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 "ctree.h"
  32 #include "disk-io.h"
  33 #include "export.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "print-tree.h"
  37 #include "volumes.h"
  38 #include "locking.h"
  39 #include "backref.h"
  40 #include "rcu-string.h"
  41 #include "send.h"
  42 #include "dev-replace.h"
  43 #include "props.h"
  44 #include "sysfs.h"
  45 #include "qgroup.h"
  46 #include "tree-log.h"
  47 #include "compression.h"
  48 #include "space-info.h"
  49 #include "delalloc-space.h"
  50 #include "block-group.h"
  51 #include "subpage.h"
  52
  53 #ifdef CONFIG_64BIT
  54 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  55  * structures are incorrect, as the timespec structure from userspace
  56  * is 4 bytes too small. We define these alternatives here to teach
  57  * the kernel about the 32-bit struct packing.
  58  */
  59 struct btrfs_ioctl_timespec_32 {
  60         __u64 sec;
  61         __u32 nsec;
  62 } __attribute__ ((__packed__));
  63
  64 struct btrfs_ioctl_received_subvol_args_32 {
  65         char    uuid[BTRFS_UUID_SIZE];  /* in */
  66         __u64   stransid;               /* in */
  67         __u64   rtransid;               /* out */
  68         struct btrfs_ioctl_timespec_32 stime; /* in */
  69         struct btrfs_ioctl_timespec_32 rtime; /* out */
  70         __u64   flags;                  /* in */
  71         __u64   reserved[16];           /* in */
  72 } __attribute__ ((__packed__));
  73
  74 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  75                                 struct btrfs_ioctl_received_subvol_args_32)
  76 #endif
  77
  78 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  79 struct btrfs_ioctl_send_args_32 {
  80         __s64 send_fd;                  /* in */
  81         __u64 clone_sources_count;      /* in */
  82         compat_uptr_t clone_sources;    /* in */
  83         __u64 parent_root;              /* in */
  84         __u64 flags;                    /* in */
  85         __u32 version;                  /* in */
  86         __u8  reserved[28];             /* in */
  87 } __attribute__ ((__packed__));
  88
  89 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  90                                struct btrfs_ioctl_send_args_32)
  91 #endif
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  95                 unsigned int flags)
  96 {
  97         if (S_ISDIR(inode->i_mode))
  98                 return flags;
  99         else if (S_ISREG(inode->i_mode))
 100                 return flags & ~FS_DIRSYNC_FL;
 101         else
 102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103 }
 104
 105 /*
 106  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 107  * ioctl.
 108  */
 109 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 110 {
 111         unsigned int iflags = 0;
 112         u32 flags = binode->flags;
 113         u32 ro_flags = binode->ro_flags;
 114
 115         if (flags & BTRFS_INODE_SYNC)
 116                 iflags |= FS_SYNC_FL;
 117         if (flags & BTRFS_INODE_IMMUTABLE)
 118                 iflags |= FS_IMMUTABLE_FL;
 119         if (flags & BTRFS_INODE_APPEND)
 120                 iflags |= FS_APPEND_FL;
 121         if (flags & BTRFS_INODE_NODUMP)
 122                 iflags |= FS_NODUMP_FL;
 123         if (flags & BTRFS_INODE_NOATIME)
 124                 iflags |= FS_NOATIME_FL;
 125         if (flags & BTRFS_INODE_DIRSYNC)
 126                 iflags |= FS_DIRSYNC_FL;
 127         if (flags & BTRFS_INODE_NODATACOW)
 128                 iflags |= FS_NOCOW_FL;
 129         if (ro_flags & BTRFS_INODE_RO_VERITY)
 130                 iflags |= FS_VERITY_FL;
 131
 132         if (flags & BTRFS_INODE_NOCOMPRESS)
 133                 iflags |= FS_NOCOMP_FL;
 134         else if (flags & BTRFS_INODE_COMPRESS)
 135                 iflags |= FS_COMPR_FL;
 136
 137         return iflags;
 138 }
 139
 140 /*
 141  * Update inode->i_flags based on the btrfs internal flags.
 142  */
 143 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 144 {
 145         struct btrfs_inode *binode = BTRFS_I(inode);
 146         unsigned int new_fl = 0;
 147
 148         if (binode->flags & BTRFS_INODE_SYNC)
 149                 new_fl |= S_SYNC;
 150         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 151                 new_fl |= S_IMMUTABLE;
 152         if (binode->flags & BTRFS_INODE_APPEND)
 153                 new_fl |= S_APPEND;
 154         if (binode->flags & BTRFS_INODE_NOATIME)
 155                 new_fl |= S_NOATIME;
 156         if (binode->flags & BTRFS_INODE_DIRSYNC)
 157                 new_fl |= S_DIRSYNC;
 158         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 159                 new_fl |= S_VERITY;
 160
 161         set_mask_bits(&inode->i_flags,
 162                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 163                       S_VERITY, new_fl);
 164 }
 165
 166 /*
 167  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 168  * the old and new flags are not conflicting
 169  */
 170 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 171 {
 172         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 173                       FS_NOATIME_FL | FS_NODUMP_FL | \
 174                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 175                       FS_NOCOMP_FL | FS_COMPR_FL |
 176                       FS_NOCOW_FL))
 177                 return -EOPNOTSUPP;
 178
 179         /* COMPR and NOCOMP on new/old are valid */
 180         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 181                 return -EINVAL;
 182
 183         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 184                 return -EINVAL;
 185
 186         /* NOCOW and compression options are mutually exclusive */
 187         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 188                 return -EINVAL;
 189         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 190                 return -EINVAL;
 191
 192         return 0;
 193 }
 194
 195 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 196                                     unsigned int flags)
 197 {
 198         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 199                 return -EPERM;
 200
 201         return 0;
 202 }
 203
 204 /*
 205  * Set flags/xflags from the internal inode flags. The remaining items of
 206  * fsxattr are zeroed.
 207  */
 208 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 209 {
 210         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 211
 212         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 213         return 0;
 214 }
 215
 216 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 217                        struct dentry *dentry, struct fileattr *fa)
 218 {
 219         struct inode *inode = d_inode(dentry);
 220         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 221         struct btrfs_inode *binode = BTRFS_I(inode);
 222         struct btrfs_root *root = binode->root;
 223         struct btrfs_trans_handle *trans;
 224         unsigned int fsflags, old_fsflags;
 225         int ret;
 226         const char *comp = NULL;
 227         u32 binode_flags;
 228
 229         if (btrfs_root_readonly(root))
 230                 return -EROFS;
 231
 232         if (fileattr_has_fsx(fa))
 233                 return -EOPNOTSUPP;
 234
 235         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 236         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 237         ret = check_fsflags(old_fsflags, fsflags);
 238         if (ret)
 239                 return ret;
 240
 241         ret = check_fsflags_compatible(fs_info, fsflags);
 242         if (ret)
 243                 return ret;
 244
 245         binode_flags = binode->flags;
 246         if (fsflags & FS_SYNC_FL)
 247                 binode_flags |= BTRFS_INODE_SYNC;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_SYNC;
 250         if (fsflags & FS_IMMUTABLE_FL)
 251                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 254         if (fsflags & FS_APPEND_FL)
 255                 binode_flags |= BTRFS_INODE_APPEND;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_APPEND;
 258         if (fsflags & FS_NODUMP_FL)
 259                 binode_flags |= BTRFS_INODE_NODUMP;
 260         else
 261                 binode_flags &= ~BTRFS_INODE_NODUMP;
 262         if (fsflags & FS_NOATIME_FL)
 263                 binode_flags |= BTRFS_INODE_NOATIME;
 264         else
 265                 binode_flags &= ~BTRFS_INODE_NOATIME;
 266
 267         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 268         if (!fa->flags_valid) {
 269                 /* 1 item for the inode */
 270                 trans = btrfs_start_transaction(root, 1);
 271                 if (IS_ERR(trans))
 272                         return PTR_ERR(trans);
 273                 goto update_flags;
 274         }
 275
 276         if (fsflags & FS_DIRSYNC_FL)
 277                 binode_flags |= BTRFS_INODE_DIRSYNC;
 278         else
 279                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 280         if (fsflags & FS_NOCOW_FL) {
 281                 if (S_ISREG(inode->i_mode)) {
 282                         /*
 283                          * It's safe to turn csums off here, no extents exist.
 284                          * Otherwise we want the flag to reflect the real COW
 285                          * status of the file and will not set it.
 286                          */
 287                         if (inode->i_size == 0)
 288                                 binode_flags |= BTRFS_INODE_NODATACOW |
 289                                                 BTRFS_INODE_NODATASUM;
 290                 } else {
 291                         binode_flags |= BTRFS_INODE_NODATACOW;
 292                 }
 293         } else {
 294                 /*
 295                  * Revert back under same assumptions as above
 296                  */
 297                 if (S_ISREG(inode->i_mode)) {
 298                         if (inode->i_size == 0)
 299                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 300                                                   BTRFS_INODE_NODATASUM);
 301                 } else {
 302                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 303                 }
 304         }
 305
 306         /*
 307          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 308          * flag may be changed automatically if compression code won't make
 309          * things smaller.
 310          */
 311         if (fsflags & FS_NOCOMP_FL) {
 312                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 313                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 314         } else if (fsflags & FS_COMPR_FL) {
 315
 316                 if (IS_SWAPFILE(inode))
 317                         return -ETXTBSY;
 318
 319                 binode_flags |= BTRFS_INODE_COMPRESS;
 320                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 321
 322                 comp = btrfs_compress_type2str(fs_info->compress_type);
 323                 if (!comp || comp[0] == 0)
 324                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 325         } else {
 326                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 327         }
 328
 329         /*
 330          * 1 for inode item
 331          * 2 for properties
 332          */
 333         trans = btrfs_start_transaction(root, 3);
 334         if (IS_ERR(trans))
 335                 return PTR_ERR(trans);
 336
 337         if (comp) {
 338                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 339                                      strlen(comp), 0);
 340                 if (ret) {
 341                         btrfs_abort_transaction(trans, ret);
 342                         goto out_end_trans;
 343                 }
 344         } else {
 345                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 346                                      0, 0);
 347                 if (ret && ret != -ENODATA) {
 348                         btrfs_abort_transaction(trans, ret);
 349                         goto out_end_trans;
 350                 }
 351         }
 352
 353 update_flags:
 354         binode->flags = binode_flags;
 355         btrfs_sync_inode_flags_to_i_flags(inode);
 356         inode_inc_iversion(inode);
 357         inode->i_ctime = current_time(inode);
 358         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 359
 360  out_end_trans:
 361         btrfs_end_transaction(trans);
 362         return ret;
 363 }
 364
 365 /*
 366  * Start exclusive operation @type, return true on success
 367  */
 368 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 369                         enum btrfs_exclusive_operation type)
 370 {
 371         bool ret = false;
 372
 373         spin_lock(&fs_info->super_lock);
 374         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 375                 fs_info->exclusive_operation = type;
 376                 ret = true;
 377         }
 378         spin_unlock(&fs_info->super_lock);
 379
 380         return ret;
 381 }
 382
 383 /*
 384  * Conditionally allow to enter the exclusive operation in case it's compatible
 385  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 386  * btrfs_exclop_finish.
 387  *
 388  * Compatibility:
 389  * - the same type is already running
 390  * - when trying to add a device and balance has been paused
 391  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 392  *   must check the condition first that would allow none -> @type
 393  */
 394 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 395                                  enum btrfs_exclusive_operation type)
 396 {
 397         spin_lock(&fs_info->super_lock);
 398         if (fs_info->exclusive_operation == type ||
 399             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 400              type == BTRFS_EXCLOP_DEV_ADD))
 401                 return true;
 402
 403         spin_unlock(&fs_info->super_lock);
 404         return false;
 405 }
 406
 407 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 408 {
 409         spin_unlock(&fs_info->super_lock);
 410 }
 411
 412 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 413 {
 414         spin_lock(&fs_info->super_lock);
 415         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 416         spin_unlock(&fs_info->super_lock);
 417         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 418 }
 419
 420 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 421                           enum btrfs_exclusive_operation op)
 422 {
 423         switch (op) {
 424         case BTRFS_EXCLOP_BALANCE_PAUSED:
 425                 spin_lock(&fs_info->super_lock);
 426                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 427                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD);
 428                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 429                 spin_unlock(&fs_info->super_lock);
 430                 break;
 431         case BTRFS_EXCLOP_BALANCE:
 432                 spin_lock(&fs_info->super_lock);
 433                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 434                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 435                 spin_unlock(&fs_info->super_lock);
 436                 break;
 437         default:
 438                 btrfs_warn(fs_info,
 439                         "invalid exclop balance operation %d requested", op);
 440         }
 441 }
 442
 443 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 444 {
 445         struct inode *inode = file_inode(file);
 446
 447         return put_user(inode->i_generation, arg);
 448 }
 449
 450 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 451                                         void __user *arg)
 452 {
 453         struct btrfs_device *device;
 454         struct request_queue *q;
 455         struct fstrim_range range;
 456         u64 minlen = ULLONG_MAX;
 457         u64 num_devices = 0;
 458         int ret;
 459
 460         if (!capable(CAP_SYS_ADMIN))
 461                 return -EPERM;
 462
 463         /*
 464          * btrfs_trim_block_group() depends on space cache, which is not
 465          * available in zoned filesystem. So, disallow fitrim on a zoned
 466          * filesystem for now.
 467          */
 468         if (btrfs_is_zoned(fs_info))
 469                 return -EOPNOTSUPP;
 470
 471         /*
 472          * If the fs is mounted with nologreplay, which requires it to be
 473          * mounted in RO mode as well, we can not allow discard on free space
 474          * inside block groups, because log trees refer to extents that are not
 475          * pinned in a block group's free space cache (pinning the extents is
 476          * precisely the first phase of replaying a log tree).
 477          */
 478         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 479                 return -EROFS;
 480
 481         rcu_read_lock();
 482         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 483                                 dev_list) {
 484                 if (!device->bdev)
 485                         continue;
 486                 q = bdev_get_queue(device->bdev);
 487                 if (blk_queue_discard(q)) {
 488                         num_devices++;
 489                         minlen = min_t(u64, q->limits.discard_granularity,
 490                                      minlen);
 491                 }
 492         }
 493         rcu_read_unlock();
 494
 495         if (!num_devices)
 496                 return -EOPNOTSUPP;
 497         if (copy_from_user(&range, arg, sizeof(range)))
 498                 return -EFAULT;
 499
 500         /*
 501          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 502          * block group is in the logical address space, which can be any
 503          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 504          */
 505         if (range.len < fs_info->sb->s_blocksize)
 506                 return -EINVAL;
 507
 508         range.minlen = max(range.minlen, minlen);
 509         ret = btrfs_trim_fs(fs_info, &range);
 510         if (ret < 0)
 511                 return ret;
 512
 513         if (copy_to_user(arg, &range, sizeof(range)))
 514                 return -EFAULT;
 515
 516         return 0;
 517 }
 518
 519 int __pure btrfs_is_empty_uuid(u8 *uuid)
 520 {
 521         int i;
 522
 523         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 524                 if (uuid[i])
 525                         return 0;
 526         }
 527         return 1;
 528 }
 529
 530 static noinline int create_subvol(struct user_namespace *mnt_userns,
 531                                   struct inode *dir, struct dentry *dentry,
 532                                   const char *name, int namelen,
 533                                   struct btrfs_qgroup_inherit *inherit)
 534 {
 535         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 536         struct btrfs_trans_handle *trans;
 537         struct btrfs_key key;
 538         struct btrfs_root_item *root_item;
 539         struct btrfs_inode_item *inode_item;
 540         struct extent_buffer *leaf;
 541         struct btrfs_root *root = BTRFS_I(dir)->root;
 542         struct btrfs_root *new_root;
 543         struct btrfs_block_rsv block_rsv;
 544         struct timespec64 cur_time = current_time(dir);
 545         struct inode *inode;
 546         int ret;
 547         dev_t anon_dev = 0;
 548         u64 objectid;
 549         u64 index = 0;
 550
 551         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 552         if (!root_item)
 553                 return -ENOMEM;
 554
 555         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 556         if (ret)
 557                 goto fail_free;
 558
 559         ret = get_anon_bdev(&anon_dev);
 560         if (ret < 0)
 561                 goto fail_free;
 562
 563         /*
 564          * Don't create subvolume whose level is not zero. Or qgroup will be
 565          * screwed up since it assumes subvolume qgroup's level to be 0.
 566          */
 567         if (btrfs_qgroup_level(objectid)) {
 568                 ret = -ENOSPC;
 569                 goto fail_free;
 570         }
 571
 572         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 573         /*
 574          * The same as the snapshot creation, please see the comment
 575          * of create_snapshot().
 576          */
 577         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 578         if (ret)
 579                 goto fail_free;
 580
 581         trans = btrfs_start_transaction(root, 0);
 582         if (IS_ERR(trans)) {
 583                 ret = PTR_ERR(trans);
 584                 btrfs_subvolume_release_metadata(root, &block_rsv);
 585                 goto fail_free;
 586         }
 587         trans->block_rsv = &block_rsv;
 588         trans->bytes_reserved = block_rsv.size;
 589
 590         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 591         if (ret)
 592                 goto fail;
 593
 594         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 595                                       BTRFS_NESTING_NORMAL);
 596         if (IS_ERR(leaf)) {
 597                 ret = PTR_ERR(leaf);
 598                 goto fail;
 599         }
 600
 601         btrfs_mark_buffer_dirty(leaf);
 602
 603         inode_item = &root_item->inode;
 604         btrfs_set_stack_inode_generation(inode_item, 1);
 605         btrfs_set_stack_inode_size(inode_item, 3);
 606         btrfs_set_stack_inode_nlink(inode_item, 1);
 607         btrfs_set_stack_inode_nbytes(inode_item,
 608                                      fs_info->nodesize);
 609         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 610
 611         btrfs_set_root_flags(root_item, 0);
 612         btrfs_set_root_limit(root_item, 0);
 613         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 614
 615         btrfs_set_root_bytenr(root_item, leaf->start);
 616         btrfs_set_root_generation(root_item, trans->transid);
 617         btrfs_set_root_level(root_item, 0);
 618         btrfs_set_root_refs(root_item, 1);
 619         btrfs_set_root_used(root_item, leaf->len);
 620         btrfs_set_root_last_snapshot(root_item, 0);
 621
 622         btrfs_set_root_generation_v2(root_item,
 623                         btrfs_root_generation(root_item));
 624         generate_random_guid(root_item->uuid);
 625         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 626         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 627         root_item->ctime = root_item->otime;
 628         btrfs_set_root_ctransid(root_item, trans->transid);
 629         btrfs_set_root_otransid(root_item, trans->transid);
 630
 631         btrfs_tree_unlock(leaf);
 632
 633         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 634
 635         key.objectid = objectid;
 636         key.offset = 0;
 637         key.type = BTRFS_ROOT_ITEM_KEY;
 638         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 639                                 root_item);
 640         if (ret) {
 641                 /*
 642                  * Since we don't abort the transaction in this case, free the
 643                  * tree block so that we don't leak space and leave the
 644                  * filesystem in an inconsistent state (an extent item in the
 645                  * extent tree with a backreference for a root that does not
 646                  * exists).
 647                  */
 648                 btrfs_tree_lock(leaf);
 649                 btrfs_clean_tree_block(leaf);
 650                 btrfs_tree_unlock(leaf);
 651                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 652                 free_extent_buffer(leaf);
 653                 goto fail;
 654         }
 655
 656         free_extent_buffer(leaf);
 657         leaf = NULL;
 658
 659         key.offset = (u64)-1;
 660         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 661         if (IS_ERR(new_root)) {
 662                 free_anon_bdev(anon_dev);
 663                 ret = PTR_ERR(new_root);
 664                 btrfs_abort_transaction(trans, ret);
 665                 goto fail;
 666         }
 667         /* Freeing will be done in btrfs_put_root() of new_root */
 668         anon_dev = 0;
 669
 670         ret = btrfs_record_root_in_trans(trans, new_root);
 671         if (ret) {
 672                 btrfs_put_root(new_root);
 673                 btrfs_abort_transaction(trans, ret);
 674                 goto fail;
 675         }
 676
 677         ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns);
 678         btrfs_put_root(new_root);
 679         if (ret) {
 680                 /* We potentially lose an unused inode item here */
 681                 btrfs_abort_transaction(trans, ret);
 682                 goto fail;
 683         }
 684
 685         /*
 686          * insert the directory item
 687          */
 688         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 689         if (ret) {
 690                 btrfs_abort_transaction(trans, ret);
 691                 goto fail;
 692         }
 693
 694         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 695                                     BTRFS_FT_DIR, index);
 696         if (ret) {
 697                 btrfs_abort_transaction(trans, ret);
 698                 goto fail;
 699         }
 700
 701         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 702         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 703         if (ret) {
 704                 btrfs_abort_transaction(trans, ret);
 705                 goto fail;
 706         }
 707
 708         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 709                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 710         if (ret) {
 711                 btrfs_abort_transaction(trans, ret);
 712                 goto fail;
 713         }
 714
 715         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 716                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 717         if (ret)
 718                 btrfs_abort_transaction(trans, ret);
 719
 720 fail:
 721         kfree(root_item);
 722         trans->block_rsv = NULL;
 723         trans->bytes_reserved = 0;
 724         btrfs_subvolume_release_metadata(root, &block_rsv);
 725
 726         if (ret)
 727                 btrfs_end_transaction(trans);
 728         else
 729                 ret = btrfs_commit_transaction(trans);
 730
 731         if (!ret) {
 732                 inode = btrfs_lookup_dentry(dir, dentry);
 733                 if (IS_ERR(inode))
 734                         return PTR_ERR(inode);
 735                 d_instantiate(dentry, inode);
 736         }
 737         return ret;
 738
 739 fail_free:
 740         if (anon_dev)
 741                 free_anon_bdev(anon_dev);
 742         kfree(root_item);
 743         return ret;
 744 }
 745
 746 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 747                            struct dentry *dentry, bool readonly,
 748                            struct btrfs_qgroup_inherit *inherit)
 749 {
 750         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 751         struct inode *inode;
 752         struct btrfs_pending_snapshot *pending_snapshot;
 753         struct btrfs_trans_handle *trans;
 754         int ret;
 755
 756         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 757                 return -EINVAL;
 758
 759         if (atomic_read(&root->nr_swapfiles)) {
 760                 btrfs_warn(fs_info,
 761                            "cannot snapshot subvolume with active swapfile");
 762                 return -ETXTBSY;
 763         }
 764
 765         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 766         if (!pending_snapshot)
 767                 return -ENOMEM;
 768
 769         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 770         if (ret < 0)
 771                 goto free_pending;
 772         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 773                         GFP_KERNEL);
 774         pending_snapshot->path = btrfs_alloc_path();
 775         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 776                 ret = -ENOMEM;
 777                 goto free_pending;
 778         }
 779
 780         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 781                              BTRFS_BLOCK_RSV_TEMP);
 782         /*
 783          * 1 - parent dir inode
 784          * 2 - dir entries
 785          * 1 - root item
 786          * 2 - root ref/backref
 787          * 1 - root of snapshot
 788          * 1 - UUID item
 789          */
 790         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 791                                         &pending_snapshot->block_rsv, 8,
 792                                         false);
 793         if (ret)
 794                 goto free_pending;
 795
 796         pending_snapshot->dentry = dentry;
 797         pending_snapshot->root = root;
 798         pending_snapshot->readonly = readonly;
 799         pending_snapshot->dir = dir;
 800         pending_snapshot->inherit = inherit;
 801
 802         trans = btrfs_start_transaction(root, 0);
 803         if (IS_ERR(trans)) {
 804                 ret = PTR_ERR(trans);
 805                 goto fail;
 806         }
 807
 808         trans->pending_snapshot = pending_snapshot;
 809
 810         ret = btrfs_commit_transaction(trans);
 811         if (ret)
 812                 goto fail;
 813
 814         ret = pending_snapshot->error;
 815         if (ret)
 816                 goto fail;
 817
 818         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 819         if (ret)
 820                 goto fail;
 821
 822         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 823         if (IS_ERR(inode)) {
 824                 ret = PTR_ERR(inode);
 825                 goto fail;
 826         }
 827
 828         d_instantiate(dentry, inode);
 829         ret = 0;
 830         pending_snapshot->anon_dev = 0;
 831 fail:
 832         /* Prevent double freeing of anon_dev */
 833         if (ret && pending_snapshot->snap)
 834                 pending_snapshot->snap->anon_dev = 0;
 835         btrfs_put_root(pending_snapshot->snap);
 836         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 837 free_pending:
 838         if (pending_snapshot->anon_dev)
 839                 free_anon_bdev(pending_snapshot->anon_dev);
 840         kfree(pending_snapshot->root_item);
 841         btrfs_free_path(pending_snapshot->path);
 842         kfree(pending_snapshot);
 843
 844         return ret;
 845 }
 846
 847 /*  copy of may_delete in fs/namei.c()
 848  *      Check whether we can remove a link victim from directory dir, check
 849  *  whether the type of victim is right.
 850  *  1. We can't do it if dir is read-only (done in permission())
 851  *  2. We should have write and exec permissions on dir
 852  *  3. We can't remove anything from append-only dir
 853  *  4. We can't do anything with immutable dir (done in permission())
 854  *  5. If the sticky bit on dir is set we should either
 855  *      a. be owner of dir, or
 856  *      b. be owner of victim, or
 857  *      c. have CAP_FOWNER capability
 858  *  6. If the victim is append-only or immutable we can't do anything with
 859  *     links pointing to it.
 860  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 861  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 862  *  9. We can't remove a root or mountpoint.
 863  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 864  *     nfs_async_unlink().
 865  */
 866
 867 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 868                             struct inode *dir, struct dentry *victim, int isdir)
 869 {
 870         int error;
 871
 872         if (d_really_is_negative(victim))
 873                 return -ENOENT;
 874
 875         BUG_ON(d_inode(victim->d_parent) != dir);
 876         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 877
 878         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 879         if (error)
 880                 return error;
 881         if (IS_APPEND(dir))
 882                 return -EPERM;
 883         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 884             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 885             IS_SWAPFILE(d_inode(victim)))
 886                 return -EPERM;
 887         if (isdir) {
 888                 if (!d_is_dir(victim))
 889                         return -ENOTDIR;
 890                 if (IS_ROOT(victim))
 891                         return -EBUSY;
 892         } else if (d_is_dir(victim))
 893                 return -EISDIR;
 894         if (IS_DEADDIR(dir))
 895                 return -ENOENT;
 896         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 897                 return -EBUSY;
 898         return 0;
 899 }
 900
 901 /* copy of may_create in fs/namei.c() */
 902 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 903                                    struct inode *dir, struct dentry *child)
 904 {
 905         if (d_really_is_positive(child))
 906                 return -EEXIST;
 907         if (IS_DEADDIR(dir))
 908                 return -ENOENT;
 909         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 910                 return -EOVERFLOW;
 911         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 912 }
 913
 914 /*
 915  * Create a new subvolume below @parent.  This is largely modeled after
 916  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 917  * inside this filesystem so it's quite a bit simpler.
 918  */
 919 static noinline int btrfs_mksubvol(const struct path *parent,
 920                                    struct user_namespace *mnt_userns,
 921                                    const char *name, int namelen,
 922                                    struct btrfs_root *snap_src,
 923                                    bool readonly,
 924                                    struct btrfs_qgroup_inherit *inherit)
 925 {
 926         struct inode *dir = d_inode(parent->dentry);
 927         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 928         struct dentry *dentry;
 929         int error;
 930
 931         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 932         if (error == -EINTR)
 933                 return error;
 934
 935         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 936         error = PTR_ERR(dentry);
 937         if (IS_ERR(dentry))
 938                 goto out_unlock;
 939
 940         error = btrfs_may_create(mnt_userns, dir, dentry);
 941         if (error)
 942                 goto out_dput;
 943
 944         /*
 945          * even if this name doesn't exist, we may get hash collisions.
 946          * check for them now when we can safely fail
 947          */
 948         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 949                                                dir->i_ino, name,
 950                                                namelen);
 951         if (error)
 952                 goto out_dput;
 953
 954         down_read(&fs_info->subvol_sem);
 955
 956         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 957                 goto out_up_read;
 958
 959         if (snap_src)
 960                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 961         else
 962                 error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit);
 963
 964         if (!error)
 965                 fsnotify_mkdir(dir, dentry);
 966 out_up_read:
 967         up_read(&fs_info->subvol_sem);
 968 out_dput:
 969         dput(dentry);
 970 out_unlock:
 971         btrfs_inode_unlock(dir, 0);
 972         return error;
 973 }
 974
 975 static noinline int btrfs_mksnapshot(const struct path *parent,
 976                                    struct user_namespace *mnt_userns,
 977                                    const char *name, int namelen,
 978                                    struct btrfs_root *root,
 979                                    bool readonly,
 980                                    struct btrfs_qgroup_inherit *inherit)
 981 {
 982         int ret;
 983         bool snapshot_force_cow = false;
 984
 985         /*
 986          * Force new buffered writes to reserve space even when NOCOW is
 987          * possible. This is to avoid later writeback (running dealloc) to
 988          * fallback to COW mode and unexpectedly fail with ENOSPC.
 989          */
 990         btrfs_drew_read_lock(&root->snapshot_lock);
 991
 992         ret = btrfs_start_delalloc_snapshot(root, false);
 993         if (ret)
 994                 goto out;
 995
 996         /*
 997          * All previous writes have started writeback in NOCOW mode, so now
 998          * we force future writes to fallback to COW mode during snapshot
 999          * creation.
1000          */
1001         atomic_inc(&root->snapshot_force_cow);
1002         snapshot_force_cow = true;
1003
1004         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1005
1006         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1007                              root, readonly, inherit);
1008 out:
1009         if (snapshot_force_cow)
1010                 atomic_dec(&root->snapshot_force_cow);
1011         btrfs_drew_read_unlock(&root->snapshot_lock);
1012         return ret;
1013 }
1014
1015 /*
1016  * Defrag specific helper to get an extent map.
1017  *
1018  * Differences between this and btrfs_get_extent() are:
1019  *
1020  * - No extent_map will be added to inode->extent_tree
1021  *   To reduce memory usage in the long run.
1022  *
1023  * - Extra optimization to skip file extents older than @newer_than
1024  *   By using btrfs_search_forward() we can skip entire file ranges that
1025  *   have extents created in past transactions, because btrfs_search_forward()
1026  *   will not visit leaves and nodes with a generation smaller than given
1027  *   minimal generation threshold (@newer_than).
1028  *
1029  * Return valid em if we find a file extent matching the requirement.
1030  * Return NULL if we can not find a file extent matching the requirement.
1031  *
1032  * Return ERR_PTR() for error.
1033  */
1034 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1035                                             u64 start, u64 newer_than)
1036 {
1037         struct btrfs_root *root = inode->root;
1038         struct btrfs_file_extent_item *fi;
1039         struct btrfs_path path = { 0 };
1040         struct extent_map *em;
1041         struct btrfs_key key;
1042         u64 ino = btrfs_ino(inode);
1043         int ret;
1044
1045         em = alloc_extent_map();
1046         if (!em) {
1047                 ret = -ENOMEM;
1048                 goto err;
1049         }
1050
1051         key.objectid = ino;
1052         key.type = BTRFS_EXTENT_DATA_KEY;
1053         key.offset = start;
1054
1055         if (newer_than) {
1056                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1057                 if (ret < 0)
1058                         goto err;
1059                 /* Can't find anything newer */
1060                 if (ret > 0)
1061                         goto not_found;
1062         } else {
1063                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1064                 if (ret < 0)
1065                         goto err;
1066         }
1067         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1068                 /*
1069                  * If btrfs_search_slot() makes path to point beyond nritems,
1070                  * we should not have an empty leaf, as this inode must at
1071                  * least have its INODE_ITEM.
1072                  */
1073                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1074                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1075         }
1076         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1077         /* Perfect match, no need to go one slot back */
1078         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1079             key.offset == start)
1080                 goto iterate;
1081
1082         /* We didn't find a perfect match, needs to go one slot back */
1083         if (path.slots[0] > 0) {
1084                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1085                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1086                         path.slots[0]--;
1087         }
1088
1089 iterate:
1090         /* Iterate through the path to find a file extent covering @start */
1091         while (true) {
1092                 u64 extent_end;
1093
1094                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1095                         goto next;
1096
1097                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1098
1099                 /*
1100                  * We may go one slot back to INODE_REF/XATTR item, then
1101                  * need to go forward until we reach an EXTENT_DATA.
1102                  * But we should still has the correct ino as key.objectid.
1103                  */
1104                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1105                         goto next;
1106
1107                 /* It's beyond our target range, definitely not extent found */
1108                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1109                         goto not_found;
1110
1111                 /*
1112                  *      |       |<- File extent ->|
1113                  *      \- start
1114                  *
1115                  * This means there is a hole between start and key.offset.
1116                  */
1117                 if (key.offset > start) {
1118                         em->start = start;
1119                         em->orig_start = start;
1120                         em->block_start = EXTENT_MAP_HOLE;
1121                         em->len = key.offset - start;
1122                         break;
1123                 }
1124
1125                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1126                                     struct btrfs_file_extent_item);
1127                 extent_end = btrfs_file_extent_end(&path);
1128
1129                 /*
1130                  *      |<- file extent ->|     |
1131                  *                              \- start
1132                  *
1133                  * We haven't reached start, search next slot.
1134                  */
1135                 if (extent_end <= start)
1136                         goto next;
1137
1138                 /* Now this extent covers @start, convert it to em */
1139                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1140                 break;
1141 next:
1142                 ret = btrfs_next_item(root, &path);
1143                 if (ret < 0)
1144                         goto err;
1145                 if (ret > 0)
1146                         goto not_found;
1147         }
1148         btrfs_release_path(&path);
1149         return em;
1150
1151 not_found:
1152         btrfs_release_path(&path);
1153         free_extent_map(em);
1154         return NULL;
1155
1156 err:
1157         btrfs_release_path(&path);
1158         free_extent_map(em);
1159         return ERR_PTR(ret);
1160 }
1161
1162 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1163                                                u64 newer_than, bool locked)
1164 {
1165         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1166         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1167         struct extent_map *em;
1168         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1169
1170         /*
1171          * hopefully we have this extent in the tree already, try without
1172          * the full extent lock
1173          */
1174         read_lock(&em_tree->lock);
1175         em = lookup_extent_mapping(em_tree, start, sectorsize);
1176         read_unlock(&em_tree->lock);
1177
1178         /*
1179          * We can get a merged extent, in that case, we need to re-search
1180          * tree to get the original em for defrag.
1181          *
1182          * If @newer_than is 0 or em::generation < newer_than, we can trust
1183          * this em, as either we don't care about the generation, or the
1184          * merged extent map will be rejected anyway.
1185          */
1186         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1187             newer_than && em->generation >= newer_than) {
1188                 free_extent_map(em);
1189                 em = NULL;
1190         }
1191
1192         if (!em) {
1193                 struct extent_state *cached = NULL;
1194                 u64 end = start + sectorsize - 1;
1195
1196                 /* get the big lock and read metadata off disk */
1197                 if (!locked)
1198                         lock_extent_bits(io_tree, start, end, &cached);
1199                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1200                 if (!locked)
1201                         unlock_extent_cached(io_tree, start, end, &cached);
1202
1203                 if (IS_ERR(em))
1204                         return NULL;
1205         }
1206
1207         return em;
1208 }
1209
1210 static u32 get_extent_max_capacity(const struct extent_map *em)
1211 {
1212         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1213                 return BTRFS_MAX_COMPRESSED;
1214         return BTRFS_MAX_EXTENT_SIZE;
1215 }
1216
1217 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1218                                      bool locked)
1219 {
1220         struct extent_map *next;
1221         bool ret = false;
1222
1223         /* this is the last extent */
1224         if (em->start + em->len >= i_size_read(inode))
1225                 return false;
1226
1227         /*
1228          * We want to check if the next extent can be merged with the current
1229          * one, which can be an extent created in a past generation, so we pass
1230          * a minimum generation of 0 to defrag_lookup_extent().
1231          */
1232         next = defrag_lookup_extent(inode, em->start + em->len, 0, locked);
1233         /* No more em or hole */
1234         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1235                 goto out;
1236         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1237                 goto out;
1238         /*
1239          * If the next extent is at its max capacity, defragging current extent
1240          * makes no sense, as the total number of extents won't change.
1241          */
1242         if (next->len >= get_extent_max_capacity(em))
1243                 goto out;
1244         ret = true;
1245 out:
1246         free_extent_map(next);
1247         return ret;
1248 }
1249
1250 /*
1251  * Prepare one page to be defragged.
1252  *
1253  * This will ensure:
1254  *
1255  * - Returned page is locked and has been set up properly.
1256  * - No ordered extent exists in the page.
1257  * - The page is uptodate.
1258  *
1259  * NOTE: Caller should also wait for page writeback after the cluster is
1260  * prepared, here we don't do writeback wait for each page.
1261  */
1262 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1263                                             pgoff_t index)
1264 {
1265         struct address_space *mapping = inode->vfs_inode.i_mapping;
1266         gfp_t mask = btrfs_alloc_write_mask(mapping);
1267         u64 page_start = (u64)index << PAGE_SHIFT;
1268         u64 page_end = page_start + PAGE_SIZE - 1;
1269         struct extent_state *cached_state = NULL;
1270         struct page *page;
1271         int ret;
1272
1273 again:
1274         page = find_or_create_page(mapping, index, mask);
1275         if (!page)
1276                 return ERR_PTR(-ENOMEM);
1277
1278         /*
1279          * Since we can defragment files opened read-only, we can encounter
1280          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1281          * can't do I/O using huge pages yet, so return an error for now.
1282          * Filesystem transparent huge pages are typically only used for
1283          * executables that explicitly enable them, so this isn't very
1284          * restrictive.
1285          */
1286         if (PageCompound(page)) {
1287                 unlock_page(page);
1288                 put_page(page);
1289                 return ERR_PTR(-ETXTBSY);
1290         }
1291
1292         ret = set_page_extent_mapped(page);
1293         if (ret < 0) {
1294                 unlock_page(page);
1295                 put_page(page);
1296                 return ERR_PTR(ret);
1297         }
1298
1299         /* Wait for any existing ordered extent in the range */
1300         while (1) {
1301                 struct btrfs_ordered_extent *ordered;
1302
1303                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1304                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1305                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1306                                      &cached_state);
1307                 if (!ordered)
1308                         break;
1309
1310                 unlock_page(page);
1311                 btrfs_start_ordered_extent(ordered, 1);
1312                 btrfs_put_ordered_extent(ordered);
1313                 lock_page(page);
1314                 /*
1315                  * We unlocked the page above, so we need check if it was
1316                  * released or not.
1317                  */
1318                 if (page->mapping != mapping || !PagePrivate(page)) {
1319                         unlock_page(page);
1320                         put_page(page);
1321                         goto again;
1322                 }
1323         }
1324
1325         /*
1326          * Now the page range has no ordered extent any more.  Read the page to
1327          * make it uptodate.
1328          */
1329         if (!PageUptodate(page)) {
1330                 btrfs_readpage(NULL, page);
1331                 lock_page(page);
1332                 if (page->mapping != mapping || !PagePrivate(page)) {
1333                         unlock_page(page);
1334                         put_page(page);
1335                         goto again;
1336                 }
1337                 if (!PageUptodate(page)) {
1338                         unlock_page(page);
1339                         put_page(page);
1340                         return ERR_PTR(-EIO);
1341                 }
1342         }
1343         return page;
1344 }
1345
1346 struct defrag_target_range {
1347         struct list_head list;
1348         u64 start;
1349         u64 len;
1350 };
1351
1352 /*
1353  * Collect all valid target extents.
1354  *
1355  * @start:         file offset to lookup
1356  * @len:           length to lookup
1357  * @extent_thresh: file extent size threshold, any extent size >= this value
1358  *                 will be ignored
1359  * @newer_than:    only defrag extents newer than this value
1360  * @do_compress:   whether the defrag is doing compression
1361  *                 if true, @extent_thresh will be ignored and all regular
1362  *                 file extents meeting @newer_than will be targets.
1363  * @locked:        if the range has already held extent lock
1364  * @target_list:   list of targets file extents
1365  */
1366 static int defrag_collect_targets(struct btrfs_inode *inode,
1367                                   u64 start, u64 len, u32 extent_thresh,
1368                                   u64 newer_than, bool do_compress,
1369                                   bool locked, struct list_head *target_list,
1370                                   u64 *last_scanned_ret)
1371 {
1372         bool last_is_target = false;
1373         u64 cur = start;
1374         int ret = 0;
1375
1376         while (cur < start + len) {
1377                 struct extent_map *em;
1378                 struct defrag_target_range *new;
1379                 bool next_mergeable = true;
1380                 u64 range_len;
1381
1382                 last_is_target = false;
1383                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1384                                           newer_than, locked);
1385                 if (!em)
1386                         break;
1387
1388                 /* Skip hole/inline/preallocated extents */
1389                 if (em->block_start >= EXTENT_MAP_LAST_BYTE ||
1390                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1391                         goto next;
1392
1393                 /* Skip older extent */
1394                 if (em->generation < newer_than)
1395                         goto next;
1396
1397                 /* This em is under writeback, no need to defrag */
1398                 if (em->generation == (u64)-1)
1399                         goto next;
1400
1401                 /*
1402                  * Our start offset might be in the middle of an existing extent
1403                  * map, so take that into account.
1404                  */
1405                 range_len = em->len - (cur - em->start);
1406                 /*
1407                  * If this range of the extent map is already flagged for delalloc,
1408                  * skip it, because:
1409                  *
1410                  * 1) We could deadlock later, when trying to reserve space for
1411                  *    delalloc, because in case we can't immediately reserve space
1412                  *    the flusher can start delalloc and wait for the respective
1413                  *    ordered extents to complete. The deadlock would happen
1414                  *    because we do the space reservation while holding the range
1415                  *    locked, and starting writeback, or finishing an ordered
1416                  *    extent, requires locking the range;
1417                  *
1418                  * 2) If there's delalloc there, it means there's dirty pages for
1419                  *    which writeback has not started yet (we clean the delalloc
1420                  *    flag when starting writeback and after creating an ordered
1421                  *    extent). If we mark pages in an adjacent range for defrag,
1422                  *    then we will have a larger contiguous range for delalloc,
1423                  *    very likely resulting in a larger extent after writeback is
1424                  *    triggered (except in a case of free space fragmentation).
1425                  */
1426                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1427                                    EXTENT_DELALLOC, 0, NULL))
1428                         goto next;
1429
1430                 /*
1431                  * For do_compress case, we want to compress all valid file
1432                  * extents, thus no @extent_thresh or mergeable check.
1433                  */
1434                 if (do_compress)
1435                         goto add;
1436
1437                 /* Skip too large extent */
1438                 if (range_len >= extent_thresh)
1439                         goto next;
1440
1441                 /*
1442                  * Skip extents already at its max capacity, this is mostly for
1443                  * compressed extents, which max cap is only 128K.
1444                  */
1445                 if (em->len >= get_extent_max_capacity(em))
1446                         goto next;
1447
1448                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1449                                                           locked);
1450                 if (!next_mergeable) {
1451                         struct defrag_target_range *last;
1452
1453                         /* Empty target list, no way to merge with last entry */
1454                         if (list_empty(target_list))
1455                                 goto next;
1456                         last = list_entry(target_list->prev,
1457                                           struct defrag_target_range, list);
1458                         /* Not mergeable with last entry */
1459                         if (last->start + last->len != cur)
1460                                 goto next;
1461
1462                         /* Mergeable, fall through to add it to @target_list. */
1463                 }
1464
1465 add:
1466                 last_is_target = true;
1467                 range_len = min(extent_map_end(em), start + len) - cur;
1468                 /*
1469                  * This one is a good target, check if it can be merged into
1470                  * last range of the target list.
1471                  */
1472                 if (!list_empty(target_list)) {
1473                         struct defrag_target_range *last;
1474
1475                         last = list_entry(target_list->prev,
1476                                           struct defrag_target_range, list);
1477                         ASSERT(last->start + last->len <= cur);
1478                         if (last->start + last->len == cur) {
1479                                 /* Mergeable, enlarge the last entry */
1480                                 last->len += range_len;
1481                                 goto next;
1482                         }
1483                         /* Fall through to allocate a new entry */
1484                 }
1485
1486                 /* Allocate new defrag_target_range */
1487                 new = kmalloc(sizeof(*new), GFP_NOFS);
1488                 if (!new) {
1489                         free_extent_map(em);
1490                         ret = -ENOMEM;
1491                         break;
1492                 }
1493                 new->start = cur;
1494                 new->len = range_len;
1495                 list_add_tail(&new->list, target_list);
1496
1497 next:
1498                 cur = extent_map_end(em);
1499                 free_extent_map(em);
1500         }
1501         if (ret < 0) {
1502                 struct defrag_target_range *entry;
1503                 struct defrag_target_range *tmp;
1504
1505                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1506                         list_del_init(&entry->list);
1507                         kfree(entry);
1508                 }
1509         }
1510         if (!ret && last_scanned_ret) {
1511                 /*
1512                  * If the last extent is not a target, the caller can skip to
1513                  * the end of that extent.
1514                  * Otherwise, we can only go the end of the specified range.
1515                  */
1516                 if (!last_is_target)
1517                         *last_scanned_ret = max(cur, *last_scanned_ret);
1518                 else
1519                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1520         }
1521         return ret;
1522 }
1523
1524 #define CLUSTER_SIZE    (SZ_256K)
1525
1526 /*
1527  * Defrag one contiguous target range.
1528  *
1529  * @inode:      target inode
1530  * @target:     target range to defrag
1531  * @pages:      locked pages covering the defrag range
1532  * @nr_pages:   number of locked pages
1533  *
1534  * Caller should ensure:
1535  *
1536  * - Pages are prepared
1537  *   Pages should be locked, no ordered extent in the pages range,
1538  *   no writeback.
1539  *
1540  * - Extent bits are locked
1541  */
1542 static int defrag_one_locked_target(struct btrfs_inode *inode,
1543                                     struct defrag_target_range *target,
1544                                     struct page **pages, int nr_pages,
1545                                     struct extent_state **cached_state)
1546 {
1547         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1548         struct extent_changeset *data_reserved = NULL;
1549         const u64 start = target->start;
1550         const u64 len = target->len;
1551         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1552         unsigned long start_index = start >> PAGE_SHIFT;
1553         unsigned long first_index = page_index(pages[0]);
1554         int ret = 0;
1555         int i;
1556
1557         ASSERT(last_index - first_index + 1 <= nr_pages);
1558
1559         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1560         if (ret < 0)
1561                 return ret;
1562         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1563                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1564                          EXTENT_DEFRAG, 0, 0, cached_state);
1565         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1566
1567         /* Update the page status */
1568         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1569                 ClearPageChecked(pages[i]);
1570                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1571         }
1572         btrfs_delalloc_release_extents(inode, len);
1573         extent_changeset_free(data_reserved);
1574
1575         return ret;
1576 }
1577
1578 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1579                             u32 extent_thresh, u64 newer_than, bool do_compress,
1580                             u64 *last_scanned_ret)
1581 {
1582         struct extent_state *cached_state = NULL;
1583         struct defrag_target_range *entry;
1584         struct defrag_target_range *tmp;
1585         LIST_HEAD(target_list);
1586         struct page **pages;
1587         const u32 sectorsize = inode->root->fs_info->sectorsize;
1588         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1589         u64 start_index = start >> PAGE_SHIFT;
1590         unsigned int nr_pages = last_index - start_index + 1;
1591         int ret = 0;
1592         int i;
1593
1594         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1595         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1596
1597         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1598         if (!pages)
1599                 return -ENOMEM;
1600
1601         /* Prepare all pages */
1602         for (i = 0; i < nr_pages; i++) {
1603                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1604                 if (IS_ERR(pages[i])) {
1605                         ret = PTR_ERR(pages[i]);
1606                         pages[i] = NULL;
1607                         goto free_pages;
1608                 }
1609         }
1610         for (i = 0; i < nr_pages; i++)
1611                 wait_on_page_writeback(pages[i]);
1612
1613         /* Lock the pages range */
1614         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1615                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1616                          &cached_state);
1617         /*
1618          * Now we have a consistent view about the extent map, re-check
1619          * which range really needs to be defragged.
1620          *
1621          * And this time we have extent locked already, pass @locked = true
1622          * so that we won't relock the extent range and cause deadlock.
1623          */
1624         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1625                                      newer_than, do_compress, true,
1626                                      &target_list, last_scanned_ret);
1627         if (ret < 0)
1628                 goto unlock_extent;
1629
1630         list_for_each_entry(entry, &target_list, list) {
1631                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1632                                                &cached_state);
1633                 if (ret < 0)
1634                         break;
1635         }
1636
1637         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1638                 list_del_init(&entry->list);
1639                 kfree(entry);
1640         }
1641 unlock_extent:
1642         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1643                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1644                              &cached_state);
1645 free_pages:
1646         for (i = 0; i < nr_pages; i++) {
1647                 if (pages[i]) {
1648                         unlock_page(pages[i]);
1649                         put_page(pages[i]);
1650                 }
1651         }
1652         kfree(pages);
1653         return ret;
1654 }
1655
1656 static int defrag_one_cluster(struct btrfs_inode *inode,
1657                               struct file_ra_state *ra,
1658                               u64 start, u32 len, u32 extent_thresh,
1659                               u64 newer_than, bool do_compress,
1660                               unsigned long *sectors_defragged,
1661                               unsigned long max_sectors,
1662                               u64 *last_scanned_ret)
1663 {
1664         const u32 sectorsize = inode->root->fs_info->sectorsize;
1665         struct defrag_target_range *entry;
1666         struct defrag_target_range *tmp;
1667         LIST_HEAD(target_list);
1668         int ret;
1669
1670         BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1671         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1672                                      newer_than, do_compress, false,
1673                                      &target_list, NULL);
1674         if (ret < 0)
1675                 goto out;
1676
1677         list_for_each_entry(entry, &target_list, list) {
1678                 u32 range_len = entry->len;
1679
1680                 /* Reached or beyond the limit */
1681                 if (max_sectors && *sectors_defragged >= max_sectors) {
1682                         ret = 1;
1683                         break;
1684                 }
1685
1686                 if (max_sectors)
1687                         range_len = min_t(u32, range_len,
1688                                 (max_sectors - *sectors_defragged) * sectorsize);
1689
1690                 /*
1691                  * If defrag_one_range() has updated last_scanned_ret,
1692                  * our range may already be invalid (e.g. hole punched).
1693                  * Skip if our range is before last_scanned_ret, as there is
1694                  * no need to defrag the range anymore.
1695                  */
1696                 if (entry->start + range_len <= *last_scanned_ret)
1697                         continue;
1698
1699                 if (ra)
1700                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1701                                 ra, NULL, entry->start >> PAGE_SHIFT,
1702                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1703                                 (entry->start >> PAGE_SHIFT) + 1);
1704                 /*
1705                  * Here we may not defrag any range if holes are punched before
1706                  * we locked the pages.
1707                  * But that's fine, it only affects the @sectors_defragged
1708                  * accounting.
1709                  */
1710                 ret = defrag_one_range(inode, entry->start, range_len,
1711                                        extent_thresh, newer_than, do_compress,
1712                                        last_scanned_ret);
1713                 if (ret < 0)
1714                         break;
1715                 *sectors_defragged += range_len >>
1716                                       inode->root->fs_info->sectorsize_bits;
1717         }
1718 out:
1719         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1720                 list_del_init(&entry->list);
1721                 kfree(entry);
1722         }
1723         if (ret >= 0)
1724                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1725         return ret;
1726 }
1727
1728 /*
1729  * Entry point to file defragmentation.
1730  *
1731  * @inode:         inode to be defragged
1732  * @ra:            readahead state (can be NUL)
1733  * @range:         defrag options including range and flags
1734  * @newer_than:    minimum transid to defrag
1735  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1736  *                 will be defragged.
1737  *
1738  * Return <0 for error.
1739  * Return >=0 for the number of sectors defragged, and range->start will be updated
1740  * to indicate the file offset where next defrag should be started at.
1741  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1742  *  defragging all the range).
1743  */
1744 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1745                       struct btrfs_ioctl_defrag_range_args *range,
1746                       u64 newer_than, unsigned long max_to_defrag)
1747 {
1748         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1749         unsigned long sectors_defragged = 0;
1750         u64 isize = i_size_read(inode);
1751         u64 cur;
1752         u64 last_byte;
1753         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1754         bool ra_allocated = false;
1755         int compress_type = BTRFS_COMPRESS_ZLIB;
1756         int ret = 0;
1757         u32 extent_thresh = range->extent_thresh;
1758         pgoff_t start_index;
1759
1760         if (isize == 0)
1761                 return 0;
1762
1763         if (range->start >= isize)
1764                 return -EINVAL;
1765
1766         if (do_compress) {
1767                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1768                         return -EINVAL;
1769                 if (range->compress_type)
1770                         compress_type = range->compress_type;
1771         }
1772
1773         if (extent_thresh == 0)
1774                 extent_thresh = SZ_256K;
1775
1776         if (range->start + range->len > range->start) {
1777                 /* Got a specific range */
1778                 last_byte = min(isize, range->start + range->len);
1779         } else {
1780                 /* Defrag until file end */
1781                 last_byte = isize;
1782         }
1783
1784         /* Align the range */
1785         cur = round_down(range->start, fs_info->sectorsize);
1786         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1787
1788         /*
1789          * If we were not given a ra, allocate a readahead context. As
1790          * readahead is just an optimization, defrag will work without it so
1791          * we don't error out.
1792          */
1793         if (!ra) {
1794                 ra_allocated = true;
1795                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1796                 if (ra)
1797                         file_ra_state_init(ra, inode->i_mapping);
1798         }
1799
1800         /*
1801          * Make writeback start from the beginning of the range, so that the
1802          * defrag range can be written sequentially.
1803          */
1804         start_index = cur >> PAGE_SHIFT;
1805         if (start_index < inode->i_mapping->writeback_index)
1806                 inode->i_mapping->writeback_index = start_index;
1807
1808         while (cur < last_byte) {
1809                 const unsigned long prev_sectors_defragged = sectors_defragged;
1810                 u64 last_scanned = cur;
1811                 u64 cluster_end;
1812
1813                 /* The cluster size 256K should always be page aligned */
1814                 BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1815
1816                 if (btrfs_defrag_cancelled(fs_info)) {
1817                         ret = -EAGAIN;
1818                         break;
1819                 }
1820
1821                 /* We want the cluster end at page boundary when possible */
1822                 cluster_end = (((cur >> PAGE_SHIFT) +
1823                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1824                 cluster_end = min(cluster_end, last_byte);
1825
1826                 btrfs_inode_lock(inode, 0);
1827                 if (IS_SWAPFILE(inode)) {
1828                         ret = -ETXTBSY;
1829                         btrfs_inode_unlock(inode, 0);
1830                         break;
1831                 }
1832                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1833                         btrfs_inode_unlock(inode, 0);
1834                         break;
1835                 }
1836                 if (do_compress)
1837                         BTRFS_I(inode)->defrag_compress = compress_type;
1838                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1839                                 cluster_end + 1 - cur, extent_thresh,
1840                                 newer_than, do_compress, &sectors_defragged,
1841                                 max_to_defrag, &last_scanned);
1842
1843                 if (sectors_defragged > prev_sectors_defragged)
1844                         balance_dirty_pages_ratelimited(inode->i_mapping);
1845
1846                 btrfs_inode_unlock(inode, 0);
1847                 if (ret < 0)
1848                         break;
1849                 cur = max(cluster_end + 1, last_scanned);
1850                 if (ret > 0) {
1851                         ret = 0;
1852                         break;
1853                 }
1854                 cond_resched();
1855         }
1856
1857         if (ra_allocated)
1858                 kfree(ra);
1859         /*
1860          * Update range.start for autodefrag, this will indicate where to start
1861          * in next run.
1862          */
1863         range->start = cur;
1864         if (sectors_defragged) {
1865                 /*
1866                  * We have defragged some sectors, for compression case they
1867                  * need to be written back immediately.
1868                  */
1869                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1870                         filemap_flush(inode->i_mapping);
1871                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1872                                      &BTRFS_I(inode)->runtime_flags))
1873                                 filemap_flush(inode->i_mapping);
1874                 }
1875                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1876                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1877                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1878                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1879                 ret = sectors_defragged;
1880         }
1881         if (do_compress) {
1882                 btrfs_inode_lock(inode, 0);
1883                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1884                 btrfs_inode_unlock(inode, 0);
1885         }
1886         return ret;
1887 }
1888
1889 /*
1890  * Try to start exclusive operation @type or cancel it if it's running.
1891  *
1892  * Return:
1893  *   0        - normal mode, newly claimed op started
1894  *  >0        - normal mode, something else is running,
1895  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1896  * ECANCELED  - cancel mode, successful cancel
1897  * ENOTCONN   - cancel mode, operation not running anymore
1898  */
1899 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1900                         enum btrfs_exclusive_operation type, bool cancel)
1901 {
1902         if (!cancel) {
1903                 /* Start normal op */
1904                 if (!btrfs_exclop_start(fs_info, type))
1905                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1906                 /* Exclusive operation is now claimed */
1907                 return 0;
1908         }
1909
1910         /* Cancel running op */
1911         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1912                 /*
1913                  * This blocks any exclop finish from setting it to NONE, so we
1914                  * request cancellation. Either it runs and we will wait for it,
1915                  * or it has finished and no waiting will happen.
1916                  */
1917                 atomic_inc(&fs_info->reloc_cancel_req);
1918                 btrfs_exclop_start_unlock(fs_info);
1919
1920                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1921                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1922                                     TASK_INTERRUPTIBLE);
1923
1924                 return -ECANCELED;
1925         }
1926
1927         /* Something else is running or none */
1928         return -ENOTCONN;
1929 }
1930
1931 static noinline int btrfs_ioctl_resize(struct file *file,
1932                                         void __user *arg)
1933 {
1934         BTRFS_DEV_LOOKUP_ARGS(args);
1935         struct inode *inode = file_inode(file);
1936         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1937         u64 new_size;
1938         u64 old_size;
1939         u64 devid = 1;
1940         struct btrfs_root *root = BTRFS_I(inode)->root;
1941         struct btrfs_ioctl_vol_args *vol_args;
1942         struct btrfs_trans_handle *trans;
1943         struct btrfs_device *device = NULL;
1944         char *sizestr;
1945         char *retptr;
1946         char *devstr = NULL;
1947         int ret = 0;
1948         int mod = 0;
1949         bool cancel;
1950
1951         if (!capable(CAP_SYS_ADMIN))
1952                 return -EPERM;
1953
1954         ret = mnt_want_write_file(file);
1955         if (ret)
1956                 return ret;
1957
1958         /*
1959          * Read the arguments before checking exclusivity to be able to
1960          * distinguish regular resize and cancel
1961          */
1962         vol_args = memdup_user(arg, sizeof(*vol_args));
1963         if (IS_ERR(vol_args)) {
1964                 ret = PTR_ERR(vol_args);
1965                 goto out_drop;
1966         }
1967         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1968         sizestr = vol_args->name;
1969         cancel = (strcmp("cancel", sizestr) == 0);
1970         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1971         if (ret)
1972                 goto out_free;
1973         /* Exclusive operation is now claimed */
1974
1975         devstr = strchr(sizestr, ':');
1976         if (devstr) {
1977                 sizestr = devstr + 1;
1978                 *devstr = '\0';
1979                 devstr = vol_args->name;
1980                 ret = kstrtoull(devstr, 10, &devid);
1981                 if (ret)
1982                         goto out_finish;
1983                 if (!devid) {
1984                         ret = -EINVAL;
1985                         goto out_finish;
1986                 }
1987                 btrfs_info(fs_info, "resizing devid %llu", devid);
1988         }
1989
1990         args.devid = devid;
1991         device = btrfs_find_device(fs_info->fs_devices, &args);
1992         if (!device) {
1993                 btrfs_info(fs_info, "resizer unable to find device %llu",
1994                            devid);
1995                 ret = -ENODEV;
1996                 goto out_finish;
1997         }
1998
1999         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2000                 btrfs_info(fs_info,
2001                            "resizer unable to apply on readonly device %llu",
2002                        devid);
2003                 ret = -EPERM;
2004                 goto out_finish;
2005         }
2006
2007         if (!strcmp(sizestr, "max"))
2008                 new_size = bdev_nr_bytes(device->bdev);
2009         else {
2010                 if (sizestr[0] == '-') {
2011                         mod = -1;
2012                         sizestr++;
2013                 } else if (sizestr[0] == '+') {
2014                         mod = 1;
2015                         sizestr++;
2016                 }
2017                 new_size = memparse(sizestr, &retptr);
2018                 if (*retptr != '\0' || new_size == 0) {
2019                         ret = -EINVAL;
2020                         goto out_finish;
2021                 }
2022         }
2023
2024         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2025                 ret = -EPERM;
2026                 goto out_finish;
2027         }
2028
2029         old_size = btrfs_device_get_total_bytes(device);
2030
2031         if (mod < 0) {
2032                 if (new_size > old_size) {
2033                         ret = -EINVAL;
2034                         goto out_finish;
2035                 }
2036                 new_size = old_size - new_size;
2037         } else if (mod > 0) {
2038                 if (new_size > ULLONG_MAX - old_size) {
2039                         ret = -ERANGE;
2040                         goto out_finish;
2041                 }
2042                 new_size = old_size + new_size;
2043         }
2044
2045         if (new_size < SZ_256M) {
2046                 ret = -EINVAL;
2047                 goto out_finish;
2048         }
2049         if (new_size > bdev_nr_bytes(device->bdev)) {
2050                 ret = -EFBIG;
2051                 goto out_finish;
2052         }
2053
2054         new_size = round_down(new_size, fs_info->sectorsize);
2055
2056         if (new_size > old_size) {
2057                 trans = btrfs_start_transaction(root, 0);
2058                 if (IS_ERR(trans)) {
2059                         ret = PTR_ERR(trans);
2060                         goto out_finish;
2061                 }
2062                 ret = btrfs_grow_device(trans, device, new_size);
2063                 btrfs_commit_transaction(trans);
2064         } else if (new_size < old_size) {
2065                 ret = btrfs_shrink_device(device, new_size);
2066         } /* equal, nothing need to do */
2067
2068         if (ret == 0 && new_size != old_size)
2069                 btrfs_info_in_rcu(fs_info,
2070                         "resize device %s (devid %llu) from %llu to %llu",
2071                         rcu_str_deref(device->name), device->devid,
2072                         old_size, new_size);
2073 out_finish:
2074         btrfs_exclop_finish(fs_info);
2075 out_free:
2076         kfree(vol_args);
2077 out_drop:
2078         mnt_drop_write_file(file);
2079         return ret;
2080 }
2081
2082 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2083                                 struct user_namespace *mnt_userns,
2084                                 const char *name, unsigned long fd, int subvol,
2085                                 bool readonly,
2086                                 struct btrfs_qgroup_inherit *inherit)
2087 {
2088         int namelen;
2089         int ret = 0;
2090
2091         if (!S_ISDIR(file_inode(file)->i_mode))
2092                 return -ENOTDIR;
2093
2094         ret = mnt_want_write_file(file);
2095         if (ret)
2096                 goto out;
2097
2098         namelen = strlen(name);
2099         if (strchr(name, '/')) {
2100                 ret = -EINVAL;
2101                 goto out_drop_write;
2102         }
2103
2104         if (name[0] == '.' &&
2105            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2106                 ret = -EEXIST;
2107                 goto out_drop_write;
2108         }
2109
2110         if (subvol) {
2111                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2112                                      namelen, NULL, readonly, inherit);
2113         } else {
2114                 struct fd src = fdget(fd);
2115                 struct inode *src_inode;
2116                 if (!src.file) {
2117                         ret = -EINVAL;
2118                         goto out_drop_write;
2119                 }
2120
2121                 src_inode = file_inode(src.file);
2122                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2123                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2124                                    "Snapshot src from another FS");
2125                         ret = -EXDEV;
2126                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2127                         /*
2128                          * Subvolume creation is not restricted, but snapshots
2129                          * are limited to own subvolumes only
2130                          */
2131                         ret = -EPERM;
2132                 } else {
2133                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2134                                                name, namelen,
2135                                                BTRFS_I(src_inode)->root,
2136                                                readonly, inherit);
2137                 }
2138                 fdput(src);
2139         }
2140 out_drop_write:
2141         mnt_drop_write_file(file);
2142 out:
2143         return ret;
2144 }
2145
2146 static noinline int btrfs_ioctl_snap_create(struct file *file,
2147                                             void __user *arg, int subvol)
2148 {
2149         struct btrfs_ioctl_vol_args *vol_args;
2150         int ret;
2151
2152         if (!S_ISDIR(file_inode(file)->i_mode))
2153                 return -ENOTDIR;
2154
2155         vol_args = memdup_user(arg, sizeof(*vol_args));
2156         if (IS_ERR(vol_args))
2157                 return PTR_ERR(vol_args);
2158         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2159
2160         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2161                                         vol_args->name, vol_args->fd, subvol,
2162                                         false, NULL);
2163
2164         kfree(vol_args);
2165         return ret;
2166 }
2167
2168 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2169                                                void __user *arg, int subvol)
2170 {
2171         struct btrfs_ioctl_vol_args_v2 *vol_args;
2172         int ret;
2173         bool readonly = false;
2174         struct btrfs_qgroup_inherit *inherit = NULL;
2175
2176         if (!S_ISDIR(file_inode(file)->i_mode))
2177                 return -ENOTDIR;
2178
2179         vol_args = memdup_user(arg, sizeof(*vol_args));
2180         if (IS_ERR(vol_args))
2181                 return PTR_ERR(vol_args);
2182         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2183
2184         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2185                 ret = -EOPNOTSUPP;
2186                 goto free_args;
2187         }
2188
2189         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2190                 readonly = true;
2191         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2192                 u64 nums;
2193
2194                 if (vol_args->size < sizeof(*inherit) ||
2195                     vol_args->size > PAGE_SIZE) {
2196                         ret = -EINVAL;
2197                         goto free_args;
2198                 }
2199                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2200                 if (IS_ERR(inherit)) {
2201                         ret = PTR_ERR(inherit);
2202                         goto free_args;
2203                 }
2204
2205                 if (inherit->num_qgroups > PAGE_SIZE ||
2206                     inherit->num_ref_copies > PAGE_SIZE ||
2207                     inherit->num_excl_copies > PAGE_SIZE) {
2208                         ret = -EINVAL;
2209                         goto free_inherit;
2210                 }
2211
2212                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2213                        2 * inherit->num_excl_copies;
2214                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2215                         ret = -EINVAL;
2216                         goto free_inherit;
2217                 }
2218         }
2219
2220         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2221                                         vol_args->name, vol_args->fd, subvol,
2222                                         readonly, inherit);
2223         if (ret)
2224                 goto free_inherit;
2225 free_inherit:
2226         kfree(inherit);
2227 free_args:
2228         kfree(vol_args);
2229         return ret;
2230 }
2231
2232 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
2233                                                 void __user *arg)
2234 {
2235         struct inode *inode = file_inode(file);
2236         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2237         struct btrfs_root *root = BTRFS_I(inode)->root;
2238         int ret = 0;
2239         u64 flags = 0;
2240
2241         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2242                 return -EINVAL;
2243
2244         down_read(&fs_info->subvol_sem);
2245         if (btrfs_root_readonly(root))
2246                 flags |= BTRFS_SUBVOL_RDONLY;
2247         up_read(&fs_info->subvol_sem);
2248
2249         if (copy_to_user(arg, &flags, sizeof(flags)))
2250                 ret = -EFAULT;
2251
2252         return ret;
2253 }
2254
2255 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2256                                               void __user *arg)
2257 {
2258         struct inode *inode = file_inode(file);
2259         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2260         struct btrfs_root *root = BTRFS_I(inode)->root;
2261         struct btrfs_trans_handle *trans;
2262         u64 root_flags;
2263         u64 flags;
2264         int ret = 0;
2265
2266         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2267                 return -EPERM;
2268
2269         ret = mnt_want_write_file(file);
2270         if (ret)
2271                 goto out;
2272
2273         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2274                 ret = -EINVAL;
2275                 goto out_drop_write;
2276         }
2277
2278         if (copy_from_user(&flags, arg, sizeof(flags))) {
2279                 ret = -EFAULT;
2280                 goto out_drop_write;
2281         }
2282
2283         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2284                 ret = -EOPNOTSUPP;
2285                 goto out_drop_write;
2286         }
2287
2288         down_write(&fs_info->subvol_sem);
2289
2290         /* nothing to do */
2291         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2292                 goto out_drop_sem;
2293
2294         root_flags = btrfs_root_flags(&root->root_item);
2295         if (flags & BTRFS_SUBVOL_RDONLY) {
2296                 btrfs_set_root_flags(&root->root_item,
2297                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2298         } else {
2299                 /*
2300                  * Block RO -> RW transition if this subvolume is involved in
2301                  * send
2302                  */
2303                 spin_lock(&root->root_item_lock);
2304                 if (root->send_in_progress == 0) {
2305                         btrfs_set_root_flags(&root->root_item,
2306                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2307                         spin_unlock(&root->root_item_lock);
2308                 } else {
2309                         spin_unlock(&root->root_item_lock);
2310                         btrfs_warn(fs_info,
2311                                    "Attempt to set subvolume %llu read-write during send",
2312                                    root->root_key.objectid);
2313                         ret = -EPERM;
2314                         goto out_drop_sem;
2315                 }
2316         }
2317
2318         trans = btrfs_start_transaction(root, 1);
2319         if (IS_ERR(trans)) {
2320                 ret = PTR_ERR(trans);
2321                 goto out_reset;
2322         }
2323
2324         ret = btrfs_update_root(trans, fs_info->tree_root,
2325                                 &root->root_key, &root->root_item);
2326         if (ret < 0) {
2327                 btrfs_end_transaction(trans);
2328                 goto out_reset;
2329         }
2330
2331         ret = btrfs_commit_transaction(trans);
2332
2333 out_reset:
2334         if (ret)
2335                 btrfs_set_root_flags(&root->root_item, root_flags);
2336 out_drop_sem:
2337         up_write(&fs_info->subvol_sem);
2338 out_drop_write:
2339         mnt_drop_write_file(file);
2340 out:
2341         return ret;
2342 }
2343
2344 static noinline int key_in_sk(struct btrfs_key *key,
2345                               struct btrfs_ioctl_search_key *sk)
2346 {
2347         struct btrfs_key test;
2348         int ret;
2349
2350         test.objectid = sk->min_objectid;
2351         test.type = sk->min_type;
2352         test.offset = sk->min_offset;
2353
2354         ret = btrfs_comp_cpu_keys(key, &test);
2355         if (ret < 0)
2356                 return 0;
2357
2358         test.objectid = sk->max_objectid;
2359         test.type = sk->max_type;
2360         test.offset = sk->max_offset;
2361
2362         ret = btrfs_comp_cpu_keys(key, &test);
2363         if (ret > 0)
2364                 return 0;
2365         return 1;
2366 }
2367
2368 static noinline int copy_to_sk(struct btrfs_path *path,
2369                                struct btrfs_key *key,
2370                                struct btrfs_ioctl_search_key *sk,
2371                                size_t *buf_size,
2372                                char __user *ubuf,
2373                                unsigned long *sk_offset,
2374                                int *num_found)
2375 {
2376         u64 found_transid;
2377         struct extent_buffer *leaf;
2378         struct btrfs_ioctl_search_header sh;
2379         struct btrfs_key test;
2380         unsigned long item_off;
2381         unsigned long item_len;
2382         int nritems;
2383         int i;
2384         int slot;
2385         int ret = 0;
2386
2387         leaf = path->nodes[0];
2388         slot = path->slots[0];
2389         nritems = btrfs_header_nritems(leaf);
2390
2391         if (btrfs_header_generation(leaf) > sk->max_transid) {
2392                 i = nritems;
2393                 goto advance_key;
2394         }
2395         found_transid = btrfs_header_generation(leaf);
2396
2397         for (i = slot; i < nritems; i++) {
2398                 item_off = btrfs_item_ptr_offset(leaf, i);
2399                 item_len = btrfs_item_size(leaf, i);
2400
2401                 btrfs_item_key_to_cpu(leaf, key, i);
2402                 if (!key_in_sk(key, sk))
2403                         continue;
2404
2405                 if (sizeof(sh) + item_len > *buf_size) {
2406                         if (*num_found) {
2407                                 ret = 1;
2408                                 goto out;
2409                         }
2410
2411                         /*
2412                          * return one empty item back for v1, which does not
2413                          * handle -EOVERFLOW
2414                          */
2415
2416                         *buf_size = sizeof(sh) + item_len;
2417                         item_len = 0;
2418                         ret = -EOVERFLOW;
2419                 }
2420
2421                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2422                         ret = 1;
2423                         goto out;
2424                 }
2425
2426                 sh.objectid = key->objectid;
2427                 sh.offset = key->offset;
2428                 sh.type = key->type;
2429                 sh.len = item_len;
2430                 sh.transid = found_transid;
2431
2432                 /*
2433                  * Copy search result header. If we fault then loop again so we
2434                  * can fault in the pages and -EFAULT there if there's a
2435                  * problem. Otherwise we'll fault and then copy the buffer in
2436                  * properly this next time through
2437                  */
2438                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2439                         ret = 0;
2440                         goto out;
2441                 }
2442
2443                 *sk_offset += sizeof(sh);
2444
2445                 if (item_len) {
2446                         char __user *up = ubuf + *sk_offset;
2447                         /*
2448                          * Copy the item, same behavior as above, but reset the
2449                          * * sk_offset so we copy the full thing again.
2450                          */
2451                         if (read_extent_buffer_to_user_nofault(leaf, up,
2452                                                 item_off, item_len)) {
2453                                 ret = 0;
2454                                 *sk_offset -= sizeof(sh);
2455                                 goto out;
2456                         }
2457
2458                         *sk_offset += item_len;
2459                 }
2460                 (*num_found)++;
2461
2462                 if (ret) /* -EOVERFLOW from above */
2463                         goto out;
2464
2465                 if (*num_found >= sk->nr_items) {
2466                         ret = 1;
2467                         goto out;
2468                 }
2469         }
2470 advance_key:
2471         ret = 0;
2472         test.objectid = sk->max_objectid;
2473         test.type = sk->max_type;
2474         test.offset = sk->max_offset;
2475         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2476                 ret = 1;
2477         else if (key->offset < (u64)-1)
2478                 key->offset++;
2479         else if (key->type < (u8)-1) {
2480                 key->offset = 0;
2481                 key->type++;
2482         } else if (key->objectid < (u64)-1) {
2483                 key->offset = 0;
2484                 key->type = 0;
2485                 key->objectid++;
2486         } else
2487                 ret = 1;
2488 out:
2489         /*
2490          *  0: all items from this leaf copied, continue with next
2491          *  1: * more items can be copied, but unused buffer is too small
2492          *     * all items were found
2493          *     Either way, it will stops the loop which iterates to the next
2494          *     leaf
2495          *  -EOVERFLOW: item was to large for buffer
2496          *  -EFAULT: could not copy extent buffer back to userspace
2497          */
2498         return ret;
2499 }
2500
2501 static noinline int search_ioctl(struct inode *inode,
2502                                  struct btrfs_ioctl_search_key *sk,
2503                                  size_t *buf_size,
2504                                  char __user *ubuf)
2505 {
2506         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2507         struct btrfs_root *root;
2508         struct btrfs_key key;
2509         struct btrfs_path *path;
2510         int ret;
2511         int num_found = 0;
2512         unsigned long sk_offset = 0;
2513
2514         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2515                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2516                 return -EOVERFLOW;
2517         }
2518
2519         path = btrfs_alloc_path();
2520         if (!path)
2521                 return -ENOMEM;
2522
2523         if (sk->tree_id == 0) {
2524                 /* search the root of the inode that was passed */
2525                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2526         } else {
2527                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2528                 if (IS_ERR(root)) {
2529                         btrfs_free_path(path);
2530                         return PTR_ERR(root);
2531                 }
2532         }
2533
2534         key.objectid = sk->min_objectid;
2535         key.type = sk->min_type;
2536         key.offset = sk->min_offset;
2537
2538         while (1) {
2539                 ret = -EFAULT;
2540                 if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
2541                         break;
2542
2543                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2544                 if (ret != 0) {
2545                         if (ret > 0)
2546                                 ret = 0;
2547                         goto err;
2548                 }
2549                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2550                                  &sk_offset, &num_found);
2551                 btrfs_release_path(path);
2552                 if (ret)
2553                         break;
2554
2555         }
2556         if (ret > 0)
2557                 ret = 0;
2558 err:
2559         sk->nr_items = num_found;
2560         btrfs_put_root(root);
2561         btrfs_free_path(path);
2562         return ret;
2563 }
2564
2565 static noinline int btrfs_ioctl_tree_search(struct file *file,
2566                                            void __user *argp)
2567 {
2568         struct btrfs_ioctl_search_args __user *uargs;
2569         struct btrfs_ioctl_search_key sk;
2570         struct inode *inode;
2571         int ret;
2572         size_t buf_size;
2573
2574         if (!capable(CAP_SYS_ADMIN))
2575                 return -EPERM;
2576
2577         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2578
2579         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2580                 return -EFAULT;
2581
2582         buf_size = sizeof(uargs->buf);
2583
2584         inode = file_inode(file);
2585         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2586
2587         /*
2588          * In the origin implementation an overflow is handled by returning a
2589          * search header with a len of zero, so reset ret.
2590          */
2591         if (ret == -EOVERFLOW)
2592                 ret = 0;
2593
2594         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2595                 ret = -EFAULT;
2596         return ret;
2597 }
2598
2599 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2600                                                void __user *argp)
2601 {
2602         struct btrfs_ioctl_search_args_v2 __user *uarg;
2603         struct btrfs_ioctl_search_args_v2 args;
2604         struct inode *inode;
2605         int ret;
2606         size_t buf_size;
2607         const size_t buf_limit = SZ_16M;
2608
2609         if (!capable(CAP_SYS_ADMIN))
2610                 return -EPERM;
2611
2612         /* copy search header and buffer size */
2613         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2614         if (copy_from_user(&args, uarg, sizeof(args)))
2615                 return -EFAULT;
2616
2617         buf_size = args.buf_size;
2618
2619         /* limit result size to 16MB */
2620         if (buf_size > buf_limit)
2621                 buf_size = buf_limit;
2622
2623         inode = file_inode(file);
2624         ret = search_ioctl(inode, &args.key, &buf_size,
2625                            (char __user *)(&uarg->buf[0]));
2626         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2627                 ret = -EFAULT;
2628         else if (ret == -EOVERFLOW &&
2629                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2630                 ret = -EFAULT;
2631
2632         return ret;
2633 }
2634
2635 /*
2636  * Search INODE_REFs to identify path name of 'dirid' directory
2637  * in a 'tree_id' tree. and sets path name to 'name'.
2638  */
2639 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2640                                 u64 tree_id, u64 dirid, char *name)
2641 {
2642         struct btrfs_root *root;
2643         struct btrfs_key key;
2644         char *ptr;
2645         int ret = -1;
2646         int slot;
2647         int len;
2648         int total_len = 0;
2649         struct btrfs_inode_ref *iref;
2650         struct extent_buffer *l;
2651         struct btrfs_path *path;
2652
2653         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2654                 name[0]='\0';
2655                 return 0;
2656         }
2657
2658         path = btrfs_alloc_path();
2659         if (!path)
2660                 return -ENOMEM;
2661
2662         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2663
2664         root = btrfs_get_fs_root(info, tree_id, true);
2665         if (IS_ERR(root)) {
2666                 ret = PTR_ERR(root);
2667                 root = NULL;
2668                 goto out;
2669         }
2670
2671         key.objectid = dirid;
2672         key.type = BTRFS_INODE_REF_KEY;
2673         key.offset = (u64)-1;
2674
2675         while (1) {
2676                 ret = btrfs_search_backwards(root, &key, path);
2677                 if (ret < 0)
2678                         goto out;
2679                 else if (ret > 0) {
2680                         ret = -ENOENT;
2681                         goto out;
2682                 }
2683
2684                 l = path->nodes[0];
2685                 slot = path->slots[0];
2686
2687                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2688                 len = btrfs_inode_ref_name_len(l, iref);
2689                 ptr -= len + 1;
2690                 total_len += len + 1;
2691                 if (ptr < name) {
2692                         ret = -ENAMETOOLONG;
2693                         goto out;
2694                 }
2695
2696                 *(ptr + len) = '/';
2697                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2698
2699                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2700                         break;
2701
2702                 btrfs_release_path(path);
2703                 key.objectid = key.offset;
2704                 key.offset = (u64)-1;
2705                 dirid = key.objectid;
2706         }
2707         memmove(name, ptr, total_len);
2708         name[total_len] = '\0';
2709         ret = 0;
2710 out:
2711         btrfs_put_root(root);
2712         btrfs_free_path(path);
2713         return ret;
2714 }
2715
2716 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2717                                 struct inode *inode,
2718                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2719 {
2720         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2721         struct super_block *sb = inode->i_sb;
2722         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2723         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2724         u64 dirid = args->dirid;
2725         unsigned long item_off;
2726         unsigned long item_len;
2727         struct btrfs_inode_ref *iref;
2728         struct btrfs_root_ref *rref;
2729         struct btrfs_root *root = NULL;
2730         struct btrfs_path *path;
2731         struct btrfs_key key, key2;
2732         struct extent_buffer *leaf;
2733         struct inode *temp_inode;
2734         char *ptr;
2735         int slot;
2736         int len;
2737         int total_len = 0;
2738         int ret;
2739
2740         path = btrfs_alloc_path();
2741         if (!path)
2742                 return -ENOMEM;
2743
2744         /*
2745          * If the bottom subvolume does not exist directly under upper_limit,
2746          * construct the path in from the bottom up.
2747          */
2748         if (dirid != upper_limit.objectid) {
2749                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2750
2751                 root = btrfs_get_fs_root(fs_info, treeid, true);
2752                 if (IS_ERR(root)) {
2753                         ret = PTR_ERR(root);
2754                         goto out;
2755                 }
2756
2757                 key.objectid = dirid;
2758                 key.type = BTRFS_INODE_REF_KEY;
2759                 key.offset = (u64)-1;
2760                 while (1) {
2761                         ret = btrfs_search_backwards(root, &key, path);
2762                         if (ret < 0)
2763                                 goto out_put;
2764                         else if (ret > 0) {
2765                                 ret = -ENOENT;
2766                                 goto out_put;
2767                         }
2768
2769                         leaf = path->nodes[0];
2770                         slot = path->slots[0];
2771
2772                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2773                         len = btrfs_inode_ref_name_len(leaf, iref);
2774                         ptr -= len + 1;
2775                         total_len += len + 1;
2776                         if (ptr < args->path) {
2777                                 ret = -ENAMETOOLONG;
2778                                 goto out_put;
2779                         }
2780
2781                         *(ptr + len) = '/';
2782                         read_extent_buffer(leaf, ptr,
2783                                         (unsigned long)(iref + 1), len);
2784
2785                         /* Check the read+exec permission of this directory */
2786                         ret = btrfs_previous_item(root, path, dirid,
2787                                                   BTRFS_INODE_ITEM_KEY);
2788                         if (ret < 0) {
2789                                 goto out_put;
2790                         } else if (ret > 0) {
2791                                 ret = -ENOENT;
2792                                 goto out_put;
2793                         }
2794
2795                         leaf = path->nodes[0];
2796                         slot = path->slots[0];
2797                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2798                         if (key2.objectid != dirid) {
2799                                 ret = -ENOENT;
2800                                 goto out_put;
2801                         }
2802
2803                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2804                         if (IS_ERR(temp_inode)) {
2805                                 ret = PTR_ERR(temp_inode);
2806                                 goto out_put;
2807                         }
2808                         ret = inode_permission(mnt_userns, temp_inode,
2809                                                MAY_READ | MAY_EXEC);
2810                         iput(temp_inode);
2811                         if (ret) {
2812                                 ret = -EACCES;
2813                                 goto out_put;
2814                         }
2815
2816                         if (key.offset == upper_limit.objectid)
2817                                 break;
2818                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2819                                 ret = -EACCES;
2820                                 goto out_put;
2821                         }
2822
2823                         btrfs_release_path(path);
2824                         key.objectid = key.offset;
2825                         key.offset = (u64)-1;
2826                         dirid = key.objectid;
2827                 }
2828
2829                 memmove(args->path, ptr, total_len);
2830                 args->path[total_len] = '\0';
2831                 btrfs_put_root(root);
2832                 root = NULL;
2833                 btrfs_release_path(path);
2834         }
2835
2836         /* Get the bottom subvolume's name from ROOT_REF */
2837         key.objectid = treeid;
2838         key.type = BTRFS_ROOT_REF_KEY;
2839         key.offset = args->treeid;
2840         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2841         if (ret < 0) {
2842                 goto out;
2843         } else if (ret > 0) {
2844                 ret = -ENOENT;
2845                 goto out;
2846         }
2847
2848         leaf = path->nodes[0];
2849         slot = path->slots[0];
2850         btrfs_item_key_to_cpu(leaf, &key, slot);
2851
2852         item_off = btrfs_item_ptr_offset(leaf, slot);
2853         item_len = btrfs_item_size(leaf, slot);
2854         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2855         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2856         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2857                 ret = -EINVAL;
2858                 goto out;
2859         }
2860
2861         /* Copy subvolume's name */
2862         item_off += sizeof(struct btrfs_root_ref);
2863         item_len -= sizeof(struct btrfs_root_ref);
2864         read_extent_buffer(leaf, args->name, item_off, item_len);
2865         args->name[item_len] = 0;
2866
2867 out_put:
2868         btrfs_put_root(root);
2869 out:
2870         btrfs_free_path(path);
2871         return ret;
2872 }
2873
2874 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2875                                            void __user *argp)
2876 {
2877         struct btrfs_ioctl_ino_lookup_args *args;
2878         struct inode *inode;
2879         int ret = 0;
2880
2881         args = memdup_user(argp, sizeof(*args));
2882         if (IS_ERR(args))
2883                 return PTR_ERR(args);
2884
2885         inode = file_inode(file);
2886
2887         /*
2888          * Unprivileged query to obtain the containing subvolume root id. The
2889          * path is reset so it's consistent with btrfs_search_path_in_tree.
2890          */
2891         if (args->treeid == 0)
2892                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2893
2894         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2895                 args->name[0] = 0;
2896                 goto out;
2897         }
2898
2899         if (!capable(CAP_SYS_ADMIN)) {
2900                 ret = -EPERM;
2901                 goto out;
2902         }
2903
2904         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2905                                         args->treeid, args->objectid,
2906                                         args->name);
2907
2908 out:
2909         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2910                 ret = -EFAULT;
2911
2912         kfree(args);
2913         return ret;
2914 }
2915
2916 /*
2917  * Version of ino_lookup ioctl (unprivileged)
2918  *
2919  * The main differences from ino_lookup ioctl are:
2920  *
2921  *   1. Read + Exec permission will be checked using inode_permission() during
2922  *      path construction. -EACCES will be returned in case of failure.
2923  *   2. Path construction will be stopped at the inode number which corresponds
2924  *      to the fd with which this ioctl is called. If constructed path does not
2925  *      exist under fd's inode, -EACCES will be returned.
2926  *   3. The name of bottom subvolume is also searched and filled.
2927  */
2928 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2929 {
2930         struct btrfs_ioctl_ino_lookup_user_args *args;
2931         struct inode *inode;
2932         int ret;
2933
2934         args = memdup_user(argp, sizeof(*args));
2935         if (IS_ERR(args))
2936                 return PTR_ERR(args);
2937
2938         inode = file_inode(file);
2939
2940         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2941             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2942                 /*
2943                  * The subvolume does not exist under fd with which this is
2944                  * called
2945                  */
2946                 kfree(args);
2947                 return -EACCES;
2948         }
2949
2950         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2951
2952         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2953                 ret = -EFAULT;
2954
2955         kfree(args);
2956         return ret;
2957 }
2958
2959 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2960 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2961 {
2962         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2963         struct btrfs_fs_info *fs_info;
2964         struct btrfs_root *root;
2965         struct btrfs_path *path;
2966         struct btrfs_key key;
2967         struct btrfs_root_item *root_item;
2968         struct btrfs_root_ref *rref;
2969         struct extent_buffer *leaf;
2970         unsigned long item_off;
2971         unsigned long item_len;
2972         struct inode *inode;
2973         int slot;
2974         int ret = 0;
2975
2976         path = btrfs_alloc_path();
2977         if (!path)
2978                 return -ENOMEM;
2979
2980         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2981         if (!subvol_info) {
2982                 btrfs_free_path(path);
2983                 return -ENOMEM;
2984         }
2985
2986         inode = file_inode(file);
2987         fs_info = BTRFS_I(inode)->root->fs_info;
2988
2989         /* Get root_item of inode's subvolume */
2990         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2991         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2992         if (IS_ERR(root)) {
2993                 ret = PTR_ERR(root);
2994                 goto out_free;
2995         }
2996         root_item = &root->root_item;
2997
2998         subvol_info->treeid = key.objectid;
2999
3000         subvol_info->generation = btrfs_root_generation(root_item);
3001         subvol_info->flags = btrfs_root_flags(root_item);
3002
3003         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3004         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3005                                                     BTRFS_UUID_SIZE);
3006         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3007                                                     BTRFS_UUID_SIZE);
3008
3009         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3010         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3011         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3012
3013         subvol_info->otransid = btrfs_root_otransid(root_item);
3014         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3015         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3016
3017         subvol_info->stransid = btrfs_root_stransid(root_item);
3018         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3019         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3020
3021         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3022         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3023         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3024
3025         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3026                 /* Search root tree for ROOT_BACKREF of this subvolume */
3027                 key.type = BTRFS_ROOT_BACKREF_KEY;
3028                 key.offset = 0;
3029                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3030                 if (ret < 0) {
3031                         goto out;
3032                 } else if (path->slots[0] >=
3033                            btrfs_header_nritems(path->nodes[0])) {
3034                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3035                         if (ret < 0) {
3036                                 goto out;
3037                         } else if (ret > 0) {
3038                                 ret = -EUCLEAN;
3039                                 goto out;
3040                         }
3041                 }
3042
3043                 leaf = path->nodes[0];
3044                 slot = path->slots[0];
3045                 btrfs_item_key_to_cpu(leaf, &key, slot);
3046                 if (key.objectid == subvol_info->treeid &&
3047                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3048                         subvol_info->parent_id = key.offset;
3049
3050                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3051                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3052
3053                         item_off = btrfs_item_ptr_offset(leaf, slot)
3054                                         + sizeof(struct btrfs_root_ref);
3055                         item_len = btrfs_item_size(leaf, slot)
3056                                         - sizeof(struct btrfs_root_ref);
3057                         read_extent_buffer(leaf, subvol_info->name,
3058                                            item_off, item_len);
3059                 } else {
3060                         ret = -ENOENT;
3061                         goto out;
3062                 }
3063         }
3064
3065         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3066                 ret = -EFAULT;
3067
3068 out:
3069         btrfs_put_root(root);
3070 out_free:
3071         btrfs_free_path(path);
3072         kfree(subvol_info);
3073         return ret;
3074 }
3075
3076 /*
3077  * Return ROOT_REF information of the subvolume containing this inode
3078  * except the subvolume name.
3079  */
3080 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
3081 {
3082         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3083         struct btrfs_root_ref *rref;
3084         struct btrfs_root *root;
3085         struct btrfs_path *path;
3086         struct btrfs_key key;
3087         struct extent_buffer *leaf;
3088         struct inode *inode;
3089         u64 objectid;
3090         int slot;
3091         int ret;
3092         u8 found;
3093
3094         path = btrfs_alloc_path();
3095         if (!path)
3096                 return -ENOMEM;
3097
3098         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3099         if (IS_ERR(rootrefs)) {
3100                 btrfs_free_path(path);
3101                 return PTR_ERR(rootrefs);
3102         }
3103
3104         inode = file_inode(file);
3105         root = BTRFS_I(inode)->root->fs_info->tree_root;
3106         objectid = BTRFS_I(inode)->root->root_key.objectid;
3107
3108         key.objectid = objectid;
3109         key.type = BTRFS_ROOT_REF_KEY;
3110         key.offset = rootrefs->min_treeid;
3111         found = 0;
3112
3113         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3114         if (ret < 0) {
3115                 goto out;
3116         } else if (path->slots[0] >=
3117                    btrfs_header_nritems(path->nodes[0])) {
3118                 ret = btrfs_next_leaf(root, path);
3119                 if (ret < 0) {
3120                         goto out;
3121                 } else if (ret > 0) {
3122                         ret = -EUCLEAN;
3123                         goto out;
3124                 }
3125         }
3126         while (1) {
3127                 leaf = path->nodes[0];
3128                 slot = path->slots[0];
3129
3130                 btrfs_item_key_to_cpu(leaf, &key, slot);
3131                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3132                         ret = 0;
3133                         goto out;
3134                 }
3135
3136                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3137                         ret = -EOVERFLOW;
3138                         goto out;
3139                 }
3140
3141                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3142                 rootrefs->rootref[found].treeid = key.offset;
3143                 rootrefs->rootref[found].dirid =
3144                                   btrfs_root_ref_dirid(leaf, rref);
3145                 found++;
3146
3147                 ret = btrfs_next_item(root, path);
3148                 if (ret < 0) {
3149                         goto out;
3150                 } else if (ret > 0) {
3151                         ret = -EUCLEAN;
3152                         goto out;
3153                 }
3154         }
3155
3156 out:
3157         if (!ret || ret == -EOVERFLOW) {
3158                 rootrefs->num_items = found;
3159                 /* update min_treeid for next search */
3160                 if (found)
3161                         rootrefs->min_treeid =
3162                                 rootrefs->rootref[found - 1].treeid + 1;
3163                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3164                         ret = -EFAULT;
3165         }
3166
3167         kfree(rootrefs);
3168         btrfs_free_path(path);
3169
3170         return ret;
3171 }
3172
3173 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3174                                              void __user *arg,
3175                                              bool destroy_v2)
3176 {
3177         struct dentry *parent = file->f_path.dentry;
3178         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3179         struct dentry *dentry;
3180         struct inode *dir = d_inode(parent);
3181         struct inode *inode;
3182         struct btrfs_root *root = BTRFS_I(dir)->root;
3183         struct btrfs_root *dest = NULL;
3184         struct btrfs_ioctl_vol_args *vol_args = NULL;
3185         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3186         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3187         char *subvol_name, *subvol_name_ptr = NULL;
3188         int subvol_namelen;
3189         int err = 0;
3190         bool destroy_parent = false;
3191
3192         if (destroy_v2) {
3193                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3194                 if (IS_ERR(vol_args2))
3195                         return PTR_ERR(vol_args2);
3196
3197                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3198                         err = -EOPNOTSUPP;
3199                         goto out;
3200                 }
3201
3202                 /*
3203                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3204                  * name, same as v1 currently does.
3205                  */
3206                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3207                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3208                         subvol_name = vol_args2->name;
3209
3210                         err = mnt_want_write_file(file);
3211                         if (err)
3212                                 goto out;
3213                 } else {
3214                         struct inode *old_dir;
3215
3216                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3217                                 err = -EINVAL;
3218                                 goto out;
3219                         }
3220
3221                         err = mnt_want_write_file(file);
3222                         if (err)
3223                                 goto out;
3224
3225                         dentry = btrfs_get_dentry(fs_info->sb,
3226                                         BTRFS_FIRST_FREE_OBJECTID,
3227                                         vol_args2->subvolid, 0, 0);
3228                         if (IS_ERR(dentry)) {
3229                                 err = PTR_ERR(dentry);
3230                                 goto out_drop_write;
3231                         }
3232
3233                         /*
3234                          * Change the default parent since the subvolume being
3235                          * deleted can be outside of the current mount point.
3236                          */
3237                         parent = btrfs_get_parent(dentry);
3238
3239                         /*
3240                          * At this point dentry->d_name can point to '/' if the
3241                          * subvolume we want to destroy is outsite of the
3242                          * current mount point, so we need to release the
3243                          * current dentry and execute the lookup to return a new
3244                          * one with ->d_name pointing to the
3245                          * <mount point>/subvol_name.
3246                          */
3247                         dput(dentry);
3248                         if (IS_ERR(parent)) {
3249                                 err = PTR_ERR(parent);
3250                                 goto out_drop_write;
3251                         }
3252                         old_dir = dir;
3253                         dir = d_inode(parent);
3254
3255                         /*
3256                          * If v2 was used with SPEC_BY_ID, a new parent was
3257                          * allocated since the subvolume can be outside of the
3258                          * current mount point. Later on we need to release this
3259                          * new parent dentry.
3260                          */
3261                         destroy_parent = true;
3262
3263                         /*
3264                          * On idmapped mounts, deletion via subvolid is
3265                          * restricted to subvolumes that are immediate
3266                          * ancestors of the inode referenced by the file
3267                          * descriptor in the ioctl. Otherwise the idmapping
3268                          * could potentially be abused to delete subvolumes
3269                          * anywhere in the filesystem the user wouldn't be able
3270                          * to delete without an idmapped mount.
3271                          */
3272                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3273                                 err = -EOPNOTSUPP;
3274                                 goto free_parent;
3275                         }
3276
3277                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3278                                                 fs_info, vol_args2->subvolid);
3279                         if (IS_ERR(subvol_name_ptr)) {
3280                                 err = PTR_ERR(subvol_name_ptr);
3281                                 goto free_parent;
3282                         }
3283                         /* subvol_name_ptr is already nul terminated */
3284                         subvol_name = (char *)kbasename(subvol_name_ptr);
3285                 }
3286         } else {
3287                 vol_args = memdup_user(arg, sizeof(*vol_args));
3288                 if (IS_ERR(vol_args))
3289                         return PTR_ERR(vol_args);
3290
3291                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3292                 subvol_name = vol_args->name;
3293
3294                 err = mnt_want_write_file(file);
3295                 if (err)
3296                         goto out;
3297         }
3298
3299         subvol_namelen = strlen(subvol_name);
3300
3301         if (strchr(subvol_name, '/') ||
3302             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3303                 err = -EINVAL;
3304                 goto free_subvol_name;
3305         }
3306
3307         if (!S_ISDIR(dir->i_mode)) {
3308                 err = -ENOTDIR;
3309                 goto free_subvol_name;
3310         }
3311
3312         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3313         if (err == -EINTR)
3314                 goto free_subvol_name;
3315         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3316         if (IS_ERR(dentry)) {
3317                 err = PTR_ERR(dentry);
3318                 goto out_unlock_dir;
3319         }
3320
3321         if (d_really_is_negative(dentry)) {
3322                 err = -ENOENT;
3323                 goto out_dput;
3324         }
3325
3326         inode = d_inode(dentry);
3327         dest = BTRFS_I(inode)->root;
3328         if (!capable(CAP_SYS_ADMIN)) {
3329                 /*
3330                  * Regular user.  Only allow this with a special mount
3331                  * option, when the user has write+exec access to the
3332                  * subvol root, and when rmdir(2) would have been
3333                  * allowed.
3334                  *
3335                  * Note that this is _not_ check that the subvol is
3336                  * empty or doesn't contain data that we wouldn't
3337                  * otherwise be able to delete.
3338                  *
3339                  * Users who want to delete empty subvols should try
3340                  * rmdir(2).
3341                  */
3342                 err = -EPERM;
3343                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3344                         goto out_dput;
3345
3346                 /*
3347                  * Do not allow deletion if the parent dir is the same
3348                  * as the dir to be deleted.  That means the ioctl
3349                  * must be called on the dentry referencing the root
3350                  * of the subvol, not a random directory contained
3351                  * within it.
3352                  */
3353                 err = -EINVAL;
3354                 if (root == dest)
3355                         goto out_dput;
3356
3357                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3358                 if (err)
3359                         goto out_dput;
3360         }
3361
3362         /* check if subvolume may be deleted by a user */
3363         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3364         if (err)
3365                 goto out_dput;
3366
3367         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3368                 err = -EINVAL;
3369                 goto out_dput;
3370         }
3371
3372         btrfs_inode_lock(inode, 0);
3373         err = btrfs_delete_subvolume(dir, dentry);
3374         btrfs_inode_unlock(inode, 0);
3375         if (!err)
3376                 d_delete_notify(dir, dentry);
3377
3378 out_dput:
3379         dput(dentry);
3380 out_unlock_dir:
3381         btrfs_inode_unlock(dir, 0);
3382 free_subvol_name:
3383         kfree(subvol_name_ptr);
3384 free_parent:
3385         if (destroy_parent)
3386                 dput(parent);
3387 out_drop_write:
3388         mnt_drop_write_file(file);
3389 out:
3390         kfree(vol_args2);
3391         kfree(vol_args);
3392         return err;
3393 }
3394
3395 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3396 {
3397         struct inode *inode = file_inode(file);
3398         struct btrfs_root *root = BTRFS_I(inode)->root;
3399         struct btrfs_ioctl_defrag_range_args range = {0};
3400         int ret;
3401
3402         ret = mnt_want_write_file(file);
3403         if (ret)
3404                 return ret;
3405
3406         if (btrfs_root_readonly(root)) {
3407                 ret = -EROFS;
3408                 goto out;
3409         }
3410
3411         switch (inode->i_mode & S_IFMT) {
3412         case S_IFDIR:
3413                 if (!capable(CAP_SYS_ADMIN)) {
3414                         ret = -EPERM;
3415                         goto out;
3416                 }
3417                 ret = btrfs_defrag_root(root);
3418                 break;
3419         case S_IFREG:
3420                 /*
3421                  * Note that this does not check the file descriptor for write
3422                  * access. This prevents defragmenting executables that are
3423                  * running and allows defrag on files open in read-only mode.
3424                  */
3425                 if (!capable(CAP_SYS_ADMIN) &&
3426                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3427                         ret = -EPERM;
3428                         goto out;
3429                 }
3430
3431                 if (argp) {
3432                         if (copy_from_user(&range, argp, sizeof(range))) {
3433                                 ret = -EFAULT;
3434                                 goto out;
3435                         }
3436                         /* compression requires us to start the IO */
3437                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3438                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3439                                 range.extent_thresh = (u32)-1;
3440                         }
3441                 } else {
3442                         /* the rest are all set to zero by kzalloc */
3443                         range.len = (u64)-1;
3444                 }
3445                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3446                                         &range, BTRFS_OLDEST_GENERATION, 0);
3447                 if (ret > 0)
3448                         ret = 0;
3449                 break;
3450         default:
3451                 ret = -EINVAL;
3452         }
3453 out:
3454         mnt_drop_write_file(file);
3455         return ret;
3456 }
3457
3458 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3459 {
3460         struct btrfs_ioctl_vol_args *vol_args;
3461         bool restore_op = false;
3462         int ret;
3463
3464         if (!capable(CAP_SYS_ADMIN))
3465                 return -EPERM;
3466
3467         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3468                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3469                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3470
3471                 /*
3472                  * We can do the device add because we have a paused balanced,
3473                  * change the exclusive op type and remember we should bring
3474                  * back the paused balance
3475                  */
3476                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3477                 btrfs_exclop_start_unlock(fs_info);
3478                 restore_op = true;
3479         }
3480
3481         vol_args = memdup_user(arg, sizeof(*vol_args));
3482         if (IS_ERR(vol_args)) {
3483                 ret = PTR_ERR(vol_args);
3484                 goto out;
3485         }
3486
3487         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3488         ret = btrfs_init_new_device(fs_info, vol_args->name);
3489
3490         if (!ret)
3491                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3492
3493         kfree(vol_args);
3494 out:
3495         if (restore_op)
3496                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3497         else
3498                 btrfs_exclop_finish(fs_info);
3499         return ret;
3500 }
3501
3502 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3503 {
3504         BTRFS_DEV_LOOKUP_ARGS(args);
3505         struct inode *inode = file_inode(file);
3506         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3507         struct btrfs_ioctl_vol_args_v2 *vol_args;
3508         struct block_device *bdev = NULL;
3509         fmode_t mode;
3510         int ret;
3511         bool cancel = false;
3512
3513         if (!capable(CAP_SYS_ADMIN))
3514                 return -EPERM;
3515
3516         vol_args = memdup_user(arg, sizeof(*vol_args));
3517         if (IS_ERR(vol_args))
3518                 return PTR_ERR(vol_args);
3519
3520         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3521                 ret = -EOPNOTSUPP;
3522                 goto out;
3523         }
3524
3525         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3526         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3527                 args.devid = vol_args->devid;
3528         } else if (!strcmp("cancel", vol_args->name)) {
3529                 cancel = true;
3530         } else {
3531                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3532                 if (ret)
3533                         goto out;
3534         }
3535
3536         ret = mnt_want_write_file(file);
3537         if (ret)
3538                 goto out;
3539
3540         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3541                                            cancel);
3542         if (ret)
3543                 goto err_drop;
3544
3545         /* Exclusive operation is now claimed */
3546         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3547
3548         btrfs_exclop_finish(fs_info);
3549
3550         if (!ret) {
3551                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3552                         btrfs_info(fs_info, "device deleted: id %llu",
3553                                         vol_args->devid);
3554                 else
3555                         btrfs_info(fs_info, "device deleted: %s",
3556                                         vol_args->name);
3557         }
3558 err_drop:
3559         mnt_drop_write_file(file);
3560         if (bdev)
3561                 blkdev_put(bdev, mode);
3562 out:
3563         btrfs_put_dev_args_from_path(&args);
3564         kfree(vol_args);
3565         return ret;
3566 }
3567
3568 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3569 {
3570         BTRFS_DEV_LOOKUP_ARGS(args);
3571         struct inode *inode = file_inode(file);
3572         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3573         struct btrfs_ioctl_vol_args *vol_args;
3574         struct block_device *bdev = NULL;
3575         fmode_t mode;
3576         int ret;
3577         bool cancel = false;
3578
3579         if (!capable(CAP_SYS_ADMIN))
3580                 return -EPERM;
3581
3582         vol_args = memdup_user(arg, sizeof(*vol_args));
3583         if (IS_ERR(vol_args))
3584                 return PTR_ERR(vol_args);
3585
3586         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3587         if (!strcmp("cancel", vol_args->name)) {
3588                 cancel = true;
3589         } else {
3590                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3591                 if (ret)
3592                         goto out;
3593         }
3594
3595         ret = mnt_want_write_file(file);
3596         if (ret)
3597                 goto out;
3598
3599         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3600                                            cancel);
3601         if (ret == 0) {
3602                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3603                 if (!ret)
3604                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3605                 btrfs_exclop_finish(fs_info);
3606         }
3607
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_fs_info(struct btrfs_fs_info *fs_info,
3618                                 void __user *arg)
3619 {
3620         struct btrfs_ioctl_fs_info_args *fi_args;
3621         struct btrfs_device *device;
3622         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3623         u64 flags_in;
3624         int ret = 0;
3625
3626         fi_args = memdup_user(arg, sizeof(*fi_args));
3627         if (IS_ERR(fi_args))
3628                 return PTR_ERR(fi_args);
3629
3630         flags_in = fi_args->flags;
3631         memset(fi_args, 0, sizeof(*fi_args));
3632
3633         rcu_read_lock();
3634         fi_args->num_devices = fs_devices->num_devices;
3635
3636         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3637                 if (device->devid > fi_args->max_id)
3638                         fi_args->max_id = device->devid;
3639         }
3640         rcu_read_unlock();
3641
3642         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3643         fi_args->nodesize = fs_info->nodesize;
3644         fi_args->sectorsize = fs_info->sectorsize;
3645         fi_args->clone_alignment = fs_info->sectorsize;
3646
3647         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3648                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3649                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3650                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3651         }
3652
3653         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3654                 fi_args->generation = fs_info->generation;
3655                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3656         }
3657
3658         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3659                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3660                        sizeof(fi_args->metadata_uuid));
3661                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3662         }
3663
3664         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3665                 ret = -EFAULT;
3666
3667         kfree(fi_args);
3668         return ret;
3669 }
3670
3671 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3672                                  void __user *arg)
3673 {
3674         BTRFS_DEV_LOOKUP_ARGS(args);
3675         struct btrfs_ioctl_dev_info_args *di_args;
3676         struct btrfs_device *dev;
3677         int ret = 0;
3678
3679         di_args = memdup_user(arg, sizeof(*di_args));
3680         if (IS_ERR(di_args))
3681                 return PTR_ERR(di_args);
3682
3683         args.devid = di_args->devid;
3684         if (!btrfs_is_empty_uuid(di_args->uuid))
3685                 args.uuid = di_args->uuid;
3686
3687         rcu_read_lock();
3688         dev = btrfs_find_device(fs_info->fs_devices, &args);
3689         if (!dev) {
3690                 ret = -ENODEV;
3691                 goto out;
3692         }
3693
3694         di_args->devid = dev->devid;
3695         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3696         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3697         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3698         if (dev->name) {
3699                 strncpy(di_args->path, rcu_str_deref(dev->name),
3700                                 sizeof(di_args->path) - 1);
3701                 di_args->path[sizeof(di_args->path) - 1] = 0;
3702         } else {
3703                 di_args->path[0] = '\0';
3704         }
3705
3706 out:
3707         rcu_read_unlock();
3708         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3709                 ret = -EFAULT;
3710
3711         kfree(di_args);
3712         return ret;
3713 }
3714
3715 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3716 {
3717         struct inode *inode = file_inode(file);
3718         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3719         struct btrfs_root *root = BTRFS_I(inode)->root;
3720         struct btrfs_root *new_root;
3721         struct btrfs_dir_item *di;
3722         struct btrfs_trans_handle *trans;
3723         struct btrfs_path *path = NULL;
3724         struct btrfs_disk_key disk_key;
3725         u64 objectid = 0;
3726         u64 dir_id;
3727         int ret;
3728
3729         if (!capable(CAP_SYS_ADMIN))
3730                 return -EPERM;
3731
3732         ret = mnt_want_write_file(file);
3733         if (ret)
3734                 return ret;
3735
3736         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3737                 ret = -EFAULT;
3738                 goto out;
3739         }
3740
3741         if (!objectid)
3742                 objectid = BTRFS_FS_TREE_OBJECTID;
3743
3744         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3745         if (IS_ERR(new_root)) {
3746                 ret = PTR_ERR(new_root);
3747                 goto out;
3748         }
3749         if (!is_fstree(new_root->root_key.objectid)) {
3750                 ret = -ENOENT;
3751                 goto out_free;
3752         }
3753
3754         path = btrfs_alloc_path();
3755         if (!path) {
3756                 ret = -ENOMEM;
3757                 goto out_free;
3758         }
3759
3760         trans = btrfs_start_transaction(root, 1);
3761         if (IS_ERR(trans)) {
3762                 ret = PTR_ERR(trans);
3763                 goto out_free;
3764         }
3765
3766         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3767         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3768                                    dir_id, "default", 7, 1);
3769         if (IS_ERR_OR_NULL(di)) {
3770                 btrfs_release_path(path);
3771                 btrfs_end_transaction(trans);
3772                 btrfs_err(fs_info,
3773                           "Umm, you don't have the default diritem, this isn't going to work");
3774                 ret = -ENOENT;
3775                 goto out_free;
3776         }
3777
3778         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3779         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3780         btrfs_mark_buffer_dirty(path->nodes[0]);
3781         btrfs_release_path(path);
3782
3783         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3784         btrfs_end_transaction(trans);
3785 out_free:
3786         btrfs_put_root(new_root);
3787         btrfs_free_path(path);
3788 out:
3789         mnt_drop_write_file(file);
3790         return ret;
3791 }
3792
3793 static void get_block_group_info(struct list_head *groups_list,
3794                                  struct btrfs_ioctl_space_info *space)
3795 {
3796         struct btrfs_block_group *block_group;
3797
3798         space->total_bytes = 0;
3799         space->used_bytes = 0;
3800         space->flags = 0;
3801         list_for_each_entry(block_group, groups_list, list) {
3802                 space->flags = block_group->flags;
3803                 space->total_bytes += block_group->length;
3804                 space->used_bytes += block_group->used;
3805         }
3806 }
3807
3808 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3809                                    void __user *arg)
3810 {
3811         struct btrfs_ioctl_space_args space_args;
3812         struct btrfs_ioctl_space_info space;
3813         struct btrfs_ioctl_space_info *dest;
3814         struct btrfs_ioctl_space_info *dest_orig;
3815         struct btrfs_ioctl_space_info __user *user_dest;
3816         struct btrfs_space_info *info;
3817         static const u64 types[] = {
3818                 BTRFS_BLOCK_GROUP_DATA,
3819                 BTRFS_BLOCK_GROUP_SYSTEM,
3820                 BTRFS_BLOCK_GROUP_METADATA,
3821                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3822         };
3823         int num_types = 4;
3824         int alloc_size;
3825         int ret = 0;
3826         u64 slot_count = 0;
3827         int i, c;
3828
3829         if (copy_from_user(&space_args,
3830                            (struct btrfs_ioctl_space_args __user *)arg,
3831                            sizeof(space_args)))
3832                 return -EFAULT;
3833
3834         for (i = 0; i < num_types; i++) {
3835                 struct btrfs_space_info *tmp;
3836
3837                 info = NULL;
3838                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3839                         if (tmp->flags == types[i]) {
3840                                 info = tmp;
3841                                 break;
3842                         }
3843                 }
3844
3845                 if (!info)
3846                         continue;
3847
3848                 down_read(&info->groups_sem);
3849                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3850                         if (!list_empty(&info->block_groups[c]))
3851                                 slot_count++;
3852                 }
3853                 up_read(&info->groups_sem);
3854         }
3855
3856         /*
3857          * Global block reserve, exported as a space_info
3858          */
3859         slot_count++;
3860
3861         /* space_slots == 0 means they are asking for a count */
3862         if (space_args.space_slots == 0) {
3863                 space_args.total_spaces = slot_count;
3864                 goto out;
3865         }
3866
3867         slot_count = min_t(u64, space_args.space_slots, slot_count);
3868
3869         alloc_size = sizeof(*dest) * slot_count;
3870
3871         /* we generally have at most 6 or so space infos, one for each raid
3872          * level.  So, a whole page should be more than enough for everyone
3873          */
3874         if (alloc_size > PAGE_SIZE)
3875                 return -ENOMEM;
3876
3877         space_args.total_spaces = 0;
3878         dest = kmalloc(alloc_size, GFP_KERNEL);
3879         if (!dest)
3880                 return -ENOMEM;
3881         dest_orig = dest;
3882
3883         /* now we have a buffer to copy into */
3884         for (i = 0; i < num_types; i++) {
3885                 struct btrfs_space_info *tmp;
3886
3887                 if (!slot_count)
3888                         break;
3889
3890                 info = NULL;
3891                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3892                         if (tmp->flags == types[i]) {
3893                                 info = tmp;
3894                                 break;
3895                         }
3896                 }
3897
3898                 if (!info)
3899                         continue;
3900                 down_read(&info->groups_sem);
3901                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3902                         if (!list_empty(&info->block_groups[c])) {
3903                                 get_block_group_info(&info->block_groups[c],
3904                                                      &space);
3905                                 memcpy(dest, &space, sizeof(space));
3906                                 dest++;
3907                                 space_args.total_spaces++;
3908                                 slot_count--;
3909                         }
3910                         if (!slot_count)
3911                                 break;
3912                 }
3913                 up_read(&info->groups_sem);
3914         }
3915
3916         /*
3917          * Add global block reserve
3918          */
3919         if (slot_count) {
3920                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3921
3922                 spin_lock(&block_rsv->lock);
3923                 space.total_bytes = block_rsv->size;
3924                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3925                 spin_unlock(&block_rsv->lock);
3926                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3927                 memcpy(dest, &space, sizeof(space));
3928                 space_args.total_spaces++;
3929         }
3930
3931         user_dest = (struct btrfs_ioctl_space_info __user *)
3932                 (arg + sizeof(struct btrfs_ioctl_space_args));
3933
3934         if (copy_to_user(user_dest, dest_orig, alloc_size))
3935                 ret = -EFAULT;
3936
3937         kfree(dest_orig);
3938 out:
3939         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3940                 ret = -EFAULT;
3941
3942         return ret;
3943 }
3944
3945 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3946                                             void __user *argp)
3947 {
3948         struct btrfs_trans_handle *trans;
3949         u64 transid;
3950
3951         trans = btrfs_attach_transaction_barrier(root);
3952         if (IS_ERR(trans)) {
3953                 if (PTR_ERR(trans) != -ENOENT)
3954                         return PTR_ERR(trans);
3955
3956                 /* No running transaction, don't bother */
3957                 transid = root->fs_info->last_trans_committed;
3958                 goto out;
3959         }
3960         transid = trans->transid;
3961         btrfs_commit_transaction_async(trans);
3962 out:
3963         if (argp)
3964                 if (copy_to_user(argp, &transid, sizeof(transid)))
3965                         return -EFAULT;
3966         return 0;
3967 }
3968
3969 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3970                                            void __user *argp)
3971 {
3972         u64 transid;
3973
3974         if (argp) {
3975                 if (copy_from_user(&transid, argp, sizeof(transid)))
3976                         return -EFAULT;
3977         } else {
3978                 transid = 0;  /* current trans */
3979         }
3980         return btrfs_wait_for_commit(fs_info, transid);
3981 }
3982
3983 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3984 {
3985         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3986         struct btrfs_ioctl_scrub_args *sa;
3987         int ret;
3988
3989         if (!capable(CAP_SYS_ADMIN))
3990                 return -EPERM;
3991
3992         sa = memdup_user(arg, sizeof(*sa));
3993         if (IS_ERR(sa))
3994                 return PTR_ERR(sa);
3995
3996         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3997                 ret = mnt_want_write_file(file);
3998                 if (ret)
3999                         goto out;
4000         }
4001
4002         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4003                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4004                               0);
4005
4006         /*
4007          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4008          * error. This is important as it allows user space to know how much
4009          * progress scrub has done. For example, if scrub is canceled we get
4010          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4011          * space. Later user space can inspect the progress from the structure
4012          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4013          * previously (btrfs-progs does this).
4014          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4015          * then return -EFAULT to signal the structure was not copied or it may
4016          * be corrupt and unreliable due to a partial copy.
4017          */
4018         if (copy_to_user(arg, sa, sizeof(*sa)))
4019                 ret = -EFAULT;
4020
4021         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4022                 mnt_drop_write_file(file);
4023 out:
4024         kfree(sa);
4025         return ret;
4026 }
4027
4028 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4029 {
4030         if (!capable(CAP_SYS_ADMIN))
4031                 return -EPERM;
4032
4033         return btrfs_scrub_cancel(fs_info);
4034 }
4035
4036 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4037                                        void __user *arg)
4038 {
4039         struct btrfs_ioctl_scrub_args *sa;
4040         int ret;
4041
4042         if (!capable(CAP_SYS_ADMIN))
4043                 return -EPERM;
4044
4045         sa = memdup_user(arg, sizeof(*sa));
4046         if (IS_ERR(sa))
4047                 return PTR_ERR(sa);
4048
4049         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4050
4051         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4052                 ret = -EFAULT;
4053
4054         kfree(sa);
4055         return ret;
4056 }
4057
4058 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4059                                       void __user *arg)
4060 {
4061         struct btrfs_ioctl_get_dev_stats *sa;
4062         int ret;
4063
4064         sa = memdup_user(arg, sizeof(*sa));
4065         if (IS_ERR(sa))
4066                 return PTR_ERR(sa);
4067
4068         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4069                 kfree(sa);
4070                 return -EPERM;
4071         }
4072
4073         ret = btrfs_get_dev_stats(fs_info, sa);
4074
4075         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4076                 ret = -EFAULT;
4077
4078         kfree(sa);
4079         return ret;
4080 }
4081
4082 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4083                                     void __user *arg)
4084 {
4085         struct btrfs_ioctl_dev_replace_args *p;
4086         int ret;
4087
4088         if (!capable(CAP_SYS_ADMIN))
4089                 return -EPERM;
4090
4091         p = memdup_user(arg, sizeof(*p));
4092         if (IS_ERR(p))
4093                 return PTR_ERR(p);
4094
4095         switch (p->cmd) {
4096         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4097                 if (sb_rdonly(fs_info->sb)) {
4098                         ret = -EROFS;
4099                         goto out;
4100                 }
4101                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4102                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4103                 } else {
4104                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4105                         btrfs_exclop_finish(fs_info);
4106                 }
4107                 break;
4108         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4109                 btrfs_dev_replace_status(fs_info, p);
4110                 ret = 0;
4111                 break;
4112         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4113                 p->result = btrfs_dev_replace_cancel(fs_info);
4114                 ret = 0;
4115                 break;
4116         default:
4117                 ret = -EINVAL;
4118                 break;
4119         }
4120
4121         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4122                 ret = -EFAULT;
4123 out:
4124         kfree(p);
4125         return ret;
4126 }
4127
4128 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4129 {
4130         int ret = 0;
4131         int i;
4132         u64 rel_ptr;
4133         int size;
4134         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4135         struct inode_fs_paths *ipath = NULL;
4136         struct btrfs_path *path;
4137
4138         if (!capable(CAP_DAC_READ_SEARCH))
4139                 return -EPERM;
4140
4141         path = btrfs_alloc_path();
4142         if (!path) {
4143                 ret = -ENOMEM;
4144                 goto out;
4145         }
4146
4147         ipa = memdup_user(arg, sizeof(*ipa));
4148         if (IS_ERR(ipa)) {
4149                 ret = PTR_ERR(ipa);
4150                 ipa = NULL;
4151                 goto out;
4152         }
4153
4154         size = min_t(u32, ipa->size, 4096);
4155         ipath = init_ipath(size, root, path);
4156         if (IS_ERR(ipath)) {
4157                 ret = PTR_ERR(ipath);
4158                 ipath = NULL;
4159                 goto out;
4160         }
4161
4162         ret = paths_from_inode(ipa->inum, ipath);
4163         if (ret < 0)
4164                 goto out;
4165
4166         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4167                 rel_ptr = ipath->fspath->val[i] -
4168                           (u64)(unsigned long)ipath->fspath->val;
4169                 ipath->fspath->val[i] = rel_ptr;
4170         }
4171
4172         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4173                            ipath->fspath, size);
4174         if (ret) {
4175                 ret = -EFAULT;
4176                 goto out;
4177         }
4178
4179 out:
4180         btrfs_free_path(path);
4181         free_ipath(ipath);
4182         kfree(ipa);
4183
4184         return ret;
4185 }
4186
4187 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4188 {
4189         struct btrfs_data_container *inodes = ctx;
4190         const size_t c = 3 * sizeof(u64);
4191
4192         if (inodes->bytes_left >= c) {
4193                 inodes->bytes_left -= c;
4194                 inodes->val[inodes->elem_cnt] = inum;
4195                 inodes->val[inodes->elem_cnt + 1] = offset;
4196                 inodes->val[inodes->elem_cnt + 2] = root;
4197                 inodes->elem_cnt += 3;
4198         } else {
4199                 inodes->bytes_missing += c - inodes->bytes_left;
4200                 inodes->bytes_left = 0;
4201                 inodes->elem_missed += 3;
4202         }
4203
4204         return 0;
4205 }
4206
4207 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4208                                         void __user *arg, int version)
4209 {
4210         int ret = 0;
4211         int size;
4212         struct btrfs_ioctl_logical_ino_args *loi;
4213         struct btrfs_data_container *inodes = NULL;
4214         struct btrfs_path *path = NULL;
4215         bool ignore_offset;
4216
4217         if (!capable(CAP_SYS_ADMIN))
4218                 return -EPERM;
4219
4220         loi = memdup_user(arg, sizeof(*loi));
4221         if (IS_ERR(loi))
4222                 return PTR_ERR(loi);
4223
4224         if (version == 1) {
4225                 ignore_offset = false;
4226                 size = min_t(u32, loi->size, SZ_64K);
4227         } else {
4228                 /* All reserved bits must be 0 for now */
4229                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4230                         ret = -EINVAL;
4231                         goto out_loi;
4232                 }
4233                 /* Only accept flags we have defined so far */
4234                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4235                         ret = -EINVAL;
4236                         goto out_loi;
4237                 }
4238                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4239                 size = min_t(u32, loi->size, SZ_16M);
4240         }
4241
4242         path = btrfs_alloc_path();
4243         if (!path) {
4244                 ret = -ENOMEM;
4245                 goto out;
4246         }
4247
4248         inodes = init_data_container(size);
4249         if (IS_ERR(inodes)) {
4250                 ret = PTR_ERR(inodes);
4251                 inodes = NULL;
4252                 goto out;
4253         }
4254
4255         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4256                                           build_ino_list, inodes, ignore_offset);
4257         if (ret == -EINVAL)
4258                 ret = -ENOENT;
4259         if (ret < 0)
4260                 goto out;
4261
4262         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4263                            size);
4264         if (ret)
4265                 ret = -EFAULT;
4266
4267 out:
4268         btrfs_free_path(path);
4269         kvfree(inodes);
4270 out_loi:
4271         kfree(loi);
4272
4273         return ret;
4274 }
4275
4276 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4277                                struct btrfs_ioctl_balance_args *bargs)
4278 {
4279         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4280
4281         bargs->flags = bctl->flags;
4282
4283         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4284                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4285         if (atomic_read(&fs_info->balance_pause_req))
4286                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4287         if (atomic_read(&fs_info->balance_cancel_req))
4288                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4289
4290         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4291         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4292         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4293
4294         spin_lock(&fs_info->balance_lock);
4295         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4296         spin_unlock(&fs_info->balance_lock);
4297 }
4298
4299 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4300 {
4301         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4302         struct btrfs_fs_info *fs_info = root->fs_info;
4303         struct btrfs_ioctl_balance_args *bargs;
4304         struct btrfs_balance_control *bctl;
4305         bool need_unlock; /* for mut. excl. ops lock */
4306         int ret;
4307
4308         if (!arg)
4309                 btrfs_warn(fs_info,
4310         "IOC_BALANCE ioctl (v1) is deprecated and will be removed in kernel 5.18");
4311
4312         if (!capable(CAP_SYS_ADMIN))
4313                 return -EPERM;
4314
4315         ret = mnt_want_write_file(file);
4316         if (ret)
4317                 return ret;
4318
4319 again:
4320         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4321                 mutex_lock(&fs_info->balance_mutex);
4322                 need_unlock = true;
4323                 goto locked;
4324         }
4325
4326         /*
4327          * mut. excl. ops lock is locked.  Three possibilities:
4328          *   (1) some other op is running
4329          *   (2) balance is running
4330          *   (3) balance is paused -- special case (think resume)
4331          */
4332         mutex_lock(&fs_info->balance_mutex);
4333         if (fs_info->balance_ctl) {
4334                 /* this is either (2) or (3) */
4335                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4336                         mutex_unlock(&fs_info->balance_mutex);
4337                         /*
4338                          * Lock released to allow other waiters to continue,
4339                          * we'll reexamine the status again.
4340                          */
4341                         mutex_lock(&fs_info->balance_mutex);
4342
4343                         if (fs_info->balance_ctl &&
4344                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4345                                 /* this is (3) */
4346                                 need_unlock = false;
4347                                 goto locked;
4348                         }
4349
4350                         mutex_unlock(&fs_info->balance_mutex);
4351                         goto again;
4352                 } else {
4353                         /* this is (2) */
4354                         mutex_unlock(&fs_info->balance_mutex);
4355                         ret = -EINPROGRESS;
4356                         goto out;
4357                 }
4358         } else {
4359                 /* this is (1) */
4360                 mutex_unlock(&fs_info->balance_mutex);
4361                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4362                 goto out;
4363         }
4364
4365 locked:
4366
4367         if (arg) {
4368                 bargs = memdup_user(arg, sizeof(*bargs));
4369                 if (IS_ERR(bargs)) {
4370                         ret = PTR_ERR(bargs);
4371                         goto out_unlock;
4372                 }
4373
4374                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4375                         if (!fs_info->balance_ctl) {
4376                                 ret = -ENOTCONN;
4377                                 goto out_bargs;
4378                         }
4379
4380                         bctl = fs_info->balance_ctl;
4381                         spin_lock(&fs_info->balance_lock);
4382                         bctl->flags |= BTRFS_BALANCE_RESUME;
4383                         spin_unlock(&fs_info->balance_lock);
4384                         btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4385
4386                         goto do_balance;
4387                 }
4388         } else {
4389                 bargs = NULL;
4390         }
4391
4392         if (fs_info->balance_ctl) {
4393                 ret = -EINPROGRESS;
4394                 goto out_bargs;
4395         }
4396
4397         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4398         if (!bctl) {
4399                 ret = -ENOMEM;
4400                 goto out_bargs;
4401         }
4402
4403         if (arg) {
4404                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4405                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4406                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4407
4408                 bctl->flags = bargs->flags;
4409         } else {
4410                 /* balance everything - no filters */
4411                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4412         }
4413
4414         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4415                 ret = -EINVAL;
4416                 goto out_bctl;
4417         }
4418
4419 do_balance:
4420         /*
4421          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4422          * bctl is freed in reset_balance_state, or, if restriper was paused
4423          * all the way until unmount, in free_fs_info.  The flag should be
4424          * cleared after reset_balance_state.
4425          */
4426         need_unlock = false;
4427
4428         ret = btrfs_balance(fs_info, bctl, bargs);
4429         bctl = NULL;
4430
4431         if ((ret == 0 || ret == -ECANCELED) && arg) {
4432                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4433                         ret = -EFAULT;
4434         }
4435
4436 out_bctl:
4437         kfree(bctl);
4438 out_bargs:
4439         kfree(bargs);
4440 out_unlock:
4441         mutex_unlock(&fs_info->balance_mutex);
4442         if (need_unlock)
4443                 btrfs_exclop_finish(fs_info);
4444 out:
4445         mnt_drop_write_file(file);
4446         return ret;
4447 }
4448
4449 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4450 {
4451         if (!capable(CAP_SYS_ADMIN))
4452                 return -EPERM;
4453
4454         switch (cmd) {
4455         case BTRFS_BALANCE_CTL_PAUSE:
4456                 return btrfs_pause_balance(fs_info);
4457         case BTRFS_BALANCE_CTL_CANCEL:
4458                 return btrfs_cancel_balance(fs_info);
4459         }
4460
4461         return -EINVAL;
4462 }
4463
4464 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4465                                          void __user *arg)
4466 {
4467         struct btrfs_ioctl_balance_args *bargs;
4468         int ret = 0;
4469
4470         if (!capable(CAP_SYS_ADMIN))
4471                 return -EPERM;
4472
4473         mutex_lock(&fs_info->balance_mutex);
4474         if (!fs_info->balance_ctl) {
4475                 ret = -ENOTCONN;
4476                 goto out;
4477         }
4478
4479         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4480         if (!bargs) {
4481                 ret = -ENOMEM;
4482                 goto out;
4483         }
4484
4485         btrfs_update_ioctl_balance_args(fs_info, bargs);
4486
4487         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4488                 ret = -EFAULT;
4489
4490         kfree(bargs);
4491 out:
4492         mutex_unlock(&fs_info->balance_mutex);
4493         return ret;
4494 }
4495
4496 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4497 {
4498         struct inode *inode = file_inode(file);
4499         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4500         struct btrfs_ioctl_quota_ctl_args *sa;
4501         int ret;
4502
4503         if (!capable(CAP_SYS_ADMIN))
4504                 return -EPERM;
4505
4506         ret = mnt_want_write_file(file);
4507         if (ret)
4508                 return ret;
4509
4510         sa = memdup_user(arg, sizeof(*sa));
4511         if (IS_ERR(sa)) {
4512                 ret = PTR_ERR(sa);
4513                 goto drop_write;
4514         }
4515
4516         down_write(&fs_info->subvol_sem);
4517
4518         switch (sa->cmd) {
4519         case BTRFS_QUOTA_CTL_ENABLE:
4520                 ret = btrfs_quota_enable(fs_info);
4521                 break;
4522         case BTRFS_QUOTA_CTL_DISABLE:
4523                 ret = btrfs_quota_disable(fs_info);
4524                 break;
4525         default:
4526                 ret = -EINVAL;
4527                 break;
4528         }
4529
4530         kfree(sa);
4531         up_write(&fs_info->subvol_sem);
4532 drop_write:
4533         mnt_drop_write_file(file);
4534         return ret;
4535 }
4536
4537 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4538 {
4539         struct inode *inode = file_inode(file);
4540         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4541         struct btrfs_root *root = BTRFS_I(inode)->root;
4542         struct btrfs_ioctl_qgroup_assign_args *sa;
4543         struct btrfs_trans_handle *trans;
4544         int ret;
4545         int err;
4546
4547         if (!capable(CAP_SYS_ADMIN))
4548                 return -EPERM;
4549
4550         ret = mnt_want_write_file(file);
4551         if (ret)
4552                 return ret;
4553
4554         sa = memdup_user(arg, sizeof(*sa));
4555         if (IS_ERR(sa)) {
4556                 ret = PTR_ERR(sa);
4557                 goto drop_write;
4558         }
4559
4560         trans = btrfs_join_transaction(root);
4561         if (IS_ERR(trans)) {
4562                 ret = PTR_ERR(trans);
4563                 goto out;
4564         }
4565
4566         if (sa->assign) {
4567                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4568         } else {
4569                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4570         }
4571
4572         /* update qgroup status and info */
4573         err = btrfs_run_qgroups(trans);
4574         if (err < 0)
4575                 btrfs_handle_fs_error(fs_info, err,
4576                                       "failed to update qgroup status and info");
4577         err = btrfs_end_transaction(trans);
4578         if (err && !ret)
4579                 ret = err;
4580
4581 out:
4582         kfree(sa);
4583 drop_write:
4584         mnt_drop_write_file(file);
4585         return ret;
4586 }
4587
4588 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4589 {
4590         struct inode *inode = file_inode(file);
4591         struct btrfs_root *root = BTRFS_I(inode)->root;
4592         struct btrfs_ioctl_qgroup_create_args *sa;
4593         struct btrfs_trans_handle *trans;
4594         int ret;
4595         int err;
4596
4597         if (!capable(CAP_SYS_ADMIN))
4598                 return -EPERM;
4599
4600         ret = mnt_want_write_file(file);
4601         if (ret)
4602                 return ret;
4603
4604         sa = memdup_user(arg, sizeof(*sa));
4605         if (IS_ERR(sa)) {
4606                 ret = PTR_ERR(sa);
4607                 goto drop_write;
4608         }
4609
4610         if (!sa->qgroupid) {
4611                 ret = -EINVAL;
4612                 goto out;
4613         }
4614
4615         trans = btrfs_join_transaction(root);
4616         if (IS_ERR(trans)) {
4617                 ret = PTR_ERR(trans);
4618                 goto out;
4619         }
4620
4621         if (sa->create) {
4622                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4623         } else {
4624                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4625         }
4626
4627         err = btrfs_end_transaction(trans);
4628         if (err && !ret)
4629                 ret = err;
4630
4631 out:
4632         kfree(sa);
4633 drop_write:
4634         mnt_drop_write_file(file);
4635         return ret;
4636 }
4637
4638 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4639 {
4640         struct inode *inode = file_inode(file);
4641         struct btrfs_root *root = BTRFS_I(inode)->root;
4642         struct btrfs_ioctl_qgroup_limit_args *sa;
4643         struct btrfs_trans_handle *trans;
4644         int ret;
4645         int err;
4646         u64 qgroupid;
4647
4648         if (!capable(CAP_SYS_ADMIN))
4649                 return -EPERM;
4650
4651         ret = mnt_want_write_file(file);
4652         if (ret)
4653                 return ret;
4654
4655         sa = memdup_user(arg, sizeof(*sa));
4656         if (IS_ERR(sa)) {
4657                 ret = PTR_ERR(sa);
4658                 goto drop_write;
4659         }
4660
4661         trans = btrfs_join_transaction(root);
4662         if (IS_ERR(trans)) {
4663                 ret = PTR_ERR(trans);
4664                 goto out;
4665         }
4666
4667         qgroupid = sa->qgroupid;
4668         if (!qgroupid) {
4669                 /* take the current subvol as qgroup */
4670                 qgroupid = root->root_key.objectid;
4671         }
4672
4673         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4674
4675         err = btrfs_end_transaction(trans);
4676         if (err && !ret)
4677                 ret = err;
4678
4679 out:
4680         kfree(sa);
4681 drop_write:
4682         mnt_drop_write_file(file);
4683         return ret;
4684 }
4685
4686 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4687 {
4688         struct inode *inode = file_inode(file);
4689         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4690         struct btrfs_ioctl_quota_rescan_args *qsa;
4691         int ret;
4692
4693         if (!capable(CAP_SYS_ADMIN))
4694                 return -EPERM;
4695
4696         ret = mnt_want_write_file(file);
4697         if (ret)
4698                 return ret;
4699
4700         qsa = memdup_user(arg, sizeof(*qsa));
4701         if (IS_ERR(qsa)) {
4702                 ret = PTR_ERR(qsa);
4703                 goto drop_write;
4704         }
4705
4706         if (qsa->flags) {
4707                 ret = -EINVAL;
4708                 goto out;
4709         }
4710
4711         ret = btrfs_qgroup_rescan(fs_info);
4712
4713 out:
4714         kfree(qsa);
4715 drop_write:
4716         mnt_drop_write_file(file);
4717         return ret;
4718 }
4719
4720 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4721                                                 void __user *arg)
4722 {
4723         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4724
4725         if (!capable(CAP_SYS_ADMIN))
4726                 return -EPERM;
4727
4728         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4729                 qsa.flags = 1;
4730                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4731         }
4732
4733         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4734                 return -EFAULT;
4735
4736         return 0;
4737 }
4738
4739 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4740                                                 void __user *arg)
4741 {
4742         if (!capable(CAP_SYS_ADMIN))
4743                 return -EPERM;
4744
4745         return btrfs_qgroup_wait_for_completion(fs_info, true);
4746 }
4747
4748 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4749                                             struct user_namespace *mnt_userns,
4750                                             struct btrfs_ioctl_received_subvol_args *sa)
4751 {
4752         struct inode *inode = file_inode(file);
4753         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4754         struct btrfs_root *root = BTRFS_I(inode)->root;
4755         struct btrfs_root_item *root_item = &root->root_item;
4756         struct btrfs_trans_handle *trans;
4757         struct timespec64 ct = current_time(inode);
4758         int ret = 0;
4759         int received_uuid_changed;
4760
4761         if (!inode_owner_or_capable(mnt_userns, inode))
4762                 return -EPERM;
4763
4764         ret = mnt_want_write_file(file);
4765         if (ret < 0)
4766                 return ret;
4767
4768         down_write(&fs_info->subvol_sem);
4769
4770         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4771                 ret = -EINVAL;
4772                 goto out;
4773         }
4774
4775         if (btrfs_root_readonly(root)) {
4776                 ret = -EROFS;
4777                 goto out;
4778         }
4779
4780         /*
4781          * 1 - root item
4782          * 2 - uuid items (received uuid + subvol uuid)
4783          */
4784         trans = btrfs_start_transaction(root, 3);
4785         if (IS_ERR(trans)) {
4786                 ret = PTR_ERR(trans);
4787                 trans = NULL;
4788                 goto out;
4789         }
4790
4791         sa->rtransid = trans->transid;
4792         sa->rtime.sec = ct.tv_sec;
4793         sa->rtime.nsec = ct.tv_nsec;
4794
4795         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4796                                        BTRFS_UUID_SIZE);
4797         if (received_uuid_changed &&
4798             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4799                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4800                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4801                                           root->root_key.objectid);
4802                 if (ret && ret != -ENOENT) {
4803                         btrfs_abort_transaction(trans, ret);
4804                         btrfs_end_transaction(trans);
4805                         goto out;
4806                 }
4807         }
4808         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4809         btrfs_set_root_stransid(root_item, sa->stransid);
4810         btrfs_set_root_rtransid(root_item, sa->rtransid);
4811         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4812         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4813         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4814         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4815
4816         ret = btrfs_update_root(trans, fs_info->tree_root,
4817                                 &root->root_key, &root->root_item);
4818         if (ret < 0) {
4819                 btrfs_end_transaction(trans);
4820                 goto out;
4821         }
4822         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4823                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4824                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4825                                           root->root_key.objectid);
4826                 if (ret < 0 && ret != -EEXIST) {
4827                         btrfs_abort_transaction(trans, ret);
4828                         btrfs_end_transaction(trans);
4829                         goto out;
4830                 }
4831         }
4832         ret = btrfs_commit_transaction(trans);
4833 out:
4834         up_write(&fs_info->subvol_sem);
4835         mnt_drop_write_file(file);
4836         return ret;
4837 }
4838
4839 #ifdef CONFIG_64BIT
4840 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4841                                                 void __user *arg)
4842 {
4843         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4844         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4845         int ret = 0;
4846
4847         args32 = memdup_user(arg, sizeof(*args32));
4848         if (IS_ERR(args32))
4849                 return PTR_ERR(args32);
4850
4851         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4852         if (!args64) {
4853                 ret = -ENOMEM;
4854                 goto out;
4855         }
4856
4857         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4858         args64->stransid = args32->stransid;
4859         args64->rtransid = args32->rtransid;
4860         args64->stime.sec = args32->stime.sec;
4861         args64->stime.nsec = args32->stime.nsec;
4862         args64->rtime.sec = args32->rtime.sec;
4863         args64->rtime.nsec = args32->rtime.nsec;
4864         args64->flags = args32->flags;
4865
4866         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4867         if (ret)
4868                 goto out;
4869
4870         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4871         args32->stransid = args64->stransid;
4872         args32->rtransid = args64->rtransid;
4873         args32->stime.sec = args64->stime.sec;
4874         args32->stime.nsec = args64->stime.nsec;
4875         args32->rtime.sec = args64->rtime.sec;
4876         args32->rtime.nsec = args64->rtime.nsec;
4877         args32->flags = args64->flags;
4878
4879         ret = copy_to_user(arg, args32, sizeof(*args32));
4880         if (ret)
4881                 ret = -EFAULT;
4882
4883 out:
4884         kfree(args32);
4885         kfree(args64);
4886         return ret;
4887 }
4888 #endif
4889
4890 static long btrfs_ioctl_set_received_subvol(struct file *file,
4891                                             void __user *arg)
4892 {
4893         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4894         int ret = 0;
4895
4896         sa = memdup_user(arg, sizeof(*sa));
4897         if (IS_ERR(sa))
4898                 return PTR_ERR(sa);
4899
4900         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4901
4902         if (ret)
4903                 goto out;
4904
4905         ret = copy_to_user(arg, sa, sizeof(*sa));
4906         if (ret)
4907                 ret = -EFAULT;
4908
4909 out:
4910         kfree(sa);
4911         return ret;
4912 }
4913
4914 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4915                                         void __user *arg)
4916 {
4917         size_t len;
4918         int ret;
4919         char label[BTRFS_LABEL_SIZE];
4920
4921         spin_lock(&fs_info->super_lock);
4922         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4923         spin_unlock(&fs_info->super_lock);
4924
4925         len = strnlen(label, BTRFS_LABEL_SIZE);
4926
4927         if (len == BTRFS_LABEL_SIZE) {
4928                 btrfs_warn(fs_info,
4929                            "label is too long, return the first %zu bytes",
4930                            --len);
4931         }
4932
4933         ret = copy_to_user(arg, label, len);
4934
4935         return ret ? -EFAULT : 0;
4936 }
4937
4938 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4939 {
4940         struct inode *inode = file_inode(file);
4941         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4942         struct btrfs_root *root = BTRFS_I(inode)->root;
4943         struct btrfs_super_block *super_block = fs_info->super_copy;
4944         struct btrfs_trans_handle *trans;
4945         char label[BTRFS_LABEL_SIZE];
4946         int ret;
4947
4948         if (!capable(CAP_SYS_ADMIN))
4949                 return -EPERM;
4950
4951         if (copy_from_user(label, arg, sizeof(label)))
4952                 return -EFAULT;
4953
4954         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4955                 btrfs_err(fs_info,
4956                           "unable to set label with more than %d bytes",
4957                           BTRFS_LABEL_SIZE - 1);
4958                 return -EINVAL;
4959         }
4960
4961         ret = mnt_want_write_file(file);
4962         if (ret)
4963                 return ret;
4964
4965         trans = btrfs_start_transaction(root, 0);
4966         if (IS_ERR(trans)) {
4967                 ret = PTR_ERR(trans);
4968                 goto out_unlock;
4969         }
4970
4971         spin_lock(&fs_info->super_lock);
4972         strcpy(super_block->label, label);
4973         spin_unlock(&fs_info->super_lock);
4974         ret = btrfs_commit_transaction(trans);
4975
4976 out_unlock:
4977         mnt_drop_write_file(file);
4978         return ret;
4979 }
4980
4981 #define INIT_FEATURE_FLAGS(suffix) \
4982         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4983           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4984           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4985
4986 int btrfs_ioctl_get_supported_features(void __user *arg)
4987 {
4988         static const struct btrfs_ioctl_feature_flags features[3] = {
4989                 INIT_FEATURE_FLAGS(SUPP),
4990                 INIT_FEATURE_FLAGS(SAFE_SET),
4991                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4992         };
4993
4994         if (copy_to_user(arg, &features, sizeof(features)))
4995                 return -EFAULT;
4996
4997         return 0;
4998 }
4999
5000 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5001                                         void __user *arg)
5002 {
5003         struct btrfs_super_block *super_block = fs_info->super_copy;
5004         struct btrfs_ioctl_feature_flags features;
5005
5006         features.compat_flags = btrfs_super_compat_flags(super_block);
5007         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5008         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5009
5010         if (copy_to_user(arg, &features, sizeof(features)))
5011                 return -EFAULT;
5012
5013         return 0;
5014 }
5015
5016 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5017                               enum btrfs_feature_set set,
5018                               u64 change_mask, u64 flags, u64 supported_flags,
5019                               u64 safe_set, u64 safe_clear)
5020 {
5021         const char *type = btrfs_feature_set_name(set);
5022         char *names;
5023         u64 disallowed, unsupported;
5024         u64 set_mask = flags & change_mask;
5025         u64 clear_mask = ~flags & change_mask;
5026
5027         unsupported = set_mask & ~supported_flags;
5028         if (unsupported) {
5029                 names = btrfs_printable_features(set, unsupported);
5030                 if (names) {
5031                         btrfs_warn(fs_info,
5032                                    "this kernel does not support the %s feature bit%s",
5033                                    names, strchr(names, ',') ? "s" : "");
5034                         kfree(names);
5035                 } else
5036                         btrfs_warn(fs_info,
5037                                    "this kernel does not support %s bits 0x%llx",
5038                                    type, unsupported);
5039                 return -EOPNOTSUPP;
5040         }
5041
5042         disallowed = set_mask & ~safe_set;
5043         if (disallowed) {
5044                 names = btrfs_printable_features(set, disallowed);
5045                 if (names) {
5046                         btrfs_warn(fs_info,
5047                                    "can't set the %s feature bit%s while mounted",
5048                                    names, strchr(names, ',') ? "s" : "");
5049                         kfree(names);
5050                 } else
5051                         btrfs_warn(fs_info,
5052                                    "can't set %s bits 0x%llx while mounted",
5053                                    type, disallowed);
5054                 return -EPERM;
5055         }
5056
5057         disallowed = clear_mask & ~safe_clear;
5058         if (disallowed) {
5059                 names = btrfs_printable_features(set, disallowed);
5060                 if (names) {
5061                         btrfs_warn(fs_info,
5062                                    "can't clear the %s feature bit%s while mounted",
5063                                    names, strchr(names, ',') ? "s" : "");
5064                         kfree(names);
5065                 } else
5066                         btrfs_warn(fs_info,
5067                                    "can't clear %s bits 0x%llx while mounted",
5068                                    type, disallowed);
5069                 return -EPERM;
5070         }
5071
5072         return 0;
5073 }
5074
5075 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5076 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5077                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5078                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5079                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5080
5081 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5082 {
5083         struct inode *inode = file_inode(file);
5084         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5085         struct btrfs_root *root = BTRFS_I(inode)->root;
5086         struct btrfs_super_block *super_block = fs_info->super_copy;
5087         struct btrfs_ioctl_feature_flags flags[2];
5088         struct btrfs_trans_handle *trans;
5089         u64 newflags;
5090         int ret;
5091
5092         if (!capable(CAP_SYS_ADMIN))
5093                 return -EPERM;
5094
5095         if (copy_from_user(flags, arg, sizeof(flags)))
5096                 return -EFAULT;
5097
5098         /* Nothing to do */
5099         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5100             !flags[0].incompat_flags)
5101                 return 0;
5102
5103         ret = check_feature(fs_info, flags[0].compat_flags,
5104                             flags[1].compat_flags, COMPAT);
5105         if (ret)
5106                 return ret;
5107
5108         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5109                             flags[1].compat_ro_flags, COMPAT_RO);
5110         if (ret)
5111                 return ret;
5112
5113         ret = check_feature(fs_info, flags[0].incompat_flags,
5114                             flags[1].incompat_flags, INCOMPAT);
5115         if (ret)
5116                 return ret;
5117
5118         ret = mnt_want_write_file(file);
5119         if (ret)
5120                 return ret;
5121
5122         trans = btrfs_start_transaction(root, 0);
5123         if (IS_ERR(trans)) {
5124                 ret = PTR_ERR(trans);
5125                 goto out_drop_write;
5126         }
5127
5128         spin_lock(&fs_info->super_lock);
5129         newflags = btrfs_super_compat_flags(super_block);
5130         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5131         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5132         btrfs_set_super_compat_flags(super_block, newflags);
5133
5134         newflags = btrfs_super_compat_ro_flags(super_block);
5135         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5136         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5137         btrfs_set_super_compat_ro_flags(super_block, newflags);
5138
5139         newflags = btrfs_super_incompat_flags(super_block);
5140         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5141         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5142         btrfs_set_super_incompat_flags(super_block, newflags);
5143         spin_unlock(&fs_info->super_lock);
5144
5145         ret = btrfs_commit_transaction(trans);
5146 out_drop_write:
5147         mnt_drop_write_file(file);
5148
5149         return ret;
5150 }
5151
5152 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5153 {
5154         struct btrfs_ioctl_send_args *arg;
5155         int ret;
5156
5157         if (compat) {
5158 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5159                 struct btrfs_ioctl_send_args_32 args32;
5160
5161                 ret = copy_from_user(&args32, argp, sizeof(args32));
5162                 if (ret)
5163                         return -EFAULT;
5164                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5165                 if (!arg)
5166                         return -ENOMEM;
5167                 arg->send_fd = args32.send_fd;
5168                 arg->clone_sources_count = args32.clone_sources_count;
5169                 arg->clone_sources = compat_ptr(args32.clone_sources);
5170                 arg->parent_root = args32.parent_root;
5171                 arg->flags = args32.flags;
5172                 memcpy(arg->reserved, args32.reserved,
5173                        sizeof(args32.reserved));
5174 #else
5175                 return -ENOTTY;
5176 #endif
5177         } else {
5178                 arg = memdup_user(argp, sizeof(*arg));
5179                 if (IS_ERR(arg))
5180                         return PTR_ERR(arg);
5181         }
5182         ret = btrfs_ioctl_send(file, arg);
5183         kfree(arg);
5184         return ret;
5185 }
5186
5187 long btrfs_ioctl(struct file *file, unsigned int
5188                 cmd, unsigned long arg)
5189 {
5190         struct inode *inode = file_inode(file);
5191         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5192         struct btrfs_root *root = BTRFS_I(inode)->root;
5193         void __user *argp = (void __user *)arg;
5194
5195         switch (cmd) {
5196         case FS_IOC_GETVERSION:
5197                 return btrfs_ioctl_getversion(file, argp);
5198         case FS_IOC_GETFSLABEL:
5199                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5200         case FS_IOC_SETFSLABEL:
5201                 return btrfs_ioctl_set_fslabel(file, argp);
5202         case FITRIM:
5203                 return btrfs_ioctl_fitrim(fs_info, argp);
5204         case BTRFS_IOC_SNAP_CREATE:
5205                 return btrfs_ioctl_snap_create(file, argp, 0);
5206         case BTRFS_IOC_SNAP_CREATE_V2:
5207                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5208         case BTRFS_IOC_SUBVOL_CREATE:
5209                 return btrfs_ioctl_snap_create(file, argp, 1);
5210         case BTRFS_IOC_SUBVOL_CREATE_V2:
5211                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5212         case BTRFS_IOC_SNAP_DESTROY:
5213                 return btrfs_ioctl_snap_destroy(file, argp, false);
5214         case BTRFS_IOC_SNAP_DESTROY_V2:
5215                 return btrfs_ioctl_snap_destroy(file, argp, true);
5216         case BTRFS_IOC_SUBVOL_GETFLAGS:
5217                 return btrfs_ioctl_subvol_getflags(file, argp);
5218         case BTRFS_IOC_SUBVOL_SETFLAGS:
5219                 return btrfs_ioctl_subvol_setflags(file, argp);
5220         case BTRFS_IOC_DEFAULT_SUBVOL:
5221                 return btrfs_ioctl_default_subvol(file, argp);
5222         case BTRFS_IOC_DEFRAG:
5223                 return btrfs_ioctl_defrag(file, NULL);
5224         case BTRFS_IOC_DEFRAG_RANGE:
5225                 return btrfs_ioctl_defrag(file, argp);
5226         case BTRFS_IOC_RESIZE:
5227                 return btrfs_ioctl_resize(file, argp);
5228         case BTRFS_IOC_ADD_DEV:
5229                 return btrfs_ioctl_add_dev(fs_info, argp);
5230         case BTRFS_IOC_RM_DEV:
5231                 return btrfs_ioctl_rm_dev(file, argp);
5232         case BTRFS_IOC_RM_DEV_V2:
5233                 return btrfs_ioctl_rm_dev_v2(file, argp);
5234         case BTRFS_IOC_FS_INFO:
5235                 return btrfs_ioctl_fs_info(fs_info, argp);
5236         case BTRFS_IOC_DEV_INFO:
5237                 return btrfs_ioctl_dev_info(fs_info, argp);
5238         case BTRFS_IOC_BALANCE:
5239                 return btrfs_ioctl_balance(file, NULL);
5240         case BTRFS_IOC_TREE_SEARCH:
5241                 return btrfs_ioctl_tree_search(file, argp);
5242         case BTRFS_IOC_TREE_SEARCH_V2:
5243                 return btrfs_ioctl_tree_search_v2(file, argp);
5244         case BTRFS_IOC_INO_LOOKUP:
5245                 return btrfs_ioctl_ino_lookup(file, argp);
5246         case BTRFS_IOC_INO_PATHS:
5247                 return btrfs_ioctl_ino_to_path(root, argp);
5248         case BTRFS_IOC_LOGICAL_INO:
5249                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5250         case BTRFS_IOC_LOGICAL_INO_V2:
5251                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5252         case BTRFS_IOC_SPACE_INFO:
5253                 return btrfs_ioctl_space_info(fs_info, argp);
5254         case BTRFS_IOC_SYNC: {
5255                 int ret;
5256
5257                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5258                 if (ret)
5259                         return ret;
5260                 ret = btrfs_sync_fs(inode->i_sb, 1);
5261                 /*
5262                  * The transaction thread may want to do more work,
5263                  * namely it pokes the cleaner kthread that will start
5264                  * processing uncleaned subvols.
5265                  */
5266                 wake_up_process(fs_info->transaction_kthread);
5267                 return ret;
5268         }
5269         case BTRFS_IOC_START_SYNC:
5270                 return btrfs_ioctl_start_sync(root, argp);
5271         case BTRFS_IOC_WAIT_SYNC:
5272                 return btrfs_ioctl_wait_sync(fs_info, argp);
5273         case BTRFS_IOC_SCRUB:
5274                 return btrfs_ioctl_scrub(file, argp);
5275         case BTRFS_IOC_SCRUB_CANCEL:
5276                 return btrfs_ioctl_scrub_cancel(fs_info);
5277         case BTRFS_IOC_SCRUB_PROGRESS:
5278                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5279         case BTRFS_IOC_BALANCE_V2:
5280                 return btrfs_ioctl_balance(file, argp);
5281         case BTRFS_IOC_BALANCE_CTL:
5282                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5283         case BTRFS_IOC_BALANCE_PROGRESS:
5284                 return btrfs_ioctl_balance_progress(fs_info, argp);
5285         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5286                 return btrfs_ioctl_set_received_subvol(file, argp);
5287 #ifdef CONFIG_64BIT
5288         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5289                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5290 #endif
5291         case BTRFS_IOC_SEND:
5292                 return _btrfs_ioctl_send(file, argp, false);
5293 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5294         case BTRFS_IOC_SEND_32:
5295                 return _btrfs_ioctl_send(file, argp, true);
5296 #endif
5297         case BTRFS_IOC_GET_DEV_STATS:
5298                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5299         case BTRFS_IOC_QUOTA_CTL:
5300                 return btrfs_ioctl_quota_ctl(file, argp);
5301         case BTRFS_IOC_QGROUP_ASSIGN:
5302                 return btrfs_ioctl_qgroup_assign(file, argp);
5303         case BTRFS_IOC_QGROUP_CREATE:
5304                 return btrfs_ioctl_qgroup_create(file, argp);
5305         case BTRFS_IOC_QGROUP_LIMIT:
5306                 return btrfs_ioctl_qgroup_limit(file, argp);
5307         case BTRFS_IOC_QUOTA_RESCAN:
5308                 return btrfs_ioctl_quota_rescan(file, argp);
5309         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5310                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5311         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5312                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5313         case BTRFS_IOC_DEV_REPLACE:
5314                 return btrfs_ioctl_dev_replace(fs_info, argp);
5315         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5316                 return btrfs_ioctl_get_supported_features(argp);
5317         case BTRFS_IOC_GET_FEATURES:
5318                 return btrfs_ioctl_get_features(fs_info, argp);
5319         case BTRFS_IOC_SET_FEATURES:
5320                 return btrfs_ioctl_set_features(file, argp);
5321         case BTRFS_IOC_GET_SUBVOL_INFO:
5322                 return btrfs_ioctl_get_subvol_info(file, argp);
5323         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5324                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5325         case BTRFS_IOC_INO_LOOKUP_USER:
5326                 return btrfs_ioctl_ino_lookup_user(file, argp);
5327         case FS_IOC_ENABLE_VERITY:
5328                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5329         case FS_IOC_MEASURE_VERITY:
5330                 return fsverity_ioctl_measure(file, argp);
5331         }
5332
5333         return -ENOTTY;
5334 }
5335
5336 #ifdef CONFIG_COMPAT
5337 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5338 {
5339         /*
5340          * These all access 32-bit values anyway so no further
5341          * handling is necessary.
5342          */
5343         switch (cmd) {
5344         case FS_IOC32_GETVERSION:
5345                 cmd = FS_IOC_GETVERSION;
5346                 break;
5347         }
5348
5349         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5350 }
5351 #endif