fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include <linux/fileattr.h>
  30 #include <linux/fsverity.h>
  31 #include <linux/sched/xacct.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "export.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "print-tree.h"
  38 #include "volumes.h"
  39 #include "locking.h"
  40 #include "backref.h"
  41 #include "rcu-string.h"
  42 #include "send.h"
  43 #include "dev-replace.h"
  44 #include "props.h"
  45 #include "sysfs.h"
  46 #include "qgroup.h"
  47 #include "tree-log.h"
  48 #include "compression.h"
  49 #include "space-info.h"
  50 #include "delalloc-space.h"
  51 #include "block-group.h"
  52 #include "subpage.h"
  53
  54 #ifdef CONFIG_64BIT
  55 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  56  * structures are incorrect, as the timespec structure from userspace
  57  * is 4 bytes too small. We define these alternatives here to teach
  58  * the kernel about the 32-bit struct packing.
  59  */
  60 struct btrfs_ioctl_timespec_32 {
  61         __u64 sec;
  62         __u32 nsec;
  63 } __attribute__ ((__packed__));
  64
  65 struct btrfs_ioctl_received_subvol_args_32 {
  66         char    uuid[BTRFS_UUID_SIZE];  /* in */
  67         __u64   stransid;               /* in */
  68         __u64   rtransid;               /* out */
  69         struct btrfs_ioctl_timespec_32 stime; /* in */
  70         struct btrfs_ioctl_timespec_32 rtime; /* out */
  71         __u64   flags;                  /* in */
  72         __u64   reserved[16];           /* in */
  73 } __attribute__ ((__packed__));
  74
  75 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  76                                 struct btrfs_ioctl_received_subvol_args_32)
  77 #endif
  78
  79 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  80 struct btrfs_ioctl_send_args_32 {
  81         __s64 send_fd;                  /* in */
  82         __u64 clone_sources_count;      /* in */
  83         compat_uptr_t clone_sources;    /* in */
  84         __u64 parent_root;              /* in */
  85         __u64 flags;                    /* in */
  86         __u32 version;                  /* in */
  87         __u8  reserved[28];             /* in */
  88 } __attribute__ ((__packed__));
  89
  90 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  91                                struct btrfs_ioctl_send_args_32)
  92
  93 struct btrfs_ioctl_encoded_io_args_32 {
  94         compat_uptr_t iov;
  95         compat_ulong_t iovcnt;
  96         __s64 offset;
  97         __u64 flags;
  98         __u64 len;
  99         __u64 unencoded_len;
 100         __u64 unencoded_offset;
 101         __u32 compression;
 102         __u32 encryption;
 103         __u8 reserved[64];
 104 };
 105
 106 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
 107                                        struct btrfs_ioctl_encoded_io_args_32)
 108 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
 109                                         struct btrfs_ioctl_encoded_io_args_32)
 110 #endif
 111
 112 /* Mask out flags that are inappropriate for the given type of inode. */
 113 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
 114                 unsigned int flags)
 115 {
 116         if (S_ISDIR(inode->i_mode))
 117                 return flags;
 118         else if (S_ISREG(inode->i_mode))
 119                 return flags & ~FS_DIRSYNC_FL;
 120         else
 121                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 122 }
 123
 124 /*
 125  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 126  * ioctl.
 127  */
 128 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 129 {
 130         unsigned int iflags = 0;
 131         u32 flags = binode->flags;
 132         u32 ro_flags = binode->ro_flags;
 133
 134         if (flags & BTRFS_INODE_SYNC)
 135                 iflags |= FS_SYNC_FL;
 136         if (flags & BTRFS_INODE_IMMUTABLE)
 137                 iflags |= FS_IMMUTABLE_FL;
 138         if (flags & BTRFS_INODE_APPEND)
 139                 iflags |= FS_APPEND_FL;
 140         if (flags & BTRFS_INODE_NODUMP)
 141                 iflags |= FS_NODUMP_FL;
 142         if (flags & BTRFS_INODE_NOATIME)
 143                 iflags |= FS_NOATIME_FL;
 144         if (flags & BTRFS_INODE_DIRSYNC)
 145                 iflags |= FS_DIRSYNC_FL;
 146         if (flags & BTRFS_INODE_NODATACOW)
 147                 iflags |= FS_NOCOW_FL;
 148         if (ro_flags & BTRFS_INODE_RO_VERITY)
 149                 iflags |= FS_VERITY_FL;
 150
 151         if (flags & BTRFS_INODE_NOCOMPRESS)
 152                 iflags |= FS_NOCOMP_FL;
 153         else if (flags & BTRFS_INODE_COMPRESS)
 154                 iflags |= FS_COMPR_FL;
 155
 156         return iflags;
 157 }
 158
 159 /*
 160  * Update inode->i_flags based on the btrfs internal flags.
 161  */
 162 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 163 {
 164         struct btrfs_inode *binode = BTRFS_I(inode);
 165         unsigned int new_fl = 0;
 166
 167         if (binode->flags & BTRFS_INODE_SYNC)
 168                 new_fl |= S_SYNC;
 169         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 170                 new_fl |= S_IMMUTABLE;
 171         if (binode->flags & BTRFS_INODE_APPEND)
 172                 new_fl |= S_APPEND;
 173         if (binode->flags & BTRFS_INODE_NOATIME)
 174                 new_fl |= S_NOATIME;
 175         if (binode->flags & BTRFS_INODE_DIRSYNC)
 176                 new_fl |= S_DIRSYNC;
 177         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 178                 new_fl |= S_VERITY;
 179
 180         set_mask_bits(&inode->i_flags,
 181                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 182                       S_VERITY, new_fl);
 183 }
 184
 185 /*
 186  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 187  * the old and new flags are not conflicting
 188  */
 189 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 190 {
 191         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 192                       FS_NOATIME_FL | FS_NODUMP_FL | \
 193                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 194                       FS_NOCOMP_FL | FS_COMPR_FL |
 195                       FS_NOCOW_FL))
 196                 return -EOPNOTSUPP;
 197
 198         /* COMPR and NOCOMP on new/old are valid */
 199         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 200                 return -EINVAL;
 201
 202         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 203                 return -EINVAL;
 204
 205         /* NOCOW and compression options are mutually exclusive */
 206         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 207                 return -EINVAL;
 208         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 215                                     unsigned int flags)
 216 {
 217         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 218                 return -EPERM;
 219
 220         return 0;
 221 }
 222
 223 /*
 224  * Set flags/xflags from the internal inode flags. The remaining items of
 225  * fsxattr are zeroed.
 226  */
 227 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 228 {
 229         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 230
 231         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 232         return 0;
 233 }
 234
 235 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 236                        struct dentry *dentry, struct fileattr *fa)
 237 {
 238         struct inode *inode = d_inode(dentry);
 239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 240         struct btrfs_inode *binode = BTRFS_I(inode);
 241         struct btrfs_root *root = binode->root;
 242         struct btrfs_trans_handle *trans;
 243         unsigned int fsflags, old_fsflags;
 244         int ret;
 245         const char *comp = NULL;
 246         u32 binode_flags;
 247
 248         if (btrfs_root_readonly(root))
 249                 return -EROFS;
 250
 251         if (fileattr_has_fsx(fa))
 252                 return -EOPNOTSUPP;
 253
 254         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 255         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 256         ret = check_fsflags(old_fsflags, fsflags);
 257         if (ret)
 258                 return ret;
 259
 260         ret = check_fsflags_compatible(fs_info, fsflags);
 261         if (ret)
 262                 return ret;
 263
 264         binode_flags = binode->flags;
 265         if (fsflags & FS_SYNC_FL)
 266                 binode_flags |= BTRFS_INODE_SYNC;
 267         else
 268                 binode_flags &= ~BTRFS_INODE_SYNC;
 269         if (fsflags & FS_IMMUTABLE_FL)
 270                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 271         else
 272                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 273         if (fsflags & FS_APPEND_FL)
 274                 binode_flags |= BTRFS_INODE_APPEND;
 275         else
 276                 binode_flags &= ~BTRFS_INODE_APPEND;
 277         if (fsflags & FS_NODUMP_FL)
 278                 binode_flags |= BTRFS_INODE_NODUMP;
 279         else
 280                 binode_flags &= ~BTRFS_INODE_NODUMP;
 281         if (fsflags & FS_NOATIME_FL)
 282                 binode_flags |= BTRFS_INODE_NOATIME;
 283         else
 284                 binode_flags &= ~BTRFS_INODE_NOATIME;
 285
 286         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 287         if (!fa->flags_valid) {
 288                 /* 1 item for the inode */
 289                 trans = btrfs_start_transaction(root, 1);
 290                 if (IS_ERR(trans))
 291                         return PTR_ERR(trans);
 292                 goto update_flags;
 293         }
 294
 295         if (fsflags & FS_DIRSYNC_FL)
 296                 binode_flags |= BTRFS_INODE_DIRSYNC;
 297         else
 298                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 299         if (fsflags & FS_NOCOW_FL) {
 300                 if (S_ISREG(inode->i_mode)) {
 301                         /*
 302                          * It's safe to turn csums off here, no extents exist.
 303                          * Otherwise we want the flag to reflect the real COW
 304                          * status of the file and will not set it.
 305                          */
 306                         if (inode->i_size == 0)
 307                                 binode_flags |= BTRFS_INODE_NODATACOW |
 308                                                 BTRFS_INODE_NODATASUM;
 309                 } else {
 310                         binode_flags |= BTRFS_INODE_NODATACOW;
 311                 }
 312         } else {
 313                 /*
 314                  * Revert back under same assumptions as above
 315                  */
 316                 if (S_ISREG(inode->i_mode)) {
 317                         if (inode->i_size == 0)
 318                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 319                                                   BTRFS_INODE_NODATASUM);
 320                 } else {
 321                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 322                 }
 323         }
 324
 325         /*
 326          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 327          * flag may be changed automatically if compression code won't make
 328          * things smaller.
 329          */
 330         if (fsflags & FS_NOCOMP_FL) {
 331                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 332                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 333         } else if (fsflags & FS_COMPR_FL) {
 334
 335                 if (IS_SWAPFILE(inode))
 336                         return -ETXTBSY;
 337
 338                 binode_flags |= BTRFS_INODE_COMPRESS;
 339                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 340
 341                 comp = btrfs_compress_type2str(fs_info->compress_type);
 342                 if (!comp || comp[0] == 0)
 343                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 344         } else {
 345                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 346         }
 347
 348         /*
 349          * 1 for inode item
 350          * 2 for properties
 351          */
 352         trans = btrfs_start_transaction(root, 3);
 353         if (IS_ERR(trans))
 354                 return PTR_ERR(trans);
 355
 356         if (comp) {
 357                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 358                                      strlen(comp), 0);
 359                 if (ret) {
 360                         btrfs_abort_transaction(trans, ret);
 361                         goto out_end_trans;
 362                 }
 363         } else {
 364                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 365                                      0, 0);
 366                 if (ret && ret != -ENODATA) {
 367                         btrfs_abort_transaction(trans, ret);
 368                         goto out_end_trans;
 369                 }
 370         }
 371
 372 update_flags:
 373         binode->flags = binode_flags;
 374         btrfs_sync_inode_flags_to_i_flags(inode);
 375         inode_inc_iversion(inode);
 376         inode->i_ctime = current_time(inode);
 377         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 378
 379  out_end_trans:
 380         btrfs_end_transaction(trans);
 381         return ret;
 382 }
 383
 384 /*
 385  * Start exclusive operation @type, return true on success
 386  */
 387 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 388                         enum btrfs_exclusive_operation type)
 389 {
 390         bool ret = false;
 391
 392         spin_lock(&fs_info->super_lock);
 393         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 394                 fs_info->exclusive_operation = type;
 395                 ret = true;
 396         }
 397         spin_unlock(&fs_info->super_lock);
 398
 399         return ret;
 400 }
 401
 402 /*
 403  * Conditionally allow to enter the exclusive operation in case it's compatible
 404  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 405  * btrfs_exclop_finish.
 406  *
 407  * Compatibility:
 408  * - the same type is already running
 409  * - when trying to add a device and balance has been paused
 410  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 411  *   must check the condition first that would allow none -> @type
 412  */
 413 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 414                                  enum btrfs_exclusive_operation type)
 415 {
 416         spin_lock(&fs_info->super_lock);
 417         if (fs_info->exclusive_operation == type ||
 418             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 419              type == BTRFS_EXCLOP_DEV_ADD))
 420                 return true;
 421
 422         spin_unlock(&fs_info->super_lock);
 423         return false;
 424 }
 425
 426 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 427 {
 428         spin_unlock(&fs_info->super_lock);
 429 }
 430
 431 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 432 {
 433         spin_lock(&fs_info->super_lock);
 434         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 435         spin_unlock(&fs_info->super_lock);
 436         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 437 }
 438
 439 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 440                           enum btrfs_exclusive_operation op)
 441 {
 442         switch (op) {
 443         case BTRFS_EXCLOP_BALANCE_PAUSED:
 444                 spin_lock(&fs_info->super_lock);
 445                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 446                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD);
 447                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 448                 spin_unlock(&fs_info->super_lock);
 449                 break;
 450         case BTRFS_EXCLOP_BALANCE:
 451                 spin_lock(&fs_info->super_lock);
 452                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 453                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 454                 spin_unlock(&fs_info->super_lock);
 455                 break;
 456         default:
 457                 btrfs_warn(fs_info,
 458                         "invalid exclop balance operation %d requested", op);
 459         }
 460 }
 461
 462 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
 463 {
 464         return put_user(inode->i_generation, arg);
 465 }
 466
 467 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 468                                         void __user *arg)
 469 {
 470         struct btrfs_device *device;
 471         struct fstrim_range range;
 472         u64 minlen = ULLONG_MAX;
 473         u64 num_devices = 0;
 474         int ret;
 475
 476         if (!capable(CAP_SYS_ADMIN))
 477                 return -EPERM;
 478
 479         /*
 480          * btrfs_trim_block_group() depends on space cache, which is not
 481          * available in zoned filesystem. So, disallow fitrim on a zoned
 482          * filesystem for now.
 483          */
 484         if (btrfs_is_zoned(fs_info))
 485                 return -EOPNOTSUPP;
 486
 487         /*
 488          * If the fs is mounted with nologreplay, which requires it to be
 489          * mounted in RO mode as well, we can not allow discard on free space
 490          * inside block groups, because log trees refer to extents that are not
 491          * pinned in a block group's free space cache (pinning the extents is
 492          * precisely the first phase of replaying a log tree).
 493          */
 494         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 495                 return -EROFS;
 496
 497         rcu_read_lock();
 498         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 499                                 dev_list) {
 500                 if (!device->bdev || !bdev_max_discard_sectors(device->bdev))
 501                         continue;
 502                 num_devices++;
 503                 minlen = min_t(u64, bdev_discard_granularity(device->bdev),
 504                                     minlen);
 505         }
 506         rcu_read_unlock();
 507
 508         if (!num_devices)
 509                 return -EOPNOTSUPP;
 510         if (copy_from_user(&range, arg, sizeof(range)))
 511                 return -EFAULT;
 512
 513         /*
 514          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 515          * block group is in the logical address space, which can be any
 516          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 517          */
 518         if (range.len < fs_info->sb->s_blocksize)
 519                 return -EINVAL;
 520
 521         range.minlen = max(range.minlen, minlen);
 522         ret = btrfs_trim_fs(fs_info, &range);
 523         if (ret < 0)
 524                 return ret;
 525
 526         if (copy_to_user(arg, &range, sizeof(range)))
 527                 return -EFAULT;
 528
 529         return 0;
 530 }
 531
 532 int __pure btrfs_is_empty_uuid(u8 *uuid)
 533 {
 534         int i;
 535
 536         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 537                 if (uuid[i])
 538                         return 0;
 539         }
 540         return 1;
 541 }
 542
 543 static noinline int create_subvol(struct user_namespace *mnt_userns,
 544                                   struct inode *dir, struct dentry *dentry,
 545                                   const char *name, int namelen,
 546                                   struct btrfs_qgroup_inherit *inherit)
 547 {
 548         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 549         struct btrfs_trans_handle *trans;
 550         struct btrfs_key key;
 551         struct btrfs_root_item *root_item;
 552         struct btrfs_inode_item *inode_item;
 553         struct extent_buffer *leaf;
 554         struct btrfs_root *root = BTRFS_I(dir)->root;
 555         struct btrfs_root *new_root;
 556         struct btrfs_block_rsv block_rsv;
 557         struct timespec64 cur_time = current_time(dir);
 558         struct inode *inode;
 559         int ret;
 560         dev_t anon_dev = 0;
 561         u64 objectid;
 562         u64 index = 0;
 563
 564         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 565         if (!root_item)
 566                 return -ENOMEM;
 567
 568         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 569         if (ret)
 570                 goto fail_free;
 571
 572         ret = get_anon_bdev(&anon_dev);
 573         if (ret < 0)
 574                 goto fail_free;
 575
 576         /*
 577          * Don't create subvolume whose level is not zero. Or qgroup will be
 578          * screwed up since it assumes subvolume qgroup's level to be 0.
 579          */
 580         if (btrfs_qgroup_level(objectid)) {
 581                 ret = -ENOSPC;
 582                 goto fail_free;
 583         }
 584
 585         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 586         /*
 587          * The same as the snapshot creation, please see the comment
 588          * of create_snapshot().
 589          */
 590         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 591         if (ret)
 592                 goto fail_free;
 593
 594         trans = btrfs_start_transaction(root, 0);
 595         if (IS_ERR(trans)) {
 596                 ret = PTR_ERR(trans);
 597                 btrfs_subvolume_release_metadata(root, &block_rsv);
 598                 goto fail_free;
 599         }
 600         trans->block_rsv = &block_rsv;
 601         trans->bytes_reserved = block_rsv.size;
 602
 603         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 604         if (ret)
 605                 goto fail;
 606
 607         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 608                                       BTRFS_NESTING_NORMAL);
 609         if (IS_ERR(leaf)) {
 610                 ret = PTR_ERR(leaf);
 611                 goto fail;
 612         }
 613
 614         btrfs_mark_buffer_dirty(leaf);
 615
 616         inode_item = &root_item->inode;
 617         btrfs_set_stack_inode_generation(inode_item, 1);
 618         btrfs_set_stack_inode_size(inode_item, 3);
 619         btrfs_set_stack_inode_nlink(inode_item, 1);
 620         btrfs_set_stack_inode_nbytes(inode_item,
 621                                      fs_info->nodesize);
 622         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 623
 624         btrfs_set_root_flags(root_item, 0);
 625         btrfs_set_root_limit(root_item, 0);
 626         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 627
 628         btrfs_set_root_bytenr(root_item, leaf->start);
 629         btrfs_set_root_generation(root_item, trans->transid);
 630         btrfs_set_root_level(root_item, 0);
 631         btrfs_set_root_refs(root_item, 1);
 632         btrfs_set_root_used(root_item, leaf->len);
 633         btrfs_set_root_last_snapshot(root_item, 0);
 634
 635         btrfs_set_root_generation_v2(root_item,
 636                         btrfs_root_generation(root_item));
 637         generate_random_guid(root_item->uuid);
 638         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 639         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 640         root_item->ctime = root_item->otime;
 641         btrfs_set_root_ctransid(root_item, trans->transid);
 642         btrfs_set_root_otransid(root_item, trans->transid);
 643
 644         btrfs_tree_unlock(leaf);
 645
 646         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 647
 648         key.objectid = objectid;
 649         key.offset = 0;
 650         key.type = BTRFS_ROOT_ITEM_KEY;
 651         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 652                                 root_item);
 653         if (ret) {
 654                 /*
 655                  * Since we don't abort the transaction in this case, free the
 656                  * tree block so that we don't leak space and leave the
 657                  * filesystem in an inconsistent state (an extent item in the
 658                  * extent tree with a backreference for a root that does not
 659                  * exists).
 660                  */
 661                 btrfs_tree_lock(leaf);
 662                 btrfs_clean_tree_block(leaf);
 663                 btrfs_tree_unlock(leaf);
 664                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 665                 free_extent_buffer(leaf);
 666                 goto fail;
 667         }
 668
 669         free_extent_buffer(leaf);
 670         leaf = NULL;
 671
 672         key.offset = (u64)-1;
 673         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 674         if (IS_ERR(new_root)) {
 675                 free_anon_bdev(anon_dev);
 676                 ret = PTR_ERR(new_root);
 677                 btrfs_abort_transaction(trans, ret);
 678                 goto fail;
 679         }
 680         /* Freeing will be done in btrfs_put_root() of new_root */
 681         anon_dev = 0;
 682
 683         ret = btrfs_record_root_in_trans(trans, new_root);
 684         if (ret) {
 685                 btrfs_put_root(new_root);
 686                 btrfs_abort_transaction(trans, ret);
 687                 goto fail;
 688         }
 689
 690         ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns);
 691         btrfs_put_root(new_root);
 692         if (ret) {
 693                 /* We potentially lose an unused inode item here */
 694                 btrfs_abort_transaction(trans, ret);
 695                 goto fail;
 696         }
 697
 698         /*
 699          * insert the directory item
 700          */
 701         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 702         if (ret) {
 703                 btrfs_abort_transaction(trans, ret);
 704                 goto fail;
 705         }
 706
 707         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 708                                     BTRFS_FT_DIR, index);
 709         if (ret) {
 710                 btrfs_abort_transaction(trans, ret);
 711                 goto fail;
 712         }
 713
 714         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 715         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 716         if (ret) {
 717                 btrfs_abort_transaction(trans, ret);
 718                 goto fail;
 719         }
 720
 721         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 722                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 723         if (ret) {
 724                 btrfs_abort_transaction(trans, ret);
 725                 goto fail;
 726         }
 727
 728         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 729                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 730         if (ret)
 731                 btrfs_abort_transaction(trans, ret);
 732
 733 fail:
 734         kfree(root_item);
 735         trans->block_rsv = NULL;
 736         trans->bytes_reserved = 0;
 737         btrfs_subvolume_release_metadata(root, &block_rsv);
 738
 739         if (ret)
 740                 btrfs_end_transaction(trans);
 741         else
 742                 ret = btrfs_commit_transaction(trans);
 743
 744         if (!ret) {
 745                 inode = btrfs_lookup_dentry(dir, dentry);
 746                 if (IS_ERR(inode))
 747                         return PTR_ERR(inode);
 748                 d_instantiate(dentry, inode);
 749         }
 750         return ret;
 751
 752 fail_free:
 753         if (anon_dev)
 754                 free_anon_bdev(anon_dev);
 755         kfree(root_item);
 756         return ret;
 757 }
 758
 759 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 760                            struct dentry *dentry, bool readonly,
 761                            struct btrfs_qgroup_inherit *inherit)
 762 {
 763         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 764         struct inode *inode;
 765         struct btrfs_pending_snapshot *pending_snapshot;
 766         struct btrfs_trans_handle *trans;
 767         int ret;
 768
 769         /* We do not support snapshotting right now. */
 770         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 771                 btrfs_warn(fs_info,
 772                            "extent tree v2 doesn't support snapshotting yet");
 773                 return -EOPNOTSUPP;
 774         }
 775
 776         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 777                 return -EINVAL;
 778
 779         if (atomic_read(&root->nr_swapfiles)) {
 780                 btrfs_warn(fs_info,
 781                            "cannot snapshot subvolume with active swapfile");
 782                 return -ETXTBSY;
 783         }
 784
 785         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 786         if (!pending_snapshot)
 787                 return -ENOMEM;
 788
 789         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 790         if (ret < 0)
 791                 goto free_pending;
 792         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 793                         GFP_KERNEL);
 794         pending_snapshot->path = btrfs_alloc_path();
 795         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 796                 ret = -ENOMEM;
 797                 goto free_pending;
 798         }
 799
 800         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 801                              BTRFS_BLOCK_RSV_TEMP);
 802         /*
 803          * 1 - parent dir inode
 804          * 2 - dir entries
 805          * 1 - root item
 806          * 2 - root ref/backref
 807          * 1 - root of snapshot
 808          * 1 - UUID item
 809          */
 810         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 811                                         &pending_snapshot->block_rsv, 8,
 812                                         false);
 813         if (ret)
 814                 goto free_pending;
 815
 816         pending_snapshot->dentry = dentry;
 817         pending_snapshot->root = root;
 818         pending_snapshot->readonly = readonly;
 819         pending_snapshot->dir = dir;
 820         pending_snapshot->inherit = inherit;
 821
 822         trans = btrfs_start_transaction(root, 0);
 823         if (IS_ERR(trans)) {
 824                 ret = PTR_ERR(trans);
 825                 goto fail;
 826         }
 827
 828         trans->pending_snapshot = pending_snapshot;
 829
 830         ret = btrfs_commit_transaction(trans);
 831         if (ret)
 832                 goto fail;
 833
 834         ret = pending_snapshot->error;
 835         if (ret)
 836                 goto fail;
 837
 838         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 839         if (ret)
 840                 goto fail;
 841
 842         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 843         if (IS_ERR(inode)) {
 844                 ret = PTR_ERR(inode);
 845                 goto fail;
 846         }
 847
 848         d_instantiate(dentry, inode);
 849         ret = 0;
 850         pending_snapshot->anon_dev = 0;
 851 fail:
 852         /* Prevent double freeing of anon_dev */
 853         if (ret && pending_snapshot->snap)
 854                 pending_snapshot->snap->anon_dev = 0;
 855         btrfs_put_root(pending_snapshot->snap);
 856         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 857 free_pending:
 858         if (pending_snapshot->anon_dev)
 859                 free_anon_bdev(pending_snapshot->anon_dev);
 860         kfree(pending_snapshot->root_item);
 861         btrfs_free_path(pending_snapshot->path);
 862         kfree(pending_snapshot);
 863
 864         return ret;
 865 }
 866
 867 /*  copy of may_delete in fs/namei.c()
 868  *      Check whether we can remove a link victim from directory dir, check
 869  *  whether the type of victim is right.
 870  *  1. We can't do it if dir is read-only (done in permission())
 871  *  2. We should have write and exec permissions on dir
 872  *  3. We can't remove anything from append-only dir
 873  *  4. We can't do anything with immutable dir (done in permission())
 874  *  5. If the sticky bit on dir is set we should either
 875  *      a. be owner of dir, or
 876  *      b. be owner of victim, or
 877  *      c. have CAP_FOWNER capability
 878  *  6. If the victim is append-only or immutable we can't do anything with
 879  *     links pointing to it.
 880  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 881  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 882  *  9. We can't remove a root or mountpoint.
 883  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 884  *     nfs_async_unlink().
 885  */
 886
 887 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 888                             struct inode *dir, struct dentry *victim, int isdir)
 889 {
 890         int error;
 891
 892         if (d_really_is_negative(victim))
 893                 return -ENOENT;
 894
 895         BUG_ON(d_inode(victim->d_parent) != dir);
 896         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 897
 898         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 899         if (error)
 900                 return error;
 901         if (IS_APPEND(dir))
 902                 return -EPERM;
 903         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 904             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 905             IS_SWAPFILE(d_inode(victim)))
 906                 return -EPERM;
 907         if (isdir) {
 908                 if (!d_is_dir(victim))
 909                         return -ENOTDIR;
 910                 if (IS_ROOT(victim))
 911                         return -EBUSY;
 912         } else if (d_is_dir(victim))
 913                 return -EISDIR;
 914         if (IS_DEADDIR(dir))
 915                 return -ENOENT;
 916         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 917                 return -EBUSY;
 918         return 0;
 919 }
 920
 921 /* copy of may_create in fs/namei.c() */
 922 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 923                                    struct inode *dir, struct dentry *child)
 924 {
 925         if (d_really_is_positive(child))
 926                 return -EEXIST;
 927         if (IS_DEADDIR(dir))
 928                 return -ENOENT;
 929         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 930                 return -EOVERFLOW;
 931         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 932 }
 933
 934 /*
 935  * Create a new subvolume below @parent.  This is largely modeled after
 936  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 937  * inside this filesystem so it's quite a bit simpler.
 938  */
 939 static noinline int btrfs_mksubvol(const struct path *parent,
 940                                    struct user_namespace *mnt_userns,
 941                                    const char *name, int namelen,
 942                                    struct btrfs_root *snap_src,
 943                                    bool readonly,
 944                                    struct btrfs_qgroup_inherit *inherit)
 945 {
 946         struct inode *dir = d_inode(parent->dentry);
 947         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 948         struct dentry *dentry;
 949         int error;
 950
 951         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 952         if (error == -EINTR)
 953                 return error;
 954
 955         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 956         error = PTR_ERR(dentry);
 957         if (IS_ERR(dentry))
 958                 goto out_unlock;
 959
 960         error = btrfs_may_create(mnt_userns, dir, dentry);
 961         if (error)
 962                 goto out_dput;
 963
 964         /*
 965          * even if this name doesn't exist, we may get hash collisions.
 966          * check for them now when we can safely fail
 967          */
 968         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 969                                                dir->i_ino, name,
 970                                                namelen);
 971         if (error)
 972                 goto out_dput;
 973
 974         down_read(&fs_info->subvol_sem);
 975
 976         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 977                 goto out_up_read;
 978
 979         if (snap_src)
 980                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 981         else
 982                 error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit);
 983
 984         if (!error)
 985                 fsnotify_mkdir(dir, dentry);
 986 out_up_read:
 987         up_read(&fs_info->subvol_sem);
 988 out_dput:
 989         dput(dentry);
 990 out_unlock:
 991         btrfs_inode_unlock(dir, 0);
 992         return error;
 993 }
 994
 995 static noinline int btrfs_mksnapshot(const struct path *parent,
 996                                    struct user_namespace *mnt_userns,
 997                                    const char *name, int namelen,
 998                                    struct btrfs_root *root,
 999                                    bool readonly,
1000                                    struct btrfs_qgroup_inherit *inherit)
1001 {
1002         int ret;
1003         bool snapshot_force_cow = false;
1004
1005         /*
1006          * Force new buffered writes to reserve space even when NOCOW is
1007          * possible. This is to avoid later writeback (running dealloc) to
1008          * fallback to COW mode and unexpectedly fail with ENOSPC.
1009          */
1010         btrfs_drew_read_lock(&root->snapshot_lock);
1011
1012         ret = btrfs_start_delalloc_snapshot(root, false);
1013         if (ret)
1014                 goto out;
1015
1016         /*
1017          * All previous writes have started writeback in NOCOW mode, so now
1018          * we force future writes to fallback to COW mode during snapshot
1019          * creation.
1020          */
1021         atomic_inc(&root->snapshot_force_cow);
1022         snapshot_force_cow = true;
1023
1024         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1025
1026         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1027                              root, readonly, inherit);
1028 out:
1029         if (snapshot_force_cow)
1030                 atomic_dec(&root->snapshot_force_cow);
1031         btrfs_drew_read_unlock(&root->snapshot_lock);
1032         return ret;
1033 }
1034
1035 /*
1036  * Defrag specific helper to get an extent map.
1037  *
1038  * Differences between this and btrfs_get_extent() are:
1039  *
1040  * - No extent_map will be added to inode->extent_tree
1041  *   To reduce memory usage in the long run.
1042  *
1043  * - Extra optimization to skip file extents older than @newer_than
1044  *   By using btrfs_search_forward() we can skip entire file ranges that
1045  *   have extents created in past transactions, because btrfs_search_forward()
1046  *   will not visit leaves and nodes with a generation smaller than given
1047  *   minimal generation threshold (@newer_than).
1048  *
1049  * Return valid em if we find a file extent matching the requirement.
1050  * Return NULL if we can not find a file extent matching the requirement.
1051  *
1052  * Return ERR_PTR() for error.
1053  */
1054 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1055                                             u64 start, u64 newer_than)
1056 {
1057         struct btrfs_root *root = inode->root;
1058         struct btrfs_file_extent_item *fi;
1059         struct btrfs_path path = { 0 };
1060         struct extent_map *em;
1061         struct btrfs_key key;
1062         u64 ino = btrfs_ino(inode);
1063         int ret;
1064
1065         em = alloc_extent_map();
1066         if (!em) {
1067                 ret = -ENOMEM;
1068                 goto err;
1069         }
1070
1071         key.objectid = ino;
1072         key.type = BTRFS_EXTENT_DATA_KEY;
1073         key.offset = start;
1074
1075         if (newer_than) {
1076                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1077                 if (ret < 0)
1078                         goto err;
1079                 /* Can't find anything newer */
1080                 if (ret > 0)
1081                         goto not_found;
1082         } else {
1083                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1084                 if (ret < 0)
1085                         goto err;
1086         }
1087         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1088                 /*
1089                  * If btrfs_search_slot() makes path to point beyond nritems,
1090                  * we should not have an empty leaf, as this inode must at
1091                  * least have its INODE_ITEM.
1092                  */
1093                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1094                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1095         }
1096         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1097         /* Perfect match, no need to go one slot back */
1098         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1099             key.offset == start)
1100                 goto iterate;
1101
1102         /* We didn't find a perfect match, needs to go one slot back */
1103         if (path.slots[0] > 0) {
1104                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1105                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1106                         path.slots[0]--;
1107         }
1108
1109 iterate:
1110         /* Iterate through the path to find a file extent covering @start */
1111         while (true) {
1112                 u64 extent_end;
1113
1114                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1115                         goto next;
1116
1117                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1118
1119                 /*
1120                  * We may go one slot back to INODE_REF/XATTR item, then
1121                  * need to go forward until we reach an EXTENT_DATA.
1122                  * But we should still has the correct ino as key.objectid.
1123                  */
1124                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1125                         goto next;
1126
1127                 /* It's beyond our target range, definitely not extent found */
1128                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1129                         goto not_found;
1130
1131                 /*
1132                  *      |       |<- File extent ->|
1133                  *      \- start
1134                  *
1135                  * This means there is a hole between start and key.offset.
1136                  */
1137                 if (key.offset > start) {
1138                         em->start = start;
1139                         em->orig_start = start;
1140                         em->block_start = EXTENT_MAP_HOLE;
1141                         em->len = key.offset - start;
1142                         break;
1143                 }
1144
1145                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1146                                     struct btrfs_file_extent_item);
1147                 extent_end = btrfs_file_extent_end(&path);
1148
1149                 /*
1150                  *      |<- file extent ->|     |
1151                  *                              \- start
1152                  *
1153                  * We haven't reached start, search next slot.
1154                  */
1155                 if (extent_end <= start)
1156                         goto next;
1157
1158                 /* Now this extent covers @start, convert it to em */
1159                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1160                 break;
1161 next:
1162                 ret = btrfs_next_item(root, &path);
1163                 if (ret < 0)
1164                         goto err;
1165                 if (ret > 0)
1166                         goto not_found;
1167         }
1168         btrfs_release_path(&path);
1169         return em;
1170
1171 not_found:
1172         btrfs_release_path(&path);
1173         free_extent_map(em);
1174         return NULL;
1175
1176 err:
1177         btrfs_release_path(&path);
1178         free_extent_map(em);
1179         return ERR_PTR(ret);
1180 }
1181
1182 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1183                                                u64 newer_than, bool locked)
1184 {
1185         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1186         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1187         struct extent_map *em;
1188         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1189
1190         /*
1191          * hopefully we have this extent in the tree already, try without
1192          * the full extent lock
1193          */
1194         read_lock(&em_tree->lock);
1195         em = lookup_extent_mapping(em_tree, start, sectorsize);
1196         read_unlock(&em_tree->lock);
1197
1198         /*
1199          * We can get a merged extent, in that case, we need to re-search
1200          * tree to get the original em for defrag.
1201          *
1202          * If @newer_than is 0 or em::generation < newer_than, we can trust
1203          * this em, as either we don't care about the generation, or the
1204          * merged extent map will be rejected anyway.
1205          */
1206         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1207             newer_than && em->generation >= newer_than) {
1208                 free_extent_map(em);
1209                 em = NULL;
1210         }
1211
1212         if (!em) {
1213                 struct extent_state *cached = NULL;
1214                 u64 end = start + sectorsize - 1;
1215
1216                 /* get the big lock and read metadata off disk */
1217                 if (!locked)
1218                         lock_extent_bits(io_tree, start, end, &cached);
1219                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1220                 if (!locked)
1221                         unlock_extent_cached(io_tree, start, end, &cached);
1222
1223                 if (IS_ERR(em))
1224                         return NULL;
1225         }
1226
1227         return em;
1228 }
1229
1230 static u32 get_extent_max_capacity(const struct extent_map *em)
1231 {
1232         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1233                 return BTRFS_MAX_COMPRESSED;
1234         return BTRFS_MAX_EXTENT_SIZE;
1235 }
1236
1237 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1238                                      u32 extent_thresh, u64 newer_than, bool locked)
1239 {
1240         struct extent_map *next;
1241         bool ret = false;
1242
1243         /* this is the last extent */
1244         if (em->start + em->len >= i_size_read(inode))
1245                 return false;
1246
1247         /*
1248          * Here we need to pass @newer_then when checking the next extent, or
1249          * we will hit a case we mark current extent for defrag, but the next
1250          * one will not be a target.
1251          * This will just cause extra IO without really reducing the fragments.
1252          */
1253         next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked);
1254         /* No more em or hole */
1255         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1256                 goto out;
1257         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1258                 goto out;
1259         /*
1260          * If the next extent is at its max capacity, defragging current extent
1261          * makes no sense, as the total number of extents won't change.
1262          */
1263         if (next->len >= get_extent_max_capacity(em))
1264                 goto out;
1265         /* Skip older extent */
1266         if (next->generation < newer_than)
1267                 goto out;
1268         /* Also check extent size */
1269         if (next->len >= extent_thresh)
1270                 goto out;
1271
1272         ret = true;
1273 out:
1274         free_extent_map(next);
1275         return ret;
1276 }
1277
1278 /*
1279  * Prepare one page to be defragged.
1280  *
1281  * This will ensure:
1282  *
1283  * - Returned page is locked and has been set up properly.
1284  * - No ordered extent exists in the page.
1285  * - The page is uptodate.
1286  *
1287  * NOTE: Caller should also wait for page writeback after the cluster is
1288  * prepared, here we don't do writeback wait for each page.
1289  */
1290 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1291                                             pgoff_t index)
1292 {
1293         struct address_space *mapping = inode->vfs_inode.i_mapping;
1294         gfp_t mask = btrfs_alloc_write_mask(mapping);
1295         u64 page_start = (u64)index << PAGE_SHIFT;
1296         u64 page_end = page_start + PAGE_SIZE - 1;
1297         struct extent_state *cached_state = NULL;
1298         struct page *page;
1299         int ret;
1300
1301 again:
1302         page = find_or_create_page(mapping, index, mask);
1303         if (!page)
1304                 return ERR_PTR(-ENOMEM);
1305
1306         /*
1307          * Since we can defragment files opened read-only, we can encounter
1308          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1309          * can't do I/O using huge pages yet, so return an error for now.
1310          * Filesystem transparent huge pages are typically only used for
1311          * executables that explicitly enable them, so this isn't very
1312          * restrictive.
1313          */
1314         if (PageCompound(page)) {
1315                 unlock_page(page);
1316                 put_page(page);
1317                 return ERR_PTR(-ETXTBSY);
1318         }
1319
1320         ret = set_page_extent_mapped(page);
1321         if (ret < 0) {
1322                 unlock_page(page);
1323                 put_page(page);
1324                 return ERR_PTR(ret);
1325         }
1326
1327         /* Wait for any existing ordered extent in the range */
1328         while (1) {
1329                 struct btrfs_ordered_extent *ordered;
1330
1331                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1332                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1333                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1334                                      &cached_state);
1335                 if (!ordered)
1336                         break;
1337
1338                 unlock_page(page);
1339                 btrfs_start_ordered_extent(ordered, 1);
1340                 btrfs_put_ordered_extent(ordered);
1341                 lock_page(page);
1342                 /*
1343                  * We unlocked the page above, so we need check if it was
1344                  * released or not.
1345                  */
1346                 if (page->mapping != mapping || !PagePrivate(page)) {
1347                         unlock_page(page);
1348                         put_page(page);
1349                         goto again;
1350                 }
1351         }
1352
1353         /*
1354          * Now the page range has no ordered extent any more.  Read the page to
1355          * make it uptodate.
1356          */
1357         if (!PageUptodate(page)) {
1358                 btrfs_readpage(NULL, page);
1359                 lock_page(page);
1360                 if (page->mapping != mapping || !PagePrivate(page)) {
1361                         unlock_page(page);
1362                         put_page(page);
1363                         goto again;
1364                 }
1365                 if (!PageUptodate(page)) {
1366                         unlock_page(page);
1367                         put_page(page);
1368                         return ERR_PTR(-EIO);
1369                 }
1370         }
1371         return page;
1372 }
1373
1374 struct defrag_target_range {
1375         struct list_head list;
1376         u64 start;
1377         u64 len;
1378 };
1379
1380 /*
1381  * Collect all valid target extents.
1382  *
1383  * @start:         file offset to lookup
1384  * @len:           length to lookup
1385  * @extent_thresh: file extent size threshold, any extent size >= this value
1386  *                 will be ignored
1387  * @newer_than:    only defrag extents newer than this value
1388  * @do_compress:   whether the defrag is doing compression
1389  *                 if true, @extent_thresh will be ignored and all regular
1390  *                 file extents meeting @newer_than will be targets.
1391  * @locked:        if the range has already held extent lock
1392  * @target_list:   list of targets file extents
1393  */
1394 static int defrag_collect_targets(struct btrfs_inode *inode,
1395                                   u64 start, u64 len, u32 extent_thresh,
1396                                   u64 newer_than, bool do_compress,
1397                                   bool locked, struct list_head *target_list,
1398                                   u64 *last_scanned_ret)
1399 {
1400         bool last_is_target = false;
1401         u64 cur = start;
1402         int ret = 0;
1403
1404         while (cur < start + len) {
1405                 struct extent_map *em;
1406                 struct defrag_target_range *new;
1407                 bool next_mergeable = true;
1408                 u64 range_len;
1409
1410                 last_is_target = false;
1411                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1412                                           newer_than, locked);
1413                 if (!em)
1414                         break;
1415
1416                 /* Skip hole/inline/preallocated extents */
1417                 if (em->block_start >= EXTENT_MAP_LAST_BYTE ||
1418                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1419                         goto next;
1420
1421                 /* Skip older extent */
1422                 if (em->generation < newer_than)
1423                         goto next;
1424
1425                 /* This em is under writeback, no need to defrag */
1426                 if (em->generation == (u64)-1)
1427                         goto next;
1428
1429                 /*
1430                  * Our start offset might be in the middle of an existing extent
1431                  * map, so take that into account.
1432                  */
1433                 range_len = em->len - (cur - em->start);
1434                 /*
1435                  * If this range of the extent map is already flagged for delalloc,
1436                  * skip it, because:
1437                  *
1438                  * 1) We could deadlock later, when trying to reserve space for
1439                  *    delalloc, because in case we can't immediately reserve space
1440                  *    the flusher can start delalloc and wait for the respective
1441                  *    ordered extents to complete. The deadlock would happen
1442                  *    because we do the space reservation while holding the range
1443                  *    locked, and starting writeback, or finishing an ordered
1444                  *    extent, requires locking the range;
1445                  *
1446                  * 2) If there's delalloc there, it means there's dirty pages for
1447                  *    which writeback has not started yet (we clean the delalloc
1448                  *    flag when starting writeback and after creating an ordered
1449                  *    extent). If we mark pages in an adjacent range for defrag,
1450                  *    then we will have a larger contiguous range for delalloc,
1451                  *    very likely resulting in a larger extent after writeback is
1452                  *    triggered (except in a case of free space fragmentation).
1453                  */
1454                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1455                                    EXTENT_DELALLOC, 0, NULL))
1456                         goto next;
1457
1458                 /*
1459                  * For do_compress case, we want to compress all valid file
1460                  * extents, thus no @extent_thresh or mergeable check.
1461                  */
1462                 if (do_compress)
1463                         goto add;
1464
1465                 /* Skip too large extent */
1466                 if (range_len >= extent_thresh)
1467                         goto next;
1468
1469                 /*
1470                  * Skip extents already at its max capacity, this is mostly for
1471                  * compressed extents, which max cap is only 128K.
1472                  */
1473                 if (em->len >= get_extent_max_capacity(em))
1474                         goto next;
1475
1476                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1477                                                 extent_thresh, newer_than, locked);
1478                 if (!next_mergeable) {
1479                         struct defrag_target_range *last;
1480
1481                         /* Empty target list, no way to merge with last entry */
1482                         if (list_empty(target_list))
1483                                 goto next;
1484                         last = list_entry(target_list->prev,
1485                                           struct defrag_target_range, list);
1486                         /* Not mergeable with last entry */
1487                         if (last->start + last->len != cur)
1488                                 goto next;
1489
1490                         /* Mergeable, fall through to add it to @target_list. */
1491                 }
1492
1493 add:
1494                 last_is_target = true;
1495                 range_len = min(extent_map_end(em), start + len) - cur;
1496                 /*
1497                  * This one is a good target, check if it can be merged into
1498                  * last range of the target list.
1499                  */
1500                 if (!list_empty(target_list)) {
1501                         struct defrag_target_range *last;
1502
1503                         last = list_entry(target_list->prev,
1504                                           struct defrag_target_range, list);
1505                         ASSERT(last->start + last->len <= cur);
1506                         if (last->start + last->len == cur) {
1507                                 /* Mergeable, enlarge the last entry */
1508                                 last->len += range_len;
1509                                 goto next;
1510                         }
1511                         /* Fall through to allocate a new entry */
1512                 }
1513
1514                 /* Allocate new defrag_target_range */
1515                 new = kmalloc(sizeof(*new), GFP_NOFS);
1516                 if (!new) {
1517                         free_extent_map(em);
1518                         ret = -ENOMEM;
1519                         break;
1520                 }
1521                 new->start = cur;
1522                 new->len = range_len;
1523                 list_add_tail(&new->list, target_list);
1524
1525 next:
1526                 cur = extent_map_end(em);
1527                 free_extent_map(em);
1528         }
1529         if (ret < 0) {
1530                 struct defrag_target_range *entry;
1531                 struct defrag_target_range *tmp;
1532
1533                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1534                         list_del_init(&entry->list);
1535                         kfree(entry);
1536                 }
1537         }
1538         if (!ret && last_scanned_ret) {
1539                 /*
1540                  * If the last extent is not a target, the caller can skip to
1541                  * the end of that extent.
1542                  * Otherwise, we can only go the end of the specified range.
1543                  */
1544                 if (!last_is_target)
1545                         *last_scanned_ret = max(cur, *last_scanned_ret);
1546                 else
1547                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1548         }
1549         return ret;
1550 }
1551
1552 #define CLUSTER_SIZE    (SZ_256K)
1553 static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1554
1555 /*
1556  * Defrag one contiguous target range.
1557  *
1558  * @inode:      target inode
1559  * @target:     target range to defrag
1560  * @pages:      locked pages covering the defrag range
1561  * @nr_pages:   number of locked pages
1562  *
1563  * Caller should ensure:
1564  *
1565  * - Pages are prepared
1566  *   Pages should be locked, no ordered extent in the pages range,
1567  *   no writeback.
1568  *
1569  * - Extent bits are locked
1570  */
1571 static int defrag_one_locked_target(struct btrfs_inode *inode,
1572                                     struct defrag_target_range *target,
1573                                     struct page **pages, int nr_pages,
1574                                     struct extent_state **cached_state)
1575 {
1576         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1577         struct extent_changeset *data_reserved = NULL;
1578         const u64 start = target->start;
1579         const u64 len = target->len;
1580         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1581         unsigned long start_index = start >> PAGE_SHIFT;
1582         unsigned long first_index = page_index(pages[0]);
1583         int ret = 0;
1584         int i;
1585
1586         ASSERT(last_index - first_index + 1 <= nr_pages);
1587
1588         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1589         if (ret < 0)
1590                 return ret;
1591         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1592                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1593                          EXTENT_DEFRAG, 0, 0, cached_state);
1594         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1595
1596         /* Update the page status */
1597         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1598                 ClearPageChecked(pages[i]);
1599                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1600         }
1601         btrfs_delalloc_release_extents(inode, len);
1602         extent_changeset_free(data_reserved);
1603
1604         return ret;
1605 }
1606
1607 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1608                             u32 extent_thresh, u64 newer_than, bool do_compress,
1609                             u64 *last_scanned_ret)
1610 {
1611         struct extent_state *cached_state = NULL;
1612         struct defrag_target_range *entry;
1613         struct defrag_target_range *tmp;
1614         LIST_HEAD(target_list);
1615         struct page **pages;
1616         const u32 sectorsize = inode->root->fs_info->sectorsize;
1617         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1618         u64 start_index = start >> PAGE_SHIFT;
1619         unsigned int nr_pages = last_index - start_index + 1;
1620         int ret = 0;
1621         int i;
1622
1623         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1624         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1625
1626         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1627         if (!pages)
1628                 return -ENOMEM;
1629
1630         /* Prepare all pages */
1631         for (i = 0; i < nr_pages; i++) {
1632                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1633                 if (IS_ERR(pages[i])) {
1634                         ret = PTR_ERR(pages[i]);
1635                         pages[i] = NULL;
1636                         goto free_pages;
1637                 }
1638         }
1639         for (i = 0; i < nr_pages; i++)
1640                 wait_on_page_writeback(pages[i]);
1641
1642         /* Lock the pages range */
1643         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1644                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1645                          &cached_state);
1646         /*
1647          * Now we have a consistent view about the extent map, re-check
1648          * which range really needs to be defragged.
1649          *
1650          * And this time we have extent locked already, pass @locked = true
1651          * so that we won't relock the extent range and cause deadlock.
1652          */
1653         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1654                                      newer_than, do_compress, true,
1655                                      &target_list, last_scanned_ret);
1656         if (ret < 0)
1657                 goto unlock_extent;
1658
1659         list_for_each_entry(entry, &target_list, list) {
1660                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1661                                                &cached_state);
1662                 if (ret < 0)
1663                         break;
1664         }
1665
1666         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1667                 list_del_init(&entry->list);
1668                 kfree(entry);
1669         }
1670 unlock_extent:
1671         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1672                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1673                              &cached_state);
1674 free_pages:
1675         for (i = 0; i < nr_pages; i++) {
1676                 if (pages[i]) {
1677                         unlock_page(pages[i]);
1678                         put_page(pages[i]);
1679                 }
1680         }
1681         kfree(pages);
1682         return ret;
1683 }
1684
1685 static int defrag_one_cluster(struct btrfs_inode *inode,
1686                               struct file_ra_state *ra,
1687                               u64 start, u32 len, u32 extent_thresh,
1688                               u64 newer_than, bool do_compress,
1689                               unsigned long *sectors_defragged,
1690                               unsigned long max_sectors,
1691                               u64 *last_scanned_ret)
1692 {
1693         const u32 sectorsize = inode->root->fs_info->sectorsize;
1694         struct defrag_target_range *entry;
1695         struct defrag_target_range *tmp;
1696         LIST_HEAD(target_list);
1697         int ret;
1698
1699         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1700                                      newer_than, do_compress, false,
1701                                      &target_list, NULL);
1702         if (ret < 0)
1703                 goto out;
1704
1705         list_for_each_entry(entry, &target_list, list) {
1706                 u32 range_len = entry->len;
1707
1708                 /* Reached or beyond the limit */
1709                 if (max_sectors && *sectors_defragged >= max_sectors) {
1710                         ret = 1;
1711                         break;
1712                 }
1713
1714                 if (max_sectors)
1715                         range_len = min_t(u32, range_len,
1716                                 (max_sectors - *sectors_defragged) * sectorsize);
1717
1718                 /*
1719                  * If defrag_one_range() has updated last_scanned_ret,
1720                  * our range may already be invalid (e.g. hole punched).
1721                  * Skip if our range is before last_scanned_ret, as there is
1722                  * no need to defrag the range anymore.
1723                  */
1724                 if (entry->start + range_len <= *last_scanned_ret)
1725                         continue;
1726
1727                 if (ra)
1728                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1729                                 ra, NULL, entry->start >> PAGE_SHIFT,
1730                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1731                                 (entry->start >> PAGE_SHIFT) + 1);
1732                 /*
1733                  * Here we may not defrag any range if holes are punched before
1734                  * we locked the pages.
1735                  * But that's fine, it only affects the @sectors_defragged
1736                  * accounting.
1737                  */
1738                 ret = defrag_one_range(inode, entry->start, range_len,
1739                                        extent_thresh, newer_than, do_compress,
1740                                        last_scanned_ret);
1741                 if (ret < 0)
1742                         break;
1743                 *sectors_defragged += range_len >>
1744                                       inode->root->fs_info->sectorsize_bits;
1745         }
1746 out:
1747         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1748                 list_del_init(&entry->list);
1749                 kfree(entry);
1750         }
1751         if (ret >= 0)
1752                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1753         return ret;
1754 }
1755
1756 /*
1757  * Entry point to file defragmentation.
1758  *
1759  * @inode:         inode to be defragged
1760  * @ra:            readahead state (can be NUL)
1761  * @range:         defrag options including range and flags
1762  * @newer_than:    minimum transid to defrag
1763  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1764  *                 will be defragged.
1765  *
1766  * Return <0 for error.
1767  * Return >=0 for the number of sectors defragged, and range->start will be updated
1768  * to indicate the file offset where next defrag should be started at.
1769  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1770  *  defragging all the range).
1771  */
1772 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1773                       struct btrfs_ioctl_defrag_range_args *range,
1774                       u64 newer_than, unsigned long max_to_defrag)
1775 {
1776         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1777         unsigned long sectors_defragged = 0;
1778         u64 isize = i_size_read(inode);
1779         u64 cur;
1780         u64 last_byte;
1781         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1782         bool ra_allocated = false;
1783         int compress_type = BTRFS_COMPRESS_ZLIB;
1784         int ret = 0;
1785         u32 extent_thresh = range->extent_thresh;
1786         pgoff_t start_index;
1787
1788         if (isize == 0)
1789                 return 0;
1790
1791         if (range->start >= isize)
1792                 return -EINVAL;
1793
1794         if (do_compress) {
1795                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1796                         return -EINVAL;
1797                 if (range->compress_type)
1798                         compress_type = range->compress_type;
1799         }
1800
1801         if (extent_thresh == 0)
1802                 extent_thresh = SZ_256K;
1803
1804         if (range->start + range->len > range->start) {
1805                 /* Got a specific range */
1806                 last_byte = min(isize, range->start + range->len);
1807         } else {
1808                 /* Defrag until file end */
1809                 last_byte = isize;
1810         }
1811
1812         /* Align the range */
1813         cur = round_down(range->start, fs_info->sectorsize);
1814         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1815
1816         /*
1817          * If we were not given a ra, allocate a readahead context. As
1818          * readahead is just an optimization, defrag will work without it so
1819          * we don't error out.
1820          */
1821         if (!ra) {
1822                 ra_allocated = true;
1823                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1824                 if (ra)
1825                         file_ra_state_init(ra, inode->i_mapping);
1826         }
1827
1828         /*
1829          * Make writeback start from the beginning of the range, so that the
1830          * defrag range can be written sequentially.
1831          */
1832         start_index = cur >> PAGE_SHIFT;
1833         if (start_index < inode->i_mapping->writeback_index)
1834                 inode->i_mapping->writeback_index = start_index;
1835
1836         while (cur < last_byte) {
1837                 const unsigned long prev_sectors_defragged = sectors_defragged;
1838                 u64 last_scanned = cur;
1839                 u64 cluster_end;
1840
1841                 if (btrfs_defrag_cancelled(fs_info)) {
1842                         ret = -EAGAIN;
1843                         break;
1844                 }
1845
1846                 /* We want the cluster end at page boundary when possible */
1847                 cluster_end = (((cur >> PAGE_SHIFT) +
1848                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1849                 cluster_end = min(cluster_end, last_byte);
1850
1851                 btrfs_inode_lock(inode, 0);
1852                 if (IS_SWAPFILE(inode)) {
1853                         ret = -ETXTBSY;
1854                         btrfs_inode_unlock(inode, 0);
1855                         break;
1856                 }
1857                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1858                         btrfs_inode_unlock(inode, 0);
1859                         break;
1860                 }
1861                 if (do_compress)
1862                         BTRFS_I(inode)->defrag_compress = compress_type;
1863                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1864                                 cluster_end + 1 - cur, extent_thresh,
1865                                 newer_than, do_compress, &sectors_defragged,
1866                                 max_to_defrag, &last_scanned);
1867
1868                 if (sectors_defragged > prev_sectors_defragged)
1869                         balance_dirty_pages_ratelimited(inode->i_mapping);
1870
1871                 btrfs_inode_unlock(inode, 0);
1872                 if (ret < 0)
1873                         break;
1874                 cur = max(cluster_end + 1, last_scanned);
1875                 if (ret > 0) {
1876                         ret = 0;
1877                         break;
1878                 }
1879                 cond_resched();
1880         }
1881
1882         if (ra_allocated)
1883                 kfree(ra);
1884         /*
1885          * Update range.start for autodefrag, this will indicate where to start
1886          * in next run.
1887          */
1888         range->start = cur;
1889         if (sectors_defragged) {
1890                 /*
1891                  * We have defragged some sectors, for compression case they
1892                  * need to be written back immediately.
1893                  */
1894                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1895                         filemap_flush(inode->i_mapping);
1896                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1897                                      &BTRFS_I(inode)->runtime_flags))
1898                                 filemap_flush(inode->i_mapping);
1899                 }
1900                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1901                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1902                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1903                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1904                 ret = sectors_defragged;
1905         }
1906         if (do_compress) {
1907                 btrfs_inode_lock(inode, 0);
1908                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1909                 btrfs_inode_unlock(inode, 0);
1910         }
1911         return ret;
1912 }
1913
1914 /*
1915  * Try to start exclusive operation @type or cancel it if it's running.
1916  *
1917  * Return:
1918  *   0        - normal mode, newly claimed op started
1919  *  >0        - normal mode, something else is running,
1920  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1921  * ECANCELED  - cancel mode, successful cancel
1922  * ENOTCONN   - cancel mode, operation not running anymore
1923  */
1924 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1925                         enum btrfs_exclusive_operation type, bool cancel)
1926 {
1927         if (!cancel) {
1928                 /* Start normal op */
1929                 if (!btrfs_exclop_start(fs_info, type))
1930                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1931                 /* Exclusive operation is now claimed */
1932                 return 0;
1933         }
1934
1935         /* Cancel running op */
1936         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1937                 /*
1938                  * This blocks any exclop finish from setting it to NONE, so we
1939                  * request cancellation. Either it runs and we will wait for it,
1940                  * or it has finished and no waiting will happen.
1941                  */
1942                 atomic_inc(&fs_info->reloc_cancel_req);
1943                 btrfs_exclop_start_unlock(fs_info);
1944
1945                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1946                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1947                                     TASK_INTERRUPTIBLE);
1948
1949                 return -ECANCELED;
1950         }
1951
1952         /* Something else is running or none */
1953         return -ENOTCONN;
1954 }
1955
1956 static noinline int btrfs_ioctl_resize(struct file *file,
1957                                         void __user *arg)
1958 {
1959         BTRFS_DEV_LOOKUP_ARGS(args);
1960         struct inode *inode = file_inode(file);
1961         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1962         u64 new_size;
1963         u64 old_size;
1964         u64 devid = 1;
1965         struct btrfs_root *root = BTRFS_I(inode)->root;
1966         struct btrfs_ioctl_vol_args *vol_args;
1967         struct btrfs_trans_handle *trans;
1968         struct btrfs_device *device = NULL;
1969         char *sizestr;
1970         char *retptr;
1971         char *devstr = NULL;
1972         int ret = 0;
1973         int mod = 0;
1974         bool cancel;
1975
1976         if (!capable(CAP_SYS_ADMIN))
1977                 return -EPERM;
1978
1979         ret = mnt_want_write_file(file);
1980         if (ret)
1981                 return ret;
1982
1983         /*
1984          * Read the arguments before checking exclusivity to be able to
1985          * distinguish regular resize and cancel
1986          */
1987         vol_args = memdup_user(arg, sizeof(*vol_args));
1988         if (IS_ERR(vol_args)) {
1989                 ret = PTR_ERR(vol_args);
1990                 goto out_drop;
1991         }
1992         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1993         sizestr = vol_args->name;
1994         cancel = (strcmp("cancel", sizestr) == 0);
1995         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1996         if (ret)
1997                 goto out_free;
1998         /* Exclusive operation is now claimed */
1999
2000         devstr = strchr(sizestr, ':');
2001         if (devstr) {
2002                 sizestr = devstr + 1;
2003                 *devstr = '\0';
2004                 devstr = vol_args->name;
2005                 ret = kstrtoull(devstr, 10, &devid);
2006                 if (ret)
2007                         goto out_finish;
2008                 if (!devid) {
2009                         ret = -EINVAL;
2010                         goto out_finish;
2011                 }
2012                 btrfs_info(fs_info, "resizing devid %llu", devid);
2013         }
2014
2015         args.devid = devid;
2016         device = btrfs_find_device(fs_info->fs_devices, &args);
2017         if (!device) {
2018                 btrfs_info(fs_info, "resizer unable to find device %llu",
2019                            devid);
2020                 ret = -ENODEV;
2021                 goto out_finish;
2022         }
2023
2024         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2025                 btrfs_info(fs_info,
2026                            "resizer unable to apply on readonly device %llu",
2027                        devid);
2028                 ret = -EPERM;
2029                 goto out_finish;
2030         }
2031
2032         if (!strcmp(sizestr, "max"))
2033                 new_size = bdev_nr_bytes(device->bdev);
2034         else {
2035                 if (sizestr[0] == '-') {
2036                         mod = -1;
2037                         sizestr++;
2038                 } else if (sizestr[0] == '+') {
2039                         mod = 1;
2040                         sizestr++;
2041                 }
2042                 new_size = memparse(sizestr, &retptr);
2043                 if (*retptr != '\0' || new_size == 0) {
2044                         ret = -EINVAL;
2045                         goto out_finish;
2046                 }
2047         }
2048
2049         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2050                 ret = -EPERM;
2051                 goto out_finish;
2052         }
2053
2054         old_size = btrfs_device_get_total_bytes(device);
2055
2056         if (mod < 0) {
2057                 if (new_size > old_size) {
2058                         ret = -EINVAL;
2059                         goto out_finish;
2060                 }
2061                 new_size = old_size - new_size;
2062         } else if (mod > 0) {
2063                 if (new_size > ULLONG_MAX - old_size) {
2064                         ret = -ERANGE;
2065                         goto out_finish;
2066                 }
2067                 new_size = old_size + new_size;
2068         }
2069
2070         if (new_size < SZ_256M) {
2071                 ret = -EINVAL;
2072                 goto out_finish;
2073         }
2074         if (new_size > bdev_nr_bytes(device->bdev)) {
2075                 ret = -EFBIG;
2076                 goto out_finish;
2077         }
2078
2079         new_size = round_down(new_size, fs_info->sectorsize);
2080
2081         if (new_size > old_size) {
2082                 trans = btrfs_start_transaction(root, 0);
2083                 if (IS_ERR(trans)) {
2084                         ret = PTR_ERR(trans);
2085                         goto out_finish;
2086                 }
2087                 ret = btrfs_grow_device(trans, device, new_size);
2088                 btrfs_commit_transaction(trans);
2089         } else if (new_size < old_size) {
2090                 ret = btrfs_shrink_device(device, new_size);
2091         } /* equal, nothing need to do */
2092
2093         if (ret == 0 && new_size != old_size)
2094                 btrfs_info_in_rcu(fs_info,
2095                         "resize device %s (devid %llu) from %llu to %llu",
2096                         rcu_str_deref(device->name), device->devid,
2097                         old_size, new_size);
2098 out_finish:
2099         btrfs_exclop_finish(fs_info);
2100 out_free:
2101         kfree(vol_args);
2102 out_drop:
2103         mnt_drop_write_file(file);
2104         return ret;
2105 }
2106
2107 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2108                                 struct user_namespace *mnt_userns,
2109                                 const char *name, unsigned long fd, int subvol,
2110                                 bool readonly,
2111                                 struct btrfs_qgroup_inherit *inherit)
2112 {
2113         int namelen;
2114         int ret = 0;
2115
2116         if (!S_ISDIR(file_inode(file)->i_mode))
2117                 return -ENOTDIR;
2118
2119         ret = mnt_want_write_file(file);
2120         if (ret)
2121                 goto out;
2122
2123         namelen = strlen(name);
2124         if (strchr(name, '/')) {
2125                 ret = -EINVAL;
2126                 goto out_drop_write;
2127         }
2128
2129         if (name[0] == '.' &&
2130            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2131                 ret = -EEXIST;
2132                 goto out_drop_write;
2133         }
2134
2135         if (subvol) {
2136                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2137                                      namelen, NULL, readonly, inherit);
2138         } else {
2139                 struct fd src = fdget(fd);
2140                 struct inode *src_inode;
2141                 if (!src.file) {
2142                         ret = -EINVAL;
2143                         goto out_drop_write;
2144                 }
2145
2146                 src_inode = file_inode(src.file);
2147                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2148                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2149                                    "Snapshot src from another FS");
2150                         ret = -EXDEV;
2151                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2152                         /*
2153                          * Subvolume creation is not restricted, but snapshots
2154                          * are limited to own subvolumes only
2155                          */
2156                         ret = -EPERM;
2157                 } else {
2158                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2159                                                name, namelen,
2160                                                BTRFS_I(src_inode)->root,
2161                                                readonly, inherit);
2162                 }
2163                 fdput(src);
2164         }
2165 out_drop_write:
2166         mnt_drop_write_file(file);
2167 out:
2168         return ret;
2169 }
2170
2171 static noinline int btrfs_ioctl_snap_create(struct file *file,
2172                                             void __user *arg, int subvol)
2173 {
2174         struct btrfs_ioctl_vol_args *vol_args;
2175         int ret;
2176
2177         if (!S_ISDIR(file_inode(file)->i_mode))
2178                 return -ENOTDIR;
2179
2180         vol_args = memdup_user(arg, sizeof(*vol_args));
2181         if (IS_ERR(vol_args))
2182                 return PTR_ERR(vol_args);
2183         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2184
2185         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2186                                         vol_args->name, vol_args->fd, subvol,
2187                                         false, NULL);
2188
2189         kfree(vol_args);
2190         return ret;
2191 }
2192
2193 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2194                                                void __user *arg, int subvol)
2195 {
2196         struct btrfs_ioctl_vol_args_v2 *vol_args;
2197         int ret;
2198         bool readonly = false;
2199         struct btrfs_qgroup_inherit *inherit = NULL;
2200
2201         if (!S_ISDIR(file_inode(file)->i_mode))
2202                 return -ENOTDIR;
2203
2204         vol_args = memdup_user(arg, sizeof(*vol_args));
2205         if (IS_ERR(vol_args))
2206                 return PTR_ERR(vol_args);
2207         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2208
2209         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2210                 ret = -EOPNOTSUPP;
2211                 goto free_args;
2212         }
2213
2214         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2215                 readonly = true;
2216         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2217                 u64 nums;
2218
2219                 if (vol_args->size < sizeof(*inherit) ||
2220                     vol_args->size > PAGE_SIZE) {
2221                         ret = -EINVAL;
2222                         goto free_args;
2223                 }
2224                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2225                 if (IS_ERR(inherit)) {
2226                         ret = PTR_ERR(inherit);
2227                         goto free_args;
2228                 }
2229
2230                 if (inherit->num_qgroups > PAGE_SIZE ||
2231                     inherit->num_ref_copies > PAGE_SIZE ||
2232                     inherit->num_excl_copies > PAGE_SIZE) {
2233                         ret = -EINVAL;
2234                         goto free_inherit;
2235                 }
2236
2237                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2238                        2 * inherit->num_excl_copies;
2239                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2240                         ret = -EINVAL;
2241                         goto free_inherit;
2242                 }
2243         }
2244
2245         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2246                                         vol_args->name, vol_args->fd, subvol,
2247                                         readonly, inherit);
2248         if (ret)
2249                 goto free_inherit;
2250 free_inherit:
2251         kfree(inherit);
2252 free_args:
2253         kfree(vol_args);
2254         return ret;
2255 }
2256
2257 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
2258                                                 void __user *arg)
2259 {
2260         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2261         struct btrfs_root *root = BTRFS_I(inode)->root;
2262         int ret = 0;
2263         u64 flags = 0;
2264
2265         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2266                 return -EINVAL;
2267
2268         down_read(&fs_info->subvol_sem);
2269         if (btrfs_root_readonly(root))
2270                 flags |= BTRFS_SUBVOL_RDONLY;
2271         up_read(&fs_info->subvol_sem);
2272
2273         if (copy_to_user(arg, &flags, sizeof(flags)))
2274                 ret = -EFAULT;
2275
2276         return ret;
2277 }
2278
2279 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2280                                               void __user *arg)
2281 {
2282         struct inode *inode = file_inode(file);
2283         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2284         struct btrfs_root *root = BTRFS_I(inode)->root;
2285         struct btrfs_trans_handle *trans;
2286         u64 root_flags;
2287         u64 flags;
2288         int ret = 0;
2289
2290         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2291                 return -EPERM;
2292
2293         ret = mnt_want_write_file(file);
2294         if (ret)
2295                 goto out;
2296
2297         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2298                 ret = -EINVAL;
2299                 goto out_drop_write;
2300         }
2301
2302         if (copy_from_user(&flags, arg, sizeof(flags))) {
2303                 ret = -EFAULT;
2304                 goto out_drop_write;
2305         }
2306
2307         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2308                 ret = -EOPNOTSUPP;
2309                 goto out_drop_write;
2310         }
2311
2312         down_write(&fs_info->subvol_sem);
2313
2314         /* nothing to do */
2315         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2316                 goto out_drop_sem;
2317
2318         root_flags = btrfs_root_flags(&root->root_item);
2319         if (flags & BTRFS_SUBVOL_RDONLY) {
2320                 btrfs_set_root_flags(&root->root_item,
2321                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2322         } else {
2323                 /*
2324                  * Block RO -> RW transition if this subvolume is involved in
2325                  * send
2326                  */
2327                 spin_lock(&root->root_item_lock);
2328                 if (root->send_in_progress == 0) {
2329                         btrfs_set_root_flags(&root->root_item,
2330                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2331                         spin_unlock(&root->root_item_lock);
2332                 } else {
2333                         spin_unlock(&root->root_item_lock);
2334                         btrfs_warn(fs_info,
2335                                    "Attempt to set subvolume %llu read-write during send",
2336                                    root->root_key.objectid);
2337                         ret = -EPERM;
2338                         goto out_drop_sem;
2339                 }
2340         }
2341
2342         trans = btrfs_start_transaction(root, 1);
2343         if (IS_ERR(trans)) {
2344                 ret = PTR_ERR(trans);
2345                 goto out_reset;
2346         }
2347
2348         ret = btrfs_update_root(trans, fs_info->tree_root,
2349                                 &root->root_key, &root->root_item);
2350         if (ret < 0) {
2351                 btrfs_end_transaction(trans);
2352                 goto out_reset;
2353         }
2354
2355         ret = btrfs_commit_transaction(trans);
2356
2357 out_reset:
2358         if (ret)
2359                 btrfs_set_root_flags(&root->root_item, root_flags);
2360 out_drop_sem:
2361         up_write(&fs_info->subvol_sem);
2362 out_drop_write:
2363         mnt_drop_write_file(file);
2364 out:
2365         return ret;
2366 }
2367
2368 static noinline int key_in_sk(struct btrfs_key *key,
2369                               struct btrfs_ioctl_search_key *sk)
2370 {
2371         struct btrfs_key test;
2372         int ret;
2373
2374         test.objectid = sk->min_objectid;
2375         test.type = sk->min_type;
2376         test.offset = sk->min_offset;
2377
2378         ret = btrfs_comp_cpu_keys(key, &test);
2379         if (ret < 0)
2380                 return 0;
2381
2382         test.objectid = sk->max_objectid;
2383         test.type = sk->max_type;
2384         test.offset = sk->max_offset;
2385
2386         ret = btrfs_comp_cpu_keys(key, &test);
2387         if (ret > 0)
2388                 return 0;
2389         return 1;
2390 }
2391
2392 static noinline int copy_to_sk(struct btrfs_path *path,
2393                                struct btrfs_key *key,
2394                                struct btrfs_ioctl_search_key *sk,
2395                                size_t *buf_size,
2396                                char __user *ubuf,
2397                                unsigned long *sk_offset,
2398                                int *num_found)
2399 {
2400         u64 found_transid;
2401         struct extent_buffer *leaf;
2402         struct btrfs_ioctl_search_header sh;
2403         struct btrfs_key test;
2404         unsigned long item_off;
2405         unsigned long item_len;
2406         int nritems;
2407         int i;
2408         int slot;
2409         int ret = 0;
2410
2411         leaf = path->nodes[0];
2412         slot = path->slots[0];
2413         nritems = btrfs_header_nritems(leaf);
2414
2415         if (btrfs_header_generation(leaf) > sk->max_transid) {
2416                 i = nritems;
2417                 goto advance_key;
2418         }
2419         found_transid = btrfs_header_generation(leaf);
2420
2421         for (i = slot; i < nritems; i++) {
2422                 item_off = btrfs_item_ptr_offset(leaf, i);
2423                 item_len = btrfs_item_size(leaf, i);
2424
2425                 btrfs_item_key_to_cpu(leaf, key, i);
2426                 if (!key_in_sk(key, sk))
2427                         continue;
2428
2429                 if (sizeof(sh) + item_len > *buf_size) {
2430                         if (*num_found) {
2431                                 ret = 1;
2432                                 goto out;
2433                         }
2434
2435                         /*
2436                          * return one empty item back for v1, which does not
2437                          * handle -EOVERFLOW
2438                          */
2439
2440                         *buf_size = sizeof(sh) + item_len;
2441                         item_len = 0;
2442                         ret = -EOVERFLOW;
2443                 }
2444
2445                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2446                         ret = 1;
2447                         goto out;
2448                 }
2449
2450                 sh.objectid = key->objectid;
2451                 sh.offset = key->offset;
2452                 sh.type = key->type;
2453                 sh.len = item_len;
2454                 sh.transid = found_transid;
2455
2456                 /*
2457                  * Copy search result header. If we fault then loop again so we
2458                  * can fault in the pages and -EFAULT there if there's a
2459                  * problem. Otherwise we'll fault and then copy the buffer in
2460                  * properly this next time through
2461                  */
2462                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2463                         ret = 0;
2464                         goto out;
2465                 }
2466
2467                 *sk_offset += sizeof(sh);
2468
2469                 if (item_len) {
2470                         char __user *up = ubuf + *sk_offset;
2471                         /*
2472                          * Copy the item, same behavior as above, but reset the
2473                          * * sk_offset so we copy the full thing again.
2474                          */
2475                         if (read_extent_buffer_to_user_nofault(leaf, up,
2476                                                 item_off, item_len)) {
2477                                 ret = 0;
2478                                 *sk_offset -= sizeof(sh);
2479                                 goto out;
2480                         }
2481
2482                         *sk_offset += item_len;
2483                 }
2484                 (*num_found)++;
2485
2486                 if (ret) /* -EOVERFLOW from above */
2487                         goto out;
2488
2489                 if (*num_found >= sk->nr_items) {
2490                         ret = 1;
2491                         goto out;
2492                 }
2493         }
2494 advance_key:
2495         ret = 0;
2496         test.objectid = sk->max_objectid;
2497         test.type = sk->max_type;
2498         test.offset = sk->max_offset;
2499         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2500                 ret = 1;
2501         else if (key->offset < (u64)-1)
2502                 key->offset++;
2503         else if (key->type < (u8)-1) {
2504                 key->offset = 0;
2505                 key->type++;
2506         } else if (key->objectid < (u64)-1) {
2507                 key->offset = 0;
2508                 key->type = 0;
2509                 key->objectid++;
2510         } else
2511                 ret = 1;
2512 out:
2513         /*
2514          *  0: all items from this leaf copied, continue with next
2515          *  1: * more items can be copied, but unused buffer is too small
2516          *     * all items were found
2517          *     Either way, it will stops the loop which iterates to the next
2518          *     leaf
2519          *  -EOVERFLOW: item was to large for buffer
2520          *  -EFAULT: could not copy extent buffer back to userspace
2521          */
2522         return ret;
2523 }
2524
2525 static noinline int search_ioctl(struct inode *inode,
2526                                  struct btrfs_ioctl_search_key *sk,
2527                                  size_t *buf_size,
2528                                  char __user *ubuf)
2529 {
2530         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2531         struct btrfs_root *root;
2532         struct btrfs_key key;
2533         struct btrfs_path *path;
2534         int ret;
2535         int num_found = 0;
2536         unsigned long sk_offset = 0;
2537
2538         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2539                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2540                 return -EOVERFLOW;
2541         }
2542
2543         path = btrfs_alloc_path();
2544         if (!path)
2545                 return -ENOMEM;
2546
2547         if (sk->tree_id == 0) {
2548                 /* search the root of the inode that was passed */
2549                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2550         } else {
2551                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2552                 if (IS_ERR(root)) {
2553                         btrfs_free_path(path);
2554                         return PTR_ERR(root);
2555                 }
2556         }
2557
2558         key.objectid = sk->min_objectid;
2559         key.type = sk->min_type;
2560         key.offset = sk->min_offset;
2561
2562         while (1) {
2563                 ret = -EFAULT;
2564                 /*
2565                  * Ensure that the whole user buffer is faulted in at sub-page
2566                  * granularity, otherwise the loop may live-lock.
2567                  */
2568                 if (fault_in_subpage_writeable(ubuf + sk_offset,
2569                                                *buf_size - sk_offset))
2570                         break;
2571
2572                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2573                 if (ret != 0) {
2574                         if (ret > 0)
2575                                 ret = 0;
2576                         goto err;
2577                 }
2578                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2579                                  &sk_offset, &num_found);
2580                 btrfs_release_path(path);
2581                 if (ret)
2582                         break;
2583
2584         }
2585         if (ret > 0)
2586                 ret = 0;
2587 err:
2588         sk->nr_items = num_found;
2589         btrfs_put_root(root);
2590         btrfs_free_path(path);
2591         return ret;
2592 }
2593
2594 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
2595                                             void __user *argp)
2596 {
2597         struct btrfs_ioctl_search_args __user *uargs;
2598         struct btrfs_ioctl_search_key sk;
2599         int ret;
2600         size_t buf_size;
2601
2602         if (!capable(CAP_SYS_ADMIN))
2603                 return -EPERM;
2604
2605         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2606
2607         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2608                 return -EFAULT;
2609
2610         buf_size = sizeof(uargs->buf);
2611
2612         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2613
2614         /*
2615          * In the origin implementation an overflow is handled by returning a
2616          * search header with a len of zero, so reset ret.
2617          */
2618         if (ret == -EOVERFLOW)
2619                 ret = 0;
2620
2621         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2622                 ret = -EFAULT;
2623         return ret;
2624 }
2625
2626 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
2627                                                void __user *argp)
2628 {
2629         struct btrfs_ioctl_search_args_v2 __user *uarg;
2630         struct btrfs_ioctl_search_args_v2 args;
2631         int ret;
2632         size_t buf_size;
2633         const size_t buf_limit = SZ_16M;
2634
2635         if (!capable(CAP_SYS_ADMIN))
2636                 return -EPERM;
2637
2638         /* copy search header and buffer size */
2639         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2640         if (copy_from_user(&args, uarg, sizeof(args)))
2641                 return -EFAULT;
2642
2643         buf_size = args.buf_size;
2644
2645         /* limit result size to 16MB */
2646         if (buf_size > buf_limit)
2647                 buf_size = buf_limit;
2648
2649         ret = search_ioctl(inode, &args.key, &buf_size,
2650                            (char __user *)(&uarg->buf[0]));
2651         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2652                 ret = -EFAULT;
2653         else if (ret == -EOVERFLOW &&
2654                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2655                 ret = -EFAULT;
2656
2657         return ret;
2658 }
2659
2660 /*
2661  * Search INODE_REFs to identify path name of 'dirid' directory
2662  * in a 'tree_id' tree. and sets path name to 'name'.
2663  */
2664 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2665                                 u64 tree_id, u64 dirid, char *name)
2666 {
2667         struct btrfs_root *root;
2668         struct btrfs_key key;
2669         char *ptr;
2670         int ret = -1;
2671         int slot;
2672         int len;
2673         int total_len = 0;
2674         struct btrfs_inode_ref *iref;
2675         struct extent_buffer *l;
2676         struct btrfs_path *path;
2677
2678         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2679                 name[0]='\0';
2680                 return 0;
2681         }
2682
2683         path = btrfs_alloc_path();
2684         if (!path)
2685                 return -ENOMEM;
2686
2687         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2688
2689         root = btrfs_get_fs_root(info, tree_id, true);
2690         if (IS_ERR(root)) {
2691                 ret = PTR_ERR(root);
2692                 root = NULL;
2693                 goto out;
2694         }
2695
2696         key.objectid = dirid;
2697         key.type = BTRFS_INODE_REF_KEY;
2698         key.offset = (u64)-1;
2699
2700         while (1) {
2701                 ret = btrfs_search_backwards(root, &key, path);
2702                 if (ret < 0)
2703                         goto out;
2704                 else if (ret > 0) {
2705                         ret = -ENOENT;
2706                         goto out;
2707                 }
2708
2709                 l = path->nodes[0];
2710                 slot = path->slots[0];
2711
2712                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2713                 len = btrfs_inode_ref_name_len(l, iref);
2714                 ptr -= len + 1;
2715                 total_len += len + 1;
2716                 if (ptr < name) {
2717                         ret = -ENAMETOOLONG;
2718                         goto out;
2719                 }
2720
2721                 *(ptr + len) = '/';
2722                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2723
2724                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2725                         break;
2726
2727                 btrfs_release_path(path);
2728                 key.objectid = key.offset;
2729                 key.offset = (u64)-1;
2730                 dirid = key.objectid;
2731         }
2732         memmove(name, ptr, total_len);
2733         name[total_len] = '\0';
2734         ret = 0;
2735 out:
2736         btrfs_put_root(root);
2737         btrfs_free_path(path);
2738         return ret;
2739 }
2740
2741 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2742                                 struct inode *inode,
2743                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2744 {
2745         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2746         struct super_block *sb = inode->i_sb;
2747         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2748         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2749         u64 dirid = args->dirid;
2750         unsigned long item_off;
2751         unsigned long item_len;
2752         struct btrfs_inode_ref *iref;
2753         struct btrfs_root_ref *rref;
2754         struct btrfs_root *root = NULL;
2755         struct btrfs_path *path;
2756         struct btrfs_key key, key2;
2757         struct extent_buffer *leaf;
2758         struct inode *temp_inode;
2759         char *ptr;
2760         int slot;
2761         int len;
2762         int total_len = 0;
2763         int ret;
2764
2765         path = btrfs_alloc_path();
2766         if (!path)
2767                 return -ENOMEM;
2768
2769         /*
2770          * If the bottom subvolume does not exist directly under upper_limit,
2771          * construct the path in from the bottom up.
2772          */
2773         if (dirid != upper_limit.objectid) {
2774                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2775
2776                 root = btrfs_get_fs_root(fs_info, treeid, true);
2777                 if (IS_ERR(root)) {
2778                         ret = PTR_ERR(root);
2779                         goto out;
2780                 }
2781
2782                 key.objectid = dirid;
2783                 key.type = BTRFS_INODE_REF_KEY;
2784                 key.offset = (u64)-1;
2785                 while (1) {
2786                         ret = btrfs_search_backwards(root, &key, path);
2787                         if (ret < 0)
2788                                 goto out_put;
2789                         else if (ret > 0) {
2790                                 ret = -ENOENT;
2791                                 goto out_put;
2792                         }
2793
2794                         leaf = path->nodes[0];
2795                         slot = path->slots[0];
2796
2797                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2798                         len = btrfs_inode_ref_name_len(leaf, iref);
2799                         ptr -= len + 1;
2800                         total_len += len + 1;
2801                         if (ptr < args->path) {
2802                                 ret = -ENAMETOOLONG;
2803                                 goto out_put;
2804                         }
2805
2806                         *(ptr + len) = '/';
2807                         read_extent_buffer(leaf, ptr,
2808                                         (unsigned long)(iref + 1), len);
2809
2810                         /* Check the read+exec permission of this directory */
2811                         ret = btrfs_previous_item(root, path, dirid,
2812                                                   BTRFS_INODE_ITEM_KEY);
2813                         if (ret < 0) {
2814                                 goto out_put;
2815                         } else if (ret > 0) {
2816                                 ret = -ENOENT;
2817                                 goto out_put;
2818                         }
2819
2820                         leaf = path->nodes[0];
2821                         slot = path->slots[0];
2822                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2823                         if (key2.objectid != dirid) {
2824                                 ret = -ENOENT;
2825                                 goto out_put;
2826                         }
2827
2828                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2829                         if (IS_ERR(temp_inode)) {
2830                                 ret = PTR_ERR(temp_inode);
2831                                 goto out_put;
2832                         }
2833                         ret = inode_permission(mnt_userns, temp_inode,
2834                                                MAY_READ | MAY_EXEC);
2835                         iput(temp_inode);
2836                         if (ret) {
2837                                 ret = -EACCES;
2838                                 goto out_put;
2839                         }
2840
2841                         if (key.offset == upper_limit.objectid)
2842                                 break;
2843                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2844                                 ret = -EACCES;
2845                                 goto out_put;
2846                         }
2847
2848                         btrfs_release_path(path);
2849                         key.objectid = key.offset;
2850                         key.offset = (u64)-1;
2851                         dirid = key.objectid;
2852                 }
2853
2854                 memmove(args->path, ptr, total_len);
2855                 args->path[total_len] = '\0';
2856                 btrfs_put_root(root);
2857                 root = NULL;
2858                 btrfs_release_path(path);
2859         }
2860
2861         /* Get the bottom subvolume's name from ROOT_REF */
2862         key.objectid = treeid;
2863         key.type = BTRFS_ROOT_REF_KEY;
2864         key.offset = args->treeid;
2865         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2866         if (ret < 0) {
2867                 goto out;
2868         } else if (ret > 0) {
2869                 ret = -ENOENT;
2870                 goto out;
2871         }
2872
2873         leaf = path->nodes[0];
2874         slot = path->slots[0];
2875         btrfs_item_key_to_cpu(leaf, &key, slot);
2876
2877         item_off = btrfs_item_ptr_offset(leaf, slot);
2878         item_len = btrfs_item_size(leaf, slot);
2879         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2880         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2881         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2882                 ret = -EINVAL;
2883                 goto out;
2884         }
2885
2886         /* Copy subvolume's name */
2887         item_off += sizeof(struct btrfs_root_ref);
2888         item_len -= sizeof(struct btrfs_root_ref);
2889         read_extent_buffer(leaf, args->name, item_off, item_len);
2890         args->name[item_len] = 0;
2891
2892 out_put:
2893         btrfs_put_root(root);
2894 out:
2895         btrfs_free_path(path);
2896         return ret;
2897 }
2898
2899 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2900                                            void __user *argp)
2901 {
2902         struct btrfs_ioctl_ino_lookup_args *args;
2903         int ret = 0;
2904
2905         args = memdup_user(argp, sizeof(*args));
2906         if (IS_ERR(args))
2907                 return PTR_ERR(args);
2908
2909         /*
2910          * Unprivileged query to obtain the containing subvolume root id. The
2911          * path is reset so it's consistent with btrfs_search_path_in_tree.
2912          */
2913         if (args->treeid == 0)
2914                 args->treeid = root->root_key.objectid;
2915
2916         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2917                 args->name[0] = 0;
2918                 goto out;
2919         }
2920
2921         if (!capable(CAP_SYS_ADMIN)) {
2922                 ret = -EPERM;
2923                 goto out;
2924         }
2925
2926         ret = btrfs_search_path_in_tree(root->fs_info,
2927                                         args->treeid, args->objectid,
2928                                         args->name);
2929
2930 out:
2931         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2932                 ret = -EFAULT;
2933
2934         kfree(args);
2935         return ret;
2936 }
2937
2938 /*
2939  * Version of ino_lookup ioctl (unprivileged)
2940  *
2941  * The main differences from ino_lookup ioctl are:
2942  *
2943  *   1. Read + Exec permission will be checked using inode_permission() during
2944  *      path construction. -EACCES will be returned in case of failure.
2945  *   2. Path construction will be stopped at the inode number which corresponds
2946  *      to the fd with which this ioctl is called. If constructed path does not
2947  *      exist under fd's inode, -EACCES will be returned.
2948  *   3. The name of bottom subvolume is also searched and filled.
2949  */
2950 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2951 {
2952         struct btrfs_ioctl_ino_lookup_user_args *args;
2953         struct inode *inode;
2954         int ret;
2955
2956         args = memdup_user(argp, sizeof(*args));
2957         if (IS_ERR(args))
2958                 return PTR_ERR(args);
2959
2960         inode = file_inode(file);
2961
2962         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2963             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2964                 /*
2965                  * The subvolume does not exist under fd with which this is
2966                  * called
2967                  */
2968                 kfree(args);
2969                 return -EACCES;
2970         }
2971
2972         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2973
2974         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2975                 ret = -EFAULT;
2976
2977         kfree(args);
2978         return ret;
2979 }
2980
2981 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2982 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2983 {
2984         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2985         struct btrfs_fs_info *fs_info;
2986         struct btrfs_root *root;
2987         struct btrfs_path *path;
2988         struct btrfs_key key;
2989         struct btrfs_root_item *root_item;
2990         struct btrfs_root_ref *rref;
2991         struct extent_buffer *leaf;
2992         unsigned long item_off;
2993         unsigned long item_len;
2994         int slot;
2995         int ret = 0;
2996
2997         path = btrfs_alloc_path();
2998         if (!path)
2999                 return -ENOMEM;
3000
3001         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
3002         if (!subvol_info) {
3003                 btrfs_free_path(path);
3004                 return -ENOMEM;
3005         }
3006
3007         fs_info = BTRFS_I(inode)->root->fs_info;
3008
3009         /* Get root_item of inode's subvolume */
3010         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
3011         root = btrfs_get_fs_root(fs_info, key.objectid, true);
3012         if (IS_ERR(root)) {
3013                 ret = PTR_ERR(root);
3014                 goto out_free;
3015         }
3016         root_item = &root->root_item;
3017
3018         subvol_info->treeid = key.objectid;
3019
3020         subvol_info->generation = btrfs_root_generation(root_item);
3021         subvol_info->flags = btrfs_root_flags(root_item);
3022
3023         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3024         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3025                                                     BTRFS_UUID_SIZE);
3026         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3027                                                     BTRFS_UUID_SIZE);
3028
3029         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3030         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3031         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3032
3033         subvol_info->otransid = btrfs_root_otransid(root_item);
3034         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3035         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3036
3037         subvol_info->stransid = btrfs_root_stransid(root_item);
3038         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3039         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3040
3041         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3042         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3043         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3044
3045         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3046                 /* Search root tree for ROOT_BACKREF of this subvolume */
3047                 key.type = BTRFS_ROOT_BACKREF_KEY;
3048                 key.offset = 0;
3049                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3050                 if (ret < 0) {
3051                         goto out;
3052                 } else if (path->slots[0] >=
3053                            btrfs_header_nritems(path->nodes[0])) {
3054                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3055                         if (ret < 0) {
3056                                 goto out;
3057                         } else if (ret > 0) {
3058                                 ret = -EUCLEAN;
3059                                 goto out;
3060                         }
3061                 }
3062
3063                 leaf = path->nodes[0];
3064                 slot = path->slots[0];
3065                 btrfs_item_key_to_cpu(leaf, &key, slot);
3066                 if (key.objectid == subvol_info->treeid &&
3067                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3068                         subvol_info->parent_id = key.offset;
3069
3070                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3071                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3072
3073                         item_off = btrfs_item_ptr_offset(leaf, slot)
3074                                         + sizeof(struct btrfs_root_ref);
3075                         item_len = btrfs_item_size(leaf, slot)
3076                                         - sizeof(struct btrfs_root_ref);
3077                         read_extent_buffer(leaf, subvol_info->name,
3078                                            item_off, item_len);
3079                 } else {
3080                         ret = -ENOENT;
3081                         goto out;
3082                 }
3083         }
3084
3085         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3086                 ret = -EFAULT;
3087
3088 out:
3089         btrfs_put_root(root);
3090 out_free:
3091         btrfs_free_path(path);
3092         kfree(subvol_info);
3093         return ret;
3094 }
3095
3096 /*
3097  * Return ROOT_REF information of the subvolume containing this inode
3098  * except the subvolume name.
3099  */
3100 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3101                                           void __user *argp)
3102 {
3103         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3104         struct btrfs_root_ref *rref;
3105         struct btrfs_path *path;
3106         struct btrfs_key key;
3107         struct extent_buffer *leaf;
3108         u64 objectid;
3109         int slot;
3110         int ret;
3111         u8 found;
3112
3113         path = btrfs_alloc_path();
3114         if (!path)
3115                 return -ENOMEM;
3116
3117         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3118         if (IS_ERR(rootrefs)) {
3119                 btrfs_free_path(path);
3120                 return PTR_ERR(rootrefs);
3121         }
3122
3123         objectid = root->root_key.objectid;
3124         key.objectid = objectid;
3125         key.type = BTRFS_ROOT_REF_KEY;
3126         key.offset = rootrefs->min_treeid;
3127         found = 0;
3128
3129         root = root->fs_info->tree_root;
3130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3131         if (ret < 0) {
3132                 goto out;
3133         } else if (path->slots[0] >=
3134                    btrfs_header_nritems(path->nodes[0])) {
3135                 ret = btrfs_next_leaf(root, path);
3136                 if (ret < 0) {
3137                         goto out;
3138                 } else if (ret > 0) {
3139                         ret = -EUCLEAN;
3140                         goto out;
3141                 }
3142         }
3143         while (1) {
3144                 leaf = path->nodes[0];
3145                 slot = path->slots[0];
3146
3147                 btrfs_item_key_to_cpu(leaf, &key, slot);
3148                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3149                         ret = 0;
3150                         goto out;
3151                 }
3152
3153                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3154                         ret = -EOVERFLOW;
3155                         goto out;
3156                 }
3157
3158                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3159                 rootrefs->rootref[found].treeid = key.offset;
3160                 rootrefs->rootref[found].dirid =
3161                                   btrfs_root_ref_dirid(leaf, rref);
3162                 found++;
3163
3164                 ret = btrfs_next_item(root, path);
3165                 if (ret < 0) {
3166                         goto out;
3167                 } else if (ret > 0) {
3168                         ret = -EUCLEAN;
3169                         goto out;
3170                 }
3171         }
3172
3173 out:
3174         if (!ret || ret == -EOVERFLOW) {
3175                 rootrefs->num_items = found;
3176                 /* update min_treeid for next search */
3177                 if (found)
3178                         rootrefs->min_treeid =
3179                                 rootrefs->rootref[found - 1].treeid + 1;
3180                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3181                         ret = -EFAULT;
3182         }
3183
3184         kfree(rootrefs);
3185         btrfs_free_path(path);
3186
3187         return ret;
3188 }
3189
3190 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3191                                              void __user *arg,
3192                                              bool destroy_v2)
3193 {
3194         struct dentry *parent = file->f_path.dentry;
3195         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3196         struct dentry *dentry;
3197         struct inode *dir = d_inode(parent);
3198         struct inode *inode;
3199         struct btrfs_root *root = BTRFS_I(dir)->root;
3200         struct btrfs_root *dest = NULL;
3201         struct btrfs_ioctl_vol_args *vol_args = NULL;
3202         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3203         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3204         char *subvol_name, *subvol_name_ptr = NULL;
3205         int subvol_namelen;
3206         int err = 0;
3207         bool destroy_parent = false;
3208
3209         /* We don't support snapshots with extent tree v2 yet. */
3210         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3211                 btrfs_err(fs_info,
3212                           "extent tree v2 doesn't support snapshot deletion yet");
3213                 return -EOPNOTSUPP;
3214         }
3215
3216         if (destroy_v2) {
3217                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3218                 if (IS_ERR(vol_args2))
3219                         return PTR_ERR(vol_args2);
3220
3221                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3222                         err = -EOPNOTSUPP;
3223                         goto out;
3224                 }
3225
3226                 /*
3227                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3228                  * name, same as v1 currently does.
3229                  */
3230                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3231                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3232                         subvol_name = vol_args2->name;
3233
3234                         err = mnt_want_write_file(file);
3235                         if (err)
3236                                 goto out;
3237                 } else {
3238                         struct inode *old_dir;
3239
3240                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3241                                 err = -EINVAL;
3242                                 goto out;
3243                         }
3244
3245                         err = mnt_want_write_file(file);
3246                         if (err)
3247                                 goto out;
3248
3249                         dentry = btrfs_get_dentry(fs_info->sb,
3250                                         BTRFS_FIRST_FREE_OBJECTID,
3251                                         vol_args2->subvolid, 0, 0);
3252                         if (IS_ERR(dentry)) {
3253                                 err = PTR_ERR(dentry);
3254                                 goto out_drop_write;
3255                         }
3256
3257                         /*
3258                          * Change the default parent since the subvolume being
3259                          * deleted can be outside of the current mount point.
3260                          */
3261                         parent = btrfs_get_parent(dentry);
3262
3263                         /*
3264                          * At this point dentry->d_name can point to '/' if the
3265                          * subvolume we want to destroy is outsite of the
3266                          * current mount point, so we need to release the
3267                          * current dentry and execute the lookup to return a new
3268                          * one with ->d_name pointing to the
3269                          * <mount point>/subvol_name.
3270                          */
3271                         dput(dentry);
3272                         if (IS_ERR(parent)) {
3273                                 err = PTR_ERR(parent);
3274                                 goto out_drop_write;
3275                         }
3276                         old_dir = dir;
3277                         dir = d_inode(parent);
3278
3279                         /*
3280                          * If v2 was used with SPEC_BY_ID, a new parent was
3281                          * allocated since the subvolume can be outside of the
3282                          * current mount point. Later on we need to release this
3283                          * new parent dentry.
3284                          */
3285                         destroy_parent = true;
3286
3287                         /*
3288                          * On idmapped mounts, deletion via subvolid is
3289                          * restricted to subvolumes that are immediate
3290                          * ancestors of the inode referenced by the file
3291                          * descriptor in the ioctl. Otherwise the idmapping
3292                          * could potentially be abused to delete subvolumes
3293                          * anywhere in the filesystem the user wouldn't be able
3294                          * to delete without an idmapped mount.
3295                          */
3296                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3297                                 err = -EOPNOTSUPP;
3298                                 goto free_parent;
3299                         }
3300
3301                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3302                                                 fs_info, vol_args2->subvolid);
3303                         if (IS_ERR(subvol_name_ptr)) {
3304                                 err = PTR_ERR(subvol_name_ptr);
3305                                 goto free_parent;
3306                         }
3307                         /* subvol_name_ptr is already nul terminated */
3308                         subvol_name = (char *)kbasename(subvol_name_ptr);
3309                 }
3310         } else {
3311                 vol_args = memdup_user(arg, sizeof(*vol_args));
3312                 if (IS_ERR(vol_args))
3313                         return PTR_ERR(vol_args);
3314
3315                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3316                 subvol_name = vol_args->name;
3317
3318                 err = mnt_want_write_file(file);
3319                 if (err)
3320                         goto out;
3321         }
3322
3323         subvol_namelen = strlen(subvol_name);
3324
3325         if (strchr(subvol_name, '/') ||
3326             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3327                 err = -EINVAL;
3328                 goto free_subvol_name;
3329         }
3330
3331         if (!S_ISDIR(dir->i_mode)) {
3332                 err = -ENOTDIR;
3333                 goto free_subvol_name;
3334         }
3335
3336         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3337         if (err == -EINTR)
3338                 goto free_subvol_name;
3339         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3340         if (IS_ERR(dentry)) {
3341                 err = PTR_ERR(dentry);
3342                 goto out_unlock_dir;
3343         }
3344
3345         if (d_really_is_negative(dentry)) {
3346                 err = -ENOENT;
3347                 goto out_dput;
3348         }
3349
3350         inode = d_inode(dentry);
3351         dest = BTRFS_I(inode)->root;
3352         if (!capable(CAP_SYS_ADMIN)) {
3353                 /*
3354                  * Regular user.  Only allow this with a special mount
3355                  * option, when the user has write+exec access to the
3356                  * subvol root, and when rmdir(2) would have been
3357                  * allowed.
3358                  *
3359                  * Note that this is _not_ check that the subvol is
3360                  * empty or doesn't contain data that we wouldn't
3361                  * otherwise be able to delete.
3362                  *
3363                  * Users who want to delete empty subvols should try
3364                  * rmdir(2).
3365                  */
3366                 err = -EPERM;
3367                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3368                         goto out_dput;
3369
3370                 /*
3371                  * Do not allow deletion if the parent dir is the same
3372                  * as the dir to be deleted.  That means the ioctl
3373                  * must be called on the dentry referencing the root
3374                  * of the subvol, not a random directory contained
3375                  * within it.
3376                  */
3377                 err = -EINVAL;
3378                 if (root == dest)
3379                         goto out_dput;
3380
3381                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3382                 if (err)
3383                         goto out_dput;
3384         }
3385
3386         /* check if subvolume may be deleted by a user */
3387         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3388         if (err)
3389                 goto out_dput;
3390
3391         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3392                 err = -EINVAL;
3393                 goto out_dput;
3394         }
3395
3396         btrfs_inode_lock(inode, 0);
3397         err = btrfs_delete_subvolume(dir, dentry);
3398         btrfs_inode_unlock(inode, 0);
3399         if (!err)
3400                 d_delete_notify(dir, dentry);
3401
3402 out_dput:
3403         dput(dentry);
3404 out_unlock_dir:
3405         btrfs_inode_unlock(dir, 0);
3406 free_subvol_name:
3407         kfree(subvol_name_ptr);
3408 free_parent:
3409         if (destroy_parent)
3410                 dput(parent);
3411 out_drop_write:
3412         mnt_drop_write_file(file);
3413 out:
3414         kfree(vol_args2);
3415         kfree(vol_args);
3416         return err;
3417 }
3418
3419 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3420 {
3421         struct inode *inode = file_inode(file);
3422         struct btrfs_root *root = BTRFS_I(inode)->root;
3423         struct btrfs_ioctl_defrag_range_args range = {0};
3424         int ret;
3425
3426         ret = mnt_want_write_file(file);
3427         if (ret)
3428                 return ret;
3429
3430         if (btrfs_root_readonly(root)) {
3431                 ret = -EROFS;
3432                 goto out;
3433         }
3434
3435         switch (inode->i_mode & S_IFMT) {
3436         case S_IFDIR:
3437                 if (!capable(CAP_SYS_ADMIN)) {
3438                         ret = -EPERM;
3439                         goto out;
3440                 }
3441                 ret = btrfs_defrag_root(root);
3442                 break;
3443         case S_IFREG:
3444                 /*
3445                  * Note that this does not check the file descriptor for write
3446                  * access. This prevents defragmenting executables that are
3447                  * running and allows defrag on files open in read-only mode.
3448                  */
3449                 if (!capable(CAP_SYS_ADMIN) &&
3450                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3451                         ret = -EPERM;
3452                         goto out;
3453                 }
3454
3455                 if (argp) {
3456                         if (copy_from_user(&range, argp, sizeof(range))) {
3457                                 ret = -EFAULT;
3458                                 goto out;
3459                         }
3460                         /* compression requires us to start the IO */
3461                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3462                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3463                                 range.extent_thresh = (u32)-1;
3464                         }
3465                 } else {
3466                         /* the rest are all set to zero by kzalloc */
3467                         range.len = (u64)-1;
3468                 }
3469                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3470                                         &range, BTRFS_OLDEST_GENERATION, 0);
3471                 if (ret > 0)
3472                         ret = 0;
3473                 break;
3474         default:
3475                 ret = -EINVAL;
3476         }
3477 out:
3478         mnt_drop_write_file(file);
3479         return ret;
3480 }
3481
3482 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3483 {
3484         struct btrfs_ioctl_vol_args *vol_args;
3485         bool restore_op = false;
3486         int ret;
3487
3488         if (!capable(CAP_SYS_ADMIN))
3489                 return -EPERM;
3490
3491         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3492                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3493                 return -EINVAL;
3494         }
3495
3496         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3497                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3498                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3499
3500                 /*
3501                  * We can do the device add because we have a paused balanced,
3502                  * change the exclusive op type and remember we should bring
3503                  * back the paused balance
3504                  */
3505                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3506                 btrfs_exclop_start_unlock(fs_info);
3507                 restore_op = true;
3508         }
3509
3510         vol_args = memdup_user(arg, sizeof(*vol_args));
3511         if (IS_ERR(vol_args)) {
3512                 ret = PTR_ERR(vol_args);
3513                 goto out;
3514         }
3515
3516         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3517         ret = btrfs_init_new_device(fs_info, vol_args->name);
3518
3519         if (!ret)
3520                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3521
3522         kfree(vol_args);
3523 out:
3524         if (restore_op)
3525                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3526         else
3527                 btrfs_exclop_finish(fs_info);
3528         return ret;
3529 }
3530
3531 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3532 {
3533         BTRFS_DEV_LOOKUP_ARGS(args);
3534         struct inode *inode = file_inode(file);
3535         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3536         struct btrfs_ioctl_vol_args_v2 *vol_args;
3537         struct block_device *bdev = NULL;
3538         fmode_t mode;
3539         int ret;
3540         bool cancel = false;
3541
3542         if (!capable(CAP_SYS_ADMIN))
3543                 return -EPERM;
3544
3545         vol_args = memdup_user(arg, sizeof(*vol_args));
3546         if (IS_ERR(vol_args))
3547                 return PTR_ERR(vol_args);
3548
3549         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3550                 ret = -EOPNOTSUPP;
3551                 goto out;
3552         }
3553
3554         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3555         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3556                 args.devid = vol_args->devid;
3557         } else if (!strcmp("cancel", vol_args->name)) {
3558                 cancel = true;
3559         } else {
3560                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3561                 if (ret)
3562                         goto out;
3563         }
3564
3565         ret = mnt_want_write_file(file);
3566         if (ret)
3567                 goto out;
3568
3569         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3570                                            cancel);
3571         if (ret)
3572                 goto err_drop;
3573
3574         /* Exclusive operation is now claimed */
3575         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3576
3577         btrfs_exclop_finish(fs_info);
3578
3579         if (!ret) {
3580                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3581                         btrfs_info(fs_info, "device deleted: id %llu",
3582                                         vol_args->devid);
3583                 else
3584                         btrfs_info(fs_info, "device deleted: %s",
3585                                         vol_args->name);
3586         }
3587 err_drop:
3588         mnt_drop_write_file(file);
3589         if (bdev)
3590                 blkdev_put(bdev, mode);
3591 out:
3592         btrfs_put_dev_args_from_path(&args);
3593         kfree(vol_args);
3594         return ret;
3595 }
3596
3597 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3598 {
3599         BTRFS_DEV_LOOKUP_ARGS(args);
3600         struct inode *inode = file_inode(file);
3601         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3602         struct btrfs_ioctl_vol_args *vol_args;
3603         struct block_device *bdev = NULL;
3604         fmode_t mode;
3605         int ret;
3606         bool cancel = false;
3607
3608         if (!capable(CAP_SYS_ADMIN))
3609                 return -EPERM;
3610
3611         vol_args = memdup_user(arg, sizeof(*vol_args));
3612         if (IS_ERR(vol_args))
3613                 return PTR_ERR(vol_args);
3614
3615         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3616         if (!strcmp("cancel", vol_args->name)) {
3617                 cancel = true;
3618         } else {
3619                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3620                 if (ret)
3621                         goto out;
3622         }
3623
3624         ret = mnt_want_write_file(file);
3625         if (ret)
3626                 goto out;
3627
3628         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3629                                            cancel);
3630         if (ret == 0) {
3631                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3632                 if (!ret)
3633                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3634                 btrfs_exclop_finish(fs_info);
3635         }
3636
3637         mnt_drop_write_file(file);
3638         if (bdev)
3639                 blkdev_put(bdev, mode);
3640 out:
3641         btrfs_put_dev_args_from_path(&args);
3642         kfree(vol_args);
3643         return ret;
3644 }
3645
3646 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3647                                 void __user *arg)
3648 {
3649         struct btrfs_ioctl_fs_info_args *fi_args;
3650         struct btrfs_device *device;
3651         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3652         u64 flags_in;
3653         int ret = 0;
3654
3655         fi_args = memdup_user(arg, sizeof(*fi_args));
3656         if (IS_ERR(fi_args))
3657                 return PTR_ERR(fi_args);
3658
3659         flags_in = fi_args->flags;
3660         memset(fi_args, 0, sizeof(*fi_args));
3661
3662         rcu_read_lock();
3663         fi_args->num_devices = fs_devices->num_devices;
3664
3665         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3666                 if (device->devid > fi_args->max_id)
3667                         fi_args->max_id = device->devid;
3668         }
3669         rcu_read_unlock();
3670
3671         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3672         fi_args->nodesize = fs_info->nodesize;
3673         fi_args->sectorsize = fs_info->sectorsize;
3674         fi_args->clone_alignment = fs_info->sectorsize;
3675
3676         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3677                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3678                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3679                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3680         }
3681
3682         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3683                 fi_args->generation = fs_info->generation;
3684                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3685         }
3686
3687         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3688                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3689                        sizeof(fi_args->metadata_uuid));
3690                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3691         }
3692
3693         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3694                 ret = -EFAULT;
3695
3696         kfree(fi_args);
3697         return ret;
3698 }
3699
3700 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3701                                  void __user *arg)
3702 {
3703         BTRFS_DEV_LOOKUP_ARGS(args);
3704         struct btrfs_ioctl_dev_info_args *di_args;
3705         struct btrfs_device *dev;
3706         int ret = 0;
3707
3708         di_args = memdup_user(arg, sizeof(*di_args));
3709         if (IS_ERR(di_args))
3710                 return PTR_ERR(di_args);
3711
3712         args.devid = di_args->devid;
3713         if (!btrfs_is_empty_uuid(di_args->uuid))
3714                 args.uuid = di_args->uuid;
3715
3716         rcu_read_lock();
3717         dev = btrfs_find_device(fs_info->fs_devices, &args);
3718         if (!dev) {
3719                 ret = -ENODEV;
3720                 goto out;
3721         }
3722
3723         di_args->devid = dev->devid;
3724         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3725         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3726         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3727         if (dev->name) {
3728                 strncpy(di_args->path, rcu_str_deref(dev->name),
3729                                 sizeof(di_args->path) - 1);
3730                 di_args->path[sizeof(di_args->path) - 1] = 0;
3731         } else {
3732                 di_args->path[0] = '\0';
3733         }
3734
3735 out:
3736         rcu_read_unlock();
3737         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3738                 ret = -EFAULT;
3739
3740         kfree(di_args);
3741         return ret;
3742 }
3743
3744 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3745 {
3746         struct inode *inode = file_inode(file);
3747         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3748         struct btrfs_root *root = BTRFS_I(inode)->root;
3749         struct btrfs_root *new_root;
3750         struct btrfs_dir_item *di;
3751         struct btrfs_trans_handle *trans;
3752         struct btrfs_path *path = NULL;
3753         struct btrfs_disk_key disk_key;
3754         u64 objectid = 0;
3755         u64 dir_id;
3756         int ret;
3757
3758         if (!capable(CAP_SYS_ADMIN))
3759                 return -EPERM;
3760
3761         ret = mnt_want_write_file(file);
3762         if (ret)
3763                 return ret;
3764
3765         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3766                 ret = -EFAULT;
3767                 goto out;
3768         }
3769
3770         if (!objectid)
3771                 objectid = BTRFS_FS_TREE_OBJECTID;
3772
3773         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3774         if (IS_ERR(new_root)) {
3775                 ret = PTR_ERR(new_root);
3776                 goto out;
3777         }
3778         if (!is_fstree(new_root->root_key.objectid)) {
3779                 ret = -ENOENT;
3780                 goto out_free;
3781         }
3782
3783         path = btrfs_alloc_path();
3784         if (!path) {
3785                 ret = -ENOMEM;
3786                 goto out_free;
3787         }
3788
3789         trans = btrfs_start_transaction(root, 1);
3790         if (IS_ERR(trans)) {
3791                 ret = PTR_ERR(trans);
3792                 goto out_free;
3793         }
3794
3795         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3796         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3797                                    dir_id, "default", 7, 1);
3798         if (IS_ERR_OR_NULL(di)) {
3799                 btrfs_release_path(path);
3800                 btrfs_end_transaction(trans);
3801                 btrfs_err(fs_info,
3802                           "Umm, you don't have the default diritem, this isn't going to work");
3803                 ret = -ENOENT;
3804                 goto out_free;
3805         }
3806
3807         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3808         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3809         btrfs_mark_buffer_dirty(path->nodes[0]);
3810         btrfs_release_path(path);
3811
3812         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3813         btrfs_end_transaction(trans);
3814 out_free:
3815         btrfs_put_root(new_root);
3816         btrfs_free_path(path);
3817 out:
3818         mnt_drop_write_file(file);
3819         return ret;
3820 }
3821
3822 static void get_block_group_info(struct list_head *groups_list,
3823                                  struct btrfs_ioctl_space_info *space)
3824 {
3825         struct btrfs_block_group *block_group;
3826
3827         space->total_bytes = 0;
3828         space->used_bytes = 0;
3829         space->flags = 0;
3830         list_for_each_entry(block_group, groups_list, list) {
3831                 space->flags = block_group->flags;
3832                 space->total_bytes += block_group->length;
3833                 space->used_bytes += block_group->used;
3834         }
3835 }
3836
3837 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3838                                    void __user *arg)
3839 {
3840         struct btrfs_ioctl_space_args space_args;
3841         struct btrfs_ioctl_space_info space;
3842         struct btrfs_ioctl_space_info *dest;
3843         struct btrfs_ioctl_space_info *dest_orig;
3844         struct btrfs_ioctl_space_info __user *user_dest;
3845         struct btrfs_space_info *info;
3846         static const u64 types[] = {
3847                 BTRFS_BLOCK_GROUP_DATA,
3848                 BTRFS_BLOCK_GROUP_SYSTEM,
3849                 BTRFS_BLOCK_GROUP_METADATA,
3850                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3851         };
3852         int num_types = 4;
3853         int alloc_size;
3854         int ret = 0;
3855         u64 slot_count = 0;
3856         int i, c;
3857
3858         if (copy_from_user(&space_args,
3859                            (struct btrfs_ioctl_space_args __user *)arg,
3860                            sizeof(space_args)))
3861                 return -EFAULT;
3862
3863         for (i = 0; i < num_types; i++) {
3864                 struct btrfs_space_info *tmp;
3865
3866                 info = NULL;
3867                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3868                         if (tmp->flags == types[i]) {
3869                                 info = tmp;
3870                                 break;
3871                         }
3872                 }
3873
3874                 if (!info)
3875                         continue;
3876
3877                 down_read(&info->groups_sem);
3878                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3879                         if (!list_empty(&info->block_groups[c]))
3880                                 slot_count++;
3881                 }
3882                 up_read(&info->groups_sem);
3883         }
3884
3885         /*
3886          * Global block reserve, exported as a space_info
3887          */
3888         slot_count++;
3889
3890         /* space_slots == 0 means they are asking for a count */
3891         if (space_args.space_slots == 0) {
3892                 space_args.total_spaces = slot_count;
3893                 goto out;
3894         }
3895
3896         slot_count = min_t(u64, space_args.space_slots, slot_count);
3897
3898         alloc_size = sizeof(*dest) * slot_count;
3899
3900         /* we generally have at most 6 or so space infos, one for each raid
3901          * level.  So, a whole page should be more than enough for everyone
3902          */
3903         if (alloc_size > PAGE_SIZE)
3904                 return -ENOMEM;
3905
3906         space_args.total_spaces = 0;
3907         dest = kmalloc(alloc_size, GFP_KERNEL);
3908         if (!dest)
3909                 return -ENOMEM;
3910         dest_orig = dest;
3911
3912         /* now we have a buffer to copy into */
3913         for (i = 0; i < num_types; i++) {
3914                 struct btrfs_space_info *tmp;
3915
3916                 if (!slot_count)
3917                         break;
3918
3919                 info = NULL;
3920                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3921                         if (tmp->flags == types[i]) {
3922                                 info = tmp;
3923                                 break;
3924                         }
3925                 }
3926
3927                 if (!info)
3928                         continue;
3929                 down_read(&info->groups_sem);
3930                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3931                         if (!list_empty(&info->block_groups[c])) {
3932                                 get_block_group_info(&info->block_groups[c],
3933                                                      &space);
3934                                 memcpy(dest, &space, sizeof(space));
3935                                 dest++;
3936                                 space_args.total_spaces++;
3937                                 slot_count--;
3938                         }
3939                         if (!slot_count)
3940                                 break;
3941                 }
3942                 up_read(&info->groups_sem);
3943         }
3944
3945         /*
3946          * Add global block reserve
3947          */
3948         if (slot_count) {
3949                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3950
3951                 spin_lock(&block_rsv->lock);
3952                 space.total_bytes = block_rsv->size;
3953                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3954                 spin_unlock(&block_rsv->lock);
3955                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3956                 memcpy(dest, &space, sizeof(space));
3957                 space_args.total_spaces++;
3958         }
3959
3960         user_dest = (struct btrfs_ioctl_space_info __user *)
3961                 (arg + sizeof(struct btrfs_ioctl_space_args));
3962
3963         if (copy_to_user(user_dest, dest_orig, alloc_size))
3964                 ret = -EFAULT;
3965
3966         kfree(dest_orig);
3967 out:
3968         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3969                 ret = -EFAULT;
3970
3971         return ret;
3972 }
3973
3974 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3975                                             void __user *argp)
3976 {
3977         struct btrfs_trans_handle *trans;
3978         u64 transid;
3979
3980         trans = btrfs_attach_transaction_barrier(root);
3981         if (IS_ERR(trans)) {
3982                 if (PTR_ERR(trans) != -ENOENT)
3983                         return PTR_ERR(trans);
3984
3985                 /* No running transaction, don't bother */
3986                 transid = root->fs_info->last_trans_committed;
3987                 goto out;
3988         }
3989         transid = trans->transid;
3990         btrfs_commit_transaction_async(trans);
3991 out:
3992         if (argp)
3993                 if (copy_to_user(argp, &transid, sizeof(transid)))
3994                         return -EFAULT;
3995         return 0;
3996 }
3997
3998 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3999                                            void __user *argp)
4000 {
4001         u64 transid;
4002
4003         if (argp) {
4004                 if (copy_from_user(&transid, argp, sizeof(transid)))
4005                         return -EFAULT;
4006         } else {
4007                 transid = 0;  /* current trans */
4008         }
4009         return btrfs_wait_for_commit(fs_info, transid);
4010 }
4011
4012 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4013 {
4014         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4015         struct btrfs_ioctl_scrub_args *sa;
4016         int ret;
4017
4018         if (!capable(CAP_SYS_ADMIN))
4019                 return -EPERM;
4020
4021         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4022                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4023                 return -EINVAL;
4024         }
4025
4026         sa = memdup_user(arg, sizeof(*sa));
4027         if (IS_ERR(sa))
4028                 return PTR_ERR(sa);
4029
4030         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4031                 ret = mnt_want_write_file(file);
4032                 if (ret)
4033                         goto out;
4034         }
4035
4036         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4037                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4038                               0);
4039
4040         /*
4041          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4042          * error. This is important as it allows user space to know how much
4043          * progress scrub has done. For example, if scrub is canceled we get
4044          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4045          * space. Later user space can inspect the progress from the structure
4046          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4047          * previously (btrfs-progs does this).
4048          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4049          * then return -EFAULT to signal the structure was not copied or it may
4050          * be corrupt and unreliable due to a partial copy.
4051          */
4052         if (copy_to_user(arg, sa, sizeof(*sa)))
4053                 ret = -EFAULT;
4054
4055         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4056                 mnt_drop_write_file(file);
4057 out:
4058         kfree(sa);
4059         return ret;
4060 }
4061
4062 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4063 {
4064         if (!capable(CAP_SYS_ADMIN))
4065                 return -EPERM;
4066
4067         return btrfs_scrub_cancel(fs_info);
4068 }
4069
4070 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4071                                        void __user *arg)
4072 {
4073         struct btrfs_ioctl_scrub_args *sa;
4074         int ret;
4075
4076         if (!capable(CAP_SYS_ADMIN))
4077                 return -EPERM;
4078
4079         sa = memdup_user(arg, sizeof(*sa));
4080         if (IS_ERR(sa))
4081                 return PTR_ERR(sa);
4082
4083         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4084
4085         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4086                 ret = -EFAULT;
4087
4088         kfree(sa);
4089         return ret;
4090 }
4091
4092 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4093                                       void __user *arg)
4094 {
4095         struct btrfs_ioctl_get_dev_stats *sa;
4096         int ret;
4097
4098         sa = memdup_user(arg, sizeof(*sa));
4099         if (IS_ERR(sa))
4100                 return PTR_ERR(sa);
4101
4102         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4103                 kfree(sa);
4104                 return -EPERM;
4105         }
4106
4107         ret = btrfs_get_dev_stats(fs_info, sa);
4108
4109         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4110                 ret = -EFAULT;
4111
4112         kfree(sa);
4113         return ret;
4114 }
4115
4116 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4117                                     void __user *arg)
4118 {
4119         struct btrfs_ioctl_dev_replace_args *p;
4120         int ret;
4121
4122         if (!capable(CAP_SYS_ADMIN))
4123                 return -EPERM;
4124
4125         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4126                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4127                 return -EINVAL;
4128         }
4129
4130         p = memdup_user(arg, sizeof(*p));
4131         if (IS_ERR(p))
4132                 return PTR_ERR(p);
4133
4134         switch (p->cmd) {
4135         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4136                 if (sb_rdonly(fs_info->sb)) {
4137                         ret = -EROFS;
4138                         goto out;
4139                 }
4140                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4141                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4142                 } else {
4143                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4144                         btrfs_exclop_finish(fs_info);
4145                 }
4146                 break;
4147         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4148                 btrfs_dev_replace_status(fs_info, p);
4149                 ret = 0;
4150                 break;
4151         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4152                 p->result = btrfs_dev_replace_cancel(fs_info);
4153                 ret = 0;
4154                 break;
4155         default:
4156                 ret = -EINVAL;
4157                 break;
4158         }
4159
4160         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4161                 ret = -EFAULT;
4162 out:
4163         kfree(p);
4164         return ret;
4165 }
4166
4167 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4168 {
4169         int ret = 0;
4170         int i;
4171         u64 rel_ptr;
4172         int size;
4173         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4174         struct inode_fs_paths *ipath = NULL;
4175         struct btrfs_path *path;
4176
4177         if (!capable(CAP_DAC_READ_SEARCH))
4178                 return -EPERM;
4179
4180         path = btrfs_alloc_path();
4181         if (!path) {
4182                 ret = -ENOMEM;
4183                 goto out;
4184         }
4185
4186         ipa = memdup_user(arg, sizeof(*ipa));
4187         if (IS_ERR(ipa)) {
4188                 ret = PTR_ERR(ipa);
4189                 ipa = NULL;
4190                 goto out;
4191         }
4192
4193         size = min_t(u32, ipa->size, 4096);
4194         ipath = init_ipath(size, root, path);
4195         if (IS_ERR(ipath)) {
4196                 ret = PTR_ERR(ipath);
4197                 ipath = NULL;
4198                 goto out;
4199         }
4200
4201         ret = paths_from_inode(ipa->inum, ipath);
4202         if (ret < 0)
4203                 goto out;
4204
4205         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4206                 rel_ptr = ipath->fspath->val[i] -
4207                           (u64)(unsigned long)ipath->fspath->val;
4208                 ipath->fspath->val[i] = rel_ptr;
4209         }
4210
4211         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4212                            ipath->fspath, size);
4213         if (ret) {
4214                 ret = -EFAULT;
4215                 goto out;
4216         }
4217
4218 out:
4219         btrfs_free_path(path);
4220         free_ipath(ipath);
4221         kfree(ipa);
4222
4223         return ret;
4224 }
4225
4226 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4227 {
4228         struct btrfs_data_container *inodes = ctx;
4229         const size_t c = 3 * sizeof(u64);
4230
4231         if (inodes->bytes_left >= c) {
4232                 inodes->bytes_left -= c;
4233                 inodes->val[inodes->elem_cnt] = inum;
4234                 inodes->val[inodes->elem_cnt + 1] = offset;
4235                 inodes->val[inodes->elem_cnt + 2] = root;
4236                 inodes->elem_cnt += 3;
4237         } else {
4238                 inodes->bytes_missing += c - inodes->bytes_left;
4239                 inodes->bytes_left = 0;
4240                 inodes->elem_missed += 3;
4241         }
4242
4243         return 0;
4244 }
4245
4246 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4247                                         void __user *arg, int version)
4248 {
4249         int ret = 0;
4250         int size;
4251         struct btrfs_ioctl_logical_ino_args *loi;
4252         struct btrfs_data_container *inodes = NULL;
4253         struct btrfs_path *path = NULL;
4254         bool ignore_offset;
4255
4256         if (!capable(CAP_SYS_ADMIN))
4257                 return -EPERM;
4258
4259         loi = memdup_user(arg, sizeof(*loi));
4260         if (IS_ERR(loi))
4261                 return PTR_ERR(loi);
4262
4263         if (version == 1) {
4264                 ignore_offset = false;
4265                 size = min_t(u32, loi->size, SZ_64K);
4266         } else {
4267                 /* All reserved bits must be 0 for now */
4268                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4269                         ret = -EINVAL;
4270                         goto out_loi;
4271                 }
4272                 /* Only accept flags we have defined so far */
4273                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4274                         ret = -EINVAL;
4275                         goto out_loi;
4276                 }
4277                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4278                 size = min_t(u32, loi->size, SZ_16M);
4279         }
4280
4281         path = btrfs_alloc_path();
4282         if (!path) {
4283                 ret = -ENOMEM;
4284                 goto out;
4285         }
4286
4287         inodes = init_data_container(size);
4288         if (IS_ERR(inodes)) {
4289                 ret = PTR_ERR(inodes);
4290                 inodes = NULL;
4291                 goto out;
4292         }
4293
4294         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4295                                           build_ino_list, inodes, ignore_offset);
4296         if (ret == -EINVAL)
4297                 ret = -ENOENT;
4298         if (ret < 0)
4299                 goto out;
4300
4301         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4302                            size);
4303         if (ret)
4304                 ret = -EFAULT;
4305
4306 out:
4307         btrfs_free_path(path);
4308         kvfree(inodes);
4309 out_loi:
4310         kfree(loi);
4311
4312         return ret;
4313 }
4314
4315 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4316                                struct btrfs_ioctl_balance_args *bargs)
4317 {
4318         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4319
4320         bargs->flags = bctl->flags;
4321
4322         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4323                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4324         if (atomic_read(&fs_info->balance_pause_req))
4325                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4326         if (atomic_read(&fs_info->balance_cancel_req))
4327                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4328
4329         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4330         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4331         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4332
4333         spin_lock(&fs_info->balance_lock);
4334         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4335         spin_unlock(&fs_info->balance_lock);
4336 }
4337
4338 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4339 {
4340         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4341         struct btrfs_fs_info *fs_info = root->fs_info;
4342         struct btrfs_ioctl_balance_args *bargs;
4343         struct btrfs_balance_control *bctl;
4344         bool need_unlock; /* for mut. excl. ops lock */
4345         int ret;
4346
4347         if (!arg)
4348                 btrfs_warn(fs_info,
4349         "IOC_BALANCE ioctl (v1) is deprecated and will be removed in kernel 5.18");
4350
4351         if (!capable(CAP_SYS_ADMIN))
4352                 return -EPERM;
4353
4354         ret = mnt_want_write_file(file);
4355         if (ret)
4356                 return ret;
4357
4358 again:
4359         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4360                 mutex_lock(&fs_info->balance_mutex);
4361                 need_unlock = true;
4362                 goto locked;
4363         }
4364
4365         /*
4366          * mut. excl. ops lock is locked.  Three possibilities:
4367          *   (1) some other op is running
4368          *   (2) balance is running
4369          *   (3) balance is paused -- special case (think resume)
4370          */
4371         mutex_lock(&fs_info->balance_mutex);
4372         if (fs_info->balance_ctl) {
4373                 /* this is either (2) or (3) */
4374                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4375                         mutex_unlock(&fs_info->balance_mutex);
4376                         /*
4377                          * Lock released to allow other waiters to continue,
4378                          * we'll reexamine the status again.
4379                          */
4380                         mutex_lock(&fs_info->balance_mutex);
4381
4382                         if (fs_info->balance_ctl &&
4383                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4384                                 /* this is (3) */
4385                                 need_unlock = false;
4386                                 goto locked;
4387                         }
4388
4389                         mutex_unlock(&fs_info->balance_mutex);
4390                         goto again;
4391                 } else {
4392                         /* this is (2) */
4393                         mutex_unlock(&fs_info->balance_mutex);
4394                         ret = -EINPROGRESS;
4395                         goto out;
4396                 }
4397         } else {
4398                 /* this is (1) */
4399                 mutex_unlock(&fs_info->balance_mutex);
4400                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4401                 goto out;
4402         }
4403
4404 locked:
4405
4406         if (arg) {
4407                 bargs = memdup_user(arg, sizeof(*bargs));
4408                 if (IS_ERR(bargs)) {
4409                         ret = PTR_ERR(bargs);
4410                         goto out_unlock;
4411                 }
4412
4413                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4414                         if (!fs_info->balance_ctl) {
4415                                 ret = -ENOTCONN;
4416                                 goto out_bargs;
4417                         }
4418
4419                         bctl = fs_info->balance_ctl;
4420                         spin_lock(&fs_info->balance_lock);
4421                         bctl->flags |= BTRFS_BALANCE_RESUME;
4422                         spin_unlock(&fs_info->balance_lock);
4423                         btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4424
4425                         goto do_balance;
4426                 }
4427         } else {
4428                 bargs = NULL;
4429         }
4430
4431         if (fs_info->balance_ctl) {
4432                 ret = -EINPROGRESS;
4433                 goto out_bargs;
4434         }
4435
4436         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4437         if (!bctl) {
4438                 ret = -ENOMEM;
4439                 goto out_bargs;
4440         }
4441
4442         if (arg) {
4443                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4444                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4445                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4446
4447                 bctl->flags = bargs->flags;
4448         } else {
4449                 /* balance everything - no filters */
4450                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4451         }
4452
4453         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4454                 ret = -EINVAL;
4455                 goto out_bctl;
4456         }
4457
4458 do_balance:
4459         /*
4460          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4461          * bctl is freed in reset_balance_state, or, if restriper was paused
4462          * all the way until unmount, in free_fs_info.  The flag should be
4463          * cleared after reset_balance_state.
4464          */
4465         need_unlock = false;
4466
4467         ret = btrfs_balance(fs_info, bctl, bargs);
4468         bctl = NULL;
4469
4470         if ((ret == 0 || ret == -ECANCELED) && arg) {
4471                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4472                         ret = -EFAULT;
4473         }
4474
4475 out_bctl:
4476         kfree(bctl);
4477 out_bargs:
4478         kfree(bargs);
4479 out_unlock:
4480         mutex_unlock(&fs_info->balance_mutex);
4481         if (need_unlock)
4482                 btrfs_exclop_finish(fs_info);
4483 out:
4484         mnt_drop_write_file(file);
4485         return ret;
4486 }
4487
4488 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4489 {
4490         if (!capable(CAP_SYS_ADMIN))
4491                 return -EPERM;
4492
4493         switch (cmd) {
4494         case BTRFS_BALANCE_CTL_PAUSE:
4495                 return btrfs_pause_balance(fs_info);
4496         case BTRFS_BALANCE_CTL_CANCEL:
4497                 return btrfs_cancel_balance(fs_info);
4498         }
4499
4500         return -EINVAL;
4501 }
4502
4503 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4504                                          void __user *arg)
4505 {
4506         struct btrfs_ioctl_balance_args *bargs;
4507         int ret = 0;
4508
4509         if (!capable(CAP_SYS_ADMIN))
4510                 return -EPERM;
4511
4512         mutex_lock(&fs_info->balance_mutex);
4513         if (!fs_info->balance_ctl) {
4514                 ret = -ENOTCONN;
4515                 goto out;
4516         }
4517
4518         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4519         if (!bargs) {
4520                 ret = -ENOMEM;
4521                 goto out;
4522         }
4523
4524         btrfs_update_ioctl_balance_args(fs_info, bargs);
4525
4526         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4527                 ret = -EFAULT;
4528
4529         kfree(bargs);
4530 out:
4531         mutex_unlock(&fs_info->balance_mutex);
4532         return ret;
4533 }
4534
4535 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4536 {
4537         struct inode *inode = file_inode(file);
4538         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4539         struct btrfs_ioctl_quota_ctl_args *sa;
4540         int ret;
4541
4542         if (!capable(CAP_SYS_ADMIN))
4543                 return -EPERM;
4544
4545         ret = mnt_want_write_file(file);
4546         if (ret)
4547                 return ret;
4548
4549         sa = memdup_user(arg, sizeof(*sa));
4550         if (IS_ERR(sa)) {
4551                 ret = PTR_ERR(sa);
4552                 goto drop_write;
4553         }
4554
4555         down_write(&fs_info->subvol_sem);
4556
4557         switch (sa->cmd) {
4558         case BTRFS_QUOTA_CTL_ENABLE:
4559                 ret = btrfs_quota_enable(fs_info);
4560                 break;
4561         case BTRFS_QUOTA_CTL_DISABLE:
4562                 ret = btrfs_quota_disable(fs_info);
4563                 break;
4564         default:
4565                 ret = -EINVAL;
4566                 break;
4567         }
4568
4569         kfree(sa);
4570         up_write(&fs_info->subvol_sem);
4571 drop_write:
4572         mnt_drop_write_file(file);
4573         return ret;
4574 }
4575
4576 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4577 {
4578         struct inode *inode = file_inode(file);
4579         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4580         struct btrfs_root *root = BTRFS_I(inode)->root;
4581         struct btrfs_ioctl_qgroup_assign_args *sa;
4582         struct btrfs_trans_handle *trans;
4583         int ret;
4584         int err;
4585
4586         if (!capable(CAP_SYS_ADMIN))
4587                 return -EPERM;
4588
4589         ret = mnt_want_write_file(file);
4590         if (ret)
4591                 return ret;
4592
4593         sa = memdup_user(arg, sizeof(*sa));
4594         if (IS_ERR(sa)) {
4595                 ret = PTR_ERR(sa);
4596                 goto drop_write;
4597         }
4598
4599         trans = btrfs_join_transaction(root);
4600         if (IS_ERR(trans)) {
4601                 ret = PTR_ERR(trans);
4602                 goto out;
4603         }
4604
4605         if (sa->assign) {
4606                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4607         } else {
4608                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4609         }
4610
4611         /* update qgroup status and info */
4612         err = btrfs_run_qgroups(trans);
4613         if (err < 0)
4614                 btrfs_handle_fs_error(fs_info, err,
4615                                       "failed to update qgroup status and info");
4616         err = btrfs_end_transaction(trans);
4617         if (err && !ret)
4618                 ret = err;
4619
4620 out:
4621         kfree(sa);
4622 drop_write:
4623         mnt_drop_write_file(file);
4624         return ret;
4625 }
4626
4627 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4628 {
4629         struct inode *inode = file_inode(file);
4630         struct btrfs_root *root = BTRFS_I(inode)->root;
4631         struct btrfs_ioctl_qgroup_create_args *sa;
4632         struct btrfs_trans_handle *trans;
4633         int ret;
4634         int err;
4635
4636         if (!capable(CAP_SYS_ADMIN))
4637                 return -EPERM;
4638
4639         ret = mnt_want_write_file(file);
4640         if (ret)
4641                 return ret;
4642
4643         sa = memdup_user(arg, sizeof(*sa));
4644         if (IS_ERR(sa)) {
4645                 ret = PTR_ERR(sa);
4646                 goto drop_write;
4647         }
4648
4649         if (!sa->qgroupid) {
4650                 ret = -EINVAL;
4651                 goto out;
4652         }
4653
4654         trans = btrfs_join_transaction(root);
4655         if (IS_ERR(trans)) {
4656                 ret = PTR_ERR(trans);
4657                 goto out;
4658         }
4659
4660         if (sa->create) {
4661                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4662         } else {
4663                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4664         }
4665
4666         err = btrfs_end_transaction(trans);
4667         if (err && !ret)
4668                 ret = err;
4669
4670 out:
4671         kfree(sa);
4672 drop_write:
4673         mnt_drop_write_file(file);
4674         return ret;
4675 }
4676
4677 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4678 {
4679         struct inode *inode = file_inode(file);
4680         struct btrfs_root *root = BTRFS_I(inode)->root;
4681         struct btrfs_ioctl_qgroup_limit_args *sa;
4682         struct btrfs_trans_handle *trans;
4683         int ret;
4684         int err;
4685         u64 qgroupid;
4686
4687         if (!capable(CAP_SYS_ADMIN))
4688                 return -EPERM;
4689
4690         ret = mnt_want_write_file(file);
4691         if (ret)
4692                 return ret;
4693
4694         sa = memdup_user(arg, sizeof(*sa));
4695         if (IS_ERR(sa)) {
4696                 ret = PTR_ERR(sa);
4697                 goto drop_write;
4698         }
4699
4700         trans = btrfs_join_transaction(root);
4701         if (IS_ERR(trans)) {
4702                 ret = PTR_ERR(trans);
4703                 goto out;
4704         }
4705
4706         qgroupid = sa->qgroupid;
4707         if (!qgroupid) {
4708                 /* take the current subvol as qgroup */
4709                 qgroupid = root->root_key.objectid;
4710         }
4711
4712         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4713
4714         err = btrfs_end_transaction(trans);
4715         if (err && !ret)
4716                 ret = err;
4717
4718 out:
4719         kfree(sa);
4720 drop_write:
4721         mnt_drop_write_file(file);
4722         return ret;
4723 }
4724
4725 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4726 {
4727         struct inode *inode = file_inode(file);
4728         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4729         struct btrfs_ioctl_quota_rescan_args *qsa;
4730         int ret;
4731
4732         if (!capable(CAP_SYS_ADMIN))
4733                 return -EPERM;
4734
4735         ret = mnt_want_write_file(file);
4736         if (ret)
4737                 return ret;
4738
4739         qsa = memdup_user(arg, sizeof(*qsa));
4740         if (IS_ERR(qsa)) {
4741                 ret = PTR_ERR(qsa);
4742                 goto drop_write;
4743         }
4744
4745         if (qsa->flags) {
4746                 ret = -EINVAL;
4747                 goto out;
4748         }
4749
4750         ret = btrfs_qgroup_rescan(fs_info);
4751
4752 out:
4753         kfree(qsa);
4754 drop_write:
4755         mnt_drop_write_file(file);
4756         return ret;
4757 }
4758
4759 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4760                                                 void __user *arg)
4761 {
4762         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4763
4764         if (!capable(CAP_SYS_ADMIN))
4765                 return -EPERM;
4766
4767         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4768                 qsa.flags = 1;
4769                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4770         }
4771
4772         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4773                 return -EFAULT;
4774
4775         return 0;
4776 }
4777
4778 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4779                                                 void __user *arg)
4780 {
4781         if (!capable(CAP_SYS_ADMIN))
4782                 return -EPERM;
4783
4784         return btrfs_qgroup_wait_for_completion(fs_info, true);
4785 }
4786
4787 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4788                                             struct user_namespace *mnt_userns,
4789                                             struct btrfs_ioctl_received_subvol_args *sa)
4790 {
4791         struct inode *inode = file_inode(file);
4792         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4793         struct btrfs_root *root = BTRFS_I(inode)->root;
4794         struct btrfs_root_item *root_item = &root->root_item;
4795         struct btrfs_trans_handle *trans;
4796         struct timespec64 ct = current_time(inode);
4797         int ret = 0;
4798         int received_uuid_changed;
4799
4800         if (!inode_owner_or_capable(mnt_userns, inode))
4801                 return -EPERM;
4802
4803         ret = mnt_want_write_file(file);
4804         if (ret < 0)
4805                 return ret;
4806
4807         down_write(&fs_info->subvol_sem);
4808
4809         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4810                 ret = -EINVAL;
4811                 goto out;
4812         }
4813
4814         if (btrfs_root_readonly(root)) {
4815                 ret = -EROFS;
4816                 goto out;
4817         }
4818
4819         /*
4820          * 1 - root item
4821          * 2 - uuid items (received uuid + subvol uuid)
4822          */
4823         trans = btrfs_start_transaction(root, 3);
4824         if (IS_ERR(trans)) {
4825                 ret = PTR_ERR(trans);
4826                 trans = NULL;
4827                 goto out;
4828         }
4829
4830         sa->rtransid = trans->transid;
4831         sa->rtime.sec = ct.tv_sec;
4832         sa->rtime.nsec = ct.tv_nsec;
4833
4834         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4835                                        BTRFS_UUID_SIZE);
4836         if (received_uuid_changed &&
4837             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4838                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4839                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4840                                           root->root_key.objectid);
4841                 if (ret && ret != -ENOENT) {
4842                         btrfs_abort_transaction(trans, ret);
4843                         btrfs_end_transaction(trans);
4844                         goto out;
4845                 }
4846         }
4847         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4848         btrfs_set_root_stransid(root_item, sa->stransid);
4849         btrfs_set_root_rtransid(root_item, sa->rtransid);
4850         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4851         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4852         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4853         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4854
4855         ret = btrfs_update_root(trans, fs_info->tree_root,
4856                                 &root->root_key, &root->root_item);
4857         if (ret < 0) {
4858                 btrfs_end_transaction(trans);
4859                 goto out;
4860         }
4861         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4862                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4863                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4864                                           root->root_key.objectid);
4865                 if (ret < 0 && ret != -EEXIST) {
4866                         btrfs_abort_transaction(trans, ret);
4867                         btrfs_end_transaction(trans);
4868                         goto out;
4869                 }
4870         }
4871         ret = btrfs_commit_transaction(trans);
4872 out:
4873         up_write(&fs_info->subvol_sem);
4874         mnt_drop_write_file(file);
4875         return ret;
4876 }
4877
4878 #ifdef CONFIG_64BIT
4879 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4880                                                 void __user *arg)
4881 {
4882         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4883         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4884         int ret = 0;
4885
4886         args32 = memdup_user(arg, sizeof(*args32));
4887         if (IS_ERR(args32))
4888                 return PTR_ERR(args32);
4889
4890         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4891         if (!args64) {
4892                 ret = -ENOMEM;
4893                 goto out;
4894         }
4895
4896         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4897         args64->stransid = args32->stransid;
4898         args64->rtransid = args32->rtransid;
4899         args64->stime.sec = args32->stime.sec;
4900         args64->stime.nsec = args32->stime.nsec;
4901         args64->rtime.sec = args32->rtime.sec;
4902         args64->rtime.nsec = args32->rtime.nsec;
4903         args64->flags = args32->flags;
4904
4905         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4906         if (ret)
4907                 goto out;
4908
4909         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4910         args32->stransid = args64->stransid;
4911         args32->rtransid = args64->rtransid;
4912         args32->stime.sec = args64->stime.sec;
4913         args32->stime.nsec = args64->stime.nsec;
4914         args32->rtime.sec = args64->rtime.sec;
4915         args32->rtime.nsec = args64->rtime.nsec;
4916         args32->flags = args64->flags;
4917
4918         ret = copy_to_user(arg, args32, sizeof(*args32));
4919         if (ret)
4920                 ret = -EFAULT;
4921
4922 out:
4923         kfree(args32);
4924         kfree(args64);
4925         return ret;
4926 }
4927 #endif
4928
4929 static long btrfs_ioctl_set_received_subvol(struct file *file,
4930                                             void __user *arg)
4931 {
4932         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4933         int ret = 0;
4934
4935         sa = memdup_user(arg, sizeof(*sa));
4936         if (IS_ERR(sa))
4937                 return PTR_ERR(sa);
4938
4939         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4940
4941         if (ret)
4942                 goto out;
4943
4944         ret = copy_to_user(arg, sa, sizeof(*sa));
4945         if (ret)
4946                 ret = -EFAULT;
4947
4948 out:
4949         kfree(sa);
4950         return ret;
4951 }
4952
4953 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4954                                         void __user *arg)
4955 {
4956         size_t len;
4957         int ret;
4958         char label[BTRFS_LABEL_SIZE];
4959
4960         spin_lock(&fs_info->super_lock);
4961         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4962         spin_unlock(&fs_info->super_lock);
4963
4964         len = strnlen(label, BTRFS_LABEL_SIZE);
4965
4966         if (len == BTRFS_LABEL_SIZE) {
4967                 btrfs_warn(fs_info,
4968                            "label is too long, return the first %zu bytes",
4969                            --len);
4970         }
4971
4972         ret = copy_to_user(arg, label, len);
4973
4974         return ret ? -EFAULT : 0;
4975 }
4976
4977 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4978 {
4979         struct inode *inode = file_inode(file);
4980         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4981         struct btrfs_root *root = BTRFS_I(inode)->root;
4982         struct btrfs_super_block *super_block = fs_info->super_copy;
4983         struct btrfs_trans_handle *trans;
4984         char label[BTRFS_LABEL_SIZE];
4985         int ret;
4986
4987         if (!capable(CAP_SYS_ADMIN))
4988                 return -EPERM;
4989
4990         if (copy_from_user(label, arg, sizeof(label)))
4991                 return -EFAULT;
4992
4993         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4994                 btrfs_err(fs_info,
4995                           "unable to set label with more than %d bytes",
4996                           BTRFS_LABEL_SIZE - 1);
4997                 return -EINVAL;
4998         }
4999
5000         ret = mnt_want_write_file(file);
5001         if (ret)
5002                 return ret;
5003
5004         trans = btrfs_start_transaction(root, 0);
5005         if (IS_ERR(trans)) {
5006                 ret = PTR_ERR(trans);
5007                 goto out_unlock;
5008         }
5009
5010         spin_lock(&fs_info->super_lock);
5011         strcpy(super_block->label, label);
5012         spin_unlock(&fs_info->super_lock);
5013         ret = btrfs_commit_transaction(trans);
5014
5015 out_unlock:
5016         mnt_drop_write_file(file);
5017         return ret;
5018 }
5019
5020 #define INIT_FEATURE_FLAGS(suffix) \
5021         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5022           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5023           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5024
5025 int btrfs_ioctl_get_supported_features(void __user *arg)
5026 {
5027         static const struct btrfs_ioctl_feature_flags features[3] = {
5028                 INIT_FEATURE_FLAGS(SUPP),
5029                 INIT_FEATURE_FLAGS(SAFE_SET),
5030                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5031         };
5032
5033         if (copy_to_user(arg, &features, sizeof(features)))
5034                 return -EFAULT;
5035
5036         return 0;
5037 }
5038
5039 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5040                                         void __user *arg)
5041 {
5042         struct btrfs_super_block *super_block = fs_info->super_copy;
5043         struct btrfs_ioctl_feature_flags features;
5044
5045         features.compat_flags = btrfs_super_compat_flags(super_block);
5046         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5047         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5048
5049         if (copy_to_user(arg, &features, sizeof(features)))
5050                 return -EFAULT;
5051
5052         return 0;
5053 }
5054
5055 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5056                               enum btrfs_feature_set set,
5057                               u64 change_mask, u64 flags, u64 supported_flags,
5058                               u64 safe_set, u64 safe_clear)
5059 {
5060         const char *type = btrfs_feature_set_name(set);
5061         char *names;
5062         u64 disallowed, unsupported;
5063         u64 set_mask = flags & change_mask;
5064         u64 clear_mask = ~flags & change_mask;
5065
5066         unsupported = set_mask & ~supported_flags;
5067         if (unsupported) {
5068                 names = btrfs_printable_features(set, unsupported);
5069                 if (names) {
5070                         btrfs_warn(fs_info,
5071                                    "this kernel does not support the %s feature bit%s",
5072                                    names, strchr(names, ',') ? "s" : "");
5073                         kfree(names);
5074                 } else
5075                         btrfs_warn(fs_info,
5076                                    "this kernel does not support %s bits 0x%llx",
5077                                    type, unsupported);
5078                 return -EOPNOTSUPP;
5079         }
5080
5081         disallowed = set_mask & ~safe_set;
5082         if (disallowed) {
5083                 names = btrfs_printable_features(set, disallowed);
5084                 if (names) {
5085                         btrfs_warn(fs_info,
5086                                    "can't set the %s feature bit%s while mounted",
5087                                    names, strchr(names, ',') ? "s" : "");
5088                         kfree(names);
5089                 } else
5090                         btrfs_warn(fs_info,
5091                                    "can't set %s bits 0x%llx while mounted",
5092                                    type, disallowed);
5093                 return -EPERM;
5094         }
5095
5096         disallowed = clear_mask & ~safe_clear;
5097         if (disallowed) {
5098                 names = btrfs_printable_features(set, disallowed);
5099                 if (names) {
5100                         btrfs_warn(fs_info,
5101                                    "can't clear the %s feature bit%s while mounted",
5102                                    names, strchr(names, ',') ? "s" : "");
5103                         kfree(names);
5104                 } else
5105                         btrfs_warn(fs_info,
5106                                    "can't clear %s bits 0x%llx while mounted",
5107                                    type, disallowed);
5108                 return -EPERM;
5109         }
5110
5111         return 0;
5112 }
5113
5114 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5115 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5116                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5117                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5118                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5119
5120 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5121 {
5122         struct inode *inode = file_inode(file);
5123         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5124         struct btrfs_root *root = BTRFS_I(inode)->root;
5125         struct btrfs_super_block *super_block = fs_info->super_copy;
5126         struct btrfs_ioctl_feature_flags flags[2];
5127         struct btrfs_trans_handle *trans;
5128         u64 newflags;
5129         int ret;
5130
5131         if (!capable(CAP_SYS_ADMIN))
5132                 return -EPERM;
5133
5134         if (copy_from_user(flags, arg, sizeof(flags)))
5135                 return -EFAULT;
5136
5137         /* Nothing to do */
5138         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5139             !flags[0].incompat_flags)
5140                 return 0;
5141
5142         ret = check_feature(fs_info, flags[0].compat_flags,
5143                             flags[1].compat_flags, COMPAT);
5144         if (ret)
5145                 return ret;
5146
5147         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5148                             flags[1].compat_ro_flags, COMPAT_RO);
5149         if (ret)
5150                 return ret;
5151
5152         ret = check_feature(fs_info, flags[0].incompat_flags,
5153                             flags[1].incompat_flags, INCOMPAT);
5154         if (ret)
5155                 return ret;
5156
5157         ret = mnt_want_write_file(file);
5158         if (ret)
5159                 return ret;
5160
5161         trans = btrfs_start_transaction(root, 0);
5162         if (IS_ERR(trans)) {
5163                 ret = PTR_ERR(trans);
5164                 goto out_drop_write;
5165         }
5166
5167         spin_lock(&fs_info->super_lock);
5168         newflags = btrfs_super_compat_flags(super_block);
5169         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5170         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5171         btrfs_set_super_compat_flags(super_block, newflags);
5172
5173         newflags = btrfs_super_compat_ro_flags(super_block);
5174         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5175         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5176         btrfs_set_super_compat_ro_flags(super_block, newflags);
5177
5178         newflags = btrfs_super_incompat_flags(super_block);
5179         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5180         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5181         btrfs_set_super_incompat_flags(super_block, newflags);
5182         spin_unlock(&fs_info->super_lock);
5183
5184         ret = btrfs_commit_transaction(trans);
5185 out_drop_write:
5186         mnt_drop_write_file(file);
5187
5188         return ret;
5189 }
5190
5191 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5192 {
5193         struct btrfs_ioctl_send_args *arg;
5194         int ret;
5195
5196         if (compat) {
5197 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5198                 struct btrfs_ioctl_send_args_32 args32;
5199
5200                 ret = copy_from_user(&args32, argp, sizeof(args32));
5201                 if (ret)
5202                         return -EFAULT;
5203                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5204                 if (!arg)
5205                         return -ENOMEM;
5206                 arg->send_fd = args32.send_fd;
5207                 arg->clone_sources_count = args32.clone_sources_count;
5208                 arg->clone_sources = compat_ptr(args32.clone_sources);
5209                 arg->parent_root = args32.parent_root;
5210                 arg->flags = args32.flags;
5211                 memcpy(arg->reserved, args32.reserved,
5212                        sizeof(args32.reserved));
5213 #else
5214                 return -ENOTTY;
5215 #endif
5216         } else {
5217                 arg = memdup_user(argp, sizeof(*arg));
5218                 if (IS_ERR(arg))
5219                         return PTR_ERR(arg);
5220         }
5221         ret = btrfs_ioctl_send(inode, arg);
5222         kfree(arg);
5223         return ret;
5224 }
5225
5226 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5227                                     bool compat)
5228 {
5229         struct btrfs_ioctl_encoded_io_args args = { 0 };
5230         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5231                                              flags);
5232         size_t copy_end;
5233         struct iovec iovstack[UIO_FASTIOV];
5234         struct iovec *iov = iovstack;
5235         struct iov_iter iter;
5236         loff_t pos;
5237         struct kiocb kiocb;
5238         ssize_t ret;
5239
5240         if (!capable(CAP_SYS_ADMIN)) {
5241                 ret = -EPERM;
5242                 goto out_acct;
5243         }
5244
5245         if (compat) {
5246 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5247                 struct btrfs_ioctl_encoded_io_args_32 args32;
5248
5249                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5250                                        flags);
5251                 if (copy_from_user(&args32, argp, copy_end)) {
5252                         ret = -EFAULT;
5253                         goto out_acct;
5254                 }
5255                 args.iov = compat_ptr(args32.iov);
5256                 args.iovcnt = args32.iovcnt;
5257                 args.offset = args32.offset;
5258                 args.flags = args32.flags;
5259 #else
5260                 return -ENOTTY;
5261 #endif
5262         } else {
5263                 copy_end = copy_end_kernel;
5264                 if (copy_from_user(&args, argp, copy_end)) {
5265                         ret = -EFAULT;
5266                         goto out_acct;
5267                 }
5268         }
5269         if (args.flags != 0) {
5270                 ret = -EINVAL;
5271                 goto out_acct;
5272         }
5273
5274         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5275                            &iov, &iter);
5276         if (ret < 0)
5277                 goto out_acct;
5278
5279         if (iov_iter_count(&iter) == 0) {
5280                 ret = 0;
5281                 goto out_iov;
5282         }
5283         pos = args.offset;
5284         ret = rw_verify_area(READ, file, &pos, args.len);
5285         if (ret < 0)
5286                 goto out_iov;
5287
5288         init_sync_kiocb(&kiocb, file);
5289         kiocb.ki_pos = pos;
5290
5291         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5292         if (ret >= 0) {
5293                 fsnotify_access(file);
5294                 if (copy_to_user(argp + copy_end,
5295                                  (char *)&args + copy_end_kernel,
5296                                  sizeof(args) - copy_end_kernel))
5297                         ret = -EFAULT;
5298         }
5299
5300 out_iov:
5301         kfree(iov);
5302 out_acct:
5303         if (ret > 0)
5304                 add_rchar(current, ret);
5305         inc_syscr(current);
5306         return ret;
5307 }
5308
5309 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5310 {
5311         struct btrfs_ioctl_encoded_io_args args;
5312         struct iovec iovstack[UIO_FASTIOV];
5313         struct iovec *iov = iovstack;
5314         struct iov_iter iter;
5315         loff_t pos;
5316         struct kiocb kiocb;
5317         ssize_t ret;
5318
5319         if (!capable(CAP_SYS_ADMIN)) {
5320                 ret = -EPERM;
5321                 goto out_acct;
5322         }
5323
5324         if (!(file->f_mode & FMODE_WRITE)) {
5325                 ret = -EBADF;
5326                 goto out_acct;
5327         }
5328
5329         if (compat) {
5330 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5331                 struct btrfs_ioctl_encoded_io_args_32 args32;
5332
5333                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5334                         ret = -EFAULT;
5335                         goto out_acct;
5336                 }
5337                 args.iov = compat_ptr(args32.iov);
5338                 args.iovcnt = args32.iovcnt;
5339                 args.offset = args32.offset;
5340                 args.flags = args32.flags;
5341                 args.len = args32.len;
5342                 args.unencoded_len = args32.unencoded_len;
5343                 args.unencoded_offset = args32.unencoded_offset;
5344                 args.compression = args32.compression;
5345                 args.encryption = args32.encryption;
5346                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5347 #else
5348                 return -ENOTTY;
5349 #endif
5350         } else {
5351                 if (copy_from_user(&args, argp, sizeof(args))) {
5352                         ret = -EFAULT;
5353                         goto out_acct;
5354                 }
5355         }
5356
5357         ret = -EINVAL;
5358         if (args.flags != 0)
5359                 goto out_acct;
5360         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5361                 goto out_acct;
5362         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5363             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5364                 goto out_acct;
5365         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5366             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5367                 goto out_acct;
5368         if (args.unencoded_offset > args.unencoded_len)
5369                 goto out_acct;
5370         if (args.len > args.unencoded_len - args.unencoded_offset)
5371                 goto out_acct;
5372
5373         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5374                            &iov, &iter);
5375         if (ret < 0)
5376                 goto out_acct;
5377
5378         file_start_write(file);
5379
5380         if (iov_iter_count(&iter) == 0) {
5381                 ret = 0;
5382                 goto out_end_write;
5383         }
5384         pos = args.offset;
5385         ret = rw_verify_area(WRITE, file, &pos, args.len);
5386         if (ret < 0)
5387                 goto out_end_write;
5388
5389         init_sync_kiocb(&kiocb, file);
5390         ret = kiocb_set_rw_flags(&kiocb, 0);
5391         if (ret)
5392                 goto out_end_write;
5393         kiocb.ki_pos = pos;
5394
5395         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5396         if (ret > 0)
5397                 fsnotify_modify(file);
5398
5399 out_end_write:
5400         file_end_write(file);
5401         kfree(iov);
5402 out_acct:
5403         if (ret > 0)
5404                 add_wchar(current, ret);
5405         inc_syscw(current);
5406         return ret;
5407 }
5408
5409 long btrfs_ioctl(struct file *file, unsigned int
5410                 cmd, unsigned long arg)
5411 {
5412         struct inode *inode = file_inode(file);
5413         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5414         struct btrfs_root *root = BTRFS_I(inode)->root;
5415         void __user *argp = (void __user *)arg;
5416
5417         switch (cmd) {
5418         case FS_IOC_GETVERSION:
5419                 return btrfs_ioctl_getversion(inode, argp);
5420         case FS_IOC_GETFSLABEL:
5421                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5422         case FS_IOC_SETFSLABEL:
5423                 return btrfs_ioctl_set_fslabel(file, argp);
5424         case FITRIM:
5425                 return btrfs_ioctl_fitrim(fs_info, argp);
5426         case BTRFS_IOC_SNAP_CREATE:
5427                 return btrfs_ioctl_snap_create(file, argp, 0);
5428         case BTRFS_IOC_SNAP_CREATE_V2:
5429                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5430         case BTRFS_IOC_SUBVOL_CREATE:
5431                 return btrfs_ioctl_snap_create(file, argp, 1);
5432         case BTRFS_IOC_SUBVOL_CREATE_V2:
5433                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5434         case BTRFS_IOC_SNAP_DESTROY:
5435                 return btrfs_ioctl_snap_destroy(file, argp, false);
5436         case BTRFS_IOC_SNAP_DESTROY_V2:
5437                 return btrfs_ioctl_snap_destroy(file, argp, true);
5438         case BTRFS_IOC_SUBVOL_GETFLAGS:
5439                 return btrfs_ioctl_subvol_getflags(inode, argp);
5440         case BTRFS_IOC_SUBVOL_SETFLAGS:
5441                 return btrfs_ioctl_subvol_setflags(file, argp);
5442         case BTRFS_IOC_DEFAULT_SUBVOL:
5443                 return btrfs_ioctl_default_subvol(file, argp);
5444         case BTRFS_IOC_DEFRAG:
5445                 return btrfs_ioctl_defrag(file, NULL);
5446         case BTRFS_IOC_DEFRAG_RANGE:
5447                 return btrfs_ioctl_defrag(file, argp);
5448         case BTRFS_IOC_RESIZE:
5449                 return btrfs_ioctl_resize(file, argp);
5450         case BTRFS_IOC_ADD_DEV:
5451                 return btrfs_ioctl_add_dev(fs_info, argp);
5452         case BTRFS_IOC_RM_DEV:
5453                 return btrfs_ioctl_rm_dev(file, argp);
5454         case BTRFS_IOC_RM_DEV_V2:
5455                 return btrfs_ioctl_rm_dev_v2(file, argp);
5456         case BTRFS_IOC_FS_INFO:
5457                 return btrfs_ioctl_fs_info(fs_info, argp);
5458         case BTRFS_IOC_DEV_INFO:
5459                 return btrfs_ioctl_dev_info(fs_info, argp);
5460         case BTRFS_IOC_TREE_SEARCH:
5461                 return btrfs_ioctl_tree_search(inode, argp);
5462         case BTRFS_IOC_TREE_SEARCH_V2:
5463                 return btrfs_ioctl_tree_search_v2(inode, argp);
5464         case BTRFS_IOC_INO_LOOKUP:
5465                 return btrfs_ioctl_ino_lookup(root, argp);
5466         case BTRFS_IOC_INO_PATHS:
5467                 return btrfs_ioctl_ino_to_path(root, argp);
5468         case BTRFS_IOC_LOGICAL_INO:
5469                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5470         case BTRFS_IOC_LOGICAL_INO_V2:
5471                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5472         case BTRFS_IOC_SPACE_INFO:
5473                 return btrfs_ioctl_space_info(fs_info, argp);
5474         case BTRFS_IOC_SYNC: {
5475                 int ret;
5476
5477                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5478                 if (ret)
5479                         return ret;
5480                 ret = btrfs_sync_fs(inode->i_sb, 1);
5481                 /*
5482                  * The transaction thread may want to do more work,
5483                  * namely it pokes the cleaner kthread that will start
5484                  * processing uncleaned subvols.
5485                  */
5486                 wake_up_process(fs_info->transaction_kthread);
5487                 return ret;
5488         }
5489         case BTRFS_IOC_START_SYNC:
5490                 return btrfs_ioctl_start_sync(root, argp);
5491         case BTRFS_IOC_WAIT_SYNC:
5492                 return btrfs_ioctl_wait_sync(fs_info, argp);
5493         case BTRFS_IOC_SCRUB:
5494                 return btrfs_ioctl_scrub(file, argp);
5495         case BTRFS_IOC_SCRUB_CANCEL:
5496                 return btrfs_ioctl_scrub_cancel(fs_info);
5497         case BTRFS_IOC_SCRUB_PROGRESS:
5498                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5499         case BTRFS_IOC_BALANCE_V2:
5500                 return btrfs_ioctl_balance(file, argp);
5501         case BTRFS_IOC_BALANCE_CTL:
5502                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5503         case BTRFS_IOC_BALANCE_PROGRESS:
5504                 return btrfs_ioctl_balance_progress(fs_info, argp);
5505         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5506                 return btrfs_ioctl_set_received_subvol(file, argp);
5507 #ifdef CONFIG_64BIT
5508         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5509                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5510 #endif
5511         case BTRFS_IOC_SEND:
5512                 return _btrfs_ioctl_send(inode, argp, false);
5513 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5514         case BTRFS_IOC_SEND_32:
5515                 return _btrfs_ioctl_send(inode, argp, true);
5516 #endif
5517         case BTRFS_IOC_GET_DEV_STATS:
5518                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5519         case BTRFS_IOC_QUOTA_CTL:
5520                 return btrfs_ioctl_quota_ctl(file, argp);
5521         case BTRFS_IOC_QGROUP_ASSIGN:
5522                 return btrfs_ioctl_qgroup_assign(file, argp);
5523         case BTRFS_IOC_QGROUP_CREATE:
5524                 return btrfs_ioctl_qgroup_create(file, argp);
5525         case BTRFS_IOC_QGROUP_LIMIT:
5526                 return btrfs_ioctl_qgroup_limit(file, argp);
5527         case BTRFS_IOC_QUOTA_RESCAN:
5528                 return btrfs_ioctl_quota_rescan(file, argp);
5529         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5530                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5531         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5532                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5533         case BTRFS_IOC_DEV_REPLACE:
5534                 return btrfs_ioctl_dev_replace(fs_info, argp);
5535         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5536                 return btrfs_ioctl_get_supported_features(argp);
5537         case BTRFS_IOC_GET_FEATURES:
5538                 return btrfs_ioctl_get_features(fs_info, argp);
5539         case BTRFS_IOC_SET_FEATURES:
5540                 return btrfs_ioctl_set_features(file, argp);
5541         case BTRFS_IOC_GET_SUBVOL_INFO:
5542                 return btrfs_ioctl_get_subvol_info(inode, argp);
5543         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5544                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5545         case BTRFS_IOC_INO_LOOKUP_USER:
5546                 return btrfs_ioctl_ino_lookup_user(file, argp);
5547         case FS_IOC_ENABLE_VERITY:
5548                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5549         case FS_IOC_MEASURE_VERITY:
5550                 return fsverity_ioctl_measure(file, argp);
5551         case BTRFS_IOC_ENCODED_READ:
5552                 return btrfs_ioctl_encoded_read(file, argp, false);
5553         case BTRFS_IOC_ENCODED_WRITE:
5554                 return btrfs_ioctl_encoded_write(file, argp, false);
5555 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5556         case BTRFS_IOC_ENCODED_READ_32:
5557                 return btrfs_ioctl_encoded_read(file, argp, true);
5558         case BTRFS_IOC_ENCODED_WRITE_32:
5559                 return btrfs_ioctl_encoded_write(file, argp, true);
5560 #endif
5561         }
5562
5563         return -ENOTTY;
5564 }
5565
5566 #ifdef CONFIG_COMPAT
5567 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5568 {
5569         /*
5570          * These all access 32-bit values anyway so no further
5571          * handling is necessary.
5572          */
5573         switch (cmd) {
5574         case FS_IOC32_GETVERSION:
5575                 cmd = FS_IOC_GETVERSION;
5576                 break;
5577         }
5578
5579         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5580 }
5581 #endif