Merge tag 'upstream-4.8-rc1' of git://git.infradead.org/linux-ubifs
[sfrench/cifs-2.6.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include <linux/uuid.h>
45 #include <linux/btrfs.h>
46 #include <linux/uaccess.h>
47 #include "ctree.h"
48 #include "disk-io.h"
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "volumes.h"
53 #include "locking.h"
54 #include "inode-map.h"
55 #include "backref.h"
56 #include "rcu-string.h"
57 #include "send.h"
58 #include "dev-replace.h"
59 #include "props.h"
60 #include "sysfs.h"
61 #include "qgroup.h"
62 #include "tree-log.h"
63 #include "compression.h"
64
65 #ifdef CONFIG_64BIT
66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
67  * structures are incorrect, as the timespec structure from userspace
68  * is 4 bytes too small. We define these alternatives here to teach
69  * the kernel about the 32-bit struct packing.
70  */
71 struct btrfs_ioctl_timespec_32 {
72         __u64 sec;
73         __u32 nsec;
74 } __attribute__ ((__packed__));
75
76 struct btrfs_ioctl_received_subvol_args_32 {
77         char    uuid[BTRFS_UUID_SIZE];  /* in */
78         __u64   stransid;               /* in */
79         __u64   rtransid;               /* out */
80         struct btrfs_ioctl_timespec_32 stime; /* in */
81         struct btrfs_ioctl_timespec_32 rtime; /* out */
82         __u64   flags;                  /* in */
83         __u64   reserved[16];           /* in */
84 } __attribute__ ((__packed__));
85
86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
87                                 struct btrfs_ioctl_received_subvol_args_32)
88 #endif
89
90
91 static int btrfs_clone(struct inode *src, struct inode *inode,
92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
93                        int no_time_update);
94
95 /* Mask out flags that are inappropriate for the given type of inode. */
96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
97 {
98         if (S_ISDIR(mode))
99                 return flags;
100         else if (S_ISREG(mode))
101                 return flags & ~FS_DIRSYNC_FL;
102         else
103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
104 }
105
106 /*
107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
108  */
109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
110 {
111         unsigned int iflags = 0;
112
113         if (flags & BTRFS_INODE_SYNC)
114                 iflags |= FS_SYNC_FL;
115         if (flags & BTRFS_INODE_IMMUTABLE)
116                 iflags |= FS_IMMUTABLE_FL;
117         if (flags & BTRFS_INODE_APPEND)
118                 iflags |= FS_APPEND_FL;
119         if (flags & BTRFS_INODE_NODUMP)
120                 iflags |= FS_NODUMP_FL;
121         if (flags & BTRFS_INODE_NOATIME)
122                 iflags |= FS_NOATIME_FL;
123         if (flags & BTRFS_INODE_DIRSYNC)
124                 iflags |= FS_DIRSYNC_FL;
125         if (flags & BTRFS_INODE_NODATACOW)
126                 iflags |= FS_NOCOW_FL;
127
128         if (flags & BTRFS_INODE_NOCOMPRESS)
129                 iflags |= FS_NOCOMP_FL;
130         else if (flags & BTRFS_INODE_COMPRESS)
131                 iflags |= FS_COMPR_FL;
132
133         return iflags;
134 }
135
136 /*
137  * Update inode->i_flags based on the btrfs internal flags.
138  */
139 void btrfs_update_iflags(struct inode *inode)
140 {
141         struct btrfs_inode *ip = BTRFS_I(inode);
142         unsigned int new_fl = 0;
143
144         if (ip->flags & BTRFS_INODE_SYNC)
145                 new_fl |= S_SYNC;
146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
147                 new_fl |= S_IMMUTABLE;
148         if (ip->flags & BTRFS_INODE_APPEND)
149                 new_fl |= S_APPEND;
150         if (ip->flags & BTRFS_INODE_NOATIME)
151                 new_fl |= S_NOATIME;
152         if (ip->flags & BTRFS_INODE_DIRSYNC)
153                 new_fl |= S_DIRSYNC;
154
155         set_mask_bits(&inode->i_flags,
156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
157                       new_fl);
158 }
159
160 /*
161  * Inherit flags from the parent inode.
162  *
163  * Currently only the compression flags and the cow flags are inherited.
164  */
165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
166 {
167         unsigned int flags;
168
169         if (!dir)
170                 return;
171
172         flags = BTRFS_I(dir)->flags;
173
174         if (flags & BTRFS_INODE_NOCOMPRESS) {
175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
177         } else if (flags & BTRFS_INODE_COMPRESS) {
178                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
179                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
180         }
181
182         if (flags & BTRFS_INODE_NODATACOW) {
183                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
184                 if (S_ISREG(inode->i_mode))
185                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
186         }
187
188         btrfs_update_iflags(inode);
189 }
190
191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
192 {
193         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
194         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
195
196         if (copy_to_user(arg, &flags, sizeof(flags)))
197                 return -EFAULT;
198         return 0;
199 }
200
201 static int check_flags(unsigned int flags)
202 {
203         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
204                       FS_NOATIME_FL | FS_NODUMP_FL | \
205                       FS_SYNC_FL | FS_DIRSYNC_FL | \
206                       FS_NOCOMP_FL | FS_COMPR_FL |
207                       FS_NOCOW_FL))
208                 return -EOPNOTSUPP;
209
210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
211                 return -EINVAL;
212
213         return 0;
214 }
215
216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
217 {
218         struct inode *inode = file_inode(file);
219         struct btrfs_inode *ip = BTRFS_I(inode);
220         struct btrfs_root *root = ip->root;
221         struct btrfs_trans_handle *trans;
222         unsigned int flags, oldflags;
223         int ret;
224         u64 ip_oldflags;
225         unsigned int i_oldflags;
226         umode_t mode;
227
228         if (!inode_owner_or_capable(inode))
229                 return -EPERM;
230
231         if (btrfs_root_readonly(root))
232                 return -EROFS;
233
234         if (copy_from_user(&flags, arg, sizeof(flags)))
235                 return -EFAULT;
236
237         ret = check_flags(flags);
238         if (ret)
239                 return ret;
240
241         ret = mnt_want_write_file(file);
242         if (ret)
243                 return ret;
244
245         inode_lock(inode);
246
247         ip_oldflags = ip->flags;
248         i_oldflags = inode->i_flags;
249         mode = inode->i_mode;
250
251         flags = btrfs_mask_flags(inode->i_mode, flags);
252         oldflags = btrfs_flags_to_ioctl(ip->flags);
253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
255                         ret = -EPERM;
256                         goto out_unlock;
257                 }
258         }
259
260         if (flags & FS_SYNC_FL)
261                 ip->flags |= BTRFS_INODE_SYNC;
262         else
263                 ip->flags &= ~BTRFS_INODE_SYNC;
264         if (flags & FS_IMMUTABLE_FL)
265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
266         else
267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
268         if (flags & FS_APPEND_FL)
269                 ip->flags |= BTRFS_INODE_APPEND;
270         else
271                 ip->flags &= ~BTRFS_INODE_APPEND;
272         if (flags & FS_NODUMP_FL)
273                 ip->flags |= BTRFS_INODE_NODUMP;
274         else
275                 ip->flags &= ~BTRFS_INODE_NODUMP;
276         if (flags & FS_NOATIME_FL)
277                 ip->flags |= BTRFS_INODE_NOATIME;
278         else
279                 ip->flags &= ~BTRFS_INODE_NOATIME;
280         if (flags & FS_DIRSYNC_FL)
281                 ip->flags |= BTRFS_INODE_DIRSYNC;
282         else
283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
284         if (flags & FS_NOCOW_FL) {
285                 if (S_ISREG(mode)) {
286                         /*
287                          * It's safe to turn csums off here, no extents exist.
288                          * Otherwise we want the flag to reflect the real COW
289                          * status of the file and will not set it.
290                          */
291                         if (inode->i_size == 0)
292                                 ip->flags |= BTRFS_INODE_NODATACOW
293                                            | BTRFS_INODE_NODATASUM;
294                 } else {
295                         ip->flags |= BTRFS_INODE_NODATACOW;
296                 }
297         } else {
298                 /*
299                  * Revert back under same assumptions as above
300                  */
301                 if (S_ISREG(mode)) {
302                         if (inode->i_size == 0)
303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
304                                              | BTRFS_INODE_NODATASUM);
305                 } else {
306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
307                 }
308         }
309
310         /*
311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
312          * flag may be changed automatically if compression code won't make
313          * things smaller.
314          */
315         if (flags & FS_NOCOMP_FL) {
316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
318
319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
320                 if (ret && ret != -ENODATA)
321                         goto out_drop;
322         } else if (flags & FS_COMPR_FL) {
323                 const char *comp;
324
325                 ip->flags |= BTRFS_INODE_COMPRESS;
326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
327
328                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
329                         comp = "lzo";
330                 else
331                         comp = "zlib";
332                 ret = btrfs_set_prop(inode, "btrfs.compression",
333                                      comp, strlen(comp), 0);
334                 if (ret)
335                         goto out_drop;
336
337         } else {
338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
339                 if (ret && ret != -ENODATA)
340                         goto out_drop;
341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
342         }
343
344         trans = btrfs_start_transaction(root, 1);
345         if (IS_ERR(trans)) {
346                 ret = PTR_ERR(trans);
347                 goto out_drop;
348         }
349
350         btrfs_update_iflags(inode);
351         inode_inc_iversion(inode);
352         inode->i_ctime = current_fs_time(inode->i_sb);
353         ret = btrfs_update_inode(trans, root, inode);
354
355         btrfs_end_transaction(trans, root);
356  out_drop:
357         if (ret) {
358                 ip->flags = ip_oldflags;
359                 inode->i_flags = i_oldflags;
360         }
361
362  out_unlock:
363         inode_unlock(inode);
364         mnt_drop_write_file(file);
365         return ret;
366 }
367
368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
369 {
370         struct inode *inode = file_inode(file);
371
372         return put_user(inode->i_generation, arg);
373 }
374
375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
376 {
377         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
378         struct btrfs_device *device;
379         struct request_queue *q;
380         struct fstrim_range range;
381         u64 minlen = ULLONG_MAX;
382         u64 num_devices = 0;
383         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
384         int ret;
385
386         if (!capable(CAP_SYS_ADMIN))
387                 return -EPERM;
388
389         rcu_read_lock();
390         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
391                                 dev_list) {
392                 if (!device->bdev)
393                         continue;
394                 q = bdev_get_queue(device->bdev);
395                 if (blk_queue_discard(q)) {
396                         num_devices++;
397                         minlen = min((u64)q->limits.discard_granularity,
398                                      minlen);
399                 }
400         }
401         rcu_read_unlock();
402
403         if (!num_devices)
404                 return -EOPNOTSUPP;
405         if (copy_from_user(&range, arg, sizeof(range)))
406                 return -EFAULT;
407         if (range.start > total_bytes ||
408             range.len < fs_info->sb->s_blocksize)
409                 return -EINVAL;
410
411         range.len = min(range.len, total_bytes - range.start);
412         range.minlen = max(range.minlen, minlen);
413         ret = btrfs_trim_fs(fs_info->tree_root, &range);
414         if (ret < 0)
415                 return ret;
416
417         if (copy_to_user(arg, &range, sizeof(range)))
418                 return -EFAULT;
419
420         return 0;
421 }
422
423 int btrfs_is_empty_uuid(u8 *uuid)
424 {
425         int i;
426
427         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
428                 if (uuid[i])
429                         return 0;
430         }
431         return 1;
432 }
433
434 static noinline int create_subvol(struct inode *dir,
435                                   struct dentry *dentry,
436                                   char *name, int namelen,
437                                   u64 *async_transid,
438                                   struct btrfs_qgroup_inherit *inherit)
439 {
440         struct btrfs_trans_handle *trans;
441         struct btrfs_key key;
442         struct btrfs_root_item *root_item;
443         struct btrfs_inode_item *inode_item;
444         struct extent_buffer *leaf;
445         struct btrfs_root *root = BTRFS_I(dir)->root;
446         struct btrfs_root *new_root;
447         struct btrfs_block_rsv block_rsv;
448         struct timespec cur_time = current_fs_time(dir->i_sb);
449         struct inode *inode;
450         int ret;
451         int err;
452         u64 objectid;
453         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
454         u64 index = 0;
455         u64 qgroup_reserved;
456         uuid_le new_uuid;
457
458         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
459         if (!root_item)
460                 return -ENOMEM;
461
462         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
463         if (ret)
464                 goto fail_free;
465
466         /*
467          * Don't create subvolume whose level is not zero. Or qgroup will be
468          * screwed up since it assumes subvolume qgroup's level to be 0.
469          */
470         if (btrfs_qgroup_level(objectid)) {
471                 ret = -ENOSPC;
472                 goto fail_free;
473         }
474
475         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
476         /*
477          * The same as the snapshot creation, please see the comment
478          * of create_snapshot().
479          */
480         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
481                                                8, &qgroup_reserved, false);
482         if (ret)
483                 goto fail_free;
484
485         trans = btrfs_start_transaction(root, 0);
486         if (IS_ERR(trans)) {
487                 ret = PTR_ERR(trans);
488                 btrfs_subvolume_release_metadata(root, &block_rsv,
489                                                  qgroup_reserved);
490                 goto fail_free;
491         }
492         trans->block_rsv = &block_rsv;
493         trans->bytes_reserved = block_rsv.size;
494
495         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
496         if (ret)
497                 goto fail;
498
499         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
500         if (IS_ERR(leaf)) {
501                 ret = PTR_ERR(leaf);
502                 goto fail;
503         }
504
505         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
506         btrfs_set_header_bytenr(leaf, leaf->start);
507         btrfs_set_header_generation(leaf, trans->transid);
508         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
509         btrfs_set_header_owner(leaf, objectid);
510
511         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
512                             BTRFS_FSID_SIZE);
513         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
514                             btrfs_header_chunk_tree_uuid(leaf),
515                             BTRFS_UUID_SIZE);
516         btrfs_mark_buffer_dirty(leaf);
517
518         inode_item = &root_item->inode;
519         btrfs_set_stack_inode_generation(inode_item, 1);
520         btrfs_set_stack_inode_size(inode_item, 3);
521         btrfs_set_stack_inode_nlink(inode_item, 1);
522         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
523         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
524
525         btrfs_set_root_flags(root_item, 0);
526         btrfs_set_root_limit(root_item, 0);
527         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
528
529         btrfs_set_root_bytenr(root_item, leaf->start);
530         btrfs_set_root_generation(root_item, trans->transid);
531         btrfs_set_root_level(root_item, 0);
532         btrfs_set_root_refs(root_item, 1);
533         btrfs_set_root_used(root_item, leaf->len);
534         btrfs_set_root_last_snapshot(root_item, 0);
535
536         btrfs_set_root_generation_v2(root_item,
537                         btrfs_root_generation(root_item));
538         uuid_le_gen(&new_uuid);
539         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
540         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
541         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
542         root_item->ctime = root_item->otime;
543         btrfs_set_root_ctransid(root_item, trans->transid);
544         btrfs_set_root_otransid(root_item, trans->transid);
545
546         btrfs_tree_unlock(leaf);
547         free_extent_buffer(leaf);
548         leaf = NULL;
549
550         btrfs_set_root_dirid(root_item, new_dirid);
551
552         key.objectid = objectid;
553         key.offset = 0;
554         key.type = BTRFS_ROOT_ITEM_KEY;
555         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
556                                 root_item);
557         if (ret)
558                 goto fail;
559
560         key.offset = (u64)-1;
561         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
562         if (IS_ERR(new_root)) {
563                 ret = PTR_ERR(new_root);
564                 btrfs_abort_transaction(trans, root, ret);
565                 goto fail;
566         }
567
568         btrfs_record_root_in_trans(trans, new_root);
569
570         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
571         if (ret) {
572                 /* We potentially lose an unused inode item here */
573                 btrfs_abort_transaction(trans, root, ret);
574                 goto fail;
575         }
576
577         mutex_lock(&new_root->objectid_mutex);
578         new_root->highest_objectid = new_dirid;
579         mutex_unlock(&new_root->objectid_mutex);
580
581         /*
582          * insert the directory item
583          */
584         ret = btrfs_set_inode_index(dir, &index);
585         if (ret) {
586                 btrfs_abort_transaction(trans, root, ret);
587                 goto fail;
588         }
589
590         ret = btrfs_insert_dir_item(trans, root,
591                                     name, namelen, dir, &key,
592                                     BTRFS_FT_DIR, index);
593         if (ret) {
594                 btrfs_abort_transaction(trans, root, ret);
595                 goto fail;
596         }
597
598         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
599         ret = btrfs_update_inode(trans, root, dir);
600         BUG_ON(ret);
601
602         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
603                                  objectid, root->root_key.objectid,
604                                  btrfs_ino(dir), index, name, namelen);
605         BUG_ON(ret);
606
607         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
608                                   root_item->uuid, BTRFS_UUID_KEY_SUBVOL,
609                                   objectid);
610         if (ret)
611                 btrfs_abort_transaction(trans, root, ret);
612
613 fail:
614         kfree(root_item);
615         trans->block_rsv = NULL;
616         trans->bytes_reserved = 0;
617         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
618
619         if (async_transid) {
620                 *async_transid = trans->transid;
621                 err = btrfs_commit_transaction_async(trans, root, 1);
622                 if (err)
623                         err = btrfs_commit_transaction(trans, root);
624         } else {
625                 err = btrfs_commit_transaction(trans, root);
626         }
627         if (err && !ret)
628                 ret = err;
629
630         if (!ret) {
631                 inode = btrfs_lookup_dentry(dir, dentry);
632                 if (IS_ERR(inode))
633                         return PTR_ERR(inode);
634                 d_instantiate(dentry, inode);
635         }
636         return ret;
637
638 fail_free:
639         kfree(root_item);
640         return ret;
641 }
642
643 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
644 {
645         s64 writers;
646         DEFINE_WAIT(wait);
647
648         do {
649                 prepare_to_wait(&root->subv_writers->wait, &wait,
650                                 TASK_UNINTERRUPTIBLE);
651
652                 writers = percpu_counter_sum(&root->subv_writers->counter);
653                 if (writers)
654                         schedule();
655
656                 finish_wait(&root->subv_writers->wait, &wait);
657         } while (writers);
658 }
659
660 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
661                            struct dentry *dentry, char *name, int namelen,
662                            u64 *async_transid, bool readonly,
663                            struct btrfs_qgroup_inherit *inherit)
664 {
665         struct inode *inode;
666         struct btrfs_pending_snapshot *pending_snapshot;
667         struct btrfs_trans_handle *trans;
668         int ret;
669
670         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
671                 return -EINVAL;
672
673         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
674         if (!pending_snapshot)
675                 return -ENOMEM;
676
677         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
678                         GFP_NOFS);
679         pending_snapshot->path = btrfs_alloc_path();
680         if (!pending_snapshot->root_item || !pending_snapshot->path) {
681                 ret = -ENOMEM;
682                 goto free_pending;
683         }
684
685         atomic_inc(&root->will_be_snapshoted);
686         smp_mb__after_atomic();
687         btrfs_wait_for_no_snapshoting_writes(root);
688
689         ret = btrfs_start_delalloc_inodes(root, 0);
690         if (ret)
691                 goto dec_and_free;
692
693         btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
694
695         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
696                              BTRFS_BLOCK_RSV_TEMP);
697         /*
698          * 1 - parent dir inode
699          * 2 - dir entries
700          * 1 - root item
701          * 2 - root ref/backref
702          * 1 - root of snapshot
703          * 1 - UUID item
704          */
705         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
706                                         &pending_snapshot->block_rsv, 8,
707                                         &pending_snapshot->qgroup_reserved,
708                                         false);
709         if (ret)
710                 goto dec_and_free;
711
712         pending_snapshot->dentry = dentry;
713         pending_snapshot->root = root;
714         pending_snapshot->readonly = readonly;
715         pending_snapshot->dir = dir;
716         pending_snapshot->inherit = inherit;
717
718         trans = btrfs_start_transaction(root, 0);
719         if (IS_ERR(trans)) {
720                 ret = PTR_ERR(trans);
721                 goto fail;
722         }
723
724         spin_lock(&root->fs_info->trans_lock);
725         list_add(&pending_snapshot->list,
726                  &trans->transaction->pending_snapshots);
727         spin_unlock(&root->fs_info->trans_lock);
728         if (async_transid) {
729                 *async_transid = trans->transid;
730                 ret = btrfs_commit_transaction_async(trans,
731                                      root->fs_info->extent_root, 1);
732                 if (ret)
733                         ret = btrfs_commit_transaction(trans, root);
734         } else {
735                 ret = btrfs_commit_transaction(trans,
736                                                root->fs_info->extent_root);
737         }
738         if (ret)
739                 goto fail;
740
741         ret = pending_snapshot->error;
742         if (ret)
743                 goto fail;
744
745         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
746         if (ret)
747                 goto fail;
748
749         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
750         if (IS_ERR(inode)) {
751                 ret = PTR_ERR(inode);
752                 goto fail;
753         }
754
755         d_instantiate(dentry, inode);
756         ret = 0;
757 fail:
758         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
759                                          &pending_snapshot->block_rsv,
760                                          pending_snapshot->qgroup_reserved);
761 dec_and_free:
762         if (atomic_dec_and_test(&root->will_be_snapshoted))
763                 wake_up_atomic_t(&root->will_be_snapshoted);
764 free_pending:
765         kfree(pending_snapshot->root_item);
766         btrfs_free_path(pending_snapshot->path);
767         kfree(pending_snapshot);
768
769         return ret;
770 }
771
772 /*  copy of may_delete in fs/namei.c()
773  *      Check whether we can remove a link victim from directory dir, check
774  *  whether the type of victim is right.
775  *  1. We can't do it if dir is read-only (done in permission())
776  *  2. We should have write and exec permissions on dir
777  *  3. We can't remove anything from append-only dir
778  *  4. We can't do anything with immutable dir (done in permission())
779  *  5. If the sticky bit on dir is set we should either
780  *      a. be owner of dir, or
781  *      b. be owner of victim, or
782  *      c. have CAP_FOWNER capability
783  *  6. If the victim is append-only or immutable we can't do anything with
784  *     links pointing to it.
785  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
786  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
787  *  9. We can't remove a root or mountpoint.
788  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
789  *     nfs_async_unlink().
790  */
791
792 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
793 {
794         int error;
795
796         if (d_really_is_negative(victim))
797                 return -ENOENT;
798
799         BUG_ON(d_inode(victim->d_parent) != dir);
800         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
801
802         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
803         if (error)
804                 return error;
805         if (IS_APPEND(dir))
806                 return -EPERM;
807         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
808             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
809                 return -EPERM;
810         if (isdir) {
811                 if (!d_is_dir(victim))
812                         return -ENOTDIR;
813                 if (IS_ROOT(victim))
814                         return -EBUSY;
815         } else if (d_is_dir(victim))
816                 return -EISDIR;
817         if (IS_DEADDIR(dir))
818                 return -ENOENT;
819         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
820                 return -EBUSY;
821         return 0;
822 }
823
824 /* copy of may_create in fs/namei.c() */
825 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
826 {
827         if (d_really_is_positive(child))
828                 return -EEXIST;
829         if (IS_DEADDIR(dir))
830                 return -ENOENT;
831         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
832 }
833
834 /*
835  * Create a new subvolume below @parent.  This is largely modeled after
836  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
837  * inside this filesystem so it's quite a bit simpler.
838  */
839 static noinline int btrfs_mksubvol(struct path *parent,
840                                    char *name, int namelen,
841                                    struct btrfs_root *snap_src,
842                                    u64 *async_transid, bool readonly,
843                                    struct btrfs_qgroup_inherit *inherit)
844 {
845         struct inode *dir  = d_inode(parent->dentry);
846         struct dentry *dentry;
847         int error;
848
849         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
850         if (error == -EINTR)
851                 return error;
852
853         dentry = lookup_one_len(name, parent->dentry, namelen);
854         error = PTR_ERR(dentry);
855         if (IS_ERR(dentry))
856                 goto out_unlock;
857
858         error = btrfs_may_create(dir, dentry);
859         if (error)
860                 goto out_dput;
861
862         /*
863          * even if this name doesn't exist, we may get hash collisions.
864          * check for them now when we can safely fail
865          */
866         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
867                                                dir->i_ino, name,
868                                                namelen);
869         if (error)
870                 goto out_dput;
871
872         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
873
874         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
875                 goto out_up_read;
876
877         if (snap_src) {
878                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
879                                         async_transid, readonly, inherit);
880         } else {
881                 error = create_subvol(dir, dentry, name, namelen,
882                                       async_transid, inherit);
883         }
884         if (!error)
885                 fsnotify_mkdir(dir, dentry);
886 out_up_read:
887         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
888 out_dput:
889         dput(dentry);
890 out_unlock:
891         inode_unlock(dir);
892         return error;
893 }
894
895 /*
896  * When we're defragging a range, we don't want to kick it off again
897  * if it is really just waiting for delalloc to send it down.
898  * If we find a nice big extent or delalloc range for the bytes in the
899  * file you want to defrag, we return 0 to let you know to skip this
900  * part of the file
901  */
902 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
903 {
904         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
905         struct extent_map *em = NULL;
906         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
907         u64 end;
908
909         read_lock(&em_tree->lock);
910         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
911         read_unlock(&em_tree->lock);
912
913         if (em) {
914                 end = extent_map_end(em);
915                 free_extent_map(em);
916                 if (end - offset > thresh)
917                         return 0;
918         }
919         /* if we already have a nice delalloc here, just stop */
920         thresh /= 2;
921         end = count_range_bits(io_tree, &offset, offset + thresh,
922                                thresh, EXTENT_DELALLOC, 1);
923         if (end >= thresh)
924                 return 0;
925         return 1;
926 }
927
928 /*
929  * helper function to walk through a file and find extents
930  * newer than a specific transid, and smaller than thresh.
931  *
932  * This is used by the defragging code to find new and small
933  * extents
934  */
935 static int find_new_extents(struct btrfs_root *root,
936                             struct inode *inode, u64 newer_than,
937                             u64 *off, u32 thresh)
938 {
939         struct btrfs_path *path;
940         struct btrfs_key min_key;
941         struct extent_buffer *leaf;
942         struct btrfs_file_extent_item *extent;
943         int type;
944         int ret;
945         u64 ino = btrfs_ino(inode);
946
947         path = btrfs_alloc_path();
948         if (!path)
949                 return -ENOMEM;
950
951         min_key.objectid = ino;
952         min_key.type = BTRFS_EXTENT_DATA_KEY;
953         min_key.offset = *off;
954
955         while (1) {
956                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
957                 if (ret != 0)
958                         goto none;
959 process_slot:
960                 if (min_key.objectid != ino)
961                         goto none;
962                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
963                         goto none;
964
965                 leaf = path->nodes[0];
966                 extent = btrfs_item_ptr(leaf, path->slots[0],
967                                         struct btrfs_file_extent_item);
968
969                 type = btrfs_file_extent_type(leaf, extent);
970                 if (type == BTRFS_FILE_EXTENT_REG &&
971                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
972                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
973                         *off = min_key.offset;
974                         btrfs_free_path(path);
975                         return 0;
976                 }
977
978                 path->slots[0]++;
979                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
980                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
981                         goto process_slot;
982                 }
983
984                 if (min_key.offset == (u64)-1)
985                         goto none;
986
987                 min_key.offset++;
988                 btrfs_release_path(path);
989         }
990 none:
991         btrfs_free_path(path);
992         return -ENOENT;
993 }
994
995 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
996 {
997         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
998         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
999         struct extent_map *em;
1000         u64 len = PAGE_SIZE;
1001
1002         /*
1003          * hopefully we have this extent in the tree already, try without
1004          * the full extent lock
1005          */
1006         read_lock(&em_tree->lock);
1007         em = lookup_extent_mapping(em_tree, start, len);
1008         read_unlock(&em_tree->lock);
1009
1010         if (!em) {
1011                 struct extent_state *cached = NULL;
1012                 u64 end = start + len - 1;
1013
1014                 /* get the big lock and read metadata off disk */
1015                 lock_extent_bits(io_tree, start, end, &cached);
1016                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1017                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1018
1019                 if (IS_ERR(em))
1020                         return NULL;
1021         }
1022
1023         return em;
1024 }
1025
1026 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1027 {
1028         struct extent_map *next;
1029         bool ret = true;
1030
1031         /* this is the last extent */
1032         if (em->start + em->len >= i_size_read(inode))
1033                 return false;
1034
1035         next = defrag_lookup_extent(inode, em->start + em->len);
1036         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1037                 ret = false;
1038         else if ((em->block_start + em->block_len == next->block_start) &&
1039                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1040                 ret = false;
1041
1042         free_extent_map(next);
1043         return ret;
1044 }
1045
1046 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1047                                u64 *last_len, u64 *skip, u64 *defrag_end,
1048                                int compress)
1049 {
1050         struct extent_map *em;
1051         int ret = 1;
1052         bool next_mergeable = true;
1053         bool prev_mergeable = true;
1054
1055         /*
1056          * make sure that once we start defragging an extent, we keep on
1057          * defragging it
1058          */
1059         if (start < *defrag_end)
1060                 return 1;
1061
1062         *skip = 0;
1063
1064         em = defrag_lookup_extent(inode, start);
1065         if (!em)
1066                 return 0;
1067
1068         /* this will cover holes, and inline extents */
1069         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1070                 ret = 0;
1071                 goto out;
1072         }
1073
1074         if (!*defrag_end)
1075                 prev_mergeable = false;
1076
1077         next_mergeable = defrag_check_next_extent(inode, em);
1078         /*
1079          * we hit a real extent, if it is big or the next extent is not a
1080          * real extent, don't bother defragging it
1081          */
1082         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1083             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1084                 ret = 0;
1085 out:
1086         /*
1087          * last_len ends up being a counter of how many bytes we've defragged.
1088          * every time we choose not to defrag an extent, we reset *last_len
1089          * so that the next tiny extent will force a defrag.
1090          *
1091          * The end result of this is that tiny extents before a single big
1092          * extent will force at least part of that big extent to be defragged.
1093          */
1094         if (ret) {
1095                 *defrag_end = extent_map_end(em);
1096         } else {
1097                 *last_len = 0;
1098                 *skip = extent_map_end(em);
1099                 *defrag_end = 0;
1100         }
1101
1102         free_extent_map(em);
1103         return ret;
1104 }
1105
1106 /*
1107  * it doesn't do much good to defrag one or two pages
1108  * at a time.  This pulls in a nice chunk of pages
1109  * to COW and defrag.
1110  *
1111  * It also makes sure the delalloc code has enough
1112  * dirty data to avoid making new small extents as part
1113  * of the defrag
1114  *
1115  * It's a good idea to start RA on this range
1116  * before calling this.
1117  */
1118 static int cluster_pages_for_defrag(struct inode *inode,
1119                                     struct page **pages,
1120                                     unsigned long start_index,
1121                                     unsigned long num_pages)
1122 {
1123         unsigned long file_end;
1124         u64 isize = i_size_read(inode);
1125         u64 page_start;
1126         u64 page_end;
1127         u64 page_cnt;
1128         int ret;
1129         int i;
1130         int i_done;
1131         struct btrfs_ordered_extent *ordered;
1132         struct extent_state *cached_state = NULL;
1133         struct extent_io_tree *tree;
1134         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1135
1136         file_end = (isize - 1) >> PAGE_SHIFT;
1137         if (!isize || start_index > file_end)
1138                 return 0;
1139
1140         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1141
1142         ret = btrfs_delalloc_reserve_space(inode,
1143                         start_index << PAGE_SHIFT,
1144                         page_cnt << PAGE_SHIFT);
1145         if (ret)
1146                 return ret;
1147         i_done = 0;
1148         tree = &BTRFS_I(inode)->io_tree;
1149
1150         /* step one, lock all the pages */
1151         for (i = 0; i < page_cnt; i++) {
1152                 struct page *page;
1153 again:
1154                 page = find_or_create_page(inode->i_mapping,
1155                                            start_index + i, mask);
1156                 if (!page)
1157                         break;
1158
1159                 page_start = page_offset(page);
1160                 page_end = page_start + PAGE_SIZE - 1;
1161                 while (1) {
1162                         lock_extent_bits(tree, page_start, page_end,
1163                                          &cached_state);
1164                         ordered = btrfs_lookup_ordered_extent(inode,
1165                                                               page_start);
1166                         unlock_extent_cached(tree, page_start, page_end,
1167                                              &cached_state, GFP_NOFS);
1168                         if (!ordered)
1169                                 break;
1170
1171                         unlock_page(page);
1172                         btrfs_start_ordered_extent(inode, ordered, 1);
1173                         btrfs_put_ordered_extent(ordered);
1174                         lock_page(page);
1175                         /*
1176                          * we unlocked the page above, so we need check if
1177                          * it was released or not.
1178                          */
1179                         if (page->mapping != inode->i_mapping) {
1180                                 unlock_page(page);
1181                                 put_page(page);
1182                                 goto again;
1183                         }
1184                 }
1185
1186                 if (!PageUptodate(page)) {
1187                         btrfs_readpage(NULL, page);
1188                         lock_page(page);
1189                         if (!PageUptodate(page)) {
1190                                 unlock_page(page);
1191                                 put_page(page);
1192                                 ret = -EIO;
1193                                 break;
1194                         }
1195                 }
1196
1197                 if (page->mapping != inode->i_mapping) {
1198                         unlock_page(page);
1199                         put_page(page);
1200                         goto again;
1201                 }
1202
1203                 pages[i] = page;
1204                 i_done++;
1205         }
1206         if (!i_done || ret)
1207                 goto out;
1208
1209         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1210                 goto out;
1211
1212         /*
1213          * so now we have a nice long stream of locked
1214          * and up to date pages, lets wait on them
1215          */
1216         for (i = 0; i < i_done; i++)
1217                 wait_on_page_writeback(pages[i]);
1218
1219         page_start = page_offset(pages[0]);
1220         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1221
1222         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1223                          page_start, page_end - 1, &cached_state);
1224         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1225                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1226                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1227                           &cached_state, GFP_NOFS);
1228
1229         if (i_done != page_cnt) {
1230                 spin_lock(&BTRFS_I(inode)->lock);
1231                 BTRFS_I(inode)->outstanding_extents++;
1232                 spin_unlock(&BTRFS_I(inode)->lock);
1233                 btrfs_delalloc_release_space(inode,
1234                                 start_index << PAGE_SHIFT,
1235                                 (page_cnt - i_done) << PAGE_SHIFT);
1236         }
1237
1238
1239         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1240                           &cached_state);
1241
1242         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1243                              page_start, page_end - 1, &cached_state,
1244                              GFP_NOFS);
1245
1246         for (i = 0; i < i_done; i++) {
1247                 clear_page_dirty_for_io(pages[i]);
1248                 ClearPageChecked(pages[i]);
1249                 set_page_extent_mapped(pages[i]);
1250                 set_page_dirty(pages[i]);
1251                 unlock_page(pages[i]);
1252                 put_page(pages[i]);
1253         }
1254         return i_done;
1255 out:
1256         for (i = 0; i < i_done; i++) {
1257                 unlock_page(pages[i]);
1258                 put_page(pages[i]);
1259         }
1260         btrfs_delalloc_release_space(inode,
1261                         start_index << PAGE_SHIFT,
1262                         page_cnt << PAGE_SHIFT);
1263         return ret;
1264
1265 }
1266
1267 int btrfs_defrag_file(struct inode *inode, struct file *file,
1268                       struct btrfs_ioctl_defrag_range_args *range,
1269                       u64 newer_than, unsigned long max_to_defrag)
1270 {
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272         struct file_ra_state *ra = NULL;
1273         unsigned long last_index;
1274         u64 isize = i_size_read(inode);
1275         u64 last_len = 0;
1276         u64 skip = 0;
1277         u64 defrag_end = 0;
1278         u64 newer_off = range->start;
1279         unsigned long i;
1280         unsigned long ra_index = 0;
1281         int ret;
1282         int defrag_count = 0;
1283         int compress_type = BTRFS_COMPRESS_ZLIB;
1284         u32 extent_thresh = range->extent_thresh;
1285         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1286         unsigned long cluster = max_cluster;
1287         u64 new_align = ~((u64)SZ_128K - 1);
1288         struct page **pages = NULL;
1289
1290         if (isize == 0)
1291                 return 0;
1292
1293         if (range->start >= isize)
1294                 return -EINVAL;
1295
1296         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1297                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1298                         return -EINVAL;
1299                 if (range->compress_type)
1300                         compress_type = range->compress_type;
1301         }
1302
1303         if (extent_thresh == 0)
1304                 extent_thresh = SZ_256K;
1305
1306         /*
1307          * if we were not given a file, allocate a readahead
1308          * context
1309          */
1310         if (!file) {
1311                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1312                 if (!ra)
1313                         return -ENOMEM;
1314                 file_ra_state_init(ra, inode->i_mapping);
1315         } else {
1316                 ra = &file->f_ra;
1317         }
1318
1319         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1320                         GFP_NOFS);
1321         if (!pages) {
1322                 ret = -ENOMEM;
1323                 goto out_ra;
1324         }
1325
1326         /* find the last page to defrag */
1327         if (range->start + range->len > range->start) {
1328                 last_index = min_t(u64, isize - 1,
1329                          range->start + range->len - 1) >> PAGE_SHIFT;
1330         } else {
1331                 last_index = (isize - 1) >> PAGE_SHIFT;
1332         }
1333
1334         if (newer_than) {
1335                 ret = find_new_extents(root, inode, newer_than,
1336                                        &newer_off, SZ_64K);
1337                 if (!ret) {
1338                         range->start = newer_off;
1339                         /*
1340                          * we always align our defrag to help keep
1341                          * the extents in the file evenly spaced
1342                          */
1343                         i = (newer_off & new_align) >> PAGE_SHIFT;
1344                 } else
1345                         goto out_ra;
1346         } else {
1347                 i = range->start >> PAGE_SHIFT;
1348         }
1349         if (!max_to_defrag)
1350                 max_to_defrag = last_index - i + 1;
1351
1352         /*
1353          * make writeback starts from i, so the defrag range can be
1354          * written sequentially.
1355          */
1356         if (i < inode->i_mapping->writeback_index)
1357                 inode->i_mapping->writeback_index = i;
1358
1359         while (i <= last_index && defrag_count < max_to_defrag &&
1360                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1361                 /*
1362                  * make sure we stop running if someone unmounts
1363                  * the FS
1364                  */
1365                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1366                         break;
1367
1368                 if (btrfs_defrag_cancelled(root->fs_info)) {
1369                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1370                         ret = -EAGAIN;
1371                         break;
1372                 }
1373
1374                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1375                                          extent_thresh, &last_len, &skip,
1376                                          &defrag_end, range->flags &
1377                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1378                         unsigned long next;
1379                         /*
1380                          * the should_defrag function tells us how much to skip
1381                          * bump our counter by the suggested amount
1382                          */
1383                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1384                         i = max(i + 1, next);
1385                         continue;
1386                 }
1387
1388                 if (!newer_than) {
1389                         cluster = (PAGE_ALIGN(defrag_end) >>
1390                                    PAGE_SHIFT) - i;
1391                         cluster = min(cluster, max_cluster);
1392                 } else {
1393                         cluster = max_cluster;
1394                 }
1395
1396                 if (i + cluster > ra_index) {
1397                         ra_index = max(i, ra_index);
1398                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1399                                        cluster);
1400                         ra_index += cluster;
1401                 }
1402
1403                 inode_lock(inode);
1404                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1405                         BTRFS_I(inode)->force_compress = compress_type;
1406                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1407                 if (ret < 0) {
1408                         inode_unlock(inode);
1409                         goto out_ra;
1410                 }
1411
1412                 defrag_count += ret;
1413                 balance_dirty_pages_ratelimited(inode->i_mapping);
1414                 inode_unlock(inode);
1415
1416                 if (newer_than) {
1417                         if (newer_off == (u64)-1)
1418                                 break;
1419
1420                         if (ret > 0)
1421                                 i += ret;
1422
1423                         newer_off = max(newer_off + 1,
1424                                         (u64)i << PAGE_SHIFT);
1425
1426                         ret = find_new_extents(root, inode, newer_than,
1427                                                &newer_off, SZ_64K);
1428                         if (!ret) {
1429                                 range->start = newer_off;
1430                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1431                         } else {
1432                                 break;
1433                         }
1434                 } else {
1435                         if (ret > 0) {
1436                                 i += ret;
1437                                 last_len += ret << PAGE_SHIFT;
1438                         } else {
1439                                 i++;
1440                                 last_len = 0;
1441                         }
1442                 }
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1446                 filemap_flush(inode->i_mapping);
1447                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1448                              &BTRFS_I(inode)->runtime_flags))
1449                         filemap_flush(inode->i_mapping);
1450         }
1451
1452         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1453                 /* the filemap_flush will queue IO into the worker threads, but
1454                  * we have to make sure the IO is actually started and that
1455                  * ordered extents get created before we return
1456                  */
1457                 atomic_inc(&root->fs_info->async_submit_draining);
1458                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1459                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1460                         wait_event(root->fs_info->async_submit_wait,
1461                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1462                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1463                 }
1464                 atomic_dec(&root->fs_info->async_submit_draining);
1465         }
1466
1467         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1468                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1469         }
1470
1471         ret = defrag_count;
1472
1473 out_ra:
1474         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1475                 inode_lock(inode);
1476                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1477                 inode_unlock(inode);
1478         }
1479         if (!file)
1480                 kfree(ra);
1481         kfree(pages);
1482         return ret;
1483 }
1484
1485 static noinline int btrfs_ioctl_resize(struct file *file,
1486                                         void __user *arg)
1487 {
1488         u64 new_size;
1489         u64 old_size;
1490         u64 devid = 1;
1491         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1492         struct btrfs_ioctl_vol_args *vol_args;
1493         struct btrfs_trans_handle *trans;
1494         struct btrfs_device *device = NULL;
1495         char *sizestr;
1496         char *retptr;
1497         char *devstr = NULL;
1498         int ret = 0;
1499         int mod = 0;
1500
1501         if (!capable(CAP_SYS_ADMIN))
1502                 return -EPERM;
1503
1504         ret = mnt_want_write_file(file);
1505         if (ret)
1506                 return ret;
1507
1508         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1509                         1)) {
1510                 mnt_drop_write_file(file);
1511                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1512         }
1513
1514         mutex_lock(&root->fs_info->volume_mutex);
1515         vol_args = memdup_user(arg, sizeof(*vol_args));
1516         if (IS_ERR(vol_args)) {
1517                 ret = PTR_ERR(vol_args);
1518                 goto out;
1519         }
1520
1521         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1522
1523         sizestr = vol_args->name;
1524         devstr = strchr(sizestr, ':');
1525         if (devstr) {
1526                 sizestr = devstr + 1;
1527                 *devstr = '\0';
1528                 devstr = vol_args->name;
1529                 ret = kstrtoull(devstr, 10, &devid);
1530                 if (ret)
1531                         goto out_free;
1532                 if (!devid) {
1533                         ret = -EINVAL;
1534                         goto out_free;
1535                 }
1536                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1537         }
1538
1539         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1540         if (!device) {
1541                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1542                        devid);
1543                 ret = -ENODEV;
1544                 goto out_free;
1545         }
1546
1547         if (!device->writeable) {
1548                 btrfs_info(root->fs_info,
1549                            "resizer unable to apply on readonly device %llu",
1550                        devid);
1551                 ret = -EPERM;
1552                 goto out_free;
1553         }
1554
1555         if (!strcmp(sizestr, "max"))
1556                 new_size = device->bdev->bd_inode->i_size;
1557         else {
1558                 if (sizestr[0] == '-') {
1559                         mod = -1;
1560                         sizestr++;
1561                 } else if (sizestr[0] == '+') {
1562                         mod = 1;
1563                         sizestr++;
1564                 }
1565                 new_size = memparse(sizestr, &retptr);
1566                 if (*retptr != '\0' || new_size == 0) {
1567                         ret = -EINVAL;
1568                         goto out_free;
1569                 }
1570         }
1571
1572         if (device->is_tgtdev_for_dev_replace) {
1573                 ret = -EPERM;
1574                 goto out_free;
1575         }
1576
1577         old_size = btrfs_device_get_total_bytes(device);
1578
1579         if (mod < 0) {
1580                 if (new_size > old_size) {
1581                         ret = -EINVAL;
1582                         goto out_free;
1583                 }
1584                 new_size = old_size - new_size;
1585         } else if (mod > 0) {
1586                 if (new_size > ULLONG_MAX - old_size) {
1587                         ret = -ERANGE;
1588                         goto out_free;
1589                 }
1590                 new_size = old_size + new_size;
1591         }
1592
1593         if (new_size < SZ_256M) {
1594                 ret = -EINVAL;
1595                 goto out_free;
1596         }
1597         if (new_size > device->bdev->bd_inode->i_size) {
1598                 ret = -EFBIG;
1599                 goto out_free;
1600         }
1601
1602         new_size = div_u64(new_size, root->sectorsize);
1603         new_size *= root->sectorsize;
1604
1605         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1606                       rcu_str_deref(device->name), new_size);
1607
1608         if (new_size > old_size) {
1609                 trans = btrfs_start_transaction(root, 0);
1610                 if (IS_ERR(trans)) {
1611                         ret = PTR_ERR(trans);
1612                         goto out_free;
1613                 }
1614                 ret = btrfs_grow_device(trans, device, new_size);
1615                 btrfs_commit_transaction(trans, root);
1616         } else if (new_size < old_size) {
1617                 ret = btrfs_shrink_device(device, new_size);
1618         } /* equal, nothing need to do */
1619
1620 out_free:
1621         kfree(vol_args);
1622 out:
1623         mutex_unlock(&root->fs_info->volume_mutex);
1624         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1625         mnt_drop_write_file(file);
1626         return ret;
1627 }
1628
1629 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1630                                 char *name, unsigned long fd, int subvol,
1631                                 u64 *transid, bool readonly,
1632                                 struct btrfs_qgroup_inherit *inherit)
1633 {
1634         int namelen;
1635         int ret = 0;
1636
1637         ret = mnt_want_write_file(file);
1638         if (ret)
1639                 goto out;
1640
1641         namelen = strlen(name);
1642         if (strchr(name, '/')) {
1643                 ret = -EINVAL;
1644                 goto out_drop_write;
1645         }
1646
1647         if (name[0] == '.' &&
1648            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1649                 ret = -EEXIST;
1650                 goto out_drop_write;
1651         }
1652
1653         if (subvol) {
1654                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1655                                      NULL, transid, readonly, inherit);
1656         } else {
1657                 struct fd src = fdget(fd);
1658                 struct inode *src_inode;
1659                 if (!src.file) {
1660                         ret = -EINVAL;
1661                         goto out_drop_write;
1662                 }
1663
1664                 src_inode = file_inode(src.file);
1665                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1666                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1667                                    "Snapshot src from another FS");
1668                         ret = -EXDEV;
1669                 } else if (!inode_owner_or_capable(src_inode)) {
1670                         /*
1671                          * Subvolume creation is not restricted, but snapshots
1672                          * are limited to own subvolumes only
1673                          */
1674                         ret = -EPERM;
1675                 } else {
1676                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1677                                              BTRFS_I(src_inode)->root,
1678                                              transid, readonly, inherit);
1679                 }
1680                 fdput(src);
1681         }
1682 out_drop_write:
1683         mnt_drop_write_file(file);
1684 out:
1685         return ret;
1686 }
1687
1688 static noinline int btrfs_ioctl_snap_create(struct file *file,
1689                                             void __user *arg, int subvol)
1690 {
1691         struct btrfs_ioctl_vol_args *vol_args;
1692         int ret;
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_PATH_NAME_MAX] = '\0';
1698
1699         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1700                                               vol_args->fd, subvol,
1701                                               NULL, false, NULL);
1702
1703         kfree(vol_args);
1704         return ret;
1705 }
1706
1707 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1708                                                void __user *arg, int subvol)
1709 {
1710         struct btrfs_ioctl_vol_args_v2 *vol_args;
1711         int ret;
1712         u64 transid = 0;
1713         u64 *ptr = NULL;
1714         bool readonly = false;
1715         struct btrfs_qgroup_inherit *inherit = NULL;
1716
1717         vol_args = memdup_user(arg, sizeof(*vol_args));
1718         if (IS_ERR(vol_args))
1719                 return PTR_ERR(vol_args);
1720         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1721
1722         if (vol_args->flags &
1723             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1724               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1725                 ret = -EOPNOTSUPP;
1726                 goto free_args;
1727         }
1728
1729         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1730                 ptr = &transid;
1731         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1732                 readonly = true;
1733         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1734                 if (vol_args->size > PAGE_SIZE) {
1735                         ret = -EINVAL;
1736                         goto free_args;
1737                 }
1738                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1739                 if (IS_ERR(inherit)) {
1740                         ret = PTR_ERR(inherit);
1741                         goto free_args;
1742                 }
1743         }
1744
1745         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1746                                               vol_args->fd, subvol, ptr,
1747                                               readonly, inherit);
1748         if (ret)
1749                 goto free_inherit;
1750
1751         if (ptr && copy_to_user(arg +
1752                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1753                                         transid),
1754                                 ptr, sizeof(*ptr)))
1755                 ret = -EFAULT;
1756
1757 free_inherit:
1758         kfree(inherit);
1759 free_args:
1760         kfree(vol_args);
1761         return ret;
1762 }
1763
1764 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1765                                                 void __user *arg)
1766 {
1767         struct inode *inode = file_inode(file);
1768         struct btrfs_root *root = BTRFS_I(inode)->root;
1769         int ret = 0;
1770         u64 flags = 0;
1771
1772         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1773                 return -EINVAL;
1774
1775         down_read(&root->fs_info->subvol_sem);
1776         if (btrfs_root_readonly(root))
1777                 flags |= BTRFS_SUBVOL_RDONLY;
1778         up_read(&root->fs_info->subvol_sem);
1779
1780         if (copy_to_user(arg, &flags, sizeof(flags)))
1781                 ret = -EFAULT;
1782
1783         return ret;
1784 }
1785
1786 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1787                                               void __user *arg)
1788 {
1789         struct inode *inode = file_inode(file);
1790         struct btrfs_root *root = BTRFS_I(inode)->root;
1791         struct btrfs_trans_handle *trans;
1792         u64 root_flags;
1793         u64 flags;
1794         int ret = 0;
1795
1796         if (!inode_owner_or_capable(inode))
1797                 return -EPERM;
1798
1799         ret = mnt_want_write_file(file);
1800         if (ret)
1801                 goto out;
1802
1803         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1804                 ret = -EINVAL;
1805                 goto out_drop_write;
1806         }
1807
1808         if (copy_from_user(&flags, arg, sizeof(flags))) {
1809                 ret = -EFAULT;
1810                 goto out_drop_write;
1811         }
1812
1813         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1814                 ret = -EINVAL;
1815                 goto out_drop_write;
1816         }
1817
1818         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1819                 ret = -EOPNOTSUPP;
1820                 goto out_drop_write;
1821         }
1822
1823         down_write(&root->fs_info->subvol_sem);
1824
1825         /* nothing to do */
1826         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1827                 goto out_drop_sem;
1828
1829         root_flags = btrfs_root_flags(&root->root_item);
1830         if (flags & BTRFS_SUBVOL_RDONLY) {
1831                 btrfs_set_root_flags(&root->root_item,
1832                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1833         } else {
1834                 /*
1835                  * Block RO -> RW transition if this subvolume is involved in
1836                  * send
1837                  */
1838                 spin_lock(&root->root_item_lock);
1839                 if (root->send_in_progress == 0) {
1840                         btrfs_set_root_flags(&root->root_item,
1841                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1842                         spin_unlock(&root->root_item_lock);
1843                 } else {
1844                         spin_unlock(&root->root_item_lock);
1845                         btrfs_warn(root->fs_info,
1846                         "Attempt to set subvolume %llu read-write during send",
1847                                         root->root_key.objectid);
1848                         ret = -EPERM;
1849                         goto out_drop_sem;
1850                 }
1851         }
1852
1853         trans = btrfs_start_transaction(root, 1);
1854         if (IS_ERR(trans)) {
1855                 ret = PTR_ERR(trans);
1856                 goto out_reset;
1857         }
1858
1859         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1860                                 &root->root_key, &root->root_item);
1861
1862         btrfs_commit_transaction(trans, root);
1863 out_reset:
1864         if (ret)
1865                 btrfs_set_root_flags(&root->root_item, root_flags);
1866 out_drop_sem:
1867         up_write(&root->fs_info->subvol_sem);
1868 out_drop_write:
1869         mnt_drop_write_file(file);
1870 out:
1871         return ret;
1872 }
1873
1874 /*
1875  * helper to check if the subvolume references other subvolumes
1876  */
1877 static noinline int may_destroy_subvol(struct btrfs_root *root)
1878 {
1879         struct btrfs_path *path;
1880         struct btrfs_dir_item *di;
1881         struct btrfs_key key;
1882         u64 dir_id;
1883         int ret;
1884
1885         path = btrfs_alloc_path();
1886         if (!path)
1887                 return -ENOMEM;
1888
1889         /* Make sure this root isn't set as the default subvol */
1890         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1891         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1892                                    dir_id, "default", 7, 0);
1893         if (di && !IS_ERR(di)) {
1894                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1895                 if (key.objectid == root->root_key.objectid) {
1896                         ret = -EPERM;
1897                         btrfs_err(root->fs_info, "deleting default subvolume "
1898                                   "%llu is not allowed", key.objectid);
1899                         goto out;
1900                 }
1901                 btrfs_release_path(path);
1902         }
1903
1904         key.objectid = root->root_key.objectid;
1905         key.type = BTRFS_ROOT_REF_KEY;
1906         key.offset = (u64)-1;
1907
1908         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1909                                 &key, path, 0, 0);
1910         if (ret < 0)
1911                 goto out;
1912         BUG_ON(ret == 0);
1913
1914         ret = 0;
1915         if (path->slots[0] > 0) {
1916                 path->slots[0]--;
1917                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1918                 if (key.objectid == root->root_key.objectid &&
1919                     key.type == BTRFS_ROOT_REF_KEY)
1920                         ret = -ENOTEMPTY;
1921         }
1922 out:
1923         btrfs_free_path(path);
1924         return ret;
1925 }
1926
1927 static noinline int key_in_sk(struct btrfs_key *key,
1928                               struct btrfs_ioctl_search_key *sk)
1929 {
1930         struct btrfs_key test;
1931         int ret;
1932
1933         test.objectid = sk->min_objectid;
1934         test.type = sk->min_type;
1935         test.offset = sk->min_offset;
1936
1937         ret = btrfs_comp_cpu_keys(key, &test);
1938         if (ret < 0)
1939                 return 0;
1940
1941         test.objectid = sk->max_objectid;
1942         test.type = sk->max_type;
1943         test.offset = sk->max_offset;
1944
1945         ret = btrfs_comp_cpu_keys(key, &test);
1946         if (ret > 0)
1947                 return 0;
1948         return 1;
1949 }
1950
1951 static noinline int copy_to_sk(struct btrfs_root *root,
1952                                struct btrfs_path *path,
1953                                struct btrfs_key *key,
1954                                struct btrfs_ioctl_search_key *sk,
1955                                size_t *buf_size,
1956                                char __user *ubuf,
1957                                unsigned long *sk_offset,
1958                                int *num_found)
1959 {
1960         u64 found_transid;
1961         struct extent_buffer *leaf;
1962         struct btrfs_ioctl_search_header sh;
1963         struct btrfs_key test;
1964         unsigned long item_off;
1965         unsigned long item_len;
1966         int nritems;
1967         int i;
1968         int slot;
1969         int ret = 0;
1970
1971         leaf = path->nodes[0];
1972         slot = path->slots[0];
1973         nritems = btrfs_header_nritems(leaf);
1974
1975         if (btrfs_header_generation(leaf) > sk->max_transid) {
1976                 i = nritems;
1977                 goto advance_key;
1978         }
1979         found_transid = btrfs_header_generation(leaf);
1980
1981         for (i = slot; i < nritems; i++) {
1982                 item_off = btrfs_item_ptr_offset(leaf, i);
1983                 item_len = btrfs_item_size_nr(leaf, i);
1984
1985                 btrfs_item_key_to_cpu(leaf, key, i);
1986                 if (!key_in_sk(key, sk))
1987                         continue;
1988
1989                 if (sizeof(sh) + item_len > *buf_size) {
1990                         if (*num_found) {
1991                                 ret = 1;
1992                                 goto out;
1993                         }
1994
1995                         /*
1996                          * return one empty item back for v1, which does not
1997                          * handle -EOVERFLOW
1998                          */
1999
2000                         *buf_size = sizeof(sh) + item_len;
2001                         item_len = 0;
2002                         ret = -EOVERFLOW;
2003                 }
2004
2005                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2006                         ret = 1;
2007                         goto out;
2008                 }
2009
2010                 sh.objectid = key->objectid;
2011                 sh.offset = key->offset;
2012                 sh.type = key->type;
2013                 sh.len = item_len;
2014                 sh.transid = found_transid;
2015
2016                 /* copy search result header */
2017                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2018                         ret = -EFAULT;
2019                         goto out;
2020                 }
2021
2022                 *sk_offset += sizeof(sh);
2023
2024                 if (item_len) {
2025                         char __user *up = ubuf + *sk_offset;
2026                         /* copy the item */
2027                         if (read_extent_buffer_to_user(leaf, up,
2028                                                        item_off, item_len)) {
2029                                 ret = -EFAULT;
2030                                 goto out;
2031                         }
2032
2033                         *sk_offset += item_len;
2034                 }
2035                 (*num_found)++;
2036
2037                 if (ret) /* -EOVERFLOW from above */
2038                         goto out;
2039
2040                 if (*num_found >= sk->nr_items) {
2041                         ret = 1;
2042                         goto out;
2043                 }
2044         }
2045 advance_key:
2046         ret = 0;
2047         test.objectid = sk->max_objectid;
2048         test.type = sk->max_type;
2049         test.offset = sk->max_offset;
2050         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2051                 ret = 1;
2052         else if (key->offset < (u64)-1)
2053                 key->offset++;
2054         else if (key->type < (u8)-1) {
2055                 key->offset = 0;
2056                 key->type++;
2057         } else if (key->objectid < (u64)-1) {
2058                 key->offset = 0;
2059                 key->type = 0;
2060                 key->objectid++;
2061         } else
2062                 ret = 1;
2063 out:
2064         /*
2065          *  0: all items from this leaf copied, continue with next
2066          *  1: * more items can be copied, but unused buffer is too small
2067          *     * all items were found
2068          *     Either way, it will stops the loop which iterates to the next
2069          *     leaf
2070          *  -EOVERFLOW: item was to large for buffer
2071          *  -EFAULT: could not copy extent buffer back to userspace
2072          */
2073         return ret;
2074 }
2075
2076 static noinline int search_ioctl(struct inode *inode,
2077                                  struct btrfs_ioctl_search_key *sk,
2078                                  size_t *buf_size,
2079                                  char __user *ubuf)
2080 {
2081         struct btrfs_root *root;
2082         struct btrfs_key key;
2083         struct btrfs_path *path;
2084         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2085         int ret;
2086         int num_found = 0;
2087         unsigned long sk_offset = 0;
2088
2089         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2090                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2091                 return -EOVERFLOW;
2092         }
2093
2094         path = btrfs_alloc_path();
2095         if (!path)
2096                 return -ENOMEM;
2097
2098         if (sk->tree_id == 0) {
2099                 /* search the root of the inode that was passed */
2100                 root = BTRFS_I(inode)->root;
2101         } else {
2102                 key.objectid = sk->tree_id;
2103                 key.type = BTRFS_ROOT_ITEM_KEY;
2104                 key.offset = (u64)-1;
2105                 root = btrfs_read_fs_root_no_name(info, &key);
2106                 if (IS_ERR(root)) {
2107                         btrfs_free_path(path);
2108                         return -ENOENT;
2109                 }
2110         }
2111
2112         key.objectid = sk->min_objectid;
2113         key.type = sk->min_type;
2114         key.offset = sk->min_offset;
2115
2116         while (1) {
2117                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2118                 if (ret != 0) {
2119                         if (ret > 0)
2120                                 ret = 0;
2121                         goto err;
2122                 }
2123                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2124                                  &sk_offset, &num_found);
2125                 btrfs_release_path(path);
2126                 if (ret)
2127                         break;
2128
2129         }
2130         if (ret > 0)
2131                 ret = 0;
2132 err:
2133         sk->nr_items = num_found;
2134         btrfs_free_path(path);
2135         return ret;
2136 }
2137
2138 static noinline int btrfs_ioctl_tree_search(struct file *file,
2139                                            void __user *argp)
2140 {
2141         struct btrfs_ioctl_search_args __user *uargs;
2142         struct btrfs_ioctl_search_key sk;
2143         struct inode *inode;
2144         int ret;
2145         size_t buf_size;
2146
2147         if (!capable(CAP_SYS_ADMIN))
2148                 return -EPERM;
2149
2150         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2151
2152         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2153                 return -EFAULT;
2154
2155         buf_size = sizeof(uargs->buf);
2156
2157         inode = file_inode(file);
2158         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2159
2160         /*
2161          * In the origin implementation an overflow is handled by returning a
2162          * search header with a len of zero, so reset ret.
2163          */
2164         if (ret == -EOVERFLOW)
2165                 ret = 0;
2166
2167         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2168                 ret = -EFAULT;
2169         return ret;
2170 }
2171
2172 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2173                                                void __user *argp)
2174 {
2175         struct btrfs_ioctl_search_args_v2 __user *uarg;
2176         struct btrfs_ioctl_search_args_v2 args;
2177         struct inode *inode;
2178         int ret;
2179         size_t buf_size;
2180         const size_t buf_limit = SZ_16M;
2181
2182         if (!capable(CAP_SYS_ADMIN))
2183                 return -EPERM;
2184
2185         /* copy search header and buffer size */
2186         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2187         if (copy_from_user(&args, uarg, sizeof(args)))
2188                 return -EFAULT;
2189
2190         buf_size = args.buf_size;
2191
2192         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2193                 return -EOVERFLOW;
2194
2195         /* limit result size to 16MB */
2196         if (buf_size > buf_limit)
2197                 buf_size = buf_limit;
2198
2199         inode = file_inode(file);
2200         ret = search_ioctl(inode, &args.key, &buf_size,
2201                            (char *)(&uarg->buf[0]));
2202         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2203                 ret = -EFAULT;
2204         else if (ret == -EOVERFLOW &&
2205                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2206                 ret = -EFAULT;
2207
2208         return ret;
2209 }
2210
2211 /*
2212  * Search INODE_REFs to identify path name of 'dirid' directory
2213  * in a 'tree_id' tree. and sets path name to 'name'.
2214  */
2215 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2216                                 u64 tree_id, u64 dirid, char *name)
2217 {
2218         struct btrfs_root *root;
2219         struct btrfs_key key;
2220         char *ptr;
2221         int ret = -1;
2222         int slot;
2223         int len;
2224         int total_len = 0;
2225         struct btrfs_inode_ref *iref;
2226         struct extent_buffer *l;
2227         struct btrfs_path *path;
2228
2229         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2230                 name[0]='\0';
2231                 return 0;
2232         }
2233
2234         path = btrfs_alloc_path();
2235         if (!path)
2236                 return -ENOMEM;
2237
2238         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2239
2240         key.objectid = tree_id;
2241         key.type = BTRFS_ROOT_ITEM_KEY;
2242         key.offset = (u64)-1;
2243         root = btrfs_read_fs_root_no_name(info, &key);
2244         if (IS_ERR(root)) {
2245                 btrfs_err(info, "could not find root %llu", tree_id);
2246                 ret = -ENOENT;
2247                 goto out;
2248         }
2249
2250         key.objectid = dirid;
2251         key.type = BTRFS_INODE_REF_KEY;
2252         key.offset = (u64)-1;
2253
2254         while (1) {
2255                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2256                 if (ret < 0)
2257                         goto out;
2258                 else if (ret > 0) {
2259                         ret = btrfs_previous_item(root, path, dirid,
2260                                                   BTRFS_INODE_REF_KEY);
2261                         if (ret < 0)
2262                                 goto out;
2263                         else if (ret > 0) {
2264                                 ret = -ENOENT;
2265                                 goto out;
2266                         }
2267                 }
2268
2269                 l = path->nodes[0];
2270                 slot = path->slots[0];
2271                 btrfs_item_key_to_cpu(l, &key, slot);
2272
2273                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2274                 len = btrfs_inode_ref_name_len(l, iref);
2275                 ptr -= len + 1;
2276                 total_len += len + 1;
2277                 if (ptr < name) {
2278                         ret = -ENAMETOOLONG;
2279                         goto out;
2280                 }
2281
2282                 *(ptr + len) = '/';
2283                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2284
2285                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2286                         break;
2287
2288                 btrfs_release_path(path);
2289                 key.objectid = key.offset;
2290                 key.offset = (u64)-1;
2291                 dirid = key.objectid;
2292         }
2293         memmove(name, ptr, total_len);
2294         name[total_len] = '\0';
2295         ret = 0;
2296 out:
2297         btrfs_free_path(path);
2298         return ret;
2299 }
2300
2301 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2302                                            void __user *argp)
2303 {
2304          struct btrfs_ioctl_ino_lookup_args *args;
2305          struct inode *inode;
2306         int ret = 0;
2307
2308         args = memdup_user(argp, sizeof(*args));
2309         if (IS_ERR(args))
2310                 return PTR_ERR(args);
2311
2312         inode = file_inode(file);
2313
2314         /*
2315          * Unprivileged query to obtain the containing subvolume root id. The
2316          * path is reset so it's consistent with btrfs_search_path_in_tree.
2317          */
2318         if (args->treeid == 0)
2319                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2320
2321         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2322                 args->name[0] = 0;
2323                 goto out;
2324         }
2325
2326         if (!capable(CAP_SYS_ADMIN)) {
2327                 ret = -EPERM;
2328                 goto out;
2329         }
2330
2331         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2332                                         args->treeid, args->objectid,
2333                                         args->name);
2334
2335 out:
2336         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2337                 ret = -EFAULT;
2338
2339         kfree(args);
2340         return ret;
2341 }
2342
2343 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2344                                              void __user *arg)
2345 {
2346         struct dentry *parent = file->f_path.dentry;
2347         struct dentry *dentry;
2348         struct inode *dir = d_inode(parent);
2349         struct inode *inode;
2350         struct btrfs_root *root = BTRFS_I(dir)->root;
2351         struct btrfs_root *dest = NULL;
2352         struct btrfs_ioctl_vol_args *vol_args;
2353         struct btrfs_trans_handle *trans;
2354         struct btrfs_block_rsv block_rsv;
2355         u64 root_flags;
2356         u64 qgroup_reserved;
2357         int namelen;
2358         int ret;
2359         int err = 0;
2360
2361         vol_args = memdup_user(arg, sizeof(*vol_args));
2362         if (IS_ERR(vol_args))
2363                 return PTR_ERR(vol_args);
2364
2365         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2366         namelen = strlen(vol_args->name);
2367         if (strchr(vol_args->name, '/') ||
2368             strncmp(vol_args->name, "..", namelen) == 0) {
2369                 err = -EINVAL;
2370                 goto out;
2371         }
2372
2373         err = mnt_want_write_file(file);
2374         if (err)
2375                 goto out;
2376
2377
2378         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2379         if (err == -EINTR)
2380                 goto out_drop_write;
2381         dentry = lookup_one_len(vol_args->name, parent, namelen);
2382         if (IS_ERR(dentry)) {
2383                 err = PTR_ERR(dentry);
2384                 goto out_unlock_dir;
2385         }
2386
2387         if (d_really_is_negative(dentry)) {
2388                 err = -ENOENT;
2389                 goto out_dput;
2390         }
2391
2392         inode = d_inode(dentry);
2393         dest = BTRFS_I(inode)->root;
2394         if (!capable(CAP_SYS_ADMIN)) {
2395                 /*
2396                  * Regular user.  Only allow this with a special mount
2397                  * option, when the user has write+exec access to the
2398                  * subvol root, and when rmdir(2) would have been
2399                  * allowed.
2400                  *
2401                  * Note that this is _not_ check that the subvol is
2402                  * empty or doesn't contain data that we wouldn't
2403                  * otherwise be able to delete.
2404                  *
2405                  * Users who want to delete empty subvols should try
2406                  * rmdir(2).
2407                  */
2408                 err = -EPERM;
2409                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2410                         goto out_dput;
2411
2412                 /*
2413                  * Do not allow deletion if the parent dir is the same
2414                  * as the dir to be deleted.  That means the ioctl
2415                  * must be called on the dentry referencing the root
2416                  * of the subvol, not a random directory contained
2417                  * within it.
2418                  */
2419                 err = -EINVAL;
2420                 if (root == dest)
2421                         goto out_dput;
2422
2423                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2424                 if (err)
2425                         goto out_dput;
2426         }
2427
2428         /* check if subvolume may be deleted by a user */
2429         err = btrfs_may_delete(dir, dentry, 1);
2430         if (err)
2431                 goto out_dput;
2432
2433         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2434                 err = -EINVAL;
2435                 goto out_dput;
2436         }
2437
2438         inode_lock(inode);
2439
2440         /*
2441          * Don't allow to delete a subvolume with send in progress. This is
2442          * inside the i_mutex so the error handling that has to drop the bit
2443          * again is not run concurrently.
2444          */
2445         spin_lock(&dest->root_item_lock);
2446         root_flags = btrfs_root_flags(&dest->root_item);
2447         if (dest->send_in_progress == 0) {
2448                 btrfs_set_root_flags(&dest->root_item,
2449                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2450                 spin_unlock(&dest->root_item_lock);
2451         } else {
2452                 spin_unlock(&dest->root_item_lock);
2453                 btrfs_warn(root->fs_info,
2454                         "Attempt to delete subvolume %llu during send",
2455                         dest->root_key.objectid);
2456                 err = -EPERM;
2457                 goto out_unlock_inode;
2458         }
2459
2460         down_write(&root->fs_info->subvol_sem);
2461
2462         err = may_destroy_subvol(dest);
2463         if (err)
2464                 goto out_up_write;
2465
2466         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2467         /*
2468          * One for dir inode, two for dir entries, two for root
2469          * ref/backref.
2470          */
2471         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2472                                                5, &qgroup_reserved, true);
2473         if (err)
2474                 goto out_up_write;
2475
2476         trans = btrfs_start_transaction(root, 0);
2477         if (IS_ERR(trans)) {
2478                 err = PTR_ERR(trans);
2479                 goto out_release;
2480         }
2481         trans->block_rsv = &block_rsv;
2482         trans->bytes_reserved = block_rsv.size;
2483
2484         btrfs_record_snapshot_destroy(trans, dir);
2485
2486         ret = btrfs_unlink_subvol(trans, root, dir,
2487                                 dest->root_key.objectid,
2488                                 dentry->d_name.name,
2489                                 dentry->d_name.len);
2490         if (ret) {
2491                 err = ret;
2492                 btrfs_abort_transaction(trans, root, ret);
2493                 goto out_end_trans;
2494         }
2495
2496         btrfs_record_root_in_trans(trans, dest);
2497
2498         memset(&dest->root_item.drop_progress, 0,
2499                 sizeof(dest->root_item.drop_progress));
2500         dest->root_item.drop_level = 0;
2501         btrfs_set_root_refs(&dest->root_item, 0);
2502
2503         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2504                 ret = btrfs_insert_orphan_item(trans,
2505                                         root->fs_info->tree_root,
2506                                         dest->root_key.objectid);
2507                 if (ret) {
2508                         btrfs_abort_transaction(trans, root, ret);
2509                         err = ret;
2510                         goto out_end_trans;
2511                 }
2512         }
2513
2514         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2515                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2516                                   dest->root_key.objectid);
2517         if (ret && ret != -ENOENT) {
2518                 btrfs_abort_transaction(trans, root, ret);
2519                 err = ret;
2520                 goto out_end_trans;
2521         }
2522         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2523                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2524                                           dest->root_item.received_uuid,
2525                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2526                                           dest->root_key.objectid);
2527                 if (ret && ret != -ENOENT) {
2528                         btrfs_abort_transaction(trans, root, ret);
2529                         err = ret;
2530                         goto out_end_trans;
2531                 }
2532         }
2533
2534 out_end_trans:
2535         trans->block_rsv = NULL;
2536         trans->bytes_reserved = 0;
2537         ret = btrfs_end_transaction(trans, root);
2538         if (ret && !err)
2539                 err = ret;
2540         inode->i_flags |= S_DEAD;
2541 out_release:
2542         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2543 out_up_write:
2544         up_write(&root->fs_info->subvol_sem);
2545         if (err) {
2546                 spin_lock(&dest->root_item_lock);
2547                 root_flags = btrfs_root_flags(&dest->root_item);
2548                 btrfs_set_root_flags(&dest->root_item,
2549                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2550                 spin_unlock(&dest->root_item_lock);
2551         }
2552 out_unlock_inode:
2553         inode_unlock(inode);
2554         if (!err) {
2555                 d_invalidate(dentry);
2556                 btrfs_invalidate_inodes(dest);
2557                 d_delete(dentry);
2558                 ASSERT(dest->send_in_progress == 0);
2559
2560                 /* the last ref */
2561                 if (dest->ino_cache_inode) {
2562                         iput(dest->ino_cache_inode);
2563                         dest->ino_cache_inode = NULL;
2564                 }
2565         }
2566 out_dput:
2567         dput(dentry);
2568 out_unlock_dir:
2569         inode_unlock(dir);
2570 out_drop_write:
2571         mnt_drop_write_file(file);
2572 out:
2573         kfree(vol_args);
2574         return err;
2575 }
2576
2577 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2578 {
2579         struct inode *inode = file_inode(file);
2580         struct btrfs_root *root = BTRFS_I(inode)->root;
2581         struct btrfs_ioctl_defrag_range_args *range;
2582         int ret;
2583
2584         ret = mnt_want_write_file(file);
2585         if (ret)
2586                 return ret;
2587
2588         if (btrfs_root_readonly(root)) {
2589                 ret = -EROFS;
2590                 goto out;
2591         }
2592
2593         switch (inode->i_mode & S_IFMT) {
2594         case S_IFDIR:
2595                 if (!capable(CAP_SYS_ADMIN)) {
2596                         ret = -EPERM;
2597                         goto out;
2598                 }
2599                 ret = btrfs_defrag_root(root);
2600                 if (ret)
2601                         goto out;
2602                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2603                 break;
2604         case S_IFREG:
2605                 if (!(file->f_mode & FMODE_WRITE)) {
2606                         ret = -EINVAL;
2607                         goto out;
2608                 }
2609
2610                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2611                 if (!range) {
2612                         ret = -ENOMEM;
2613                         goto out;
2614                 }
2615
2616                 if (argp) {
2617                         if (copy_from_user(range, argp,
2618                                            sizeof(*range))) {
2619                                 ret = -EFAULT;
2620                                 kfree(range);
2621                                 goto out;
2622                         }
2623                         /* compression requires us to start the IO */
2624                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2625                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2626                                 range->extent_thresh = (u32)-1;
2627                         }
2628                 } else {
2629                         /* the rest are all set to zero by kzalloc */
2630                         range->len = (u64)-1;
2631                 }
2632                 ret = btrfs_defrag_file(file_inode(file), file,
2633                                         range, 0, 0);
2634                 if (ret > 0)
2635                         ret = 0;
2636                 kfree(range);
2637                 break;
2638         default:
2639                 ret = -EINVAL;
2640         }
2641 out:
2642         mnt_drop_write_file(file);
2643         return ret;
2644 }
2645
2646 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2647 {
2648         struct btrfs_ioctl_vol_args *vol_args;
2649         int ret;
2650
2651         if (!capable(CAP_SYS_ADMIN))
2652                 return -EPERM;
2653
2654         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2655                         1)) {
2656                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2657         }
2658
2659         mutex_lock(&root->fs_info->volume_mutex);
2660         vol_args = memdup_user(arg, sizeof(*vol_args));
2661         if (IS_ERR(vol_args)) {
2662                 ret = PTR_ERR(vol_args);
2663                 goto out;
2664         }
2665
2666         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2667         ret = btrfs_init_new_device(root, vol_args->name);
2668
2669         if (!ret)
2670                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2671
2672         kfree(vol_args);
2673 out:
2674         mutex_unlock(&root->fs_info->volume_mutex);
2675         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2676         return ret;
2677 }
2678
2679 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2680 {
2681         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2682         struct btrfs_ioctl_vol_args_v2 *vol_args;
2683         int ret;
2684
2685         if (!capable(CAP_SYS_ADMIN))
2686                 return -EPERM;
2687
2688         ret = mnt_want_write_file(file);
2689         if (ret)
2690                 return ret;
2691
2692         vol_args = memdup_user(arg, sizeof(*vol_args));
2693         if (IS_ERR(vol_args)) {
2694                 ret = PTR_ERR(vol_args);
2695                 goto err_drop;
2696         }
2697
2698         /* Check for compatibility reject unknown flags */
2699         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2700                 return -EOPNOTSUPP;
2701
2702         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2703                         1)) {
2704                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2705                 goto out;
2706         }
2707
2708         mutex_lock(&root->fs_info->volume_mutex);
2709         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2710                 ret = btrfs_rm_device(root, NULL, vol_args->devid);
2711         } else {
2712                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2713                 ret = btrfs_rm_device(root, vol_args->name, 0);
2714         }
2715         mutex_unlock(&root->fs_info->volume_mutex);
2716         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2717
2718         if (!ret) {
2719                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2720                         btrfs_info(root->fs_info, "device deleted: id %llu",
2721                                         vol_args->devid);
2722                 else
2723                         btrfs_info(root->fs_info, "device deleted: %s",
2724                                         vol_args->name);
2725         }
2726 out:
2727         kfree(vol_args);
2728 err_drop:
2729         mnt_drop_write_file(file);
2730         return ret;
2731 }
2732
2733 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2734 {
2735         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2736         struct btrfs_ioctl_vol_args *vol_args;
2737         int ret;
2738
2739         if (!capable(CAP_SYS_ADMIN))
2740                 return -EPERM;
2741
2742         ret = mnt_want_write_file(file);
2743         if (ret)
2744                 return ret;
2745
2746         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2747                         1)) {
2748                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2749                 goto out_drop_write;
2750         }
2751
2752         vol_args = memdup_user(arg, sizeof(*vol_args));
2753         if (IS_ERR(vol_args)) {
2754                 ret = PTR_ERR(vol_args);
2755                 goto out;
2756         }
2757
2758         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2759         mutex_lock(&root->fs_info->volume_mutex);
2760         ret = btrfs_rm_device(root, vol_args->name, 0);
2761         mutex_unlock(&root->fs_info->volume_mutex);
2762
2763         if (!ret)
2764                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2765         kfree(vol_args);
2766 out:
2767         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2768 out_drop_write:
2769         mnt_drop_write_file(file);
2770
2771         return ret;
2772 }
2773
2774 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2775 {
2776         struct btrfs_ioctl_fs_info_args *fi_args;
2777         struct btrfs_device *device;
2778         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2779         int ret = 0;
2780
2781         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2782         if (!fi_args)
2783                 return -ENOMEM;
2784
2785         mutex_lock(&fs_devices->device_list_mutex);
2786         fi_args->num_devices = fs_devices->num_devices;
2787         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2788
2789         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2790                 if (device->devid > fi_args->max_id)
2791                         fi_args->max_id = device->devid;
2792         }
2793         mutex_unlock(&fs_devices->device_list_mutex);
2794
2795         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2796         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2797         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2798
2799         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2800                 ret = -EFAULT;
2801
2802         kfree(fi_args);
2803         return ret;
2804 }
2805
2806 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2807 {
2808         struct btrfs_ioctl_dev_info_args *di_args;
2809         struct btrfs_device *dev;
2810         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2811         int ret = 0;
2812         char *s_uuid = NULL;
2813
2814         di_args = memdup_user(arg, sizeof(*di_args));
2815         if (IS_ERR(di_args))
2816                 return PTR_ERR(di_args);
2817
2818         if (!btrfs_is_empty_uuid(di_args->uuid))
2819                 s_uuid = di_args->uuid;
2820
2821         mutex_lock(&fs_devices->device_list_mutex);
2822         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2823
2824         if (!dev) {
2825                 ret = -ENODEV;
2826                 goto out;
2827         }
2828
2829         di_args->devid = dev->devid;
2830         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2831         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2832         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2833         if (dev->name) {
2834                 struct rcu_string *name;
2835
2836                 rcu_read_lock();
2837                 name = rcu_dereference(dev->name);
2838                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2839                 rcu_read_unlock();
2840                 di_args->path[sizeof(di_args->path) - 1] = 0;
2841         } else {
2842                 di_args->path[0] = '\0';
2843         }
2844
2845 out:
2846         mutex_unlock(&fs_devices->device_list_mutex);
2847         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2848                 ret = -EFAULT;
2849
2850         kfree(di_args);
2851         return ret;
2852 }
2853
2854 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2855 {
2856         struct page *page;
2857
2858         page = grab_cache_page(inode->i_mapping, index);
2859         if (!page)
2860                 return ERR_PTR(-ENOMEM);
2861
2862         if (!PageUptodate(page)) {
2863                 int ret;
2864
2865                 ret = btrfs_readpage(NULL, page);
2866                 if (ret)
2867                         return ERR_PTR(ret);
2868                 lock_page(page);
2869                 if (!PageUptodate(page)) {
2870                         unlock_page(page);
2871                         put_page(page);
2872                         return ERR_PTR(-EIO);
2873                 }
2874                 if (page->mapping != inode->i_mapping) {
2875                         unlock_page(page);
2876                         put_page(page);
2877                         return ERR_PTR(-EAGAIN);
2878                 }
2879         }
2880
2881         return page;
2882 }
2883
2884 static int gather_extent_pages(struct inode *inode, struct page **pages,
2885                                int num_pages, u64 off)
2886 {
2887         int i;
2888         pgoff_t index = off >> PAGE_SHIFT;
2889
2890         for (i = 0; i < num_pages; i++) {
2891 again:
2892                 pages[i] = extent_same_get_page(inode, index + i);
2893                 if (IS_ERR(pages[i])) {
2894                         int err = PTR_ERR(pages[i]);
2895
2896                         if (err == -EAGAIN)
2897                                 goto again;
2898                         pages[i] = NULL;
2899                         return err;
2900                 }
2901         }
2902         return 0;
2903 }
2904
2905 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2906                              bool retry_range_locking)
2907 {
2908         /*
2909          * Do any pending delalloc/csum calculations on inode, one way or
2910          * another, and lock file content.
2911          * The locking order is:
2912          *
2913          *   1) pages
2914          *   2) range in the inode's io tree
2915          */
2916         while (1) {
2917                 struct btrfs_ordered_extent *ordered;
2918                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2919                 ordered = btrfs_lookup_first_ordered_extent(inode,
2920                                                             off + len - 1);
2921                 if ((!ordered ||
2922                      ordered->file_offset + ordered->len <= off ||
2923                      ordered->file_offset >= off + len) &&
2924                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2925                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2926                         if (ordered)
2927                                 btrfs_put_ordered_extent(ordered);
2928                         break;
2929                 }
2930                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2931                 if (ordered)
2932                         btrfs_put_ordered_extent(ordered);
2933                 if (!retry_range_locking)
2934                         return -EAGAIN;
2935                 btrfs_wait_ordered_range(inode, off, len);
2936         }
2937         return 0;
2938 }
2939
2940 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2941 {
2942         inode_unlock(inode1);
2943         inode_unlock(inode2);
2944 }
2945
2946 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2947 {
2948         if (inode1 < inode2)
2949                 swap(inode1, inode2);
2950
2951         inode_lock_nested(inode1, I_MUTEX_PARENT);
2952         inode_lock_nested(inode2, I_MUTEX_CHILD);
2953 }
2954
2955 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2956                                       struct inode *inode2, u64 loff2, u64 len)
2957 {
2958         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2959         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2960 }
2961
2962 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2963                                     struct inode *inode2, u64 loff2, u64 len,
2964                                     bool retry_range_locking)
2965 {
2966         int ret;
2967
2968         if (inode1 < inode2) {
2969                 swap(inode1, inode2);
2970                 swap(loff1, loff2);
2971         }
2972         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2973         if (ret)
2974                 return ret;
2975         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2976         if (ret)
2977                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2978                               loff1 + len - 1);
2979         return ret;
2980 }
2981
2982 struct cmp_pages {
2983         int             num_pages;
2984         struct page     **src_pages;
2985         struct page     **dst_pages;
2986 };
2987
2988 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2989 {
2990         int i;
2991         struct page *pg;
2992
2993         for (i = 0; i < cmp->num_pages; i++) {
2994                 pg = cmp->src_pages[i];
2995                 if (pg) {
2996                         unlock_page(pg);
2997                         put_page(pg);
2998                 }
2999                 pg = cmp->dst_pages[i];
3000                 if (pg) {
3001                         unlock_page(pg);
3002                         put_page(pg);
3003                 }
3004         }
3005         kfree(cmp->src_pages);
3006         kfree(cmp->dst_pages);
3007 }
3008
3009 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3010                                   struct inode *dst, u64 dst_loff,
3011                                   u64 len, struct cmp_pages *cmp)
3012 {
3013         int ret;
3014         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3015         struct page **src_pgarr, **dst_pgarr;
3016
3017         /*
3018          * We must gather up all the pages before we initiate our
3019          * extent locking. We use an array for the page pointers. Size
3020          * of the array is bounded by len, which is in turn bounded by
3021          * BTRFS_MAX_DEDUPE_LEN.
3022          */
3023         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3024         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3025         if (!src_pgarr || !dst_pgarr) {
3026                 kfree(src_pgarr);
3027                 kfree(dst_pgarr);
3028                 return -ENOMEM;
3029         }
3030         cmp->num_pages = num_pages;
3031         cmp->src_pages = src_pgarr;
3032         cmp->dst_pages = dst_pgarr;
3033
3034         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3035         if (ret)
3036                 goto out;
3037
3038         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3039
3040 out:
3041         if (ret)
3042                 btrfs_cmp_data_free(cmp);
3043         return 0;
3044 }
3045
3046 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3047                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
3048 {
3049         int ret = 0;
3050         int i;
3051         struct page *src_page, *dst_page;
3052         unsigned int cmp_len = PAGE_SIZE;
3053         void *addr, *dst_addr;
3054
3055         i = 0;
3056         while (len) {
3057                 if (len < PAGE_SIZE)
3058                         cmp_len = len;
3059
3060                 BUG_ON(i >= cmp->num_pages);
3061
3062                 src_page = cmp->src_pages[i];
3063                 dst_page = cmp->dst_pages[i];
3064                 ASSERT(PageLocked(src_page));
3065                 ASSERT(PageLocked(dst_page));
3066
3067                 addr = kmap_atomic(src_page);
3068                 dst_addr = kmap_atomic(dst_page);
3069
3070                 flush_dcache_page(src_page);
3071                 flush_dcache_page(dst_page);
3072
3073                 if (memcmp(addr, dst_addr, cmp_len))
3074                         ret = -EBADE;
3075
3076                 kunmap_atomic(addr);
3077                 kunmap_atomic(dst_addr);
3078
3079                 if (ret)
3080                         break;
3081
3082                 len -= cmp_len;
3083                 i++;
3084         }
3085
3086         return ret;
3087 }
3088
3089 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3090                                      u64 olen)
3091 {
3092         u64 len = *plen;
3093         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3094
3095         if (off + olen > inode->i_size || off + olen < off)
3096                 return -EINVAL;
3097
3098         /* if we extend to eof, continue to block boundary */
3099         if (off + len == inode->i_size)
3100                 *plen = len = ALIGN(inode->i_size, bs) - off;
3101
3102         /* Check that we are block aligned - btrfs_clone() requires this */
3103         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3104                 return -EINVAL;
3105
3106         return 0;
3107 }
3108
3109 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3110                              struct inode *dst, u64 dst_loff)
3111 {
3112         int ret;
3113         u64 len = olen;
3114         struct cmp_pages cmp;
3115         int same_inode = 0;
3116         u64 same_lock_start = 0;
3117         u64 same_lock_len = 0;
3118
3119         if (src == dst)
3120                 same_inode = 1;
3121
3122         if (len == 0)
3123                 return 0;
3124
3125         if (same_inode) {
3126                 inode_lock(src);
3127
3128                 ret = extent_same_check_offsets(src, loff, &len, olen);
3129                 if (ret)
3130                         goto out_unlock;
3131                 ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3132                 if (ret)
3133                         goto out_unlock;
3134
3135                 /*
3136                  * Single inode case wants the same checks, except we
3137                  * don't want our length pushed out past i_size as
3138                  * comparing that data range makes no sense.
3139                  *
3140                  * extent_same_check_offsets() will do this for an
3141                  * unaligned length at i_size, so catch it here and
3142                  * reject the request.
3143                  *
3144                  * This effectively means we require aligned extents
3145                  * for the single-inode case, whereas the other cases
3146                  * allow an unaligned length so long as it ends at
3147                  * i_size.
3148                  */
3149                 if (len != olen) {
3150                         ret = -EINVAL;
3151                         goto out_unlock;
3152                 }
3153
3154                 /* Check for overlapping ranges */
3155                 if (dst_loff + len > loff && dst_loff < loff + len) {
3156                         ret = -EINVAL;
3157                         goto out_unlock;
3158                 }
3159
3160                 same_lock_start = min_t(u64, loff, dst_loff);
3161                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3162         } else {
3163                 btrfs_double_inode_lock(src, dst);
3164
3165                 ret = extent_same_check_offsets(src, loff, &len, olen);
3166                 if (ret)
3167                         goto out_unlock;
3168
3169                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3170                 if (ret)
3171                         goto out_unlock;
3172         }
3173
3174         /* don't make the dst file partly checksummed */
3175         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3176             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3177                 ret = -EINVAL;
3178                 goto out_unlock;
3179         }
3180
3181 again:
3182         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3183         if (ret)
3184                 goto out_unlock;
3185
3186         if (same_inode)
3187                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3188                                         false);
3189         else
3190                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3191                                                false);
3192         /*
3193          * If one of the inodes has dirty pages in the respective range or
3194          * ordered extents, we need to flush dellaloc and wait for all ordered
3195          * extents in the range. We must unlock the pages and the ranges in the
3196          * io trees to avoid deadlocks when flushing delalloc (requires locking
3197          * pages) and when waiting for ordered extents to complete (they require
3198          * range locking).
3199          */
3200         if (ret == -EAGAIN) {
3201                 /*
3202                  * Ranges in the io trees already unlocked. Now unlock all
3203                  * pages before waiting for all IO to complete.
3204                  */
3205                 btrfs_cmp_data_free(&cmp);
3206                 if (same_inode) {
3207                         btrfs_wait_ordered_range(src, same_lock_start,
3208                                                  same_lock_len);
3209                 } else {
3210                         btrfs_wait_ordered_range(src, loff, len);
3211                         btrfs_wait_ordered_range(dst, dst_loff, len);
3212                 }
3213                 goto again;
3214         }
3215         ASSERT(ret == 0);
3216         if (WARN_ON(ret)) {
3217                 /* ranges in the io trees already unlocked */
3218                 btrfs_cmp_data_free(&cmp);
3219                 return ret;
3220         }
3221
3222         /* pass original length for comparison so we stay within i_size */
3223         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3224         if (ret == 0)
3225                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3226
3227         if (same_inode)
3228                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3229                               same_lock_start + same_lock_len - 1);
3230         else
3231                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3232
3233         btrfs_cmp_data_free(&cmp);
3234 out_unlock:
3235         if (same_inode)
3236                 inode_unlock(src);
3237         else
3238                 btrfs_double_inode_unlock(src, dst);
3239
3240         return ret;
3241 }
3242
3243 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3244
3245 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3246                                 struct file *dst_file, u64 dst_loff)
3247 {
3248         struct inode *src = file_inode(src_file);
3249         struct inode *dst = file_inode(dst_file);
3250         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3251         ssize_t res;
3252
3253         if (olen > BTRFS_MAX_DEDUPE_LEN)
3254                 olen = BTRFS_MAX_DEDUPE_LEN;
3255
3256         if (WARN_ON_ONCE(bs < PAGE_SI