Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
[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 check_sticky in fs/namei.c()
769 * It's inline, so penalty for filesystems that don't use sticky bit is
770 * minimal.
771 */
772 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
773 {
774         kuid_t fsuid = current_fsuid();
775
776         if (!(dir->i_mode & S_ISVTX))
777                 return 0;
778         if (uid_eq(inode->i_uid, fsuid))
779                 return 0;
780         if (uid_eq(dir->i_uid, fsuid))
781                 return 0;
782         return !capable(CAP_FOWNER);
783 }
784
785 /*  copy of may_delete in fs/namei.c()
786  *      Check whether we can remove a link victim from directory dir, check
787  *  whether the type of victim is right.
788  *  1. We can't do it if dir is read-only (done in permission())
789  *  2. We should have write and exec permissions on dir
790  *  3. We can't remove anything from append-only dir
791  *  4. We can't do anything with immutable dir (done in permission())
792  *  5. If the sticky bit on dir is set we should either
793  *      a. be owner of dir, or
794  *      b. be owner of victim, or
795  *      c. have CAP_FOWNER capability
796  *  6. If the victim is append-only or immutable we can't do antyhing with
797  *     links pointing to it.
798  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
799  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
800  *  9. We can't remove a root or mountpoint.
801  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
802  *     nfs_async_unlink().
803  */
804
805 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
806 {
807         int error;
808
809         if (!victim->d_inode)
810                 return -ENOENT;
811
812         BUG_ON(victim->d_parent->d_inode != dir);
813         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
814
815         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
816         if (error)
817                 return error;
818         if (IS_APPEND(dir))
819                 return -EPERM;
820         if (btrfs_check_sticky(dir, victim->d_inode)||
821                 IS_APPEND(victim->d_inode)||
822             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
823                 return -EPERM;
824         if (isdir) {
825                 if (!S_ISDIR(victim->d_inode->i_mode))
826                         return -ENOTDIR;
827                 if (IS_ROOT(victim))
828                         return -EBUSY;
829         } else if (S_ISDIR(victim->d_inode->i_mode))
830                 return -EISDIR;
831         if (IS_DEADDIR(dir))
832                 return -ENOENT;
833         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
834                 return -EBUSY;
835         return 0;
836 }
837
838 /* copy of may_create in fs/namei.c() */
839 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
840 {
841         if (child->d_inode)
842                 return -EEXIST;
843         if (IS_DEADDIR(dir))
844                 return -ENOENT;
845         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
846 }
847
848 /*
849  * Create a new subvolume below @parent.  This is largely modeled after
850  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
851  * inside this filesystem so it's quite a bit simpler.
852  */
853 static noinline int btrfs_mksubvol(struct path *parent,
854                                    char *name, int namelen,
855                                    struct btrfs_root *snap_src,
856                                    u64 *async_transid, bool readonly,
857                                    struct btrfs_qgroup_inherit *inherit)
858 {
859         struct inode *dir  = parent->dentry->d_inode;
860         struct dentry *dentry;
861         int error;
862
863         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
864         if (error == -EINTR)
865                 return error;
866
867         dentry = lookup_one_len(name, parent->dentry, namelen);
868         error = PTR_ERR(dentry);
869         if (IS_ERR(dentry))
870                 goto out_unlock;
871
872         error = -EEXIST;
873         if (dentry->d_inode)
874                 goto out_dput;
875
876         error = btrfs_may_create(dir, dentry);
877         if (error)
878                 goto out_dput;
879
880         /*
881          * even if this name doesn't exist, we may get hash collisions.
882          * check for them now when we can safely fail
883          */
884         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
885                                                dir->i_ino, name,
886                                                namelen);
887         if (error)
888                 goto out_dput;
889
890         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
891
892         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
893                 goto out_up_read;
894
895         if (snap_src) {
896                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
897                                         async_transid, readonly, inherit);
898         } else {
899                 error = create_subvol(dir, dentry, name, namelen,
900                                       async_transid, inherit);
901         }
902         if (!error)
903                 fsnotify_mkdir(dir, dentry);
904 out_up_read:
905         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
906 out_dput:
907         dput(dentry);
908 out_unlock:
909         mutex_unlock(&dir->i_mutex);
910         return error;
911 }
912
913 /*
914  * When we're defragging a range, we don't want to kick it off again
915  * if it is really just waiting for delalloc to send it down.
916  * If we find a nice big extent or delalloc range for the bytes in the
917  * file you want to defrag, we return 0 to let you know to skip this
918  * part of the file
919  */
920 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
921 {
922         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
923         struct extent_map *em = NULL;
924         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
925         u64 end;
926
927         read_lock(&em_tree->lock);
928         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
929         read_unlock(&em_tree->lock);
930
931         if (em) {
932                 end = extent_map_end(em);
933                 free_extent_map(em);
934                 if (end - offset > thresh)
935                         return 0;
936         }
937         /* if we already have a nice delalloc here, just stop */
938         thresh /= 2;
939         end = count_range_bits(io_tree, &offset, offset + thresh,
940                                thresh, EXTENT_DELALLOC, 1);
941         if (end >= thresh)
942                 return 0;
943         return 1;
944 }
945
946 /*
947  * helper function to walk through a file and find extents
948  * newer than a specific transid, and smaller than thresh.
949  *
950  * This is used by the defragging code to find new and small
951  * extents
952  */
953 static int find_new_extents(struct btrfs_root *root,
954                             struct inode *inode, u64 newer_than,
955                             u64 *off, u32 thresh)
956 {
957         struct btrfs_path *path;
958         struct btrfs_key min_key;
959         struct extent_buffer *leaf;
960         struct btrfs_file_extent_item *extent;
961         int type;
962         int ret;
963         u64 ino = btrfs_ino(inode);
964
965         path = btrfs_alloc_path();
966         if (!path)
967                 return -ENOMEM;
968
969         min_key.objectid = ino;
970         min_key.type = BTRFS_EXTENT_DATA_KEY;
971         min_key.offset = *off;
972
973         while (1) {
974                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
975                 if (ret != 0)
976                         goto none;
977 process_slot:
978                 if (min_key.objectid != ino)
979                         goto none;
980                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
981                         goto none;
982
983                 leaf = path->nodes[0];
984                 extent = btrfs_item_ptr(leaf, path->slots[0],
985                                         struct btrfs_file_extent_item);
986
987                 type = btrfs_file_extent_type(leaf, extent);
988                 if (type == BTRFS_FILE_EXTENT_REG &&
989                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
990                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
991                         *off = min_key.offset;
992                         btrfs_free_path(path);
993                         return 0;
994                 }
995
996                 path->slots[0]++;
997                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
998                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
999                         goto process_slot;
1000                 }
1001
1002                 if (min_key.offset == (u64)-1)
1003                         goto none;
1004
1005                 min_key.offset++;
1006                 btrfs_release_path(path);
1007         }
1008 none:
1009         btrfs_free_path(path);
1010         return -ENOENT;
1011 }
1012
1013 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1014 {
1015         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1016         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1017         struct extent_map *em;
1018         u64 len = PAGE_CACHE_SIZE;
1019
1020         /*
1021          * hopefully we have this extent in the tree already, try without
1022          * the full extent lock
1023          */
1024         read_lock(&em_tree->lock);
1025         em = lookup_extent_mapping(em_tree, start, len);
1026         read_unlock(&em_tree->lock);
1027
1028         if (!em) {
1029                 struct extent_state *cached = NULL;
1030                 u64 end = start + len - 1;
1031
1032                 /* get the big lock and read metadata off disk */
1033                 lock_extent_bits(io_tree, start, end, 0, &cached);
1034                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1035                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1036
1037                 if (IS_ERR(em))
1038                         return NULL;
1039         }
1040
1041         return em;
1042 }
1043
1044 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1045 {
1046         struct extent_map *next;
1047         bool ret = true;
1048
1049         /* this is the last extent */
1050         if (em->start + em->len >= i_size_read(inode))
1051                 return false;
1052
1053         next = defrag_lookup_extent(inode, em->start + em->len);
1054         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1055                 ret = false;
1056         else if ((em->block_start + em->block_len == next->block_start) &&
1057                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1058                 ret = false;
1059
1060         free_extent_map(next);
1061         return ret;
1062 }
1063
1064 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1065                                u64 *last_len, u64 *skip, u64 *defrag_end,
1066                                int compress)
1067 {
1068         struct extent_map *em;
1069         int ret = 1;
1070         bool next_mergeable = true;
1071
1072         /*
1073          * make sure that once we start defragging an extent, we keep on
1074          * defragging it
1075          */
1076         if (start < *defrag_end)
1077                 return 1;
1078
1079         *skip = 0;
1080
1081         em = defrag_lookup_extent(inode, start);
1082         if (!em)
1083                 return 0;
1084
1085         /* this will cover holes, and inline extents */
1086         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1087                 ret = 0;
1088                 goto out;
1089         }
1090
1091         next_mergeable = defrag_check_next_extent(inode, em);
1092         /*
1093          * we hit a real extent, if it is big or the next extent is not a
1094          * real extent, don't bother defragging it
1095          */
1096         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1097             (em->len >= thresh || !next_mergeable))
1098                 ret = 0;
1099 out:
1100         /*
1101          * last_len ends up being a counter of how many bytes we've defragged.
1102          * every time we choose not to defrag an extent, we reset *last_len
1103          * so that the next tiny extent will force a defrag.
1104          *
1105          * The end result of this is that tiny extents before a single big
1106          * extent will force at least part of that big extent to be defragged.
1107          */
1108         if (ret) {
1109                 *defrag_end = extent_map_end(em);
1110         } else {
1111                 *last_len = 0;
1112                 *skip = extent_map_end(em);
1113                 *defrag_end = 0;
1114         }
1115
1116         free_extent_map(em);
1117         return ret;
1118 }
1119
1120 /*
1121  * it doesn't do much good to defrag one or two pages
1122  * at a time.  This pulls in a nice chunk of pages
1123  * to COW and defrag.
1124  *
1125  * It also makes sure the delalloc code has enough
1126  * dirty data to avoid making new small extents as part
1127  * of the defrag
1128  *
1129  * It's a good idea to start RA on this range
1130  * before calling this.
1131  */
1132 static int cluster_pages_for_defrag(struct inode *inode,
1133                                     struct page **pages,
1134                                     unsigned long start_index,
1135                                     unsigned long num_pages)
1136 {
1137         unsigned long file_end;
1138         u64 isize = i_size_read(inode);
1139         u64 page_start;
1140         u64 page_end;
1141         u64 page_cnt;
1142         int ret;
1143         int i;
1144         int i_done;
1145         struct btrfs_ordered_extent *ordered;
1146         struct extent_state *cached_state = NULL;
1147         struct extent_io_tree *tree;
1148         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1149
1150         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1151         if (!isize || start_index > file_end)
1152                 return 0;
1153
1154         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1155
1156         ret = btrfs_delalloc_reserve_space(inode,
1157                                            page_cnt << PAGE_CACHE_SHIFT);
1158         if (ret)
1159                 return ret;
1160         i_done = 0;
1161         tree = &BTRFS_I(inode)->io_tree;
1162
1163         /* step one, lock all the pages */
1164         for (i = 0; i < page_cnt; i++) {
1165                 struct page *page;
1166 again:
1167                 page = find_or_create_page(inode->i_mapping,
1168                                            start_index + i, mask);
1169                 if (!page)
1170                         break;
1171
1172                 page_start = page_offset(page);
1173                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1174                 while (1) {
1175                         lock_extent_bits(tree, page_start, page_end,
1176                                          0, &cached_state);
1177                         ordered = btrfs_lookup_ordered_extent(inode,
1178                                                               page_start);
1179                         unlock_extent_cached(tree, page_start, page_end,
1180                                              &cached_state, GFP_NOFS);
1181                         if (!ordered)
1182                                 break;
1183
1184                         unlock_page(page);
1185                         btrfs_start_ordered_extent(inode, ordered, 1);
1186                         btrfs_put_ordered_extent(ordered);
1187                         lock_page(page);
1188                         /*
1189                          * we unlocked the page above, so we need check if
1190                          * it was released or not.
1191                          */
1192                         if (page->mapping != inode->i_mapping) {
1193                                 unlock_page(page);
1194                                 page_cache_release(page);
1195                                 goto again;
1196                         }
1197                 }
1198
1199                 if (!PageUptodate(page)) {
1200                         btrfs_readpage(NULL, page);
1201                         lock_page(page);
1202                         if (!PageUptodate(page)) {
1203                                 unlock_page(page);
1204                                 page_cache_release(page);
1205                                 ret = -EIO;
1206                                 break;
1207                         }
1208                 }
1209
1210                 if (page->mapping != inode->i_mapping) {
1211                         unlock_page(page);
1212                         page_cache_release(page);
1213                         goto again;
1214                 }
1215
1216                 pages[i] = page;
1217                 i_done++;
1218         }
1219         if (!i_done || ret)
1220                 goto out;
1221
1222         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1223                 goto out;
1224
1225         /*
1226          * so now we have a nice long stream of locked
1227          * and up to date pages, lets wait on them
1228          */
1229         for (i = 0; i < i_done; i++)
1230                 wait_on_page_writeback(pages[i]);
1231
1232         page_start = page_offset(pages[0]);
1233         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1234
1235         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1236                          page_start, page_end - 1, 0, &cached_state);
1237         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1238                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1239                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1240                           &cached_state, GFP_NOFS);
1241
1242         if (i_done != page_cnt) {
1243                 spin_lock(&BTRFS_I(inode)->lock);
1244                 BTRFS_I(inode)->outstanding_extents++;
1245                 spin_unlock(&BTRFS_I(inode)->lock);
1246                 btrfs_delalloc_release_space(inode,
1247                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1248         }
1249
1250
1251         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1252                           &cached_state, GFP_NOFS);
1253
1254         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1255                              page_start, page_end - 1, &cached_state,
1256                              GFP_NOFS);
1257
1258         for (i = 0; i < i_done; i++) {
1259                 clear_page_dirty_for_io(pages[i]);
1260                 ClearPageChecked(pages[i]);
1261                 set_page_extent_mapped(pages[i]);
1262                 set_page_dirty(pages[i]);
1263                 unlock_page(pages[i]);
1264                 page_cache_release(pages[i]);
1265         }
1266         return i_done;
1267 out:
1268         for (i = 0; i < i_done; i++) {
1269                 unlock_page(pages[i]);
1270                 page_cache_release(pages[i]);
1271         }
1272         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1273         return ret;
1274
1275 }
1276
1277 int btrfs_defrag_file(struct inode *inode, struct file *file,
1278                       struct btrfs_ioctl_defrag_range_args *range,
1279                       u64 newer_than, unsigned long max_to_defrag)
1280 {
1281         struct btrfs_root *root = BTRFS_I(inode)->root;
1282         struct file_ra_state *ra = NULL;
1283         unsigned long last_index;
1284         u64 isize = i_size_read(inode);
1285         u64 last_len = 0;
1286         u64 skip = 0;
1287         u64 defrag_end = 0;
1288         u64 newer_off = range->start;
1289         unsigned long i;
1290         unsigned long ra_index = 0;
1291         int ret;
1292         int defrag_count = 0;
1293         int compress_type = BTRFS_COMPRESS_ZLIB;
1294         u32 extent_thresh = range->extent_thresh;
1295         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1296         unsigned long cluster = max_cluster;
1297         u64 new_align = ~((u64)128 * 1024 - 1);
1298         struct page **pages = NULL;
1299
1300         if (isize == 0)
1301                 return 0;
1302
1303         if (range->start >= isize)
1304                 return -EINVAL;
1305
1306         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1307                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1308                         return -EINVAL;
1309                 if (range->compress_type)
1310                         compress_type = range->compress_type;
1311         }
1312
1313         if (extent_thresh == 0)
1314                 extent_thresh = 256 * 1024;
1315
1316         /*
1317          * if we were not given a file, allocate a readahead
1318          * context
1319          */
1320         if (!file) {
1321                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1322                 if (!ra)
1323                         return -ENOMEM;
1324                 file_ra_state_init(ra, inode->i_mapping);
1325         } else {
1326                 ra = &file->f_ra;
1327         }
1328
1329         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1330                         GFP_NOFS);
1331         if (!pages) {
1332                 ret = -ENOMEM;
1333                 goto out_ra;
1334         }
1335
1336         /* find the last page to defrag */
1337         if (range->start + range->len > range->start) {
1338                 last_index = min_t(u64, isize - 1,
1339                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1340         } else {
1341                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1342         }
1343
1344         if (newer_than) {
1345                 ret = find_new_extents(root, inode, newer_than,
1346                                        &newer_off, 64 * 1024);
1347                 if (!ret) {
1348                         range->start = newer_off;
1349                         /*
1350                          * we always align our defrag to help keep
1351                          * the extents in the file evenly spaced
1352                          */
1353                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1354                 } else
1355                         goto out_ra;
1356         } else {
1357                 i = range->start >> PAGE_CACHE_SHIFT;
1358         }
1359         if (!max_to_defrag)
1360                 max_to_defrag = last_index + 1;
1361
1362         /*
1363          * make writeback starts from i, so the defrag range can be
1364          * written sequentially.
1365          */
1366         if (i < inode->i_mapping->writeback_index)
1367                 inode->i_mapping->writeback_index = i;
1368
1369         while (i <= last_index && defrag_count < max_to_defrag &&
1370                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1371                 /*
1372                  * make sure we stop running if someone unmounts
1373                  * the FS
1374                  */
1375                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1376                         break;
1377
1378                 if (btrfs_defrag_cancelled(root->fs_info)) {
1379                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1380                         ret = -EAGAIN;
1381                         break;
1382                 }
1383
1384                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1385                                          extent_thresh, &last_len, &skip,
1386                                          &defrag_end, range->flags &
1387                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1388                         unsigned long next;
1389                         /*
1390                          * the should_defrag function tells us how much to skip
1391                          * bump our counter by the suggested amount
1392                          */
1393                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1394                         i = max(i + 1, next);
1395                         continue;
1396                 }
1397
1398                 if (!newer_than) {
1399                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1400                                    PAGE_CACHE_SHIFT) - i;
1401                         cluster = min(cluster, max_cluster);
1402                 } else {
1403                         cluster = max_cluster;
1404                 }
1405
1406                 if (i + cluster > ra_index) {
1407                         ra_index = max(i, ra_index);
1408                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1409                                        cluster);
1410                         ra_index += max_cluster;
1411                 }
1412
1413                 mutex_lock(&inode->i_mutex);
1414                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1415                         BTRFS_I(inode)->force_compress = compress_type;
1416                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1417                 if (ret < 0) {
1418                         mutex_unlock(&inode->i_mutex);
1419                         goto out_ra;
1420                 }
1421
1422                 defrag_count += ret;
1423                 balance_dirty_pages_ratelimited(inode->i_mapping);
1424                 mutex_unlock(&inode->i_mutex);
1425
1426                 if (newer_than) {
1427                         if (newer_off == (u64)-1)
1428                                 break;
1429
1430                         if (ret > 0)
1431                                 i += ret;
1432
1433                         newer_off = max(newer_off + 1,
1434                                         (u64)i << PAGE_CACHE_SHIFT);
1435
1436                         ret = find_new_extents(root, inode,
1437                                                newer_than, &newer_off,
1438                                                64 * 1024);
1439                         if (!ret) {
1440                                 range->start = newer_off;
1441                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1442                         } else {
1443                                 break;
1444                         }
1445                 } else {
1446                         if (ret > 0) {
1447                                 i += ret;
1448                                 last_len += ret << PAGE_CACHE_SHIFT;
1449                         } else {
1450                                 i++;
1451                                 last_len = 0;
1452                         }
1453                 }
1454         }
1455
1456         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1457                 filemap_flush(inode->i_mapping);
1458                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1459                              &BTRFS_I(inode)->runtime_flags))
1460                         filemap_flush(inode->i_mapping);
1461         }
1462
1463         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1464                 /* the filemap_flush will queue IO into the worker threads, but
1465                  * we have to make sure the IO is actually started and that
1466                  * ordered extents get created before we return
1467                  */
1468                 atomic_inc(&root->fs_info->async_submit_draining);
1469                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1470                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1471                         wait_event(root->fs_info->async_submit_wait,
1472                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1473                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1474                 }
1475                 atomic_dec(&root->fs_info->async_submit_draining);
1476         }
1477
1478         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1479                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1480         }
1481
1482         ret = defrag_count;
1483
1484 out_ra:
1485         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1486                 mutex_lock(&inode->i_mutex);
1487                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1488                 mutex_unlock(&inode->i_mutex);
1489         }
1490         if (!file)
1491                 kfree(ra);
1492         kfree(pages);
1493         return ret;
1494 }
1495
1496 static noinline int btrfs_ioctl_resize(struct file *file,
1497                                         void __user *arg)
1498 {
1499         u64 new_size;
1500         u64 old_size;
1501         u64 devid = 1;
1502         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1503         struct btrfs_ioctl_vol_args *vol_args;
1504         struct btrfs_trans_handle *trans;
1505         struct btrfs_device *device = NULL;
1506         char *sizestr;
1507         char *retptr;
1508         char *devstr = NULL;
1509         int ret = 0;
1510         int mod = 0;
1511
1512         if (!capable(CAP_SYS_ADMIN))
1513                 return -EPERM;
1514
1515         ret = mnt_want_write_file(file);
1516         if (ret)
1517                 return ret;
1518
1519         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1520                         1)) {
1521                 mnt_drop_write_file(file);
1522                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1523         }
1524
1525         mutex_lock(&root->fs_info->volume_mutex);
1526         vol_args = memdup_user(arg, sizeof(*vol_args));
1527         if (IS_ERR(vol_args)) {
1528                 ret = PTR_ERR(vol_args);
1529                 goto out;
1530         }
1531
1532         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1533
1534         sizestr = vol_args->name;
1535         devstr = strchr(sizestr, ':');
1536         if (devstr) {
1537                 sizestr = devstr + 1;
1538                 *devstr = '\0';
1539                 devstr = vol_args->name;
1540                 ret = kstrtoull(devstr, 10, &devid);
1541                 if (ret)
1542                         goto out_free;
1543                 if (!devid) {
1544                         ret = -EINVAL;
1545                         goto out_free;
1546                 }
1547                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1548         }
1549
1550         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1551         if (!device) {
1552                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1553                        devid);
1554                 ret = -ENODEV;
1555                 goto out_free;
1556         }
1557
1558         if (!device->writeable) {
1559                 btrfs_info(root->fs_info,
1560                            "resizer unable to apply on readonly device %llu",
1561                        devid);
1562                 ret = -EPERM;
1563                 goto out_free;
1564         }
1565
1566         if (!strcmp(sizestr, "max"))
1567                 new_size = device->bdev->bd_inode->i_size;
1568         else {
1569                 if (sizestr[0] == '-') {
1570                         mod = -1;
1571                         sizestr++;
1572                 } else if (sizestr[0] == '+') {
1573                         mod = 1;
1574                         sizestr++;
1575                 }
1576                 new_size = memparse(sizestr, &retptr);
1577                 if (*retptr != '\0' || new_size == 0) {
1578                         ret = -EINVAL;
1579                         goto out_free;
1580                 }
1581         }
1582
1583         if (device->is_tgtdev_for_dev_replace) {
1584                 ret = -EPERM;
1585                 goto out_free;
1586         }
1587
1588         old_size = btrfs_device_get_total_bytes(device);
1589
1590         if (mod < 0) {
1591                 if (new_size > old_size) {
1592                         ret = -EINVAL;
1593                         goto out_free;
1594                 }
1595                 new_size = old_size - new_size;
1596         } else if (mod > 0) {
1597                 if (new_size > ULLONG_MAX - old_size) {
1598                         ret = -ERANGE;
1599                         goto out_free;
1600                 }
1601                 new_size = old_size + new_size;
1602         }
1603
1604         if (new_size < 256 * 1024 * 1024) {
1605                 ret = -EINVAL;
1606                 goto out_free;
1607         }
1608         if (new_size > device->bdev->bd_inode->i_size) {
1609                 ret = -EFBIG;
1610                 goto out_free;
1611         }
1612
1613         do_div(new_size, root->sectorsize);
1614         new_size *= root->sectorsize;
1615
1616         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1617                       rcu_str_deref(device->name), new_size);
1618
1619         if (new_size > old_size) {
1620                 trans = btrfs_start_transaction(root, 0);
1621                 if (IS_ERR(trans)) {
1622                         ret = PTR_ERR(trans);
1623                         goto out_free;
1624                 }
1625                 ret = btrfs_grow_device(trans, device, new_size);
1626                 btrfs_commit_transaction(trans, root);
1627         } else if (new_size < old_size) {
1628                 ret = btrfs_shrink_device(device, new_size);
1629         } /* equal, nothing need to do */
1630
1631 out_free:
1632         kfree(vol_args);
1633 out:
1634         mutex_unlock(&root->fs_info->volume_mutex);
1635         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1636         mnt_drop_write_file(file);
1637         return ret;
1638 }
1639
1640 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1641                                 char *name, unsigned long fd, int subvol,
1642                                 u64 *transid, bool readonly,
1643                                 struct btrfs_qgroup_inherit *inherit)
1644 {
1645         int namelen;
1646         int ret = 0;
1647
1648         ret = mnt_want_write_file(file);
1649         if (ret)
1650                 goto out;
1651
1652         namelen = strlen(name);
1653         if (strchr(name, '/')) {
1654                 ret = -EINVAL;
1655                 goto out_drop_write;
1656         }
1657
1658         if (name[0] == '.' &&
1659            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1660                 ret = -EEXIST;
1661                 goto out_drop_write;
1662         }
1663
1664         if (subvol) {
1665                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1666                                      NULL, transid, readonly, inherit);
1667         } else {
1668                 struct fd src = fdget(fd);
1669                 struct inode *src_inode;
1670                 if (!src.file) {
1671                         ret = -EINVAL;
1672                         goto out_drop_write;
1673                 }
1674
1675                 src_inode = file_inode(src.file);
1676                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1677                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1678                                    "Snapshot src from another FS");
1679                         ret = -EXDEV;
1680                 } else if (!inode_owner_or_capable(src_inode)) {
1681                         /*
1682                          * Subvolume creation is not restricted, but snapshots
1683                          * are limited to own subvolumes only
1684                          */
1685                         ret = -EPERM;
1686                 } else {
1687                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1688                                              BTRFS_I(src_inode)->root,
1689                                              transid, readonly, inherit);
1690                 }
1691                 fdput(src);
1692         }
1693 out_drop_write:
1694         mnt_drop_write_file(file);
1695 out:
1696         return ret;
1697 }
1698
1699 static noinline int btrfs_ioctl_snap_create(struct file *file,
1700                                             void __user *arg, int subvol)
1701 {
1702         struct btrfs_ioctl_vol_args *vol_args;
1703         int ret;
1704
1705         vol_args = memdup_user(arg, sizeof(*vol_args));
1706         if (IS_ERR(vol_args))
1707                 return PTR_ERR(vol_args);
1708         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1709
1710         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1711                                               vol_args->fd, subvol,
1712                                               NULL, false, NULL);
1713
1714         kfree(vol_args);
1715         return ret;
1716 }
1717
1718 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1719                                                void __user *arg, int subvol)
1720 {
1721         struct btrfs_ioctl_vol_args_v2 *vol_args;
1722         int ret;
1723         u64 transid = 0;
1724         u64 *ptr = NULL;
1725         bool readonly = false;
1726         struct btrfs_qgroup_inherit *inherit = NULL;
1727
1728         vol_args = memdup_user(arg, sizeof(*vol_args));
1729         if (IS_ERR(vol_args))
1730                 return PTR_ERR(vol_args);
1731         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1732
1733         if (vol_args->flags &
1734             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1735               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1736                 ret = -EOPNOTSUPP;
1737                 goto free_args;
1738         }
1739
1740         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1741                 ptr = &transid;
1742         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1743                 readonly = true;
1744         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1745                 if (vol_args->size > PAGE_CACHE_SIZE) {
1746                         ret = -EINVAL;
1747                         goto free_args;
1748                 }
1749                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1750                 if (IS_ERR(inherit)) {
1751                         ret = PTR_ERR(inherit);
1752                         goto free_args;
1753                 }
1754         }
1755
1756         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1757                                               vol_args->fd, subvol, ptr,
1758                                               readonly, inherit);
1759         if (ret)
1760                 goto free_inherit;
1761
1762         if (ptr && copy_to_user(arg +
1763                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1764                                         transid),
1765                                 ptr, sizeof(*ptr)))
1766                 ret = -EFAULT;
1767
1768 free_inherit:
1769         kfree(inherit);
1770 free_args:
1771         kfree(vol_args);
1772         return ret;
1773 }
1774
1775 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1776                                                 void __user *arg)
1777 {
1778         struct inode *inode = file_inode(file);
1779         struct btrfs_root *root = BTRFS_I(inode)->root;
1780         int ret = 0;
1781         u64 flags = 0;
1782
1783         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1784                 return -EINVAL;
1785
1786         down_read(&root->fs_info->subvol_sem);
1787         if (btrfs_root_readonly(root))
1788                 flags |= BTRFS_SUBVOL_RDONLY;
1789         up_read(&root->fs_info->subvol_sem);
1790
1791         if (copy_to_user(arg, &flags, sizeof(flags)))
1792                 ret = -EFAULT;
1793
1794         return ret;
1795 }
1796
1797 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1798                                               void __user *arg)
1799 {
1800         struct inode *inode = file_inode(file);
1801         struct btrfs_root *root = BTRFS_I(inode)->root;
1802         struct btrfs_trans_handle *trans;
1803         u64 root_flags;
1804         u64 flags;
1805         int ret = 0;
1806
1807         if (!inode_owner_or_capable(inode))
1808                 return -EPERM;
1809
1810         ret = mnt_want_write_file(file);
1811         if (ret)
1812                 goto out;
1813
1814         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1815                 ret = -EINVAL;
1816                 goto out_drop_write;
1817         }
1818
1819         if (copy_from_user(&flags, arg, sizeof(flags))) {
1820                 ret = -EFAULT;
1821                 goto out_drop_write;
1822         }
1823
1824         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1825                 ret = -EINVAL;
1826                 goto out_drop_write;
1827         }
1828
1829         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1830                 ret = -EOPNOTSUPP;
1831                 goto out_drop_write;
1832         }
1833
1834         down_write(&root->fs_info->subvol_sem);
1835
1836         /* nothing to do */
1837         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1838                 goto out_drop_sem;
1839
1840         root_flags = btrfs_root_flags(&root->root_item);
1841         if (flags & BTRFS_SUBVOL_RDONLY) {
1842                 btrfs_set_root_flags(&root->root_item,
1843                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1844         } else {
1845                 /*
1846                  * Block RO -> RW transition if this subvolume is involved in
1847                  * send
1848                  */
1849                 spin_lock(&root->root_item_lock);
1850                 if (root->send_in_progress == 0) {
1851                         btrfs_set_root_flags(&root->root_item,
1852                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1853                         spin_unlock(&root->root_item_lock);
1854                 } else {
1855                         spin_unlock(&root->root_item_lock);
1856                         btrfs_warn(root->fs_info,
1857                         "Attempt to set subvolume %llu read-write during send",
1858                                         root->root_key.objectid);
1859                         ret = -EPERM;
1860                         goto out_drop_sem;
1861                 }
1862         }
1863
1864         trans = btrfs_start_transaction(root, 1);
1865         if (IS_ERR(trans)) {
1866                 ret = PTR_ERR(trans);
1867                 goto out_reset;
1868         }
1869
1870         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1871                                 &root->root_key, &root->root_item);
1872
1873         btrfs_commit_transaction(trans, root);
1874 out_reset:
1875         if (ret)
1876                 btrfs_set_root_flags(&root->root_item, root_flags);
1877 out_drop_sem:
1878         up_write(&root->fs_info->subvol_sem);
1879 out_drop_write:
1880         mnt_drop_write_file(file);
1881 out:
1882         return ret;
1883 }
1884
1885 /*
1886  * helper to check if the subvolume references other subvolumes
1887  */
1888 static noinline int may_destroy_subvol(struct btrfs_root *root)
1889 {
1890         struct btrfs_path *path;
1891         struct btrfs_dir_item *di;
1892         struct btrfs_key key;
1893         u64 dir_id;
1894         int ret;
1895
1896         path = btrfs_alloc_path();
1897         if (!path)
1898                 return -ENOMEM;
1899
1900         /* Make sure this root isn't set as the default subvol */
1901         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1902         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1903                                    dir_id, "default", 7, 0);
1904         if (di && !IS_ERR(di)) {
1905                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1906                 if (key.objectid == root->root_key.objectid) {
1907                         ret = -EPERM;
1908                         btrfs_err(root->fs_info, "deleting default subvolume "
1909                                   "%llu is not allowed", key.objectid);
1910                         goto out;
1911                 }
1912                 btrfs_release_path(path);
1913         }
1914
1915         key.objectid = root->root_key.objectid;
1916         key.type = BTRFS_ROOT_REF_KEY;
1917         key.offset = (u64)-1;
1918
1919         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1920                                 &key, path, 0, 0);
1921         if (ret < 0)
1922                 goto out;
1923         BUG_ON(ret == 0);
1924
1925         ret = 0;
1926         if (path->slots[0] > 0) {
1927                 path->slots[0]--;
1928                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1929                 if (key.objectid == root->root_key.objectid &&
1930                     key.type == BTRFS_ROOT_REF_KEY)
1931                         ret = -ENOTEMPTY;
1932         }
1933 out:
1934         btrfs_free_path(path);
1935         return ret;
1936 }
1937
1938 static noinline int key_in_sk(struct btrfs_key *key,
1939                               struct btrfs_ioctl_search_key *sk)
1940 {
1941         struct btrfs_key test;
1942         int ret;
1943
1944         test.objectid = sk->min_objectid;
1945         test.type = sk->min_type;
1946         test.offset = sk->min_offset;
1947
1948         ret = btrfs_comp_cpu_keys(key, &test);
1949         if (ret < 0)
1950                 return 0;
1951
1952         test.objectid = sk->max_objectid;
1953         test.type = sk->max_type;
1954         test.offset = sk->max_offset;
1955
1956         ret = btrfs_comp_cpu_keys(key, &test);
1957         if (ret > 0)
1958                 return 0;
1959         return 1;
1960 }
1961
1962 static noinline int copy_to_sk(struct btrfs_root *root,
1963                                struct btrfs_path *path,
1964                                struct btrfs_key *key,
1965                                struct btrfs_ioctl_search_key *sk,
1966                                size_t *buf_size,
1967                                char __user *ubuf,
1968                                unsigned long *sk_offset,
1969                                int *num_found)
1970 {
1971         u64 found_transid;
1972         struct extent_buffer *leaf;
1973         struct btrfs_ioctl_search_header sh;
1974         unsigned long item_off;
1975         unsigned long item_len;
1976         int nritems;
1977         int i;
1978         int slot;
1979         int ret = 0;
1980
1981         leaf = path->nodes[0];
1982         slot = path->slots[0];
1983         nritems = btrfs_header_nritems(leaf);
1984
1985         if (btrfs_header_generation(leaf) > sk->max_transid) {
1986                 i = nritems;
1987                 goto advance_key;
1988         }
1989         found_transid = btrfs_header_generation(leaf);
1990
1991         for (i = slot; i < nritems; i++) {
1992                 item_off = btrfs_item_ptr_offset(leaf, i);
1993                 item_len = btrfs_item_size_nr(leaf, i);
1994
1995                 btrfs_item_key_to_cpu(leaf, key, i);
1996                 if (!key_in_sk(key, sk))
1997                         continue;
1998
1999                 if (sizeof(sh) + item_len > *buf_size) {
2000                         if (*num_found) {
2001                                 ret = 1;
2002                                 goto out;
2003                         }
2004
2005                         /*
2006                          * return one empty item back for v1, which does not
2007                          * handle -EOVERFLOW
2008                          */
2009
2010                         *buf_size = sizeof(sh) + item_len;
2011                         item_len = 0;
2012                         ret = -EOVERFLOW;
2013                 }
2014
2015                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2016                         ret = 1;
2017                         goto out;
2018                 }
2019
2020                 sh.objectid = key->objectid;
2021                 sh.offset = key->offset;
2022                 sh.type = key->type;
2023                 sh.len = item_len;
2024                 sh.transid = found_transid;
2025
2026                 /* copy search result header */
2027                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2028                         ret = -EFAULT;
2029                         goto out;
2030                 }
2031
2032                 *sk_offset += sizeof(sh);
2033
2034                 if (item_len) {
2035                         char __user *up = ubuf + *sk_offset;
2036                         /* copy the item */
2037                         if (read_extent_buffer_to_user(leaf, up,
2038                                                        item_off, item_len)) {
2039                                 ret = -EFAULT;
2040                                 goto out;
2041                         }
2042
2043                         *sk_offset += item_len;
2044                 }
2045                 (*num_found)++;
2046
2047                 if (ret) /* -EOVERFLOW from above */
2048                         goto out;
2049
2050                 if (*num_found >= sk->nr_items) {
2051                         ret = 1;
2052                         goto out;
2053                 }
2054         }
2055 advance_key:
2056         ret = 0;
2057         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2058                 key->offset++;
2059         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2060                 key->offset = 0;
2061                 key->type++;
2062         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2063                 key->offset = 0;
2064                 key->type = 0;
2065                 key->objectid++;
2066         } else
2067                 ret = 1;
2068 out:
2069         /*
2070          *  0: all items from this leaf copied, continue with next
2071          *  1: * more items can be copied, but unused buffer is too small
2072          *     * all items were found
2073          *     Either way, it will stops the loop which iterates to the next
2074          *     leaf
2075          *  -EOVERFLOW: item was to large for buffer
2076          *  -EFAULT: could not copy extent buffer back to userspace
2077          */
2078         return ret;
2079 }
2080
2081 static noinline int search_ioctl(struct inode *inode,
2082                                  struct btrfs_ioctl_search_key *sk,
2083                                  size_t *buf_size,
2084                                  char __user *ubuf)
2085 {
2086         struct btrfs_root *root;
2087         struct btrfs_key key;
2088         struct btrfs_path *path;
2089         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2090         int ret;
2091         int num_found = 0;
2092         unsigned long sk_offset = 0;
2093
2094         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2095                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2096                 return -EOVERFLOW;
2097         }
2098
2099         path = btrfs_alloc_path();
2100         if (!path)
2101                 return -ENOMEM;
2102
2103         if (sk->tree_id == 0) {
2104                 /* search the root of the inode that was passed */
2105                 root = BTRFS_I(inode)->root;
2106         } else {
2107                 key.objectid = sk->tree_id;
2108                 key.type = BTRFS_ROOT_ITEM_KEY;
2109                 key.offset = (u64)-1;
2110                 root = btrfs_read_fs_root_no_name(info, &key);
2111                 if (IS_ERR(root)) {
2112                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2113                                sk->tree_id);
2114                         btrfs_free_path(path);
2115                         return -ENOENT;
2116                 }
2117         }
2118
2119         key.objectid = sk->min_objectid;
2120         key.type = sk->min_type;
2121         key.offset = sk->min_offset;
2122
2123         while (1) {
2124                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2125                 if (ret != 0) {
2126                         if (ret > 0)
2127                                 ret = 0;
2128                         goto err;
2129                 }
2130                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2131                                  &sk_offset, &num_found);
2132                 btrfs_release_path(path);
2133                 if (ret)
2134                         break;
2135
2136         }
2137         if (ret > 0)
2138                 ret = 0;
2139 err:
2140         sk->nr_items = num_found;
2141         btrfs_free_path(path);
2142         return ret;
2143 }
2144
2145 static noinline int btrfs_ioctl_tree_search(struct file *file,
2146                                            void __user *argp)
2147 {
2148         struct btrfs_ioctl_search_args __user *uargs;
2149         struct btrfs_ioctl_search_key sk;
2150         struct inode *inode;
2151         int ret;
2152         size_t buf_size;
2153
2154         if (!capable(CAP_SYS_ADMIN))
2155                 return -EPERM;
2156
2157         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2158
2159         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2160                 return -EFAULT;
2161
2162         buf_size = sizeof(uargs->buf);
2163
2164         inode = file_inode(file);
2165         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2166
2167         /*
2168          * In the origin implementation an overflow is handled by returning a
2169          * search header with a len of zero, so reset ret.
2170          */
2171         if (ret == -EOVERFLOW)
2172                 ret = 0;
2173
2174         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2175                 ret = -EFAULT;
2176         return ret;
2177 }
2178
2179 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2180                                                void __user *argp)
2181 {
2182         struct btrfs_ioctl_search_args_v2 __user *uarg;
2183         struct btrfs_ioctl_search_args_v2 args;
2184         struct inode *inode;
2185         int ret;
2186         size_t buf_size;
2187         const size_t buf_limit = 16 * 1024 * 1024;
2188
2189         if (!capable(CAP_SYS_ADMIN))
2190                 return -EPERM;
2191
2192         /* copy search header and buffer size */
2193         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2194         if (copy_from_user(&args, uarg, sizeof(args)))
2195                 return -EFAULT;
2196
2197         buf_size = args.buf_size;
2198
2199         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2200                 return -EOVERFLOW;
2201
2202         /* limit result size to 16MB */
2203         if (buf_size > buf_limit)
2204                 buf_size = buf_limit;
2205
2206         inode = file_inode(file);
2207         ret = search_ioctl(inode, &args.key, &buf_size,
2208                            (char *)(&uarg->buf[0]));
2209         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2210                 ret = -EFAULT;
2211         else if (ret == -EOVERFLOW &&
2212                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2213                 ret = -EFAULT;
2214
2215         return ret;
2216 }
2217
2218 /*
2219  * Search INODE_REFs to identify path name of 'dirid' directory
2220  * in a 'tree_id' tree. and sets path name to 'name'.
2221  */
2222 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2223                                 u64 tree_id, u64 dirid, char *name)
2224 {
2225         struct btrfs_root *root;
2226         struct btrfs_key key;
2227         char *ptr;
2228         int ret = -1;
2229         int slot;
2230         int len;
2231         int total_len = 0;
2232         struct btrfs_inode_ref *iref;
2233         struct extent_buffer *l;
2234         struct btrfs_path *path;
2235
2236         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2237                 name[0]='\0';
2238                 return 0;
2239         }
2240
2241         path = btrfs_alloc_path();
2242         if (!path)
2243                 return -ENOMEM;
2244
2245         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2246
2247         key.objectid = tree_id;
2248         key.type = BTRFS_ROOT_ITEM_KEY;
2249         key.offset = (u64)-1;
2250         root = btrfs_read_fs_root_no_name(info, &key);
2251         if (IS_ERR(root)) {
2252                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2253                 ret = -ENOENT;
2254                 goto out;
2255         }
2256
2257         key.objectid = dirid;
2258         key.type = BTRFS_INODE_REF_KEY;
2259         key.offset = (u64)-1;
2260
2261         while (1) {
2262                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2263                 if (ret < 0)
2264                         goto out;
2265                 else if (ret > 0) {
2266                         ret = btrfs_previous_item(root, path, dirid,
2267                                                   BTRFS_INODE_REF_KEY);
2268                         if (ret < 0)
2269                                 goto out;
2270                         else if (ret > 0) {
2271                                 ret = -ENOENT;
2272                                 goto out;
2273                         }
2274                 }
2275
2276                 l = path->nodes[0];
2277                 slot = path->slots[0];
2278                 btrfs_item_key_to_cpu(l, &key, slot);
2279
2280                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2281                 len = btrfs_inode_ref_name_len(l, iref);
2282                 ptr -= len + 1;
2283                 total_len += len + 1;
2284                 if (ptr < name) {
2285                         ret = -ENAMETOOLONG;
2286                         goto out;
2287                 }
2288
2289                 *(ptr + len) = '/';
2290                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2291
2292                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2293                         break;
2294
2295                 btrfs_release_path(path);
2296                 key.objectid = key.offset;
2297                 key.offset = (u64)-1;
2298                 dirid = key.objectid;
2299         }
2300         memmove(name, ptr, total_len);
2301         name[total_len] = '\0';
2302         ret = 0;
2303 out:
2304         btrfs_free_path(path);
2305         return ret;
2306 }
2307
2308 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2309                                            void __user *argp)
2310 {
2311          struct btrfs_ioctl_ino_lookup_args *args;
2312          struct inode *inode;
2313          int ret;
2314
2315         if (!capable(CAP_SYS_ADMIN))
2316                 return -EPERM;
2317
2318         args = memdup_user(argp, sizeof(*args));
2319         if (IS_ERR(args))
2320                 return PTR_ERR(args);
2321
2322         inode = file_inode(file);
2323
2324         if (args->treeid == 0)
2325                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2326
2327         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2328                                         args->treeid, args->objectid,
2329                                         args->name);
2330
2331         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2332                 ret = -EFAULT;
2333
2334         kfree(args);
2335         return ret;
2336 }
2337
2338 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2339                                              void __user *arg)
2340 {
2341         struct dentry *parent = file->f_path.dentry;
2342         struct dentry *dentry;
2343         struct inode *dir = parent->d_inode;
2344         struct inode *inode;
2345         struct btrfs_root *root = BTRFS_I(dir)->root;
2346         struct btrfs_root *dest = NULL;
2347         struct btrfs_ioctl_vol_args *vol_args;
2348         struct btrfs_trans_handle *trans;
2349         struct btrfs_block_rsv block_rsv;
2350         u64 root_flags;
2351         u64 qgroup_reserved;
2352         int namelen;
2353         int ret;
2354         int err = 0;
2355
2356         vol_args = memdup_user(arg, sizeof(*vol_args));
2357         if (IS_ERR(vol_args))
2358                 return PTR_ERR(vol_args);
2359
2360         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2361         namelen = strlen(vol_args->name);
2362         if (strchr(vol_args->name, '/') ||
2363             strncmp(vol_args->name, "..", namelen) == 0) {
2364                 err = -EINVAL;
2365                 goto out;
2366         }
2367
2368         err = mnt_want_write_file(file);
2369         if (err)
2370                 goto out;
2371
2372
2373         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2374         if (err == -EINTR)
2375                 goto out_drop_write;
2376         dentry = lookup_one_len(vol_args->name, parent, namelen);
2377         if (IS_ERR(dentry)) {
2378                 err = PTR_ERR(dentry);
2379                 goto out_unlock_dir;
2380         }
2381
2382         if (!dentry->d_inode) {
2383                 err = -ENOENT;
2384                 goto out_dput;
2385         }
2386
2387         inode = dentry->d_inode;
2388         dest = BTRFS_I(inode)->root;
2389         if (!capable(CAP_SYS_ADMIN)) {
2390                 /*
2391                  * Regular user.  Only allow this with a special mount
2392                  * option, when the user has write+exec access to the
2393                  * subvol root, and when rmdir(2) would have been
2394                  * allowed.
2395                  *
2396                  * Note that this is _not_ check that the subvol is
2397                  * empty or doesn't contain data that we wouldn't
2398                  * otherwise be able to delete.
2399                  *
2400                  * Users who want to delete empty subvols should try
2401                  * rmdir(2).
2402                  */
2403                 err = -EPERM;
2404                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2405                         goto out_dput;
2406
2407                 /*
2408                  * Do not allow deletion if the parent dir is the same
2409                  * as the dir to be deleted.  That means the ioctl
2410                  * must be called on the dentry referencing the root
2411                  * of the subvol, not a random directory contained
2412                  * within it.
2413                  */
2414                 err = -EINVAL;
2415                 if (root == dest)
2416                         goto out_dput;
2417
2418                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2419                 if (err)
2420                         goto out_dput;
2421         }
2422
2423         /* check if subvolume may be deleted by a user */
2424         err = btrfs_may_delete(dir, dentry, 1);
2425         if (err)
2426                 goto out_dput;
2427
2428         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2429                 err = -EINVAL;
2430                 goto out_dput;
2431         }
2432
2433         mutex_lock(&inode->i_mutex);
2434
2435         /*
2436          * Don't allow to delete a subvolume with send in progress. This is
2437          * inside the i_mutex so the error handling that has to drop the bit
2438          * again is not run concurrently.
2439          */
2440         spin_lock(&dest->root_item_lock);
2441         root_flags = btrfs_root_flags(&dest->root_item);
2442         if (dest->send_in_progress == 0) {
2443                 btrfs_set_root_flags(&dest->root_item,
2444                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2445                 spin_unlock(&dest->root_item_lock);
2446         } else {
2447                 spin_unlock(&dest->root_item_lock);
2448                 btrfs_warn(root->fs_info,
2449                         "Attempt to delete subvolume %llu during send",
2450                         dest->root_key.objectid);
2451                 err = -EPERM;
2452                 goto out_dput;
2453         }
2454
2455         d_invalidate(dentry);
2456
2457         down_write(&root->fs_info->subvol_sem);
2458
2459         err = may_destroy_subvol(dest);
2460         if (err)
2461                 goto out_up_write;
2462
2463         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2464         /*
2465          * One for dir inode, two for dir entries, two for root
2466          * ref/backref.
2467          */
2468         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2469                                                5, &qgroup_reserved, true);
2470         if (err)
2471                 goto out_up_write;
2472
2473         trans = btrfs_start_transaction(root, 0);
2474         if (IS_ERR(trans)) {
2475                 err = PTR_ERR(trans);
2476                 goto out_release;
2477         }
2478         trans->block_rsv = &block_rsv;
2479         trans->bytes_reserved = block_rsv.size;
2480
2481         ret = btrfs_unlink_subvol(trans, root, dir,
2482                                 dest->root_key.objectid,
2483                                 dentry->d_name.name,
2484                                 dentry->d_name.len);
2485         if (ret) {
2486                 err = ret;
2487                 btrfs_abort_transaction(trans, root, ret);
2488                 goto out_end_trans;
2489         }
2490
2491         btrfs_record_root_in_trans(trans, dest);
2492
2493         memset(&dest->root_item.drop_progress, 0,
2494                 sizeof(dest->root_item.drop_progress));
2495         dest->root_item.drop_level = 0;
2496         btrfs_set_root_refs(&dest->root_item, 0);
2497
2498         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2499                 ret = btrfs_insert_orphan_item(trans,
2500                                         root->fs_info->tree_root,
2501                                         dest->root_key.objectid);
2502                 if (ret) {
2503                         btrfs_abort_transaction(trans, root, ret);
2504                         err = ret;
2505                         goto out_end_trans;
2506                 }
2507         }
2508
2509         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2510                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2511                                   dest->root_key.objectid);
2512         if (ret && ret != -ENOENT) {
2513                 btrfs_abort_transaction(trans, root, ret);
2514                 err = ret;
2515                 goto out_end_trans;
2516         }
2517         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2518                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2519                                           dest->root_item.received_uuid,
2520                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2521                                           dest->root_key.objectid);
2522                 if (ret && ret != -ENOENT) {
2523                         btrfs_abort_transaction(trans, root, ret);
2524                         err = ret;
2525                         goto out_end_trans;
2526                 }
2527         }
2528
2529 out_end_trans:
2530         trans->block_rsv = NULL;
2531         trans->bytes_reserved = 0;
2532         ret = btrfs_end_transaction(trans, root);
2533         if (ret && !err)
2534                 err = ret;
2535         inode->i_flags |= S_DEAD;
2536 out_release:
2537         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2538 out_up_write:
2539         up_write(&root->fs_info->subvol_sem);
2540         if (err) {
2541                 spin_lock(&dest->root_item_lock);
2542                 root_flags = btrfs_root_flags(&dest->root_item);
2543                 btrfs_set_root_flags(&dest->root_item,
2544                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2545                 spin_unlock(&dest->root_item_lock);
2546         }
2547         mutex_unlock(&inode->i_mutex);
2548         if (!err) {
2549                 shrink_dcache_sb(root->fs_info->sb);
2550                 btrfs_invalidate_inodes(dest);
2551                 d_delete(dentry);
2552                 ASSERT(dest->send_in_progress == 0);
2553
2554                 /* the last ref */
2555                 if (dest->ino_cache_inode) {
2556                         iput(dest->ino_cache_inode);
2557                         dest->ino_cache_inode = NULL;
2558                 }
2559         }
2560 out_dput:
2561         dput(dentry);
2562 out_unlock_dir:
2563         mutex_unlock(&dir->i_mutex);
2564 out_drop_write:
2565         mnt_drop_write_file(file);
2566 out:
2567         kfree(vol_args);
2568         return err;
2569 }
2570
2571 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2572 {
2573         struct inode *inode = file_inode(file);
2574         struct btrfs_root *root = BTRFS_I(inode)->root;
2575         struct btrfs_ioctl_defrag_range_args *range;
2576         int ret;
2577
2578         ret = mnt_want_write_file(file);
2579         if (ret)
2580                 return ret;
2581
2582         if (btrfs_root_readonly(root)) {
2583                 ret = -EROFS;
2584                 goto out;
2585         }
2586
2587         switch (inode->i_mode & S_IFMT) {
2588         case S_IFDIR:
2589                 if (!capable(CAP_SYS_ADMIN)) {
2590                         ret = -EPERM;
2591                         goto out;
2592                 }
2593                 ret = btrfs_defrag_root(root);
2594                 if (ret)
2595                         goto out;
2596                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2597                 break;
2598         case S_IFREG:
2599                 if (!(file->f_mode & FMODE_WRITE)) {
2600                         ret = -EINVAL;
2601                         goto out;
2602                 }
2603
2604                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2605                 if (!range) {
2606                         ret = -ENOMEM;
2607                         goto out;
2608                 }
2609
2610                 if (argp) {
2611                         if (copy_from_user(range, argp,
2612                                            sizeof(*range))) {
2613                                 ret = -EFAULT;
2614                                 kfree(range);
2615                                 goto out;
2616                         }
2617                         /* compression requires us to start the IO */
2618                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2619                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2620                                 range->extent_thresh = (u32)-1;
2621                         }
2622                 } else {
2623                         /* the rest are all set to zero by kzalloc */
2624                         range->len = (u64)-1;
2625                 }
2626                 ret = btrfs_defrag_file(file_inode(file), file,
2627                                         range, 0, 0);
2628                 if (ret > 0)
2629                         ret = 0;
2630                 kfree(range);
2631                 break;
2632         default:
2633                 ret = -EINVAL;
2634         }
2635 out:
2636         mnt_drop_write_file(file);
2637         return ret;
2638 }
2639
2640 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2641 {
2642         struct btrfs_ioctl_vol_args *vol_args;
2643         int ret;
2644
2645         if (!capable(CAP_SYS_ADMIN))
2646                 return -EPERM;
2647
2648         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2649                         1)) {
2650                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2651         }
2652
2653         mutex_lock(&root->fs_info->volume_mutex);
2654         vol_args = memdup_user(arg, sizeof(*vol_args));
2655         if (IS_ERR(vol_args)) {
2656                 ret = PTR_ERR(vol_args);
2657                 goto out;
2658         }
2659
2660         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2661         ret = btrfs_init_new_device(root, vol_args->name);
2662
2663         if (!ret)
2664                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2665
2666         kfree(vol_args);
2667 out:
2668         mutex_unlock(&root->fs_info->volume_mutex);
2669         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2670         return ret;
2671 }
2672
2673 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2674 {
2675         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2676         struct btrfs_ioctl_vol_args *vol_args;
2677         int ret;
2678
2679         if (!capable(CAP_SYS_ADMIN))
2680                 return -EPERM;
2681
2682         ret = mnt_want_write_file(file);
2683         if (ret)
2684                 return ret;
2685
2686         vol_args = memdup_user(arg, sizeof(*vol_args));
2687         if (IS_ERR(vol_args)) {
2688                 ret = PTR_ERR(vol_args);
2689                 goto err_drop;
2690         }
2691
2692         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2693
2694         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2695                         1)) {
2696                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2697                 goto out;
2698         }
2699
2700         mutex_lock(&root->fs_info->volume_mutex);
2701         ret = btrfs_rm_device(root, vol_args->name);
2702         mutex_unlock(&root->fs_info->volume_mutex);
2703         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2704
2705         if (!ret)
2706                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2707
2708 out:
2709         kfree(vol_args);
2710 err_drop:
2711         mnt_drop_write_file(file);
2712         return ret;
2713 }
2714
2715 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2716 {
2717         struct btrfs_ioctl_fs_info_args *fi_args;
2718         struct btrfs_device *device;
2719         struct btrfs_device *next;
2720         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2721         int ret = 0;
2722
2723         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2724         if (!fi_args)
2725                 return -ENOMEM;
2726
2727         mutex_lock(&fs_devices->device_list_mutex);
2728         fi_args->num_devices = fs_devices->num_devices;
2729         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2730
2731         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2732                 if (device->devid > fi_args->max_id)
2733                         fi_args->max_id = device->devid;
2734         }
2735         mutex_unlock(&fs_devices->device_list_mutex);
2736
2737         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2738         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2739         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2740
2741         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2742                 ret = -EFAULT;
2743
2744         kfree(fi_args);
2745         return ret;
2746 }
2747
2748 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2749 {
2750         struct btrfs_ioctl_dev_info_args *di_args;
2751         struct btrfs_device *dev;
2752         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2753         int ret = 0;
2754         char *s_uuid = NULL;
2755
2756         di_args = memdup_user(arg, sizeof(*di_args));
2757         if (IS_ERR(di_args))
2758                 return PTR_ERR(di_args);
2759
2760         if (!btrfs_is_empty_uuid(di_args->uuid))
2761                 s_uuid = di_args->uuid;
2762
2763         mutex_lock(&fs_devices->device_list_mutex);
2764         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2765
2766         if (!dev) {
2767                 ret = -ENODEV;
2768                 goto out;
2769         }
2770
2771         di_args->devid = dev->devid;
2772         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2773         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2774         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2775         if (dev->name) {
2776                 struct rcu_string *name;
2777
2778                 rcu_read_lock();
2779                 name = rcu_dereference(dev->name);
2780                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2781                 rcu_read_unlock();
2782                 di_args->path[sizeof(di_args->path) - 1] = 0;
2783         } else {
2784                 di_args->path[0] = '\0';
2785         }
2786
2787 out:
2788         mutex_unlock(&fs_devices->device_list_mutex);
2789         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2790                 ret = -EFAULT;
2791
2792         kfree(di_args);
2793         return ret;
2794 }
2795
2796 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2797 {
2798         struct page *page;
2799         pgoff_t index;
2800         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2801
2802         index = off >> PAGE_CACHE_SHIFT;
2803
2804         page = grab_cache_page(inode->i_mapping, index);
2805         if (!page)
2806                 return NULL;
2807
2808         if (!PageUptodate(page)) {
2809                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2810                                                  0))
2811                         return NULL;
2812                 lock_page(page);
2813                 if (!PageUptodate(page)) {
2814                         unlock_page(page);
2815                         page_cache_release(page);
2816                         return NULL;
2817                 }
2818         }
2819         unlock_page(page);
2820
2821         return page;
2822 }
2823
2824 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2825 {
2826         /* do any pending delalloc/csum calc on src, one way or
2827            another, and lock file content */
2828         while (1) {
2829                 struct btrfs_ordered_extent *ordered;
2830                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2831                 ordered = btrfs_lookup_first_ordered_extent(inode,
2832                                                             off + len - 1);
2833                 if ((!ordered ||
2834                      ordered->file_offset + ordered->len <= off ||
2835                      ordered->file_offset >= off + len) &&
2836                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2837                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2838                         if (ordered)
2839                                 btrfs_put_ordered_extent(ordered);
2840                         break;
2841                 }
2842                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2843                 if (ordered)
2844                         btrfs_put_ordered_extent(ordered);
2845                 btrfs_wait_ordered_range(inode, off, len);
2846         }
2847 }
2848
2849 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2850                                 struct inode *inode2, u64 loff2, u64 len)
2851 {
2852         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2853         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2854
2855         mutex_unlock(&inode1->i_mutex);
2856         mutex_unlock(&inode2->i_mutex);
2857 }
2858
2859 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2860                               struct inode *inode2, u64 loff2, u64 len)
2861 {
2862         if (inode1 < inode2) {
2863                 swap(inode1, inode2);
2864                 swap(loff1, loff2);
2865         }
2866
2867         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2868         lock_extent_range(inode1, loff1, len);
2869         if (inode1 != inode2) {
2870                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2871                 lock_extent_range(inode2, loff2, len);
2872         }
2873 }
2874
2875 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2876                           u64 dst_loff, u64 len)
2877 {
2878         int ret = 0;
2879         struct page *src_page, *dst_page;
2880         unsigned int cmp_len = PAGE_CACHE_SIZE;
2881         void *addr, *dst_addr;
2882
2883         while (len) {
2884                 if (len < PAGE_CACHE_SIZE)
2885                         cmp_len = len;
2886
2887                 src_page = extent_same_get_page(src, loff);
2888                 if (!src_page)
2889                         return -EINVAL;
2890                 dst_page = extent_same_get_page(dst, dst_loff);
2891                 if (!dst_page) {
2892                         page_cache_release(src_page);
2893                         return -EINVAL;
2894                 }
2895                 addr = kmap_atomic(src_page);
2896                 dst_addr = kmap_atomic(dst_page);
2897
2898                 flush_dcache_page(src_page);
2899                 flush_dcache_page(dst_page);
2900
2901                 if (memcmp(addr, dst_addr, cmp_len))
2902                         ret = BTRFS_SAME_DATA_DIFFERS;
2903
2904                 kunmap_atomic(addr);
2905                 kunmap_atomic(dst_addr);
2906                 page_cache_release(src_page);
2907                 page_cache_release(dst_page);
2908
2909                 if (ret)
2910                         break;
2911
2912                 loff += cmp_len;
2913                 dst_loff += cmp_len;
2914                 len -= cmp_len;
2915         }
2916
2917         return ret;
2918 }
2919
2920 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2921 {
2922         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2923
2924         if (off + len > inode->i_size || off + len < off)
2925                 return -EINVAL;
2926         /* Check that we are block aligned - btrfs_clone() requires this */
2927         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2928                 return -EINVAL;
2929
2930         return 0;
2931 }
2932
2933 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2934                              struct inode *dst, u64 dst_loff)
2935 {
2936         int ret;
2937
2938         /*
2939          * btrfs_clone() can't handle extents in the same file
2940          * yet. Once that works, we can drop this check and replace it
2941          * with a check for the same inode, but overlapping extents.
2942          */
2943         if (src == dst)
2944                 return -EINVAL;
2945
2946         btrfs_double_lock(src, loff, dst, dst_loff, len);
2947
2948         ret = extent_same_check_offsets(src, loff, len);
2949         if (ret)
2950                 goto out_unlock;
2951
2952         ret = extent_same_check_offsets(dst, dst_loff, len);
2953         if (ret)
2954                 goto out_unlock;
2955
2956         /* don't make the dst file partly checksummed */
2957         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2958             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2959                 ret = -EINVAL;
2960                 goto out_unlock;
2961         }
2962
2963         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2964         if (ret == 0)
2965                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2966
2967 out_unlock:
2968         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2969
2970         return ret;
2971 }
2972
2973 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2974
2975 static long btrfs_ioctl_file_extent_same(struct file *file,
2976                         struct btrfs_ioctl_same_args __user *argp)
2977 {
2978         struct btrfs_ioctl_same_args *same;
2979         struct btrfs_ioctl_same_extent_info *info;
2980         struct inode *src = file_inode(file);
2981         u64 off;
2982         u64 len;
2983         int i;
2984         int ret;
2985         unsigned long size;
2986         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2987         bool is_admin = capable(CAP_SYS_ADMIN);
2988         u16 count;
2989
2990         if (!(file->f_mode & FMODE_READ))
2991                 return -EINVAL;
2992
2993         ret = mnt_want_write_file(file);
2994         if (ret)
2995                 return ret;
2996
2997         if (get_user(count, &argp->dest_count)) {
2998                 ret = -EFAULT;
2999                 goto out;
3000         }
3001
3002         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3003
3004         same = memdup_user(argp, size);
3005
3006         if (IS_ERR(same)) {
3007                 ret = PTR_ERR(same);
3008                 goto out;
3009         }
3010
3011         off = same->logical_offset;
3012         len = same->length;
3013
3014         /*
3015          * Limit the total length we will dedupe for each operation.
3016          * This is intended to bound the total time spent in this
3017          * ioctl to something sane.
3018          */
3019         if (len > BTRFS_MAX_DEDUPE_LEN)
3020                 len = BTRFS_MAX_DEDUPE_LEN;
3021
3022         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3023                 /*
3024                  * Btrfs does not support blocksize < page_size. As a
3025                  * result, btrfs_cmp_data() won't correctly handle
3026                  * this situation without an update.
3027                  */
3028                 ret = -EINVAL;
3029                 goto out;
3030         }
3031
3032         ret = -EISDIR;
3033         if (S_ISDIR(src->i_mode))
3034                 goto out;
3035
3036         ret = -EACCES;
3037         if (!S_ISREG(src->i_mode))
3038                 goto out;
3039
3040         /* pre-format output fields to sane values */
3041         for (i = 0; i < count; i++) {
3042                 same->info[i].bytes_deduped = 0ULL;
3043                 same->info[i].status = 0;
3044         }
3045
3046         for (i = 0, info = same->info; i < count; i++, info++) {
3047                 struct inode *dst;
3048                 struct fd dst_file = fdget(info->fd);
3049                 if (!dst_file.file) {
3050                         info->status = -EBADF;
3051                         continue;
3052                 }
3053                 dst = file_inode(dst_file.file);
3054
3055                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3056                         info->status = -EINVAL;
3057                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3058                         info->status = -EXDEV;
3059                 } else if (S_ISDIR(dst->i_mode)) {
3060                         info->status = -EISDIR;
3061                 } else if (!S_ISREG(dst->i_mode)) {
3062                         info->status = -EACCES;
3063                 } else {
3064                         info->status = btrfs_extent_same(src, off, len, dst,
3065                                                         info->logical_offset);
3066                         if (info->status == 0)
3067                                 info->bytes_deduped += len;
3068                 }
3069                 fdput(dst_file);
3070         }
3071
3072         ret = copy_to_user(argp, same, size);
3073         if (ret)
3074                 ret = -EFAULT;
3075
3076 out:
3077         mnt_drop_write_file(file);
3078         return ret;
3079 }
3080
3081 /* Helper to check and see if this root currently has a ref on the given disk
3082  * bytenr.  If it does then we need to update the quota for this root.  This
3083  * doesn't do anything if quotas aren't enabled.
3084  */
3085 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3086                      u64 disko)
3087 {
3088         struct seq_list tree_mod_seq_elem = {};
3089         struct ulist *roots;
3090         struct ulist_iterator uiter;
3091         struct ulist_node *root_node = NULL;
3092         int ret;
3093
3094         if (!root->fs_info->quota_enabled)
3095                 return 1;
3096
3097         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3098         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3099                                    tree_mod_seq_elem.seq, &roots);
3100         if (ret < 0)
3101                 goto out;
3102         ret = 0;
3103         ULIST_ITER_INIT(&uiter);
3104         while ((root_node = ulist_next(roots, &uiter))) {
3105                 if (root_node->val == root->objectid) {
3106                         ret = 1;
3107                         break;
3108                 }
3109         }
3110         ulist_free(roots);
3111 out:
3112         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3113         return ret;
3114 }
3115
3116 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3117                                      struct inode *inode,
3118                                      u64 endoff,
3119                                      const u64 destoff,
3120                                      const u64 olen)
3121 {
3122         struct btrfs_root *root = BTRFS_I(inode)->root;
3123         int ret;
3124
3125         inode_inc_iversion(inode);
3126         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3127         /*
3128          * We round up to the block size at eof when determining which
3129          * extents to clone above, but shouldn't round up the file size.
3130          */
3131         if (endoff > destoff + olen)
3132                 endoff = destoff + olen;
3133         if (endoff > inode->i_size)
3134                 btrfs_i_size_write(inode, endoff);
3135
3136         ret = btrfs_update_inode(trans, root, inode);
3137         if (ret) {
3138                 btrfs_abort_transaction(trans, root, ret);
3139                 btrfs_end_transaction(trans, root);
3140                 goto out;
3141         }
3142         ret = btrfs_end_transaction(trans, root);
3143 out:
3144         return ret;
3145 }
3146
3147 static void clone_update_extent_map(struct inode *inode,
3148                                     const struct btrfs_trans_handle *trans,
3149                                     const struct btrfs_path *path,
3150                                     const u64 hole_offset,
3151                                     const u64 hole_len)
3152 {
3153         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3154         struct extent_map *em;
3155         int ret;
3156
3157         em = alloc_extent_map();
3158         if (!em) {
3159                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3160                         &BTRFS_I(inode)->runtime_flags);
3161                 return;
3162         }
3163
3164         if (path) {
3165                 struct btrfs_file_extent_item *fi;
3166
3167                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3168                                     struct btrfs_file_extent_item);
3169                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3170                 em->generation = -1;
3171                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3172                     BTRFS_FILE_EXTENT_INLINE)
3173                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3174                                 &BTRFS_I(inode)->runtime_flags);
3175         } else {
3176                 em->start = hole_offset;
3177                 em->len = hole_len;
3178                 em->ram_bytes = em->len;
3179                 em->orig_start = hole_offset;
3180                 em->block_start = EXTENT_MAP_HOLE;
3181                 em->block_len = 0;
3182                 em->orig_block_len = 0;
3183                 em->compress_type = BTRFS_COMPRESS_NONE;
3184                 em->generation = trans->transid;
3185         }
3186
3187         while (1) {
3188                 write_lock(&em_tree->lock);
3189                 ret = add_extent_mapping(em_tree, em, 1);
3190                 write_unlock(&em_tree->lock);
3191                 if (ret != -EEXIST) {
3192                         free_extent_map(em);
3193                         break;
3194                 }
3195                 btrfs_drop_extent_cache(inode, em->start,
3196                                         em->start + em->len - 1, 0);
3197         }
3198
3199         if (ret)
3200                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3201                         &BTRFS_I(inode)->runtime_flags);
3202 }
3203
3204 /**
3205  * btrfs_clone() - clone a range from inode file to another
3206  *
3207  * @src: Inode to clone from
3208  * @inode: Inode to clone to
3209  * @off: Offset within source to start clone from
3210  * @olen: Original length, passed by user, of range to clone
3211  * @olen_aligned: Block-aligned value of olen, extent_same uses
3212  *               identical values here
3213  * @destoff: Offset within @inode to start clone
3214  */
3215 static int btrfs_clone(struct inode *src, struct inode *inode,
3216                        const u64 off, const u64 olen, const u64 olen_aligned,
3217                        const u64 destoff)
3218 {
3219         struct btrfs_root *root = BTRFS_I(inode)->root;
3220         struct btrfs_path *path = NULL;
3221         struct extent_buffer *leaf;
3222         struct btrfs_trans_handle *trans;
3223         char *buf = NULL;
3224         struct btrfs_key key;
3225         u32 nritems;
3226         int slot;
3227         int ret;
3228         int no_quota;
3229         const u64 len = olen_aligned;
3230         u64 last_disko = 0;
3231         u64 last_dest_end = destoff;
3232
3233         ret = -ENOMEM;
3234         buf = vmalloc(root->nodesize);
3235         if (!buf)
3236                 return ret;
3237
3238         path = btrfs_alloc_path();
3239         if (!path) {
3240                 vfree(buf);
3241                 return ret;
3242         }
3243
3244         path->reada = 2;
3245         /* clone data */
3246         key.objectid = btrfs_ino(src);
3247         key.type = BTRFS_EXTENT_DATA_KEY;
3248         key.offset = off;
3249
3250         while (1) {
3251                 /*