3e3e6130637fa7268e6834c6fafe129de84345a7
[muen/linux.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                        int no_time_update);
92
93 /* Mask out flags that are inappropriate for the given type of inode. */
94 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
95 {
96         if (S_ISDIR(mode))
97                 return flags;
98         else if (S_ISREG(mode))
99                 return flags & ~FS_DIRSYNC_FL;
100         else
101                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
102 }
103
104 /*
105  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
106  */
107 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
108 {
109         unsigned int iflags = 0;
110
111         if (flags & BTRFS_INODE_SYNC)
112                 iflags |= FS_SYNC_FL;
113         if (flags & BTRFS_INODE_IMMUTABLE)
114                 iflags |= FS_IMMUTABLE_FL;
115         if (flags & BTRFS_INODE_APPEND)
116                 iflags |= FS_APPEND_FL;
117         if (flags & BTRFS_INODE_NODUMP)
118                 iflags |= FS_NODUMP_FL;
119         if (flags & BTRFS_INODE_NOATIME)
120                 iflags |= FS_NOATIME_FL;
121         if (flags & BTRFS_INODE_DIRSYNC)
122                 iflags |= FS_DIRSYNC_FL;
123         if (flags & BTRFS_INODE_NODATACOW)
124                 iflags |= FS_NOCOW_FL;
125
126         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
127                 iflags |= FS_COMPR_FL;
128         else if (flags & BTRFS_INODE_NOCOMPRESS)
129                 iflags |= FS_NOCOMP_FL;
130
131         return iflags;
132 }
133
134 /*
135  * Update inode->i_flags based on the btrfs internal flags.
136  */
137 void btrfs_update_iflags(struct inode *inode)
138 {
139         struct btrfs_inode *ip = BTRFS_I(inode);
140         unsigned int new_fl = 0;
141
142         if (ip->flags & BTRFS_INODE_SYNC)
143                 new_fl |= S_SYNC;
144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
145                 new_fl |= S_IMMUTABLE;
146         if (ip->flags & BTRFS_INODE_APPEND)
147                 new_fl |= S_APPEND;
148         if (ip->flags & BTRFS_INODE_NOATIME)
149                 new_fl |= S_NOATIME;
150         if (ip->flags & BTRFS_INODE_DIRSYNC)
151                 new_fl |= S_DIRSYNC;
152
153         set_mask_bits(&inode->i_flags,
154                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
155                       new_fl);
156 }
157
158 /*
159  * Inherit flags from the parent inode.
160  *
161  * Currently only the compression flags and the cow flags are inherited.
162  */
163 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
164 {
165         unsigned int flags;
166
167         if (!dir)
168                 return;
169
170         flags = BTRFS_I(dir)->flags;
171
172         if (flags & BTRFS_INODE_NOCOMPRESS) {
173                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
174                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
175         } else if (flags & BTRFS_INODE_COMPRESS) {
176                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
177                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
178         }
179
180         if (flags & BTRFS_INODE_NODATACOW) {
181                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
182                 if (S_ISREG(inode->i_mode))
183                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
184         }
185
186         btrfs_update_iflags(inode);
187 }
188
189 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
190 {
191         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
192         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
193
194         if (copy_to_user(arg, &flags, sizeof(flags)))
195                 return -EFAULT;
196         return 0;
197 }
198
199 static int check_flags(unsigned int flags)
200 {
201         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
202                       FS_NOATIME_FL | FS_NODUMP_FL | \
203                       FS_SYNC_FL | FS_DIRSYNC_FL | \
204                       FS_NOCOMP_FL | FS_COMPR_FL |
205                       FS_NOCOW_FL))
206                 return -EOPNOTSUPP;
207
208         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
209                 return -EINVAL;
210
211         return 0;
212 }
213
214 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
215 {
216         struct inode *inode = file_inode(file);
217         struct btrfs_inode *ip = BTRFS_I(inode);
218         struct btrfs_root *root = ip->root;
219         struct btrfs_trans_handle *trans;
220         unsigned int flags, oldflags;
221         int ret;
222         u64 ip_oldflags;
223         unsigned int i_oldflags;
224         umode_t mode;
225
226         if (!inode_owner_or_capable(inode))
227                 return -EPERM;
228
229         if (btrfs_root_readonly(root))
230                 return -EROFS;
231
232         if (copy_from_user(&flags, arg, sizeof(flags)))
233                 return -EFAULT;
234
235         ret = check_flags(flags);
236         if (ret)
237                 return ret;
238
239         ret = mnt_want_write_file(file);
240         if (ret)
241                 return ret;
242
243         mutex_lock(&inode->i_mutex);
244
245         ip_oldflags = ip->flags;
246         i_oldflags = inode->i_flags;
247         mode = inode->i_mode;
248
249         flags = btrfs_mask_flags(inode->i_mode, flags);
250         oldflags = btrfs_flags_to_ioctl(ip->flags);
251         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
252                 if (!capable(CAP_LINUX_IMMUTABLE)) {
253                         ret = -EPERM;
254                         goto out_unlock;
255                 }
256         }
257
258         if (flags & FS_SYNC_FL)
259                 ip->flags |= BTRFS_INODE_SYNC;
260         else
261                 ip->flags &= ~BTRFS_INODE_SYNC;
262         if (flags & FS_IMMUTABLE_FL)
263                 ip->flags |= BTRFS_INODE_IMMUTABLE;
264         else
265                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
266         if (flags & FS_APPEND_FL)
267                 ip->flags |= BTRFS_INODE_APPEND;
268         else
269                 ip->flags &= ~BTRFS_INODE_APPEND;
270         if (flags & FS_NODUMP_FL)
271                 ip->flags |= BTRFS_INODE_NODUMP;
272         else
273                 ip->flags &= ~BTRFS_INODE_NODUMP;
274         if (flags & FS_NOATIME_FL)
275                 ip->flags |= BTRFS_INODE_NOATIME;
276         else
277                 ip->flags &= ~BTRFS_INODE_NOATIME;
278         if (flags & FS_DIRSYNC_FL)
279                 ip->flags |= BTRFS_INODE_DIRSYNC;
280         else
281                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
282         if (flags & FS_NOCOW_FL) {
283                 if (S_ISREG(mode)) {
284                         /*
285                          * It's safe to turn csums off here, no extents exist.
286                          * Otherwise we want the flag to reflect the real COW
287                          * status of the file and will not set it.
288                          */
289                         if (inode->i_size == 0)
290                                 ip->flags |= BTRFS_INODE_NODATACOW
291                                            | BTRFS_INODE_NODATASUM;
292                 } else {
293                         ip->flags |= BTRFS_INODE_NODATACOW;
294                 }
295         } else {
296                 /*
297                  * Revert back under same assuptions as above
298                  */
299                 if (S_ISREG(mode)) {
300                         if (inode->i_size == 0)
301                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
302                                              | BTRFS_INODE_NODATASUM);
303                 } else {
304                         ip->flags &= ~BTRFS_INODE_NODATACOW;
305                 }
306         }
307
308         /*
309          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
310          * flag may be changed automatically if compression code won't make
311          * things smaller.
312          */
313         if (flags & FS_NOCOMP_FL) {
314                 ip->flags &= ~BTRFS_INODE_COMPRESS;
315                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
316
317                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
318                 if (ret && ret != -ENODATA)
319                         goto out_drop;
320         } else if (flags & FS_COMPR_FL) {
321                 const char *comp;
322
323                 ip->flags |= BTRFS_INODE_COMPRESS;
324                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
325
326                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
327                         comp = "lzo";
328                 else
329                         comp = "zlib";
330                 ret = btrfs_set_prop(inode, "btrfs.compression",
331                                      comp, strlen(comp), 0);
332                 if (ret)
333                         goto out_drop;
334
335         } else {
336                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
337                 if (ret && ret != -ENODATA)
338                         goto out_drop;
339                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
340         }
341
342         trans = btrfs_start_transaction(root, 1);
343         if (IS_ERR(trans)) {
344                 ret = PTR_ERR(trans);
345                 goto out_drop;
346         }
347
348         btrfs_update_iflags(inode);
349         inode_inc_iversion(inode);
350         inode->i_ctime = CURRENT_TIME;
351         ret = btrfs_update_inode(trans, root, inode);
352
353         btrfs_end_transaction(trans, root);
354  out_drop:
355         if (ret) {
356                 ip->flags = ip_oldflags;
357                 inode->i_flags = i_oldflags;
358         }
359
360  out_unlock:
361         mutex_unlock(&inode->i_mutex);
362         mnt_drop_write_file(file);
363         return ret;
364 }
365
366 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
367 {
368         struct inode *inode = file_inode(file);
369
370         return put_user(inode->i_generation, arg);
371 }
372
373 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
374 {
375         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
376         struct btrfs_device *device;
377         struct request_queue *q;
378         struct fstrim_range range;
379         u64 minlen = ULLONG_MAX;
380         u64 num_devices = 0;
381         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
382         int ret;
383
384         if (!capable(CAP_SYS_ADMIN))
385                 return -EPERM;
386
387         rcu_read_lock();
388         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
389                                 dev_list) {
390                 if (!device->bdev)
391                         continue;
392                 q = bdev_get_queue(device->bdev);
393                 if (blk_queue_discard(q)) {
394                         num_devices++;
395                         minlen = min((u64)q->limits.discard_granularity,
396                                      minlen);
397                 }
398         }
399         rcu_read_unlock();
400
401         if (!num_devices)
402                 return -EOPNOTSUPP;
403         if (copy_from_user(&range, arg, sizeof(range)))
404                 return -EFAULT;
405         if (range.start > total_bytes ||
406             range.len < fs_info->sb->s_blocksize)
407                 return -EINVAL;
408
409         range.len = min(range.len, total_bytes - range.start);
410         range.minlen = max(range.minlen, minlen);
411         ret = btrfs_trim_fs(fs_info->tree_root, &range);
412         if (ret < 0)
413                 return ret;
414
415         if (copy_to_user(arg, &range, sizeof(range)))
416                 return -EFAULT;
417
418         return 0;
419 }
420
421 int btrfs_is_empty_uuid(u8 *uuid)
422 {
423         int i;
424
425         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
426                 if (uuid[i])
427                         return 0;
428         }
429         return 1;
430 }
431
432 static noinline int create_subvol(struct inode *dir,
433                                   struct dentry *dentry,
434                                   char *name, int namelen,
435                                   u64 *async_transid,
436                                   struct btrfs_qgroup_inherit *inherit)
437 {
438         struct btrfs_trans_handle *trans;
439         struct btrfs_key key;
440         struct btrfs_root_item root_item;
441         struct btrfs_inode_item *inode_item;
442         struct extent_buffer *leaf;
443         struct btrfs_root *root = BTRFS_I(dir)->root;
444         struct btrfs_root *new_root;
445         struct btrfs_block_rsv block_rsv;
446         struct timespec cur_time = CURRENT_TIME;
447         struct inode *inode;
448         int ret;
449         int err;
450         u64 objectid;
451         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
452         u64 index = 0;
453         u64 qgroup_reserved;
454         uuid_le new_uuid;
455
456         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
457         if (ret)
458                 return ret;
459
460         /*
461          * Don't create subvolume whose level is not zero. Or qgroup will be
462          * screwed up since it assume subvolme qgroup's level to be 0.
463          */
464         if (btrfs_qgroup_level(objectid))
465                 return -ENOSPC;
466
467         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
468         /*
469          * The same as the snapshot creation, please see the comment
470          * of create_snapshot().
471          */
472         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
473                                                8, &qgroup_reserved, false);
474         if (ret)
475                 return ret;
476
477         trans = btrfs_start_transaction(root, 0);
478         if (IS_ERR(trans)) {
479                 ret = PTR_ERR(trans);
480                 btrfs_subvolume_release_metadata(root, &block_rsv,
481                                                  qgroup_reserved);
482                 return ret;
483         }
484         trans->block_rsv = &block_rsv;
485         trans->bytes_reserved = block_rsv.size;
486
487         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
488         if (ret)
489                 goto fail;
490
491         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
492         if (IS_ERR(leaf)) {
493                 ret = PTR_ERR(leaf);
494                 goto fail;
495         }
496
497         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
498         btrfs_set_header_bytenr(leaf, leaf->start);
499         btrfs_set_header_generation(leaf, trans->transid);
500         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
501         btrfs_set_header_owner(leaf, objectid);
502
503         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
504                             BTRFS_FSID_SIZE);
505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
506                             btrfs_header_chunk_tree_uuid(leaf),
507                             BTRFS_UUID_SIZE);
508         btrfs_mark_buffer_dirty(leaf);
509
510         memset(&root_item, 0, sizeof(root_item));
511
512         inode_item = &root_item.inode;
513         btrfs_set_stack_inode_generation(inode_item, 1);
514         btrfs_set_stack_inode_size(inode_item, 3);
515         btrfs_set_stack_inode_nlink(inode_item, 1);
516         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
517         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
518
519         btrfs_set_root_flags(&root_item, 0);
520         btrfs_set_root_limit(&root_item, 0);
521         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
522
523         btrfs_set_root_bytenr(&root_item, leaf->start);
524         btrfs_set_root_generation(&root_item, trans->transid);
525         btrfs_set_root_level(&root_item, 0);
526         btrfs_set_root_refs(&root_item, 1);
527         btrfs_set_root_used(&root_item, leaf->len);
528         btrfs_set_root_last_snapshot(&root_item, 0);
529
530         btrfs_set_root_generation_v2(&root_item,
531                         btrfs_root_generation(&root_item));
532         uuid_le_gen(&new_uuid);
533         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
534         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
535         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
536         root_item.ctime = root_item.otime;
537         btrfs_set_root_ctransid(&root_item, trans->transid);
538         btrfs_set_root_otransid(&root_item, trans->transid);
539
540         btrfs_tree_unlock(leaf);
541         free_extent_buffer(leaf);
542         leaf = NULL;
543
544         btrfs_set_root_dirid(&root_item, new_dirid);
545
546         key.objectid = objectid;
547         key.offset = 0;
548         key.type = BTRFS_ROOT_ITEM_KEY;
549         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
550                                 &root_item);
551         if (ret)
552                 goto fail;
553
554         key.offset = (u64)-1;
555         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
556         if (IS_ERR(new_root)) {
557                 ret = PTR_ERR(new_root);
558                 btrfs_abort_transaction(trans, root, ret);
559                 goto fail;
560         }
561
562         btrfs_record_root_in_trans(trans, new_root);
563
564         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
565         if (ret) {
566                 /* We potentially lose an unused inode item here */
567                 btrfs_abort_transaction(trans, root, ret);
568                 goto fail;
569         }
570
571         /*
572          * insert the directory item
573          */
574         ret = btrfs_set_inode_index(dir, &index);
575         if (ret) {
576                 btrfs_abort_transaction(trans, root, ret);
577                 goto fail;
578         }
579
580         ret = btrfs_insert_dir_item(trans, root,
581                                     name, namelen, dir, &key,
582                                     BTRFS_FT_DIR, index);
583         if (ret) {
584                 btrfs_abort_transaction(trans, root, ret);
585                 goto fail;
586         }
587
588         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
589         ret = btrfs_update_inode(trans, root, dir);
590         BUG_ON(ret);
591
592         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
593                                  objectid, root->root_key.objectid,
594                                  btrfs_ino(dir), index, name, namelen);
595         BUG_ON(ret);
596
597         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
598                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
599                                   objectid);
600         if (ret)
601                 btrfs_abort_transaction(trans, root, ret);
602
603 fail:
604         trans->block_rsv = NULL;
605         trans->bytes_reserved = 0;
606         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
607
608         if (async_transid) {
609                 *async_transid = trans->transid;
610                 err = btrfs_commit_transaction_async(trans, root, 1);
611                 if (err)
612                         err = btrfs_commit_transaction(trans, root);
613         } else {
614                 err = btrfs_commit_transaction(trans, root);
615         }
616         if (err && !ret)
617                 ret = err;
618
619         if (!ret) {
620                 inode = btrfs_lookup_dentry(dir, dentry);
621                 if (IS_ERR(inode))
622                         return PTR_ERR(inode);
623                 d_instantiate(dentry, inode);
624         }
625         return ret;
626 }
627
628 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
629 {
630         s64 writers;
631         DEFINE_WAIT(wait);
632
633         do {
634                 prepare_to_wait(&root->subv_writers->wait, &wait,
635                                 TASK_UNINTERRUPTIBLE);
636
637                 writers = percpu_counter_sum(&root->subv_writers->counter);
638                 if (writers)
639                         schedule();
640
641                 finish_wait(&root->subv_writers->wait, &wait);
642         } while (writers);
643 }
644
645 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
646                            struct dentry *dentry, char *name, int namelen,
647                            u64 *async_transid, bool readonly,
648                            struct btrfs_qgroup_inherit *inherit)
649 {
650         struct inode *inode;
651         struct btrfs_pending_snapshot *pending_snapshot;
652         struct btrfs_trans_handle *trans;
653         int ret;
654
655         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
656                 return -EINVAL;
657
658         atomic_inc(&root->will_be_snapshoted);
659         smp_mb__after_atomic();
660         btrfs_wait_for_no_snapshoting_writes(root);
661
662         ret = btrfs_start_delalloc_inodes(root, 0);
663         if (ret)
664                 goto out;
665
666         btrfs_wait_ordered_extents(root, -1);
667
668         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
669         if (!pending_snapshot) {
670                 ret = -ENOMEM;
671                 goto out;
672         }
673
674         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
675                              BTRFS_BLOCK_RSV_TEMP);
676         /*
677          * 1 - parent dir inode
678          * 2 - dir entries
679          * 1 - root item
680          * 2 - root ref/backref
681          * 1 - root of snapshot
682          * 1 - UUID item
683          */
684         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
685                                         &pending_snapshot->block_rsv, 8,
686                                         &pending_snapshot->qgroup_reserved,
687                                         false);
688         if (ret)
689                 goto free;
690
691         pending_snapshot->dentry = dentry;
692         pending_snapshot->root = root;
693         pending_snapshot->readonly = readonly;
694         pending_snapshot->dir = dir;
695         pending_snapshot->inherit = inherit;
696
697         trans = btrfs_start_transaction(root, 0);
698         if (IS_ERR(trans)) {
699                 ret = PTR_ERR(trans);
700                 goto fail;
701         }
702
703         spin_lock(&root->fs_info->trans_lock);
704         list_add(&pending_snapshot->list,
705                  &trans->transaction->pending_snapshots);
706         spin_unlock(&root->fs_info->trans_lock);
707         if (async_transid) {
708                 *async_transid = trans->transid;
709                 ret = btrfs_commit_transaction_async(trans,
710                                      root->fs_info->extent_root, 1);
711                 if (ret)
712                         ret = btrfs_commit_transaction(trans, root);
713         } else {
714                 ret = btrfs_commit_transaction(trans,
715                                                root->fs_info->extent_root);
716         }
717         if (ret)
718                 goto fail;
719
720         ret = pending_snapshot->error;
721         if (ret)
722                 goto fail;
723
724         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
725         if (ret)
726                 goto fail;
727
728         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
729         if (IS_ERR(inode)) {
730                 ret = PTR_ERR(inode);
731                 goto fail;
732         }
733
734         d_instantiate(dentry, inode);
735         ret = 0;
736 fail:
737         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
738                                          &pending_snapshot->block_rsv,
739                                          pending_snapshot->qgroup_reserved);
740 free:
741         kfree(pending_snapshot);
742 out:
743         if (atomic_dec_and_test(&root->will_be_snapshoted))
744                 wake_up_atomic_t(&root->will_be_snapshoted);
745         return ret;
746 }
747
748 /*  copy of may_delete in fs/namei.c()
749  *      Check whether we can remove a link victim from directory dir, check
750  *  whether the type of victim is right.
751  *  1. We can't do it if dir is read-only (done in permission())
752  *  2. We should have write and exec permissions on dir
753  *  3. We can't remove anything from append-only dir
754  *  4. We can't do anything with immutable dir (done in permission())
755  *  5. If the sticky bit on dir is set we should either
756  *      a. be owner of dir, or
757  *      b. be owner of victim, or
758  *      c. have CAP_FOWNER capability
759  *  6. If the victim is append-only or immutable we can't do antyhing with
760  *     links pointing to it.
761  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
762  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
763  *  9. We can't remove a root or mountpoint.
764  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
765  *     nfs_async_unlink().
766  */
767
768 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
769 {
770         int error;
771
772         if (d_really_is_negative(victim))
773                 return -ENOENT;
774
775         BUG_ON(d_inode(victim->d_parent) != dir);
776         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
777
778         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
779         if (error)
780                 return error;
781         if (IS_APPEND(dir))
782                 return -EPERM;
783         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
784             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
785                 return -EPERM;
786         if (isdir) {
787                 if (!d_is_dir(victim))
788                         return -ENOTDIR;
789                 if (IS_ROOT(victim))
790                         return -EBUSY;
791         } else if (d_is_dir(victim))
792                 return -EISDIR;
793         if (IS_DEADDIR(dir))
794                 return -ENOENT;
795         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
796                 return -EBUSY;
797         return 0;
798 }
799
800 /* copy of may_create in fs/namei.c() */
801 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
802 {
803         if (d_really_is_positive(child))
804                 return -EEXIST;
805         if (IS_DEADDIR(dir))
806                 return -ENOENT;
807         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
808 }
809
810 /*
811  * Create a new subvolume below @parent.  This is largely modeled after
812  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
813  * inside this filesystem so it's quite a bit simpler.
814  */
815 static noinline int btrfs_mksubvol(struct path *parent,
816                                    char *name, int namelen,
817                                    struct btrfs_root *snap_src,
818                                    u64 *async_transid, bool readonly,
819                                    struct btrfs_qgroup_inherit *inherit)
820 {
821         struct inode *dir  = d_inode(parent->dentry);
822         struct dentry *dentry;
823         int error;
824
825         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
826         if (error == -EINTR)
827                 return error;
828
829         dentry = lookup_one_len(name, parent->dentry, namelen);
830         error = PTR_ERR(dentry);
831         if (IS_ERR(dentry))
832                 goto out_unlock;
833
834         error = -EEXIST;
835         if (d_really_is_positive(dentry))
836                 goto out_dput;
837
838         error = btrfs_may_create(dir, dentry);
839         if (error)
840                 goto out_dput;
841
842         /*
843          * even if this name doesn't exist, we may get hash collisions.
844          * check for them now when we can safely fail
845          */
846         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
847                                                dir->i_ino, name,
848                                                namelen);
849         if (error)
850                 goto out_dput;
851
852         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
853
854         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
855                 goto out_up_read;
856
857         if (snap_src) {
858                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
859                                         async_transid, readonly, inherit);
860         } else {
861                 error = create_subvol(dir, dentry, name, namelen,
862                                       async_transid, inherit);
863         }
864         if (!error)
865                 fsnotify_mkdir(dir, dentry);
866 out_up_read:
867         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
868 out_dput:
869         dput(dentry);
870 out_unlock:
871         mutex_unlock(&dir->i_mutex);
872         return error;
873 }
874
875 /*
876  * When we're defragging a range, we don't want to kick it off again
877  * if it is really just waiting for delalloc to send it down.
878  * If we find a nice big extent or delalloc range for the bytes in the
879  * file you want to defrag, we return 0 to let you know to skip this
880  * part of the file
881  */
882 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
883 {
884         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
885         struct extent_map *em = NULL;
886         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
887         u64 end;
888
889         read_lock(&em_tree->lock);
890         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
891         read_unlock(&em_tree->lock);
892
893         if (em) {
894                 end = extent_map_end(em);
895                 free_extent_map(em);
896                 if (end - offset > thresh)
897                         return 0;
898         }
899         /* if we already have a nice delalloc here, just stop */
900         thresh /= 2;
901         end = count_range_bits(io_tree, &offset, offset + thresh,
902                                thresh, EXTENT_DELALLOC, 1);
903         if (end >= thresh)
904                 return 0;
905         return 1;
906 }
907
908 /*
909  * helper function to walk through a file and find extents
910  * newer than a specific transid, and smaller than thresh.
911  *
912  * This is used by the defragging code to find new and small
913  * extents
914  */
915 static int find_new_extents(struct btrfs_root *root,
916                             struct inode *inode, u64 newer_than,
917                             u64 *off, u32 thresh)
918 {
919         struct btrfs_path *path;
920         struct btrfs_key min_key;
921         struct extent_buffer *leaf;
922         struct btrfs_file_extent_item *extent;
923         int type;
924         int ret;
925         u64 ino = btrfs_ino(inode);
926
927         path = btrfs_alloc_path();
928         if (!path)
929                 return -ENOMEM;
930
931         min_key.objectid = ino;
932         min_key.type = BTRFS_EXTENT_DATA_KEY;
933         min_key.offset = *off;
934
935         while (1) {
936                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
937                 if (ret != 0)
938                         goto none;
939 process_slot:
940                 if (min_key.objectid != ino)
941                         goto none;
942                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
943                         goto none;
944
945                 leaf = path->nodes[0];
946                 extent = btrfs_item_ptr(leaf, path->slots[0],
947                                         struct btrfs_file_extent_item);
948
949                 type = btrfs_file_extent_type(leaf, extent);
950                 if (type == BTRFS_FILE_EXTENT_REG &&
951                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
952                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
953                         *off = min_key.offset;
954                         btrfs_free_path(path);
955                         return 0;
956                 }
957
958                 path->slots[0]++;
959                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
960                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
961                         goto process_slot;
962                 }
963
964                 if (min_key.offset == (u64)-1)
965                         goto none;
966
967                 min_key.offset++;
968                 btrfs_release_path(path);
969         }
970 none:
971         btrfs_free_path(path);
972         return -ENOENT;
973 }
974
975 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
976 {
977         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
978         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
979         struct extent_map *em;
980         u64 len = PAGE_CACHE_SIZE;
981
982         /*
983          * hopefully we have this extent in the tree already, try without
984          * the full extent lock
985          */
986         read_lock(&em_tree->lock);
987         em = lookup_extent_mapping(em_tree, start, len);
988         read_unlock(&em_tree->lock);
989
990         if (!em) {
991                 struct extent_state *cached = NULL;
992                 u64 end = start + len - 1;
993
994                 /* get the big lock and read metadata off disk */
995                 lock_extent_bits(io_tree, start, end, 0, &cached);
996                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
997                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
998
999                 if (IS_ERR(em))
1000                         return NULL;
1001         }
1002
1003         return em;
1004 }
1005
1006 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1007 {
1008         struct extent_map *next;
1009         bool ret = true;
1010
1011         /* this is the last extent */
1012         if (em->start + em->len >= i_size_read(inode))
1013                 return false;
1014
1015         next = defrag_lookup_extent(inode, em->start + em->len);
1016         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1017                 ret = false;
1018         else if ((em->block_start + em->block_len == next->block_start) &&
1019                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1020                 ret = false;
1021
1022         free_extent_map(next);
1023         return ret;
1024 }
1025
1026 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1027                                u64 *last_len, u64 *skip, u64 *defrag_end,
1028                                int compress)
1029 {
1030         struct extent_map *em;
1031         int ret = 1;
1032         bool next_mergeable = true;
1033         bool prev_mergeable = true;
1034
1035         /*
1036          * make sure that once we start defragging an extent, we keep on
1037          * defragging it
1038          */
1039         if (start < *defrag_end)
1040                 return 1;
1041
1042         *skip = 0;
1043
1044         em = defrag_lookup_extent(inode, start);
1045         if (!em)
1046                 return 0;
1047
1048         /* this will cover holes, and inline extents */
1049         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1050                 ret = 0;
1051                 goto out;
1052         }
1053
1054         if (!*defrag_end)
1055                 prev_mergeable = false;
1056
1057         next_mergeable = defrag_check_next_extent(inode, em);
1058         /*
1059          * we hit a real extent, if it is big or the next extent is not a
1060          * real extent, don't bother defragging it
1061          */
1062         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1063             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1064                 ret = 0;
1065 out:
1066         /*
1067          * last_len ends up being a counter of how many bytes we've defragged.
1068          * every time we choose not to defrag an extent, we reset *last_len
1069          * so that the next tiny extent will force a defrag.
1070          *
1071          * The end result of this is that tiny extents before a single big
1072          * extent will force at least part of that big extent to be defragged.
1073          */
1074         if (ret) {
1075                 *defrag_end = extent_map_end(em);
1076         } else {
1077                 *last_len = 0;
1078                 *skip = extent_map_end(em);
1079                 *defrag_end = 0;
1080         }
1081
1082         free_extent_map(em);
1083         return ret;
1084 }
1085
1086 /*
1087  * it doesn't do much good to defrag one or two pages
1088  * at a time.  This pulls in a nice chunk of pages
1089  * to COW and defrag.
1090  *
1091  * It also makes sure the delalloc code has enough
1092  * dirty data to avoid making new small extents as part
1093  * of the defrag
1094  *
1095  * It's a good idea to start RA on this range
1096  * before calling this.
1097  */
1098 static int cluster_pages_for_defrag(struct inode *inode,
1099                                     struct page **pages,
1100                                     unsigned long start_index,
1101                                     unsigned long num_pages)
1102 {
1103         unsigned long file_end;
1104         u64 isize = i_size_read(inode);
1105         u64 page_start;
1106         u64 page_end;
1107         u64 page_cnt;
1108         int ret;
1109         int i;
1110         int i_done;
1111         struct btrfs_ordered_extent *ordered;
1112         struct extent_state *cached_state = NULL;
1113         struct extent_io_tree *tree;
1114         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1115
1116         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1117         if (!isize || start_index > file_end)
1118                 return 0;
1119
1120         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1121
1122         ret = btrfs_delalloc_reserve_space(inode,
1123                                            page_cnt << PAGE_CACHE_SHIFT);
1124         if (ret)
1125                 return ret;
1126         i_done = 0;
1127         tree = &BTRFS_I(inode)->io_tree;
1128
1129         /* step one, lock all the pages */
1130         for (i = 0; i < page_cnt; i++) {
1131                 struct page *page;
1132 again:
1133                 page = find_or_create_page(inode->i_mapping,
1134                                            start_index + i, mask);
1135                 if (!page)
1136                         break;
1137
1138                 page_start = page_offset(page);
1139                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1140                 while (1) {
1141                         lock_extent_bits(tree, page_start, page_end,
1142                                          0, &cached_state);
1143                         ordered = btrfs_lookup_ordered_extent(inode,
1144                                                               page_start);
1145                         unlock_extent_cached(tree, page_start, page_end,
1146                                              &cached_state, GFP_NOFS);
1147                         if (!ordered)
1148                                 break;
1149
1150                         unlock_page(page);
1151                         btrfs_start_ordered_extent(inode, ordered, 1);
1152                         btrfs_put_ordered_extent(ordered);
1153                         lock_page(page);
1154                         /*
1155                          * we unlocked the page above, so we need check if
1156                          * it was released or not.
1157                          */
1158                         if (page->mapping != inode->i_mapping) {
1159                                 unlock_page(page);
1160                                 page_cache_release(page);
1161                                 goto again;
1162                         }
1163                 }
1164
1165                 if (!PageUptodate(page)) {
1166                         btrfs_readpage(NULL, page);
1167                         lock_page(page);
1168                         if (!PageUptodate(page)) {
1169                                 unlock_page(page);
1170                                 page_cache_release(page);
1171                                 ret = -EIO;
1172                                 break;
1173                         }
1174                 }
1175
1176                 if (page->mapping != inode->i_mapping) {
1177                         unlock_page(page);
1178                         page_cache_release(page);
1179                         goto again;
1180                 }
1181
1182                 pages[i] = page;
1183                 i_done++;
1184         }
1185         if (!i_done || ret)
1186                 goto out;
1187
1188         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1189                 goto out;
1190
1191         /*
1192          * so now we have a nice long stream of locked
1193          * and up to date pages, lets wait on them
1194          */
1195         for (i = 0; i < i_done; i++)
1196                 wait_on_page_writeback(pages[i]);
1197
1198         page_start = page_offset(pages[0]);
1199         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1200
1201         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1202                          page_start, page_end - 1, 0, &cached_state);
1203         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1204                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1205                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1206                           &cached_state, GFP_NOFS);
1207
1208         if (i_done != page_cnt) {
1209                 spin_lock(&BTRFS_I(inode)->lock);
1210                 BTRFS_I(inode)->outstanding_extents++;
1211                 spin_unlock(&BTRFS_I(inode)->lock);
1212                 btrfs_delalloc_release_space(inode,
1213                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1214         }
1215
1216
1217         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1218                           &cached_state, GFP_NOFS);
1219
1220         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1221                              page_start, page_end - 1, &cached_state,
1222                              GFP_NOFS);
1223
1224         for (i = 0; i < i_done; i++) {
1225                 clear_page_dirty_for_io(pages[i]);
1226                 ClearPageChecked(pages[i]);
1227                 set_page_extent_mapped(pages[i]);
1228                 set_page_dirty(pages[i]);
1229                 unlock_page(pages[i]);
1230                 page_cache_release(pages[i]);
1231         }
1232         return i_done;
1233 out:
1234         for (i = 0; i < i_done; i++) {
1235                 unlock_page(pages[i]);
1236                 page_cache_release(pages[i]);
1237         }
1238         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1239         return ret;
1240
1241 }
1242
1243 int btrfs_defrag_file(struct inode *inode, struct file *file,
1244                       struct btrfs_ioctl_defrag_range_args *range,
1245                       u64 newer_than, unsigned long max_to_defrag)
1246 {
1247         struct btrfs_root *root = BTRFS_I(inode)->root;
1248         struct file_ra_state *ra = NULL;
1249         unsigned long last_index;
1250         u64 isize = i_size_read(inode);
1251         u64 last_len = 0;
1252         u64 skip = 0;
1253         u64 defrag_end = 0;
1254         u64 newer_off = range->start;
1255         unsigned long i;
1256         unsigned long ra_index = 0;
1257         int ret;
1258         int defrag_count = 0;
1259         int compress_type = BTRFS_COMPRESS_ZLIB;
1260         u32 extent_thresh = range->extent_thresh;
1261         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1262         unsigned long cluster = max_cluster;
1263         u64 new_align = ~((u64)128 * 1024 - 1);
1264         struct page **pages = NULL;
1265
1266         if (isize == 0)
1267                 return 0;
1268
1269         if (range->start >= isize)
1270                 return -EINVAL;
1271
1272         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1273                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1274                         return -EINVAL;
1275                 if (range->compress_type)
1276                         compress_type = range->compress_type;
1277         }
1278
1279         if (extent_thresh == 0)
1280                 extent_thresh = 256 * 1024;
1281
1282         /*
1283          * if we were not given a file, allocate a readahead
1284          * context
1285          */
1286         if (!file) {
1287                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1288                 if (!ra)
1289                         return -ENOMEM;
1290                 file_ra_state_init(ra, inode->i_mapping);
1291         } else {
1292                 ra = &file->f_ra;
1293         }
1294
1295         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1296                         GFP_NOFS);
1297         if (!pages) {
1298                 ret = -ENOMEM;
1299                 goto out_ra;
1300         }
1301
1302         /* find the last page to defrag */
1303         if (range->start + range->len > range->start) {
1304                 last_index = min_t(u64, isize - 1,
1305                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1306         } else {
1307                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1308         }
1309
1310         if (newer_than) {
1311                 ret = find_new_extents(root, inode, newer_than,
1312                                        &newer_off, 64 * 1024);
1313                 if (!ret) {
1314                         range->start = newer_off;
1315                         /*
1316                          * we always align our defrag to help keep
1317                          * the extents in the file evenly spaced
1318                          */
1319                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1320                 } else
1321                         goto out_ra;
1322         } else {
1323                 i = range->start >> PAGE_CACHE_SHIFT;
1324         }
1325         if (!max_to_defrag)
1326                 max_to_defrag = last_index - i + 1;
1327
1328         /*
1329          * make writeback starts from i, so the defrag range can be
1330          * written sequentially.
1331          */
1332         if (i < inode->i_mapping->writeback_index)
1333                 inode->i_mapping->writeback_index = i;
1334
1335         while (i <= last_index && defrag_count < max_to_defrag &&
1336                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1337                 /*
1338                  * make sure we stop running if someone unmounts
1339                  * the FS
1340                  */
1341                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1342                         break;
1343
1344                 if (btrfs_defrag_cancelled(root->fs_info)) {
1345                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1346                         ret = -EAGAIN;
1347                         break;
1348                 }
1349
1350                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1351                                          extent_thresh, &last_len, &skip,
1352                                          &defrag_end, range->flags &
1353                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1354                         unsigned long next;
1355                         /*
1356                          * the should_defrag function tells us how much to skip
1357                          * bump our counter by the suggested amount
1358                          */
1359                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1360                         i = max(i + 1, next);
1361                         continue;
1362                 }
1363
1364                 if (!newer_than) {
1365                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1366                                    PAGE_CACHE_SHIFT) - i;
1367                         cluster = min(cluster, max_cluster);
1368                 } else {
1369                         cluster = max_cluster;
1370                 }
1371
1372                 if (i + cluster > ra_index) {
1373                         ra_index = max(i, ra_index);
1374                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1375                                        cluster);
1376                         ra_index += cluster;
1377                 }
1378
1379                 mutex_lock(&inode->i_mutex);
1380                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1381                         BTRFS_I(inode)->force_compress = compress_type;
1382                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1383                 if (ret < 0) {
1384                         mutex_unlock(&inode->i_mutex);
1385                         goto out_ra;
1386                 }
1387
1388                 defrag_count += ret;
1389                 balance_dirty_pages_ratelimited(inode->i_mapping);
1390                 mutex_unlock(&inode->i_mutex);
1391
1392                 if (newer_than) {
1393                         if (newer_off == (u64)-1)
1394                                 break;
1395
1396                         if (ret > 0)
1397                                 i += ret;
1398
1399                         newer_off = max(newer_off + 1,
1400                                         (u64)i << PAGE_CACHE_SHIFT);
1401
1402                         ret = find_new_extents(root, inode,
1403                                                newer_than, &newer_off,
1404                                                64 * 1024);
1405                         if (!ret) {
1406                                 range->start = newer_off;
1407                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1408                         } else {
1409                                 break;
1410                         }
1411                 } else {
1412                         if (ret > 0) {
1413                                 i += ret;
1414                                 last_len += ret << PAGE_CACHE_SHIFT;
1415                         } else {
1416                                 i++;
1417                                 last_len = 0;
1418                         }
1419                 }
1420         }
1421
1422         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1423                 filemap_flush(inode->i_mapping);
1424                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1425                              &BTRFS_I(inode)->runtime_flags))
1426                         filemap_flush(inode->i_mapping);
1427         }
1428
1429         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1430                 /* the filemap_flush will queue IO into the worker threads, but
1431                  * we have to make sure the IO is actually started and that
1432                  * ordered extents get created before we return
1433                  */
1434                 atomic_inc(&root->fs_info->async_submit_draining);
1435                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1436                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1437                         wait_event(root->fs_info->async_submit_wait,
1438                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1439                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1440                 }
1441                 atomic_dec(&root->fs_info->async_submit_draining);
1442         }
1443
1444         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1445                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1446         }
1447
1448         ret = defrag_count;
1449
1450 out_ra:
1451         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1452                 mutex_lock(&inode->i_mutex);
1453                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1454                 mutex_unlock(&inode->i_mutex);
1455         }
1456         if (!file)
1457                 kfree(ra);
1458         kfree(pages);
1459         return ret;
1460 }
1461
1462 static noinline int btrfs_ioctl_resize(struct file *file,
1463                                         void __user *arg)
1464 {
1465         u64 new_size;
1466         u64 old_size;
1467         u64 devid = 1;
1468         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1469         struct btrfs_ioctl_vol_args *vol_args;
1470         struct btrfs_trans_handle *trans;
1471         struct btrfs_device *device = NULL;
1472         char *sizestr;
1473         char *retptr;
1474         char *devstr = NULL;
1475         int ret = 0;
1476         int mod = 0;
1477
1478         if (!capable(CAP_SYS_ADMIN))
1479                 return -EPERM;
1480
1481         ret = mnt_want_write_file(file);
1482         if (ret)
1483                 return ret;
1484
1485         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1486                         1)) {
1487                 mnt_drop_write_file(file);
1488                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1489         }
1490
1491         mutex_lock(&root->fs_info->volume_mutex);
1492         vol_args = memdup_user(arg, sizeof(*vol_args));
1493         if (IS_ERR(vol_args)) {
1494                 ret = PTR_ERR(vol_args);
1495                 goto out;
1496         }
1497
1498         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1499
1500         sizestr = vol_args->name;
1501         devstr = strchr(sizestr, ':');
1502         if (devstr) {
1503                 sizestr = devstr + 1;
1504                 *devstr = '\0';
1505                 devstr = vol_args->name;
1506                 ret = kstrtoull(devstr, 10, &devid);
1507                 if (ret)
1508                         goto out_free;
1509                 if (!devid) {
1510                         ret = -EINVAL;
1511                         goto out_free;
1512                 }
1513                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1514         }
1515
1516         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1517         if (!device) {
1518                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1519                        devid);
1520                 ret = -ENODEV;
1521                 goto out_free;
1522         }
1523
1524         if (!device->writeable) {
1525                 btrfs_info(root->fs_info,
1526                            "resizer unable to apply on readonly device %llu",
1527                        devid);
1528                 ret = -EPERM;
1529                 goto out_free;
1530         }
1531
1532         if (!strcmp(sizestr, "max"))
1533                 new_size = device->bdev->bd_inode->i_size;
1534         else {
1535                 if (sizestr[0] == '-') {
1536                         mod = -1;
1537                         sizestr++;
1538                 } else if (sizestr[0] == '+') {
1539                         mod = 1;
1540                         sizestr++;
1541                 }
1542                 new_size = memparse(sizestr, &retptr);
1543                 if (*retptr != '\0' || new_size == 0) {
1544                         ret = -EINVAL;
1545                         goto out_free;
1546                 }
1547         }
1548
1549         if (device->is_tgtdev_for_dev_replace) {
1550                 ret = -EPERM;
1551                 goto out_free;
1552         }
1553
1554         old_size = btrfs_device_get_total_bytes(device);
1555
1556         if (mod < 0) {
1557                 if (new_size > old_size) {
1558                         ret = -EINVAL;
1559                         goto out_free;
1560                 }
1561                 new_size = old_size - new_size;
1562         } else if (mod > 0) {
1563                 if (new_size > ULLONG_MAX - old_size) {
1564                         ret = -ERANGE;
1565                         goto out_free;
1566                 }
1567                 new_size = old_size + new_size;
1568         }
1569
1570         if (new_size < 256 * 1024 * 1024) {
1571                 ret = -EINVAL;
1572                 goto out_free;
1573         }
1574         if (new_size > device->bdev->bd_inode->i_size) {
1575                 ret = -EFBIG;
1576                 goto out_free;
1577         }
1578
1579         new_size = div_u64(new_size, root->sectorsize);
1580         new_size *= root->sectorsize;
1581
1582         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1583                       rcu_str_deref(device->name), new_size);
1584
1585         if (new_size > old_size) {
1586                 trans = btrfs_start_transaction(root, 0);
1587                 if (IS_ERR(trans)) {
1588                         ret = PTR_ERR(trans);
1589                         goto out_free;
1590                 }
1591                 ret = btrfs_grow_device(trans, device, new_size);
1592                 btrfs_commit_transaction(trans, root);
1593         } else if (new_size < old_size) {
1594                 ret = btrfs_shrink_device(device, new_size);
1595         } /* equal, nothing need to do */
1596
1597 out_free:
1598         kfree(vol_args);
1599 out:
1600         mutex_unlock(&root->fs_info->volume_mutex);
1601         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1602         mnt_drop_write_file(file);
1603         return ret;
1604 }
1605
1606 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1607                                 char *name, unsigned long fd, int subvol,
1608                                 u64 *transid, bool readonly,
1609                                 struct btrfs_qgroup_inherit *inherit)
1610 {
1611         int namelen;
1612         int ret = 0;
1613
1614         ret = mnt_want_write_file(file);
1615         if (ret)
1616                 goto out;
1617
1618         namelen = strlen(name);
1619         if (strchr(name, '/')) {
1620                 ret = -EINVAL;
1621                 goto out_drop_write;
1622         }
1623
1624         if (name[0] == '.' &&
1625            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1626                 ret = -EEXIST;
1627                 goto out_drop_write;
1628         }
1629
1630         if (subvol) {
1631                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1632                                      NULL, transid, readonly, inherit);
1633         } else {
1634                 struct fd src = fdget(fd);
1635                 struct inode *src_inode;
1636                 if (!src.file) {
1637                         ret = -EINVAL;
1638                         goto out_drop_write;
1639                 }
1640
1641                 src_inode = file_inode(src.file);
1642                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1643                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1644                                    "Snapshot src from another FS");
1645                         ret = -EXDEV;
1646                 } else if (!inode_owner_or_capable(src_inode)) {
1647                         /*
1648                          * Subvolume creation is not restricted, but snapshots
1649                          * are limited to own subvolumes only
1650                          */
1651                         ret = -EPERM;
1652                 } else {
1653                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1654                                              BTRFS_I(src_inode)->root,
1655                                              transid, readonly, inherit);
1656                 }
1657                 fdput(src);
1658         }
1659 out_drop_write:
1660         mnt_drop_write_file(file);
1661 out:
1662         return ret;
1663 }
1664
1665 static noinline int btrfs_ioctl_snap_create(struct file *file,
1666                                             void __user *arg, int subvol)
1667 {
1668         struct btrfs_ioctl_vol_args *vol_args;
1669         int ret;
1670
1671         vol_args = memdup_user(arg, sizeof(*vol_args));
1672         if (IS_ERR(vol_args))
1673                 return PTR_ERR(vol_args);
1674         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1675
1676         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1677                                               vol_args->fd, subvol,
1678                                               NULL, false, NULL);
1679
1680         kfree(vol_args);
1681         return ret;
1682 }
1683
1684 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1685                                                void __user *arg, int subvol)
1686 {
1687         struct btrfs_ioctl_vol_args_v2 *vol_args;
1688         int ret;
1689         u64 transid = 0;
1690         u64 *ptr = NULL;
1691         bool readonly = false;
1692         struct btrfs_qgroup_inherit *inherit = NULL;
1693
1694         vol_args = memdup_user(arg, sizeof(*vol_args));
1695         if (IS_ERR(vol_args))
1696                 return PTR_ERR(vol_args);
1697         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1698
1699         if (vol_args->flags &
1700             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1701               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1702                 ret = -EOPNOTSUPP;
1703                 goto free_args;
1704         }
1705
1706         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1707                 ptr = &transid;
1708         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1709                 readonly = true;
1710         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1711                 if (vol_args->size > PAGE_CACHE_SIZE) {
1712                         ret = -EINVAL;
1713                         goto free_args;
1714                 }
1715                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1716                 if (IS_ERR(inherit)) {
1717                         ret = PTR_ERR(inherit);
1718                         goto free_args;
1719                 }
1720         }
1721
1722         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1723                                               vol_args->fd, subvol, ptr,
1724                                               readonly, inherit);
1725         if (ret)
1726                 goto free_inherit;
1727
1728         if (ptr && copy_to_user(arg +
1729                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1730                                         transid),
1731                                 ptr, sizeof(*ptr)))
1732                 ret = -EFAULT;
1733
1734 free_inherit:
1735         kfree(inherit);
1736 free_args:
1737         kfree(vol_args);
1738         return ret;
1739 }
1740
1741 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1742                                                 void __user *arg)
1743 {
1744         struct inode *inode = file_inode(file);
1745         struct btrfs_root *root = BTRFS_I(inode)->root;
1746         int ret = 0;
1747         u64 flags = 0;
1748
1749         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1750                 return -EINVAL;
1751
1752         down_read(&root->fs_info->subvol_sem);
1753         if (btrfs_root_readonly(root))
1754                 flags |= BTRFS_SUBVOL_RDONLY;
1755         up_read(&root->fs_info->subvol_sem);
1756
1757         if (copy_to_user(arg, &flags, sizeof(flags)))
1758                 ret = -EFAULT;
1759
1760         return ret;
1761 }
1762
1763 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1764                                               void __user *arg)
1765 {
1766         struct inode *inode = file_inode(file);
1767         struct btrfs_root *root = BTRFS_I(inode)->root;
1768         struct btrfs_trans_handle *trans;
1769         u64 root_flags;
1770         u64 flags;
1771         int ret = 0;
1772
1773         if (!inode_owner_or_capable(inode))
1774                 return -EPERM;
1775
1776         ret = mnt_want_write_file(file);
1777         if (ret)
1778                 goto out;
1779
1780         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1781                 ret = -EINVAL;
1782                 goto out_drop_write;
1783         }
1784
1785         if (copy_from_user(&flags, arg, sizeof(flags))) {
1786                 ret = -EFAULT;
1787                 goto out_drop_write;
1788         }
1789
1790         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1791                 ret = -EINVAL;
1792                 goto out_drop_write;
1793         }
1794
1795         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1796                 ret = -EOPNOTSUPP;
1797                 goto out_drop_write;
1798         }
1799
1800         down_write(&root->fs_info->subvol_sem);
1801
1802         /* nothing to do */
1803         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1804                 goto out_drop_sem;
1805
1806         root_flags = btrfs_root_flags(&root->root_item);
1807         if (flags & BTRFS_SUBVOL_RDONLY) {
1808                 btrfs_set_root_flags(&root->root_item,
1809                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1810         } else {
1811                 /*
1812                  * Block RO -> RW transition if this subvolume is involved in
1813                  * send
1814                  */
1815                 spin_lock(&root->root_item_lock);
1816                 if (root->send_in_progress == 0) {
1817                         btrfs_set_root_flags(&root->root_item,
1818                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1819                         spin_unlock(&root->root_item_lock);
1820                 } else {
1821                         spin_unlock(&root->root_item_lock);
1822                         btrfs_warn(root->fs_info,
1823                         "Attempt to set subvolume %llu read-write during send",
1824                                         root->root_key.objectid);
1825                         ret = -EPERM;
1826                         goto out_drop_sem;
1827                 }
1828         }
1829
1830         trans = btrfs_start_transaction(root, 1);
1831         if (IS_ERR(trans)) {
1832                 ret = PTR_ERR(trans);
1833                 goto out_reset;
1834         }
1835
1836         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1837                                 &root->root_key, &root->root_item);
1838
1839         btrfs_commit_transaction(trans, root);
1840 out_reset:
1841         if (ret)
1842                 btrfs_set_root_flags(&root->root_item, root_flags);
1843 out_drop_sem:
1844         up_write(&root->fs_info->subvol_sem);
1845 out_drop_write:
1846         mnt_drop_write_file(file);
1847 out:
1848         return ret;
1849 }
1850
1851 /*
1852  * helper to check if the subvolume references other subvolumes
1853  */
1854 static noinline int may_destroy_subvol(struct btrfs_root *root)
1855 {
1856         struct btrfs_path *path;
1857         struct btrfs_dir_item *di;
1858         struct btrfs_key key;
1859         u64 dir_id;
1860         int ret;
1861
1862         path = btrfs_alloc_path();
1863         if (!path)
1864                 return -ENOMEM;
1865
1866         /* Make sure this root isn't set as the default subvol */
1867         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1868         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1869                                    dir_id, "default", 7, 0);
1870         if (di && !IS_ERR(di)) {
1871                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1872                 if (key.objectid == root->root_key.objectid) {
1873                         ret = -EPERM;
1874                         btrfs_err(root->fs_info, "deleting default subvolume "
1875                                   "%llu is not allowed", key.objectid);
1876                         goto out;
1877                 }
1878                 btrfs_release_path(path);
1879         }
1880
1881         key.objectid = root->root_key.objectid;
1882         key.type = BTRFS_ROOT_REF_KEY;
1883         key.offset = (u64)-1;
1884
1885         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1886                                 &key, path, 0, 0);
1887         if (ret < 0)
1888                 goto out;
1889         BUG_ON(ret == 0);
1890
1891         ret = 0;
1892         if (path->slots[0] > 0) {
1893                 path->slots[0]--;
1894                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1895                 if (key.objectid == root->root_key.objectid &&
1896                     key.type == BTRFS_ROOT_REF_KEY)
1897                         ret = -ENOTEMPTY;
1898         }
1899 out:
1900         btrfs_free_path(path);
1901         return ret;
1902 }
1903
1904 static noinline int key_in_sk(struct btrfs_key *key,
1905                               struct btrfs_ioctl_search_key *sk)
1906 {
1907         struct btrfs_key test;
1908         int ret;
1909
1910         test.objectid = sk->min_objectid;
1911         test.type = sk->min_type;
1912         test.offset = sk->min_offset;
1913
1914         ret = btrfs_comp_cpu_keys(key, &test);
1915         if (ret < 0)
1916                 return 0;
1917
1918         test.objectid = sk->max_objectid;
1919         test.type = sk->max_type;
1920         test.offset = sk->max_offset;
1921
1922         ret = btrfs_comp_cpu_keys(key, &test);
1923         if (ret > 0)
1924                 return 0;
1925         return 1;
1926 }
1927
1928 static noinline int copy_to_sk(struct btrfs_root *root,
1929                                struct btrfs_path *path,
1930                                struct btrfs_key *key,
1931                                struct btrfs_ioctl_search_key *sk,
1932                                size_t *buf_size,
1933                                char __user *ubuf,
1934                                unsigned long *sk_offset,
1935                                int *num_found)
1936 {
1937         u64 found_transid;
1938         struct extent_buffer *leaf;
1939         struct btrfs_ioctl_search_header sh;
1940         struct btrfs_key test;
1941         unsigned long item_off;
1942         unsigned long item_len;
1943         int nritems;
1944         int i;
1945         int slot;
1946         int ret = 0;
1947
1948         leaf = path->nodes[0];
1949         slot = path->slots[0];
1950         nritems = btrfs_header_nritems(leaf);
1951
1952         if (btrfs_header_generation(leaf) > sk->max_transid) {
1953                 i = nritems;
1954                 goto advance_key;
1955         }
1956         found_transid = btrfs_header_generation(leaf);
1957
1958         for (i = slot; i < nritems; i++) {
1959                 item_off = btrfs_item_ptr_offset(leaf, i);
1960                 item_len = btrfs_item_size_nr(leaf, i);
1961
1962                 btrfs_item_key_to_cpu(leaf, key, i);
1963                 if (!key_in_sk(key, sk))
1964                         continue;
1965
1966                 if (sizeof(sh) + item_len > *buf_size) {
1967                         if (*num_found) {
1968                                 ret = 1;
1969                                 goto out;
1970                         }
1971
1972                         /*
1973                          * return one empty item back for v1, which does not
1974                          * handle -EOVERFLOW
1975                          */
1976
1977                         *buf_size = sizeof(sh) + item_len;
1978                         item_len = 0;
1979                         ret = -EOVERFLOW;
1980                 }
1981
1982                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1983                         ret = 1;
1984                         goto out;
1985                 }
1986
1987                 sh.objectid = key->objectid;
1988                 sh.offset = key->offset;
1989                 sh.type = key->type;
1990                 sh.len = item_len;
1991                 sh.transid = found_transid;
1992
1993                 /* copy search result header */
1994                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1995                         ret = -EFAULT;
1996                         goto out;
1997                 }
1998
1999                 *sk_offset += sizeof(sh);
2000
2001                 if (item_len) {
2002                         char __user *up = ubuf + *sk_offset;
2003                         /* copy the item */
2004                         if (read_extent_buffer_to_user(leaf, up,
2005                                                        item_off, item_len)) {
2006                                 ret = -EFAULT;
2007                                 goto out;
2008                         }
2009
2010                         *sk_offset += item_len;
2011                 }
2012                 (*num_found)++;
2013
2014                 if (ret) /* -EOVERFLOW from above */
2015                         goto out;
2016
2017                 if (*num_found >= sk->nr_items) {
2018                         ret = 1;
2019                         goto out;
2020                 }
2021         }
2022 advance_key:
2023         ret = 0;
2024         test.objectid = sk->max_objectid;
2025         test.type = sk->max_type;
2026         test.offset = sk->max_offset;
2027         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2028                 ret = 1;
2029         else if (key->offset < (u64)-1)
2030                 key->offset++;
2031         else if (key->type < (u8)-1) {
2032                 key->offset = 0;
2033                 key->type++;
2034         } else if (key->objectid < (u64)-1) {
2035                 key->offset = 0;
2036                 key->type = 0;
2037                 key->objectid++;
2038         } else
2039                 ret = 1;
2040 out:
2041         /*
2042          *  0: all items from this leaf copied, continue with next
2043          *  1: * more items can be copied, but unused buffer is too small
2044          *     * all items were found
2045          *     Either way, it will stops the loop which iterates to the next
2046          *     leaf
2047          *  -EOVERFLOW: item was to large for buffer
2048          *  -EFAULT: could not copy extent buffer back to userspace
2049          */
2050         return ret;
2051 }
2052
2053 static noinline int search_ioctl(struct inode *inode,
2054                                  struct btrfs_ioctl_search_key *sk,
2055                                  size_t *buf_size,
2056                                  char __user *ubuf)
2057 {
2058         struct btrfs_root *root;
2059         struct btrfs_key key;
2060         struct btrfs_path *path;
2061         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2062         int ret;
2063         int num_found = 0;
2064         unsigned long sk_offset = 0;
2065
2066         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2067                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2068                 return -EOVERFLOW;
2069         }
2070
2071         path = btrfs_alloc_path();
2072         if (!path)
2073                 return -ENOMEM;
2074
2075         if (sk->tree_id == 0) {
2076                 /* search the root of the inode that was passed */
2077                 root = BTRFS_I(inode)->root;
2078         } else {
2079                 key.objectid = sk->tree_id;
2080                 key.type = BTRFS_ROOT_ITEM_KEY;
2081                 key.offset = (u64)-1;
2082                 root = btrfs_read_fs_root_no_name(info, &key);
2083                 if (IS_ERR(root)) {
2084                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2085                                sk->tree_id);
2086                         btrfs_free_path(path);
2087                         return -ENOENT;
2088                 }
2089         }
2090
2091         key.objectid = sk->min_objectid;
2092         key.type = sk->min_type;
2093         key.offset = sk->min_offset;
2094
2095         while (1) {
2096                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2097                 if (ret != 0) {
2098                         if (ret > 0)
2099                                 ret = 0;
2100                         goto err;
2101                 }
2102                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2103                                  &sk_offset, &num_found);
2104                 btrfs_release_path(path);
2105                 if (ret)
2106                         break;
2107
2108         }
2109         if (ret > 0)
2110                 ret = 0;
2111 err:
2112         sk->nr_items = num_found;
2113         btrfs_free_path(path);
2114         return ret;
2115 }
2116
2117 static noinline int btrfs_ioctl_tree_search(struct file *file,
2118                                            void __user *argp)
2119 {
2120         struct btrfs_ioctl_search_args __user *uargs;
2121         struct btrfs_ioctl_search_key sk;
2122         struct inode *inode;
2123         int ret;
2124         size_t buf_size;
2125
2126         if (!capable(CAP_SYS_ADMIN))
2127                 return -EPERM;
2128
2129         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2130
2131         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2132                 return -EFAULT;
2133
2134         buf_size = sizeof(uargs->buf);
2135
2136         inode = file_inode(file);
2137         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2138
2139         /*
2140          * In the origin implementation an overflow is handled by returning a
2141          * search header with a len of zero, so reset ret.
2142          */
2143         if (ret == -EOVERFLOW)
2144                 ret = 0;
2145
2146         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2147                 ret = -EFAULT;
2148         return ret;
2149 }
2150
2151 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2152                                                void __user *argp)
2153 {
2154         struct btrfs_ioctl_search_args_v2 __user *uarg;
2155         struct btrfs_ioctl_search_args_v2 args;
2156         struct inode *inode;
2157         int ret;
2158         size_t buf_size;
2159         const size_t buf_limit = 16 * 1024 * 1024;
2160
2161         if (!capable(CAP_SYS_ADMIN))
2162                 return -EPERM;
2163
2164         /* copy search header and buffer size */
2165         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2166         if (copy_from_user(&args, uarg, sizeof(args)))
2167                 return -EFAULT;
2168
2169         buf_size = args.buf_size;
2170
2171         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2172                 return -EOVERFLOW;
2173
2174         /* limit result size to 16MB */
2175         if (buf_size > buf_limit)
2176                 buf_size = buf_limit;
2177
2178         inode = file_inode(file);
2179         ret = search_ioctl(inode, &args.key, &buf_size,
2180                            (char *)(&uarg->buf[0]));
2181         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2182                 ret = -EFAULT;
2183         else if (ret == -EOVERFLOW &&
2184                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2185                 ret = -EFAULT;
2186
2187         return ret;
2188 }
2189
2190 /*
2191  * Search INODE_REFs to identify path name of 'dirid' directory
2192  * in a 'tree_id' tree. and sets path name to 'name'.
2193  */
2194 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2195                                 u64 tree_id, u64 dirid, char *name)
2196 {
2197         struct btrfs_root *root;
2198         struct btrfs_key key;
2199         char *ptr;
2200         int ret = -1;
2201         int slot;
2202         int len;
2203         int total_len = 0;
2204         struct btrfs_inode_ref *iref;
2205         struct extent_buffer *l;
2206         struct btrfs_path *path;
2207
2208         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2209                 name[0]='\0';
2210                 return 0;
2211         }
2212
2213         path = btrfs_alloc_path();
2214         if (!path)
2215                 return -ENOMEM;
2216
2217         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2218
2219         key.objectid = tree_id;
2220         key.type = BTRFS_ROOT_ITEM_KEY;
2221         key.offset = (u64)-1;
2222         root = btrfs_read_fs_root_no_name(info, &key);
2223         if (IS_ERR(root)) {
2224                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2225                 ret = -ENOENT;
2226                 goto out;
2227         }
2228
2229         key.objectid = dirid;
2230         key.type = BTRFS_INODE_REF_KEY;
2231         key.offset = (u64)-1;
2232
2233         while (1) {
2234                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2235                 if (ret < 0)
2236                         goto out;
2237                 else if (ret > 0) {
2238                         ret = btrfs_previous_item(root, path, dirid,
2239                                                   BTRFS_INODE_REF_KEY);
2240                         if (ret < 0)
2241                                 goto out;
2242                         else if (ret > 0) {
2243                                 ret = -ENOENT;
2244                                 goto out;
2245                         }
2246                 }
2247
2248                 l = path->nodes[0];
2249                 slot = path->slots[0];
2250                 btrfs_item_key_to_cpu(l, &key, slot);
2251
2252                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2253                 len = btrfs_inode_ref_name_len(l, iref);
2254                 ptr -= len + 1;
2255                 total_len += len + 1;
2256                 if (ptr < name) {
2257                         ret = -ENAMETOOLONG;
2258                         goto out;
2259                 }
2260
2261                 *(ptr + len) = '/';
2262                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2263
2264                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2265                         break;
2266
2267                 btrfs_release_path(path);
2268                 key.objectid = key.offset;
2269                 key.offset = (u64)-1;
2270                 dirid = key.objectid;
2271         }
2272         memmove(name, ptr, total_len);
2273         name[total_len] = '\0';
2274         ret = 0;
2275 out:
2276         btrfs_free_path(path);
2277         return ret;
2278 }
2279
2280 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2281                                            void __user *argp)
2282 {
2283          struct btrfs_ioctl_ino_lookup_args *args;
2284          struct inode *inode;
2285         int ret = 0;
2286
2287         args = memdup_user(argp, sizeof(*args));
2288         if (IS_ERR(args))
2289                 return PTR_ERR(args);
2290
2291         inode = file_inode(file);
2292
2293         /*
2294          * Unprivileged query to obtain the containing subvolume root id. The
2295          * path is reset so it's consistent with btrfs_search_path_in_tree.
2296          */
2297         if (args->treeid == 0)
2298                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2299
2300         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2301                 args->name[0] = 0;
2302                 goto out;
2303         }
2304
2305         if (!capable(CAP_SYS_ADMIN)) {
2306                 ret = -EPERM;
2307                 goto out;
2308         }
2309
2310         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2311                                         args->treeid, args->objectid,
2312                                         args->name);
2313
2314 out:
2315         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2316                 ret = -EFAULT;
2317
2318         kfree(args);
2319         return ret;
2320 }
2321
2322 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2323                                              void __user *arg)
2324 {
2325         struct dentry *parent = file->f_path.dentry;
2326         struct dentry *dentry;
2327         struct inode *dir = d_inode(parent);
2328         struct inode *inode;
2329         struct btrfs_root *root = BTRFS_I(dir)->root;
2330         struct btrfs_root *dest = NULL;
2331         struct btrfs_ioctl_vol_args *vol_args;
2332         struct btrfs_trans_handle *trans;
2333         struct btrfs_block_rsv block_rsv;
2334         u64 root_flags;
2335         u64 qgroup_reserved;
2336         int namelen;
2337         int ret;
2338         int err = 0;
2339
2340         vol_args = memdup_user(arg, sizeof(*vol_args));
2341         if (IS_ERR(vol_args))
2342                 return PTR_ERR(vol_args);
2343
2344         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2345         namelen = strlen(vol_args->name);
2346         if (strchr(vol_args->name, '/') ||
2347             strncmp(vol_args->name, "..", namelen) == 0) {
2348                 err = -EINVAL;
2349                 goto out;
2350         }
2351
2352         err = mnt_want_write_file(file);
2353         if (err)
2354                 goto out;
2355
2356
2357         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2358         if (err == -EINTR)
2359                 goto out_drop_write;
2360         dentry = lookup_one_len(vol_args->name, parent, namelen);
2361         if (IS_ERR(dentry)) {
2362                 err = PTR_ERR(dentry);
2363                 goto out_unlock_dir;
2364         }
2365
2366         if (d_really_is_negative(dentry)) {
2367                 err = -ENOENT;
2368                 goto out_dput;
2369         }
2370
2371         inode = d_inode(dentry);
2372         dest = BTRFS_I(inode)->root;
2373         if (!capable(CAP_SYS_ADMIN)) {
2374                 /*
2375                  * Regular user.  Only allow this with a special mount
2376                  * option, when the user has write+exec access to the
2377                  * subvol root, and when rmdir(2) would have been
2378                  * allowed.
2379                  *
2380                  * Note that this is _not_ check that the subvol is
2381                  * empty or doesn't contain data that we wouldn't
2382                  * otherwise be able to delete.
2383                  *
2384                  * Users who want to delete empty subvols should try
2385                  * rmdir(2).
2386                  */
2387                 err = -EPERM;
2388                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2389                         goto out_dput;
2390
2391                 /*
2392                  * Do not allow deletion if the parent dir is the same
2393                  * as the dir to be deleted.  That means the ioctl
2394                  * must be called on the dentry referencing the root
2395                  * of the subvol, not a random directory contained
2396                  * within it.
2397                  */
2398                 err = -EINVAL;
2399                 if (root == dest)
2400                         goto out_dput;
2401
2402                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2403                 if (err)
2404                         goto out_dput;
2405         }
2406
2407         /* check if subvolume may be deleted by a user */
2408         err = btrfs_may_delete(dir, dentry, 1);
2409         if (err)
2410                 goto out_dput;
2411
2412         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2413                 err = -EINVAL;
2414                 goto out_dput;
2415         }
2416
2417         mutex_lock(&inode->i_mutex);
2418
2419         /*
2420          * Don't allow to delete a subvolume with send in progress. This is
2421          * inside the i_mutex so the error handling that has to drop the bit
2422          * again is not run concurrently.
2423          */
2424         spin_lock(&dest->root_item_lock);
2425         root_flags = btrfs_root_flags(&dest->root_item);
2426         if (dest->send_in_progress == 0) {
2427                 btrfs_set_root_flags(&dest->root_item,
2428                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2429                 spin_unlock(&dest->root_item_lock);
2430         } else {
2431                 spin_unlock(&dest->root_item_lock);
2432                 btrfs_warn(root->fs_info,
2433                         "Attempt to delete subvolume %llu during send",
2434                         dest->root_key.objectid);
2435                 err = -EPERM;
2436                 goto out_unlock_inode;
2437         }
2438
2439         down_write(&root->fs_info->subvol_sem);
2440
2441         err = may_destroy_subvol(dest);
2442         if (err)
2443                 goto out_up_write;
2444
2445         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2446         /*
2447          * One for dir inode, two for dir entries, two for root
2448          * ref/backref.
2449          */
2450         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2451                                                5, &qgroup_reserved, true);
2452         if (err)
2453                 goto out_up_write;
2454
2455         trans = btrfs_start_transaction(root, 0);
2456         if (IS_ERR(trans)) {
2457                 err = PTR_ERR(trans);
2458                 goto out_release;
2459         }
2460         trans->block_rsv = &block_rsv;
2461         trans->bytes_reserved = block_rsv.size;
2462
2463         ret = btrfs_unlink_subvol(trans, root, dir,
2464                                 dest->root_key.objectid,
2465                                 dentry->d_name.name,
2466                                 dentry->d_name.len);
2467         if (ret) {
2468                 err = ret;
2469                 btrfs_abort_transaction(trans, root, ret);
2470                 goto out_end_trans;
2471         }
2472
2473         btrfs_record_root_in_trans(trans, dest);
2474
2475         memset(&dest->root_item.drop_progress, 0,
2476                 sizeof(dest->root_item.drop_progress));
2477         dest->root_item.drop_level = 0;
2478         btrfs_set_root_refs(&dest->root_item, 0);
2479
2480         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2481                 ret = btrfs_insert_orphan_item(trans,
2482                                         root->fs_info->tree_root,
2483                                         dest->root_key.objectid);
2484                 if (ret) {
2485                         btrfs_abort_transaction(trans, root, ret);
2486                         err = ret;
2487                         goto out_end_trans;
2488                 }
2489         }
2490
2491         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2492                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2493                                   dest->root_key.objectid);
2494         if (ret && ret != -ENOENT) {
2495                 btrfs_abort_transaction(trans, root, ret);
2496                 err = ret;
2497                 goto out_end_trans;
2498         }
2499         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2500                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2501                                           dest->root_item.received_uuid,
2502                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2503                                           dest->root_key.objectid);
2504                 if (ret && ret != -ENOENT) {
2505                         btrfs_abort_transaction(trans, root, ret);
2506                         err = ret;
2507                         goto out_end_trans;
2508                 }
2509         }
2510
2511 out_end_trans:
2512         trans->block_rsv = NULL;
2513         trans->bytes_reserved = 0;
2514         ret = btrfs_end_transaction(trans, root);
2515         if (ret && !err)
2516                 err = ret;
2517         inode->i_flags |= S_DEAD;
2518 out_release:
2519         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2520 out_up_write:
2521         up_write(&root->fs_info->subvol_sem);
2522         if (err) {
2523                 spin_lock(&dest->root_item_lock);
2524                 root_flags = btrfs_root_flags(&dest->root_item);
2525                 btrfs_set_root_flags(&dest->root_item,
2526                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2527                 spin_unlock(&dest->root_item_lock);
2528         }
2529 out_unlock_inode:
2530         mutex_unlock(&inode->i_mutex);
2531         if (!err) {
2532                 d_invalidate(dentry);
2533                 btrfs_invalidate_inodes(dest);
2534                 d_delete(dentry);
2535                 ASSERT(dest->send_in_progress == 0);
2536
2537                 /* the last ref */
2538                 if (dest->ino_cache_inode) {
2539                         iput(dest->ino_cache_inode);
2540                         dest->ino_cache_inode = NULL;
2541                 }
2542         }
2543 out_dput:
2544         dput(dentry);
2545 out_unlock_dir:
2546         mutex_unlock(&dir->i_mutex);
2547 out_drop_write:
2548         mnt_drop_write_file(file);
2549 out:
2550         kfree(vol_args);
2551         return err;
2552 }
2553
2554 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2555 {
2556         struct inode *inode = file_inode(file);
2557         struct btrfs_root *root = BTRFS_I(inode)->root;
2558         struct btrfs_ioctl_defrag_range_args *range;
2559         int ret;
2560
2561         ret = mnt_want_write_file(file);
2562         if (ret)
2563                 return ret;
2564
2565         if (btrfs_root_readonly(root)) {
2566                 ret = -EROFS;
2567                 goto out;
2568         }
2569
2570         switch (inode->i_mode & S_IFMT) {
2571         case S_IFDIR:
2572                 if (!capable(CAP_SYS_ADMIN)) {
2573                         ret = -EPERM;
2574                         goto out;
2575                 }
2576                 ret = btrfs_defrag_root(root);
2577                 if (ret)
2578                         goto out;
2579                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2580                 break;
2581         case S_IFREG:
2582                 if (!(file->f_mode & FMODE_WRITE)) {
2583                         ret = -EINVAL;
2584                         goto out;
2585                 }
2586
2587                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2588                 if (!range) {
2589                         ret = -ENOMEM;
2590                         goto out;
2591                 }
2592
2593                 if (argp) {
2594                         if (copy_from_user(range, argp,
2595                                            sizeof(*range))) {
2596                                 ret = -EFAULT;
2597                                 kfree(range);
2598                                 goto out;
2599                         }
2600                         /* compression requires us to start the IO */
2601                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2602                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2603                                 range->extent_thresh = (u32)-1;
2604                         }
2605                 } else {
2606                         /* the rest are all set to zero by kzalloc */
2607                         range->len = (u64)-1;
2608                 }
2609                 ret = btrfs_defrag_file(file_inode(file), file,
2610                                         range, 0, 0);
2611                 if (ret > 0)
2612                         ret = 0;
2613                 kfree(range);
2614                 break;
2615         default:
2616                 ret = -EINVAL;
2617         }
2618 out:
2619         mnt_drop_write_file(file);
2620         return ret;
2621 }
2622
2623 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2624 {
2625         struct btrfs_ioctl_vol_args *vol_args;
2626         int ret;
2627
2628         if (!capable(CAP_SYS_ADMIN))
2629                 return -EPERM;
2630
2631         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2632                         1)) {
2633                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2634         }
2635
2636         mutex_lock(&root->fs_info->volume_mutex);
2637         vol_args = memdup_user(arg, sizeof(*vol_args));
2638         if (IS_ERR(vol_args)) {
2639                 ret = PTR_ERR(vol_args);
2640                 goto out;
2641         }
2642
2643         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2644         ret = btrfs_init_new_device(root, vol_args->name);
2645
2646         if (!ret)
2647                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2648
2649         kfree(vol_args);
2650 out:
2651         mutex_unlock(&root->fs_info->volume_mutex);
2652         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2653         return ret;
2654 }
2655
2656 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2657 {
2658         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2659         struct btrfs_ioctl_vol_args *vol_args;
2660         int ret;
2661
2662         if (!capable(CAP_SYS_ADMIN))
2663                 return -EPERM;
2664
2665         ret = mnt_want_write_file(file);
2666         if (ret)
2667                 return ret;
2668
2669         vol_args = memdup_user(arg, sizeof(*vol_args));
2670         if (IS_ERR(vol_args)) {
2671                 ret = PTR_ERR(vol_args);
2672                 goto err_drop;
2673         }
2674
2675         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2676
2677         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2678                         1)) {
2679                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2680                 goto out;
2681         }
2682
2683         mutex_lock(&root->fs_info->volume_mutex);
2684         ret = btrfs_rm_device(root, vol_args->name);
2685         mutex_unlock(&root->fs_info->volume_mutex);
2686         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2687
2688         if (!ret)
2689                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2690
2691 out:
2692         kfree(vol_args);
2693 err_drop:
2694         mnt_drop_write_file(file);
2695         return ret;
2696 }
2697
2698 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2699 {
2700         struct btrfs_ioctl_fs_info_args *fi_args;
2701         struct btrfs_device *device;
2702         struct btrfs_device *next;
2703         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2704         int ret = 0;
2705
2706         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2707         if (!fi_args)
2708                 return -ENOMEM;
2709
2710         mutex_lock(&fs_devices->device_list_mutex);
2711         fi_args->num_devices = fs_devices->num_devices;
2712         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2713
2714         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2715                 if (device->devid > fi_args->max_id)
2716                         fi_args->max_id = device->devid;
2717         }
2718         mutex_unlock(&fs_devices->device_list_mutex);
2719
2720         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2721         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2722         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2723
2724         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2725                 ret = -EFAULT;
2726
2727         kfree(fi_args);
2728         return ret;
2729 }
2730
2731 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2732 {
2733         struct btrfs_ioctl_dev_info_args *di_args;
2734         struct btrfs_device *dev;
2735         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2736         int ret = 0;
2737         char *s_uuid = NULL;
2738
2739         di_args = memdup_user(arg, sizeof(*di_args));
2740         if (IS_ERR(di_args))
2741                 return PTR_ERR(di_args);
2742
2743         if (!btrfs_is_empty_uuid(di_args->uuid))
2744                 s_uuid = di_args->uuid;
2745
2746         mutex_lock(&fs_devices->device_list_mutex);
2747         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2748
2749         if (!dev) {
2750                 ret = -ENODEV;
2751                 goto out;
2752         }
2753
2754         di_args->devid = dev->devid;
2755         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2756         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2757         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2758         if (dev->name) {
2759                 struct rcu_string *name;
2760
2761                 rcu_read_lock();
2762                 name = rcu_dereference(dev->name);
2763                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2764                 rcu_read_unlock();
2765                 di_args->path[sizeof(di_args->path) - 1] = 0;
2766         } else {
2767                 di_args->path[0] = '\0';
2768         }
2769
2770 out:
2771         mutex_unlock(&fs_devices->device_list_mutex);
2772         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2773                 ret = -EFAULT;
2774
2775         kfree(di_args);
2776         return ret;
2777 }
2778
2779 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2780 {
2781         struct page *page;
2782         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2783
2784         page = grab_cache_page(inode->i_mapping, index);
2785         if (!page)
2786                 return NULL;
2787
2788         if (!PageUptodate(page)) {
2789                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2790                                                  0))
2791                         return NULL;
2792                 lock_page(page);
2793                 if (!PageUptodate(page)) {
2794                         unlock_page(page);
2795                         page_cache_release(page);
2796                         return NULL;
2797                 }
2798         }
2799         unlock_page(page);
2800
2801         return page;
2802 }
2803
2804 static int gather_extent_pages(struct inode *inode, struct page **pages,
2805                                int num_pages, u64 off)
2806 {
2807         int i;
2808         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2809
2810         for (i = 0; i < num_pages; i++) {
2811                 pages[i] = extent_same_get_page(inode, index + i);
2812                 if (!pages[i])
2813                         return -ENOMEM;
2814         }
2815         return 0;
2816 }
2817
2818 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2819 {
2820         /* do any pending delalloc/csum calc on src, one way or
2821            another, and lock file content */
2822         while (1) {
2823                 struct btrfs_ordered_extent *ordered;
2824                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2825                 ordered = btrfs_lookup_first_ordered_extent(inode,
2826                                                             off + len - 1);
2827                 if ((!ordered ||
2828                      ordered->file_offset + ordered->len <= off ||
2829                      ordered->file_offset >= off + len) &&
2830                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2831                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2832                         if (ordered)
2833                                 btrfs_put_ordered_extent(ordered);
2834                         break;
2835                 }
2836                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2837                 if (ordered)
2838                         btrfs_put_ordered_extent(ordered);
2839                 btrfs_wait_ordered_range(inode, off, len);
2840         }
2841 }
2842
2843 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2844 {
2845         mutex_unlock(&inode1->i_mutex);
2846         mutex_unlock(&inode2->i_mutex);
2847 }
2848
2849 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2850 {
2851         if (inode1 < inode2)
2852                 swap(inode1, inode2);
2853
2854         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2855         mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2856 }
2857
2858 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2859                                       struct inode *inode2, u64 loff2, u64 len)
2860 {
2861         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2862         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2863 }
2864
2865 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2866                                      struct inode *inode2, u64 loff2, u64 len)
2867 {
2868         if (inode1 < inode2) {
2869                 swap(inode1, inode2);
2870                 swap(loff1, loff2);
2871         }
2872         lock_extent_range(inode1, loff1, len);
2873         lock_extent_range(inode2, loff2, len);
2874 }
2875
2876 struct cmp_pages {
2877         int             num_pages;
2878         struct page     **src_pages;
2879         struct page     **dst_pages;
2880 };
2881
2882 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2883 {
2884         int i;
2885         struct page *pg;
2886
2887         for (i = 0; i < cmp->num_pages; i++) {
2888                 pg = cmp->src_pages[i];
2889                 if (pg)
2890                         page_cache_release(pg);
2891                 pg = cmp->dst_pages[i];
2892                 if (pg)
2893                         page_cache_release(pg);
2894         }
2895         kfree(cmp->src_pages);
2896         kfree(cmp->dst_pages);
2897 }
2898
2899 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2900                                   struct inode *dst, u64 dst_loff,
2901                                   u64 len, struct cmp_pages *cmp)
2902 {
2903         int ret;
2904         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2905         struct page **src_pgarr, **dst_pgarr;
2906
2907         /*
2908          * We must gather up all the pages before we initiate our
2909          * extent locking. We use an array for the page pointers. Size
2910          * of the array is bounded by len, which is in turn bounded by
2911          * BTRFS_MAX_DEDUPE_LEN.
2912          */
2913         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2914         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2915         if (!src_pgarr || !dst_pgarr) {
2916                 kfree(src_pgarr);
2917                 kfree(dst_pgarr);
2918                 return -ENOMEM;
2919         }
2920         cmp->num_pages = num_pages;
2921         cmp->src_pages = src_pgarr;
2922         cmp->dst_pages = dst_pgarr;
2923
2924         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2925         if (ret)
2926                 goto out;
2927
2928         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2929
2930 out:
2931         if (ret)
2932                 btrfs_cmp_data_free(cmp);
2933         return 0;
2934 }
2935
2936 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2937                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2938 {
2939         int ret = 0;
2940         int i;
2941         struct page *src_page, *dst_page;
2942         unsigned int cmp_len = PAGE_CACHE_SIZE;
2943         void *addr, *dst_addr;
2944
2945         i = 0;
2946         while (len) {
2947                 if (len < PAGE_CACHE_SIZE)
2948                         cmp_len = len;
2949
2950                 BUG_ON(i >= cmp->num_pages);
2951
2952                 src_page = cmp->src_pages[i];
2953                 dst_page = cmp->dst_pages[i];
2954
2955                 addr = kmap_atomic(src_page);
2956                 dst_addr = kmap_atomic(dst_page);
2957
2958                 flush_dcache_page(src_page);
2959                 flush_dcache_page(dst_page);
2960
2961                 if (memcmp(addr, dst_addr, cmp_len))
2962                         ret = BTRFS_SAME_DATA_DIFFERS;
2963
2964                 kunmap_atomic(addr);
2965                 kunmap_atomic(dst_addr);
2966
2967                 if (ret)
2968                         break;
2969
2970                 len -= cmp_len;
2971                 i++;
2972         }
2973
2974         return ret;
2975 }
2976
2977 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
2978                                      u64 olen)
2979 {
2980         u64 len = *plen;
2981         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2982
2983         if (off + olen > inode->i_size || off + olen < off)
2984                 return -EINVAL;
2985
2986         /* if we extend to eof, continue to block boundary */
2987         if (off + len == inode->i_size)
2988                 *plen = len = ALIGN(inode->i_size, bs) - off;
2989
2990         /* Check that we are block aligned - btrfs_clone() requires this */
2991         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2992                 return -EINVAL;
2993
2994         return 0;
2995 }
2996
2997 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
2998                              struct inode *dst, u64 dst_loff)
2999 {
3000         int ret;
3001         u64 len = olen;
3002         struct cmp_pages cmp;
3003         int same_inode = 0;
3004         u64 same_lock_start = 0;
3005         u64 same_lock_len = 0;
3006
3007         if (src == dst)
3008                 same_inode = 1;
3009
3010         if (len == 0)
3011                 return 0;
3012
3013         if (same_inode) {
3014                 mutex_lock(&src->i_mutex);
3015
3016                 ret = extent_same_check_offsets(src, loff, &len, olen);
3017                 if (ret)
3018                         goto out_unlock;
3019
3020                 /*
3021                  * Single inode case wants the same checks, except we
3022                  * don't want our length pushed out past i_size as
3023                  * comparing that data range makes no sense.
3024                  *
3025                  * extent_same_check_offsets() will do this for an
3026                  * unaligned length at i_size, so catch it here and
3027                  * reject the request.
3028                  *
3029                  * This effectively means we require aligned extents
3030                  * for the single-inode case, whereas the other cases
3031                  * allow an unaligned length so long as it ends at
3032                  * i_size.
3033                  */
3034                 if (len != olen) {
3035                         ret = -EINVAL;
3036                         goto out_unlock;
3037                 }
3038
3039                 /* Check for overlapping ranges */
3040                 if (dst_loff + len > loff && dst_loff < loff + len) {
3041                         ret = -EINVAL;
3042                         goto out_unlock;
3043                 }
3044
3045                 same_lock_start = min_t(u64, loff, dst_loff);
3046                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3047         } else {
3048                 btrfs_double_inode_lock(src, dst);
3049
3050                 ret = extent_same_check_offsets(src, loff, &len, olen);
3051                 if (ret)
3052                         goto out_unlock;
3053
3054                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3055                 if (ret)
3056                         goto out_unlock;
3057         }
3058
3059         /* don't make the dst file partly checksummed */
3060         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3061             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3062                 ret = -EINVAL;
3063                 goto out_unlock;
3064         }
3065
3066         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3067         if (ret)
3068                 goto out_unlock;
3069
3070         if (same_inode)
3071                 lock_extent_range(src, same_lock_start, same_lock_len);
3072         else
3073                 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3074
3075         /* pass original length for comparison so we stay within i_size */
3076         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3077         if (ret == 0)
3078                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3079
3080         if (same_inode)
3081                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3082                               same_lock_start + same_lock_len - 1);
3083         else
3084                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3085
3086         btrfs_cmp_data_free(&cmp);
3087 out_unlock:
3088         if (same_inode)
3089                 mutex_unlock(&src->i_mutex);
3090         else
3091                 btrfs_double_inode_unlock(src, dst);
3092
3093         return ret;
3094 }
3095
3096 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
3097
3098 static long btrfs_ioctl_file_extent_same(struct file *file,
3099                         struct btrfs_ioctl_same_args __user *argp)
3100 {
3101         struct btrfs_ioctl_same_args *same = NULL;
3102         struct btrfs_ioctl_same_extent_info *info;
3103         struct inode *src = file_inode(file);
3104         u64 off;
3105         u64 len;
3106         int i;
3107         int ret;
3108         unsigned long size;
3109         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3110         bool is_admin = capable(CAP_SYS_ADMIN);
3111         u16 count;
3112
3113         if (!(file->f_mode & FMODE_READ))
3114                 return -EINVAL;
3115
3116         ret = mnt_want_write_file(file);
3117         if (ret)
3118                 return ret;
3119
3120         if (get_user(count, &argp->dest_count)) {
3121                 ret = -EFAULT;
3122                 goto out;
3123         }
3124
3125         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3126
3127         same = memdup_user(argp, size);
3128
3129         if (IS_ERR(same)) {
3130                 ret = PTR_ERR(same);
3131                 same = NULL;
3132                 goto out;
3133         }
3134
3135         off = same->logical_offset;
3136         len = same->length;
3137
3138         /*
3139          * Limit the total length we will dedupe for each operation.
3140          * This is intended to bound the total time spent in this
3141          * ioctl to something sane.
3142          */
3143         if (len > BTRFS_MAX_DEDUPE_LEN)
3144                 len = BTRFS_MAX_DEDUPE_LEN;
3145
3146         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3147                 /*
3148                  * Btrfs does not support blocksize < page_size. As a
3149                  * result, btrfs_cmp_data() won't correctly handle
3150                  * this situation without an update.
3151                  */
3152                 ret = -EINVAL;
3153                 goto out;
3154         }
3155
3156         ret = -EISDIR;
3157         if (S_ISDIR(src->i_mode))
3158                 goto out;
3159
3160         ret = -EACCES;
3161         if (!S_ISREG(src->i_mode))
3162                 goto out;
3163
3164         /* pre-format output fields to sane values */
3165         for (i = 0; i < count; i++) {
3166                 same->info[i].bytes_deduped = 0ULL;
3167                 same->info[i].status = 0;
3168         }
3169
3170         for (i = 0, info = same->info; i < count; i++, info++) {
3171                 struct inode *dst;
3172                 struct fd dst_file = fdget(info->fd);
3173                 if (!dst_file.file) {
3174                         info->status = -EBADF;
3175                         continue;
3176                 }
3177                 dst = file_inode(dst_file.file);
3178
3179                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3180                         info->status = -EINVAL;
3181                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3182                         info->status = -EXDEV;
3183                 } else if (S_ISDIR(dst->i_mode)) {
3184                         info->status = -EISDIR;
3185                 } else if (!S_ISREG(dst->i_mode)) {
3186                         info->status = -EACCES;
3187                 } else {
3188                         info->status = btrfs_extent_same(src, off, len, dst,
3189                                                         info->logical_offset);
3190                         if (info->status == 0)
3191                                 info->bytes_deduped += len;
3192                 }
3193                 fdput(dst_file);
3194         }
3195
3196         ret = copy_to_user(argp, same, size);
3197         if (ret)
3198                 ret = -EFAULT;
3199
3200 out:
3201         mnt_drop_write_file(file);
3202         kfree(same);
3203         return ret;
3204 }
3205
3206 /* Helper to check and see if this root currently has a ref on the given disk
3207  * bytenr.  If it does then we need to update the quota for this root.  This
3208  * doesn't do anything if quotas aren't enabled.
3209  */
3210 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3211                      u64 disko)
3212 {
3213         struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
3214         struct ulist *roots;
3215         struct ulist_iterator uiter;
3216         struct ulist_node *root_node = NULL;
3217         int ret;
3218
3219         if (!root->fs_info->quota_enabled)
3220                 return 1;
3221
3222         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3223         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
3224                                    tree_mod_seq_elem.seq, &roots);
3225         if (ret < 0)
3226                 goto out;
3227         ret = 0;
3228         ULIST_ITER_INIT(&uiter);
3229         while ((root_node = ulist_next(roots, &uiter))) {
3230                 if (root_node->val == root->objectid) {
3231                         ret = 1;
3232                         break;
3233                 }
3234         }
3235         ulist_free(roots);
3236 out:
3237         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3238         return ret;
3239 }
3240
3241 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3242                                      struct inode *inode,
3243                                      u64 endoff,
3244                                      const u64 destoff,
3245                                      const u64 olen,
3246                                      int no_time_update)
3247 {
3248         struct btrfs_root *root = BTRFS_I(inode)->root;
3249         int ret;
3250
3251         inode_inc_iversion(inode);
3252         if (!no_time_update)
3253                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3254         /*
3255          * We round up to the block size at eof when determining which
3256          * extents to clone above, but shouldn't round up the file size.
3257          */
3258         if (endoff > destoff + olen)
3259                 endoff = destoff + olen;
3260         if (endoff > inode->i_size)
3261                 btrfs_i_size_write(inode, endoff);
3262
3263         ret = btrfs_update_inode(trans, root, inode);
3264         if (ret) {
3265                 btrfs_abort_transaction(trans, root, ret);
3266                 btrfs_end_transaction(trans, root);
3267                 goto out;
3268         }
3269         ret = btrfs_end_transaction(trans, root);
3270 out:
3271         return ret;
3272 }
3273
3274 static void clone_update_extent_map(struct inode *inode,
3275                                     const struct btrfs_trans_handle *trans,
3276                                     const struct btrfs_path *path,
3277                                     const u64 hole_offset,
3278                                     const u64 hole_len)
3279 {
3280         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3281         struct extent_map *em;
3282         int ret;
3283
3284         em = alloc_extent_map();
3285         if (!em) {
3286                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3287                         &BTRFS_I(inode)->runtime_flags);
3288                 return;
3289         }
3290
3291         if (path) {
3292                 struct btrfs_file_extent_item *fi;
3293
3294                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3295                                     struct btrfs_file_extent_item);
3296                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3297                 em->generation = -1;
3298                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3299                     BTRFS_FILE_EXTENT_INLINE)
3300                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3301                                 &BTRFS_I(inode)->runtime_flags);
3302         } else {
3303                 em->start = hole_offset;
3304                 em->len = hole_len;
3305                 em->ram_bytes = em->len;
3306                 em->orig_start = hole_offset;
3307                 em->block_start = EXTENT_MAP_HOLE;
3308                 em->block_len = 0;
3309                 em->orig_block_len = 0;
3310                 em->compress_type = BTRFS_COMPRESS_NONE;
3311                 em->generation = trans->transid;
3312         }
3313
3314         while (1) {
3315                 write_lock(&em_tree->lock);
3316                 ret = add_extent_mapping(em_tree, em, 1);
3317                 write_unlock(&em_tree->lock);
3318                 if (ret != -EEXIST) {
3319                         free_extent_map(em);
3320                         break;
3321                 }
3322                 btrfs_drop_extent_cache(inode, em->start,
3323                                         em->start + em->len - 1, 0);
3324         }
3325
3326         if (ret)
3327                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3328                         &BTRFS_I(inode)->runtime_flags);
3329 }
3330
3331 /**
3332  * btrfs_clone() - clone a range from inode file to another
3333  *
3334  * @src: Inode to clone from
3335  * @inode: Inode to clone to
3336  * @off: Offset within source to start clone from
3337  * @olen: Original length, passed by user, of range to clone
3338  * @olen_aligned: Block-aligned value of olen
3339  * @destoff: Offset within @inode to start clone
3340  * @no_time_update: Whether to update mtime/ctime on the target inode
3341  */
3342 static int btrfs_clone(struct inode *src, struct inode *inode,
3343                        const u64 off, const u64 olen, const u64 olen_aligned,
3344                        const u64 destoff, int no_time_update)
3345 {
3346         struct btrfs_root *root = BTRFS_I(inode)->root;
3347         struct btrfs_path *path = NULL;
3348         struct extent_buffer *leaf;
3349         struct btrfs_trans_handle *trans;
3350         char *buf = NULL;
3351         struct btrfs_key key;
3352         u32 nritems;
3353         int slot;
3354         int ret;
3355         int no_quota;
3356         const u64 len = olen_aligned;
3357         u64 last_disko = 0;
3358         u64 last_dest_end = destoff;
3359
3360         ret = -ENOMEM;
3361         buf = vmalloc(root->nodesize);
3362         if (!buf)
3363                 return ret;
3364
3365         path = btrfs_alloc_path();
3366         if (!path) {
3367                 vfree(buf);
3368                 return ret;
3369         }
3370
3371         path->reada = 2;
3372         /* clone data */
3373         key.objectid = btrfs_ino(src);
3374         key.type = BTRFS_EXTENT_DATA_KEY;
3375         key.offset = off;
3376
3377         while (1) {
3378                 u64 next_key_min_offset = key.offset + 1;
3379
3380                 /*
3381                  * note the key will change type as we walk through the
3382                  * tree.
3383                  */
3384                 path->leave_spinning = 1;
3385                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3386                                 0, 0);
3387                 if (ret < 0)
3388                         goto out;
3389                 /*
3390                  * First search, if no extent item that starts at offset off was
3391                  * found but the previous item is an extent item, it's possible
3392                  * it might overlap our target range, therefore process it.
3393                  */
3394                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3395                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3396                                               path->slots[0] - 1);
3397                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3398                                 path->slots[0]--;
3399                 }
3400
3401                 nritems = btrfs_header_nritems(path->nodes[0]);
3402 process_slot:
3403                 no_quota = 1;
3404                 if (path->slots[0] >= nritems) {
3405                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3406                         if (ret < 0)
3407                                 goto out;
3408                         if (ret > 0)
3409                                 break;
3410                         nritems = btrfs_header_nritems(path->nodes[0]);
3411                 }
3412                 leaf = path->nodes[0];
3413                 slot = path->slots[0];
3414
3415                 btrfs_item_key_to_cpu(leaf, &key, slot);
3416                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3417                     key.objectid != btrfs_ino(src))
3418                         break;
3419
3420                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3421                         struct btrfs_file_extent_item *extent;
3422                         int type;
3423                         u32 size;
3424                         struct btrfs_key new_key;
3425                         u64 disko = 0, diskl = 0;
3426                         u64 datao = 0, datal = 0;
3427                         u8 comp;
3428                         u64 drop_start;
3429
3430                         extent = btrfs_item_ptr(leaf, slot,
3431                                                 struct btrfs_file_extent_item);
3432                         comp = btrfs_file_extent_compression(leaf, extent);
3433                         type = btrfs_file_extent_type(leaf, extent);
3434                         if (type == BTRFS_FILE_EXTENT_REG ||
3435                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3436                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3437                                                                       extent);
3438                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3439                                                                  extent);
3440                                 datao = btrfs_file_extent_offset(leaf, extent);
3441                                 datal = btrfs_file_extent_num_bytes(leaf,
3442                                                                     extent);
3443                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3444                                 /* take upper bound, may be compressed */
3445                                 datal = btrfs_file_extent_ram_bytes(leaf,
3446                                                                     extent);
3447                         }
3448
3449                         /*
3450                          * The first search might have left us at an extent
3451                          * item that ends before our target range's start, can
3452                          * happen if we have holes and NO_HOLES feature enabled.
3453                          */
3454                         if (key.offset + datal <= off) {
3455                                 path->slots[0]++;
3456                                 goto process_slot;
3457                         } else if (key.offset >= off + len) {
3458                                 break;
3459                         }
3460                         next_key_min_offset = key.offset + datal;
3461                         size = btrfs_item_size_nr(leaf, slot);
3462                         read_extent_buffer(leaf, buf,
3463                                            btrfs_item_ptr_offset(leaf, slot),
3464                                            size);
3465
3466                         btrfs_release_path(path);
3467                         path->leave_spinning = 0;
3468
3469                         memcpy(&new_key, &key, sizeof(new_key));
3470                         new_key.objectid = btrfs_ino(inode);
3471                         if (off <= key.offset)
3472                                 new_key.offset = key.offset + destoff - off;
3473                         else
3474                                 new_key.offset = destoff;
3475
3476                         /*
3477                          * Deal with a hole that doesn't have an extent item
3478                          * that represents it (NO_HOLES feature enabled).
3479                          * This hole is either in the middle of the cloning
3480                          * range or at the beginning (fully overlaps it or
3481                          * partially overlaps it).
3482                          */
3483                         if (new_key.offset != last_dest_end)
3484                                 drop_start = last_dest_end;
3485                         else
3486                                 drop_start = new_key.offset;
3487
3488                         /*
3489                          * 1 - adjusting old extent (we may have to split it)
3490                          * 1 - add new extent
3491                          * 1 - inode update
3492                          */
3493                         trans = btrfs_start_transaction(root, 3);
3494                         if (IS_ERR(trans)) {
3495                                 ret = PTR_ERR(trans);
3496                                 goto out;
3497                         }
3498
3499                         if (type == BTRFS_FILE_EXTENT_REG ||
3500                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3501                                 /*
3502                                  *    a  | --- range to clone ---|  b
3503                                  * | ------------- extent ------------- |
3504                                  */
3505
3506                                 /* subtract range b */
3507                                 if (key.offset + datal > off + len)
3508                                         datal = off + len - key.offset;
3509
3510                                 /* subtract range a */
3511                                 if (off > key.offset) {
3512                                         datao += off - key.offset;
3513                                         datal -= off - key.offset;
3514                                 }
3515
3516                                 ret = btrfs_drop_extents(trans, root, inode,
3517                                                          drop_start,
3518                                                          new_key.offset + datal,
3519                                                          1);
3520                                 if (ret) {
3521                                         if (ret != -EOPNOTSUPP)
3522                                                 btrfs_abort_transaction(trans,
3523                                                                 root, ret);
3524                                         btrfs_end_transaction(trans, root);
3525                                         goto out;
3526                                 }
3527
3528                                 ret = btrfs_insert_empty_item(trans, root, path,
3529                                                               &new_key, size);
3530                                 if (ret) {
3531                                         btrfs_abort_transaction(trans, root,
3532                                                                 ret);
3533                                         btrfs_end_transaction(trans, root);
3534                                         goto out;
3535                                 }
3536
3537                                 leaf = path->nodes[0];
3538                                 slot = path->slots[0];
3539                                 write_extent_buffer(leaf, buf,
3540                                             btrfs_item_ptr_offset(leaf, slot),
3541                                             size);
3542
3543                                 extent = btrfs_item_ptr(leaf, slot,
3544                                                 struct btrfs_file_extent_item);
3545
3546                                 /* disko == 0 means it's a hole */
3547                                 if (!disko)
3548                                         datao = 0;
3549
3550                                 btrfs_set_file_extent_offset(leaf, extent,
3551                                                              datao);
3552                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3553                                                                 datal);
3554
3555                                 /*
3556                                  * We need to look up the roots that point at
3557                                  * this bytenr and see if the new root does.  If
3558                                  * it does not we need to make sure we update
3559                                  * quotas appropriately.
3560                                  */
3561                                 if (disko && root != BTRFS_I(src)->root &&
3562                                     disko != last_disko) {
3563                                         no_quota = check_ref(trans, root,
3564                                                              disko);
3565                                         if (no_quota < 0) {
3566                                                 btrfs_abort_transaction(trans,
3567                                                                         root,
3568                                                                         ret);
3569                                                 btrfs_end_transaction(trans,
3570                                                                       root);
3571                                                 ret = no_quota;
3572                                                 goto out;
3573                                         }
3574                                 }
3575
3576                                 if (disko) {
3577                                         inode_add_bytes(inode, datal);
3578                                         ret = btrfs_inc_extent_ref(trans, root,
3579                                                         disko, diskl, 0,
3580                                                         root->root_key.objectid,
3581                                                         btrfs_ino(inode),
3582                                                         new_key.offset - datao,
3583                                                         no_quota);
3584                                         if (ret) {
3585                                                 btrfs_abort_transaction(trans,
3586                                                                         root,
3587                                                                         ret);
3588                                                 btrfs_end_transaction(trans,
3589                                                                       root);
3590                                                 goto out;
3591
3592                                         }
3593                                 }
3594                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3595                                 u64 skip = 0;
3596                                 u64 trim = 0;
3597                                 u64 aligned_end = 0;
3598
3599                                 /*
3600                                  * Don't copy an inline extent into an offset
3601                                  * greater than zero. Having an inline extent
3602                                  * at such an offset results in chaos as btrfs
3603                                  * isn't prepared for such cases. Just skip
3604                                  * this case for the same reasons as commented
3605                                  * at btrfs_ioctl_clone().
3606                                  */
3607                                 if (last_dest_end > 0) {
3608                                         ret = -EOPNOTSUPP;
3609                                         btrfs_end_transaction(trans, root);
3610                                         goto out;
3611                                 }
3612
3613                                 if (off > key.offset) {
3614                                         skip = off - key.offset;
3615                                         new_key.offset += skip;
3616                                 }
3617
3618                                 if (key.offset + datal > off + len)
3619                                         trim = key.offset + datal - (off + len);
3620
3621                                 if (comp && (skip || trim)) {
3622                                         ret = -EINVAL;
3623                                         btrfs_end_transaction(trans, root);
3624                                         goto out;
3625                                 }
3626                                 size -= skip + trim;
3627                                 datal -= skip + trim;
3628
3629                                 aligned_end = ALIGN(new_key.offset + datal,
3630                                                     root->sectorsize);
3631                                 ret = btrfs_drop_extents(trans, root, inode,
3632                                                          drop_start,
3633                                                          aligned_end,
3634                                                          1);
3635                                 if (ret) {
3636                                         if (ret != -EOPNOTSUPP)
3637                                                 btrfs_abort_transaction(trans,
3638                                                         root, ret);
3639                                         btrfs_end_transaction(trans, root);
3640                                         goto out;
3641                                 }
3642
3643                                 ret = btrfs_insert_empty_item(trans, root, path,
3644                                                               &new_key, size);
3645                                 if (ret) {
3646                                         btrfs_abort_transaction(trans, root,
3647                                                                 ret);
3648                                         btrfs_end_transaction(trans, root);
3649                                         goto out;
3650                                 }
3651
3652                                 if (skip) {
3653                                         u32 start =
3654                                           btrfs_file_extent_calc_inline_size(0);
3655                                         memmove(buf+start, buf+start+skip,
3656                                                 datal);
3657                                 }
3658
3659                                 leaf = path->nodes[0];
3660                                 slot = path->slots[0];
3661                                 write_extent_buffer(leaf, buf,
3662                                             btrfs_item_ptr_offset(leaf, slot),
3663                                             size);
3664                                 inode_add_bytes(inode, datal);
3665                         }
3666
3667                         /* If we have an implicit hole (NO_HOLES feature). */
3668                         if (drop_start < new_key.offset)
3669                                 clone_update_extent_map(inode, trans,
3670                                                 NULL, drop_start,
3671                                                 new_key.offset - drop_start);
3672
3673                         clone_update_extent_map(inode, trans, path, 0, 0);
3674
3675                         btrfs_mark_buffer_dirty(leaf);
3676                         btrfs_release_path(path);
3677
3678                         last_dest_end = ALIGN(new_key.offset + datal,
3679                                               root->sectorsize);
3680                         ret = clone_finish_inode_update(trans, inode,
3681                                                         last_dest_end,
3682                                                         destoff, olen,
3683                                                         no_time_update);
3684                         if (ret)
3685                                 goto out;
3686                         if (new_key.offset + datal >= destoff + len)
3687                                 break;
3688                 }
3689                 btrfs_release_path(path);
3690                 key.offset = next_key_min_offset;
3691         }
3692         ret = 0;
3693
3694         if (last_dest_end < destoff + len) {
3695                 /*
3696                  * We have an implicit hole (NO_HOLES feature is enabled) that
3697                  * fully or partially overlaps our cloning range at its end.
3698                  */
3699                 btrfs_release_path(path);
3700
3701                 /*
3702                  * 1 - remove extent(s)
3703                  * 1 - inode update
3704                  */
3705                 trans = btrfs_start_transaction(root, 2);
3706                 if (IS_ERR(trans)) {
3707                         ret = PTR_ERR(trans);
3708                         goto out;
3709                 }
3710                 ret = btrfs_drop_extents(trans, root, inode,
3711                                          last_dest_end, destoff + len, 1);
3712                 if (ret) {
3713                         if (ret != -EOPNOTSUPP)
3714                                 btrfs_abort_transaction(trans, root, ret);
3715                         btrfs_end_transaction(trans, root);
3716                         goto out;
3717                 }
3718                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3719                                         destoff + len - last_dest_end);
3720                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3721                                                 destoff, olen, no_time_update);
3722         }
3723
3724 out:
3725         btrfs_free_path(path);
3726         vfree(buf);
3727         return ret;
3728 }
3729
3730 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3731                                        u64 off, u64 olen, u64 destoff)
3732 {
3733         struct inode *inode = file_inode(file);
3734         struct btrfs_root *root = BTRFS_I(inode)->root;
3735         struct fd src_file;
3736         struct inode *src;
3737         int ret;
3738         u64 len = olen;
3739         u64 bs = root->fs_info->sb->s_blocksize;
3740         int same_inode = 0;
3741
3742         /*
3743          * TODO:
3744          * - split compressed inline extents.  annoying: we need to
3745          *   decompress into destination's address_space (the file offset
3746          *   may change, so source mapping won't do), then recompress (or
3747          *   otherwise reinsert) a subrange.
3748          *
3749          * - split destination inode's inline extents.  The inline extents can
3750          *   be either compressed or non-compressed.
3751          */
3752
3753         /* the destination must be opened for writing */
3754         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3755                 return -EINVAL;
3756
3757         if (btrfs_root_readonly(root))
3758                 return -EROFS;
3759
3760         ret = mnt_want_write_file(file);
3761         if (ret)
3762                 return ret;
3763
3764         src_file = fdget(srcfd);
3765         if (!src_file.file) {
3766                 ret = -EBADF;
3767                 goto out_drop_write;
3768         }
3769
3770         ret = -EXDEV;
3771         if (src_file.file->f_path.mnt != file->f_path.mnt)
3772                 goto out_fput;
3773
3774         src = file_inode(src_file.file);
3775
3776         ret = -EINVAL;
3777         if (src == inode)
3778                 same_inode = 1;
3779
3780         /* the src must be open for reading */
3781         if (!(src_file.file->f_mode & FMODE_READ))
3782                 goto out_fput;
3783
3784         /* don't make the dst file partly checksummed */
3785         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3786             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3787                 goto out_fput;
3788
3789         ret = -EISDIR;
3790         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3791                 goto out_fput;
3792
3793         ret = -EXDEV;
3794         if (src->i_sb != inode->i_sb)
3795                 goto out_fput;
3796
3797         if (!same_inode) {
3798                 btrfs_double_inode_lock(src, inode);
3799         } else {
3800                 mutex_lock(&src->i_mutex);
3801         }
3802
3803         /* determine range to clone */
3804         ret = -EINVAL;
3805         if (off + len > src->i_size || off + len < off)
3806                 goto out_unlock;
3807         if (len == 0)
3808                 olen = len = src->i_size - off;
3809         /* if we extend to eof, continue to block boundary */
3810         if (off + len == src->i_size)
3811                 len = ALIGN(src->i_size, bs) - off;
3812
3813         if (len == 0) {
3814                 ret = 0;
3815                 goto out_unlock;
3816         }
3817
3818         /* verify the end result is block aligned */
3819         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3820             !IS_ALIGNED(destoff, bs))
3821                 goto out_unlock;
3822
3823         /* verify if ranges are overlapped within the same file */
3824         if (same_inode) {
3825                 if (destoff + len > off && destoff < off + len)
3826                         goto out_unlock;
3827         }
3828
3829         if (destoff > inode->i_size) {
3830                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3831                 if (ret)
3832                         goto out_unlock;
3833         }
3834
3835         /*
3836          * Lock the target range too. Right after we replace the file extent
3837          * items in the fs tree (which now point to the cloned data), we might
3838          * have a worker replace them with extent items relative to a write
3839          * operation that was issued before this clone operation (i.e. confront
3840          * with inode.c:btrfs_finish_ordered_io).
3841          */
3842         if (same_inode) {
3843                 u64 lock_start = min_t(u64, off, destoff);
3844                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3845
3846                 lock_extent_range(src, lock_start, lock_len);
3847         } else {
3848                 btrfs_double_extent_lock(src, off, inode, destoff, len);
3849         }
3850
3851         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3852
3853         if (same_inode) {
3854                 u64 lock_start = min_t(u64, off, destoff);
3855                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3856
3857                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3858         } else {
3859                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3860         }
3861         /*
3862          * Truncate page cache pages so that future reads will see the cloned
3863          * data immediately and not the previous data.
3864          */
3865         truncate_inode_pages_range(&inode->i_data, destoff,
3866                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3867 out_unlock:
3868         if (!same_inode)