Merge branch 'for-4.14' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
[sfrench/cifs-2.6.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/compat.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/security.h>
39 #include <linux/xattr.h>
40 #include <linux/mm.h>
41 #include <linux/slab.h>
42 #include <linux/blkdev.h>
43 #include <linux/uuid.h>
44 #include <linux/btrfs.h>
45 #include <linux/uaccess.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55 #include "rcu-string.h"
56 #include "send.h"
57 #include "dev-replace.h"
58 #include "props.h"
59 #include "sysfs.h"
60 #include "qgroup.h"
61 #include "tree-log.h"
62 #include "compression.h"
63
64 #ifdef CONFIG_64BIT
65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
66  * structures are incorrect, as the timespec structure from userspace
67  * is 4 bytes too small. We define these alternatives here to teach
68  * the kernel about the 32-bit struct packing.
69  */
70 struct btrfs_ioctl_timespec_32 {
71         __u64 sec;
72         __u32 nsec;
73 } __attribute__ ((__packed__));
74
75 struct btrfs_ioctl_received_subvol_args_32 {
76         char    uuid[BTRFS_UUID_SIZE];  /* in */
77         __u64   stransid;               /* in */
78         __u64   rtransid;               /* out */
79         struct btrfs_ioctl_timespec_32 stime; /* in */
80         struct btrfs_ioctl_timespec_32 rtime; /* out */
81         __u64   flags;                  /* in */
82         __u64   reserved[16];           /* in */
83 } __attribute__ ((__packed__));
84
85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
86                                 struct btrfs_ioctl_received_subvol_args_32)
87 #endif
88
89
90 static int btrfs_clone(struct inode *src, struct inode *inode,
91                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
92                        int no_time_update);
93
94 /* Mask out flags that are inappropriate for the given type of inode. */
95 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
96 {
97         if (S_ISDIR(mode))
98                 return flags;
99         else if (S_ISREG(mode))
100                 return flags & ~FS_DIRSYNC_FL;
101         else
102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
103 }
104
105 /*
106  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
107  */
108 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
109 {
110         unsigned int iflags = 0;
111
112         if (flags & BTRFS_INODE_SYNC)
113                 iflags |= FS_SYNC_FL;
114         if (flags & BTRFS_INODE_IMMUTABLE)
115                 iflags |= FS_IMMUTABLE_FL;
116         if (flags & BTRFS_INODE_APPEND)
117                 iflags |= FS_APPEND_FL;
118         if (flags & BTRFS_INODE_NODUMP)
119                 iflags |= FS_NODUMP_FL;
120         if (flags & BTRFS_INODE_NOATIME)
121                 iflags |= FS_NOATIME_FL;
122         if (flags & BTRFS_INODE_DIRSYNC)
123                 iflags |= FS_DIRSYNC_FL;
124         if (flags & BTRFS_INODE_NODATACOW)
125                 iflags |= FS_NOCOW_FL;
126
127         if (flags & BTRFS_INODE_NOCOMPRESS)
128                 iflags |= FS_NOCOMP_FL;
129         else if (flags & BTRFS_INODE_COMPRESS)
130                 iflags |= FS_COMPR_FL;
131
132         return iflags;
133 }
134
135 /*
136  * Update inode->i_flags based on the btrfs internal flags.
137  */
138 void btrfs_update_iflags(struct inode *inode)
139 {
140         struct btrfs_inode *ip = BTRFS_I(inode);
141         unsigned int new_fl = 0;
142
143         if (ip->flags & BTRFS_INODE_SYNC)
144                 new_fl |= S_SYNC;
145         if (ip->flags & BTRFS_INODE_IMMUTABLE)
146                 new_fl |= S_IMMUTABLE;
147         if (ip->flags & BTRFS_INODE_APPEND)
148                 new_fl |= S_APPEND;
149         if (ip->flags & BTRFS_INODE_NOATIME)
150                 new_fl |= S_NOATIME;
151         if (ip->flags & BTRFS_INODE_DIRSYNC)
152                 new_fl |= S_DIRSYNC;
153
154         set_mask_bits(&inode->i_flags,
155                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
156                       new_fl);
157 }
158
159 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
160 {
161         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
162         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
163
164         if (copy_to_user(arg, &flags, sizeof(flags)))
165                 return -EFAULT;
166         return 0;
167 }
168
169 static int check_flags(unsigned int flags)
170 {
171         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
172                       FS_NOATIME_FL | FS_NODUMP_FL | \
173                       FS_SYNC_FL | FS_DIRSYNC_FL | \
174                       FS_NOCOMP_FL | FS_COMPR_FL |
175                       FS_NOCOW_FL))
176                 return -EOPNOTSUPP;
177
178         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
179                 return -EINVAL;
180
181         return 0;
182 }
183
184 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
185 {
186         struct inode *inode = file_inode(file);
187         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
188         struct btrfs_inode *ip = BTRFS_I(inode);
189         struct btrfs_root *root = ip->root;
190         struct btrfs_trans_handle *trans;
191         unsigned int flags, oldflags;
192         int ret;
193         u64 ip_oldflags;
194         unsigned int i_oldflags;
195         umode_t mode;
196
197         if (!inode_owner_or_capable(inode))
198                 return -EPERM;
199
200         if (btrfs_root_readonly(root))
201                 return -EROFS;
202
203         if (copy_from_user(&flags, arg, sizeof(flags)))
204                 return -EFAULT;
205
206         ret = check_flags(flags);
207         if (ret)
208                 return ret;
209
210         ret = mnt_want_write_file(file);
211         if (ret)
212                 return ret;
213
214         inode_lock(inode);
215
216         ip_oldflags = ip->flags;
217         i_oldflags = inode->i_flags;
218         mode = inode->i_mode;
219
220         flags = btrfs_mask_flags(inode->i_mode, flags);
221         oldflags = btrfs_flags_to_ioctl(ip->flags);
222         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
223                 if (!capable(CAP_LINUX_IMMUTABLE)) {
224                         ret = -EPERM;
225                         goto out_unlock;
226                 }
227         }
228
229         if (flags & FS_SYNC_FL)
230                 ip->flags |= BTRFS_INODE_SYNC;
231         else
232                 ip->flags &= ~BTRFS_INODE_SYNC;
233         if (flags & FS_IMMUTABLE_FL)
234                 ip->flags |= BTRFS_INODE_IMMUTABLE;
235         else
236                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
237         if (flags & FS_APPEND_FL)
238                 ip->flags |= BTRFS_INODE_APPEND;
239         else
240                 ip->flags &= ~BTRFS_INODE_APPEND;
241         if (flags & FS_NODUMP_FL)
242                 ip->flags |= BTRFS_INODE_NODUMP;
243         else
244                 ip->flags &= ~BTRFS_INODE_NODUMP;
245         if (flags & FS_NOATIME_FL)
246                 ip->flags |= BTRFS_INODE_NOATIME;
247         else
248                 ip->flags &= ~BTRFS_INODE_NOATIME;
249         if (flags & FS_DIRSYNC_FL)
250                 ip->flags |= BTRFS_INODE_DIRSYNC;
251         else
252                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
253         if (flags & FS_NOCOW_FL) {
254                 if (S_ISREG(mode)) {
255                         /*
256                          * It's safe to turn csums off here, no extents exist.
257                          * Otherwise we want the flag to reflect the real COW
258                          * status of the file and will not set it.
259                          */
260                         if (inode->i_size == 0)
261                                 ip->flags |= BTRFS_INODE_NODATACOW
262                                            | BTRFS_INODE_NODATASUM;
263                 } else {
264                         ip->flags |= BTRFS_INODE_NODATACOW;
265                 }
266         } else {
267                 /*
268                  * Revert back under same assumptions as above
269                  */
270                 if (S_ISREG(mode)) {
271                         if (inode->i_size == 0)
272                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
273                                              | BTRFS_INODE_NODATASUM);
274                 } else {
275                         ip->flags &= ~BTRFS_INODE_NODATACOW;
276                 }
277         }
278
279         /*
280          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
281          * flag may be changed automatically if compression code won't make
282          * things smaller.
283          */
284         if (flags & FS_NOCOMP_FL) {
285                 ip->flags &= ~BTRFS_INODE_COMPRESS;
286                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
287
288                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
289                 if (ret && ret != -ENODATA)
290                         goto out_drop;
291         } else if (flags & FS_COMPR_FL) {
292                 const char *comp;
293
294                 ip->flags |= BTRFS_INODE_COMPRESS;
295                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
296
297                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
298                         comp = "lzo";
299                 else
300                         comp = "zlib";
301                 ret = btrfs_set_prop(inode, "btrfs.compression",
302                                      comp, strlen(comp), 0);
303                 if (ret)
304                         goto out_drop;
305
306         } else {
307                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
308                 if (ret && ret != -ENODATA)
309                         goto out_drop;
310                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
311         }
312
313         trans = btrfs_start_transaction(root, 1);
314         if (IS_ERR(trans)) {
315                 ret = PTR_ERR(trans);
316                 goto out_drop;
317         }
318
319         btrfs_update_iflags(inode);
320         inode_inc_iversion(inode);
321         inode->i_ctime = current_time(inode);
322         ret = btrfs_update_inode(trans, root, inode);
323
324         btrfs_end_transaction(trans);
325  out_drop:
326         if (ret) {
327                 ip->flags = ip_oldflags;
328                 inode->i_flags = i_oldflags;
329         }
330
331  out_unlock:
332         inode_unlock(inode);
333         mnt_drop_write_file(file);
334         return ret;
335 }
336
337 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
338 {
339         struct inode *inode = file_inode(file);
340
341         return put_user(inode->i_generation, arg);
342 }
343
344 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
345 {
346         struct inode *inode = file_inode(file);
347         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
348         struct btrfs_device *device;
349         struct request_queue *q;
350         struct fstrim_range range;
351         u64 minlen = ULLONG_MAX;
352         u64 num_devices = 0;
353         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
354         int ret;
355
356         if (!capable(CAP_SYS_ADMIN))
357                 return -EPERM;
358
359         rcu_read_lock();
360         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
361                                 dev_list) {
362                 if (!device->bdev)
363                         continue;
364                 q = bdev_get_queue(device->bdev);
365                 if (blk_queue_discard(q)) {
366                         num_devices++;
367                         minlen = min_t(u64, q->limits.discard_granularity,
368                                      minlen);
369                 }
370         }
371         rcu_read_unlock();
372
373         if (!num_devices)
374                 return -EOPNOTSUPP;
375         if (copy_from_user(&range, arg, sizeof(range)))
376                 return -EFAULT;
377         if (range.start > total_bytes ||
378             range.len < fs_info->sb->s_blocksize)
379                 return -EINVAL;
380
381         range.len = min(range.len, total_bytes - range.start);
382         range.minlen = max(range.minlen, minlen);
383         ret = btrfs_trim_fs(fs_info, &range);
384         if (ret < 0)
385                 return ret;
386
387         if (copy_to_user(arg, &range, sizeof(range)))
388                 return -EFAULT;
389
390         return 0;
391 }
392
393 int btrfs_is_empty_uuid(u8 *uuid)
394 {
395         int i;
396
397         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
398                 if (uuid[i])
399                         return 0;
400         }
401         return 1;
402 }
403
404 static noinline int create_subvol(struct inode *dir,
405                                   struct dentry *dentry,
406                                   const char *name, int namelen,
407                                   u64 *async_transid,
408                                   struct btrfs_qgroup_inherit *inherit)
409 {
410         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
411         struct btrfs_trans_handle *trans;
412         struct btrfs_key key;
413         struct btrfs_root_item *root_item;
414         struct btrfs_inode_item *inode_item;
415         struct extent_buffer *leaf;
416         struct btrfs_root *root = BTRFS_I(dir)->root;
417         struct btrfs_root *new_root;
418         struct btrfs_block_rsv block_rsv;
419         struct timespec cur_time = current_time(dir);
420         struct inode *inode;
421         int ret;
422         int err;
423         u64 objectid;
424         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
425         u64 index = 0;
426         u64 qgroup_reserved;
427         uuid_le new_uuid;
428
429         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
430         if (!root_item)
431                 return -ENOMEM;
432
433         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
434         if (ret)
435                 goto fail_free;
436
437         /*
438          * Don't create subvolume whose level is not zero. Or qgroup will be
439          * screwed up since it assumes subvolume qgroup's level to be 0.
440          */
441         if (btrfs_qgroup_level(objectid)) {
442                 ret = -ENOSPC;
443                 goto fail_free;
444         }
445
446         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
447         /*
448          * The same as the snapshot creation, please see the comment
449          * of create_snapshot().
450          */
451         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
452                                                8, &qgroup_reserved, false);
453         if (ret)
454                 goto fail_free;
455
456         trans = btrfs_start_transaction(root, 0);
457         if (IS_ERR(trans)) {
458                 ret = PTR_ERR(trans);
459                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
460                 goto fail_free;
461         }
462         trans->block_rsv = &block_rsv;
463         trans->bytes_reserved = block_rsv.size;
464
465         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
466         if (ret)
467                 goto fail;
468
469         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
470         if (IS_ERR(leaf)) {
471                 ret = PTR_ERR(leaf);
472                 goto fail;
473         }
474
475         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
476         btrfs_set_header_bytenr(leaf, leaf->start);
477         btrfs_set_header_generation(leaf, trans->transid);
478         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
479         btrfs_set_header_owner(leaf, objectid);
480
481         write_extent_buffer_fsid(leaf, fs_info->fsid);
482         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
483         btrfs_mark_buffer_dirty(leaf);
484
485         inode_item = &root_item->inode;
486         btrfs_set_stack_inode_generation(inode_item, 1);
487         btrfs_set_stack_inode_size(inode_item, 3);
488         btrfs_set_stack_inode_nlink(inode_item, 1);
489         btrfs_set_stack_inode_nbytes(inode_item,
490                                      fs_info->nodesize);
491         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
492
493         btrfs_set_root_flags(root_item, 0);
494         btrfs_set_root_limit(root_item, 0);
495         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
496
497         btrfs_set_root_bytenr(root_item, leaf->start);
498         btrfs_set_root_generation(root_item, trans->transid);
499         btrfs_set_root_level(root_item, 0);
500         btrfs_set_root_refs(root_item, 1);
501         btrfs_set_root_used(root_item, leaf->len);
502         btrfs_set_root_last_snapshot(root_item, 0);
503
504         btrfs_set_root_generation_v2(root_item,
505                         btrfs_root_generation(root_item));
506         uuid_le_gen(&new_uuid);
507         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
508         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
509         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
510         root_item->ctime = root_item->otime;
511         btrfs_set_root_ctransid(root_item, trans->transid);
512         btrfs_set_root_otransid(root_item, trans->transid);
513
514         btrfs_tree_unlock(leaf);
515         free_extent_buffer(leaf);
516         leaf = NULL;
517
518         btrfs_set_root_dirid(root_item, new_dirid);
519
520         key.objectid = objectid;
521         key.offset = 0;
522         key.type = BTRFS_ROOT_ITEM_KEY;
523         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
524                                 root_item);
525         if (ret)
526                 goto fail;
527
528         key.offset = (u64)-1;
529         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
530         if (IS_ERR(new_root)) {
531                 ret = PTR_ERR(new_root);
532                 btrfs_abort_transaction(trans, ret);
533                 goto fail;
534         }
535
536         btrfs_record_root_in_trans(trans, new_root);
537
538         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
539         if (ret) {
540                 /* We potentially lose an unused inode item here */
541                 btrfs_abort_transaction(trans, ret);
542                 goto fail;
543         }
544
545         mutex_lock(&new_root->objectid_mutex);
546         new_root->highest_objectid = new_dirid;
547         mutex_unlock(&new_root->objectid_mutex);
548
549         /*
550          * insert the directory item
551          */
552         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
553         if (ret) {
554                 btrfs_abort_transaction(trans, ret);
555                 goto fail;
556         }
557
558         ret = btrfs_insert_dir_item(trans, root,
559                                     name, namelen, BTRFS_I(dir), &key,
560                                     BTRFS_FT_DIR, index);
561         if (ret) {
562                 btrfs_abort_transaction(trans, ret);
563                 goto fail;
564         }
565
566         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
567         ret = btrfs_update_inode(trans, root, dir);
568         BUG_ON(ret);
569
570         ret = btrfs_add_root_ref(trans, fs_info,
571                                  objectid, root->root_key.objectid,
572                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
573         BUG_ON(ret);
574
575         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
576                                   BTRFS_UUID_KEY_SUBVOL, objectid);
577         if (ret)
578                 btrfs_abort_transaction(trans, ret);
579
580 fail:
581         kfree(root_item);
582         trans->block_rsv = NULL;
583         trans->bytes_reserved = 0;
584         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
585
586         if (async_transid) {
587                 *async_transid = trans->transid;
588                 err = btrfs_commit_transaction_async(trans, 1);
589                 if (err)
590                         err = btrfs_commit_transaction(trans);
591         } else {
592                 err = btrfs_commit_transaction(trans);
593         }
594         if (err && !ret)
595                 ret = err;
596
597         if (!ret) {
598                 inode = btrfs_lookup_dentry(dir, dentry);
599                 if (IS_ERR(inode))
600                         return PTR_ERR(inode);
601                 d_instantiate(dentry, inode);
602         }
603         return ret;
604
605 fail_free:
606         kfree(root_item);
607         return ret;
608 }
609
610 static void btrfs_wait_for_no_snapshotting_writes(struct btrfs_root *root)
611 {
612         s64 writers;
613         DEFINE_WAIT(wait);
614
615         do {
616                 prepare_to_wait(&root->subv_writers->wait, &wait,
617                                 TASK_UNINTERRUPTIBLE);
618
619                 writers = percpu_counter_sum(&root->subv_writers->counter);
620                 if (writers)
621                         schedule();
622
623                 finish_wait(&root->subv_writers->wait, &wait);
624         } while (writers);
625 }
626
627 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
628                            struct dentry *dentry,
629                            u64 *async_transid, bool readonly,
630                            struct btrfs_qgroup_inherit *inherit)
631 {
632         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
633         struct inode *inode;
634         struct btrfs_pending_snapshot *pending_snapshot;
635         struct btrfs_trans_handle *trans;
636         int ret;
637
638         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
639                 return -EINVAL;
640
641         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
642         if (!pending_snapshot)
643                 return -ENOMEM;
644
645         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
646                         GFP_KERNEL);
647         pending_snapshot->path = btrfs_alloc_path();
648         if (!pending_snapshot->root_item || !pending_snapshot->path) {
649                 ret = -ENOMEM;
650                 goto free_pending;
651         }
652
653         atomic_inc(&root->will_be_snapshotted);
654         smp_mb__after_atomic();
655         btrfs_wait_for_no_snapshotting_writes(root);
656
657         ret = btrfs_start_delalloc_inodes(root, 0);
658         if (ret)
659                 goto dec_and_free;
660
661         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
662
663         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
664                              BTRFS_BLOCK_RSV_TEMP);
665         /*
666          * 1 - parent dir inode
667          * 2 - dir entries
668          * 1 - root item
669          * 2 - root ref/backref
670          * 1 - root of snapshot
671          * 1 - UUID item
672          */
673         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
674                                         &pending_snapshot->block_rsv, 8,
675                                         &pending_snapshot->qgroup_reserved,
676                                         false);
677         if (ret)
678                 goto dec_and_free;
679
680         pending_snapshot->dentry = dentry;
681         pending_snapshot->root = root;
682         pending_snapshot->readonly = readonly;
683         pending_snapshot->dir = dir;
684         pending_snapshot->inherit = inherit;
685
686         trans = btrfs_start_transaction(root, 0);
687         if (IS_ERR(trans)) {
688                 ret = PTR_ERR(trans);
689                 goto fail;
690         }
691
692         spin_lock(&fs_info->trans_lock);
693         list_add(&pending_snapshot->list,
694                  &trans->transaction->pending_snapshots);
695         spin_unlock(&fs_info->trans_lock);
696         if (async_transid) {
697                 *async_transid = trans->transid;
698                 ret = btrfs_commit_transaction_async(trans, 1);
699                 if (ret)
700                         ret = btrfs_commit_transaction(trans);
701         } else {
702                 ret = btrfs_commit_transaction(trans);
703         }
704         if (ret)
705                 goto fail;
706
707         ret = pending_snapshot->error;
708         if (ret)
709                 goto fail;
710
711         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
712         if (ret)
713                 goto fail;
714
715         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
716         if (IS_ERR(inode)) {
717                 ret = PTR_ERR(inode);
718                 goto fail;
719         }
720
721         d_instantiate(dentry, inode);
722         ret = 0;
723 fail:
724         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
725 dec_and_free:
726         if (atomic_dec_and_test(&root->will_be_snapshotted))
727                 wake_up_atomic_t(&root->will_be_snapshotted);
728 free_pending:
729         kfree(pending_snapshot->root_item);
730         btrfs_free_path(pending_snapshot->path);
731         kfree(pending_snapshot);
732
733         return ret;
734 }
735
736 /*  copy of may_delete in fs/namei.c()
737  *      Check whether we can remove a link victim from directory dir, check
738  *  whether the type of victim is right.
739  *  1. We can't do it if dir is read-only (done in permission())
740  *  2. We should have write and exec permissions on dir
741  *  3. We can't remove anything from append-only dir
742  *  4. We can't do anything with immutable dir (done in permission())
743  *  5. If the sticky bit on dir is set we should either
744  *      a. be owner of dir, or
745  *      b. be owner of victim, or
746  *      c. have CAP_FOWNER capability
747  *  6. If the victim is append-only or immutable we can't do anything with
748  *     links pointing to it.
749  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
750  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
751  *  9. We can't remove a root or mountpoint.
752  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
753  *     nfs_async_unlink().
754  */
755
756 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
757 {
758         int error;
759
760         if (d_really_is_negative(victim))
761                 return -ENOENT;
762
763         BUG_ON(d_inode(victim->d_parent) != dir);
764         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
765
766         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
767         if (error)
768                 return error;
769         if (IS_APPEND(dir))
770                 return -EPERM;
771         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
772             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
773                 return -EPERM;
774         if (isdir) {
775                 if (!d_is_dir(victim))
776                         return -ENOTDIR;
777                 if (IS_ROOT(victim))
778                         return -EBUSY;
779         } else if (d_is_dir(victim))
780                 return -EISDIR;
781         if (IS_DEADDIR(dir))
782                 return -ENOENT;
783         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
784                 return -EBUSY;
785         return 0;
786 }
787
788 /* copy of may_create in fs/namei.c() */
789 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
790 {
791         if (d_really_is_positive(child))
792                 return -EEXIST;
793         if (IS_DEADDIR(dir))
794                 return -ENOENT;
795         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
796 }
797
798 /*
799  * Create a new subvolume below @parent.  This is largely modeled after
800  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
801  * inside this filesystem so it's quite a bit simpler.
802  */
803 static noinline int btrfs_mksubvol(const struct path *parent,
804                                    const char *name, int namelen,
805                                    struct btrfs_root *snap_src,
806                                    u64 *async_transid, bool readonly,
807                                    struct btrfs_qgroup_inherit *inherit)
808 {
809         struct inode *dir = d_inode(parent->dentry);
810         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
811         struct dentry *dentry;
812         int error;
813
814         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
815         if (error == -EINTR)
816                 return error;
817
818         dentry = lookup_one_len(name, parent->dentry, namelen);
819         error = PTR_ERR(dentry);
820         if (IS_ERR(dentry))
821                 goto out_unlock;
822
823         error = btrfs_may_create(dir, dentry);
824         if (error)
825                 goto out_dput;
826
827         /*
828          * even if this name doesn't exist, we may get hash collisions.
829          * check for them now when we can safely fail
830          */
831         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
832                                                dir->i_ino, name,
833                                                namelen);
834         if (error)
835                 goto out_dput;
836
837         down_read(&fs_info->subvol_sem);
838
839         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
840                 goto out_up_read;
841
842         if (snap_src) {
843                 error = create_snapshot(snap_src, dir, dentry,
844                                         async_transid, readonly, inherit);
845         } else {
846                 error = create_subvol(dir, dentry, name, namelen,
847                                       async_transid, inherit);
848         }
849         if (!error)
850                 fsnotify_mkdir(dir, dentry);
851 out_up_read:
852         up_read(&fs_info->subvol_sem);
853 out_dput:
854         dput(dentry);
855 out_unlock:
856         inode_unlock(dir);
857         return error;
858 }
859
860 /*
861  * When we're defragging a range, we don't want to kick it off again
862  * if it is really just waiting for delalloc to send it down.
863  * If we find a nice big extent or delalloc range for the bytes in the
864  * file you want to defrag, we return 0 to let you know to skip this
865  * part of the file
866  */
867 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
868 {
869         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
870         struct extent_map *em = NULL;
871         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
872         u64 end;
873
874         read_lock(&em_tree->lock);
875         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
876         read_unlock(&em_tree->lock);
877
878         if (em) {
879                 end = extent_map_end(em);
880                 free_extent_map(em);
881                 if (end - offset > thresh)
882                         return 0;
883         }
884         /* if we already have a nice delalloc here, just stop */
885         thresh /= 2;
886         end = count_range_bits(io_tree, &offset, offset + thresh,
887                                thresh, EXTENT_DELALLOC, 1);
888         if (end >= thresh)
889                 return 0;
890         return 1;
891 }
892
893 /*
894  * helper function to walk through a file and find extents
895  * newer than a specific transid, and smaller than thresh.
896  *
897  * This is used by the defragging code to find new and small
898  * extents
899  */
900 static int find_new_extents(struct btrfs_root *root,
901                             struct inode *inode, u64 newer_than,
902                             u64 *off, u32 thresh)
903 {
904         struct btrfs_path *path;
905         struct btrfs_key min_key;
906         struct extent_buffer *leaf;
907         struct btrfs_file_extent_item *extent;
908         int type;
909         int ret;
910         u64 ino = btrfs_ino(BTRFS_I(inode));
911
912         path = btrfs_alloc_path();
913         if (!path)
914                 return -ENOMEM;
915
916         min_key.objectid = ino;
917         min_key.type = BTRFS_EXTENT_DATA_KEY;
918         min_key.offset = *off;
919
920         while (1) {
921                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
922                 if (ret != 0)
923                         goto none;
924 process_slot:
925                 if (min_key.objectid != ino)
926                         goto none;
927                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
928                         goto none;
929
930                 leaf = path->nodes[0];
931                 extent = btrfs_item_ptr(leaf, path->slots[0],
932                                         struct btrfs_file_extent_item);
933
934                 type = btrfs_file_extent_type(leaf, extent);
935                 if (type == BTRFS_FILE_EXTENT_REG &&
936                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
937                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
938                         *off = min_key.offset;
939                         btrfs_free_path(path);
940                         return 0;
941                 }
942
943                 path->slots[0]++;
944                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
945                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
946                         goto process_slot;
947                 }
948
949                 if (min_key.offset == (u64)-1)
950                         goto none;
951
952                 min_key.offset++;
953                 btrfs_release_path(path);
954         }
955 none:
956         btrfs_free_path(path);
957         return -ENOENT;
958 }
959
960 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
961 {
962         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
963         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
964         struct extent_map *em;
965         u64 len = PAGE_SIZE;
966
967         /*
968          * hopefully we have this extent in the tree already, try without
969          * the full extent lock
970          */
971         read_lock(&em_tree->lock);
972         em = lookup_extent_mapping(em_tree, start, len);
973         read_unlock(&em_tree->lock);
974
975         if (!em) {
976                 struct extent_state *cached = NULL;
977                 u64 end = start + len - 1;
978
979                 /* get the big lock and read metadata off disk */
980                 lock_extent_bits(io_tree, start, end, &cached);
981                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
982                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
983
984                 if (IS_ERR(em))
985                         return NULL;
986         }
987
988         return em;
989 }
990
991 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
992 {
993         struct extent_map *next;
994         bool ret = true;
995
996         /* this is the last extent */
997         if (em->start + em->len >= i_size_read(inode))
998                 return false;
999
1000         next = defrag_lookup_extent(inode, em->start + em->len);
1001         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1002                 ret = false;
1003         else if ((em->block_start + em->block_len == next->block_start) &&
1004                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1005                 ret = false;
1006
1007         free_extent_map(next);
1008         return ret;
1009 }
1010
1011 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1012                                u64 *last_len, u64 *skip, u64 *defrag_end,
1013                                int compress)
1014 {
1015         struct extent_map *em;
1016         int ret = 1;
1017         bool next_mergeable = true;
1018         bool prev_mergeable = true;
1019
1020         /*
1021          * make sure that once we start defragging an extent, we keep on
1022          * defragging it
1023          */
1024         if (start < *defrag_end)
1025                 return 1;
1026
1027         *skip = 0;
1028
1029         em = defrag_lookup_extent(inode, start);
1030         if (!em)
1031                 return 0;
1032
1033         /* this will cover holes, and inline extents */
1034         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035                 ret = 0;
1036                 goto out;
1037         }
1038
1039         if (!*defrag_end)
1040                 prev_mergeable = false;
1041
1042         next_mergeable = defrag_check_next_extent(inode, em);
1043         /*
1044          * we hit a real extent, if it is big or the next extent is not a
1045          * real extent, don't bother defragging it
1046          */
1047         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1048             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1049                 ret = 0;
1050 out:
1051         /*
1052          * last_len ends up being a counter of how many bytes we've defragged.
1053          * every time we choose not to defrag an extent, we reset *last_len
1054          * so that the next tiny extent will force a defrag.
1055          *
1056          * The end result of this is that tiny extents before a single big
1057          * extent will force at least part of that big extent to be defragged.
1058          */
1059         if (ret) {
1060                 *defrag_end = extent_map_end(em);
1061         } else {
1062                 *last_len = 0;
1063                 *skip = extent_map_end(em);
1064                 *defrag_end = 0;
1065         }
1066
1067         free_extent_map(em);
1068         return ret;
1069 }
1070
1071 /*
1072  * it doesn't do much good to defrag one or two pages
1073  * at a time.  This pulls in a nice chunk of pages
1074  * to COW and defrag.
1075  *
1076  * It also makes sure the delalloc code has enough
1077  * dirty data to avoid making new small extents as part
1078  * of the defrag
1079  *
1080  * It's a good idea to start RA on this range
1081  * before calling this.
1082  */
1083 static int cluster_pages_for_defrag(struct inode *inode,
1084                                     struct page **pages,
1085                                     unsigned long start_index,
1086                                     unsigned long num_pages)
1087 {
1088         unsigned long file_end;
1089         u64 isize = i_size_read(inode);
1090         u64 page_start;
1091         u64 page_end;
1092         u64 page_cnt;
1093         int ret;
1094         int i;
1095         int i_done;
1096         struct btrfs_ordered_extent *ordered;
1097         struct extent_state *cached_state = NULL;
1098         struct extent_io_tree *tree;
1099         struct extent_changeset *data_reserved = NULL;
1100         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1101
1102         file_end = (isize - 1) >> PAGE_SHIFT;
1103         if (!isize || start_index > file_end)
1104                 return 0;
1105
1106         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1107
1108         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1109                         start_index << PAGE_SHIFT,
1110                         page_cnt << PAGE_SHIFT);
1111         if (ret)
1112                 return ret;
1113         i_done = 0;
1114         tree = &BTRFS_I(inode)->io_tree;
1115
1116         /* step one, lock all the pages */
1117         for (i = 0; i < page_cnt; i++) {
1118                 struct page *page;
1119 again:
1120                 page = find_or_create_page(inode->i_mapping,
1121                                            start_index + i, mask);
1122                 if (!page)
1123                         break;
1124
1125                 page_start = page_offset(page);
1126                 page_end = page_start + PAGE_SIZE - 1;
1127                 while (1) {
1128                         lock_extent_bits(tree, page_start, page_end,
1129                                          &cached_state);
1130                         ordered = btrfs_lookup_ordered_extent(inode,
1131                                                               page_start);
1132                         unlock_extent_cached(tree, page_start, page_end,
1133                                              &cached_state, GFP_NOFS);
1134                         if (!ordered)
1135                                 break;
1136
1137                         unlock_page(page);
1138                         btrfs_start_ordered_extent(inode, ordered, 1);
1139                         btrfs_put_ordered_extent(ordered);
1140                         lock_page(page);
1141                         /*
1142                          * we unlocked the page above, so we need check if
1143                          * it was released or not.
1144                          */
1145                         if (page->mapping != inode->i_mapping) {
1146                                 unlock_page(page);
1147                                 put_page(page);
1148                                 goto again;
1149                         }
1150                 }
1151
1152                 if (!PageUptodate(page)) {
1153                         btrfs_readpage(NULL, page);
1154                         lock_page(page);
1155                         if (!PageUptodate(page)) {
1156                                 unlock_page(page);
1157                                 put_page(page);
1158                                 ret = -EIO;
1159                                 break;
1160                         }
1161                 }
1162
1163                 if (page->mapping != inode->i_mapping) {
1164                         unlock_page(page);
1165                         put_page(page);
1166                         goto again;
1167                 }
1168
1169                 pages[i] = page;
1170                 i_done++;
1171         }
1172         if (!i_done || ret)
1173                 goto out;
1174
1175         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1176                 goto out;
1177
1178         /*
1179          * so now we have a nice long stream of locked
1180          * and up to date pages, lets wait on them
1181          */
1182         for (i = 0; i < i_done; i++)
1183                 wait_on_page_writeback(pages[i]);
1184
1185         page_start = page_offset(pages[0]);
1186         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1187
1188         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1189                          page_start, page_end - 1, &cached_state);
1190         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1191                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1192                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1193                           &cached_state, GFP_NOFS);
1194
1195         if (i_done != page_cnt) {
1196                 spin_lock(&BTRFS_I(inode)->lock);
1197                 BTRFS_I(inode)->outstanding_extents++;
1198                 spin_unlock(&BTRFS_I(inode)->lock);
1199                 btrfs_delalloc_release_space(inode, data_reserved,
1200                                 start_index << PAGE_SHIFT,
1201                                 (page_cnt - i_done) << PAGE_SHIFT);
1202         }
1203
1204
1205         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1206                           &cached_state);
1207
1208         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1209                              page_start, page_end - 1, &cached_state,
1210                              GFP_NOFS);
1211
1212         for (i = 0; i < i_done; i++) {
1213                 clear_page_dirty_for_io(pages[i]);
1214                 ClearPageChecked(pages[i]);
1215                 set_page_extent_mapped(pages[i]);
1216                 set_page_dirty(pages[i]);
1217                 unlock_page(pages[i]);
1218                 put_page(pages[i]);
1219         }
1220         extent_changeset_free(data_reserved);
1221         return i_done;
1222 out:
1223         for (i = 0; i < i_done; i++) {
1224                 unlock_page(pages[i]);
1225                 put_page(pages[i]);
1226         }
1227         btrfs_delalloc_release_space(inode, data_reserved,
1228                         start_index << PAGE_SHIFT,
1229                         page_cnt << PAGE_SHIFT);
1230         extent_changeset_free(data_reserved);
1231         return ret;
1232
1233 }
1234
1235 int btrfs_defrag_file(struct inode *inode, struct file *file,
1236                       struct btrfs_ioctl_defrag_range_args *range,
1237                       u64 newer_than, unsigned long max_to_defrag)
1238 {
1239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1240         struct btrfs_root *root = BTRFS_I(inode)->root;
1241         struct file_ra_state *ra = NULL;
1242         unsigned long last_index;
1243         u64 isize = i_size_read(inode);
1244         u64 last_len = 0;
1245         u64 skip = 0;
1246         u64 defrag_end = 0;
1247         u64 newer_off = range->start;
1248         unsigned long i;
1249         unsigned long ra_index = 0;
1250         int ret;
1251         int defrag_count = 0;
1252         int compress_type = BTRFS_COMPRESS_ZLIB;
1253         u32 extent_thresh = range->extent_thresh;
1254         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1255         unsigned long cluster = max_cluster;
1256         u64 new_align = ~((u64)SZ_128K - 1);
1257         struct page **pages = NULL;
1258         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1259
1260         if (isize == 0)
1261                 return 0;
1262
1263         if (range->start >= isize)
1264                 return -EINVAL;
1265
1266         if (do_compress) {
1267                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1268                         return -EINVAL;
1269                 if (range->compress_type)
1270                         compress_type = range->compress_type;
1271         }
1272
1273         if (extent_thresh == 0)
1274                 extent_thresh = SZ_256K;
1275
1276         /*
1277          * If we were not given a file, allocate a readahead context. As
1278          * readahead is just an optimization, defrag will work without it so
1279          * we don't error out.
1280          */
1281         if (!file) {
1282                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1283                 if (ra)
1284                         file_ra_state_init(ra, inode->i_mapping);
1285         } else {
1286                 ra = &file->f_ra;
1287         }
1288
1289         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1290         if (!pages) {
1291                 ret = -ENOMEM;
1292                 goto out_ra;
1293         }
1294
1295         /* find the last page to defrag */
1296         if (range->start + range->len > range->start) {
1297                 last_index = min_t(u64, isize - 1,
1298                          range->start + range->len - 1) >> PAGE_SHIFT;
1299         } else {
1300                 last_index = (isize - 1) >> PAGE_SHIFT;
1301         }
1302
1303         if (newer_than) {
1304                 ret = find_new_extents(root, inode, newer_than,
1305                                        &newer_off, SZ_64K);
1306                 if (!ret) {
1307                         range->start = newer_off;
1308                         /*
1309                          * we always align our defrag to help keep
1310                          * the extents in the file evenly spaced
1311                          */
1312                         i = (newer_off & new_align) >> PAGE_SHIFT;
1313                 } else
1314                         goto out_ra;
1315         } else {
1316                 i = range->start >> PAGE_SHIFT;
1317         }
1318         if (!max_to_defrag)
1319                 max_to_defrag = last_index - i + 1;
1320
1321         /*
1322          * make writeback starts from i, so the defrag range can be
1323          * written sequentially.
1324          */
1325         if (i < inode->i_mapping->writeback_index)
1326                 inode->i_mapping->writeback_index = i;
1327
1328         while (i <= last_index && defrag_count < max_to_defrag &&
1329                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1330                 /*
1331                  * make sure we stop running if someone unmounts
1332                  * the FS
1333                  */
1334                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1335                         break;
1336
1337                 if (btrfs_defrag_cancelled(fs_info)) {
1338                         btrfs_debug(fs_info, "defrag_file cancelled");
1339                         ret = -EAGAIN;
1340                         break;
1341                 }
1342
1343                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1344                                          extent_thresh, &last_len, &skip,
1345                                          &defrag_end, do_compress)){
1346                         unsigned long next;
1347                         /*
1348                          * the should_defrag function tells us how much to skip
1349                          * bump our counter by the suggested amount
1350                          */
1351                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1352                         i = max(i + 1, next);
1353                         continue;
1354                 }
1355
1356                 if (!newer_than) {
1357                         cluster = (PAGE_ALIGN(defrag_end) >>
1358                                    PAGE_SHIFT) - i;
1359                         cluster = min(cluster, max_cluster);
1360                 } else {
1361                         cluster = max_cluster;
1362                 }
1363
1364                 if (i + cluster > ra_index) {
1365                         ra_index = max(i, ra_index);
1366                         if (ra)
1367                                 page_cache_sync_readahead(inode->i_mapping, ra,
1368                                                 file, ra_index, cluster);
1369                         ra_index += cluster;
1370                 }
1371
1372                 inode_lock(inode);
1373                 if (do_compress)
1374                         BTRFS_I(inode)->defrag_compress = compress_type;
1375                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1376                 if (ret < 0) {
1377                         inode_unlock(inode);
1378                         goto out_ra;
1379                 }
1380
1381                 defrag_count += ret;
1382                 balance_dirty_pages_ratelimited(inode->i_mapping);
1383                 inode_unlock(inode);
1384
1385                 if (newer_than) {
1386                         if (newer_off == (u64)-1)
1387                                 break;
1388
1389                         if (ret > 0)
1390                                 i += ret;
1391
1392                         newer_off = max(newer_off + 1,
1393                                         (u64)i << PAGE_SHIFT);
1394
1395                         ret = find_new_extents(root, inode, newer_than,
1396                                                &newer_off, SZ_64K);
1397                         if (!ret) {
1398                                 range->start = newer_off;
1399                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1400                         } else {
1401                                 break;
1402                         }
1403                 } else {
1404                         if (ret > 0) {
1405                                 i += ret;
1406                                 last_len += ret << PAGE_SHIFT;
1407                         } else {
1408                                 i++;
1409                                 last_len = 0;
1410                         }
1411                 }
1412         }
1413
1414         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1415                 filemap_flush(inode->i_mapping);
1416                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1417                              &BTRFS_I(inode)->runtime_flags))
1418                         filemap_flush(inode->i_mapping);
1419         }
1420
1421         if (do_compress) {
1422                 /* the filemap_flush will queue IO into the worker threads, but
1423                  * we have to make sure the IO is actually started and that
1424                  * ordered extents get created before we return
1425                  */
1426                 atomic_inc(&fs_info->async_submit_draining);
1427                 while (atomic_read(&fs_info->nr_async_submits) ||
1428                        atomic_read(&fs_info->async_delalloc_pages)) {
1429                         wait_event(fs_info->async_submit_wait,
1430                                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
1431                                     atomic_read(&fs_info->async_delalloc_pages) == 0));
1432                 }
1433                 atomic_dec(&fs_info->async_submit_draining);
1434         }
1435
1436         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1437                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1438         }
1439
1440         ret = defrag_count;
1441
1442 out_ra:
1443         if (do_compress) {
1444                 inode_lock(inode);
1445                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1446                 inode_unlock(inode);
1447         }
1448         if (!file)
1449                 kfree(ra);
1450         kfree(pages);
1451         return ret;
1452 }
1453
1454 static noinline int btrfs_ioctl_resize(struct file *file,
1455                                         void __user *arg)
1456 {
1457         struct inode *inode = file_inode(file);
1458         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1459         u64 new_size;
1460         u64 old_size;
1461         u64 devid = 1;
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         struct btrfs_ioctl_vol_args *vol_args;
1464         struct btrfs_trans_handle *trans;
1465         struct btrfs_device *device = NULL;
1466         char *sizestr;
1467         char *retptr;
1468         char *devstr = NULL;
1469         int ret = 0;
1470         int mod = 0;
1471
1472         if (!capable(CAP_SYS_ADMIN))
1473                 return -EPERM;
1474
1475         ret = mnt_want_write_file(file);
1476         if (ret)
1477                 return ret;
1478
1479         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1480                 mnt_drop_write_file(file);
1481                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1482         }
1483
1484         mutex_lock(&fs_info->volume_mutex);
1485         vol_args = memdup_user(arg, sizeof(*vol_args));
1486         if (IS_ERR(vol_args)) {
1487                 ret = PTR_ERR(vol_args);
1488                 goto out;
1489         }
1490
1491         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1492
1493         sizestr = vol_args->name;
1494         devstr = strchr(sizestr, ':');
1495         if (devstr) {
1496                 sizestr = devstr + 1;
1497                 *devstr = '\0';
1498                 devstr = vol_args->name;
1499                 ret = kstrtoull(devstr, 10, &devid);
1500                 if (ret)
1501                         goto out_free;
1502                 if (!devid) {
1503                         ret = -EINVAL;
1504                         goto out_free;
1505                 }
1506                 btrfs_info(fs_info, "resizing devid %llu", devid);
1507         }
1508
1509         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1510         if (!device) {
1511                 btrfs_info(fs_info, "resizer unable to find device %llu",
1512                            devid);
1513                 ret = -ENODEV;
1514                 goto out_free;
1515         }
1516
1517         if (!device->writeable) {
1518                 btrfs_info(fs_info,
1519                            "resizer unable to apply on readonly device %llu",
1520                        devid);
1521                 ret = -EPERM;
1522                 goto out_free;
1523         }
1524
1525         if (!strcmp(sizestr, "max"))
1526                 new_size = device->bdev->bd_inode->i_size;
1527         else {
1528                 if (sizestr[0] == '-') {
1529                         mod = -1;
1530                         sizestr++;
1531                 } else if (sizestr[0] == '+') {
1532                         mod = 1;
1533                         sizestr++;
1534                 }
1535                 new_size = memparse(sizestr, &retptr);
1536                 if (*retptr != '\0' || new_size == 0) {
1537                         ret = -EINVAL;
1538                         goto out_free;
1539                 }
1540         }
1541
1542         if (device->is_tgtdev_for_dev_replace) {
1543                 ret = -EPERM;
1544                 goto out_free;
1545         }
1546
1547         old_size = btrfs_device_get_total_bytes(device);
1548
1549         if (mod < 0) {
1550                 if (new_size > old_size) {
1551                         ret = -EINVAL;
1552                         goto out_free;
1553                 }
1554                 new_size = old_size - new_size;
1555         } else if (mod > 0) {
1556                 if (new_size > ULLONG_MAX - old_size) {
1557                         ret = -ERANGE;
1558                         goto out_free;
1559                 }
1560                 new_size = old_size + new_size;
1561         }
1562
1563         if (new_size < SZ_256M) {
1564                 ret = -EINVAL;
1565                 goto out_free;
1566         }
1567         if (new_size > device->bdev->bd_inode->i_size) {
1568                 ret = -EFBIG;
1569                 goto out_free;
1570         }
1571
1572         new_size = round_down(new_size, fs_info->sectorsize);
1573
1574         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1575                           rcu_str_deref(device->name), new_size);
1576
1577         if (new_size > old_size) {
1578                 trans = btrfs_start_transaction(root, 0);
1579                 if (IS_ERR(trans)) {
1580                         ret = PTR_ERR(trans);
1581                         goto out_free;
1582                 }
1583                 ret = btrfs_grow_device(trans, device, new_size);
1584                 btrfs_commit_transaction(trans);
1585         } else if (new_size < old_size) {
1586                 ret = btrfs_shrink_device(device, new_size);
1587         } /* equal, nothing need to do */
1588
1589 out_free:
1590         kfree(vol_args);
1591 out:
1592         mutex_unlock(&fs_info->volume_mutex);
1593         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1594         mnt_drop_write_file(file);
1595         return ret;
1596 }
1597
1598 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1599                                 const char *name, unsigned long fd, int subvol,
1600                                 u64 *transid, bool readonly,
1601                                 struct btrfs_qgroup_inherit *inherit)
1602 {
1603         int namelen;
1604         int ret = 0;
1605
1606         if (!S_ISDIR(file_inode(file)->i_mode))
1607                 return -ENOTDIR;
1608
1609         ret = mnt_want_write_file(file);
1610         if (ret)
1611                 goto out;
1612
1613         namelen = strlen(name);
1614         if (strchr(name, '/')) {
1615                 ret = -EINVAL;
1616                 goto out_drop_write;
1617         }
1618
1619         if (name[0] == '.' &&
1620            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1621                 ret = -EEXIST;
1622                 goto out_drop_write;
1623         }
1624
1625         if (subvol) {
1626                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1627                                      NULL, transid, readonly, inherit);
1628         } else {
1629                 struct fd src = fdget(fd);
1630                 struct inode *src_inode;
1631                 if (!src.file) {
1632                         ret = -EINVAL;
1633                         goto out_drop_write;
1634                 }
1635
1636                 src_inode = file_inode(src.file);
1637                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1638                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1639                                    "Snapshot src from another FS");
1640                         ret = -EXDEV;
1641                 } else if (!inode_owner_or_capable(src_inode)) {
1642                         /*
1643                          * Subvolume creation is not restricted, but snapshots
1644                          * are limited to own subvolumes only
1645                          */
1646                         ret = -EPERM;
1647                 } else {
1648                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1649                                              BTRFS_I(src_inode)->root,
1650                                              transid, readonly, inherit);
1651                 }
1652                 fdput(src);
1653         }
1654 out_drop_write:
1655         mnt_drop_write_file(file);
1656 out:
1657         return ret;
1658 }
1659
1660 static noinline int btrfs_ioctl_snap_create(struct file *file,
1661                                             void __user *arg, int subvol)
1662 {
1663         struct btrfs_ioctl_vol_args *vol_args;
1664         int ret;
1665
1666         if (!S_ISDIR(file_inode(file)->i_mode))
1667                 return -ENOTDIR;
1668
1669         vol_args = memdup_user(arg, sizeof(*vol_args));
1670         if (IS_ERR(vol_args))
1671                 return PTR_ERR(vol_args);
1672         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1673
1674         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1675                                               vol_args->fd, subvol,
1676                                               NULL, false, NULL);
1677
1678         kfree(vol_args);
1679         return ret;
1680 }
1681
1682 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1683                                                void __user *arg, int subvol)
1684 {
1685         struct btrfs_ioctl_vol_args_v2 *vol_args;
1686         int ret;
1687         u64 transid = 0;
1688         u64 *ptr = NULL;
1689         bool readonly = false;
1690         struct btrfs_qgroup_inherit *inherit = NULL;
1691
1692         if (!S_ISDIR(file_inode(file)->i_mode))
1693                 return -ENOTDIR;
1694
1695         vol_args = memdup_user(arg, sizeof(*vol_args));
1696         if (IS_ERR(vol_args))
1697                 return PTR_ERR(vol_args);
1698         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1699
1700         if (vol_args->flags &
1701             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1702               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1703                 ret = -EOPNOTSUPP;
1704                 goto free_args;
1705         }
1706
1707         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1708                 ptr = &transid;
1709         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1710                 readonly = true;
1711         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1712                 if (vol_args->size > PAGE_SIZE) {
1713                         ret = -EINVAL;
1714                         goto free_args;
1715                 }
1716                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1717                 if (IS_ERR(inherit)) {
1718                         ret = PTR_ERR(inherit);
1719                         goto free_args;
1720                 }
1721         }
1722
1723         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1724                                               vol_args->fd, subvol, ptr,
1725                                               readonly, inherit);
1726         if (ret)
1727                 goto free_inherit;
1728
1729         if (ptr && copy_to_user(arg +
1730                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1731                                         transid),
1732                                 ptr, sizeof(*ptr)))
1733                 ret = -EFAULT;
1734
1735 free_inherit:
1736         kfree(inherit);
1737 free_args:
1738         kfree(vol_args);
1739         return ret;
1740 }
1741
1742 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1743                                                 void __user *arg)
1744 {
1745         struct inode *inode = file_inode(file);
1746         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1747         struct btrfs_root *root = BTRFS_I(inode)->root;
1748         int ret = 0;
1749         u64 flags = 0;
1750
1751         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1752                 return -EINVAL;
1753
1754         down_read(&fs_info->subvol_sem);
1755         if (btrfs_root_readonly(root))
1756                 flags |= BTRFS_SUBVOL_RDONLY;
1757         up_read(&fs_info->subvol_sem);
1758
1759         if (copy_to_user(arg, &flags, sizeof(flags)))
1760                 ret = -EFAULT;
1761
1762         return ret;
1763 }
1764
1765 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1766                                               void __user *arg)
1767 {
1768         struct inode *inode = file_inode(file);
1769         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1770         struct btrfs_root *root = BTRFS_I(inode)->root;
1771         struct btrfs_trans_handle *trans;
1772         u64 root_flags;
1773         u64 flags;
1774         int ret = 0;
1775
1776         if (!inode_owner_or_capable(inode))
1777                 return -EPERM;
1778
1779         ret = mnt_want_write_file(file);
1780         if (ret)
1781                 goto out;
1782
1783         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1784                 ret = -EINVAL;
1785                 goto out_drop_write;
1786         }
1787
1788         if (copy_from_user(&flags, arg, sizeof(flags))) {
1789                 ret = -EFAULT;
1790                 goto out_drop_write;
1791         }
1792
1793         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1794                 ret = -EINVAL;
1795                 goto out_drop_write;
1796         }
1797
1798         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1799                 ret = -EOPNOTSUPP;
1800                 goto out_drop_write;
1801         }
1802
1803         down_write(&fs_info->subvol_sem);
1804
1805         /* nothing to do */
1806         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1807                 goto out_drop_sem;
1808
1809         root_flags = btrfs_root_flags(&root->root_item);
1810         if (flags & BTRFS_SUBVOL_RDONLY) {
1811                 btrfs_set_root_flags(&root->root_item,
1812                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1813         } else {
1814                 /*
1815                  * Block RO -> RW transition if this subvolume is involved in
1816                  * send
1817                  */
1818                 spin_lock(&root->root_item_lock);
1819                 if (root->send_in_progress == 0) {
1820                         btrfs_set_root_flags(&root->root_item,
1821                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1822                         spin_unlock(&root->root_item_lock);
1823                 } else {
1824                         spin_unlock(&root->root_item_lock);
1825                         btrfs_warn(fs_info,
1826                                    "Attempt to set subvolume %llu read-write during send",
1827                                    root->root_key.objectid);
1828                         ret = -EPERM;
1829                         goto out_drop_sem;
1830                 }
1831         }
1832
1833         trans = btrfs_start_transaction(root, 1);
1834         if (IS_ERR(trans)) {
1835                 ret = PTR_ERR(trans);
1836                 goto out_reset;
1837         }
1838
1839         ret = btrfs_update_root(trans, fs_info->tree_root,
1840                                 &root->root_key, &root->root_item);
1841
1842         btrfs_commit_transaction(trans);
1843 out_reset:
1844         if (ret)
1845                 btrfs_set_root_flags(&root->root_item, root_flags);
1846 out_drop_sem:
1847         up_write(&fs_info->subvol_sem);
1848 out_drop_write:
1849         mnt_drop_write_file(file);
1850 out:
1851         return ret;
1852 }
1853
1854 /*
1855  * helper to check if the subvolume references other subvolumes
1856  */
1857 static noinline int may_destroy_subvol(struct btrfs_root *root)
1858 {
1859         struct btrfs_fs_info *fs_info = root->fs_info;
1860         struct btrfs_path *path;
1861         struct btrfs_dir_item *di;
1862         struct btrfs_key key;
1863         u64 dir_id;
1864         int ret;
1865
1866         path = btrfs_alloc_path();
1867         if (!path)
1868                 return -ENOMEM;
1869
1870         /* Make sure this root isn't set as the default subvol */
1871         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1872         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1873                                    dir_id, "default", 7, 0);
1874         if (di && !IS_ERR(di)) {
1875                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1876                 if (key.objectid == root->root_key.objectid) {
1877                         ret = -EPERM;
1878                         btrfs_err(fs_info,
1879                                   "deleting default subvolume %llu is not allowed",
1880                                   key.objectid);
1881                         goto out;
1882                 }
1883                 btrfs_release_path(path);
1884         }
1885
1886         key.objectid = root->root_key.objectid;
1887         key.type = BTRFS_ROOT_REF_KEY;
1888         key.offset = (u64)-1;
1889
1890         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1891         if (ret < 0)
1892                 goto out;
1893         BUG_ON(ret == 0);
1894
1895         ret = 0;
1896         if (path->slots[0] > 0) {
1897                 path->slots[0]--;
1898                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1899                 if (key.objectid == root->root_key.objectid &&
1900                     key.type == BTRFS_ROOT_REF_KEY)
1901                         ret = -ENOTEMPTY;
1902         }
1903 out:
1904         btrfs_free_path(path);
1905         return ret;
1906 }
1907
1908 static noinline int key_in_sk(struct btrfs_key *key,
1909                               struct btrfs_ioctl_search_key *sk)
1910 {
1911         struct btrfs_key test;
1912         int ret;
1913
1914         test.objectid = sk->min_objectid;
1915         test.type = sk->min_type;
1916         test.offset = sk->min_offset;
1917
1918         ret = btrfs_comp_cpu_keys(key, &test);
1919         if (ret < 0)
1920                 return 0;
1921
1922         test.objectid = sk->max_objectid;
1923         test.type = sk->max_type;
1924         test.offset = sk->max_offset;
1925
1926         ret = btrfs_comp_cpu_keys(key, &test);
1927         if (ret > 0)
1928                 return 0;
1929         return 1;
1930 }
1931
1932 static noinline int copy_to_sk(struct btrfs_path *path,
1933                                struct btrfs_key *key,
1934                                struct btrfs_ioctl_search_key *sk,
1935                                size_t *buf_size,
1936                                char __user *ubuf,
1937                                unsigned long *sk_offset,
1938                                int *num_found)
1939 {
1940         u64 found_transid;
1941         struct extent_buffer *leaf;
1942         struct btrfs_ioctl_search_header sh;
1943         struct btrfs_key test;
1944         unsigned long item_off;
1945         unsigned long item_len;
1946         int nritems;
1947         int i;
1948         int slot;
1949         int ret = 0;
1950
1951         leaf = path->nodes[0];
1952         slot = path->slots[0];
1953         nritems = btrfs_header_nritems(leaf);
1954
1955         if (btrfs_header_generation(leaf) > sk->max_transid) {
1956                 i = nritems;
1957                 goto advance_key;
1958         }
1959         found_transid = btrfs_header_generation(leaf);
1960
1961         for (i = slot; i < nritems; i++) {
1962                 item_off = btrfs_item_ptr_offset(leaf, i);
1963                 item_len = btrfs_item_size_nr(leaf, i);
1964
1965                 btrfs_item_key_to_cpu(leaf, key, i);
1966                 if (!key_in_sk(key, sk))
1967                         continue;
1968
1969                 if (sizeof(sh) + item_len > *buf_size) {
1970                         if (*num_found) {
1971                                 ret = 1;
1972                                 goto out;
1973                         }
1974
1975                         /*
1976                          * return one empty item back for v1, which does not
1977                          * handle -EOVERFLOW
1978                          */
1979
1980                         *buf_size = sizeof(sh) + item_len;
1981                         item_len = 0;
1982                         ret = -EOVERFLOW;
1983                 }
1984
1985                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1986                         ret = 1;
1987                         goto out;
1988                 }
1989
1990                 sh.objectid = key->objectid;
1991                 sh.offset = key->offset;
1992                 sh.type = key->type;
1993                 sh.len = item_len;
1994                 sh.transid = found_transid;
1995
1996                 /* copy search result header */
1997                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1998                         ret = -EFAULT;
1999                         goto out;
2000                 }
2001
2002                 *sk_offset += sizeof(sh);
2003
2004                 if (item_len) {
2005                         char __user *up = ubuf + *sk_offset;
2006                         /* copy the item */
2007                         if (read_extent_buffer_to_user(leaf, up,
2008                                                        item_off, item_len)) {
2009                                 ret = -EFAULT;
2010                                 goto out;
2011                         }
2012
2013                         *sk_offset += item_len;
2014                 }
2015                 (*num_found)++;
2016
2017                 if (ret) /* -EOVERFLOW from above */
2018                         goto out;
2019
2020                 if (*num_found >= sk->nr_items) {
2021                         ret = 1;
2022                         goto out;
2023                 }
2024         }
2025 advance_key:
2026         ret = 0;
2027         test.objectid = sk->max_objectid;
2028         test.type = sk->max_type;
2029         test.offset = sk->max_offset;
2030         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2031                 ret = 1;
2032         else if (key->offset < (u64)-1)
2033                 key->offset++;
2034         else if (key->type < (u8)-1) {
2035                 key->offset = 0;
2036                 key->type++;
2037         } else if (key->objectid < (u64)-1) {
2038                 key->offset = 0;
2039                 key->type = 0;
2040                 key->objectid++;
2041         } else
2042                 ret = 1;
2043 out:
2044         /*
2045          *  0: all items from this leaf copied, continue with next
2046          *  1: * more items can be copied, but unused buffer is too small
2047          *     * all items were found
2048          *     Either way, it will stops the loop which iterates to the next
2049          *     leaf
2050          *  -EOVERFLOW: item was to large for buffer
2051          *  -EFAULT: could not copy extent buffer back to userspace
2052          */
2053         return ret;
2054 }
2055
2056 static noinline int search_ioctl(struct inode *inode,
2057                                  struct btrfs_ioctl_search_key *sk,
2058                                  size_t *buf_size,
2059                                  char __user *ubuf)
2060 {
2061         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2062         struct btrfs_root *root;
2063         struct btrfs_key key;
2064         struct btrfs_path *path;
2065         int ret;
2066         int num_found = 0;
2067         unsigned long sk_offset = 0;
2068
2069         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2070                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2071                 return -EOVERFLOW;
2072         }
2073
2074         path = btrfs_alloc_path();
2075         if (!path)
2076                 return -ENOMEM;
2077
2078         if (sk->tree_id == 0) {
2079                 /* search the root of the inode that was passed */
2080                 root = BTRFS_I(inode)->root;
2081         } else {
2082                 key.objectid = sk->tree_id;
2083                 key.type = BTRFS_ROOT_ITEM_KEY;
2084                 key.offset = (u64)-1;
2085                 root = btrfs_read_fs_root_no_name(info, &key);
2086                 if (IS_ERR(root)) {
2087                         btrfs_free_path(path);
2088                         return -ENOENT;
2089                 }
2090         }
2091
2092         key.objectid = sk->min_objectid;
2093         key.type = sk->min_type;
2094         key.offset = sk->min_offset;
2095
2096         while (1) {
2097                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2098                 if (ret != 0) {
2099                         if (ret > 0)
2100                                 ret = 0;
2101                         goto err;
2102                 }
2103                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2104                                  &sk_offset, &num_found);
2105                 btrfs_release_path(path);
2106                 if (ret)
2107                         break;
2108
2109         }
2110         if (ret > 0)
2111                 ret = 0;
2112 err:
2113         sk->nr_items = num_found;
2114         btrfs_free_path(path);
2115         return ret;
2116 }
2117
2118 static noinline int btrfs_ioctl_tree_search(struct file *file,
2119                                            void __user *argp)
2120 {
2121         struct btrfs_ioctl_search_args __user *uargs;
2122         struct btrfs_ioctl_search_key sk;
2123         struct inode *inode;
2124         int ret;
2125         size_t buf_size;
2126
2127         if (!capable(CAP_SYS_ADMIN))
2128                 return -EPERM;
2129
2130         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2131
2132         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2133                 return -EFAULT;
2134
2135         buf_size = sizeof(uargs->buf);
2136
2137         inode = file_inode(file);
2138         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2139
2140         /*
2141          * In the origin implementation an overflow is handled by returning a
2142          * search header with a len of zero, so reset ret.
2143          */
2144         if (ret == -EOVERFLOW)
2145                 ret = 0;
2146
2147         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2148                 ret = -EFAULT;
2149         return ret;
2150 }
2151
2152 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2153                                                void __user *argp)
2154 {
2155         struct btrfs_ioctl_search_args_v2 __user *uarg;
2156         struct btrfs_ioctl_search_args_v2 args;
2157         struct inode *inode;
2158         int ret;
2159         size_t buf_size;
2160         const size_t buf_limit = SZ_16M;
2161
2162         if (!capable(CAP_SYS_ADMIN))
2163                 return -EPERM;
2164
2165         /* copy search header and buffer size */
2166         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2167         if (copy_from_user(&args, uarg, sizeof(args)))
2168                 return -EFAULT;
2169
2170         buf_size = args.buf_size;
2171
2172         /* limit result size to 16MB */
2173         if (buf_size > buf_limit)
2174                 buf_size = buf_limit;
2175
2176         inode = file_inode(file);
2177         ret = search_ioctl(inode, &args.key, &buf_size,
2178                            (char *)(&uarg->buf[0]));
2179         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2180                 ret = -EFAULT;
2181         else if (ret == -EOVERFLOW &&
2182                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2183                 ret = -EFAULT;
2184
2185         return ret;
2186 }
2187
2188 /*
2189  * Search INODE_REFs to identify path name of 'dirid' directory
2190  * in a 'tree_id' tree. and sets path name to 'name'.
2191  */
2192 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2193                                 u64 tree_id, u64 dirid, char *name)
2194 {
2195         struct btrfs_root *root;
2196         struct btrfs_key key;
2197         char *ptr;
2198         int ret = -1;
2199         int slot;
2200         int len;
2201         int total_len = 0;
2202         struct btrfs_inode_ref *iref;
2203         struct extent_buffer *l;
2204         struct btrfs_path *path;
2205
2206         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2207                 name[0]='\0';
2208                 return 0;
2209         }
2210
2211         path = btrfs_alloc_path();
2212         if (!path)
2213                 return -ENOMEM;
2214
2215         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2216
2217         key.objectid = tree_id;
2218         key.type = BTRFS_ROOT_ITEM_KEY;
2219         key.offset = (u64)-1;
2220         root = btrfs_read_fs_root_no_name(info, &key);
2221         if (IS_ERR(root)) {
2222                 btrfs_err(info, "could not find root %llu", tree_id);
2223                 ret = -ENOENT;
2224                 goto out;
2225         }
2226
2227         key.objectid = dirid;
2228         key.type = BTRFS_INODE_REF_KEY;
2229         key.offset = (u64)-1;
2230
2231         while (1) {
2232                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2233                 if (ret < 0)
2234                         goto out;
2235                 else if (ret > 0) {
2236                         ret = btrfs_previous_item(root, path, dirid,
2237                                                   BTRFS_INODE_REF_KEY);
2238                         if (ret < 0)
2239                                 goto out;
2240                         else if (ret > 0) {
2241                                 ret = -ENOENT;
2242                                 goto out;
2243                         }
2244                 }
2245
2246                 l = path->nodes[0];
2247                 slot = path->slots[0];
2248                 btrfs_item_key_to_cpu(l, &key, slot);
2249
2250                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2251                 len = btrfs_inode_ref_name_len(l, iref);
2252                 ptr -= len + 1;
2253                 total_len += len + 1;
2254                 if (ptr < name) {
2255                         ret = -ENAMETOOLONG;
2256                         goto out;
2257                 }
2258
2259                 *(ptr + len) = '/';
2260                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2261
2262                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2263                         break;
2264
2265                 btrfs_release_path(path);
2266                 key.objectid = key.offset;
2267                 key.offset = (u64)-1;
2268                 dirid = key.objectid;
2269         }
2270         memmove(name, ptr, total_len);
2271         name[total_len] = '\0';
2272         ret = 0;
2273 out:
2274         btrfs_free_path(path);
2275         return ret;
2276 }
2277
2278 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2279                                            void __user *argp)
2280 {
2281          struct btrfs_ioctl_ino_lookup_args *args;
2282          struct inode *inode;
2283         int ret = 0;
2284
2285         args = memdup_user(argp, sizeof(*args));
2286         if (IS_ERR(args))
2287                 return PTR_ERR(args);
2288
2289         inode = file_inode(file);
2290
2291         /*
2292          * Unprivileged query to obtain the containing subvolume root id. The
2293          * path is reset so it's consistent with btrfs_search_path_in_tree.
2294          */
2295         if (args->treeid == 0)
2296                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2297
2298         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2299                 args->name[0] = 0;
2300                 goto out;
2301         }
2302
2303         if (!capable(CAP_SYS_ADMIN)) {
2304                 ret = -EPERM;
2305                 goto out;
2306         }
2307
2308         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2309                                         args->treeid, args->objectid,
2310                                         args->name);
2311
2312 out:
2313         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2314                 ret = -EFAULT;
2315
2316         kfree(args);
2317         return ret;
2318 }
2319
2320 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2321                                              void __user *arg)
2322 {
2323         struct dentry *parent = file->f_path.dentry;
2324         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2325         struct dentry *dentry;
2326         struct inode *dir = d_inode(parent);
2327         struct inode *inode;
2328         struct btrfs_root *root = BTRFS_I(dir)->root;
2329         struct btrfs_root *dest = NULL;
2330         struct btrfs_ioctl_vol_args *vol_args;
2331         struct btrfs_trans_handle *trans;
2332         struct btrfs_block_rsv block_rsv;
2333         u64 root_flags;
2334         u64 qgroup_reserved;
2335         int namelen;
2336         int ret;
2337         int err = 0;
2338
2339         if (!S_ISDIR(dir->i_mode))
2340                 return -ENOTDIR;
2341
2342         vol_args = memdup_user(arg, sizeof(*vol_args));
2343         if (IS_ERR(vol_args))
2344                 return PTR_ERR(vol_args);
2345
2346         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2347         namelen = strlen(vol_args->name);
2348         if (strchr(vol_args->name, '/') ||
2349             strncmp(vol_args->name, "..", namelen) == 0) {
2350                 err = -EINVAL;
2351                 goto out;
2352         }
2353
2354         err = mnt_want_write_file(file);
2355         if (err)
2356                 goto out;
2357
2358
2359         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2360         if (err == -EINTR)
2361                 goto out_drop_write;
2362         dentry = lookup_one_len(vol_args->name, parent, namelen);
2363         if (IS_ERR(dentry)) {
2364                 err = PTR_ERR(dentry);
2365                 goto out_unlock_dir;
2366         }
2367
2368         if (d_really_is_negative(dentry)) {
2369                 err = -ENOENT;
2370                 goto out_dput;
2371         }
2372
2373         inode = d_inode(dentry);
2374         dest = BTRFS_I(inode)->root;
2375         if (!capable(CAP_SYS_ADMIN)) {
2376                 /*
2377                  * Regular user.  Only allow this with a special mount
2378                  * option, when the user has write+exec access to the
2379                  * subvol root, and when rmdir(2) would have been
2380                  * allowed.
2381                  *
2382                  * Note that this is _not_ check that the subvol is
2383                  * empty or doesn't contain data that we wouldn't
2384                  * otherwise be able to delete.
2385                  *
2386                  * Users who want to delete empty subvols should try
2387                  * rmdir(2).
2388                  */
2389                 err = -EPERM;
2390                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2391                         goto out_dput;
2392
2393                 /*
2394                  * Do not allow deletion if the parent dir is the same
2395                  * as the dir to be deleted.  That means the ioctl
2396                  * must be called on the dentry referencing the root
2397                  * of the subvol, not a random directory contained
2398                  * within it.
2399                  */
2400                 err = -EINVAL;
2401                 if (root == dest)
2402                         goto out_dput;
2403
2404                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2405                 if (err)
2406                         goto out_dput;
2407         }
2408
2409         /* check if subvolume may be deleted by a user */
2410         err = btrfs_may_delete(dir, dentry, 1);
2411         if (err)
2412                 goto out_dput;
2413
2414         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2415                 err = -EINVAL;
2416                 goto out_dput;
2417         }
2418
2419         inode_lock(inode);
2420
2421         /*
2422          * Don't allow to delete a subvolume with send in progress. This is
2423          * inside the i_mutex so the error handling that has to drop the bit
2424          * again is not run concurrently.
2425          */
2426         spin_lock(&dest->root_item_lock);
2427         root_flags = btrfs_root_flags(&dest->root_item);
2428         if (dest->send_in_progress == 0) {
2429                 btrfs_set_root_flags(&dest->root_item,
2430                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2431                 spin_unlock(&dest->root_item_lock);
2432         } else {
2433                 spin_unlock(&dest->root_item_lock);
2434                 btrfs_warn(fs_info,
2435                            "Attempt to delete subvolume %llu during send",
2436                            dest->root_key.objectid);
2437                 err = -EPERM;
2438                 goto out_unlock_inode;
2439         }
2440
2441         down_write(&fs_info->subvol_sem);
2442
2443         err = may_destroy_subvol(dest);
2444         if (err)
2445                 goto out_up_write;
2446
2447         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2448         /*
2449          * One for dir inode, two for dir entries, two for root
2450          * ref/backref.
2451          */
2452         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2453                                                5, &qgroup_reserved, true);
2454         if (err)
2455                 goto out_up_write;
2456
2457         trans = btrfs_start_transaction(root, 0);
2458         if (IS_ERR(trans)) {
2459                 err = PTR_ERR(trans);
2460                 goto out_release;
2461         }
2462         trans->block_rsv = &block_rsv;
2463         trans->bytes_reserved = block_rsv.size;
2464
2465         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2466
2467         ret = btrfs_unlink_subvol(trans, root, dir,
2468                                 dest->root_key.objectid,
2469                                 dentry->d_name.name,
2470                                 dentry->d_name.len);
2471         if (ret) {
2472                 err = ret;
2473                 btrfs_abort_transaction(trans, ret);
2474                 goto out_end_trans;
2475         }
2476
2477         btrfs_record_root_in_trans(trans, dest);
2478
2479         memset(&dest->root_item.drop_progress, 0,
2480                 sizeof(dest->root_item.drop_progress));
2481         dest->root_item.drop_level = 0;
2482         btrfs_set_root_refs(&dest->root_item, 0);
2483
2484         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2485                 ret = btrfs_insert_orphan_item(trans,
2486                                         fs_info->tree_root,
2487                                         dest->root_key.objectid);
2488                 if (ret) {
2489                         btrfs_abort_transaction(trans, ret);
2490                         err = ret;
2491                         goto out_end_trans;
2492                 }
2493         }
2494
2495         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2496                                   BTRFS_UUID_KEY_SUBVOL,
2497                                   dest->root_key.objectid);
2498         if (ret && ret != -ENOENT) {
2499                 btrfs_abort_transaction(trans, ret);
2500                 err = ret;
2501                 goto out_end_trans;
2502         }
2503         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2504                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2505                                           dest->root_item.received_uuid,
2506                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2507                                           dest->root_key.objectid);
2508                 if (ret && ret != -ENOENT) {
2509                         btrfs_abort_transaction(trans, ret);
2510                         err = ret;
2511                         goto out_end_trans;
2512                 }
2513         }
2514
2515 out_end_trans:
2516         trans->block_rsv = NULL;
2517         trans->bytes_reserved = 0;
2518         ret = btrfs_end_transaction(trans);
2519         if (ret && !err)
2520                 err = ret;
2521         inode->i_flags |= S_DEAD;
2522 out_release:
2523         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2524 out_up_write:
2525         up_write(&fs_info->subvol_sem);
2526         if (err) {
2527                 spin_lock(&dest->root_item_lock);
2528                 root_flags = btrfs_root_flags(&dest->root_item);
2529                 btrfs_set_root_flags(&dest->root_item,
2530                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2531                 spin_unlock(&dest->root_item_lock);
2532         }
2533 out_unlock_inode:
2534         inode_unlock(inode);
2535         if (!err) {
2536                 d_invalidate(dentry);
2537                 btrfs_invalidate_inodes(dest);
2538                 d_delete(dentry);
2539                 ASSERT(dest->send_in_progress == 0);
2540
2541                 /* the last ref */
2542                 if (dest->ino_cache_inode) {
2543                         iput(dest->ino_cache_inode);
2544                         dest->ino_cache_inode = NULL;
2545                 }
2546         }
2547 out_dput:
2548         dput(dentry);
2549 out_unlock_dir:
2550         inode_unlock(dir);
2551 out_drop_write:
2552         mnt_drop_write_file(file);
2553 out:
2554         kfree(vol_args);
2555         return err;
2556 }
2557
2558 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2559 {
2560         struct inode *inode = file_inode(file);
2561         struct btrfs_root *root = BTRFS_I(inode)->root;
2562         struct btrfs_ioctl_defrag_range_args *range;
2563         int ret;
2564
2565         ret = mnt_want_write_file(file);
2566         if (ret)
2567                 return ret;
2568
2569         if (btrfs_root_readonly(root)) {
2570                 ret = -EROFS;
2571                 goto out;
2572         }
2573
2574         switch (inode->i_mode & S_IFMT) {
2575         case S_IFDIR:
2576                 if (!capable(CAP_SYS_ADMIN)) {
2577                         ret = -EPERM;
2578                         goto out;
2579                 }
2580                 ret = btrfs_defrag_root(root);
2581                 break;
2582         case S_IFREG:
2583                 if (!(file->f_mode & FMODE_WRITE)) {
2584                         ret = -EINVAL;
2585                         goto out;
2586                 }
2587
2588                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2589                 if (!range) {
2590                         ret = -ENOMEM;
2591                         goto out;
2592                 }
2593
2594                 if (argp) {
2595                         if (copy_from_user(range, argp,
2596                                            sizeof(*range))) {
2597                                 ret = -EFAULT;
2598                                 kfree(range);
2599                                 goto out;
2600                         }
2601                         /* compression requires us to start the IO */
2602                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2603                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2604                                 range->extent_thresh = (u32)-1;
2605                         }
2606                 } else {
2607                         /* the rest are all set to zero by kzalloc */
2608                         range->len = (u64)-1;
2609                 }
2610                 ret = btrfs_defrag_file(file_inode(file), file,
2611                                         range, 0, 0);
2612                 if (ret > 0)
2613                         ret = 0;
2614                 kfree(range);
2615                 break;
2616         default:
2617                 ret = -EINVAL;
2618         }
2619 out:
2620         mnt_drop_write_file(file);
2621         return ret;
2622 }
2623
2624 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2625 {
2626         struct btrfs_ioctl_vol_args *vol_args;
2627         int ret;
2628
2629         if (!capable(CAP_SYS_ADMIN))
2630                 return -EPERM;
2631
2632         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2633                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2634
2635         mutex_lock(&fs_info->volume_mutex);
2636         vol_args = memdup_user(arg, sizeof(*vol_args));
2637         if (IS_ERR(vol_args)) {
2638                 ret = PTR_ERR(vol_args);
2639                 goto out;
2640         }
2641
2642         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2643         ret = btrfs_init_new_device(fs_info, vol_args->name);
2644
2645         if (!ret)
2646                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2647
2648         kfree(vol_args);
2649 out:
2650         mutex_unlock(&fs_info->volume_mutex);
2651         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2652         return ret;
2653 }
2654
2655 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2656 {
2657         struct inode *inode = file_inode(file);
2658         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2659         struct btrfs_ioctl_vol_args_v2 *vol_args;
2660         int ret;
2661
2662         if (!capable(CAP_SYS_ADMIN))
2663                 return -EPERM;
2664
2665         ret = mnt_want_write_file(file);
2666         if (ret)
2667                 return ret;
2668
2669         vol_args = memdup_user(arg, sizeof(*vol_args));
2670         if (IS_ERR(vol_args)) {
2671                 ret = PTR_ERR(vol_args);
2672                 goto err_drop;
2673         }
2674
2675         /* Check for compatibility reject unknown flags */
2676         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2677                 return -EOPNOTSUPP;
2678
2679         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2680                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2681                 goto out;
2682         }
2683
2684         mutex_lock(&fs_info->volume_mutex);
2685         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2686                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2687         } else {
2688                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2689                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2690         }
2691         mutex_unlock(&fs_info->volume_mutex);
2692         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2693
2694         if (!ret) {
2695                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2696                         btrfs_info(fs_info, "device deleted: id %llu",
2697                                         vol_args->devid);
2698                 else
2699                         btrfs_info(fs_info, "device deleted: %s",
2700                                         vol_args->name);
2701         }
2702 out:
2703         kfree(vol_args);
2704 err_drop:
2705         mnt_drop_write_file(file);
2706         return ret;
2707 }
2708
2709 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2710 {
2711         struct inode *inode = file_inode(file);
2712         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2713         struct btrfs_ioctl_vol_args *vol_args;
2714         int ret;
2715
2716         if (!capable(CAP_SYS_ADMIN))
2717                 return -EPERM;
2718
2719         ret = mnt_want_write_file(file);
2720         if (ret)
2721                 return ret;
2722
2723         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2724                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2725                 goto out_drop_write;
2726         }
2727
2728         vol_args = memdup_user(arg, sizeof(*vol_args));
2729         if (IS_ERR(vol_args)) {
2730                 ret = PTR_ERR(vol_args);
2731                 goto out;
2732         }
2733
2734         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2735         mutex_lock(&fs_info->volume_mutex);
2736         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2737         mutex_unlock(&fs_info->volume_mutex);
2738
2739         if (!ret)
2740                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2741         kfree(vol_args);
2742 out:
2743         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2744 out_drop_write:
2745         mnt_drop_write_file(file);
2746
2747         return ret;
2748 }
2749
2750 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2751                                 void __user *arg)
2752 {
2753         struct btrfs_ioctl_fs_info_args *fi_args;
2754         struct btrfs_device *device;
2755         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2756         int ret = 0;
2757
2758         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2759         if (!fi_args)
2760                 return -ENOMEM;
2761
2762         mutex_lock(&fs_devices->device_list_mutex);
2763         fi_args->num_devices = fs_devices->num_devices;
2764         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2765
2766         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2767                 if (device->devid > fi_args->max_id)
2768                         fi_args->max_id = device->devid;
2769         }
2770         mutex_unlock(&fs_devices->device_list_mutex);
2771
2772         fi_args->nodesize = fs_info->super_copy->nodesize;
2773         fi_args->sectorsize = fs_info->super_copy->sectorsize;
2774         fi_args->clone_alignment = fs_info->super_copy->sectorsize;
2775
2776         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2777                 ret = -EFAULT;
2778
2779         kfree(fi_args);
2780         return ret;
2781 }
2782
2783 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2784                                  void __user *arg)
2785 {
2786         struct btrfs_ioctl_dev_info_args *di_args;
2787         struct btrfs_device *dev;
2788         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2789         int ret = 0;
2790         char *s_uuid = NULL;
2791
2792         di_args = memdup_user(arg, sizeof(*di_args));
2793         if (IS_ERR(di_args))
2794                 return PTR_ERR(di_args);
2795
2796         if (!btrfs_is_empty_uuid(di_args->uuid))
2797                 s_uuid = di_args->uuid;
2798
2799         mutex_lock(&fs_devices->device_list_mutex);
2800         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2801
2802         if (!dev) {
2803                 ret = -ENODEV;
2804                 goto out;
2805         }
2806
2807         di_args->devid = dev->devid;
2808         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2809         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2810         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2811         if (dev->name) {
2812                 struct rcu_string *name;
2813
2814                 rcu_read_lock();
2815                 name = rcu_dereference(dev->name);
2816                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2817                 rcu_read_unlock();
2818                 di_args->path[sizeof(di_args->path) - 1] = 0;
2819         } else {
2820                 di_args->path[0] = '\0';
2821         }
2822
2823 out:
2824         mutex_unlock(&fs_devices->device_list_mutex);
2825         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2826                 ret = -EFAULT;
2827
2828         kfree(di_args);
2829         return ret;
2830 }
2831
2832 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2833 {
2834         struct page *page;
2835
2836         page = grab_cache_page(inode->i_mapping, index);
2837         if (!page)
2838                 return ERR_PTR(-ENOMEM);
2839
2840         if (!PageUptodate(page)) {
2841                 int ret;
2842
2843                 ret = btrfs_readpage(NULL, page);
2844                 if (ret)
2845                         return ERR_PTR(ret);
2846                 lock_page(page);
2847                 if (!PageUptodate(page)) {
2848                         unlock_page(page);
2849                         put_page(page);
2850                         return ERR_PTR(-EIO);
2851                 }
2852                 if (page->mapping != inode->i_mapping) {
2853                         unlock_page(page);
2854                         put_page(page);
2855                         return ERR_PTR(-EAGAIN);
2856                 }
2857         }
2858
2859         return page;
2860 }
2861
2862 static int gather_extent_pages(struct inode *inode, struct page **pages,
2863                                int num_pages, u64 off)
2864 {
2865         int i;
2866         pgoff_t index = off >> PAGE_SHIFT;
2867
2868         for (i = 0; i < num_pages; i++) {
2869 again:
2870                 pages[i] = extent_same_get_page(inode, index + i);
2871                 if (IS_ERR(pages[i])) {
2872                         int err = PTR_ERR(pages[i]);
2873
2874                         if (err == -EAGAIN)
2875                                 goto again;
2876                         pages[i] = NULL;
2877                         return err;
2878                 }
2879         }
2880         return 0;
2881 }
2882
2883 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2884                              bool retry_range_locking)
2885 {
2886         /*
2887          * Do any pending delalloc/csum calculations on inode, one way or
2888          * another, and lock file content.
2889          * The locking order is:
2890          *
2891          *   1) pages
2892          *   2) range in the inode's io tree
2893          */
2894         while (1) {
2895                 struct btrfs_ordered_extent *ordered;
2896                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2897                 ordered = btrfs_lookup_first_ordered_extent(inode,
2898                                                             off + len - 1);
2899                 if ((!ordered ||
2900                      ordered->file_offset + ordered->len <= off ||
2901                      ordered->file_offset >= off + len) &&
2902                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2903                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2904                         if (ordered)
2905                                 btrfs_put_ordered_extent(ordered);
2906                         break;
2907                 }
2908                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2909                 if (ordered)
2910                         btrfs_put_ordered_extent(ordered);
2911                 if (!retry_range_locking)
2912                         return -EAGAIN;
2913                 btrfs_wait_ordered_range(inode, off, len);
2914         }
2915         return 0;
2916 }
2917
2918 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2919 {
2920         inode_unlock(inode1);
2921         inode_unlock(inode2);
2922 }
2923
2924 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2925 {
2926         if (inode1 < inode2)
2927                 swap(inode1, inode2);
2928
2929         inode_lock_nested(inode1, I_MUTEX_PARENT);
2930         inode_lock_nested(inode2, I_MUTEX_CHILD);
2931 }
2932
2933 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2934                                       struct inode *inode2, u64 loff2, u64 len)
2935 {
2936         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2937         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2938 }
2939
2940 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2941                                     struct inode *inode2, u64 loff2, u64 len,
2942                                     bool retry_range_locking)
2943 {
2944         int ret;
2945
2946         if (inode1 < inode2) {
2947                 swap(inode1, inode2);
2948                 swap(loff1, loff2);
2949         }
2950         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2951         if (ret)
2952                 return ret;
2953         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2954         if (ret)
2955                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2956                               loff1 + len - 1);
2957         return ret;
2958 }
2959
2960 struct cmp_pages {
2961         int             num_pages;
2962         struct page     **src_pages;
2963         struct page     **dst_pages;
2964 };
2965
2966 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2967 {
2968         int i;
2969         struct page *pg;
2970
2971         for (i = 0; i < cmp->num_pages; i++) {
2972                 pg = cmp->src_pages[i];
2973                 if (pg) {
2974                         unlock_page(pg);
2975                         put_page(pg);
2976                 }
2977                 pg = cmp->dst_pages[i];
2978                 if (pg) {
2979                         unlock_page(pg);
2980                         put_page(pg);
2981                 }
2982         }
2983         kfree(cmp->src_pages);
2984         kfree(cmp->dst_pages);
2985 }
2986
2987 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2988                                   struct inode *dst, u64 dst_loff,
2989                                   u64 len, struct cmp_pages *cmp)
2990 {
2991         int ret;
2992         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2993         struct page **src_pgarr, **dst_pgarr;
2994
2995         /*
2996          * We must gather up all the pages before we initiate our
2997          * extent locking. We use an array for the page pointers. Size
2998          * of the array is bounded by len, which is in turn bounded by
2999          * BTRFS_MAX_DEDUPE_LEN.
3000          */
3001         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3002         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3003         if (!src_pgarr || !dst_pgarr) {
3004                 kfree(src_pgarr);
3005                 kfree(dst_pgarr);
3006                 return -ENOMEM;
3007         }
3008         cmp->num_pages = num_pages;
3009         cmp->src_pages = src_pgarr;
3010         cmp->dst_pages = dst_pgarr;
3011
3012         /*
3013          * If deduping ranges in the same inode, locking rules make it mandatory
3014          * to always lock pages in ascending order to avoid deadlocks with
3015          * concurrent tasks (such as starting writeback/delalloc).
3016          */
3017         if (src == dst && dst_loff < loff) {
3018                 swap(src_pgarr, dst_pgarr);
3019                 swap(loff, dst_loff);
3020         }
3021
3022         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3023         if (ret)
3024                 goto out;
3025
3026         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3027
3028 out:
3029         if (ret)
3030                 btrfs_cmp_data_free(cmp);
3031         return 0;
3032 }
3033
3034 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3035 {
3036         int ret = 0;
3037         int i;
3038         struct page *src_page, *dst_page;
3039         unsigned int cmp_len = PAGE_SIZE;
3040         void *addr, *dst_addr;
3041
3042         i = 0;
3043         while (len) {
3044                 if (len < PAGE_SIZE)
3045                         cmp_len = len;
3046
3047                 BUG_ON(i >= cmp->num_pages);
3048
3049                 src_page = cmp->src_pages[i];
3050                 dst_page = cmp->dst_pages[i];
3051                 ASSERT(PageLocked(src_page));
3052                 ASSERT(PageLocked(dst_page));
3053
3054                 addr = kmap_atomic(src_page);
3055                 dst_addr = kmap_atomic(dst_page);
3056
3057                 flush_dcache_page(src_page);
3058                 flush_dcache_page(dst_page);
3059
3060                 if (memcmp(addr, dst_addr, cmp_len))
3061                         ret = -EBADE;
3062
3063                 kunmap_atomic(addr);
3064                 kunmap_atomic(dst_addr);
3065
3066                 if (ret)
3067                         break;
3068
3069                 len -= cmp_len;
3070                 i++;
3071         }
3072
3073         return ret;
3074 }
3075
3076 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3077                                      u64 olen)
3078 {
3079         u64 len = *plen;
3080         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3081
3082         if (off + olen > inode->i_size || off + olen < off)
3083                 return -EINVAL;
3084
3085         /* if we extend to eof, continue to block boundary */
3086         if (off + len == inode->i_size)
3087                 *plen = len = ALIGN(inode->i_size, bs) - off;
3088
3089         /* Check that we are block aligned - btrfs_clone() requires this */
3090         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3091                 return -EINVAL;
3092
3093         return 0;
3094 }
3095
3096 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3097                              struct inode *dst, u64 dst_loff)
3098 {
3099         int ret;
3100         u64 len = olen;
3101         struct cmp_pages cmp;
3102         bool same_inode = (src == dst);
3103         u64 same_lock_start = 0;
3104         u64 same_lock_len = 0;
3105
3106         if (len == 0)
3107                 return 0;
3108
3109         if (same_inode)
3110                 inode_lock(src);
3111         else
3112                 btrfs_double_inode_lock(src, dst);
3113
3114         ret = extent_same_check_offsets(src, loff, &len, olen);
3115         if (ret)
3116                 goto out_unlock;
3117
3118         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3119         if (ret)
3120                 goto out_unlock;
3121
3122         if (same_inode) {
3123                 /*
3124                  * Single inode case wants the same checks, except we
3125                  * don't want our length pushed out past i_size as
3126                  * comparing that data range makes no sense.
3127                  *
3128                  * extent_same_check_offsets() will do this for an
3129                  * unaligned length at i_size, so catch it here and
3130                  * reject the request.
3131                  *
3132                  * This effectively means we require aligned extents
3133                  * for the single-inode case, whereas the other cases
3134                  * allow an unaligned length so long as it ends at
3135                  * i_size.
3136                  */
3137                 if (len != olen) {
3138                         ret = -EINVAL;
3139                         goto out_unlock;
3140                 }
3141
3142                 /* Check for overlapping ranges */
3143                 if (dst_loff + len > loff && dst_loff < loff + len) {
3144                         ret = -EINVAL;
3145                         goto out_unlock;
3146                 }
3147
3148                 same_lock_start = min_t(u64, loff, dst_loff);
3149                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3150         }
3151
3152         /* don't make the dst file partly checksummed */
3153         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3154             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3155                 ret = -EINVAL;
3156                 goto out_unlock;
3157         }
3158
3159 again:
3160         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3161         if (ret)
3162                 goto out_unlock;
3163
3164         if (same_inode)
3165                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3166                                         false);
3167         else
3168                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3169                                                false);
3170         /*
3171          * If one of the inodes has dirty pages in the respective range or
3172          * ordered extents, we need to flush dellaloc and wait for all ordered
3173          * extents in the range. We must unlock the pages and the ranges in the
3174          * io trees to avoid deadlocks when flushing delalloc (requires locking
3175          * pages) and when waiting for ordered extents to complete (they require
3176          * range locking).
3177          */
3178         if (ret == -EAGAIN) {
3179                 /*
3180                  * Ranges in the io trees already unlocked. Now unlock all
3181                  * pages before waiting for all IO to complete.
3182                  */
3183                 btrfs_cmp_data_free(&cmp);
3184                 if (same_inode) {
3185                         btrfs_wait_ordered_range(src, same_lock_start,
3186                                                  same_lock_len);
3187                 } else {
3188                         btrfs_wait_ordered_range(src, loff, len);
3189                         btrfs_wait_ordered_range(dst, dst_loff, len);
3190                 }
3191                 goto again;
3192         }
3193         ASSERT(ret == 0);
3194         if (WARN_ON(ret)) {
3195                 /* ranges in the io trees already unlocked */
3196                 btrfs_cmp_data_free(&cmp);
3197                 return ret;
3198         }
3199
3200         /* pass original length for comparison so we stay within i_size */
3201         ret = btrfs_cmp_data(olen, &cmp);
3202         if (ret == 0)
3203                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3204
3205         if (same_inode)
3206                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3207                               same_lock_start + same_lock_len - 1);
3208         else
3209                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3210
3211         btrfs_cmp_data_free(&cmp);
3212 out_unlock:
3213         if (same_inode)
3214                 inode_unlock(src);
3215         else
3216                 btrfs_double_inode_unlock(src, dst);
3217
3218         return ret;
3219 }
3220
3221 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3222
3223 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3224                                 struct file *dst_file, u64 dst_loff)
3225 {
3226         struct inode *src = file_inode(src_file);
3227         struct inode *dst = file_inode(dst_file);
3228         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3229         ssize_t res;
3230
3231         if (olen > BTRFS_MAX_DEDUPE_LEN)
3232                 olen = BTRFS_MAX_DEDUPE_LEN;
3233
3234         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3235                 /*
3236                  * Btrfs does not support blocksize < page_size. As a
3237                  * result, btrfs_cmp_data() won't correctly handle
3238                  * this situation without an update.
3239                  */
3240                 return -EINVAL;
3241         }
3242
3243         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3244         if (res)
3245                 return res;
3246         return olen;
3247 }
3248
3249 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3250                                      struct inode *inode,
3251                                      u64 endoff,
3252                                      const u64 destoff,
3253                                      const u64 olen,
3254                                      int no_time_update)
3255 {
3256         struct btrfs_root *root = BTRFS_I(inode)->root;
3257         int ret;
3258
3259         inode_inc_iversion(inode);
3260         if (!no_time_update)
3261                 inode->i_mtim