GNU Linux-libre 5.19-rc6-gnu
[releases.git] / fs / btrfs / ioctl.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/file.h>
9 #include <linux/fs.h>
10 #include <linux/fsnotify.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/string.h>
15 #include <linux/backing-dev.h>
16 #include <linux/mount.h>
17 #include <linux/namei.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/security.h>
21 #include <linux/xattr.h>
22 #include <linux/mm.h>
23 #include <linux/slab.h>
24 #include <linux/blkdev.h>
25 #include <linux/uuid.h>
26 #include <linux/btrfs.h>
27 #include <linux/uaccess.h>
28 #include <linux/iversion.h>
29 #include <linux/fileattr.h>
30 #include <linux/fsverity.h>
31 #include <linux/sched/xacct.h>
32 #include "ctree.h"
33 #include "disk-io.h"
34 #include "export.h"
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "print-tree.h"
38 #include "volumes.h"
39 #include "locking.h"
40 #include "backref.h"
41 #include "rcu-string.h"
42 #include "send.h"
43 #include "dev-replace.h"
44 #include "props.h"
45 #include "sysfs.h"
46 #include "qgroup.h"
47 #include "tree-log.h"
48 #include "compression.h"
49 #include "space-info.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "subpage.h"
53
54 #ifdef CONFIG_64BIT
55 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
56  * structures are incorrect, as the timespec structure from userspace
57  * is 4 bytes too small. We define these alternatives here to teach
58  * the kernel about the 32-bit struct packing.
59  */
60 struct btrfs_ioctl_timespec_32 {
61         __u64 sec;
62         __u32 nsec;
63 } __attribute__ ((__packed__));
64
65 struct btrfs_ioctl_received_subvol_args_32 {
66         char    uuid[BTRFS_UUID_SIZE];  /* in */
67         __u64   stransid;               /* in */
68         __u64   rtransid;               /* out */
69         struct btrfs_ioctl_timespec_32 stime; /* in */
70         struct btrfs_ioctl_timespec_32 rtime; /* out */
71         __u64   flags;                  /* in */
72         __u64   reserved[16];           /* in */
73 } __attribute__ ((__packed__));
74
75 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
76                                 struct btrfs_ioctl_received_subvol_args_32)
77 #endif
78
79 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
80 struct btrfs_ioctl_send_args_32 {
81         __s64 send_fd;                  /* in */
82         __u64 clone_sources_count;      /* in */
83         compat_uptr_t clone_sources;    /* in */
84         __u64 parent_root;              /* in */
85         __u64 flags;                    /* in */
86         __u32 version;                  /* in */
87         __u8  reserved[28];             /* in */
88 } __attribute__ ((__packed__));
89
90 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
91                                struct btrfs_ioctl_send_args_32)
92
93 struct btrfs_ioctl_encoded_io_args_32 {
94         compat_uptr_t iov;
95         compat_ulong_t iovcnt;
96         __s64 offset;
97         __u64 flags;
98         __u64 len;
99         __u64 unencoded_len;
100         __u64 unencoded_offset;
101         __u32 compression;
102         __u32 encryption;
103         __u8 reserved[64];
104 };
105
106 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
107                                        struct btrfs_ioctl_encoded_io_args_32)
108 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
109                                         struct btrfs_ioctl_encoded_io_args_32)
110 #endif
111
112 /* Mask out flags that are inappropriate for the given type of inode. */
113 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
114                 unsigned int flags)
115 {
116         if (S_ISDIR(inode->i_mode))
117                 return flags;
118         else if (S_ISREG(inode->i_mode))
119                 return flags & ~FS_DIRSYNC_FL;
120         else
121                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
122 }
123
124 /*
125  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
126  * ioctl.
127  */
128 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
129 {
130         unsigned int iflags = 0;
131         u32 flags = binode->flags;
132         u32 ro_flags = binode->ro_flags;
133
134         if (flags & BTRFS_INODE_SYNC)
135                 iflags |= FS_SYNC_FL;
136         if (flags & BTRFS_INODE_IMMUTABLE)
137                 iflags |= FS_IMMUTABLE_FL;
138         if (flags & BTRFS_INODE_APPEND)
139                 iflags |= FS_APPEND_FL;
140         if (flags & BTRFS_INODE_NODUMP)
141                 iflags |= FS_NODUMP_FL;
142         if (flags & BTRFS_INODE_NOATIME)
143                 iflags |= FS_NOATIME_FL;
144         if (flags & BTRFS_INODE_DIRSYNC)
145                 iflags |= FS_DIRSYNC_FL;
146         if (flags & BTRFS_INODE_NODATACOW)
147                 iflags |= FS_NOCOW_FL;
148         if (ro_flags & BTRFS_INODE_RO_VERITY)
149                 iflags |= FS_VERITY_FL;
150
151         if (flags & BTRFS_INODE_NOCOMPRESS)
152                 iflags |= FS_NOCOMP_FL;
153         else if (flags & BTRFS_INODE_COMPRESS)
154                 iflags |= FS_COMPR_FL;
155
156         return iflags;
157 }
158
159 /*
160  * Update inode->i_flags based on the btrfs internal flags.
161  */
162 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
163 {
164         struct btrfs_inode *binode = BTRFS_I(inode);
165         unsigned int new_fl = 0;
166
167         if (binode->flags & BTRFS_INODE_SYNC)
168                 new_fl |= S_SYNC;
169         if (binode->flags & BTRFS_INODE_IMMUTABLE)
170                 new_fl |= S_IMMUTABLE;
171         if (binode->flags & BTRFS_INODE_APPEND)
172                 new_fl |= S_APPEND;
173         if (binode->flags & BTRFS_INODE_NOATIME)
174                 new_fl |= S_NOATIME;
175         if (binode->flags & BTRFS_INODE_DIRSYNC)
176                 new_fl |= S_DIRSYNC;
177         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
178                 new_fl |= S_VERITY;
179
180         set_mask_bits(&inode->i_flags,
181                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
182                       S_VERITY, new_fl);
183 }
184
185 /*
186  * Check if @flags are a supported and valid set of FS_*_FL flags and that
187  * the old and new flags are not conflicting
188  */
189 static int check_fsflags(unsigned int old_flags, unsigned int flags)
190 {
191         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
192                       FS_NOATIME_FL | FS_NODUMP_FL | \
193                       FS_SYNC_FL | FS_DIRSYNC_FL | \
194                       FS_NOCOMP_FL | FS_COMPR_FL |
195                       FS_NOCOW_FL))
196                 return -EOPNOTSUPP;
197
198         /* COMPR and NOCOMP on new/old are valid */
199         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
200                 return -EINVAL;
201
202         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
203                 return -EINVAL;
204
205         /* NOCOW and compression options are mutually exclusive */
206         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
207                 return -EINVAL;
208         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
209                 return -EINVAL;
210
211         return 0;
212 }
213
214 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
215                                     unsigned int flags)
216 {
217         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
218                 return -EPERM;
219
220         return 0;
221 }
222
223 /*
224  * Set flags/xflags from the internal inode flags. The remaining items of
225  * fsxattr are zeroed.
226  */
227 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
228 {
229         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
230
231         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
232         return 0;
233 }
234
235 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
236                        struct dentry *dentry, struct fileattr *fa)
237 {
238         struct inode *inode = d_inode(dentry);
239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
240         struct btrfs_inode *binode = BTRFS_I(inode);
241         struct btrfs_root *root = binode->root;
242         struct btrfs_trans_handle *trans;
243         unsigned int fsflags, old_fsflags;
244         int ret;
245         const char *comp = NULL;
246         u32 binode_flags;
247
248         if (btrfs_root_readonly(root))
249                 return -EROFS;
250
251         if (fileattr_has_fsx(fa))
252                 return -EOPNOTSUPP;
253
254         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
255         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
256         ret = check_fsflags(old_fsflags, fsflags);
257         if (ret)
258                 return ret;
259
260         ret = check_fsflags_compatible(fs_info, fsflags);
261         if (ret)
262                 return ret;
263
264         binode_flags = binode->flags;
265         if (fsflags & FS_SYNC_FL)
266                 binode_flags |= BTRFS_INODE_SYNC;
267         else
268                 binode_flags &= ~BTRFS_INODE_SYNC;
269         if (fsflags & FS_IMMUTABLE_FL)
270                 binode_flags |= BTRFS_INODE_IMMUTABLE;
271         else
272                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
273         if (fsflags & FS_APPEND_FL)
274                 binode_flags |= BTRFS_INODE_APPEND;
275         else
276                 binode_flags &= ~BTRFS_INODE_APPEND;
277         if (fsflags & FS_NODUMP_FL)
278                 binode_flags |= BTRFS_INODE_NODUMP;
279         else
280                 binode_flags &= ~BTRFS_INODE_NODUMP;
281         if (fsflags & FS_NOATIME_FL)
282                 binode_flags |= BTRFS_INODE_NOATIME;
283         else
284                 binode_flags &= ~BTRFS_INODE_NOATIME;
285
286         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
287         if (!fa->flags_valid) {
288                 /* 1 item for the inode */
289                 trans = btrfs_start_transaction(root, 1);
290                 if (IS_ERR(trans))
291                         return PTR_ERR(trans);
292                 goto update_flags;
293         }
294
295         if (fsflags & FS_DIRSYNC_FL)
296                 binode_flags |= BTRFS_INODE_DIRSYNC;
297         else
298                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
299         if (fsflags & FS_NOCOW_FL) {
300                 if (S_ISREG(inode->i_mode)) {
301                         /*
302                          * It's safe to turn csums off here, no extents exist.
303                          * Otherwise we want the flag to reflect the real COW
304                          * status of the file and will not set it.
305                          */
306                         if (inode->i_size == 0)
307                                 binode_flags |= BTRFS_INODE_NODATACOW |
308                                                 BTRFS_INODE_NODATASUM;
309                 } else {
310                         binode_flags |= BTRFS_INODE_NODATACOW;
311                 }
312         } else {
313                 /*
314                  * Revert back under same assumptions as above
315                  */
316                 if (S_ISREG(inode->i_mode)) {
317                         if (inode->i_size == 0)
318                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
319                                                   BTRFS_INODE_NODATASUM);
320                 } else {
321                         binode_flags &= ~BTRFS_INODE_NODATACOW;
322                 }
323         }
324
325         /*
326          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
327          * flag may be changed automatically if compression code won't make
328          * things smaller.
329          */
330         if (fsflags & FS_NOCOMP_FL) {
331                 binode_flags &= ~BTRFS_INODE_COMPRESS;
332                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
333         } else if (fsflags & FS_COMPR_FL) {
334
335                 if (IS_SWAPFILE(inode))
336                         return -ETXTBSY;
337
338                 binode_flags |= BTRFS_INODE_COMPRESS;
339                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
340
341                 comp = btrfs_compress_type2str(fs_info->compress_type);
342                 if (!comp || comp[0] == 0)
343                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
344         } else {
345                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
346         }
347
348         /*
349          * 1 for inode item
350          * 2 for properties
351          */
352         trans = btrfs_start_transaction(root, 3);
353         if (IS_ERR(trans))
354                 return PTR_ERR(trans);
355
356         if (comp) {
357                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
358                                      strlen(comp), 0);
359                 if (ret) {
360                         btrfs_abort_transaction(trans, ret);
361                         goto out_end_trans;
362                 }
363         } else {
364                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
365                                      0, 0);
366                 if (ret && ret != -ENODATA) {
367                         btrfs_abort_transaction(trans, ret);
368                         goto out_end_trans;
369                 }
370         }
371
372 update_flags:
373         binode->flags = binode_flags;
374         btrfs_sync_inode_flags_to_i_flags(inode);
375         inode_inc_iversion(inode);
376         inode->i_ctime = current_time(inode);
377         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
378
379  out_end_trans:
380         btrfs_end_transaction(trans);
381         return ret;
382 }
383
384 /*
385  * Start exclusive operation @type, return true on success
386  */
387 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
388                         enum btrfs_exclusive_operation type)
389 {
390         bool ret = false;
391
392         spin_lock(&fs_info->super_lock);
393         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
394                 fs_info->exclusive_operation = type;
395                 ret = true;
396         }
397         spin_unlock(&fs_info->super_lock);
398
399         return ret;
400 }
401
402 /*
403  * Conditionally allow to enter the exclusive operation in case it's compatible
404  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
405  * btrfs_exclop_finish.
406  *
407  * Compatibility:
408  * - the same type is already running
409  * - when trying to add a device and balance has been paused
410  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
411  *   must check the condition first that would allow none -> @type
412  */
413 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
414                                  enum btrfs_exclusive_operation type)
415 {
416         spin_lock(&fs_info->super_lock);
417         if (fs_info->exclusive_operation == type ||
418             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
419              type == BTRFS_EXCLOP_DEV_ADD))
420                 return true;
421
422         spin_unlock(&fs_info->super_lock);
423         return false;
424 }
425
426 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
427 {
428         spin_unlock(&fs_info->super_lock);
429 }
430
431 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
432 {
433         spin_lock(&fs_info->super_lock);
434         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
435         spin_unlock(&fs_info->super_lock);
436         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
437 }
438
439 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
440                           enum btrfs_exclusive_operation op)
441 {
442         switch (op) {
443         case BTRFS_EXCLOP_BALANCE_PAUSED:
444                 spin_lock(&fs_info->super_lock);
445                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
446                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD);
447                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
448                 spin_unlock(&fs_info->super_lock);
449                 break;
450         case BTRFS_EXCLOP_BALANCE:
451                 spin_lock(&fs_info->super_lock);
452                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
453                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
454                 spin_unlock(&fs_info->super_lock);
455                 break;
456         default:
457                 btrfs_warn(fs_info,
458                         "invalid exclop balance operation %d requested", op);
459         }
460 }
461
462 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
463 {
464         return put_user(inode->i_generation, arg);
465 }
466
467 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
468                                         void __user *arg)
469 {
470         struct btrfs_device *device;
471         struct fstrim_range range;
472         u64 minlen = ULLONG_MAX;
473         u64 num_devices = 0;
474         int ret;
475
476         if (!capable(CAP_SYS_ADMIN))
477                 return -EPERM;
478
479         /*
480          * btrfs_trim_block_group() depends on space cache, which is not
481          * available in zoned filesystem. So, disallow fitrim on a zoned
482          * filesystem for now.
483          */
484         if (btrfs_is_zoned(fs_info))
485                 return -EOPNOTSUPP;
486
487         /*
488          * If the fs is mounted with nologreplay, which requires it to be
489          * mounted in RO mode as well, we can not allow discard on free space
490          * inside block groups, because log trees refer to extents that are not
491          * pinned in a block group's free space cache (pinning the extents is
492          * precisely the first phase of replaying a log tree).
493          */
494         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
495                 return -EROFS;
496
497         rcu_read_lock();
498         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
499                                 dev_list) {
500                 if (!device->bdev || !bdev_max_discard_sectors(device->bdev))
501                         continue;
502                 num_devices++;
503                 minlen = min_t(u64, bdev_discard_granularity(device->bdev),
504                                     minlen);
505         }
506         rcu_read_unlock();
507
508         if (!num_devices)
509                 return -EOPNOTSUPP;
510         if (copy_from_user(&range, arg, sizeof(range)))
511                 return -EFAULT;
512
513         /*
514          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
515          * block group is in the logical address space, which can be any
516          * sectorsize aligned bytenr in  the range [0, U64_MAX].
517          */
518         if (range.len < fs_info->sb->s_blocksize)
519                 return -EINVAL;
520
521         range.minlen = max(range.minlen, minlen);
522         ret = btrfs_trim_fs(fs_info, &range);
523         if (ret < 0)
524                 return ret;
525
526         if (copy_to_user(arg, &range, sizeof(range)))
527                 return -EFAULT;
528
529         return 0;
530 }
531
532 int __pure btrfs_is_empty_uuid(u8 *uuid)
533 {
534         int i;
535
536         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
537                 if (uuid[i])
538                         return 0;
539         }
540         return 1;
541 }
542
543 /*
544  * Calculate the number of transaction items to reserve for creating a subvolume
545  * or snapshot, not including the inode, directory entries, or parent directory.
546  */
547 static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit)
548 {
549         /*
550          * 1 to add root block
551          * 1 to add root item
552          * 1 to add root ref
553          * 1 to add root backref
554          * 1 to add UUID item
555          * 1 to add qgroup info
556          * 1 to add qgroup limit
557          *
558          * Ideally the last two would only be accounted if qgroups are enabled,
559          * but that can change between now and the time we would insert them.
560          */
561         unsigned int num_items = 7;
562
563         if (inherit) {
564                 /* 2 to add qgroup relations for each inherited qgroup */
565                 num_items += 2 * inherit->num_qgroups;
566         }
567         return num_items;
568 }
569
570 static noinline int create_subvol(struct user_namespace *mnt_userns,
571                                   struct inode *dir, struct dentry *dentry,
572                                   struct btrfs_qgroup_inherit *inherit)
573 {
574         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
575         struct btrfs_trans_handle *trans;
576         struct btrfs_key key;
577         struct btrfs_root_item *root_item;
578         struct btrfs_inode_item *inode_item;
579         struct extent_buffer *leaf;
580         struct btrfs_root *root = BTRFS_I(dir)->root;
581         struct btrfs_root *new_root;
582         struct btrfs_block_rsv block_rsv;
583         struct timespec64 cur_time = current_time(dir);
584         struct btrfs_new_inode_args new_inode_args = {
585                 .dir = dir,
586                 .dentry = dentry,
587                 .subvol = true,
588         };
589         unsigned int trans_num_items;
590         int ret;
591         dev_t anon_dev;
592         u64 objectid;
593
594         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
595         if (!root_item)
596                 return -ENOMEM;
597
598         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
599         if (ret)
600                 goto out_root_item;
601
602         /*
603          * Don't create subvolume whose level is not zero. Or qgroup will be
604          * screwed up since it assumes subvolume qgroup's level to be 0.
605          */
606         if (btrfs_qgroup_level(objectid)) {
607                 ret = -ENOSPC;
608                 goto out_root_item;
609         }
610
611         ret = get_anon_bdev(&anon_dev);
612         if (ret < 0)
613                 goto out_root_item;
614
615         new_inode_args.inode = btrfs_new_subvol_inode(mnt_userns, dir);
616         if (!new_inode_args.inode) {
617                 ret = -ENOMEM;
618                 goto out_anon_dev;
619         }
620         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
621         if (ret)
622                 goto out_inode;
623         trans_num_items += create_subvol_num_items(inherit);
624
625         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
626         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
627                                                trans_num_items, false);
628         if (ret)
629                 goto out_new_inode_args;
630
631         trans = btrfs_start_transaction(root, 0);
632         if (IS_ERR(trans)) {
633                 ret = PTR_ERR(trans);
634                 btrfs_subvolume_release_metadata(root, &block_rsv);
635                 goto out_new_inode_args;
636         }
637         trans->block_rsv = &block_rsv;
638         trans->bytes_reserved = block_rsv.size;
639
640         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
641         if (ret)
642                 goto out;
643
644         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
645                                       BTRFS_NESTING_NORMAL);
646         if (IS_ERR(leaf)) {
647                 ret = PTR_ERR(leaf);
648                 goto out;
649         }
650
651         btrfs_mark_buffer_dirty(leaf);
652
653         inode_item = &root_item->inode;
654         btrfs_set_stack_inode_generation(inode_item, 1);
655         btrfs_set_stack_inode_size(inode_item, 3);
656         btrfs_set_stack_inode_nlink(inode_item, 1);
657         btrfs_set_stack_inode_nbytes(inode_item,
658                                      fs_info->nodesize);
659         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
660
661         btrfs_set_root_flags(root_item, 0);
662         btrfs_set_root_limit(root_item, 0);
663         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
664
665         btrfs_set_root_bytenr(root_item, leaf->start);
666         btrfs_set_root_generation(root_item, trans->transid);
667         btrfs_set_root_level(root_item, 0);
668         btrfs_set_root_refs(root_item, 1);
669         btrfs_set_root_used(root_item, leaf->len);
670         btrfs_set_root_last_snapshot(root_item, 0);
671
672         btrfs_set_root_generation_v2(root_item,
673                         btrfs_root_generation(root_item));
674         generate_random_guid(root_item->uuid);
675         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
676         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
677         root_item->ctime = root_item->otime;
678         btrfs_set_root_ctransid(root_item, trans->transid);
679         btrfs_set_root_otransid(root_item, trans->transid);
680
681         btrfs_tree_unlock(leaf);
682
683         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
684
685         key.objectid = objectid;
686         key.offset = 0;
687         key.type = BTRFS_ROOT_ITEM_KEY;
688         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
689                                 root_item);
690         if (ret) {
691                 /*
692                  * Since we don't abort the transaction in this case, free the
693                  * tree block so that we don't leak space and leave the
694                  * filesystem in an inconsistent state (an extent item in the
695                  * extent tree with a backreference for a root that does not
696                  * exists).
697                  */
698                 btrfs_tree_lock(leaf);
699                 btrfs_clean_tree_block(leaf);
700                 btrfs_tree_unlock(leaf);
701                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
702                 free_extent_buffer(leaf);
703                 goto out;
704         }
705
706         free_extent_buffer(leaf);
707         leaf = NULL;
708
709         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
710         if (IS_ERR(new_root)) {
711                 ret = PTR_ERR(new_root);
712                 btrfs_abort_transaction(trans, ret);
713                 goto out;
714         }
715         /* anon_dev is owned by new_root now. */
716         anon_dev = 0;
717         BTRFS_I(new_inode_args.inode)->root = new_root;
718         /* ... and new_root is owned by new_inode_args.inode now. */
719
720         ret = btrfs_record_root_in_trans(trans, new_root);
721         if (ret) {
722                 btrfs_abort_transaction(trans, ret);
723                 goto out;
724         }
725
726         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
727                                   BTRFS_UUID_KEY_SUBVOL, objectid);
728         if (ret) {
729                 btrfs_abort_transaction(trans, ret);
730                 goto out;
731         }
732
733         ret = btrfs_create_new_inode(trans, &new_inode_args);
734         if (ret) {
735                 btrfs_abort_transaction(trans, ret);
736                 goto out;
737         }
738
739         d_instantiate_new(dentry, new_inode_args.inode);
740         new_inode_args.inode = NULL;
741
742 out:
743         trans->block_rsv = NULL;
744         trans->bytes_reserved = 0;
745         btrfs_subvolume_release_metadata(root, &block_rsv);
746
747         if (ret)
748                 btrfs_end_transaction(trans);
749         else
750                 ret = btrfs_commit_transaction(trans);
751 out_new_inode_args:
752         btrfs_new_inode_args_destroy(&new_inode_args);
753 out_inode:
754         iput(new_inode_args.inode);
755 out_anon_dev:
756         if (anon_dev)
757                 free_anon_bdev(anon_dev);
758 out_root_item:
759         kfree(root_item);
760         return ret;
761 }
762
763 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
764                            struct dentry *dentry, bool readonly,
765                            struct btrfs_qgroup_inherit *inherit)
766 {
767         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
768         struct inode *inode;
769         struct btrfs_pending_snapshot *pending_snapshot;
770         unsigned int trans_num_items;
771         struct btrfs_trans_handle *trans;
772         int ret;
773
774         /* We do not support snapshotting right now. */
775         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
776                 btrfs_warn(fs_info,
777                            "extent tree v2 doesn't support snapshotting yet");
778                 return -EOPNOTSUPP;
779         }
780
781         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
782                 return -EINVAL;
783
784         if (atomic_read(&root->nr_swapfiles)) {
785                 btrfs_warn(fs_info,
786                            "cannot snapshot subvolume with active swapfile");
787                 return -ETXTBSY;
788         }
789
790         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
791         if (!pending_snapshot)
792                 return -ENOMEM;
793
794         ret = get_anon_bdev(&pending_snapshot->anon_dev);
795         if (ret < 0)
796                 goto free_pending;
797         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
798                         GFP_KERNEL);
799         pending_snapshot->path = btrfs_alloc_path();
800         if (!pending_snapshot->root_item || !pending_snapshot->path) {
801                 ret = -ENOMEM;
802                 goto free_pending;
803         }
804
805         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
806                              BTRFS_BLOCK_RSV_TEMP);
807         /*
808          * 1 to add dir item
809          * 1 to add dir index
810          * 1 to update parent inode item
811          */
812         trans_num_items = create_subvol_num_items(inherit) + 3;
813         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
814                                                &pending_snapshot->block_rsv,
815                                                trans_num_items, false);
816         if (ret)
817                 goto free_pending;
818
819         pending_snapshot->dentry = dentry;
820         pending_snapshot->root = root;
821         pending_snapshot->readonly = readonly;
822         pending_snapshot->dir = dir;
823         pending_snapshot->inherit = inherit;
824
825         trans = btrfs_start_transaction(root, 0);
826         if (IS_ERR(trans)) {
827                 ret = PTR_ERR(trans);
828                 goto fail;
829         }
830
831         trans->pending_snapshot = pending_snapshot;
832
833         ret = btrfs_commit_transaction(trans);
834         if (ret)
835                 goto fail;
836
837         ret = pending_snapshot->error;
838         if (ret)
839                 goto fail;
840
841         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
842         if (ret)
843                 goto fail;
844
845         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
846         if (IS_ERR(inode)) {
847                 ret = PTR_ERR(inode);
848                 goto fail;
849         }
850
851         d_instantiate(dentry, inode);
852         ret = 0;
853         pending_snapshot->anon_dev = 0;
854 fail:
855         /* Prevent double freeing of anon_dev */
856         if (ret && pending_snapshot->snap)
857                 pending_snapshot->snap->anon_dev = 0;
858         btrfs_put_root(pending_snapshot->snap);
859         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
860 free_pending:
861         if (pending_snapshot->anon_dev)
862                 free_anon_bdev(pending_snapshot->anon_dev);
863         kfree(pending_snapshot->root_item);
864         btrfs_free_path(pending_snapshot->path);
865         kfree(pending_snapshot);
866
867         return ret;
868 }
869
870 /*  copy of may_delete in fs/namei.c()
871  *      Check whether we can remove a link victim from directory dir, check
872  *  whether the type of victim is right.
873  *  1. We can't do it if dir is read-only (done in permission())
874  *  2. We should have write and exec permissions on dir
875  *  3. We can't remove anything from append-only dir
876  *  4. We can't do anything with immutable dir (done in permission())
877  *  5. If the sticky bit on dir is set we should either
878  *      a. be owner of dir, or
879  *      b. be owner of victim, or
880  *      c. have CAP_FOWNER capability
881  *  6. If the victim is append-only or immutable we can't do anything with
882  *     links pointing to it.
883  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
884  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
885  *  9. We can't remove a root or mountpoint.
886  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
887  *     nfs_async_unlink().
888  */
889
890 static int btrfs_may_delete(struct user_namespace *mnt_userns,
891                             struct inode *dir, struct dentry *victim, int isdir)
892 {
893         int error;
894
895         if (d_really_is_negative(victim))
896                 return -ENOENT;
897
898         BUG_ON(d_inode(victim->d_parent) != dir);
899         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
900
901         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
902         if (error)
903                 return error;
904         if (IS_APPEND(dir))
905                 return -EPERM;
906         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
907             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
908             IS_SWAPFILE(d_inode(victim)))
909                 return -EPERM;
910         if (isdir) {
911                 if (!d_is_dir(victim))
912                         return -ENOTDIR;
913                 if (IS_ROOT(victim))
914                         return -EBUSY;
915         } else if (d_is_dir(victim))
916                 return -EISDIR;
917         if (IS_DEADDIR(dir))
918                 return -ENOENT;
919         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
920                 return -EBUSY;
921         return 0;
922 }
923
924 /* copy of may_create in fs/namei.c() */
925 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
926                                    struct inode *dir, struct dentry *child)
927 {
928         if (d_really_is_positive(child))
929                 return -EEXIST;
930         if (IS_DEADDIR(dir))
931                 return -ENOENT;
932         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
933                 return -EOVERFLOW;
934         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
935 }
936
937 /*
938  * Create a new subvolume below @parent.  This is largely modeled after
939  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
940  * inside this filesystem so it's quite a bit simpler.
941  */
942 static noinline int btrfs_mksubvol(const struct path *parent,
943                                    struct user_namespace *mnt_userns,
944                                    const char *name, int namelen,
945                                    struct btrfs_root *snap_src,
946                                    bool readonly,
947                                    struct btrfs_qgroup_inherit *inherit)
948 {
949         struct inode *dir = d_inode(parent->dentry);
950         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
951         struct dentry *dentry;
952         int error;
953
954         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
955         if (error == -EINTR)
956                 return error;
957
958         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
959         error = PTR_ERR(dentry);
960         if (IS_ERR(dentry))
961                 goto out_unlock;
962
963         error = btrfs_may_create(mnt_userns, dir, dentry);
964         if (error)
965                 goto out_dput;
966
967         /*
968          * even if this name doesn't exist, we may get hash collisions.
969          * check for them now when we can safely fail
970          */
971         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
972                                                dir->i_ino, name,
973                                                namelen);
974         if (error)
975                 goto out_dput;
976
977         down_read(&fs_info->subvol_sem);
978
979         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
980                 goto out_up_read;
981
982         if (snap_src)
983                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
984         else
985                 error = create_subvol(mnt_userns, dir, dentry, inherit);
986
987         if (!error)
988                 fsnotify_mkdir(dir, dentry);
989 out_up_read:
990         up_read(&fs_info->subvol_sem);
991 out_dput:
992         dput(dentry);
993 out_unlock:
994         btrfs_inode_unlock(dir, 0);
995         return error;
996 }
997
998 static noinline int btrfs_mksnapshot(const struct path *parent,
999                                    struct user_namespace *mnt_userns,
1000                                    const char *name, int namelen,
1001                                    struct btrfs_root *root,
1002                                    bool readonly,
1003                                    struct btrfs_qgroup_inherit *inherit)
1004 {
1005         int ret;
1006         bool snapshot_force_cow = false;
1007
1008         /*
1009          * Force new buffered writes to reserve space even when NOCOW is
1010          * possible. This is to avoid later writeback (running dealloc) to
1011          * fallback to COW mode and unexpectedly fail with ENOSPC.
1012          */
1013         btrfs_drew_read_lock(&root->snapshot_lock);
1014
1015         ret = btrfs_start_delalloc_snapshot(root, false);
1016         if (ret)
1017                 goto out;
1018
1019         /*
1020          * All previous writes have started writeback in NOCOW mode, so now
1021          * we force future writes to fallback to COW mode during snapshot
1022          * creation.
1023          */
1024         atomic_inc(&root->snapshot_force_cow);
1025         snapshot_force_cow = true;
1026
1027         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1028
1029         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1030                              root, readonly, inherit);
1031 out:
1032         if (snapshot_force_cow)
1033                 atomic_dec(&root->snapshot_force_cow);
1034         btrfs_drew_read_unlock(&root->snapshot_lock);
1035         return ret;
1036 }
1037
1038 /*
1039  * Defrag specific helper to get an extent map.
1040  *
1041  * Differences between this and btrfs_get_extent() are:
1042  *
1043  * - No extent_map will be added to inode->extent_tree
1044  *   To reduce memory usage in the long run.
1045  *
1046  * - Extra optimization to skip file extents older than @newer_than
1047  *   By using btrfs_search_forward() we can skip entire file ranges that
1048  *   have extents created in past transactions, because btrfs_search_forward()
1049  *   will not visit leaves and nodes with a generation smaller than given
1050  *   minimal generation threshold (@newer_than).
1051  *
1052  * Return valid em if we find a file extent matching the requirement.
1053  * Return NULL if we can not find a file extent matching the requirement.
1054  *
1055  * Return ERR_PTR() for error.
1056  */
1057 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1058                                             u64 start, u64 newer_than)
1059 {
1060         struct btrfs_root *root = inode->root;
1061         struct btrfs_file_extent_item *fi;
1062         struct btrfs_path path = { 0 };
1063         struct extent_map *em;
1064         struct btrfs_key key;
1065         u64 ino = btrfs_ino(inode);
1066         int ret;
1067
1068         em = alloc_extent_map();
1069         if (!em) {
1070                 ret = -ENOMEM;
1071                 goto err;
1072         }
1073
1074         key.objectid = ino;
1075         key.type = BTRFS_EXTENT_DATA_KEY;
1076         key.offset = start;
1077
1078         if (newer_than) {
1079                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1080                 if (ret < 0)
1081                         goto err;
1082                 /* Can't find anything newer */
1083                 if (ret > 0)
1084                         goto not_found;
1085         } else {
1086                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1087                 if (ret < 0)
1088                         goto err;
1089         }
1090         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1091                 /*
1092                  * If btrfs_search_slot() makes path to point beyond nritems,
1093                  * we should not have an empty leaf, as this inode must at
1094                  * least have its INODE_ITEM.
1095                  */
1096                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1097                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1098         }
1099         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1100         /* Perfect match, no need to go one slot back */
1101         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1102             key.offset == start)
1103                 goto iterate;
1104
1105         /* We didn't find a perfect match, needs to go one slot back */
1106         if (path.slots[0] > 0) {
1107                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1108                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1109                         path.slots[0]--;
1110         }
1111
1112 iterate:
1113         /* Iterate through the path to find a file extent covering @start */
1114         while (true) {
1115                 u64 extent_end;
1116
1117                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1118                         goto next;
1119
1120                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1121
1122                 /*
1123                  * We may go one slot back to INODE_REF/XATTR item, then
1124                  * need to go forward until we reach an EXTENT_DATA.
1125                  * But we should still has the correct ino as key.objectid.
1126                  */
1127                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1128                         goto next;
1129
1130                 /* It's beyond our target range, definitely not extent found */
1131                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1132                         goto not_found;
1133
1134                 /*
1135                  *      |       |<- File extent ->|
1136                  *      \- start
1137                  *
1138                  * This means there is a hole between start and key.offset.
1139                  */
1140                 if (key.offset > start) {
1141                         em->start = start;
1142                         em->orig_start = start;
1143                         em->block_start = EXTENT_MAP_HOLE;
1144                         em->len = key.offset - start;
1145                         break;
1146                 }
1147
1148                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1149                                     struct btrfs_file_extent_item);
1150                 extent_end = btrfs_file_extent_end(&path);
1151
1152                 /*
1153                  *      |<- file extent ->|     |
1154                  *                              \- start
1155                  *
1156                  * We haven't reached start, search next slot.
1157                  */
1158                 if (extent_end <= start)
1159                         goto next;
1160
1161                 /* Now this extent covers @start, convert it to em */
1162                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1163                 break;
1164 next:
1165                 ret = btrfs_next_item(root, &path);
1166                 if (ret < 0)
1167                         goto err;
1168                 if (ret > 0)
1169                         goto not_found;
1170         }
1171         btrfs_release_path(&path);
1172         return em;
1173
1174 not_found:
1175         btrfs_release_path(&path);
1176         free_extent_map(em);
1177         return NULL;
1178
1179 err:
1180         btrfs_release_path(&path);
1181         free_extent_map(em);
1182         return ERR_PTR(ret);
1183 }
1184
1185 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1186                                                u64 newer_than, bool locked)
1187 {
1188         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1189         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1190         struct extent_map *em;
1191         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1192
1193         /*
1194          * hopefully we have this extent in the tree already, try without
1195          * the full extent lock
1196          */
1197         read_lock(&em_tree->lock);
1198         em = lookup_extent_mapping(em_tree, start, sectorsize);
1199         read_unlock(&em_tree->lock);
1200
1201         /*
1202          * We can get a merged extent, in that case, we need to re-search
1203          * tree to get the original em for defrag.
1204          *
1205          * If @newer_than is 0 or em::generation < newer_than, we can trust
1206          * this em, as either we don't care about the generation, or the
1207          * merged extent map will be rejected anyway.
1208          */
1209         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1210             newer_than && em->generation >= newer_than) {
1211                 free_extent_map(em);
1212                 em = NULL;
1213         }
1214
1215         if (!em) {
1216                 struct extent_state *cached = NULL;
1217                 u64 end = start + sectorsize - 1;
1218
1219                 /* get the big lock and read metadata off disk */
1220                 if (!locked)
1221                         lock_extent_bits(io_tree, start, end, &cached);
1222                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1223                 if (!locked)
1224                         unlock_extent_cached(io_tree, start, end, &cached);
1225
1226                 if (IS_ERR(em))
1227                         return NULL;
1228         }
1229
1230         return em;
1231 }
1232
1233 static u32 get_extent_max_capacity(const struct extent_map *em)
1234 {
1235         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1236                 return BTRFS_MAX_COMPRESSED;
1237         return BTRFS_MAX_EXTENT_SIZE;
1238 }
1239
1240 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1241                                      u32 extent_thresh, u64 newer_than, bool locked)
1242 {
1243         struct extent_map *next;
1244         bool ret = false;
1245
1246         /* this is the last extent */
1247         if (em->start + em->len >= i_size_read(inode))
1248                 return false;
1249
1250         /*
1251          * Here we need to pass @newer_then when checking the next extent, or
1252          * we will hit a case we mark current extent for defrag, but the next
1253          * one will not be a target.
1254          * This will just cause extra IO without really reducing the fragments.
1255          */
1256         next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked);
1257         /* No more em or hole */
1258         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1259                 goto out;
1260         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1261                 goto out;
1262         /*
1263          * If the next extent is at its max capacity, defragging current extent
1264          * makes no sense, as the total number of extents won't change.
1265          */
1266         if (next->len >= get_extent_max_capacity(em))
1267                 goto out;
1268         /* Skip older extent */
1269         if (next->generation < newer_than)
1270                 goto out;
1271         /* Also check extent size */
1272         if (next->len >= extent_thresh)
1273                 goto out;
1274
1275         ret = true;
1276 out:
1277         free_extent_map(next);
1278         return ret;
1279 }
1280
1281 /*
1282  * Prepare one page to be defragged.
1283  *
1284  * This will ensure:
1285  *
1286  * - Returned page is locked and has been set up properly.
1287  * - No ordered extent exists in the page.
1288  * - The page is uptodate.
1289  *
1290  * NOTE: Caller should also wait for page writeback after the cluster is
1291  * prepared, here we don't do writeback wait for each page.
1292  */
1293 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1294                                             pgoff_t index)
1295 {
1296         struct address_space *mapping = inode->vfs_inode.i_mapping;
1297         gfp_t mask = btrfs_alloc_write_mask(mapping);
1298         u64 page_start = (u64)index << PAGE_SHIFT;
1299         u64 page_end = page_start + PAGE_SIZE - 1;
1300         struct extent_state *cached_state = NULL;
1301         struct page *page;
1302         int ret;
1303
1304 again:
1305         page = find_or_create_page(mapping, index, mask);
1306         if (!page)
1307                 return ERR_PTR(-ENOMEM);
1308
1309         /*
1310          * Since we can defragment files opened read-only, we can encounter
1311          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1312          * can't do I/O using huge pages yet, so return an error for now.
1313          * Filesystem transparent huge pages are typically only used for
1314          * executables that explicitly enable them, so this isn't very
1315          * restrictive.
1316          */
1317         if (PageCompound(page)) {
1318                 unlock_page(page);
1319                 put_page(page);
1320                 return ERR_PTR(-ETXTBSY);
1321         }
1322
1323         ret = set_page_extent_mapped(page);
1324         if (ret < 0) {
1325                 unlock_page(page);
1326                 put_page(page);
1327                 return ERR_PTR(ret);
1328         }
1329
1330         /* Wait for any existing ordered extent in the range */
1331         while (1) {
1332                 struct btrfs_ordered_extent *ordered;
1333
1334                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1335                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1336                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1337                                      &cached_state);
1338                 if (!ordered)
1339                         break;
1340
1341                 unlock_page(page);
1342                 btrfs_start_ordered_extent(ordered, 1);
1343                 btrfs_put_ordered_extent(ordered);
1344                 lock_page(page);
1345                 /*
1346                  * We unlocked the page above, so we need check if it was
1347                  * released or not.
1348                  */
1349                 if (page->mapping != mapping || !PagePrivate(page)) {
1350                         unlock_page(page);
1351                         put_page(page);
1352                         goto again;
1353                 }
1354         }
1355
1356         /*
1357          * Now the page range has no ordered extent any more.  Read the page to
1358          * make it uptodate.
1359          */
1360         if (!PageUptodate(page)) {
1361                 btrfs_read_folio(NULL, page_folio(page));
1362                 lock_page(page);
1363                 if (page->mapping != mapping || !PagePrivate(page)) {
1364                         unlock_page(page);
1365                         put_page(page);
1366                         goto again;
1367                 }
1368                 if (!PageUptodate(page)) {
1369                         unlock_page(page);
1370                         put_page(page);
1371                         return ERR_PTR(-EIO);
1372                 }
1373         }
1374         return page;
1375 }
1376
1377 struct defrag_target_range {
1378         struct list_head list;
1379         u64 start;
1380         u64 len;
1381 };
1382
1383 /*
1384  * Collect all valid target extents.
1385  *
1386  * @start:         file offset to lookup
1387  * @len:           length to lookup
1388  * @extent_thresh: file extent size threshold, any extent size >= this value
1389  *                 will be ignored
1390  * @newer_than:    only defrag extents newer than this value
1391  * @do_compress:   whether the defrag is doing compression
1392  *                 if true, @extent_thresh will be ignored and all regular
1393  *                 file extents meeting @newer_than will be targets.
1394  * @locked:        if the range has already held extent lock
1395  * @target_list:   list of targets file extents
1396  */
1397 static int defrag_collect_targets(struct btrfs_inode *inode,
1398                                   u64 start, u64 len, u32 extent_thresh,
1399                                   u64 newer_than, bool do_compress,
1400                                   bool locked, struct list_head *target_list,
1401                                   u64 *last_scanned_ret)
1402 {
1403         bool last_is_target = false;
1404         u64 cur = start;
1405         int ret = 0;
1406
1407         while (cur < start + len) {
1408                 struct extent_map *em;
1409                 struct defrag_target_range *new;
1410                 bool next_mergeable = true;
1411                 u64 range_len;
1412
1413                 last_is_target = false;
1414                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1415                                           newer_than, locked);
1416                 if (!em)
1417                         break;
1418
1419                 /*
1420                  * If the file extent is an inlined one, we may still want to
1421                  * defrag it (fallthrough) if it will cause a regular extent.
1422                  * This is for users who want to convert inline extents to
1423                  * regular ones through max_inline= mount option.
1424                  */
1425                 if (em->block_start == EXTENT_MAP_INLINE &&
1426                     em->len <= inode->root->fs_info->max_inline)
1427                         goto next;
1428
1429                 /* Skip hole/delalloc/preallocated extents */
1430                 if (em->block_start == EXTENT_MAP_HOLE ||
1431                     em->block_start == EXTENT_MAP_DELALLOC ||
1432                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1433                         goto next;
1434
1435                 /* Skip older extent */
1436                 if (em->generation < newer_than)
1437                         goto next;
1438
1439                 /* This em is under writeback, no need to defrag */
1440                 if (em->generation == (u64)-1)
1441                         goto next;
1442
1443                 /*
1444                  * Our start offset might be in the middle of an existing extent
1445                  * map, so take that into account.
1446                  */
1447                 range_len = em->len - (cur - em->start);
1448                 /*
1449                  * If this range of the extent map is already flagged for delalloc,
1450                  * skip it, because:
1451                  *
1452                  * 1) We could deadlock later, when trying to reserve space for
1453                  *    delalloc, because in case we can't immediately reserve space
1454                  *    the flusher can start delalloc and wait for the respective
1455                  *    ordered extents to complete. The deadlock would happen
1456                  *    because we do the space reservation while holding the range
1457                  *    locked, and starting writeback, or finishing an ordered
1458                  *    extent, requires locking the range;
1459                  *
1460                  * 2) If there's delalloc there, it means there's dirty pages for
1461                  *    which writeback has not started yet (we clean the delalloc
1462                  *    flag when starting writeback and after creating an ordered
1463                  *    extent). If we mark pages in an adjacent range for defrag,
1464                  *    then we will have a larger contiguous range for delalloc,
1465                  *    very likely resulting in a larger extent after writeback is
1466                  *    triggered (except in a case of free space fragmentation).
1467                  */
1468                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1469                                    EXTENT_DELALLOC, 0, NULL))
1470                         goto next;
1471
1472                 /*
1473                  * For do_compress case, we want to compress all valid file
1474                  * extents, thus no @extent_thresh or mergeable check.
1475                  */
1476                 if (do_compress)
1477                         goto add;
1478
1479                 /* Skip too large extent */
1480                 if (range_len >= extent_thresh)
1481                         goto next;
1482
1483                 /*
1484                  * Skip extents already at its max capacity, this is mostly for
1485                  * compressed extents, which max cap is only 128K.
1486                  */
1487                 if (em->len >= get_extent_max_capacity(em))
1488                         goto next;
1489
1490                 /*
1491                  * Normally there are no more extents after an inline one, thus
1492                  * @next_mergeable will normally be false and not defragged.
1493                  * So if an inline extent passed all above checks, just add it
1494                  * for defrag, and be converted to regular extents.
1495                  */
1496                 if (em->block_start == EXTENT_MAP_INLINE)
1497                         goto add;
1498
1499                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1500                                                 extent_thresh, newer_than, locked);
1501                 if (!next_mergeable) {
1502                         struct defrag_target_range *last;
1503
1504                         /* Empty target list, no way to merge with last entry */
1505                         if (list_empty(target_list))
1506                                 goto next;
1507                         last = list_entry(target_list->prev,
1508                                           struct defrag_target_range, list);
1509                         /* Not mergeable with last entry */
1510                         if (last->start + last->len != cur)
1511                                 goto next;
1512
1513                         /* Mergeable, fall through to add it to @target_list. */
1514                 }
1515
1516 add:
1517                 last_is_target = true;
1518                 range_len = min(extent_map_end(em), start + len) - cur;
1519                 /*
1520                  * This one is a good target, check if it can be merged into
1521                  * last range of the target list.
1522                  */
1523                 if (!list_empty(target_list)) {
1524                         struct defrag_target_range *last;
1525
1526                         last = list_entry(target_list->prev,
1527                                           struct defrag_target_range, list);
1528                         ASSERT(last->start + last->len <= cur);
1529                         if (last->start + last->len == cur) {
1530                                 /* Mergeable, enlarge the last entry */
1531                                 last->len += range_len;
1532                                 goto next;
1533                         }
1534                         /* Fall through to allocate a new entry */
1535                 }
1536
1537                 /* Allocate new defrag_target_range */
1538                 new = kmalloc(sizeof(*new), GFP_NOFS);
1539                 if (!new) {
1540                         free_extent_map(em);
1541                         ret = -ENOMEM;
1542                         break;
1543                 }
1544                 new->start = cur;
1545                 new->len = range_len;
1546                 list_add_tail(&new->list, target_list);
1547
1548 next:
1549                 cur = extent_map_end(em);
1550                 free_extent_map(em);
1551         }
1552         if (ret < 0) {
1553                 struct defrag_target_range *entry;
1554                 struct defrag_target_range *tmp;
1555
1556                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1557                         list_del_init(&entry->list);
1558                         kfree(entry);
1559                 }
1560         }
1561         if (!ret && last_scanned_ret) {
1562                 /*
1563                  * If the last extent is not a target, the caller can skip to
1564                  * the end of that extent.
1565                  * Otherwise, we can only go the end of the specified range.
1566                  */
1567                 if (!last_is_target)
1568                         *last_scanned_ret = max(cur, *last_scanned_ret);
1569                 else
1570                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1571         }
1572         return ret;
1573 }
1574
1575 #define CLUSTER_SIZE    (SZ_256K)
1576 static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1577
1578 /*
1579  * Defrag one contiguous target range.
1580  *
1581  * @inode:      target inode
1582  * @target:     target range to defrag
1583  * @pages:      locked pages covering the defrag range
1584  * @nr_pages:   number of locked pages
1585  *
1586  * Caller should ensure:
1587  *
1588  * - Pages are prepared
1589  *   Pages should be locked, no ordered extent in the pages range,
1590  *   no writeback.
1591  *
1592  * - Extent bits are locked
1593  */
1594 static int defrag_one_locked_target(struct btrfs_inode *inode,
1595                                     struct defrag_target_range *target,
1596                                     struct page **pages, int nr_pages,
1597                                     struct extent_state **cached_state)
1598 {
1599         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1600         struct extent_changeset *data_reserved = NULL;
1601         const u64 start = target->start;
1602         const u64 len = target->len;
1603         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1604         unsigned long start_index = start >> PAGE_SHIFT;
1605         unsigned long first_index = page_index(pages[0]);
1606         int ret = 0;
1607         int i;
1608
1609         ASSERT(last_index - first_index + 1 <= nr_pages);
1610
1611         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1612         if (ret < 0)
1613                 return ret;
1614         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1615                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1616                          EXTENT_DEFRAG, 0, 0, cached_state);
1617         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1618
1619         /* Update the page status */
1620         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1621                 ClearPageChecked(pages[i]);
1622                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1623         }
1624         btrfs_delalloc_release_extents(inode, len);
1625         extent_changeset_free(data_reserved);
1626
1627         return ret;
1628 }
1629
1630 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1631                             u32 extent_thresh, u64 newer_than, bool do_compress,
1632                             u64 *last_scanned_ret)
1633 {
1634         struct extent_state *cached_state = NULL;
1635         struct defrag_target_range *entry;
1636         struct defrag_target_range *tmp;
1637         LIST_HEAD(target_list);
1638         struct page **pages;
1639         const u32 sectorsize = inode->root->fs_info->sectorsize;
1640         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1641         u64 start_index = start >> PAGE_SHIFT;
1642         unsigned int nr_pages = last_index - start_index + 1;
1643         int ret = 0;
1644         int i;
1645
1646         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1647         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1648
1649         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1650         if (!pages)
1651                 return -ENOMEM;
1652
1653         /* Prepare all pages */
1654         for (i = 0; i < nr_pages; i++) {
1655                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1656                 if (IS_ERR(pages[i])) {
1657                         ret = PTR_ERR(pages[i]);
1658                         pages[i] = NULL;
1659                         goto free_pages;
1660                 }
1661         }
1662         for (i = 0; i < nr_pages; i++)
1663                 wait_on_page_writeback(pages[i]);
1664
1665         /* Lock the pages range */
1666         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1667                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1668                          &cached_state);
1669         /*
1670          * Now we have a consistent view about the extent map, re-check
1671          * which range really needs to be defragged.
1672          *
1673          * And this time we have extent locked already, pass @locked = true
1674          * so that we won't relock the extent range and cause deadlock.
1675          */
1676         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1677                                      newer_than, do_compress, true,
1678                                      &target_list, last_scanned_ret);
1679         if (ret < 0)
1680                 goto unlock_extent;
1681
1682         list_for_each_entry(entry, &target_list, list) {
1683                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1684                                                &cached_state);
1685                 if (ret < 0)
1686                         break;
1687         }
1688
1689         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1690                 list_del_init(&entry->list);
1691                 kfree(entry);
1692         }
1693 unlock_extent:
1694         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1695                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1696                              &cached_state);
1697 free_pages:
1698         for (i = 0; i < nr_pages; i++) {
1699                 if (pages[i]) {
1700                         unlock_page(pages[i]);
1701                         put_page(pages[i]);
1702                 }
1703         }
1704         kfree(pages);
1705         return ret;
1706 }
1707
1708 static int defrag_one_cluster(struct btrfs_inode *inode,
1709                               struct file_ra_state *ra,
1710                               u64 start, u32 len, u32 extent_thresh,
1711                               u64 newer_than, bool do_compress,
1712                               unsigned long *sectors_defragged,
1713                               unsigned long max_sectors,
1714                               u64 *last_scanned_ret)
1715 {
1716         const u32 sectorsize = inode->root->fs_info->sectorsize;
1717         struct defrag_target_range *entry;
1718         struct defrag_target_range *tmp;
1719         LIST_HEAD(target_list);
1720         int ret;
1721
1722         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1723                                      newer_than, do_compress, false,
1724                                      &target_list, NULL);
1725         if (ret < 0)
1726                 goto out;
1727
1728         list_for_each_entry(entry, &target_list, list) {
1729                 u32 range_len = entry->len;
1730
1731                 /* Reached or beyond the limit */
1732                 if (max_sectors && *sectors_defragged >= max_sectors) {
1733                         ret = 1;
1734                         break;
1735                 }
1736
1737                 if (max_sectors)
1738                         range_len = min_t(u32, range_len,
1739                                 (max_sectors - *sectors_defragged) * sectorsize);
1740
1741                 /*
1742                  * If defrag_one_range() has updated last_scanned_ret,
1743                  * our range may already be invalid (e.g. hole punched).
1744                  * Skip if our range is before last_scanned_ret, as there is
1745                  * no need to defrag the range anymore.
1746                  */
1747                 if (entry->start + range_len <= *last_scanned_ret)
1748                         continue;
1749
1750                 if (ra)
1751                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1752                                 ra, NULL, entry->start >> PAGE_SHIFT,
1753                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1754                                 (entry->start >> PAGE_SHIFT) + 1);
1755                 /*
1756                  * Here we may not defrag any range if holes are punched before
1757                  * we locked the pages.
1758                  * But that's fine, it only affects the @sectors_defragged
1759                  * accounting.
1760                  */
1761                 ret = defrag_one_range(inode, entry->start, range_len,
1762                                        extent_thresh, newer_than, do_compress,
1763                                        last_scanned_ret);
1764                 if (ret < 0)
1765                         break;
1766                 *sectors_defragged += range_len >>
1767                                       inode->root->fs_info->sectorsize_bits;
1768         }
1769 out:
1770         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1771                 list_del_init(&entry->list);
1772                 kfree(entry);
1773         }
1774         if (ret >= 0)
1775                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1776         return ret;
1777 }
1778
1779 /*
1780  * Entry point to file defragmentation.
1781  *
1782  * @inode:         inode to be defragged
1783  * @ra:            readahead state (can be NUL)
1784  * @range:         defrag options including range and flags
1785  * @newer_than:    minimum transid to defrag
1786  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1787  *                 will be defragged.
1788  *
1789  * Return <0 for error.
1790  * Return >=0 for the number of sectors defragged, and range->start will be updated
1791  * to indicate the file offset where next defrag should be started at.
1792  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1793  *  defragging all the range).
1794  */
1795 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1796                       struct btrfs_ioctl_defrag_range_args *range,
1797                       u64 newer_than, unsigned long max_to_defrag)
1798 {
1799         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1800         unsigned long sectors_defragged = 0;
1801         u64 isize = i_size_read(inode);
1802         u64 cur;
1803         u64 last_byte;
1804         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1805         bool ra_allocated = false;
1806         int compress_type = BTRFS_COMPRESS_ZLIB;
1807         int ret = 0;
1808         u32 extent_thresh = range->extent_thresh;
1809         pgoff_t start_index;
1810
1811         if (isize == 0)
1812                 return 0;
1813
1814         if (range->start >= isize)
1815                 return -EINVAL;
1816
1817         if (do_compress) {
1818                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1819                         return -EINVAL;
1820                 if (range->compress_type)
1821                         compress_type = range->compress_type;
1822         }
1823
1824         if (extent_thresh == 0)
1825                 extent_thresh = SZ_256K;
1826
1827         if (range->start + range->len > range->start) {
1828                 /* Got a specific range */
1829                 last_byte = min(isize, range->start + range->len);
1830         } else {
1831                 /* Defrag until file end */
1832                 last_byte = isize;
1833         }
1834
1835         /* Align the range */
1836         cur = round_down(range->start, fs_info->sectorsize);
1837         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1838
1839         /*
1840          * If we were not given a ra, allocate a readahead context. As
1841          * readahead is just an optimization, defrag will work without it so
1842          * we don't error out.
1843          */
1844         if (!ra) {
1845                 ra_allocated = true;
1846                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1847                 if (ra)
1848                         file_ra_state_init(ra, inode->i_mapping);
1849         }
1850
1851         /*
1852          * Make writeback start from the beginning of the range, so that the
1853          * defrag range can be written sequentially.
1854          */
1855         start_index = cur >> PAGE_SHIFT;
1856         if (start_index < inode->i_mapping->writeback_index)
1857                 inode->i_mapping->writeback_index = start_index;
1858
1859         while (cur < last_byte) {
1860                 const unsigned long prev_sectors_defragged = sectors_defragged;
1861                 u64 last_scanned = cur;
1862                 u64 cluster_end;
1863
1864                 if (btrfs_defrag_cancelled(fs_info)) {
1865                         ret = -EAGAIN;
1866                         break;
1867                 }
1868
1869                 /* We want the cluster end at page boundary when possible */
1870                 cluster_end = (((cur >> PAGE_SHIFT) +
1871                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1872                 cluster_end = min(cluster_end, last_byte);
1873
1874                 btrfs_inode_lock(inode, 0);
1875                 if (IS_SWAPFILE(inode)) {
1876                         ret = -ETXTBSY;
1877                         btrfs_inode_unlock(inode, 0);
1878                         break;
1879                 }
1880                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1881                         btrfs_inode_unlock(inode, 0);
1882                         break;
1883                 }
1884                 if (do_compress)
1885                         BTRFS_I(inode)->defrag_compress = compress_type;
1886                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1887                                 cluster_end + 1 - cur, extent_thresh,
1888                                 newer_than, do_compress, &sectors_defragged,
1889                                 max_to_defrag, &last_scanned);
1890
1891                 if (sectors_defragged > prev_sectors_defragged)
1892                         balance_dirty_pages_ratelimited(inode->i_mapping);
1893
1894                 btrfs_inode_unlock(inode, 0);
1895                 if (ret < 0)
1896                         break;
1897                 cur = max(cluster_end + 1, last_scanned);
1898                 if (ret > 0) {
1899                         ret = 0;
1900                         break;
1901                 }
1902                 cond_resched();
1903         }
1904
1905         if (ra_allocated)
1906                 kfree(ra);
1907         /*
1908          * Update range.start for autodefrag, this will indicate where to start
1909          * in next run.
1910          */
1911         range->start = cur;
1912         if (sectors_defragged) {
1913                 /*
1914                  * We have defragged some sectors, for compression case they
1915                  * need to be written back immediately.
1916                  */
1917                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1918                         filemap_flush(inode->i_mapping);
1919                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1920                                      &BTRFS_I(inode)->runtime_flags))
1921                                 filemap_flush(inode->i_mapping);
1922                 }
1923                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1924                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1925                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1926                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1927                 ret = sectors_defragged;
1928         }
1929         if (do_compress) {
1930                 btrfs_inode_lock(inode, 0);
1931                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1932                 btrfs_inode_unlock(inode, 0);
1933         }
1934         return ret;
1935 }
1936
1937 /*
1938  * Try to start exclusive operation @type or cancel it if it's running.
1939  *
1940  * Return:
1941  *   0        - normal mode, newly claimed op started
1942  *  >0        - normal mode, something else is running,
1943  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1944  * ECANCELED  - cancel mode, successful cancel
1945  * ENOTCONN   - cancel mode, operation not running anymore
1946  */
1947 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1948                         enum btrfs_exclusive_operation type, bool cancel)
1949 {
1950         if (!cancel) {
1951                 /* Start normal op */
1952                 if (!btrfs_exclop_start(fs_info, type))
1953                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1954                 /* Exclusive operation is now claimed */
1955                 return 0;
1956         }
1957
1958         /* Cancel running op */
1959         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1960                 /*
1961                  * This blocks any exclop finish from setting it to NONE, so we
1962                  * request cancellation. Either it runs and we will wait for it,
1963                  * or it has finished and no waiting will happen.
1964                  */
1965                 atomic_inc(&fs_info->reloc_cancel_req);
1966                 btrfs_exclop_start_unlock(fs_info);
1967
1968                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1969                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1970                                     TASK_INTERRUPTIBLE);
1971
1972                 return -ECANCELED;
1973         }
1974
1975         /* Something else is running or none */
1976         return -ENOTCONN;
1977 }
1978
1979 static noinline int btrfs_ioctl_resize(struct file *file,
1980                                         void __user *arg)
1981 {
1982         BTRFS_DEV_LOOKUP_ARGS(args);
1983         struct inode *inode = file_inode(file);
1984         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1985         u64 new_size;
1986         u64 old_size;
1987         u64 devid = 1;
1988         struct btrfs_root *root = BTRFS_I(inode)->root;
1989         struct btrfs_ioctl_vol_args *vol_args;
1990         struct btrfs_trans_handle *trans;
1991         struct btrfs_device *device = NULL;
1992         char *sizestr;
1993         char *retptr;
1994         char *devstr = NULL;
1995         int ret = 0;
1996         int mod = 0;
1997         bool cancel;
1998
1999         if (!capable(CAP_SYS_ADMIN))
2000                 return -EPERM;
2001
2002         ret = mnt_want_write_file(file);
2003         if (ret)
2004                 return ret;
2005
2006         /*
2007          * Read the arguments before checking exclusivity to be able to
2008          * distinguish regular resize and cancel
2009          */
2010         vol_args = memdup_user(arg, sizeof(*vol_args));
2011         if (IS_ERR(vol_args)) {
2012                 ret = PTR_ERR(vol_args);
2013                 goto out_drop;
2014         }
2015         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2016         sizestr = vol_args->name;
2017         cancel = (strcmp("cancel", sizestr) == 0);
2018         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
2019         if (ret)
2020                 goto out_free;
2021         /* Exclusive operation is now claimed */
2022
2023         devstr = strchr(sizestr, ':');
2024         if (devstr) {
2025                 sizestr = devstr + 1;
2026                 *devstr = '\0';
2027                 devstr = vol_args->name;
2028                 ret = kstrtoull(devstr, 10, &devid);
2029                 if (ret)
2030                         goto out_finish;
2031                 if (!devid) {
2032                         ret = -EINVAL;
2033                         goto out_finish;
2034                 }
2035                 btrfs_info(fs_info, "resizing devid %llu", devid);
2036         }
2037
2038         args.devid = devid;
2039         device = btrfs_find_device(fs_info->fs_devices, &args);
2040         if (!device) {
2041                 btrfs_info(fs_info, "resizer unable to find device %llu",
2042                            devid);
2043                 ret = -ENODEV;
2044                 goto out_finish;
2045         }
2046
2047         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2048                 btrfs_info(fs_info,
2049                            "resizer unable to apply on readonly device %llu",
2050                        devid);
2051                 ret = -EPERM;
2052                 goto out_finish;
2053         }
2054
2055         if (!strcmp(sizestr, "max"))
2056                 new_size = bdev_nr_bytes(device->bdev);
2057         else {
2058                 if (sizestr[0] == '-') {
2059                         mod = -1;
2060                         sizestr++;
2061                 } else if (sizestr[0] == '+') {
2062                         mod = 1;
2063                         sizestr++;
2064                 }
2065                 new_size = memparse(sizestr, &retptr);
2066                 if (*retptr != '\0' || new_size == 0) {
2067                         ret = -EINVAL;
2068                         goto out_finish;
2069                 }
2070         }
2071
2072         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2073                 ret = -EPERM;
2074                 goto out_finish;
2075         }
2076
2077         old_size = btrfs_device_get_total_bytes(device);
2078
2079         if (mod < 0) {
2080                 if (new_size > old_size) {
2081                         ret = -EINVAL;
2082                         goto out_finish;
2083                 }
2084                 new_size = old_size - new_size;
2085         } else if (mod > 0) {
2086                 if (new_size > ULLONG_MAX - old_size) {
2087                         ret = -ERANGE;
2088                         goto out_finish;
2089                 }
2090                 new_size = old_size + new_size;
2091         }
2092
2093         if (new_size < SZ_256M) {
2094                 ret = -EINVAL;
2095                 goto out_finish;
2096         }
2097         if (new_size > bdev_nr_bytes(device->bdev)) {
2098                 ret = -EFBIG;
2099                 goto out_finish;
2100         }
2101
2102         new_size = round_down(new_size, fs_info->sectorsize);
2103
2104         if (new_size > old_size) {
2105                 trans = btrfs_start_transaction(root, 0);
2106                 if (IS_ERR(trans)) {
2107                         ret = PTR_ERR(trans);
2108                         goto out_finish;
2109                 }
2110                 ret = btrfs_grow_device(trans, device, new_size);
2111                 btrfs_commit_transaction(trans);
2112         } else if (new_size < old_size) {
2113                 ret = btrfs_shrink_device(device, new_size);
2114         } /* equal, nothing need to do */
2115
2116         if (ret == 0 && new_size != old_size)
2117                 btrfs_info_in_rcu(fs_info,
2118                         "resize device %s (devid %llu) from %llu to %llu",
2119                         rcu_str_deref(device->name), device->devid,
2120                         old_size, new_size);
2121 out_finish:
2122         btrfs_exclop_finish(fs_info);
2123 out_free:
2124         kfree(vol_args);
2125 out_drop:
2126         mnt_drop_write_file(file);
2127         return ret;
2128 }
2129
2130 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2131                                 struct user_namespace *mnt_userns,
2132                                 const char *name, unsigned long fd, int subvol,
2133                                 bool readonly,
2134                                 struct btrfs_qgroup_inherit *inherit)
2135 {
2136         int namelen;
2137         int ret = 0;
2138
2139         if (!S_ISDIR(file_inode(file)->i_mode))
2140                 return -ENOTDIR;
2141
2142         ret = mnt_want_write_file(file);
2143         if (ret)
2144                 goto out;
2145
2146         namelen = strlen(name);
2147         if (strchr(name, '/')) {
2148                 ret = -EINVAL;
2149                 goto out_drop_write;
2150         }
2151
2152         if (name[0] == '.' &&
2153            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2154                 ret = -EEXIST;
2155                 goto out_drop_write;
2156         }
2157
2158         if (subvol) {
2159                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2160                                      namelen, NULL, readonly, inherit);
2161         } else {
2162                 struct fd src = fdget(fd);
2163                 struct inode *src_inode;
2164                 if (!src.file) {
2165                         ret = -EINVAL;
2166                         goto out_drop_write;
2167                 }
2168
2169                 src_inode = file_inode(src.file);
2170                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2171                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2172                                    "Snapshot src from another FS");
2173                         ret = -EXDEV;
2174                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2175                         /*
2176                          * Subvolume creation is not restricted, but snapshots
2177                          * are limited to own subvolumes only
2178                          */
2179                         ret = -EPERM;
2180                 } else {
2181                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2182                                                name, namelen,
2183                                                BTRFS_I(src_inode)->root,
2184                                                readonly, inherit);
2185                 }
2186                 fdput(src);
2187         }
2188 out_drop_write:
2189         mnt_drop_write_file(file);
2190 out:
2191         return ret;
2192 }
2193
2194 static noinline int btrfs_ioctl_snap_create(struct file *file,
2195                                             void __user *arg, int subvol)
2196 {
2197         struct btrfs_ioctl_vol_args *vol_args;
2198         int ret;
2199
2200         if (!S_ISDIR(file_inode(file)->i_mode))
2201                 return -ENOTDIR;
2202
2203         vol_args = memdup_user(arg, sizeof(*vol_args));
2204         if (IS_ERR(vol_args))
2205                 return PTR_ERR(vol_args);
2206         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2207
2208         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2209                                         vol_args->name, vol_args->fd, subvol,
2210                                         false, NULL);
2211
2212         kfree(vol_args);
2213         return ret;
2214 }
2215
2216 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2217                                                void __user *arg, int subvol)
2218 {
2219         struct btrfs_ioctl_vol_args_v2 *vol_args;
2220         int ret;
2221         bool readonly = false;
2222         struct btrfs_qgroup_inherit *inherit = NULL;
2223
2224         if (!S_ISDIR(file_inode(file)->i_mode))
2225                 return -ENOTDIR;
2226
2227         vol_args = memdup_user(arg, sizeof(*vol_args));
2228         if (IS_ERR(vol_args))
2229                 return PTR_ERR(vol_args);
2230         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2231
2232         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2233                 ret = -EOPNOTSUPP;
2234                 goto free_args;
2235         }
2236
2237         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2238                 readonly = true;
2239         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2240                 u64 nums;
2241
2242                 if (vol_args->size < sizeof(*inherit) ||
2243                     vol_args->size > PAGE_SIZE) {
2244                         ret = -EINVAL;
2245                         goto free_args;
2246                 }
2247                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2248                 if (IS_ERR(inherit)) {
2249                         ret = PTR_ERR(inherit);
2250                         goto free_args;
2251                 }
2252
2253                 if (inherit->num_qgroups > PAGE_SIZE ||
2254                     inherit->num_ref_copies > PAGE_SIZE ||
2255                     inherit->num_excl_copies > PAGE_SIZE) {
2256                         ret = -EINVAL;
2257                         goto free_inherit;
2258                 }
2259
2260                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2261                        2 * inherit->num_excl_copies;
2262                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2263                         ret = -EINVAL;
2264                         goto free_inherit;
2265                 }
2266         }
2267
2268         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2269                                         vol_args->name, vol_args->fd, subvol,
2270                                         readonly, inherit);
2271         if (ret)
2272                 goto free_inherit;
2273 free_inherit:
2274         kfree(inherit);
2275 free_args:
2276         kfree(vol_args);
2277         return ret;
2278 }
2279
2280 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
2281                                                 void __user *arg)
2282 {
2283         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2284         struct btrfs_root *root = BTRFS_I(inode)->root;
2285         int ret = 0;
2286         u64 flags = 0;
2287
2288         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2289                 return -EINVAL;
2290
2291         down_read(&fs_info->subvol_sem);
2292         if (btrfs_root_readonly(root))
2293                 flags |= BTRFS_SUBVOL_RDONLY;
2294         up_read(&fs_info->subvol_sem);
2295
2296         if (copy_to_user(arg, &flags, sizeof(flags)))
2297                 ret = -EFAULT;
2298
2299         return ret;
2300 }
2301
2302 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2303                                               void __user *arg)
2304 {
2305         struct inode *inode = file_inode(file);
2306         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2307         struct btrfs_root *root = BTRFS_I(inode)->root;
2308         struct btrfs_trans_handle *trans;
2309         u64 root_flags;
2310         u64 flags;
2311         int ret = 0;
2312
2313         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2314                 return -EPERM;
2315
2316         ret = mnt_want_write_file(file);
2317         if (ret)
2318                 goto out;
2319
2320         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2321                 ret = -EINVAL;
2322                 goto out_drop_write;
2323         }
2324
2325         if (copy_from_user(&flags, arg, sizeof(flags))) {
2326                 ret = -EFAULT;
2327                 goto out_drop_write;
2328         }
2329
2330         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2331                 ret = -EOPNOTSUPP;
2332                 goto out_drop_write;
2333         }
2334
2335         down_write(&fs_info->subvol_sem);
2336
2337         /* nothing to do */
2338         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2339                 goto out_drop_sem;
2340
2341         root_flags = btrfs_root_flags(&root->root_item);
2342         if (flags & BTRFS_SUBVOL_RDONLY) {
2343                 btrfs_set_root_flags(&root->root_item,
2344                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2345         } else {
2346                 /*
2347                  * Block RO -> RW transition if this subvolume is involved in
2348                  * send
2349                  */
2350                 spin_lock(&root->root_item_lock);
2351                 if (root->send_in_progress == 0) {
2352                         btrfs_set_root_flags(&root->root_item,
2353                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2354                         spin_unlock(&root->root_item_lock);
2355                 } else {
2356                         spin_unlock(&root->root_item_lock);
2357                         btrfs_warn(fs_info,
2358                                    "Attempt to set subvolume %llu read-write during send",
2359                                    root->root_key.objectid);
2360                         ret = -EPERM;
2361                         goto out_drop_sem;
2362                 }
2363         }
2364
2365         trans = btrfs_start_transaction(root, 1);
2366         if (IS_ERR(trans)) {
2367                 ret = PTR_ERR(trans);
2368                 goto out_reset;
2369         }
2370
2371         ret = btrfs_update_root(trans, fs_info->tree_root,
2372                                 &root->root_key, &root->root_item);
2373         if (ret < 0) {
2374                 btrfs_end_transaction(trans);
2375                 goto out_reset;
2376         }
2377
2378         ret = btrfs_commit_transaction(trans);
2379
2380 out_reset:
2381         if (ret)
2382                 btrfs_set_root_flags(&root->root_item, root_flags);
2383 out_drop_sem:
2384         up_write(&fs_info->subvol_sem);
2385 out_drop_write:
2386         mnt_drop_write_file(file);
2387 out:
2388         return ret;
2389 }
2390
2391 static noinline int key_in_sk(struct btrfs_key *key,
2392                               struct btrfs_ioctl_search_key *sk)
2393 {
2394         struct btrfs_key test;
2395         int ret;
2396
2397         test.objectid = sk->min_objectid;
2398         test.type = sk->min_type;
2399         test.offset = sk->min_offset;
2400
2401         ret = btrfs_comp_cpu_keys(key, &test);
2402         if (ret < 0)
2403                 return 0;
2404
2405         test.objectid = sk->max_objectid;
2406         test.type = sk->max_type;
2407         test.offset = sk->max_offset;
2408
2409         ret = btrfs_comp_cpu_keys(key, &test);
2410         if (ret > 0)
2411                 return 0;
2412         return 1;
2413 }
2414
2415 static noinline int copy_to_sk(struct btrfs_path *path,
2416                                struct btrfs_key *key,
2417                                struct btrfs_ioctl_search_key *sk,
2418                                size_t *buf_size,
2419                                char __user *ubuf,
2420                                unsigned long *sk_offset,
2421                                int *num_found)
2422 {
2423         u64 found_transid;
2424         struct extent_buffer *leaf;
2425         struct btrfs_ioctl_search_header sh;
2426         struct btrfs_key test;
2427         unsigned long item_off;
2428         unsigned long item_len;
2429         int nritems;
2430         int i;
2431         int slot;
2432         int ret = 0;
2433
2434         leaf = path->nodes[0];
2435         slot = path->slots[0];
2436         nritems = btrfs_header_nritems(leaf);
2437
2438         if (btrfs_header_generation(leaf) > sk->max_transid) {
2439                 i = nritems;
2440                 goto advance_key;
2441         }
2442         found_transid = btrfs_header_generation(leaf);
2443
2444         for (i = slot; i < nritems; i++) {
2445                 item_off = btrfs_item_ptr_offset(leaf, i);
2446                 item_len = btrfs_item_size(leaf, i);
2447
2448                 btrfs_item_key_to_cpu(leaf, key, i);
2449                 if (!key_in_sk(key, sk))
2450                         continue;
2451
2452                 if (sizeof(sh) + item_len > *buf_size) {
2453                         if (*num_found) {
2454                                 ret = 1;
2455                                 goto out;
2456                         }
2457
2458                         /*
2459                          * return one empty item back for v1, which does not
2460                          * handle -EOVERFLOW
2461                          */
2462
2463                         *buf_size = sizeof(sh) + item_len;
2464                         item_len = 0;
2465                         ret = -EOVERFLOW;
2466                 }
2467
2468                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2469                         ret = 1;
2470                         goto out;
2471                 }
2472
2473                 sh.objectid = key->objectid;
2474                 sh.offset = key->offset;
2475                 sh.type = key->type;
2476                 sh.len = item_len;
2477                 sh.transid = found_transid;
2478
2479                 /*
2480                  * Copy search result header. If we fault then loop again so we
2481                  * can fault in the pages and -EFAULT there if there's a
2482                  * problem. Otherwise we'll fault and then copy the buffer in
2483                  * properly this next time through
2484                  */
2485                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2486                         ret = 0;
2487                         goto out;
2488                 }
2489
2490                 *sk_offset += sizeof(sh);
2491
2492                 if (item_len) {
2493                         char __user *up = ubuf + *sk_offset;
2494                         /*
2495                          * Copy the item, same behavior as above, but reset the
2496                          * * sk_offset so we copy the full thing again.
2497                          */
2498                         if (read_extent_buffer_to_user_nofault(leaf, up,
2499                                                 item_off, item_len)) {
2500                                 ret = 0;
2501                                 *sk_offset -= sizeof(sh);
2502                                 goto out;
2503                         }
2504
2505                         *sk_offset += item_len;
2506                 }
2507                 (*num_found)++;
2508
2509                 if (ret) /* -EOVERFLOW from above */
2510                         goto out;
2511
2512                 if (*num_found >= sk->nr_items) {
2513                         ret = 1;
2514                         goto out;
2515                 }
2516         }
2517 advance_key:
2518         ret = 0;
2519         test.objectid = sk->max_objectid;
2520         test.type = sk->max_type;
2521         test.offset = sk->max_offset;
2522         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2523                 ret = 1;
2524         else if (key->offset < (u64)-1)
2525                 key->offset++;
2526         else if (key->type < (u8)-1) {
2527                 key->offset = 0;
2528                 key->type++;
2529         } else if (key->objectid < (u64)-1) {
2530                 key->offset = 0;
2531                 key->type = 0;
2532                 key->objectid++;
2533         } else
2534                 ret = 1;
2535 out:
2536         /*
2537          *  0: all items from this leaf copied, continue with next
2538          *  1: * more items can be copied, but unused buffer is too small
2539          *     * all items were found
2540          *     Either way, it will stops the loop which iterates to the next
2541          *     leaf
2542          *  -EOVERFLOW: item was to large for buffer
2543          *  -EFAULT: could not copy extent buffer back to userspace
2544          */
2545         return ret;
2546 }
2547
2548 static noinline int search_ioctl(struct inode *inode,
2549                                  struct btrfs_ioctl_search_key *sk,
2550                                  size_t *buf_size,
2551                                  char __user *ubuf)
2552 {
2553         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2554         struct btrfs_root *root;
2555         struct btrfs_key key;
2556         struct btrfs_path *path;
2557         int ret;
2558         int num_found = 0;
2559         unsigned long sk_offset = 0;
2560
2561         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2562                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2563                 return -EOVERFLOW;
2564         }
2565
2566         path = btrfs_alloc_path();
2567         if (!path)
2568                 return -ENOMEM;
2569
2570         if (sk->tree_id == 0) {
2571                 /* search the root of the inode that was passed */
2572                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2573         } else {
2574                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2575                 if (IS_ERR(root)) {
2576                         btrfs_free_path(path);
2577                         return PTR_ERR(root);
2578                 }
2579         }
2580
2581         key.objectid = sk->min_objectid;
2582         key.type = sk->min_type;
2583         key.offset = sk->min_offset;
2584
2585         while (1) {
2586                 ret = -EFAULT;
2587                 /*
2588                  * Ensure that the whole user buffer is faulted in at sub-page
2589                  * granularity, otherwise the loop may live-lock.
2590                  */
2591                 if (fault_in_subpage_writeable(ubuf + sk_offset,
2592                                                *buf_size - sk_offset))
2593                         break;
2594
2595                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2596                 if (ret != 0) {
2597                         if (ret > 0)
2598                                 ret = 0;
2599                         goto err;
2600                 }
2601                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2602                                  &sk_offset, &num_found);
2603                 btrfs_release_path(path);
2604                 if (ret)
2605                         break;
2606
2607         }
2608         if (ret > 0)
2609                 ret = 0;
2610 err:
2611         sk->nr_items = num_found;
2612         btrfs_put_root(root);
2613         btrfs_free_path(path);
2614         return ret;
2615 }
2616
2617 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
2618                                             void __user *argp)
2619 {
2620         struct btrfs_ioctl_search_args __user *uargs = argp;
2621         struct btrfs_ioctl_search_key sk;
2622         int ret;
2623         size_t buf_size;
2624
2625         if (!capable(CAP_SYS_ADMIN))
2626                 return -EPERM;
2627
2628         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2629                 return -EFAULT;
2630
2631         buf_size = sizeof(uargs->buf);
2632
2633         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2634
2635         /*
2636          * In the origin implementation an overflow is handled by returning a
2637          * search header with a len of zero, so reset ret.
2638          */
2639         if (ret == -EOVERFLOW)
2640                 ret = 0;
2641
2642         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2643                 ret = -EFAULT;
2644         return ret;
2645 }
2646
2647 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
2648                                                void __user *argp)
2649 {
2650         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
2651         struct btrfs_ioctl_search_args_v2 args;
2652         int ret;
2653         size_t buf_size;
2654         const size_t buf_limit = SZ_16M;
2655
2656         if (!capable(CAP_SYS_ADMIN))
2657                 return -EPERM;
2658
2659         /* copy search header and buffer size */
2660         if (copy_from_user(&args, uarg, sizeof(args)))
2661                 return -EFAULT;
2662
2663         buf_size = args.buf_size;
2664
2665         /* limit result size to 16MB */
2666         if (buf_size > buf_limit)
2667                 buf_size = buf_limit;
2668
2669         ret = search_ioctl(inode, &args.key, &buf_size,
2670                            (char __user *)(&uarg->buf[0]));
2671         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2672                 ret = -EFAULT;
2673         else if (ret == -EOVERFLOW &&
2674                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2675                 ret = -EFAULT;
2676
2677         return ret;
2678 }
2679
2680 /*
2681  * Search INODE_REFs to identify path name of 'dirid' directory
2682  * in a 'tree_id' tree. and sets path name to 'name'.
2683  */
2684 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2685                                 u64 tree_id, u64 dirid, char *name)
2686 {
2687         struct btrfs_root *root;
2688         struct btrfs_key key;
2689         char *ptr;
2690         int ret = -1;
2691         int slot;
2692         int len;
2693         int total_len = 0;
2694         struct btrfs_inode_ref *iref;
2695         struct extent_buffer *l;
2696         struct btrfs_path *path;
2697
2698         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2699                 name[0]='\0';
2700                 return 0;
2701         }
2702
2703         path = btrfs_alloc_path();
2704         if (!path)
2705                 return -ENOMEM;
2706
2707         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2708
2709         root = btrfs_get_fs_root(info, tree_id, true);
2710         if (IS_ERR(root)) {
2711                 ret = PTR_ERR(root);
2712                 root = NULL;
2713                 goto out;
2714         }
2715
2716         key.objectid = dirid;
2717         key.type = BTRFS_INODE_REF_KEY;
2718         key.offset = (u64)-1;
2719
2720         while (1) {
2721                 ret = btrfs_search_backwards(root, &key, path);
2722                 if (ret < 0)
2723                         goto out;
2724                 else if (ret > 0) {
2725                         ret = -ENOENT;
2726                         goto out;
2727                 }
2728
2729                 l = path->nodes[0];
2730                 slot = path->slots[0];
2731
2732                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2733                 len = btrfs_inode_ref_name_len(l, iref);
2734                 ptr -= len + 1;
2735                 total_len += len + 1;
2736                 if (ptr < name) {
2737                         ret = -ENAMETOOLONG;
2738                         goto out;
2739                 }
2740
2741                 *(ptr + len) = '/';
2742                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2743
2744                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2745                         break;
2746
2747                 btrfs_release_path(path);
2748                 key.objectid = key.offset;
2749                 key.offset = (u64)-1;
2750                 dirid = key.objectid;
2751         }
2752         memmove(name, ptr, total_len);
2753         name[total_len] = '\0';
2754         ret = 0;
2755 out:
2756         btrfs_put_root(root);
2757         btrfs_free_path(path);
2758         return ret;
2759 }
2760
2761 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2762                                 struct inode *inode,
2763                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2764 {
2765         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2766         struct super_block *sb = inode->i_sb;
2767         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2768         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2769         u64 dirid = args->dirid;
2770         unsigned long item_off;
2771         unsigned long item_len;
2772         struct btrfs_inode_ref *iref;
2773         struct btrfs_root_ref *rref;
2774         struct btrfs_root *root = NULL;
2775         struct btrfs_path *path;
2776         struct btrfs_key key, key2;
2777         struct extent_buffer *leaf;
2778         struct inode *temp_inode;
2779         char *ptr;
2780         int slot;
2781         int len;
2782         int total_len = 0;
2783         int ret;
2784
2785         path = btrfs_alloc_path();
2786         if (!path)
2787                 return -ENOMEM;
2788
2789         /*
2790          * If the bottom subvolume does not exist directly under upper_limit,
2791          * construct the path in from the bottom up.
2792          */
2793         if (dirid != upper_limit.objectid) {
2794                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2795
2796                 root = btrfs_get_fs_root(fs_info, treeid, true);
2797                 if (IS_ERR(root)) {
2798                         ret = PTR_ERR(root);
2799                         goto out;
2800                 }
2801
2802                 key.objectid = dirid;
2803                 key.type = BTRFS_INODE_REF_KEY;
2804                 key.offset = (u64)-1;
2805                 while (1) {
2806                         ret = btrfs_search_backwards(root, &key, path);
2807                         if (ret < 0)
2808                                 goto out_put;
2809                         else if (ret > 0) {
2810                                 ret = -ENOENT;
2811                                 goto out_put;
2812                         }
2813
2814                         leaf = path->nodes[0];
2815                         slot = path->slots[0];
2816
2817                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2818                         len = btrfs_inode_ref_name_len(leaf, iref);
2819                         ptr -= len + 1;
2820                         total_len += len + 1;
2821                         if (ptr < args->path) {
2822                                 ret = -ENAMETOOLONG;
2823                                 goto out_put;
2824                         }
2825
2826                         *(ptr + len) = '/';
2827                         read_extent_buffer(leaf, ptr,
2828                                         (unsigned long)(iref + 1), len);
2829
2830                         /* Check the read+exec permission of this directory */
2831                         ret = btrfs_previous_item(root, path, dirid,
2832                                                   BTRFS_INODE_ITEM_KEY);
2833                         if (ret < 0) {
2834                                 goto out_put;
2835                         } else if (ret > 0) {
2836                                 ret = -ENOENT;
2837                                 goto out_put;
2838                         }
2839
2840                         leaf = path->nodes[0];
2841                         slot = path->slots[0];
2842                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2843                         if (key2.objectid != dirid) {
2844                                 ret = -ENOENT;
2845                                 goto out_put;
2846                         }
2847
2848                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2849                         if (IS_ERR(temp_inode)) {
2850                                 ret = PTR_ERR(temp_inode);
2851                                 goto out_put;
2852                         }
2853                         ret = inode_permission(mnt_userns, temp_inode,
2854                                                MAY_READ | MAY_EXEC);
2855                         iput(temp_inode);
2856                         if (ret) {
2857                                 ret = -EACCES;
2858                                 goto out_put;
2859                         }
2860
2861                         if (key.offset == upper_limit.objectid)
2862                                 break;
2863                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2864                                 ret = -EACCES;
2865                                 goto out_put;
2866                         }
2867
2868                         btrfs_release_path(path);
2869                         key.objectid = key.offset;
2870                         key.offset = (u64)-1;
2871                         dirid = key.objectid;
2872                 }
2873
2874                 memmove(args->path, ptr, total_len);
2875                 args->path[total_len] = '\0';
2876                 btrfs_put_root(root);
2877                 root = NULL;
2878                 btrfs_release_path(path);
2879         }
2880
2881         /* Get the bottom subvolume's name from ROOT_REF */
2882         key.objectid = treeid;
2883         key.type = BTRFS_ROOT_REF_KEY;
2884         key.offset = args->treeid;
2885         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2886         if (ret < 0) {
2887                 goto out;
2888         } else if (ret > 0) {
2889                 ret = -ENOENT;
2890                 goto out;
2891         }
2892
2893         leaf = path->nodes[0];
2894         slot = path->slots[0];
2895         btrfs_item_key_to_cpu(leaf, &key, slot);
2896
2897         item_off = btrfs_item_ptr_offset(leaf, slot);
2898         item_len = btrfs_item_size(leaf, slot);
2899         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2900         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2901         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2902                 ret = -EINVAL;
2903                 goto out;
2904         }
2905
2906         /* Copy subvolume's name */
2907         item_off += sizeof(struct btrfs_root_ref);
2908         item_len -= sizeof(struct btrfs_root_ref);
2909         read_extent_buffer(leaf, args->name, item_off, item_len);
2910         args->name[item_len] = 0;
2911
2912 out_put:
2913         btrfs_put_root(root);
2914 out:
2915         btrfs_free_path(path);
2916         return ret;
2917 }
2918
2919 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2920                                            void __user *argp)
2921 {
2922         struct btrfs_ioctl_ino_lookup_args *args;
2923         int ret = 0;
2924
2925         args = memdup_user(argp, sizeof(*args));
2926         if (IS_ERR(args))
2927                 return PTR_ERR(args);
2928
2929         /*
2930          * Unprivileged query to obtain the containing subvolume root id. The
2931          * path is reset so it's consistent with btrfs_search_path_in_tree.
2932          */
2933         if (args->treeid == 0)
2934                 args->treeid = root->root_key.objectid;
2935
2936         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2937                 args->name[0] = 0;
2938                 goto out;
2939         }
2940
2941         if (!capable(CAP_SYS_ADMIN)) {
2942                 ret = -EPERM;
2943                 goto out;
2944         }
2945
2946         ret = btrfs_search_path_in_tree(root->fs_info,
2947                                         args->treeid, args->objectid,
2948                                         args->name);
2949
2950 out:
2951         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2952                 ret = -EFAULT;
2953
2954         kfree(args);
2955         return ret;
2956 }
2957
2958 /*
2959  * Version of ino_lookup ioctl (unprivileged)
2960  *
2961  * The main differences from ino_lookup ioctl are:
2962  *
2963  *   1. Read + Exec permission will be checked using inode_permission() during
2964  *      path construction. -EACCES will be returned in case of failure.
2965  *   2. Path construction will be stopped at the inode number which corresponds
2966  *      to the fd with which this ioctl is called. If constructed path does not
2967  *      exist under fd's inode, -EACCES will be returned.
2968  *   3. The name of bottom subvolume is also searched and filled.
2969  */
2970 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2971 {
2972         struct btrfs_ioctl_ino_lookup_user_args *args;
2973         struct inode *inode;
2974         int ret;
2975
2976         args = memdup_user(argp, sizeof(*args));
2977         if (IS_ERR(args))
2978                 return PTR_ERR(args);
2979
2980         inode = file_inode(file);
2981
2982         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2983             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2984                 /*
2985                  * The subvolume does not exist under fd with which this is
2986                  * called
2987                  */
2988                 kfree(args);
2989                 return -EACCES;
2990         }
2991
2992         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2993
2994         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2995                 ret = -EFAULT;
2996
2997         kfree(args);
2998         return ret;
2999 }
3000
3001 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
3002 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
3003 {
3004         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
3005         struct btrfs_fs_info *fs_info;
3006         struct btrfs_root *root;
3007         struct btrfs_path *path;
3008         struct btrfs_key key;
3009         struct btrfs_root_item *root_item;
3010         struct btrfs_root_ref *rref;
3011         struct extent_buffer *leaf;
3012         unsigned long item_off;
3013         unsigned long item_len;
3014         int slot;
3015         int ret = 0;
3016
3017         path = btrfs_alloc_path();
3018         if (!path)
3019                 return -ENOMEM;
3020
3021         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
3022         if (!subvol_info) {
3023                 btrfs_free_path(path);
3024                 return -ENOMEM;
3025         }
3026
3027         fs_info = BTRFS_I(inode)->root->fs_info;
3028
3029         /* Get root_item of inode's subvolume */
3030         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
3031         root = btrfs_get_fs_root(fs_info, key.objectid, true);
3032         if (IS_ERR(root)) {
3033                 ret = PTR_ERR(root);
3034                 goto out_free;
3035         }
3036         root_item = &root->root_item;
3037
3038         subvol_info->treeid = key.objectid;
3039
3040         subvol_info->generation = btrfs_root_generation(root_item);
3041         subvol_info->flags = btrfs_root_flags(root_item);
3042
3043         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3044         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3045                                                     BTRFS_UUID_SIZE);
3046         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3047                                                     BTRFS_UUID_SIZE);
3048
3049         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3050         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3051         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3052
3053         subvol_info->otransid = btrfs_root_otransid(root_item);
3054         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3055         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3056
3057         subvol_info->stransid = btrfs_root_stransid(root_item);
3058         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3059         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3060
3061         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3062         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3063         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3064
3065         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3066                 /* Search root tree for ROOT_BACKREF of this subvolume */
3067                 key.type = BTRFS_ROOT_BACKREF_KEY;
3068                 key.offset = 0;
3069                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3070                 if (ret < 0) {
3071                         goto out;
3072                 } else if (path->slots[0] >=
3073                            btrfs_header_nritems(path->nodes[0])) {
3074                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3075                         if (ret < 0) {
3076                                 goto out;
3077                         } else if (ret > 0) {
3078                                 ret = -EUCLEAN;
3079                                 goto out;
3080                         }
3081                 }
3082
3083                 leaf = path->nodes[0];
3084                 slot = path->slots[0];
3085                 btrfs_item_key_to_cpu(leaf, &key, slot);
3086                 if (key.objectid == subvol_info->treeid &&
3087                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3088                         subvol_info->parent_id = key.offset;
3089
3090                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3091                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3092
3093                         item_off = btrfs_item_ptr_offset(leaf, slot)
3094                                         + sizeof(struct btrfs_root_ref);
3095                         item_len = btrfs_item_size(leaf, slot)
3096                                         - sizeof(struct btrfs_root_ref);
3097                         read_extent_buffer(leaf, subvol_info->name,
3098                                            item_off, item_len);
3099                 } else {
3100                         ret = -ENOENT;
3101                         goto out;
3102                 }
3103         }
3104
3105         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3106                 ret = -EFAULT;
3107
3108 out:
3109         btrfs_put_root(root);
3110 out_free:
3111         btrfs_free_path(path);
3112         kfree(subvol_info);
3113         return ret;
3114 }
3115
3116 /*
3117  * Return ROOT_REF information of the subvolume containing this inode
3118  * except the subvolume name.
3119  */
3120 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3121                                           void __user *argp)
3122 {
3123         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3124         struct btrfs_root_ref *rref;
3125         struct btrfs_path *path;
3126         struct btrfs_key key;
3127         struct extent_buffer *leaf;
3128         u64 objectid;
3129         int slot;
3130         int ret;
3131         u8 found;
3132
3133         path = btrfs_alloc_path();
3134         if (!path)
3135                 return -ENOMEM;
3136
3137         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3138         if (IS_ERR(rootrefs)) {
3139                 btrfs_free_path(path);
3140                 return PTR_ERR(rootrefs);
3141         }
3142
3143         objectid = root->root_key.objectid;
3144         key.objectid = objectid;
3145         key.type = BTRFS_ROOT_REF_KEY;
3146         key.offset = rootrefs->min_treeid;
3147         found = 0;
3148
3149         root = root->fs_info->tree_root;
3150         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3151         if (ret < 0) {
3152                 goto out;
3153         } else if (path->slots[0] >=
3154                    btrfs_header_nritems(path->nodes[0])) {
3155                 ret = btrfs_next_leaf(root, path);
3156                 if (ret < 0) {
3157                         goto out;
3158                 } else if (ret > 0) {
3159                         ret = -EUCLEAN;
3160                         goto out;
3161                 }
3162         }
3163         while (1) {
3164                 leaf = path->nodes[0];
3165                 slot = path->slots[0];
3166
3167                 btrfs_item_key_to_cpu(leaf, &key, slot);
3168                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3169                         ret = 0;
3170                         goto out;
3171                 }
3172
3173                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3174                         ret = -EOVERFLOW;
3175                         goto out;
3176                 }
3177
3178                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3179                 rootrefs->rootref[found].treeid = key.offset;
3180                 rootrefs->rootref[found].dirid =
3181                                   btrfs_root_ref_dirid(leaf, rref);
3182                 found++;
3183
3184                 ret = btrfs_next_item(root, path);
3185                 if (ret < 0) {
3186                         goto out;
3187                 } else if (ret > 0) {
3188                         ret = -EUCLEAN;
3189                         goto out;
3190                 }
3191         }
3192
3193 out:
3194         if (!ret || ret == -EOVERFLOW) {
3195                 rootrefs->num_items = found;
3196                 /* update min_treeid for next search */
3197                 if (found)
3198                         rootrefs->min_treeid =
3199                                 rootrefs->rootref[found - 1].treeid + 1;
3200                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3201                         ret = -EFAULT;
3202         }
3203
3204         kfree(rootrefs);
3205         btrfs_free_path(path);
3206
3207         return ret;
3208 }
3209
3210 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3211                                              void __user *arg,
3212                                              bool destroy_v2)
3213 {
3214         struct dentry *parent = file->f_path.dentry;
3215         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3216         struct dentry *dentry;
3217         struct inode *dir = d_inode(parent);
3218         struct inode *inode;
3219         struct btrfs_root *root = BTRFS_I(dir)->root;
3220         struct btrfs_root *dest = NULL;
3221         struct btrfs_ioctl_vol_args *vol_args = NULL;
3222         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3223         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3224         char *subvol_name, *subvol_name_ptr = NULL;
3225         int subvol_namelen;
3226         int err = 0;
3227         bool destroy_parent = false;
3228
3229         /* We don't support snapshots with extent tree v2 yet. */
3230         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3231                 btrfs_err(fs_info,
3232                           "extent tree v2 doesn't support snapshot deletion yet");
3233                 return -EOPNOTSUPP;
3234         }
3235
3236         if (destroy_v2) {
3237                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3238                 if (IS_ERR(vol_args2))
3239                         return PTR_ERR(vol_args2);
3240
3241                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3242                         err = -EOPNOTSUPP;
3243                         goto out;
3244                 }
3245
3246                 /*
3247                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3248                  * name, same as v1 currently does.
3249                  */
3250                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3251                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3252                         subvol_name = vol_args2->name;
3253
3254                         err = mnt_want_write_file(file);
3255                         if (err)
3256                                 goto out;
3257                 } else {
3258                         struct inode *old_dir;
3259
3260                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3261                                 err = -EINVAL;
3262                                 goto out;
3263                         }
3264
3265                         err = mnt_want_write_file(file);
3266                         if (err)
3267                                 goto out;
3268
3269                         dentry = btrfs_get_dentry(fs_info->sb,
3270                                         BTRFS_FIRST_FREE_OBJECTID,
3271                                         vol_args2->subvolid, 0, 0);
3272                         if (IS_ERR(dentry)) {
3273                                 err = PTR_ERR(dentry);
3274                                 goto out_drop_write;
3275                         }
3276
3277                         /*
3278                          * Change the default parent since the subvolume being
3279                          * deleted can be outside of the current mount point.
3280                          */
3281                         parent = btrfs_get_parent(dentry);
3282
3283                         /*
3284                          * At this point dentry->d_name can point to '/' if the
3285                          * subvolume we want to destroy is outsite of the
3286                          * current mount point, so we need to release the
3287                          * current dentry and execute the lookup to return a new
3288                          * one with ->d_name pointing to the
3289                          * <mount point>/subvol_name.
3290                          */
3291                         dput(dentry);
3292                         if (IS_ERR(parent)) {
3293                                 err = PTR_ERR(parent);
3294                                 goto out_drop_write;
3295                         }
3296                         old_dir = dir;
3297                         dir = d_inode(parent);
3298
3299                         /*
3300                          * If v2 was used with SPEC_BY_ID, a new parent was
3301                          * allocated since the subvolume can be outside of the
3302                          * current mount point. Later on we need to release this
3303                          * new parent dentry.
3304                          */
3305                         destroy_parent = true;
3306
3307                         /*
3308                          * On idmapped mounts, deletion via subvolid is
3309                          * restricted to subvolumes that are immediate
3310                          * ancestors of the inode referenced by the file
3311                          * descriptor in the ioctl. Otherwise the idmapping
3312                          * could potentially be abused to delete subvolumes
3313                          * anywhere in the filesystem the user wouldn't be able
3314                          * to delete without an idmapped mount.
3315                          */
3316                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3317                                 err = -EOPNOTSUPP;
3318                                 goto free_parent;
3319                         }
3320
3321                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3322                                                 fs_info, vol_args2->subvolid);
3323                         if (IS_ERR(subvol_name_ptr)) {
3324                                 err = PTR_ERR(subvol_name_ptr);
3325                                 goto free_parent;
3326                         }
3327                         /* subvol_name_ptr is already nul terminated */
3328                         subvol_name = (char *)kbasename(subvol_name_ptr);
3329                 }
3330         } else {
3331                 vol_args = memdup_user(arg, sizeof(*vol_args));
3332                 if (IS_ERR(vol_args))
3333                         return PTR_ERR(vol_args);
3334
3335                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3336                 subvol_name = vol_args->name;
3337
3338                 err = mnt_want_write_file(file);
3339                 if (err)
3340                         goto out;
3341         }
3342
3343         subvol_namelen = strlen(subvol_name);
3344
3345         if (strchr(subvol_name, '/') ||
3346             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3347                 err = -EINVAL;
3348                 goto free_subvol_name;
3349         }
3350
3351         if (!S_ISDIR(dir->i_mode)) {
3352                 err = -ENOTDIR;
3353                 goto free_subvol_name;
3354         }
3355
3356         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3357         if (err == -EINTR)
3358                 goto free_subvol_name;
3359         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3360         if (IS_ERR(dentry)) {
3361                 err = PTR_ERR(dentry);
3362                 goto out_unlock_dir;
3363         }
3364
3365         if (d_really_is_negative(dentry)) {
3366                 err = -ENOENT;
3367                 goto out_dput;
3368         }
3369
3370         inode = d_inode(dentry);
3371         dest = BTRFS_I(inode)->root;
3372         if (!capable(CAP_SYS_ADMIN)) {
3373                 /*
3374                  * Regular user.  Only allow this with a special mount
3375                  * option, when the user has write+exec access to the
3376                  * subvol root, and when rmdir(2) would have been
3377                  * allowed.
3378                  *
3379                  * Note that this is _not_ check that the subvol is
3380                  * empty or doesn't contain data that we wouldn't
3381                  * otherwise be able to delete.
3382                  *
3383                  * Users who want to delete empty subvols should try
3384                  * rmdir(2).
3385                  */
3386                 err = -EPERM;
3387                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3388                         goto out_dput;
3389
3390                 /*
3391                  * Do not allow deletion if the parent dir is the same
3392                  * as the dir to be deleted.  That means the ioctl
3393                  * must be called on the dentry referencing the root
3394                  * of the subvol, not a random directory contained
3395                  * within it.
3396                  */
3397                 err = -EINVAL;
3398                 if (root == dest)
3399                         goto out_dput;
3400
3401                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3402                 if (err)
3403                         goto out_dput;
3404         }
3405
3406         /* check if subvolume may be deleted by a user */
3407         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3408         if (err)
3409                 goto out_dput;
3410
3411         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3412                 err = -EINVAL;
3413                 goto out_dput;
3414         }
3415
3416         btrfs_inode_lock(inode, 0);
3417         err = btrfs_delete_subvolume(dir, dentry);
3418         btrfs_inode_unlock(inode, 0);
3419         if (!err)
3420                 d_delete_notify(dir, dentry);
3421
3422 out_dput:
3423         dput(dentry);
3424 out_unlock_dir:
3425         btrfs_inode_unlock(dir, 0);
3426 free_subvol_name:
3427         kfree(subvol_name_ptr);
3428 free_parent:
3429         if (destroy_parent)
3430                 dput(parent);
3431 out_drop_write:
3432         mnt_drop_write_file(file);
3433 out:
3434         kfree(vol_args2);
3435         kfree(vol_args);
3436         return err;
3437 }
3438
3439 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3440 {
3441         struct inode *inode = file_inode(file);
3442         struct btrfs_root *root = BTRFS_I(inode)->root;
3443         struct btrfs_ioctl_defrag_range_args range = {0};
3444         int ret;
3445
3446         ret = mnt_want_write_file(file);
3447         if (ret)
3448                 return ret;
3449
3450         if (btrfs_root_readonly(root)) {
3451                 ret = -EROFS;
3452                 goto out;
3453         }
3454
3455         switch (inode->i_mode & S_IFMT) {
3456         case S_IFDIR:
3457                 if (!capable(CAP_SYS_ADMIN)) {
3458                         ret = -EPERM;
3459                         goto out;
3460                 }
3461                 ret = btrfs_defrag_root(root);
3462                 break;
3463         case S_IFREG:
3464                 /*
3465                  * Note that this does not check the file descriptor for write
3466                  * access. This prevents defragmenting executables that are
3467                  * running and allows defrag on files open in read-only mode.
3468                  */
3469                 if (!capable(CAP_SYS_ADMIN) &&
3470                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3471                         ret = -EPERM;
3472                         goto out;
3473                 }
3474
3475                 if (argp) {
3476                         if (copy_from_user(&range, argp, sizeof(range))) {
3477                                 ret = -EFAULT;
3478                                 goto out;
3479                         }
3480                         /* compression requires us to start the IO */
3481                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3482                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3483                                 range.extent_thresh = (u32)-1;
3484                         }
3485                 } else {
3486                         /* the rest are all set to zero by kzalloc */
3487                         range.len = (u64)-1;
3488                 }
3489                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3490                                         &range, BTRFS_OLDEST_GENERATION, 0);
3491                 if (ret > 0)
3492                         ret = 0;
3493                 break;
3494         default:
3495                 ret = -EINVAL;
3496         }
3497 out:
3498         mnt_drop_write_file(file);
3499         return ret;
3500 }
3501
3502 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3503 {
3504         struct btrfs_ioctl_vol_args *vol_args;
3505         bool restore_op = false;
3506         int ret;
3507
3508         if (!capable(CAP_SYS_ADMIN))
3509                 return -EPERM;
3510
3511         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3512                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3513                 return -EINVAL;
3514         }
3515
3516         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3517                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3518                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3519
3520                 /*
3521                  * We can do the device add because we have a paused balanced,
3522                  * change the exclusive op type and remember we should bring
3523                  * back the paused balance
3524                  */
3525                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3526                 btrfs_exclop_start_unlock(fs_info);
3527                 restore_op = true;
3528         }
3529
3530         vol_args = memdup_user(arg, sizeof(*vol_args));
3531         if (IS_ERR(vol_args)) {
3532                 ret = PTR_ERR(vol_args);
3533                 goto out;
3534         }
3535
3536         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3537         ret = btrfs_init_new_device(fs_info, vol_args->name);
3538
3539         if (!ret)
3540                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3541
3542         kfree(vol_args);
3543 out:
3544         if (restore_op)
3545                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3546         else
3547                 btrfs_exclop_finish(fs_info);
3548         return ret;
3549 }
3550
3551 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3552 {
3553         BTRFS_DEV_LOOKUP_ARGS(args);
3554         struct inode *inode = file_inode(file);
3555         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3556         struct btrfs_ioctl_vol_args_v2 *vol_args;
3557         struct block_device *bdev = NULL;
3558         fmode_t mode;
3559         int ret;
3560         bool cancel = false;
3561
3562         if (!capable(CAP_SYS_ADMIN))
3563                 return -EPERM;
3564
3565         vol_args = memdup_user(arg, sizeof(*vol_args));
3566         if (IS_ERR(vol_args))
3567                 return PTR_ERR(vol_args);
3568
3569         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3570                 ret = -EOPNOTSUPP;
3571                 goto out;
3572         }
3573
3574         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3575         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3576                 args.devid = vol_args->devid;
3577         } else if (!strcmp("cancel", vol_args->name)) {
3578                 cancel = true;
3579         } else {
3580                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3581                 if (ret)
3582                         goto out;
3583         }
3584
3585         ret = mnt_want_write_file(file);
3586         if (ret)
3587                 goto out;
3588
3589         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3590                                            cancel);
3591         if (ret)
3592                 goto err_drop;
3593
3594         /* Exclusive operation is now claimed */
3595         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3596
3597         btrfs_exclop_finish(fs_info);
3598
3599         if (!ret) {
3600                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3601                         btrfs_info(fs_info, "device deleted: id %llu",
3602                                         vol_args->devid);
3603                 else
3604                         btrfs_info(fs_info, "device deleted: %s",
3605                                         vol_args->name);
3606         }
3607 err_drop:
3608         mnt_drop_write_file(file);
3609         if (bdev)
3610                 blkdev_put(bdev, mode);
3611 out:
3612         btrfs_put_dev_args_from_path(&args);
3613         kfree(vol_args);
3614         return ret;
3615 }
3616
3617 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3618 {
3619         BTRFS_DEV_LOOKUP_ARGS(args);
3620         struct inode *inode = file_inode(file);
3621         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3622         struct btrfs_ioctl_vol_args *vol_args;
3623         struct block_device *bdev = NULL;
3624         fmode_t mode;
3625         int ret;
3626         bool cancel = false;
3627
3628         if (!capable(CAP_SYS_ADMIN))
3629                 return -EPERM;
3630
3631         vol_args = memdup_user(arg, sizeof(*vol_args));
3632         if (IS_ERR(vol_args))
3633                 return PTR_ERR(vol_args);
3634
3635         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3636         if (!strcmp("cancel", vol_args->name)) {
3637                 cancel = true;
3638         } else {
3639                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3640                 if (ret)
3641                         goto out;
3642         }
3643
3644         ret = mnt_want_write_file(file);
3645         if (ret)
3646                 goto out;
3647
3648         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3649                                            cancel);
3650         if (ret == 0) {
3651                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3652                 if (!ret)
3653                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3654                 btrfs_exclop_finish(fs_info);
3655         }
3656
3657         mnt_drop_write_file(file);
3658         if (bdev)
3659                 blkdev_put(bdev, mode);
3660 out:
3661         btrfs_put_dev_args_from_path(&args);
3662         kfree(vol_args);
3663         return ret;
3664 }
3665
3666 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3667                                 void __user *arg)
3668 {
3669         struct btrfs_ioctl_fs_info_args *fi_args;
3670         struct btrfs_device *device;
3671         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3672         u64 flags_in;
3673         int ret = 0;
3674
3675         fi_args = memdup_user(arg, sizeof(*fi_args));
3676         if (IS_ERR(fi_args))
3677                 return PTR_ERR(fi_args);
3678
3679         flags_in = fi_args->flags;
3680         memset(fi_args, 0, sizeof(*fi_args));
3681
3682         rcu_read_lock();
3683         fi_args->num_devices = fs_devices->num_devices;
3684
3685         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3686                 if (device->devid > fi_args->max_id)
3687                         fi_args->max_id = device->devid;
3688         }
3689         rcu_read_unlock();
3690
3691         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3692         fi_args->nodesize = fs_info->nodesize;
3693         fi_args->sectorsize = fs_info->sectorsize;
3694         fi_args->clone_alignment = fs_info->sectorsize;
3695
3696         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3697                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3698                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3699                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3700         }
3701
3702         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3703                 fi_args->generation = fs_info->generation;
3704                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3705         }
3706
3707         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3708                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3709                        sizeof(fi_args->metadata_uuid));
3710                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3711         }
3712
3713         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3714                 ret = -EFAULT;
3715
3716         kfree(fi_args);
3717         return ret;
3718 }
3719
3720 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3721                                  void __user *arg)
3722 {
3723         BTRFS_DEV_LOOKUP_ARGS(args);
3724         struct btrfs_ioctl_dev_info_args *di_args;
3725         struct btrfs_device *dev;
3726         int ret = 0;
3727
3728         di_args = memdup_user(arg, sizeof(*di_args));
3729         if (IS_ERR(di_args))
3730                 return PTR_ERR(di_args);
3731
3732         args.devid = di_args->devid;
3733         if (!btrfs_is_empty_uuid(di_args->uuid))
3734                 args.uuid = di_args->uuid;
3735
3736         rcu_read_lock();
3737         dev = btrfs_find_device(fs_info->fs_devices, &args);
3738         if (!dev) {
3739                 ret = -ENODEV;
3740                 goto out;
3741         }
3742
3743         di_args->devid = dev->devid;
3744         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3745         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3746         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3747         if (dev->name) {
3748                 strncpy(di_args->path, rcu_str_deref(dev->name),
3749                                 sizeof(di_args->path) - 1);
3750                 di_args->path[sizeof(di_args->path) - 1] = 0;
3751         } else {
3752                 di_args->path[0] = '\0';
3753         }
3754
3755 out:
3756         rcu_read_unlock();
3757         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3758                 ret = -EFAULT;
3759
3760         kfree(di_args);
3761         return ret;
3762 }
3763
3764 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3765 {
3766         struct inode *inode = file_inode(file);
3767         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3768         struct btrfs_root *root = BTRFS_I(inode)->root;
3769         struct btrfs_root *new_root;
3770         struct btrfs_dir_item *di;
3771         struct btrfs_trans_handle *trans;
3772         struct btrfs_path *path = NULL;
3773         struct btrfs_disk_key disk_key;
3774         u64 objectid = 0;
3775         u64 dir_id;
3776         int ret;
3777
3778         if (!capable(CAP_SYS_ADMIN))
3779                 return -EPERM;
3780
3781         ret = mnt_want_write_file(file);
3782         if (ret)
3783                 return ret;
3784
3785         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3786                 ret = -EFAULT;
3787                 goto out;
3788         }
3789
3790         if (!objectid)
3791                 objectid = BTRFS_FS_TREE_OBJECTID;
3792
3793         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3794         if (IS_ERR(new_root)) {
3795                 ret = PTR_ERR(new_root);
3796                 goto out;
3797         }
3798         if (!is_fstree(new_root->root_key.objectid)) {
3799                 ret = -ENOENT;
3800                 goto out_free;
3801         }
3802
3803         path = btrfs_alloc_path();
3804         if (!path) {
3805                 ret = -ENOMEM;
3806                 goto out_free;
3807         }
3808
3809         trans = btrfs_start_transaction(root, 1);
3810         if (IS_ERR(trans)) {
3811                 ret = PTR_ERR(trans);
3812                 goto out_free;
3813         }
3814
3815         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3816         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3817                                    dir_id, "default", 7, 1);
3818         if (IS_ERR_OR_NULL(di)) {
3819                 btrfs_release_path(path);
3820                 btrfs_end_transaction(trans);
3821                 btrfs_err(fs_info,
3822                           "Umm, you don't have the default diritem, this isn't going to work");
3823                 ret = -ENOENT;
3824                 goto out_free;
3825         }
3826
3827         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3828         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3829         btrfs_mark_buffer_dirty(path->nodes[0]);
3830         btrfs_release_path(path);
3831
3832         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3833         btrfs_end_transaction(trans);
3834 out_free:
3835         btrfs_put_root(new_root);
3836         btrfs_free_path(path);
3837 out:
3838         mnt_drop_write_file(file);
3839         return ret;
3840 }
3841
3842 static void get_block_group_info(struct list_head *groups_list,
3843                                  struct btrfs_ioctl_space_info *space)
3844 {
3845         struct btrfs_block_group *block_group;
3846
3847         space->total_bytes = 0;
3848         space->used_bytes = 0;
3849         space->flags = 0;
3850         list_for_each_entry(block_group, groups_list, list) {
3851                 space->flags = block_group->flags;
3852                 space->total_bytes += block_group->length;
3853                 space->used_bytes += block_group->used;
3854         }
3855 }
3856
3857 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3858                                    void __user *arg)
3859 {
3860         struct btrfs_ioctl_space_args space_args;
3861         struct btrfs_ioctl_space_info space;
3862         struct btrfs_ioctl_space_info *dest;
3863         struct btrfs_ioctl_space_info *dest_orig;
3864         struct btrfs_ioctl_space_info __user *user_dest;
3865         struct btrfs_space_info *info;
3866         static const u64 types[] = {
3867                 BTRFS_BLOCK_GROUP_DATA,
3868                 BTRFS_BLOCK_GROUP_SYSTEM,
3869                 BTRFS_BLOCK_GROUP_METADATA,
3870                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3871         };
3872         int num_types = 4;
3873         int alloc_size;
3874         int ret = 0;
3875         u64 slot_count = 0;
3876         int i, c;
3877
3878         if (copy_from_user(&space_args,
3879                            (struct btrfs_ioctl_space_args __user *)arg,
3880                            sizeof(space_args)))
3881                 return -EFAULT;
3882
3883         for (i = 0; i < num_types; i++) {
3884                 struct btrfs_space_info *tmp;
3885
3886                 info = NULL;
3887                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3888                         if (tmp->flags == types[i]) {
3889                                 info = tmp;
3890                                 break;
3891                         }
3892                 }
3893
3894                 if (!info)
3895                         continue;
3896
3897                 down_read(&info->groups_sem);
3898                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3899                         if (!list_empty(&info->block_groups[c]))
3900                                 slot_count++;
3901                 }
3902                 up_read(&info->groups_sem);
3903         }
3904
3905         /*
3906          * Global block reserve, exported as a space_info
3907          */
3908         slot_count++;
3909
3910         /* space_slots == 0 means they are asking for a count */
3911         if (space_args.space_slots == 0) {
3912                 space_args.total_spaces = slot_count;
3913                 goto out;
3914         }
3915
3916         slot_count = min_t(u64, space_args.space_slots, slot_count);
3917
3918         alloc_size = sizeof(*dest) * slot_count;
3919
3920         /* we generally have at most 6 or so space infos, one for each raid
3921          * level.  So, a whole page should be more than enough for everyone
3922          */
3923         if (alloc_size > PAGE_SIZE)
3924                 return -ENOMEM;
3925
3926         space_args.total_spaces = 0;
3927         dest = kmalloc(alloc_size, GFP_KERNEL);
3928         if (!dest)
3929                 return -ENOMEM;
3930         dest_orig = dest;
3931
3932         /* now we have a buffer to copy into */
3933         for (i = 0; i < num_types; i++) {
3934                 struct btrfs_space_info *tmp;
3935
3936                 if (!slot_count)
3937                         break;
3938
3939                 info = NULL;
3940                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3941                         if (tmp->flags == types[i]) {
3942                                 info = tmp;
3943                                 break;
3944                         }
3945                 }
3946
3947                 if (!info)
3948                         continue;
3949                 down_read(&info->groups_sem);
3950                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3951                         if (!list_empty(&info->block_groups[c])) {
3952                                 get_block_group_info(&info->block_groups[c],
3953                                                      &space);
3954                                 memcpy(dest, &space, sizeof(space));
3955                                 dest++;
3956                                 space_args.total_spaces++;
3957                                 slot_count--;
3958                         }
3959                         if (!slot_count)
3960                                 break;
3961                 }
3962                 up_read(&info->groups_sem);
3963         }
3964
3965         /*
3966          * Add global block reserve
3967          */
3968         if (slot_count) {
3969                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3970
3971                 spin_lock(&block_rsv->lock);
3972                 space.total_bytes = block_rsv->size;
3973                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3974                 spin_unlock(&block_rsv->lock);
3975                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3976                 memcpy(dest, &space, sizeof(space));
3977                 space_args.total_spaces++;
3978         }
3979
3980         user_dest = (struct btrfs_ioctl_space_info __user *)
3981                 (arg + sizeof(struct btrfs_ioctl_space_args));
3982
3983         if (copy_to_user(user_dest, dest_orig, alloc_size))
3984                 ret = -EFAULT;
3985
3986         kfree(dest_orig);
3987 out:
3988         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3989                 ret = -EFAULT;
3990
3991         return ret;
3992 }
3993
3994 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3995                                             void __user *argp)
3996 {
3997         struct btrfs_trans_handle *trans;
3998         u64 transid;
3999
4000         trans = btrfs_attach_transaction_barrier(root);
4001         if (IS_ERR(trans)) {
4002                 if (PTR_ERR(trans) != -ENOENT)
4003                         return PTR_ERR(trans);
4004
4005                 /* No running transaction, don't bother */
4006                 transid = root->fs_info->last_trans_committed;
4007                 goto out;
4008         }
4009         transid = trans->transid;
4010         btrfs_commit_transaction_async(trans);
4011 out:
4012         if (argp)
4013                 if (copy_to_user(argp, &transid, sizeof(transid)))
4014                         return -EFAULT;
4015         return 0;
4016 }
4017
4018 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4019                                            void __user *argp)
4020 {
4021         u64 transid;
4022
4023         if (argp) {
4024                 if (copy_from_user(&transid, argp, sizeof(transid)))
4025                         return -EFAULT;
4026         } else {
4027                 transid = 0;  /* current trans */
4028         }
4029         return btrfs_wait_for_commit(fs_info, transid);
4030 }
4031
4032 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4033 {
4034         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4035         struct btrfs_ioctl_scrub_args *sa;
4036         int ret;
4037
4038         if (!capable(CAP_SYS_ADMIN))
4039                 return -EPERM;
4040
4041         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4042                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4043                 return -EINVAL;
4044         }
4045
4046         sa = memdup_user(arg, sizeof(*sa));
4047         if (IS_ERR(sa))
4048                 return PTR_ERR(sa);
4049
4050         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4051                 ret = mnt_want_write_file(file);
4052                 if (ret)
4053                         goto out;
4054         }
4055
4056         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4057                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4058                               0);
4059
4060         /*
4061          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4062          * error. This is important as it allows user space to know how much
4063          * progress scrub has done. For example, if scrub is canceled we get
4064          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4065          * space. Later user space can inspect the progress from the structure
4066          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4067          * previously (btrfs-progs does this).
4068          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4069          * then return -EFAULT to signal the structure was not copied or it may
4070          * be corrupt and unreliable due to a partial copy.
4071          */
4072         if (copy_to_user(arg, sa, sizeof(*sa)))
4073                 ret = -EFAULT;
4074
4075         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4076                 mnt_drop_write_file(file);
4077 out:
4078         kfree(sa);
4079         return ret;
4080 }
4081
4082 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4083 {
4084         if (!capable(CAP_SYS_ADMIN))
4085                 return -EPERM;
4086
4087         return btrfs_scrub_cancel(fs_info);
4088 }
4089
4090 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4091                                        void __user *arg)
4092 {
4093         struct btrfs_ioctl_scrub_args *sa;
4094         int ret;
4095
4096         if (!capable(CAP_SYS_ADMIN))
4097                 return -EPERM;
4098
4099         sa = memdup_user(arg, sizeof(*sa));
4100         if (IS_ERR(sa))
4101                 return PTR_ERR(sa);
4102
4103         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4104
4105         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4106                 ret = -EFAULT;
4107
4108         kfree(sa);
4109         return ret;
4110 }
4111
4112 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4113                                       void __user *arg)
4114 {
4115         struct btrfs_ioctl_get_dev_stats *sa;
4116         int ret;
4117
4118         sa = memdup_user(arg, sizeof(*sa));
4119         if (IS_ERR(sa))
4120                 return PTR_ERR(sa);
4121
4122         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4123                 kfree(sa);
4124                 return -EPERM;
4125         }
4126
4127         ret = btrfs_get_dev_stats(fs_info, sa);
4128
4129         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4130                 ret = -EFAULT;
4131
4132         kfree(sa);
4133         return ret;
4134 }
4135
4136 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4137                                     void __user *arg)
4138 {
4139         struct btrfs_ioctl_dev_replace_args *p;
4140         int ret;
4141
4142         if (!capable(CAP_SYS_ADMIN))
4143                 return -EPERM;
4144
4145         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4146                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4147                 return -EINVAL;
4148         }
4149
4150         p = memdup_user(arg, sizeof(*p));
4151         if (IS_ERR(p))
4152                 return PTR_ERR(p);
4153
4154         switch (p->cmd) {
4155         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4156                 if (sb_rdonly(fs_info->sb)) {
4157                         ret = -EROFS;
4158                         goto out;
4159                 }
4160                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4161                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4162                 } else {
4163                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4164                         btrfs_exclop_finish(fs_info);
4165                 }
4166                 break;
4167         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4168                 btrfs_dev_replace_status(fs_info, p);
4169                 ret = 0;
4170                 break;
4171         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4172                 p->result = btrfs_dev_replace_cancel(fs_info);
4173                 ret = 0;
4174                 break;
4175         default:
4176                 ret = -EINVAL;
4177                 break;
4178         }
4179
4180         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4181                 ret = -EFAULT;
4182 out:
4183         kfree(p);
4184         return ret;
4185 }
4186
4187 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4188 {
4189         int ret = 0;
4190         int i;
4191         u64 rel_ptr;
4192         int size;
4193         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4194         struct inode_fs_paths *ipath = NULL;
4195         struct btrfs_path *path;
4196
4197         if (!capable(CAP_DAC_READ_SEARCH))
4198                 return -EPERM;
4199
4200         path = btrfs_alloc_path();
4201         if (!path) {
4202                 ret = -ENOMEM;
4203                 goto out;
4204         }
4205
4206         ipa = memdup_user(arg, sizeof(*ipa));
4207         if (IS_ERR(ipa)) {
4208                 ret = PTR_ERR(ipa);
4209                 ipa = NULL;
4210                 goto out;
4211         }
4212
4213         size = min_t(u32, ipa->size, 4096);
4214         ipath = init_ipath(size, root, path);
4215         if (IS_ERR(ipath)) {
4216                 ret = PTR_ERR(ipath);
4217                 ipath = NULL;
4218                 goto out;
4219         }
4220
4221         ret = paths_from_inode(ipa->inum, ipath);
4222         if (ret < 0)
4223                 goto out;
4224
4225         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4226                 rel_ptr = ipath->fspath->val[i] -
4227                           (u64)(unsigned long)ipath->fspath->val;
4228                 ipath->fspath->val[i] = rel_ptr;
4229         }
4230
4231         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4232                            ipath->fspath, size);
4233         if (ret) {
4234                 ret = -EFAULT;
4235                 goto out;
4236         }
4237
4238 out:
4239         btrfs_free_path(path);
4240         free_ipath(ipath);
4241         kfree(ipa);
4242
4243         return ret;
4244 }
4245
4246 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4247 {
4248         struct btrfs_data_container *inodes = ctx;
4249         const size_t c = 3 * sizeof(u64);
4250
4251         if (inodes->bytes_left >= c) {
4252                 inodes->bytes_left -= c;
4253                 inodes->val[inodes->elem_cnt] = inum;
4254                 inodes->val[inodes->elem_cnt + 1] = offset;
4255                 inodes->val[inodes->elem_cnt + 2] = root;
4256                 inodes->elem_cnt += 3;
4257         } else {
4258                 inodes->bytes_missing += c - inodes->bytes_left;
4259                 inodes->bytes_left = 0;
4260                 inodes->elem_missed += 3;
4261         }
4262
4263         return 0;
4264 }
4265
4266 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4267                                         void __user *arg, int version)
4268 {
4269         int ret = 0;
4270         int size;
4271         struct btrfs_ioctl_logical_ino_args *loi;
4272         struct btrfs_data_container *inodes = NULL;
4273         struct btrfs_path *path = NULL;
4274         bool ignore_offset;
4275
4276         if (!capable(CAP_SYS_ADMIN))
4277                 return -EPERM;
4278
4279         loi = memdup_user(arg, sizeof(*loi));
4280         if (IS_ERR(loi))
4281                 return PTR_ERR(loi);
4282
4283         if (version == 1) {
4284                 ignore_offset = false;
4285                 size = min_t(u32, loi->size, SZ_64K);
4286         } else {
4287                 /* All reserved bits must be 0 for now */
4288                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4289                         ret = -EINVAL;
4290                         goto out_loi;
4291                 }
4292                 /* Only accept flags we have defined so far */
4293                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4294                         ret = -EINVAL;
4295                         goto out_loi;
4296                 }
4297                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4298                 size = min_t(u32, loi->size, SZ_16M);
4299         }
4300
4301         path = btrfs_alloc_path();
4302         if (!path) {
4303                 ret = -ENOMEM;
4304                 goto out;
4305         }
4306
4307         inodes = init_data_container(size);
4308         if (IS_ERR(inodes)) {
4309                 ret = PTR_ERR(inodes);
4310                 inodes = NULL;
4311                 goto out;
4312         }
4313
4314         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4315                                           build_ino_list, inodes, ignore_offset);
4316         if (ret == -EINVAL)
4317                 ret = -ENOENT;
4318         if (ret < 0)
4319                 goto out;
4320
4321         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4322                            size);
4323         if (ret)
4324                 ret = -EFAULT;
4325
4326 out:
4327         btrfs_free_path(path);
4328         kvfree(inodes);
4329 out_loi:
4330         kfree(loi);
4331
4332         return ret;
4333 }
4334
4335 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4336                                struct btrfs_ioctl_balance_args *bargs)
4337 {
4338         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4339
4340         bargs->flags = bctl->flags;
4341
4342         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4343                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4344         if (atomic_read(&fs_info->balance_pause_req))
4345                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4346         if (atomic_read(&fs_info->balance_cancel_req))
4347                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4348
4349         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4350         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4351         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4352
4353         spin_lock(&fs_info->balance_lock);
4354         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4355         spin_unlock(&fs_info->balance_lock);
4356 }
4357
4358 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4359 {
4360         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4361         struct btrfs_fs_info *fs_info = root->fs_info;
4362         struct btrfs_ioctl_balance_args *bargs;
4363         struct btrfs_balance_control *bctl;
4364         bool need_unlock; /* for mut. excl. ops lock */
4365         int ret;
4366
4367         if (!capable(CAP_SYS_ADMIN))
4368                 return -EPERM;
4369
4370         ret = mnt_want_write_file(file);
4371         if (ret)
4372                 return ret;
4373
4374         bargs = memdup_user(arg, sizeof(*bargs));
4375         if (IS_ERR(bargs)) {
4376                 ret = PTR_ERR(bargs);
4377                 bargs = NULL;
4378                 goto out;
4379         }
4380
4381 again:
4382         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4383                 mutex_lock(&fs_info->balance_mutex);
4384                 need_unlock = true;
4385                 goto locked;
4386         }
4387
4388         /*
4389          * mut. excl. ops lock is locked.  Three possibilities:
4390          *   (1) some other op is running
4391          *   (2) balance is running
4392          *   (3) balance is paused -- special case (think resume)
4393          */
4394         mutex_lock(&fs_info->balance_mutex);
4395         if (fs_info->balance_ctl) {
4396                 /* this is either (2) or (3) */
4397                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4398                         mutex_unlock(&fs_info->balance_mutex);
4399                         /*
4400                          * Lock released to allow other waiters to continue,
4401                          * we'll reexamine the status again.
4402                          */
4403                         mutex_lock(&fs_info->balance_mutex);
4404
4405                         if (fs_info->balance_ctl &&
4406                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4407                                 /* this is (3) */
4408                                 need_unlock = false;
4409                                 goto locked;
4410                         }
4411
4412                         mutex_unlock(&fs_info->balance_mutex);
4413                         goto again;
4414                 } else {
4415                         /* this is (2) */
4416                         mutex_unlock(&fs_info->balance_mutex);
4417                         ret = -EINPROGRESS;
4418                         goto out;
4419                 }
4420         } else {
4421                 /* this is (1) */
4422                 mutex_unlock(&fs_info->balance_mutex);
4423                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4424                 goto out;
4425         }
4426
4427 locked:
4428         if (bargs->flags & BTRFS_BALANCE_RESUME) {
4429                 if (!fs_info->balance_ctl) {
4430                         ret = -ENOTCONN;
4431                         goto out_unlock;
4432                 }
4433
4434                 bctl = fs_info->balance_ctl;
4435                 spin_lock(&fs_info->balance_lock);
4436                 bctl->flags |= BTRFS_BALANCE_RESUME;
4437                 spin_unlock(&fs_info->balance_lock);
4438                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4439
4440                 goto do_balance;
4441         }
4442
4443         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4444                 ret = -EINVAL;
4445                 goto out_unlock;
4446         }
4447
4448         if (fs_info->balance_ctl) {
4449                 ret = -EINPROGRESS;
4450                 goto out_unlock;
4451         }
4452
4453         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4454         if (!bctl) {
4455                 ret = -ENOMEM;
4456                 goto out_unlock;
4457         }
4458
4459         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4460         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4461         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4462
4463         bctl->flags = bargs->flags;
4464 do_balance:
4465         /*
4466          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4467          * bctl is freed in reset_balance_state, or, if restriper was paused
4468          * all the way until unmount, in free_fs_info.  The flag should be
4469          * cleared after reset_balance_state.
4470          */
4471         need_unlock = false;
4472
4473         ret = btrfs_balance(fs_info, bctl, bargs);
4474         bctl = NULL;
4475
4476         if (ret == 0 || ret == -ECANCELED) {
4477                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4478                         ret = -EFAULT;
4479         }
4480
4481         kfree(bctl);
4482 out_unlock:
4483         mutex_unlock(&fs_info->balance_mutex);
4484         if (need_unlock)
4485                 btrfs_exclop_finish(fs_info);
4486 out:
4487         mnt_drop_write_file(file);
4488         kfree(bargs);
4489         return ret;
4490 }
4491
4492 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4493 {
4494         if (!capable(CAP_SYS_ADMIN))
4495                 return -EPERM;
4496
4497         switch (cmd) {
4498         case BTRFS_BALANCE_CTL_PAUSE:
4499                 return btrfs_pause_balance(fs_info);
4500         case BTRFS_BALANCE_CTL_CANCEL:
4501                 return btrfs_cancel_balance(fs_info);
4502         }
4503
4504         return -EINVAL;
4505 }
4506
4507 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4508                                          void __user *arg)
4509 {
4510         struct btrfs_ioctl_balance_args *bargs;
4511         int ret = 0;
4512
4513         if (!capable(CAP_SYS_ADMIN))
4514                 return -EPERM;
4515
4516         mutex_lock(&fs_info->balance_mutex);
4517         if (!fs_info->balance_ctl) {
4518                 ret = -ENOTCONN;
4519                 goto out;
4520         }
4521
4522         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4523         if (!bargs) {
4524                 ret = -ENOMEM;
4525                 goto out;
4526         }
4527
4528         btrfs_update_ioctl_balance_args(fs_info, bargs);
4529
4530         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4531                 ret = -EFAULT;
4532
4533         kfree(bargs);
4534 out:
4535         mutex_unlock(&fs_info->balance_mutex);
4536         return ret;
4537 }
4538
4539 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4540 {
4541         struct inode *inode = file_inode(file);
4542         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4543         struct btrfs_ioctl_quota_ctl_args *sa;
4544         int ret;
4545
4546         if (!capable(CAP_SYS_ADMIN))
4547                 return -EPERM;
4548
4549         ret = mnt_want_write_file(file);
4550         if (ret)
4551                 return ret;
4552
4553         sa = memdup_user(arg, sizeof(*sa));
4554         if (IS_ERR(sa)) {
4555                 ret = PTR_ERR(sa);
4556                 goto drop_write;
4557         }
4558
4559         down_write(&fs_info->subvol_sem);
4560
4561         switch (sa->cmd) {
4562         case BTRFS_QUOTA_CTL_ENABLE:
4563                 ret = btrfs_quota_enable(fs_info);
4564                 break;
4565         case BTRFS_QUOTA_CTL_DISABLE:
4566                 ret = btrfs_quota_disable(fs_info);
4567                 break;
4568         default:
4569                 ret = -EINVAL;
4570                 break;
4571         }
4572
4573         kfree(sa);
4574         up_write(&fs_info->subvol_sem);
4575 drop_write:
4576         mnt_drop_write_file(file);
4577         return ret;
4578 }
4579
4580 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4581 {
4582         struct inode *inode = file_inode(file);
4583         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4584         struct btrfs_root *root = BTRFS_I(inode)->root;
4585         struct btrfs_ioctl_qgroup_assign_args *sa;
4586         struct btrfs_trans_handle *trans;
4587         int ret;
4588         int err;
4589
4590         if (!capable(CAP_SYS_ADMIN))
4591                 return -EPERM;
4592
4593         ret = mnt_want_write_file(file);
4594         if (ret)
4595                 return ret;
4596
4597         sa = memdup_user(arg, sizeof(*sa));
4598         if (IS_ERR(sa)) {
4599                 ret = PTR_ERR(sa);
4600                 goto drop_write;
4601         }
4602
4603         trans = btrfs_join_transaction(root);
4604         if (IS_ERR(trans)) {
4605                 ret = PTR_ERR(trans);
4606                 goto out;
4607         }
4608
4609         if (sa->assign) {
4610                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4611         } else {
4612                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4613         }
4614
4615         /* update qgroup status and info */
4616         err = btrfs_run_qgroups(trans);
4617         if (err < 0)
4618                 btrfs_handle_fs_error(fs_info, err,
4619                                       "failed to update qgroup status and info");
4620         err = btrfs_end_transaction(trans);
4621         if (err && !ret)
4622                 ret = err;
4623
4624 out:
4625         kfree(sa);
4626 drop_write:
4627         mnt_drop_write_file(file);
4628         return ret;
4629 }
4630
4631 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4632 {
4633         struct inode *inode = file_inode(file);
4634         struct btrfs_root *root = BTRFS_I(inode)->root;
4635         struct btrfs_ioctl_qgroup_create_args *sa;
4636         struct btrfs_trans_handle *trans;
4637         int ret;
4638         int err;
4639
4640         if (!capable(CAP_SYS_ADMIN))
4641                 return -EPERM;
4642
4643         ret = mnt_want_write_file(file);
4644         if (ret)
4645                 return ret;
4646
4647         sa = memdup_user(arg, sizeof(*sa));
4648         if (IS_ERR(sa)) {
4649                 ret = PTR_ERR(sa);
4650                 goto drop_write;
4651         }
4652
4653         if (!sa->qgroupid) {
4654                 ret = -EINVAL;
4655                 goto out;
4656         }
4657
4658         trans = btrfs_join_transaction(root);
4659         if (IS_ERR(trans)) {
4660                 ret = PTR_ERR(trans);
4661                 goto out;
4662         }
4663
4664         if (sa->create) {
4665                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4666         } else {
4667                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4668         }
4669
4670         err = btrfs_end_transaction(trans);
4671         if (err && !ret)
4672                 ret = err;
4673
4674 out:
4675         kfree(sa);
4676 drop_write:
4677         mnt_drop_write_file(file);
4678         return ret;
4679 }
4680
4681 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4682 {
4683         struct inode *inode = file_inode(file);
4684         struct btrfs_root *root = BTRFS_I(inode)->root;
4685         struct btrfs_ioctl_qgroup_limit_args *sa;
4686         struct btrfs_trans_handle *trans;
4687         int ret;
4688         int err;
4689         u64 qgroupid;
4690
4691         if (!capable(CAP_SYS_ADMIN))
4692                 return -EPERM;
4693
4694         ret = mnt_want_write_file(file);
4695         if (ret)
4696                 return ret;
4697
4698         sa = memdup_user(arg, sizeof(*sa));
4699         if (IS_ERR(sa)) {
4700                 ret = PTR_ERR(sa);
4701                 goto drop_write;
4702         }
4703
4704         trans = btrfs_join_transaction(root);
4705         if (IS_ERR(trans)) {
4706                 ret = PTR_ERR(trans);
4707                 goto out;
4708         }
4709
4710         qgroupid = sa->qgroupid;
4711         if (!qgroupid) {
4712                 /* take the current subvol as qgroup */
4713                 qgroupid = root->root_key.objectid;
4714         }
4715
4716         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4717
4718         err = btrfs_end_transaction(trans);
4719         if (err && !ret)
4720                 ret = err;
4721
4722 out:
4723         kfree(sa);
4724 drop_write:
4725         mnt_drop_write_file(file);
4726         return ret;
4727 }
4728
4729 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4730 {
4731         struct inode *inode = file_inode(file);
4732         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4733         struct btrfs_ioctl_quota_rescan_args *qsa;
4734         int ret;
4735
4736         if (!capable(CAP_SYS_ADMIN))
4737                 return -EPERM;
4738
4739         ret = mnt_want_write_file(file);
4740         if (ret)
4741                 return ret;
4742
4743         qsa = memdup_user(arg, sizeof(*qsa));
4744         if (IS_ERR(qsa)) {
4745                 ret = PTR_ERR(qsa);
4746                 goto drop_write;
4747         }
4748
4749         if (qsa->flags) {
4750                 ret = -EINVAL;
4751                 goto out;
4752         }
4753
4754         ret = btrfs_qgroup_rescan(fs_info);
4755
4756 out:
4757         kfree(qsa);
4758 drop_write:
4759         mnt_drop_write_file(file);
4760         return ret;
4761 }
4762
4763 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4764                                                 void __user *arg)
4765 {
4766         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4767
4768         if (!capable(CAP_SYS_ADMIN))
4769                 return -EPERM;
4770
4771         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4772                 qsa.flags = 1;
4773                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4774         }
4775
4776         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4777                 return -EFAULT;
4778
4779         return 0;
4780 }
4781
4782 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4783                                                 void __user *arg)
4784 {
4785         if (!capable(CAP_SYS_ADMIN))
4786                 return -EPERM;
4787
4788         return btrfs_qgroup_wait_for_completion(fs_info, true);
4789 }
4790
4791 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4792                                             struct user_namespace *mnt_userns,
4793                                             struct btrfs_ioctl_received_subvol_args *sa)
4794 {
4795         struct inode *inode = file_inode(file);
4796         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4797         struct btrfs_root *root = BTRFS_I(inode)->root;
4798         struct btrfs_root_item *root_item = &root->root_item;
4799         struct btrfs_trans_handle *trans;
4800         struct timespec64 ct = current_time(inode);
4801         int ret = 0;
4802         int received_uuid_changed;
4803
4804         if (!inode_owner_or_capable(mnt_userns, inode))
4805                 return -EPERM;
4806
4807         ret = mnt_want_write_file(file);
4808         if (ret < 0)
4809                 return ret;
4810
4811         down_write(&fs_info->subvol_sem);
4812
4813         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4814                 ret = -EINVAL;
4815                 goto out;
4816         }
4817
4818         if (btrfs_root_readonly(root)) {
4819                 ret = -EROFS;
4820                 goto out;
4821         }
4822
4823         /*
4824          * 1 - root item
4825          * 2 - uuid items (received uuid + subvol uuid)
4826          */
4827         trans = btrfs_start_transaction(root, 3);
4828         if (IS_ERR(trans)) {
4829                 ret = PTR_ERR(trans);
4830                 trans = NULL;
4831                 goto out;
4832         }
4833
4834         sa->rtransid = trans->transid;
4835         sa->rtime.sec = ct.tv_sec;
4836         sa->rtime.nsec = ct.tv_nsec;
4837
4838         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4839                                        BTRFS_UUID_SIZE);
4840         if (received_uuid_changed &&
4841             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4842                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4843                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4844                                           root->root_key.objectid);
4845                 if (ret && ret != -ENOENT) {
4846                         btrfs_abort_transaction(trans, ret);
4847                         btrfs_end_transaction(trans);
4848                         goto out;
4849                 }
4850         }
4851         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4852         btrfs_set_root_stransid(root_item, sa->stransid);
4853         btrfs_set_root_rtransid(root_item, sa->rtransid);
4854         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4855         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4856         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4857         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4858
4859         ret = btrfs_update_root(trans, fs_info->tree_root,
4860                                 &root->root_key, &root->root_item);
4861         if (ret < 0) {
4862                 btrfs_end_transaction(trans);
4863                 goto out;
4864         }
4865         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4866                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4867                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4868                                           root->root_key.objectid);
4869                 if (ret < 0 && ret != -EEXIST) {
4870                         btrfs_abort_transaction(trans, ret);
4871                         btrfs_end_transaction(trans);
4872                         goto out;
4873                 }
4874         }
4875         ret = btrfs_commit_transaction(trans);
4876 out:
4877         up_write(&fs_info->subvol_sem);
4878         mnt_drop_write_file(file);
4879         return ret;
4880 }
4881
4882 #ifdef CONFIG_64BIT
4883 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4884                                                 void __user *arg)
4885 {
4886         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4887         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4888         int ret = 0;
4889
4890         args32 = memdup_user(arg, sizeof(*args32));
4891         if (IS_ERR(args32))
4892                 return PTR_ERR(args32);
4893
4894         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4895         if (!args64) {
4896                 ret = -ENOMEM;
4897                 goto out;
4898         }
4899
4900         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4901         args64->stransid = args32->stransid;
4902         args64->rtransid = args32->rtransid;
4903         args64->stime.sec = args32->stime.sec;
4904         args64->stime.nsec = args32->stime.nsec;
4905         args64->rtime.sec = args32->rtime.sec;
4906         args64->rtime.nsec = args32->rtime.nsec;
4907         args64->flags = args32->flags;
4908
4909         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4910         if (ret)
4911                 goto out;
4912
4913         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4914         args32->stransid = args64->stransid;
4915         args32->rtransid = args64->rtransid;
4916         args32->stime.sec = args64->stime.sec;
4917         args32->stime.nsec = args64->stime.nsec;
4918         args32->rtime.sec = args64->rtime.sec;
4919         args32->rtime.nsec = args64->rtime.nsec;
4920         args32->flags = args64->flags;
4921
4922         ret = copy_to_user(arg, args32, sizeof(*args32));
4923         if (ret)
4924                 ret = -EFAULT;
4925
4926 out:
4927         kfree(args32);
4928         kfree(args64);
4929         return ret;
4930 }
4931 #endif
4932
4933 static long btrfs_ioctl_set_received_subvol(struct file *file,
4934                                             void __user *arg)
4935 {
4936         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4937         int ret = 0;
4938
4939         sa = memdup_user(arg, sizeof(*sa));
4940         if (IS_ERR(sa))
4941                 return PTR_ERR(sa);
4942
4943         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4944
4945         if (ret)
4946                 goto out;
4947
4948         ret = copy_to_user(arg, sa, sizeof(*sa));
4949         if (ret)
4950                 ret = -EFAULT;
4951
4952 out:
4953         kfree(sa);
4954         return ret;
4955 }
4956
4957 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4958                                         void __user *arg)
4959 {
4960         size_t len;
4961         int ret;
4962         char label[BTRFS_LABEL_SIZE];
4963
4964         spin_lock(&fs_info->super_lock);
4965         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4966         spin_unlock(&fs_info->super_lock);
4967
4968         len = strnlen(label, BTRFS_LABEL_SIZE);
4969
4970         if (len == BTRFS_LABEL_SIZE) {
4971                 btrfs_warn(fs_info,
4972                            "label is too long, return the first %zu bytes",
4973                            --len);
4974         }
4975
4976         ret = copy_to_user(arg, label, len);
4977
4978         return ret ? -EFAULT : 0;
4979 }
4980
4981 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4982 {
4983         struct inode *inode = file_inode(file);
4984         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4985         struct btrfs_root *root = BTRFS_I(inode)->root;
4986         struct btrfs_super_block *super_block = fs_info->super_copy;
4987         struct btrfs_trans_handle *trans;
4988         char label[BTRFS_LABEL_SIZE];
4989         int ret;
4990
4991         if (!capable(CAP_SYS_ADMIN))
4992                 return -EPERM;
4993
4994         if (copy_from_user(label, arg, sizeof(label)))
4995                 return -EFAULT;
4996
4997         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4998                 btrfs_err(fs_info,
4999                           "unable to set label with more than %d bytes",
5000                           BTRFS_LABEL_SIZE - 1);
5001                 return -EINVAL;
5002         }
5003
5004         ret = mnt_want_write_file(file);
5005         if (ret)
5006                 return ret;
5007
5008         trans = btrfs_start_transaction(root, 0);
5009         if (IS_ERR(trans)) {
5010                 ret = PTR_ERR(trans);
5011                 goto out_unlock;
5012         }
5013
5014         spin_lock(&fs_info->super_lock);
5015         strcpy(super_block->label, label);
5016         spin_unlock(&fs_info->super_lock);
5017         ret = btrfs_commit_transaction(trans);
5018
5019 out_unlock:
5020         mnt_drop_write_file(file);
5021         return ret;
5022 }
5023
5024 #define INIT_FEATURE_FLAGS(suffix) \
5025         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5026           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5027           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5028
5029 int btrfs_ioctl_get_supported_features(void __user *arg)
5030 {
5031         static const struct btrfs_ioctl_feature_flags features[3] = {
5032                 INIT_FEATURE_FLAGS(SUPP),
5033                 INIT_FEATURE_FLAGS(SAFE_SET),
5034                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5035         };
5036
5037         if (copy_to_user(arg, &features, sizeof(features)))
5038                 return -EFAULT;
5039
5040         return 0;
5041 }
5042
5043 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5044                                         void __user *arg)
5045 {
5046         struct btrfs_super_block *super_block = fs_info->super_copy;
5047         struct btrfs_ioctl_feature_flags features;
5048
5049         features.compat_flags = btrfs_super_compat_flags(super_block);
5050         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5051         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5052
5053         if (copy_to_user(arg, &features, sizeof(features)))
5054                 return -EFAULT;
5055
5056         return 0;
5057 }
5058
5059 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5060                               enum btrfs_feature_set set,
5061                               u64 change_mask, u64 flags, u64 supported_flags,
5062                               u64 safe_set, u64 safe_clear)
5063 {
5064         const char *type = btrfs_feature_set_name(set);
5065         char *names;
5066         u64 disallowed, unsupported;
5067         u64 set_mask = flags & change_mask;
5068         u64 clear_mask = ~flags & change_mask;
5069
5070         unsupported = set_mask & ~supported_flags;
5071         if (unsupported) {
5072                 names = btrfs_printable_features(set, unsupported);
5073                 if (names) {
5074                         btrfs_warn(fs_info,
5075                                    "this kernel does not support the %s feature bit%s",
5076                                    names, strchr(names, ',') ? "s" : "");
5077                         kfree(names);
5078                 } else
5079                         btrfs_warn(fs_info,
5080                                    "this kernel does not support %s bits 0x%llx",
5081                                    type, unsupported);
5082                 return -EOPNOTSUPP;
5083         }
5084
5085         disallowed = set_mask & ~safe_set;
5086         if (disallowed) {
5087                 names = btrfs_printable_features(set, disallowed);
5088                 if (names) {
5089                         btrfs_warn(fs_info,
5090                                    "can't set the %s feature bit%s while mounted",
5091                                    names, strchr(names, ',') ? "s" : "");
5092                         kfree(names);
5093                 } else
5094                         btrfs_warn(fs_info,
5095                                    "can't set %s bits 0x%llx while mounted",
5096                                    type, disallowed);
5097                 return -EPERM;
5098         }
5099
5100         disallowed = clear_mask & ~safe_clear;
5101         if (disallowed) {
5102                 names = btrfs_printable_features(set, disallowed);
5103                 if (names) {
5104                         btrfs_warn(fs_info,
5105                                    "can't clear the %s feature bit%s while mounted",
5106                                    names, strchr(names, ',') ? "s" : "");
5107                         kfree(names);
5108                 } else
5109                         btrfs_warn(fs_info,
5110                                    "can't clear %s bits 0x%llx while mounted",
5111                                    type, disallowed);
5112                 return -EPERM;
5113         }
5114
5115         return 0;
5116 }
5117
5118 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5119 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5120                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5121                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5122                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5123
5124 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5125 {
5126         struct inode *inode = file_inode(file);
5127         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5128         struct btrfs_root *root = BTRFS_I(inode)->root;
5129         struct btrfs_super_block *super_block = fs_info->super_copy;
5130         struct btrfs_ioctl_feature_flags flags[2];
5131         struct btrfs_trans_handle *trans;
5132         u64 newflags;
5133         int ret;
5134
5135         if (!capable(CAP_SYS_ADMIN))
5136                 return -EPERM;
5137
5138         if (copy_from_user(flags, arg, sizeof(flags)))
5139                 return -EFAULT;
5140
5141         /* Nothing to do */
5142         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5143             !flags[0].incompat_flags)
5144                 return 0;
5145
5146         ret = check_feature(fs_info, flags[0].compat_flags,
5147                             flags[1].compat_flags, COMPAT);
5148         if (ret)
5149                 return ret;
5150
5151         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5152                             flags[1].compat_ro_flags, COMPAT_RO);
5153         if (ret)
5154                 return ret;
5155
5156         ret = check_feature(fs_info, flags[0].incompat_flags,
5157                             flags[1].incompat_flags, INCOMPAT);
5158         if (ret)
5159                 return ret;
5160
5161         ret = mnt_want_write_file(file);
5162         if (ret)
5163                 return ret;
5164
5165         trans = btrfs_start_transaction(root, 0);
5166         if (IS_ERR(trans)) {
5167                 ret = PTR_ERR(trans);
5168                 goto out_drop_write;
5169         }
5170
5171         spin_lock(&fs_info->super_lock);
5172         newflags = btrfs_super_compat_flags(super_block);
5173         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5174         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5175         btrfs_set_super_compat_flags(super_block, newflags);
5176
5177         newflags = btrfs_super_compat_ro_flags(super_block);
5178         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5179         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5180         btrfs_set_super_compat_ro_flags(super_block, newflags);
5181
5182         newflags = btrfs_super_incompat_flags(super_block);
5183         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5184         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5185         btrfs_set_super_incompat_flags(super_block, newflags);
5186         spin_unlock(&fs_info->super_lock);
5187
5188         ret = btrfs_commit_transaction(trans);
5189 out_drop_write:
5190         mnt_drop_write_file(file);
5191
5192         return ret;
5193 }
5194
5195 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5196 {
5197         struct btrfs_ioctl_send_args *arg;
5198         int ret;
5199
5200         if (compat) {
5201 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5202                 struct btrfs_ioctl_send_args_32 args32;
5203
5204                 ret = copy_from_user(&args32, argp, sizeof(args32));
5205                 if (ret)
5206                         return -EFAULT;
5207                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5208                 if (!arg)
5209                         return -ENOMEM;
5210                 arg->send_fd = args32.send_fd;
5211                 arg->clone_sources_count = args32.clone_sources_count;
5212                 arg->clone_sources = compat_ptr(args32.clone_sources);
5213                 arg->parent_root = args32.parent_root;
5214                 arg->flags = args32.flags;
5215                 memcpy(arg->reserved, args32.reserved,
5216                        sizeof(args32.reserved));
5217 #else
5218                 return -ENOTTY;
5219 #endif
5220         } else {
5221                 arg = memdup_user(argp, sizeof(*arg));
5222                 if (IS_ERR(arg))
5223                         return PTR_ERR(arg);
5224         }
5225         ret = btrfs_ioctl_send(inode, arg);
5226         kfree(arg);
5227         return ret;
5228 }
5229
5230 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5231                                     bool compat)
5232 {
5233         struct btrfs_ioctl_encoded_io_args args = { 0 };
5234         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5235                                              flags);
5236         size_t copy_end;
5237         struct iovec iovstack[UIO_FASTIOV];
5238         struct iovec *iov = iovstack;
5239         struct iov_iter iter;
5240         loff_t pos;
5241         struct kiocb kiocb;
5242         ssize_t ret;
5243
5244         if (!capable(CAP_SYS_ADMIN)) {
5245                 ret = -EPERM;
5246                 goto out_acct;
5247         }
5248
5249         if (compat) {
5250 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5251                 struct btrfs_ioctl_encoded_io_args_32 args32;
5252
5253                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5254                                        flags);
5255                 if (copy_from_user(&args32, argp, copy_end)) {
5256                         ret = -EFAULT;
5257                         goto out_acct;
5258                 }
5259                 args.iov = compat_ptr(args32.iov);
5260                 args.iovcnt = args32.iovcnt;
5261                 args.offset = args32.offset;
5262                 args.flags = args32.flags;
5263 #else
5264                 return -ENOTTY;
5265 #endif
5266         } else {
5267                 copy_end = copy_end_kernel;
5268                 if (copy_from_user(&args, argp, copy_end)) {
5269                         ret = -EFAULT;
5270                         goto out_acct;
5271                 }
5272         }
5273         if (args.flags != 0) {
5274                 ret = -EINVAL;
5275                 goto out_acct;
5276         }
5277
5278         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5279                            &iov, &iter);
5280         if (ret < 0)
5281                 goto out_acct;
5282
5283         if (iov_iter_count(&iter) == 0) {
5284                 ret = 0;
5285                 goto out_iov;
5286         }
5287         pos = args.offset;
5288         ret = rw_verify_area(READ, file, &pos, args.len);
5289         if (ret < 0)
5290                 goto out_iov;
5291
5292         init_sync_kiocb(&kiocb, file);
5293         kiocb.ki_pos = pos;
5294
5295         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5296         if (ret >= 0) {
5297                 fsnotify_access(file);
5298                 if (copy_to_user(argp + copy_end,
5299                                  (char *)&args + copy_end_kernel,
5300                                  sizeof(args) - copy_end_kernel))
5301                         ret = -EFAULT;
5302         }
5303
5304 out_iov:
5305         kfree(iov);
5306 out_acct:
5307         if (ret > 0)
5308                 add_rchar(current, ret);
5309         inc_syscr(current);
5310         return ret;
5311 }
5312
5313 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5314 {
5315         struct btrfs_ioctl_encoded_io_args args;
5316         struct iovec iovstack[UIO_FASTIOV];
5317         struct iovec *iov = iovstack;
5318         struct iov_iter iter;
5319         loff_t pos;
5320         struct kiocb kiocb;
5321         ssize_t ret;
5322
5323         if (!capable(CAP_SYS_ADMIN)) {
5324                 ret = -EPERM;
5325                 goto out_acct;
5326         }
5327
5328         if (!(file->f_mode & FMODE_WRITE)) {
5329                 ret = -EBADF;
5330                 goto out_acct;
5331         }
5332
5333         if (compat) {
5334 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5335                 struct btrfs_ioctl_encoded_io_args_32 args32;
5336
5337                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5338                         ret = -EFAULT;
5339                         goto out_acct;
5340                 }
5341                 args.iov = compat_ptr(args32.iov);
5342                 args.iovcnt = args32.iovcnt;
5343                 args.offset = args32.offset;
5344                 args.flags = args32.flags;
5345                 args.len = args32.len;
5346                 args.unencoded_len = args32.unencoded_len;
5347                 args.unencoded_offset = args32.unencoded_offset;
5348                 args.compression = args32.compression;
5349                 args.encryption = args32.encryption;
5350                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5351 #else
5352                 return -ENOTTY;
5353 #endif
5354         } else {
5355                 if (copy_from_user(&args, argp, sizeof(args))) {
5356                         ret = -EFAULT;
5357                         goto out_acct;
5358                 }
5359         }
5360
5361         ret = -EINVAL;
5362         if (args.flags != 0)
5363                 goto out_acct;
5364         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5365                 goto out_acct;
5366         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5367             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5368                 goto out_acct;
5369         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5370             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5371                 goto out_acct;
5372         if (args.unencoded_offset > args.unencoded_len)
5373                 goto out_acct;
5374         if (args.len > args.unencoded_len - args.unencoded_offset)
5375                 goto out_acct;
5376
5377         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5378                            &iov, &iter);
5379         if (ret < 0)
5380                 goto out_acct;
5381
5382         file_start_write(file);
5383
5384         if (iov_iter_count(&iter) == 0) {
5385                 ret = 0;
5386                 goto out_end_write;
5387         }
5388         pos = args.offset;
5389         ret = rw_verify_area(WRITE, file, &pos, args.len);
5390         if (ret < 0)
5391                 goto out_end_write;
5392
5393         init_sync_kiocb(&kiocb, file);
5394         ret = kiocb_set_rw_flags(&kiocb, 0);
5395         if (ret)
5396                 goto out_end_write;
5397         kiocb.ki_pos = pos;
5398
5399         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5400         if (ret > 0)
5401                 fsnotify_modify(file);
5402
5403 out_end_write:
5404         file_end_write(file);
5405         kfree(iov);
5406 out_acct:
5407         if (ret > 0)
5408                 add_wchar(current, ret);
5409         inc_syscw(current);
5410         return ret;
5411 }
5412
5413 long btrfs_ioctl(struct file *file, unsigned int
5414                 cmd, unsigned long arg)
5415 {
5416         struct inode *inode = file_inode(file);
5417         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5418         struct btrfs_root *root = BTRFS_I(inode)->root;
5419         void __user *argp = (void __user *)arg;
5420
5421         switch (cmd) {
5422         case FS_IOC_GETVERSION:
5423                 return btrfs_ioctl_getversion(inode, argp);
5424         case FS_IOC_GETFSLABEL:
5425                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5426         case FS_IOC_SETFSLABEL:
5427                 return btrfs_ioctl_set_fslabel(file, argp);
5428         case FITRIM:
5429                 return btrfs_ioctl_fitrim(fs_info, argp);
5430         case BTRFS_IOC_SNAP_CREATE:
5431                 return btrfs_ioctl_snap_create(file, argp, 0);
5432         case BTRFS_IOC_SNAP_CREATE_V2:
5433                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5434         case BTRFS_IOC_SUBVOL_CREATE:
5435                 return btrfs_ioctl_snap_create(file, argp, 1);
5436         case BTRFS_IOC_SUBVOL_CREATE_V2:
5437                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5438         case BTRFS_IOC_SNAP_DESTROY:
5439                 return btrfs_ioctl_snap_destroy(file, argp, false);
5440         case BTRFS_IOC_SNAP_DESTROY_V2:
5441                 return btrfs_ioctl_snap_destroy(file, argp, true);
5442         case BTRFS_IOC_SUBVOL_GETFLAGS:
5443                 return btrfs_ioctl_subvol_getflags(inode, argp);
5444         case BTRFS_IOC_SUBVOL_SETFLAGS:
5445                 return btrfs_ioctl_subvol_setflags(file, argp);
5446         case BTRFS_IOC_DEFAULT_SUBVOL:
5447                 return btrfs_ioctl_default_subvol(file, argp);
5448         case BTRFS_IOC_DEFRAG:
5449                 return btrfs_ioctl_defrag(file, NULL);
5450         case BTRFS_IOC_DEFRAG_RANGE:
5451                 return btrfs_ioctl_defrag(file, argp);
5452         case BTRFS_IOC_RESIZE:
5453                 return btrfs_ioctl_resize(file, argp);
5454         case BTRFS_IOC_ADD_DEV:
5455                 return btrfs_ioctl_add_dev(fs_info, argp);
5456         case BTRFS_IOC_RM_DEV:
5457                 return btrfs_ioctl_rm_dev(file, argp);
5458         case BTRFS_IOC_RM_DEV_V2:
5459                 return btrfs_ioctl_rm_dev_v2(file, argp);
5460         case BTRFS_IOC_FS_INFO:
5461                 return btrfs_ioctl_fs_info(fs_info, argp);
5462         case BTRFS_IOC_DEV_INFO:
5463                 return btrfs_ioctl_dev_info(fs_info, argp);
5464         case BTRFS_IOC_TREE_SEARCH:
5465                 return btrfs_ioctl_tree_search(inode, argp);
5466         case BTRFS_IOC_TREE_SEARCH_V2:
5467                 return btrfs_ioctl_tree_search_v2(inode, argp);
5468         case BTRFS_IOC_INO_LOOKUP:
5469                 return btrfs_ioctl_ino_lookup(root, argp);
5470         case BTRFS_IOC_INO_PATHS:
5471                 return btrfs_ioctl_ino_to_path(root, argp);
5472         case BTRFS_IOC_LOGICAL_INO:
5473                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5474         case BTRFS_IOC_LOGICAL_INO_V2:
5475                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5476         case BTRFS_IOC_SPACE_INFO:
5477                 return btrfs_ioctl_space_info(fs_info, argp);
5478         case BTRFS_IOC_SYNC: {
5479                 int ret;
5480
5481                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5482                 if (ret)
5483                         return ret;
5484                 ret = btrfs_sync_fs(inode->i_sb, 1);
5485                 /*
5486                  * The transaction thread may want to do more work,
5487                  * namely it pokes the cleaner kthread that will start
5488                  * processing uncleaned subvols.
5489                  */
5490                 wake_up_process(fs_info->transaction_kthread);
5491                 return ret;
5492         }
5493         case BTRFS_IOC_START_SYNC:
5494                 return btrfs_ioctl_start_sync(root, argp);
5495         case BTRFS_IOC_WAIT_SYNC:
5496                 return btrfs_ioctl_wait_sync(fs_info, argp);
5497         case BTRFS_IOC_SCRUB:
5498                 return btrfs_ioctl_scrub(file, argp);
5499         case BTRFS_IOC_SCRUB_CANCEL:
5500                 return btrfs_ioctl_scrub_cancel(fs_info);
5501         case BTRFS_IOC_SCRUB_PROGRESS:
5502                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5503         case BTRFS_IOC_BALANCE_V2:
5504                 return btrfs_ioctl_balance(file, argp);
5505         case BTRFS_IOC_BALANCE_CTL:
5506                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5507         case BTRFS_IOC_BALANCE_PROGRESS:
5508                 return btrfs_ioctl_balance_progress(fs_info, argp);
5509         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5510                 return btrfs_ioctl_set_received_subvol(file, argp);
5511 #ifdef CONFIG_64BIT
5512         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5513                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5514 #endif
5515         case BTRFS_IOC_SEND:
5516                 return _btrfs_ioctl_send(inode, argp, false);
5517 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5518         case BTRFS_IOC_SEND_32:
5519                 return _btrfs_ioctl_send(inode, argp, true);
5520 #endif
5521         case BTRFS_IOC_GET_DEV_STATS:
5522                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5523         case BTRFS_IOC_QUOTA_CTL:
5524                 return btrfs_ioctl_quota_ctl(file, argp);
5525         case BTRFS_IOC_QGROUP_ASSIGN:
5526                 return btrfs_ioctl_qgroup_assign(file, argp);
5527         case BTRFS_IOC_QGROUP_CREATE:
5528                 return btrfs_ioctl_qgroup_create(file, argp);
5529         case BTRFS_IOC_QGROUP_LIMIT:
5530                 return btrfs_ioctl_qgroup_limit(file, argp);
5531         case BTRFS_IOC_QUOTA_RESCAN:
5532                 return btrfs_ioctl_quota_rescan(file, argp);
5533         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5534                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5535         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5536                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5537         case BTRFS_IOC_DEV_REPLACE:
5538                 return btrfs_ioctl_dev_replace(fs_info, argp);
5539         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5540                 return btrfs_ioctl_get_supported_features(argp);
5541         case BTRFS_IOC_GET_FEATURES:
5542                 return btrfs_ioctl_get_features(fs_info, argp);
5543         case BTRFS_IOC_SET_FEATURES:
5544                 return btrfs_ioctl_set_features(file, argp);
5545         case BTRFS_IOC_GET_SUBVOL_INFO:
5546                 return btrfs_ioctl_get_subvol_info(inode, argp);
5547         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5548                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5549         case BTRFS_IOC_INO_LOOKUP_USER:
5550                 return btrfs_ioctl_ino_lookup_user(file, argp);
5551         case FS_IOC_ENABLE_VERITY:
5552                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5553         case FS_IOC_MEASURE_VERITY:
5554                 return fsverity_ioctl_measure(file, argp);
5555         case BTRFS_IOC_ENCODED_READ:
5556                 return btrfs_ioctl_encoded_read(file, argp, false);
5557         case BTRFS_IOC_ENCODED_WRITE:
5558                 return btrfs_ioctl_encoded_write(file, argp, false);
5559 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5560         case BTRFS_IOC_ENCODED_READ_32:
5561                 return btrfs_ioctl_encoded_read(file, argp, true);
5562         case BTRFS_IOC_ENCODED_WRITE_32:
5563                 return btrfs_ioctl_encoded_write(file, argp, true);
5564 #endif
5565         }
5566
5567         return -ENOTTY;
5568 }
5569
5570 #ifdef CONFIG_COMPAT
5571 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5572 {
5573         /*
5574          * These all access 32-bit values anyway so no further
5575          * handling is necessary.
5576          */
5577         switch (cmd) {
5578         case FS_IOC32_GETVERSION:
5579                 cmd = FS_IOC_GETVERSION;
5580                 break;
5581         }
5582
5583         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5584 }
5585 #endif