fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include <linux/fileattr.h>
  30 #include <linux/fsverity.h>
  31 #include <linux/sched/xacct.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "export.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "print-tree.h"
  38 #include "volumes.h"
  39 #include "locking.h"
  40 #include "backref.h"
  41 #include "rcu-string.h"
  42 #include "send.h"
  43 #include "dev-replace.h"
  44 #include "props.h"
  45 #include "sysfs.h"
  46 #include "qgroup.h"
  47 #include "tree-log.h"
  48 #include "compression.h"
  49 #include "space-info.h"
  50 #include "delalloc-space.h"
  51 #include "block-group.h"
  52 #include "subpage.h"
  53
  54 #ifdef CONFIG_64BIT
  55 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  56  * structures are incorrect, as the timespec structure from userspace
  57  * is 4 bytes too small. We define these alternatives here to teach
  58  * the kernel about the 32-bit struct packing.
  59  */
  60 struct btrfs_ioctl_timespec_32 {
  61         __u64 sec;
  62         __u32 nsec;
  63 } __attribute__ ((__packed__));
  64
  65 struct btrfs_ioctl_received_subvol_args_32 {
  66         char    uuid[BTRFS_UUID_SIZE];  /* in */
  67         __u64   stransid;               /* in */
  68         __u64   rtransid;               /* out */
  69         struct btrfs_ioctl_timespec_32 stime; /* in */
  70         struct btrfs_ioctl_timespec_32 rtime; /* out */
  71         __u64   flags;                  /* in */
  72         __u64   reserved[16];           /* in */
  73 } __attribute__ ((__packed__));
  74
  75 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  76                                 struct btrfs_ioctl_received_subvol_args_32)
  77 #endif
  78
  79 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  80 struct btrfs_ioctl_send_args_32 {
  81         __s64 send_fd;                  /* in */
  82         __u64 clone_sources_count;      /* in */
  83         compat_uptr_t clone_sources;    /* in */
  84         __u64 parent_root;              /* in */
  85         __u64 flags;                    /* in */
  86         __u32 version;                  /* in */
  87         __u8  reserved[28];             /* in */
  88 } __attribute__ ((__packed__));
  89
  90 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  91                                struct btrfs_ioctl_send_args_32)
  92
  93 struct btrfs_ioctl_encoded_io_args_32 {
  94         compat_uptr_t iov;
  95         compat_ulong_t iovcnt;
  96         __s64 offset;
  97         __u64 flags;
  98         __u64 len;
  99         __u64 unencoded_len;
 100         __u64 unencoded_offset;
 101         __u32 compression;
 102         __u32 encryption;
 103         __u8 reserved[64];
 104 };
 105
 106 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
 107                                        struct btrfs_ioctl_encoded_io_args_32)
 108 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
 109                                         struct btrfs_ioctl_encoded_io_args_32)
 110 #endif
 111
 112 /* Mask out flags that are inappropriate for the given type of inode. */
 113 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
 114                 unsigned int flags)
 115 {
 116         if (S_ISDIR(inode->i_mode))
 117                 return flags;
 118         else if (S_ISREG(inode->i_mode))
 119                 return flags & ~FS_DIRSYNC_FL;
 120         else
 121                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 122 }
 123
 124 /*
 125  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 126  * ioctl.
 127  */
 128 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 129 {
 130         unsigned int iflags = 0;
 131         u32 flags = binode->flags;
 132         u32 ro_flags = binode->ro_flags;
 133
 134         if (flags & BTRFS_INODE_SYNC)
 135                 iflags |= FS_SYNC_FL;
 136         if (flags & BTRFS_INODE_IMMUTABLE)
 137                 iflags |= FS_IMMUTABLE_FL;
 138         if (flags & BTRFS_INODE_APPEND)
 139                 iflags |= FS_APPEND_FL;
 140         if (flags & BTRFS_INODE_NODUMP)
 141                 iflags |= FS_NODUMP_FL;
 142         if (flags & BTRFS_INODE_NOATIME)
 143                 iflags |= FS_NOATIME_FL;
 144         if (flags & BTRFS_INODE_DIRSYNC)
 145                 iflags |= FS_DIRSYNC_FL;
 146         if (flags & BTRFS_INODE_NODATACOW)
 147                 iflags |= FS_NOCOW_FL;
 148         if (ro_flags & BTRFS_INODE_RO_VERITY)
 149                 iflags |= FS_VERITY_FL;
 150
 151         if (flags & BTRFS_INODE_NOCOMPRESS)
 152                 iflags |= FS_NOCOMP_FL;
 153         else if (flags & BTRFS_INODE_COMPRESS)
 154                 iflags |= FS_COMPR_FL;
 155
 156         return iflags;
 157 }
 158
 159 /*
 160  * Update inode->i_flags based on the btrfs internal flags.
 161  */
 162 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 163 {
 164         struct btrfs_inode *binode = BTRFS_I(inode);
 165         unsigned int new_fl = 0;
 166
 167         if (binode->flags & BTRFS_INODE_SYNC)
 168                 new_fl |= S_SYNC;
 169         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 170                 new_fl |= S_IMMUTABLE;
 171         if (binode->flags & BTRFS_INODE_APPEND)
 172                 new_fl |= S_APPEND;
 173         if (binode->flags & BTRFS_INODE_NOATIME)
 174                 new_fl |= S_NOATIME;
 175         if (binode->flags & BTRFS_INODE_DIRSYNC)
 176                 new_fl |= S_DIRSYNC;
 177         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 178                 new_fl |= S_VERITY;
 179
 180         set_mask_bits(&inode->i_flags,
 181                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 182                       S_VERITY, new_fl);
 183 }
 184
 185 /*
 186  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 187  * the old and new flags are not conflicting
 188  */
 189 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 190 {
 191         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 192                       FS_NOATIME_FL | FS_NODUMP_FL | \
 193                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 194                       FS_NOCOMP_FL | FS_COMPR_FL |
 195                       FS_NOCOW_FL))
 196                 return -EOPNOTSUPP;
 197
 198         /* COMPR and NOCOMP on new/old are valid */
 199         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 200                 return -EINVAL;
 201
 202         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 203                 return -EINVAL;
 204
 205         /* NOCOW and compression options are mutually exclusive */
 206         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 207                 return -EINVAL;
 208         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 215                                     unsigned int flags)
 216 {
 217         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 218                 return -EPERM;
 219
 220         return 0;
 221 }
 222
 223 /*
 224  * Set flags/xflags from the internal inode flags. The remaining items of
 225  * fsxattr are zeroed.
 226  */
 227 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 228 {
 229         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 230
 231         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 232         return 0;
 233 }
 234
 235 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 236                        struct dentry *dentry, struct fileattr *fa)
 237 {
 238         struct inode *inode = d_inode(dentry);
 239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 240         struct btrfs_inode *binode = BTRFS_I(inode);
 241         struct btrfs_root *root = binode->root;
 242         struct btrfs_trans_handle *trans;
 243         unsigned int fsflags, old_fsflags;
 244         int ret;
 245         const char *comp = NULL;
 246         u32 binode_flags;
 247
 248         if (btrfs_root_readonly(root))
 249                 return -EROFS;
 250
 251         if (fileattr_has_fsx(fa))
 252                 return -EOPNOTSUPP;
 253
 254         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 255         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 256         ret = check_fsflags(old_fsflags, fsflags);
 257         if (ret)
 258                 return ret;
 259
 260         ret = check_fsflags_compatible(fs_info, fsflags);
 261         if (ret)
 262                 return ret;
 263
 264         binode_flags = binode->flags;
 265         if (fsflags & FS_SYNC_FL)
 266                 binode_flags |= BTRFS_INODE_SYNC;
 267         else
 268                 binode_flags &= ~BTRFS_INODE_SYNC;
 269         if (fsflags & FS_IMMUTABLE_FL)
 270                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 271         else
 272                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 273         if (fsflags & FS_APPEND_FL)
 274                 binode_flags |= BTRFS_INODE_APPEND;
 275         else
 276                 binode_flags &= ~BTRFS_INODE_APPEND;
 277         if (fsflags & FS_NODUMP_FL)
 278                 binode_flags |= BTRFS_INODE_NODUMP;
 279         else
 280                 binode_flags &= ~BTRFS_INODE_NODUMP;
 281         if (fsflags & FS_NOATIME_FL)
 282                 binode_flags |= BTRFS_INODE_NOATIME;
 283         else
 284                 binode_flags &= ~BTRFS_INODE_NOATIME;
 285
 286         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 287         if (!fa->flags_valid) {
 288                 /* 1 item for the inode */
 289                 trans = btrfs_start_transaction(root, 1);
 290                 if (IS_ERR(trans))
 291                         return PTR_ERR(trans);
 292                 goto update_flags;
 293         }
 294
 295         if (fsflags & FS_DIRSYNC_FL)
 296                 binode_flags |= BTRFS_INODE_DIRSYNC;
 297         else
 298                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 299         if (fsflags & FS_NOCOW_FL) {
 300                 if (S_ISREG(inode->i_mode)) {
 301                         /*
 302                          * It's safe to turn csums off here, no extents exist.
 303                          * Otherwise we want the flag to reflect the real COW
 304                          * status of the file and will not set it.
 305                          */
 306                         if (inode->i_size == 0)
 307                                 binode_flags |= BTRFS_INODE_NODATACOW |
 308                                                 BTRFS_INODE_NODATASUM;
 309                 } else {
 310                         binode_flags |= BTRFS_INODE_NODATACOW;
 311                 }
 312         } else {
 313                 /*
 314                  * Revert back under same assumptions as above
 315                  */
 316                 if (S_ISREG(inode->i_mode)) {
 317                         if (inode->i_size == 0)
 318                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 319                                                   BTRFS_INODE_NODATASUM);
 320                 } else {
 321                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 322                 }
 323         }
 324
 325         /*
 326          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 327          * flag may be changed automatically if compression code won't make
 328          * things smaller.
 329          */
 330         if (fsflags & FS_NOCOMP_FL) {
 331                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 332                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 333         } else if (fsflags & FS_COMPR_FL) {
 334
 335                 if (IS_SWAPFILE(inode))
 336                         return -ETXTBSY;
 337
 338                 binode_flags |= BTRFS_INODE_COMPRESS;
 339                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 340
 341                 comp = btrfs_compress_type2str(fs_info->compress_type);
 342                 if (!comp || comp[0] == 0)
 343                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 344         } else {
 345                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 346         }
 347
 348         /*
 349          * 1 for inode item
 350          * 2 for properties
 351          */
 352         trans = btrfs_start_transaction(root, 3);
 353         if (IS_ERR(trans))
 354                 return PTR_ERR(trans);
 355
 356         if (comp) {
 357                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 358                                      strlen(comp), 0);
 359                 if (ret) {
 360                         btrfs_abort_transaction(trans, ret);
 361                         goto out_end_trans;
 362                 }
 363         } else {
 364                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 365                                      0, 0);
 366                 if (ret && ret != -ENODATA) {
 367                         btrfs_abort_transaction(trans, ret);
 368                         goto out_end_trans;
 369                 }
 370         }
 371
 372 update_flags:
 373         binode->flags = binode_flags;
 374         btrfs_sync_inode_flags_to_i_flags(inode);
 375         inode_inc_iversion(inode);
 376         inode->i_ctime = current_time(inode);
 377         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 378
 379  out_end_trans:
 380         btrfs_end_transaction(trans);
 381         return ret;
 382 }
 383
 384 /*
 385  * Start exclusive operation @type, return true on success
 386  */
 387 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 388                         enum btrfs_exclusive_operation type)
 389 {
 390         bool ret = false;
 391
 392         spin_lock(&fs_info->super_lock);
 393         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 394                 fs_info->exclusive_operation = type;
 395                 ret = true;
 396         }
 397         spin_unlock(&fs_info->super_lock);
 398
 399         return ret;
 400 }
 401
 402 /*
 403  * Conditionally allow to enter the exclusive operation in case it's compatible
 404  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 405  * btrfs_exclop_finish.
 406  *
 407  * Compatibility:
 408  * - the same type is already running
 409  * - when trying to add a device and balance has been paused
 410  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 411  *   must check the condition first that would allow none -> @type
 412  */
 413 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 414                                  enum btrfs_exclusive_operation type)
 415 {
 416         spin_lock(&fs_info->super_lock);
 417         if (fs_info->exclusive_operation == type ||
 418             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 419              type == BTRFS_EXCLOP_DEV_ADD))
 420                 return true;
 421
 422         spin_unlock(&fs_info->super_lock);
 423         return false;
 424 }
 425
 426 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 427 {
 428         spin_unlock(&fs_info->super_lock);
 429 }
 430
 431 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 432 {
 433         spin_lock(&fs_info->super_lock);
 434         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 435         spin_unlock(&fs_info->super_lock);
 436         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 437 }
 438
 439 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 440                           enum btrfs_exclusive_operation op)
 441 {
 442         switch (op) {
 443         case BTRFS_EXCLOP_BALANCE_PAUSED:
 444                 spin_lock(&fs_info->super_lock);
 445                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 446                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD);
 447                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 448                 spin_unlock(&fs_info->super_lock);
 449                 break;
 450         case BTRFS_EXCLOP_BALANCE:
 451                 spin_lock(&fs_info->super_lock);
 452                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 453                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 454                 spin_unlock(&fs_info->super_lock);
 455                 break;
 456         default:
 457                 btrfs_warn(fs_info,
 458                         "invalid exclop balance operation %d requested", op);
 459         }
 460 }
 461
 462 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
 463 {
 464         return put_user(inode->i_generation, arg);
 465 }
 466
 467 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 468                                         void __user *arg)
 469 {
 470         struct btrfs_device *device;
 471         struct request_queue *q;
 472         struct fstrim_range range;
 473         u64 minlen = ULLONG_MAX;
 474         u64 num_devices = 0;
 475         int ret;
 476
 477         if (!capable(CAP_SYS_ADMIN))
 478                 return -EPERM;
 479
 480         /*
 481          * btrfs_trim_block_group() depends on space cache, which is not
 482          * available in zoned filesystem. So, disallow fitrim on a zoned
 483          * filesystem for now.
 484          */
 485         if (btrfs_is_zoned(fs_info))
 486                 return -EOPNOTSUPP;
 487
 488         /*
 489          * If the fs is mounted with nologreplay, which requires it to be
 490          * mounted in RO mode as well, we can not allow discard on free space
 491          * inside block groups, because log trees refer to extents that are not
 492          * pinned in a block group's free space cache (pinning the extents is
 493          * precisely the first phase of replaying a log tree).
 494          */
 495         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 496                 return -EROFS;
 497
 498         rcu_read_lock();
 499         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 500                                 dev_list) {
 501                 if (!device->bdev)
 502                         continue;
 503                 q = bdev_get_queue(device->bdev);
 504                 if (blk_queue_discard(q)) {
 505                         num_devices++;
 506                         minlen = min_t(u64, q->limits.discard_granularity,
 507                                      minlen);
 508                 }
 509         }
 510         rcu_read_unlock();
 511
 512         if (!num_devices)
 513                 return -EOPNOTSUPP;
 514         if (copy_from_user(&range, arg, sizeof(range)))
 515                 return -EFAULT;
 516
 517         /*
 518          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 519          * block group is in the logical address space, which can be any
 520          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 521          */
 522         if (range.len < fs_info->sb->s_blocksize)
 523                 return -EINVAL;
 524
 525         range.minlen = max(range.minlen, minlen);
 526         ret = btrfs_trim_fs(fs_info, &range);
 527         if (ret < 0)
 528                 return ret;
 529
 530         if (copy_to_user(arg, &range, sizeof(range)))
 531                 return -EFAULT;
 532
 533         return 0;
 534 }
 535
 536 int __pure btrfs_is_empty_uuid(u8 *uuid)
 537 {
 538         int i;
 539
 540         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 541                 if (uuid[i])
 542                         return 0;
 543         }
 544         return 1;
 545 }
 546
 547 static noinline int create_subvol(struct user_namespace *mnt_userns,
 548                                   struct inode *dir, struct dentry *dentry,
 549                                   const char *name, int namelen,
 550                                   struct btrfs_qgroup_inherit *inherit)
 551 {
 552         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 553         struct btrfs_trans_handle *trans;
 554         struct btrfs_key key;
 555         struct btrfs_root_item *root_item;
 556         struct btrfs_inode_item *inode_item;
 557         struct extent_buffer *leaf;
 558         struct btrfs_root *root = BTRFS_I(dir)->root;
 559         struct btrfs_root *new_root;
 560         struct btrfs_block_rsv block_rsv;
 561         struct timespec64 cur_time = current_time(dir);
 562         struct inode *inode;
 563         int ret;
 564         dev_t anon_dev = 0;
 565         u64 objectid;
 566         u64 index = 0;
 567
 568         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 569         if (!root_item)
 570                 return -ENOMEM;
 571
 572         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 573         if (ret)
 574                 goto fail_free;
 575
 576         ret = get_anon_bdev(&anon_dev);
 577         if (ret < 0)
 578                 goto fail_free;
 579
 580         /*
 581          * Don't create subvolume whose level is not zero. Or qgroup will be
 582          * screwed up since it assumes subvolume qgroup's level to be 0.
 583          */
 584         if (btrfs_qgroup_level(objectid)) {
 585                 ret = -ENOSPC;
 586                 goto fail_free;
 587         }
 588
 589         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 590         /*
 591          * The same as the snapshot creation, please see the comment
 592          * of create_snapshot().
 593          */
 594         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 595         if (ret)
 596                 goto fail_free;
 597
 598         trans = btrfs_start_transaction(root, 0);
 599         if (IS_ERR(trans)) {
 600                 ret = PTR_ERR(trans);
 601                 btrfs_subvolume_release_metadata(root, &block_rsv);
 602                 goto fail_free;
 603         }
 604         trans->block_rsv = &block_rsv;
 605         trans->bytes_reserved = block_rsv.size;
 606
 607         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 608         if (ret)
 609                 goto fail;
 610
 611         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 612                                       BTRFS_NESTING_NORMAL);
 613         if (IS_ERR(leaf)) {
 614                 ret = PTR_ERR(leaf);
 615                 goto fail;
 616         }
 617
 618         btrfs_mark_buffer_dirty(leaf);
 619
 620         inode_item = &root_item->inode;
 621         btrfs_set_stack_inode_generation(inode_item, 1);
 622         btrfs_set_stack_inode_size(inode_item, 3);
 623         btrfs_set_stack_inode_nlink(inode_item, 1);
 624         btrfs_set_stack_inode_nbytes(inode_item,
 625                                      fs_info->nodesize);
 626         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 627
 628         btrfs_set_root_flags(root_item, 0);
 629         btrfs_set_root_limit(root_item, 0);
 630         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 631
 632         btrfs_set_root_bytenr(root_item, leaf->start);
 633         btrfs_set_root_generation(root_item, trans->transid);
 634         btrfs_set_root_level(root_item, 0);
 635         btrfs_set_root_refs(root_item, 1);
 636         btrfs_set_root_used(root_item, leaf->len);
 637         btrfs_set_root_last_snapshot(root_item, 0);
 638
 639         btrfs_set_root_generation_v2(root_item,
 640                         btrfs_root_generation(root_item));
 641         generate_random_guid(root_item->uuid);
 642         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 643         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 644         root_item->ctime = root_item->otime;
 645         btrfs_set_root_ctransid(root_item, trans->transid);
 646         btrfs_set_root_otransid(root_item, trans->transid);
 647
 648         btrfs_tree_unlock(leaf);
 649
 650         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 651
 652         key.objectid = objectid;
 653         key.offset = 0;
 654         key.type = BTRFS_ROOT_ITEM_KEY;
 655         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 656                                 root_item);
 657         if (ret) {
 658                 /*
 659                  * Since we don't abort the transaction in this case, free the
 660                  * tree block so that we don't leak space and leave the
 661                  * filesystem in an inconsistent state (an extent item in the
 662                  * extent tree with a backreference for a root that does not
 663                  * exists).
 664                  */
 665                 btrfs_tree_lock(leaf);
 666                 btrfs_clean_tree_block(leaf);
 667                 btrfs_tree_unlock(leaf);
 668                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 669                 free_extent_buffer(leaf);
 670                 goto fail;
 671         }
 672
 673         free_extent_buffer(leaf);
 674         leaf = NULL;
 675
 676         key.offset = (u64)-1;
 677         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 678         if (IS_ERR(new_root)) {
 679                 free_anon_bdev(anon_dev);
 680                 ret = PTR_ERR(new_root);
 681                 btrfs_abort_transaction(trans, ret);
 682                 goto fail;
 683         }
 684         /* Freeing will be done in btrfs_put_root() of new_root */
 685         anon_dev = 0;
 686
 687         ret = btrfs_record_root_in_trans(trans, new_root);
 688         if (ret) {
 689                 btrfs_put_root(new_root);
 690                 btrfs_abort_transaction(trans, ret);
 691                 goto fail;
 692         }
 693
 694         ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns);
 695         btrfs_put_root(new_root);
 696         if (ret) {
 697                 /* We potentially lose an unused inode item here */
 698                 btrfs_abort_transaction(trans, ret);
 699                 goto fail;
 700         }
 701
 702         /*
 703          * insert the directory item
 704          */
 705         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 706         if (ret) {
 707                 btrfs_abort_transaction(trans, ret);
 708                 goto fail;
 709         }
 710
 711         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 712                                     BTRFS_FT_DIR, index);
 713         if (ret) {
 714                 btrfs_abort_transaction(trans, ret);
 715                 goto fail;
 716         }
 717
 718         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 719         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 720         if (ret) {
 721                 btrfs_abort_transaction(trans, ret);
 722                 goto fail;
 723         }
 724
 725         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 726                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 727         if (ret) {
 728                 btrfs_abort_transaction(trans, ret);
 729                 goto fail;
 730         }
 731
 732         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 733                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 734         if (ret)
 735                 btrfs_abort_transaction(trans, ret);
 736
 737 fail:
 738         kfree(root_item);
 739         trans->block_rsv = NULL;
 740         trans->bytes_reserved = 0;
 741         btrfs_subvolume_release_metadata(root, &block_rsv);
 742
 743         if (ret)
 744                 btrfs_end_transaction(trans);
 745         else
 746                 ret = btrfs_commit_transaction(trans);
 747
 748         if (!ret) {
 749                 inode = btrfs_lookup_dentry(dir, dentry);
 750                 if (IS_ERR(inode))
 751                         return PTR_ERR(inode);
 752                 d_instantiate(dentry, inode);
 753         }
 754         return ret;
 755
 756 fail_free:
 757         if (anon_dev)
 758                 free_anon_bdev(anon_dev);
 759         kfree(root_item);
 760         return ret;
 761 }
 762
 763 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 764                            struct dentry *dentry, bool readonly,
 765                            struct btrfs_qgroup_inherit *inherit)
 766 {
 767         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 768         struct inode *inode;
 769         struct btrfs_pending_snapshot *pending_snapshot;
 770         struct btrfs_trans_handle *trans;
 771         int ret;
 772
 773         /* We do not support snapshotting right now. */
 774         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 775                 btrfs_warn(fs_info,
 776                            "extent tree v2 doesn't support snapshotting yet");
 777                 return -EOPNOTSUPP;
 778         }
 779
 780         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 781                 return -EINVAL;
 782
 783         if (atomic_read(&root->nr_swapfiles)) {
 784                 btrfs_warn(fs_info,
 785                            "cannot snapshot subvolume with active swapfile");
 786                 return -ETXTBSY;
 787         }
 788
 789         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 790         if (!pending_snapshot)
 791                 return -ENOMEM;
 792
 793         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 794         if (ret < 0)
 795                 goto free_pending;
 796         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 797                         GFP_KERNEL);
 798         pending_snapshot->path = btrfs_alloc_path();
 799         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 800                 ret = -ENOMEM;
 801                 goto free_pending;
 802         }
 803
 804         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 805                              BTRFS_BLOCK_RSV_TEMP);
 806         /*
 807          * 1 - parent dir inode
 808          * 2 - dir entries
 809          * 1 - root item
 810          * 2 - root ref/backref
 811          * 1 - root of snapshot
 812          * 1 - UUID item
 813          */
 814         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 815                                         &pending_snapshot->block_rsv, 8,
 816                                         false);
 817         if (ret)
 818                 goto free_pending;
 819
 820         pending_snapshot->dentry = dentry;
 821         pending_snapshot->root = root;
 822         pending_snapshot->readonly = readonly;
 823         pending_snapshot->dir = dir;
 824         pending_snapshot->inherit = inherit;
 825
 826         trans = btrfs_start_transaction(root, 0);
 827         if (IS_ERR(trans)) {
 828                 ret = PTR_ERR(trans);
 829                 goto fail;
 830         }
 831
 832         trans->pending_snapshot = pending_snapshot;
 833
 834         ret = btrfs_commit_transaction(trans);
 835         if (ret)
 836                 goto fail;
 837
 838         ret = pending_snapshot->error;
 839         if (ret)
 840                 goto fail;
 841
 842         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 843         if (ret)
 844                 goto fail;
 845
 846         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 847         if (IS_ERR(inode)) {
 848                 ret = PTR_ERR(inode);
 849                 goto fail;
 850         }
 851
 852         d_instantiate(dentry, inode);
 853         ret = 0;
 854         pending_snapshot->anon_dev = 0;
 855 fail:
 856         /* Prevent double freeing of anon_dev */
 857         if (ret && pending_snapshot->snap)
 858                 pending_snapshot->snap->anon_dev = 0;
 859         btrfs_put_root(pending_snapshot->snap);
 860         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 861 free_pending:
 862         if (pending_snapshot->anon_dev)
 863                 free_anon_bdev(pending_snapshot->anon_dev);
 864         kfree(pending_snapshot->root_item);
 865         btrfs_free_path(pending_snapshot->path);
 866         kfree(pending_snapshot);
 867
 868         return ret;
 869 }
 870
 871 /*  copy of may_delete in fs/namei.c()
 872  *      Check whether we can remove a link victim from directory dir, check
 873  *  whether the type of victim is right.
 874  *  1. We can't do it if dir is read-only (done in permission())
 875  *  2. We should have write and exec permissions on dir
 876  *  3. We can't remove anything from append-only dir
 877  *  4. We can't do anything with immutable dir (done in permission())
 878  *  5. If the sticky bit on dir is set we should either
 879  *      a. be owner of dir, or
 880  *      b. be owner of victim, or
 881  *      c. have CAP_FOWNER capability
 882  *  6. If the victim is append-only or immutable we can't do anything with
 883  *     links pointing to it.
 884  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 885  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 886  *  9. We can't remove a root or mountpoint.
 887  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 888  *     nfs_async_unlink().
 889  */
 890
 891 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 892                             struct inode *dir, struct dentry *victim, int isdir)
 893 {
 894         int error;
 895
 896         if (d_really_is_negative(victim))
 897                 return -ENOENT;
 898
 899         BUG_ON(d_inode(victim->d_parent) != dir);
 900         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 901
 902         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 903         if (error)
 904                 return error;
 905         if (IS_APPEND(dir))
 906                 return -EPERM;
 907         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 908             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 909             IS_SWAPFILE(d_inode(victim)))
 910                 return -EPERM;
 911         if (isdir) {
 912                 if (!d_is_dir(victim))
 913                         return -ENOTDIR;
 914                 if (IS_ROOT(victim))
 915                         return -EBUSY;
 916         } else if (d_is_dir(victim))
 917                 return -EISDIR;
 918         if (IS_DEADDIR(dir))
 919                 return -ENOENT;
 920         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 921                 return -EBUSY;
 922         return 0;
 923 }
 924
 925 /* copy of may_create in fs/namei.c() */
 926 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 927                                    struct inode *dir, struct dentry *child)
 928 {
 929         if (d_really_is_positive(child))
 930                 return -EEXIST;
 931         if (IS_DEADDIR(dir))
 932                 return -ENOENT;
 933         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 934                 return -EOVERFLOW;
 935         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 936 }
 937
 938 /*
 939  * Create a new subvolume below @parent.  This is largely modeled after
 940  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 941  * inside this filesystem so it's quite a bit simpler.
 942  */
 943 static noinline int btrfs_mksubvol(const struct path *parent,
 944                                    struct user_namespace *mnt_userns,
 945                                    const char *name, int namelen,
 946                                    struct btrfs_root *snap_src,
 947                                    bool readonly,
 948                                    struct btrfs_qgroup_inherit *inherit)
 949 {
 950         struct inode *dir = d_inode(parent->dentry);
 951         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 952         struct dentry *dentry;
 953         int error;
 954
 955         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 956         if (error == -EINTR)
 957                 return error;
 958
 959         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 960         error = PTR_ERR(dentry);
 961         if (IS_ERR(dentry))
 962                 goto out_unlock;
 963
 964         error = btrfs_may_create(mnt_userns, dir, dentry);
 965         if (error)
 966                 goto out_dput;
 967
 968         /*
 969          * even if this name doesn't exist, we may get hash collisions.
 970          * check for them now when we can safely fail
 971          */
 972         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 973                                                dir->i_ino, name,
 974                                                namelen);
 975         if (error)
 976                 goto out_dput;
 977
 978         down_read(&fs_info->subvol_sem);
 979
 980         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 981                 goto out_up_read;
 982
 983         if (snap_src)
 984                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 985         else
 986                 error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit);
 987
 988         if (!error)
 989                 fsnotify_mkdir(dir, dentry);
 990 out_up_read:
 991         up_read(&fs_info->subvol_sem);
 992 out_dput:
 993         dput(dentry);
 994 out_unlock:
 995         btrfs_inode_unlock(dir, 0);
 996         return error;
 997 }
 998
 999 static noinline int btrfs_mksnapshot(const struct path *parent,
1000                                    struct user_namespace *mnt_userns,
1001                                    const char *name, int namelen,
1002                                    struct btrfs_root *root,
1003                                    bool readonly,
1004                                    struct btrfs_qgroup_inherit *inherit)
1005 {
1006         int ret;
1007         bool snapshot_force_cow = false;
1008
1009         /*
1010          * Force new buffered writes to reserve space even when NOCOW is
1011          * possible. This is to avoid later writeback (running dealloc) to
1012          * fallback to COW mode and unexpectedly fail with ENOSPC.
1013          */
1014         btrfs_drew_read_lock(&root->snapshot_lock);
1015
1016         ret = btrfs_start_delalloc_snapshot(root, false);
1017         if (ret)
1018                 goto out;
1019
1020         /*
1021          * All previous writes have started writeback in NOCOW mode, so now
1022          * we force future writes to fallback to COW mode during snapshot
1023          * creation.
1024          */
1025         atomic_inc(&root->snapshot_force_cow);
1026         snapshot_force_cow = true;
1027
1028         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1029
1030         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1031                              root, readonly, inherit);
1032 out:
1033         if (snapshot_force_cow)
1034                 atomic_dec(&root->snapshot_force_cow);
1035         btrfs_drew_read_unlock(&root->snapshot_lock);
1036         return ret;
1037 }
1038
1039 /*
1040  * Defrag specific helper to get an extent map.
1041  *
1042  * Differences between this and btrfs_get_extent() are:
1043  *
1044  * - No extent_map will be added to inode->extent_tree
1045  *   To reduce memory usage in the long run.
1046  *
1047  * - Extra optimization to skip file extents older than @newer_than
1048  *   By using btrfs_search_forward() we can skip entire file ranges that
1049  *   have extents created in past transactions, because btrfs_search_forward()
1050  *   will not visit leaves and nodes with a generation smaller than given
1051  *   minimal generation threshold (@newer_than).
1052  *
1053  * Return valid em if we find a file extent matching the requirement.
1054  * Return NULL if we can not find a file extent matching the requirement.
1055  *
1056  * Return ERR_PTR() for error.
1057  */
1058 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1059                                             u64 start, u64 newer_than)
1060 {
1061         struct btrfs_root *root = inode->root;
1062         struct btrfs_file_extent_item *fi;
1063         struct btrfs_path path = { 0 };
1064         struct extent_map *em;
1065         struct btrfs_key key;
1066         u64 ino = btrfs_ino(inode);
1067         int ret;
1068
1069         em = alloc_extent_map();
1070         if (!em) {
1071                 ret = -ENOMEM;
1072                 goto err;
1073         }
1074
1075         key.objectid = ino;
1076         key.type = BTRFS_EXTENT_DATA_KEY;
1077         key.offset = start;
1078
1079         if (newer_than) {
1080                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1081                 if (ret < 0)
1082                         goto err;
1083                 /* Can't find anything newer */
1084                 if (ret > 0)
1085                         goto not_found;
1086         } else {
1087                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1088                 if (ret < 0)
1089                         goto err;
1090         }
1091         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1092                 /*
1093                  * If btrfs_search_slot() makes path to point beyond nritems,
1094                  * we should not have an empty leaf, as this inode must at
1095                  * least have its INODE_ITEM.
1096                  */
1097                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1098                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1099         }
1100         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1101         /* Perfect match, no need to go one slot back */
1102         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1103             key.offset == start)
1104                 goto iterate;
1105
1106         /* We didn't find a perfect match, needs to go one slot back */
1107         if (path.slots[0] > 0) {
1108                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1109                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1110                         path.slots[0]--;
1111         }
1112
1113 iterate:
1114         /* Iterate through the path to find a file extent covering @start */
1115         while (true) {
1116                 u64 extent_end;
1117
1118                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1119                         goto next;
1120
1121                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1122
1123                 /*
1124                  * We may go one slot back to INODE_REF/XATTR item, then
1125                  * need to go forward until we reach an EXTENT_DATA.
1126                  * But we should still has the correct ino as key.objectid.
1127                  */
1128                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1129                         goto next;
1130
1131                 /* It's beyond our target range, definitely not extent found */
1132                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1133                         goto not_found;
1134
1135                 /*
1136                  *      |       |<- File extent ->|
1137                  *      \- start
1138                  *
1139                  * This means there is a hole between start and key.offset.
1140                  */
1141                 if (key.offset > start) {
1142                         em->start = start;
1143                         em->orig_start = start;
1144                         em->block_start = EXTENT_MAP_HOLE;
1145                         em->len = key.offset - start;
1146                         break;
1147                 }
1148
1149                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1150                                     struct btrfs_file_extent_item);
1151                 extent_end = btrfs_file_extent_end(&path);
1152
1153                 /*
1154                  *      |<- file extent ->|     |
1155                  *                              \- start
1156                  *
1157                  * We haven't reached start, search next slot.
1158                  */
1159                 if (extent_end <= start)
1160                         goto next;
1161
1162                 /* Now this extent covers @start, convert it to em */
1163                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1164                 break;
1165 next:
1166                 ret = btrfs_next_item(root, &path);
1167                 if (ret < 0)
1168                         goto err;
1169                 if (ret > 0)
1170                         goto not_found;
1171         }
1172         btrfs_release_path(&path);
1173         return em;
1174
1175 not_found:
1176         btrfs_release_path(&path);
1177         free_extent_map(em);
1178         return NULL;
1179
1180 err:
1181         btrfs_release_path(&path);
1182         free_extent_map(em);
1183         return ERR_PTR(ret);
1184 }
1185
1186 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1187                                                u64 newer_than, bool locked)
1188 {
1189         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1190         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1191         struct extent_map *em;
1192         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1193
1194         /*
1195          * hopefully we have this extent in the tree already, try without
1196          * the full extent lock
1197          */
1198         read_lock(&em_tree->lock);
1199         em = lookup_extent_mapping(em_tree, start, sectorsize);
1200         read_unlock(&em_tree->lock);
1201
1202         /*
1203          * We can get a merged extent, in that case, we need to re-search
1204          * tree to get the original em for defrag.
1205          *
1206          * If @newer_than is 0 or em::generation < newer_than, we can trust
1207          * this em, as either we don't care about the generation, or the
1208          * merged extent map will be rejected anyway.
1209          */
1210         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1211             newer_than && em->generation >= newer_than) {
1212                 free_extent_map(em);
1213                 em = NULL;
1214         }
1215
1216         if (!em) {
1217                 struct extent_state *cached = NULL;
1218                 u64 end = start + sectorsize - 1;
1219
1220                 /* get the big lock and read metadata off disk */
1221                 if (!locked)
1222                         lock_extent_bits(io_tree, start, end, &cached);
1223                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1224                 if (!locked)
1225                         unlock_extent_cached(io_tree, start, end, &cached);
1226
1227                 if (IS_ERR(em))
1228                         return NULL;
1229         }
1230
1231         return em;
1232 }
1233
1234 static u32 get_extent_max_capacity(const struct extent_map *em)
1235 {
1236         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1237                 return BTRFS_MAX_COMPRESSED;
1238         return BTRFS_MAX_EXTENT_SIZE;
1239 }
1240
1241 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1242                                      bool locked)
1243 {
1244         struct extent_map *next;
1245         bool ret = false;
1246
1247         /* this is the last extent */
1248         if (em->start + em->len >= i_size_read(inode))
1249                 return false;
1250
1251         /*
1252          * We want to check if the next extent can be merged with the current
1253          * one, which can be an extent created in a past generation, so we pass
1254          * a minimum generation of 0 to defrag_lookup_extent().
1255          */
1256         next = defrag_lookup_extent(inode, em->start + em->len, 0, locked);
1257         /* No more em or hole */
1258         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1259                 goto out;
1260         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1261                 goto out;
1262         /*
1263          * If the next extent is at its max capacity, defragging current extent
1264          * makes no sense, as the total number of extents won't change.
1265          */
1266         if (next->len >= get_extent_max_capacity(em))
1267                 goto out;
1268         ret = true;
1269 out:
1270         free_extent_map(next);
1271         return ret;
1272 }
1273
1274 /*
1275  * Prepare one page to be defragged.
1276  *
1277  * This will ensure:
1278  *
1279  * - Returned page is locked and has been set up properly.
1280  * - No ordered extent exists in the page.
1281  * - The page is uptodate.
1282  *
1283  * NOTE: Caller should also wait for page writeback after the cluster is
1284  * prepared, here we don't do writeback wait for each page.
1285  */
1286 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1287                                             pgoff_t index)
1288 {
1289         struct address_space *mapping = inode->vfs_inode.i_mapping;
1290         gfp_t mask = btrfs_alloc_write_mask(mapping);
1291         u64 page_start = (u64)index << PAGE_SHIFT;
1292         u64 page_end = page_start + PAGE_SIZE - 1;
1293         struct extent_state *cached_state = NULL;
1294         struct page *page;
1295         int ret;
1296
1297 again:
1298         page = find_or_create_page(mapping, index, mask);
1299         if (!page)
1300                 return ERR_PTR(-ENOMEM);
1301
1302         /*
1303          * Since we can defragment files opened read-only, we can encounter
1304          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1305          * can't do I/O using huge pages yet, so return an error for now.
1306          * Filesystem transparent huge pages are typically only used for
1307          * executables that explicitly enable them, so this isn't very
1308          * restrictive.
1309          */
1310         if (PageCompound(page)) {
1311                 unlock_page(page);
1312                 put_page(page);
1313                 return ERR_PTR(-ETXTBSY);
1314         }
1315
1316         ret = set_page_extent_mapped(page);
1317         if (ret < 0) {
1318                 unlock_page(page);
1319                 put_page(page);
1320                 return ERR_PTR(ret);
1321         }
1322
1323         /* Wait for any existing ordered extent in the range */
1324         while (1) {
1325                 struct btrfs_ordered_extent *ordered;
1326
1327                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1328                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1329                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1330                                      &cached_state);
1331                 if (!ordered)
1332                         break;
1333
1334                 unlock_page(page);
1335                 btrfs_start_ordered_extent(ordered, 1);
1336                 btrfs_put_ordered_extent(ordered);
1337                 lock_page(page);
1338                 /*
1339                  * We unlocked the page above, so we need check if it was
1340                  * released or not.
1341                  */
1342                 if (page->mapping != mapping || !PagePrivate(page)) {
1343                         unlock_page(page);
1344                         put_page(page);
1345                         goto again;
1346                 }
1347         }
1348
1349         /*
1350          * Now the page range has no ordered extent any more.  Read the page to
1351          * make it uptodate.
1352          */
1353         if (!PageUptodate(page)) {
1354                 btrfs_readpage(NULL, page);
1355                 lock_page(page);
1356                 if (page->mapping != mapping || !PagePrivate(page)) {
1357                         unlock_page(page);
1358                         put_page(page);
1359                         goto again;
1360                 }
1361                 if (!PageUptodate(page)) {
1362                         unlock_page(page);
1363                         put_page(page);
1364                         return ERR_PTR(-EIO);
1365                 }
1366         }
1367         return page;
1368 }
1369
1370 struct defrag_target_range {
1371         struct list_head list;
1372         u64 start;
1373         u64 len;
1374 };
1375
1376 /*
1377  * Collect all valid target extents.
1378  *
1379  * @start:         file offset to lookup
1380  * @len:           length to lookup
1381  * @extent_thresh: file extent size threshold, any extent size >= this value
1382  *                 will be ignored
1383  * @newer_than:    only defrag extents newer than this value
1384  * @do_compress:   whether the defrag is doing compression
1385  *                 if true, @extent_thresh will be ignored and all regular
1386  *                 file extents meeting @newer_than will be targets.
1387  * @locked:        if the range has already held extent lock
1388  * @target_list:   list of targets file extents
1389  */
1390 static int defrag_collect_targets(struct btrfs_inode *inode,
1391                                   u64 start, u64 len, u32 extent_thresh,
1392                                   u64 newer_than, bool do_compress,
1393                                   bool locked, struct list_head *target_list,
1394                                   u64 *last_scanned_ret)
1395 {
1396         bool last_is_target = false;
1397         u64 cur = start;
1398         int ret = 0;
1399
1400         while (cur < start + len) {
1401                 struct extent_map *em;
1402                 struct defrag_target_range *new;
1403                 bool next_mergeable = true;
1404                 u64 range_len;
1405
1406                 last_is_target = false;
1407                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1408                                           newer_than, locked);
1409                 if (!em)
1410                         break;
1411
1412                 /* Skip hole/inline/preallocated extents */
1413                 if (em->block_start >= EXTENT_MAP_LAST_BYTE ||
1414                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1415                         goto next;
1416
1417                 /* Skip older extent */
1418                 if (em->generation < newer_than)
1419                         goto next;
1420
1421                 /* This em is under writeback, no need to defrag */
1422                 if (em->generation == (u64)-1)
1423                         goto next;
1424
1425                 /*
1426                  * Our start offset might be in the middle of an existing extent
1427                  * map, so take that into account.
1428                  */
1429                 range_len = em->len - (cur - em->start);
1430                 /*
1431                  * If this range of the extent map is already flagged for delalloc,
1432                  * skip it, because:
1433                  *
1434                  * 1) We could deadlock later, when trying to reserve space for
1435                  *    delalloc, because in case we can't immediately reserve space
1436                  *    the flusher can start delalloc and wait for the respective
1437                  *    ordered extents to complete. The deadlock would happen
1438                  *    because we do the space reservation while holding the range
1439                  *    locked, and starting writeback, or finishing an ordered
1440                  *    extent, requires locking the range;
1441                  *
1442                  * 2) If there's delalloc there, it means there's dirty pages for
1443                  *    which writeback has not started yet (we clean the delalloc
1444                  *    flag when starting writeback and after creating an ordered
1445                  *    extent). If we mark pages in an adjacent range for defrag,
1446                  *    then we will have a larger contiguous range for delalloc,
1447                  *    very likely resulting in a larger extent after writeback is
1448                  *    triggered (except in a case of free space fragmentation).
1449                  */
1450                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1451                                    EXTENT_DELALLOC, 0, NULL))
1452                         goto next;
1453
1454                 /*
1455                  * For do_compress case, we want to compress all valid file
1456                  * extents, thus no @extent_thresh or mergeable check.
1457                  */
1458                 if (do_compress)
1459                         goto add;
1460
1461                 /* Skip too large extent */
1462                 if (range_len >= extent_thresh)
1463                         goto next;
1464
1465                 /*
1466                  * Skip extents already at its max capacity, this is mostly for
1467                  * compressed extents, which max cap is only 128K.
1468                  */
1469                 if (em->len >= get_extent_max_capacity(em))
1470                         goto next;
1471
1472                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1473                                                           locked);
1474                 if (!next_mergeable) {
1475                         struct defrag_target_range *last;
1476
1477                         /* Empty target list, no way to merge with last entry */
1478                         if (list_empty(target_list))
1479                                 goto next;
1480                         last = list_entry(target_list->prev,
1481                                           struct defrag_target_range, list);
1482                         /* Not mergeable with last entry */
1483                         if (last->start + last->len != cur)
1484                                 goto next;
1485
1486                         /* Mergeable, fall through to add it to @target_list. */
1487                 }
1488
1489 add:
1490                 last_is_target = true;
1491                 range_len = min(extent_map_end(em), start + len) - cur;
1492                 /*
1493                  * This one is a good target, check if it can be merged into
1494                  * last range of the target list.
1495                  */
1496                 if (!list_empty(target_list)) {
1497                         struct defrag_target_range *last;
1498
1499                         last = list_entry(target_list->prev,
1500                                           struct defrag_target_range, list);
1501                         ASSERT(last->start + last->len <= cur);
1502                         if (last->start + last->len == cur) {
1503                                 /* Mergeable, enlarge the last entry */
1504                                 last->len += range_len;
1505                                 goto next;
1506                         }
1507                         /* Fall through to allocate a new entry */
1508                 }
1509
1510                 /* Allocate new defrag_target_range */
1511                 new = kmalloc(sizeof(*new), GFP_NOFS);
1512                 if (!new) {
1513                         free_extent_map(em);
1514                         ret = -ENOMEM;
1515                         break;
1516                 }
1517                 new->start = cur;
1518                 new->len = range_len;
1519                 list_add_tail(&new->list, target_list);
1520
1521 next:
1522                 cur = extent_map_end(em);
1523                 free_extent_map(em);
1524         }
1525         if (ret < 0) {
1526                 struct defrag_target_range *entry;
1527                 struct defrag_target_range *tmp;
1528
1529                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1530                         list_del_init(&entry->list);
1531                         kfree(entry);
1532                 }
1533         }
1534         if (!ret && last_scanned_ret) {
1535                 /*
1536                  * If the last extent is not a target, the caller can skip to
1537                  * the end of that extent.
1538                  * Otherwise, we can only go the end of the specified range.
1539                  */
1540                 if (!last_is_target)
1541                         *last_scanned_ret = max(cur, *last_scanned_ret);
1542                 else
1543                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1544         }
1545         return ret;
1546 }
1547
1548 #define CLUSTER_SIZE    (SZ_256K)
1549 static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1550
1551 /*
1552  * Defrag one contiguous target range.
1553  *
1554  * @inode:      target inode
1555  * @target:     target range to defrag
1556  * @pages:      locked pages covering the defrag range
1557  * @nr_pages:   number of locked pages
1558  *
1559  * Caller should ensure:
1560  *
1561  * - Pages are prepared
1562  *   Pages should be locked, no ordered extent in the pages range,
1563  *   no writeback.
1564  *
1565  * - Extent bits are locked
1566  */
1567 static int defrag_one_locked_target(struct btrfs_inode *inode,
1568                                     struct defrag_target_range *target,
1569                                     struct page **pages, int nr_pages,
1570                                     struct extent_state **cached_state)
1571 {
1572         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1573         struct extent_changeset *data_reserved = NULL;
1574         const u64 start = target->start;
1575         const u64 len = target->len;
1576         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1577         unsigned long start_index = start >> PAGE_SHIFT;
1578         unsigned long first_index = page_index(pages[0]);
1579         int ret = 0;
1580         int i;
1581
1582         ASSERT(last_index - first_index + 1 <= nr_pages);
1583
1584         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1585         if (ret < 0)
1586                 return ret;
1587         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1588                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1589                          EXTENT_DEFRAG, 0, 0, cached_state);
1590         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1591
1592         /* Update the page status */
1593         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1594                 ClearPageChecked(pages[i]);
1595                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1596         }
1597         btrfs_delalloc_release_extents(inode, len);
1598         extent_changeset_free(data_reserved);
1599
1600         return ret;
1601 }
1602
1603 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1604                             u32 extent_thresh, u64 newer_than, bool do_compress,
1605                             u64 *last_scanned_ret)
1606 {
1607         struct extent_state *cached_state = NULL;
1608         struct defrag_target_range *entry;
1609         struct defrag_target_range *tmp;
1610         LIST_HEAD(target_list);
1611         struct page **pages;
1612         const u32 sectorsize = inode->root->fs_info->sectorsize;
1613         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1614         u64 start_index = start >> PAGE_SHIFT;
1615         unsigned int nr_pages = last_index - start_index + 1;
1616         int ret = 0;
1617         int i;
1618
1619         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1620         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1621
1622         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1623         if (!pages)
1624                 return -ENOMEM;
1625
1626         /* Prepare all pages */
1627         for (i = 0; i < nr_pages; i++) {
1628                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1629                 if (IS_ERR(pages[i])) {
1630                         ret = PTR_ERR(pages[i]);
1631                         pages[i] = NULL;
1632                         goto free_pages;
1633                 }
1634         }
1635         for (i = 0; i < nr_pages; i++)
1636                 wait_on_page_writeback(pages[i]);
1637
1638         /* Lock the pages range */
1639         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1640                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1641                          &cached_state);
1642         /*
1643          * Now we have a consistent view about the extent map, re-check
1644          * which range really needs to be defragged.
1645          *
1646          * And this time we have extent locked already, pass @locked = true
1647          * so that we won't relock the extent range and cause deadlock.
1648          */
1649         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1650                                      newer_than, do_compress, true,
1651                                      &target_list, last_scanned_ret);
1652         if (ret < 0)
1653                 goto unlock_extent;
1654
1655         list_for_each_entry(entry, &target_list, list) {
1656                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1657                                                &cached_state);
1658                 if (ret < 0)
1659                         break;
1660         }
1661
1662         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1663                 list_del_init(&entry->list);
1664                 kfree(entry);
1665         }
1666 unlock_extent:
1667         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1668                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1669                              &cached_state);
1670 free_pages:
1671         for (i = 0; i < nr_pages; i++) {
1672                 if (pages[i]) {
1673                         unlock_page(pages[i]);
1674                         put_page(pages[i]);
1675                 }
1676         }
1677         kfree(pages);
1678         return ret;
1679 }
1680
1681 static int defrag_one_cluster(struct btrfs_inode *inode,
1682                               struct file_ra_state *ra,
1683                               u64 start, u32 len, u32 extent_thresh,
1684                               u64 newer_than, bool do_compress,
1685                               unsigned long *sectors_defragged,
1686                               unsigned long max_sectors,
1687                               u64 *last_scanned_ret)
1688 {
1689         const u32 sectorsize = inode->root->fs_info->sectorsize;
1690         struct defrag_target_range *entry;
1691         struct defrag_target_range *tmp;
1692         LIST_HEAD(target_list);
1693         int ret;
1694
1695         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1696                                      newer_than, do_compress, false,
1697                                      &target_list, NULL);
1698         if (ret < 0)
1699                 goto out;
1700
1701         list_for_each_entry(entry, &target_list, list) {
1702                 u32 range_len = entry->len;
1703
1704                 /* Reached or beyond the limit */
1705                 if (max_sectors && *sectors_defragged >= max_sectors) {
1706                         ret = 1;
1707                         break;
1708                 }
1709
1710                 if (max_sectors)
1711                         range_len = min_t(u32, range_len,
1712                                 (max_sectors - *sectors_defragged) * sectorsize);
1713
1714                 /*
1715                  * If defrag_one_range() has updated last_scanned_ret,
1716                  * our range may already be invalid (e.g. hole punched).
1717                  * Skip if our range is before last_scanned_ret, as there is
1718                  * no need to defrag the range anymore.
1719                  */
1720                 if (entry->start + range_len <= *last_scanned_ret)
1721                         continue;
1722
1723                 if (ra)
1724                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1725                                 ra, NULL, entry->start >> PAGE_SHIFT,
1726                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1727                                 (entry->start >> PAGE_SHIFT) + 1);
1728                 /*
1729                  * Here we may not defrag any range if holes are punched before
1730                  * we locked the pages.
1731                  * But that's fine, it only affects the @sectors_defragged
1732                  * accounting.
1733                  */
1734                 ret = defrag_one_range(inode, entry->start, range_len,
1735                                        extent_thresh, newer_than, do_compress,
1736                                        last_scanned_ret);
1737                 if (ret < 0)
1738                         break;
1739                 *sectors_defragged += range_len >>
1740                                       inode->root->fs_info->sectorsize_bits;
1741         }
1742 out:
1743         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1744                 list_del_init(&entry->list);
1745                 kfree(entry);
1746         }
1747         if (ret >= 0)
1748                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1749         return ret;
1750 }
1751
1752 /*
1753  * Entry point to file defragmentation.
1754  *
1755  * @inode:         inode to be defragged
1756  * @ra:            readahead state (can be NUL)
1757  * @range:         defrag options including range and flags
1758  * @newer_than:    minimum transid to defrag
1759  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1760  *                 will be defragged.
1761  *
1762  * Return <0 for error.
1763  * Return >=0 for the number of sectors defragged, and range->start will be updated
1764  * to indicate the file offset where next defrag should be started at.
1765  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1766  *  defragging all the range).
1767  */
1768 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1769                       struct btrfs_ioctl_defrag_range_args *range,
1770                       u64 newer_than, unsigned long max_to_defrag)
1771 {
1772         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1773         unsigned long sectors_defragged = 0;
1774         u64 isize = i_size_read(inode);
1775         u64 cur;
1776         u64 last_byte;
1777         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1778         bool ra_allocated = false;
1779         int compress_type = BTRFS_COMPRESS_ZLIB;
1780         int ret = 0;
1781         u32 extent_thresh = range->extent_thresh;
1782         pgoff_t start_index;
1783
1784         if (isize == 0)
1785                 return 0;
1786
1787         if (range->start >= isize)
1788                 return -EINVAL;
1789
1790         if (do_compress) {
1791                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1792                         return -EINVAL;
1793                 if (range->compress_type)
1794                         compress_type = range->compress_type;
1795         }
1796
1797         if (extent_thresh == 0)
1798                 extent_thresh = SZ_256K;
1799
1800         if (range->start + range->len > range->start) {
1801                 /* Got a specific range */
1802                 last_byte = min(isize, range->start + range->len);
1803         } else {
1804                 /* Defrag until file end */
1805                 last_byte = isize;
1806         }
1807
1808         /* Align the range */
1809         cur = round_down(range->start, fs_info->sectorsize);
1810         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1811
1812         /*
1813          * If we were not given a ra, allocate a readahead context. As
1814          * readahead is just an optimization, defrag will work without it so
1815          * we don't error out.
1816          */
1817         if (!ra) {
1818                 ra_allocated = true;
1819                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1820                 if (ra)
1821                         file_ra_state_init(ra, inode->i_mapping);
1822         }
1823
1824         /*
1825          * Make writeback start from the beginning of the range, so that the
1826          * defrag range can be written sequentially.
1827          */
1828         start_index = cur >> PAGE_SHIFT;
1829         if (start_index < inode->i_mapping->writeback_index)
1830                 inode->i_mapping->writeback_index = start_index;
1831
1832         while (cur < last_byte) {
1833                 const unsigned long prev_sectors_defragged = sectors_defragged;
1834                 u64 last_scanned = cur;
1835                 u64 cluster_end;
1836
1837                 if (btrfs_defrag_cancelled(fs_info)) {
1838                         ret = -EAGAIN;
1839                         break;
1840                 }
1841
1842                 /* We want the cluster end at page boundary when possible */
1843                 cluster_end = (((cur >> PAGE_SHIFT) +
1844                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1845                 cluster_end = min(cluster_end, last_byte);
1846
1847                 btrfs_inode_lock(inode, 0);
1848                 if (IS_SWAPFILE(inode)) {
1849                         ret = -ETXTBSY;
1850                         btrfs_inode_unlock(inode, 0);
1851                         break;
1852                 }
1853                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1854                         btrfs_inode_unlock(inode, 0);
1855                         break;
1856                 }
1857                 if (do_compress)
1858                         BTRFS_I(inode)->defrag_compress = compress_type;
1859                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1860                                 cluster_end + 1 - cur, extent_thresh,
1861                                 newer_than, do_compress, &sectors_defragged,
1862                                 max_to_defrag, &last_scanned);
1863
1864                 if (sectors_defragged > prev_sectors_defragged)
1865                         balance_dirty_pages_ratelimited(inode->i_mapping);
1866
1867                 btrfs_inode_unlock(inode, 0);
1868                 if (ret < 0)
1869                         break;
1870                 cur = max(cluster_end + 1, last_scanned);
1871                 if (ret > 0) {
1872                         ret = 0;
1873                         break;
1874                 }
1875                 cond_resched();
1876         }
1877
1878         if (ra_allocated)
1879                 kfree(ra);
1880         /*
1881          * Update range.start for autodefrag, this will indicate where to start
1882          * in next run.
1883          */
1884         range->start = cur;
1885         if (sectors_defragged) {
1886                 /*
1887                  * We have defragged some sectors, for compression case they
1888                  * need to be written back immediately.
1889                  */
1890                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1891                         filemap_flush(inode->i_mapping);
1892                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1893                                      &BTRFS_I(inode)->runtime_flags))
1894                                 filemap_flush(inode->i_mapping);
1895                 }
1896                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1897                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1898                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1899                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1900                 ret = sectors_defragged;
1901         }
1902         if (do_compress) {
1903                 btrfs_inode_lock(inode, 0);
1904                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1905                 btrfs_inode_unlock(inode, 0);
1906         }
1907         return ret;
1908 }
1909
1910 /*
1911  * Try to start exclusive operation @type or cancel it if it's running.
1912  *
1913  * Return:
1914  *   0        - normal mode, newly claimed op started
1915  *  >0        - normal mode, something else is running,
1916  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1917  * ECANCELED  - cancel mode, successful cancel
1918  * ENOTCONN   - cancel mode, operation not running anymore
1919  */
1920 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1921                         enum btrfs_exclusive_operation type, bool cancel)
1922 {
1923         if (!cancel) {
1924                 /* Start normal op */
1925                 if (!btrfs_exclop_start(fs_info, type))
1926                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1927                 /* Exclusive operation is now claimed */
1928                 return 0;
1929         }
1930
1931         /* Cancel running op */
1932         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1933                 /*
1934                  * This blocks any exclop finish from setting it to NONE, so we
1935                  * request cancellation. Either it runs and we will wait for it,
1936                  * or it has finished and no waiting will happen.
1937                  */
1938                 atomic_inc(&fs_info->reloc_cancel_req);
1939                 btrfs_exclop_start_unlock(fs_info);
1940
1941                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1942                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1943                                     TASK_INTERRUPTIBLE);
1944
1945                 return -ECANCELED;
1946         }
1947
1948         /* Something else is running or none */
1949         return -ENOTCONN;
1950 }
1951
1952 static noinline int btrfs_ioctl_resize(struct file *file,
1953                                         void __user *arg)
1954 {
1955         BTRFS_DEV_LOOKUP_ARGS(args);
1956         struct inode *inode = file_inode(file);
1957         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1958         u64 new_size;
1959         u64 old_size;
1960         u64 devid = 1;
1961         struct btrfs_root *root = BTRFS_I(inode)->root;
1962         struct btrfs_ioctl_vol_args *vol_args;
1963         struct btrfs_trans_handle *trans;
1964         struct btrfs_device *device = NULL;
1965         char *sizestr;
1966         char *retptr;
1967         char *devstr = NULL;
1968         int ret = 0;
1969         int mod = 0;
1970         bool cancel;
1971
1972         if (!capable(CAP_SYS_ADMIN))
1973                 return -EPERM;
1974
1975         ret = mnt_want_write_file(file);
1976         if (ret)
1977                 return ret;
1978
1979         /*
1980          * Read the arguments before checking exclusivity to be able to
1981          * distinguish regular resize and cancel
1982          */
1983         vol_args = memdup_user(arg, sizeof(*vol_args));
1984         if (IS_ERR(vol_args)) {
1985                 ret = PTR_ERR(vol_args);
1986                 goto out_drop;
1987         }
1988         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1989         sizestr = vol_args->name;
1990         cancel = (strcmp("cancel", sizestr) == 0);
1991         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1992         if (ret)
1993                 goto out_free;
1994         /* Exclusive operation is now claimed */
1995
1996         devstr = strchr(sizestr, ':');
1997         if (devstr) {
1998                 sizestr = devstr + 1;
1999                 *devstr = '\0';
2000                 devstr = vol_args->name;
2001                 ret = kstrtoull(devstr, 10, &devid);
2002                 if (ret)
2003                         goto out_finish;
2004                 if (!devid) {
2005                         ret = -EINVAL;
2006                         goto out_finish;
2007                 }
2008                 btrfs_info(fs_info, "resizing devid %llu", devid);
2009         }
2010
2011         args.devid = devid;
2012         device = btrfs_find_device(fs_info->fs_devices, &args);
2013         if (!device) {
2014                 btrfs_info(fs_info, "resizer unable to find device %llu",
2015                            devid);
2016                 ret = -ENODEV;
2017                 goto out_finish;
2018         }
2019
2020         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2021                 btrfs_info(fs_info,
2022                            "resizer unable to apply on readonly device %llu",
2023                        devid);
2024                 ret = -EPERM;
2025                 goto out_finish;
2026         }
2027
2028         if (!strcmp(sizestr, "max"))
2029                 new_size = bdev_nr_bytes(device->bdev);
2030         else {
2031                 if (sizestr[0] == '-') {
2032                         mod = -1;
2033                         sizestr++;
2034                 } else if (sizestr[0] == '+') {
2035                         mod = 1;
2036                         sizestr++;
2037                 }
2038                 new_size = memparse(sizestr, &retptr);
2039                 if (*retptr != '\0' || new_size == 0) {
2040                         ret = -EINVAL;
2041                         goto out_finish;
2042                 }
2043         }
2044
2045         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2046                 ret = -EPERM;
2047                 goto out_finish;
2048         }
2049
2050         old_size = btrfs_device_get_total_bytes(device);
2051
2052         if (mod < 0) {
2053                 if (new_size > old_size) {
2054                         ret = -EINVAL;
2055                         goto out_finish;
2056                 }
2057                 new_size = old_size - new_size;
2058         } else if (mod > 0) {
2059                 if (new_size > ULLONG_MAX - old_size) {
2060                         ret = -ERANGE;
2061                         goto out_finish;
2062                 }
2063                 new_size = old_size + new_size;
2064         }
2065
2066         if (new_size < SZ_256M) {
2067                 ret = -EINVAL;
2068                 goto out_finish;
2069         }
2070         if (new_size > bdev_nr_bytes(device->bdev)) {
2071                 ret = -EFBIG;
2072                 goto out_finish;
2073         }
2074
2075         new_size = round_down(new_size, fs_info->sectorsize);
2076
2077         if (new_size > old_size) {
2078                 trans = btrfs_start_transaction(root, 0);
2079                 if (IS_ERR(trans)) {
2080                         ret = PTR_ERR(trans);
2081                         goto out_finish;
2082                 }
2083                 ret = btrfs_grow_device(trans, device, new_size);
2084                 btrfs_commit_transaction(trans);
2085         } else if (new_size < old_size) {
2086                 ret = btrfs_shrink_device(device, new_size);
2087         } /* equal, nothing need to do */
2088
2089         if (ret == 0 && new_size != old_size)
2090                 btrfs_info_in_rcu(fs_info,
2091                         "resize device %s (devid %llu) from %llu to %llu",
2092                         rcu_str_deref(device->name), device->devid,
2093                         old_size, new_size);
2094 out_finish:
2095         btrfs_exclop_finish(fs_info);
2096 out_free:
2097         kfree(vol_args);
2098 out_drop:
2099         mnt_drop_write_file(file);
2100         return ret;
2101 }
2102
2103 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2104                                 struct user_namespace *mnt_userns,
2105                                 const char *name, unsigned long fd, int subvol,
2106                                 bool readonly,
2107                                 struct btrfs_qgroup_inherit *inherit)
2108 {
2109         int namelen;
2110         int ret = 0;
2111
2112         if (!S_ISDIR(file_inode(file)->i_mode))
2113                 return -ENOTDIR;
2114
2115         ret = mnt_want_write_file(file);
2116         if (ret)
2117                 goto out;
2118
2119         namelen = strlen(name);
2120         if (strchr(name, '/')) {
2121                 ret = -EINVAL;
2122                 goto out_drop_write;
2123         }
2124
2125         if (name[0] == '.' &&
2126            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2127                 ret = -EEXIST;
2128                 goto out_drop_write;
2129         }
2130
2131         if (subvol) {
2132                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2133                                      namelen, NULL, readonly, inherit);
2134         } else {
2135                 struct fd src = fdget(fd);
2136                 struct inode *src_inode;
2137                 if (!src.file) {
2138                         ret = -EINVAL;
2139                         goto out_drop_write;
2140                 }
2141
2142                 src_inode = file_inode(src.file);
2143                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2144                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2145                                    "Snapshot src from another FS");
2146                         ret = -EXDEV;
2147                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2148                         /*
2149                          * Subvolume creation is not restricted, but snapshots
2150                          * are limited to own subvolumes only
2151                          */
2152                         ret = -EPERM;
2153                 } else {
2154                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2155                                                name, namelen,
2156                                                BTRFS_I(src_inode)->root,
2157                                                readonly, inherit);
2158                 }
2159                 fdput(src);
2160         }
2161 out_drop_write:
2162         mnt_drop_write_file(file);
2163 out:
2164         return ret;
2165 }
2166
2167 static noinline int btrfs_ioctl_snap_create(struct file *file,
2168                                             void __user *arg, int subvol)
2169 {
2170         struct btrfs_ioctl_vol_args *vol_args;
2171         int ret;
2172
2173         if (!S_ISDIR(file_inode(file)->i_mode))
2174                 return -ENOTDIR;
2175
2176         vol_args = memdup_user(arg, sizeof(*vol_args));
2177         if (IS_ERR(vol_args))
2178                 return PTR_ERR(vol_args);
2179         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2180
2181         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2182                                         vol_args->name, vol_args->fd, subvol,
2183                                         false, NULL);
2184
2185         kfree(vol_args);
2186         return ret;
2187 }
2188
2189 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2190                                                void __user *arg, int subvol)
2191 {
2192         struct btrfs_ioctl_vol_args_v2 *vol_args;
2193         int ret;
2194         bool readonly = false;
2195         struct btrfs_qgroup_inherit *inherit = NULL;
2196
2197         if (!S_ISDIR(file_inode(file)->i_mode))
2198                 return -ENOTDIR;
2199
2200         vol_args = memdup_user(arg, sizeof(*vol_args));
2201         if (IS_ERR(vol_args))
2202                 return PTR_ERR(vol_args);
2203         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2204
2205         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2206                 ret = -EOPNOTSUPP;
2207                 goto free_args;
2208         }
2209
2210         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2211                 readonly = true;
2212         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2213                 u64 nums;
2214
2215                 if (vol_args->size < sizeof(*inherit) ||
2216                     vol_args->size > PAGE_SIZE) {
2217                         ret = -EINVAL;
2218                         goto free_args;
2219                 }
2220                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2221                 if (IS_ERR(inherit)) {
2222                         ret = PTR_ERR(inherit);
2223                         goto free_args;
2224                 }
2225
2226                 if (inherit->num_qgroups > PAGE_SIZE ||
2227                     inherit->num_ref_copies > PAGE_SIZE ||
2228                     inherit->num_excl_copies > PAGE_SIZE) {
2229                         ret = -EINVAL;
2230                         goto free_inherit;
2231                 }
2232
2233                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2234                        2 * inherit->num_excl_copies;
2235                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2236                         ret = -EINVAL;
2237                         goto free_inherit;
2238                 }
2239         }
2240
2241         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2242                                         vol_args->name, vol_args->fd, subvol,
2243                                         readonly, inherit);
2244         if (ret)
2245                 goto free_inherit;
2246 free_inherit:
2247         kfree(inherit);
2248 free_args:
2249         kfree(vol_args);
2250         return ret;
2251 }
2252
2253 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
2254                                                 void __user *arg)
2255 {
2256         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2257         struct btrfs_root *root = BTRFS_I(inode)->root;
2258         int ret = 0;
2259         u64 flags = 0;
2260
2261         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2262                 return -EINVAL;
2263
2264         down_read(&fs_info->subvol_sem);
2265         if (btrfs_root_readonly(root))
2266                 flags |= BTRFS_SUBVOL_RDONLY;
2267         up_read(&fs_info->subvol_sem);
2268
2269         if (copy_to_user(arg, &flags, sizeof(flags)))
2270                 ret = -EFAULT;
2271
2272         return ret;
2273 }
2274
2275 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2276                                               void __user *arg)
2277 {
2278         struct inode *inode = file_inode(file);
2279         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2280         struct btrfs_root *root = BTRFS_I(inode)->root;
2281         struct btrfs_trans_handle *trans;
2282         u64 root_flags;
2283         u64 flags;
2284         int ret = 0;
2285
2286         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2287                 return -EPERM;
2288
2289         ret = mnt_want_write_file(file);
2290         if (ret)
2291                 goto out;
2292
2293         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2294                 ret = -EINVAL;
2295                 goto out_drop_write;
2296         }
2297
2298         if (copy_from_user(&flags, arg, sizeof(flags))) {
2299                 ret = -EFAULT;
2300                 goto out_drop_write;
2301         }
2302
2303         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2304                 ret = -EOPNOTSUPP;
2305                 goto out_drop_write;
2306         }
2307
2308         down_write(&fs_info->subvol_sem);
2309
2310         /* nothing to do */
2311         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2312                 goto out_drop_sem;
2313
2314         root_flags = btrfs_root_flags(&root->root_item);
2315         if (flags & BTRFS_SUBVOL_RDONLY) {
2316                 btrfs_set_root_flags(&root->root_item,
2317                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2318         } else {
2319                 /*
2320                  * Block RO -> RW transition if this subvolume is involved in
2321                  * send
2322                  */
2323                 spin_lock(&root->root_item_lock);
2324                 if (root->send_in_progress == 0) {
2325                         btrfs_set_root_flags(&root->root_item,
2326                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2327                         spin_unlock(&root->root_item_lock);
2328                 } else {
2329                         spin_unlock(&root->root_item_lock);
2330                         btrfs_warn(fs_info,
2331                                    "Attempt to set subvolume %llu read-write during send",
2332                                    root->root_key.objectid);
2333                         ret = -EPERM;
2334                         goto out_drop_sem;
2335                 }
2336         }
2337
2338         trans = btrfs_start_transaction(root, 1);
2339         if (IS_ERR(trans)) {
2340                 ret = PTR_ERR(trans);
2341                 goto out_reset;
2342         }
2343
2344         ret = btrfs_update_root(trans, fs_info->tree_root,
2345                                 &root->root_key, &root->root_item);
2346         if (ret < 0) {
2347                 btrfs_end_transaction(trans);
2348                 goto out_reset;
2349         }
2350
2351         ret = btrfs_commit_transaction(trans);
2352
2353 out_reset:
2354         if (ret)
2355                 btrfs_set_root_flags(&root->root_item, root_flags);
2356 out_drop_sem:
2357         up_write(&fs_info->subvol_sem);
2358 out_drop_write:
2359         mnt_drop_write_file(file);
2360 out:
2361         return ret;
2362 }
2363
2364 static noinline int key_in_sk(struct btrfs_key *key,
2365                               struct btrfs_ioctl_search_key *sk)
2366 {
2367         struct btrfs_key test;
2368         int ret;
2369
2370         test.objectid = sk->min_objectid;
2371         test.type = sk->min_type;
2372         test.offset = sk->min_offset;
2373
2374         ret = btrfs_comp_cpu_keys(key, &test);
2375         if (ret < 0)
2376                 return 0;
2377
2378         test.objectid = sk->max_objectid;
2379         test.type = sk->max_type;
2380         test.offset = sk->max_offset;
2381
2382         ret = btrfs_comp_cpu_keys(key, &test);
2383         if (ret > 0)
2384                 return 0;
2385         return 1;
2386 }
2387
2388 static noinline int copy_to_sk(struct btrfs_path *path,
2389                                struct btrfs_key *key,
2390                                struct btrfs_ioctl_search_key *sk,
2391                                size_t *buf_size,
2392                                char __user *ubuf,
2393                                unsigned long *sk_offset,
2394                                int *num_found)
2395 {
2396         u64 found_transid;
2397         struct extent_buffer *leaf;
2398         struct btrfs_ioctl_search_header sh;
2399         struct btrfs_key test;
2400         unsigned long item_off;
2401         unsigned long item_len;
2402         int nritems;
2403         int i;
2404         int slot;
2405         int ret = 0;
2406
2407         leaf = path->nodes[0];
2408         slot = path->slots[0];
2409         nritems = btrfs_header_nritems(leaf);
2410
2411         if (btrfs_header_generation(leaf) > sk->max_transid) {
2412                 i = nritems;
2413                 goto advance_key;
2414         }
2415         found_transid = btrfs_header_generation(leaf);
2416
2417         for (i = slot; i < nritems; i++) {
2418                 item_off = btrfs_item_ptr_offset(leaf, i);
2419                 item_len = btrfs_item_size(leaf, i);
2420
2421                 btrfs_item_key_to_cpu(leaf, key, i);
2422                 if (!key_in_sk(key, sk))
2423                         continue;
2424
2425                 if (sizeof(sh) + item_len > *buf_size) {
2426                         if (*num_found) {
2427                                 ret = 1;
2428                                 goto out;
2429                         }
2430
2431                         /*
2432                          * return one empty item back for v1, which does not
2433                          * handle -EOVERFLOW
2434                          */
2435
2436                         *buf_size = sizeof(sh) + item_len;
2437                         item_len = 0;
2438                         ret = -EOVERFLOW;
2439                 }
2440
2441                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2442                         ret = 1;
2443                         goto out;
2444                 }
2445
2446                 sh.objectid = key->objectid;
2447                 sh.offset = key->offset;
2448                 sh.type = key->type;
2449                 sh.len = item_len;
2450                 sh.transid = found_transid;
2451
2452                 /*
2453                  * Copy search result header. If we fault then loop again so we
2454                  * can fault in the pages and -EFAULT there if there's a
2455                  * problem. Otherwise we'll fault and then copy the buffer in
2456                  * properly this next time through
2457                  */
2458                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2459                         ret = 0;
2460                         goto out;
2461                 }
2462
2463                 *sk_offset += sizeof(sh);
2464
2465                 if (item_len) {
2466                         char __user *up = ubuf + *sk_offset;
2467                         /*
2468                          * Copy the item, same behavior as above, but reset the
2469                          * * sk_offset so we copy the full thing again.
2470                          */
2471                         if (read_extent_buffer_to_user_nofault(leaf, up,
2472                                                 item_off, item_len)) {
2473                                 ret = 0;
2474                                 *sk_offset -= sizeof(sh);
2475                                 goto out;
2476                         }
2477
2478                         *sk_offset += item_len;
2479                 }
2480                 (*num_found)++;
2481
2482                 if (ret) /* -EOVERFLOW from above */
2483                         goto out;
2484
2485                 if (*num_found >= sk->nr_items) {
2486                         ret = 1;
2487                         goto out;
2488                 }
2489         }
2490 advance_key:
2491         ret = 0;
2492         test.objectid = sk->max_objectid;
2493         test.type = sk->max_type;
2494         test.offset = sk->max_offset;
2495         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2496                 ret = 1;
2497         else if (key->offset < (u64)-1)
2498                 key->offset++;
2499         else if (key->type < (u8)-1) {
2500                 key->offset = 0;
2501                 key->type++;
2502         } else if (key->objectid < (u64)-1) {
2503                 key->offset = 0;
2504                 key->type = 0;
2505                 key->objectid++;
2506         } else
2507                 ret = 1;
2508 out:
2509         /*
2510          *  0: all items from this leaf copied, continue with next
2511          *  1: * more items can be copied, but unused buffer is too small
2512          *     * all items were found
2513          *     Either way, it will stops the loop which iterates to the next
2514          *     leaf
2515          *  -EOVERFLOW: item was to large for buffer
2516          *  -EFAULT: could not copy extent buffer back to userspace
2517          */
2518         return ret;
2519 }
2520
2521 static noinline int search_ioctl(struct inode *inode,
2522                                  struct btrfs_ioctl_search_key *sk,
2523                                  size_t *buf_size,
2524                                  char __user *ubuf)
2525 {
2526         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2527         struct btrfs_root *root;
2528         struct btrfs_key key;
2529         struct btrfs_path *path;
2530         int ret;
2531         int num_found = 0;
2532         unsigned long sk_offset = 0;
2533
2534         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2535                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2536                 return -EOVERFLOW;
2537         }
2538
2539         path = btrfs_alloc_path();
2540         if (!path)
2541                 return -ENOMEM;
2542
2543         if (sk->tree_id == 0) {
2544                 /* search the root of the inode that was passed */
2545                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2546         } else {
2547                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2548                 if (IS_ERR(root)) {
2549                         btrfs_free_path(path);
2550                         return PTR_ERR(root);
2551                 }
2552         }
2553
2554         key.objectid = sk->min_objectid;
2555         key.type = sk->min_type;
2556         key.offset = sk->min_offset;
2557
2558         while (1) {
2559                 ret = -EFAULT;
2560                 if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
2561                         break;
2562
2563                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2564                 if (ret != 0) {
2565                         if (ret > 0)
2566                                 ret = 0;
2567                         goto err;
2568                 }
2569                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2570                                  &sk_offset, &num_found);
2571                 btrfs_release_path(path);
2572                 if (ret)
2573                         break;
2574
2575         }
2576         if (ret > 0)
2577                 ret = 0;
2578 err:
2579         sk->nr_items = num_found;
2580         btrfs_put_root(root);
2581         btrfs_free_path(path);
2582         return ret;
2583 }
2584
2585 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
2586                                             void __user *argp)
2587 {
2588         struct btrfs_ioctl_search_args __user *uargs;
2589         struct btrfs_ioctl_search_key sk;
2590         int ret;
2591         size_t buf_size;
2592
2593         if (!capable(CAP_SYS_ADMIN))
2594                 return -EPERM;
2595
2596         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2597
2598         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2599                 return -EFAULT;
2600
2601         buf_size = sizeof(uargs->buf);
2602
2603         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2604
2605         /*
2606          * In the origin implementation an overflow is handled by returning a
2607          * search header with a len of zero, so reset ret.
2608          */
2609         if (ret == -EOVERFLOW)
2610                 ret = 0;
2611
2612         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2613                 ret = -EFAULT;
2614         return ret;
2615 }
2616
2617 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
2618                                                void __user *argp)
2619 {
2620         struct btrfs_ioctl_search_args_v2 __user *uarg;
2621         struct btrfs_ioctl_search_args_v2 args;
2622         int ret;
2623         size_t buf_size;
2624         const size_t buf_limit = SZ_16M;
2625
2626         if (!capable(CAP_SYS_ADMIN))
2627                 return -EPERM;
2628
2629         /* copy search header and buffer size */
2630         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2631         if (copy_from_user(&args, uarg, sizeof(args)))
2632                 return -EFAULT;
2633
2634         buf_size = args.buf_size;
2635
2636         /* limit result size to 16MB */
2637         if (buf_size > buf_limit)
2638                 buf_size = buf_limit;
2639
2640         ret = search_ioctl(inode, &args.key, &buf_size,
2641                            (char __user *)(&uarg->buf[0]));
2642         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2643                 ret = -EFAULT;
2644         else if (ret == -EOVERFLOW &&
2645                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2646                 ret = -EFAULT;
2647
2648         return ret;
2649 }
2650
2651 /*
2652  * Search INODE_REFs to identify path name of 'dirid' directory
2653  * in a 'tree_id' tree. and sets path name to 'name'.
2654  */
2655 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2656                                 u64 tree_id, u64 dirid, char *name)
2657 {
2658         struct btrfs_root *root;
2659         struct btrfs_key key;
2660         char *ptr;
2661         int ret = -1;
2662         int slot;
2663         int len;
2664         int total_len = 0;
2665         struct btrfs_inode_ref *iref;
2666         struct extent_buffer *l;
2667         struct btrfs_path *path;
2668
2669         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2670                 name[0]='\0';
2671                 return 0;
2672         }
2673
2674         path = btrfs_alloc_path();
2675         if (!path)
2676                 return -ENOMEM;
2677
2678         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2679
2680         root = btrfs_get_fs_root(info, tree_id, true);
2681         if (IS_ERR(root)) {
2682                 ret = PTR_ERR(root);
2683                 root = NULL;
2684                 goto out;
2685         }
2686
2687         key.objectid = dirid;
2688         key.type = BTRFS_INODE_REF_KEY;
2689         key.offset = (u64)-1;
2690
2691         while (1) {
2692                 ret = btrfs_search_backwards(root, &key, path);
2693                 if (ret < 0)
2694                         goto out;
2695                 else if (ret > 0) {
2696                         ret = -ENOENT;
2697                         goto out;
2698                 }
2699
2700                 l = path->nodes[0];
2701                 slot = path->slots[0];
2702
2703                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2704                 len = btrfs_inode_ref_name_len(l, iref);
2705                 ptr -= len + 1;
2706                 total_len += len + 1;
2707                 if (ptr < name) {
2708                         ret = -ENAMETOOLONG;
2709                         goto out;
2710                 }
2711
2712                 *(ptr + len) = '/';
2713                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2714
2715                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2716                         break;
2717
2718                 btrfs_release_path(path);
2719                 key.objectid = key.offset;
2720                 key.offset = (u64)-1;
2721                 dirid = key.objectid;
2722         }
2723         memmove(name, ptr, total_len);
2724         name[total_len] = '\0';
2725         ret = 0;
2726 out:
2727         btrfs_put_root(root);
2728         btrfs_free_path(path);
2729         return ret;
2730 }
2731
2732 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2733                                 struct inode *inode,
2734                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2735 {
2736         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2737         struct super_block *sb = inode->i_sb;
2738         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2739         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2740         u64 dirid = args->dirid;
2741         unsigned long item_off;
2742         unsigned long item_len;
2743         struct btrfs_inode_ref *iref;
2744         struct btrfs_root_ref *rref;
2745         struct btrfs_root *root = NULL;
2746         struct btrfs_path *path;
2747         struct btrfs_key key, key2;
2748         struct extent_buffer *leaf;
2749         struct inode *temp_inode;
2750         char *ptr;
2751         int slot;
2752         int len;
2753         int total_len = 0;
2754         int ret;
2755
2756         path = btrfs_alloc_path();
2757         if (!path)
2758                 return -ENOMEM;
2759
2760         /*
2761          * If the bottom subvolume does not exist directly under upper_limit,
2762          * construct the path in from the bottom up.
2763          */
2764         if (dirid != upper_limit.objectid) {
2765                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2766
2767                 root = btrfs_get_fs_root(fs_info, treeid, true);
2768                 if (IS_ERR(root)) {
2769                         ret = PTR_ERR(root);
2770                         goto out;
2771                 }
2772
2773                 key.objectid = dirid;
2774                 key.type = BTRFS_INODE_REF_KEY;
2775                 key.offset = (u64)-1;
2776                 while (1) {
2777                         ret = btrfs_search_backwards(root, &key, path);
2778                         if (ret < 0)
2779                                 goto out_put;
2780                         else if (ret > 0) {
2781                                 ret = -ENOENT;
2782                                 goto out_put;
2783                         }
2784
2785                         leaf = path->nodes[0];
2786                         slot = path->slots[0];
2787
2788                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2789                         len = btrfs_inode_ref_name_len(leaf, iref);
2790                         ptr -= len + 1;
2791                         total_len += len + 1;
2792                         if (ptr < args->path) {
2793                                 ret = -ENAMETOOLONG;
2794                                 goto out_put;
2795                         }
2796
2797                         *(ptr + len) = '/';
2798                         read_extent_buffer(leaf, ptr,
2799                                         (unsigned long)(iref + 1), len);
2800
2801                         /* Check the read+exec permission of this directory */
2802                         ret = btrfs_previous_item(root, path, dirid,
2803                                                   BTRFS_INODE_ITEM_KEY);
2804                         if (ret < 0) {
2805                                 goto out_put;
2806                         } else if (ret > 0) {
2807                                 ret = -ENOENT;
2808                                 goto out_put;
2809                         }
2810
2811                         leaf = path->nodes[0];
2812                         slot = path->slots[0];
2813                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2814                         if (key2.objectid != dirid) {
2815                                 ret = -ENOENT;
2816                                 goto out_put;
2817                         }
2818
2819                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2820                         if (IS_ERR(temp_inode)) {
2821                                 ret = PTR_ERR(temp_inode);
2822                                 goto out_put;
2823                         }
2824                         ret = inode_permission(mnt_userns, temp_inode,
2825                                                MAY_READ | MAY_EXEC);
2826                         iput(temp_inode);
2827                         if (ret) {
2828                                 ret = -EACCES;
2829                                 goto out_put;
2830                         }
2831
2832                         if (key.offset == upper_limit.objectid)
2833                                 break;
2834                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2835                                 ret = -EACCES;
2836                                 goto out_put;
2837                         }
2838
2839                         btrfs_release_path(path);
2840                         key.objectid = key.offset;
2841                         key.offset = (u64)-1;
2842                         dirid = key.objectid;
2843                 }
2844
2845                 memmove(args->path, ptr, total_len);
2846                 args->path[total_len] = '\0';
2847                 btrfs_put_root(root);
2848                 root = NULL;
2849                 btrfs_release_path(path);
2850         }
2851
2852         /* Get the bottom subvolume's name from ROOT_REF */
2853         key.objectid = treeid;
2854         key.type = BTRFS_ROOT_REF_KEY;
2855         key.offset = args->treeid;
2856         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2857         if (ret < 0) {
2858                 goto out;
2859         } else if (ret > 0) {
2860                 ret = -ENOENT;
2861                 goto out;
2862         }
2863
2864         leaf = path->nodes[0];
2865         slot = path->slots[0];
2866         btrfs_item_key_to_cpu(leaf, &key, slot);
2867
2868         item_off = btrfs_item_ptr_offset(leaf, slot);
2869         item_len = btrfs_item_size(leaf, slot);
2870         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2871         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2872         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2873                 ret = -EINVAL;
2874                 goto out;
2875         }
2876
2877         /* Copy subvolume's name */
2878         item_off += sizeof(struct btrfs_root_ref);
2879         item_len -= sizeof(struct btrfs_root_ref);
2880         read_extent_buffer(leaf, args->name, item_off, item_len);
2881         args->name[item_len] = 0;
2882
2883 out_put:
2884         btrfs_put_root(root);
2885 out:
2886         btrfs_free_path(path);
2887         return ret;
2888 }
2889
2890 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2891                                            void __user *argp)
2892 {
2893         struct btrfs_ioctl_ino_lookup_args *args;
2894         int ret = 0;
2895
2896         args = memdup_user(argp, sizeof(*args));
2897         if (IS_ERR(args))
2898                 return PTR_ERR(args);
2899
2900         /*
2901          * Unprivileged query to obtain the containing subvolume root id. The
2902          * path is reset so it's consistent with btrfs_search_path_in_tree.
2903          */
2904         if (args->treeid == 0)
2905                 args->treeid = root->root_key.objectid;
2906
2907         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2908                 args->name[0] = 0;
2909                 goto out;
2910         }
2911
2912         if (!capable(CAP_SYS_ADMIN)) {
2913                 ret = -EPERM;
2914                 goto out;
2915         }
2916
2917         ret = btrfs_search_path_in_tree(root->fs_info,
2918                                         args->treeid, args->objectid,
2919                                         args->name);
2920
2921 out:
2922         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2923                 ret = -EFAULT;
2924
2925         kfree(args);
2926         return ret;
2927 }
2928
2929 /*
2930  * Version of ino_lookup ioctl (unprivileged)
2931  *
2932  * The main differences from ino_lookup ioctl are:
2933  *
2934  *   1. Read + Exec permission will be checked using inode_permission() during
2935  *      path construction. -EACCES will be returned in case of failure.
2936  *   2. Path construction will be stopped at the inode number which corresponds
2937  *      to the fd with which this ioctl is called. If constructed path does not
2938  *      exist under fd's inode, -EACCES will be returned.
2939  *   3. The name of bottom subvolume is also searched and filled.
2940  */
2941 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2942 {
2943         struct btrfs_ioctl_ino_lookup_user_args *args;
2944         struct inode *inode;
2945         int ret;
2946
2947         args = memdup_user(argp, sizeof(*args));
2948         if (IS_ERR(args))
2949                 return PTR_ERR(args);
2950
2951         inode = file_inode(file);
2952
2953         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2954             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2955                 /*
2956                  * The subvolume does not exist under fd with which this is
2957                  * called
2958                  */
2959                 kfree(args);
2960                 return -EACCES;
2961         }
2962
2963         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2964
2965         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2966                 ret = -EFAULT;
2967
2968         kfree(args);
2969         return ret;
2970 }
2971
2972 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2973 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2974 {
2975         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2976         struct btrfs_fs_info *fs_info;
2977         struct btrfs_root *root;
2978         struct btrfs_path *path;
2979         struct btrfs_key key;
2980         struct btrfs_root_item *root_item;
2981         struct btrfs_root_ref *rref;
2982         struct extent_buffer *leaf;
2983         unsigned long item_off;
2984         unsigned long item_len;
2985         int slot;
2986         int ret = 0;
2987
2988         path = btrfs_alloc_path();
2989         if (!path)
2990                 return -ENOMEM;
2991
2992         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2993         if (!subvol_info) {
2994                 btrfs_free_path(path);
2995                 return -ENOMEM;
2996         }
2997
2998         fs_info = BTRFS_I(inode)->root->fs_info;
2999
3000         /* Get root_item of inode's subvolume */
3001         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
3002         root = btrfs_get_fs_root(fs_info, key.objectid, true);
3003         if (IS_ERR(root)) {
3004                 ret = PTR_ERR(root);
3005                 goto out_free;
3006         }
3007         root_item = &root->root_item;
3008
3009         subvol_info->treeid = key.objectid;
3010
3011         subvol_info->generation = btrfs_root_generation(root_item);
3012         subvol_info->flags = btrfs_root_flags(root_item);
3013
3014         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3015         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3016                                                     BTRFS_UUID_SIZE);
3017         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3018                                                     BTRFS_UUID_SIZE);
3019
3020         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3021         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3022         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3023
3024         subvol_info->otransid = btrfs_root_otransid(root_item);
3025         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3026         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3027
3028         subvol_info->stransid = btrfs_root_stransid(root_item);
3029         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3030         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3031
3032         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3033         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3034         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3035
3036         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3037                 /* Search root tree for ROOT_BACKREF of this subvolume */
3038                 key.type = BTRFS_ROOT_BACKREF_KEY;
3039                 key.offset = 0;
3040                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3041                 if (ret < 0) {
3042                         goto out;
3043                 } else if (path->slots[0] >=
3044                            btrfs_header_nritems(path->nodes[0])) {
3045                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3046                         if (ret < 0) {
3047                                 goto out;
3048                         } else if (ret > 0) {
3049                                 ret = -EUCLEAN;
3050                                 goto out;
3051                         }
3052                 }
3053
3054                 leaf = path->nodes[0];
3055                 slot = path->slots[0];
3056                 btrfs_item_key_to_cpu(leaf, &key, slot);
3057                 if (key.objectid == subvol_info->treeid &&
3058                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3059                         subvol_info->parent_id = key.offset;
3060
3061                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3062                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3063
3064                         item_off = btrfs_item_ptr_offset(leaf, slot)
3065                                         + sizeof(struct btrfs_root_ref);
3066                         item_len = btrfs_item_size(leaf, slot)
3067                                         - sizeof(struct btrfs_root_ref);
3068                         read_extent_buffer(leaf, subvol_info->name,
3069                                            item_off, item_len);
3070                 } else {
3071                         ret = -ENOENT;
3072                         goto out;
3073                 }
3074         }
3075
3076         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3077                 ret = -EFAULT;
3078
3079 out:
3080         btrfs_put_root(root);
3081 out_free:
3082         btrfs_free_path(path);
3083         kfree(subvol_info);
3084         return ret;
3085 }
3086
3087 /*
3088  * Return ROOT_REF information of the subvolume containing this inode
3089  * except the subvolume name.
3090  */
3091 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3092                                           void __user *argp)
3093 {
3094         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3095         struct btrfs_root_ref *rref;
3096         struct btrfs_path *path;
3097         struct btrfs_key key;
3098         struct extent_buffer *leaf;
3099         u64 objectid;
3100         int slot;
3101         int ret;
3102         u8 found;
3103
3104         path = btrfs_alloc_path();
3105         if (!path)
3106                 return -ENOMEM;
3107
3108         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3109         if (IS_ERR(rootrefs)) {
3110                 btrfs_free_path(path);
3111                 return PTR_ERR(rootrefs);
3112         }
3113
3114         objectid = root->root_key.objectid;
3115         key.objectid = objectid;
3116         key.type = BTRFS_ROOT_REF_KEY;
3117         key.offset = rootrefs->min_treeid;
3118         found = 0;
3119
3120         root = root->fs_info->tree_root;
3121         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3122         if (ret < 0) {
3123                 goto out;
3124         } else if (path->slots[0] >=
3125                    btrfs_header_nritems(path->nodes[0])) {
3126                 ret = btrfs_next_leaf(root, path);
3127                 if (ret < 0) {
3128                         goto out;
3129                 } else if (ret > 0) {
3130                         ret = -EUCLEAN;
3131                         goto out;
3132                 }
3133         }
3134         while (1) {
3135                 leaf = path->nodes[0];
3136                 slot = path->slots[0];
3137
3138                 btrfs_item_key_to_cpu(leaf, &key, slot);
3139                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3140                         ret = 0;
3141                         goto out;
3142                 }
3143
3144                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3145                         ret = -EOVERFLOW;
3146                         goto out;
3147                 }
3148
3149                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3150                 rootrefs->rootref[found].treeid = key.offset;
3151                 rootrefs->rootref[found].dirid =
3152                                   btrfs_root_ref_dirid(leaf, rref);
3153                 found++;
3154
3155                 ret = btrfs_next_item(root, path);
3156                 if (ret < 0) {
3157                         goto out;
3158                 } else if (ret > 0) {
3159                         ret = -EUCLEAN;
3160                         goto out;
3161                 }
3162         }
3163
3164 out:
3165         if (!ret || ret == -EOVERFLOW) {
3166                 rootrefs->num_items = found;
3167                 /* update min_treeid for next search */
3168                 if (found)
3169                         rootrefs->min_treeid =
3170                                 rootrefs->rootref[found - 1].treeid + 1;
3171                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3172                         ret = -EFAULT;
3173         }
3174
3175         kfree(rootrefs);
3176         btrfs_free_path(path);
3177
3178         return ret;
3179 }
3180
3181 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3182                                              void __user *arg,
3183                                              bool destroy_v2)
3184 {
3185         struct dentry *parent = file->f_path.dentry;
3186         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3187         struct dentry *dentry;
3188         struct inode *dir = d_inode(parent);
3189         struct inode *inode;
3190         struct btrfs_root *root = BTRFS_I(dir)->root;
3191         struct btrfs_root *dest = NULL;
3192         struct btrfs_ioctl_vol_args *vol_args = NULL;
3193         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3194         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3195         char *subvol_name, *subvol_name_ptr = NULL;
3196         int subvol_namelen;
3197         int err = 0;
3198         bool destroy_parent = false;
3199
3200         /* We don't support snapshots with extent tree v2 yet. */
3201         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3202                 btrfs_err(fs_info,
3203                           "extent tree v2 doesn't support snapshot deletion yet");
3204                 return -EOPNOTSUPP;
3205         }
3206
3207         if (destroy_v2) {
3208                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3209                 if (IS_ERR(vol_args2))
3210                         return PTR_ERR(vol_args2);
3211
3212                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3213                         err = -EOPNOTSUPP;
3214                         goto out;
3215                 }
3216
3217                 /*
3218                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3219                  * name, same as v1 currently does.
3220                  */
3221                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3222                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3223                         subvol_name = vol_args2->name;
3224
3225                         err = mnt_want_write_file(file);
3226                         if (err)
3227                                 goto out;
3228                 } else {
3229                         struct inode *old_dir;
3230
3231                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3232                                 err = -EINVAL;
3233                                 goto out;
3234                         }
3235
3236                         err = mnt_want_write_file(file);
3237                         if (err)
3238                                 goto out;
3239
3240                         dentry = btrfs_get_dentry(fs_info->sb,
3241                                         BTRFS_FIRST_FREE_OBJECTID,
3242                                         vol_args2->subvolid, 0, 0);
3243                         if (IS_ERR(dentry)) {
3244                                 err = PTR_ERR(dentry);
3245                                 goto out_drop_write;
3246                         }
3247
3248                         /*
3249                          * Change the default parent since the subvolume being
3250                          * deleted can be outside of the current mount point.
3251                          */
3252                         parent = btrfs_get_parent(dentry);
3253
3254                         /*
3255                          * At this point dentry->d_name can point to '/' if the
3256                          * subvolume we want to destroy is outsite of the
3257                          * current mount point, so we need to release the
3258                          * current dentry and execute the lookup to return a new
3259                          * one with ->d_name pointing to the
3260                          * <mount point>/subvol_name.
3261                          */
3262                         dput(dentry);
3263                         if (IS_ERR(parent)) {
3264                                 err = PTR_ERR(parent);
3265                                 goto out_drop_write;
3266                         }
3267                         old_dir = dir;
3268                         dir = d_inode(parent);
3269
3270                         /*
3271                          * If v2 was used with SPEC_BY_ID, a new parent was
3272                          * allocated since the subvolume can be outside of the
3273                          * current mount point. Later on we need to release this
3274                          * new parent dentry.
3275                          */
3276                         destroy_parent = true;
3277
3278                         /*
3279                          * On idmapped mounts, deletion via subvolid is
3280                          * restricted to subvolumes that are immediate
3281                          * ancestors of the inode referenced by the file
3282                          * descriptor in the ioctl. Otherwise the idmapping
3283                          * could potentially be abused to delete subvolumes
3284                          * anywhere in the filesystem the user wouldn't be able
3285                          * to delete without an idmapped mount.
3286                          */
3287                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3288                                 err = -EOPNOTSUPP;
3289                                 goto free_parent;
3290                         }
3291
3292                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3293                                                 fs_info, vol_args2->subvolid);
3294                         if (IS_ERR(subvol_name_ptr)) {
3295                                 err = PTR_ERR(subvol_name_ptr);
3296                                 goto free_parent;
3297                         }
3298                         /* subvol_name_ptr is already nul terminated */
3299                         subvol_name = (char *)kbasename(subvol_name_ptr);
3300                 }
3301         } else {
3302                 vol_args = memdup_user(arg, sizeof(*vol_args));
3303                 if (IS_ERR(vol_args))
3304                         return PTR_ERR(vol_args);
3305
3306                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3307                 subvol_name = vol_args->name;
3308
3309                 err = mnt_want_write_file(file);
3310                 if (err)
3311                         goto out;
3312         }
3313
3314         subvol_namelen = strlen(subvol_name);
3315
3316         if (strchr(subvol_name, '/') ||
3317             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3318                 err = -EINVAL;
3319                 goto free_subvol_name;
3320         }
3321
3322         if (!S_ISDIR(dir->i_mode)) {
3323                 err = -ENOTDIR;
3324                 goto free_subvol_name;
3325         }
3326
3327         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3328         if (err == -EINTR)
3329                 goto free_subvol_name;
3330         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3331         if (IS_ERR(dentry)) {
3332                 err = PTR_ERR(dentry);
3333                 goto out_unlock_dir;
3334         }
3335
3336         if (d_really_is_negative(dentry)) {
3337                 err = -ENOENT;
3338                 goto out_dput;
3339         }
3340
3341         inode = d_inode(dentry);
3342         dest = BTRFS_I(inode)->root;
3343         if (!capable(CAP_SYS_ADMIN)) {
3344                 /*
3345                  * Regular user.  Only allow this with a special mount
3346                  * option, when the user has write+exec access to the
3347                  * subvol root, and when rmdir(2) would have been
3348                  * allowed.
3349                  *
3350                  * Note that this is _not_ check that the subvol is
3351                  * empty or doesn't contain data that we wouldn't
3352                  * otherwise be able to delete.
3353                  *
3354                  * Users who want to delete empty subvols should try
3355                  * rmdir(2).
3356                  */
3357                 err = -EPERM;
3358                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3359                         goto out_dput;
3360
3361                 /*
3362                  * Do not allow deletion if the parent dir is the same
3363                  * as the dir to be deleted.  That means the ioctl
3364                  * must be called on the dentry referencing the root
3365                  * of the subvol, not a random directory contained
3366                  * within it.
3367                  */
3368                 err = -EINVAL;
3369                 if (root == dest)
3370                         goto out_dput;
3371
3372                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3373                 if (err)
3374                         goto out_dput;
3375         }
3376
3377         /* check if subvolume may be deleted by a user */
3378         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3379         if (err)
3380                 goto out_dput;
3381
3382         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3383                 err = -EINVAL;
3384                 goto out_dput;
3385         }
3386
3387         btrfs_inode_lock(inode, 0);
3388         err = btrfs_delete_subvolume(dir, dentry);
3389         btrfs_inode_unlock(inode, 0);
3390         if (!err)
3391                 d_delete_notify(dir, dentry);
3392
3393 out_dput:
3394         dput(dentry);
3395 out_unlock_dir:
3396         btrfs_inode_unlock(dir, 0);
3397 free_subvol_name:
3398         kfree(subvol_name_ptr);
3399 free_parent:
3400         if (destroy_parent)
3401                 dput(parent);
3402 out_drop_write:
3403         mnt_drop_write_file(file);
3404 out:
3405         kfree(vol_args2);
3406         kfree(vol_args);
3407         return err;
3408 }
3409
3410 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3411 {
3412         struct inode *inode = file_inode(file);
3413         struct btrfs_root *root = BTRFS_I(inode)->root;
3414         struct btrfs_ioctl_defrag_range_args range = {0};
3415         int ret;
3416
3417         ret = mnt_want_write_file(file);
3418         if (ret)
3419                 return ret;
3420
3421         if (btrfs_root_readonly(root)) {
3422                 ret = -EROFS;
3423                 goto out;
3424         }
3425
3426         switch (inode->i_mode & S_IFMT) {
3427         case S_IFDIR:
3428                 if (!capable(CAP_SYS_ADMIN)) {
3429                         ret = -EPERM;
3430                         goto out;
3431                 }
3432                 ret = btrfs_defrag_root(root);
3433                 break;
3434         case S_IFREG:
3435                 /*
3436                  * Note that this does not check the file descriptor for write
3437                  * access. This prevents defragmenting executables that are
3438                  * running and allows defrag on files open in read-only mode.
3439                  */
3440                 if (!capable(CAP_SYS_ADMIN) &&
3441                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3442                         ret = -EPERM;
3443                         goto out;
3444                 }
3445
3446                 if (argp) {
3447                         if (copy_from_user(&range, argp, sizeof(range))) {
3448                                 ret = -EFAULT;
3449                                 goto out;
3450                         }
3451                         /* compression requires us to start the IO */
3452                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3453                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3454                                 range.extent_thresh = (u32)-1;
3455                         }
3456                 } else {
3457                         /* the rest are all set to zero by kzalloc */
3458                         range.len = (u64)-1;
3459                 }
3460                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3461                                         &range, BTRFS_OLDEST_GENERATION, 0);
3462                 if (ret > 0)
3463                         ret = 0;
3464                 break;
3465         default:
3466                 ret = -EINVAL;
3467         }
3468 out:
3469         mnt_drop_write_file(file);
3470         return ret;
3471 }
3472
3473 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3474 {
3475         struct btrfs_ioctl_vol_args *vol_args;
3476         bool restore_op = false;
3477         int ret;
3478
3479         if (!capable(CAP_SYS_ADMIN))
3480                 return -EPERM;
3481
3482         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3483                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3484                 return -EINVAL;
3485         }
3486
3487         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3488                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3489                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3490
3491                 /*
3492                  * We can do the device add because we have a paused balanced,
3493                  * change the exclusive op type and remember we should bring
3494                  * back the paused balance
3495                  */
3496                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3497                 btrfs_exclop_start_unlock(fs_info);
3498                 restore_op = true;
3499         }
3500
3501         vol_args = memdup_user(arg, sizeof(*vol_args));
3502         if (IS_ERR(vol_args)) {
3503                 ret = PTR_ERR(vol_args);
3504                 goto out;
3505         }
3506
3507         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3508         ret = btrfs_init_new_device(fs_info, vol_args->name);
3509
3510         if (!ret)
3511                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3512
3513         kfree(vol_args);
3514 out:
3515         if (restore_op)
3516                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3517         else
3518                 btrfs_exclop_finish(fs_info);
3519         return ret;
3520 }
3521
3522 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3523 {
3524         BTRFS_DEV_LOOKUP_ARGS(args);
3525         struct inode *inode = file_inode(file);
3526         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3527         struct btrfs_ioctl_vol_args_v2 *vol_args;
3528         struct block_device *bdev = NULL;
3529         fmode_t mode;
3530         int ret;
3531         bool cancel = false;
3532
3533         if (!capable(CAP_SYS_ADMIN))
3534                 return -EPERM;
3535
3536         vol_args = memdup_user(arg, sizeof(*vol_args));
3537         if (IS_ERR(vol_args))
3538                 return PTR_ERR(vol_args);
3539
3540         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3541                 ret = -EOPNOTSUPP;
3542                 goto out;
3543         }
3544
3545         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3546         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3547                 args.devid = vol_args->devid;
3548         } else if (!strcmp("cancel", vol_args->name)) {
3549                 cancel = true;
3550         } else {
3551                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3552                 if (ret)
3553                         goto out;
3554         }
3555
3556         ret = mnt_want_write_file(file);
3557         if (ret)
3558                 goto out;
3559
3560         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3561                                            cancel);
3562         if (ret)
3563                 goto err_drop;
3564
3565         /* Exclusive operation is now claimed */
3566         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3567
3568         btrfs_exclop_finish(fs_info);
3569
3570         if (!ret) {
3571                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3572                         btrfs_info(fs_info, "device deleted: id %llu",
3573                                         vol_args->devid);
3574                 else
3575                         btrfs_info(fs_info, "device deleted: %s",
3576                                         vol_args->name);
3577         }
3578 err_drop:
3579         mnt_drop_write_file(file);
3580         if (bdev)
3581                 blkdev_put(bdev, mode);
3582 out:
3583         btrfs_put_dev_args_from_path(&args);
3584         kfree(vol_args);
3585         return ret;
3586 }
3587
3588 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3589 {
3590         BTRFS_DEV_LOOKUP_ARGS(args);
3591         struct inode *inode = file_inode(file);
3592         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3593         struct btrfs_ioctl_vol_args *vol_args;
3594         struct block_device *bdev = NULL;
3595         fmode_t mode;
3596         int ret;
3597         bool cancel = false;
3598
3599         if (!capable(CAP_SYS_ADMIN))
3600                 return -EPERM;
3601
3602         vol_args = memdup_user(arg, sizeof(*vol_args));
3603         if (IS_ERR(vol_args))
3604                 return PTR_ERR(vol_args);
3605
3606         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3607         if (!strcmp("cancel", vol_args->name)) {
3608                 cancel = true;
3609         } else {
3610                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3611                 if (ret)
3612                         goto out;
3613         }
3614
3615         ret = mnt_want_write_file(file);
3616         if (ret)
3617                 goto out;
3618
3619         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3620                                            cancel);
3621         if (ret == 0) {
3622                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3623                 if (!ret)
3624                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3625                 btrfs_exclop_finish(fs_info);
3626         }
3627
3628         mnt_drop_write_file(file);
3629         if (bdev)
3630                 blkdev_put(bdev, mode);
3631 out:
3632         btrfs_put_dev_args_from_path(&args);
3633         kfree(vol_args);
3634         return ret;
3635 }
3636
3637 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3638                                 void __user *arg)
3639 {
3640         struct btrfs_ioctl_fs_info_args *fi_args;
3641         struct btrfs_device *device;
3642         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3643         u64 flags_in;
3644         int ret = 0;
3645
3646         fi_args = memdup_user(arg, sizeof(*fi_args));
3647         if (IS_ERR(fi_args))
3648                 return PTR_ERR(fi_args);
3649
3650         flags_in = fi_args->flags;
3651         memset(fi_args, 0, sizeof(*fi_args));
3652
3653         rcu_read_lock();
3654         fi_args->num_devices = fs_devices->num_devices;
3655
3656         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3657                 if (device->devid > fi_args->max_id)
3658                         fi_args->max_id = device->devid;
3659         }
3660         rcu_read_unlock();
3661
3662         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3663         fi_args->nodesize = fs_info->nodesize;
3664         fi_args->sectorsize = fs_info->sectorsize;
3665         fi_args->clone_alignment = fs_info->sectorsize;
3666
3667         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3668                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3669                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3670                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3671         }
3672
3673         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3674                 fi_args->generation = fs_info->generation;
3675                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3676         }
3677
3678         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3679                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3680                        sizeof(fi_args->metadata_uuid));
3681                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3682         }
3683
3684         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3685                 ret = -EFAULT;
3686
3687         kfree(fi_args);
3688         return ret;
3689 }
3690
3691 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3692                                  void __user *arg)
3693 {
3694         BTRFS_DEV_LOOKUP_ARGS(args);
3695         struct btrfs_ioctl_dev_info_args *di_args;
3696         struct btrfs_device *dev;
3697         int ret = 0;
3698
3699         di_args = memdup_user(arg, sizeof(*di_args));
3700         if (IS_ERR(di_args))
3701                 return PTR_ERR(di_args);
3702
3703         args.devid = di_args->devid;
3704         if (!btrfs_is_empty_uuid(di_args->uuid))
3705                 args.uuid = di_args->uuid;
3706
3707         rcu_read_lock();
3708         dev = btrfs_find_device(fs_info->fs_devices, &args);
3709         if (!dev) {
3710                 ret = -ENODEV;
3711                 goto out;
3712         }
3713
3714         di_args->devid = dev->devid;
3715         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3716         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3717         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3718         if (dev->name) {
3719                 strncpy(di_args->path, rcu_str_deref(dev->name),
3720                                 sizeof(di_args->path) - 1);
3721                 di_args->path[sizeof(di_args->path) - 1] = 0;
3722         } else {
3723                 di_args->path[0] = '\0';
3724         }
3725
3726 out:
3727         rcu_read_unlock();
3728         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3729                 ret = -EFAULT;
3730
3731         kfree(di_args);
3732         return ret;
3733 }
3734
3735 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3736 {
3737         struct inode *inode = file_inode(file);
3738         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3739         struct btrfs_root *root = BTRFS_I(inode)->root;
3740         struct btrfs_root *new_root;
3741         struct btrfs_dir_item *di;
3742         struct btrfs_trans_handle *trans;
3743         struct btrfs_path *path = NULL;
3744         struct btrfs_disk_key disk_key;
3745         u64 objectid = 0;
3746         u64 dir_id;
3747         int ret;
3748
3749         if (!capable(CAP_SYS_ADMIN))
3750                 return -EPERM;
3751
3752         ret = mnt_want_write_file(file);
3753         if (ret)
3754                 return ret;
3755
3756         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3757                 ret = -EFAULT;
3758                 goto out;
3759         }
3760
3761         if (!objectid)
3762                 objectid = BTRFS_FS_TREE_OBJECTID;
3763
3764         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3765         if (IS_ERR(new_root)) {
3766                 ret = PTR_ERR(new_root);
3767                 goto out;
3768         }
3769         if (!is_fstree(new_root->root_key.objectid)) {
3770                 ret = -ENOENT;
3771                 goto out_free;
3772         }
3773
3774         path = btrfs_alloc_path();
3775         if (!path) {
3776                 ret = -ENOMEM;
3777                 goto out_free;
3778         }
3779
3780         trans = btrfs_start_transaction(root, 1);
3781         if (IS_ERR(trans)) {
3782                 ret = PTR_ERR(trans);
3783                 goto out_free;
3784         }
3785
3786         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3787         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3788                                    dir_id, "default", 7, 1);
3789         if (IS_ERR_OR_NULL(di)) {
3790                 btrfs_release_path(path);
3791                 btrfs_end_transaction(trans);
3792                 btrfs_err(fs_info,
3793                           "Umm, you don't have the default diritem, this isn't going to work");
3794                 ret = -ENOENT;
3795                 goto out_free;
3796         }
3797
3798         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3799         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3800         btrfs_mark_buffer_dirty(path->nodes[0]);
3801         btrfs_release_path(path);
3802
3803         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3804         btrfs_end_transaction(trans);
3805 out_free:
3806         btrfs_put_root(new_root);
3807         btrfs_free_path(path);
3808 out:
3809         mnt_drop_write_file(file);
3810         return ret;
3811 }
3812
3813 static void get_block_group_info(struct list_head *groups_list,
3814                                  struct btrfs_ioctl_space_info *space)
3815 {
3816         struct btrfs_block_group *block_group;
3817
3818         space->total_bytes = 0;
3819         space->used_bytes = 0;
3820         space->flags = 0;
3821         list_for_each_entry(block_group, groups_list, list) {
3822                 space->flags = block_group->flags;
3823                 space->total_bytes += block_group->length;
3824                 space->used_bytes += block_group->used;
3825         }
3826 }
3827
3828 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3829                                    void __user *arg)
3830 {
3831         struct btrfs_ioctl_space_args space_args;
3832         struct btrfs_ioctl_space_info space;
3833         struct btrfs_ioctl_space_info *dest;
3834         struct btrfs_ioctl_space_info *dest_orig;
3835         struct btrfs_ioctl_space_info __user *user_dest;
3836         struct btrfs_space_info *info;
3837         static const u64 types[] = {
3838                 BTRFS_BLOCK_GROUP_DATA,
3839                 BTRFS_BLOCK_GROUP_SYSTEM,
3840                 BTRFS_BLOCK_GROUP_METADATA,
3841                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3842         };
3843         int num_types = 4;
3844         int alloc_size;
3845         int ret = 0;
3846         u64 slot_count = 0;
3847         int i, c;
3848
3849         if (copy_from_user(&space_args,
3850                            (struct btrfs_ioctl_space_args __user *)arg,
3851                            sizeof(space_args)))
3852                 return -EFAULT;
3853
3854         for (i = 0; i < num_types; i++) {
3855                 struct btrfs_space_info *tmp;
3856
3857                 info = NULL;
3858                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3859                         if (tmp->flags == types[i]) {
3860                                 info = tmp;
3861                                 break;
3862                         }
3863                 }
3864
3865                 if (!info)
3866                         continue;
3867
3868                 down_read(&info->groups_sem);
3869                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3870                         if (!list_empty(&info->block_groups[c]))
3871                                 slot_count++;
3872                 }
3873                 up_read(&info->groups_sem);
3874         }
3875
3876         /*
3877          * Global block reserve, exported as a space_info
3878          */
3879         slot_count++;
3880
3881         /* space_slots == 0 means they are asking for a count */
3882         if (space_args.space_slots == 0) {
3883                 space_args.total_spaces = slot_count;
3884                 goto out;
3885         }
3886
3887         slot_count = min_t(u64, space_args.space_slots, slot_count);
3888
3889         alloc_size = sizeof(*dest) * slot_count;
3890
3891         /* we generally have at most 6 or so space infos, one for each raid
3892          * level.  So, a whole page should be more than enough for everyone
3893          */
3894         if (alloc_size > PAGE_SIZE)
3895                 return -ENOMEM;
3896
3897         space_args.total_spaces = 0;
3898         dest = kmalloc(alloc_size, GFP_KERNEL);
3899         if (!dest)
3900                 return -ENOMEM;
3901         dest_orig = dest;
3902
3903         /* now we have a buffer to copy into */
3904         for (i = 0; i < num_types; i++) {
3905                 struct btrfs_space_info *tmp;
3906
3907                 if (!slot_count)
3908                         break;
3909
3910                 info = NULL;
3911                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3912                         if (tmp->flags == types[i]) {
3913                                 info = tmp;
3914                                 break;
3915                         }
3916                 }
3917
3918                 if (!info)
3919                         continue;
3920                 down_read(&info->groups_sem);
3921                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3922                         if (!list_empty(&info->block_groups[c])) {
3923                                 get_block_group_info(&info->block_groups[c],
3924                                                      &space);
3925                                 memcpy(dest, &space, sizeof(space));
3926                                 dest++;
3927                                 space_args.total_spaces++;
3928                                 slot_count--;
3929                         }
3930                         if (!slot_count)
3931                                 break;
3932                 }
3933                 up_read(&info->groups_sem);
3934         }
3935
3936         /*
3937          * Add global block reserve
3938          */
3939         if (slot_count) {
3940                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3941
3942                 spin_lock(&block_rsv->lock);
3943                 space.total_bytes = block_rsv->size;
3944                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3945                 spin_unlock(&block_rsv->lock);
3946                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3947                 memcpy(dest, &space, sizeof(space));
3948                 space_args.total_spaces++;
3949         }
3950
3951         user_dest = (struct btrfs_ioctl_space_info __user *)
3952                 (arg + sizeof(struct btrfs_ioctl_space_args));
3953
3954         if (copy_to_user(user_dest, dest_orig, alloc_size))
3955                 ret = -EFAULT;
3956
3957         kfree(dest_orig);
3958 out:
3959         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3960                 ret = -EFAULT;
3961
3962         return ret;
3963 }
3964
3965 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3966                                             void __user *argp)
3967 {
3968         struct btrfs_trans_handle *trans;
3969         u64 transid;
3970
3971         trans = btrfs_attach_transaction_barrier(root);
3972         if (IS_ERR(trans)) {
3973                 if (PTR_ERR(trans) != -ENOENT)
3974                         return PTR_ERR(trans);
3975
3976                 /* No running transaction, don't bother */
3977                 transid = root->fs_info->last_trans_committed;
3978                 goto out;
3979         }
3980         transid = trans->transid;
3981         btrfs_commit_transaction_async(trans);
3982 out:
3983         if (argp)
3984                 if (copy_to_user(argp, &transid, sizeof(transid)))
3985                         return -EFAULT;
3986         return 0;
3987 }
3988
3989 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3990                                            void __user *argp)
3991 {
3992         u64 transid;
3993
3994         if (argp) {
3995                 if (copy_from_user(&transid, argp, sizeof(transid)))
3996                         return -EFAULT;
3997         } else {
3998                 transid = 0;  /* current trans */
3999         }
4000         return btrfs_wait_for_commit(fs_info, transid);
4001 }
4002
4003 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4004 {
4005         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4006         struct btrfs_ioctl_scrub_args *sa;
4007         int ret;
4008
4009         if (!capable(CAP_SYS_ADMIN))
4010                 return -EPERM;
4011
4012         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4013                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4014                 return -EINVAL;
4015         }
4016
4017         sa = memdup_user(arg, sizeof(*sa));
4018         if (IS_ERR(sa))
4019                 return PTR_ERR(sa);
4020
4021         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4022                 ret = mnt_want_write_file(file);
4023                 if (ret)
4024                         goto out;
4025         }
4026
4027         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4028                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4029                               0);
4030
4031         /*
4032          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4033          * error. This is important as it allows user space to know how much
4034          * progress scrub has done. For example, if scrub is canceled we get
4035          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4036          * space. Later user space can inspect the progress from the structure
4037          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4038          * previously (btrfs-progs does this).
4039          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4040          * then return -EFAULT to signal the structure was not copied or it may
4041          * be corrupt and unreliable due to a partial copy.
4042          */
4043         if (copy_to_user(arg, sa, sizeof(*sa)))
4044                 ret = -EFAULT;
4045
4046         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4047                 mnt_drop_write_file(file);
4048 out:
4049         kfree(sa);
4050         return ret;
4051 }
4052
4053 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4054 {
4055         if (!capable(CAP_SYS_ADMIN))
4056                 return -EPERM;
4057
4058         return btrfs_scrub_cancel(fs_info);
4059 }
4060
4061 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4062                                        void __user *arg)
4063 {
4064         struct btrfs_ioctl_scrub_args *sa;
4065         int ret;
4066
4067         if (!capable(CAP_SYS_ADMIN))
4068                 return -EPERM;
4069
4070         sa = memdup_user(arg, sizeof(*sa));
4071         if (IS_ERR(sa))
4072                 return PTR_ERR(sa);
4073
4074         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4075
4076         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4077                 ret = -EFAULT;
4078
4079         kfree(sa);
4080         return ret;
4081 }
4082
4083 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4084                                       void __user *arg)
4085 {
4086         struct btrfs_ioctl_get_dev_stats *sa;
4087         int ret;
4088
4089         sa = memdup_user(arg, sizeof(*sa));
4090         if (IS_ERR(sa))
4091                 return PTR_ERR(sa);
4092
4093         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4094                 kfree(sa);
4095                 return -EPERM;
4096         }
4097
4098         ret = btrfs_get_dev_stats(fs_info, sa);
4099
4100         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4101                 ret = -EFAULT;
4102
4103         kfree(sa);
4104         return ret;
4105 }
4106
4107 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4108                                     void __user *arg)
4109 {
4110         struct btrfs_ioctl_dev_replace_args *p;
4111         int ret;
4112
4113         if (!capable(CAP_SYS_ADMIN))
4114                 return -EPERM;
4115
4116         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4117                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4118                 return -EINVAL;
4119         }
4120
4121         p = memdup_user(arg, sizeof(*p));
4122         if (IS_ERR(p))
4123                 return PTR_ERR(p);
4124
4125         switch (p->cmd) {
4126         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4127                 if (sb_rdonly(fs_info->sb)) {
4128                         ret = -EROFS;
4129                         goto out;
4130                 }
4131                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4132                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4133                 } else {
4134                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4135                         btrfs_exclop_finish(fs_info);
4136                 }
4137                 break;
4138         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4139                 btrfs_dev_replace_status(fs_info, p);
4140                 ret = 0;
4141                 break;
4142         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4143                 p->result = btrfs_dev_replace_cancel(fs_info);
4144                 ret = 0;
4145                 break;
4146         default:
4147                 ret = -EINVAL;
4148                 break;
4149         }
4150
4151         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4152                 ret = -EFAULT;
4153 out:
4154         kfree(p);
4155         return ret;
4156 }
4157
4158 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4159 {
4160         int ret = 0;
4161         int i;
4162         u64 rel_ptr;
4163         int size;
4164         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4165         struct inode_fs_paths *ipath = NULL;
4166         struct btrfs_path *path;
4167
4168         if (!capable(CAP_DAC_READ_SEARCH))
4169                 return -EPERM;
4170
4171         path = btrfs_alloc_path();
4172         if (!path) {
4173                 ret = -ENOMEM;
4174                 goto out;
4175         }
4176
4177         ipa = memdup_user(arg, sizeof(*ipa));
4178         if (IS_ERR(ipa)) {
4179                 ret = PTR_ERR(ipa);
4180                 ipa = NULL;
4181                 goto out;
4182         }
4183
4184         size = min_t(u32, ipa->size, 4096);
4185         ipath = init_ipath(size, root, path);
4186         if (IS_ERR(ipath)) {
4187                 ret = PTR_ERR(ipath);
4188                 ipath = NULL;
4189                 goto out;
4190         }
4191
4192         ret = paths_from_inode(ipa->inum, ipath);
4193         if (ret < 0)
4194                 goto out;
4195
4196         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4197                 rel_ptr = ipath->fspath->val[i] -
4198                           (u64)(unsigned long)ipath->fspath->val;
4199                 ipath->fspath->val[i] = rel_ptr;
4200         }
4201
4202         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4203                            ipath->fspath, size);
4204         if (ret) {
4205                 ret = -EFAULT;
4206                 goto out;
4207         }
4208
4209 out:
4210         btrfs_free_path(path);
4211         free_ipath(ipath);
4212         kfree(ipa);
4213
4214         return ret;
4215 }
4216
4217 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4218 {
4219         struct btrfs_data_container *inodes = ctx;
4220         const size_t c = 3 * sizeof(u64);
4221
4222         if (inodes->bytes_left >= c) {
4223                 inodes->bytes_left -= c;
4224                 inodes->val[inodes->elem_cnt] = inum;
4225                 inodes->val[inodes->elem_cnt + 1] = offset;
4226                 inodes->val[inodes->elem_cnt + 2] = root;
4227                 inodes->elem_cnt += 3;
4228         } else {
4229                 inodes->bytes_missing += c - inodes->bytes_left;
4230                 inodes->bytes_left = 0;
4231                 inodes->elem_missed += 3;
4232         }
4233
4234         return 0;
4235 }
4236
4237 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4238                                         void __user *arg, int version)
4239 {
4240         int ret = 0;
4241         int size;
4242         struct btrfs_ioctl_logical_ino_args *loi;
4243         struct btrfs_data_container *inodes = NULL;
4244         struct btrfs_path *path = NULL;
4245         bool ignore_offset;
4246
4247         if (!capable(CAP_SYS_ADMIN))
4248                 return -EPERM;
4249
4250         loi = memdup_user(arg, sizeof(*loi));
4251         if (IS_ERR(loi))
4252                 return PTR_ERR(loi);
4253
4254         if (version == 1) {
4255                 ignore_offset = false;
4256                 size = min_t(u32, loi->size, SZ_64K);
4257         } else {
4258                 /* All reserved bits must be 0 for now */
4259                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4260                         ret = -EINVAL;
4261                         goto out_loi;
4262                 }
4263                 /* Only accept flags we have defined so far */
4264                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4265                         ret = -EINVAL;
4266                         goto out_loi;
4267                 }
4268                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4269                 size = min_t(u32, loi->size, SZ_16M);
4270         }
4271
4272         path = btrfs_alloc_path();
4273         if (!path) {
4274                 ret = -ENOMEM;
4275                 goto out;
4276         }
4277
4278         inodes = init_data_container(size);
4279         if (IS_ERR(inodes)) {
4280                 ret = PTR_ERR(inodes);
4281                 inodes = NULL;
4282                 goto out;
4283         }
4284
4285         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4286                                           build_ino_list, inodes, ignore_offset);
4287         if (ret == -EINVAL)
4288                 ret = -ENOENT;
4289         if (ret < 0)
4290                 goto out;
4291
4292         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4293                            size);
4294         if (ret)
4295                 ret = -EFAULT;
4296
4297 out:
4298         btrfs_free_path(path);
4299         kvfree(inodes);
4300 out_loi:
4301         kfree(loi);
4302
4303         return ret;
4304 }
4305
4306 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4307                                struct btrfs_ioctl_balance_args *bargs)
4308 {
4309         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4310
4311         bargs->flags = bctl->flags;
4312
4313         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4314                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4315         if (atomic_read(&fs_info->balance_pause_req))
4316                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4317         if (atomic_read(&fs_info->balance_cancel_req))
4318                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4319
4320         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4321         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4322         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4323
4324         spin_lock(&fs_info->balance_lock);
4325         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4326         spin_unlock(&fs_info->balance_lock);
4327 }
4328
4329 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4330 {
4331         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4332         struct btrfs_fs_info *fs_info = root->fs_info;
4333         struct btrfs_ioctl_balance_args *bargs;
4334         struct btrfs_balance_control *bctl;
4335         bool need_unlock; /* for mut. excl. ops lock */
4336         int ret;
4337
4338         if (!arg)
4339                 btrfs_warn(fs_info,
4340         "IOC_BALANCE ioctl (v1) is deprecated and will be removed in kernel 5.18");
4341
4342         if (!capable(CAP_SYS_ADMIN))
4343                 return -EPERM;
4344
4345         ret = mnt_want_write_file(file);
4346         if (ret)
4347                 return ret;
4348
4349 again:
4350         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4351                 mutex_lock(&fs_info->balance_mutex);
4352                 need_unlock = true;
4353                 goto locked;
4354         }
4355
4356         /*
4357          * mut. excl. ops lock is locked.  Three possibilities:
4358          *   (1) some other op is running
4359          *   (2) balance is running
4360          *   (3) balance is paused -- special case (think resume)
4361          */
4362         mutex_lock(&fs_info->balance_mutex);
4363         if (fs_info->balance_ctl) {
4364                 /* this is either (2) or (3) */
4365                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4366                         mutex_unlock(&fs_info->balance_mutex);
4367                         /*
4368                          * Lock released to allow other waiters to continue,
4369                          * we'll reexamine the status again.
4370                          */
4371                         mutex_lock(&fs_info->balance_mutex);
4372
4373                         if (fs_info->balance_ctl &&
4374                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4375                                 /* this is (3) */
4376                                 need_unlock = false;
4377                                 goto locked;
4378                         }
4379
4380                         mutex_unlock(&fs_info->balance_mutex);
4381                         goto again;
4382                 } else {
4383                         /* this is (2) */
4384                         mutex_unlock(&fs_info->balance_mutex);
4385                         ret = -EINPROGRESS;
4386                         goto out;
4387                 }
4388         } else {
4389                 /* this is (1) */
4390                 mutex_unlock(&fs_info->balance_mutex);
4391                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4392                 goto out;
4393         }
4394
4395 locked:
4396
4397         if (arg) {
4398                 bargs = memdup_user(arg, sizeof(*bargs));
4399                 if (IS_ERR(bargs)) {
4400                         ret = PTR_ERR(bargs);
4401                         goto out_unlock;
4402                 }
4403
4404                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4405                         if (!fs_info->balance_ctl) {
4406                                 ret = -ENOTCONN;
4407                                 goto out_bargs;
4408                         }
4409
4410                         bctl = fs_info->balance_ctl;
4411                         spin_lock(&fs_info->balance_lock);
4412                         bctl->flags |= BTRFS_BALANCE_RESUME;
4413                         spin_unlock(&fs_info->balance_lock);
4414                         btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4415
4416                         goto do_balance;
4417                 }
4418         } else {
4419                 bargs = NULL;
4420         }
4421
4422         if (fs_info->balance_ctl) {
4423                 ret = -EINPROGRESS;
4424                 goto out_bargs;
4425         }
4426
4427         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4428         if (!bctl) {
4429                 ret = -ENOMEM;
4430                 goto out_bargs;
4431         }
4432
4433         if (arg) {
4434                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4435                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4436                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4437
4438                 bctl->flags = bargs->flags;
4439         } else {
4440                 /* balance everything - no filters */
4441                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4442         }
4443
4444         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4445                 ret = -EINVAL;
4446                 goto out_bctl;
4447         }
4448
4449 do_balance:
4450         /*
4451          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4452          * bctl is freed in reset_balance_state, or, if restriper was paused
4453          * all the way until unmount, in free_fs_info.  The flag should be
4454          * cleared after reset_balance_state.
4455          */
4456         need_unlock = false;
4457
4458         ret = btrfs_balance(fs_info, bctl, bargs);
4459         bctl = NULL;
4460
4461         if ((ret == 0 || ret == -ECANCELED) && arg) {
4462                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4463                         ret = -EFAULT;
4464         }
4465
4466 out_bctl:
4467         kfree(bctl);
4468 out_bargs:
4469         kfree(bargs);
4470 out_unlock:
4471         mutex_unlock(&fs_info->balance_mutex);
4472         if (need_unlock)
4473                 btrfs_exclop_finish(fs_info);
4474 out:
4475         mnt_drop_write_file(file);
4476         return ret;
4477 }
4478
4479 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4480 {
4481         if (!capable(CAP_SYS_ADMIN))
4482                 return -EPERM;
4483
4484         switch (cmd) {
4485         case BTRFS_BALANCE_CTL_PAUSE:
4486                 return btrfs_pause_balance(fs_info);
4487         case BTRFS_BALANCE_CTL_CANCEL:
4488                 return btrfs_cancel_balance(fs_info);
4489         }
4490
4491         return -EINVAL;
4492 }
4493
4494 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4495                                          void __user *arg)
4496 {
4497         struct btrfs_ioctl_balance_args *bargs;
4498         int ret = 0;
4499
4500         if (!capable(CAP_SYS_ADMIN))
4501                 return -EPERM;
4502
4503         mutex_lock(&fs_info->balance_mutex);
4504         if (!fs_info->balance_ctl) {
4505                 ret = -ENOTCONN;
4506                 goto out;
4507         }
4508
4509         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4510         if (!bargs) {
4511                 ret = -ENOMEM;
4512                 goto out;
4513         }
4514
4515         btrfs_update_ioctl_balance_args(fs_info, bargs);
4516
4517         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4518                 ret = -EFAULT;
4519
4520         kfree(bargs);
4521 out:
4522         mutex_unlock(&fs_info->balance_mutex);
4523         return ret;
4524 }
4525
4526 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4527 {
4528         struct inode *inode = file_inode(file);
4529         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4530         struct btrfs_ioctl_quota_ctl_args *sa;
4531         int ret;
4532
4533         if (!capable(CAP_SYS_ADMIN))
4534                 return -EPERM;
4535
4536         ret = mnt_want_write_file(file);
4537         if (ret)
4538                 return ret;
4539
4540         sa = memdup_user(arg, sizeof(*sa));
4541         if (IS_ERR(sa)) {
4542                 ret = PTR_ERR(sa);
4543                 goto drop_write;
4544         }
4545
4546         down_write(&fs_info->subvol_sem);
4547
4548         switch (sa->cmd) {
4549         case BTRFS_QUOTA_CTL_ENABLE:
4550                 ret = btrfs_quota_enable(fs_info);
4551                 break;
4552         case BTRFS_QUOTA_CTL_DISABLE:
4553                 ret = btrfs_quota_disable(fs_info);
4554                 break;
4555         default:
4556                 ret = -EINVAL;
4557                 break;
4558         }
4559
4560         kfree(sa);
4561         up_write(&fs_info->subvol_sem);
4562 drop_write:
4563         mnt_drop_write_file(file);
4564         return ret;
4565 }
4566
4567 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4568 {
4569         struct inode *inode = file_inode(file);
4570         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4571         struct btrfs_root *root = BTRFS_I(inode)->root;
4572         struct btrfs_ioctl_qgroup_assign_args *sa;
4573         struct btrfs_trans_handle *trans;
4574         int ret;
4575         int err;
4576
4577         if (!capable(CAP_SYS_ADMIN))
4578                 return -EPERM;
4579
4580         ret = mnt_want_write_file(file);
4581         if (ret)
4582                 return ret;
4583
4584         sa = memdup_user(arg, sizeof(*sa));
4585         if (IS_ERR(sa)) {
4586                 ret = PTR_ERR(sa);
4587                 goto drop_write;
4588         }
4589
4590         trans = btrfs_join_transaction(root);
4591         if (IS_ERR(trans)) {
4592                 ret = PTR_ERR(trans);
4593                 goto out;
4594         }
4595
4596         if (sa->assign) {
4597                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4598         } else {
4599                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4600         }
4601
4602         /* update qgroup status and info */
4603         err = btrfs_run_qgroups(trans);
4604         if (err < 0)
4605                 btrfs_handle_fs_error(fs_info, err,
4606                                       "failed to update qgroup status and info");
4607         err = btrfs_end_transaction(trans);
4608         if (err && !ret)
4609                 ret = err;
4610
4611 out:
4612         kfree(sa);
4613 drop_write:
4614         mnt_drop_write_file(file);
4615         return ret;
4616 }
4617
4618 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4619 {
4620         struct inode *inode = file_inode(file);
4621         struct btrfs_root *root = BTRFS_I(inode)->root;
4622         struct btrfs_ioctl_qgroup_create_args *sa;
4623         struct btrfs_trans_handle *trans;
4624         int ret;
4625         int err;
4626
4627         if (!capable(CAP_SYS_ADMIN))
4628                 return -EPERM;
4629
4630         ret = mnt_want_write_file(file);
4631         if (ret)
4632                 return ret;
4633
4634         sa = memdup_user(arg, sizeof(*sa));
4635         if (IS_ERR(sa)) {
4636                 ret = PTR_ERR(sa);
4637                 goto drop_write;
4638         }
4639
4640         if (!sa->qgroupid) {
4641                 ret = -EINVAL;
4642                 goto out;
4643         }
4644
4645         trans = btrfs_join_transaction(root);
4646         if (IS_ERR(trans)) {
4647                 ret = PTR_ERR(trans);
4648                 goto out;
4649         }
4650
4651         if (sa->create) {
4652                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4653         } else {
4654                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4655         }
4656
4657         err = btrfs_end_transaction(trans);
4658         if (err && !ret)
4659                 ret = err;
4660
4661 out:
4662         kfree(sa);
4663 drop_write:
4664         mnt_drop_write_file(file);
4665         return ret;
4666 }
4667
4668 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4669 {
4670         struct inode *inode = file_inode(file);
4671         struct btrfs_root *root = BTRFS_I(inode)->root;
4672         struct btrfs_ioctl_qgroup_limit_args *sa;
4673         struct btrfs_trans_handle *trans;
4674         int ret;
4675         int err;
4676         u64 qgroupid;
4677
4678         if (!capable(CAP_SYS_ADMIN))
4679                 return -EPERM;
4680
4681         ret = mnt_want_write_file(file);
4682         if (ret)
4683                 return ret;
4684
4685         sa = memdup_user(arg, sizeof(*sa));
4686         if (IS_ERR(sa)) {
4687                 ret = PTR_ERR(sa);
4688                 goto drop_write;
4689         }
4690
4691         trans = btrfs_join_transaction(root);
4692         if (IS_ERR(trans)) {
4693                 ret = PTR_ERR(trans);
4694                 goto out;
4695         }
4696
4697         qgroupid = sa->qgroupid;
4698         if (!qgroupid) {
4699                 /* take the current subvol as qgroup */
4700                 qgroupid = root->root_key.objectid;
4701         }
4702
4703         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4704
4705         err = btrfs_end_transaction(trans);
4706         if (err && !ret)
4707                 ret = err;
4708
4709 out:
4710         kfree(sa);
4711 drop_write:
4712         mnt_drop_write_file(file);
4713         return ret;
4714 }
4715
4716 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4717 {
4718         struct inode *inode = file_inode(file);
4719         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4720         struct btrfs_ioctl_quota_rescan_args *qsa;
4721         int ret;
4722
4723         if (!capable(CAP_SYS_ADMIN))
4724                 return -EPERM;
4725
4726         ret = mnt_want_write_file(file);
4727         if (ret)
4728                 return ret;
4729
4730         qsa = memdup_user(arg, sizeof(*qsa));
4731         if (IS_ERR(qsa)) {
4732                 ret = PTR_ERR(qsa);
4733                 goto drop_write;
4734         }
4735
4736         if (qsa->flags) {
4737                 ret = -EINVAL;
4738                 goto out;
4739         }
4740
4741         ret = btrfs_qgroup_rescan(fs_info);
4742
4743 out:
4744         kfree(qsa);
4745 drop_write:
4746         mnt_drop_write_file(file);
4747         return ret;
4748 }
4749
4750 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4751                                                 void __user *arg)
4752 {
4753         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4754
4755         if (!capable(CAP_SYS_ADMIN))
4756                 return -EPERM;
4757
4758         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4759                 qsa.flags = 1;
4760                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4761         }
4762
4763         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4764                 return -EFAULT;
4765
4766         return 0;
4767 }
4768
4769 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4770                                                 void __user *arg)
4771 {
4772         if (!capable(CAP_SYS_ADMIN))
4773                 return -EPERM;
4774
4775         return btrfs_qgroup_wait_for_completion(fs_info, true);
4776 }
4777
4778 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4779                                             struct user_namespace *mnt_userns,
4780                                             struct btrfs_ioctl_received_subvol_args *sa)
4781 {
4782         struct inode *inode = file_inode(file);
4783         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4784         struct btrfs_root *root = BTRFS_I(inode)->root;
4785         struct btrfs_root_item *root_item = &root->root_item;
4786         struct btrfs_trans_handle *trans;
4787         struct timespec64 ct = current_time(inode);
4788         int ret = 0;
4789         int received_uuid_changed;
4790
4791         if (!inode_owner_or_capable(mnt_userns, inode))
4792                 return -EPERM;
4793
4794         ret = mnt_want_write_file(file);
4795         if (ret < 0)
4796                 return ret;
4797
4798         down_write(&fs_info->subvol_sem);
4799
4800         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4801                 ret = -EINVAL;
4802                 goto out;
4803         }
4804
4805         if (btrfs_root_readonly(root)) {
4806                 ret = -EROFS;
4807                 goto out;
4808         }
4809
4810         /*
4811          * 1 - root item
4812          * 2 - uuid items (received uuid + subvol uuid)
4813          */
4814         trans = btrfs_start_transaction(root, 3);
4815         if (IS_ERR(trans)) {
4816                 ret = PTR_ERR(trans);
4817                 trans = NULL;
4818                 goto out;
4819         }
4820
4821         sa->rtransid = trans->transid;
4822         sa->rtime.sec = ct.tv_sec;
4823         sa->rtime.nsec = ct.tv_nsec;
4824
4825         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4826                                        BTRFS_UUID_SIZE);
4827         if (received_uuid_changed &&
4828             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4829                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4830                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4831                                           root->root_key.objectid);
4832                 if (ret && ret != -ENOENT) {
4833                         btrfs_abort_transaction(trans, ret);
4834                         btrfs_end_transaction(trans);
4835                         goto out;
4836                 }
4837         }
4838         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4839         btrfs_set_root_stransid(root_item, sa->stransid);
4840         btrfs_set_root_rtransid(root_item, sa->rtransid);
4841         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4842         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4843         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4844         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4845
4846         ret = btrfs_update_root(trans, fs_info->tree_root,
4847                                 &root->root_key, &root->root_item);
4848         if (ret < 0) {
4849                 btrfs_end_transaction(trans);
4850                 goto out;
4851         }
4852         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4853                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4854                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4855                                           root->root_key.objectid);
4856                 if (ret < 0 && ret != -EEXIST) {
4857                         btrfs_abort_transaction(trans, ret);
4858                         btrfs_end_transaction(trans);
4859                         goto out;
4860                 }
4861         }
4862         ret = btrfs_commit_transaction(trans);
4863 out:
4864         up_write(&fs_info->subvol_sem);
4865         mnt_drop_write_file(file);
4866         return ret;
4867 }
4868
4869 #ifdef CONFIG_64BIT
4870 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4871                                                 void __user *arg)
4872 {
4873         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4874         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4875         int ret = 0;
4876
4877         args32 = memdup_user(arg, sizeof(*args32));
4878         if (IS_ERR(args32))
4879                 return PTR_ERR(args32);
4880
4881         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4882         if (!args64) {
4883                 ret = -ENOMEM;
4884                 goto out;
4885         }
4886
4887         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4888         args64->stransid = args32->stransid;
4889         args64->rtransid = args32->rtransid;
4890         args64->stime.sec = args32->stime.sec;
4891         args64->stime.nsec = args32->stime.nsec;
4892         args64->rtime.sec = args32->rtime.sec;
4893         args64->rtime.nsec = args32->rtime.nsec;
4894         args64->flags = args32->flags;
4895
4896         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4897         if (ret)
4898                 goto out;
4899
4900         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4901         args32->stransid = args64->stransid;
4902         args32->rtransid = args64->rtransid;
4903         args32->stime.sec = args64->stime.sec;
4904         args32->stime.nsec = args64->stime.nsec;
4905         args32->rtime.sec = args64->rtime.sec;
4906         args32->rtime.nsec = args64->rtime.nsec;
4907         args32->flags = args64->flags;
4908
4909         ret = copy_to_user(arg, args32, sizeof(*args32));
4910         if (ret)
4911                 ret = -EFAULT;
4912
4913 out:
4914         kfree(args32);
4915         kfree(args64);
4916         return ret;
4917 }
4918 #endif
4919
4920 static long btrfs_ioctl_set_received_subvol(struct file *file,
4921                                             void __user *arg)
4922 {
4923         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4924         int ret = 0;
4925
4926         sa = memdup_user(arg, sizeof(*sa));
4927         if (IS_ERR(sa))
4928                 return PTR_ERR(sa);
4929
4930         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4931
4932         if (ret)
4933                 goto out;
4934
4935         ret = copy_to_user(arg, sa, sizeof(*sa));
4936         if (ret)
4937                 ret = -EFAULT;
4938
4939 out:
4940         kfree(sa);
4941         return ret;
4942 }
4943
4944 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4945                                         void __user *arg)
4946 {
4947         size_t len;
4948         int ret;
4949         char label[BTRFS_LABEL_SIZE];
4950
4951         spin_lock(&fs_info->super_lock);
4952         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4953         spin_unlock(&fs_info->super_lock);
4954
4955         len = strnlen(label, BTRFS_LABEL_SIZE);
4956
4957         if (len == BTRFS_LABEL_SIZE) {
4958                 btrfs_warn(fs_info,
4959                            "label is too long, return the first %zu bytes",
4960                            --len);
4961         }
4962
4963         ret = copy_to_user(arg, label, len);
4964
4965         return ret ? -EFAULT : 0;
4966 }
4967
4968 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4969 {
4970         struct inode *inode = file_inode(file);
4971         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4972         struct btrfs_root *root = BTRFS_I(inode)->root;
4973         struct btrfs_super_block *super_block = fs_info->super_copy;
4974         struct btrfs_trans_handle *trans;
4975         char label[BTRFS_LABEL_SIZE];
4976         int ret;
4977
4978         if (!capable(CAP_SYS_ADMIN))
4979                 return -EPERM;
4980
4981         if (copy_from_user(label, arg, sizeof(label)))
4982                 return -EFAULT;
4983
4984         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4985                 btrfs_err(fs_info,
4986                           "unable to set label with more than %d bytes",
4987                           BTRFS_LABEL_SIZE - 1);
4988                 return -EINVAL;
4989         }
4990
4991         ret = mnt_want_write_file(file);
4992         if (ret)
4993                 return ret;
4994
4995         trans = btrfs_start_transaction(root, 0);
4996         if (IS_ERR(trans)) {
4997                 ret = PTR_ERR(trans);
4998                 goto out_unlock;
4999         }
5000
5001         spin_lock(&fs_info->super_lock);
5002         strcpy(super_block->label, label);
5003         spin_unlock(&fs_info->super_lock);
5004         ret = btrfs_commit_transaction(trans);
5005
5006 out_unlock:
5007         mnt_drop_write_file(file);
5008         return ret;
5009 }
5010
5011 #define INIT_FEATURE_FLAGS(suffix) \
5012         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5013           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5014           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5015
5016 int btrfs_ioctl_get_supported_features(void __user *arg)
5017 {
5018         static const struct btrfs_ioctl_feature_flags features[3] = {
5019                 INIT_FEATURE_FLAGS(SUPP),
5020                 INIT_FEATURE_FLAGS(SAFE_SET),
5021                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5022         };
5023
5024         if (copy_to_user(arg, &features, sizeof(features)))
5025                 return -EFAULT;
5026
5027         return 0;
5028 }
5029
5030 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5031                                         void __user *arg)
5032 {
5033         struct btrfs_super_block *super_block = fs_info->super_copy;
5034         struct btrfs_ioctl_feature_flags features;
5035
5036         features.compat_flags = btrfs_super_compat_flags(super_block);
5037         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5038         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5039
5040         if (copy_to_user(arg, &features, sizeof(features)))
5041                 return -EFAULT;
5042
5043         return 0;
5044 }
5045
5046 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5047                               enum btrfs_feature_set set,
5048                               u64 change_mask, u64 flags, u64 supported_flags,
5049                               u64 safe_set, u64 safe_clear)
5050 {
5051         const char *type = btrfs_feature_set_name(set);
5052         char *names;
5053         u64 disallowed, unsupported;
5054         u64 set_mask = flags & change_mask;
5055         u64 clear_mask = ~flags & change_mask;
5056
5057         unsupported = set_mask & ~supported_flags;
5058         if (unsupported) {
5059                 names = btrfs_printable_features(set, unsupported);
5060                 if (names) {
5061                         btrfs_warn(fs_info,
5062                                    "this kernel does not support the %s feature bit%s",
5063                                    names, strchr(names, ',') ? "s" : "");
5064                         kfree(names);
5065                 } else
5066                         btrfs_warn(fs_info,
5067                                    "this kernel does not support %s bits 0x%llx",
5068                                    type, unsupported);
5069                 return -EOPNOTSUPP;
5070         }
5071
5072         disallowed = set_mask & ~safe_set;
5073         if (disallowed) {
5074                 names = btrfs_printable_features(set, disallowed);
5075                 if (names) {
5076                         btrfs_warn(fs_info,
5077                                    "can't set the %s feature bit%s while mounted",
5078                                    names, strchr(names, ',') ? "s" : "");
5079                         kfree(names);
5080                 } else
5081                         btrfs_warn(fs_info,
5082                                    "can't set %s bits 0x%llx while mounted",
5083                                    type, disallowed);
5084                 return -EPERM;
5085         }
5086
5087         disallowed = clear_mask & ~safe_clear;
5088         if (disallowed) {
5089                 names = btrfs_printable_features(set, disallowed);
5090                 if (names) {
5091                         btrfs_warn(fs_info,
5092                                    "can't clear the %s feature bit%s while mounted",
5093                                    names, strchr(names, ',') ? "s" : "");
5094                         kfree(names);
5095                 } else
5096                         btrfs_warn(fs_info,
5097                                    "can't clear %s bits 0x%llx while mounted",
5098                                    type, disallowed);
5099                 return -EPERM;
5100         }
5101
5102         return 0;
5103 }
5104
5105 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5106 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5107                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5108                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5109                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5110
5111 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5112 {
5113         struct inode *inode = file_inode(file);
5114         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5115         struct btrfs_root *root = BTRFS_I(inode)->root;
5116         struct btrfs_super_block *super_block = fs_info->super_copy;
5117         struct btrfs_ioctl_feature_flags flags[2];
5118         struct btrfs_trans_handle *trans;
5119         u64 newflags;
5120         int ret;
5121
5122         if (!capable(CAP_SYS_ADMIN))
5123                 return -EPERM;
5124
5125         if (copy_from_user(flags, arg, sizeof(flags)))
5126                 return -EFAULT;
5127
5128         /* Nothing to do */
5129         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5130             !flags[0].incompat_flags)
5131                 return 0;
5132
5133         ret = check_feature(fs_info, flags[0].compat_flags,
5134                             flags[1].compat_flags, COMPAT);
5135         if (ret)
5136                 return ret;
5137
5138         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5139                             flags[1].compat_ro_flags, COMPAT_RO);
5140         if (ret)
5141                 return ret;
5142
5143         ret = check_feature(fs_info, flags[0].incompat_flags,
5144                             flags[1].incompat_flags, INCOMPAT);
5145         if (ret)
5146                 return ret;
5147
5148         ret = mnt_want_write_file(file);
5149         if (ret)
5150                 return ret;
5151
5152         trans = btrfs_start_transaction(root, 0);
5153         if (IS_ERR(trans)) {
5154                 ret = PTR_ERR(trans);
5155                 goto out_drop_write;
5156         }
5157
5158         spin_lock(&fs_info->super_lock);
5159         newflags = btrfs_super_compat_flags(super_block);
5160         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5161         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5162         btrfs_set_super_compat_flags(super_block, newflags);
5163
5164         newflags = btrfs_super_compat_ro_flags(super_block);
5165         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5166         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5167         btrfs_set_super_compat_ro_flags(super_block, newflags);
5168
5169         newflags = btrfs_super_incompat_flags(super_block);
5170         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5171         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5172         btrfs_set_super_incompat_flags(super_block, newflags);
5173         spin_unlock(&fs_info->super_lock);
5174
5175         ret = btrfs_commit_transaction(trans);
5176 out_drop_write:
5177         mnt_drop_write_file(file);
5178
5179         return ret;
5180 }
5181
5182 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5183 {
5184         struct btrfs_ioctl_send_args *arg;
5185         int ret;
5186
5187         if (compat) {
5188 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5189                 struct btrfs_ioctl_send_args_32 args32;
5190
5191                 ret = copy_from_user(&args32, argp, sizeof(args32));
5192                 if (ret)
5193                         return -EFAULT;
5194                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5195                 if (!arg)
5196                         return -ENOMEM;
5197                 arg->send_fd = args32.send_fd;
5198                 arg->clone_sources_count = args32.clone_sources_count;
5199                 arg->clone_sources = compat_ptr(args32.clone_sources);
5200                 arg->parent_root = args32.parent_root;
5201                 arg->flags = args32.flags;
5202                 memcpy(arg->reserved, args32.reserved,
5203                        sizeof(args32.reserved));
5204 #else
5205                 return -ENOTTY;
5206 #endif
5207         } else {
5208                 arg = memdup_user(argp, sizeof(*arg));
5209                 if (IS_ERR(arg))
5210                         return PTR_ERR(arg);
5211         }
5212         ret = btrfs_ioctl_send(inode, arg);
5213         kfree(arg);
5214         return ret;
5215 }
5216
5217 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5218                                     bool compat)
5219 {
5220         struct btrfs_ioctl_encoded_io_args args = { 0 };
5221         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5222                                              flags);
5223         size_t copy_end;
5224         struct iovec iovstack[UIO_FASTIOV];
5225         struct iovec *iov = iovstack;
5226         struct iov_iter iter;
5227         loff_t pos;
5228         struct kiocb kiocb;
5229         ssize_t ret;
5230
5231         if (!capable(CAP_SYS_ADMIN)) {
5232                 ret = -EPERM;
5233                 goto out_acct;
5234         }
5235
5236         if (compat) {
5237 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5238                 struct btrfs_ioctl_encoded_io_args_32 args32;
5239
5240                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5241                                        flags);
5242                 if (copy_from_user(&args32, argp, copy_end)) {
5243                         ret = -EFAULT;
5244                         goto out_acct;
5245                 }
5246                 args.iov = compat_ptr(args32.iov);
5247                 args.iovcnt = args32.iovcnt;
5248                 args.offset = args32.offset;
5249                 args.flags = args32.flags;
5250 #else
5251                 return -ENOTTY;
5252 #endif
5253         } else {
5254                 copy_end = copy_end_kernel;
5255                 if (copy_from_user(&args, argp, copy_end)) {
5256                         ret = -EFAULT;
5257                         goto out_acct;
5258                 }
5259         }
5260         if (args.flags != 0) {
5261                 ret = -EINVAL;
5262                 goto out_acct;
5263         }
5264
5265         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5266                            &iov, &iter);
5267         if (ret < 0)
5268                 goto out_acct;
5269
5270         if (iov_iter_count(&iter) == 0) {
5271                 ret = 0;
5272                 goto out_iov;
5273         }
5274         pos = args.offset;
5275         ret = rw_verify_area(READ, file, &pos, args.len);
5276         if (ret < 0)
5277                 goto out_iov;
5278
5279         init_sync_kiocb(&kiocb, file);
5280         kiocb.ki_pos = pos;
5281
5282         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5283         if (ret >= 0) {
5284                 fsnotify_access(file);
5285                 if (copy_to_user(argp + copy_end,
5286                                  (char *)&args + copy_end_kernel,
5287                                  sizeof(args) - copy_end_kernel))
5288                         ret = -EFAULT;
5289         }
5290
5291 out_iov:
5292         kfree(iov);
5293 out_acct:
5294         if (ret > 0)
5295                 add_rchar(current, ret);
5296         inc_syscr(current);
5297         return ret;
5298 }
5299
5300 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5301 {
5302         struct btrfs_ioctl_encoded_io_args args;
5303         struct iovec iovstack[UIO_FASTIOV];
5304         struct iovec *iov = iovstack;
5305         struct iov_iter iter;
5306         loff_t pos;
5307         struct kiocb kiocb;
5308         ssize_t ret;
5309
5310         if (!capable(CAP_SYS_ADMIN)) {
5311                 ret = -EPERM;
5312                 goto out_acct;
5313         }
5314
5315         if (!(file->f_mode & FMODE_WRITE)) {
5316                 ret = -EBADF;
5317                 goto out_acct;
5318         }
5319
5320         if (compat) {
5321 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5322                 struct btrfs_ioctl_encoded_io_args_32 args32;
5323
5324                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5325                         ret = -EFAULT;
5326                         goto out_acct;
5327                 }
5328                 args.iov = compat_ptr(args32.iov);
5329                 args.iovcnt = args32.iovcnt;
5330                 args.offset = args32.offset;
5331                 args.flags = args32.flags;
5332                 args.len = args32.len;
5333                 args.unencoded_len = args32.unencoded_len;
5334                 args.unencoded_offset = args32.unencoded_offset;
5335                 args.compression = args32.compression;
5336                 args.encryption = args32.encryption;
5337                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5338 #else
5339                 return -ENOTTY;
5340 #endif
5341         } else {
5342                 if (copy_from_user(&args, argp, sizeof(args))) {
5343                         ret = -EFAULT;
5344                         goto out_acct;
5345                 }
5346         }
5347
5348         ret = -EINVAL;
5349         if (args.flags != 0)
5350                 goto out_acct;
5351         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5352                 goto out_acct;
5353         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5354             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5355                 goto out_acct;
5356         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5357             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5358                 goto out_acct;
5359         if (args.unencoded_offset > args.unencoded_len)
5360                 goto out_acct;
5361         if (args.len > args.unencoded_len - args.unencoded_offset)
5362                 goto out_acct;
5363
5364         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5365                            &iov, &iter);
5366         if (ret < 0)
5367                 goto out_acct;
5368
5369         file_start_write(file);
5370
5371         if (iov_iter_count(&iter) == 0) {
5372                 ret = 0;
5373                 goto out_end_write;
5374         }
5375         pos = args.offset;
5376         ret = rw_verify_area(WRITE, file, &pos, args.len);
5377         if (ret < 0)
5378                 goto out_end_write;
5379
5380         init_sync_kiocb(&kiocb, file);
5381         ret = kiocb_set_rw_flags(&kiocb, 0);
5382         if (ret)
5383                 goto out_end_write;
5384         kiocb.ki_pos = pos;
5385
5386         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5387         if (ret > 0)
5388                 fsnotify_modify(file);
5389
5390 out_end_write:
5391         file_end_write(file);
5392         kfree(iov);
5393 out_acct:
5394         if (ret > 0)
5395                 add_wchar(current, ret);
5396         inc_syscw(current);
5397         return ret;
5398 }
5399
5400 long btrfs_ioctl(struct file *file, unsigned int
5401                 cmd, unsigned long arg)
5402 {
5403         struct inode *inode = file_inode(file);
5404         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5405         struct btrfs_root *root = BTRFS_I(inode)->root;
5406         void __user *argp = (void __user *)arg;
5407
5408         switch (cmd) {
5409         case FS_IOC_GETVERSION:
5410                 return btrfs_ioctl_getversion(inode, argp);
5411         case FS_IOC_GETFSLABEL:
5412                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5413         case FS_IOC_SETFSLABEL:
5414                 return btrfs_ioctl_set_fslabel(file, argp);
5415         case FITRIM:
5416                 return btrfs_ioctl_fitrim(fs_info, argp);
5417         case BTRFS_IOC_SNAP_CREATE:
5418                 return btrfs_ioctl_snap_create(file, argp, 0);
5419         case BTRFS_IOC_SNAP_CREATE_V2:
5420                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5421         case BTRFS_IOC_SUBVOL_CREATE:
5422                 return btrfs_ioctl_snap_create(file, argp, 1);
5423         case BTRFS_IOC_SUBVOL_CREATE_V2:
5424                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5425         case BTRFS_IOC_SNAP_DESTROY:
5426                 return btrfs_ioctl_snap_destroy(file, argp, false);
5427         case BTRFS_IOC_SNAP_DESTROY_V2:
5428                 return btrfs_ioctl_snap_destroy(file, argp, true);
5429         case BTRFS_IOC_SUBVOL_GETFLAGS:
5430                 return btrfs_ioctl_subvol_getflags(inode, argp);
5431         case BTRFS_IOC_SUBVOL_SETFLAGS:
5432                 return btrfs_ioctl_subvol_setflags(file, argp);
5433         case BTRFS_IOC_DEFAULT_SUBVOL:
5434                 return btrfs_ioctl_default_subvol(file, argp);
5435         case BTRFS_IOC_DEFRAG:
5436                 return btrfs_ioctl_defrag(file, NULL);
5437         case BTRFS_IOC_DEFRAG_RANGE:
5438                 return btrfs_ioctl_defrag(file, argp);
5439         case BTRFS_IOC_RESIZE:
5440                 return btrfs_ioctl_resize(file, argp);
5441         case BTRFS_IOC_ADD_DEV:
5442                 return btrfs_ioctl_add_dev(fs_info, argp);
5443         case BTRFS_IOC_RM_DEV:
5444                 return btrfs_ioctl_rm_dev(file, argp);
5445         case BTRFS_IOC_RM_DEV_V2:
5446                 return btrfs_ioctl_rm_dev_v2(file, argp);
5447         case BTRFS_IOC_FS_INFO:
5448                 return btrfs_ioctl_fs_info(fs_info, argp);
5449         case BTRFS_IOC_DEV_INFO:
5450                 return btrfs_ioctl_dev_info(fs_info, argp);
5451         case BTRFS_IOC_BALANCE:
5452                 return btrfs_ioctl_balance(file, NULL);
5453         case BTRFS_IOC_TREE_SEARCH:
5454                 return btrfs_ioctl_tree_search(inode, argp);
5455         case BTRFS_IOC_TREE_SEARCH_V2:
5456                 return btrfs_ioctl_tree_search_v2(inode, argp);
5457         case BTRFS_IOC_INO_LOOKUP:
5458                 return btrfs_ioctl_ino_lookup(root, argp);
5459         case BTRFS_IOC_INO_PATHS:
5460                 return btrfs_ioctl_ino_to_path(root, argp);
5461         case BTRFS_IOC_LOGICAL_INO:
5462                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5463         case BTRFS_IOC_LOGICAL_INO_V2:
5464                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5465         case BTRFS_IOC_SPACE_INFO:
5466                 return btrfs_ioctl_space_info(fs_info, argp);
5467         case BTRFS_IOC_SYNC: {
5468                 int ret;
5469
5470                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5471                 if (ret)
5472                         return ret;
5473                 ret = btrfs_sync_fs(inode->i_sb, 1);
5474                 /*
5475                  * The transaction thread may want to do more work,
5476                  * namely it pokes the cleaner kthread that will start
5477                  * processing uncleaned subvols.
5478                  */
5479                 wake_up_process(fs_info->transaction_kthread);
5480                 return ret;
5481         }
5482         case BTRFS_IOC_START_SYNC:
5483                 return btrfs_ioctl_start_sync(root, argp);
5484         case BTRFS_IOC_WAIT_SYNC:
5485                 return btrfs_ioctl_wait_sync(fs_info, argp);
5486         case BTRFS_IOC_SCRUB:
5487                 return btrfs_ioctl_scrub(file, argp);
5488         case BTRFS_IOC_SCRUB_CANCEL:
5489                 return btrfs_ioctl_scrub_cancel(fs_info);
5490         case BTRFS_IOC_SCRUB_PROGRESS:
5491                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5492         case BTRFS_IOC_BALANCE_V2:
5493                 return btrfs_ioctl_balance(file, argp);
5494         case BTRFS_IOC_BALANCE_CTL:
5495                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5496         case BTRFS_IOC_BALANCE_PROGRESS:
5497                 return btrfs_ioctl_balance_progress(fs_info, argp);
5498         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5499                 return btrfs_ioctl_set_received_subvol(file, argp);
5500 #ifdef CONFIG_64BIT
5501         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5502                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5503 #endif
5504         case BTRFS_IOC_SEND:
5505                 return _btrfs_ioctl_send(inode, argp, false);
5506 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5507         case BTRFS_IOC_SEND_32:
5508                 return _btrfs_ioctl_send(inode, argp, true);
5509 #endif
5510         case BTRFS_IOC_GET_DEV_STATS:
5511                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5512         case BTRFS_IOC_QUOTA_CTL:
5513                 return btrfs_ioctl_quota_ctl(file, argp);
5514         case BTRFS_IOC_QGROUP_ASSIGN:
5515                 return btrfs_ioctl_qgroup_assign(file, argp);
5516         case BTRFS_IOC_QGROUP_CREATE:
5517                 return btrfs_ioctl_qgroup_create(file, argp);
5518         case BTRFS_IOC_QGROUP_LIMIT:
5519                 return btrfs_ioctl_qgroup_limit(file, argp);
5520         case BTRFS_IOC_QUOTA_RESCAN:
5521                 return btrfs_ioctl_quota_rescan(file, argp);
5522         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5523                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5524         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5525                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5526         case BTRFS_IOC_DEV_REPLACE:
5527                 return btrfs_ioctl_dev_replace(fs_info, argp);
5528         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5529                 return btrfs_ioctl_get_supported_features(argp);
5530         case BTRFS_IOC_GET_FEATURES:
5531                 return btrfs_ioctl_get_features(fs_info, argp);
5532         case BTRFS_IOC_SET_FEATURES:
5533                 return btrfs_ioctl_set_features(file, argp);
5534         case BTRFS_IOC_GET_SUBVOL_INFO:
5535                 return btrfs_ioctl_get_subvol_info(inode, argp);
5536         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5537                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5538         case BTRFS_IOC_INO_LOOKUP_USER:
5539                 return btrfs_ioctl_ino_lookup_user(file, argp);
5540         case FS_IOC_ENABLE_VERITY:
5541                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5542         case FS_IOC_MEASURE_VERITY:
5543                 return fsverity_ioctl_measure(file, argp);
5544         case BTRFS_IOC_ENCODED_READ:
5545                 return btrfs_ioctl_encoded_read(file, argp, false);
5546         case BTRFS_IOC_ENCODED_WRITE:
5547                 return btrfs_ioctl_encoded_write(file, argp, false);
5548 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5549         case BTRFS_IOC_ENCODED_READ_32:
5550                 return btrfs_ioctl_encoded_read(file, argp, true);
5551         case BTRFS_IOC_ENCODED_WRITE_32:
5552                 return btrfs_ioctl_encoded_write(file, argp, true);
5553 #endif
5554         }
5555
5556         return -ENOTTY;
5557 }
5558
5559 #ifdef CONFIG_COMPAT
5560 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5561 {
5562         /*
5563          * These all access 32-bit values anyway so no further
5564          * handling is necessary.
5565          */
5566         switch (cmd) {
5567         case FS_IOC32_GETVERSION:
5568                 cmd = FS_IOC_GETVERSION;
5569                 break;
5570         }
5571
5572         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5573 }
5574 #endif