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 /*
 548  * Calculate the number of transaction items to reserve for creating a subvolume
 549  * or snapshot, not including the inode, directory entries, or parent directory.
 550  */
 551 static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit)
 552 {
 553         /*
 554          * 1 to add root block
 555          * 1 to add root item
 556          * 1 to add root ref
 557          * 1 to add root backref
 558          * 1 to add UUID item
 559          * 1 to add qgroup info
 560          * 1 to add qgroup limit
 561          *
 562          * Ideally the last two would only be accounted if qgroups are enabled,
 563          * but that can change between now and the time we would insert them.
 564          */
 565         unsigned int num_items = 7;
 566
 567         if (inherit) {
 568                 /* 2 to add qgroup relations for each inherited qgroup */
 569                 num_items += 2 * inherit->num_qgroups;
 570         }
 571         return num_items;
 572 }
 573
 574 static noinline int create_subvol(struct user_namespace *mnt_userns,
 575                                   struct inode *dir, struct dentry *dentry,
 576                                   struct btrfs_qgroup_inherit *inherit)
 577 {
 578         const char *name = dentry->d_name.name;
 579         int namelen = dentry->d_name.len;
 580         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 581         struct btrfs_trans_handle *trans;
 582         struct btrfs_key key;
 583         struct btrfs_root_item *root_item;
 584         struct btrfs_inode_item *inode_item;
 585         struct extent_buffer *leaf;
 586         struct btrfs_root *root = BTRFS_I(dir)->root;
 587         struct btrfs_root *new_root;
 588         struct btrfs_block_rsv block_rsv;
 589         struct timespec64 cur_time = current_time(dir);
 590         struct btrfs_new_inode_args new_inode_args = {
 591                 .dir = dir,
 592                 .dentry = dentry,
 593                 .subvol = true,
 594         };
 595         unsigned int trans_num_items;
 596         int ret;
 597         dev_t anon_dev;
 598         u64 objectid;
 599         u64 index = 0;
 600
 601         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 602         if (!root_item)
 603                 return -ENOMEM;
 604
 605         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 606         if (ret)
 607                 goto out_root_item;
 608
 609         /*
 610          * Don't create subvolume whose level is not zero. Or qgroup will be
 611          * screwed up since it assumes subvolume qgroup's level to be 0.
 612          */
 613         if (btrfs_qgroup_level(objectid)) {
 614                 ret = -ENOSPC;
 615                 goto out_root_item;
 616         }
 617
 618         ret = get_anon_bdev(&anon_dev);
 619         if (ret < 0)
 620                 goto out_root_item;
 621
 622         new_inode_args.inode = btrfs_new_subvol_inode(mnt_userns, dir);
 623         if (!new_inode_args.inode) {
 624                 ret = -ENOMEM;
 625                 goto out_anon_dev;
 626         }
 627         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
 628         if (ret)
 629                 goto out_inode;
 630         trans_num_items += create_subvol_num_items(inherit);
 631
 632         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 633         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 634                                                trans_num_items, false);
 635         if (ret)
 636                 goto out_new_inode_args;
 637
 638         trans = btrfs_start_transaction(root, 0);
 639         if (IS_ERR(trans)) {
 640                 ret = PTR_ERR(trans);
 641                 btrfs_subvolume_release_metadata(root, &block_rsv);
 642                 goto out_new_inode_args;
 643         }
 644         trans->block_rsv = &block_rsv;
 645         trans->bytes_reserved = block_rsv.size;
 646
 647         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 648         if (ret)
 649                 goto out;
 650
 651         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 652                                       BTRFS_NESTING_NORMAL);
 653         if (IS_ERR(leaf)) {
 654                 ret = PTR_ERR(leaf);
 655                 goto out;
 656         }
 657
 658         btrfs_mark_buffer_dirty(leaf);
 659
 660         inode_item = &root_item->inode;
 661         btrfs_set_stack_inode_generation(inode_item, 1);
 662         btrfs_set_stack_inode_size(inode_item, 3);
 663         btrfs_set_stack_inode_nlink(inode_item, 1);
 664         btrfs_set_stack_inode_nbytes(inode_item,
 665                                      fs_info->nodesize);
 666         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 667
 668         btrfs_set_root_flags(root_item, 0);
 669         btrfs_set_root_limit(root_item, 0);
 670         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 671
 672         btrfs_set_root_bytenr(root_item, leaf->start);
 673         btrfs_set_root_generation(root_item, trans->transid);
 674         btrfs_set_root_level(root_item, 0);
 675         btrfs_set_root_refs(root_item, 1);
 676         btrfs_set_root_used(root_item, leaf->len);
 677         btrfs_set_root_last_snapshot(root_item, 0);
 678
 679         btrfs_set_root_generation_v2(root_item,
 680                         btrfs_root_generation(root_item));
 681         generate_random_guid(root_item->uuid);
 682         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 683         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 684         root_item->ctime = root_item->otime;
 685         btrfs_set_root_ctransid(root_item, trans->transid);
 686         btrfs_set_root_otransid(root_item, trans->transid);
 687
 688         btrfs_tree_unlock(leaf);
 689
 690         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 691
 692         key.objectid = objectid;
 693         key.offset = 0;
 694         key.type = BTRFS_ROOT_ITEM_KEY;
 695         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 696                                 root_item);
 697         if (ret) {
 698                 /*
 699                  * Since we don't abort the transaction in this case, free the
 700                  * tree block so that we don't leak space and leave the
 701                  * filesystem in an inconsistent state (an extent item in the
 702                  * extent tree with a backreference for a root that does not
 703                  * exists).
 704                  */
 705                 btrfs_tree_lock(leaf);
 706                 btrfs_clean_tree_block(leaf);
 707                 btrfs_tree_unlock(leaf);
 708                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 709                 free_extent_buffer(leaf);
 710                 goto out;
 711         }
 712
 713         free_extent_buffer(leaf);
 714         leaf = NULL;
 715
 716         key.offset = (u64)-1;
 717         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 718         if (IS_ERR(new_root)) {
 719                 ret = PTR_ERR(new_root);
 720                 btrfs_abort_transaction(trans, ret);
 721                 goto out;
 722         }
 723         /* anon_dev is owned by new_root now. */
 724         anon_dev = 0;
 725         BTRFS_I(new_inode_args.inode)->root = new_root;
 726         /* ... and new_root is owned by new_inode_args.inode now. */
 727
 728         ret = btrfs_record_root_in_trans(trans, new_root);
 729         if (ret) {
 730                 btrfs_abort_transaction(trans, ret);
 731                 goto out;
 732         }
 733
 734         ret = btrfs_create_subvol_root(trans, root, &new_inode_args);
 735         if (ret) {
 736                 /* We potentially lose an unused inode item here */
 737                 btrfs_abort_transaction(trans, ret);
 738                 goto out;
 739         }
 740
 741         /*
 742          * insert the directory item
 743          */
 744         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 745         if (ret) {
 746                 btrfs_abort_transaction(trans, ret);
 747                 goto out;
 748         }
 749
 750         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 751                                     BTRFS_FT_DIR, index);
 752         if (ret) {
 753                 btrfs_abort_transaction(trans, ret);
 754                 goto out;
 755         }
 756
 757         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 758         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 759         if (ret) {
 760                 btrfs_abort_transaction(trans, ret);
 761                 goto out;
 762         }
 763
 764         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 765                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 766         if (ret) {
 767                 btrfs_abort_transaction(trans, ret);
 768                 goto out;
 769         }
 770
 771         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 772                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 773         if (ret)
 774                 btrfs_abort_transaction(trans, ret);
 775
 776 out:
 777         trans->block_rsv = NULL;
 778         trans->bytes_reserved = 0;
 779         btrfs_subvolume_release_metadata(root, &block_rsv);
 780
 781         if (ret)
 782                 btrfs_end_transaction(trans);
 783         else
 784                 ret = btrfs_commit_transaction(trans);
 785
 786         if (!ret) {
 787                 d_instantiate(dentry, new_inode_args.inode);
 788                 new_inode_args.inode = NULL;
 789         }
 790 out_new_inode_args:
 791         btrfs_new_inode_args_destroy(&new_inode_args);
 792 out_inode:
 793         iput(new_inode_args.inode);
 794 out_anon_dev:
 795         if (anon_dev)
 796                 free_anon_bdev(anon_dev);
 797 out_root_item:
 798         kfree(root_item);
 799         return ret;
 800 }
 801
 802 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 803                            struct dentry *dentry, bool readonly,
 804                            struct btrfs_qgroup_inherit *inherit)
 805 {
 806         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 807         struct inode *inode;
 808         struct btrfs_pending_snapshot *pending_snapshot;
 809         unsigned int trans_num_items;
 810         struct btrfs_trans_handle *trans;
 811         int ret;
 812
 813         /* We do not support snapshotting right now. */
 814         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 815                 btrfs_warn(fs_info,
 816                            "extent tree v2 doesn't support snapshotting yet");
 817                 return -EOPNOTSUPP;
 818         }
 819
 820         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 821                 return -EINVAL;
 822
 823         if (atomic_read(&root->nr_swapfiles)) {
 824                 btrfs_warn(fs_info,
 825                            "cannot snapshot subvolume with active swapfile");
 826                 return -ETXTBSY;
 827         }
 828
 829         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 830         if (!pending_snapshot)
 831                 return -ENOMEM;
 832
 833         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 834         if (ret < 0)
 835                 goto free_pending;
 836         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 837                         GFP_KERNEL);
 838         pending_snapshot->path = btrfs_alloc_path();
 839         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 840                 ret = -ENOMEM;
 841                 goto free_pending;
 842         }
 843
 844         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 845                              BTRFS_BLOCK_RSV_TEMP);
 846         /*
 847          * 1 to add dir item
 848          * 1 to add dir index
 849          * 1 to update parent inode item
 850          */
 851         trans_num_items = create_subvol_num_items(inherit) + 3;
 852         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 853                                                &pending_snapshot->block_rsv,
 854                                                trans_num_items, false);
 855         if (ret)
 856                 goto free_pending;
 857
 858         pending_snapshot->dentry = dentry;
 859         pending_snapshot->root = root;
 860         pending_snapshot->readonly = readonly;
 861         pending_snapshot->dir = dir;
 862         pending_snapshot->inherit = inherit;
 863
 864         trans = btrfs_start_transaction(root, 0);
 865         if (IS_ERR(trans)) {
 866                 ret = PTR_ERR(trans);
 867                 goto fail;
 868         }
 869
 870         trans->pending_snapshot = pending_snapshot;
 871
 872         ret = btrfs_commit_transaction(trans);
 873         if (ret)
 874                 goto fail;
 875
 876         ret = pending_snapshot->error;
 877         if (ret)
 878                 goto fail;
 879
 880         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 881         if (ret)
 882                 goto fail;
 883
 884         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 885         if (IS_ERR(inode)) {
 886                 ret = PTR_ERR(inode);
 887                 goto fail;
 888         }
 889
 890         d_instantiate(dentry, inode);
 891         ret = 0;
 892         pending_snapshot->anon_dev = 0;
 893 fail:
 894         /* Prevent double freeing of anon_dev */
 895         if (ret && pending_snapshot->snap)
 896                 pending_snapshot->snap->anon_dev = 0;
 897         btrfs_put_root(pending_snapshot->snap);
 898         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 899 free_pending:
 900         if (pending_snapshot->anon_dev)
 901                 free_anon_bdev(pending_snapshot->anon_dev);
 902         kfree(pending_snapshot->root_item);
 903         btrfs_free_path(pending_snapshot->path);
 904         kfree(pending_snapshot);
 905
 906         return ret;
 907 }
 908
 909 /*  copy of may_delete in fs/namei.c()
 910  *      Check whether we can remove a link victim from directory dir, check
 911  *  whether the type of victim is right.
 912  *  1. We can't do it if dir is read-only (done in permission())
 913  *  2. We should have write and exec permissions on dir
 914  *  3. We can't remove anything from append-only dir
 915  *  4. We can't do anything with immutable dir (done in permission())
 916  *  5. If the sticky bit on dir is set we should either
 917  *      a. be owner of dir, or
 918  *      b. be owner of victim, or
 919  *      c. have CAP_FOWNER capability
 920  *  6. If the victim is append-only or immutable we can't do anything with
 921  *     links pointing to it.
 922  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 923  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 924  *  9. We can't remove a root or mountpoint.
 925  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 926  *     nfs_async_unlink().
 927  */
 928
 929 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 930                             struct inode *dir, struct dentry *victim, int isdir)
 931 {
 932         int error;
 933
 934         if (d_really_is_negative(victim))
 935                 return -ENOENT;
 936
 937         BUG_ON(d_inode(victim->d_parent) != dir);
 938         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 939
 940         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 941         if (error)
 942                 return error;
 943         if (IS_APPEND(dir))
 944                 return -EPERM;
 945         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 946             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 947             IS_SWAPFILE(d_inode(victim)))
 948                 return -EPERM;
 949         if (isdir) {
 950                 if (!d_is_dir(victim))
 951                         return -ENOTDIR;
 952                 if (IS_ROOT(victim))
 953                         return -EBUSY;
 954         } else if (d_is_dir(victim))
 955                 return -EISDIR;
 956         if (IS_DEADDIR(dir))
 957                 return -ENOENT;
 958         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 959                 return -EBUSY;
 960         return 0;
 961 }
 962
 963 /* copy of may_create in fs/namei.c() */
 964 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 965                                    struct inode *dir, struct dentry *child)
 966 {
 967         if (d_really_is_positive(child))
 968                 return -EEXIST;
 969         if (IS_DEADDIR(dir))
 970                 return -ENOENT;
 971         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 972                 return -EOVERFLOW;
 973         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 974 }
 975
 976 /*
 977  * Create a new subvolume below @parent.  This is largely modeled after
 978  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 979  * inside this filesystem so it's quite a bit simpler.
 980  */
 981 static noinline int btrfs_mksubvol(const struct path *parent,
 982                                    struct user_namespace *mnt_userns,
 983                                    const char *name, int namelen,
 984                                    struct btrfs_root *snap_src,
 985                                    bool readonly,
 986                                    struct btrfs_qgroup_inherit *inherit)
 987 {
 988         struct inode *dir = d_inode(parent->dentry);
 989         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 990         struct dentry *dentry;
 991         int error;
 992
 993         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 994         if (error == -EINTR)
 995                 return error;
 996
 997         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 998         error = PTR_ERR(dentry);
 999         if (IS_ERR(dentry))
1000                 goto out_unlock;
1001
1002         error = btrfs_may_create(mnt_userns, dir, dentry);
1003         if (error)
1004                 goto out_dput;
1005
1006         /*
1007          * even if this name doesn't exist, we may get hash collisions.
1008          * check for them now when we can safely fail
1009          */
1010         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
1011                                                dir->i_ino, name,
1012                                                namelen);
1013         if (error)
1014                 goto out_dput;
1015
1016         down_read(&fs_info->subvol_sem);
1017
1018         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1019                 goto out_up_read;
1020
1021         if (snap_src)
1022                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1023         else
1024                 error = create_subvol(mnt_userns, dir, dentry, inherit);
1025
1026         if (!error)
1027                 fsnotify_mkdir(dir, dentry);
1028 out_up_read:
1029         up_read(&fs_info->subvol_sem);
1030 out_dput:
1031         dput(dentry);
1032 out_unlock:
1033         btrfs_inode_unlock(dir, 0);
1034         return error;
1035 }
1036
1037 static noinline int btrfs_mksnapshot(const struct path *parent,
1038                                    struct user_namespace *mnt_userns,
1039                                    const char *name, int namelen,
1040                                    struct btrfs_root *root,
1041                                    bool readonly,
1042                                    struct btrfs_qgroup_inherit *inherit)
1043 {
1044         int ret;
1045         bool snapshot_force_cow = false;
1046
1047         /*
1048          * Force new buffered writes to reserve space even when NOCOW is
1049          * possible. This is to avoid later writeback (running dealloc) to
1050          * fallback to COW mode and unexpectedly fail with ENOSPC.
1051          */
1052         btrfs_drew_read_lock(&root->snapshot_lock);
1053
1054         ret = btrfs_start_delalloc_snapshot(root, false);
1055         if (ret)
1056                 goto out;
1057
1058         /*
1059          * All previous writes have started writeback in NOCOW mode, so now
1060          * we force future writes to fallback to COW mode during snapshot
1061          * creation.
1062          */
1063         atomic_inc(&root->snapshot_force_cow);
1064         snapshot_force_cow = true;
1065
1066         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1067
1068         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1069                              root, readonly, inherit);
1070 out:
1071         if (snapshot_force_cow)
1072                 atomic_dec(&root->snapshot_force_cow);
1073         btrfs_drew_read_unlock(&root->snapshot_lock);
1074         return ret;
1075 }
1076
1077 /*
1078  * Defrag specific helper to get an extent map.
1079  *
1080  * Differences between this and btrfs_get_extent() are:
1081  *
1082  * - No extent_map will be added to inode->extent_tree
1083  *   To reduce memory usage in the long run.
1084  *
1085  * - Extra optimization to skip file extents older than @newer_than
1086  *   By using btrfs_search_forward() we can skip entire file ranges that
1087  *   have extents created in past transactions, because btrfs_search_forward()
1088  *   will not visit leaves and nodes with a generation smaller than given
1089  *   minimal generation threshold (@newer_than).
1090  *
1091  * Return valid em if we find a file extent matching the requirement.
1092  * Return NULL if we can not find a file extent matching the requirement.
1093  *
1094  * Return ERR_PTR() for error.
1095  */
1096 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1097                                             u64 start, u64 newer_than)
1098 {
1099         struct btrfs_root *root = inode->root;
1100         struct btrfs_file_extent_item *fi;
1101         struct btrfs_path path = { 0 };
1102         struct extent_map *em;
1103         struct btrfs_key key;
1104         u64 ino = btrfs_ino(inode);
1105         int ret;
1106
1107         em = alloc_extent_map();
1108         if (!em) {
1109                 ret = -ENOMEM;
1110                 goto err;
1111         }
1112
1113         key.objectid = ino;
1114         key.type = BTRFS_EXTENT_DATA_KEY;
1115         key.offset = start;
1116
1117         if (newer_than) {
1118                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1119                 if (ret < 0)
1120                         goto err;
1121                 /* Can't find anything newer */
1122                 if (ret > 0)
1123                         goto not_found;
1124         } else {
1125                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1126                 if (ret < 0)
1127                         goto err;
1128         }
1129         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1130                 /*
1131                  * If btrfs_search_slot() makes path to point beyond nritems,
1132                  * we should not have an empty leaf, as this inode must at
1133                  * least have its INODE_ITEM.
1134                  */
1135                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1136                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1137         }
1138         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1139         /* Perfect match, no need to go one slot back */
1140         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1141             key.offset == start)
1142                 goto iterate;
1143
1144         /* We didn't find a perfect match, needs to go one slot back */
1145         if (path.slots[0] > 0) {
1146                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1147                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1148                         path.slots[0]--;
1149         }
1150
1151 iterate:
1152         /* Iterate through the path to find a file extent covering @start */
1153         while (true) {
1154                 u64 extent_end;
1155
1156                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1157                         goto next;
1158
1159                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1160
1161                 /*
1162                  * We may go one slot back to INODE_REF/XATTR item, then
1163                  * need to go forward until we reach an EXTENT_DATA.
1164                  * But we should still has the correct ino as key.objectid.
1165                  */
1166                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1167                         goto next;
1168
1169                 /* It's beyond our target range, definitely not extent found */
1170                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1171                         goto not_found;
1172
1173                 /*
1174                  *      |       |<- File extent ->|
1175                  *      \- start
1176                  *
1177                  * This means there is a hole between start and key.offset.
1178                  */
1179                 if (key.offset > start) {
1180                         em->start = start;
1181                         em->orig_start = start;
1182                         em->block_start = EXTENT_MAP_HOLE;
1183                         em->len = key.offset - start;
1184                         break;
1185                 }
1186
1187                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1188                                     struct btrfs_file_extent_item);
1189                 extent_end = btrfs_file_extent_end(&path);
1190
1191                 /*
1192                  *      |<- file extent ->|     |
1193                  *                              \- start
1194                  *
1195                  * We haven't reached start, search next slot.
1196                  */
1197                 if (extent_end <= start)
1198                         goto next;
1199
1200                 /* Now this extent covers @start, convert it to em */
1201                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1202                 break;
1203 next:
1204                 ret = btrfs_next_item(root, &path);
1205                 if (ret < 0)
1206                         goto err;
1207                 if (ret > 0)
1208                         goto not_found;
1209         }
1210         btrfs_release_path(&path);
1211         return em;
1212
1213 not_found:
1214         btrfs_release_path(&path);
1215         free_extent_map(em);
1216         return NULL;
1217
1218 err:
1219         btrfs_release_path(&path);
1220         free_extent_map(em);
1221         return ERR_PTR(ret);
1222 }
1223
1224 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1225                                                u64 newer_than, bool locked)
1226 {
1227         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1228         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1229         struct extent_map *em;
1230         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1231
1232         /*
1233          * hopefully we have this extent in the tree already, try without
1234          * the full extent lock
1235          */
1236         read_lock(&em_tree->lock);
1237         em = lookup_extent_mapping(em_tree, start, sectorsize);
1238         read_unlock(&em_tree->lock);
1239
1240         /*
1241          * We can get a merged extent, in that case, we need to re-search
1242          * tree to get the original em for defrag.
1243          *
1244          * If @newer_than is 0 or em::generation < newer_than, we can trust
1245          * this em, as either we don't care about the generation, or the
1246          * merged extent map will be rejected anyway.
1247          */
1248         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1249             newer_than && em->generation >= newer_than) {
1250                 free_extent_map(em);
1251                 em = NULL;
1252         }
1253
1254         if (!em) {
1255                 struct extent_state *cached = NULL;
1256                 u64 end = start + sectorsize - 1;
1257
1258                 /* get the big lock and read metadata off disk */
1259                 if (!locked)
1260                         lock_extent_bits(io_tree, start, end, &cached);
1261                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1262                 if (!locked)
1263                         unlock_extent_cached(io_tree, start, end, &cached);
1264
1265                 if (IS_ERR(em))
1266                         return NULL;
1267         }
1268
1269         return em;
1270 }
1271
1272 static u32 get_extent_max_capacity(const struct extent_map *em)
1273 {
1274         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1275                 return BTRFS_MAX_COMPRESSED;
1276         return BTRFS_MAX_EXTENT_SIZE;
1277 }
1278
1279 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1280                                      u32 extent_thresh, u64 newer_than, bool locked)
1281 {
1282         struct extent_map *next;
1283         bool ret = false;
1284
1285         /* this is the last extent */
1286         if (em->start + em->len >= i_size_read(inode))
1287                 return false;
1288
1289         /*
1290          * Here we need to pass @newer_then when checking the next extent, or
1291          * we will hit a case we mark current extent for defrag, but the next
1292          * one will not be a target.
1293          * This will just cause extra IO without really reducing the fragments.
1294          */
1295         next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked);
1296         /* No more em or hole */
1297         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1298                 goto out;
1299         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1300                 goto out;
1301         /*
1302          * If the next extent is at its max capacity, defragging current extent
1303          * makes no sense, as the total number of extents won't change.
1304          */
1305         if (next->len >= get_extent_max_capacity(em))
1306                 goto out;
1307         /* Skip older extent */
1308         if (next->generation < newer_than)
1309                 goto out;
1310         /* Also check extent size */
1311         if (next->len >= extent_thresh)
1312                 goto out;
1313
1314         ret = true;
1315 out:
1316         free_extent_map(next);
1317         return ret;
1318 }
1319
1320 /*
1321  * Prepare one page to be defragged.
1322  *
1323  * This will ensure:
1324  *
1325  * - Returned page is locked and has been set up properly.
1326  * - No ordered extent exists in the page.
1327  * - The page is uptodate.
1328  *
1329  * NOTE: Caller should also wait for page writeback after the cluster is
1330  * prepared, here we don't do writeback wait for each page.
1331  */
1332 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1333                                             pgoff_t index)
1334 {
1335         struct address_space *mapping = inode->vfs_inode.i_mapping;
1336         gfp_t mask = btrfs_alloc_write_mask(mapping);
1337         u64 page_start = (u64)index << PAGE_SHIFT;
1338         u64 page_end = page_start + PAGE_SIZE - 1;
1339         struct extent_state *cached_state = NULL;
1340         struct page *page;
1341         int ret;
1342
1343 again:
1344         page = find_or_create_page(mapping, index, mask);
1345         if (!page)
1346                 return ERR_PTR(-ENOMEM);
1347
1348         /*
1349          * Since we can defragment files opened read-only, we can encounter
1350          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1351          * can't do I/O using huge pages yet, so return an error for now.
1352          * Filesystem transparent huge pages are typically only used for
1353          * executables that explicitly enable them, so this isn't very
1354          * restrictive.
1355          */
1356         if (PageCompound(page)) {
1357                 unlock_page(page);
1358                 put_page(page);
1359                 return ERR_PTR(-ETXTBSY);
1360         }
1361
1362         ret = set_page_extent_mapped(page);
1363         if (ret < 0) {
1364                 unlock_page(page);
1365                 put_page(page);
1366                 return ERR_PTR(ret);
1367         }
1368
1369         /* Wait for any existing ordered extent in the range */
1370         while (1) {
1371                 struct btrfs_ordered_extent *ordered;
1372
1373                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1374                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1375                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1376                                      &cached_state);
1377                 if (!ordered)
1378                         break;
1379
1380                 unlock_page(page);
1381                 btrfs_start_ordered_extent(ordered, 1);
1382                 btrfs_put_ordered_extent(ordered);
1383                 lock_page(page);
1384                 /*
1385                  * We unlocked the page above, so we need check if it was
1386                  * released or not.
1387                  */
1388                 if (page->mapping != mapping || !PagePrivate(page)) {
1389                         unlock_page(page);
1390                         put_page(page);
1391                         goto again;
1392                 }
1393         }
1394
1395         /*
1396          * Now the page range has no ordered extent any more.  Read the page to
1397          * make it uptodate.
1398          */
1399         if (!PageUptodate(page)) {
1400                 btrfs_readpage(NULL, page);
1401                 lock_page(page);
1402                 if (page->mapping != mapping || !PagePrivate(page)) {
1403                         unlock_page(page);
1404                         put_page(page);
1405                         goto again;
1406                 }
1407                 if (!PageUptodate(page)) {
1408                         unlock_page(page);
1409                         put_page(page);
1410                         return ERR_PTR(-EIO);
1411                 }
1412         }
1413         return page;
1414 }
1415
1416 struct defrag_target_range {
1417         struct list_head list;
1418         u64 start;
1419         u64 len;
1420 };
1421
1422 /*
1423  * Collect all valid target extents.
1424  *
1425  * @start:         file offset to lookup
1426  * @len:           length to lookup
1427  * @extent_thresh: file extent size threshold, any extent size >= this value
1428  *                 will be ignored
1429  * @newer_than:    only defrag extents newer than this value
1430  * @do_compress:   whether the defrag is doing compression
1431  *                 if true, @extent_thresh will be ignored and all regular
1432  *                 file extents meeting @newer_than will be targets.
1433  * @locked:        if the range has already held extent lock
1434  * @target_list:   list of targets file extents
1435  */
1436 static int defrag_collect_targets(struct btrfs_inode *inode,
1437                                   u64 start, u64 len, u32 extent_thresh,
1438                                   u64 newer_than, bool do_compress,
1439                                   bool locked, struct list_head *target_list,
1440                                   u64 *last_scanned_ret)
1441 {
1442         bool last_is_target = false;
1443         u64 cur = start;
1444         int ret = 0;
1445
1446         while (cur < start + len) {
1447                 struct extent_map *em;
1448                 struct defrag_target_range *new;
1449                 bool next_mergeable = true;
1450                 u64 range_len;
1451
1452                 last_is_target = false;
1453                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1454                                           newer_than, locked);
1455                 if (!em)
1456                         break;
1457
1458                 /* Skip hole/inline/preallocated extents */
1459                 if (em->block_start >= EXTENT_MAP_LAST_BYTE ||
1460                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1461                         goto next;
1462
1463                 /* Skip older extent */
1464                 if (em->generation < newer_than)
1465                         goto next;
1466
1467                 /* This em is under writeback, no need to defrag */
1468                 if (em->generation == (u64)-1)
1469                         goto next;
1470
1471                 /*
1472                  * Our start offset might be in the middle of an existing extent
1473                  * map, so take that into account.
1474                  */
1475                 range_len = em->len - (cur - em->start);
1476                 /*
1477                  * If this range of the extent map is already flagged for delalloc,
1478                  * skip it, because:
1479                  *
1480                  * 1) We could deadlock later, when trying to reserve space for
1481                  *    delalloc, because in case we can't immediately reserve space
1482                  *    the flusher can start delalloc and wait for the respective
1483                  *    ordered extents to complete. The deadlock would happen
1484                  *    because we do the space reservation while holding the range
1485                  *    locked, and starting writeback, or finishing an ordered
1486                  *    extent, requires locking the range;
1487                  *
1488                  * 2) If there's delalloc there, it means there's dirty pages for
1489                  *    which writeback has not started yet (we clean the delalloc
1490                  *    flag when starting writeback and after creating an ordered
1491                  *    extent). If we mark pages in an adjacent range for defrag,
1492                  *    then we will have a larger contiguous range for delalloc,
1493                  *    very likely resulting in a larger extent after writeback is
1494                  *    triggered (except in a case of free space fragmentation).
1495                  */
1496                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1497                                    EXTENT_DELALLOC, 0, NULL))
1498                         goto next;
1499
1500                 /*
1501                  * For do_compress case, we want to compress all valid file
1502                  * extents, thus no @extent_thresh or mergeable check.
1503                  */
1504                 if (do_compress)
1505                         goto add;
1506
1507                 /* Skip too large extent */
1508                 if (range_len >= extent_thresh)
1509                         goto next;
1510
1511                 /*
1512                  * Skip extents already at its max capacity, this is mostly for
1513                  * compressed extents, which max cap is only 128K.
1514                  */
1515                 if (em->len >= get_extent_max_capacity(em))
1516                         goto next;
1517
1518                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1519                                                 extent_thresh, newer_than, locked);
1520                 if (!next_mergeable) {
1521                         struct defrag_target_range *last;
1522
1523                         /* Empty target list, no way to merge with last entry */
1524                         if (list_empty(target_list))
1525                                 goto next;
1526                         last = list_entry(target_list->prev,
1527                                           struct defrag_target_range, list);
1528                         /* Not mergeable with last entry */
1529                         if (last->start + last->len != cur)
1530                                 goto next;
1531
1532                         /* Mergeable, fall through to add it to @target_list. */
1533                 }
1534
1535 add:
1536                 last_is_target = true;
1537                 range_len = min(extent_map_end(em), start + len) - cur;
1538                 /*
1539                  * This one is a good target, check if it can be merged into
1540                  * last range of the target list.
1541                  */
1542                 if (!list_empty(target_list)) {
1543                         struct defrag_target_range *last;
1544
1545                         last = list_entry(target_list->prev,
1546                                           struct defrag_target_range, list);
1547                         ASSERT(last->start + last->len <= cur);
1548                         if (last->start + last->len == cur) {
1549                                 /* Mergeable, enlarge the last entry */
1550                                 last->len += range_len;
1551                                 goto next;
1552                         }
1553                         /* Fall through to allocate a new entry */
1554                 }
1555
1556                 /* Allocate new defrag_target_range */
1557                 new = kmalloc(sizeof(*new), GFP_NOFS);
1558                 if (!new) {
1559                         free_extent_map(em);
1560                         ret = -ENOMEM;
1561                         break;
1562                 }
1563                 new->start = cur;
1564                 new->len = range_len;
1565                 list_add_tail(&new->list, target_list);
1566
1567 next:
1568                 cur = extent_map_end(em);
1569                 free_extent_map(em);
1570         }
1571         if (ret < 0) {
1572                 struct defrag_target_range *entry;
1573                 struct defrag_target_range *tmp;
1574
1575                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1576                         list_del_init(&entry->list);
1577                         kfree(entry);
1578                 }
1579         }
1580         if (!ret && last_scanned_ret) {
1581                 /*
1582                  * If the last extent is not a target, the caller can skip to
1583                  * the end of that extent.
1584                  * Otherwise, we can only go the end of the specified range.
1585                  */
1586                 if (!last_is_target)
1587                         *last_scanned_ret = max(cur, *last_scanned_ret);
1588                 else
1589                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1590         }
1591         return ret;
1592 }
1593
1594 #define CLUSTER_SIZE    (SZ_256K)
1595 static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1596
1597 /*
1598  * Defrag one contiguous target range.
1599  *
1600  * @inode:      target inode
1601  * @target:     target range to defrag
1602  * @pages:      locked pages covering the defrag range
1603  * @nr_pages:   number of locked pages
1604  *
1605  * Caller should ensure:
1606  *
1607  * - Pages are prepared
1608  *   Pages should be locked, no ordered extent in the pages range,
1609  *   no writeback.
1610  *
1611  * - Extent bits are locked
1612  */
1613 static int defrag_one_locked_target(struct btrfs_inode *inode,
1614                                     struct defrag_target_range *target,
1615                                     struct page **pages, int nr_pages,
1616                                     struct extent_state **cached_state)
1617 {
1618         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1619         struct extent_changeset *data_reserved = NULL;
1620         const u64 start = target->start;
1621         const u64 len = target->len;
1622         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1623         unsigned long start_index = start >> PAGE_SHIFT;
1624         unsigned long first_index = page_index(pages[0]);
1625         int ret = 0;
1626         int i;
1627
1628         ASSERT(last_index - first_index + 1 <= nr_pages);
1629
1630         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1631         if (ret < 0)
1632                 return ret;
1633         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1634                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1635                          EXTENT_DEFRAG, 0, 0, cached_state);
1636         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1637
1638         /* Update the page status */
1639         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1640                 ClearPageChecked(pages[i]);
1641                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1642         }
1643         btrfs_delalloc_release_extents(inode, len);
1644         extent_changeset_free(data_reserved);
1645
1646         return ret;
1647 }
1648
1649 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1650                             u32 extent_thresh, u64 newer_than, bool do_compress,
1651                             u64 *last_scanned_ret)
1652 {
1653         struct extent_state *cached_state = NULL;
1654         struct defrag_target_range *entry;
1655         struct defrag_target_range *tmp;
1656         LIST_HEAD(target_list);
1657         struct page **pages;
1658         const u32 sectorsize = inode->root->fs_info->sectorsize;
1659         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1660         u64 start_index = start >> PAGE_SHIFT;
1661         unsigned int nr_pages = last_index - start_index + 1;
1662         int ret = 0;
1663         int i;
1664
1665         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1666         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1667
1668         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1669         if (!pages)
1670                 return -ENOMEM;
1671
1672         /* Prepare all pages */
1673         for (i = 0; i < nr_pages; i++) {
1674                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1675                 if (IS_ERR(pages[i])) {
1676                         ret = PTR_ERR(pages[i]);
1677                         pages[i] = NULL;
1678                         goto free_pages;
1679                 }
1680         }
1681         for (i = 0; i < nr_pages; i++)
1682                 wait_on_page_writeback(pages[i]);
1683
1684         /* Lock the pages range */
1685         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1686                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1687                          &cached_state);
1688         /*
1689          * Now we have a consistent view about the extent map, re-check
1690          * which range really needs to be defragged.
1691          *
1692          * And this time we have extent locked already, pass @locked = true
1693          * so that we won't relock the extent range and cause deadlock.
1694          */
1695         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1696                                      newer_than, do_compress, true,
1697                                      &target_list, last_scanned_ret);
1698         if (ret < 0)
1699                 goto unlock_extent;
1700
1701         list_for_each_entry(entry, &target_list, list) {
1702                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1703                                                &cached_state);
1704                 if (ret < 0)
1705                         break;
1706         }
1707
1708         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1709                 list_del_init(&entry->list);
1710                 kfree(entry);
1711         }
1712 unlock_extent:
1713         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1714                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1715                              &cached_state);
1716 free_pages:
1717         for (i = 0; i < nr_pages; i++) {
1718                 if (pages[i]) {
1719                         unlock_page(pages[i]);
1720                         put_page(pages[i]);
1721                 }
1722         }
1723         kfree(pages);
1724         return ret;
1725 }
1726
1727 static int defrag_one_cluster(struct btrfs_inode *inode,
1728                               struct file_ra_state *ra,
1729                               u64 start, u32 len, u32 extent_thresh,
1730                               u64 newer_than, bool do_compress,
1731                               unsigned long *sectors_defragged,
1732                               unsigned long max_sectors,
1733                               u64 *last_scanned_ret)
1734 {
1735         const u32 sectorsize = inode->root->fs_info->sectorsize;
1736         struct defrag_target_range *entry;
1737         struct defrag_target_range *tmp;
1738         LIST_HEAD(target_list);
1739         int ret;
1740
1741         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1742                                      newer_than, do_compress, false,
1743                                      &target_list, NULL);
1744         if (ret < 0)
1745                 goto out;
1746
1747         list_for_each_entry(entry, &target_list, list) {
1748                 u32 range_len = entry->len;
1749
1750                 /* Reached or beyond the limit */
1751                 if (max_sectors && *sectors_defragged >= max_sectors) {
1752                         ret = 1;
1753                         break;
1754                 }
1755
1756                 if (max_sectors)
1757                         range_len = min_t(u32, range_len,
1758                                 (max_sectors - *sectors_defragged) * sectorsize);
1759
1760                 /*
1761                  * If defrag_one_range() has updated last_scanned_ret,
1762                  * our range may already be invalid (e.g. hole punched).
1763                  * Skip if our range is before last_scanned_ret, as there is
1764                  * no need to defrag the range anymore.
1765                  */
1766                 if (entry->start + range_len <= *last_scanned_ret)
1767                         continue;
1768
1769                 if (ra)
1770                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1771                                 ra, NULL, entry->start >> PAGE_SHIFT,
1772                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1773                                 (entry->start >> PAGE_SHIFT) + 1);
1774                 /*
1775                  * Here we may not defrag any range if holes are punched before
1776                  * we locked the pages.
1777                  * But that's fine, it only affects the @sectors_defragged
1778                  * accounting.
1779                  */
1780                 ret = defrag_one_range(inode, entry->start, range_len,
1781                                        extent_thresh, newer_than, do_compress,
1782                                        last_scanned_ret);
1783                 if (ret < 0)
1784                         break;
1785                 *sectors_defragged += range_len >>
1786                                       inode->root->fs_info->sectorsize_bits;
1787         }
1788 out:
1789         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1790                 list_del_init(&entry->list);
1791                 kfree(entry);
1792         }
1793         if (ret >= 0)
1794                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1795         return ret;
1796 }
1797
1798 /*
1799  * Entry point to file defragmentation.
1800  *
1801  * @inode:         inode to be defragged
1802  * @ra:            readahead state (can be NUL)
1803  * @range:         defrag options including range and flags
1804  * @newer_than:    minimum transid to defrag
1805  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1806  *                 will be defragged.
1807  *
1808  * Return <0 for error.
1809  * Return >=0 for the number of sectors defragged, and range->start will be updated
1810  * to indicate the file offset where next defrag should be started at.
1811  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1812  *  defragging all the range).
1813  */
1814 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1815                       struct btrfs_ioctl_defrag_range_args *range,
1816                       u64 newer_than, unsigned long max_to_defrag)
1817 {
1818         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1819         unsigned long sectors_defragged = 0;
1820         u64 isize = i_size_read(inode);
1821         u64 cur;
1822         u64 last_byte;
1823         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1824         bool ra_allocated = false;
1825         int compress_type = BTRFS_COMPRESS_ZLIB;
1826         int ret = 0;
1827         u32 extent_thresh = range->extent_thresh;
1828         pgoff_t start_index;
1829
1830         if (isize == 0)
1831                 return 0;
1832
1833         if (range->start >= isize)
1834                 return -EINVAL;
1835
1836         if (do_compress) {
1837                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1838                         return -EINVAL;
1839                 if (range->compress_type)
1840                         compress_type = range->compress_type;
1841         }
1842
1843         if (extent_thresh == 0)
1844                 extent_thresh = SZ_256K;
1845
1846         if (range->start + range->len > range->start) {
1847                 /* Got a specific range */
1848                 last_byte = min(isize, range->start + range->len);
1849         } else {
1850                 /* Defrag until file end */
1851                 last_byte = isize;
1852         }
1853
1854         /* Align the range */
1855         cur = round_down(range->start, fs_info->sectorsize);
1856         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1857
1858         /*
1859          * If we were not given a ra, allocate a readahead context. As
1860          * readahead is just an optimization, defrag will work without it so
1861          * we don't error out.
1862          */
1863         if (!ra) {
1864                 ra_allocated = true;
1865                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1866                 if (ra)
1867                         file_ra_state_init(ra, inode->i_mapping);
1868         }
1869
1870         /*
1871          * Make writeback start from the beginning of the range, so that the
1872          * defrag range can be written sequentially.
1873          */
1874         start_index = cur >> PAGE_SHIFT;
1875         if (start_index < inode->i_mapping->writeback_index)
1876                 inode->i_mapping->writeback_index = start_index;
1877
1878         while (cur < last_byte) {
1879                 const unsigned long prev_sectors_defragged = sectors_defragged;
1880                 u64 last_scanned = cur;
1881                 u64 cluster_end;
1882
1883                 if (btrfs_defrag_cancelled(fs_info)) {
1884                         ret = -EAGAIN;
1885                         break;
1886                 }
1887
1888                 /* We want the cluster end at page boundary when possible */
1889                 cluster_end = (((cur >> PAGE_SHIFT) +
1890                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1891                 cluster_end = min(cluster_end, last_byte);
1892
1893                 btrfs_inode_lock(inode, 0);
1894                 if (IS_SWAPFILE(inode)) {
1895                         ret = -ETXTBSY;
1896                         btrfs_inode_unlock(inode, 0);
1897                         break;
1898                 }
1899                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1900                         btrfs_inode_unlock(inode, 0);
1901                         break;
1902                 }
1903                 if (do_compress)
1904                         BTRFS_I(inode)->defrag_compress = compress_type;
1905                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1906                                 cluster_end + 1 - cur, extent_thresh,
1907                                 newer_than, do_compress, &sectors_defragged,
1908                                 max_to_defrag, &last_scanned);
1909
1910                 if (sectors_defragged > prev_sectors_defragged)
1911                         balance_dirty_pages_ratelimited(inode->i_mapping);
1912
1913                 btrfs_inode_unlock(inode, 0);
1914                 if (ret < 0)
1915                         break;
1916                 cur = max(cluster_end + 1, last_scanned);
1917                 if (ret > 0) {
1918                         ret = 0;
1919                         break;
1920                 }
1921                 cond_resched();
1922         }
1923
1924         if (ra_allocated)
1925                 kfree(ra);
1926         /*
1927          * Update range.start for autodefrag, this will indicate where to start
1928          * in next run.
1929          */
1930         range->start = cur;
1931         if (sectors_defragged) {
1932                 /*
1933                  * We have defragged some sectors, for compression case they
1934                  * need to be written back immediately.
1935                  */
1936                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1937                         filemap_flush(inode->i_mapping);
1938                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1939                                      &BTRFS_I(inode)->runtime_flags))
1940                                 filemap_flush(inode->i_mapping);
1941                 }
1942                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1943                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1944                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1945                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1946                 ret = sectors_defragged;
1947         }
1948         if (do_compress) {
1949                 btrfs_inode_lock(inode, 0);
1950                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1951                 btrfs_inode_unlock(inode, 0);
1952         }
1953         return ret;
1954 }
1955
1956 /*
1957  * Try to start exclusive operation @type or cancel it if it's running.
1958  *
1959  * Return:
1960  *   0        - normal mode, newly claimed op started
1961  *  >0        - normal mode, something else is running,
1962  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1963  * ECANCELED  - cancel mode, successful cancel
1964  * ENOTCONN   - cancel mode, operation not running anymore
1965  */
1966 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1967                         enum btrfs_exclusive_operation type, bool cancel)
1968 {
1969         if (!cancel) {
1970                 /* Start normal op */
1971                 if (!btrfs_exclop_start(fs_info, type))
1972                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1973                 /* Exclusive operation is now claimed */
1974                 return 0;
1975         }
1976
1977         /* Cancel running op */
1978         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1979                 /*
1980                  * This blocks any exclop finish from setting it to NONE, so we
1981                  * request cancellation. Either it runs and we will wait for it,
1982                  * or it has finished and no waiting will happen.
1983                  */
1984                 atomic_inc(&fs_info->reloc_cancel_req);
1985                 btrfs_exclop_start_unlock(fs_info);
1986
1987                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1988                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1989                                     TASK_INTERRUPTIBLE);
1990
1991                 return -ECANCELED;
1992         }
1993
1994         /* Something else is running or none */
1995         return -ENOTCONN;
1996 }
1997
1998 static noinline int btrfs_ioctl_resize(struct file *file,
1999                                         void __user *arg)
2000 {
2001         BTRFS_DEV_LOOKUP_ARGS(args);
2002         struct inode *inode = file_inode(file);
2003         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2004         u64 new_size;
2005         u64 old_size;
2006         u64 devid = 1;
2007         struct btrfs_root *root = BTRFS_I(inode)->root;
2008         struct btrfs_ioctl_vol_args *vol_args;
2009         struct btrfs_trans_handle *trans;
2010         struct btrfs_device *device = NULL;
2011         char *sizestr;
2012         char *retptr;
2013         char *devstr = NULL;
2014         int ret = 0;
2015         int mod = 0;
2016         bool cancel;
2017
2018         if (!capable(CAP_SYS_ADMIN))
2019                 return -EPERM;
2020
2021         ret = mnt_want_write_file(file);
2022         if (ret)
2023                 return ret;
2024
2025         /*
2026          * Read the arguments before checking exclusivity to be able to
2027          * distinguish regular resize and cancel
2028          */
2029         vol_args = memdup_user(arg, sizeof(*vol_args));
2030         if (IS_ERR(vol_args)) {
2031                 ret = PTR_ERR(vol_args);
2032                 goto out_drop;
2033         }
2034         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2035         sizestr = vol_args->name;
2036         cancel = (strcmp("cancel", sizestr) == 0);
2037         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
2038         if (ret)
2039                 goto out_free;
2040         /* Exclusive operation is now claimed */
2041
2042         devstr = strchr(sizestr, ':');
2043         if (devstr) {
2044                 sizestr = devstr + 1;
2045                 *devstr = '\0';
2046                 devstr = vol_args->name;
2047                 ret = kstrtoull(devstr, 10, &devid);
2048                 if (ret)
2049                         goto out_finish;
2050                 if (!devid) {
2051                         ret = -EINVAL;
2052                         goto out_finish;
2053                 }
2054                 btrfs_info(fs_info, "resizing devid %llu", devid);
2055         }
2056
2057         args.devid = devid;
2058         device = btrfs_find_device(fs_info->fs_devices, &args);
2059         if (!device) {
2060                 btrfs_info(fs_info, "resizer unable to find device %llu",
2061                            devid);
2062                 ret = -ENODEV;
2063                 goto out_finish;
2064         }
2065
2066         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2067                 btrfs_info(fs_info,
2068                            "resizer unable to apply on readonly device %llu",
2069                        devid);
2070                 ret = -EPERM;
2071                 goto out_finish;
2072         }
2073
2074         if (!strcmp(sizestr, "max"))
2075                 new_size = bdev_nr_bytes(device->bdev);
2076         else {
2077                 if (sizestr[0] == '-') {
2078                         mod = -1;
2079                         sizestr++;
2080                 } else if (sizestr[0] == '+') {
2081                         mod = 1;
2082                         sizestr++;
2083                 }
2084                 new_size = memparse(sizestr, &retptr);
2085                 if (*retptr != '\0' || new_size == 0) {
2086                         ret = -EINVAL;
2087                         goto out_finish;
2088                 }
2089         }
2090
2091         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2092                 ret = -EPERM;
2093                 goto out_finish;
2094         }
2095
2096         old_size = btrfs_device_get_total_bytes(device);
2097
2098         if (mod < 0) {
2099                 if (new_size > old_size) {
2100                         ret = -EINVAL;
2101                         goto out_finish;
2102                 }
2103                 new_size = old_size - new_size;
2104         } else if (mod > 0) {
2105                 if (new_size > ULLONG_MAX - old_size) {
2106                         ret = -ERANGE;
2107                         goto out_finish;
2108                 }
2109                 new_size = old_size + new_size;
2110         }
2111
2112         if (new_size < SZ_256M) {
2113                 ret = -EINVAL;
2114                 goto out_finish;
2115         }
2116         if (new_size > bdev_nr_bytes(device->bdev)) {
2117                 ret = -EFBIG;
2118                 goto out_finish;
2119         }
2120
2121         new_size = round_down(new_size, fs_info->sectorsize);
2122
2123         if (new_size > old_size) {
2124                 trans = btrfs_start_transaction(root, 0);
2125                 if (IS_ERR(trans)) {
2126                         ret = PTR_ERR(trans);
2127                         goto out_finish;
2128                 }
2129                 ret = btrfs_grow_device(trans, device, new_size);
2130                 btrfs_commit_transaction(trans);
2131         } else if (new_size < old_size) {
2132                 ret = btrfs_shrink_device(device, new_size);
2133         } /* equal, nothing need to do */
2134
2135         if (ret == 0 && new_size != old_size)
2136                 btrfs_info_in_rcu(fs_info,
2137                         "resize device %s (devid %llu) from %llu to %llu",
2138                         rcu_str_deref(device->name), device->devid,
2139                         old_size, new_size);
2140 out_finish:
2141         btrfs_exclop_finish(fs_info);
2142 out_free:
2143         kfree(vol_args);
2144 out_drop:
2145         mnt_drop_write_file(file);
2146         return ret;
2147 }
2148
2149 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2150                                 struct user_namespace *mnt_userns,
2151                                 const char *name, unsigned long fd, int subvol,
2152                                 bool readonly,
2153                                 struct btrfs_qgroup_inherit *inherit)
2154 {
2155         int namelen;
2156         int ret = 0;
2157
2158         if (!S_ISDIR(file_inode(file)->i_mode))
2159                 return -ENOTDIR;
2160
2161         ret = mnt_want_write_file(file);
2162         if (ret)
2163                 goto out;
2164
2165         namelen = strlen(name);
2166         if (strchr(name, '/')) {
2167                 ret = -EINVAL;
2168                 goto out_drop_write;
2169         }
2170
2171         if (name[0] == '.' &&
2172            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2173                 ret = -EEXIST;
2174                 goto out_drop_write;
2175         }
2176
2177         if (subvol) {
2178                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2179                                      namelen, NULL, readonly, inherit);
2180         } else {
2181                 struct fd src = fdget(fd);
2182                 struct inode *src_inode;
2183                 if (!src.file) {
2184                         ret = -EINVAL;
2185                         goto out_drop_write;
2186                 }
2187
2188                 src_inode = file_inode(src.file);
2189                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2190                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2191                                    "Snapshot src from another FS");
2192                         ret = -EXDEV;
2193                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2194                         /*
2195                          * Subvolume creation is not restricted, but snapshots
2196                          * are limited to own subvolumes only
2197                          */
2198                         ret = -EPERM;
2199                 } else {
2200                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2201                                                name, namelen,
2202                                                BTRFS_I(src_inode)->root,
2203                                                readonly, inherit);
2204                 }
2205                 fdput(src);
2206         }
2207 out_drop_write:
2208         mnt_drop_write_file(file);
2209 out:
2210         return ret;
2211 }
2212
2213 static noinline int btrfs_ioctl_snap_create(struct file *file,
2214                                             void __user *arg, int subvol)
2215 {
2216         struct btrfs_ioctl_vol_args *vol_args;
2217         int ret;
2218
2219         if (!S_ISDIR(file_inode(file)->i_mode))
2220                 return -ENOTDIR;
2221
2222         vol_args = memdup_user(arg, sizeof(*vol_args));
2223         if (IS_ERR(vol_args))
2224                 return PTR_ERR(vol_args);
2225         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2226
2227         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2228                                         vol_args->name, vol_args->fd, subvol,
2229                                         false, NULL);
2230
2231         kfree(vol_args);
2232         return ret;
2233 }
2234
2235 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2236                                                void __user *arg, int subvol)
2237 {
2238         struct btrfs_ioctl_vol_args_v2 *vol_args;
2239         int ret;
2240         bool readonly = false;
2241         struct btrfs_qgroup_inherit *inherit = NULL;
2242
2243         if (!S_ISDIR(file_inode(file)->i_mode))
2244                 return -ENOTDIR;
2245
2246         vol_args = memdup_user(arg, sizeof(*vol_args));
2247         if (IS_ERR(vol_args))
2248                 return PTR_ERR(vol_args);
2249         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2250
2251         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2252                 ret = -EOPNOTSUPP;
2253                 goto free_args;
2254         }
2255
2256         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2257                 readonly = true;
2258         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2259                 u64 nums;
2260
2261                 if (vol_args->size < sizeof(*inherit) ||
2262                     vol_args->size > PAGE_SIZE) {
2263                         ret = -EINVAL;
2264                         goto free_args;
2265                 }
2266                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2267                 if (IS_ERR(inherit)) {
2268                         ret = PTR_ERR(inherit);
2269                         goto free_args;
2270                 }
2271
2272                 if (inherit->num_qgroups > PAGE_SIZE ||
2273                     inherit->num_ref_copies > PAGE_SIZE ||
2274                     inherit->num_excl_copies > PAGE_SIZE) {
2275                         ret = -EINVAL;
2276                         goto free_inherit;
2277                 }
2278
2279                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2280                        2 * inherit->num_excl_copies;
2281                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2282                         ret = -EINVAL;
2283                         goto free_inherit;
2284                 }
2285         }
2286
2287         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2288                                         vol_args->name, vol_args->fd, subvol,
2289                                         readonly, inherit);
2290         if (ret)
2291                 goto free_inherit;
2292 free_inherit:
2293         kfree(inherit);
2294 free_args:
2295         kfree(vol_args);
2296         return ret;
2297 }
2298
2299 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
2300                                                 void __user *arg)
2301 {
2302         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2303         struct btrfs_root *root = BTRFS_I(inode)->root;
2304         int ret = 0;
2305         u64 flags = 0;
2306
2307         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2308                 return -EINVAL;
2309
2310         down_read(&fs_info->subvol_sem);
2311         if (btrfs_root_readonly(root))
2312                 flags |= BTRFS_SUBVOL_RDONLY;
2313         up_read(&fs_info->subvol_sem);
2314
2315         if (copy_to_user(arg, &flags, sizeof(flags)))
2316                 ret = -EFAULT;
2317
2318         return ret;
2319 }
2320
2321 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2322                                               void __user *arg)
2323 {
2324         struct inode *inode = file_inode(file);
2325         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2326         struct btrfs_root *root = BTRFS_I(inode)->root;
2327         struct btrfs_trans_handle *trans;
2328         u64 root_flags;
2329         u64 flags;
2330         int ret = 0;
2331
2332         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2333                 return -EPERM;
2334
2335         ret = mnt_want_write_file(file);
2336         if (ret)
2337                 goto out;
2338
2339         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2340                 ret = -EINVAL;
2341                 goto out_drop_write;
2342         }
2343
2344         if (copy_from_user(&flags, arg, sizeof(flags))) {
2345                 ret = -EFAULT;
2346                 goto out_drop_write;
2347         }
2348
2349         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2350                 ret = -EOPNOTSUPP;
2351                 goto out_drop_write;
2352         }
2353
2354         down_write(&fs_info->subvol_sem);
2355
2356         /* nothing to do */
2357         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2358                 goto out_drop_sem;
2359
2360         root_flags = btrfs_root_flags(&root->root_item);
2361         if (flags & BTRFS_SUBVOL_RDONLY) {
2362                 btrfs_set_root_flags(&root->root_item,
2363                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2364         } else {
2365                 /*
2366                  * Block RO -> RW transition if this subvolume is involved in
2367                  * send
2368                  */
2369                 spin_lock(&root->root_item_lock);
2370                 if (root->send_in_progress == 0) {
2371                         btrfs_set_root_flags(&root->root_item,
2372                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2373                         spin_unlock(&root->root_item_lock);
2374                 } else {
2375                         spin_unlock(&root->root_item_lock);
2376                         btrfs_warn(fs_info,
2377                                    "Attempt to set subvolume %llu read-write during send",
2378                                    root->root_key.objectid);
2379                         ret = -EPERM;
2380                         goto out_drop_sem;
2381                 }
2382         }
2383
2384         trans = btrfs_start_transaction(root, 1);
2385         if (IS_ERR(trans)) {
2386                 ret = PTR_ERR(trans);
2387                 goto out_reset;
2388         }
2389
2390         ret = btrfs_update_root(trans, fs_info->tree_root,
2391                                 &root->root_key, &root->root_item);
2392         if (ret < 0) {
2393                 btrfs_end_transaction(trans);
2394                 goto out_reset;
2395         }
2396
2397         ret = btrfs_commit_transaction(trans);
2398
2399 out_reset:
2400         if (ret)
2401                 btrfs_set_root_flags(&root->root_item, root_flags);
2402 out_drop_sem:
2403         up_write(&fs_info->subvol_sem);
2404 out_drop_write:
2405         mnt_drop_write_file(file);
2406 out:
2407         return ret;
2408 }
2409
2410 static noinline int key_in_sk(struct btrfs_key *key,
2411                               struct btrfs_ioctl_search_key *sk)
2412 {
2413         struct btrfs_key test;
2414         int ret;
2415
2416         test.objectid = sk->min_objectid;
2417         test.type = sk->min_type;
2418         test.offset = sk->min_offset;
2419
2420         ret = btrfs_comp_cpu_keys(key, &test);
2421         if (ret < 0)
2422                 return 0;
2423
2424         test.objectid = sk->max_objectid;
2425         test.type = sk->max_type;
2426         test.offset = sk->max_offset;
2427
2428         ret = btrfs_comp_cpu_keys(key, &test);
2429         if (ret > 0)
2430                 return 0;
2431         return 1;
2432 }
2433
2434 static noinline int copy_to_sk(struct btrfs_path *path,
2435                                struct btrfs_key *key,
2436                                struct btrfs_ioctl_search_key *sk,
2437                                size_t *buf_size,
2438                                char __user *ubuf,
2439                                unsigned long *sk_offset,
2440                                int *num_found)
2441 {
2442         u64 found_transid;
2443         struct extent_buffer *leaf;
2444         struct btrfs_ioctl_search_header sh;
2445         struct btrfs_key test;
2446         unsigned long item_off;
2447         unsigned long item_len;
2448         int nritems;
2449         int i;
2450         int slot;
2451         int ret = 0;
2452
2453         leaf = path->nodes[0];
2454         slot = path->slots[0];
2455         nritems = btrfs_header_nritems(leaf);
2456
2457         if (btrfs_header_generation(leaf) > sk->max_transid) {
2458                 i = nritems;
2459                 goto advance_key;
2460         }
2461         found_transid = btrfs_header_generation(leaf);
2462
2463         for (i = slot; i < nritems; i++) {
2464                 item_off = btrfs_item_ptr_offset(leaf, i);
2465                 item_len = btrfs_item_size(leaf, i);
2466
2467                 btrfs_item_key_to_cpu(leaf, key, i);
2468                 if (!key_in_sk(key, sk))
2469                         continue;
2470
2471                 if (sizeof(sh) + item_len > *buf_size) {
2472                         if (*num_found) {
2473                                 ret = 1;
2474                                 goto out;
2475                         }
2476
2477                         /*
2478                          * return one empty item back for v1, which does not
2479                          * handle -EOVERFLOW
2480                          */
2481
2482                         *buf_size = sizeof(sh) + item_len;
2483                         item_len = 0;
2484                         ret = -EOVERFLOW;
2485                 }
2486
2487                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2488                         ret = 1;
2489                         goto out;
2490                 }
2491
2492                 sh.objectid = key->objectid;
2493                 sh.offset = key->offset;
2494                 sh.type = key->type;
2495                 sh.len = item_len;
2496                 sh.transid = found_transid;
2497
2498                 /*
2499                  * Copy search result header. If we fault then loop again so we
2500                  * can fault in the pages and -EFAULT there if there's a
2501                  * problem. Otherwise we'll fault and then copy the buffer in
2502                  * properly this next time through
2503                  */
2504                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2505                         ret = 0;
2506                         goto out;
2507                 }
2508
2509                 *sk_offset += sizeof(sh);
2510
2511                 if (item_len) {
2512                         char __user *up = ubuf + *sk_offset;
2513                         /*
2514                          * Copy the item, same behavior as above, but reset the
2515                          * * sk_offset so we copy the full thing again.
2516                          */
2517                         if (read_extent_buffer_to_user_nofault(leaf, up,
2518                                                 item_off, item_len)) {
2519                                 ret = 0;
2520                                 *sk_offset -= sizeof(sh);
2521                                 goto out;
2522                         }
2523
2524                         *sk_offset += item_len;
2525                 }
2526                 (*num_found)++;
2527
2528                 if (ret) /* -EOVERFLOW from above */
2529                         goto out;
2530
2531                 if (*num_found >= sk->nr_items) {
2532                         ret = 1;
2533                         goto out;
2534                 }
2535         }
2536 advance_key:
2537         ret = 0;
2538         test.objectid = sk->max_objectid;
2539         test.type = sk->max_type;
2540         test.offset = sk->max_offset;
2541         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2542                 ret = 1;
2543         else if (key->offset < (u64)-1)
2544                 key->offset++;
2545         else if (key->type < (u8)-1) {
2546                 key->offset = 0;
2547                 key->type++;
2548         } else if (key->objectid < (u64)-1) {
2549                 key->offset = 0;
2550                 key->type = 0;
2551                 key->objectid++;
2552         } else
2553                 ret = 1;
2554 out:
2555         /*
2556          *  0: all items from this leaf copied, continue with next
2557          *  1: * more items can be copied, but unused buffer is too small
2558          *     * all items were found
2559          *     Either way, it will stops the loop which iterates to the next
2560          *     leaf
2561          *  -EOVERFLOW: item was to large for buffer
2562          *  -EFAULT: could not copy extent buffer back to userspace
2563          */
2564         return ret;
2565 }
2566
2567 static noinline int search_ioctl(struct inode *inode,
2568                                  struct btrfs_ioctl_search_key *sk,
2569                                  size_t *buf_size,
2570                                  char __user *ubuf)
2571 {
2572         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2573         struct btrfs_root *root;
2574         struct btrfs_key key;
2575         struct btrfs_path *path;
2576         int ret;
2577         int num_found = 0;
2578         unsigned long sk_offset = 0;
2579
2580         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2581                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2582                 return -EOVERFLOW;
2583         }
2584
2585         path = btrfs_alloc_path();
2586         if (!path)
2587                 return -ENOMEM;
2588
2589         if (sk->tree_id == 0) {
2590                 /* search the root of the inode that was passed */
2591                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2592         } else {
2593                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2594                 if (IS_ERR(root)) {
2595                         btrfs_free_path(path);
2596                         return PTR_ERR(root);
2597                 }
2598         }
2599
2600         key.objectid = sk->min_objectid;
2601         key.type = sk->min_type;
2602         key.offset = sk->min_offset;
2603
2604         while (1) {
2605                 ret = -EFAULT;
2606                 if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
2607                         break;
2608
2609                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2610                 if (ret != 0) {
2611                         if (ret > 0)
2612                                 ret = 0;
2613                         goto err;
2614                 }
2615                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2616                                  &sk_offset, &num_found);
2617                 btrfs_release_path(path);
2618                 if (ret)
2619                         break;
2620
2621         }
2622         if (ret > 0)
2623                 ret = 0;
2624 err:
2625         sk->nr_items = num_found;
2626         btrfs_put_root(root);
2627         btrfs_free_path(path);
2628         return ret;
2629 }
2630
2631 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
2632                                             void __user *argp)
2633 {
2634         struct btrfs_ioctl_search_args __user *uargs;
2635         struct btrfs_ioctl_search_key sk;
2636         int ret;
2637         size_t buf_size;
2638
2639         if (!capable(CAP_SYS_ADMIN))
2640                 return -EPERM;
2641
2642         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2643
2644         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2645                 return -EFAULT;
2646
2647         buf_size = sizeof(uargs->buf);
2648
2649         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2650
2651         /*
2652          * In the origin implementation an overflow is handled by returning a
2653          * search header with a len of zero, so reset ret.
2654          */
2655         if (ret == -EOVERFLOW)
2656                 ret = 0;
2657
2658         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2659                 ret = -EFAULT;
2660         return ret;
2661 }
2662
2663 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
2664                                                void __user *argp)
2665 {
2666         struct btrfs_ioctl_search_args_v2 __user *uarg;
2667         struct btrfs_ioctl_search_args_v2 args;
2668         int ret;
2669         size_t buf_size;
2670         const size_t buf_limit = SZ_16M;
2671
2672         if (!capable(CAP_SYS_ADMIN))
2673                 return -EPERM;
2674
2675         /* copy search header and buffer size */
2676         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2677         if (copy_from_user(&args, uarg, sizeof(args)))
2678                 return -EFAULT;
2679
2680         buf_size = args.buf_size;
2681
2682         /* limit result size to 16MB */
2683         if (buf_size > buf_limit)
2684                 buf_size = buf_limit;
2685
2686         ret = search_ioctl(inode, &args.key, &buf_size,
2687                            (char __user *)(&uarg->buf[0]));
2688         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2689                 ret = -EFAULT;
2690         else if (ret == -EOVERFLOW &&
2691                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2692                 ret = -EFAULT;
2693
2694         return ret;
2695 }
2696
2697 /*
2698  * Search INODE_REFs to identify path name of 'dirid' directory
2699  * in a 'tree_id' tree. and sets path name to 'name'.
2700  */
2701 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2702                                 u64 tree_id, u64 dirid, char *name)
2703 {
2704         struct btrfs_root *root;
2705         struct btrfs_key key;
2706         char *ptr;
2707         int ret = -1;
2708         int slot;
2709         int len;
2710         int total_len = 0;
2711         struct btrfs_inode_ref *iref;
2712         struct extent_buffer *l;
2713         struct btrfs_path *path;
2714
2715         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2716                 name[0]='\0';
2717                 return 0;
2718         }
2719
2720         path = btrfs_alloc_path();
2721         if (!path)
2722                 return -ENOMEM;
2723
2724         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2725
2726         root = btrfs_get_fs_root(info, tree_id, true);
2727         if (IS_ERR(root)) {
2728                 ret = PTR_ERR(root);
2729                 root = NULL;
2730                 goto out;
2731         }
2732
2733         key.objectid = dirid;
2734         key.type = BTRFS_INODE_REF_KEY;
2735         key.offset = (u64)-1;
2736
2737         while (1) {
2738                 ret = btrfs_search_backwards(root, &key, path);
2739                 if (ret < 0)
2740                         goto out;
2741                 else if (ret > 0) {
2742                         ret = -ENOENT;
2743                         goto out;
2744                 }
2745
2746                 l = path->nodes[0];
2747                 slot = path->slots[0];
2748
2749                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2750                 len = btrfs_inode_ref_name_len(l, iref);
2751                 ptr -= len + 1;
2752                 total_len += len + 1;
2753                 if (ptr < name) {
2754                         ret = -ENAMETOOLONG;
2755                         goto out;
2756                 }
2757
2758                 *(ptr + len) = '/';
2759                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2760
2761                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2762                         break;
2763
2764                 btrfs_release_path(path);
2765                 key.objectid = key.offset;
2766                 key.offset = (u64)-1;
2767                 dirid = key.objectid;
2768         }
2769         memmove(name, ptr, total_len);
2770         name[total_len] = '\0';
2771         ret = 0;
2772 out:
2773         btrfs_put_root(root);
2774         btrfs_free_path(path);
2775         return ret;
2776 }
2777
2778 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2779                                 struct inode *inode,
2780                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2781 {
2782         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2783         struct super_block *sb = inode->i_sb;
2784         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2785         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2786         u64 dirid = args->dirid;
2787         unsigned long item_off;
2788         unsigned long item_len;
2789         struct btrfs_inode_ref *iref;
2790         struct btrfs_root_ref *rref;
2791         struct btrfs_root *root = NULL;
2792         struct btrfs_path *path;
2793         struct btrfs_key key, key2;
2794         struct extent_buffer *leaf;
2795         struct inode *temp_inode;
2796         char *ptr;
2797         int slot;
2798         int len;
2799         int total_len = 0;
2800         int ret;
2801
2802         path = btrfs_alloc_path();
2803         if (!path)
2804                 return -ENOMEM;
2805
2806         /*
2807          * If the bottom subvolume does not exist directly under upper_limit,
2808          * construct the path in from the bottom up.
2809          */
2810         if (dirid != upper_limit.objectid) {
2811                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2812
2813                 root = btrfs_get_fs_root(fs_info, treeid, true);
2814                 if (IS_ERR(root)) {
2815                         ret = PTR_ERR(root);
2816                         goto out;
2817                 }
2818
2819                 key.objectid = dirid;
2820                 key.type = BTRFS_INODE_REF_KEY;
2821                 key.offset = (u64)-1;
2822                 while (1) {
2823                         ret = btrfs_search_backwards(root, &key, path);
2824                         if (ret < 0)
2825                                 goto out_put;
2826                         else if (ret > 0) {
2827                                 ret = -ENOENT;
2828                                 goto out_put;
2829                         }
2830
2831                         leaf = path->nodes[0];
2832                         slot = path->slots[0];
2833
2834                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2835                         len = btrfs_inode_ref_name_len(leaf, iref);
2836                         ptr -= len + 1;
2837                         total_len += len + 1;
2838                         if (ptr < args->path) {
2839                                 ret = -ENAMETOOLONG;
2840                                 goto out_put;
2841                         }
2842
2843                         *(ptr + len) = '/';
2844                         read_extent_buffer(leaf, ptr,
2845                                         (unsigned long)(iref + 1), len);
2846
2847                         /* Check the read+exec permission of this directory */
2848                         ret = btrfs_previous_item(root, path, dirid,
2849                                                   BTRFS_INODE_ITEM_KEY);
2850                         if (ret < 0) {
2851                                 goto out_put;
2852                         } else if (ret > 0) {
2853                                 ret = -ENOENT;
2854                                 goto out_put;
2855                         }
2856
2857                         leaf = path->nodes[0];
2858                         slot = path->slots[0];
2859                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2860                         if (key2.objectid != dirid) {
2861                                 ret = -ENOENT;
2862                                 goto out_put;
2863                         }
2864
2865                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2866                         if (IS_ERR(temp_inode)) {
2867                                 ret = PTR_ERR(temp_inode);
2868                                 goto out_put;
2869                         }
2870                         ret = inode_permission(mnt_userns, temp_inode,
2871                                                MAY_READ | MAY_EXEC);
2872                         iput(temp_inode);
2873                         if (ret) {
2874                                 ret = -EACCES;
2875                                 goto out_put;
2876                         }
2877
2878                         if (key.offset == upper_limit.objectid)
2879                                 break;
2880                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2881                                 ret = -EACCES;
2882                                 goto out_put;
2883                         }
2884
2885                         btrfs_release_path(path);
2886                         key.objectid = key.offset;
2887                         key.offset = (u64)-1;
2888                         dirid = key.objectid;
2889                 }
2890
2891                 memmove(args->path, ptr, total_len);
2892                 args->path[total_len] = '\0';
2893                 btrfs_put_root(root);
2894                 root = NULL;
2895                 btrfs_release_path(path);
2896         }
2897
2898         /* Get the bottom subvolume's name from ROOT_REF */
2899         key.objectid = treeid;
2900         key.type = BTRFS_ROOT_REF_KEY;
2901         key.offset = args->treeid;
2902         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2903         if (ret < 0) {
2904                 goto out;
2905         } else if (ret > 0) {
2906                 ret = -ENOENT;
2907                 goto out;
2908         }
2909
2910         leaf = path->nodes[0];
2911         slot = path->slots[0];
2912         btrfs_item_key_to_cpu(leaf, &key, slot);
2913
2914         item_off = btrfs_item_ptr_offset(leaf, slot);
2915         item_len = btrfs_item_size(leaf, slot);
2916         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2917         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2918         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2919                 ret = -EINVAL;
2920                 goto out;
2921         }
2922
2923         /* Copy subvolume's name */
2924         item_off += sizeof(struct btrfs_root_ref);
2925         item_len -= sizeof(struct btrfs_root_ref);
2926         read_extent_buffer(leaf, args->name, item_off, item_len);
2927         args->name[item_len] = 0;
2928
2929 out_put:
2930         btrfs_put_root(root);
2931 out:
2932         btrfs_free_path(path);
2933         return ret;
2934 }
2935
2936 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2937                                            void __user *argp)
2938 {
2939         struct btrfs_ioctl_ino_lookup_args *args;
2940         int ret = 0;
2941
2942         args = memdup_user(argp, sizeof(*args));
2943         if (IS_ERR(args))
2944                 return PTR_ERR(args);
2945
2946         /*
2947          * Unprivileged query to obtain the containing subvolume root id. The
2948          * path is reset so it's consistent with btrfs_search_path_in_tree.
2949          */
2950         if (args->treeid == 0)
2951                 args->treeid = root->root_key.objectid;
2952
2953         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2954                 args->name[0] = 0;
2955                 goto out;
2956         }
2957
2958         if (!capable(CAP_SYS_ADMIN)) {
2959                 ret = -EPERM;
2960                 goto out;
2961         }
2962
2963         ret = btrfs_search_path_in_tree(root->fs_info,
2964                                         args->treeid, args->objectid,
2965                                         args->name);
2966
2967 out:
2968         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2969                 ret = -EFAULT;
2970
2971         kfree(args);
2972         return ret;
2973 }
2974
2975 /*
2976  * Version of ino_lookup ioctl (unprivileged)
2977  *
2978  * The main differences from ino_lookup ioctl are:
2979  *
2980  *   1. Read + Exec permission will be checked using inode_permission() during
2981  *      path construction. -EACCES will be returned in case of failure.
2982  *   2. Path construction will be stopped at the inode number which corresponds
2983  *      to the fd with which this ioctl is called. If constructed path does not
2984  *      exist under fd's inode, -EACCES will be returned.
2985  *   3. The name of bottom subvolume is also searched and filled.
2986  */
2987 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2988 {
2989         struct btrfs_ioctl_ino_lookup_user_args *args;
2990         struct inode *inode;
2991         int ret;
2992
2993         args = memdup_user(argp, sizeof(*args));
2994         if (IS_ERR(args))
2995                 return PTR_ERR(args);
2996
2997         inode = file_inode(file);
2998
2999         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
3000             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
3001                 /*
3002                  * The subvolume does not exist under fd with which this is
3003                  * called
3004                  */
3005                 kfree(args);
3006                 return -EACCES;
3007         }
3008
3009         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
3010
3011         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
3012                 ret = -EFAULT;
3013
3014         kfree(args);
3015         return ret;
3016 }
3017
3018 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
3019 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
3020 {
3021         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
3022         struct btrfs_fs_info *fs_info;
3023         struct btrfs_root *root;
3024         struct btrfs_path *path;
3025         struct btrfs_key key;
3026         struct btrfs_root_item *root_item;
3027         struct btrfs_root_ref *rref;
3028         struct extent_buffer *leaf;
3029         unsigned long item_off;
3030         unsigned long item_len;
3031         int slot;
3032         int ret = 0;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
3039         if (!subvol_info) {
3040                 btrfs_free_path(path);
3041                 return -ENOMEM;
3042         }
3043
3044         fs_info = BTRFS_I(inode)->root->fs_info;
3045
3046         /* Get root_item of inode's subvolume */
3047         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
3048         root = btrfs_get_fs_root(fs_info, key.objectid, true);
3049         if (IS_ERR(root)) {
3050                 ret = PTR_ERR(root);
3051                 goto out_free;
3052         }
3053         root_item = &root->root_item;
3054
3055         subvol_info->treeid = key.objectid;
3056
3057         subvol_info->generation = btrfs_root_generation(root_item);
3058         subvol_info->flags = btrfs_root_flags(root_item);
3059
3060         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3061         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3062                                                     BTRFS_UUID_SIZE);
3063         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3064                                                     BTRFS_UUID_SIZE);
3065
3066         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3067         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3068         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3069
3070         subvol_info->otransid = btrfs_root_otransid(root_item);
3071         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3072         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3073
3074         subvol_info->stransid = btrfs_root_stransid(root_item);
3075         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3076         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3077
3078         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3079         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3080         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3081
3082         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3083                 /* Search root tree for ROOT_BACKREF of this subvolume */
3084                 key.type = BTRFS_ROOT_BACKREF_KEY;
3085                 key.offset = 0;
3086                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3087                 if (ret < 0) {
3088                         goto out;
3089                 } else if (path->slots[0] >=
3090                            btrfs_header_nritems(path->nodes[0])) {
3091                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3092                         if (ret < 0) {
3093                                 goto out;
3094                         } else if (ret > 0) {
3095                                 ret = -EUCLEAN;
3096                                 goto out;
3097                         }
3098                 }
3099
3100                 leaf = path->nodes[0];
3101                 slot = path->slots[0];
3102                 btrfs_item_key_to_cpu(leaf, &key, slot);
3103                 if (key.objectid == subvol_info->treeid &&
3104                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3105                         subvol_info->parent_id = key.offset;
3106
3107                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3108                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3109
3110                         item_off = btrfs_item_ptr_offset(leaf, slot)
3111                                         + sizeof(struct btrfs_root_ref);
3112                         item_len = btrfs_item_size(leaf, slot)
3113                                         - sizeof(struct btrfs_root_ref);
3114                         read_extent_buffer(leaf, subvol_info->name,
3115                                            item_off, item_len);
3116                 } else {
3117                         ret = -ENOENT;
3118                         goto out;
3119                 }
3120         }
3121
3122         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3123                 ret = -EFAULT;
3124
3125 out:
3126         btrfs_put_root(root);
3127 out_free:
3128         btrfs_free_path(path);
3129         kfree(subvol_info);
3130         return ret;
3131 }
3132
3133 /*
3134  * Return ROOT_REF information of the subvolume containing this inode
3135  * except the subvolume name.
3136  */
3137 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3138                                           void __user *argp)
3139 {
3140         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3141         struct btrfs_root_ref *rref;
3142         struct btrfs_path *path;
3143         struct btrfs_key key;
3144         struct extent_buffer *leaf;
3145         u64 objectid;
3146         int slot;
3147         int ret;
3148         u8 found;
3149
3150         path = btrfs_alloc_path();
3151         if (!path)
3152                 return -ENOMEM;
3153
3154         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3155         if (IS_ERR(rootrefs)) {
3156                 btrfs_free_path(path);
3157                 return PTR_ERR(rootrefs);
3158         }
3159
3160         objectid = root->root_key.objectid;
3161         key.objectid = objectid;
3162         key.type = BTRFS_ROOT_REF_KEY;
3163         key.offset = rootrefs->min_treeid;
3164         found = 0;
3165
3166         root = root->fs_info->tree_root;
3167         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3168         if (ret < 0) {
3169                 goto out;
3170         } else if (path->slots[0] >=
3171                    btrfs_header_nritems(path->nodes[0])) {
3172                 ret = btrfs_next_leaf(root, path);
3173                 if (ret < 0) {
3174                         goto out;
3175                 } else if (ret > 0) {
3176                         ret = -EUCLEAN;
3177                         goto out;
3178                 }
3179         }
3180         while (1) {
3181                 leaf = path->nodes[0];
3182                 slot = path->slots[0];
3183
3184                 btrfs_item_key_to_cpu(leaf, &key, slot);
3185                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3186                         ret = 0;
3187                         goto out;
3188                 }
3189
3190                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3191                         ret = -EOVERFLOW;
3192                         goto out;
3193                 }
3194
3195                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3196                 rootrefs->rootref[found].treeid = key.offset;
3197                 rootrefs->rootref[found].dirid =
3198                                   btrfs_root_ref_dirid(leaf, rref);
3199                 found++;
3200
3201                 ret = btrfs_next_item(root, path);
3202                 if (ret < 0) {
3203                         goto out;
3204                 } else if (ret > 0) {
3205                         ret = -EUCLEAN;
3206                         goto out;
3207                 }
3208         }
3209
3210 out:
3211         if (!ret || ret == -EOVERFLOW) {
3212                 rootrefs->num_items = found;
3213                 /* update min_treeid for next search */
3214                 if (found)
3215                         rootrefs->min_treeid =
3216                                 rootrefs->rootref[found - 1].treeid + 1;
3217                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3218                         ret = -EFAULT;
3219         }
3220
3221         kfree(rootrefs);
3222         btrfs_free_path(path);
3223
3224         return ret;
3225 }
3226
3227 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3228                                              void __user *arg,
3229                                              bool destroy_v2)
3230 {
3231         struct dentry *parent = file->f_path.dentry;
3232         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3233         struct dentry *dentry;
3234         struct inode *dir = d_inode(parent);
3235         struct inode *inode;
3236         struct btrfs_root *root = BTRFS_I(dir)->root;
3237         struct btrfs_root *dest = NULL;
3238         struct btrfs_ioctl_vol_args *vol_args = NULL;
3239         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3240         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3241         char *subvol_name, *subvol_name_ptr = NULL;
3242         int subvol_namelen;
3243         int err = 0;
3244         bool destroy_parent = false;
3245
3246         /* We don't support snapshots with extent tree v2 yet. */
3247         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3248                 btrfs_err(fs_info,
3249                           "extent tree v2 doesn't support snapshot deletion yet");
3250                 return -EOPNOTSUPP;
3251         }
3252
3253         if (destroy_v2) {
3254                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3255                 if (IS_ERR(vol_args2))
3256                         return PTR_ERR(vol_args2);
3257
3258                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3259                         err = -EOPNOTSUPP;
3260                         goto out;
3261                 }
3262
3263                 /*
3264                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3265                  * name, same as v1 currently does.
3266                  */
3267                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3268                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3269                         subvol_name = vol_args2->name;
3270
3271                         err = mnt_want_write_file(file);
3272                         if (err)
3273                                 goto out;
3274                 } else {
3275                         struct inode *old_dir;
3276
3277                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3278                                 err = -EINVAL;
3279                                 goto out;
3280                         }
3281
3282                         err = mnt_want_write_file(file);
3283                         if (err)
3284                                 goto out;
3285
3286                         dentry = btrfs_get_dentry(fs_info->sb,
3287                                         BTRFS_FIRST_FREE_OBJECTID,
3288                                         vol_args2->subvolid, 0, 0);
3289                         if (IS_ERR(dentry)) {
3290                                 err = PTR_ERR(dentry);
3291                                 goto out_drop_write;
3292                         }
3293
3294                         /*
3295                          * Change the default parent since the subvolume being
3296                          * deleted can be outside of the current mount point.
3297                          */
3298                         parent = btrfs_get_parent(dentry);
3299
3300                         /*
3301                          * At this point dentry->d_name can point to '/' if the
3302                          * subvolume we want to destroy is outsite of the
3303                          * current mount point, so we need to release the
3304                          * current dentry and execute the lookup to return a new
3305                          * one with ->d_name pointing to the
3306                          * <mount point>/subvol_name.
3307                          */
3308                         dput(dentry);
3309                         if (IS_ERR(parent)) {
3310                                 err = PTR_ERR(parent);
3311                                 goto out_drop_write;
3312                         }
3313                         old_dir = dir;
3314                         dir = d_inode(parent);
3315
3316                         /*
3317                          * If v2 was used with SPEC_BY_ID, a new parent was
3318                          * allocated since the subvolume can be outside of the
3319                          * current mount point. Later on we need to release this
3320                          * new parent dentry.
3321                          */
3322                         destroy_parent = true;
3323
3324                         /*
3325                          * On idmapped mounts, deletion via subvolid is
3326                          * restricted to subvolumes that are immediate
3327                          * ancestors of the inode referenced by the file
3328                          * descriptor in the ioctl. Otherwise the idmapping
3329                          * could potentially be abused to delete subvolumes
3330                          * anywhere in the filesystem the user wouldn't be able
3331                          * to delete without an idmapped mount.
3332                          */
3333                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3334                                 err = -EOPNOTSUPP;
3335                                 goto free_parent;
3336                         }
3337
3338                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3339                                                 fs_info, vol_args2->subvolid);
3340                         if (IS_ERR(subvol_name_ptr)) {
3341                                 err = PTR_ERR(subvol_name_ptr);
3342                                 goto free_parent;
3343                         }
3344                         /* subvol_name_ptr is already nul terminated */
3345                         subvol_name = (char *)kbasename(subvol_name_ptr);
3346                 }
3347         } else {
3348                 vol_args = memdup_user(arg, sizeof(*vol_args));
3349                 if (IS_ERR(vol_args))
3350                         return PTR_ERR(vol_args);
3351
3352                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3353                 subvol_name = vol_args->name;
3354
3355                 err = mnt_want_write_file(file);
3356                 if (err)
3357                         goto out;
3358         }
3359
3360         subvol_namelen = strlen(subvol_name);
3361
3362         if (strchr(subvol_name, '/') ||
3363             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3364                 err = -EINVAL;
3365                 goto free_subvol_name;
3366         }
3367
3368         if (!S_ISDIR(dir->i_mode)) {
3369                 err = -ENOTDIR;
3370                 goto free_subvol_name;
3371         }
3372
3373         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3374         if (err == -EINTR)
3375                 goto free_subvol_name;
3376         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3377         if (IS_ERR(dentry)) {
3378                 err = PTR_ERR(dentry);
3379                 goto out_unlock_dir;
3380         }
3381
3382         if (d_really_is_negative(dentry)) {
3383                 err = -ENOENT;
3384                 goto out_dput;
3385         }
3386
3387         inode = d_inode(dentry);
3388         dest = BTRFS_I(inode)->root;
3389         if (!capable(CAP_SYS_ADMIN)) {
3390                 /*
3391                  * Regular user.  Only allow this with a special mount
3392                  * option, when the user has write+exec access to the
3393                  * subvol root, and when rmdir(2) would have been
3394                  * allowed.
3395                  *
3396                  * Note that this is _not_ check that the subvol is
3397                  * empty or doesn't contain data that we wouldn't
3398                  * otherwise be able to delete.
3399                  *
3400                  * Users who want to delete empty subvols should try
3401                  * rmdir(2).
3402                  */
3403                 err = -EPERM;
3404                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3405                         goto out_dput;
3406
3407                 /*
3408                  * Do not allow deletion if the parent dir is the same
3409                  * as the dir to be deleted.  That means the ioctl
3410                  * must be called on the dentry referencing the root
3411                  * of the subvol, not a random directory contained
3412                  * within it.
3413                  */
3414                 err = -EINVAL;
3415                 if (root == dest)
3416                         goto out_dput;
3417
3418                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3419                 if (err)
3420                         goto out_dput;
3421         }
3422
3423         /* check if subvolume may be deleted by a user */
3424         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3425         if (err)
3426                 goto out_dput;
3427
3428         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3429                 err = -EINVAL;
3430                 goto out_dput;
3431         }
3432
3433         btrfs_inode_lock(inode, 0);
3434         err = btrfs_delete_subvolume(dir, dentry);
3435         btrfs_inode_unlock(inode, 0);
3436         if (!err)
3437                 d_delete_notify(dir, dentry);
3438
3439 out_dput:
3440         dput(dentry);
3441 out_unlock_dir:
3442         btrfs_inode_unlock(dir, 0);
3443 free_subvol_name:
3444         kfree(subvol_name_ptr);
3445 free_parent:
3446         if (destroy_parent)
3447                 dput(parent);
3448 out_drop_write:
3449         mnt_drop_write_file(file);
3450 out:
3451         kfree(vol_args2);
3452         kfree(vol_args);
3453         return err;
3454 }
3455
3456 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3457 {
3458         struct inode *inode = file_inode(file);
3459         struct btrfs_root *root = BTRFS_I(inode)->root;
3460         struct btrfs_ioctl_defrag_range_args range = {0};
3461         int ret;
3462
3463         ret = mnt_want_write_file(file);
3464         if (ret)
3465                 return ret;
3466
3467         if (btrfs_root_readonly(root)) {
3468                 ret = -EROFS;
3469                 goto out;
3470         }
3471
3472         switch (inode->i_mode & S_IFMT) {
3473         case S_IFDIR:
3474                 if (!capable(CAP_SYS_ADMIN)) {
3475                         ret = -EPERM;
3476                         goto out;
3477                 }
3478                 ret = btrfs_defrag_root(root);
3479                 break;
3480         case S_IFREG:
3481                 /*
3482                  * Note that this does not check the file descriptor for write
3483                  * access. This prevents defragmenting executables that are
3484                  * running and allows defrag on files open in read-only mode.
3485                  */
3486                 if (!capable(CAP_SYS_ADMIN) &&
3487                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3488                         ret = -EPERM;
3489                         goto out;
3490                 }
3491
3492                 if (argp) {
3493                         if (copy_from_user(&range, argp, sizeof(range))) {
3494                                 ret = -EFAULT;
3495                                 goto out;
3496                         }
3497                         /* compression requires us to start the IO */
3498                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3499                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3500                                 range.extent_thresh = (u32)-1;
3501                         }
3502                 } else {
3503                         /* the rest are all set to zero by kzalloc */
3504                         range.len = (u64)-1;
3505                 }
3506                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3507                                         &range, BTRFS_OLDEST_GENERATION, 0);
3508                 if (ret > 0)
3509                         ret = 0;
3510                 break;
3511         default:
3512                 ret = -EINVAL;
3513         }
3514 out:
3515         mnt_drop_write_file(file);
3516         return ret;
3517 }
3518
3519 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3520 {
3521         struct btrfs_ioctl_vol_args *vol_args;
3522         bool restore_op = false;
3523         int ret;
3524
3525         if (!capable(CAP_SYS_ADMIN))
3526                 return -EPERM;
3527
3528         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3529                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3530                 return -EINVAL;
3531         }
3532
3533         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3534                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3535                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3536
3537                 /*
3538                  * We can do the device add because we have a paused balanced,
3539                  * change the exclusive op type and remember we should bring
3540                  * back the paused balance
3541                  */
3542                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3543                 btrfs_exclop_start_unlock(fs_info);
3544                 restore_op = true;
3545         }
3546
3547         vol_args = memdup_user(arg, sizeof(*vol_args));
3548         if (IS_ERR(vol_args)) {
3549                 ret = PTR_ERR(vol_args);
3550                 goto out;
3551         }
3552
3553         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3554         ret = btrfs_init_new_device(fs_info, vol_args->name);
3555
3556         if (!ret)
3557                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3558
3559         kfree(vol_args);
3560 out:
3561         if (restore_op)
3562                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3563         else
3564                 btrfs_exclop_finish(fs_info);
3565         return ret;
3566 }
3567
3568 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3569 {
3570         BTRFS_DEV_LOOKUP_ARGS(args);
3571         struct inode *inode = file_inode(file);
3572         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3573         struct btrfs_ioctl_vol_args_v2 *vol_args;
3574         struct block_device *bdev = NULL;
3575         fmode_t mode;
3576         int ret;
3577         bool cancel = false;
3578
3579         if (!capable(CAP_SYS_ADMIN))
3580                 return -EPERM;
3581
3582         vol_args = memdup_user(arg, sizeof(*vol_args));
3583         if (IS_ERR(vol_args))
3584                 return PTR_ERR(vol_args);
3585
3586         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3587                 ret = -EOPNOTSUPP;
3588                 goto out;
3589         }
3590
3591         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3592         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3593                 args.devid = vol_args->devid;
3594         } else if (!strcmp("cancel", vol_args->name)) {
3595                 cancel = true;
3596         } else {
3597                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3598                 if (ret)
3599                         goto out;
3600         }
3601
3602         ret = mnt_want_write_file(file);
3603         if (ret)
3604                 goto out;
3605
3606         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3607                                            cancel);
3608         if (ret)
3609                 goto err_drop;
3610
3611         /* Exclusive operation is now claimed */
3612         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3613
3614         btrfs_exclop_finish(fs_info);
3615
3616         if (!ret) {
3617                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3618                         btrfs_info(fs_info, "device deleted: id %llu",
3619                                         vol_args->devid);
3620                 else
3621                         btrfs_info(fs_info, "device deleted: %s",
3622                                         vol_args->name);
3623         }
3624 err_drop:
3625         mnt_drop_write_file(file);
3626         if (bdev)
3627                 blkdev_put(bdev, mode);
3628 out:
3629         btrfs_put_dev_args_from_path(&args);
3630         kfree(vol_args);
3631         return ret;
3632 }
3633
3634 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3635 {
3636         BTRFS_DEV_LOOKUP_ARGS(args);
3637         struct inode *inode = file_inode(file);
3638         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3639         struct btrfs_ioctl_vol_args *vol_args;
3640         struct block_device *bdev = NULL;
3641         fmode_t mode;
3642         int ret;
3643         bool cancel = false;
3644
3645         if (!capable(CAP_SYS_ADMIN))
3646                 return -EPERM;
3647
3648         vol_args = memdup_user(arg, sizeof(*vol_args));
3649         if (IS_ERR(vol_args))
3650                 return PTR_ERR(vol_args);
3651
3652         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3653         if (!strcmp("cancel", vol_args->name)) {
3654                 cancel = true;
3655         } else {
3656                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3657                 if (ret)
3658                         goto out;
3659         }
3660
3661         ret = mnt_want_write_file(file);
3662         if (ret)
3663                 goto out;
3664
3665         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3666                                            cancel);
3667         if (ret == 0) {
3668                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3669                 if (!ret)
3670                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3671                 btrfs_exclop_finish(fs_info);
3672         }
3673
3674         mnt_drop_write_file(file);
3675         if (bdev)
3676                 blkdev_put(bdev, mode);
3677 out:
3678         btrfs_put_dev_args_from_path(&args);
3679         kfree(vol_args);
3680         return ret;
3681 }
3682
3683 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3684                                 void __user *arg)
3685 {
3686         struct btrfs_ioctl_fs_info_args *fi_args;
3687         struct btrfs_device *device;
3688         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3689         u64 flags_in;
3690         int ret = 0;
3691
3692         fi_args = memdup_user(arg, sizeof(*fi_args));
3693         if (IS_ERR(fi_args))
3694                 return PTR_ERR(fi_args);
3695
3696         flags_in = fi_args->flags;
3697         memset(fi_args, 0, sizeof(*fi_args));
3698
3699         rcu_read_lock();
3700         fi_args->num_devices = fs_devices->num_devices;
3701
3702         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3703                 if (device->devid > fi_args->max_id)
3704                         fi_args->max_id = device->devid;
3705         }
3706         rcu_read_unlock();
3707
3708         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3709         fi_args->nodesize = fs_info->nodesize;
3710         fi_args->sectorsize = fs_info->sectorsize;
3711         fi_args->clone_alignment = fs_info->sectorsize;
3712
3713         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3714                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3715                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3716                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3717         }
3718
3719         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3720                 fi_args->generation = fs_info->generation;
3721                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3722         }
3723
3724         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3725                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3726                        sizeof(fi_args->metadata_uuid));
3727                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3728         }
3729
3730         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3731                 ret = -EFAULT;
3732
3733         kfree(fi_args);
3734         return ret;
3735 }
3736
3737 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3738                                  void __user *arg)
3739 {
3740         BTRFS_DEV_LOOKUP_ARGS(args);
3741         struct btrfs_ioctl_dev_info_args *di_args;
3742         struct btrfs_device *dev;
3743         int ret = 0;
3744
3745         di_args = memdup_user(arg, sizeof(*di_args));
3746         if (IS_ERR(di_args))
3747                 return PTR_ERR(di_args);
3748
3749         args.devid = di_args->devid;
3750         if (!btrfs_is_empty_uuid(di_args->uuid))
3751                 args.uuid = di_args->uuid;
3752
3753         rcu_read_lock();
3754         dev = btrfs_find_device(fs_info->fs_devices, &args);
3755         if (!dev) {
3756                 ret = -ENODEV;
3757                 goto out;
3758         }
3759
3760         di_args->devid = dev->devid;
3761         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3762         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3763         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3764         if (dev->name) {
3765                 strncpy(di_args->path, rcu_str_deref(dev->name),
3766                                 sizeof(di_args->path) - 1);
3767                 di_args->path[sizeof(di_args->path) - 1] = 0;
3768         } else {
3769                 di_args->path[0] = '\0';
3770         }
3771
3772 out:
3773         rcu_read_unlock();
3774         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3775                 ret = -EFAULT;
3776
3777         kfree(di_args);
3778         return ret;
3779 }
3780
3781 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3782 {
3783         struct inode *inode = file_inode(file);
3784         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3785         struct btrfs_root *root = BTRFS_I(inode)->root;
3786         struct btrfs_root *new_root;
3787         struct btrfs_dir_item *di;
3788         struct btrfs_trans_handle *trans;
3789         struct btrfs_path *path = NULL;
3790         struct btrfs_disk_key disk_key;
3791         u64 objectid = 0;
3792         u64 dir_id;
3793         int ret;
3794
3795         if (!capable(CAP_SYS_ADMIN))
3796                 return -EPERM;
3797
3798         ret = mnt_want_write_file(file);
3799         if (ret)
3800                 return ret;
3801
3802         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3803                 ret = -EFAULT;
3804                 goto out;
3805         }
3806
3807         if (!objectid)
3808                 objectid = BTRFS_FS_TREE_OBJECTID;
3809
3810         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3811         if (IS_ERR(new_root)) {
3812                 ret = PTR_ERR(new_root);
3813                 goto out;
3814         }
3815         if (!is_fstree(new_root->root_key.objectid)) {
3816                 ret = -ENOENT;
3817                 goto out_free;
3818         }
3819
3820         path = btrfs_alloc_path();
3821         if (!path) {
3822                 ret = -ENOMEM;
3823                 goto out_free;
3824         }
3825
3826         trans = btrfs_start_transaction(root, 1);
3827         if (IS_ERR(trans)) {
3828                 ret = PTR_ERR(trans);
3829                 goto out_free;
3830         }
3831
3832         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3833         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3834                                    dir_id, "default", 7, 1);
3835         if (IS_ERR_OR_NULL(di)) {
3836                 btrfs_release_path(path);
3837                 btrfs_end_transaction(trans);
3838                 btrfs_err(fs_info,
3839                           "Umm, you don't have the default diritem, this isn't going to work");
3840                 ret = -ENOENT;
3841                 goto out_free;
3842         }
3843
3844         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3845         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3846         btrfs_mark_buffer_dirty(path->nodes[0]);
3847         btrfs_release_path(path);
3848
3849         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3850         btrfs_end_transaction(trans);
3851 out_free:
3852         btrfs_put_root(new_root);
3853         btrfs_free_path(path);
3854 out:
3855         mnt_drop_write_file(file);
3856         return ret;
3857 }
3858
3859 static void get_block_group_info(struct list_head *groups_list,
3860                                  struct btrfs_ioctl_space_info *space)
3861 {
3862         struct btrfs_block_group *block_group;
3863
3864         space->total_bytes = 0;
3865         space->used_bytes = 0;
3866         space->flags = 0;
3867         list_for_each_entry(block_group, groups_list, list) {
3868                 space->flags = block_group->flags;
3869                 space->total_bytes += block_group->length;
3870                 space->used_bytes += block_group->used;
3871         }
3872 }
3873
3874 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3875                                    void __user *arg)
3876 {
3877         struct btrfs_ioctl_space_args space_args;
3878         struct btrfs_ioctl_space_info space;
3879         struct btrfs_ioctl_space_info *dest;
3880         struct btrfs_ioctl_space_info *dest_orig;
3881         struct btrfs_ioctl_space_info __user *user_dest;
3882         struct btrfs_space_info *info;
3883         static const u64 types[] = {
3884                 BTRFS_BLOCK_GROUP_DATA,
3885                 BTRFS_BLOCK_GROUP_SYSTEM,
3886                 BTRFS_BLOCK_GROUP_METADATA,
3887                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3888         };
3889         int num_types = 4;
3890         int alloc_size;
3891         int ret = 0;
3892         u64 slot_count = 0;
3893         int i, c;
3894
3895         if (copy_from_user(&space_args,
3896                            (struct btrfs_ioctl_space_args __user *)arg,
3897                            sizeof(space_args)))
3898                 return -EFAULT;
3899
3900         for (i = 0; i < num_types; i++) {
3901                 struct btrfs_space_info *tmp;
3902
3903                 info = NULL;
3904                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3905                         if (tmp->flags == types[i]) {
3906                                 info = tmp;
3907                                 break;
3908                         }
3909                 }
3910
3911                 if (!info)
3912                         continue;
3913
3914                 down_read(&info->groups_sem);
3915                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3916                         if (!list_empty(&info->block_groups[c]))
3917                                 slot_count++;
3918                 }
3919                 up_read(&info->groups_sem);
3920         }
3921
3922         /*
3923          * Global block reserve, exported as a space_info
3924          */
3925         slot_count++;
3926
3927         /* space_slots == 0 means they are asking for a count */
3928         if (space_args.space_slots == 0) {
3929                 space_args.total_spaces = slot_count;
3930                 goto out;
3931         }
3932
3933         slot_count = min_t(u64, space_args.space_slots, slot_count);
3934
3935         alloc_size = sizeof(*dest) * slot_count;
3936
3937         /* we generally have at most 6 or so space infos, one for each raid
3938          * level.  So, a whole page should be more than enough for everyone
3939          */
3940         if (alloc_size > PAGE_SIZE)
3941                 return -ENOMEM;
3942
3943         space_args.total_spaces = 0;
3944         dest = kmalloc(alloc_size, GFP_KERNEL);
3945         if (!dest)
3946                 return -ENOMEM;
3947         dest_orig = dest;
3948
3949         /* now we have a buffer to copy into */
3950         for (i = 0; i < num_types; i++) {
3951                 struct btrfs_space_info *tmp;
3952
3953                 if (!slot_count)
3954                         break;
3955
3956                 info = NULL;
3957                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3958                         if (tmp->flags == types[i]) {
3959                                 info = tmp;
3960                                 break;
3961                         }
3962                 }
3963
3964                 if (!info)
3965                         continue;
3966                 down_read(&info->groups_sem);
3967                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3968                         if (!list_empty(&info->block_groups[c])) {
3969                                 get_block_group_info(&info->block_groups[c],
3970                                                      &space);
3971                                 memcpy(dest, &space, sizeof(space));
3972                                 dest++;
3973                                 space_args.total_spaces++;
3974                                 slot_count--;
3975                         }
3976                         if (!slot_count)
3977                                 break;
3978                 }
3979                 up_read(&info->groups_sem);
3980         }
3981
3982         /*
3983          * Add global block reserve
3984          */
3985         if (slot_count) {
3986                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3987
3988                 spin_lock(&block_rsv->lock);
3989                 space.total_bytes = block_rsv->size;
3990                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3991                 spin_unlock(&block_rsv->lock);
3992                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3993                 memcpy(dest, &space, sizeof(space));
3994                 space_args.total_spaces++;
3995         }
3996
3997         user_dest = (struct btrfs_ioctl_space_info __user *)
3998                 (arg + sizeof(struct btrfs_ioctl_space_args));
3999
4000         if (copy_to_user(user_dest, dest_orig, alloc_size))
4001                 ret = -EFAULT;
4002
4003         kfree(dest_orig);
4004 out:
4005         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4006                 ret = -EFAULT;
4007
4008         return ret;
4009 }
4010
4011 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4012                                             void __user *argp)
4013 {
4014         struct btrfs_trans_handle *trans;
4015         u64 transid;
4016
4017         trans = btrfs_attach_transaction_barrier(root);
4018         if (IS_ERR(trans)) {
4019                 if (PTR_ERR(trans) != -ENOENT)
4020                         return PTR_ERR(trans);
4021
4022                 /* No running transaction, don't bother */
4023                 transid = root->fs_info->last_trans_committed;
4024                 goto out;
4025         }
4026         transid = trans->transid;
4027         btrfs_commit_transaction_async(trans);
4028 out:
4029         if (argp)
4030                 if (copy_to_user(argp, &transid, sizeof(transid)))
4031                         return -EFAULT;
4032         return 0;
4033 }
4034
4035 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4036                                            void __user *argp)
4037 {
4038         u64 transid;
4039
4040         if (argp) {
4041                 if (copy_from_user(&transid, argp, sizeof(transid)))
4042                         return -EFAULT;
4043         } else {
4044                 transid = 0;  /* current trans */
4045         }
4046         return btrfs_wait_for_commit(fs_info, transid);
4047 }
4048
4049 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4050 {
4051         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4052         struct btrfs_ioctl_scrub_args *sa;
4053         int ret;
4054
4055         if (!capable(CAP_SYS_ADMIN))
4056                 return -EPERM;
4057
4058         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4059                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4060                 return -EINVAL;
4061         }
4062
4063         sa = memdup_user(arg, sizeof(*sa));
4064         if (IS_ERR(sa))
4065                 return PTR_ERR(sa);
4066
4067         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4068                 ret = mnt_want_write_file(file);
4069                 if (ret)
4070                         goto out;
4071         }
4072
4073         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4074                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4075                               0);
4076
4077         /*
4078          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4079          * error. This is important as it allows user space to know how much
4080          * progress scrub has done. For example, if scrub is canceled we get
4081          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4082          * space. Later user space can inspect the progress from the structure
4083          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4084          * previously (btrfs-progs does this).
4085          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4086          * then return -EFAULT to signal the structure was not copied or it may
4087          * be corrupt and unreliable due to a partial copy.
4088          */
4089         if (copy_to_user(arg, sa, sizeof(*sa)))
4090                 ret = -EFAULT;
4091
4092         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4093                 mnt_drop_write_file(file);
4094 out:
4095         kfree(sa);
4096         return ret;
4097 }
4098
4099 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4100 {
4101         if (!capable(CAP_SYS_ADMIN))
4102                 return -EPERM;
4103
4104         return btrfs_scrub_cancel(fs_info);
4105 }
4106
4107 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4108                                        void __user *arg)
4109 {
4110         struct btrfs_ioctl_scrub_args *sa;
4111         int ret;
4112
4113         if (!capable(CAP_SYS_ADMIN))
4114                 return -EPERM;
4115
4116         sa = memdup_user(arg, sizeof(*sa));
4117         if (IS_ERR(sa))
4118                 return PTR_ERR(sa);
4119
4120         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4121
4122         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4123                 ret = -EFAULT;
4124
4125         kfree(sa);
4126         return ret;
4127 }
4128
4129 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4130                                       void __user *arg)
4131 {
4132         struct btrfs_ioctl_get_dev_stats *sa;
4133         int ret;
4134
4135         sa = memdup_user(arg, sizeof(*sa));
4136         if (IS_ERR(sa))
4137                 return PTR_ERR(sa);
4138
4139         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4140                 kfree(sa);
4141                 return -EPERM;
4142         }
4143
4144         ret = btrfs_get_dev_stats(fs_info, sa);
4145
4146         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4147                 ret = -EFAULT;
4148
4149         kfree(sa);
4150         return ret;
4151 }
4152
4153 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4154                                     void __user *arg)
4155 {
4156         struct btrfs_ioctl_dev_replace_args *p;
4157         int ret;
4158
4159         if (!capable(CAP_SYS_ADMIN))
4160                 return -EPERM;
4161
4162         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4163                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4164                 return -EINVAL;
4165         }
4166
4167         p = memdup_user(arg, sizeof(*p));
4168         if (IS_ERR(p))
4169                 return PTR_ERR(p);
4170
4171         switch (p->cmd) {
4172         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4173                 if (sb_rdonly(fs_info->sb)) {
4174                         ret = -EROFS;
4175                         goto out;
4176                 }
4177                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4178                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4179                 } else {
4180                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4181                         btrfs_exclop_finish(fs_info);
4182                 }
4183                 break;
4184         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4185                 btrfs_dev_replace_status(fs_info, p);
4186                 ret = 0;
4187                 break;
4188         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4189                 p->result = btrfs_dev_replace_cancel(fs_info);
4190                 ret = 0;
4191                 break;
4192         default:
4193                 ret = -EINVAL;
4194                 break;
4195         }
4196
4197         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4198                 ret = -EFAULT;
4199 out:
4200         kfree(p);
4201         return ret;
4202 }
4203
4204 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4205 {
4206         int ret = 0;
4207         int i;
4208         u64 rel_ptr;
4209         int size;
4210         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4211         struct inode_fs_paths *ipath = NULL;
4212         struct btrfs_path *path;
4213
4214         if (!capable(CAP_DAC_READ_SEARCH))
4215                 return -EPERM;
4216
4217         path = btrfs_alloc_path();
4218         if (!path) {
4219                 ret = -ENOMEM;
4220                 goto out;
4221         }
4222
4223         ipa = memdup_user(arg, sizeof(*ipa));
4224         if (IS_ERR(ipa)) {
4225                 ret = PTR_ERR(ipa);
4226                 ipa = NULL;
4227                 goto out;
4228         }
4229
4230         size = min_t(u32, ipa->size, 4096);
4231         ipath = init_ipath(size, root, path);
4232         if (IS_ERR(ipath)) {
4233                 ret = PTR_ERR(ipath);
4234                 ipath = NULL;
4235                 goto out;
4236         }
4237
4238         ret = paths_from_inode(ipa->inum, ipath);
4239         if (ret < 0)
4240                 goto out;
4241
4242         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4243                 rel_ptr = ipath->fspath->val[i] -
4244                           (u64)(unsigned long)ipath->fspath->val;
4245                 ipath->fspath->val[i] = rel_ptr;
4246         }
4247
4248         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4249                            ipath->fspath, size);
4250         if (ret) {
4251                 ret = -EFAULT;
4252                 goto out;
4253         }
4254
4255 out:
4256         btrfs_free_path(path);
4257         free_ipath(ipath);
4258         kfree(ipa);
4259
4260         return ret;
4261 }
4262
4263 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4264 {
4265         struct btrfs_data_container *inodes = ctx;
4266         const size_t c = 3 * sizeof(u64);
4267
4268         if (inodes->bytes_left >= c) {
4269                 inodes->bytes_left -= c;
4270                 inodes->val[inodes->elem_cnt] = inum;
4271                 inodes->val[inodes->elem_cnt + 1] = offset;
4272                 inodes->val[inodes->elem_cnt + 2] = root;
4273                 inodes->elem_cnt += 3;
4274         } else {
4275                 inodes->bytes_missing += c - inodes->bytes_left;
4276                 inodes->bytes_left = 0;
4277                 inodes->elem_missed += 3;
4278         }
4279
4280         return 0;
4281 }
4282
4283 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4284                                         void __user *arg, int version)
4285 {
4286         int ret = 0;
4287         int size;
4288         struct btrfs_ioctl_logical_ino_args *loi;
4289         struct btrfs_data_container *inodes = NULL;
4290         struct btrfs_path *path = NULL;
4291         bool ignore_offset;
4292
4293         if (!capable(CAP_SYS_ADMIN))
4294                 return -EPERM;
4295
4296         loi = memdup_user(arg, sizeof(*loi));
4297         if (IS_ERR(loi))
4298                 return PTR_ERR(loi);
4299
4300         if (version == 1) {
4301                 ignore_offset = false;
4302                 size = min_t(u32, loi->size, SZ_64K);
4303         } else {
4304                 /* All reserved bits must be 0 for now */
4305                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4306                         ret = -EINVAL;
4307                         goto out_loi;
4308                 }
4309                 /* Only accept flags we have defined so far */
4310                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4311                         ret = -EINVAL;
4312                         goto out_loi;
4313                 }
4314                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4315                 size = min_t(u32, loi->size, SZ_16M);
4316         }
4317
4318         path = btrfs_alloc_path();
4319         if (!path) {
4320                 ret = -ENOMEM;
4321                 goto out;
4322         }
4323
4324         inodes = init_data_container(size);
4325         if (IS_ERR(inodes)) {
4326                 ret = PTR_ERR(inodes);
4327                 inodes = NULL;
4328                 goto out;
4329         }
4330
4331         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4332                                           build_ino_list, inodes, ignore_offset);
4333         if (ret == -EINVAL)
4334                 ret = -ENOENT;
4335         if (ret < 0)
4336                 goto out;
4337
4338         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4339                            size);
4340         if (ret)
4341                 ret = -EFAULT;
4342
4343 out:
4344         btrfs_free_path(path);
4345         kvfree(inodes);
4346 out_loi:
4347         kfree(loi);
4348
4349         return ret;
4350 }
4351
4352 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4353                                struct btrfs_ioctl_balance_args *bargs)
4354 {
4355         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4356
4357         bargs->flags = bctl->flags;
4358
4359         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4360                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4361         if (atomic_read(&fs_info->balance_pause_req))
4362                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4363         if (atomic_read(&fs_info->balance_cancel_req))
4364                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4365
4366         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4367         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4368         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4369
4370         spin_lock(&fs_info->balance_lock);
4371         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4372         spin_unlock(&fs_info->balance_lock);
4373 }
4374
4375 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4376 {
4377         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4378         struct btrfs_fs_info *fs_info = root->fs_info;
4379         struct btrfs_ioctl_balance_args *bargs;
4380         struct btrfs_balance_control *bctl;
4381         bool need_unlock; /* for mut. excl. ops lock */
4382         int ret;
4383
4384         if (!arg)
4385                 btrfs_warn(fs_info,
4386         "IOC_BALANCE ioctl (v1) is deprecated and will be removed in kernel 5.18");
4387
4388         if (!capable(CAP_SYS_ADMIN))
4389                 return -EPERM;
4390
4391         ret = mnt_want_write_file(file);
4392         if (ret)
4393                 return ret;
4394
4395 again:
4396         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4397                 mutex_lock(&fs_info->balance_mutex);
4398                 need_unlock = true;
4399                 goto locked;
4400         }
4401
4402         /*
4403          * mut. excl. ops lock is locked.  Three possibilities:
4404          *   (1) some other op is running
4405          *   (2) balance is running
4406          *   (3) balance is paused -- special case (think resume)
4407          */
4408         mutex_lock(&fs_info->balance_mutex);
4409         if (fs_info->balance_ctl) {
4410                 /* this is either (2) or (3) */
4411                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4412                         mutex_unlock(&fs_info->balance_mutex);
4413                         /*
4414                          * Lock released to allow other waiters to continue,
4415                          * we'll reexamine the status again.
4416                          */
4417                         mutex_lock(&fs_info->balance_mutex);
4418
4419                         if (fs_info->balance_ctl &&
4420                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4421                                 /* this is (3) */
4422                                 need_unlock = false;
4423                                 goto locked;
4424                         }
4425
4426                         mutex_unlock(&fs_info->balance_mutex);
4427                         goto again;
4428                 } else {
4429                         /* this is (2) */
4430                         mutex_unlock(&fs_info->balance_mutex);
4431                         ret = -EINPROGRESS;
4432                         goto out;
4433                 }
4434         } else {
4435                 /* this is (1) */
4436                 mutex_unlock(&fs_info->balance_mutex);
4437                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4438                 goto out;
4439         }
4440
4441 locked:
4442
4443         if (arg) {
4444                 bargs = memdup_user(arg, sizeof(*bargs));
4445                 if (IS_ERR(bargs)) {
4446                         ret = PTR_ERR(bargs);
4447                         goto out_unlock;
4448                 }
4449
4450                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4451                         if (!fs_info->balance_ctl) {
4452                                 ret = -ENOTCONN;
4453                                 goto out_bargs;
4454                         }
4455
4456                         bctl = fs_info->balance_ctl;
4457                         spin_lock(&fs_info->balance_lock);
4458                         bctl->flags |= BTRFS_BALANCE_RESUME;
4459                         spin_unlock(&fs_info->balance_lock);
4460                         btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4461
4462                         goto do_balance;
4463                 }
4464         } else {
4465                 bargs = NULL;
4466         }
4467
4468         if (fs_info->balance_ctl) {
4469                 ret = -EINPROGRESS;
4470                 goto out_bargs;
4471         }
4472
4473         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4474         if (!bctl) {
4475                 ret = -ENOMEM;
4476                 goto out_bargs;
4477         }
4478
4479         if (arg) {
4480                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4481                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4482                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4483
4484                 bctl->flags = bargs->flags;
4485         } else {
4486                 /* balance everything - no filters */
4487                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4488         }
4489
4490         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4491                 ret = -EINVAL;
4492                 goto out_bctl;
4493         }
4494
4495 do_balance:
4496         /*
4497          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4498          * bctl is freed in reset_balance_state, or, if restriper was paused
4499          * all the way until unmount, in free_fs_info.  The flag should be
4500          * cleared after reset_balance_state.
4501          */
4502         need_unlock = false;
4503
4504         ret = btrfs_balance(fs_info, bctl, bargs);
4505         bctl = NULL;
4506
4507         if ((ret == 0 || ret == -ECANCELED) && arg) {
4508                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4509                         ret = -EFAULT;
4510         }
4511
4512 out_bctl:
4513         kfree(bctl);
4514 out_bargs:
4515         kfree(bargs);
4516 out_unlock:
4517         mutex_unlock(&fs_info->balance_mutex);
4518         if (need_unlock)
4519                 btrfs_exclop_finish(fs_info);
4520 out:
4521         mnt_drop_write_file(file);
4522         return ret;
4523 }
4524
4525 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4526 {
4527         if (!capable(CAP_SYS_ADMIN))
4528                 return -EPERM;
4529
4530         switch (cmd) {
4531         case BTRFS_BALANCE_CTL_PAUSE:
4532                 return btrfs_pause_balance(fs_info);
4533         case BTRFS_BALANCE_CTL_CANCEL:
4534                 return btrfs_cancel_balance(fs_info);
4535         }
4536
4537         return -EINVAL;
4538 }
4539
4540 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4541                                          void __user *arg)
4542 {
4543         struct btrfs_ioctl_balance_args *bargs;
4544         int ret = 0;
4545
4546         if (!capable(CAP_SYS_ADMIN))
4547                 return -EPERM;
4548
4549         mutex_lock(&fs_info->balance_mutex);
4550         if (!fs_info->balance_ctl) {
4551                 ret = -ENOTCONN;
4552                 goto out;
4553         }
4554
4555         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4556         if (!bargs) {
4557                 ret = -ENOMEM;
4558                 goto out;
4559         }
4560
4561         btrfs_update_ioctl_balance_args(fs_info, bargs);
4562
4563         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4564                 ret = -EFAULT;
4565
4566         kfree(bargs);
4567 out:
4568         mutex_unlock(&fs_info->balance_mutex);
4569         return ret;
4570 }
4571
4572 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4573 {
4574         struct inode *inode = file_inode(file);
4575         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4576         struct btrfs_ioctl_quota_ctl_args *sa;
4577         int ret;
4578
4579         if (!capable(CAP_SYS_ADMIN))
4580                 return -EPERM;
4581
4582         ret = mnt_want_write_file(file);
4583         if (ret)
4584                 return ret;
4585
4586         sa = memdup_user(arg, sizeof(*sa));
4587         if (IS_ERR(sa)) {
4588                 ret = PTR_ERR(sa);
4589                 goto drop_write;
4590         }
4591
4592         down_write(&fs_info->subvol_sem);
4593
4594         switch (sa->cmd) {
4595         case BTRFS_QUOTA_CTL_ENABLE:
4596                 ret = btrfs_quota_enable(fs_info);
4597                 break;
4598         case BTRFS_QUOTA_CTL_DISABLE:
4599                 ret = btrfs_quota_disable(fs_info);
4600                 break;
4601         default:
4602                 ret = -EINVAL;
4603                 break;
4604         }
4605
4606         kfree(sa);
4607         up_write(&fs_info->subvol_sem);
4608 drop_write:
4609         mnt_drop_write_file(file);
4610         return ret;
4611 }
4612
4613 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4614 {
4615         struct inode *inode = file_inode(file);
4616         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4617         struct btrfs_root *root = BTRFS_I(inode)->root;
4618         struct btrfs_ioctl_qgroup_assign_args *sa;
4619         struct btrfs_trans_handle *trans;
4620         int ret;
4621         int err;
4622
4623         if (!capable(CAP_SYS_ADMIN))
4624                 return -EPERM;
4625
4626         ret = mnt_want_write_file(file);
4627         if (ret)
4628                 return ret;
4629
4630         sa = memdup_user(arg, sizeof(*sa));
4631         if (IS_ERR(sa)) {
4632                 ret = PTR_ERR(sa);
4633                 goto drop_write;
4634         }
4635
4636         trans = btrfs_join_transaction(root);
4637         if (IS_ERR(trans)) {
4638                 ret = PTR_ERR(trans);
4639                 goto out;
4640         }
4641
4642         if (sa->assign) {
4643                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4644         } else {
4645                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4646         }
4647
4648         /* update qgroup status and info */
4649         err = btrfs_run_qgroups(trans);
4650         if (err < 0)
4651                 btrfs_handle_fs_error(fs_info, err,
4652                                       "failed to update qgroup status and info");
4653         err = btrfs_end_transaction(trans);
4654         if (err && !ret)
4655                 ret = err;
4656
4657 out:
4658         kfree(sa);
4659 drop_write:
4660         mnt_drop_write_file(file);
4661         return ret;
4662 }
4663
4664 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4665 {
4666         struct inode *inode = file_inode(file);
4667         struct btrfs_root *root = BTRFS_I(inode)->root;
4668         struct btrfs_ioctl_qgroup_create_args *sa;
4669         struct btrfs_trans_handle *trans;
4670         int ret;
4671         int err;
4672
4673         if (!capable(CAP_SYS_ADMIN))
4674                 return -EPERM;
4675
4676         ret = mnt_want_write_file(file);
4677         if (ret)
4678                 return ret;
4679
4680         sa = memdup_user(arg, sizeof(*sa));
4681         if (IS_ERR(sa)) {
4682                 ret = PTR_ERR(sa);
4683                 goto drop_write;
4684         }
4685
4686         if (!sa->qgroupid) {
4687                 ret = -EINVAL;
4688                 goto out;
4689         }
4690
4691         trans = btrfs_join_transaction(root);
4692         if (IS_ERR(trans)) {
4693                 ret = PTR_ERR(trans);
4694                 goto out;
4695         }
4696
4697         if (sa->create) {
4698                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4699         } else {
4700                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4701         }
4702
4703         err = btrfs_end_transaction(trans);
4704         if (err && !ret)
4705                 ret = err;
4706
4707 out:
4708         kfree(sa);
4709 drop_write:
4710         mnt_drop_write_file(file);
4711         return ret;
4712 }
4713
4714 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4715 {
4716         struct inode *inode = file_inode(file);
4717         struct btrfs_root *root = BTRFS_I(inode)->root;
4718         struct btrfs_ioctl_qgroup_limit_args *sa;
4719         struct btrfs_trans_handle *trans;
4720         int ret;
4721         int err;
4722         u64 qgroupid;
4723
4724         if (!capable(CAP_SYS_ADMIN))
4725                 return -EPERM;
4726
4727         ret = mnt_want_write_file(file);
4728         if (ret)
4729                 return ret;
4730
4731         sa = memdup_user(arg, sizeof(*sa));
4732         if (IS_ERR(sa)) {
4733                 ret = PTR_ERR(sa);
4734                 goto drop_write;
4735         }
4736
4737         trans = btrfs_join_transaction(root);
4738         if (IS_ERR(trans)) {
4739                 ret = PTR_ERR(trans);
4740                 goto out;
4741         }
4742
4743         qgroupid = sa->qgroupid;
4744         if (!qgroupid) {
4745                 /* take the current subvol as qgroup */
4746                 qgroupid = root->root_key.objectid;
4747         }
4748
4749         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4750
4751         err = btrfs_end_transaction(trans);
4752         if (err && !ret)
4753                 ret = err;
4754
4755 out:
4756         kfree(sa);
4757 drop_write:
4758         mnt_drop_write_file(file);
4759         return ret;
4760 }
4761
4762 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4763 {
4764         struct inode *inode = file_inode(file);
4765         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4766         struct btrfs_ioctl_quota_rescan_args *qsa;
4767         int ret;
4768
4769         if (!capable(CAP_SYS_ADMIN))
4770                 return -EPERM;
4771
4772         ret = mnt_want_write_file(file);
4773         if (ret)
4774                 return ret;
4775
4776         qsa = memdup_user(arg, sizeof(*qsa));
4777         if (IS_ERR(qsa)) {
4778                 ret = PTR_ERR(qsa);
4779                 goto drop_write;
4780         }
4781
4782         if (qsa->flags) {
4783                 ret = -EINVAL;
4784                 goto out;
4785         }
4786
4787         ret = btrfs_qgroup_rescan(fs_info);
4788
4789 out:
4790         kfree(qsa);
4791 drop_write:
4792         mnt_drop_write_file(file);
4793         return ret;
4794 }
4795
4796 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4797                                                 void __user *arg)
4798 {
4799         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4800
4801         if (!capable(CAP_SYS_ADMIN))
4802                 return -EPERM;
4803
4804         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4805                 qsa.flags = 1;
4806                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4807         }
4808
4809         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4810                 return -EFAULT;
4811
4812         return 0;
4813 }
4814
4815 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4816                                                 void __user *arg)
4817 {
4818         if (!capable(CAP_SYS_ADMIN))
4819                 return -EPERM;
4820
4821         return btrfs_qgroup_wait_for_completion(fs_info, true);
4822 }
4823
4824 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4825                                             struct user_namespace *mnt_userns,
4826                                             struct btrfs_ioctl_received_subvol_args *sa)
4827 {
4828         struct inode *inode = file_inode(file);
4829         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4830         struct btrfs_root *root = BTRFS_I(inode)->root;
4831         struct btrfs_root_item *root_item = &root->root_item;
4832         struct btrfs_trans_handle *trans;
4833         struct timespec64 ct = current_time(inode);
4834         int ret = 0;
4835         int received_uuid_changed;
4836
4837         if (!inode_owner_or_capable(mnt_userns, inode))
4838                 return -EPERM;
4839
4840         ret = mnt_want_write_file(file);
4841         if (ret < 0)
4842                 return ret;
4843
4844         down_write(&fs_info->subvol_sem);
4845
4846         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4847                 ret = -EINVAL;
4848                 goto out;
4849         }
4850
4851         if (btrfs_root_readonly(root)) {
4852                 ret = -EROFS;
4853                 goto out;
4854         }
4855
4856         /*
4857          * 1 - root item
4858          * 2 - uuid items (received uuid + subvol uuid)
4859          */
4860         trans = btrfs_start_transaction(root, 3);
4861         if (IS_ERR(trans)) {
4862                 ret = PTR_ERR(trans);
4863                 trans = NULL;
4864                 goto out;
4865         }
4866
4867         sa->rtransid = trans->transid;
4868         sa->rtime.sec = ct.tv_sec;
4869         sa->rtime.nsec = ct.tv_nsec;
4870
4871         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4872                                        BTRFS_UUID_SIZE);
4873         if (received_uuid_changed &&
4874             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4875                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4876                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4877                                           root->root_key.objectid);
4878                 if (ret && ret != -ENOENT) {
4879                         btrfs_abort_transaction(trans, ret);
4880                         btrfs_end_transaction(trans);
4881                         goto out;
4882                 }
4883         }
4884         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4885         btrfs_set_root_stransid(root_item, sa->stransid);
4886         btrfs_set_root_rtransid(root_item, sa->rtransid);
4887         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4888         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4889         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4890         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4891
4892         ret = btrfs_update_root(trans, fs_info->tree_root,
4893                                 &root->root_key, &root->root_item);
4894         if (ret < 0) {
4895                 btrfs_end_transaction(trans);
4896                 goto out;
4897         }
4898         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4899                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4900                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4901                                           root->root_key.objectid);
4902                 if (ret < 0 && ret != -EEXIST) {
4903                         btrfs_abort_transaction(trans, ret);
4904                         btrfs_end_transaction(trans);
4905                         goto out;
4906                 }
4907         }
4908         ret = btrfs_commit_transaction(trans);
4909 out:
4910         up_write(&fs_info->subvol_sem);
4911         mnt_drop_write_file(file);
4912         return ret;
4913 }
4914
4915 #ifdef CONFIG_64BIT
4916 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4917                                                 void __user *arg)
4918 {
4919         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4920         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4921         int ret = 0;
4922
4923         args32 = memdup_user(arg, sizeof(*args32));
4924         if (IS_ERR(args32))
4925                 return PTR_ERR(args32);
4926
4927         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4928         if (!args64) {
4929                 ret = -ENOMEM;
4930                 goto out;
4931         }
4932
4933         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4934         args64->stransid = args32->stransid;
4935         args64->rtransid = args32->rtransid;
4936         args64->stime.sec = args32->stime.sec;
4937         args64->stime.nsec = args32->stime.nsec;
4938         args64->rtime.sec = args32->rtime.sec;
4939         args64->rtime.nsec = args32->rtime.nsec;
4940         args64->flags = args32->flags;
4941
4942         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4943         if (ret)
4944                 goto out;
4945
4946         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4947         args32->stransid = args64->stransid;
4948         args32->rtransid = args64->rtransid;
4949         args32->stime.sec = args64->stime.sec;
4950         args32->stime.nsec = args64->stime.nsec;
4951         args32->rtime.sec = args64->rtime.sec;
4952         args32->rtime.nsec = args64->rtime.nsec;
4953         args32->flags = args64->flags;
4954
4955         ret = copy_to_user(arg, args32, sizeof(*args32));
4956         if (ret)
4957                 ret = -EFAULT;
4958
4959 out:
4960         kfree(args32);
4961         kfree(args64);
4962         return ret;
4963 }
4964 #endif
4965
4966 static long btrfs_ioctl_set_received_subvol(struct file *file,
4967                                             void __user *arg)
4968 {
4969         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4970         int ret = 0;
4971
4972         sa = memdup_user(arg, sizeof(*sa));
4973         if (IS_ERR(sa))
4974                 return PTR_ERR(sa);
4975
4976         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4977
4978         if (ret)
4979                 goto out;
4980
4981         ret = copy_to_user(arg, sa, sizeof(*sa));
4982         if (ret)
4983                 ret = -EFAULT;
4984
4985 out:
4986         kfree(sa);
4987         return ret;
4988 }
4989
4990 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4991                                         void __user *arg)
4992 {
4993         size_t len;
4994         int ret;
4995         char label[BTRFS_LABEL_SIZE];
4996
4997         spin_lock(&fs_info->super_lock);
4998         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4999         spin_unlock(&fs_info->super_lock);
5000
5001         len = strnlen(label, BTRFS_LABEL_SIZE);
5002
5003         if (len == BTRFS_LABEL_SIZE) {
5004                 btrfs_warn(fs_info,
5005                            "label is too long, return the first %zu bytes",
5006                            --len);
5007         }
5008
5009         ret = copy_to_user(arg, label, len);
5010
5011         return ret ? -EFAULT : 0;
5012 }
5013
5014 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5015 {
5016         struct inode *inode = file_inode(file);
5017         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5018         struct btrfs_root *root = BTRFS_I(inode)->root;
5019         struct btrfs_super_block *super_block = fs_info->super_copy;
5020         struct btrfs_trans_handle *trans;
5021         char label[BTRFS_LABEL_SIZE];
5022         int ret;
5023
5024         if (!capable(CAP_SYS_ADMIN))
5025                 return -EPERM;
5026
5027         if (copy_from_user(label, arg, sizeof(label)))
5028                 return -EFAULT;
5029
5030         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5031                 btrfs_err(fs_info,
5032                           "unable to set label with more than %d bytes",
5033                           BTRFS_LABEL_SIZE - 1);
5034                 return -EINVAL;
5035         }
5036
5037         ret = mnt_want_write_file(file);
5038         if (ret)
5039                 return ret;
5040
5041         trans = btrfs_start_transaction(root, 0);
5042         if (IS_ERR(trans)) {
5043                 ret = PTR_ERR(trans);
5044                 goto out_unlock;
5045         }
5046
5047         spin_lock(&fs_info->super_lock);
5048         strcpy(super_block->label, label);
5049         spin_unlock(&fs_info->super_lock);
5050         ret = btrfs_commit_transaction(trans);
5051
5052 out_unlock:
5053         mnt_drop_write_file(file);
5054         return ret;
5055 }
5056
5057 #define INIT_FEATURE_FLAGS(suffix) \
5058         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5059           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5060           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5061
5062 int btrfs_ioctl_get_supported_features(void __user *arg)
5063 {
5064         static const struct btrfs_ioctl_feature_flags features[3] = {
5065                 INIT_FEATURE_FLAGS(SUPP),
5066                 INIT_FEATURE_FLAGS(SAFE_SET),
5067                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5068         };
5069
5070         if (copy_to_user(arg, &features, sizeof(features)))
5071                 return -EFAULT;
5072
5073         return 0;
5074 }
5075
5076 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5077                                         void __user *arg)
5078 {
5079         struct btrfs_super_block *super_block = fs_info->super_copy;
5080         struct btrfs_ioctl_feature_flags features;
5081
5082         features.compat_flags = btrfs_super_compat_flags(super_block);
5083         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5084         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5085
5086         if (copy_to_user(arg, &features, sizeof(features)))
5087                 return -EFAULT;
5088
5089         return 0;
5090 }
5091
5092 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5093                               enum btrfs_feature_set set,
5094                               u64 change_mask, u64 flags, u64 supported_flags,
5095                               u64 safe_set, u64 safe_clear)
5096 {
5097         const char *type = btrfs_feature_set_name(set);
5098         char *names;
5099         u64 disallowed, unsupported;
5100         u64 set_mask = flags & change_mask;
5101         u64 clear_mask = ~flags & change_mask;
5102
5103         unsupported = set_mask & ~supported_flags;
5104         if (unsupported) {
5105                 names = btrfs_printable_features(set, unsupported);
5106                 if (names) {
5107                         btrfs_warn(fs_info,
5108                                    "this kernel does not support the %s feature bit%s",
5109                                    names, strchr(names, ',') ? "s" : "");
5110                         kfree(names);
5111                 } else
5112                         btrfs_warn(fs_info,
5113                                    "this kernel does not support %s bits 0x%llx",
5114                                    type, unsupported);
5115                 return -EOPNOTSUPP;
5116         }
5117
5118         disallowed = set_mask & ~safe_set;
5119         if (disallowed) {
5120                 names = btrfs_printable_features(set, disallowed);
5121                 if (names) {
5122                         btrfs_warn(fs_info,
5123                                    "can't set the %s feature bit%s while mounted",
5124                                    names, strchr(names, ',') ? "s" : "");
5125                         kfree(names);
5126                 } else
5127                         btrfs_warn(fs_info,
5128                                    "can't set %s bits 0x%llx while mounted",
5129                                    type, disallowed);
5130                 return -EPERM;
5131         }
5132
5133         disallowed = clear_mask & ~safe_clear;
5134         if (disallowed) {
5135                 names = btrfs_printable_features(set, disallowed);
5136                 if (names) {
5137                         btrfs_warn(fs_info,
5138                                    "can't clear the %s feature bit%s while mounted",
5139                                    names, strchr(names, ',') ? "s" : "");
5140                         kfree(names);
5141                 } else
5142                         btrfs_warn(fs_info,
5143                                    "can't clear %s bits 0x%llx while mounted",
5144                                    type, disallowed);
5145                 return -EPERM;
5146         }
5147
5148         return 0;
5149 }
5150
5151 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5152 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5153                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5154                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5155                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5156
5157 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5158 {
5159         struct inode *inode = file_inode(file);
5160         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5161         struct btrfs_root *root = BTRFS_I(inode)->root;
5162         struct btrfs_super_block *super_block = fs_info->super_copy;
5163         struct btrfs_ioctl_feature_flags flags[2];
5164         struct btrfs_trans_handle *trans;
5165         u64 newflags;
5166         int ret;
5167
5168         if (!capable(CAP_SYS_ADMIN))
5169                 return -EPERM;
5170
5171         if (copy_from_user(flags, arg, sizeof(flags)))
5172                 return -EFAULT;
5173
5174         /* Nothing to do */
5175         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5176             !flags[0].incompat_flags)
5177                 return 0;
5178
5179         ret = check_feature(fs_info, flags[0].compat_flags,
5180                             flags[1].compat_flags, COMPAT);
5181         if (ret)
5182                 return ret;
5183
5184         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5185                             flags[1].compat_ro_flags, COMPAT_RO);
5186         if (ret)
5187                 return ret;
5188
5189         ret = check_feature(fs_info, flags[0].incompat_flags,
5190                             flags[1].incompat_flags, INCOMPAT);
5191         if (ret)
5192                 return ret;
5193
5194         ret = mnt_want_write_file(file);
5195         if (ret)
5196                 return ret;
5197
5198         trans = btrfs_start_transaction(root, 0);
5199         if (IS_ERR(trans)) {
5200                 ret = PTR_ERR(trans);
5201                 goto out_drop_write;
5202         }
5203
5204         spin_lock(&fs_info->super_lock);
5205         newflags = btrfs_super_compat_flags(super_block);
5206         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5207         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5208         btrfs_set_super_compat_flags(super_block, newflags);
5209
5210         newflags = btrfs_super_compat_ro_flags(super_block);
5211         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5212         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5213         btrfs_set_super_compat_ro_flags(super_block, newflags);
5214
5215         newflags = btrfs_super_incompat_flags(super_block);
5216         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5217         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5218         btrfs_set_super_incompat_flags(super_block, newflags);
5219         spin_unlock(&fs_info->super_lock);
5220
5221         ret = btrfs_commit_transaction(trans);
5222 out_drop_write:
5223         mnt_drop_write_file(file);
5224
5225         return ret;
5226 }
5227
5228 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5229 {
5230         struct btrfs_ioctl_send_args *arg;
5231         int ret;
5232
5233         if (compat) {
5234 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5235                 struct btrfs_ioctl_send_args_32 args32;
5236
5237                 ret = copy_from_user(&args32, argp, sizeof(args32));
5238                 if (ret)
5239                         return -EFAULT;
5240                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5241                 if (!arg)
5242                         return -ENOMEM;
5243                 arg->send_fd = args32.send_fd;
5244                 arg->clone_sources_count = args32.clone_sources_count;
5245                 arg->clone_sources = compat_ptr(args32.clone_sources);
5246                 arg->parent_root = args32.parent_root;
5247                 arg->flags = args32.flags;
5248                 memcpy(arg->reserved, args32.reserved,
5249                        sizeof(args32.reserved));
5250 #else
5251                 return -ENOTTY;
5252 #endif
5253         } else {
5254                 arg = memdup_user(argp, sizeof(*arg));
5255                 if (IS_ERR(arg))
5256                         return PTR_ERR(arg);
5257         }
5258         ret = btrfs_ioctl_send(inode, arg);
5259         kfree(arg);
5260         return ret;
5261 }
5262
5263 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5264                                     bool compat)
5265 {
5266         struct btrfs_ioctl_encoded_io_args args = { 0 };
5267         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5268                                              flags);
5269         size_t copy_end;
5270         struct iovec iovstack[UIO_FASTIOV];
5271         struct iovec *iov = iovstack;
5272         struct iov_iter iter;
5273         loff_t pos;
5274         struct kiocb kiocb;
5275         ssize_t ret;
5276
5277         if (!capable(CAP_SYS_ADMIN)) {
5278                 ret = -EPERM;
5279                 goto out_acct;
5280         }
5281
5282         if (compat) {
5283 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5284                 struct btrfs_ioctl_encoded_io_args_32 args32;
5285
5286                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5287                                        flags);
5288                 if (copy_from_user(&args32, argp, copy_end)) {
5289                         ret = -EFAULT;
5290                         goto out_acct;
5291                 }
5292                 args.iov = compat_ptr(args32.iov);
5293                 args.iovcnt = args32.iovcnt;
5294                 args.offset = args32.offset;
5295                 args.flags = args32.flags;
5296 #else
5297                 return -ENOTTY;
5298 #endif
5299         } else {
5300                 copy_end = copy_end_kernel;
5301                 if (copy_from_user(&args, argp, copy_end)) {
5302                         ret = -EFAULT;
5303                         goto out_acct;
5304                 }
5305         }
5306         if (args.flags != 0) {
5307                 ret = -EINVAL;
5308                 goto out_acct;
5309         }
5310
5311         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5312                            &iov, &iter);
5313         if (ret < 0)
5314                 goto out_acct;
5315
5316         if (iov_iter_count(&iter) == 0) {
5317                 ret = 0;
5318                 goto out_iov;
5319         }
5320         pos = args.offset;
5321         ret = rw_verify_area(READ, file, &pos, args.len);
5322         if (ret < 0)
5323                 goto out_iov;
5324
5325         init_sync_kiocb(&kiocb, file);
5326         kiocb.ki_pos = pos;
5327
5328         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5329         if (ret >= 0) {
5330                 fsnotify_access(file);
5331                 if (copy_to_user(argp + copy_end,
5332                                  (char *)&args + copy_end_kernel,
5333                                  sizeof(args) - copy_end_kernel))
5334                         ret = -EFAULT;
5335         }
5336
5337 out_iov:
5338         kfree(iov);
5339 out_acct:
5340         if (ret > 0)
5341                 add_rchar(current, ret);
5342         inc_syscr(current);
5343         return ret;
5344 }
5345
5346 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5347 {
5348         struct btrfs_ioctl_encoded_io_args args;
5349         struct iovec iovstack[UIO_FASTIOV];
5350         struct iovec *iov = iovstack;
5351         struct iov_iter iter;
5352         loff_t pos;
5353         struct kiocb kiocb;
5354         ssize_t ret;
5355
5356         if (!capable(CAP_SYS_ADMIN)) {
5357                 ret = -EPERM;
5358                 goto out_acct;
5359         }
5360
5361         if (!(file->f_mode & FMODE_WRITE)) {
5362                 ret = -EBADF;
5363                 goto out_acct;
5364         }
5365
5366         if (compat) {
5367 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5368                 struct btrfs_ioctl_encoded_io_args_32 args32;
5369
5370                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5371                         ret = -EFAULT;
5372                         goto out_acct;
5373                 }
5374                 args.iov = compat_ptr(args32.iov);
5375                 args.iovcnt = args32.iovcnt;
5376                 args.offset = args32.offset;
5377                 args.flags = args32.flags;
5378                 args.len = args32.len;
5379                 args.unencoded_len = args32.unencoded_len;
5380                 args.unencoded_offset = args32.unencoded_offset;
5381                 args.compression = args32.compression;
5382                 args.encryption = args32.encryption;
5383                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5384 #else
5385                 return -ENOTTY;
5386 #endif
5387         } else {
5388                 if (copy_from_user(&args, argp, sizeof(args))) {
5389                         ret = -EFAULT;
5390                         goto out_acct;
5391                 }
5392         }
5393
5394         ret = -EINVAL;
5395         if (args.flags != 0)
5396                 goto out_acct;
5397         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5398                 goto out_acct;
5399         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5400             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5401                 goto out_acct;
5402         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5403             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5404                 goto out_acct;
5405         if (args.unencoded_offset > args.unencoded_len)
5406                 goto out_acct;
5407         if (args.len > args.unencoded_len - args.unencoded_offset)
5408                 goto out_acct;
5409
5410         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5411                            &iov, &iter);
5412         if (ret < 0)
5413                 goto out_acct;
5414
5415         file_start_write(file);
5416
5417         if (iov_iter_count(&iter) == 0) {
5418                 ret = 0;
5419                 goto out_end_write;
5420         }
5421         pos = args.offset;
5422         ret = rw_verify_area(WRITE, file, &pos, args.len);
5423         if (ret < 0)
5424                 goto out_end_write;
5425
5426         init_sync_kiocb(&kiocb, file);
5427         ret = kiocb_set_rw_flags(&kiocb, 0);
5428         if (ret)
5429                 goto out_end_write;
5430         kiocb.ki_pos = pos;
5431
5432         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5433         if (ret > 0)
5434                 fsnotify_modify(file);
5435
5436 out_end_write:
5437         file_end_write(file);
5438         kfree(iov);
5439 out_acct:
5440         if (ret > 0)
5441                 add_wchar(current, ret);
5442         inc_syscw(current);
5443         return ret;
5444 }
5445
5446 long btrfs_ioctl(struct file *file, unsigned int
5447                 cmd, unsigned long arg)
5448 {
5449         struct inode *inode = file_inode(file);
5450         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5451         struct btrfs_root *root = BTRFS_I(inode)->root;
5452         void __user *argp = (void __user *)arg;
5453
5454         switch (cmd) {
5455         case FS_IOC_GETVERSION:
5456                 return btrfs_ioctl_getversion(inode, argp);
5457         case FS_IOC_GETFSLABEL:
5458                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5459         case FS_IOC_SETFSLABEL:
5460                 return btrfs_ioctl_set_fslabel(file, argp);
5461         case FITRIM:
5462                 return btrfs_ioctl_fitrim(fs_info, argp);
5463         case BTRFS_IOC_SNAP_CREATE:
5464                 return btrfs_ioctl_snap_create(file, argp, 0);
5465         case BTRFS_IOC_SNAP_CREATE_V2:
5466                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5467         case BTRFS_IOC_SUBVOL_CREATE:
5468                 return btrfs_ioctl_snap_create(file, argp, 1);
5469         case BTRFS_IOC_SUBVOL_CREATE_V2:
5470                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5471         case BTRFS_IOC_SNAP_DESTROY:
5472                 return btrfs_ioctl_snap_destroy(file, argp, false);
5473         case BTRFS_IOC_SNAP_DESTROY_V2:
5474                 return btrfs_ioctl_snap_destroy(file, argp, true);
5475         case BTRFS_IOC_SUBVOL_GETFLAGS:
5476                 return btrfs_ioctl_subvol_getflags(inode, argp);
5477         case BTRFS_IOC_SUBVOL_SETFLAGS:
5478                 return btrfs_ioctl_subvol_setflags(file, argp);
5479         case BTRFS_IOC_DEFAULT_SUBVOL:
5480                 return btrfs_ioctl_default_subvol(file, argp);
5481         case BTRFS_IOC_DEFRAG:
5482                 return btrfs_ioctl_defrag(file, NULL);
5483         case BTRFS_IOC_DEFRAG_RANGE:
5484                 return btrfs_ioctl_defrag(file, argp);
5485         case BTRFS_IOC_RESIZE:
5486                 return btrfs_ioctl_resize(file, argp);
5487         case BTRFS_IOC_ADD_DEV:
5488                 return btrfs_ioctl_add_dev(fs_info, argp);
5489         case BTRFS_IOC_RM_DEV:
5490                 return btrfs_ioctl_rm_dev(file, argp);
5491         case BTRFS_IOC_RM_DEV_V2:
5492                 return btrfs_ioctl_rm_dev_v2(file, argp);
5493         case BTRFS_IOC_FS_INFO:
5494                 return btrfs_ioctl_fs_info(fs_info, argp);
5495         case BTRFS_IOC_DEV_INFO:
5496                 return btrfs_ioctl_dev_info(fs_info, argp);
5497         case BTRFS_IOC_TREE_SEARCH:
5498                 return btrfs_ioctl_tree_search(inode, argp);
5499         case BTRFS_IOC_TREE_SEARCH_V2:
5500                 return btrfs_ioctl_tree_search_v2(inode, argp);
5501         case BTRFS_IOC_INO_LOOKUP:
5502                 return btrfs_ioctl_ino_lookup(root, argp);
5503         case BTRFS_IOC_INO_PATHS:
5504                 return btrfs_ioctl_ino_to_path(root, argp);
5505         case BTRFS_IOC_LOGICAL_INO:
5506                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5507         case BTRFS_IOC_LOGICAL_INO_V2:
5508                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5509         case BTRFS_IOC_SPACE_INFO:
5510                 return btrfs_ioctl_space_info(fs_info, argp);
5511         case BTRFS_IOC_SYNC: {
5512                 int ret;
5513
5514                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5515                 if (ret)
5516                         return ret;
5517                 ret = btrfs_sync_fs(inode->i_sb, 1);
5518                 /*
5519                  * The transaction thread may want to do more work,
5520                  * namely it pokes the cleaner kthread that will start
5521                  * processing uncleaned subvols.
5522                  */
5523                 wake_up_process(fs_info->transaction_kthread);
5524                 return ret;
5525         }
5526         case BTRFS_IOC_START_SYNC:
5527                 return btrfs_ioctl_start_sync(root, argp);
5528         case BTRFS_IOC_WAIT_SYNC:
5529                 return btrfs_ioctl_wait_sync(fs_info, argp);
5530         case BTRFS_IOC_SCRUB:
5531                 return btrfs_ioctl_scrub(file, argp);
5532         case BTRFS_IOC_SCRUB_CANCEL:
5533                 return btrfs_ioctl_scrub_cancel(fs_info);
5534         case BTRFS_IOC_SCRUB_PROGRESS:
5535                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5536         case BTRFS_IOC_BALANCE_V2:
5537                 return btrfs_ioctl_balance(file, argp);
5538         case BTRFS_IOC_BALANCE_CTL:
5539                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5540         case BTRFS_IOC_BALANCE_PROGRESS:
5541                 return btrfs_ioctl_balance_progress(fs_info, argp);
5542         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5543                 return btrfs_ioctl_set_received_subvol(file, argp);
5544 #ifdef CONFIG_64BIT
5545         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5546                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5547 #endif
5548         case BTRFS_IOC_SEND:
5549                 return _btrfs_ioctl_send(inode, argp, false);
5550 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5551         case BTRFS_IOC_SEND_32:
5552                 return _btrfs_ioctl_send(inode, argp, true);
5553 #endif
5554         case BTRFS_IOC_GET_DEV_STATS:
5555                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5556         case BTRFS_IOC_QUOTA_CTL:
5557                 return btrfs_ioctl_quota_ctl(file, argp);
5558         case BTRFS_IOC_QGROUP_ASSIGN:
5559                 return btrfs_ioctl_qgroup_assign(file, argp);
5560         case BTRFS_IOC_QGROUP_CREATE:
5561                 return btrfs_ioctl_qgroup_create(file, argp);
5562         case BTRFS_IOC_QGROUP_LIMIT:
5563                 return btrfs_ioctl_qgroup_limit(file, argp);
5564         case BTRFS_IOC_QUOTA_RESCAN:
5565                 return btrfs_ioctl_quota_rescan(file, argp);
5566         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5567                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5568         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5569                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5570         case BTRFS_IOC_DEV_REPLACE:
5571                 return btrfs_ioctl_dev_replace(fs_info, argp);
5572         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5573                 return btrfs_ioctl_get_supported_features(argp);
5574         case BTRFS_IOC_GET_FEATURES:
5575                 return btrfs_ioctl_get_features(fs_info, argp);
5576         case BTRFS_IOC_SET_FEATURES:
5577                 return btrfs_ioctl_set_features(file, argp);
5578         case BTRFS_IOC_GET_SUBVOL_INFO:
5579                 return btrfs_ioctl_get_subvol_info(inode, argp);
5580         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5581                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5582         case BTRFS_IOC_INO_LOOKUP_USER:
5583                 return btrfs_ioctl_ino_lookup_user(file, argp);
5584         case FS_IOC_ENABLE_VERITY:
5585                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5586         case FS_IOC_MEASURE_VERITY:
5587                 return fsverity_ioctl_measure(file, argp);
5588         case BTRFS_IOC_ENCODED_READ:
5589                 return btrfs_ioctl_encoded_read(file, argp, false);
5590         case BTRFS_IOC_ENCODED_WRITE:
5591                 return btrfs_ioctl_encoded_write(file, argp, false);
5592 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5593         case BTRFS_IOC_ENCODED_READ_32:
5594                 return btrfs_ioctl_encoded_read(file, argp, true);
5595         case BTRFS_IOC_ENCODED_WRITE_32:
5596                 return btrfs_ioctl_encoded_write(file, argp, true);
5597 #endif
5598         }
5599
5600         return -ENOTTY;
5601 }
5602
5603 #ifdef CONFIG_COMPAT
5604 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5605 {
5606         /*
5607          * These all access 32-bit values anyway so no further
5608          * handling is necessary.
5609          */
5610         switch (cmd) {
5611         case FS_IOC32_GETVERSION:
5612                 cmd = FS_IOC_GETVERSION;
5613                 break;
5614         }
5615
5616         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5617 }
5618 #endif