Merge tag 'for-4.19-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[muen/linux.git] / fs / btrfs / ioctl.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/file.h>
9 #include <linux/fs.h>
10 #include <linux/fsnotify.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/string.h>
15 #include <linux/backing-dev.h>
16 #include <linux/mount.h>
17 #include <linux/namei.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/security.h>
21 #include <linux/xattr.h>
22 #include <linux/mm.h>
23 #include <linux/slab.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
26 #include <linux/btrfs.h>
27 #include <linux/uaccess.h>
28 #include <linux/iversion.h>
29 #include "ctree.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "btrfs_inode.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "locking.h"
36 #include "inode-map.h"
37 #include "backref.h"
38 #include "rcu-string.h"
39 #include "send.h"
40 #include "dev-replace.h"
41 #include "props.h"
42 #include "sysfs.h"
43 #include "qgroup.h"
44 #include "tree-log.h"
45 #include "compression.h"
46
47 #ifdef CONFIG_64BIT
48 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
49  * structures are incorrect, as the timespec structure from userspace
50  * is 4 bytes too small. We define these alternatives here to teach
51  * the kernel about the 32-bit struct packing.
52  */
53 struct btrfs_ioctl_timespec_32 {
54         __u64 sec;
55         __u32 nsec;
56 } __attribute__ ((__packed__));
57
58 struct btrfs_ioctl_received_subvol_args_32 {
59         char    uuid[BTRFS_UUID_SIZE];  /* in */
60         __u64   stransid;               /* in */
61         __u64   rtransid;               /* out */
62         struct btrfs_ioctl_timespec_32 stime; /* in */
63         struct btrfs_ioctl_timespec_32 rtime; /* out */
64         __u64   flags;                  /* in */
65         __u64   reserved[16];           /* in */
66 } __attribute__ ((__packed__));
67
68 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
69                                 struct btrfs_ioctl_received_subvol_args_32)
70 #endif
71
72 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
73 struct btrfs_ioctl_send_args_32 {
74         __s64 send_fd;                  /* in */
75         __u64 clone_sources_count;      /* in */
76         compat_uptr_t clone_sources;    /* in */
77         __u64 parent_root;              /* in */
78         __u64 flags;                    /* in */
79         __u64 reserved[4];              /* in */
80 } __attribute__ ((__packed__));
81
82 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
83                                struct btrfs_ioctl_send_args_32)
84 #endif
85
86 static int btrfs_clone(struct inode *src, struct inode *inode,
87                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
88                        int no_time_update);
89
90 /* Mask out flags that are inappropriate for the given type of inode. */
91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
92                 unsigned int flags)
93 {
94         if (S_ISDIR(inode->i_mode))
95                 return flags;
96         else if (S_ISREG(inode->i_mode))
97                 return flags & ~FS_DIRSYNC_FL;
98         else
99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
100 }
101
102 /*
103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
104  * ioctl.
105  */
106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
107 {
108         unsigned int iflags = 0;
109
110         if (flags & BTRFS_INODE_SYNC)
111                 iflags |= FS_SYNC_FL;
112         if (flags & BTRFS_INODE_IMMUTABLE)
113                 iflags |= FS_IMMUTABLE_FL;
114         if (flags & BTRFS_INODE_APPEND)
115                 iflags |= FS_APPEND_FL;
116         if (flags & BTRFS_INODE_NODUMP)
117                 iflags |= FS_NODUMP_FL;
118         if (flags & BTRFS_INODE_NOATIME)
119                 iflags |= FS_NOATIME_FL;
120         if (flags & BTRFS_INODE_DIRSYNC)
121                 iflags |= FS_DIRSYNC_FL;
122         if (flags & BTRFS_INODE_NODATACOW)
123                 iflags |= FS_NOCOW_FL;
124
125         if (flags & BTRFS_INODE_NOCOMPRESS)
126                 iflags |= FS_NOCOMP_FL;
127         else if (flags & BTRFS_INODE_COMPRESS)
128                 iflags |= FS_COMPR_FL;
129
130         return iflags;
131 }
132
133 /*
134  * Update inode->i_flags based on the btrfs internal flags.
135  */
136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
137 {
138         struct btrfs_inode *binode = BTRFS_I(inode);
139         unsigned int new_fl = 0;
140
141         if (binode->flags & BTRFS_INODE_SYNC)
142                 new_fl |= S_SYNC;
143         if (binode->flags & BTRFS_INODE_IMMUTABLE)
144                 new_fl |= S_IMMUTABLE;
145         if (binode->flags & BTRFS_INODE_APPEND)
146                 new_fl |= S_APPEND;
147         if (binode->flags & BTRFS_INODE_NOATIME)
148                 new_fl |= S_NOATIME;
149         if (binode->flags & BTRFS_INODE_DIRSYNC)
150                 new_fl |= S_DIRSYNC;
151
152         set_mask_bits(&inode->i_flags,
153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
154                       new_fl);
155 }
156
157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
158 {
159         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
160         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
161
162         if (copy_to_user(arg, &flags, sizeof(flags)))
163                 return -EFAULT;
164         return 0;
165 }
166
167 /* Check if @flags are a supported and valid set of FS_*_FL flags */
168 static int check_fsflags(unsigned int flags)
169 {
170         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
171                       FS_NOATIME_FL | FS_NODUMP_FL | \
172                       FS_SYNC_FL | FS_DIRSYNC_FL | \
173                       FS_NOCOMP_FL | FS_COMPR_FL |
174                       FS_NOCOW_FL))
175                 return -EOPNOTSUPP;
176
177         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
178                 return -EINVAL;
179
180         return 0;
181 }
182
183 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
184 {
185         struct inode *inode = file_inode(file);
186         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
187         struct btrfs_inode *binode = BTRFS_I(inode);
188         struct btrfs_root *root = binode->root;
189         struct btrfs_trans_handle *trans;
190         unsigned int fsflags, old_fsflags;
191         int ret;
192         u64 old_flags;
193         unsigned int old_i_flags;
194         umode_t mode;
195
196         if (!inode_owner_or_capable(inode))
197                 return -EPERM;
198
199         if (btrfs_root_readonly(root))
200                 return -EROFS;
201
202         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
203                 return -EFAULT;
204
205         ret = check_fsflags(fsflags);
206         if (ret)
207                 return ret;
208
209         ret = mnt_want_write_file(file);
210         if (ret)
211                 return ret;
212
213         inode_lock(inode);
214
215         old_flags = binode->flags;
216         old_i_flags = inode->i_flags;
217         mode = inode->i_mode;
218
219         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
220         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
221         if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
222                 if (!capable(CAP_LINUX_IMMUTABLE)) {
223                         ret = -EPERM;
224                         goto out_unlock;
225                 }
226         }
227
228         if (fsflags & FS_SYNC_FL)
229                 binode->flags |= BTRFS_INODE_SYNC;
230         else
231                 binode->flags &= ~BTRFS_INODE_SYNC;
232         if (fsflags & FS_IMMUTABLE_FL)
233                 binode->flags |= BTRFS_INODE_IMMUTABLE;
234         else
235                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
236         if (fsflags & FS_APPEND_FL)
237                 binode->flags |= BTRFS_INODE_APPEND;
238         else
239                 binode->flags &= ~BTRFS_INODE_APPEND;
240         if (fsflags & FS_NODUMP_FL)
241                 binode->flags |= BTRFS_INODE_NODUMP;
242         else
243                 binode->flags &= ~BTRFS_INODE_NODUMP;
244         if (fsflags & FS_NOATIME_FL)
245                 binode->flags |= BTRFS_INODE_NOATIME;
246         else
247                 binode->flags &= ~BTRFS_INODE_NOATIME;
248         if (fsflags & FS_DIRSYNC_FL)
249                 binode->flags |= BTRFS_INODE_DIRSYNC;
250         else
251                 binode->flags &= ~BTRFS_INODE_DIRSYNC;
252         if (fsflags & FS_NOCOW_FL) {
253                 if (S_ISREG(mode)) {
254                         /*
255                          * It's safe to turn csums off here, no extents exist.
256                          * Otherwise we want the flag to reflect the real COW
257                          * status of the file and will not set it.
258                          */
259                         if (inode->i_size == 0)
260                                 binode->flags |= BTRFS_INODE_NODATACOW
261                                               | BTRFS_INODE_NODATASUM;
262                 } else {
263                         binode->flags |= BTRFS_INODE_NODATACOW;
264                 }
265         } else {
266                 /*
267                  * Revert back under same assumptions as above
268                  */
269                 if (S_ISREG(mode)) {
270                         if (inode->i_size == 0)
271                                 binode->flags &= ~(BTRFS_INODE_NODATACOW
272                                              | BTRFS_INODE_NODATASUM);
273                 } else {
274                         binode->flags &= ~BTRFS_INODE_NODATACOW;
275                 }
276         }
277
278         /*
279          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
280          * flag may be changed automatically if compression code won't make
281          * things smaller.
282          */
283         if (fsflags & FS_NOCOMP_FL) {
284                 binode->flags &= ~BTRFS_INODE_COMPRESS;
285                 binode->flags |= BTRFS_INODE_NOCOMPRESS;
286
287                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
288                 if (ret && ret != -ENODATA)
289                         goto out_drop;
290         } else if (fsflags & FS_COMPR_FL) {
291                 const char *comp;
292
293                 binode->flags |= BTRFS_INODE_COMPRESS;
294                 binode->flags &= ~BTRFS_INODE_NOCOMPRESS;
295
296                 comp = btrfs_compress_type2str(fs_info->compress_type);
297                 if (!comp || comp[0] == 0)
298                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
299
300                 ret = btrfs_set_prop(inode, "btrfs.compression",
301                                      comp, strlen(comp), 0);
302                 if (ret)
303                         goto out_drop;
304
305         } else {
306                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
307                 if (ret && ret != -ENODATA)
308                         goto out_drop;
309                 binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
310         }
311
312         trans = btrfs_start_transaction(root, 1);
313         if (IS_ERR(trans)) {
314                 ret = PTR_ERR(trans);
315                 goto out_drop;
316         }
317
318         btrfs_sync_inode_flags_to_i_flags(inode);
319         inode_inc_iversion(inode);
320         inode->i_ctime = current_time(inode);
321         ret = btrfs_update_inode(trans, root, inode);
322
323         btrfs_end_transaction(trans);
324  out_drop:
325         if (ret) {
326                 binode->flags = old_flags;
327                 inode->i_flags = old_i_flags;
328         }
329
330  out_unlock:
331         inode_unlock(inode);
332         mnt_drop_write_file(file);
333         return ret;
334 }
335
336 /*
337  * Translate btrfs internal inode flags to xflags as expected by the
338  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
339  * silently dropped.
340  */
341 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
342 {
343         unsigned int xflags = 0;
344
345         if (flags & BTRFS_INODE_APPEND)
346                 xflags |= FS_XFLAG_APPEND;
347         if (flags & BTRFS_INODE_IMMUTABLE)
348                 xflags |= FS_XFLAG_IMMUTABLE;
349         if (flags & BTRFS_INODE_NOATIME)
350                 xflags |= FS_XFLAG_NOATIME;
351         if (flags & BTRFS_INODE_NODUMP)
352                 xflags |= FS_XFLAG_NODUMP;
353         if (flags & BTRFS_INODE_SYNC)
354                 xflags |= FS_XFLAG_SYNC;
355
356         return xflags;
357 }
358
359 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
360 static int check_xflags(unsigned int flags)
361 {
362         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
363                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
364                 return -EOPNOTSUPP;
365         return 0;
366 }
367
368 /*
369  * Set the xflags from the internal inode flags. The remaining items of fsxattr
370  * are zeroed.
371  */
372 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
373 {
374         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
375         struct fsxattr fa;
376
377         memset(&fa, 0, sizeof(fa));
378         fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
379
380         if (copy_to_user(arg, &fa, sizeof(fa)))
381                 return -EFAULT;
382
383         return 0;
384 }
385
386 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
387 {
388         struct inode *inode = file_inode(file);
389         struct btrfs_inode *binode = BTRFS_I(inode);
390         struct btrfs_root *root = binode->root;
391         struct btrfs_trans_handle *trans;
392         struct fsxattr fa;
393         unsigned old_flags;
394         unsigned old_i_flags;
395         int ret = 0;
396
397         if (!inode_owner_or_capable(inode))
398                 return -EPERM;
399
400         if (btrfs_root_readonly(root))
401                 return -EROFS;
402
403         memset(&fa, 0, sizeof(fa));
404         if (copy_from_user(&fa, arg, sizeof(fa)))
405                 return -EFAULT;
406
407         ret = check_xflags(fa.fsx_xflags);
408         if (ret)
409                 return ret;
410
411         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
412                 return -EOPNOTSUPP;
413
414         ret = mnt_want_write_file(file);
415         if (ret)
416                 return ret;
417
418         inode_lock(inode);
419
420         old_flags = binode->flags;
421         old_i_flags = inode->i_flags;
422
423         /* We need the capabilities to change append-only or immutable inode */
424         if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) ||
425              (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) &&
426             !capable(CAP_LINUX_IMMUTABLE)) {
427                 ret = -EPERM;
428                 goto out_unlock;
429         }
430
431         if (fa.fsx_xflags & FS_XFLAG_SYNC)
432                 binode->flags |= BTRFS_INODE_SYNC;
433         else
434                 binode->flags &= ~BTRFS_INODE_SYNC;
435         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
436                 binode->flags |= BTRFS_INODE_IMMUTABLE;
437         else
438                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
439         if (fa.fsx_xflags & FS_XFLAG_APPEND)
440                 binode->flags |= BTRFS_INODE_APPEND;
441         else
442                 binode->flags &= ~BTRFS_INODE_APPEND;
443         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
444                 binode->flags |= BTRFS_INODE_NODUMP;
445         else
446                 binode->flags &= ~BTRFS_INODE_NODUMP;
447         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
448                 binode->flags |= BTRFS_INODE_NOATIME;
449         else
450                 binode->flags &= ~BTRFS_INODE_NOATIME;
451
452         /* 1 item for the inode */
453         trans = btrfs_start_transaction(root, 1);
454         if (IS_ERR(trans)) {
455                 ret = PTR_ERR(trans);
456                 goto out_unlock;
457         }
458
459         btrfs_sync_inode_flags_to_i_flags(inode);
460         inode_inc_iversion(inode);
461         inode->i_ctime = current_time(inode);
462         ret = btrfs_update_inode(trans, root, inode);
463
464         btrfs_end_transaction(trans);
465
466 out_unlock:
467         if (ret) {
468                 binode->flags = old_flags;
469                 inode->i_flags = old_i_flags;
470         }
471
472         inode_unlock(inode);
473         mnt_drop_write_file(file);
474
475         return ret;
476 }
477
478 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
479 {
480         struct inode *inode = file_inode(file);
481
482         return put_user(inode->i_generation, arg);
483 }
484
485 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
486 {
487         struct inode *inode = file_inode(file);
488         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
489         struct btrfs_device *device;
490         struct request_queue *q;
491         struct fstrim_range range;
492         u64 minlen = ULLONG_MAX;
493         u64 num_devices = 0;
494         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
495         int ret;
496
497         if (!capable(CAP_SYS_ADMIN))
498                 return -EPERM;
499
500         rcu_read_lock();
501         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
502                                 dev_list) {
503                 if (!device->bdev)
504                         continue;
505                 q = bdev_get_queue(device->bdev);
506                 if (blk_queue_discard(q)) {
507                         num_devices++;
508                         minlen = min_t(u64, q->limits.discard_granularity,
509                                      minlen);
510                 }
511         }
512         rcu_read_unlock();
513
514         if (!num_devices)
515                 return -EOPNOTSUPP;
516         if (copy_from_user(&range, arg, sizeof(range)))
517                 return -EFAULT;
518         if (range.start > total_bytes ||
519             range.len < fs_info->sb->s_blocksize)
520                 return -EINVAL;
521
522         range.len = min(range.len, total_bytes - range.start);
523         range.minlen = max(range.minlen, minlen);
524         ret = btrfs_trim_fs(fs_info, &range);
525         if (ret < 0)
526                 return ret;
527
528         if (copy_to_user(arg, &range, sizeof(range)))
529                 return -EFAULT;
530
531         return 0;
532 }
533
534 int btrfs_is_empty_uuid(u8 *uuid)
535 {
536         int i;
537
538         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
539                 if (uuid[i])
540                         return 0;
541         }
542         return 1;
543 }
544
545 static noinline int create_subvol(struct inode *dir,
546                                   struct dentry *dentry,
547                                   const char *name, int namelen,
548                                   u64 *async_transid,
549                                   struct btrfs_qgroup_inherit *inherit)
550 {
551         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
552         struct btrfs_trans_handle *trans;
553         struct btrfs_key key;
554         struct btrfs_root_item *root_item;
555         struct btrfs_inode_item *inode_item;
556         struct extent_buffer *leaf;
557         struct btrfs_root *root = BTRFS_I(dir)->root;
558         struct btrfs_root *new_root;
559         struct btrfs_block_rsv block_rsv;
560         struct timespec64 cur_time = current_time(dir);
561         struct inode *inode;
562         int ret;
563         int err;
564         u64 objectid;
565         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
566         u64 index = 0;
567         uuid_le new_uuid;
568
569         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
570         if (!root_item)
571                 return -ENOMEM;
572
573         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
574         if (ret)
575                 goto fail_free;
576
577         /*
578          * Don't create subvolume whose level is not zero. Or qgroup will be
579          * screwed up since it assumes subvolume qgroup's level to be 0.
580          */
581         if (btrfs_qgroup_level(objectid)) {
582                 ret = -ENOSPC;
583                 goto fail_free;
584         }
585
586         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
587         /*
588          * The same as the snapshot creation, please see the comment
589          * of create_snapshot().
590          */
591         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
592         if (ret)
593                 goto fail_free;
594
595         trans = btrfs_start_transaction(root, 0);
596         if (IS_ERR(trans)) {
597                 ret = PTR_ERR(trans);
598                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
599                 goto fail_free;
600         }
601         trans->block_rsv = &block_rsv;
602         trans->bytes_reserved = block_rsv.size;
603
604         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
605         if (ret)
606                 goto fail;
607
608         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
609         if (IS_ERR(leaf)) {
610                 ret = PTR_ERR(leaf);
611                 goto fail;
612         }
613
614         btrfs_mark_buffer_dirty(leaf);
615
616         inode_item = &root_item->inode;
617         btrfs_set_stack_inode_generation(inode_item, 1);
618         btrfs_set_stack_inode_size(inode_item, 3);
619         btrfs_set_stack_inode_nlink(inode_item, 1);
620         btrfs_set_stack_inode_nbytes(inode_item,
621                                      fs_info->nodesize);
622         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
623
624         btrfs_set_root_flags(root_item, 0);
625         btrfs_set_root_limit(root_item, 0);
626         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
627
628         btrfs_set_root_bytenr(root_item, leaf->start);
629         btrfs_set_root_generation(root_item, trans->transid);
630         btrfs_set_root_level(root_item, 0);
631         btrfs_set_root_refs(root_item, 1);
632         btrfs_set_root_used(root_item, leaf->len);
633         btrfs_set_root_last_snapshot(root_item, 0);
634
635         btrfs_set_root_generation_v2(root_item,
636                         btrfs_root_generation(root_item));
637         uuid_le_gen(&new_uuid);
638         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
639         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
640         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
641         root_item->ctime = root_item->otime;
642         btrfs_set_root_ctransid(root_item, trans->transid);
643         btrfs_set_root_otransid(root_item, trans->transid);
644
645         btrfs_tree_unlock(leaf);
646         free_extent_buffer(leaf);
647         leaf = NULL;
648
649         btrfs_set_root_dirid(root_item, new_dirid);
650
651         key.objectid = objectid;
652         key.offset = 0;
653         key.type = BTRFS_ROOT_ITEM_KEY;
654         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
655                                 root_item);
656         if (ret)
657                 goto fail;
658
659         key.offset = (u64)-1;
660         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
661         if (IS_ERR(new_root)) {
662                 ret = PTR_ERR(new_root);
663                 btrfs_abort_transaction(trans, ret);
664                 goto fail;
665         }
666
667         btrfs_record_root_in_trans(trans, new_root);
668
669         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
670         if (ret) {
671                 /* We potentially lose an unused inode item here */
672                 btrfs_abort_transaction(trans, ret);
673                 goto fail;
674         }
675
676         mutex_lock(&new_root->objectid_mutex);
677         new_root->highest_objectid = new_dirid;
678         mutex_unlock(&new_root->objectid_mutex);
679
680         /*
681          * insert the directory item
682          */
683         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
684         if (ret) {
685                 btrfs_abort_transaction(trans, ret);
686                 goto fail;
687         }
688
689         ret = btrfs_insert_dir_item(trans, root,
690                                     name, namelen, BTRFS_I(dir), &key,
691                                     BTRFS_FT_DIR, index);
692         if (ret) {
693                 btrfs_abort_transaction(trans, ret);
694                 goto fail;
695         }
696
697         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
698         ret = btrfs_update_inode(trans, root, dir);
699         BUG_ON(ret);
700
701         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
702                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
703         BUG_ON(ret);
704
705         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
706                                   BTRFS_UUID_KEY_SUBVOL, objectid);
707         if (ret)
708                 btrfs_abort_transaction(trans, ret);
709
710 fail:
711         kfree(root_item);
712         trans->block_rsv = NULL;
713         trans->bytes_reserved = 0;
714         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
715
716         if (async_transid) {
717                 *async_transid = trans->transid;
718                 err = btrfs_commit_transaction_async(trans, 1);
719                 if (err)
720                         err = btrfs_commit_transaction(trans);
721         } else {
722                 err = btrfs_commit_transaction(trans);
723         }
724         if (err && !ret)
725                 ret = err;
726
727         if (!ret) {
728                 inode = btrfs_lookup_dentry(dir, dentry);
729                 if (IS_ERR(inode))
730                         return PTR_ERR(inode);
731                 d_instantiate(dentry, inode);
732         }
733         return ret;
734
735 fail_free:
736         kfree(root_item);
737         return ret;
738 }
739
740 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
741                            struct dentry *dentry,
742                            u64 *async_transid, bool readonly,
743                            struct btrfs_qgroup_inherit *inherit)
744 {
745         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
746         struct inode *inode;
747         struct btrfs_pending_snapshot *pending_snapshot;
748         struct btrfs_trans_handle *trans;
749         int ret;
750         bool snapshot_force_cow = false;
751
752         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
753                 return -EINVAL;
754
755         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
756         if (!pending_snapshot)
757                 return -ENOMEM;
758
759         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
760                         GFP_KERNEL);
761         pending_snapshot->path = btrfs_alloc_path();
762         if (!pending_snapshot->root_item || !pending_snapshot->path) {
763                 ret = -ENOMEM;
764                 goto free_pending;
765         }
766
767         /*
768          * Force new buffered writes to reserve space even when NOCOW is
769          * possible. This is to avoid later writeback (running dealloc) to
770          * fallback to COW mode and unexpectedly fail with ENOSPC.
771          */
772         atomic_inc(&root->will_be_snapshotted);
773         smp_mb__after_atomic();
774         /* wait for no snapshot writes */
775         wait_event(root->subv_writers->wait,
776                    percpu_counter_sum(&root->subv_writers->counter) == 0);
777
778         ret = btrfs_start_delalloc_inodes(root);
779         if (ret)
780                 goto dec_and_free;
781
782         /*
783          * All previous writes have started writeback in NOCOW mode, so now
784          * we force future writes to fallback to COW mode during snapshot
785          * creation.
786          */
787         atomic_inc(&root->snapshot_force_cow);
788         snapshot_force_cow = true;
789
790         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
791
792         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
793                              BTRFS_BLOCK_RSV_TEMP);
794         /*
795          * 1 - parent dir inode
796          * 2 - dir entries
797          * 1 - root item
798          * 2 - root ref/backref
799          * 1 - root of snapshot
800          * 1 - UUID item
801          */
802         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
803                                         &pending_snapshot->block_rsv, 8,
804                                         false);
805         if (ret)
806                 goto dec_and_free;
807
808         pending_snapshot->dentry = dentry;
809         pending_snapshot->root = root;
810         pending_snapshot->readonly = readonly;
811         pending_snapshot->dir = dir;
812         pending_snapshot->inherit = inherit;
813
814         trans = btrfs_start_transaction(root, 0);
815         if (IS_ERR(trans)) {
816                 ret = PTR_ERR(trans);
817                 goto fail;
818         }
819
820         spin_lock(&fs_info->trans_lock);
821         list_add(&pending_snapshot->list,
822                  &trans->transaction->pending_snapshots);
823         spin_unlock(&fs_info->trans_lock);
824         if (async_transid) {
825                 *async_transid = trans->transid;
826                 ret = btrfs_commit_transaction_async(trans, 1);
827                 if (ret)
828                         ret = btrfs_commit_transaction(trans);
829         } else {
830                 ret = btrfs_commit_transaction(trans);
831         }
832         if (ret)
833                 goto fail;
834
835         ret = pending_snapshot->error;
836         if (ret)
837                 goto fail;
838
839         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
840         if (ret)
841                 goto fail;
842
843         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
844         if (IS_ERR(inode)) {
845                 ret = PTR_ERR(inode);
846                 goto fail;
847         }
848
849         d_instantiate(dentry, inode);
850         ret = 0;
851 fail:
852         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
853 dec_and_free:
854         if (snapshot_force_cow)
855                 atomic_dec(&root->snapshot_force_cow);
856         if (atomic_dec_and_test(&root->will_be_snapshotted))
857                 wake_up_var(&root->will_be_snapshotted);
858 free_pending:
859         kfree(pending_snapshot->root_item);
860         btrfs_free_path(pending_snapshot->path);
861         kfree(pending_snapshot);
862
863         return ret;
864 }
865
866 /*  copy of may_delete in fs/namei.c()
867  *      Check whether we can remove a link victim from directory dir, check
868  *  whether the type of victim is right.
869  *  1. We can't do it if dir is read-only (done in permission())
870  *  2. We should have write and exec permissions on dir
871  *  3. We can't remove anything from append-only dir
872  *  4. We can't do anything with immutable dir (done in permission())
873  *  5. If the sticky bit on dir is set we should either
874  *      a. be owner of dir, or
875  *      b. be owner of victim, or
876  *      c. have CAP_FOWNER capability
877  *  6. If the victim is append-only or immutable we can't do anything with
878  *     links pointing to it.
879  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
880  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
881  *  9. We can't remove a root or mountpoint.
882  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
883  *     nfs_async_unlink().
884  */
885
886 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
887 {
888         int error;
889
890         if (d_really_is_negative(victim))
891                 return -ENOENT;
892
893         BUG_ON(d_inode(victim->d_parent) != dir);
894         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
895
896         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
897         if (error)
898                 return error;
899         if (IS_APPEND(dir))
900                 return -EPERM;
901         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
902             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
903                 return -EPERM;
904         if (isdir) {
905                 if (!d_is_dir(victim))
906                         return -ENOTDIR;
907                 if (IS_ROOT(victim))
908                         return -EBUSY;
909         } else if (d_is_dir(victim))
910                 return -EISDIR;
911         if (IS_DEADDIR(dir))
912                 return -ENOENT;
913         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
914                 return -EBUSY;
915         return 0;
916 }
917
918 /* copy of may_create in fs/namei.c() */
919 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
920 {
921         if (d_really_is_positive(child))
922                 return -EEXIST;
923         if (IS_DEADDIR(dir))
924                 return -ENOENT;
925         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
926 }
927
928 /*
929  * Create a new subvolume below @parent.  This is largely modeled after
930  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
931  * inside this filesystem so it's quite a bit simpler.
932  */
933 static noinline int btrfs_mksubvol(const struct path *parent,
934                                    const char *name, int namelen,
935                                    struct btrfs_root *snap_src,
936                                    u64 *async_transid, bool readonly,
937                                    struct btrfs_qgroup_inherit *inherit)
938 {
939         struct inode *dir = d_inode(parent->dentry);
940         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
941         struct dentry *dentry;
942         int error;
943
944         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
945         if (error == -EINTR)
946                 return error;
947
948         dentry = lookup_one_len(name, parent->dentry, namelen);
949         error = PTR_ERR(dentry);
950         if (IS_ERR(dentry))
951                 goto out_unlock;
952
953         error = btrfs_may_create(dir, dentry);
954         if (error)
955                 goto out_dput;
956
957         /*
958          * even if this name doesn't exist, we may get hash collisions.
959          * check for them now when we can safely fail
960          */
961         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
962                                                dir->i_ino, name,
963                                                namelen);
964         if (error)
965                 goto out_dput;
966
967         down_read(&fs_info->subvol_sem);
968
969         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
970                 goto out_up_read;
971
972         if (snap_src) {
973                 error = create_snapshot(snap_src, dir, dentry,
974                                         async_transid, readonly, inherit);
975         } else {
976                 error = create_subvol(dir, dentry, name, namelen,
977                                       async_transid, inherit);
978         }
979         if (!error)
980                 fsnotify_mkdir(dir, dentry);
981 out_up_read:
982         up_read(&fs_info->subvol_sem);
983 out_dput:
984         dput(dentry);
985 out_unlock:
986         inode_unlock(dir);
987         return error;
988 }
989
990 /*
991  * When we're defragging a range, we don't want to kick it off again
992  * if it is really just waiting for delalloc to send it down.
993  * If we find a nice big extent or delalloc range for the bytes in the
994  * file you want to defrag, we return 0 to let you know to skip this
995  * part of the file
996  */
997 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
998 {
999         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1000         struct extent_map *em = NULL;
1001         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1002         u64 end;
1003
1004         read_lock(&em_tree->lock);
1005         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1006         read_unlock(&em_tree->lock);
1007
1008         if (em) {
1009                 end = extent_map_end(em);
1010                 free_extent_map(em);
1011                 if (end - offset > thresh)
1012                         return 0;
1013         }
1014         /* if we already have a nice delalloc here, just stop */
1015         thresh /= 2;
1016         end = count_range_bits(io_tree, &offset, offset + thresh,
1017                                thresh, EXTENT_DELALLOC, 1);
1018         if (end >= thresh)
1019                 return 0;
1020         return 1;
1021 }
1022
1023 /*
1024  * helper function to walk through a file and find extents
1025  * newer than a specific transid, and smaller than thresh.
1026  *
1027  * This is used by the defragging code to find new and small
1028  * extents
1029  */
1030 static int find_new_extents(struct btrfs_root *root,
1031                             struct inode *inode, u64 newer_than,
1032                             u64 *off, u32 thresh)
1033 {
1034         struct btrfs_path *path;
1035         struct btrfs_key min_key;
1036         struct extent_buffer *leaf;
1037         struct btrfs_file_extent_item *extent;
1038         int type;
1039         int ret;
1040         u64 ino = btrfs_ino(BTRFS_I(inode));
1041
1042         path = btrfs_alloc_path();
1043         if (!path)
1044                 return -ENOMEM;
1045
1046         min_key.objectid = ino;
1047         min_key.type = BTRFS_EXTENT_DATA_KEY;
1048         min_key.offset = *off;
1049
1050         while (1) {
1051                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1052                 if (ret != 0)
1053                         goto none;
1054 process_slot:
1055                 if (min_key.objectid != ino)
1056                         goto none;
1057                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1058                         goto none;
1059
1060                 leaf = path->nodes[0];
1061                 extent = btrfs_item_ptr(leaf, path->slots[0],
1062                                         struct btrfs_file_extent_item);
1063
1064                 type = btrfs_file_extent_type(leaf, extent);
1065                 if (type == BTRFS_FILE_EXTENT_REG &&
1066                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1067                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1068                         *off = min_key.offset;
1069                         btrfs_free_path(path);
1070                         return 0;
1071                 }
1072
1073                 path->slots[0]++;
1074                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1075                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1076                         goto process_slot;
1077                 }
1078
1079                 if (min_key.offset == (u64)-1)
1080                         goto none;
1081
1082                 min_key.offset++;
1083                 btrfs_release_path(path);
1084         }
1085 none:
1086         btrfs_free_path(path);
1087         return -ENOENT;
1088 }
1089
1090 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1091 {
1092         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1093         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1094         struct extent_map *em;
1095         u64 len = PAGE_SIZE;
1096
1097         /*
1098          * hopefully we have this extent in the tree already, try without
1099          * the full extent lock
1100          */
1101         read_lock(&em_tree->lock);
1102         em = lookup_extent_mapping(em_tree, start, len);
1103         read_unlock(&em_tree->lock);
1104
1105         if (!em) {
1106                 struct extent_state *cached = NULL;
1107                 u64 end = start + len - 1;
1108
1109                 /* get the big lock and read metadata off disk */
1110                 lock_extent_bits(io_tree, start, end, &cached);
1111                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1112                 unlock_extent_cached(io_tree, start, end, &cached);
1113
1114                 if (IS_ERR(em))
1115                         return NULL;
1116         }
1117
1118         return em;
1119 }
1120
1121 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1122 {
1123         struct extent_map *next;
1124         bool ret = true;
1125
1126         /* this is the last extent */
1127         if (em->start + em->len >= i_size_read(inode))
1128                 return false;
1129
1130         next = defrag_lookup_extent(inode, em->start + em->len);
1131         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1132                 ret = false;
1133         else if ((em->block_start + em->block_len == next->block_start) &&
1134                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1135                 ret = false;
1136
1137         free_extent_map(next);
1138         return ret;
1139 }
1140
1141 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1142                                u64 *last_len, u64 *skip, u64 *defrag_end,
1143                                int compress)
1144 {
1145         struct extent_map *em;
1146         int ret = 1;
1147         bool next_mergeable = true;
1148         bool prev_mergeable = true;
1149
1150         /*
1151          * make sure that once we start defragging an extent, we keep on
1152          * defragging it
1153          */
1154         if (start < *defrag_end)
1155                 return 1;
1156
1157         *skip = 0;
1158
1159         em = defrag_lookup_extent(inode, start);
1160         if (!em)
1161                 return 0;
1162
1163         /* this will cover holes, and inline extents */
1164         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1165                 ret = 0;
1166                 goto out;
1167         }
1168
1169         if (!*defrag_end)
1170                 prev_mergeable = false;
1171
1172         next_mergeable = defrag_check_next_extent(inode, em);
1173         /*
1174          * we hit a real extent, if it is big or the next extent is not a
1175          * real extent, don't bother defragging it
1176          */
1177         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1178             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1179                 ret = 0;
1180 out:
1181         /*
1182          * last_len ends up being a counter of how many bytes we've defragged.
1183          * every time we choose not to defrag an extent, we reset *last_len
1184          * so that the next tiny extent will force a defrag.
1185          *
1186          * The end result of this is that tiny extents before a single big
1187          * extent will force at least part of that big extent to be defragged.
1188          */
1189         if (ret) {
1190                 *defrag_end = extent_map_end(em);
1191         } else {
1192                 *last_len = 0;
1193                 *skip = extent_map_end(em);
1194                 *defrag_end = 0;
1195         }
1196
1197         free_extent_map(em);
1198         return ret;
1199 }
1200
1201 /*
1202  * it doesn't do much good to defrag one or two pages
1203  * at a time.  This pulls in a nice chunk of pages
1204  * to COW and defrag.
1205  *
1206  * It also makes sure the delalloc code has enough
1207  * dirty data to avoid making new small extents as part
1208  * of the defrag
1209  *
1210  * It's a good idea to start RA on this range
1211  * before calling this.
1212  */
1213 static int cluster_pages_for_defrag(struct inode *inode,
1214                                     struct page **pages,
1215                                     unsigned long start_index,
1216                                     unsigned long num_pages)
1217 {
1218         unsigned long file_end;
1219         u64 isize = i_size_read(inode);
1220         u64 page_start;
1221         u64 page_end;
1222         u64 page_cnt;
1223         int ret;
1224         int i;
1225         int i_done;
1226         struct btrfs_ordered_extent *ordered;
1227         struct extent_state *cached_state = NULL;
1228         struct extent_io_tree *tree;
1229         struct extent_changeset *data_reserved = NULL;
1230         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1231
1232         file_end = (isize - 1) >> PAGE_SHIFT;
1233         if (!isize || start_index > file_end)
1234                 return 0;
1235
1236         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1237
1238         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1239                         start_index << PAGE_SHIFT,
1240                         page_cnt << PAGE_SHIFT);
1241         if (ret)
1242                 return ret;
1243         i_done = 0;
1244         tree = &BTRFS_I(inode)->io_tree;
1245
1246         /* step one, lock all the pages */
1247         for (i = 0; i < page_cnt; i++) {
1248                 struct page *page;
1249 again:
1250                 page = find_or_create_page(inode->i_mapping,
1251                                            start_index + i, mask);
1252                 if (!page)
1253                         break;
1254
1255                 page_start = page_offset(page);
1256                 page_end = page_start + PAGE_SIZE - 1;
1257                 while (1) {
1258                         lock_extent_bits(tree, page_start, page_end,
1259                                          &cached_state);
1260                         ordered = btrfs_lookup_ordered_extent(inode,
1261                                                               page_start);
1262                         unlock_extent_cached(tree, page_start, page_end,
1263                                              &cached_state);
1264                         if (!ordered)
1265                                 break;
1266
1267                         unlock_page(page);
1268                         btrfs_start_ordered_extent(inode, ordered, 1);
1269                         btrfs_put_ordered_extent(ordered);
1270                         lock_page(page);
1271                         /*
1272                          * we unlocked the page above, so we need check if
1273                          * it was released or not.
1274                          */
1275                         if (page->mapping != inode->i_mapping) {
1276                                 unlock_page(page);
1277                                 put_page(page);
1278                                 goto again;
1279                         }
1280                 }
1281
1282                 if (!PageUptodate(page)) {
1283                         btrfs_readpage(NULL, page);
1284                         lock_page(page);
1285                         if (!PageUptodate(page)) {
1286                                 unlock_page(page);
1287                                 put_page(page);
1288                                 ret = -EIO;
1289                                 break;
1290                         }
1291                 }
1292
1293                 if (page->mapping != inode->i_mapping) {
1294                         unlock_page(page);
1295                         put_page(page);
1296                         goto again;
1297                 }
1298
1299                 pages[i] = page;
1300                 i_done++;
1301         }
1302         if (!i_done || ret)
1303                 goto out;
1304
1305         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1306                 goto out;
1307
1308         /*
1309          * so now we have a nice long stream of locked
1310          * and up to date pages, lets wait on them
1311          */
1312         for (i = 0; i < i_done; i++)
1313                 wait_on_page_writeback(pages[i]);
1314
1315         page_start = page_offset(pages[0]);
1316         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1317
1318         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1319                          page_start, page_end - 1, &cached_state);
1320         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1321                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1322                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1323                           &cached_state);
1324
1325         if (i_done != page_cnt) {
1326                 spin_lock(&BTRFS_I(inode)->lock);
1327                 BTRFS_I(inode)->outstanding_extents++;
1328                 spin_unlock(&BTRFS_I(inode)->lock);
1329                 btrfs_delalloc_release_space(inode, data_reserved,
1330                                 start_index << PAGE_SHIFT,
1331                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1332         }
1333
1334
1335         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1336                           &cached_state);
1337
1338         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1339                              page_start, page_end - 1, &cached_state);
1340
1341         for (i = 0; i < i_done; i++) {
1342                 clear_page_dirty_for_io(pages[i]);
1343                 ClearPageChecked(pages[i]);
1344                 set_page_extent_mapped(pages[i]);
1345                 set_page_dirty(pages[i]);
1346                 unlock_page(pages[i]);
1347                 put_page(pages[i]);
1348         }
1349         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1350                                        false);
1351         extent_changeset_free(data_reserved);
1352         return i_done;
1353 out:
1354         for (i = 0; i < i_done; i++) {
1355                 unlock_page(pages[i]);
1356                 put_page(pages[i]);
1357         }
1358         btrfs_delalloc_release_space(inode, data_reserved,
1359                         start_index << PAGE_SHIFT,
1360                         page_cnt << PAGE_SHIFT, true);
1361         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1362                                        true);
1363         extent_changeset_free(data_reserved);
1364         return ret;
1365
1366 }
1367
1368 int btrfs_defrag_file(struct inode *inode, struct file *file,
1369                       struct btrfs_ioctl_defrag_range_args *range,
1370                       u64 newer_than, unsigned long max_to_defrag)
1371 {
1372         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1373         struct btrfs_root *root = BTRFS_I(inode)->root;
1374         struct file_ra_state *ra = NULL;
1375         unsigned long last_index;
1376         u64 isize = i_size_read(inode);
1377         u64 last_len = 0;
1378         u64 skip = 0;
1379         u64 defrag_end = 0;
1380         u64 newer_off = range->start;
1381         unsigned long i;
1382         unsigned long ra_index = 0;
1383         int ret;
1384         int defrag_count = 0;
1385         int compress_type = BTRFS_COMPRESS_ZLIB;
1386         u32 extent_thresh = range->extent_thresh;
1387         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1388         unsigned long cluster = max_cluster;
1389         u64 new_align = ~((u64)SZ_128K - 1);
1390         struct page **pages = NULL;
1391         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1392
1393         if (isize == 0)
1394                 return 0;
1395
1396         if (range->start >= isize)
1397                 return -EINVAL;
1398
1399         if (do_compress) {
1400                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1401                         return -EINVAL;
1402                 if (range->compress_type)
1403                         compress_type = range->compress_type;
1404         }
1405
1406         if (extent_thresh == 0)
1407                 extent_thresh = SZ_256K;
1408
1409         /*
1410          * If we were not given a file, allocate a readahead context. As
1411          * readahead is just an optimization, defrag will work without it so
1412          * we don't error out.
1413          */
1414         if (!file) {
1415                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1416                 if (ra)
1417                         file_ra_state_init(ra, inode->i_mapping);
1418         } else {
1419                 ra = &file->f_ra;
1420         }
1421
1422         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1423         if (!pages) {
1424                 ret = -ENOMEM;
1425                 goto out_ra;
1426         }
1427
1428         /* find the last page to defrag */
1429         if (range->start + range->len > range->start) {
1430                 last_index = min_t(u64, isize - 1,
1431                          range->start + range->len - 1) >> PAGE_SHIFT;
1432         } else {
1433                 last_index = (isize - 1) >> PAGE_SHIFT;
1434         }
1435
1436         if (newer_than) {
1437                 ret = find_new_extents(root, inode, newer_than,
1438                                        &newer_off, SZ_64K);
1439                 if (!ret) {
1440                         range->start = newer_off;
1441                         /*
1442                          * we always align our defrag to help keep
1443                          * the extents in the file evenly spaced
1444                          */
1445                         i = (newer_off & new_align) >> PAGE_SHIFT;
1446                 } else
1447                         goto out_ra;
1448         } else {
1449                 i = range->start >> PAGE_SHIFT;
1450         }
1451         if (!max_to_defrag)
1452                 max_to_defrag = last_index - i + 1;
1453
1454         /*
1455          * make writeback starts from i, so the defrag range can be
1456          * written sequentially.
1457          */
1458         if (i < inode->i_mapping->writeback_index)
1459                 inode->i_mapping->writeback_index = i;
1460
1461         while (i <= last_index && defrag_count < max_to_defrag &&
1462                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1463                 /*
1464                  * make sure we stop running if someone unmounts
1465                  * the FS
1466                  */
1467                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1468                         break;
1469
1470                 if (btrfs_defrag_cancelled(fs_info)) {
1471                         btrfs_debug(fs_info, "defrag_file cancelled");
1472                         ret = -EAGAIN;
1473                         break;
1474                 }
1475
1476                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1477                                          extent_thresh, &last_len, &skip,
1478                                          &defrag_end, do_compress)){
1479                         unsigned long next;
1480                         /*
1481                          * the should_defrag function tells us how much to skip
1482                          * bump our counter by the suggested amount
1483                          */
1484                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1485                         i = max(i + 1, next);
1486                         continue;
1487                 }
1488
1489                 if (!newer_than) {
1490                         cluster = (PAGE_ALIGN(defrag_end) >>
1491                                    PAGE_SHIFT) - i;
1492                         cluster = min(cluster, max_cluster);
1493                 } else {
1494                         cluster = max_cluster;
1495                 }
1496
1497                 if (i + cluster > ra_index) {
1498                         ra_index = max(i, ra_index);
1499                         if (ra)
1500                                 page_cache_sync_readahead(inode->i_mapping, ra,
1501                                                 file, ra_index, cluster);
1502                         ra_index += cluster;
1503                 }
1504
1505                 inode_lock(inode);
1506                 if (do_compress)
1507                         BTRFS_I(inode)->defrag_compress = compress_type;
1508                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1509                 if (ret < 0) {
1510                         inode_unlock(inode);
1511                         goto out_ra;
1512                 }
1513
1514                 defrag_count += ret;
1515                 balance_dirty_pages_ratelimited(inode->i_mapping);
1516                 inode_unlock(inode);
1517
1518                 if (newer_than) {
1519                         if (newer_off == (u64)-1)
1520                                 break;
1521
1522                         if (ret > 0)
1523                                 i += ret;
1524
1525                         newer_off = max(newer_off + 1,
1526                                         (u64)i << PAGE_SHIFT);
1527
1528                         ret = find_new_extents(root, inode, newer_than,
1529                                                &newer_off, SZ_64K);
1530                         if (!ret) {
1531                                 range->start = newer_off;
1532                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1533                         } else {
1534                                 break;
1535                         }
1536                 } else {
1537                         if (ret > 0) {
1538                                 i += ret;
1539                                 last_len += ret << PAGE_SHIFT;
1540                         } else {
1541                                 i++;
1542                                 last_len = 0;
1543                         }
1544                 }
1545         }
1546
1547         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1548                 filemap_flush(inode->i_mapping);
1549                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1550                              &BTRFS_I(inode)->runtime_flags))
1551                         filemap_flush(inode->i_mapping);
1552         }
1553
1554         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1555                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1556         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1557                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1558         }
1559
1560         ret = defrag_count;
1561
1562 out_ra:
1563         if (do_compress) {
1564                 inode_lock(inode);
1565                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1566                 inode_unlock(inode);
1567         }
1568         if (!file)
1569                 kfree(ra);
1570         kfree(pages);
1571         return ret;
1572 }
1573
1574 static noinline int btrfs_ioctl_resize(struct file *file,
1575                                         void __user *arg)
1576 {
1577         struct inode *inode = file_inode(file);
1578         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1579         u64 new_size;
1580         u64 old_size;
1581         u64 devid = 1;
1582         struct btrfs_root *root = BTRFS_I(inode)->root;
1583         struct btrfs_ioctl_vol_args *vol_args;
1584         struct btrfs_trans_handle *trans;
1585         struct btrfs_device *device = NULL;
1586         char *sizestr;
1587         char *retptr;
1588         char *devstr = NULL;
1589         int ret = 0;
1590         int mod = 0;
1591
1592         if (!capable(CAP_SYS_ADMIN))
1593                 return -EPERM;
1594
1595         ret = mnt_want_write_file(file);
1596         if (ret)
1597                 return ret;
1598
1599         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1600                 mnt_drop_write_file(file);
1601                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1602         }
1603
1604         vol_args = memdup_user(arg, sizeof(*vol_args));
1605         if (IS_ERR(vol_args)) {
1606                 ret = PTR_ERR(vol_args);
1607                 goto out;
1608         }
1609
1610         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1611
1612         sizestr = vol_args->name;
1613         devstr = strchr(sizestr, ':');
1614         if (devstr) {
1615                 sizestr = devstr + 1;
1616                 *devstr = '\0';
1617                 devstr = vol_args->name;
1618                 ret = kstrtoull(devstr, 10, &devid);
1619                 if (ret)
1620                         goto out_free;
1621                 if (!devid) {
1622                         ret = -EINVAL;
1623                         goto out_free;
1624                 }
1625                 btrfs_info(fs_info, "resizing devid %llu", devid);
1626         }
1627
1628         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1629         if (!device) {
1630                 btrfs_info(fs_info, "resizer unable to find device %llu",
1631                            devid);
1632                 ret = -ENODEV;
1633                 goto out_free;
1634         }
1635
1636         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1637                 btrfs_info(fs_info,
1638                            "resizer unable to apply on readonly device %llu",
1639                        devid);
1640                 ret = -EPERM;
1641                 goto out_free;
1642         }
1643
1644         if (!strcmp(sizestr, "max"))
1645                 new_size = device->bdev->bd_inode->i_size;
1646         else {
1647                 if (sizestr[0] == '-') {
1648                         mod = -1;
1649                         sizestr++;
1650                 } else if (sizestr[0] == '+') {
1651                         mod = 1;
1652                         sizestr++;
1653                 }
1654                 new_size = memparse(sizestr, &retptr);
1655                 if (*retptr != '\0' || new_size == 0) {
1656                         ret = -EINVAL;
1657                         goto out_free;
1658                 }
1659         }
1660
1661         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1662                 ret = -EPERM;
1663                 goto out_free;
1664         }
1665
1666         old_size = btrfs_device_get_total_bytes(device);
1667
1668         if (mod < 0) {
1669                 if (new_size > old_size) {
1670                         ret = -EINVAL;
1671                         goto out_free;
1672                 }
1673                 new_size = old_size - new_size;
1674         } else if (mod > 0) {
1675                 if (new_size > ULLONG_MAX - old_size) {
1676                         ret = -ERANGE;
1677                         goto out_free;
1678                 }
1679                 new_size = old_size + new_size;
1680         }
1681
1682         if (new_size < SZ_256M) {
1683                 ret = -EINVAL;
1684                 goto out_free;
1685         }
1686         if (new_size > device->bdev->bd_inode->i_size) {
1687                 ret = -EFBIG;
1688                 goto out_free;
1689         }
1690
1691         new_size = round_down(new_size, fs_info->sectorsize);
1692
1693         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1694                           rcu_str_deref(device->name), new_size);
1695
1696         if (new_size > old_size) {
1697                 trans = btrfs_start_transaction(root, 0);
1698                 if (IS_ERR(trans)) {
1699                         ret = PTR_ERR(trans);
1700                         goto out_free;
1701                 }
1702                 ret = btrfs_grow_device(trans, device, new_size);
1703                 btrfs_commit_transaction(trans);
1704         } else if (new_size < old_size) {
1705                 ret = btrfs_shrink_device(device, new_size);
1706         } /* equal, nothing need to do */
1707
1708 out_free:
1709         kfree(vol_args);
1710 out:
1711         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1712         mnt_drop_write_file(file);
1713         return ret;
1714 }
1715
1716 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1717                                 const char *name, unsigned long fd, int subvol,
1718                                 u64 *transid, bool readonly,
1719                                 struct btrfs_qgroup_inherit *inherit)
1720 {
1721         int namelen;
1722         int ret = 0;
1723
1724         if (!S_ISDIR(file_inode(file)->i_mode))
1725                 return -ENOTDIR;
1726
1727         ret = mnt_want_write_file(file);
1728         if (ret)
1729                 goto out;
1730
1731         namelen = strlen(name);
1732         if (strchr(name, '/')) {
1733                 ret = -EINVAL;
1734                 goto out_drop_write;
1735         }
1736
1737         if (name[0] == '.' &&
1738            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1739                 ret = -EEXIST;
1740                 goto out_drop_write;
1741         }
1742
1743         if (subvol) {
1744                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1745                                      NULL, transid, readonly, inherit);
1746         } else {
1747                 struct fd src = fdget(fd);
1748                 struct inode *src_inode;
1749                 if (!src.file) {
1750                         ret = -EINVAL;
1751                         goto out_drop_write;
1752                 }
1753
1754                 src_inode = file_inode(src.file);
1755                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1756                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1757                                    "Snapshot src from another FS");
1758                         ret = -EXDEV;
1759                 } else if (!inode_owner_or_capable(src_inode)) {
1760                         /*
1761                          * Subvolume creation is not restricted, but snapshots
1762                          * are limited to own subvolumes only
1763                          */
1764                         ret = -EPERM;
1765                 } else {
1766                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1767                                              BTRFS_I(src_inode)->root,
1768                                              transid, readonly, inherit);
1769                 }
1770                 fdput(src);
1771         }
1772 out_drop_write:
1773         mnt_drop_write_file(file);
1774 out:
1775         return ret;
1776 }
1777
1778 static noinline int btrfs_ioctl_snap_create(struct file *file,
1779                                             void __user *arg, int subvol)
1780 {
1781         struct btrfs_ioctl_vol_args *vol_args;
1782         int ret;
1783
1784         if (!S_ISDIR(file_inode(file)->i_mode))
1785                 return -ENOTDIR;
1786
1787         vol_args = memdup_user(arg, sizeof(*vol_args));
1788         if (IS_ERR(vol_args))
1789                 return PTR_ERR(vol_args);
1790         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1791
1792         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1793                                               vol_args->fd, subvol,
1794                                               NULL, false, NULL);
1795
1796         kfree(vol_args);
1797         return ret;
1798 }
1799
1800 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1801                                                void __user *arg, int subvol)
1802 {
1803         struct btrfs_ioctl_vol_args_v2 *vol_args;
1804         int ret;
1805         u64 transid = 0;
1806         u64 *ptr = NULL;
1807         bool readonly = false;
1808         struct btrfs_qgroup_inherit *inherit = NULL;
1809
1810         if (!S_ISDIR(file_inode(file)->i_mode))
1811                 return -ENOTDIR;
1812
1813         vol_args = memdup_user(arg, sizeof(*vol_args));
1814         if (IS_ERR(vol_args))
1815                 return PTR_ERR(vol_args);
1816         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1817
1818         if (vol_args->flags &
1819             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1820               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1821                 ret = -EOPNOTSUPP;
1822                 goto free_args;
1823         }
1824
1825         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1826                 ptr = &transid;
1827         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1828                 readonly = true;
1829         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1830                 if (vol_args->size > PAGE_SIZE) {
1831                         ret = -EINVAL;
1832                         goto free_args;
1833                 }
1834                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1835                 if (IS_ERR(inherit)) {
1836                         ret = PTR_ERR(inherit);
1837                         goto free_args;
1838                 }
1839         }
1840
1841         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1842                                               vol_args->fd, subvol, ptr,
1843                                               readonly, inherit);
1844         if (ret)
1845                 goto free_inherit;
1846
1847         if (ptr && copy_to_user(arg +
1848                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1849                                         transid),
1850                                 ptr, sizeof(*ptr)))
1851                 ret = -EFAULT;
1852
1853 free_inherit:
1854         kfree(inherit);
1855 free_args:
1856         kfree(vol_args);
1857         return ret;
1858 }
1859
1860 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1861                                                 void __user *arg)
1862 {
1863         struct inode *inode = file_inode(file);
1864         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1865         struct btrfs_root *root = BTRFS_I(inode)->root;
1866         int ret = 0;
1867         u64 flags = 0;
1868
1869         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1870                 return -EINVAL;
1871
1872         down_read(&fs_info->subvol_sem);
1873         if (btrfs_root_readonly(root))
1874                 flags |= BTRFS_SUBVOL_RDONLY;
1875         up_read(&fs_info->subvol_sem);
1876
1877         if (copy_to_user(arg, &flags, sizeof(flags)))
1878                 ret = -EFAULT;
1879
1880         return ret;
1881 }
1882
1883 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1884                                               void __user *arg)
1885 {
1886         struct inode *inode = file_inode(file);
1887         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1888         struct btrfs_root *root = BTRFS_I(inode)->root;
1889         struct btrfs_trans_handle *trans;
1890         u64 root_flags;
1891         u64 flags;
1892         int ret = 0;
1893
1894         if (!inode_owner_or_capable(inode))
1895                 return -EPERM;
1896
1897         ret = mnt_want_write_file(file);
1898         if (ret)
1899                 goto out;
1900
1901         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1902                 ret = -EINVAL;
1903                 goto out_drop_write;
1904         }
1905
1906         if (copy_from_user(&flags, arg, sizeof(flags))) {
1907                 ret = -EFAULT;
1908                 goto out_drop_write;
1909         }
1910
1911         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1912                 ret = -EINVAL;
1913                 goto out_drop_write;
1914         }
1915
1916         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1917                 ret = -EOPNOTSUPP;
1918                 goto out_drop_write;
1919         }
1920
1921         down_write(&fs_info->subvol_sem);
1922
1923         /* nothing to do */
1924         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1925                 goto out_drop_sem;
1926
1927         root_flags = btrfs_root_flags(&root->root_item);
1928         if (flags & BTRFS_SUBVOL_RDONLY) {
1929                 btrfs_set_root_flags(&root->root_item,
1930                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1931         } else {
1932                 /*
1933                  * Block RO -> RW transition if this subvolume is involved in
1934                  * send
1935                  */
1936                 spin_lock(&root->root_item_lock);
1937                 if (root->send_in_progress == 0) {
1938                         btrfs_set_root_flags(&root->root_item,
1939                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1940                         spin_unlock(&root->root_item_lock);
1941                 } else {
1942                         spin_unlock(&root->root_item_lock);
1943                         btrfs_warn(fs_info,
1944                                    "Attempt to set subvolume %llu read-write during send",
1945                                    root->root_key.objectid);
1946                         ret = -EPERM;
1947                         goto out_drop_sem;
1948                 }
1949         }
1950
1951         trans = btrfs_start_transaction(root, 1);
1952         if (IS_ERR(trans)) {
1953                 ret = PTR_ERR(trans);
1954                 goto out_reset;
1955         }
1956
1957         ret = btrfs_update_root(trans, fs_info->tree_root,
1958                                 &root->root_key, &root->root_item);
1959         if (ret < 0) {
1960                 btrfs_end_transaction(trans);
1961                 goto out_reset;
1962         }
1963
1964         ret = btrfs_commit_transaction(trans);
1965
1966 out_reset:
1967         if (ret)
1968                 btrfs_set_root_flags(&root->root_item, root_flags);
1969 out_drop_sem:
1970         up_write(&fs_info->subvol_sem);
1971 out_drop_write:
1972         mnt_drop_write_file(file);
1973 out:
1974         return ret;
1975 }
1976
1977 static noinline int key_in_sk(struct btrfs_key *key,
1978                               struct btrfs_ioctl_search_key *sk)
1979 {
1980         struct btrfs_key test;
1981         int ret;
1982
1983         test.objectid = sk->min_objectid;
1984         test.type = sk->min_type;
1985         test.offset = sk->min_offset;
1986
1987         ret = btrfs_comp_cpu_keys(key, &test);
1988         if (ret < 0)
1989                 return 0;
1990
1991         test.objectid = sk->max_objectid;
1992         test.type = sk->max_type;
1993         test.offset = sk->max_offset;
1994
1995         ret = btrfs_comp_cpu_keys(key, &test);
1996         if (ret > 0)
1997                 return 0;
1998         return 1;
1999 }
2000
2001 static noinline int copy_to_sk(struct btrfs_path *path,
2002                                struct btrfs_key *key,
2003                                struct btrfs_ioctl_search_key *sk,
2004                                size_t *buf_size,
2005                                char __user *ubuf,
2006                                unsigned long *sk_offset,
2007                                int *num_found)
2008 {
2009         u64 found_transid;
2010         struct extent_buffer *leaf;
2011         struct btrfs_ioctl_search_header sh;
2012         struct btrfs_key test;
2013         unsigned long item_off;
2014         unsigned long item_len;
2015         int nritems;
2016         int i;
2017         int slot;
2018         int ret = 0;
2019
2020         leaf = path->nodes[0];
2021         slot = path->slots[0];
2022         nritems = btrfs_header_nritems(leaf);
2023
2024         if (btrfs_header_generation(leaf) > sk->max_transid) {
2025                 i = nritems;
2026                 goto advance_key;
2027         }
2028         found_transid = btrfs_header_generation(leaf);
2029
2030         for (i = slot; i < nritems; i++) {
2031                 item_off = btrfs_item_ptr_offset(leaf, i);
2032                 item_len = btrfs_item_size_nr(leaf, i);
2033
2034                 btrfs_item_key_to_cpu(leaf, key, i);
2035                 if (!key_in_sk(key, sk))
2036                         continue;
2037
2038                 if (sizeof(sh) + item_len > *buf_size) {
2039                         if (*num_found) {
2040                                 ret = 1;
2041                                 goto out;
2042                         }
2043
2044                         /*
2045                          * return one empty item back for v1, which does not
2046                          * handle -EOVERFLOW
2047                          */
2048
2049                         *buf_size = sizeof(sh) + item_len;
2050                         item_len = 0;
2051                         ret = -EOVERFLOW;
2052                 }
2053
2054                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2055                         ret = 1;
2056                         goto out;
2057                 }
2058
2059                 sh.objectid = key->objectid;
2060                 sh.offset = key->offset;
2061                 sh.type = key->type;
2062                 sh.len = item_len;
2063                 sh.transid = found_transid;
2064
2065                 /* copy search result header */
2066                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2067                         ret = -EFAULT;
2068                         goto out;
2069                 }
2070
2071                 *sk_offset += sizeof(sh);
2072
2073                 if (item_len) {
2074                         char __user *up = ubuf + *sk_offset;
2075                         /* copy the item */
2076                         if (read_extent_buffer_to_user(leaf, up,
2077                                                        item_off, item_len)) {
2078                                 ret = -EFAULT;
2079                                 goto out;
2080                         }
2081
2082                         *sk_offset += item_len;
2083                 }
2084                 (*num_found)++;
2085
2086                 if (ret) /* -EOVERFLOW from above */
2087                         goto out;
2088
2089                 if (*num_found >= sk->nr_items) {
2090                         ret = 1;
2091                         goto out;
2092                 }
2093         }
2094 advance_key:
2095         ret = 0;
2096         test.objectid = sk->max_objectid;
2097         test.type = sk->max_type;
2098         test.offset = sk->max_offset;
2099         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2100                 ret = 1;
2101         else if (key->offset < (u64)-1)
2102                 key->offset++;
2103         else if (key->type < (u8)-1) {
2104                 key->offset = 0;
2105                 key->type++;
2106         } else if (key->objectid < (u64)-1) {
2107                 key->offset = 0;
2108                 key->type = 0;
2109                 key->objectid++;
2110         } else
2111                 ret = 1;
2112 out:
2113         /*
2114          *  0: all items from this leaf copied, continue with next
2115          *  1: * more items can be copied, but unused buffer is too small
2116          *     * all items were found
2117          *     Either way, it will stops the loop which iterates to the next
2118          *     leaf
2119          *  -EOVERFLOW: item was to large for buffer
2120          *  -EFAULT: could not copy extent buffer back to userspace
2121          */
2122         return ret;
2123 }
2124
2125 static noinline int search_ioctl(struct inode *inode,
2126                                  struct btrfs_ioctl_search_key *sk,
2127                                  size_t *buf_size,
2128                                  char __user *ubuf)
2129 {
2130         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2131         struct btrfs_root *root;
2132         struct btrfs_key key;
2133         struct btrfs_path *path;
2134         int ret;
2135         int num_found = 0;
2136         unsigned long sk_offset = 0;
2137
2138         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2139                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2140                 return -EOVERFLOW;
2141         }
2142
2143         path = btrfs_alloc_path();
2144         if (!path)
2145                 return -ENOMEM;
2146
2147         if (sk->tree_id == 0) {
2148                 /* search the root of the inode that was passed */
2149                 root = BTRFS_I(inode)->root;
2150         } else {
2151                 key.objectid = sk->tree_id;
2152                 key.type = BTRFS_ROOT_ITEM_KEY;
2153                 key.offset = (u64)-1;
2154                 root = btrfs_read_fs_root_no_name(info, &key);
2155                 if (IS_ERR(root)) {
2156                         btrfs_free_path(path);
2157                         return PTR_ERR(root);
2158                 }
2159         }
2160
2161         key.objectid = sk->min_objectid;
2162         key.type = sk->min_type;
2163         key.offset = sk->min_offset;
2164
2165         while (1) {
2166                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2167                 if (ret != 0) {
2168                         if (ret > 0)
2169                                 ret = 0;
2170                         goto err;
2171                 }
2172                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2173                                  &sk_offset, &num_found);
2174                 btrfs_release_path(path);
2175                 if (ret)
2176                         break;
2177
2178         }
2179         if (ret > 0)
2180                 ret = 0;
2181 err:
2182         sk->nr_items = num_found;
2183         btrfs_free_path(path);
2184         return ret;
2185 }
2186
2187 static noinline int btrfs_ioctl_tree_search(struct file *file,
2188                                            void __user *argp)
2189 {
2190         struct btrfs_ioctl_search_args __user *uargs;
2191         struct btrfs_ioctl_search_key sk;
2192         struct inode *inode;
2193         int ret;
2194         size_t buf_size;
2195
2196         if (!capable(CAP_SYS_ADMIN))
2197                 return -EPERM;
2198
2199         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2200
2201         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2202                 return -EFAULT;
2203
2204         buf_size = sizeof(uargs->buf);
2205
2206         inode = file_inode(file);
2207         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2208
2209         /*
2210          * In the origin implementation an overflow is handled by returning a
2211          * search header with a len of zero, so reset ret.
2212          */
2213         if (ret == -EOVERFLOW)
2214                 ret = 0;
2215
2216         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2217                 ret = -EFAULT;
2218         return ret;
2219 }
2220
2221 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2222                                                void __user *argp)
2223 {
2224         struct btrfs_ioctl_search_args_v2 __user *uarg;
2225         struct btrfs_ioctl_search_args_v2 args;
2226         struct inode *inode;
2227         int ret;
2228         size_t buf_size;
2229         const size_t buf_limit = SZ_16M;
2230
2231         if (!capable(CAP_SYS_ADMIN))
2232                 return -EPERM;
2233
2234         /* copy search header and buffer size */
2235         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2236         if (copy_from_user(&args, uarg, sizeof(args)))
2237                 return -EFAULT;
2238
2239         buf_size = args.buf_size;
2240
2241         /* limit result size to 16MB */
2242         if (buf_size > buf_limit)
2243                 buf_size = buf_limit;
2244
2245         inode = file_inode(file);
2246         ret = search_ioctl(inode, &args.key, &buf_size,
2247                            (char __user *)(&uarg->buf[0]));
2248         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2249                 ret = -EFAULT;
2250         else if (ret == -EOVERFLOW &&
2251                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2252                 ret = -EFAULT;
2253
2254         return ret;
2255 }
2256
2257 /*
2258  * Search INODE_REFs to identify path name of 'dirid' directory
2259  * in a 'tree_id' tree. and sets path name to 'name'.
2260  */
2261 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2262                                 u64 tree_id, u64 dirid, char *name)
2263 {
2264         struct btrfs_root *root;
2265         struct btrfs_key key;
2266         char *ptr;
2267         int ret = -1;
2268         int slot;
2269         int len;
2270         int total_len = 0;
2271         struct btrfs_inode_ref *iref;
2272         struct extent_buffer *l;
2273         struct btrfs_path *path;
2274
2275         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2276                 name[0]='\0';
2277                 return 0;
2278         }
2279
2280         path = btrfs_alloc_path();
2281         if (!path)
2282                 return -ENOMEM;
2283
2284         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2285
2286         key.objectid = tree_id;
2287         key.type = BTRFS_ROOT_ITEM_KEY;
2288         key.offset = (u64)-1;
2289         root = btrfs_read_fs_root_no_name(info, &key);
2290         if (IS_ERR(root)) {
2291                 ret = PTR_ERR(root);
2292                 goto out;
2293         }
2294
2295         key.objectid = dirid;
2296         key.type = BTRFS_INODE_REF_KEY;
2297         key.offset = (u64)-1;
2298
2299         while (1) {
2300                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2301                 if (ret < 0)
2302                         goto out;
2303                 else if (ret > 0) {
2304                         ret = btrfs_previous_item(root, path, dirid,
2305                                                   BTRFS_INODE_REF_KEY);
2306                         if (ret < 0)
2307                                 goto out;
2308                         else if (ret > 0) {
2309                                 ret = -ENOENT;
2310                                 goto out;
2311                         }
2312                 }
2313
2314                 l = path->nodes[0];
2315                 slot = path->slots[0];
2316                 btrfs_item_key_to_cpu(l, &key, slot);
2317
2318                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2319                 len = btrfs_inode_ref_name_len(l, iref);
2320                 ptr -= len + 1;
2321                 total_len += len + 1;
2322                 if (ptr < name) {
2323                         ret = -ENAMETOOLONG;
2324                         goto out;
2325                 }
2326
2327                 *(ptr + len) = '/';
2328                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2329
2330                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2331                         break;
2332
2333                 btrfs_release_path(path);
2334                 key.objectid = key.offset;
2335                 key.offset = (u64)-1;
2336                 dirid = key.objectid;
2337         }
2338         memmove(name, ptr, total_len);
2339         name[total_len] = '\0';
2340         ret = 0;
2341 out:
2342         btrfs_free_path(path);
2343         return ret;
2344 }
2345
2346 static int btrfs_search_path_in_tree_user(struct inode *inode,
2347                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2348 {
2349         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2350         struct super_block *sb = inode->i_sb;
2351         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2352         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2353         u64 dirid = args->dirid;
2354         unsigned long item_off;
2355         unsigned long item_len;
2356         struct btrfs_inode_ref *iref;
2357         struct btrfs_root_ref *rref;
2358         struct btrfs_root *root;
2359         struct btrfs_path *path;
2360         struct btrfs_key key, key2;
2361         struct extent_buffer *leaf;
2362         struct inode *temp_inode;
2363         char *ptr;
2364         int slot;
2365         int len;
2366         int total_len = 0;
2367         int ret;
2368
2369         path = btrfs_alloc_path();
2370         if (!path)
2371                 return -ENOMEM;
2372
2373         /*
2374          * If the bottom subvolume does not exist directly under upper_limit,
2375          * construct the path in from the bottom up.
2376          */
2377         if (dirid != upper_limit.objectid) {
2378                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2379
2380                 key.objectid = treeid;
2381                 key.type = BTRFS_ROOT_ITEM_KEY;
2382                 key.offset = (u64)-1;
2383                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2384                 if (IS_ERR(root)) {
2385                         ret = PTR_ERR(root);
2386                         goto out;
2387                 }
2388
2389                 key.objectid = dirid;
2390                 key.type = BTRFS_INODE_REF_KEY;
2391                 key.offset = (u64)-1;
2392                 while (1) {
2393                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2394                         if (ret < 0) {
2395                                 goto out;
2396                         } else if (ret > 0) {
2397                                 ret = btrfs_previous_item(root, path, dirid,
2398                                                           BTRFS_INODE_REF_KEY);
2399                                 if (ret < 0) {
2400                                         goto out;
2401                                 } else if (ret > 0) {
2402                                         ret = -ENOENT;
2403                                         goto out;
2404                                 }
2405                         }
2406
2407                         leaf = path->nodes[0];
2408                         slot = path->slots[0];
2409                         btrfs_item_key_to_cpu(leaf, &key, slot);
2410
2411                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2412                         len = btrfs_inode_ref_name_len(leaf, iref);
2413                         ptr -= len + 1;
2414                         total_len += len + 1;
2415                         if (ptr < args->path) {
2416                                 ret = -ENAMETOOLONG;
2417                                 goto out;
2418                         }
2419
2420                         *(ptr + len) = '/';
2421                         read_extent_buffer(leaf, ptr,
2422                                         (unsigned long)(iref + 1), len);
2423
2424                         /* Check the read+exec permission of this directory */
2425                         ret = btrfs_previous_item(root, path, dirid,
2426                                                   BTRFS_INODE_ITEM_KEY);
2427                         if (ret < 0) {
2428                                 goto out;
2429                         } else if (ret > 0) {
2430                                 ret = -ENOENT;
2431                                 goto out;
2432                         }
2433
2434                         leaf = path->nodes[0];
2435                         slot = path->slots[0];
2436                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2437                         if (key2.objectid != dirid) {
2438                                 ret = -ENOENT;
2439                                 goto out;
2440                         }
2441
2442                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2443                         if (IS_ERR(temp_inode)) {
2444                                 ret = PTR_ERR(temp_inode);
2445                                 goto out;
2446                         }
2447                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2448                         iput(temp_inode);
2449                         if (ret) {
2450                                 ret = -EACCES;
2451                                 goto out;
2452                         }
2453
2454                         if (key.offset == upper_limit.objectid)
2455                                 break;
2456                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2457                                 ret = -EACCES;
2458                                 goto out;
2459                         }
2460
2461                         btrfs_release_path(path);
2462                         key.objectid = key.offset;
2463                         key.offset = (u64)-1;
2464                         dirid = key.objectid;
2465                 }
2466
2467                 memmove(args->path, ptr, total_len);
2468                 args->path[total_len] = '\0';
2469                 btrfs_release_path(path);
2470         }
2471
2472         /* Get the bottom subvolume's name from ROOT_REF */
2473         root = fs_info->tree_root;
2474         key.objectid = treeid;
2475         key.type = BTRFS_ROOT_REF_KEY;
2476         key.offset = args->treeid;
2477         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2478         if (ret < 0) {
2479                 goto out;
2480         } else if (ret > 0) {
2481                 ret = -ENOENT;
2482                 goto out;
2483         }
2484
2485         leaf = path->nodes[0];
2486         slot = path->slots[0];
2487         btrfs_item_key_to_cpu(leaf, &key, slot);
2488
2489         item_off = btrfs_item_ptr_offset(leaf, slot);
2490         item_len = btrfs_item_size_nr(leaf, slot);
2491         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2492         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2493         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2494                 ret = -EINVAL;
2495                 goto out;
2496         }
2497
2498         /* Copy subvolume's name */
2499         item_off += sizeof(struct btrfs_root_ref);
2500         item_len -= sizeof(struct btrfs_root_ref);
2501         read_extent_buffer(leaf, args->name, item_off, item_len);
2502         args->name[item_len] = 0;
2503
2504 out:
2505         btrfs_free_path(path);
2506         return ret;
2507 }
2508
2509 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2510                                            void __user *argp)
2511 {
2512         struct btrfs_ioctl_ino_lookup_args *args;
2513         struct inode *inode;
2514         int ret = 0;
2515
2516         args = memdup_user(argp, sizeof(*args));
2517         if (IS_ERR(args))
2518                 return PTR_ERR(args);
2519
2520         inode = file_inode(file);
2521
2522         /*
2523          * Unprivileged query to obtain the containing subvolume root id. The
2524          * path is reset so it's consistent with btrfs_search_path_in_tree.
2525          */
2526         if (args->treeid == 0)
2527                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2528
2529         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2530                 args->name[0] = 0;
2531                 goto out;
2532         }
2533
2534         if (!capable(CAP_SYS_ADMIN)) {
2535                 ret = -EPERM;
2536                 goto out;
2537         }
2538
2539         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2540                                         args->treeid, args->objectid,
2541                                         args->name);
2542
2543 out:
2544         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2545                 ret = -EFAULT;
2546
2547         kfree(args);
2548         return ret;
2549 }
2550
2551 /*
2552  * Version of ino_lookup ioctl (unprivileged)
2553  *
2554  * The main differences from ino_lookup ioctl are:
2555  *
2556  *   1. Read + Exec permission will be checked using inode_permission() during
2557  *      path construction. -EACCES will be returned in case of failure.
2558  *   2. Path construction will be stopped at the inode number which corresponds
2559  *      to the fd with which this ioctl is called. If constructed path does not
2560  *      exist under fd's inode, -EACCES will be returned.
2561  *   3. The name of bottom subvolume is also searched and filled.
2562  */
2563 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2564 {
2565         struct btrfs_ioctl_ino_lookup_user_args *args;
2566         struct inode *inode;
2567         int ret;
2568
2569         args = memdup_user(argp, sizeof(*args));
2570         if (IS_ERR(args))
2571                 return PTR_ERR(args);
2572
2573         inode = file_inode(file);
2574
2575         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2576             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2577                 /*
2578                  * The subvolume does not exist under fd with which this is
2579                  * called
2580                  */
2581                 kfree(args);
2582                 return -EACCES;
2583         }
2584
2585         ret = btrfs_search_path_in_tree_user(inode, args);
2586
2587         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2588                 ret = -EFAULT;
2589
2590         kfree(args);
2591         return ret;
2592 }
2593
2594 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2595 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2596 {
2597         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2598         struct btrfs_fs_info *fs_info;
2599         struct btrfs_root *root;
2600         struct btrfs_path *path;
2601         struct btrfs_key key;
2602         struct btrfs_root_item *root_item;
2603         struct btrfs_root_ref *rref;
2604         struct extent_buffer *leaf;
2605         unsigned long item_off;
2606         unsigned long item_len;
2607         struct inode *inode;
2608         int slot;
2609         int ret = 0;
2610
2611         path = btrfs_alloc_path();
2612         if (!path)
2613                 return -ENOMEM;
2614
2615         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2616         if (!subvol_info) {
2617                 btrfs_free_path(path);
2618                 return -ENOMEM;
2619         }
2620
2621         inode = file_inode(file);
2622         fs_info = BTRFS_I(inode)->root->fs_info;
2623
2624         /* Get root_item of inode's subvolume */
2625         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2626         key.type = BTRFS_ROOT_ITEM_KEY;
2627         key.offset = (u64)-1;
2628         root = btrfs_read_fs_root_no_name(fs_info, &key);
2629         if (IS_ERR(root)) {
2630                 ret = PTR_ERR(root);
2631                 goto out;
2632         }
2633         root_item = &root->root_item;
2634
2635         subvol_info->treeid = key.objectid;
2636
2637         subvol_info->generation = btrfs_root_generation(root_item);
2638         subvol_info->flags = btrfs_root_flags(root_item);
2639
2640         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2641         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2642                                                     BTRFS_UUID_SIZE);
2643         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2644                                                     BTRFS_UUID_SIZE);
2645
2646         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2647         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2648         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2649
2650         subvol_info->otransid = btrfs_root_otransid(root_item);
2651         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2652         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2653
2654         subvol_info->stransid = btrfs_root_stransid(root_item);
2655         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2656         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2657
2658         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2659         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2660         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2661
2662         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2663                 /* Search root tree for ROOT_BACKREF of this subvolume */
2664                 root = fs_info->tree_root;
2665
2666                 key.type = BTRFS_ROOT_BACKREF_KEY;
2667                 key.offset = 0;
2668                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2669                 if (ret < 0) {
2670                         goto out;
2671                 } else if (path->slots[0] >=
2672                            btrfs_header_nritems(path->nodes[0])) {
2673                         ret = btrfs_next_leaf(root, path);
2674                         if (ret < 0) {
2675                                 goto out;
2676                         } else if (ret > 0) {
2677                                 ret = -EUCLEAN;
2678                                 goto out;
2679                         }
2680                 }
2681
2682                 leaf = path->nodes[0];
2683                 slot = path->slots[0];
2684                 btrfs_item_key_to_cpu(leaf, &key, slot);
2685                 if (key.objectid == subvol_info->treeid &&
2686                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2687                         subvol_info->parent_id = key.offset;
2688
2689                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2690                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2691
2692                         item_off = btrfs_item_ptr_offset(leaf, slot)
2693                                         + sizeof(struct btrfs_root_ref);
2694                         item_len = btrfs_item_size_nr(leaf, slot)
2695                                         - sizeof(struct btrfs_root_ref);
2696                         read_extent_buffer(leaf, subvol_info->name,
2697                                            item_off, item_len);
2698                 } else {
2699                         ret = -ENOENT;
2700                         goto out;
2701                 }
2702         }
2703
2704         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2705                 ret = -EFAULT;
2706
2707 out:
2708         btrfs_free_path(path);
2709         kzfree(subvol_info);
2710         return ret;
2711 }
2712
2713 /*
2714  * Return ROOT_REF information of the subvolume containing this inode
2715  * except the subvolume name.
2716  */
2717 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2718 {
2719         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2720         struct btrfs_root_ref *rref;
2721         struct btrfs_root *root;
2722         struct btrfs_path *path;
2723         struct btrfs_key key;
2724         struct extent_buffer *leaf;
2725         struct inode *inode;
2726         u64 objectid;
2727         int slot;
2728         int ret;
2729         u8 found;
2730
2731         path = btrfs_alloc_path();
2732         if (!path)
2733                 return -ENOMEM;
2734
2735         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2736         if (IS_ERR(rootrefs)) {
2737                 btrfs_free_path(path);
2738                 return PTR_ERR(rootrefs);
2739         }
2740
2741         inode = file_inode(file);
2742         root = BTRFS_I(inode)->root->fs_info->tree_root;
2743         objectid = BTRFS_I(inode)->root->root_key.objectid;
2744
2745         key.objectid = objectid;
2746         key.type = BTRFS_ROOT_REF_KEY;
2747         key.offset = rootrefs->min_treeid;
2748         found = 0;
2749
2750         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2751         if (ret < 0) {
2752                 goto out;
2753         } else if (path->slots[0] >=
2754                    btrfs_header_nritems(path->nodes[0])) {
2755                 ret = btrfs_next_leaf(root, path);
2756                 if (ret < 0) {
2757                         goto out;
2758                 } else if (ret > 0) {
2759                         ret = -EUCLEAN;
2760                         goto out;
2761                 }
2762         }
2763         while (1) {
2764                 leaf = path->nodes[0];
2765                 slot = path->slots[0];
2766
2767                 btrfs_item_key_to_cpu(leaf, &key, slot);
2768                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2769                         ret = 0;
2770                         goto out;
2771                 }
2772
2773                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2774                         ret = -EOVERFLOW;
2775                         goto out;
2776                 }
2777
2778                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2779                 rootrefs->rootref[found].treeid = key.offset;
2780                 rootrefs->rootref[found].dirid =
2781                                   btrfs_root_ref_dirid(leaf, rref);
2782                 found++;
2783
2784                 ret = btrfs_next_item(root, path);
2785                 if (ret < 0) {
2786                         goto out;
2787                 } else if (ret > 0) {
2788                         ret = -EUCLEAN;
2789                         goto out;
2790                 }
2791         }
2792
2793 out:
2794         if (!ret || ret == -EOVERFLOW) {
2795                 rootrefs->num_items = found;
2796                 /* update min_treeid for next search */
2797                 if (found)
2798                         rootrefs->min_treeid =
2799                                 rootrefs->rootref[found - 1].treeid + 1;
2800                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2801                         ret = -EFAULT;
2802         }
2803
2804         kfree(rootrefs);
2805         btrfs_free_path(path);
2806
2807         return ret;
2808 }
2809
2810 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2811                                              void __user *arg)
2812 {
2813         struct dentry *parent = file->f_path.dentry;
2814         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2815         struct dentry *dentry;
2816         struct inode *dir = d_inode(parent);
2817         struct inode *inode;
2818         struct btrfs_root *root = BTRFS_I(dir)->root;
2819         struct btrfs_root *dest = NULL;
2820         struct btrfs_ioctl_vol_args *vol_args;
2821         int namelen;
2822         int err = 0;
2823
2824         if (!S_ISDIR(dir->i_mode))
2825                 return -ENOTDIR;
2826
2827         vol_args = memdup_user(arg, sizeof(*vol_args));
2828         if (IS_ERR(vol_args))
2829                 return PTR_ERR(vol_args);
2830
2831         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2832         namelen = strlen(vol_args->name);
2833         if (strchr(vol_args->name, '/') ||
2834             strncmp(vol_args->name, "..", namelen) == 0) {
2835                 err = -EINVAL;
2836                 goto out;
2837         }
2838
2839         err = mnt_want_write_file(file);
2840         if (err)
2841                 goto out;
2842
2843
2844         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2845         if (err == -EINTR)
2846                 goto out_drop_write;
2847         dentry = lookup_one_len(vol_args->name, parent, namelen);
2848         if (IS_ERR(dentry)) {
2849                 err = PTR_ERR(dentry);
2850                 goto out_unlock_dir;
2851         }
2852
2853         if (d_really_is_negative(dentry)) {
2854                 err = -ENOENT;
2855                 goto out_dput;
2856         }
2857
2858         inode = d_inode(dentry);
2859         dest = BTRFS_I(inode)->root;
2860         if (!capable(CAP_SYS_ADMIN)) {
2861                 /*
2862                  * Regular user.  Only allow this with a special mount
2863                  * option, when the user has write+exec access to the
2864                  * subvol root, and when rmdir(2) would have been
2865                  * allowed.
2866                  *
2867                  * Note that this is _not_ check that the subvol is
2868                  * empty or doesn't contain data that we wouldn't
2869                  * otherwise be able to delete.
2870                  *
2871                  * Users who want to delete empty subvols should try
2872                  * rmdir(2).
2873                  */
2874                 err = -EPERM;
2875                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2876                         goto out_dput;
2877
2878                 /*
2879                  * Do not allow deletion if the parent dir is the same
2880                  * as the dir to be deleted.  That means the ioctl
2881                  * must be called on the dentry referencing the root
2882                  * of the subvol, not a random directory contained
2883                  * within it.
2884                  */
2885                 err = -EINVAL;
2886                 if (root == dest)
2887                         goto out_dput;
2888
2889                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2890                 if (err)
2891                         goto out_dput;
2892         }
2893
2894         /* check if subvolume may be deleted by a user */
2895         err = btrfs_may_delete(dir, dentry, 1);
2896         if (err)
2897                 goto out_dput;
2898
2899         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2900                 err = -EINVAL;
2901                 goto out_dput;
2902         }
2903
2904         inode_lock(inode);
2905         err = btrfs_delete_subvolume(dir, dentry);
2906         inode_unlock(inode);
2907         if (!err)
2908                 d_delete(dentry);
2909
2910 out_dput:
2911         dput(dentry);
2912 out_unlock_dir:
2913         inode_unlock(dir);
2914 out_drop_write:
2915         mnt_drop_write_file(file);
2916 out:
2917         kfree(vol_args);
2918         return err;
2919 }
2920
2921 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2922 {
2923         struct inode *inode = file_inode(file);
2924         struct btrfs_root *root = BTRFS_I(inode)->root;
2925         struct btrfs_ioctl_defrag_range_args *range;
2926         int ret;
2927
2928         ret = mnt_want_write_file(file);
2929         if (ret)
2930                 return ret;
2931
2932         if (btrfs_root_readonly(root)) {
2933                 ret = -EROFS;
2934                 goto out;
2935         }
2936
2937         switch (inode->i_mode & S_IFMT) {
2938         case S_IFDIR:
2939                 if (!capable(CAP_SYS_ADMIN)) {
2940                         ret = -EPERM;
2941                         goto out;
2942                 }
2943                 ret = btrfs_defrag_root(root);
2944                 break;
2945         case S_IFREG:
2946                 /*
2947                  * Note that this does not check the file descriptor for write
2948                  * access. This prevents defragmenting executables that are
2949                  * running and allows defrag on files open in read-only mode.
2950                  */
2951                 if (!capable(CAP_SYS_ADMIN) &&
2952                     inode_permission(inode, MAY_WRITE)) {
2953                         ret = -EPERM;
2954                         goto out;
2955                 }
2956
2957                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2958                 if (!range) {
2959                         ret = -ENOMEM;
2960                         goto out;
2961                 }
2962
2963                 if (argp) {
2964                         if (copy_from_user(range, argp,
2965                                            sizeof(*range))) {
2966                                 ret = -EFAULT;
2967                                 kfree(range);
2968                                 goto out;
2969                         }
2970                         /* compression requires us to start the IO */
2971                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2972                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2973                                 range->extent_thresh = (u32)-1;
2974                         }
2975                 } else {
2976                         /* the rest are all set to zero by kzalloc */
2977                         range->len = (u64)-1;
2978                 }
2979                 ret = btrfs_defrag_file(file_inode(file), file,
2980                                         range, BTRFS_OLDEST_GENERATION, 0);
2981                 if (ret > 0)
2982                         ret = 0;
2983                 kfree(range);
2984                 break;
2985         default:
2986                 ret = -EINVAL;
2987         }
2988 out:
2989         mnt_drop_write_file(file);
2990         return ret;
2991 }
2992
2993 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2994 {
2995         struct btrfs_ioctl_vol_args *vol_args;
2996         int ret;
2997
2998         if (!capable(CAP_SYS_ADMIN))
2999                 return -EPERM;
3000
3001         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3002                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3003
3004         vol_args = memdup_user(arg, sizeof(*vol_args));
3005         if (IS_ERR(vol_args)) {
3006                 ret = PTR_ERR(vol_args);
3007                 goto out;
3008         }
3009
3010         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3011         ret = btrfs_init_new_device(fs_info, vol_args->name);
3012
3013         if (!ret)
3014                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3015
3016         kfree(vol_args);
3017 out:
3018         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3019         return ret;
3020 }
3021
3022 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3023 {
3024         struct inode *inode = file_inode(file);
3025         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3026         struct btrfs_ioctl_vol_args_v2 *vol_args;
3027         int ret;
3028
3029         if (!capable(CAP_SYS_ADMIN))
3030                 return -EPERM;
3031
3032         ret = mnt_want_write_file(file);
3033         if (ret)
3034                 return ret;
3035
3036         vol_args = memdup_user(arg, sizeof(*vol_args));
3037         if (IS_ERR(vol_args)) {
3038                 ret = PTR_ERR(vol_args);
3039                 goto err_drop;
3040         }
3041
3042         /* Check for compatibility reject unknown flags */
3043         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3044                 ret = -EOPNOTSUPP;
3045                 goto out;
3046         }
3047
3048         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3049                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3050                 goto out;
3051         }
3052
3053         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3054                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3055         } else {
3056                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3057                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3058         }
3059         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3060
3061         if (!ret) {
3062                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3063                         btrfs_info(fs_info, "device deleted: id %llu",
3064                                         vol_args->devid);
3065                 else
3066                         btrfs_info(fs_info, "device deleted: %s",
3067                                         vol_args->name);
3068         }
3069 out:
3070         kfree(vol_args);
3071 err_drop:
3072         mnt_drop_write_file(file);
3073         return ret;
3074 }
3075
3076 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3077 {
3078         struct inode *inode = file_inode(file);
3079         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3080         struct btrfs_ioctl_vol_args *vol_args;
3081         int ret;
3082
3083         if (!capable(CAP_SYS_ADMIN))
3084                 return -EPERM;
3085
3086         ret = mnt_want_write_file(file);
3087         if (ret)
3088                 return ret;
3089
3090         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3091                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3092                 goto out_drop_write;
3093         }
3094
3095         vol_args = memdup_user(arg, sizeof(*vol_args));
3096         if (IS_ERR(vol_args)) {
3097                 ret = PTR_ERR(vol_args);
3098                 goto out;
3099         }
3100
3101         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3102         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3103
3104         if (!ret)
3105                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3106         kfree(vol_args);
3107 out:
3108         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3109 out_drop_write:
3110         mnt_drop_write_file(file);
3111
3112         return ret;
3113 }
3114
3115 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3116                                 void __user *arg)
3117 {
3118         struct btrfs_ioctl_fs_info_args *fi_args;
3119         struct btrfs_device *device;
3120         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3121         int ret = 0;
3122
3123         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3124         if (!fi_args)
3125                 return -ENOMEM;
3126
3127         rcu_read_lock();
3128         fi_args->num_devices = fs_devices->num_devices;
3129
3130         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3131                 if (device->devid > fi_args->max_id)
3132                         fi_args->max_id = device->devid;
3133         }
3134         rcu_read_unlock();
3135
3136         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3137         fi_args->nodesize = fs_info->nodesize;
3138         fi_args->sectorsize = fs_info->sectorsize;
3139         fi_args->clone_alignment = fs_info->sectorsize;
3140
3141         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3142                 ret = -EFAULT;
3143
3144         kfree(fi_args);
3145         return ret;
3146 }
3147
3148 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3149                                  void __user *arg)
3150 {
3151         struct btrfs_ioctl_dev_info_args *di_args;
3152         struct btrfs_device *dev;
3153         int ret = 0;
3154         char *s_uuid = NULL;
3155
3156         di_args = memdup_user(arg, sizeof(*di_args));
3157         if (IS_ERR(di_args))
3158                 return PTR_ERR(di_args);
3159
3160         if (!btrfs_is_empty_uuid(di_args->uuid))
3161                 s_uuid = di_args->uuid;
3162
3163         rcu_read_lock();
3164         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
3165
3166         if (!dev) {
3167                 ret = -ENODEV;
3168                 goto out;
3169         }
3170
3171         di_args->devid = dev->devid;
3172         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3173         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3174         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3175         if (dev->name) {
3176                 strncpy(di_args->path, rcu_str_deref(dev->name),
3177                                 sizeof(di_args->path) - 1);
3178                 di_args->path[sizeof(di_args->path) - 1] = 0;
3179         } else {
3180                 di_args->path[0] = '\0';
3181         }
3182
3183 out:
3184         rcu_read_unlock();
3185         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3186                 ret = -EFAULT;
3187
3188         kfree(di_args);
3189         return ret;
3190 }
3191
3192 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3193 {
3194         struct page *page;
3195
3196         page = grab_cache_page(inode->i_mapping, index);
3197         if (!page)
3198                 return ERR_PTR(-ENOMEM);
3199
3200         if (!PageUptodate(page)) {
3201                 int ret;
3202
3203                 ret = btrfs_readpage(NULL, page);
3204                 if (ret)
3205                         return ERR_PTR(ret);
3206                 lock_page(page);
3207                 if (!PageUptodate(page)) {
3208                         unlock_page(page);
3209                         put_page(page);
3210                         return ERR_PTR(-EIO);
3211                 }
3212                 if (page->mapping != inode->i_mapping) {
3213                         unlock_page(page);
3214                         put_page(page);
3215                         return ERR_PTR(-EAGAIN);
3216                 }
3217         }
3218
3219         return page;
3220 }
3221
3222 static int gather_extent_pages(struct inode *inode, struct page **pages,
3223                                int num_pages, u64 off)
3224 {
3225         int i;
3226         pgoff_t index = off >> PAGE_SHIFT;
3227
3228         for (i = 0; i < num_pages; i++) {
3229 again:
3230                 pages[i] = extent_same_get_page(inode, index + i);
3231                 if (IS_ERR(pages[i])) {
3232                         int err = PTR_ERR(pages[i]);
3233
3234                         if (err == -EAGAIN)
3235                                 goto again;
3236                         pages[i] = NULL;
3237                         return err;
3238                 }
3239         }
3240         return 0;
3241 }
3242
3243 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3244                              bool retry_range_locking)
3245 {
3246         /*
3247          * Do any pending delalloc/csum calculations on inode, one way or
3248          * another, and lock file content.
3249          * The locking order is:
3250          *
3251          *   1) pages
3252          *   2) range in the inode's io tree
3253          */
3254         while (1) {
3255                 struct btrfs_ordered_extent *ordered;
3256                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3257                 ordered = btrfs_lookup_first_ordered_extent(inode,
3258                                                             off + len - 1);
3259                 if ((!ordered ||
3260                      ordered->file_offset + ordered->len <= off ||
3261                      ordered->file_offset >= off + len) &&
3262                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3263                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3264                         if (ordered)
3265                                 btrfs_put_ordered_extent(ordered);
3266                         break;
3267                 }
3268                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3269                 if (ordered)
3270                         btrfs_put_ordered_extent(ordered);
3271                 if (!retry_range_locking)
3272                         return -EAGAIN;
3273                 btrfs_wait_ordered_range(inode, off, len);
3274         }
3275         return 0;
3276 }
3277
3278 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3279 {
3280         inode_unlock(inode1);
3281         inode_unlock(inode2);
3282 }
3283
3284 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3285 {
3286         if (inode1 < inode2)
3287                 swap(inode1, inode2);
3288
3289         inode_lock_nested(inode1, I_MUTEX_PARENT);
3290         inode_lock_nested(inode2, I_MUTEX_CHILD);
3291 }
3292
3293 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3294                                       struct inode *inode2, u64 loff2, u64 len)
3295 {
3296         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3297         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3298 }
3299
3300 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3301                                     struct inode *inode2, u64 loff2, u64 len,
3302                                     bool retry_range_locking)
3303 {
3304         int ret;
3305
3306         if (inode1 < inode2) {
3307                 swap(inode1, inode2);
3308                 swap(loff1, loff2);
3309         }
3310         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3311         if (ret)
3312                 return ret;
3313         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3314         if (ret)
3315                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3316                               loff1 + len - 1);
3317         return ret;
3318 }
3319
3320 struct cmp_pages {
3321         int             num_pages;
3322         struct page     **src_pages;
3323         struct page     **dst_pages;
3324 };
3325
3326 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3327 {
3328         int i;
3329         struct page *pg;
3330
3331         for (i = 0; i < cmp->num_pages; i++) {
3332                 pg = cmp->src_pages[i];
3333                 if (pg) {
3334                         unlock_page(pg);
3335                         put_page(pg);
3336                         cmp->src_pages[i] = NULL;
3337                 }
3338                 pg = cmp->dst_pages[i];
3339                 if (pg) {
3340                         unlock_page(pg);
3341                         put_page(pg);
3342                         cmp->dst_pages[i] = NULL;
3343                 }
3344         }
3345 }
3346
3347 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3348                                   struct inode *dst, u64 dst_loff,
3349                                   u64 len, struct cmp_pages *cmp)
3350 {
3351         int ret;
3352         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3353
3354         cmp->num_pages = num_pages;
3355
3356         ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3357         if (ret)
3358                 goto out;
3359
3360         ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3361
3362 out:
3363         if (ret)
3364                 btrfs_cmp_data_free(cmp);
3365         return ret;
3366 }
3367
3368 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3369 {
3370         int ret = 0;
3371         int i;
3372         struct page *src_page, *dst_page;
3373         unsigned int cmp_len = PAGE_SIZE;
3374         void *addr, *dst_addr;
3375
3376         i = 0;
3377         while (len) {
3378                 if (len < PAGE_SIZE)
3379                         cmp_len = len;
3380
3381                 BUG_ON(i >= cmp->num_pages);
3382
3383                 src_page = cmp->src_pages[i];
3384                 dst_page = cmp->dst_pages[i];
3385                 ASSERT(PageLocked(src_page));
3386                 ASSERT(PageLocked(dst_page));
3387
3388                 addr = kmap_atomic(src_page);
3389                 dst_addr = kmap_atomic(dst_page);
3390
3391                 flush_dcache_page(src_page);
3392                 flush_dcache_page(dst_page);
3393
3394                 if (memcmp(addr, dst_addr, cmp_len))
3395                         ret = -EBADE;
3396
3397                 kunmap_atomic(addr);
3398                 kunmap_atomic(dst_addr);
3399
3400                 if (ret)
3401                         break;
3402
3403                 len -= cmp_len;
3404                 i++;
3405         }
3406
3407         return ret;
3408 }
3409
3410 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3411                                      u64 olen)
3412 {
3413         u64 len = *plen;
3414         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3415
3416         if (off + olen > inode->i_size || off + olen < off)
3417                 return -EINVAL;
3418
3419         /* if we extend to eof, continue to block boundary */
3420         if (off + len == inode->i_size)
3421                 *plen = len = ALIGN(inode->i_size, bs) - off;
3422
3423         /* Check that we are block aligned - btrfs_clone() requires this */
3424         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3425                 return -EINVAL;
3426
3427         return 0;
3428 }
3429
3430 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3431                                    struct inode *dst, u64 dst_loff,
3432                                    struct cmp_pages *cmp)
3433 {
3434         int ret;
3435         u64 len = olen;