Merge branch 'for-4.14/block-postmerge' of git://git.kernel.dk/linux-block
[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 /*
160  * Inherit flags from the parent inode.
161  *
162  * Currently only the compression flags and the cow flags are inherited.
163  */
164 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
165 {
166         unsigned int flags;
167
168         if (!dir)
169                 return;
170
171         flags = BTRFS_I(dir)->flags;
172
173         if (flags & BTRFS_INODE_NOCOMPRESS) {
174                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
175                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
176         } else if (flags & BTRFS_INODE_COMPRESS) {
177                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
178                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
179         }
180
181         if (flags & BTRFS_INODE_NODATACOW) {
182                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
183                 if (S_ISREG(inode->i_mode))
184                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
185         }
186
187         btrfs_update_iflags(inode);
188 }
189
190 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
191 {
192         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
193         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
194
195         if (copy_to_user(arg, &flags, sizeof(flags)))
196                 return -EFAULT;
197         return 0;
198 }
199
200 static int check_flags(unsigned int flags)
201 {
202         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
203                       FS_NOATIME_FL | FS_NODUMP_FL | \
204                       FS_SYNC_FL | FS_DIRSYNC_FL | \
205                       FS_NOCOMP_FL | FS_COMPR_FL |
206                       FS_NOCOW_FL))
207                 return -EOPNOTSUPP;
208
209         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
210                 return -EINVAL;
211
212         return 0;
213 }
214
215 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
216 {
217         struct inode *inode = file_inode(file);
218         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
219         struct btrfs_inode *ip = BTRFS_I(inode);
220         struct btrfs_root *root = ip->root;
221         struct btrfs_trans_handle *trans;
222         unsigned int flags, oldflags;
223         int ret;
224         u64 ip_oldflags;
225         unsigned int i_oldflags;
226         umode_t mode;
227
228         if (!inode_owner_or_capable(inode))
229                 return -EPERM;
230
231         if (btrfs_root_readonly(root))
232                 return -EROFS;
233
234         if (copy_from_user(&flags, arg, sizeof(flags)))
235                 return -EFAULT;
236
237         ret = check_flags(flags);
238         if (ret)
239                 return ret;
240
241         ret = mnt_want_write_file(file);
242         if (ret)
243                 return ret;
244
245         inode_lock(inode);
246
247         ip_oldflags = ip->flags;
248         i_oldflags = inode->i_flags;
249         mode = inode->i_mode;
250
251         flags = btrfs_mask_flags(inode->i_mode, flags);
252         oldflags = btrfs_flags_to_ioctl(ip->flags);
253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
255                         ret = -EPERM;
256                         goto out_unlock;
257                 }
258         }
259
260         if (flags & FS_SYNC_FL)
261                 ip->flags |= BTRFS_INODE_SYNC;
262         else
263                 ip->flags &= ~BTRFS_INODE_SYNC;
264         if (flags & FS_IMMUTABLE_FL)
265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
266         else
267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
268         if (flags & FS_APPEND_FL)
269                 ip->flags |= BTRFS_INODE_APPEND;
270         else
271                 ip->flags &= ~BTRFS_INODE_APPEND;
272         if (flags & FS_NODUMP_FL)
273                 ip->flags |= BTRFS_INODE_NODUMP;
274         else
275                 ip->flags &= ~BTRFS_INODE_NODUMP;
276         if (flags & FS_NOATIME_FL)
277                 ip->flags |= BTRFS_INODE_NOATIME;
278         else
279                 ip->flags &= ~BTRFS_INODE_NOATIME;
280         if (flags & FS_DIRSYNC_FL)
281                 ip->flags |= BTRFS_INODE_DIRSYNC;
282         else
283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
284         if (flags & FS_NOCOW_FL) {
285                 if (S_ISREG(mode)) {
286                         /*
287                          * It's safe to turn csums off here, no extents exist.
288                          * Otherwise we want the flag to reflect the real COW
289                          * status of the file and will not set it.
290                          */
291                         if (inode->i_size == 0)
292                                 ip->flags |= BTRFS_INODE_NODATACOW
293                                            | BTRFS_INODE_NODATASUM;
294                 } else {
295                         ip->flags |= BTRFS_INODE_NODATACOW;
296                 }
297         } else {
298                 /*
299                  * Revert back under same assumptions as above
300                  */
301                 if (S_ISREG(mode)) {
302                         if (inode->i_size == 0)
303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
304                                              | BTRFS_INODE_NODATASUM);
305                 } else {
306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
307                 }
308         }
309
310         /*
311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
312          * flag may be changed automatically if compression code won't make
313          * things smaller.
314          */
315         if (flags & FS_NOCOMP_FL) {
316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
318
319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
320                 if (ret && ret != -ENODATA)
321                         goto out_drop;
322         } else if (flags & FS_COMPR_FL) {
323                 const char *comp;
324
325                 ip->flags |= BTRFS_INODE_COMPRESS;
326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
327
328                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
329                         comp = "lzo";
330                 else
331                         comp = "zlib";
332                 ret = btrfs_set_prop(inode, "btrfs.compression",
333                                      comp, strlen(comp), 0);
334                 if (ret)
335                         goto out_drop;
336
337         } else {
338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
339                 if (ret && ret != -ENODATA)
340                         goto out_drop;
341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
342         }
343
344         trans = btrfs_start_transaction(root, 1);
345         if (IS_ERR(trans)) {
346                 ret = PTR_ERR(trans);
347                 goto out_drop;
348         }
349
350         btrfs_update_iflags(inode);
351         inode_inc_iversion(inode);
352         inode->i_ctime = current_time(inode);
353         ret = btrfs_update_inode(trans, root, inode);
354
355         btrfs_end_transaction(trans);
356  out_drop:
357         if (ret) {
358                 ip->flags = ip_oldflags;
359                 inode->i_flags = i_oldflags;
360         }
361
362  out_unlock:
363         inode_unlock(inode);
364         mnt_drop_write_file(file);
365         return ret;
366 }
367
368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
369 {
370         struct inode *inode = file_inode(file);
371
372         return put_user(inode->i_generation, arg);
373 }
374
375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
376 {
377         struct inode *inode = file_inode(file);
378         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
379         struct btrfs_device *device;
380         struct request_queue *q;
381         struct fstrim_range range;
382         u64 minlen = ULLONG_MAX;
383         u64 num_devices = 0;
384         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
385         int ret;
386
387         if (!capable(CAP_SYS_ADMIN))
388                 return -EPERM;
389
390         rcu_read_lock();
391         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
392                                 dev_list) {
393                 if (!device->bdev)
394                         continue;
395                 q = bdev_get_queue(device->bdev);
396                 if (blk_queue_discard(q)) {
397                         num_devices++;
398                         minlen = min_t(u64, q->limits.discard_granularity,
399                                      minlen);
400                 }
401         }
402         rcu_read_unlock();
403
404         if (!num_devices)
405                 return -EOPNOTSUPP;
406         if (copy_from_user(&range, arg, sizeof(range)))
407                 return -EFAULT;
408         if (range.start > total_bytes ||
409             range.len < fs_info->sb->s_blocksize)
410                 return -EINVAL;
411
412         range.len = min(range.len, total_bytes - range.start);
413         range.minlen = max(range.minlen, minlen);
414         ret = btrfs_trim_fs(fs_info, &range);
415         if (ret < 0)
416                 return ret;
417
418         if (copy_to_user(arg, &range, sizeof(range)))
419                 return -EFAULT;
420
421         return 0;
422 }
423
424 int btrfs_is_empty_uuid(u8 *uuid)
425 {
426         int i;
427
428         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
429                 if (uuid[i])
430                         return 0;
431         }
432         return 1;
433 }
434
435 static noinline int create_subvol(struct inode *dir,
436                                   struct dentry *dentry,
437                                   const char *name, int namelen,
438                                   u64 *async_transid,
439                                   struct btrfs_qgroup_inherit *inherit)
440 {
441         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
442         struct btrfs_trans_handle *trans;
443         struct btrfs_key key;
444         struct btrfs_root_item *root_item;
445         struct btrfs_inode_item *inode_item;
446         struct extent_buffer *leaf;
447         struct btrfs_root *root = BTRFS_I(dir)->root;
448         struct btrfs_root *new_root;
449         struct btrfs_block_rsv block_rsv;
450         struct timespec cur_time = current_time(dir);
451         struct inode *inode;
452         int ret;
453         int err;
454         u64 objectid;
455         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
456         u64 index = 0;
457         u64 qgroup_reserved;
458         uuid_le new_uuid;
459
460         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
461         if (!root_item)
462                 return -ENOMEM;
463
464         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
465         if (ret)
466                 goto fail_free;
467
468         /*
469          * Don't create subvolume whose level is not zero. Or qgroup will be
470          * screwed up since it assumes subvolume qgroup's level to be 0.
471          */
472         if (btrfs_qgroup_level(objectid)) {
473                 ret = -ENOSPC;
474                 goto fail_free;
475         }
476
477         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
478         /*
479          * The same as the snapshot creation, please see the comment
480          * of create_snapshot().
481          */
482         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
483                                                8, &qgroup_reserved, false);
484         if (ret)
485                 goto fail_free;
486
487         trans = btrfs_start_transaction(root, 0);
488         if (IS_ERR(trans)) {
489                 ret = PTR_ERR(trans);
490                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
491                 goto fail_free;
492         }
493         trans->block_rsv = &block_rsv;
494         trans->bytes_reserved = block_rsv.size;
495
496         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
497         if (ret)
498                 goto fail;
499
500         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
501         if (IS_ERR(leaf)) {
502                 ret = PTR_ERR(leaf);
503                 goto fail;
504         }
505
506         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
507         btrfs_set_header_bytenr(leaf, leaf->start);
508         btrfs_set_header_generation(leaf, trans->transid);
509         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
510         btrfs_set_header_owner(leaf, objectid);
511
512         write_extent_buffer_fsid(leaf, fs_info->fsid);
513         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
514         btrfs_mark_buffer_dirty(leaf);
515
516         inode_item = &root_item->inode;
517         btrfs_set_stack_inode_generation(inode_item, 1);
518         btrfs_set_stack_inode_size(inode_item, 3);
519         btrfs_set_stack_inode_nlink(inode_item, 1);
520         btrfs_set_stack_inode_nbytes(inode_item,
521                                      fs_info->nodesize);
522         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
523
524         btrfs_set_root_flags(root_item, 0);
525         btrfs_set_root_limit(root_item, 0);
526         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
527
528         btrfs_set_root_bytenr(root_item, leaf->start);
529         btrfs_set_root_generation(root_item, trans->transid);
530         btrfs_set_root_level(root_item, 0);
531         btrfs_set_root_refs(root_item, 1);
532         btrfs_set_root_used(root_item, leaf->len);
533         btrfs_set_root_last_snapshot(root_item, 0);
534
535         btrfs_set_root_generation_v2(root_item,
536                         btrfs_root_generation(root_item));
537         uuid_le_gen(&new_uuid);
538         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
539         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
540         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
541         root_item->ctime = root_item->otime;
542         btrfs_set_root_ctransid(root_item, trans->transid);
543         btrfs_set_root_otransid(root_item, trans->transid);
544
545         btrfs_tree_unlock(leaf);
546         free_extent_buffer(leaf);
547         leaf = NULL;
548
549         btrfs_set_root_dirid(root_item, new_dirid);
550
551         key.objectid = objectid;
552         key.offset = 0;
553         key.type = BTRFS_ROOT_ITEM_KEY;
554         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
555                                 root_item);
556         if (ret)
557                 goto fail;
558
559         key.offset = (u64)-1;
560         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
561         if (IS_ERR(new_root)) {
562                 ret = PTR_ERR(new_root);
563                 btrfs_abort_transaction(trans, ret);
564                 goto fail;
565         }
566
567         btrfs_record_root_in_trans(trans, new_root);
568
569         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
570         if (ret) {
571                 /* We potentially lose an unused inode item here */
572                 btrfs_abort_transaction(trans, ret);
573                 goto fail;
574         }
575
576         mutex_lock(&new_root->objectid_mutex);
577         new_root->highest_objectid = new_dirid;
578         mutex_unlock(&new_root->objectid_mutex);
579
580         /*
581          * insert the directory item
582          */
583         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
584         if (ret) {
585                 btrfs_abort_transaction(trans, ret);
586                 goto fail;
587         }
588
589         ret = btrfs_insert_dir_item(trans, root,
590                                     name, namelen, BTRFS_I(dir), &key,
591                                     BTRFS_FT_DIR, index);
592         if (ret) {
593                 btrfs_abort_transaction(trans, ret);
594                 goto fail;
595         }
596
597         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
598         ret = btrfs_update_inode(trans, root, dir);
599         BUG_ON(ret);
600
601         ret = btrfs_add_root_ref(trans, fs_info,
602                                  objectid, root->root_key.objectid,
603                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
604         BUG_ON(ret);
605
606         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
607                                   BTRFS_UUID_KEY_SUBVOL, objectid);
608         if (ret)
609                 btrfs_abort_transaction(trans, ret);
610
611 fail:
612         kfree(root_item);
613         trans->block_rsv = NULL;
614         trans->bytes_reserved = 0;
615         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
616
617         if (async_transid) {
618                 *async_transid = trans->transid;
619                 err = btrfs_commit_transaction_async(trans, 1);
620                 if (err)
621                         err = btrfs_commit_transaction(trans);
622         } else {
623                 err = btrfs_commit_transaction(trans);
624         }
625         if (err && !ret)
626                 ret = err;
627
628         if (!ret) {
629                 inode = btrfs_lookup_dentry(dir, dentry);
630                 if (IS_ERR(inode))
631                         return PTR_ERR(inode);
632                 d_instantiate(dentry, inode);
633         }
634         return ret;
635
636 fail_free:
637         kfree(root_item);
638         return ret;
639 }
640
641 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
642 {
643         s64 writers;
644         DEFINE_WAIT(wait);
645
646         do {
647                 prepare_to_wait(&root->subv_writers->wait, &wait,
648                                 TASK_UNINTERRUPTIBLE);
649
650                 writers = percpu_counter_sum(&root->subv_writers->counter);
651                 if (writers)
652                         schedule();
653
654                 finish_wait(&root->subv_writers->wait, &wait);
655         } while (writers);
656 }
657
658 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
659                            struct dentry *dentry,
660                            u64 *async_transid, bool readonly,
661                            struct btrfs_qgroup_inherit *inherit)
662 {
663         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
664         struct inode *inode;
665         struct btrfs_pending_snapshot *pending_snapshot;
666         struct btrfs_trans_handle *trans;
667         int ret;
668
669         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
670                 return -EINVAL;
671
672         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
673         if (!pending_snapshot)
674                 return -ENOMEM;
675
676         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
677                         GFP_KERNEL);
678         pending_snapshot->path = btrfs_alloc_path();
679         if (!pending_snapshot->root_item || !pending_snapshot->path) {
680                 ret = -ENOMEM;
681                 goto free_pending;
682         }
683
684         atomic_inc(&root->will_be_snapshoted);
685         smp_mb__after_atomic();
686         btrfs_wait_for_no_snapshoting_writes(root);
687
688         ret = btrfs_start_delalloc_inodes(root, 0);
689         if (ret)
690                 goto dec_and_free;
691
692         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
693
694         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
695                              BTRFS_BLOCK_RSV_TEMP);
696         /*
697          * 1 - parent dir inode
698          * 2 - dir entries
699          * 1 - root item
700          * 2 - root ref/backref
701          * 1 - root of snapshot
702          * 1 - UUID item
703          */
704         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
705                                         &pending_snapshot->block_rsv, 8,
706                                         &pending_snapshot->qgroup_reserved,
707                                         false);
708         if (ret)
709                 goto dec_and_free;
710
711         pending_snapshot->dentry = dentry;
712         pending_snapshot->root = root;
713         pending_snapshot->readonly = readonly;
714         pending_snapshot->dir = dir;
715         pending_snapshot->inherit = inherit;
716
717         trans = btrfs_start_transaction(root, 0);
718         if (IS_ERR(trans)) {
719                 ret = PTR_ERR(trans);
720                 goto fail;
721         }
722
723         spin_lock(&fs_info->trans_lock);
724         list_add(&pending_snapshot->list,
725                  &trans->transaction->pending_snapshots);
726         spin_unlock(&fs_info->trans_lock);
727         if (async_transid) {
728                 *async_transid = trans->transid;
729                 ret = btrfs_commit_transaction_async(trans, 1);
730                 if (ret)
731                         ret = btrfs_commit_transaction(trans);
732         } else {
733                 ret = btrfs_commit_transaction(trans);
734         }
735         if (ret)
736                 goto fail;
737
738         ret = pending_snapshot->error;
739         if (ret)
740                 goto fail;
741
742         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
743         if (ret)
744                 goto fail;
745
746         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
747         if (IS_ERR(inode)) {
748                 ret = PTR_ERR(inode);
749                 goto fail;
750         }
751
752         d_instantiate(dentry, inode);
753         ret = 0;
754 fail:
755         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
756 dec_and_free:
757         if (atomic_dec_and_test(&root->will_be_snapshoted))
758                 wake_up_atomic_t(&root->will_be_snapshoted);
759 free_pending:
760         kfree(pending_snapshot->root_item);
761         btrfs_free_path(pending_snapshot->path);
762         kfree(pending_snapshot);
763
764         return ret;
765 }
766
767 /*  copy of may_delete in fs/namei.c()
768  *      Check whether we can remove a link victim from directory dir, check
769  *  whether the type of victim is right.
770  *  1. We can't do it if dir is read-only (done in permission())
771  *  2. We should have write and exec permissions on dir
772  *  3. We can't remove anything from append-only dir
773  *  4. We can't do anything with immutable dir (done in permission())
774  *  5. If the sticky bit on dir is set we should either
775  *      a. be owner of dir, or
776  *      b. be owner of victim, or
777  *      c. have CAP_FOWNER capability
778  *  6. If the victim is append-only or immutable we can't do anything with
779  *     links pointing to it.
780  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
781  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
782  *  9. We can't remove a root or mountpoint.
783  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
784  *     nfs_async_unlink().
785  */
786
787 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
788 {
789         int error;
790
791         if (d_really_is_negative(victim))
792                 return -ENOENT;
793
794         BUG_ON(d_inode(victim->d_parent) != dir);
795         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
796
797         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
798         if (error)
799                 return error;
800         if (IS_APPEND(dir))
801                 return -EPERM;
802         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
803             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
804                 return -EPERM;
805         if (isdir) {
806                 if (!d_is_dir(victim))
807                         return -ENOTDIR;
808                 if (IS_ROOT(victim))
809                         return -EBUSY;
810         } else if (d_is_dir(victim))
811                 return -EISDIR;
812         if (IS_DEADDIR(dir))
813                 return -ENOENT;
814         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
815                 return -EBUSY;
816         return 0;
817 }
818
819 /* copy of may_create in fs/namei.c() */
820 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
821 {
822         if (d_really_is_positive(child))
823                 return -EEXIST;
824         if (IS_DEADDIR(dir))
825                 return -ENOENT;
826         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
827 }
828
829 /*
830  * Create a new subvolume below @parent.  This is largely modeled after
831  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
832  * inside this filesystem so it's quite a bit simpler.
833  */
834 static noinline int btrfs_mksubvol(const struct path *parent,
835                                    const char *name, int namelen,
836                                    struct btrfs_root *snap_src,
837                                    u64 *async_transid, bool readonly,
838                                    struct btrfs_qgroup_inherit *inherit)
839 {
840         struct inode *dir = d_inode(parent->dentry);
841         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
842         struct dentry *dentry;
843         int error;
844
845         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
846         if (error == -EINTR)
847                 return error;
848
849         dentry = lookup_one_len(name, parent->dentry, namelen);
850         error = PTR_ERR(dentry);
851         if (IS_ERR(dentry))
852                 goto out_unlock;
853
854         error = btrfs_may_create(dir, dentry);
855         if (error)
856                 goto out_dput;
857
858         /*
859          * even if this name doesn't exist, we may get hash collisions.
860          * check for them now when we can safely fail
861          */
862         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
863                                                dir->i_ino, name,
864                                                namelen);
865         if (error)
866                 goto out_dput;
867
868         down_read(&fs_info->subvol_sem);
869
870         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
871                 goto out_up_read;
872
873         if (snap_src) {
874                 error = create_snapshot(snap_src, dir, dentry,
875                                         async_transid, readonly, inherit);
876         } else {
877                 error = create_subvol(dir, dentry, name, namelen,
878                                       async_transid, inherit);
879         }
880         if (!error)
881                 fsnotify_mkdir(dir, dentry);
882 out_up_read:
883         up_read(&fs_info->subvol_sem);
884 out_dput:
885         dput(dentry);
886 out_unlock:
887         inode_unlock(dir);
888         return error;
889 }
890
891 /*
892  * When we're defragging a range, we don't want to kick it off again
893  * if it is really just waiting for delalloc to send it down.
894  * If we find a nice big extent or delalloc range for the bytes in the
895  * file you want to defrag, we return 0 to let you know to skip this
896  * part of the file
897  */
898 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
899 {
900         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
901         struct extent_map *em = NULL;
902         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
903         u64 end;
904
905         read_lock(&em_tree->lock);
906         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
907         read_unlock(&em_tree->lock);
908
909         if (em) {
910                 end = extent_map_end(em);
911                 free_extent_map(em);
912                 if (end - offset > thresh)
913                         return 0;
914         }
915         /* if we already have a nice delalloc here, just stop */
916         thresh /= 2;
917         end = count_range_bits(io_tree, &offset, offset + thresh,
918                                thresh, EXTENT_DELALLOC, 1);
919         if (end >= thresh)
920                 return 0;
921         return 1;
922 }
923
924 /*
925  * helper function to walk through a file and find extents
926  * newer than a specific transid, and smaller than thresh.
927  *
928  * This is used by the defragging code to find new and small
929  * extents
930  */
931 static int find_new_extents(struct btrfs_root *root,
932                             struct inode *inode, u64 newer_than,
933                             u64 *off, u32 thresh)
934 {
935         struct btrfs_path *path;
936         struct btrfs_key min_key;
937         struct extent_buffer *leaf;
938         struct btrfs_file_extent_item *extent;
939         int type;
940         int ret;
941         u64 ino = btrfs_ino(BTRFS_I(inode));
942
943         path = btrfs_alloc_path();
944         if (!path)
945                 return -ENOMEM;
946
947         min_key.objectid = ino;
948         min_key.type = BTRFS_EXTENT_DATA_KEY;
949         min_key.offset = *off;
950
951         while (1) {
952                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
953                 if (ret != 0)
954                         goto none;
955 process_slot:
956                 if (min_key.objectid != ino)
957                         goto none;
958                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
959                         goto none;
960
961                 leaf = path->nodes[0];
962                 extent = btrfs_item_ptr(leaf, path->slots[0],
963                                         struct btrfs_file_extent_item);
964
965                 type = btrfs_file_extent_type(leaf, extent);
966                 if (type == BTRFS_FILE_EXTENT_REG &&
967                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
968                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
969                         *off = min_key.offset;
970                         btrfs_free_path(path);
971                         return 0;
972                 }
973
974                 path->slots[0]++;
975                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
976                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
977                         goto process_slot;
978                 }
979
980                 if (min_key.offset == (u64)-1)
981                         goto none;
982
983                 min_key.offset++;
984                 btrfs_release_path(path);
985         }
986 none:
987         btrfs_free_path(path);
988         return -ENOENT;
989 }
990
991 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
992 {
993         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
994         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
995         struct extent_map *em;
996         u64 len = PAGE_SIZE;
997
998         /*
999          * hopefully we have this extent in the tree already, try without
1000          * the full extent lock
1001          */
1002         read_lock(&em_tree->lock);
1003         em = lookup_extent_mapping(em_tree, start, len);
1004         read_unlock(&em_tree->lock);
1005
1006         if (!em) {
1007                 struct extent_state *cached = NULL;
1008                 u64 end = start + len - 1;
1009
1010                 /* get the big lock and read metadata off disk */
1011                 lock_extent_bits(io_tree, start, end, &cached);
1012                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1013                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1014
1015                 if (IS_ERR(em))
1016                         return NULL;
1017         }
1018
1019         return em;
1020 }
1021
1022 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1023 {
1024         struct extent_map *next;
1025         bool ret = true;
1026
1027         /* this is the last extent */
1028         if (em->start + em->len >= i_size_read(inode))
1029                 return false;
1030
1031         next = defrag_lookup_extent(inode, em->start + em->len);
1032         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1033                 ret = false;
1034         else if ((em->block_start + em->block_len == next->block_start) &&
1035                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1036                 ret = false;
1037
1038         free_extent_map(next);
1039         return ret;
1040 }
1041
1042 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1043                                u64 *last_len, u64 *skip, u64 *defrag_end,
1044                                int compress)
1045 {
1046         struct extent_map *em;
1047         int ret = 1;
1048         bool next_mergeable = true;
1049         bool prev_mergeable = true;
1050
1051         /*
1052          * make sure that once we start defragging an extent, we keep on
1053          * defragging it
1054          */
1055         if (start < *defrag_end)
1056                 return 1;
1057
1058         *skip = 0;
1059
1060         em = defrag_lookup_extent(inode, start);
1061         if (!em)
1062                 return 0;
1063
1064         /* this will cover holes, and inline extents */
1065         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1066                 ret = 0;
1067                 goto out;
1068         }
1069
1070         if (!*defrag_end)
1071                 prev_mergeable = false;
1072
1073         next_mergeable = defrag_check_next_extent(inode, em);
1074         /*
1075          * we hit a real extent, if it is big or the next extent is not a
1076          * real extent, don't bother defragging it
1077          */
1078         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1079             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1080                 ret = 0;
1081 out:
1082         /*
1083          * last_len ends up being a counter of how many bytes we've defragged.
1084          * every time we choose not to defrag an extent, we reset *last_len
1085          * so that the next tiny extent will force a defrag.
1086          *
1087          * The end result of this is that tiny extents before a single big
1088          * extent will force at least part of that big extent to be defragged.
1089          */
1090         if (ret) {
1091                 *defrag_end = extent_map_end(em);
1092         } else {
1093                 *last_len = 0;
1094                 *skip = extent_map_end(em);
1095                 *defrag_end = 0;
1096         }
1097
1098         free_extent_map(em);
1099         return ret;
1100 }
1101
1102 /*
1103  * it doesn't do much good to defrag one or two pages
1104  * at a time.  This pulls in a nice chunk of pages
1105  * to COW and defrag.
1106  *
1107  * It also makes sure the delalloc code has enough
1108  * dirty data to avoid making new small extents as part
1109  * of the defrag
1110  *
1111  * It's a good idea to start RA on this range
1112  * before calling this.
1113  */
1114 static int cluster_pages_for_defrag(struct inode *inode,
1115                                     struct page **pages,
1116                                     unsigned long start_index,
1117                                     unsigned long num_pages)
1118 {
1119         unsigned long file_end;
1120         u64 isize = i_size_read(inode);
1121         u64 page_start;
1122         u64 page_end;
1123         u64 page_cnt;
1124         int ret;
1125         int i;
1126         int i_done;
1127         struct btrfs_ordered_extent *ordered;
1128         struct extent_state *cached_state = NULL;
1129         struct extent_io_tree *tree;
1130         struct extent_changeset *data_reserved = NULL;
1131         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1132
1133         file_end = (isize - 1) >> PAGE_SHIFT;
1134         if (!isize || start_index > file_end)
1135                 return 0;
1136
1137         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1138
1139         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1140                         start_index << PAGE_SHIFT,
1141                         page_cnt << PAGE_SHIFT);
1142         if (ret)
1143                 return ret;
1144         i_done = 0;
1145         tree = &BTRFS_I(inode)->io_tree;
1146
1147         /* step one, lock all the pages */
1148         for (i = 0; i < page_cnt; i++) {
1149                 struct page *page;
1150 again:
1151                 page = find_or_create_page(inode->i_mapping,
1152                                            start_index + i, mask);
1153                 if (!page)
1154                         break;
1155
1156                 page_start = page_offset(page);
1157                 page_end = page_start + PAGE_SIZE - 1;
1158                 while (1) {
1159                         lock_extent_bits(tree, page_start, page_end,
1160                                          &cached_state);
1161                         ordered = btrfs_lookup_ordered_extent(inode,
1162                                                               page_start);
1163                         unlock_extent_cached(tree, page_start, page_end,
1164                                              &cached_state, GFP_NOFS);
1165                         if (!ordered)
1166                                 break;
1167
1168                         unlock_page(page);
1169                         btrfs_start_ordered_extent(inode, ordered, 1);
1170                         btrfs_put_ordered_extent(ordered);
1171                         lock_page(page);
1172                         /*
1173                          * we unlocked the page above, so we need check if
1174                          * it was released or not.
1175                          */
1176                         if (page->mapping != inode->i_mapping) {
1177                                 unlock_page(page);
1178                                 put_page(page);
1179                                 goto again;
1180                         }
1181                 }
1182
1183                 if (!PageUptodate(page)) {
1184                         btrfs_readpage(NULL, page);
1185                         lock_page(page);
1186                         if (!PageUptodate(page)) {
1187                                 unlock_page(page);
1188                                 put_page(page);
1189                                 ret = -EIO;
1190                                 break;
1191                         }
1192                 }
1193
1194                 if (page->mapping != inode->i_mapping) {
1195                         unlock_page(page);
1196                         put_page(page);
1197                         goto again;
1198                 }
1199
1200                 pages[i] = page;
1201                 i_done++;
1202         }
1203         if (!i_done || ret)
1204                 goto out;
1205
1206         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1207                 goto out;
1208
1209         /*
1210          * so now we have a nice long stream of locked
1211          * and up to date pages, lets wait on them
1212          */
1213         for (i = 0; i < i_done; i++)
1214                 wait_on_page_writeback(pages[i]);
1215
1216         page_start = page_offset(pages[0]);
1217         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1218
1219         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1220                          page_start, page_end - 1, &cached_state);
1221         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1222                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1223                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1224                           &cached_state, GFP_NOFS);
1225
1226         if (i_done != page_cnt) {
1227                 spin_lock(&BTRFS_I(inode)->lock);
1228                 BTRFS_I(inode)->outstanding_extents++;
1229                 spin_unlock(&BTRFS_I(inode)->lock);
1230                 btrfs_delalloc_release_space(inode, data_reserved,
1231                                 start_index << PAGE_SHIFT,
1232                                 (page_cnt - i_done) << PAGE_SHIFT);
1233         }
1234
1235
1236         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1237                           &cached_state);
1238
1239         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1240                              page_start, page_end - 1, &cached_state,
1241                              GFP_NOFS);
1242
1243         for (i = 0; i < i_done; i++) {
1244                 clear_page_dirty_for_io(pages[i]);
1245                 ClearPageChecked(pages[i]);
1246                 set_page_extent_mapped(pages[i]);
1247                 set_page_dirty(pages[i]);
1248                 unlock_page(pages[i]);
1249                 put_page(pages[i]);
1250         }
1251         extent_changeset_free(data_reserved);
1252         return i_done;
1253 out:
1254         for (i = 0; i < i_done; i++) {
1255                 unlock_page(pages[i]);
1256                 put_page(pages[i]);
1257         }
1258         btrfs_delalloc_release_space(inode, data_reserved,
1259                         start_index << PAGE_SHIFT,
1260                         page_cnt << PAGE_SHIFT);
1261         extent_changeset_free(data_reserved);
1262         return ret;
1263
1264 }
1265
1266 int btrfs_defrag_file(struct inode *inode, struct file *file,
1267                       struct btrfs_ioctl_defrag_range_args *range,
1268                       u64 newer_than, unsigned long max_to_defrag)
1269 {
1270         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272         struct file_ra_state *ra = NULL;
1273         unsigned long last_index;
1274         u64 isize = i_size_read(inode);
1275         u64 last_len = 0;
1276         u64 skip = 0;
1277         u64 defrag_end = 0;
1278         u64 newer_off = range->start;
1279         unsigned long i;
1280         unsigned long ra_index = 0;
1281         int ret;
1282         int defrag_count = 0;
1283         int compress_type = BTRFS_COMPRESS_ZLIB;
1284         u32 extent_thresh = range->extent_thresh;
1285         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1286         unsigned long cluster = max_cluster;
1287         u64 new_align = ~((u64)SZ_128K - 1);
1288         struct page **pages = NULL;
1289
1290         if (isize == 0)
1291                 return 0;
1292
1293         if (range->start >= isize)
1294                 return -EINVAL;
1295
1296         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1297                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1298                         return -EINVAL;
1299                 if (range->compress_type)
1300                         compress_type = range->compress_type;
1301         }
1302
1303         if (extent_thresh == 0)
1304                 extent_thresh = SZ_256K;
1305
1306         /*
1307          * if we were not given a file, allocate a readahead
1308          * context
1309          */
1310         if (!file) {
1311                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1312                 if (!ra)
1313                         return -ENOMEM;
1314                 file_ra_state_init(ra, inode->i_mapping);
1315         } else {
1316                 ra = &file->f_ra;
1317         }
1318
1319         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1320                         GFP_NOFS);
1321         if (!pages) {
1322                 ret = -ENOMEM;
1323                 goto out_ra;
1324         }
1325
1326         /* find the last page to defrag */
1327         if (range->start + range->len > range->start) {
1328                 last_index = min_t(u64, isize - 1,
1329                          range->start + range->len - 1) >> PAGE_SHIFT;
1330         } else {
1331                 last_index = (isize - 1) >> PAGE_SHIFT;
1332         }
1333
1334         if (newer_than) {
1335                 ret = find_new_extents(root, inode, newer_than,
1336                                        &newer_off, SZ_64K);
1337                 if (!ret) {
1338                         range->start = newer_off;
1339                         /*
1340                          * we always align our defrag to help keep
1341                          * the extents in the file evenly spaced
1342                          */
1343                         i = (newer_off & new_align) >> PAGE_SHIFT;
1344                 } else
1345                         goto out_ra;
1346         } else {
1347                 i = range->start >> PAGE_SHIFT;
1348         }
1349         if (!max_to_defrag)
1350                 max_to_defrag = last_index - i + 1;
1351
1352         /*
1353          * make writeback starts from i, so the defrag range can be
1354          * written sequentially.
1355          */
1356         if (i < inode->i_mapping->writeback_index)
1357                 inode->i_mapping->writeback_index = i;
1358
1359         while (i <= last_index && defrag_count < max_to_defrag &&
1360                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1361                 /*
1362                  * make sure we stop running if someone unmounts
1363                  * the FS
1364                  */
1365                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1366                         break;
1367
1368                 if (btrfs_defrag_cancelled(fs_info)) {
1369                         btrfs_debug(fs_info, "defrag_file cancelled");
1370                         ret = -EAGAIN;
1371                         break;
1372                 }
1373
1374                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1375                                          extent_thresh, &last_len, &skip,
1376                                          &defrag_end, range->flags &
1377                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1378                         unsigned long next;
1379                         /*
1380                          * the should_defrag function tells us how much to skip
1381                          * bump our counter by the suggested amount
1382                          */
1383                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1384                         i = max(i + 1, next);
1385                         continue;
1386                 }
1387
1388                 if (!newer_than) {
1389                         cluster = (PAGE_ALIGN(defrag_end) >>
1390                                    PAGE_SHIFT) - i;
1391                         cluster = min(cluster, max_cluster);
1392                 } else {
1393                         cluster = max_cluster;
1394                 }
1395
1396                 if (i + cluster > ra_index) {
1397                         ra_index = max(i, ra_index);
1398                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1399                                        cluster);
1400                         ra_index += cluster;
1401                 }
1402
1403                 inode_lock(inode);
1404                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1405                         BTRFS_I(inode)->force_compress = compress_type;
1406                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1407                 if (ret < 0) {
1408                         inode_unlock(inode);
1409                         goto out_ra;
1410                 }
1411
1412                 defrag_count += ret;
1413                 balance_dirty_pages_ratelimited(inode->i_mapping);
1414                 inode_unlock(inode);
1415
1416                 if (newer_than) {
1417                         if (newer_off == (u64)-1)
1418                                 break;
1419
1420                         if (ret > 0)
1421                                 i += ret;
1422
1423                         newer_off = max(newer_off + 1,
1424                                         (u64)i << PAGE_SHIFT);
1425
1426                         ret = find_new_extents(root, inode, newer_than,
1427                                                &newer_off, SZ_64K);
1428                         if (!ret) {
1429                                 range->start = newer_off;
1430                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1431                         } else {
1432                                 break;
1433                         }
1434                 } else {
1435                         if (ret > 0) {
1436                                 i += ret;
1437                                 last_len += ret << PAGE_SHIFT;
1438                         } else {
1439                                 i++;
1440                                 last_len = 0;
1441                         }
1442                 }
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1446                 filemap_flush(inode->i_mapping);
1447                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1448                              &BTRFS_I(inode)->runtime_flags))
1449                         filemap_flush(inode->i_mapping);
1450         }
1451
1452         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1453                 /* the filemap_flush will queue IO into the worker threads, but
1454                  * we have to make sure the IO is actually started and that
1455                  * ordered extents get created before we return
1456                  */
1457                 atomic_inc(&fs_info->async_submit_draining);
1458                 while (atomic_read(&fs_info->nr_async_submits) ||
1459                        atomic_read(&fs_info->async_delalloc_pages)) {
1460                         wait_event(fs_info->async_submit_wait,
1461                                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
1462                                     atomic_read(&fs_info->async_delalloc_pages) == 0));
1463                 }
1464                 atomic_dec(&fs_info->async_submit_draining);
1465         }
1466
1467         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1468                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1469         }
1470
1471         ret = defrag_count;
1472
1473 out_ra:
1474         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1475                 inode_lock(inode);
1476                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1477                 inode_unlock(inode);
1478         }
1479         if (!file)
1480                 kfree(ra);
1481         kfree(pages);
1482         return ret;
1483 }
1484
1485 static noinline int btrfs_ioctl_resize(struct file *file,
1486                                         void __user *arg)
1487 {
1488         struct inode *inode = file_inode(file);
1489         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1490         u64 new_size;
1491         u64 old_size;
1492         u64 devid = 1;
1493         struct btrfs_root *root = BTRFS_I(inode)->root;
1494         struct btrfs_ioctl_vol_args *vol_args;
1495         struct btrfs_trans_handle *trans;
1496         struct btrfs_device *device = NULL;
1497         char *sizestr;
1498         char *retptr;
1499         char *devstr = NULL;
1500         int ret = 0;
1501         int mod = 0;
1502
1503         if (!capable(CAP_SYS_ADMIN))
1504                 return -EPERM;
1505
1506         ret = mnt_want_write_file(file);
1507         if (ret)
1508                 return ret;
1509
1510         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1511                 mnt_drop_write_file(file);
1512                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1513         }
1514
1515         mutex_lock(&fs_info->volume_mutex);
1516         vol_args = memdup_user(arg, sizeof(*vol_args));
1517         if (IS_ERR(vol_args)) {
1518                 ret = PTR_ERR(vol_args);
1519                 goto out;
1520         }
1521
1522         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1523
1524         sizestr = vol_args->name;
1525         devstr = strchr(sizestr, ':');
1526         if (devstr) {
1527                 sizestr = devstr + 1;
1528                 *devstr = '\0';
1529                 devstr = vol_args->name;
1530                 ret = kstrtoull(devstr, 10, &devid);
1531                 if (ret)
1532                         goto out_free;
1533                 if (!devid) {
1534                         ret = -EINVAL;
1535                         goto out_free;
1536                 }
1537                 btrfs_info(fs_info, "resizing devid %llu", devid);
1538         }
1539
1540         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1541         if (!device) {
1542                 btrfs_info(fs_info, "resizer unable to find device %llu",
1543                            devid);
1544                 ret = -ENODEV;
1545                 goto out_free;
1546         }
1547
1548         if (!device->writeable) {
1549                 btrfs_info(fs_info,
1550                            "resizer unable to apply on readonly device %llu",
1551                        devid);
1552                 ret = -EPERM;
1553                 goto out_free;
1554         }
1555
1556         if (!strcmp(sizestr, "max"))
1557                 new_size = device->bdev->bd_inode->i_size;
1558         else {
1559                 if (sizestr[0] == '-') {
1560                         mod = -1;
1561                         sizestr++;
1562                 } else if (sizestr[0] == '+') {
1563                         mod = 1;
1564                         sizestr++;
1565                 }
1566                 new_size = memparse(sizestr, &retptr);
1567                 if (*retptr != '\0' || new_size == 0) {
1568                         ret = -EINVAL;
1569                         goto out_free;
1570                 }
1571         }
1572
1573         if (device->is_tgtdev_for_dev_replace) {
1574                 ret = -EPERM;
1575                 goto out_free;
1576         }
1577
1578         old_size = btrfs_device_get_total_bytes(device);
1579
1580         if (mod < 0) {
1581                 if (new_size > old_size) {
1582                         ret = -EINVAL;
1583                         goto out_free;
1584                 }
1585                 new_size = old_size - new_size;
1586         } else if (mod > 0) {
1587                 if (new_size > ULLONG_MAX - old_size) {
1588                         ret = -ERANGE;
1589                         goto out_free;
1590                 }
1591                 new_size = old_size + new_size;
1592         }
1593
1594         if (new_size < SZ_256M) {
1595                 ret = -EINVAL;
1596                 goto out_free;
1597         }
1598         if (new_size > device->bdev->bd_inode->i_size) {
1599                 ret = -EFBIG;
1600                 goto out_free;
1601         }
1602
1603         new_size = div_u64(new_size, fs_info->sectorsize);
1604         new_size *= fs_info->sectorsize;
1605
1606         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1607                           rcu_str_deref(device->name), new_size);
1608
1609         if (new_size > old_size) {
1610                 trans = btrfs_start_transaction(root, 0);
1611                 if (IS_ERR(trans)) {
1612                         ret = PTR_ERR(trans);
1613                         goto out_free;
1614                 }
1615                 ret = btrfs_grow_device(trans, device, new_size);
1616                 btrfs_commit_transaction(trans);
1617         } else if (new_size < old_size) {
1618                 ret = btrfs_shrink_device(device, new_size);
1619         } /* equal, nothing need to do */
1620
1621 out_free:
1622         kfree(vol_args);
1623 out:
1624         mutex_unlock(&fs_info->volume_mutex);
1625         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1626         mnt_drop_write_file(file);
1627         return ret;
1628 }
1629
1630 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1631                                 const char *name, unsigned long fd, int subvol,
1632                                 u64 *transid, bool readonly,
1633                                 struct btrfs_qgroup_inherit *inherit)
1634 {
1635         int namelen;
1636         int ret = 0;
1637
1638         if (!S_ISDIR(file_inode(file)->i_mode))
1639                 return -ENOTDIR;
1640
1641         ret = mnt_want_write_file(file);
1642         if (ret)
1643                 goto out;
1644
1645         namelen = strlen(name);
1646         if (strchr(name, '/')) {
1647                 ret = -EINVAL;
1648                 goto out_drop_write;
1649         }
1650
1651         if (name[0] == '.' &&
1652            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1653                 ret = -EEXIST;
1654                 goto out_drop_write;
1655         }
1656
1657         if (subvol) {
1658                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1659                                      NULL, transid, readonly, inherit);
1660         } else {
1661                 struct fd src = fdget(fd);
1662                 struct inode *src_inode;
1663                 if (!src.file) {
1664                         ret = -EINVAL;
1665                         goto out_drop_write;
1666                 }
1667
1668                 src_inode = file_inode(src.file);
1669                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1670                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1671                                    "Snapshot src from another FS");
1672                         ret = -EXDEV;
1673                 } else if (!inode_owner_or_capable(src_inode)) {
1674                         /*
1675                          * Subvolume creation is not restricted, but snapshots
1676                          * are limited to own subvolumes only
1677                          */
1678                         ret = -EPERM;
1679                 } else {
1680                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1681                                              BTRFS_I(src_inode)->root,
1682                                              transid, readonly, inherit);
1683                 }
1684                 fdput(src);
1685         }
1686 out_drop_write:
1687         mnt_drop_write_file(file);
1688 out:
1689         return ret;
1690 }
1691
1692 static noinline int btrfs_ioctl_snap_create(struct file *file,
1693                                             void __user *arg, int subvol)
1694 {
1695         struct btrfs_ioctl_vol_args *vol_args;
1696         int ret;
1697
1698         if (!S_ISDIR(file_inode(file)->i_mode))
1699                 return -ENOTDIR;
1700
1701         vol_args = memdup_user(arg, sizeof(*vol_args));
1702         if (IS_ERR(vol_args))
1703                 return PTR_ERR(vol_args);
1704         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1705
1706         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1707                                               vol_args->fd, subvol,
1708                                               NULL, false, NULL);
1709
1710         kfree(vol_args);
1711         return ret;
1712 }
1713
1714 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1715                                                void __user *arg, int subvol)
1716 {
1717         struct btrfs_ioctl_vol_args_v2 *vol_args;
1718         int ret;
1719         u64 transid = 0;
1720         u64 *ptr = NULL;
1721         bool readonly = false;
1722         struct btrfs_qgroup_inherit *inherit = NULL;
1723
1724         if (!S_ISDIR(file_inode(file)->i_mode))
1725                 return -ENOTDIR;
1726
1727         vol_args = memdup_user(arg, sizeof(*vol_args));
1728         if (IS_ERR(vol_args))
1729                 return PTR_ERR(vol_args);
1730         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1731
1732         if (vol_args->flags &
1733             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1734               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1735                 ret = -EOPNOTSUPP;
1736                 goto free_args;
1737         }
1738
1739         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1740                 ptr = &transid;
1741         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1742                 readonly = true;
1743         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1744                 if (vol_args->size > PAGE_SIZE) {
1745                         ret = -EINVAL;
1746                         goto free_args;
1747                 }
1748                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1749                 if (IS_ERR(inherit)) {
1750                         ret = PTR_ERR(inherit);
1751                         goto free_args;
1752                 }
1753         }
1754
1755         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1756                                               vol_args->fd, subvol, ptr,
1757                                               readonly, inherit);
1758         if (ret)
1759                 goto free_inherit;
1760
1761         if (ptr && copy_to_user(arg +
1762                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1763                                         transid),
1764                                 ptr, sizeof(*ptr)))
1765                 ret = -EFAULT;
1766
1767 free_inherit:
1768         kfree(inherit);
1769 free_args:
1770         kfree(vol_args);
1771         return ret;
1772 }
1773
1774 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1775                                                 void __user *arg)
1776 {
1777         struct inode *inode = file_inode(file);
1778         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1779         struct btrfs_root *root = BTRFS_I(inode)->root;
1780         int ret = 0;
1781         u64 flags = 0;
1782
1783         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1784                 return -EINVAL;
1785
1786         down_read(&fs_info->subvol_sem);
1787         if (btrfs_root_readonly(root))
1788                 flags |= BTRFS_SUBVOL_RDONLY;
1789         up_read(&fs_info->subvol_sem);
1790
1791         if (copy_to_user(arg, &flags, sizeof(flags)))
1792                 ret = -EFAULT;
1793
1794         return ret;
1795 }
1796
1797 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1798                                               void __user *arg)
1799 {
1800         struct inode *inode = file_inode(file);
1801         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1802         struct btrfs_root *root = BTRFS_I(inode)->root;
1803         struct btrfs_trans_handle *trans;
1804         u64 root_flags;
1805         u64 flags;
1806         int ret = 0;
1807
1808         if (!inode_owner_or_capable(inode))
1809                 return -EPERM;
1810
1811         ret = mnt_want_write_file(file);
1812         if (ret)
1813                 goto out;
1814
1815         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1816                 ret = -EINVAL;
1817                 goto out_drop_write;
1818         }
1819
1820         if (copy_from_user(&flags, arg, sizeof(flags))) {
1821                 ret = -EFAULT;
1822                 goto out_drop_write;
1823         }
1824
1825         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1826                 ret = -EINVAL;
1827                 goto out_drop_write;
1828         }
1829
1830         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1831                 ret = -EOPNOTSUPP;
1832                 goto out_drop_write;
1833         }
1834
1835         down_write(&fs_info->subvol_sem);
1836
1837         /* nothing to do */
1838         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1839                 goto out_drop_sem;
1840
1841         root_flags = btrfs_root_flags(&root->root_item);
1842         if (flags & BTRFS_SUBVOL_RDONLY) {
1843                 btrfs_set_root_flags(&root->root_item,
1844                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1845         } else {
1846                 /*
1847                  * Block RO -> RW transition if this subvolume is involved in
1848                  * send
1849                  */
1850                 spin_lock(&root->root_item_lock);
1851                 if (root->send_in_progress == 0) {
1852                         btrfs_set_root_flags(&root->root_item,
1853                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1854                         spin_unlock(&root->root_item_lock);
1855                 } else {
1856                         spin_unlock(&root->root_item_lock);
1857                         btrfs_warn(fs_info,
1858                                    "Attempt to set subvolume %llu read-write during send",
1859                                    root->root_key.objectid);
1860                         ret = -EPERM;
1861                         goto out_drop_sem;
1862                 }
1863         }
1864
1865         trans = btrfs_start_transaction(root, 1);
1866         if (IS_ERR(trans)) {
1867                 ret = PTR_ERR(trans);
1868                 goto out_reset;
1869         }
1870
1871         ret = btrfs_update_root(trans, fs_info->tree_root,
1872                                 &root->root_key, &root->root_item);
1873
1874         btrfs_commit_transaction(trans);
1875 out_reset:
1876         if (ret)
1877                 btrfs_set_root_flags(&root->root_item, root_flags);
1878 out_drop_sem:
1879         up_write(&fs_info->subvol_sem);
1880 out_drop_write:
1881         mnt_drop_write_file(file);
1882 out:
1883         return ret;
1884 }
1885
1886 /*
1887  * helper to check if the subvolume references other subvolumes
1888  */
1889 static noinline int may_destroy_subvol(struct btrfs_root *root)
1890 {
1891         struct btrfs_fs_info *fs_info = root->fs_info;
1892         struct btrfs_path *path;
1893         struct btrfs_dir_item *di;
1894         struct btrfs_key key;
1895         u64 dir_id;
1896         int ret;
1897
1898         path = btrfs_alloc_path();
1899         if (!path)
1900                 return -ENOMEM;
1901
1902         /* Make sure this root isn't set as the default subvol */
1903         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1904         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1905                                    dir_id, "default", 7, 0);
1906         if (di && !IS_ERR(di)) {
1907                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1908                 if (key.objectid == root->root_key.objectid) {
1909                         ret = -EPERM;
1910                         btrfs_err(fs_info,
1911                                   "deleting default subvolume %llu is not allowed",
1912                                   key.objectid);
1913                         goto out;
1914                 }
1915                 btrfs_release_path(path);
1916         }
1917
1918         key.objectid = root->root_key.objectid;
1919         key.type = BTRFS_ROOT_REF_KEY;
1920         key.offset = (u64)-1;
1921
1922         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1923         if (ret < 0)
1924                 goto out;
1925         BUG_ON(ret == 0);
1926
1927         ret = 0;
1928         if (path->slots[0] > 0) {
1929                 path->slots[0]--;
1930                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1931                 if (key.objectid == root->root_key.objectid &&
1932                     key.type == BTRFS_ROOT_REF_KEY)
1933                         ret = -ENOTEMPTY;
1934         }
1935 out:
1936         btrfs_free_path(path);
1937         return ret;
1938 }
1939
1940 static noinline int key_in_sk(struct btrfs_key *key,
1941                               struct btrfs_ioctl_search_key *sk)
1942 {
1943         struct btrfs_key test;
1944         int ret;
1945
1946         test.objectid = sk->min_objectid;
1947         test.type = sk->min_type;
1948         test.offset = sk->min_offset;
1949
1950         ret = btrfs_comp_cpu_keys(key, &test);
1951         if (ret < 0)
1952                 return 0;
1953
1954         test.objectid = sk->max_objectid;
1955         test.type = sk->max_type;
1956         test.offset = sk->max_offset;
1957
1958         ret = btrfs_comp_cpu_keys(key, &test);
1959         if (ret > 0)
1960                 return 0;
1961         return 1;
1962 }
1963
1964 static noinline int copy_to_sk(struct btrfs_path *path,
1965                                struct btrfs_key *key,
1966                                struct btrfs_ioctl_search_key *sk,
1967                                size_t *buf_size,
1968                                char __user *ubuf,
1969                                unsigned long *sk_offset,
1970                                int *num_found)
1971 {
1972         u64 found_transid;
1973         struct extent_buffer *leaf;
1974         struct btrfs_ioctl_search_header sh;
1975         struct btrfs_key test;
1976         unsigned long item_off;
1977         unsigned long item_len;
1978         int nritems;
1979         int i;
1980         int slot;
1981         int ret = 0;
1982
1983         leaf = path->nodes[0];
1984         slot = path->slots[0];
1985         nritems = btrfs_header_nritems(leaf);
1986
1987         if (btrfs_header_generation(leaf) > sk->max_transid) {
1988                 i = nritems;
1989                 goto advance_key;
1990         }
1991         found_transid = btrfs_header_generation(leaf);
1992
1993         for (i = slot; i < nritems; i++) {
1994                 item_off = btrfs_item_ptr_offset(leaf, i);
1995                 item_len = btrfs_item_size_nr(leaf, i);
1996
1997                 btrfs_item_key_to_cpu(leaf, key, i);
1998                 if (!key_in_sk(key, sk))
1999                         continue;
2000
2001                 if (sizeof(sh) + item_len > *buf_size) {
2002                         if (*num_found) {
2003                                 ret = 1;
2004                                 goto out;
2005                         }
2006
2007                         /*
2008                          * return one empty item back for v1, which does not
2009                          * handle -EOVERFLOW
2010                          */
2011
2012                         *buf_size = sizeof(sh) + item_len;
2013                         item_len = 0;
2014                         ret = -EOVERFLOW;
2015                 }
2016
2017                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2018                         ret = 1;
2019                         goto out;
2020                 }
2021
2022                 sh.objectid = key->objectid;
2023                 sh.offset = key->offset;
2024                 sh.type = key->type;
2025                 sh.len = item_len;
2026                 sh.transid = found_transid;
2027
2028                 /* copy search result header */
2029                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2030                         ret = -EFAULT;
2031                         goto out;
2032                 }
2033
2034                 *sk_offset += sizeof(sh);
2035
2036                 if (item_len) {
2037                         char __user *up = ubuf + *sk_offset;
2038                         /* copy the item */
2039                         if (read_extent_buffer_to_user(leaf, up,
2040                                                        item_off, item_len)) {
2041                                 ret = -EFAULT;
2042                                 goto out;
2043                         }
2044
2045                         *sk_offset += item_len;
2046                 }
2047                 (*num_found)++;
2048
2049                 if (ret) /* -EOVERFLOW from above */
2050                         goto out;
2051
2052                 if (*num_found >= sk->nr_items) {
2053                         ret = 1;
2054                         goto out;
2055                 }
2056         }
2057 advance_key:
2058         ret = 0;
2059         test.objectid = sk->max_objectid;
2060         test.type = sk->max_type;
2061         test.offset = sk->max_offset;
2062         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2063                 ret = 1;
2064         else if (key->offset < (u64)-1)
2065                 key->offset++;
2066         else if (key->type < (u8)-1) {
2067                 key->offset = 0;
2068                 key->type++;
2069         } else if (key->objectid < (u64)-1) {
2070                 key->offset = 0;
2071                 key->type = 0;
2072                 key->objectid++;
2073         } else
2074                 ret = 1;
2075 out:
2076         /*
2077          *  0: all items from this leaf copied, continue with next
2078          *  1: * more items can be copied, but unused buffer is too small
2079          *     * all items were found
2080          *     Either way, it will stops the loop which iterates to the next
2081          *     leaf
2082          *  -EOVERFLOW: item was to large for buffer
2083          *  -EFAULT: could not copy extent buffer back to userspace
2084          */
2085         return ret;
2086 }
2087
2088 static noinline int search_ioctl(struct inode *inode,
2089                                  struct btrfs_ioctl_search_key *sk,
2090                                  size_t *buf_size,
2091                                  char __user *ubuf)
2092 {
2093         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2094         struct btrfs_root *root;
2095         struct btrfs_key key;
2096         struct btrfs_path *path;
2097         int ret;
2098         int num_found = 0;
2099         unsigned long sk_offset = 0;
2100
2101         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2102                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2103                 return -EOVERFLOW;
2104         }
2105
2106         path = btrfs_alloc_path();
2107         if (!path)
2108                 return -ENOMEM;
2109
2110         if (sk->tree_id == 0) {
2111                 /* search the root of the inode that was passed */
2112                 root = BTRFS_I(inode)->root;
2113         } else {
2114                 key.objectid = sk->tree_id;
2115                 key.type = BTRFS_ROOT_ITEM_KEY;
2116                 key.offset = (u64)-1;
2117                 root = btrfs_read_fs_root_no_name(info, &key);
2118                 if (IS_ERR(root)) {
2119                         btrfs_free_path(path);
2120                         return -ENOENT;
2121                 }
2122         }
2123
2124         key.objectid = sk->min_objectid;
2125         key.type = sk->min_type;
2126         key.offset = sk->min_offset;
2127
2128         while (1) {
2129                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2130                 if (ret != 0) {
2131                         if (ret > 0)
2132                                 ret = 0;
2133                         goto err;
2134                 }
2135                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2136                                  &sk_offset, &num_found);
2137                 btrfs_release_path(path);
2138                 if (ret)
2139                         break;
2140
2141         }
2142         if (ret > 0)
2143                 ret = 0;
2144 err:
2145         sk->nr_items = num_found;
2146         btrfs_free_path(path);
2147         return ret;
2148 }
2149
2150 static noinline int btrfs_ioctl_tree_search(struct file *file,
2151                                            void __user *argp)
2152 {
2153         struct btrfs_ioctl_search_args __user *uargs;
2154         struct btrfs_ioctl_search_key sk;
2155         struct inode *inode;
2156         int ret;
2157         size_t buf_size;
2158
2159         if (!capable(CAP_SYS_ADMIN))
2160                 return -EPERM;
2161
2162         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2163
2164         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2165                 return -EFAULT;
2166
2167         buf_size = sizeof(uargs->buf);
2168
2169         inode = file_inode(file);
2170         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2171
2172         /*
2173          * In the origin implementation an overflow is handled by returning a
2174          * search header with a len of zero, so reset ret.
2175          */
2176         if (ret == -EOVERFLOW)
2177                 ret = 0;
2178
2179         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2180                 ret = -EFAULT;
2181         return ret;
2182 }
2183
2184 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2185                                                void __user *argp)
2186 {
2187         struct btrfs_ioctl_search_args_v2 __user *uarg;
2188         struct btrfs_ioctl_search_args_v2 args;
2189         struct inode *inode;
2190         int ret;
2191         size_t buf_size;
2192         const size_t buf_limit = SZ_16M;
2193
2194         if (!capable(CAP_SYS_ADMIN))
2195                 return -EPERM;
2196
2197         /* copy search header and buffer size */
2198         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2199         if (copy_from_user(&args, uarg, sizeof(args)))
2200                 return -EFAULT;
2201
2202         buf_size = args.buf_size;
2203
2204         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2205                 return -EOVERFLOW;
2206
2207         /* limit result size to 16MB */
2208         if (buf_size > buf_limit)
2209                 buf_size = buf_limit;
2210
2211         inode = file_inode(file);
2212         ret = search_ioctl(inode, &args.key, &buf_size,
2213                            (char *)(&uarg->buf[0]));
2214         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2215                 ret = -EFAULT;
2216         else if (ret == -EOVERFLOW &&
2217                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2218                 ret = -EFAULT;
2219
2220         return ret;
2221 }
2222
2223 /*
2224  * Search INODE_REFs to identify path name of 'dirid' directory
2225  * in a 'tree_id' tree. and sets path name to 'name'.
2226  */
2227 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2228                                 u64 tree_id, u64 dirid, char *name)
2229 {
2230         struct btrfs_root *root;
2231         struct btrfs_key key;
2232         char *ptr;
2233         int ret = -1;
2234         int slot;
2235         int len;
2236         int total_len = 0;
2237         struct btrfs_inode_ref *iref;
2238         struct extent_buffer *l;
2239         struct btrfs_path *path;
2240
2241         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2242                 name[0]='\0';
2243                 return 0;
2244         }
2245
2246         path = btrfs_alloc_path();
2247         if (!path)
2248                 return -ENOMEM;
2249
2250         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2251
2252         key.objectid = tree_id;
2253         key.type = BTRFS_ROOT_ITEM_KEY;
2254         key.offset = (u64)-1;
2255         root = btrfs_read_fs_root_no_name(info, &key);
2256         if (IS_ERR(root)) {
2257                 btrfs_err(info, "could not find root %llu", tree_id);
2258                 ret = -ENOENT;
2259                 goto out;
2260         }
2261
2262         key.objectid = dirid;
2263         key.type = BTRFS_INODE_REF_KEY;
2264         key.offset = (u64)-1;
2265
2266         while (1) {
2267                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2268                 if (ret < 0)
2269                         goto out;
2270                 else if (ret > 0) {
2271                         ret = btrfs_previous_item(root, path, dirid,
2272                                                   BTRFS_INODE_REF_KEY);
2273                         if (ret < 0)
2274                                 goto out;
2275                         else if (ret > 0) {
2276                                 ret = -ENOENT;
2277                                 goto out;
2278                         }
2279                 }
2280
2281                 l = path->nodes[0];
2282                 slot = path->slots[0];
2283                 btrfs_item_key_to_cpu(l, &key, slot);
2284
2285                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2286                 len = btrfs_inode_ref_name_len(l, iref);
2287                 ptr -= len + 1;
2288                 total_len += len + 1;
2289                 if (ptr < name) {
2290                         ret = -ENAMETOOLONG;
2291                         goto out;
2292                 }
2293
2294                 *(ptr + len) = '/';
2295                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2296
2297                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2298                         break;
2299
2300                 btrfs_release_path(path);
2301                 key.objectid = key.offset;
2302                 key.offset = (u64)-1;
2303                 dirid = key.objectid;
2304         }
2305         memmove(name, ptr, total_len);
2306         name[total_len] = '\0';
2307         ret = 0;
2308 out:
2309         btrfs_free_path(path);
2310         return ret;
2311 }
2312
2313 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2314                                            void __user *argp)
2315 {
2316          struct btrfs_ioctl_ino_lookup_args *args;
2317          struct inode *inode;
2318         int ret = 0;
2319
2320         args = memdup_user(argp, sizeof(*args));
2321         if (IS_ERR(args))
2322                 return PTR_ERR(args);
2323
2324         inode = file_inode(file);
2325
2326         /*
2327          * Unprivileged query to obtain the containing subvolume root id. The
2328          * path is reset so it's consistent with btrfs_search_path_in_tree.
2329          */
2330         if (args->treeid == 0)
2331                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2332
2333         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2334                 args->name[0] = 0;
2335                 goto out;
2336         }
2337
2338         if (!capable(CAP_SYS_ADMIN)) {
2339                 ret = -EPERM;
2340                 goto out;
2341         }
2342
2343         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2344                                         args->treeid, args->objectid,
2345                                         args->name);
2346
2347 out:
2348         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2349                 ret = -EFAULT;
2350
2351         kfree(args);
2352         return ret;
2353 }
2354
2355 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2356                                              void __user *arg)
2357 {
2358         struct dentry *parent = file->f_path.dentry;
2359         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2360         struct dentry *dentry;
2361         struct inode *dir = d_inode(parent);
2362         struct inode *inode;
2363         struct btrfs_root *root = BTRFS_I(dir)->root;
2364         struct btrfs_root *dest = NULL;
2365         struct btrfs_ioctl_vol_args *vol_args;
2366         struct btrfs_trans_handle *trans;
2367         struct btrfs_block_rsv block_rsv;
2368         u64 root_flags;
2369         u64 qgroup_reserved;
2370         int namelen;
2371         int ret;
2372         int err = 0;
2373
2374         if (!S_ISDIR(dir->i_mode))
2375                 return -ENOTDIR;
2376
2377         vol_args = memdup_user(arg, sizeof(*vol_args));
2378         if (IS_ERR(vol_args))
2379                 return PTR_ERR(vol_args);
2380
2381         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2382         namelen = strlen(vol_args->name);
2383         if (strchr(vol_args->name, '/') ||
2384             strncmp(vol_args->name, "..", namelen) == 0) {
2385                 err = -EINVAL;
2386                 goto out;
2387         }
2388
2389         err = mnt_want_write_file(file);
2390         if (err)
2391                 goto out;
2392
2393
2394         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2395         if (err == -EINTR)
2396                 goto out_drop_write;
2397         dentry = lookup_one_len(vol_args->name, parent, namelen);
2398         if (IS_ERR(dentry)) {
2399                 err = PTR_ERR(dentry);
2400                 goto out_unlock_dir;
2401         }
2402
2403         if (d_really_is_negative(dentry)) {
2404                 err = -ENOENT;
2405                 goto out_dput;
2406         }
2407
2408         inode = d_inode(dentry);
2409         dest = BTRFS_I(inode)->root;
2410         if (!capable(CAP_SYS_ADMIN)) {
2411                 /*
2412                  * Regular user.  Only allow this with a special mount
2413                  * option, when the user has write+exec access to the
2414                  * subvol root, and when rmdir(2) would have been
2415                  * allowed.
2416                  *
2417                  * Note that this is _not_ check that the subvol is
2418                  * empty or doesn't contain data that we wouldn't
2419                  * otherwise be able to delete.
2420                  *
2421                  * Users who want to delete empty subvols should try
2422                  * rmdir(2).
2423                  */
2424                 err = -EPERM;
2425                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2426                         goto out_dput;
2427
2428                 /*
2429                  * Do not allow deletion if the parent dir is the same
2430                  * as the dir to be deleted.  That means the ioctl
2431                  * must be called on the dentry referencing the root
2432                  * of the subvol, not a random directory contained
2433                  * within it.
2434                  */
2435                 err = -EINVAL;
2436                 if (root == dest)
2437                         goto out_dput;
2438
2439                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2440                 if (err)
2441                         goto out_dput;
2442         }
2443
2444         /* check if subvolume may be deleted by a user */
2445         err = btrfs_may_delete(dir, dentry, 1);
2446         if (err)
2447                 goto out_dput;
2448
2449         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2450                 err = -EINVAL;
2451                 goto out_dput;
2452         }
2453
2454         inode_lock(inode);
2455
2456         /*
2457          * Don't allow to delete a subvolume with send in progress. This is
2458          * inside the i_mutex so the error handling that has to drop the bit
2459          * again is not run concurrently.
2460          */
2461         spin_lock(&dest->root_item_lock);
2462         root_flags = btrfs_root_flags(&dest->root_item);
2463         if (dest->send_in_progress == 0) {
2464                 btrfs_set_root_flags(&dest->root_item,
2465                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2466                 spin_unlock(&dest->root_item_lock);
2467         } else {
2468                 spin_unlock(&dest->root_item_lock);
2469                 btrfs_warn(fs_info,
2470                            "Attempt to delete subvolume %llu during send",
2471                            dest->root_key.objectid);
2472                 err = -EPERM;
2473                 goto out_unlock_inode;
2474         }
2475
2476         down_write(&fs_info->subvol_sem);
2477
2478         err = may_destroy_subvol(dest);
2479         if (err)
2480                 goto out_up_write;
2481
2482         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2483         /*
2484          * One for dir inode, two for dir entries, two for root
2485          * ref/backref.
2486          */
2487         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2488                                                5, &qgroup_reserved, true);
2489         if (err)
2490                 goto out_up_write;
2491
2492         trans = btrfs_start_transaction(root, 0);
2493         if (IS_ERR(trans)) {
2494                 err = PTR_ERR(trans);
2495                 goto out_release;
2496         }
2497         trans->block_rsv = &block_rsv;
2498         trans->bytes_reserved = block_rsv.size;
2499
2500         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2501
2502         ret = btrfs_unlink_subvol(trans, root, dir,
2503                                 dest->root_key.objectid,
2504                                 dentry->d_name.name,
2505                                 dentry->d_name.len);
2506         if (ret) {
2507                 err = ret;
2508                 btrfs_abort_transaction(trans, ret);
2509                 goto out_end_trans;
2510         }
2511
2512         btrfs_record_root_in_trans(trans, dest);
2513
2514         memset(&dest->root_item.drop_progress, 0,
2515                 sizeof(dest->root_item.drop_progress));
2516         dest->root_item.drop_level = 0;
2517         btrfs_set_root_refs(&dest->root_item, 0);
2518
2519         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2520                 ret = btrfs_insert_orphan_item(trans,
2521                                         fs_info->tree_root,
2522                                         dest->root_key.objectid);
2523                 if (ret) {
2524                         btrfs_abort_transaction(trans, ret);
2525                         err = ret;
2526                         goto out_end_trans;
2527                 }
2528         }
2529
2530         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2531                                   BTRFS_UUID_KEY_SUBVOL,
2532                                   dest->root_key.objectid);
2533         if (ret && ret != -ENOENT) {
2534                 btrfs_abort_transaction(trans, ret);
2535                 err = ret;
2536                 goto out_end_trans;
2537         }
2538         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2539                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2540                                           dest->root_item.received_uuid,
2541                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2542                                           dest->root_key.objectid);
2543                 if (ret && ret != -ENOENT) {
2544                         btrfs_abort_transaction(trans, ret);
2545                         err = ret;
2546                         goto out_end_trans;
2547                 }
2548         }
2549
2550 out_end_trans:
2551         trans->block_rsv = NULL;
2552         trans->bytes_reserved = 0;
2553         ret = btrfs_end_transaction(trans);
2554         if (ret && !err)
2555                 err = ret;
2556         inode->i_flags |= S_DEAD;
2557 out_release:
2558         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2559 out_up_write:
2560         up_write(&fs_info->subvol_sem);
2561         if (err) {
2562                 spin_lock(&dest->root_item_lock);
2563                 root_flags = btrfs_root_flags(&dest->root_item);
2564                 btrfs_set_root_flags(&dest->root_item,
2565                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2566                 spin_unlock(&dest->root_item_lock);
2567         }
2568 out_unlock_inode:
2569         inode_unlock(inode);
2570         if (!err) {
2571                 d_invalidate(dentry);
2572                 btrfs_invalidate_inodes(dest);
2573                 d_delete(dentry);
2574                 ASSERT(dest->send_in_progress == 0);
2575
2576                 /* the last ref */
2577                 if (dest->ino_cache_inode) {
2578                         iput(dest->ino_cache_inode);
2579                         dest->ino_cache_inode = NULL;
2580                 }
2581         }
2582 out_dput:
2583         dput(dentry);
2584 out_unlock_dir:
2585         inode_unlock(dir);
2586 out_drop_write:
2587         mnt_drop_write_file(file);
2588 out:
2589         kfree(vol_args);
2590         return err;
2591 }
2592
2593 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2594 {
2595         struct inode *inode = file_inode(file);
2596         struct btrfs_root *root = BTRFS_I(inode)->root;
2597         struct btrfs_ioctl_defrag_range_args *range;
2598         int ret;
2599
2600         ret = mnt_want_write_file(file);
2601         if (ret)
2602                 return ret;
2603
2604         if (btrfs_root_readonly(root)) {
2605                 ret = -EROFS;
2606                 goto out;
2607         }
2608
2609         switch (inode->i_mode & S_IFMT) {
2610         case S_IFDIR:
2611                 if (!capable(CAP_SYS_ADMIN)) {
2612                         ret = -EPERM;
2613                         goto out;
2614                 }
2615                 ret = btrfs_defrag_root(root);
2616                 break;
2617         case S_IFREG:
2618                 if (!(file->f_mode & FMODE_WRITE)) {
2619                         ret = -EINVAL;
2620                         goto out;
2621                 }
2622
2623                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2624                 if (!range) {
2625                         ret = -ENOMEM;
2626                         goto out;
2627                 }
2628
2629                 if (argp) {
2630                         if (copy_from_user(range, argp,
2631                                            sizeof(*range))) {
2632                                 ret = -EFAULT;
2633                                 kfree(range);
2634                                 goto out;
2635                         }
2636                         /* compression requires us to start the IO */
2637                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2638                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2639                                 range->extent_thresh = (u32)-1;
2640                         }
2641                 } else {
2642                         /* the rest are all set to zero by kzalloc */
2643                         range->len = (u64)-1;
2644                 }
2645                 ret = btrfs_defrag_file(file_inode(file), file,
2646                                         range, 0, 0);
2647                 if (ret > 0)
2648                         ret = 0;
2649                 kfree(range);
2650                 break;
2651         default:
2652                 ret = -EINVAL;
2653         }
2654 out:
2655         mnt_drop_write_file(file);
2656         return ret;
2657 }
2658
2659 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2660 {
2661         struct btrfs_ioctl_vol_args *vol_args;
2662         int ret;
2663
2664         if (!capable(CAP_SYS_ADMIN))
2665                 return -EPERM;
2666
2667         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2668                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2669
2670         mutex_lock(&fs_info->volume_mutex);
2671         vol_args = memdup_user(arg, sizeof(*vol_args));
2672         if (IS_ERR(vol_args)) {
2673                 ret = PTR_ERR(vol_args);
2674                 goto out;
2675         }
2676
2677         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2678         ret = btrfs_init_new_device(fs_info, vol_args->name);
2679
2680         if (!ret)
2681                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2682
2683         kfree(vol_args);
2684 out:
2685         mutex_unlock(&fs_info->volume_mutex);
2686         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2687         return ret;
2688 }
2689
2690 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2691 {
2692         struct inode *inode = file_inode(file);
2693         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2694         struct btrfs_ioctl_vol_args_v2 *vol_args;
2695         int ret;
2696
2697         if (!capable(CAP_SYS_ADMIN))
2698                 return -EPERM;
2699
2700         ret = mnt_want_write_file(file);
2701         if (ret)
2702                 return ret;
2703
2704         vol_args = memdup_user(arg, sizeof(*vol_args));
2705         if (IS_ERR(vol_args)) {
2706                 ret = PTR_ERR(vol_args);
2707                 goto err_drop;
2708         }
2709
2710         /* Check for compatibility reject unknown flags */
2711         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2712                 return -EOPNOTSUPP;
2713
2714         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2715                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2716                 goto out;
2717         }
2718
2719         mutex_lock(&fs_info->volume_mutex);
2720         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2721                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2722         } else {
2723                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2724                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2725         }
2726         mutex_unlock(&fs_info->volume_mutex);
2727         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2728
2729         if (!ret) {
2730                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2731                         btrfs_info(fs_info, "device deleted: id %llu",
2732                                         vol_args->devid);
2733                 else
2734                         btrfs_info(fs_info, "device deleted: %s",
2735                                         vol_args->name);
2736         }
2737 out:
2738         kfree(vol_args);
2739 err_drop:
2740         mnt_drop_write_file(file);
2741         return ret;
2742 }
2743
2744 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2745 {
2746         struct inode *inode = file_inode(file);
2747         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2748         struct btrfs_ioctl_vol_args *vol_args;
2749         int ret;
2750
2751         if (!capable(CAP_SYS_ADMIN))
2752                 return -EPERM;
2753
2754         ret = mnt_want_write_file(file);
2755         if (ret)
2756                 return ret;
2757
2758         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2759                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2760                 goto out_drop_write;
2761         }
2762
2763         vol_args = memdup_user(arg, sizeof(*vol_args));
2764         if (IS_ERR(vol_args)) {
2765                 ret = PTR_ERR(vol_args);
2766                 goto out;
2767         }
2768
2769         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2770         mutex_lock(&fs_info->volume_mutex);
2771         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2772         mutex_unlock(&fs_info->volume_mutex);
2773
2774         if (!ret)
2775                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2776         kfree(vol_args);
2777 out:
2778         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2779 out_drop_write:
2780         mnt_drop_write_file(file);
2781
2782         return ret;
2783 }
2784
2785 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2786                                 void __user *arg)
2787 {
2788         struct btrfs_ioctl_fs_info_args *fi_args;
2789         struct btrfs_device *device;
2790         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2791         int ret = 0;
2792
2793         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2794         if (!fi_args)
2795                 return -ENOMEM;
2796
2797         mutex_lock(&fs_devices->device_list_mutex);
2798         fi_args->num_devices = fs_devices->num_devices;
2799         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2800
2801         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2802                 if (device->devid > fi_args->max_id)
2803                         fi_args->max_id = device->devid;
2804         }
2805         mutex_unlock(&fs_devices->device_list_mutex);
2806
2807         fi_args->nodesize = fs_info->super_copy->nodesize;
2808         fi_args->sectorsize = fs_info->super_copy->sectorsize;
2809         fi_args->clone_alignment = fs_info->super_copy->sectorsize;
2810
2811         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2812                 ret = -EFAULT;
2813
2814         kfree(fi_args);
2815         return ret;
2816 }
2817
2818 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2819                                  void __user *arg)
2820 {
2821         struct btrfs_ioctl_dev_info_args *di_args;
2822         struct btrfs_device *dev;
2823         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2824         int ret = 0;
2825         char *s_uuid = NULL;
2826
2827         di_args = memdup_user(arg, sizeof(*di_args));
2828         if (IS_ERR(di_args))
2829                 return PTR_ERR(di_args);
2830
2831         if (!btrfs_is_empty_uuid(di_args->uuid))
2832                 s_uuid = di_args->uuid;
2833
2834         mutex_lock(&fs_devices->device_list_mutex);
2835         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2836
2837         if (!dev) {
2838                 ret = -ENODEV;
2839                 goto out;
2840         }
2841
2842         di_args->devid = dev->devid;
2843         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2844         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2845         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2846         if (dev->name) {
2847                 struct rcu_string *name;
2848
2849                 rcu_read_lock();
2850                 name = rcu_dereference(dev->name);
2851                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2852                 rcu_read_unlock();
2853                 di_args->path[sizeof(di_args->path) - 1] = 0;
2854         } else {
2855                 di_args->path[0] = '\0';
2856         }
2857
2858 out:
2859         mutex_unlock(&fs_devices->device_list_mutex);
2860         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2861                 ret = -EFAULT;
2862
2863         kfree(di_args);
2864         return ret;
2865 }
2866
2867 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2868 {
2869         struct page *page;
2870
2871         page = grab_cache_page(inode->i_mapping, index);
2872         if (!page)
2873                 return ERR_PTR(-ENOMEM);
2874
2875         if (!PageUptodate(page)) {
2876                 int ret;
2877
2878                 ret = btrfs_readpage(NULL, page);
2879                 if (ret)
2880                         return ERR_PTR(ret);
2881                 lock_page(page);
2882                 if (!PageUptodate(page)) {
2883                         unlock_page(page);
2884                         put_page(page);
2885                         return ERR_PTR(-EIO);
2886                 }
2887                 if (page->mapping != inode->i_mapping) {
2888                         unlock_page(page);
2889                         put_page(page);
2890                         return ERR_PTR(-EAGAIN);
2891                 }
2892         }
2893
2894         return page;
2895 }
2896
2897 static int gather_extent_pages(struct inode *inode, struct page **pages,
2898                                int num_pages, u64 off)
2899 {
2900         int i;
2901         pgoff_t index = off >> PAGE_SHIFT;
2902
2903         for (i = 0; i < num_pages; i++) {
2904 again:
2905                 pages[i] = extent_same_get_page(inode, index + i);
2906                 if (IS_ERR(pages[i])) {
2907                         int err = PTR_ERR(pages[i]);
2908
2909                         if (err == -EAGAIN)
2910                                 goto again;
2911                         pages[i] = NULL;
2912                         return err;
2913                 }
2914         }
2915         return 0;
2916 }
2917
2918 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2919                              bool retry_range_locking)
2920 {
2921         /*
2922          * Do any pending delalloc/csum calculations on inode, one way or
2923          * another, and lock file content.
2924          * The locking order is:
2925          *
2926          *   1) pages
2927          *   2) range in the inode's io tree
2928          */
2929         while (1) {
2930                 struct btrfs_ordered_extent *ordered;
2931                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2932                 ordered = btrfs_lookup_first_ordered_extent(inode,
2933                                                             off + len - 1);
2934                 if ((!ordered ||
2935                      ordered->file_offset + ordered->len <= off ||
2936                      ordered->file_offset >= off + len) &&
2937                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2938                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2939                         if (ordered)
2940                                 btrfs_put_ordered_extent(ordered);
2941                         break;
2942                 }
2943                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2944                 if (ordered)
2945                         btrfs_put_ordered_extent(ordered);
2946                 if (!retry_range_locking)
2947                         return -EAGAIN;
2948                 btrfs_wait_ordered_range(inode, off, len);
2949         }
2950         return 0;
2951 }
2952
2953 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2954 {
2955         inode_unlock(inode1);
2956         inode_unlock(inode2);
2957 }
2958
2959 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2960 {
2961         if (inode1 < inode2)
2962                 swap(inode1, inode2);
2963
2964         inode_lock_nested(inode1, I_MUTEX_PARENT);
2965         inode_lock_nested(inode2, I_MUTEX_CHILD);
2966 }
2967
2968 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2969                                       struct inode *inode2, u64 loff2, u64 len)
2970 {
2971         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2972         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2973 }
2974
2975 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2976                                     struct inode *inode2, u64 loff2, u64 len,
2977                                     bool retry_range_locking)
2978 {
2979         int ret;
2980
2981         if (inode1 < inode2) {
2982                 swap(inode1, inode2);
2983                 swap(loff1, loff2);
2984         }
2985         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2986         if (ret)
2987                 return ret;
2988         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2989         if (ret)
2990                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2991                               loff1 + len - 1);
2992         return ret;
2993 }
2994
2995 struct cmp_pages {
2996         int             num_pages;
2997         struct page     **src_pages;
2998         struct page     **dst_pages;
2999 };
3000
3001 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3002 {
3003         int i;
3004         struct page *pg;
3005
3006         for (i = 0; i < cmp->num_pages; i++) {
3007                 pg = cmp->src_pages[i];
3008                 if (pg) {
3009                         unlock_page(pg);
3010                         put_page(pg);
3011                 }
3012                 pg = cmp->dst_pages[i];
3013                 if (pg) {
3014                         unlock_page(pg);
3015                         put_page(pg);
3016                 }
3017         }
3018         kfree(cmp->src_pages);
3019         kfree(cmp->dst_pages);
3020 }
3021
3022 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3023                                   struct inode *dst, u64 dst_loff,
3024                                   u64 len, struct cmp_pages *cmp)
3025 {
3026         int ret;
3027         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3028         struct page **src_pgarr, **dst_pgarr;
3029
3030         /*
3031          * We must gather up all the pages before we initiate our
3032          * extent locking. We use an array for the page pointers. Size
3033          * of the array is bounded by len, which is in turn bounded by
3034          * BTRFS_MAX_DEDUPE_LEN.
3035          */
3036         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3037         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3038         if (!src_pgarr || !dst_pgarr) {
3039                 kfree(src_pgarr);
3040                 kfree(dst_pgarr);
3041                 return -ENOMEM;
3042         }
3043         cmp->num_pages = num_pages;
3044         cmp->src_pages = src_pgarr;
3045         cmp->dst_pages = dst_pgarr;
3046
3047         /*
3048          * If deduping ranges in the same inode, locking rules make it mandatory
3049          * to always lock pages in ascending order to avoid deadlocks with
3050          * concurrent tasks (such as starting writeback/delalloc).
3051          */
3052         if (src == dst && dst_loff < loff) {
3053                 swap(src_pgarr, dst_pgarr);
3054                 swap(loff, dst_loff);
3055         }
3056
3057         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3058         if (ret)
3059                 goto out;
3060
3061         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3062
3063 out:
3064         if (ret)
3065                 btrfs_cmp_data_free(cmp);
3066         return 0;
3067 }
3068
3069 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3070 {
3071         int ret = 0;
3072         int i;
3073         struct page *src_page, *dst_page;
3074         unsigned int cmp_len = PAGE_SIZE;
3075         void *addr, *dst_addr;
3076
3077         i = 0;
3078         while (len) {
3079                 if (len < PAGE_SIZE)
3080                         cmp_len = len;
3081
3082                 BUG_ON(i >= cmp->num_pages);
3083
3084                 src_page = cmp->src_pages[i];
3085                 dst_page = cmp->dst_pages[i];
3086                 ASSERT(PageLocked(src_page));
3087                 ASSERT(PageLocked(dst_page));
3088
3089                 addr = kmap_atomic(src_page);
3090                 dst_addr = kmap_atomic(dst_page);
3091
3092                 flush_dcache_page(src_page);
3093                 flush_dcache_page(dst_page);
3094
3095                 if (memcmp(addr, dst_addr, cmp_len))
3096                         ret = -EBADE;
3097
3098                 kunmap_atomic(addr);
3099                 kunmap_atomic(dst_addr);
3100
3101                 if (ret)
3102                         break;
3103
3104                 len -= cmp_len;
3105                 i++;
3106         }
3107
3108         return ret;
3109 }
3110
3111 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3112                                      u64 olen)
3113 {
3114         u64 len = *plen;
3115         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3116
3117         if (off + olen > inode->i_size || off + olen < off)
3118                 return -EINVAL;
3119
3120         /* if we extend to eof, continue to block boundary */
3121         if (off + len == inode->i_size)
3122                 *plen = len = ALIGN(inode->i_size, bs) - off;
3123
3124         /* Check that we are block aligned - btrfs_clone() requires this */
3125         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3126                 return -EINVAL;
3127
3128         return 0;
3129 }
3130
3131 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3132                              struct inode *dst, u64 dst_loff)
3133 {
3134         int ret;
3135         u64 len = olen;
3136         struct cmp_pages cmp;
3137         bool same_inode = (src == dst);
3138         u64 same_lock_start = 0;
3139         u64 same_lock_len = 0;
3140
3141         if (len == 0)
3142                 return 0;
3143
3144         if (same_inode)
3145                 inode_lock(src);
3146         else
3147                 btrfs_double_inode_lock(src, dst);
3148
3149         ret = extent_same_check_offsets(src, loff, &len, olen);
3150         if (ret)
3151                 goto out_unlock;
3152
3153         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3154         if (ret)
3155                 goto out_unlock;
3156
3157         if (same_inode) {
3158                 /*
3159                  * Single inode case wants the same checks, except we
3160                  * don't want our length pushed out past i_size as
3161                  * comparing that data range makes no sense.
3162                  *
3163                  * extent_same_check_offsets() will do this for an
3164                  * unaligned length at i_size, so catch it here and
3165                  * reject the request.
3166                  *
3167                  * This effectively means we require aligned extents
3168                  * for the single-inode case, whereas the other cases
3169                  * allow an unaligned length so long as it ends at
3170                  * i_size.
3171                  */
3172                 if (len != olen) {
3173                         ret = -EINVAL;
3174                         goto out_unlock;
3175                 }
3176
3177                 /* Check for overlapping ranges */
3178                 if (dst_loff + len > loff && dst_loff < loff + len) {
3179                         ret = -EINVAL;
3180                         goto out_unlock;
3181                 }
3182
3183                 same_lock_start = min_t(u64, loff, dst_loff);
3184                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3185         }
3186
3187         /* don't make the dst file partly checksummed */
3188         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3189             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3190                 ret = -EINVAL;
3191                 goto out_unlock;
3192         }
3193
3194 again:
3195         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3196         if (ret)
3197                 goto out_unlock;
3198
3199         if (same_inode)
3200                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3201                                         false);
3202         else
3203                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3204                                                false);
3205         /*
3206          * If one of the inodes has dirty pages in the respective range or
3207          * ordered extents, we need to flush dellaloc and wait for all ordered
3208          * extents in the range. We must unlock the pages and the ranges in the
3209          * io trees to avoid deadlocks when flushing delalloc (requires locking
3210          * pages) and when waiting for ordered extents to complete (they require
3211          * range locking).
3212          */
3213         if (ret == -EAGAIN) {
3214                 /*
3215                  * Ranges in the io trees already unlocked. Now unlock all
3216                  * pages before waiting for all IO to complete.
3217                  */
3218                 btrfs_cmp_data_free(&cmp);
3219                 if (same_inode) {
3220                         btrfs_wait_ordered_range(src, same_lock_start,
3221                                                  same_lock_len);
3222                 } else {
3223                         btrfs_wait_ordered_range(src, loff, len);
3224                         btrfs_wait_ordered_range(dst, dst_loff, len);
3225                 }
3226                 goto again;
3227         }
3228         ASSERT(ret == 0);
3229         if (WARN_ON(ret)) {
3230                 /* ranges in the io trees already unlocked */
3231                 btrfs_cmp_data_free(&cmp);
3232                 return ret;
3233         }
3234
3235         /* pass original length for comparison so we stay within i_size */
3236         ret = btrfs_cmp_data(olen, &cmp);
3237         if (ret == 0)
3238                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3239
3240         if (same_inode)
3241                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3242                               same_lock_start + same_lock_len - 1);
3243         else
3244                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3245
3246         btrfs_cmp_data_free(&cmp);
3247 out_unlock:
3248         if (same_inode)
3249                 inode_unlock(src);
3250         else
3251                 btrfs_double_inode_unlock(src, dst);
3252
3253         return ret;
3254 }
3255
3256 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3257
3258 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3259                                 struct file *dst_file, u64 dst_loff)
3260 {
3261         struct inode *src = file_inode(src_file);
3262         struct inode *dst = file_inode(dst_file);
3263         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3264         ssize_t res;