549 out:
3550         if (ret)
3551                 btrfs_err(root->fs_info,
3552                         "could not do orphan cleanup %d", ret);
3553         btrfs_free_path(path);
3554         return ret;
3555 }
3556 
3557 /*
3558  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3559  * don't find any xattrs, we know there can't be any acls.
3560  *
3561  * slot is the slot the inode is in, objectid is the objectid of the inode
3562  */
3563 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3564                                           int slot, u64 objectid,
3565                                           int *first_xattr_slot)
3566 {
3567         u32 nritems = btrfs_header_nritems(leaf);
3568         struct btrfs_key found_key;
3569         static u64 xattr_access = 0;
3570         static u64 xattr_default = 0;
3571         int scanned = 0;
3572 
3573         if (!xattr_access) {
3574                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3575                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3576                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3577                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3578         }
3579 
3580         slot++;
3581         *first_xattr_slot = -1;
3582         while (slot < nritems) {
3583                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3584 
3585                 /* we found a different objectid, there must not be acls */
3586                 if (found_key.objectid != objectid)
3587                         return 0;
3588 
3589                 /* we found an xattr, assume we've got an acl */
3590                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3591                         if (*first_xattr_slot == -1)
3592                                 *first_xattr_slot = slot;
3593                         if (found_key.offset == xattr_access ||
3594                             found_key.offset == xattr_default)
3595                                 return 1;
3596                 }
3597 
3598                 /*
3599                  * we found a key greater than an xattr key, there can't
3600                  * be any acls later on
3601                  */
3602                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3603                         return 0;
3604 
3605                 slot++;
3606                 scanned++;
3607 
3608                 /*
3609                  * it goes inode, inode backrefs, xattrs, extents,
3610                  * so if there are a ton of hard links to an inode there can
3611                  * be a lot of backrefs.  Don't waste time searching too hard,
3612                  * this is just an optimization
3613                  */
3614                 if (scanned >= 8)
3615                         break;
3616         }
3617         /* we hit the end of the leaf before we found an xattr or
3618          * something larger than an xattr.  We have to assume the inode
3619          * has acls
3620          */
3621         if (*first_xattr_slot == -1)
3622                 *first_xattr_slot = slot;
3623         return 1;
3624 }
3625 
3626 /*
3627  * read an inode from the btree into the in-memory inode
3628  */
3629 static void btrfs_read_locked_inode(struct inode *inode)
3630 {
3631         struct btrfs_path *path;
3632         struct extent_buffer *leaf;
3633         struct btrfs_inode_item *inode_item;
3634         struct btrfs_root *root = BTRFS_I(inode)->root;
3635         struct btrfs_key location;
3636         unsigned long ptr;
3637         int maybe_acls;
3638         u32 rdev;
3639         int ret;
3640         bool filled = false;
3641         int first_xattr_slot;
3642 
3643         ret = btrfs_fill_inode(inode, &rdev);
3644         if (!ret)
3645                 filled = true;
3646 
3647         path = btrfs_alloc_path();
3648         if (!path)
3649                 goto make_bad;
3650 
3651         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3652 
3653         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3654         if (ret)
3655                 goto make_bad;
3656 
3657         leaf = path->nodes[0];
3658 
3659         if (filled)
3660                 goto cache_index;
3661 
3662         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3663                                     struct btrfs_inode_item);
3664         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3665         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3666         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3667         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3668         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3669 
3670         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3671         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3672 
3673         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3674         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3675 
3676         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3677         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3678 
3679         BTRFS_I(inode)->i_otime.tv_sec =
3680                 btrfs_timespec_sec(leaf, &inode_item->otime);
3681         BTRFS_I(inode)->i_otime.tv_nsec =
3682                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3683 
3684         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3685         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3686         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3687 
3688         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3689         inode->i_generation = BTRFS_I(inode)->generation;
3690         inode->i_rdev = 0;
3691         rdev = btrfs_inode_rdev(leaf, inode_item);
3692 
3693         BTRFS_I(inode)->index_cnt = (u64)-1;
3694         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3695 
3696 cache_index:
3697         /*
3698          * If we were modified in the current generation and evicted from memory
3699          * and then re-read we need to do a full sync since we don't have any
3700          * idea about which extents were modified before we were evicted from
3701          * cache.
3702          *
3703          * This is required for both inode re-read from disk and delayed inode
3704          * in delayed_nodes_tree.
3705          */
3706         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3707                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3708                         &BTRFS_I(inode)->runtime_flags);
3709 
3710         /*
3711          * We don't persist the id of the transaction where an unlink operation
3712          * against the inode was last made. So here we assume the inode might
3713          * have been evicted, and therefore the exact value of last_unlink_trans
3714          * lost, and set it to last_trans to avoid metadata inconsistencies
3715          * between the inode and its parent if the inode is fsync'ed and the log
3716          * replayed. For example, in the scenario:
3717          *
3718          * touch mydir/foo
3719          * ln mydir/foo mydir/bar
3720          * sync
3721          * unlink mydir/bar
3722          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3723          * xfs_io -c fsync mydir/foo
3724          * <power failure>
3725          * mount fs, triggers fsync log replay
3726          *
3727          * We must make sure that when we fsync our inode foo we also log its
3728          * parent inode, otherwise after log replay the parent still has the
3729          * dentry with the "bar" name but our inode foo has a link count of 1
3730          * and doesn't have an inode ref with the name "bar" anymore.
3731          *
3732          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3733          * but it guarantees correctness at the expense of occasional full
3734          * transaction commits on fsync if our inode is a directory, or if our
3735          * inode is not a directory, logging its parent unnecessarily.
3736          */
3737         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3738 
3739         path->slots[0]++;
3740         if (inode->i_nlink != 1 ||
3741             path->slots[0] >= btrfs_header_nritems(leaf))
3742                 goto cache_acl;
3743 
3744         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3745         if (location.objectid != btrfs_ino(inode))
3746                 goto cache_acl;
3747 
3748         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3749         if (location.type == BTRFS_INODE_REF_KEY) {
3750                 struct btrfs_inode_ref *ref;
3751 
3752                 ref = (struct btrfs_inode_ref *)ptr;
3753                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3754         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3755                 struct btrfs_inode_extref *extref;
3756 
3757                 extref = (struct btrfs_inode_extref *)ptr;
3758                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3759                                                                      extref);
3760         }
3761 cache_acl:
3762         /*
3763          * try to precache a NULL acl entry for files that don't have
3764          * any xattrs or acls
3765          */
3766         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3767                                            btrfs_ino(inode), &first_xattr_slot);
3768         if (first_xattr_slot != -1) {
3769                 path->slots[0] = first_xattr_slot;
3770                 ret = btrfs_load_inode_props(inode, path);
3771                 if (ret)
3772                         btrfs_err(root->fs_info,
3773                                   "error loading props for ino %llu (root %llu): %d",
3774                                   btrfs_ino(inode),
3775                                   root->root_key.objectid, ret);
3776         }
3777         btrfs_free_path(path);
3778 
3779         if (!maybe_acls)
3780                 cache_no_acl(inode);
3781 
3782         switch (inode->i_mode & S_IFMT) {
3783         case S_IFREG:
3784                 inode->i_mapping->a_ops = &btrfs_aops;
3785                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3786                 inode->i_fop = &btrfs_file_operations;
3787                 inode->i_op = &btrfs_file_inode_operations;
3788                 break;
3789         case S_IFDIR:
3790                 inode->i_fop = &btrfs_dir_file_operations;
3791                 if (root == root->fs_info->tree_root)
3792                         inode->i_op = &btrfs_dir_ro_inode_operations;
3793                 else
3794                         inode->i_op = &btrfs_dir_inode_operations;
3795                 break;
3796         case S_IFLNK:
3797                 inode->i_op = &btrfs_symlink_inode_operations;
3798                 inode_nohighmem(inode);
3799                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3800                 break;
3801         default:
3802                 inode->i_op = &btrfs_special_inode_operations;
3803                 init_special_inode(inode, inode->i_mode, rdev);
3804                 break;
3805         }
3806 
3807         btrfs_update_iflags(inode);
3808         return;
3809 
3810 make_bad:
3811         btrfs_free_path(path);
3812         make_bad_inode(inode);
3813 }
3814 
3815 /*
3816  * given a leaf and an inode, copy the inode fields into the leaf
3817  */
3818 static void fill_inode_item(struct btrfs_trans_handle *trans,
3819                             struct extent_buffer *leaf,
3820                             struct btrfs_inode_item *item,
3821                             struct inode *inode)
3822 {
3823         struct btrfs_map_token token;
3824 
3825         btrfs_init_map_token(&token);
3826 
3827         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3828         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3829         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3830                                    &token);
3831         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3832         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3833 
3834         btrfs_set_token_timespec_sec(leaf, &item->atime,
3835                                      inode->i_atime.tv_sec, &token);
3836         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3837                                       inode->i_atime.tv_nsec, &token);
3838 
3839         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3840                                      inode->i_mtime.tv_sec, &token);
3841         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3842                                       inode->i_mtime.tv_nsec, &token);
3843 
3844         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3845                                      inode->i_ctime.tv_sec, &token);
3846         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3847                                       inode->i_ctime.tv_nsec, &token);
3848 
3849         btrfs_set_token_timespec_sec(leaf, &item->otime,
3850                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3851         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3852                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3853 
3854         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3855                                      &token);
3856         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3857                                          &token);
3858         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3859         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3860         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3861         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3862         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3863 }
3864 
3865 /*
3866  * copy everything in the in-memory inode into the btree.
3867  */
3868 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3869                                 struct btrfs_root *root, struct inode *inode)
3870 {
3871         struct btrfs_inode_item *inode_item;
3872         struct btrfs_path *path;
3873         struct extent_buffer *leaf;
3874         int ret;
3875 
3876         path = btrfs_alloc_path();
3877         if (!path)
3878                 return -ENOMEM;
3879 
3880         path->leave_spinning = 1;
3881         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3882                                  1);
3883         if (ret) {
3884                 if (ret > 0)
3885                         ret = -ENOENT;
3886                 goto failed;
3887         }
3888 
3889         leaf = path->nodes[0];
3890         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3891                                     struct btrfs_inode_item);
3892 
3893         fill_inode_item(trans, leaf, inode_item, inode);
3894         btrfs_mark_buffer_dirty(leaf);
3895         btrfs_set_inode_last_trans(trans, inode);
3896         ret = 0;
3897 failed:
3898         btrfs_free_path(path);
3899         return ret;
3900 }
3901 
3902 /*
3903  * copy everything in the in-memory inode into the btree.
3904  */
3905 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3906                                 struct btrfs_root *root, struct inode *inode)
3907 {
3908         int ret;
3909 
3910         /*
3911          * If the inode is a free space inode, we can deadlock during commit
3912          * if we put it into the delayed code.
3913          *
3914          * The data relocation inode should also be directly updated
3915          * without delay
3916          */
3917         if (!btrfs_is_free_space_inode(inode)
3918             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3919             && !root->fs_info->log_root_recovering) {
3920                 btrfs_update_root_times(trans, root);
3921 
3922                 ret = btrfs_delayed_update_inode(trans, root, inode);
3923                 if (!ret)
3924                         btrfs_set_inode_last_trans(trans, inode);
3925                 return ret;
3926         }
3927 
3928         return btrfs_update_inode_item(trans, root, inode);
3929 }
3930 
3931 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3932                                          struct btrfs_root *root,
3933                                          struct inode *inode)
3934 {
3935         int ret;
3936 
3937         ret = btrfs_update_inode(trans, root, inode);
3938         if (ret == -ENOSPC)
3939                 return btrfs_update_inode_item(trans, root, inode);
3940         return ret;
3941 }
3942 
3943 /*
3944  * unlink helper that gets used here in inode.c and in the tree logging
3945  * recovery code.  It remove a link in a directory with a given name, and
3946  * also drops the back refs in the inode to the directory
3947  */
3948 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3949                                 struct btrfs_root *root,
3950                                 struct inode *dir, struct inode *inode,
3951                                 const char *name, int name_len)
3952 {
3953         struct btrfs_path *path;
3954         int ret = 0;
3955         struct extent_buffer *leaf;
3956         struct btrfs_dir_item *di;
3957         struct btrfs_key key;
3958         u64 index;
3959         u64 ino = btrfs_ino(inode);
3960         u64 dir_ino = btrfs_ino(dir);
3961 
3962         path = btrfs_alloc_path();
3963         if (!path) {
3964                 ret = -ENOMEM;
3965                 goto out;
3966         }
3967 
3968         path->leave_spinning = 1;
3969         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3970                                     name, name_len, -1);
3971         if (IS_ERR(di)) {
3972                 ret = PTR_ERR(di);
3973                 goto err;
3974         }
3975         if (!di) {
3976                 ret = -ENOENT;
3977                 goto err;
3978         }
3979         leaf = path->nodes[0];
3980         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3981         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3982         if (ret)
3983                 goto err;
3984         btrfs_release_path(path);
3985 
3986         /*
3987          * If we don't have dir index, we have to get it by looking up
3988          * the inode ref, since we get the inode ref, remove it directly,
3989          * it is unnecessary to do delayed deletion.
3990          *
3991          * But if we have dir index, needn't search inode ref to get it.
3992          * Since the inode ref is close to the inode item, it is better
3993          * that we delay to delete it, and just do this deletion when
3994          * we update the inode item.
3995          */
3996         if (BTRFS_I(inode)->dir_index) {
3997                 ret = btrfs_delayed_delete_inode_ref(inode);
3998                 if (!ret) {
3999                         index = BTRFS_I(inode)->dir_index;
4000                         goto skip_backref;
4001                 }
4002         }
4003 
4004         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4005                                   dir_ino, &index);
4006         if (ret) {
4007                 btrfs_info(root->fs_info,
4008                         "failed to delete reference to %.*s, inode %llu parent %llu",
4009                         name_len, name, ino, dir_ino);
4010                 btrfs_abort_transaction(trans, root, ret);
4011                 goto err;
4012         }
4013 skip_backref:
4014         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4015         if (ret) {
4016                 btrfs_abort_transaction(trans, root, ret);
4017                 goto err;
4018         }
4019 
4020         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4021                                          inode, dir_ino);
4022         if (ret != 0 && ret != -ENOENT) {
4023                 btrfs_abort_transaction(trans, root, ret);
4024                 goto err;
4025         }
4026 
4027         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4028                                            dir, index);
4029         if (ret == -ENOENT)
4030                 ret = 0;
4031         else if (ret)
4032                 btrfs_abort_transaction(trans, root, ret);
4033 err:
4034         btrfs_free_path(path);
4035         if (ret)
4036                 goto out;
4037 
4038         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4039         inode_inc_iversion(inode);
4040         inode_inc_iversion(dir);
4041         inode->i_ctime = dir->i_mtime =
4042                 dir->i_ctime = current_fs_time(inode->i_sb);
4043         ret = btrfs_update_inode(trans, root, dir);
4044 out:
4045         return ret;
4046 }
4047 
4048 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4049                        struct btrfs_root *root,
4050                        struct inode *dir, struct inode *inode,
4051                        const char *name, int name_len)
4052 {
4053         int ret;
4054         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4055         if (!ret) {
4056                 drop_nlink(inode);
4057                 ret = btrfs_update_inode(trans, root, inode);
4058         }
4059         return ret;
4060 }
4061 
4062 /*
4063  * helper to start transaction for unlink and rmdir.
4064  *
4065  * unlink and rmdir are special in btrfs, they do not always free space, so
4066  * if we cannot make our reservations the normal way try and see if there is
4067  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4068  * allow the unlink to occur.
4069  */
4070 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4071 {
4072         struct btrfs_root *root = BTRFS_I(dir)->root;
4073 
4074         /*
4075          * 1 for the possible orphan item
4076          * 1 for the dir item
4077          * 1 for the dir index
4078          * 1 for the inode ref
4079          * 1 for the inode
4080          */
4081         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4082 }
4083 
4084 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4085 {
4086         struct btrfs_root *root = BTRFS_I(dir)->root;
4087         struct btrfs_trans_handle *trans;
4088         struct inode *inode = d_inode(dentry);
4089         int ret;
4090 
4091         trans = __unlink_start_trans(dir);
4092         if (IS_ERR(trans))
4093                 return PTR_ERR(trans);
4094 
4095         btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4096 
4097         ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4098                                  dentry->d_name.name, dentry->d_name.len);
4099         if (ret)
4100                 goto out;
4101 
4102         if (inode->i_nlink == 0) {
4103                 ret = btrfs_orphan_add(trans, inode);
4104                 if (ret)
4105                         goto out;
4106         }
4107 
4108 out:
4109         btrfs_end_transaction(trans, root);
4110         btrfs_btree_balance_dirty(root);
4111         return ret;
4112 }
4113 
4114 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4115                         struct btrfs_root *root,
4116                         struct inode *dir, u64 objectid,
4117                         const char *name, int name_len)
4118 {
4119         struct btrfs_path *path;
4120         struct extent_buffer *leaf;
4121         struct btrfs_dir_item *di;
4122         struct btrfs_key key;
4123         u64 index;
4124         int ret;
4125         u64 dir_ino = btrfs_ino(dir);
4126 
4127         path = btrfs_alloc_path();
4128         if (!path)
4129                 return -ENOMEM;
4130 
4131         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4132                                    name, name_len, -1);
4133         if (IS_ERR_OR_NULL(di)) {
4134                 if (!di)
4135                         ret = -ENOENT;
4136                 else
4137                         ret = PTR_ERR(di);
4138                 goto out;
4139         }
4140 
4141         leaf = path->nodes[0];
4142         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4143         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4144         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4145         if (ret) {
4146                 btrfs_abort_transaction(trans, root, ret);
4147                 goto out;
4148         }
4149         btrfs_release_path(path);
4150 
4151         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4152                                  objectid, root->root_key.objectid,
4153                                  dir_ino, &index, name, name_len);
4154         if (ret < 0) {
4155                 if (ret != -ENOENT) {
4156                         btrfs_abort_transaction(trans, root, ret);
4157                         goto out;
4158                 }
4159                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4160                                                  name, name_len);
4161                 if (IS_ERR_OR_NULL(di)) {
4162                         if (!di)
4163                                 ret = -ENOENT;
4164                         else
4165                                 ret = PTR_ERR(di);
4166                         btrfs_abort_transaction(trans, root, ret);
4167                         goto out;
4168                 }
4169 
4170                 leaf = path->nodes[0];
4171                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4172                 btrfs_release_path(path);
4173                 index = key.offset;
4174         }
4175         btrfs_release_path(path);
4176 
4177         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4178         if (ret) {
4179                 btrfs_abort_transaction(trans, root, ret);
4180                 goto out;
4181         }
4182 
4183         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4184         inode_inc_iversion(dir);
4185         dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4186         ret = btrfs_update_inode_fallback(trans, root, dir);
4187         if (ret)
4188                 btrfs_abort_transaction(trans, root, ret);
4189 out:
4190         btrfs_free_path(path);
4191         return ret;
4192 }
4193 
4194 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4195 {
4196         struct inode *inode = d_inode(dentry);
4197         int err = 0;
4198         struct btrfs_root *root = BTRFS_I(dir)->root;
4199         struct btrfs_trans_handle *trans;
4200 
4201         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4202                 return -ENOTEMPTY;
4203         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4204                 return -EPERM;
4205 
4206         trans = __unlink_start_trans(dir);
4207         if (IS_ERR(trans))
4208                 return PTR_ERR(trans);
4209 
4210         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4211                 err = btrfs_unlink_subvol(trans, root, dir,
4212                                           BTRFS_I(inode)->location.objectid,
4213                                           dentry->d_name.name,
4214                                           dentry->d_name.len);
4215                 goto out;
4216         }
4217 
4218         err = btrfs_orphan_add(trans, inode);
4219         if (err)
4220                 goto out;
4221 
4222         /* now the directory is empty */
4223         err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4224                                  dentry->d_name.name, dentry->d_name.len);
4225         if (!err)
4226                 btrfs_i_size_write(inode, 0);
4227 out:
4228         btrfs_end_transaction(trans, root);
4229         btrfs_btree_balance_dirty(root);
4230 
4231         return err;
4232 }
4233 
4234 static int truncate_space_check(struct btrfs_trans_handle *trans,
4235                                 struct btrfs_root *root,
4236                                 u64 bytes_deleted)
4237 {
4238         int ret;
4239 
4240         /*
4241          * This is only used to apply pressure to the enospc system, we don't
4242          * intend to use this reservation at all.
4243          */
4244         bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4245         bytes_deleted *= root->nodesize;
4246         ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4247                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4248         if (!ret) {
4249                 trace_btrfs_space_reservation(root->fs_info, "transaction",
4250                                               trans->transid,
4251                                               bytes_deleted, 1);
4252                 trans->bytes_reserved += bytes_deleted;
4253         }
4254         return ret;
4255 
4256 }
4257 
4258 static int truncate_inline_extent(struct inode *inode,
4259                                   struct btrfs_path *path,
4260                                   struct btrfs_key *found_key,
4261                                   const u64 item_end,
4262                                   const u64 new_size)
4263 {
4264         struct extent_buffer *leaf = path->nodes[0];
4265         int slot = path->slots[0];
4266         struct btrfs_file_extent_item *fi;
4267         u32 size = (u32)(new_size - found_key->offset);
4268         struct btrfs_root *root = BTRFS_I(inode)->root;
4269 
4270         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4271 
4272         if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4273                 loff_t offset = new_size;
4274                 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4275 
4276                 /*
4277                  * Zero out the remaining of the last page of our inline extent,
4278                  * instead of directly truncating our inline extent here - that
4279                  * would be much more complex (decompressing all the data, then
4280                  * compressing the truncated data, which might be bigger than
4281                  * the size of the inline extent, resize the extent, etc).
4282                  * We release the path because to get the page we might need to
4283                  * read the extent item from disk (data not in the page cache).
4284                  */
4285                 btrfs_release_path(path);
4286                 return btrfs_truncate_block(inode, offset, page_end - offset,
4287                                         0);
4288         }
4289 
4290         btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4291         size = btrfs_file_extent_calc_inline_size(size);
4292         btrfs_truncate_item(root, path, size, 1);
4293 
4294         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4295                 inode_sub_bytes(inode, item_end + 1 - new_size);
4296 
4297         return 0;
4298 }
4299 
4300 /*
4301  * this can truncate away extent items, csum items and directory items.
4302  * It starts at a high offset and removes keys until it can't find
4303  * any higher than new_size
4304  *
4305  * csum items that cross the new i_size are truncated to the new size
4306  * as well.
4307  *
4308  * min_type is the minimum key type to truncate down to.  If set to 0, this
4309  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4310  */
4311 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4312                                struct btrfs_root *root,
4313                                struct inode *inode,
4314                                u64 new_size, u32 min_type)
4315 {
4316         struct btrfs_path *path;
4317         struct extent_buffer *leaf;
4318         struct btrfs_file_extent_item *fi;
4319         struct btrfs_key key;
4320         struct btrfs_key found_key;
4321         u64 extent_start = 0;
4322         u64 extent_num_bytes = 0;
4323         u64 extent_offset = 0;
4324         u64 item_end = 0;
4325         u64 last_size = new_size;
4326         u32 found_type = (u8)-1;
4327         int found_extent;
4328         int del_item;
4329         int pending_del_nr = 0;
4330         int pending_del_slot = 0;
4331         int extent_type = -1;
4332         int ret;
4333         int err = 0;
4334         u64 ino = btrfs_ino(inode);
4335         u64 bytes_deleted = 0;
4336         bool be_nice = 0;
4337         bool should_throttle = 0;
4338         bool should_end = 0;
4339 
4340         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4341 
4342         /*
4343          * for non-free space inodes and ref cows, we want to back off from
4344          * time to time
4345          */
4346         if (!btrfs_is_free_space_inode(inode) &&
4347             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4348                 be_nice = 1;
4349 
4350         path = btrfs_alloc_path();
4351         if (!path)
4352                 return -ENOMEM;
4353         path->reada = READA_BACK;
4354 
4355         /*
4356          * We want to drop from the next block forward in case this new size is
4357          * not block aligned since we will be keeping the last block of the
4358          * extent just the way it is.
4359          */
4360         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4361             root == root->fs_info->tree_root)
4362                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4363                                         root->sectorsize), (u64)-1, 0);
4364 
4365         /*
4366          * This function is also used to drop the items in the log tree before
4367          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4368          * it is used to drop the loged items. So we shouldn't kill the delayed
4369          * items.
4370          */
4371         if (min_type == 0 && root == BTRFS_I(inode)->root)
4372                 btrfs_kill_delayed_inode_items(inode);
4373 
4374         key.objectid = ino;
4375         key.offset = (u64)-1;
4376         key.type = (u8)-1;
4377 
4378 search_again:
4379         /*
4380          * with a 16K leaf size and 128MB extents, you can actually queue
4381          * up a huge file in a single leaf.  Most of the time that
4382          * bytes_deleted is > 0, it will be huge by the time we get here
4383          */
4384         if (be_nice && bytes_deleted > SZ_32M) {
4385                 if (btrfs_should_end_transaction(trans, root)) {
4386                         err = -EAGAIN;
4387                         goto error;
4388                 }
4389         }
4390 
4391 
4392         path->leave_spinning = 1;
4393         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4394         if (ret < 0) {
4395                 err = ret;
4396                 goto out;
4397         }
4398 
4399         if (ret > 0) {
4400                 /* there are no items in the tree for us to truncate, we're
4401                  * done
4402                  */
4403                 if (path->slots[0] == 0)
4404                         goto out;
4405                 path->slots[0]--;
4406         }
4407 
4408         while (1) {
4409                 fi = NULL;
4410                 leaf = path->nodes[0];
4411                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4412                 found_type = found_key.type;
4413 
4414                 if (found_key.objectid != ino)
4415                         break;
4416 
4417                 if (found_type < min_type)
4418                         break;
4419 
4420                 item_end = found_key.offset;
4421                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4422                         fi = btrfs_item_ptr(leaf, path->slots[0],
4423                                             struct btrfs_file_extent_item);
4424                         extent_type = btrfs_file_extent_type(leaf, fi);
4425                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4426                                 item_end +=
4427                                     btrfs_file_extent_num_bytes(leaf, fi);
4428                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4429                                 item_end += btrfs_file_extent_inline_len(leaf,
4430                                                          path->slots[0], fi);
4431                         }
4432                         item_end--;
4433                 }
4434                 if (found_type > min_type) {
4435                         del_item = 1;
4436                 } else {
4437                         if (item_end < new_size)
4438                                 break;
4439                         if (found_key.offset >= new_size)
4440                                 del_item = 1;
4441                         else
4442                                 del_item = 0;
4443                 }
4444                 found_extent = 0;
4445                 /* FIXME, shrink the extent if the ref count is only 1 */
4446                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4447                         goto delete;
4448 
4449                 if (del_item)
4450                         last_size = found_key.offset;
4451                 else
4452                         last_size = new_size;
4453 
4454                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4455                         u64 num_dec;
4456                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4457                         if (!del_item) {
4458                                 u64 orig_num_bytes =
4459                                         btrfs_file_extent_num_bytes(leaf, fi);
4460                                 extent_num_bytes = ALIGN(new_size -
4461                                                 found_key.offset,
4462                                                 root->sectorsize);
4463                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4464                                                          extent_num_bytes);
4465                                 num_dec = (orig_num_bytes -
4466                                            extent_num_bytes);
4467                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4468                                              &root->state) &&
4469                                     extent_start != 0)
4470                                         inode_sub_bytes(inode, num_dec);
4471                                 btrfs_mark_buffer_dirty(leaf);
4472                         } else {
4473                                 extent_num_bytes =
4474                                         btrfs_file_extent_disk_num_bytes(leaf,
4475                                                                          fi);
4476                                 extent_offset = found_key.offset -
4477                                         btrfs_file_extent_offset(leaf, fi);
4478 
4479                                 /* FIXME blocksize != 4096 */
4480                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4481                                 if (extent_start != 0) {
4482                                         found_extent = 1;
4483                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4484                                                      &root->state))
4485                                                 inode_sub_bytes(inode, num_dec);
4486                                 }
4487                         }
4488                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4489                         /*
4490                          * we can't truncate inline items that have had
4491                          * special encodings
4492                          */
4493                         if (!del_item &&
4494                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4495                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4496 
4497                                 /*
4498                                  * Need to release path in order to truncate a
4499                                  * compressed extent. So delete any accumulated
4500                                  * extent items so far.
4501                                  */
4502                                 if (btrfs_file_extent_compression(leaf, fi) !=
4503                                     BTRFS_COMPRESS_NONE && pending_del_nr) {
4504                                         err = btrfs_del_items(trans, root, path,
4505                                                               pending_del_slot,
4506                                                               pending_del_nr);
4507                                         if (err) {
4508                                                 btrfs_abort_transaction(trans,
4509                                                                         root,
4510                                                                         err);
4511                                                 goto error;
4512                                         }
4513                                         pending_del_nr = 0;
4514                                 }
4515 
4516                                 err = truncate_inline_extent(inode, path,
4517                                                              &found_key,
4518                                                              item_end,
4519                                                              new_size);
4520                                 if (err) {
4521                                         btrfs_abort_transaction(trans,
4522                                                                 root, err);
4523                                         goto error;
4524                                 }
4525                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4526                                             &root->state)) {
4527                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4528                         }
4529                 }
4530 delete:
4531                 if (del_item) {
4532                         if (!pending_del_nr) {
4533                                 /* no pending yet, add ourselves */
4534                                 pending_del_slot = path->slots[0];
4535                                 pending_del_nr = 1;
4536                         } else if (pending_del_nr &&
4537                                    path->slots[0] + 1 == pending_del_slot) {
4538                                 /* hop on the pending chunk */
4539                                 pending_del_nr++;
4540                                 pending_del_slot = path->slots[0];
4541                         } else {
4542                                 BUG();
4543                         }
4544                 } else {
4545                         break;
4546                 }
4547                 should_throttle = 0;
4548 
4549                 if (found_extent &&
4550                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4551                      root == root->fs_info->tree_root)) {
4552                         btrfs_set_path_blocking(path);
4553                         bytes_deleted += extent_num_bytes;
4554                         ret = btrfs_free_extent(trans, root, extent_start,
4555                                                 extent_num_bytes, 0,
4556                                                 btrfs_header_owner(leaf),
4557                                                 ino, extent_offset);
4558                         BUG_ON(ret);
4559                         if (btrfs_should_throttle_delayed_refs(trans, root))
4560                                 btrfs_async_run_delayed_refs(root,
4561                                                              trans->transid,
4562                                         trans->delayed_ref_updates * 2, 0);
4563                         if (be_nice) {
4564                                 if (truncate_space_check(trans, root,
4565                                                          extent_num_bytes)) {
4566                                         should_end = 1;
4567                                 }
4568                                 if (btrfs_should_throttle_delayed_refs(trans,
4569                                                                        root)) {
4570                                         should_throttle = 1;
4571                                 }
4572                         }
4573                 }
4574 
4575                 if (found_type == BTRFS_INODE_ITEM_KEY)
4576                         break;
4577 
4578                 if (path->slots[0] == 0 ||
4579                     path->slots[0] != pending_del_slot ||
4580                     should_throttle || should_end) {
4581                         if (pending_del_nr) {
4582                                 ret = btrfs_del_items(trans, root, path,
4583                                                 pending_del_slot,
4584                                                 pending_del_nr);
4585                                 if (ret) {
4586                                         btrfs_abort_transaction(trans,
4587                                                                 root, ret);
4588                                         goto error;
4589                                 }
4590                                 pending_del_nr = 0;
4591                         }
4592                         btrfs_release_path(path);
4593                         if (should_throttle) {
4594                                 unsigned long updates = trans->delayed_ref_updates;
4595                                 if (updates) {
4596                                         trans->delayed_ref_updates = 0;
4597                                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4598                                         if (ret && !err)
4599                                                 err = ret;
4600                                 }
4601                         }
4602                         /*
4603                          * if we failed to refill our space rsv, bail out
4604                          * and let the transaction restart
4605                          */
4606                         if (should_end) {
4607                                 err = -EAGAIN;
4608                                 goto error;
4609                         }
4610                         goto search_again;
4611                 } else {
4612                         path->slots[0]--;
4613                 }
4614         }
4615 out:
4616         if (pending_del_nr) {
4617                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4618                                       pending_del_nr);
4619                 if (ret)
4620                         btrfs_abort_transaction(trans, root, ret);
4621         }
4622 error:
4623         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4624                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4625 
4626         btrfs_free_path(path);
4627 
4628         if (be_nice && bytes_deleted > SZ_32M) {
4629                 unsigned long updates = trans->delayed_ref_updates;
4630                 if (updates) {
4631                         trans->delayed_ref_updates = 0;
4632                         ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4633                         if (ret && !err)
4634                                 err = ret;
4635                 }
4636         }
4637         return err;
4638 }
4639 
4640 /*
4641  * btrfs_truncate_block - read, zero a chunk and write a block
4642  * @inode - inode that we're zeroing
4643  * @from - the offset to start zeroing
4644  * @len - the length to zero, 0 to zero the entire range respective to the
4645  *      offset
4646  * @front - zero up to the offset instead of from the offset on
4647  *
4648  * This will find the block for the "from" offset and cow the block and zero the
4649  * part we want to zero.  This is used with truncate and hole punching.
4650  */
4651 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4652                         int front)
4653 {
4654         struct address_space *mapping = inode->i_mapping;
4655         struct btrfs_root *root = BTRFS_I(inode)->root;
4656         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4657         struct btrfs_ordered_extent *ordered;
4658         struct extent_state *cached_state = NULL;
4659         char *kaddr;
4660         u32 blocksize = root->sectorsize;
4661         pgoff_t index = from >> PAGE_SHIFT;
4662         unsigned offset = from & (blocksize - 1);
4663         struct page *page;
4664         gfp_t mask = btrfs_alloc_write_mask(mapping);
4665         int ret = 0;
4666         u64 block_start;
4667         u64 block_end;
4668 
4669         if ((offset & (blocksize - 1)) == 0 &&
4670             (!len || ((len & (blocksize - 1)) == 0)))
4671                 goto out;
4672 
4673         ret = btrfs_delalloc_reserve_space(inode,
4674                         round_down(from, blocksize), blocksize);
4675         if (ret)
4676                 goto out;
4677 
4678 again:
4679         page = find_or_create_page(mapping, index, mask);
4680         if (!page) {
4681                 btrfs_delalloc_release_space(inode,
4682                                 round_down(from, blocksize),
4683                                 blocksize);
4684                 ret = -ENOMEM;
4685                 goto out;
4686         }
4687 
4688         block_start = round_down(from, blocksize);
4689         block_end = block_start + blocksize - 1;
4690 
4691         if (!PageUptodate(page)) {
4692                 ret = btrfs_readpage(NULL, page);
4693                 lock_page(page);
4694                 if (page->mapping != mapping) {
4695                         unlock_page(page);
4696                         put_page(page);
4697                         goto again;
4698                 }
4699                 if (!PageUptodate(page)) {
4700                         ret = -EIO;
4701                         goto out_unlock;
4702                 }
4703         }
4704         wait_on_page_writeback(page);
4705 
4706         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4707         set_page_extent_mapped(page);
4708 
4709         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4710         if (ordered) {
4711                 unlock_extent_cached(io_tree, block_start, block_end,
4712                                      &cached_state, GFP_NOFS);
4713                 unlock_page(page);
4714                 put_page(page);
4715                 btrfs_start_ordered_extent(inode, ordered, 1);
4716                 btrfs_put_ordered_extent(ordered);
4717                 goto again;
4718         }
4719 
4720         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4721                           EXTENT_DIRTY | EXTENT_DELALLOC |
4722                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4723                           0, 0, &cached_state, GFP_NOFS);
4724 
4725         ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4726                                         &cached_state);
4727         if (ret) {
4728                 unlock_extent_cached(io_tree, block_start, block_end,
4729                                      &cached_state, GFP_NOFS);
4730                 goto out_unlock;
4731         }
4732 
4733         if (offset != blocksize) {
4734                 if (!len)
4735                         len = blocksize - offset;
4736                 kaddr = kmap(page);
4737                 if (front)
4738                         memset(kaddr + (block_start - page_offset(page)),
4739                                 0, offset);
4740                 else
4741                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4742                                 0, len);
4743                 flush_dcache_page(page);
4744                 kunmap(page);
4745         }
4746         ClearPageChecked(page);
4747         set_page_dirty(page);
4748         unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4749                              GFP_NOFS);
4750 
4751 out_unlock:
4752         if (ret)
4753                 btrfs_delalloc_release_space(inode, block_start,
4754                                              blocksize);
4755         unlock_page(page);
4756         put_page(page);
4757 out:
4758         return ret;
4759 }
4760 
4761 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4762                              u64 offset, u64 len)
4763 {
4764         struct btrfs_trans_handle *trans;
4765         int ret;
4766 
4767         /*
4768          * Still need to make sure the inode looks like it's been updated so
4769          * that any holes get logged if we fsync.
4770          */
4771         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4772                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4773                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4774                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4775                 return 0;
4776         }
4777 
4778         /*
4779          * 1 - for the one we're dropping
4780          * 1 - for the one we're adding
4781          * 1 - for updating the inode.
4782          */
4783         trans = btrfs_start_transaction(root, 3);
4784         if (IS_ERR(trans))
4785                 return PTR_ERR(trans);
4786 
4787         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4788         if (ret) {
4789                 btrfs_abort_transaction(trans, root, ret);
4790                 btrfs_end_transaction(trans, root);
4791                 return ret;
4792         }
4793 
4794         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4795                                        0, 0, len, 0, len, 0, 0, 0);
4796         if (ret)
4797                 btrfs_abort_transaction(trans, root, ret);
4798         else
4799                 btrfs_update_inode(trans, root, inode);
4800         btrfs_end_transaction(trans, root);
4801         return ret;
4802 }
4803 
4804 /*
4805  * This function puts in dummy file extents for the area we're creating a hole
4806  * for.  So if we are truncating this file to a larger size we need to insert
4807  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4808  * the range between oldsize and size
4809  */
4810 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4811 {
4812         struct btrfs_root *root = BTRFS_I(inode)->root;
4813         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4814         struct extent_map *em = NULL;
4815         struct extent_state *cached_state = NULL;
4816         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4817         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4818         u64 block_end = ALIGN(size, root->sectorsize);
4819         u64 last_byte;
4820         u64 cur_offset;
4821         u64 hole_size;
4822         int err = 0;
4823 
4824         /*
4825          * If our size started in the middle of a block we need to zero out the
4826          * rest of the block before we expand the i_size, otherwise we could
4827          * expose stale data.
4828          */
4829         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4830         if (err)
4831                 return err;
4832 
4833         if (size <= hole_start)
4834                 return 0;
4835 
4836         while (1) {
4837                 struct btrfs_ordered_extent *ordered;
4838 
4839                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4840                                  &cached_state);
4841                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4842                                                      block_end - hole_start);
4843                 if (!ordered)
4844                         break;
4845                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4846                                      &cached_state, GFP_NOFS);
4847                 btrfs_start_ordered_extent(inode, ordered, 1);
4848                 btrfs_put_ordered_extent(ordered);
4849         }
4850 
4851         cur_offset = hole_start;
4852         while (1) {
4853                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4854                                 block_end - cur_offset, 0);
4855                 if (IS_ERR(em)) {
4856                         err = PTR_ERR(em);
4857                         em = NULL;
4858                         break;
4859                 }
4860                 last_byte = min(extent_map_end(em), block_end);
4861                 last_byte = ALIGN(last_byte , root->sectorsize);
4862                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4863                         struct extent_map *hole_em;
4864                         hole_size = last_byte - cur_offset;
4865 
4866                         err = maybe_insert_hole(root, inode, cur_offset,
4867                                                 hole_size);
4868                         if (err)
4869                                 break;
4870                         btrfs_drop_extent_cache(inode, cur_offset,
4871                                                 cur_offset + hole_size - 1, 0);
4872                         hole_em = alloc_extent_map();
4873                         if (!hole_em) {
4874                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4875                                         &BTRFS_I(inode)->runtime_flags);
4876                                 goto next;
4877                         }
4878                         hole_em->start = cur_offset;
4879                         hole_em->len = hole_size;
4880                         hole_em->orig_start = cur_offset;
4881 
4882                         hole_em->block_start = EXTENT_MAP_HOLE;
4883                         hole_em->block_len = 0;
4884                         hole_em->orig_block_len = 0;
4885                         hole_em->ram_bytes = hole_size;
4886                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4887                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4888                         hole_em->generation = root->fs_info->generation;
4889 
4890                         while (1) {
4891                                 write_lock(&em_tree->lock);
4892                                 err = add_extent_mapping(em_tree, hole_em, 1);
4893                                 write_unlock(&em_tree->lock);
4894                                 if (err != -EEXIST)
4895                                         break;
4896                                 btrfs_drop_extent_cache(inode, cur_offset,
4897                                                         cur_offset +
4898                                                         hole_size - 1, 0);
4899                         }
4900                         free_extent_map(hole_em);
4901                 }
4902 next:
4903                 free_extent_map(em);
4904                 em = NULL;
4905                 cur_offset = last_byte;
4906                 if (cur_offset >= block_end)
4907                         break;
4908         }
4909         free_extent_map(em);
4910         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4911                              GFP_NOFS);
4912         return err;
4913 }
4914 
4915 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4916 {
4917         struct btrfs_root *root = BTRFS_I(inode)->root;
4918         struct btrfs_trans_handle *trans;
4919         loff_t oldsize = i_size_read(inode);
4920         loff_t newsize = attr->ia_size;
4921         int mask = attr->ia_valid;
4922         int ret;
4923 
4924         /*
4925          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4926          * special case where we need to update the times despite not having
4927          * these flags set.  For all other operations the VFS set these flags
4928          * explicitly if it wants a timestamp update.
4929          */
4930         if (newsize != oldsize) {
4931                 inode_inc_iversion(inode);
4932                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4933                         inode->i_ctime = inode->i_mtime =
4934                                 current_fs_time(inode->i_sb);
4935         }
4936 
4937         if (newsize > oldsize) {
4938                 /*
4939                  * Don't do an expanding truncate while snapshoting is ongoing.
4940                  * This is to ensure the snapshot captures a fully consistent
4941                  * state of this file - if the snapshot captures this expanding
4942                  * truncation, it must capture all writes that happened before
4943                  * this truncation.
4944                  */
4945                 btrfs_wait_for_snapshot_creation(root);
4946                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4947                 if (ret) {
4948                         btrfs_end_write_no_snapshoting(root);
4949                         return ret;
4950                 }
4951 
4952                 trans = btrfs_start_transaction(root, 1);
4953                 if (IS_ERR(trans)) {
4954                         btrfs_end_write_no_snapshoting(root);
4955                         return PTR_ERR(trans);
4956                 }
4957 
4958                 i_size_write(inode, newsize);
4959                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4960                 pagecache_isize_extended(inode, oldsize, newsize);
4961                 ret = btrfs_update_inode(trans, root, inode);
4962                 btrfs_end_write_no_snapshoting(root);
4963                 btrfs_end_transaction(trans, root);
4964         } else {
4965 
4966                 /*
4967                  * We're truncating a file that used to have good data down to
4968                  * zero. Make sure it gets into the ordered flush list so that
4969                  * any new writes get down to disk quickly.
4970                  */
4971                 if (newsize == 0)
4972                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4973                                 &BTRFS_I(inode)->runtime_flags);
4974 
4975                 /*
4976                  * 1 for the orphan item we're going to add
4977                  * 1 for the orphan item deletion.
4978                  */
4979                 trans = btrfs_start_transaction(root, 2);
4980                 if (IS_ERR(trans))
4981                         return PTR_ERR(trans);
4982 
4983                 /*
4984                  * We need to do this in case we fail at _any_ point during the
4985                  * actual truncate.  Once we do the truncate_setsize we could
4986                  * invalidate pages which forces any outstanding ordered io to
4987                  * be instantly completed which will give us extents that need
4988                  * to be truncated.  If we fail to get an orphan inode down we
4989                  * could have left over extents that were never meant to live,
4990                  * so we need to guarantee from this point on that everything
4991                  * will be consistent.
4992                  */
4993                 ret = btrfs_orphan_add(trans, inode);
4994                 btrfs_end_transaction(trans, root);
4995                 if (ret)
4996                         return ret;
4997 
4998                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4999                 truncate_setsize(inode, newsize);
5000 
5001                 /* Disable nonlocked read DIO to avoid the end less truncate */
5002                 btrfs_inode_block_unlocked_dio(inode);
5003                 inode_dio_wait(inode);
5004                 btrfs_inode_resume_unlocked_dio(inode);
5005 
5006                 ret = btrfs_truncate(inode);
5007                 if (ret && inode->i_nlink) {
5008                         int err;
5009 
5010                         /*
5011                          * failed to truncate, disk_i_size is only adjusted down
5012                          * as we remove extents, so it should represent the true
5013                          * size of the inode, so reset the in memory size and
5014                          * delete our orphan entry.
5015                          */
5016                         trans = btrfs_join_transaction(root);
5017                         if (IS_ERR(trans)) {
5018                                 btrfs_orphan_del(NULL, inode);
5019                                 return ret;
5020                         }
5021                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5022                         err = btrfs_orphan_del(trans, inode);
5023                         if (err)
5024                                 btrfs_abort_transaction(trans, root, err);
5025                         btrfs_end_transaction(trans, root);
5026                 }
5027         }
5028 
5029         return ret;
5030 }
5031 
5032 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5033 {
5034         struct inode *inode = d_inode(dentry);
5035         struct btrfs_root *root = BTRFS_I(inode)->root;
5036         int err;
5037 
5038         if (btrfs_root_readonly(root))
5039                 return -EROFS;
5040 
5041         err = inode_change_ok(inode, attr);
5042         if (err)
5043                 return err;
5044 
5045         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5046                 err = btrfs_setsize(inode, attr);
5047                 if (err)
5048                         return err;
5049         }
5050 
5051         if (attr->ia_valid) {
5052                 setattr_copy(inode, attr);
5053                 inode_inc_iversion(inode);
5054                 err = btrfs_dirty_inode(inode);
5055 
5056                 if (!err && attr->ia_valid & ATTR_MODE)
5057                         err = posix_acl_chmod(inode, inode->i_mode);
5058         }
5059 
5060         return err;
5061 }
5062 
5063 /*
5064  * While truncating the inode pages during eviction, we get the VFS calling
5065  * btrfs_invalidatepage() against each page of the inode. This is slow because
5066  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5067  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5068  * extent_state structures over and over, wasting lots of time.
5069  *
5070  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5071  * those expensive operations on a per page basis and do only the ordered io
5072  * finishing, while we release here the extent_map and extent_state structures,
5073  * without the excessive merging and splitting.
5074  */
5075 static void evict_inode_truncate_pages(struct inode *inode)
5076 {
5077         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5078         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5079         struct rb_node *node;
5080 
5081         ASSERT(inode->i_state & I_FREEING);
5082         truncate_inode_pages_final(&inode->i_data);
5083 
5084         write_lock(&map_tree->lock);
5085         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5086                 struct extent_map *em;
5087 
5088                 node = rb_first(&map_tree->map);
5089                 em = rb_entry(node, struct extent_map, rb_node);
5090                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5091                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5092                 remove_extent_mapping(map_tree, em);
5093                 free_extent_map(em);
5094                 if (need_resched()) {
5095                         write_unlock(&map_tree->lock);
5096                         cond_resched();
5097                         write_lock(&map_tree->lock);
5098                 }
5099         }
5100         write_unlock(&map_tree->lock);
5101 
5102         /*
5103          * Keep looping until we have no more ranges in the io tree.
5104          * We can have ongoing bios started by readpages (called from readahead)
5105          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5106          * still in progress (unlocked the pages in the bio but did not yet
5107          * unlocked the ranges in the io tree). Therefore this means some
5108          * ranges can still be locked and eviction started because before
5109          * submitting those bios, which are executed by a separate task (work
5110          * queue kthread), inode references (inode->i_count) were not taken
5111          * (which would be dropped in the end io callback of each bio).
5112          * Therefore here we effectively end up waiting for those bios and
5113          * anyone else holding locked ranges without having bumped the inode's
5114          * reference count - if we don't do it, when they access the inode's
5115          * io_tree to unlock a range it may be too late, leading to an
5116          * use-after-free issue.
5117          */
5118         spin_lock(&io_tree->lock);
5119         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5120                 struct extent_state *state;
5121                 struct extent_state *cached_state = NULL;
5122                 u64 start;
5123                 u64 end;
5124 
5125                 node = rb_first(&io_tree->state);
5126                 state = rb_entry(node, struct extent_state, rb_node);
5127                 start = state->start;
5128                 end = state->end;
5129                 spin_unlock(&io_tree->lock);
5130 
5131                 lock_extent_bits(io_tree, start, end, &cached_state);
5132 
5133                 /*
5134                  * If still has DELALLOC flag, the extent didn't reach disk,
5135                  * and its reserved space won't be freed by delayed_ref.
5136                  * So we need to free its reserved space here.
5137                  * (Refer to comment in btrfs_invalidatepage, case 2)
5138                  *
5139                  * Note, end is the bytenr of last byte, so we need + 1 here.
5140                  */
5141                 if (state->state & EXTENT_DELALLOC)
5142                         btrfs_qgroup_free_data(inode, start, end - start + 1);
5143 
5144                 clear_extent_bit(io_tree, start, end,
5145                                  EXTENT_LOCKED | EXTENT_DIRTY |
5146                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5147                                  EXTENT_DEFRAG, 1, 1,
5148                                  &cached_state, GFP_NOFS);
5149 
5150                 cond_resched();
5151                 spin_lock(&io_tree->lock);
5152         }
5153         spin_unlock(&io_tree->lock);
5154 }
5155 
5156 void btrfs_evict_inode(struct inode *inode)
5157 {
5158         struct btrfs_trans_handle *trans;
5159         struct btrfs_root *root = BTRFS_I(inode)->root;
5160         struct btrfs_block_rsv *rsv, *global_rsv;
5161         int steal_from_global = 0;
5162         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5163         int ret;
5164 
5165         trace_btrfs_inode_evict(inode);
5166 
5167         evict_inode_truncate_pages(inode);
5168 
5169         if (inode->i_nlink &&
5170             ((btrfs_root_refs(&root->root_item) != 0 &&
5171               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5172              btrfs_is_free_space_inode(inode)))
5173                 goto no_delete;
5174 
5175         if (is_bad_inode(inode)) {
5176                 btrfs_orphan_del(NULL, inode);
5177                 goto no_delete;
5178         }
5179         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5180         if (!special_file(inode->i_mode))
5181                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5182 
5183         btrfs_free_io_failure_record(inode, 0, (u64)-1);
5184 
5185         if (root->fs_info->log_root_recovering) {
5186                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5187                                  &BTRFS_I(inode)->runtime_flags));
5188                 goto no_delete;
5189         }
5190 
5191         if (inode->i_nlink > 0) {
5192                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5193                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5194                 goto no_delete;
5195         }
5196 
5197         ret = btrfs_commit_inode_delayed_inode(inode);
5198         if (ret) {
5199                 btrfs_orphan_del(NULL, inode);
5200                 goto no_delete;
5201         }
5202 
5203         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5204         if (!rsv) {
5205                 btrfs_orphan_del(NULL, inode);
5206                 goto no_delete;
5207         }
5208         rsv->size = min_size;
5209         rsv->failfast = 1;
5210         global_rsv = &root->fs_info->global_block_rsv;
5211 
5212         btrfs_i_size_write(inode, 0);
5213 
5214         /*
5215          * This is a bit simpler than btrfs_truncate since we've already
5216          * reserved our space for our orphan item in the unlink, so we just
5217          * need to reserve some slack space in case we add bytes and update
5218          * inode item when doing the truncate.
5219          */
5220         while (1) {
5221                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5222                                              BTRFS_RESERVE_FLUSH_LIMIT);
5223 
5224                 /*
5225                  * Try and steal from the global reserve since we will
5226                  * likely not use this space anyway, we want to try as
5227                  * hard as possible to get this to work.
5228                  */
5229                 if (ret)
5230                         steal_from_global++;
5231                 else
5232                         steal_from_global = 0;
5233                 ret = 0;
5234 
5235                 /*
5236                  * steal_from_global == 0: we reserved stuff, hooray!
5237                  * steal_from_global == 1: we didn't reserve stuff, boo!
5238                  * steal_from_global == 2: we've committed, still not a lot of
5239                  * room but maybe we'll have room in the global reserve this
5240                  * time.
5241                  * steal_from_global == 3: abandon all hope!
5242                  */
5243                 if (steal_from_global > 2) {
5244                         btrfs_warn(root->fs_info,
5245                                 "Could not get space for a delete, will truncate on mount %d",
5246                                 ret);
5247                         btrfs_orphan_del(NULL, inode);
5248                         btrfs_free_block_rsv(root, rsv);
5249                         goto no_delete;
5250                 }
5251 
5252                 trans = btrfs_join_transaction(root);
5253                 if (IS_ERR(trans)) {
5254                         btrfs_orphan_del(NULL, inode);
5255                         btrfs_free_block_rsv(root, rsv);
5256                         goto no_delete;
5257                 }
5258 
5259                 /*
5260                  * We can't just steal from the global reserve, we need to make
5261                  * sure there is room to do it, if not we need to commit and try
5262                  * again.
5263                  */
5264                 if (steal_from_global) {
5265                         if (!btrfs_check_space_for_delayed_refs(trans, root))
5266                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5267                                                               min_size);
5268                         else
5269                                 ret = -ENOSPC;
5270                 }
5271 
5272                 /*
5273                  * Couldn't steal from the global reserve, we have too much
5274                  * pending stuff built up, commit the transaction and try it
5275                  * again.
5276                  */
5277                 if (ret) {
5278                         ret = btrfs_commit_transaction(trans, root);
5279                         if (ret) {
5280                                 btrfs_orphan_del(NULL, inode);
5281                                 btrfs_free_block_rsv(root, rsv);
5282                                 goto no_delete;
5283                         }
5284                         continue;
5285                 } else {
5286                         steal_from_global = 0;
5287                 }
5288 
5289                 trans->block_rsv = rsv;
5290 
5291                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5292                 if (ret != -ENOSPC && ret != -EAGAIN)
5293                         break;
5294 
5295                 trans->block_rsv = &root->fs_info->trans_block_rsv;
5296                 btrfs_end_transaction(trans, root);
5297                 trans = NULL;
5298                 btrfs_btree_balance_dirty(root);
5299         }
5300 
5301         btrfs_free_block_rsv(root, rsv);
5302 
5303         /*
5304          * Errors here aren't a big deal, it just means we leave orphan items
5305          * in the tree.  They will be cleaned up on the next mount.
5306          */
5307         if (ret == 0) {
5308                 trans->block_rsv = root->orphan_block_rsv;
5309                 btrfs_orphan_del(trans, inode);
5310         } else {
5311                 btrfs_orphan_del(NULL, inode);
5312         }
5313 
5314         trans->block_rsv = &root->fs_info->trans_block_rsv;
5315         if (!(root == root->fs_info->tree_root ||
5316               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5317                 btrfs_return_ino(root, btrfs_ino(inode));
5318 
5319         btrfs_end_transaction(trans, root);
5320         btrfs_btree_balance_dirty(root);
5321 no_delete:
5322         btrfs_remove_delayed_node(inode);
5323         clear_inode(inode);
5324 }
5325 
5326 /*
5327  * this returns the key found in the dir entry in the location pointer.
5328  * If no dir entries were found, location->objectid is 0.
5329  */
5330 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5331                                struct btrfs_key *location)
5332 {
5333         const char *name = dentry->d_name.name;
5334         int namelen = dentry->d_name.len;
5335         struct btrfs_dir_item *di;
5336         struct btrfs_path *path;
5337         struct btrfs_root *root = BTRFS_I(dir)->root;
5338         int ret = 0;
5339 
5340         path = btrfs_alloc_path();
5341         if (!path)
5342                 return -ENOMEM;
5343 
5344         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5345                                     namelen, 0);
5346         if (IS_ERR(di))
5347                 ret = PTR_ERR(di);
5348 
5349         if (IS_ERR_OR_NULL(di))
5350                 goto out_err;
5351 
5352         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5353 out:
5354         btrfs_free_path(path);
5355         return ret;
5356 out_err:
5357         location->objectid = 0;
5358         goto out;
5359 }
5360 
5361 /*
5362  * when we hit a tree root in a directory, the btrfs part of the inode
5363  * needs to be changed to reflect the root directory of the tree root.  This
5364  * is kind of like crossing a mount point.
5365  */
5366 static int fixup_tree_root_location(struct btrfs_root *root,
5367                                     struct inode *dir,
5368                                     struct dentry *dentry,
5369                                     struct btrfs_key *location,
5370                                     struct btrfs_root **sub_root)
5371 {
5372         struct btrfs_path *path;
5373         struct btrfs_root *new_root;
5374         struct btrfs_root_ref *ref;
5375         struct extent_buffer *leaf;
5376         struct btrfs_key key;
5377         int ret;
5378         int err = 0;
5379 
5380         path = btrfs_alloc_path();
5381         if (!path) {
5382                 err = -ENOMEM;
5383                 goto out;
5384         }
5385 
5386         err = -ENOENT;
5387         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5388         key.type = BTRFS_ROOT_REF_KEY;
5389         key.offset = location->objectid;
5390 
5391         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5392                                 0, 0);
5393         if (ret) {
5394                 if (ret < 0)
5395                         err = ret;
5396                 goto out;
5397         }
5398 
5399         leaf = path->nodes[0];
5400         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5401         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5402             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5403                 goto out;
5404 
5405         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5406                                    (unsigned long)(ref + 1),
5407                                    dentry->d_name.len);
5408         if (ret)
5409                 goto out;
5410 
5411         btrfs_release_path(path);
5412 
5413         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5414         if (IS_ERR(new_root)) {
5415                 err = PTR_ERR(new_root);
5416                 goto out;
5417         }
5418 
5419         *sub_root = new_root;
5420         location->objectid = btrfs_root_dirid(&new_root->root_item);
5421         location->type = BTRFS_INODE_ITEM_KEY;
5422         location->offset = 0;
5423         err = 0;
5424 out:
5425         btrfs_free_path(path);
5426         return err;
5427 }
5428 
5429 static void inode_tree_add(struct inode *inode)
5430 {
5431         struct btrfs_root *root = BTRFS_I(inode)->root;
5432         struct btrfs_inode *entry;
5433         struct rb_node **p;
5434         struct rb_node *parent;
5435         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5436         u64 ino = btrfs_ino(inode);
5437 
5438         if (inode_unhashed(inode))
5439                 return;
5440         parent = NULL;
5441         spin_lock(&root->inode_lock);
5442         p = &root->inode_tree.rb_node;
5443         while (*p) {
5444                 parent = *p;
5445                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5446 
5447                 if (ino < btrfs_ino(&entry->vfs_inode))
5448                         p = &parent->rb_left;
5449                 else if (ino > btrfs_ino(&entry->vfs_inode))
5450                         p = &parent->rb_right;
5451                 else {
5452                         WARN_ON(!(entry->vfs_inode.i_state &
5453                                   (I_WILL_FREE | I_FREEING)));
5454                         rb_replace_node(parent, new, &root->inode_tree);
5455                         RB_CLEAR_NODE(parent);
5456                         spin_unlock(&root->inode_lock);
5457                         return;
5458                 }
5459         }
5460         rb_link_node(new, parent, p);
5461         rb_insert_color(new, &root->inode_tree);
5462         spin_unlock(&root->inode_lock);
5463 }
5464 
5465 static void inode_tree_del(struct inode *inode)
5466 {
5467         struct btrfs_root *root = BTRFS_I(inode)->root;
5468         int empty = 0;
5469 
5470         spin_lock(&root->inode_lock);
5471         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5472                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5473                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5474                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5475         }
5476         spin_unlock(&root->inode_lock);
5477 
5478         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5479                 synchronize_srcu(&root->fs_info->subvol_srcu);
5480                 spin_lock(&root->inode_lock);
5481                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5482                 spin_unlock(&root->inode_lock);
5483                 if (empty)
5484                         btrfs_add_dead_root(root);
5485         }
5486 }
5487 
5488 void btrfs_invalidate_inodes(struct btrfs_root *root)
5489 {
5490         struct rb_node *node;
5491         struct rb_node *prev;
5492         struct btrfs_inode *entry;
5493         struct inode *inode;
5494         u64 objectid = 0;
5495 
5496         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5497                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5498 
5499         spin_lock(&root->inode_lock);
5500 again:
5501         node = root->inode_tree.rb_node;
5502         prev = NULL;
5503         while (node) {
5504                 prev = node;
5505                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5506 
5507                 if (objectid < btrfs_ino(&entry->vfs_inode))
5508                         node = node->rb_left;
5509                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5510                         node = node->rb_right;
5511                 else
5512                         break;
5513         }
5514         if (!node) {
5515                 while (prev) {
5516                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5517                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5518                                 node = prev;
5519                                 break;
5520                         }
5521                         prev = rb_next(prev);
5522                 }
5523         }
5524         while (node) {
5525                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5526                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5527                 inode = igrab(&entry->vfs_inode);
5528                 if (inode) {
5529                         spin_unlock(&root->inode_lock);
5530                         if (atomic_read(&inode->i_count) > 1)
5531                                 d_prune_aliases(inode);
5532                         /*
5533                          * btrfs_drop_inode will have it removed from
5534                          * the inode cache when its usage count
5535                          * hits zero.
5536                          */
5537                         iput(inode);
5538                         cond_resched();
5539                         spin_lock(&root->inode_lock);
5540                         goto again;
5541                 }
5542 
5543                 if (cond_resched_lock(&root->inode_lock))
5544                         goto again;
5545 
5546                 node = rb_next(node);
5547         }
5548         spin_unlock(&root->inode_lock);
5549 }
5550 
5551 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5552 {
5553         struct btrfs_iget_args *args = p;
5554         inode->i_ino = args->location->objectid;
5555         memcpy(&BTRFS_I(inode)->location, args->location,
5556                sizeof(*args->location));
5557         BTRFS_I(inode)->root = args->root;
5558         return 0;
5559 }
5560 
5561 static int btrfs_find_actor(struct inode *inode, void *opaque)
5562 {
5563         struct btrfs_iget_args *args = opaque;
5564         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5565                 args->root == BTRFS_I(inode)->root;
5566 }
5567 
5568 static struct inode *btrfs_iget_locked(struct super_block *s,
5569                                        struct btrfs_key *location,
5570                                        struct btrfs_root *root)
5571 {
5572         struct inode *inode;
5573         struct btrfs_iget_args args;
5574         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5575 
5576         args.location = location;
5577         args.root = root;
5578 
5579         inode = iget5_locked(s, hashval, btrfs_find_actor,
5580                              btrfs_init_locked_inode,
5581                              (void *)&args);
5582         return inode;
5583 }
5584 
5585 /* Get an inode object given its location and corresponding root.
5586  * Returns in *is_new if the inode was read from disk
5587  */
5588 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5589                          struct btrfs_root *root, int *new)
5590 {
5591         struct inode *inode;
5592 
5593         inode = btrfs_iget_locked(s, location, root);
5594         if (!inode)
5595                 return ERR_PTR(-ENOMEM);
5596 
5597         if (inode->i_state & I_NEW) {
5598                 btrfs_read_locked_inode(inode);
5599                 if (!is_bad_inode(inode)) {
5600                         inode_tree_add(inode);
5601                         unlock_new_inode(inode);
5602                         if (new)
5603                                 *new = 1;
5604                 } else {
5605                         unlock_new_inode(inode);
5606                         iput(inode);
5607                         inode = ERR_PTR(-ESTALE);
5608                 }
5609         }
5610 
5611         return inode;
5612 }
5613 
5614 static struct inode *new_simple_dir(struct super_block *s,
5615                                     struct btrfs_key *key,
5616                                     struct btrfs_root *root)
5617 {
5618         struct inode *inode = new_inode(s);
5619 
5620         if (!inode)
5621                 return ERR_PTR(-ENOMEM);
5622 
5623         BTRFS_I(inode)->root = root;
5624         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5625         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5626 
5627         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5628         inode->i_op = &btrfs_dir_ro_inode_operations;
5629         inode->i_fop = &simple_dir_operations;
5630         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5631         inode->i_mtime = current_fs_time(inode->i_sb);
5632         inode->i_atime = inode->i_mtime;
5633         inode->i_ctime = inode->i_mtime;
5634         BTRFS_I(inode)->i_otime = inode->i_mtime;
5635 
5636         return inode;
5637 }
5638 
5639 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5640 {
5641         struct inode *inode;
5642         struct btrfs_root *root = BTRFS_I(dir)->root;
5643         struct btrfs_root *sub_root = root;
5644         struct btrfs_key location;
5645         int index;
5646         int ret = 0;
5647 
5648         if (dentry->d_name.len > BTRFS_NAME_LEN)
5649                 return ERR_PTR(-ENAMETOOLONG);
5650 
5651         ret = btrfs_inode_by_name(dir, dentry, &location);
5652         if (ret < 0)
5653                 return ERR_PTR(ret);
5654 
5655         if (location.objectid == 0)
5656                 return ERR_PTR(-ENOENT);
5657 
5658         if (location.type == BTRFS_INODE_ITEM_KEY) {
5659                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5660                 return inode;
5661         }
5662 
5663         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5664 
5665         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5666         ret = fixup_tree_root_location(root, dir, dentry,
5667                                        &location, &sub_root);
5668         if (ret < 0) {
5669                 if (ret != -ENOENT)
5670                         inode = ERR_PTR(ret);
5671                 else
5672                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5673         } else {
5674                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5675         }
5676         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5677 
5678         if (!IS_ERR(inode) && root != sub_root) {
5679                 down_read(&root->fs_info->cleanup_work_sem);
5680                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5681                         ret = btrfs_orphan_cleanup(sub_root);
5682                 up_read(&root->fs_info->cleanup_work_sem);
5683                 if (ret) {
5684                         iput(inode);
5685                         inode = ERR_PTR(ret);
5686                 }
5687         }
5688 
5689         return inode;
5690 }
5691 
5692 static int btrfs_dentry_delete(const struct dentry *dentry)
5693 {
5694         struct btrfs_root *root;
5695         struct inode *inode = d_inode(dentry);
5696 
5697         if (!inode && !IS_ROOT(dentry))
5698                 inode = d_inode(dentry->d_parent);
5699 
5700         if (inode) {
5701                 root = BTRFS_I(inode)->root;
5702                 if (btrfs_root_refs(&root->root_item) == 0)
5703                         return 1;
5704 
5705                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5706                         return 1;
5707         }
5708         return 0;
5709 }
5710 
5711 static void btrfs_dentry_release(struct dentry *dentry)
5712 {
5713         kfree(dentry->d_fsdata);
5714 }
5715 
5716 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5717                                    unsigned int flags)
5718 {
5719         struct inode *inode;
5720 
5721         inode = btrfs_lookup_dentry(dir, dentry);
5722         if (IS_ERR(inode)) {
5723                 if (PTR_ERR(inode) == -ENOENT)
5724                         inode = NULL;
5725                 else
5726                         return ERR_CAST(inode);
5727         }
5728 
5729         return d_splice_alias(inode, dentry);
5730 }
5731 
5732 unsigned char btrfs_filetype_table[] = {
5733         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5734 };
5735 
5736 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5737 {
5738         struct inode *inode = file_inode(file);
5739         struct btrfs_root *root = BTRFS_I(inode)->root;
5740         struct btrfs_item *item;
5741         struct btrfs_dir_item *di;
5742         struct btrfs_key key;
5743         struct btrfs_key found_key;
5744         struct btrfs_path *path;
5745         struct list_head ins_list;
5746         struct list_head del_list;
5747         int ret;
5748         struct extent_buffer *leaf;
5749         int slot;
5750         unsigned char d_type;
5751         int over = 0;
5752         u32 di_cur;
5753         u32 di_total;
5754         u32 di_len;
5755         int key_type = BTRFS_DIR_INDEX_KEY;
5756         char tmp_name[32];
5757         char *name_ptr;
5758         int name_len;
5759         int is_curr = 0;        /* ctx->pos points to the current index? */
5760         bool emitted;
5761         bool put = false;
5762 
5763         /* FIXME, use a real flag for deciding about the key type */
5764         if (root->fs_info->tree_root == root)
5765                 key_type = BTRFS_DIR_ITEM_KEY;
5766 
5767         if (!dir_emit_dots(file, ctx))
5768                 return 0;
5769 
5770         path = btrfs_alloc_path();
5771         if (!path)
5772                 return -ENOMEM;
5773 
5774         path->reada = READA_FORWARD;
5775 
5776         if (key_type == BTRFS_DIR_INDEX_KEY) {
5777                 INIT_LIST_HEAD(&ins_list);
5778                 INIT_LIST_HEAD(&del_list);
5779                 put = btrfs_readdir_get_delayed_items(inode, &ins_list,
5780                                                       &del_list);
5781         }
5782 
5783         key.type = key_type;
5784         key.offset = ctx->pos;
5785         key.objectid = btrfs_ino(inode);
5786 
5787         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5788         if (ret < 0)
5789                 goto err;
5790 
5791         emitted = false;
5792         while (1) {
5793                 leaf = path->nodes[0];
5794                 slot = path->slots[0];
5795                 if (slot >= btrfs_header_nritems(leaf)) {
5796                         ret = btrfs_next_leaf(root, path);
5797                         if (ret < 0)
5798                                 goto err;
5799                         else if (ret > 0)
5800                                 break;
5801                         continue;
5802                 }
5803 
5804                 item = btrfs_item_nr(slot);
5805                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5806 
5807                 if (found_key.objectid != key.objectid)
5808                         break;
5809                 if (found_key.type != key_type)
5810                         break;
5811                 if (found_key.offset < ctx->pos)
5812                         goto next;
5813                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5814                     btrfs_should_delete_dir_index(&del_list,
5815                                                   found_key.offset))
5816                         goto next;
5817 
5818                 ctx->pos = found_key.offset;
5819                 is_curr = 1;
5820 
5821                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5822                 di_cur = 0;
5823                 di_total = btrfs_item_size(leaf, item);
5824 
5825                 while (di_cur < di_total) {
5826                         struct btrfs_key location;
5827 
5828                         if (verify_dir_item(root, leaf, di))
5829                                 break;
5830 
5831                         name_len = btrfs_dir_name_len(leaf, di);
5832                         if (name_len <= sizeof(tmp_name)) {
5833                                 name_ptr = tmp_name;
5834                         } else {
5835                                 name_ptr = kmalloc(name_len, GFP_KERNEL);
5836                                 if (!name_ptr) {
5837                                         ret = -ENOMEM;
5838                                         goto err;
5839                                 }
5840                         }
5841                         read_extent_buffer(leaf, name_ptr,
5842                                            (unsigned long)(di + 1), name_len);
5843 
5844                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5845                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5846 
5847 
5848                         /* is this a reference to our own snapshot? If so
5849                          * skip it.
5850                          *
5851                          * In contrast to old kernels, we insert the snapshot's
5852                          * dir item and dir index after it has been created, so
5853                          * we won't find a reference to our own snapshot. We
5854                          * still keep the following code for backward
5855                          * compatibility.
5856                          */
5857                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5858                             location.objectid == root->root_key.objectid) {
5859                                 over = 0;
5860                                 goto skip;
5861                         }
5862                         over = !dir_emit(ctx, name_ptr, name_len,
5863                                        location.objectid, d_type);
5864 
5865 skip:
5866                         if (name_ptr != tmp_name)
5867                                 kfree(name_ptr);
5868 
5869                         if (over)
5870                                 goto nopos;
5871                         emitted = true;
5872                         di_len = btrfs_dir_name_len(leaf, di) +
5873                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5874                         di_cur += di_len;
5875                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5876                 }
5877 next:
5878                 path->slots[0]++;
5879         }
5880 
5881         if (key_type == BTRFS_DIR_INDEX_KEY) {
5882                 if (is_curr)
5883                         ctx->pos++;
5884                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5885                 if (ret)
5886                         goto nopos;
5887         }
5888 
5889         /*
5890          * If we haven't emitted any dir entry, we must not touch ctx->pos as
5891          * it was was set to the termination value in previous call. We assume
5892          * that "." and ".." were emitted if we reach this point and set the
5893          * termination value as well for an empty directory.
5894          */
5895         if (ctx->pos > 2 && !emitted)
5896                 goto nopos;
5897 
5898         /* Reached end of directory/root. Bump pos past the last item. */
5899         ctx->pos++;
5900 
5901         /*
5902          * Stop new entries from being returned after we return the last
5903          * entry.
5904          *
5905          * New directory entries are assigned a strictly increasing
5906          * offset.  This means that new entries created during readdir
5907          * are *guaranteed* to be seen in the future by that readdir.
5908          * This has broken buggy programs which operate on names as
5909          * they're returned by readdir.  Until we re-use freed offsets
5910          * we have this hack to stop new entries from being returned
5911          * under the assumption that they'll never reach this huge
5912          * offset.
5913          *
5914          * This is being careful not to overflow 32bit loff_t unless the
5915          * last entry requires it because doing so has broken 32bit apps
5916          * in the past.
5917          */
5918         if (key_type == BTRFS_DIR_INDEX_KEY) {
5919                 if (ctx->pos >= INT_MAX)
5920                         ctx->pos = LLONG_MAX;
5921                 else
5922                         ctx->pos = INT_MAX;
5923         }
5924 nopos:
5925         ret = 0;
5926 err:
5927         if (put)
5928                 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5929         btrfs_free_path(path);
5930         return ret;
5931 }
5932 
5933 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5934 {
5935         struct btrfs_root *root = BTRFS_I(inode)->root;
5936         struct btrfs_trans_handle *trans;
5937         int ret = 0;
5938         bool nolock = false;
5939 
5940         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5941                 return 0;
5942 
5943         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5944                 nolock = true;
5945 
5946         if (wbc->sync_mode == WB_SYNC_ALL) {
5947                 if (nolock)
5948                         trans = btrfs_join_transaction_nolock(root);
5949                 else
5950                         trans = btrfs_join_transaction(root);
5951                 if (IS_ERR(trans))
5952                         return PTR_ERR(trans);
5953                 ret = btrfs_commit_transaction(trans, root);
5954         }
5955         return ret;
5956 }
5957 
5958 /*
5959  * This is somewhat expensive, updating the tree every time the
5960  * inode changes.  But, it is most likely to find the inode in cache.
5961  * FIXME, needs more benchmarking...there are no reasons other than performance
5962  * to keep or drop this code.
5963  */
5964 static int btrfs_dirty_inode(struct inode *inode)
5965 {
5966         struct btrfs_root *root = BTRFS_I(inode)->root;
5967         struct btrfs_trans_handle *trans;
5968         int ret;
5969 
5970         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5971                 return 0;
5972 
5973         trans = btrfs_join_transaction(root);
5974         if (IS_ERR(trans))
5975                 return PTR_ERR(trans);
5976 
5977         ret = btrfs_update_inode(trans, root, inode);
5978         if (ret && ret == -ENOSPC) {
5979                 /* whoops, lets try again with the full transaction */
5980                 btrfs_end_transaction(trans, root);
5981                 trans = btrfs_start_transaction(root, 1);
5982                 if (IS_ERR(trans))
5983                         return PTR_ERR(trans);
5984 
5985                 ret = btrfs_update_inode(trans, root, inode);
5986         }
5987         btrfs_end_transaction(trans, root);
5988         if (BTRFS_I(inode)->delayed_node)
5989                 btrfs_balance_delayed_items(root);
5990 
5991         return ret;
5992 }
5993 
5994 /*
5995  * This is a copy of file_update_time.  We need this so we can return error on
5996  * ENOSPC for updating the inode in the case of file write and mmap writes.
5997  */
5998 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5999                              int flags)
6000 {
6001         struct btrfs_root *root = BTRFS_I(inode)->root;
6002 
6003         if (btrfs_root_readonly(root))
6004                 return -EROFS;
6005 
6006         if (flags & S_VERSION)
6007                 inode_inc_iversion(inode);
6008         if (flags & S_CTIME)
6009                 inode->i_ctime = *now;
6010         if (flags & S_MTIME)
6011                 inode->i_mtime = *now;
6012         if (flags & S_ATIME)
6013                 inode->i_atime = *now;
6014         return btrfs_dirty_inode(inode);
6015 }
6016 
6017 /*
6018  * find the highest existing sequence number in a directory
6019  * and then set the in-memory index_cnt variable to reflect
6020  * free sequence numbers
6021  */
6022 static int btrfs_set_inode_index_count(struct inode *inode)
6023 {
6024         struct btrfs_root *root = BTRFS_I(inode)->root;
6025         struct btrfs_key key, found_key;
6026         struct btrfs_path *path;
6027         struct extent_buffer *leaf;
6028         int ret;
6029 
6030         key.objectid = btrfs_ino(inode);
6031         key.type = BTRFS_DIR_INDEX_KEY;
6032         key.offset = (u64)-1;
6033 
6034         path = btrfs_alloc_path();
6035         if (!path)
6036                 return -ENOMEM;
6037 
6038         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6039         if (ret < 0)
6040                 goto out;
6041         /* FIXME: we should be able to handle this */
6042         if (ret == 0)
6043                 goto out;
6044         ret = 0;
6045 
6046         /*
6047          * MAGIC NUMBER EXPLANATION:
6048          * since we search a directory based on f_pos we have to start at 2
6049          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6050          * else has to start at 2
6051          */
6052         if (path->slots[0] == 0) {
6053                 BTRFS_I(inode)->index_cnt = 2;
6054                 goto out;
6055         }
6056 
6057         path->slots[0]--;
6058 
6059         leaf = path->nodes[0];
6060         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6061 
6062         if (found_key.objectid != btrfs_ino(inode) ||
6063             found_key.type != BTRFS_DIR_INDEX_KEY) {
6064                 BTRFS_I(inode)->index_cnt = 2;
6065                 goto out;
6066         }
6067 
6068         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6069 out:
6070         btrfs_free_path(path);
6071         return ret;
6072 }
6073 
6074 /*
6075  * helper to find a free sequence number in a given directory.  This current
6076  * code is very simple, later versions will do smarter things in the btree
6077  */
6078 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6079 {
6080         int ret = 0;
6081 
6082         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6083                 ret = btrfs_inode_delayed_dir_index_count(dir);
6084                 if (ret) {
6085                         ret = btrfs_set_inode_index_count(dir);
6086                         if (ret)
6087                                 return ret;
6088                 }
6089         }
6090 
6091         *index = BTRFS_I(dir)->index_cnt;
6092         BTRFS_I(dir)->index_cnt++;
6093 
6094         return ret;
6095 }
6096 
6097 static int btrfs_insert_inode_locked(struct inode *inode)
6098 {
6099         struct btrfs_iget_args args;
6100         args.location = &BTRFS_I(inode)->location;
6101         args.root = BTRFS_I(inode)->root;
6102 
6103         return insert_inode_locked4(inode,
6104                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6105                    btrfs_find_actor, &args);
6106 }
6107 
6108 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6109                                      struct btrfs_root *root,
6110                                      struct inode *dir,
6111                                      const char *name, int name_len,
6112                                      u64 ref_objectid, u64 objectid,
6113                                      umode_t mode, u64 *index)
6114 {
6115         struct inode *inode;
6116         struct btrfs_inode_item *inode_item;
6117         struct btrfs_key *location;
6118         struct btrfs_path *path;
6119         struct btrfs_inode_ref *ref;
6120         struct btrfs_key key[2];
6121         u32 sizes[2];
6122         int nitems = name ? 2 : 1;
6123         unsigned long ptr;
6124         int ret;
6125 
6126         path = btrfs_alloc_path();
6127         if (!path)
6128                 return ERR_PTR(-ENOMEM);
6129 
6130         inode = new_inode(root->fs_info->sb);
6131         if (!inode) {
6132                 btrfs_free_path(path);
6133                 return ERR_PTR(-ENOMEM);
6134         }
6135 
6136         /*
6137          * O_TMPFILE, set link count to 0, so that after this point,
6138          * we fill in an inode item with the correct link count.
6139          */
6140         if (!name)
6141                 set_nlink(inode, 0);
6142 
6143         /*
6144          * we have to initialize this early, so we can reclaim the inode
6145          * number if we fail afterwards in this function.
6146          */
6147         inode->i_ino = objectid;
6148 
6149         if (dir && name) {
6150                 trace_btrfs_inode_request(dir);
6151 
6152                 ret = btrfs_set_inode_index(dir, index);
6153                 if (ret) {
6154                         btrfs_free_path(path);
6155                         iput(inode);
6156                         return ERR_PTR(ret);
6157                 }
6158         } else if (dir) {
6159                 *index = 0;
6160         }
6161         /*
6162          * index_cnt is ignored for everything but a dir,
6163          * btrfs_get_inode_index_count has an explanation for the magic
6164          * number
6165          */
6166         BTRFS_I(inode)->index_cnt = 2;
6167         BTRFS_I(inode)->dir_index = *index;
6168         BTRFS_I(inode)->root = root;
6169         BTRFS_I(inode)->generation = trans->transid;
6170         inode->i_generation