Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[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/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include <linux/uuid.h>
45 #include <linux/btrfs.h>
46 #include <linux/uaccess.h>
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "volumes.h"
53 #include "locking.h"
54 #include "inode-map.h"
55 #include "backref.h"
56 #include "rcu-string.h"
57 #include "send.h"
58 #include "dev-replace.h"
59 #include "props.h"
60 #include "sysfs.h"
61 #include "qgroup.h"
62
63 #ifdef CONFIG_64BIT
64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
65  * structures are incorrect, as the timespec structure from userspace
66  * is 4 bytes too small. We define these alternatives here to teach
67  * the kernel about the 32-bit struct packing.
68  */
69 struct btrfs_ioctl_timespec_32 {
70         __u64 sec;
71         __u32 nsec;
72 } __attribute__ ((__packed__));
73
74 struct btrfs_ioctl_received_subvol_args_32 {
75         char    uuid[BTRFS_UUID_SIZE];  /* in */
76         __u64   stransid;               /* in */
77         __u64   rtransid;               /* out */
78         struct btrfs_ioctl_timespec_32 stime; /* in */
79         struct btrfs_ioctl_timespec_32 rtime; /* out */
80         __u64   flags;                  /* in */
81         __u64   reserved[16];           /* in */
82 } __attribute__ ((__packed__));
83
84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
85                                 struct btrfs_ioctl_received_subvol_args_32)
86 #endif
87
88
89 static int btrfs_clone(struct inode *src, struct inode *inode,
90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
91
92 /* Mask out flags that are inappropriate for the given type of inode. */
93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
94 {
95         if (S_ISDIR(mode))
96                 return flags;
97         else if (S_ISREG(mode))
98                 return flags & ~FS_DIRSYNC_FL;
99         else
100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
101 }
102
103 /*
104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
105  */
106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
107 {
108         unsigned int iflags = 0;
109
110         if (flags & BTRFS_INODE_SYNC)
111                 iflags |= FS_SYNC_FL;
112         if (flags & BTRFS_INODE_IMMUTABLE)
113                 iflags |= FS_IMMUTABLE_FL;
114         if (flags & BTRFS_INODE_APPEND)
115                 iflags |= FS_APPEND_FL;
116         if (flags & BTRFS_INODE_NODUMP)
117                 iflags |= FS_NODUMP_FL;
118         if (flags & BTRFS_INODE_NOATIME)
119                 iflags |= FS_NOATIME_FL;
120         if (flags & BTRFS_INODE_DIRSYNC)
121                 iflags |= FS_DIRSYNC_FL;
122         if (flags & BTRFS_INODE_NODATACOW)
123                 iflags |= FS_NOCOW_FL;
124
125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
126                 iflags |= FS_COMPR_FL;
127         else if (flags & BTRFS_INODE_NOCOMPRESS)
128                 iflags |= FS_NOCOMP_FL;
129
130         return iflags;
131 }
132
133 /*
134  * Update inode->i_flags based on the btrfs internal flags.
135  */
136 void btrfs_update_iflags(struct inode *inode)
137 {
138         struct btrfs_inode *ip = BTRFS_I(inode);
139         unsigned int new_fl = 0;
140
141         if (ip->flags & BTRFS_INODE_SYNC)
142                 new_fl |= S_SYNC;
143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
144                 new_fl |= S_IMMUTABLE;
145         if (ip->flags & BTRFS_INODE_APPEND)
146                 new_fl |= S_APPEND;
147         if (ip->flags & BTRFS_INODE_NOATIME)
148                 new_fl |= S_NOATIME;
149         if (ip->flags & BTRFS_INODE_DIRSYNC)
150                 new_fl |= S_DIRSYNC;
151
152         set_mask_bits(&inode->i_flags,
153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
154                       new_fl);
155 }
156
157 /*
158  * Inherit flags from the parent inode.
159  *
160  * Currently only the compression flags and the cow flags are inherited.
161  */
162 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
163 {
164         unsigned int flags;
165
166         if (!dir)
167                 return;
168
169         flags = BTRFS_I(dir)->flags;
170
171         if (flags & BTRFS_INODE_NOCOMPRESS) {
172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
173                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
174         } else if (flags & BTRFS_INODE_COMPRESS) {
175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
176                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
177         }
178
179         if (flags & BTRFS_INODE_NODATACOW) {
180                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
181                 if (S_ISREG(inode->i_mode))
182                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
183         }
184
185         btrfs_update_iflags(inode);
186 }
187
188 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
189 {
190         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
191         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
192
193         if (copy_to_user(arg, &flags, sizeof(flags)))
194                 return -EFAULT;
195         return 0;
196 }
197
198 static int check_flags(unsigned int flags)
199 {
200         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
201                       FS_NOATIME_FL | FS_NODUMP_FL | \
202                       FS_SYNC_FL | FS_DIRSYNC_FL | \
203                       FS_NOCOMP_FL | FS_COMPR_FL |
204                       FS_NOCOW_FL))
205                 return -EOPNOTSUPP;
206
207         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
208                 return -EINVAL;
209
210         return 0;
211 }
212
213 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
214 {
215         struct inode *inode = file_inode(file);
216         struct btrfs_inode *ip = BTRFS_I(inode);
217         struct btrfs_root *root = ip->root;
218         struct btrfs_trans_handle *trans;
219         unsigned int flags, oldflags;
220         int ret;
221         u64 ip_oldflags;
222         unsigned int i_oldflags;
223         umode_t mode;
224
225         if (!inode_owner_or_capable(inode))
226                 return -EPERM;
227
228         if (btrfs_root_readonly(root))
229                 return -EROFS;
230
231         if (copy_from_user(&flags, arg, sizeof(flags)))
232                 return -EFAULT;
233
234         ret = check_flags(flags);
235         if (ret)
236                 return ret;
237
238         ret = mnt_want_write_file(file);
239         if (ret)
240                 return ret;
241
242         mutex_lock(&inode->i_mutex);
243
244         ip_oldflags = ip->flags;
245         i_oldflags = inode->i_flags;
246         mode = inode->i_mode;
247
248         flags = btrfs_mask_flags(inode->i_mode, flags);
249         oldflags = btrfs_flags_to_ioctl(ip->flags);
250         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
251                 if (!capable(CAP_LINUX_IMMUTABLE)) {
252                         ret = -EPERM;
253                         goto out_unlock;
254                 }
255         }
256
257         if (flags & FS_SYNC_FL)
258                 ip->flags |= BTRFS_INODE_SYNC;
259         else
260                 ip->flags &= ~BTRFS_INODE_SYNC;
261         if (flags & FS_IMMUTABLE_FL)
262                 ip->flags |= BTRFS_INODE_IMMUTABLE;
263         else
264                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
265         if (flags & FS_APPEND_FL)
266                 ip->flags |= BTRFS_INODE_APPEND;
267         else
268                 ip->flags &= ~BTRFS_INODE_APPEND;
269         if (flags & FS_NODUMP_FL)
270                 ip->flags |= BTRFS_INODE_NODUMP;
271         else
272                 ip->flags &= ~BTRFS_INODE_NODUMP;
273         if (flags & FS_NOATIME_FL)
274                 ip->flags |= BTRFS_INODE_NOATIME;
275         else
276                 ip->flags &= ~BTRFS_INODE_NOATIME;
277         if (flags & FS_DIRSYNC_FL)
278                 ip->flags |= BTRFS_INODE_DIRSYNC;
279         else
280                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
281         if (flags & FS_NOCOW_FL) {
282                 if (S_ISREG(mode)) {
283                         /*
284                          * It's safe to turn csums off here, no extents exist.
285                          * Otherwise we want the flag to reflect the real COW
286                          * status of the file and will not set it.
287                          */
288                         if (inode->i_size == 0)
289                                 ip->flags |= BTRFS_INODE_NODATACOW
290                                            | BTRFS_INODE_NODATASUM;
291                 } else {
292                         ip->flags |= BTRFS_INODE_NODATACOW;
293                 }
294         } else {
295                 /*
296                  * Revert back under same assuptions as above
297                  */
298                 if (S_ISREG(mode)) {
299                         if (inode->i_size == 0)
300                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
301                                              | BTRFS_INODE_NODATASUM);
302                 } else {
303                         ip->flags &= ~BTRFS_INODE_NODATACOW;
304                 }
305         }
306
307         /*
308          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
309          * flag may be changed automatically if compression code won't make
310          * things smaller.
311          */
312         if (flags & FS_NOCOMP_FL) {
313                 ip->flags &= ~BTRFS_INODE_COMPRESS;
314                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
315
316                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
317                 if (ret && ret != -ENODATA)
318                         goto out_drop;
319         } else if (flags & FS_COMPR_FL) {
320                 const char *comp;
321
322                 ip->flags |= BTRFS_INODE_COMPRESS;
323                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
324
325                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
326                         comp = "lzo";
327                 else
328                         comp = "zlib";
329                 ret = btrfs_set_prop(inode, "btrfs.compression",
330                                      comp, strlen(comp), 0);
331                 if (ret)
332                         goto out_drop;
333
334         } else {
335                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
336                 if (ret && ret != -ENODATA)
337                         goto out_drop;
338                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
339         }
340
341         trans = btrfs_start_transaction(root, 1);
342         if (IS_ERR(trans)) {
343                 ret = PTR_ERR(trans);
344                 goto out_drop;
345         }
346
347         btrfs_update_iflags(inode);
348         inode_inc_iversion(inode);
349         inode->i_ctime = CURRENT_TIME;
350         ret = btrfs_update_inode(trans, root, inode);
351
352         btrfs_end_transaction(trans, root);
353  out_drop:
354         if (ret) {
355                 ip->flags = ip_oldflags;
356                 inode->i_flags = i_oldflags;
357         }
358
359  out_unlock:
360         mutex_unlock(&inode->i_mutex);
361         mnt_drop_write_file(file);
362         return ret;
363 }
364
365 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
366 {
367         struct inode *inode = file_inode(file);
368
369         return put_user(inode->i_generation, arg);
370 }
371
372 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
373 {
374         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
375         struct btrfs_device *device;
376         struct request_queue *q;
377         struct fstrim_range range;
378         u64 minlen = ULLONG_MAX;
379         u64 num_devices = 0;
380         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
381         int ret;
382
383         if (!capable(CAP_SYS_ADMIN))
384                 return -EPERM;
385
386         rcu_read_lock();
387         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
388                                 dev_list) {
389                 if (!device->bdev)
390                         continue;
391                 q = bdev_get_queue(device->bdev);
392                 if (blk_queue_discard(q)) {
393                         num_devices++;
394                         minlen = min((u64)q->limits.discard_granularity,
395                                      minlen);
396                 }
397         }
398         rcu_read_unlock();
399
400         if (!num_devices)
401                 return -EOPNOTSUPP;
402         if (copy_from_user(&range, arg, sizeof(range)))
403                 return -EFAULT;
404         if (range.start > total_bytes ||
405             range.len < fs_info->sb->s_blocksize)
406                 return -EINVAL;
407
408         range.len = min(range.len, total_bytes - range.start);
409         range.minlen = max(range.minlen, minlen);
410         ret = btrfs_trim_fs(fs_info->tree_root, &range);
411         if (ret < 0)
412                 return ret;
413
414         if (copy_to_user(arg, &range, sizeof(range)))
415                 return -EFAULT;
416
417         return 0;
418 }
419
420 int btrfs_is_empty_uuid(u8 *uuid)
421 {
422         int i;
423
424         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
425                 if (uuid[i])
426                         return 0;
427         }
428         return 1;
429 }
430
431 static noinline int create_subvol(struct inode *dir,
432                                   struct dentry *dentry,
433                                   char *name, int namelen,
434                                   u64 *async_transid,
435                                   struct btrfs_qgroup_inherit *inherit)
436 {
437         struct btrfs_trans_handle *trans;
438         struct btrfs_key key;
439         struct btrfs_root_item root_item;
440         struct btrfs_inode_item *inode_item;
441         struct extent_buffer *leaf;
442         struct btrfs_root *root = BTRFS_I(dir)->root;
443         struct btrfs_root *new_root;
444         struct btrfs_block_rsv block_rsv;
445         struct timespec cur_time = CURRENT_TIME;
446         struct inode *inode;
447         int ret;
448         int err;
449         u64 objectid;
450         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
451         u64 index = 0;
452         u64 qgroup_reserved;
453         uuid_le new_uuid;
454
455         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
456         if (ret)
457                 return ret;
458
459         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
460         /*
461          * The same as the snapshot creation, please see the comment
462          * of create_snapshot().
463          */
464         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
465                                                8, &qgroup_reserved, false);
466         if (ret)
467                 return ret;
468
469         trans = btrfs_start_transaction(root, 0);
470         if (IS_ERR(trans)) {
471                 ret = PTR_ERR(trans);
472                 btrfs_subvolume_release_metadata(root, &block_rsv,
473                                                  qgroup_reserved);
474                 return ret;
475         }
476         trans->block_rsv = &block_rsv;
477         trans->bytes_reserved = block_rsv.size;
478
479         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
480         if (ret)
481                 goto fail;
482
483         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
484         if (IS_ERR(leaf)) {
485                 ret = PTR_ERR(leaf);
486                 goto fail;
487         }
488
489         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
490         btrfs_set_header_bytenr(leaf, leaf->start);
491         btrfs_set_header_generation(leaf, trans->transid);
492         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
493         btrfs_set_header_owner(leaf, objectid);
494
495         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
496                             BTRFS_FSID_SIZE);
497         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
498                             btrfs_header_chunk_tree_uuid(leaf),
499                             BTRFS_UUID_SIZE);
500         btrfs_mark_buffer_dirty(leaf);
501
502         memset(&root_item, 0, sizeof(root_item));
503
504         inode_item = &root_item.inode;
505         btrfs_set_stack_inode_generation(inode_item, 1);
506         btrfs_set_stack_inode_size(inode_item, 3);
507         btrfs_set_stack_inode_nlink(inode_item, 1);
508         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
509         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
510
511         btrfs_set_root_flags(&root_item, 0);
512         btrfs_set_root_limit(&root_item, 0);
513         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
514
515         btrfs_set_root_bytenr(&root_item, leaf->start);
516         btrfs_set_root_generation(&root_item, trans->transid);
517         btrfs_set_root_level(&root_item, 0);
518         btrfs_set_root_refs(&root_item, 1);
519         btrfs_set_root_used(&root_item, leaf->len);
520         btrfs_set_root_last_snapshot(&root_item, 0);
521
522         btrfs_set_root_generation_v2(&root_item,
523                         btrfs_root_generation(&root_item));
524         uuid_le_gen(&new_uuid);
525         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
526         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
527         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
528         root_item.ctime = root_item.otime;
529         btrfs_set_root_ctransid(&root_item, trans->transid);
530         btrfs_set_root_otransid(&root_item, trans->transid);
531
532         btrfs_tree_unlock(leaf);
533         free_extent_buffer(leaf);
534         leaf = NULL;
535
536         btrfs_set_root_dirid(&root_item, new_dirid);
537
538         key.objectid = objectid;
539         key.offset = 0;
540         key.type = BTRFS_ROOT_ITEM_KEY;
541         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
542                                 &root_item);
543         if (ret)
544                 goto fail;
545
546         key.offset = (u64)-1;
547         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
548         if (IS_ERR(new_root)) {
549                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
550                 ret = PTR_ERR(new_root);
551                 goto fail;
552         }
553
554         btrfs_record_root_in_trans(trans, new_root);
555
556         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
557         if (ret) {
558                 /* We potentially lose an unused inode item here */
559                 btrfs_abort_transaction(trans, root, ret);
560                 goto fail;
561         }
562
563         /*
564          * insert the directory item
565          */
566         ret = btrfs_set_inode_index(dir, &index);
567         if (ret) {
568                 btrfs_abort_transaction(trans, root, ret);
569                 goto fail;
570         }
571
572         ret = btrfs_insert_dir_item(trans, root,
573                                     name, namelen, dir, &key,
574                                     BTRFS_FT_DIR, index);
575         if (ret) {
576                 btrfs_abort_transaction(trans, root, ret);
577                 goto fail;
578         }
579
580         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
581         ret = btrfs_update_inode(trans, root, dir);
582         BUG_ON(ret);
583
584         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
585                                  objectid, root->root_key.objectid,
586                                  btrfs_ino(dir), index, name, namelen);
587         BUG_ON(ret);
588
589         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
590                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
591                                   objectid);
592         if (ret)
593                 btrfs_abort_transaction(trans, root, ret);
594
595 fail:
596         trans->block_rsv = NULL;
597         trans->bytes_reserved = 0;
598         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
599
600         if (async_transid) {
601                 *async_transid = trans->transid;
602                 err = btrfs_commit_transaction_async(trans, root, 1);
603                 if (err)
604                         err = btrfs_commit_transaction(trans, root);
605         } else {
606                 err = btrfs_commit_transaction(trans, root);
607         }
608         if (err && !ret)
609                 ret = err;
610
611         if (!ret) {
612                 inode = btrfs_lookup_dentry(dir, dentry);
613                 if (IS_ERR(inode))
614                         return PTR_ERR(inode);
615                 d_instantiate(dentry, inode);
616         }
617         return ret;
618 }
619
620 static void btrfs_wait_nocow_write(struct btrfs_root *root)
621 {
622         s64 writers;
623         DEFINE_WAIT(wait);
624
625         do {
626                 prepare_to_wait(&root->subv_writers->wait, &wait,
627                                 TASK_UNINTERRUPTIBLE);
628
629                 writers = percpu_counter_sum(&root->subv_writers->counter);
630                 if (writers)
631                         schedule();
632
633                 finish_wait(&root->subv_writers->wait, &wait);
634         } while (writers);
635 }
636
637 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
638                            struct dentry *dentry, char *name, int namelen,
639                            u64 *async_transid, bool readonly,
640                            struct btrfs_qgroup_inherit *inherit)
641 {
642         struct inode *inode;
643         struct btrfs_pending_snapshot *pending_snapshot;
644         struct btrfs_trans_handle *trans;
645         int ret;
646
647         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
648                 return -EINVAL;
649
650         atomic_inc(&root->will_be_snapshoted);
651         smp_mb__after_atomic();
652         btrfs_wait_nocow_write(root);
653
654         ret = btrfs_start_delalloc_inodes(root, 0);
655         if (ret)
656                 goto out;
657
658         btrfs_wait_ordered_extents(root, -1);
659
660         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
661         if (!pending_snapshot) {
662                 ret = -ENOMEM;
663                 goto out;
664         }
665
666         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
667                              BTRFS_BLOCK_RSV_TEMP);
668         /*
669          * 1 - parent dir inode
670          * 2 - dir entries
671          * 1 - root item
672          * 2 - root ref/backref
673          * 1 - root of snapshot
674          * 1 - UUID item
675          */
676         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
677                                         &pending_snapshot->block_rsv, 8,
678                                         &pending_snapshot->qgroup_reserved,
679                                         false);
680         if (ret)
681                 goto free;
682
683         pending_snapshot->dentry = dentry;
684         pending_snapshot->root = root;
685         pending_snapshot->readonly = readonly;
686         pending_snapshot->dir = dir;
687         pending_snapshot->inherit = inherit;
688
689         trans = btrfs_start_transaction(root, 0);
690         if (IS_ERR(trans)) {
691                 ret = PTR_ERR(trans);
692                 goto fail;
693         }
694
695         spin_lock(&root->fs_info->trans_lock);
696         list_add(&pending_snapshot->list,
697                  &trans->transaction->pending_snapshots);
698         spin_unlock(&root->fs_info->trans_lock);
699         if (async_transid) {
700                 *async_transid = trans->transid;
701                 ret = btrfs_commit_transaction_async(trans,
702                                      root->fs_info->extent_root, 1);
703                 if (ret)
704                         ret = btrfs_commit_transaction(trans, root);
705         } else {
706                 ret = btrfs_commit_transaction(trans,
707                                                root->fs_info->extent_root);
708         }
709         if (ret)
710                 goto fail;
711
712         ret = pending_snapshot->error;
713         if (ret)
714                 goto fail;
715
716         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
717         if (ret)
718                 goto fail;
719
720         /*
721          * If orphan cleanup did remove any orphans, it means the tree was
722          * modified and therefore the commit root is not the same as the
723          * current root anymore. This is a problem, because send uses the
724          * commit root and therefore can see inode items that don't exist
725          * in the current root anymore, and for example make calls to
726          * btrfs_iget, which will do tree lookups based on the current root
727          * and not on the commit root. Those lookups will fail, returning a
728          * -ESTALE error, and making send fail with that error. So make sure
729          * a send does not see any orphans we have just removed, and that it
730          * will see the same inodes regardless of whether a transaction
731          * commit happened before it started (meaning that the commit root
732          * will be the same as the current root) or not.
733          */
734         if (readonly && pending_snapshot->snap->node !=
735             pending_snapshot->snap->commit_root) {
736                 trans = btrfs_join_transaction(pending_snapshot->snap);
737                 if (IS_ERR(trans) && PTR_ERR(trans) != -ENOENT) {
738                         ret = PTR_ERR(trans);
739                         goto fail;
740                 }
741                 if (!IS_ERR(trans)) {
742                         ret = btrfs_commit_transaction(trans,
743                                                        pending_snapshot->snap);
744                         if (ret)
745                                 goto fail;
746                 }
747         }
748
749         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
750         if (IS_ERR(inode)) {
751                 ret = PTR_ERR(inode);
752                 goto fail;
753         }
754
755         d_instantiate(dentry, inode);
756         ret = 0;
757 fail:
758         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
759                                          &pending_snapshot->block_rsv,
760                                          pending_snapshot->qgroup_reserved);
761 free:
762         kfree(pending_snapshot);
763 out:
764         atomic_dec(&root->will_be_snapshoted);
765         return ret;
766 }
767
768 /*  copy of may_delete in fs/namei.c()
769  *      Check whether we can remove a link victim from directory dir, check
770  *  whether the type of victim is right.
771  *  1. We can't do it if dir is read-only (done in permission())
772  *  2. We should have write and exec permissions on dir
773  *  3. We can't remove anything from append-only dir
774  *  4. We can't do anything with immutable dir (done in permission())
775  *  5. If the sticky bit on dir is set we should either
776  *      a. be owner of dir, or
777  *      b. be owner of victim, or
778  *      c. have CAP_FOWNER capability
779  *  6. If the victim is append-only or immutable we can't do antyhing with
780  *     links pointing to it.
781  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
782  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
783  *  9. We can't remove a root or mountpoint.
784  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
785  *     nfs_async_unlink().
786  */
787
788 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
789 {
790         int error;
791
792         if (!victim->d_inode)
793                 return -ENOENT;
794
795         BUG_ON(victim->d_parent->d_inode != dir);
796         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
797
798         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
799         if (error)
800                 return error;
801         if (IS_APPEND(dir))
802                 return -EPERM;
803         if (check_sticky(dir, victim->d_inode) || IS_APPEND(victim->d_inode) ||
804             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
805                 return -EPERM;
806         if (isdir) {
807                 if (!S_ISDIR(victim->d_inode->i_mode))
808                         return -ENOTDIR;
809                 if (IS_ROOT(victim))
810                         return -EBUSY;
811         } else if (S_ISDIR(victim->d_inode->i_mode))
812                 return -EISDIR;
813         if (IS_DEADDIR(dir))
814                 return -ENOENT;
815         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
816                 return -EBUSY;
817         return 0;
818 }
819
820 /* copy of may_create in fs/namei.c() */
821 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
822 {
823         if (child->d_inode)
824                 return -EEXIST;
825         if (IS_DEADDIR(dir))
826                 return -ENOENT;
827         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
828 }
829
830 /*
831  * Create a new subvolume below @parent.  This is largely modeled after
832  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
833  * inside this filesystem so it's quite a bit simpler.
834  */
835 static noinline int btrfs_mksubvol(struct path *parent,
836                                    char *name, int namelen,
837                                    struct btrfs_root *snap_src,
838                                    u64 *async_transid, bool readonly,
839                                    struct btrfs_qgroup_inherit *inherit)
840 {
841         struct inode *dir  = parent->dentry->d_inode;
842         struct dentry *dentry;
843         int error;
844
845         error = mutex_lock_killable_nested(&dir->i_mutex, 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 = -EEXIST;
855         if (dentry->d_inode)
856                 goto out_dput;
857
858         error = btrfs_may_create(dir, dentry);
859         if (error)
860                 goto out_dput;
861
862         /*
863          * even if this name doesn't exist, we may get hash collisions.
864          * check for them now when we can safely fail
865          */
866         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
867                                                dir->i_ino, name,
868                                                namelen);
869         if (error)
870                 goto out_dput;
871
872         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
873
874         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
875                 goto out_up_read;
876
877         if (snap_src) {
878                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
879                                         async_transid, readonly, inherit);
880         } else {
881                 error = create_subvol(dir, dentry, name, namelen,
882                                       async_transid, inherit);
883         }
884         if (!error)
885                 fsnotify_mkdir(dir, dentry);
886 out_up_read:
887         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
888 out_dput:
889         dput(dentry);
890 out_unlock:
891         mutex_unlock(&dir->i_mutex);
892         return error;
893 }
894
895 /*
896  * When we're defragging a range, we don't want to kick it off again
897  * if it is really just waiting for delalloc to send it down.
898  * If we find a nice big extent or delalloc range for the bytes in the
899  * file you want to defrag, we return 0 to let you know to skip this
900  * part of the file
901  */
902 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
903 {
904         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
905         struct extent_map *em = NULL;
906         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
907         u64 end;
908
909         read_lock(&em_tree->lock);
910         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
911         read_unlock(&em_tree->lock);
912
913         if (em) {
914                 end = extent_map_end(em);
915                 free_extent_map(em);
916                 if (end - offset > thresh)
917                         return 0;
918         }
919         /* if we already have a nice delalloc here, just stop */
920         thresh /= 2;
921         end = count_range_bits(io_tree, &offset, offset + thresh,
922                                thresh, EXTENT_DELALLOC, 1);
923         if (end >= thresh)
924                 return 0;
925         return 1;
926 }
927
928 /*
929  * helper function to walk through a file and find extents
930  * newer than a specific transid, and smaller than thresh.
931  *
932  * This is used by the defragging code to find new and small
933  * extents
934  */
935 static int find_new_extents(struct btrfs_root *root,
936                             struct inode *inode, u64 newer_than,
937                             u64 *off, u32 thresh)
938 {
939         struct btrfs_path *path;
940         struct btrfs_key min_key;
941         struct extent_buffer *leaf;
942         struct btrfs_file_extent_item *extent;
943         int type;
944         int ret;
945         u64 ino = btrfs_ino(inode);
946
947         path = btrfs_alloc_path();
948         if (!path)
949                 return -ENOMEM;
950
951         min_key.objectid = ino;
952         min_key.type = BTRFS_EXTENT_DATA_KEY;
953         min_key.offset = *off;
954
955         while (1) {
956                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
957                 if (ret != 0)
958                         goto none;
959 process_slot:
960                 if (min_key.objectid != ino)
961                         goto none;
962                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
963                         goto none;
964
965                 leaf = path->nodes[0];
966                 extent = btrfs_item_ptr(leaf, path->slots[0],
967                                         struct btrfs_file_extent_item);
968
969                 type = btrfs_file_extent_type(leaf, extent);
970                 if (type == BTRFS_FILE_EXTENT_REG &&
971                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
972                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
973                         *off = min_key.offset;
974                         btrfs_free_path(path);
975                         return 0;
976                 }
977
978                 path->slots[0]++;
979                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
980                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
981                         goto process_slot;
982                 }
983
984                 if (min_key.offset == (u64)-1)
985                         goto none;
986
987                 min_key.offset++;
988                 btrfs_release_path(path);
989         }
990 none:
991         btrfs_free_path(path);
992         return -ENOENT;
993 }
994
995 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
996 {
997         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
998         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
999         struct extent_map *em;
1000         u64 len = PAGE_CACHE_SIZE;
1001
1002         /*
1003          * hopefully we have this extent in the tree already, try without
1004          * the full extent lock
1005          */
1006         read_lock(&em_tree->lock);
1007         em = lookup_extent_mapping(em_tree, start, len);
1008         read_unlock(&em_tree->lock);
1009
1010         if (!em) {
1011                 struct extent_state *cached = NULL;
1012                 u64 end = start + len - 1;
1013
1014                 /* get the big lock and read metadata off disk */
1015                 lock_extent_bits(io_tree, start, end, 0, &cached);
1016                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1017                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1018
1019                 if (IS_ERR(em))
1020                         return NULL;
1021         }
1022
1023         return em;
1024 }
1025
1026 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1027 {
1028         struct extent_map *next;
1029         bool ret = true;
1030
1031         /* this is the last extent */
1032         if (em->start + em->len >= i_size_read(inode))
1033                 return false;
1034
1035         next = defrag_lookup_extent(inode, em->start + em->len);
1036         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1037                 ret = false;
1038         else if ((em->block_start + em->block_len == next->block_start) &&
1039                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1040                 ret = false;
1041
1042         free_extent_map(next);
1043         return ret;
1044 }
1045
1046 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1047                                u64 *last_len, u64 *skip, u64 *defrag_end,
1048                                int compress)
1049 {
1050         struct extent_map *em;
1051         int ret = 1;
1052         bool next_mergeable = true;
1053
1054         /*
1055          * make sure that once we start defragging an extent, we keep on
1056          * defragging it
1057          */
1058         if (start < *defrag_end)
1059                 return 1;
1060
1061         *skip = 0;
1062
1063         em = defrag_lookup_extent(inode, start);
1064         if (!em)
1065                 return 0;
1066
1067         /* this will cover holes, and inline extents */
1068         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1069                 ret = 0;
1070                 goto out;
1071         }
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))
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         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1131
1132         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1133         if (!isize || start_index > file_end)
1134                 return 0;
1135
1136         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1137
1138         ret = btrfs_delalloc_reserve_space(inode,
1139                                            page_cnt << PAGE_CACHE_SHIFT);
1140         if (ret)
1141                 return ret;
1142         i_done = 0;
1143         tree = &BTRFS_I(inode)->io_tree;
1144
1145         /* step one, lock all the pages */
1146         for (i = 0; i < page_cnt; i++) {
1147                 struct page *page;
1148 again:
1149                 page = find_or_create_page(inode->i_mapping,
1150                                            start_index + i, mask);
1151                 if (!page)
1152                         break;
1153
1154                 page_start = page_offset(page);
1155                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1156                 while (1) {
1157                         lock_extent_bits(tree, page_start, page_end,
1158                                          0, &cached_state);
1159                         ordered = btrfs_lookup_ordered_extent(inode,
1160                                                               page_start);
1161                         unlock_extent_cached(tree, page_start, page_end,
1162                                              &cached_state, GFP_NOFS);
1163                         if (!ordered)
1164                                 break;
1165
1166                         unlock_page(page);
1167                         btrfs_start_ordered_extent(inode, ordered, 1);
1168                         btrfs_put_ordered_extent(ordered);
1169                         lock_page(page);
1170                         /*
1171                          * we unlocked the page above, so we need check if
1172                          * it was released or not.
1173                          */
1174                         if (page->mapping != inode->i_mapping) {
1175                                 unlock_page(page);
1176                                 page_cache_release(page);
1177                                 goto again;
1178                         }
1179                 }
1180
1181                 if (!PageUptodate(page)) {
1182                         btrfs_readpage(NULL, page);
1183                         lock_page(page);
1184                         if (!PageUptodate(page)) {
1185                                 unlock_page(page);
1186                                 page_cache_release(page);
1187                                 ret = -EIO;
1188                                 break;
1189                         }
1190                 }
1191
1192                 if (page->mapping != inode->i_mapping) {
1193                         unlock_page(page);
1194                         page_cache_release(page);
1195                         goto again;
1196                 }
1197
1198                 pages[i] = page;
1199                 i_done++;
1200         }
1201         if (!i_done || ret)
1202                 goto out;
1203
1204         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1205                 goto out;
1206
1207         /*
1208          * so now we have a nice long stream of locked
1209          * and up to date pages, lets wait on them
1210          */
1211         for (i = 0; i < i_done; i++)
1212                 wait_on_page_writeback(pages[i]);
1213
1214         page_start = page_offset(pages[0]);
1215         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1216
1217         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1218                          page_start, page_end - 1, 0, &cached_state);
1219         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1220                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1221                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1222                           &cached_state, GFP_NOFS);
1223
1224         if (i_done != page_cnt) {
1225                 spin_lock(&BTRFS_I(inode)->lock);
1226                 BTRFS_I(inode)->outstanding_extents++;
1227                 spin_unlock(&BTRFS_I(inode)->lock);
1228                 btrfs_delalloc_release_space(inode,
1229                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1230         }
1231
1232
1233         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1234                           &cached_state, GFP_NOFS);
1235
1236         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1237                              page_start, page_end - 1, &cached_state,
1238                              GFP_NOFS);
1239
1240         for (i = 0; i < i_done; i++) {
1241                 clear_page_dirty_for_io(pages[i]);
1242                 ClearPageChecked(pages[i]);
1243                 set_page_extent_mapped(pages[i]);
1244                 set_page_dirty(pages[i]);
1245                 unlock_page(pages[i]);
1246                 page_cache_release(pages[i]);
1247         }
1248         return i_done;
1249 out:
1250         for (i = 0; i < i_done; i++) {
1251                 unlock_page(pages[i]);
1252                 page_cache_release(pages[i]);
1253         }
1254         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1255         return ret;
1256
1257 }
1258
1259 int btrfs_defrag_file(struct inode *inode, struct file *file,
1260                       struct btrfs_ioctl_defrag_range_args *range,
1261                       u64 newer_than, unsigned long max_to_defrag)
1262 {
1263         struct btrfs_root *root = BTRFS_I(inode)->root;
1264         struct file_ra_state *ra = NULL;
1265         unsigned long last_index;
1266         u64 isize = i_size_read(inode);
1267         u64 last_len = 0;
1268         u64 skip = 0;
1269         u64 defrag_end = 0;
1270         u64 newer_off = range->start;
1271         unsigned long i;
1272         unsigned long ra_index = 0;
1273         int ret;
1274         int defrag_count = 0;
1275         int compress_type = BTRFS_COMPRESS_ZLIB;
1276         u32 extent_thresh = range->extent_thresh;
1277         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1278         unsigned long cluster = max_cluster;
1279         u64 new_align = ~((u64)128 * 1024 - 1);
1280         struct page **pages = NULL;
1281
1282         if (isize == 0)
1283                 return 0;
1284
1285         if (range->start >= isize)
1286                 return -EINVAL;
1287
1288         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1289                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1290                         return -EINVAL;
1291                 if (range->compress_type)
1292                         compress_type = range->compress_type;
1293         }
1294
1295         if (extent_thresh == 0)
1296                 extent_thresh = 256 * 1024;
1297
1298         /*
1299          * if we were not given a file, allocate a readahead
1300          * context
1301          */
1302         if (!file) {
1303                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1304                 if (!ra)
1305                         return -ENOMEM;
1306                 file_ra_state_init(ra, inode->i_mapping);
1307         } else {
1308                 ra = &file->f_ra;
1309         }
1310
1311         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1312                         GFP_NOFS);
1313         if (!pages) {
1314                 ret = -ENOMEM;
1315                 goto out_ra;
1316         }
1317
1318         /* find the last page to defrag */
1319         if (range->start + range->len > range->start) {
1320                 last_index = min_t(u64, isize - 1,
1321                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1322         } else {
1323                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1324         }
1325
1326         if (newer_than) {
1327                 ret = find_new_extents(root, inode, newer_than,
1328                                        &newer_off, 64 * 1024);
1329                 if (!ret) {
1330                         range->start = newer_off;
1331                         /*
1332                          * we always align our defrag to help keep
1333                          * the extents in the file evenly spaced
1334                          */
1335                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1336                 } else
1337                         goto out_ra;
1338         } else {
1339                 i = range->start >> PAGE_CACHE_SHIFT;
1340         }
1341         if (!max_to_defrag)
1342                 max_to_defrag = last_index + 1;
1343
1344         /*
1345          * make writeback starts from i, so the defrag range can be
1346          * written sequentially.
1347          */
1348         if (i < inode->i_mapping->writeback_index)
1349                 inode->i_mapping->writeback_index = i;
1350
1351         while (i <= last_index && defrag_count < max_to_defrag &&
1352                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1353                 /*
1354                  * make sure we stop running if someone unmounts
1355                  * the FS
1356                  */
1357                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1358                         break;
1359
1360                 if (btrfs_defrag_cancelled(root->fs_info)) {
1361                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1362                         ret = -EAGAIN;
1363                         break;
1364                 }
1365
1366                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1367                                          extent_thresh, &last_len, &skip,
1368                                          &defrag_end, range->flags &
1369                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1370                         unsigned long next;
1371                         /*
1372                          * the should_defrag function tells us how much to skip
1373                          * bump our counter by the suggested amount
1374                          */
1375                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1376                         i = max(i + 1, next);
1377                         continue;
1378                 }
1379
1380                 if (!newer_than) {
1381                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1382                                    PAGE_CACHE_SHIFT) - i;
1383                         cluster = min(cluster, max_cluster);
1384                 } else {
1385                         cluster = max_cluster;
1386                 }
1387
1388                 if (i + cluster > ra_index) {
1389                         ra_index = max(i, ra_index);
1390                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1391                                        cluster);
1392                         ra_index += max_cluster;
1393                 }
1394
1395                 mutex_lock(&inode->i_mutex);
1396                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1397                         BTRFS_I(inode)->force_compress = compress_type;
1398                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1399                 if (ret < 0) {
1400                         mutex_unlock(&inode->i_mutex);
1401                         goto out_ra;
1402                 }
1403
1404                 defrag_count += ret;
1405                 balance_dirty_pages_ratelimited(inode->i_mapping);
1406                 mutex_unlock(&inode->i_mutex);
1407
1408                 if (newer_than) {
1409                         if (newer_off == (u64)-1)
1410                                 break;
1411
1412                         if (ret > 0)
1413                                 i += ret;
1414
1415                         newer_off = max(newer_off + 1,
1416                                         (u64)i << PAGE_CACHE_SHIFT);
1417
1418                         ret = find_new_extents(root, inode,
1419                                                newer_than, &newer_off,
1420                                                64 * 1024);
1421                         if (!ret) {
1422                                 range->start = newer_off;
1423                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1424                         } else {
1425                                 break;
1426                         }
1427                 } else {
1428                         if (ret > 0) {
1429                                 i += ret;
1430                                 last_len += ret << PAGE_CACHE_SHIFT;
1431                         } else {
1432                                 i++;
1433                                 last_len = 0;
1434                         }
1435                 }
1436         }
1437
1438         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1439                 filemap_flush(inode->i_mapping);
1440                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1441                              &BTRFS_I(inode)->runtime_flags))
1442                         filemap_flush(inode->i_mapping);
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1446                 /* the filemap_flush will queue IO into the worker threads, but
1447                  * we have to make sure the IO is actually started and that
1448                  * ordered extents get created before we return
1449                  */
1450                 atomic_inc(&root->fs_info->async_submit_draining);
1451                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1452                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1453                         wait_event(root->fs_info->async_submit_wait,
1454                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1455                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1456                 }
1457                 atomic_dec(&root->fs_info->async_submit_draining);
1458         }
1459
1460         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1461                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1462         }
1463
1464         ret = defrag_count;
1465
1466 out_ra:
1467         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1468                 mutex_lock(&inode->i_mutex);
1469                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1470                 mutex_unlock(&inode->i_mutex);
1471         }
1472         if (!file)
1473                 kfree(ra);
1474         kfree(pages);
1475         return ret;
1476 }
1477
1478 static noinline int btrfs_ioctl_resize(struct file *file,
1479                                         void __user *arg)
1480 {
1481         u64 new_size;
1482         u64 old_size;
1483         u64 devid = 1;
1484         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1485         struct btrfs_ioctl_vol_args *vol_args;
1486         struct btrfs_trans_handle *trans;
1487         struct btrfs_device *device = NULL;
1488         char *sizestr;
1489         char *retptr;
1490         char *devstr = NULL;
1491         int ret = 0;
1492         int mod = 0;
1493
1494         if (!capable(CAP_SYS_ADMIN))
1495                 return -EPERM;
1496
1497         ret = mnt_want_write_file(file);
1498         if (ret)
1499                 return ret;
1500
1501         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1502                         1)) {
1503                 mnt_drop_write_file(file);
1504                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1505         }
1506
1507         mutex_lock(&root->fs_info->volume_mutex);
1508         vol_args = memdup_user(arg, sizeof(*vol_args));
1509         if (IS_ERR(vol_args)) {
1510                 ret = PTR_ERR(vol_args);
1511                 goto out;
1512         }
1513
1514         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1515
1516         sizestr = vol_args->name;
1517         devstr = strchr(sizestr, ':');
1518         if (devstr) {
1519                 sizestr = devstr + 1;
1520                 *devstr = '\0';
1521                 devstr = vol_args->name;
1522                 ret = kstrtoull(devstr, 10, &devid);
1523                 if (ret)
1524                         goto out_free;
1525                 if (!devid) {
1526                         ret = -EINVAL;
1527                         goto out_free;
1528                 }
1529                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1530         }
1531
1532         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1533         if (!device) {
1534                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1535                        devid);
1536                 ret = -ENODEV;
1537                 goto out_free;
1538         }
1539
1540         if (!device->writeable) {
1541                 btrfs_info(root->fs_info,
1542                            "resizer unable to apply on readonly device %llu",
1543                        devid);
1544                 ret = -EPERM;
1545                 goto out_free;
1546         }
1547
1548         if (!strcmp(sizestr, "max"))
1549                 new_size = device->bdev->bd_inode->i_size;
1550         else {
1551                 if (sizestr[0] == '-') {
1552                         mod = -1;
1553                         sizestr++;
1554                 } else if (sizestr[0] == '+') {
1555                         mod = 1;
1556                         sizestr++;
1557                 }
1558                 new_size = memparse(sizestr, &retptr);
1559                 if (*retptr != '\0' || new_size == 0) {
1560                         ret = -EINVAL;
1561                         goto out_free;
1562                 }
1563         }
1564
1565         if (device->is_tgtdev_for_dev_replace) {
1566                 ret = -EPERM;
1567                 goto out_free;
1568         }
1569
1570         old_size = btrfs_device_get_total_bytes(device);
1571
1572         if (mod < 0) {
1573                 if (new_size > old_size) {
1574                         ret = -EINVAL;
1575                         goto out_free;
1576                 }
1577                 new_size = old_size - new_size;
1578         } else if (mod > 0) {
1579                 if (new_size > ULLONG_MAX - old_size) {
1580                         ret = -ERANGE;
1581                         goto out_free;
1582                 }
1583                 new_size = old_size + new_size;
1584         }
1585
1586         if (new_size < 256 * 1024 * 1024) {
1587                 ret = -EINVAL;
1588                 goto out_free;
1589         }
1590         if (new_size > device->bdev->bd_inode->i_size) {
1591                 ret = -EFBIG;
1592                 goto out_free;
1593         }
1594
1595         do_div(new_size, root->sectorsize);
1596         new_size *= root->sectorsize;
1597
1598         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1599                       rcu_str_deref(device->name), new_size);
1600
1601         if (new_size > old_size) {
1602                 trans = btrfs_start_transaction(root, 0);
1603                 if (IS_ERR(trans)) {
1604                         ret = PTR_ERR(trans);
1605                         goto out_free;
1606                 }
1607                 ret = btrfs_grow_device(trans, device, new_size);
1608                 btrfs_commit_transaction(trans, root);
1609         } else if (new_size < old_size) {
1610                 ret = btrfs_shrink_device(device, new_size);
1611         } /* equal, nothing need to do */
1612
1613 out_free:
1614         kfree(vol_args);
1615 out:
1616         mutex_unlock(&root->fs_info->volume_mutex);
1617         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1618         mnt_drop_write_file(file);
1619         return ret;
1620 }
1621
1622 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1623                                 char *name, unsigned long fd, int subvol,
1624                                 u64 *transid, bool readonly,
1625                                 struct btrfs_qgroup_inherit *inherit)
1626 {
1627         int namelen;
1628         int ret = 0;
1629
1630         ret = mnt_want_write_file(file);
1631         if (ret)
1632                 goto out;
1633
1634         namelen = strlen(name);
1635         if (strchr(name, '/')) {
1636                 ret = -EINVAL;
1637                 goto out_drop_write;
1638         }
1639
1640         if (name[0] == '.' &&
1641            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1642                 ret = -EEXIST;
1643                 goto out_drop_write;
1644         }
1645
1646         if (subvol) {
1647                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1648                                      NULL, transid, readonly, inherit);
1649         } else {
1650                 struct fd src = fdget(fd);
1651                 struct inode *src_inode;
1652                 if (!src.file) {
1653                         ret = -EINVAL;
1654                         goto out_drop_write;
1655                 }
1656
1657                 src_inode = file_inode(src.file);
1658                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1659                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1660                                    "Snapshot src from another FS");
1661                         ret = -EXDEV;
1662                 } else if (!inode_owner_or_capable(src_inode)) {
1663                         /*
1664                          * Subvolume creation is not restricted, but snapshots
1665                          * are limited to own subvolumes only
1666                          */
1667                         ret = -EPERM;
1668                 } else {
1669                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1670                                              BTRFS_I(src_inode)->root,
1671                                              transid, readonly, inherit);
1672                 }
1673                 fdput(src);
1674         }
1675 out_drop_write:
1676         mnt_drop_write_file(file);
1677 out:
1678         return ret;
1679 }
1680
1681 static noinline int btrfs_ioctl_snap_create(struct file *file,
1682                                             void __user *arg, int subvol)
1683 {
1684         struct btrfs_ioctl_vol_args *vol_args;
1685         int ret;
1686
1687         vol_args = memdup_user(arg, sizeof(*vol_args));
1688         if (IS_ERR(vol_args))
1689                 return PTR_ERR(vol_args);
1690         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1691
1692         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1693                                               vol_args->fd, subvol,
1694                                               NULL, false, NULL);
1695
1696         kfree(vol_args);
1697         return ret;
1698 }
1699
1700 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1701                                                void __user *arg, int subvol)
1702 {
1703         struct btrfs_ioctl_vol_args_v2 *vol_args;
1704         int ret;
1705         u64 transid = 0;
1706         u64 *ptr = NULL;
1707         bool readonly = false;
1708         struct btrfs_qgroup_inherit *inherit = NULL;
1709
1710         vol_args = memdup_user(arg, sizeof(*vol_args));
1711         if (IS_ERR(vol_args))
1712                 return PTR_ERR(vol_args);
1713         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1714
1715         if (vol_args->flags &
1716             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1717               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1718                 ret = -EOPNOTSUPP;
1719                 goto free_args;
1720         }
1721
1722         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1723                 ptr = &transid;
1724         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1725                 readonly = true;
1726         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1727                 if (vol_args->size > PAGE_CACHE_SIZE) {
1728                         ret = -EINVAL;
1729                         goto free_args;
1730                 }
1731                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1732                 if (IS_ERR(inherit)) {
1733                         ret = PTR_ERR(inherit);
1734                         goto free_args;
1735                 }
1736         }
1737
1738         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1739                                               vol_args->fd, subvol, ptr,
1740                                               readonly, inherit);
1741         if (ret)
1742                 goto free_inherit;
1743
1744         if (ptr && copy_to_user(arg +
1745                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1746                                         transid),
1747                                 ptr, sizeof(*ptr)))
1748                 ret = -EFAULT;
1749
1750 free_inherit:
1751         kfree(inherit);
1752 free_args:
1753         kfree(vol_args);
1754         return ret;
1755 }
1756
1757 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1758                                                 void __user *arg)
1759 {
1760         struct inode *inode = file_inode(file);
1761         struct btrfs_root *root = BTRFS_I(inode)->root;
1762         int ret = 0;
1763         u64 flags = 0;
1764
1765         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1766                 return -EINVAL;
1767
1768         down_read(&root->fs_info->subvol_sem);
1769         if (btrfs_root_readonly(root))
1770                 flags |= BTRFS_SUBVOL_RDONLY;
1771         up_read(&root->fs_info->subvol_sem);
1772
1773         if (copy_to_user(arg, &flags, sizeof(flags)))
1774                 ret = -EFAULT;
1775
1776         return ret;
1777 }
1778
1779 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1780                                               void __user *arg)
1781 {
1782         struct inode *inode = file_inode(file);
1783         struct btrfs_root *root = BTRFS_I(inode)->root;
1784         struct btrfs_trans_handle *trans;
1785         u64 root_flags;
1786         u64 flags;
1787         int ret = 0;
1788
1789         if (!inode_owner_or_capable(inode))
1790                 return -EPERM;
1791
1792         ret = mnt_want_write_file(file);
1793         if (ret)
1794                 goto out;
1795
1796         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1797                 ret = -EINVAL;
1798                 goto out_drop_write;
1799         }
1800
1801         if (copy_from_user(&flags, arg, sizeof(flags))) {
1802                 ret = -EFAULT;
1803                 goto out_drop_write;
1804         }
1805
1806         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1807                 ret = -EINVAL;
1808                 goto out_drop_write;
1809         }
1810
1811         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1812                 ret = -EOPNOTSUPP;
1813                 goto out_drop_write;
1814         }
1815
1816         down_write(&root->fs_info->subvol_sem);
1817
1818         /* nothing to do */
1819         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1820                 goto out_drop_sem;
1821
1822         root_flags = btrfs_root_flags(&root->root_item);
1823         if (flags & BTRFS_SUBVOL_RDONLY) {
1824                 btrfs_set_root_flags(&root->root_item,
1825                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1826         } else {
1827                 /*
1828                  * Block RO -> RW transition if this subvolume is involved in
1829                  * send
1830                  */
1831                 spin_lock(&root->root_item_lock);
1832                 if (root->send_in_progress == 0) {
1833                         btrfs_set_root_flags(&root->root_item,
1834                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1835                         spin_unlock(&root->root_item_lock);
1836                 } else {
1837                         spin_unlock(&root->root_item_lock);
1838                         btrfs_warn(root->fs_info,
1839                         "Attempt to set subvolume %llu read-write during send",
1840                                         root->root_key.objectid);
1841                         ret = -EPERM;
1842                         goto out_drop_sem;
1843                 }
1844         }
1845
1846         trans = btrfs_start_transaction(root, 1);
1847         if (IS_ERR(trans)) {
1848                 ret = PTR_ERR(trans);
1849                 goto out_reset;
1850         }
1851
1852         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1853                                 &root->root_key, &root->root_item);
1854
1855         btrfs_commit_transaction(trans, root);
1856 out_reset:
1857         if (ret)
1858                 btrfs_set_root_flags(&root->root_item, root_flags);
1859 out_drop_sem:
1860         up_write(&root->fs_info->subvol_sem);
1861 out_drop_write:
1862         mnt_drop_write_file(file);
1863 out:
1864         return ret;
1865 }
1866
1867 /*
1868  * helper to check if the subvolume references other subvolumes
1869  */
1870 static noinline int may_destroy_subvol(struct btrfs_root *root)
1871 {
1872         struct btrfs_path *path;
1873         struct btrfs_dir_item *di;
1874         struct btrfs_key key;
1875         u64 dir_id;
1876         int ret;
1877
1878         path = btrfs_alloc_path();
1879         if (!path)
1880                 return -ENOMEM;
1881
1882         /* Make sure this root isn't set as the default subvol */
1883         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1884         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1885                                    dir_id, "default", 7, 0);
1886         if (di && !IS_ERR(di)) {
1887                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1888                 if (key.objectid == root->root_key.objectid) {
1889                         ret = -EPERM;
1890                         btrfs_err(root->fs_info, "deleting default subvolume "
1891                                   "%llu is not allowed", key.objectid);
1892                         goto out;
1893                 }
1894                 btrfs_release_path(path);
1895         }
1896
1897         key.objectid = root->root_key.objectid;
1898         key.type = BTRFS_ROOT_REF_KEY;
1899         key.offset = (u64)-1;
1900
1901         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1902                                 &key, path, 0, 0);
1903         if (ret < 0)
1904                 goto out;
1905         BUG_ON(ret == 0);
1906
1907         ret = 0;
1908         if (path->slots[0] > 0) {
1909                 path->slots[0]--;
1910                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1911                 if (key.objectid == root->root_key.objectid &&
1912                     key.type == BTRFS_ROOT_REF_KEY)
1913                         ret = -ENOTEMPTY;
1914         }
1915 out:
1916         btrfs_free_path(path);
1917         return ret;
1918 }
1919
1920 static noinline int key_in_sk(struct btrfs_key *key,
1921                               struct btrfs_ioctl_search_key *sk)
1922 {
1923         struct btrfs_key test;
1924         int ret;
1925
1926         test.objectid = sk->min_objectid;
1927         test.type = sk->min_type;
1928         test.offset = sk->min_offset;
1929
1930         ret = btrfs_comp_cpu_keys(key, &test);
1931         if (ret < 0)
1932                 return 0;
1933
1934         test.objectid = sk->max_objectid;
1935         test.type = sk->max_type;
1936         test.offset = sk->max_offset;
1937
1938         ret = btrfs_comp_cpu_keys(key, &test);
1939         if (ret > 0)
1940                 return 0;
1941         return 1;
1942 }
1943
1944 static noinline int copy_to_sk(struct btrfs_root *root,
1945                                struct btrfs_path *path,
1946                                struct btrfs_key *key,
1947                                struct btrfs_ioctl_search_key *sk,
1948                                size_t *buf_size,
1949                                char __user *ubuf,
1950                                unsigned long *sk_offset,
1951                                int *num_found)
1952 {
1953         u64 found_transid;
1954         struct extent_buffer *leaf;
1955         struct btrfs_ioctl_search_header sh;
1956         unsigned long item_off;
1957         unsigned long item_len;
1958         int nritems;
1959         int i;
1960         int slot;
1961         int ret = 0;
1962
1963         leaf = path->nodes[0];
1964         slot = path->slots[0];
1965         nritems = btrfs_header_nritems(leaf);
1966
1967         if (btrfs_header_generation(leaf) > sk->max_transid) {
1968                 i = nritems;
1969                 goto advance_key;
1970         }
1971         found_transid = btrfs_header_generation(leaf);
1972
1973         for (i = slot; i < nritems; i++) {
1974                 item_off = btrfs_item_ptr_offset(leaf, i);
1975                 item_len = btrfs_item_size_nr(leaf, i);
1976
1977                 btrfs_item_key_to_cpu(leaf, key, i);
1978                 if (!key_in_sk(key, sk))
1979                         continue;
1980
1981                 if (sizeof(sh) + item_len > *buf_size) {
1982                         if (*num_found) {
1983                                 ret = 1;
1984                                 goto out;
1985                         }
1986
1987                         /*
1988                          * return one empty item back for v1, which does not
1989                          * handle -EOVERFLOW
1990                          */
1991
1992                         *buf_size = sizeof(sh) + item_len;
1993                         item_len = 0;
1994                         ret = -EOVERFLOW;
1995                 }
1996
1997                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1998                         ret = 1;
1999                         goto out;
2000                 }
2001
2002                 sh.objectid = key->objectid;
2003                 sh.offset = key->offset;
2004                 sh.type = key->type;
2005                 sh.len = item_len;
2006                 sh.transid = found_transid;
2007
2008                 /* copy search result header */
2009                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2010                         ret = -EFAULT;
2011                         goto out;
2012                 }
2013
2014                 *sk_offset += sizeof(sh);
2015
2016                 if (item_len) {
2017                         char __user *up = ubuf + *sk_offset;
2018                         /* copy the item */
2019                         if (read_extent_buffer_to_user(leaf, up,
2020                                                        item_off, item_len)) {
2021                                 ret = -EFAULT;
2022                                 goto out;
2023                         }
2024
2025                         *sk_offset += item_len;
2026                 }
2027                 (*num_found)++;
2028
2029                 if (ret) /* -EOVERFLOW from above */
2030                         goto out;
2031
2032                 if (*num_found >= sk->nr_items) {
2033                         ret = 1;
2034                         goto out;
2035                 }
2036         }
2037 advance_key:
2038         ret = 0;
2039         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2040                 key->offset++;
2041         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2042                 key->offset = 0;
2043                 key->type++;
2044         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2045                 key->offset = 0;
2046                 key->type = 0;
2047                 key->objectid++;
2048         } else
2049                 ret = 1;
2050 out:
2051         /*
2052          *  0: all items from this leaf copied, continue with next
2053          *  1: * more items can be copied, but unused buffer is too small
2054          *     * all items were found
2055          *     Either way, it will stops the loop which iterates to the next
2056          *     leaf
2057          *  -EOVERFLOW: item was to large for buffer
2058          *  -EFAULT: could not copy extent buffer back to userspace
2059          */
2060         return ret;
2061 }
2062
2063 static noinline int search_ioctl(struct inode *inode,
2064                                  struct btrfs_ioctl_search_key *sk,
2065                                  size_t *buf_size,
2066                                  char __user *ubuf)
2067 {
2068         struct btrfs_root *root;
2069         struct btrfs_key key;
2070         struct btrfs_path *path;
2071         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2072         int ret;
2073         int num_found = 0;
2074         unsigned long sk_offset = 0;
2075
2076         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2077                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2078                 return -EOVERFLOW;
2079         }
2080
2081         path = btrfs_alloc_path();
2082         if (!path)
2083                 return -ENOMEM;
2084
2085         if (sk->tree_id == 0) {
2086                 /* search the root of the inode that was passed */
2087                 root = BTRFS_I(inode)->root;
2088         } else {
2089                 key.objectid = sk->tree_id;
2090                 key.type = BTRFS_ROOT_ITEM_KEY;
2091                 key.offset = (u64)-1;
2092                 root = btrfs_read_fs_root_no_name(info, &key);
2093                 if (IS_ERR(root)) {
2094                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2095                                sk->tree_id);
2096                         btrfs_free_path(path);
2097                         return -ENOENT;
2098                 }
2099         }
2100
2101         key.objectid = sk->min_objectid;
2102         key.type = sk->min_type;
2103         key.offset = sk->min_offset;
2104
2105         while (1) {
2106                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2107                 if (ret != 0) {
2108                         if (ret > 0)
2109                                 ret = 0;
2110                         goto err;
2111                 }
2112                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2113                                  &sk_offset, &num_found);
2114                 btrfs_release_path(path);
2115                 if (ret)
2116                         break;
2117
2118         }
2119         if (ret > 0)
2120                 ret = 0;
2121 err:
2122         sk->nr_items = num_found;
2123         btrfs_free_path(path);
2124         return ret;
2125 }
2126
2127 static noinline int btrfs_ioctl_tree_search(struct file *file,
2128                                            void __user *argp)
2129 {
2130         struct btrfs_ioctl_search_args __user *uargs;
2131         struct btrfs_ioctl_search_key sk;
2132         struct inode *inode;
2133         int ret;
2134         size_t buf_size;
2135
2136         if (!capable(CAP_SYS_ADMIN))
2137                 return -EPERM;
2138
2139         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2140
2141         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2142                 return -EFAULT;
2143
2144         buf_size = sizeof(uargs->buf);
2145
2146         inode = file_inode(file);
2147         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2148
2149         /*
2150          * In the origin implementation an overflow is handled by returning a
2151          * search header with a len of zero, so reset ret.
2152          */
2153         if (ret == -EOVERFLOW)
2154                 ret = 0;
2155
2156         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2157                 ret = -EFAULT;
2158         return ret;
2159 }
2160
2161 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2162                                                void __user *argp)
2163 {
2164         struct btrfs_ioctl_search_args_v2 __user *uarg;
2165         struct btrfs_ioctl_search_args_v2 args;
2166         struct inode *inode;
2167         int ret;
2168         size_t buf_size;
2169         const size_t buf_limit = 16 * 1024 * 1024;
2170
2171         if (!capable(CAP_SYS_ADMIN))
2172                 return -EPERM;
2173
2174         /* copy search header and buffer size */
2175         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2176         if (copy_from_user(&args, uarg, sizeof(args)))
2177                 return -EFAULT;
2178
2179         buf_size = args.buf_size;
2180
2181         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2182                 return -EOVERFLOW;
2183
2184         /* limit result size to 16MB */
2185         if (buf_size > buf_limit)
2186                 buf_size = buf_limit;
2187
2188         inode = file_inode(file);
2189         ret = search_ioctl(inode, &args.key, &buf_size,
2190                            (char *)(&uarg->buf[0]));
2191         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2192                 ret = -EFAULT;
2193         else if (ret == -EOVERFLOW &&
2194                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2195                 ret = -EFAULT;
2196
2197         return ret;
2198 }
2199
2200 /*
2201  * Search INODE_REFs to identify path name of 'dirid' directory
2202  * in a 'tree_id' tree. and sets path name to 'name'.
2203  */
2204 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2205                                 u64 tree_id, u64 dirid, char *name)
2206 {
2207         struct btrfs_root *root;
2208         struct btrfs_key key;
2209         char *ptr;
2210         int ret = -1;
2211         int slot;
2212         int len;
2213         int total_len = 0;
2214         struct btrfs_inode_ref *iref;
2215         struct extent_buffer *l;
2216         struct btrfs_path *path;
2217
2218         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2219                 name[0]='\0';
2220                 return 0;
2221         }
2222
2223         path = btrfs_alloc_path();
2224         if (!path)
2225                 return -ENOMEM;
2226
2227         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2228
2229         key.objectid = tree_id;
2230         key.type = BTRFS_ROOT_ITEM_KEY;
2231         key.offset = (u64)-1;
2232         root = btrfs_read_fs_root_no_name(info, &key);
2233         if (IS_ERR(root)) {
2234                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2235                 ret = -ENOENT;
2236                 goto out;
2237         }
2238
2239         key.objectid = dirid;
2240         key.type = BTRFS_INODE_REF_KEY;
2241         key.offset = (u64)-1;
2242
2243         while (1) {
2244                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2245                 if (ret < 0)
2246                         goto out;
2247                 else if (ret > 0) {
2248                         ret = btrfs_previous_item(root, path, dirid,
2249                                                   BTRFS_INODE_REF_KEY);
2250                         if (ret < 0)
2251                                 goto out;
2252                         else if (ret > 0) {
2253                                 ret = -ENOENT;
2254                                 goto out;
2255                         }
2256                 }
2257
2258                 l = path->nodes[0];
2259                 slot = path->slots[0];
2260                 btrfs_item_key_to_cpu(l, &key, slot);
2261
2262                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2263                 len = btrfs_inode_ref_name_len(l, iref);
2264                 ptr -= len + 1;
2265                 total_len += len + 1;
2266                 if (ptr < name) {
2267                         ret = -ENAMETOOLONG;
2268                         goto out;
2269                 }
2270
2271                 *(ptr + len) = '/';
2272                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2273
2274                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2275                         break;
2276
2277                 btrfs_release_path(path);
2278                 key.objectid = key.offset;
2279                 key.offset = (u64)-1;
2280                 dirid = key.objectid;
2281         }
2282         memmove(name, ptr, total_len);
2283         name[total_len] = '\0';
2284         ret = 0;
2285 out:
2286         btrfs_free_path(path);
2287         return ret;
2288 }
2289
2290 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2291                                            void __user *argp)
2292 {
2293          struct btrfs_ioctl_ino_lookup_args *args;
2294          struct inode *inode;
2295          int ret;
2296
2297         if (!capable(CAP_SYS_ADMIN))
2298                 return -EPERM;
2299
2300         args = memdup_user(argp, sizeof(*args));
2301         if (IS_ERR(args))
2302                 return PTR_ERR(args);
2303
2304         inode = file_inode(file);
2305
2306         if (args->treeid == 0)
2307                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2308
2309         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2310                                         args->treeid, args->objectid,
2311                                         args->name);
2312
2313         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2314                 ret = -EFAULT;
2315
2316         kfree(args);
2317         return ret;
2318 }
2319
2320 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2321                                              void __user *arg)
2322 {
2323         struct dentry *parent = file->f_path.dentry;
2324         struct dentry *dentry;
2325         struct inode *dir = parent->d_inode;
2326         struct inode *inode;
2327         struct btrfs_root *root = BTRFS_I(dir)->root;
2328         struct btrfs_root *dest = NULL;
2329         struct btrfs_ioctl_vol_args *vol_args;
2330         struct btrfs_trans_handle *trans;
2331         struct btrfs_block_rsv block_rsv;
2332         u64 root_flags;
2333         u64 qgroup_reserved;
2334         int namelen;
2335         int ret;
2336         int err = 0;
2337
2338         vol_args = memdup_user(arg, sizeof(*vol_args));
2339         if (IS_ERR(vol_args))
2340                 return PTR_ERR(vol_args);
2341
2342         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2343         namelen = strlen(vol_args->name);
2344         if (strchr(vol_args->name, '/') ||
2345             strncmp(vol_args->name, "..", namelen) == 0) {
2346                 err = -EINVAL;
2347                 goto out;
2348         }
2349
2350         err = mnt_want_write_file(file);
2351         if (err)
2352                 goto out;
2353
2354
2355         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2356         if (err == -EINTR)
2357                 goto out_drop_write;
2358         dentry = lookup_one_len(vol_args->name, parent, namelen);
2359         if (IS_ERR(dentry)) {
2360                 err = PTR_ERR(dentry);
2361                 goto out_unlock_dir;
2362         }
2363
2364         if (!dentry->d_inode) {
2365                 err = -ENOENT;
2366                 goto out_dput;
2367         }
2368
2369         inode = dentry->d_inode;
2370         dest = BTRFS_I(inode)->root;
2371         if (!capable(CAP_SYS_ADMIN)) {
2372                 /*
2373                  * Regular user.  Only allow this with a special mount
2374                  * option, when the user has write+exec access to the
2375                  * subvol root, and when rmdir(2) would have been
2376                  * allowed.
2377                  *
2378                  * Note that this is _not_ check that the subvol is
2379                  * empty or doesn't contain data that we wouldn't
2380                  * otherwise be able to delete.
2381                  *
2382                  * Users who want to delete empty subvols should try
2383                  * rmdir(2).
2384                  */
2385                 err = -EPERM;
2386                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2387                         goto out_dput;
2388
2389                 /*
2390                  * Do not allow deletion if the parent dir is the same
2391                  * as the dir to be deleted.  That means the ioctl
2392                  * must be called on the dentry referencing the root
2393                  * of the subvol, not a random directory contained
2394                  * within it.
2395                  */
2396                 err = -EINVAL;
2397                 if (root == dest)
2398                         goto out_dput;
2399
2400                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2401                 if (err)
2402                         goto out_dput;
2403         }
2404
2405         /* check if subvolume may be deleted by a user */
2406         err = btrfs_may_delete(dir, dentry, 1);
2407         if (err)
2408                 goto out_dput;
2409
2410         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2411                 err = -EINVAL;
2412                 goto out_dput;
2413         }
2414
2415         mutex_lock(&inode->i_mutex);
2416
2417         /*
2418          * Don't allow to delete a subvolume with send in progress. This is
2419          * inside the i_mutex so the error handling that has to drop the bit
2420          * again is not run concurrently.
2421          */
2422         spin_lock(&dest->root_item_lock);
2423         root_flags = btrfs_root_flags(&dest->root_item);
2424         if (dest->send_in_progress == 0) {
2425                 btrfs_set_root_flags(&dest->root_item,
2426                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2427                 spin_unlock(&dest->root_item_lock);
2428         } else {
2429                 spin_unlock(&dest->root_item_lock);
2430                 btrfs_warn(root->fs_info,
2431                         "Attempt to delete subvolume %llu during send",
2432                         dest->root_key.objectid);
2433                 err = -EPERM;
2434                 goto out_dput;
2435         }
2436
2437         d_invalidate(dentry);
2438
2439         down_write(&root->fs_info->subvol_sem);
2440
2441         err = may_destroy_subvol(dest);
2442         if (err)
2443                 goto out_up_write;
2444
2445         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2446         /*
2447          * One for dir inode, two for dir entries, two for root
2448          * ref/backref.
2449          */
2450         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2451                                                5, &qgroup_reserved, true);
2452         if (err)
2453                 goto out_up_write;
2454
2455         trans = btrfs_start_transaction(root, 0);
2456         if (IS_ERR(trans)) {
2457                 err = PTR_ERR(trans);
2458                 goto out_release;
2459         }
2460         trans->block_rsv = &block_rsv;
2461         trans->bytes_reserved = block_rsv.size;
2462
2463         ret = btrfs_unlink_subvol(trans, root, dir,
2464                                 dest->root_key.objectid,
2465                                 dentry->d_name.name,
2466                                 dentry->d_name.len);
2467         if (ret) {
2468                 err = ret;
2469                 btrfs_abort_transaction(trans, root, ret);
2470                 goto out_end_trans;
2471         }
2472
2473         btrfs_record_root_in_trans(trans, dest);
2474
2475         memset(&dest->root_item.drop_progress, 0,
2476                 sizeof(dest->root_item.drop_progress));
2477         dest->root_item.drop_level = 0;
2478         btrfs_set_root_refs(&dest->root_item, 0);
2479
2480         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2481                 ret = btrfs_insert_orphan_item(trans,
2482                                         root->fs_info->tree_root,
2483                                         dest->root_key.objectid);
2484                 if (ret) {
2485                         btrfs_abort_transaction(trans, root, ret);
2486                         err = ret;
2487                         goto out_end_trans;
2488                 }
2489         }
2490
2491         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2492                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2493                                   dest->root_key.objectid);
2494         if (ret && ret != -ENOENT) {
2495                 btrfs_abort_transaction(trans, root, ret);
2496                 err = ret;
2497                 goto out_end_trans;
2498         }
2499         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2500                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2501                                           dest->root_item.received_uuid,
2502                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2503                                           dest->root_key.objectid);
2504                 if (ret && ret != -ENOENT) {
2505                         btrfs_abort_transaction(trans, root, ret);
2506                         err = ret;
2507                         goto out_end_trans;
2508                 }
2509         }
2510
2511 out_end_trans:
2512         trans->block_rsv = NULL;
2513         trans->bytes_reserved = 0;
2514         ret = btrfs_end_transaction(trans, root);
2515         if (ret && !err)
2516                 err = ret;
2517         inode->i_flags |= S_DEAD;
2518 out_release:
2519         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2520 out_up_write:
2521         up_write(&root->fs_info->subvol_sem);
2522         if (err) {
2523                 spin_lock(&dest->root_item_lock);
2524                 root_flags = btrfs_root_flags(&dest->root_item);
2525                 btrfs_set_root_flags(&dest->root_item,
2526                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2527                 spin_unlock(&dest->root_item_lock);
2528         }
2529         mutex_unlock(&inode->i_mutex);
2530         if (!err) {
2531                 shrink_dcache_sb(root->fs_info->sb);
2532                 btrfs_invalidate_inodes(dest);
2533                 d_delete(dentry);
2534                 ASSERT(dest->send_in_progress == 0);
2535
2536                 /* the last ref */
2537                 if (dest->ino_cache_inode) {
2538                         iput(dest->ino_cache_inode);
2539                         dest->ino_cache_inode = NULL;
2540                 }
2541         }
2542 out_dput:
2543         dput(dentry);
2544 out_unlock_dir:
2545         mutex_unlock(&dir->i_mutex);
2546 out_drop_write:
2547         mnt_drop_write_file(file);
2548 out:
2549         kfree(vol_args);
2550         return err;
2551 }
2552
2553 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2554 {
2555         struct inode *inode = file_inode(file);
2556         struct btrfs_root *root = BTRFS_I(inode)->root;
2557         struct btrfs_ioctl_defrag_range_args *range;
2558         int ret;
2559
2560         ret = mnt_want_write_file(file);
2561         if (ret)
2562                 return ret;
2563
2564         if (btrfs_root_readonly(root)) {
2565                 ret = -EROFS;
2566                 goto out;
2567         }
2568
2569         switch (inode->i_mode & S_IFMT) {
2570         case S_IFDIR:
2571                 if (!capable(CAP_SYS_ADMIN)) {
2572                         ret = -EPERM;
2573                         goto out;
2574                 }
2575                 ret = btrfs_defrag_root(root);
2576                 if (ret)
2577                         goto out;
2578                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2579                 break;
2580         case S_IFREG:
2581                 if (!(file->f_mode & FMODE_WRITE)) {
2582                         ret = -EINVAL;
2583                         goto out;
2584                 }
2585
2586                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2587                 if (!range) {
2588                         ret = -ENOMEM;
2589                         goto out;
2590                 }
2591
2592                 if (argp) {
2593                         if (copy_from_user(range, argp,
2594                                            sizeof(*range))) {
2595                                 ret = -EFAULT;
2596                                 kfree(range);
2597                                 goto out;
2598                         }
2599                         /* compression requires us to start the IO */
2600                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2601                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2602                                 range->extent_thresh = (u32)-1;
2603                         }
2604                 } else {
2605                         /* the rest are all set to zero by kzalloc */
2606                         range->len = (u64)-1;
2607                 }
2608                 ret = btrfs_defrag_file(file_inode(file), file,
2609                                         range, 0, 0);
2610                 if (ret > 0)
2611                         ret = 0;
2612                 kfree(range);
2613                 break;
2614         default:
2615                 ret = -EINVAL;
2616         }
2617 out:
2618         mnt_drop_write_file(file);
2619         return ret;
2620 }
2621
2622 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2623 {
2624         struct btrfs_ioctl_vol_args *vol_args;
2625         int ret;
2626
2627         if (!capable(CAP_SYS_ADMIN))
2628                 return -EPERM;
2629
2630         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2631                         1)) {
2632                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2633         }
2634
2635         mutex_lock(&root->fs_info->volume_mutex);
2636         vol_args = memdup_user(arg, sizeof(*vol_args));
2637         if (IS_ERR(vol_args)) {
2638                 ret = PTR_ERR(vol_args);
2639                 goto out;
2640         }
2641
2642         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2643         ret = btrfs_init_new_device(root, vol_args->name);
2644
2645         if (!ret)
2646                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2647
2648         kfree(vol_args);
2649 out:
2650         mutex_unlock(&root->fs_info->volume_mutex);
2651         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2652         return ret;
2653 }
2654
2655 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2656 {
2657         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2658         struct btrfs_ioctl_vol_args *vol_args;
2659         int ret;
2660
2661         if (!capable(CAP_SYS_ADMIN))
2662                 return -EPERM;
2663
2664         ret = mnt_want_write_file(file);
2665         if (ret)
2666                 return ret;
2667
2668         vol_args = memdup_user(arg, sizeof(*vol_args));
2669         if (IS_ERR(vol_args)) {
2670                 ret = PTR_ERR(vol_args);
2671                 goto err_drop;
2672         }
2673
2674         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2675
2676         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2677                         1)) {
2678                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2679                 goto out;
2680         }
2681
2682         mutex_lock(&root->fs_info->volume_mutex);
2683         ret = btrfs_rm_device(root, vol_args->name);
2684         mutex_unlock(&root->fs_info->volume_mutex);
2685         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2686
2687         if (!ret)
2688                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2689
2690 out:
2691         kfree(vol_args);
2692 err_drop:
2693         mnt_drop_write_file(file);
2694         return ret;
2695 }
2696
2697 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2698 {
2699         struct btrfs_ioctl_fs_info_args *fi_args;
2700         struct btrfs_device *device;
2701         struct btrfs_device *next;
2702         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2703         int ret = 0;
2704
2705         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2706         if (!fi_args)
2707                 return -ENOMEM;
2708
2709         mutex_lock(&fs_devices->device_list_mutex);
2710         fi_args->num_devices = fs_devices->num_devices;
2711         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2712
2713         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2714                 if (device->devid > fi_args->max_id)
2715                         fi_args->max_id = device->devid;
2716         }
2717         mutex_unlock(&fs_devices->device_list_mutex);
2718
2719         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2720         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2721         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2722
2723         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2724                 ret = -EFAULT;
2725
2726         kfree(fi_args);
2727         return ret;
2728 }
2729
2730 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2731 {
2732         struct btrfs_ioctl_dev_info_args *di_args;
2733         struct btrfs_device *dev;
2734         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2735         int ret = 0;
2736         char *s_uuid = NULL;
2737
2738         di_args = memdup_user(arg, sizeof(*di_args));
2739         if (IS_ERR(di_args))
2740                 return PTR_ERR(di_args);
2741
2742         if (!btrfs_is_empty_uuid(di_args->uuid))
2743                 s_uuid = di_args->uuid;
2744
2745         mutex_lock(&fs_devices->device_list_mutex);
2746         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2747
2748         if (!dev) {
2749                 ret = -ENODEV;
2750                 goto out;
2751         }
2752
2753         di_args->devid = dev->devid;
2754         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2755         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2756         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2757         if (dev->name) {
2758                 struct rcu_string *name;
2759
2760                 rcu_read_lock();
2761                 name = rcu_dereference(dev->name);
2762                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2763                 rcu_read_unlock();
2764                 di_args->path[sizeof(di_args->path) - 1] = 0;
2765         } else {
2766                 di_args->path[0] = '\0';
2767         }
2768
2769 out:
2770         mutex_unlock(&fs_devices->device_list_mutex);
2771         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2772                 ret = -EFAULT;
2773
2774         kfree(di_args);
2775         return ret;
2776 }
2777
2778 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2779 {
2780         struct page *page;
2781         pgoff_t index;
2782         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2783
2784         index = off >> PAGE_CACHE_SHIFT;
2785
2786         page = grab_cache_page(inode->i_mapping, index);
2787         if (!page)
2788                 return NULL;
2789
2790         if (!PageUptodate(page)) {
2791                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2792                                                  0))
2793                         return NULL;
2794                 lock_page(page);
2795                 if (!PageUptodate(page)) {
2796                         unlock_page(page);
2797                         page_cache_release(page);
2798                         return NULL;
2799                 }
2800         }
2801         unlock_page(page);
2802
2803         return page;
2804 }
2805
2806 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2807 {
2808         /* do any pending delalloc/csum calc on src, one way or
2809            another, and lock file content */
2810         while (1) {
2811                 struct btrfs_ordered_extent *ordered;
2812                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2813                 ordered = btrfs_lookup_first_ordered_extent(inode,
2814                                                             off + len - 1);
2815                 if ((!ordered ||
2816                      ordered->file_offset + ordered->len <= off ||
2817                      ordered->file_offset >= off + len) &&
2818                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2819                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2820                         if (ordered)
2821                                 btrfs_put_ordered_extent(ordered);
2822                         break;
2823                 }
2824                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2825                 if (ordered)
2826                         btrfs_put_ordered_extent(ordered);
2827                 btrfs_wait_ordered_range(inode, off, len);
2828         }
2829 }
2830
2831 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2832                                 struct inode *inode2, u64 loff2, u64 len)
2833 {
2834         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2835         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2836
2837         mutex_unlock(&inode1->i_mutex);
2838         mutex_unlock(&inode2->i_mutex);
2839 }
2840
2841 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2842                               struct inode *inode2, u64 loff2, u64 len)
2843 {
2844         if (inode1 < inode2) {
2845                 swap(inode1, inode2);
2846                 swap(loff1, loff2);
2847         }
2848
2849         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2850         lock_extent_range(inode1, loff1, len);
2851         if (inode1 != inode2) {
2852                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2853                 lock_extent_range(inode2, loff2, len);
2854         }
2855 }
2856
2857 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2858                           u64 dst_loff, u64 len)
2859 {
2860         int ret = 0;
2861         struct page *src_page, *dst_page;
2862         unsigned int cmp_len = PAGE_CACHE_SIZE;
2863         void *addr, *dst_addr;
2864
2865         while (len) {
2866                 if (len < PAGE_CACHE_SIZE)
2867                         cmp_len = len;
2868
2869                 src_page = extent_same_get_page(src, loff);
2870                 if (!src_page)
2871                         return -EINVAL;
2872                 dst_page = extent_same_get_page(dst, dst_loff);
2873                 if (!dst_page) {
2874                         page_cache_release(src_page);
2875                         return -EINVAL;
2876                 }
2877                 addr = kmap_atomic(src_page);
2878                 dst_addr = kmap_atomic(dst_page);
2879
2880                 flush_dcache_page(src_page);
2881                 flush_dcache_page(dst_page);
2882
2883                 if (memcmp(addr, dst_addr, cmp_len))
2884                         ret = BTRFS_SAME_DATA_DIFFERS;
2885
2886                 kunmap_atomic(addr);
2887                 kunmap_atomic(dst_addr);
2888                 page_cache_release(src_page);
2889                 page_cache_release(dst_page);
2890
2891                 if (ret)
2892                         break;
2893
2894                 loff += cmp_len;
2895                 dst_loff += cmp_len;
2896                 len -= cmp_len;
2897         }
2898
2899         return ret;
2900 }
2901
2902 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2903 {
2904         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2905
2906         if (off + len > inode->i_size || off + len < off)
2907                 return -EINVAL;
2908         /* Check that we are block aligned - btrfs_clone() requires this */
2909         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2910                 return -EINVAL;
2911
2912         return 0;
2913 }
2914
2915 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2916                              struct inode *dst, u64 dst_loff)
2917 {
2918         int ret;
2919
2920         /*
2921          * btrfs_clone() can't handle extents in the same file
2922          * yet. Once that works, we can drop this check and replace it
2923          * with a check for the same inode, but overlapping extents.
2924          */
2925         if (src == dst)
2926                 return -EINVAL;
2927
2928         btrfs_double_lock(src, loff, dst, dst_loff, len);
2929
2930         ret = extent_same_check_offsets(src, loff, len);
2931         if (ret)
2932                 goto out_unlock;
2933
2934         ret = extent_same_check_offsets(dst, dst_loff, len);
2935         if (ret)
2936                 goto out_unlock;
2937
2938         /* don't make the dst file partly checksummed */
2939         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2940             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2941                 ret = -EINVAL;
2942                 goto out_unlock;
2943         }
2944
2945         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2946         if (ret == 0)
2947                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2948
2949 out_unlock:
2950         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2951
2952         return ret;
2953 }
2954
2955 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2956
2957 static long btrfs_ioctl_file_extent_same(struct file *file,
2958                         struct btrfs_ioctl_same_args __user *argp)
2959 {
2960         struct btrfs_ioctl_same_args *same;
2961         struct btrfs_ioctl_same_extent_info *info;
2962         struct inode *src = file_inode(file);
2963         u64 off;
2964         u64 len;
2965         int i;
2966         int ret;
2967         unsigned long size;
2968         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2969         bool is_admin = capable(CAP_SYS_ADMIN);
2970         u16 count;
2971
2972         if (!(file->f_mode & FMODE_READ))
2973                 return -EINVAL;
2974
2975         ret = mnt_want_write_file(file);
2976         if (ret)
2977                 return ret;
2978
2979         if (get_user(count, &argp->dest_count)) {
2980                 ret = -EFAULT;
2981                 goto out;
2982         }
2983
2984         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2985
2986         same = memdup_user(argp, size);
2987
2988         if (IS_ERR(same)) {
2989                 ret = PTR_ERR(same);
2990                 goto out;
2991         }
2992
2993         off = same->logical_offset;
2994         len = same->length;
2995
2996         /*
2997          * Limit the total length we will dedupe for each operation.
2998          * This is intended to bound the total time spent in this
2999          * ioctl to something sane.
3000          */
3001         if (len > BTRFS_MAX_DEDUPE_LEN)
3002                 len = BTRFS_MAX_DEDUPE_LEN;
3003
3004         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3005                 /*
3006                  * Btrfs does not support blocksize < page_size. As a
3007                  * result, btrfs_cmp_data() won't correctly handle
3008                  * this situation without an update.
3009                  */
3010                 ret = -EINVAL;
3011                 goto out;
3012         }
3013
3014         ret = -EISDIR;
3015         if (S_ISDIR(src->i_mode))
3016                 goto out;
3017
3018         ret = -EACCES;
3019         if (!S_ISREG(src->i_mode))
3020                 goto out;
3021
3022         /* pre-format output fields to sane values */
3023         for (i = 0; i < count; i++) {
3024                 same->info[i].bytes_deduped = 0ULL;
3025                 same->info[i].status = 0;
3026         }
3027
3028         for (i = 0, info = same->info; i < count; i++, info++) {
3029                 struct inode *dst;
3030                 struct fd dst_file = fdget(info->fd);
3031                 if (!dst_file.file) {
3032                         info->status = -EBADF;
3033                         continue;
3034                 }
3035                 dst = file_inode(dst_file.file);
3036
3037                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3038                         info->status = -EINVAL;
3039                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3040                         info->status = -EXDEV;
3041                 } else if (S_ISDIR(dst->i_mode)) {
3042                         info->status = -EISDIR;
3043                 } else if (!S_ISREG(dst->i_mode)) {
3044                         info->status = -EACCES;
3045                 } else {
3046                         info->status = btrfs_extent_same(src, off, len, dst,
3047                                                         info->logical_offset);
3048                         if (info->status == 0)
3049                                 info->bytes_deduped += len;
3050                 }
3051                 fdput(dst_file);
3052         }
3053
3054         ret = copy_to_user(argp, same, size);
3055         if (ret)
3056                 ret = -EFAULT;
3057
3058 out:
3059         mnt_drop_write_file(file);
3060         return ret;
3061 }
3062
3063 /* Helper to check and see if this root currently has a ref on the given disk
3064  * bytenr.  If it does then we need to update the quota for this root.  This
3065  * doesn't do anything if quotas aren't enabled.
3066  */
3067 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3068                      u64 disko)
3069 {
3070         struct seq_list tree_mod_seq_elem = {};
3071         struct ulist *roots;
3072         struct ulist_iterator uiter;
3073         struct ulist_node *root_node = NULL;
3074         int ret;
3075
3076         if (!root->fs_info->quota_enabled)
3077                 return 1;
3078
3079         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3080         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3081                                    tree_mod_seq_elem.seq, &roots);
3082         if (ret < 0)
3083                 goto out;
3084         ret = 0;
3085         ULIST_ITER_INIT(&uiter);
3086         while ((root_node = ulist_next(roots, &uiter))) {
3087                 if (root_node->val == root->objectid) {
3088                         ret = 1;
3089                         break;
3090                 }
3091         }
3092         ulist_free(roots);
3093 out:
3094         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3095         return ret;
3096 }
3097
3098 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3099                                      struct inode *inode,
3100                                      u64 endoff,
3101                                      const u64 destoff,
3102                                      const u64 olen)
3103 {
3104         struct btrfs_root *root = BTRFS_I(inode)->root;
3105         int ret;
3106
3107         inode_inc_iversion(inode);
3108         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3109         /*
3110          * We round up to the block size at eof when determining which
3111          * extents to clone above, but shouldn't round up the file size.
3112          */
3113         if (endoff > destoff + olen)
3114                 endoff = destoff + olen;
3115         if (endoff > inode->i_size)
3116                 btrfs_i_size_write(inode, endoff);
3117
3118         ret = btrfs_update_inode(trans, root, inode);
3119         if (ret) {
3120                 btrfs_abort_transaction(trans, root, ret);
3121                 btrfs_end_transaction(trans, root);
3122                 goto out;
3123         }
3124         ret = btrfs_end_transaction(trans, root);
3125 out:
3126         return ret;
3127 }
3128
3129 static void clone_update_extent_map(struct inode *inode,
3130                                     const struct btrfs_trans_handle *trans,
3131                                     const struct btrfs_path *path,
3132                                     const u64 hole_offset,
3133                                     const u64 hole_len)
3134 {
3135         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3136         struct extent_map *em;
3137         int ret;
3138
3139         em = alloc_extent_map();
3140         if (!em) {
3141                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3142                         &BTRFS_I(inode)->runtime_flags);
3143                 return;
3144         }
3145
3146         if (path) {
3147                 struct btrfs_file_extent_item *fi;
3148
3149                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3150                                     struct btrfs_file_extent_item);
3151                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3152                 em->generation = -1;
3153                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3154                     BTRFS_FILE_EXTENT_INLINE)
3155                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3156                                 &BTRFS_I(inode)->runtime_flags);
3157         } else {
3158                 em->start = hole_offset;
3159                 em->len = hole_len;
3160                 em->ram_bytes = em->len;
3161                 em->orig_start = hole_offset;
3162                 em->block_start = EXTENT_MAP_HOLE;
3163                 em->block_len = 0;
3164                 em->orig_block_len = 0;
3165                 em->compress_type = BTRFS_COMPRESS_NONE;
3166                 em->generation = trans->transid;
3167         }
3168
3169         while (1) {
3170                 write_lock(&em_tree->lock);
3171                 ret = add_extent_mapping(em_tree, em, 1);
3172                 write_unlock(&em_tree->lock);
3173                 if (ret != -EEXIST) {
3174                         free_extent_map(em);
3175                         break;
3176                 }
3177                 btrfs_drop_extent_cache(inode, em->start,
3178                                         em->start + em->len - 1, 0);
3179         }
3180
3181         if (ret)
3182                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3183                         &BTRFS_I(inode)->runtime_flags);
3184 }
3185
3186 /**
3187  * btrfs_clone() - clone a range from inode file to another
3188  *
3189  * @src: Inode to clone from
3190  * @inode: Inode to clone to
3191  * @off: Offset within source to start clone from
3192  * @olen: Original length, passed by user, of range to clone
3193  * @olen_aligned: Block-aligned value of olen, extent_same uses
3194  *               identical values here
3195  * @destoff: Offset within @inode to start clone
3196  */
3197 static int btrfs_clone(struct inode *src, struct inode *inode,
3198                        const u64 off, const u64 olen, const u64 olen_aligned,
3199                        const u64 destoff)
3200 {
3201         struct btrfs_root *root = BTRFS_I(inode)->root;
3202         struct btrfs_path *path = NULL;
3203         struct extent_buffer *leaf;
3204         struct btrfs_trans_handle *trans;
3205         char *buf = NULL;
3206         struct btrfs_key key;
3207         u32 nritems;
3208         int slot;
3209         int ret;
3210         int no_quota;
3211         const u64 len = olen_aligned;
3212         u64 last_disko = 0;
3213         u64 last_dest_end = destoff;
3214
3215         ret = -ENOMEM;
3216         buf = vmalloc(root->nodesize);
3217         if (!buf)
3218                 return ret;
3219
3220         path = btrfs_alloc_path();
3221         if (!path) {
3222                 vfree(buf);
3223                 return ret;
3224         }
3225
3226         path->reada = 2;
3227         /* clone data */
3228         key.objectid = btrfs_ino(src);
3229         key.type = BTRFS_EXTENT_DATA_KEY;
3230         key.offset = off;
3231
3232         while (1) {
3233                 /*
3234                  * note the key will change type as we walk through the
3235                  * tree.
3236                  */
3237                 path->leave_spinning = 1;
3238                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3239                                 0, 0);
3240                 if (ret < 0)
3241                         goto out;
3242                 /*
3243                  * First search, if no extent item that starts at offset off was
3244                  * found but the previous item is an extent item, it's possible
3245                  * it might overlap our target range, therefore process it.
3246                  */
3247                 if (key.offset == off && ret &g