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/buffer_head.h>
   9 #include <linux/file.h>
  10 #include <linux/fs.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <linux/time.h>
  15 #include <linux/init.h>
  16 #include <linux/string.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/mount.h>
  19 #include <linux/mpage.h>
  20 #include <linux/namei.h>
  21 #include <linux/swap.h>
  22 #include <linux/writeback.h>
  23 #include <linux/compat.h>
  24 #include <linux/bit_spinlock.h>
  25 #include <linux/security.h>
  26 #include <linux/xattr.h>
  27 #include <linux/mm.h>
  28 #include <linux/slab.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/uuid.h>
  31 #include <linux/btrfs.h>
  32 #include <linux/uaccess.h>
  33 #include <linux/iversion.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "print-tree.h"
  39 #include "volumes.h"
  40 #include "locking.h"
  41 #include "inode-map.h"
  42 #include "backref.h"
  43 #include "rcu-string.h"
  44 #include "send.h"
  45 #include "dev-replace.h"
  46 #include "props.h"
  47 #include "sysfs.h"
  48 #include "qgroup.h"
  49 #include "tree-log.h"
  50 #include "compression.h"
  51
  52 #ifdef CONFIG_64BIT
  53 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  54  * structures are incorrect, as the timespec structure from userspace
  55  * is 4 bytes too small. We define these alternatives here to teach
  56  * the kernel about the 32-bit struct packing.
  57  */
  58 struct btrfs_ioctl_timespec_32 {
  59         __u64 sec;
  60         __u32 nsec;
  61 } __attribute__ ((__packed__));
  62
  63 struct btrfs_ioctl_received_subvol_args_32 {
  64         char    uuid[BTRFS_UUID_SIZE];  /* in */
  65         __u64   stransid;               /* in */
  66         __u64   rtransid;               /* out */
  67         struct btrfs_ioctl_timespec_32 stime; /* in */
  68         struct btrfs_ioctl_timespec_32 rtime; /* out */
  69         __u64   flags;                  /* in */
  70         __u64   reserved[16];           /* in */
  71 } __attribute__ ((__packed__));
  72
  73 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  74                                 struct btrfs_ioctl_received_subvol_args_32)
  75 #endif
  76
  77 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  78 struct btrfs_ioctl_send_args_32 {
  79         __s64 send_fd;                  /* in */
  80         __u64 clone_sources_count;      /* in */
  81         compat_uptr_t clone_sources;    /* in */
  82         __u64 parent_root;              /* in */
  83         __u64 flags;                    /* in */
  84         __u64 reserved[4];              /* in */
  85 } __attribute__ ((__packed__));
  86
  87 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  88                                struct btrfs_ioctl_send_args_32)
  89 #endif
  90
  91 static int btrfs_clone(struct inode *src, struct inode *inode,
  92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  93                        int no_time_update);
  94
  95 /* Mask out flags that are inappropriate for the given type of inode. */
  96 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  97                 unsigned int flags)
  98 {
  99         if (S_ISDIR(inode->i_mode))
 100                 return flags;
 101         else if (S_ISREG(inode->i_mode))
 102                 return flags & ~FS_DIRSYNC_FL;
 103         else
 104                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 105 }
 106
 107 /*
 108  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 109  * ioctl.
 110  */
 111 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 112 {
 113         unsigned int iflags = 0;
 114
 115         if (flags & BTRFS_INODE_SYNC)
 116                 iflags |= FS_SYNC_FL;
 117         if (flags & BTRFS_INODE_IMMUTABLE)
 118                 iflags |= FS_IMMUTABLE_FL;
 119         if (flags & BTRFS_INODE_APPEND)
 120                 iflags |= FS_APPEND_FL;
 121         if (flags & BTRFS_INODE_NODUMP)
 122                 iflags |= FS_NODUMP_FL;
 123         if (flags & BTRFS_INODE_NOATIME)
 124                 iflags |= FS_NOATIME_FL;
 125         if (flags & BTRFS_INODE_DIRSYNC)
 126                 iflags |= FS_DIRSYNC_FL;
 127         if (flags & BTRFS_INODE_NODATACOW)
 128                 iflags |= FS_NOCOW_FL;
 129
 130         if (flags & BTRFS_INODE_NOCOMPRESS)
 131                 iflags |= FS_NOCOMP_FL;
 132         else if (flags & BTRFS_INODE_COMPRESS)
 133                 iflags |= FS_COMPR_FL;
 134
 135         return iflags;
 136 }
 137
 138 /*
 139  * Update inode->i_flags based on the btrfs internal flags.
 140  */
 141 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 142 {
 143         struct btrfs_inode *binode = BTRFS_I(inode);
 144         unsigned int new_fl = 0;
 145
 146         if (binode->flags & BTRFS_INODE_SYNC)
 147                 new_fl |= S_SYNC;
 148         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 149                 new_fl |= S_IMMUTABLE;
 150         if (binode->flags & BTRFS_INODE_APPEND)
 151                 new_fl |= S_APPEND;
 152         if (binode->flags & BTRFS_INODE_NOATIME)
 153                 new_fl |= S_NOATIME;
 154         if (binode->flags & BTRFS_INODE_DIRSYNC)
 155                 new_fl |= S_DIRSYNC;
 156
 157         set_mask_bits(&inode->i_flags,
 158                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 159                       new_fl);
 160 }
 161
 162 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 163 {
 164         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 165         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 166
 167         if (copy_to_user(arg, &flags, sizeof(flags)))
 168                 return -EFAULT;
 169         return 0;
 170 }
 171
 172 /* Check if @flags are a supported and valid set of FS_*_FL flags */
 173 static int check_fsflags(unsigned int flags)
 174 {
 175         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 176                       FS_NOATIME_FL | FS_NODUMP_FL | \
 177                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 178                       FS_NOCOMP_FL | FS_COMPR_FL |
 179                       FS_NOCOW_FL))
 180                 return -EOPNOTSUPP;
 181
 182         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 183                 return -EINVAL;
 184
 185         return 0;
 186 }
 187
 188 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 189 {
 190         struct inode *inode = file_inode(file);
 191         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 192         struct btrfs_inode *binode = BTRFS_I(inode);
 193         struct btrfs_root *root = binode->root;
 194         struct btrfs_trans_handle *trans;
 195         unsigned int fsflags, old_fsflags;
 196         int ret;
 197         u64 old_flags;
 198         unsigned int old_i_flags;
 199         umode_t mode;
 200
 201         if (!inode_owner_or_capable(inode))
 202                 return -EPERM;
 203
 204         if (btrfs_root_readonly(root))
 205                 return -EROFS;
 206
 207         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 208                 return -EFAULT;
 209
 210         ret = check_fsflags(fsflags);
 211         if (ret)
 212                 return ret;
 213
 214         ret = mnt_want_write_file(file);
 215         if (ret)
 216                 return ret;
 217
 218         inode_lock(inode);
 219
 220         old_flags = binode->flags;
 221         old_i_flags = inode->i_flags;
 222         mode = inode->i_mode;
 223
 224         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 225         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 226         if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 227                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 228                         ret = -EPERM;
 229                         goto out_unlock;
 230                 }
 231         }
 232
 233         if (fsflags & FS_SYNC_FL)
 234                 binode->flags |= BTRFS_INODE_SYNC;
 235         else
 236                 binode->flags &= ~BTRFS_INODE_SYNC;
 237         if (fsflags & FS_IMMUTABLE_FL)
 238                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 239         else
 240                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 241         if (fsflags & FS_APPEND_FL)
 242                 binode->flags |= BTRFS_INODE_APPEND;
 243         else
 244                 binode->flags &= ~BTRFS_INODE_APPEND;
 245         if (fsflags & FS_NODUMP_FL)
 246                 binode->flags |= BTRFS_INODE_NODUMP;
 247         else
 248                 binode->flags &= ~BTRFS_INODE_NODUMP;
 249         if (fsflags & FS_NOATIME_FL)
 250                 binode->flags |= BTRFS_INODE_NOATIME;
 251         else
 252                 binode->flags &= ~BTRFS_INODE_NOATIME;
 253         if (fsflags & FS_DIRSYNC_FL)
 254                 binode->flags |= BTRFS_INODE_DIRSYNC;
 255         else
 256                 binode->flags &= ~BTRFS_INODE_DIRSYNC;
 257         if (fsflags & FS_NOCOW_FL) {
 258                 if (S_ISREG(mode)) {
 259                         /*
 260                          * It's safe to turn csums off here, no extents exist.
 261                          * Otherwise we want the flag to reflect the real COW
 262                          * status of the file and will not set it.
 263                          */
 264                         if (inode->i_size == 0)
 265                                 binode->flags |= BTRFS_INODE_NODATACOW
 266                                               | BTRFS_INODE_NODATASUM;
 267                 } else {
 268                         binode->flags |= BTRFS_INODE_NODATACOW;
 269                 }
 270         } else {
 271                 /*
 272                  * Revert back under same assumptions as above
 273                  */
 274                 if (S_ISREG(mode)) {
 275                         if (inode->i_size == 0)
 276                                 binode->flags &= ~(BTRFS_INODE_NODATACOW
 277                                              | BTRFS_INODE_NODATASUM);
 278                 } else {
 279                         binode->flags &= ~BTRFS_INODE_NODATACOW;
 280                 }
 281         }
 282
 283         /*
 284          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 285          * flag may be changed automatically if compression code won't make
 286          * things smaller.
 287          */
 288         if (fsflags & FS_NOCOMP_FL) {
 289                 binode->flags &= ~BTRFS_INODE_COMPRESS;
 290                 binode->flags |= BTRFS_INODE_NOCOMPRESS;
 291
 292                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 293                 if (ret && ret != -ENODATA)
 294                         goto out_drop;
 295         } else if (fsflags & FS_COMPR_FL) {
 296                 const char *comp;
 297
 298                 binode->flags |= BTRFS_INODE_COMPRESS;
 299                 binode->flags &= ~BTRFS_INODE_NOCOMPRESS;
 300
 301                 comp = btrfs_compress_type2str(fs_info->compress_type);
 302                 if (!comp || comp[0] == 0)
 303                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 304
 305                 ret = btrfs_set_prop(inode, "btrfs.compression",
 306                                      comp, strlen(comp), 0);
 307                 if (ret)
 308                         goto out_drop;
 309
 310         } else {
 311                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 312                 if (ret && ret != -ENODATA)
 313                         goto out_drop;
 314                 binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 315         }
 316
 317         trans = btrfs_start_transaction(root, 1);
 318         if (IS_ERR(trans)) {
 319                 ret = PTR_ERR(trans);
 320                 goto out_drop;
 321         }
 322
 323         btrfs_sync_inode_flags_to_i_flags(inode);
 324         inode_inc_iversion(inode);
 325         inode->i_ctime = current_time(inode);
 326         ret = btrfs_update_inode(trans, root, inode);
 327
 328         btrfs_end_transaction(trans);
 329  out_drop:
 330         if (ret) {
 331                 binode->flags = old_flags;
 332                 inode->i_flags = old_i_flags;
 333         }
 334
 335  out_unlock:
 336         inode_unlock(inode);
 337         mnt_drop_write_file(file);
 338         return ret;
 339 }
 340
 341 /*
 342  * Translate btrfs internal inode flags to xflags as expected by the
 343  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 344  * silently dropped.
 345  */
 346 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 347 {
 348         unsigned int xflags = 0;
 349
 350         if (flags & BTRFS_INODE_APPEND)
 351                 xflags |= FS_XFLAG_APPEND;
 352         if (flags & BTRFS_INODE_IMMUTABLE)
 353                 xflags |= FS_XFLAG_IMMUTABLE;
 354         if (flags & BTRFS_INODE_NOATIME)
 355                 xflags |= FS_XFLAG_NOATIME;
 356         if (flags & BTRFS_INODE_NODUMP)
 357                 xflags |= FS_XFLAG_NODUMP;
 358         if (flags & BTRFS_INODE_SYNC)
 359                 xflags |= FS_XFLAG_SYNC;
 360
 361         return xflags;
 362 }
 363
 364 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 365 static int check_xflags(unsigned int flags)
 366 {
 367         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 368                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 369                 return -EOPNOTSUPP;
 370         return 0;
 371 }
 372
 373 /*
 374  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 375  * are zeroed.
 376  */
 377 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 378 {
 379         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 380         struct fsxattr fa;
 381
 382         memset(&fa, 0, sizeof(fa));
 383         fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
 384
 385         if (copy_to_user(arg, &fa, sizeof(fa)))
 386                 return -EFAULT;
 387
 388         return 0;
 389 }
 390
 391 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 392 {
 393         struct inode *inode = file_inode(file);
 394         struct btrfs_inode *binode = BTRFS_I(inode);
 395         struct btrfs_root *root = binode->root;
 396         struct btrfs_trans_handle *trans;
 397         struct fsxattr fa;
 398         unsigned old_flags;
 399         unsigned old_i_flags;
 400         int ret = 0;
 401
 402         if (!inode_owner_or_capable(inode))
 403                 return -EPERM;
 404
 405         if (btrfs_root_readonly(root))
 406                 return -EROFS;
 407
 408         memset(&fa, 0, sizeof(fa));
 409         if (copy_from_user(&fa, arg, sizeof(fa)))
 410                 return -EFAULT;
 411
 412         ret = check_xflags(fa.fsx_xflags);
 413         if (ret)
 414                 return ret;
 415
 416         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 417                 return -EOPNOTSUPP;
 418
 419         ret = mnt_want_write_file(file);
 420         if (ret)
 421                 return ret;
 422
 423         inode_lock(inode);
 424
 425         old_flags = binode->flags;
 426         old_i_flags = inode->i_flags;
 427
 428         /* We need the capabilities to change append-only or immutable inode */
 429         if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) ||
 430              (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) &&
 431             !capable(CAP_LINUX_IMMUTABLE)) {
 432                 ret = -EPERM;
 433                 goto out_unlock;
 434         }
 435
 436         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 437                 binode->flags |= BTRFS_INODE_SYNC;
 438         else
 439                 binode->flags &= ~BTRFS_INODE_SYNC;
 440         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 441                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 442         else
 443                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 444         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 445                 binode->flags |= BTRFS_INODE_APPEND;
 446         else
 447                 binode->flags &= ~BTRFS_INODE_APPEND;
 448         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 449                 binode->flags |= BTRFS_INODE_NODUMP;
 450         else
 451                 binode->flags &= ~BTRFS_INODE_NODUMP;
 452         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 453                 binode->flags |= BTRFS_INODE_NOATIME;
 454         else
 455                 binode->flags &= ~BTRFS_INODE_NOATIME;
 456
 457         /* 1 item for the inode */
 458         trans = btrfs_start_transaction(root, 1);
 459         if (IS_ERR(trans)) {
 460                 ret = PTR_ERR(trans);
 461                 goto out_unlock;
 462         }
 463
 464         btrfs_sync_inode_flags_to_i_flags(inode);
 465         inode_inc_iversion(inode);
 466         inode->i_ctime = current_time(inode);
 467         ret = btrfs_update_inode(trans, root, inode);
 468
 469         btrfs_end_transaction(trans);
 470
 471 out_unlock:
 472         if (ret) {
 473                 binode->flags = old_flags;
 474                 inode->i_flags = old_i_flags;
 475         }
 476
 477         inode_unlock(inode);
 478         mnt_drop_write_file(file);
 479
 480         return ret;
 481 }
 482
 483 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 484 {
 485         struct inode *inode = file_inode(file);
 486
 487         return put_user(inode->i_generation, arg);
 488 }
 489
 490 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 491 {
 492         struct inode *inode = file_inode(file);
 493         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 494         struct btrfs_device *device;
 495         struct request_queue *q;
 496         struct fstrim_range range;
 497         u64 minlen = ULLONG_MAX;
 498         u64 num_devices = 0;
 499         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 500         int ret;
 501
 502         if (!capable(CAP_SYS_ADMIN))
 503                 return -EPERM;
 504
 505         rcu_read_lock();
 506         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 507                                 dev_list) {
 508                 if (!device->bdev)
 509                         continue;
 510                 q = bdev_get_queue(device->bdev);
 511                 if (blk_queue_discard(q)) {
 512                         num_devices++;
 513                         minlen = min_t(u64, q->limits.discard_granularity,
 514                                      minlen);
 515                 }
 516         }
 517         rcu_read_unlock();
 518
 519         if (!num_devices)
 520                 return -EOPNOTSUPP;
 521         if (copy_from_user(&range, arg, sizeof(range)))
 522                 return -EFAULT;
 523         if (range.start > total_bytes ||
 524             range.len < fs_info->sb->s_blocksize)
 525                 return -EINVAL;
 526
 527         range.len = min(range.len, total_bytes - range.start);
 528         range.minlen = max(range.minlen, minlen);
 529         ret = btrfs_trim_fs(fs_info, &range);
 530         if (ret < 0)
 531                 return ret;
 532
 533         if (copy_to_user(arg, &range, sizeof(range)))
 534                 return -EFAULT;
 535
 536         return 0;
 537 }
 538
 539 int btrfs_is_empty_uuid(u8 *uuid)
 540 {
 541         int i;
 542
 543         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 544                 if (uuid[i])
 545                         return 0;
 546         }
 547         return 1;
 548 }
 549
 550 static noinline int create_subvol(struct inode *dir,
 551                                   struct dentry *dentry,
 552                                   const char *name, int namelen,
 553                                   u64 *async_transid,
 554                                   struct btrfs_qgroup_inherit *inherit)
 555 {
 556         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 557         struct btrfs_trans_handle *trans;
 558         struct btrfs_key key;
 559         struct btrfs_root_item *root_item;
 560         struct btrfs_inode_item *inode_item;
 561         struct extent_buffer *leaf;
 562         struct btrfs_root *root = BTRFS_I(dir)->root;
 563         struct btrfs_root *new_root;
 564         struct btrfs_block_rsv block_rsv;
 565         struct timespec cur_time = current_time(dir);
 566         struct inode *inode;
 567         int ret;
 568         int err;
 569         u64 objectid;
 570         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 571         u64 index = 0;
 572         uuid_le new_uuid;
 573
 574         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 575         if (!root_item)
 576                 return -ENOMEM;
 577
 578         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 579         if (ret)
 580                 goto fail_free;
 581
 582         /*
 583          * Don't create subvolume whose level is not zero. Or qgroup will be
 584          * screwed up since it assumes subvolume qgroup's level to be 0.
 585          */
 586         if (btrfs_qgroup_level(objectid)) {
 587                 ret = -ENOSPC;
 588                 goto fail_free;
 589         }
 590
 591         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 592         /*
 593          * The same as the snapshot creation, please see the comment
 594          * of create_snapshot().
 595          */
 596         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 597         if (ret)
 598                 goto fail_free;
 599
 600         trans = btrfs_start_transaction(root, 0);
 601         if (IS_ERR(trans)) {
 602                 ret = PTR_ERR(trans);
 603                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 604                 goto fail_free;
 605         }
 606         trans->block_rsv = &block_rsv;
 607         trans->bytes_reserved = block_rsv.size;
 608
 609         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 610         if (ret)
 611                 goto fail;
 612
 613         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 614         if (IS_ERR(leaf)) {
 615                 ret = PTR_ERR(leaf);
 616                 goto fail;
 617         }
 618
 619         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 620         btrfs_set_header_bytenr(leaf, leaf->start);
 621         btrfs_set_header_generation(leaf, trans->transid);
 622         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 623         btrfs_set_header_owner(leaf, objectid);
 624
 625         write_extent_buffer_fsid(leaf, fs_info->fsid);
 626         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 627         btrfs_mark_buffer_dirty(leaf);
 628
 629         inode_item = &root_item->inode;
 630         btrfs_set_stack_inode_generation(inode_item, 1);
 631         btrfs_set_stack_inode_size(inode_item, 3);
 632         btrfs_set_stack_inode_nlink(inode_item, 1);
 633         btrfs_set_stack_inode_nbytes(inode_item,
 634                                      fs_info->nodesize);
 635         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 636
 637         btrfs_set_root_flags(root_item, 0);
 638         btrfs_set_root_limit(root_item, 0);
 639         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 640
 641         btrfs_set_root_bytenr(root_item, leaf->start);
 642         btrfs_set_root_generation(root_item, trans->transid);
 643         btrfs_set_root_level(root_item, 0);
 644         btrfs_set_root_refs(root_item, 1);
 645         btrfs_set_root_used(root_item, leaf->len);
 646         btrfs_set_root_last_snapshot(root_item, 0);
 647
 648         btrfs_set_root_generation_v2(root_item,
 649                         btrfs_root_generation(root_item));
 650         uuid_le_gen(&new_uuid);
 651         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 652         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 653         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 654         root_item->ctime = root_item->otime;
 655         btrfs_set_root_ctransid(root_item, trans->transid);
 656         btrfs_set_root_otransid(root_item, trans->transid);
 657
 658         btrfs_tree_unlock(leaf);
 659         free_extent_buffer(leaf);
 660         leaf = NULL;
 661
 662         btrfs_set_root_dirid(root_item, new_dirid);
 663
 664         key.objectid = objectid;
 665         key.offset = 0;
 666         key.type = BTRFS_ROOT_ITEM_KEY;
 667         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 668                                 root_item);
 669         if (ret)
 670                 goto fail;
 671
 672         key.offset = (u64)-1;
 673         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 674         if (IS_ERR(new_root)) {
 675                 ret = PTR_ERR(new_root);
 676                 btrfs_abort_transaction(trans, ret);
 677                 goto fail;
 678         }
 679
 680         btrfs_record_root_in_trans(trans, new_root);
 681
 682         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 683         if (ret) {
 684                 /* We potentially lose an unused inode item here */
 685                 btrfs_abort_transaction(trans, ret);
 686                 goto fail;
 687         }
 688
 689         mutex_lock(&new_root->objectid_mutex);
 690         new_root->highest_objectid = new_dirid;
 691         mutex_unlock(&new_root->objectid_mutex);
 692
 693         /*
 694          * insert the directory item
 695          */
 696         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 697         if (ret) {
 698                 btrfs_abort_transaction(trans, ret);
 699                 goto fail;
 700         }
 701
 702         ret = btrfs_insert_dir_item(trans, root,
 703                                     name, namelen, BTRFS_I(dir), &key,
 704                                     BTRFS_FT_DIR, index);
 705         if (ret) {
 706                 btrfs_abort_transaction(trans, ret);
 707                 goto fail;
 708         }
 709
 710         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 711         ret = btrfs_update_inode(trans, root, dir);
 712         BUG_ON(ret);
 713
 714         ret = btrfs_add_root_ref(trans, fs_info,
 715                                  objectid, root->root_key.objectid,
 716                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 717         BUG_ON(ret);
 718
 719         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 720                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 721         if (ret)
 722                 btrfs_abort_transaction(trans, ret);
 723
 724 fail:
 725         kfree(root_item);
 726         trans->block_rsv = NULL;
 727         trans->bytes_reserved = 0;
 728         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 729
 730         if (async_transid) {
 731                 *async_transid = trans->transid;
 732                 err = btrfs_commit_transaction_async(trans, 1);
 733                 if (err)
 734                         err = btrfs_commit_transaction(trans);
 735         } else {
 736                 err = btrfs_commit_transaction(trans);
 737         }
 738         if (err && !ret)
 739                 ret = err;
 740
 741         if (!ret) {
 742                 inode = btrfs_lookup_dentry(dir, dentry);
 743                 if (IS_ERR(inode))
 744                         return PTR_ERR(inode);
 745                 d_instantiate(dentry, inode);
 746         }
 747         return ret;
 748
 749 fail_free:
 750         kfree(root_item);
 751         return ret;
 752 }
 753
 754 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 755                            struct dentry *dentry,
 756                            u64 *async_transid, bool readonly,
 757                            struct btrfs_qgroup_inherit *inherit)
 758 {
 759         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 760         struct inode *inode;
 761         struct btrfs_pending_snapshot *pending_snapshot;
 762         struct btrfs_trans_handle *trans;
 763         int ret;
 764
 765         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 766                 return -EINVAL;
 767
 768         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 769         if (!pending_snapshot)
 770                 return -ENOMEM;
 771
 772         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 773                         GFP_KERNEL);
 774         pending_snapshot->path = btrfs_alloc_path();
 775         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 776                 ret = -ENOMEM;
 777                 goto free_pending;
 778         }
 779
 780         atomic_inc(&root->will_be_snapshotted);
 781         smp_mb__after_atomic();
 782         /* wait for no snapshot writes */
 783         wait_event(root->subv_writers->wait,
 784                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 785
 786         ret = btrfs_start_delalloc_inodes(root);
 787         if (ret)
 788                 goto dec_and_free;
 789
 790         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 791
 792         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 793                              BTRFS_BLOCK_RSV_TEMP);
 794         /*
 795          * 1 - parent dir inode
 796          * 2 - dir entries
 797          * 1 - root item
 798          * 2 - root ref/backref
 799          * 1 - root of snapshot
 800          * 1 - UUID item
 801          */
 802         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 803                                         &pending_snapshot->block_rsv, 8,
 804                                         false);
 805         if (ret)
 806                 goto dec_and_free;
 807
 808         pending_snapshot->dentry = dentry;
 809         pending_snapshot->root = root;
 810         pending_snapshot->readonly = readonly;
 811         pending_snapshot->dir = dir;
 812         pending_snapshot->inherit = inherit;
 813
 814         trans = btrfs_start_transaction(root, 0);
 815         if (IS_ERR(trans)) {
 816                 ret = PTR_ERR(trans);
 817                 goto fail;
 818         }
 819
 820         spin_lock(&fs_info->trans_lock);
 821         list_add(&pending_snapshot->list,
 822                  &trans->transaction->pending_snapshots);
 823         spin_unlock(&fs_info->trans_lock);
 824         if (async_transid) {
 825                 *async_transid = trans->transid;
 826                 ret = btrfs_commit_transaction_async(trans, 1);
 827                 if (ret)
 828                         ret = btrfs_commit_transaction(trans);
 829         } else {
 830                 ret = btrfs_commit_transaction(trans);
 831         }
 832         if (ret)
 833                 goto fail;
 834
 835         ret = pending_snapshot->error;
 836         if (ret)
 837                 goto fail;
 838
 839         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 840         if (ret)
 841                 goto fail;
 842
 843         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 844         if (IS_ERR(inode)) {
 845                 ret = PTR_ERR(inode);
 846                 goto fail;
 847         }
 848
 849         d_instantiate(dentry, inode);
 850         ret = 0;
 851 fail:
 852         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 853 dec_and_free:
 854         if (atomic_dec_and_test(&root->will_be_snapshotted))
 855                 wake_up_var(&root->will_be_snapshotted);
 856 free_pending:
 857         kfree(pending_snapshot->root_item);
 858         btrfs_free_path(pending_snapshot->path);
 859         kfree(pending_snapshot);
 860
 861         return ret;
 862 }
 863
 864 /*  copy of may_delete in fs/namei.c()
 865  *      Check whether we can remove a link victim from directory dir, check
 866  *  whether the type of victim is right.
 867  *  1. We can't do it if dir is read-only (done in permission())
 868  *  2. We should have write and exec permissions on dir
 869  *  3. We can't remove anything from append-only dir
 870  *  4. We can't do anything with immutable dir (done in permission())
 871  *  5. If the sticky bit on dir is set we should either
 872  *      a. be owner of dir, or
 873  *      b. be owner of victim, or
 874  *      c. have CAP_FOWNER capability
 875  *  6. If the victim is append-only or immutable we can't do anything with
 876  *     links pointing to it.
 877  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 878  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 879  *  9. We can't remove a root or mountpoint.
 880  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 881  *     nfs_async_unlink().
 882  */
 883
 884 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 885 {
 886         int error;
 887
 888         if (d_really_is_negative(victim))
 889                 return -ENOENT;
 890
 891         BUG_ON(d_inode(victim->d_parent) != dir);
 892         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 893
 894         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 895         if (error)
 896                 return error;
 897         if (IS_APPEND(dir))
 898                 return -EPERM;
 899         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 900             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 901                 return -EPERM;
 902         if (isdir) {
 903                 if (!d_is_dir(victim))
 904                         return -ENOTDIR;
 905                 if (IS_ROOT(victim))
 906                         return -EBUSY;
 907         } else if (d_is_dir(victim))
 908                 return -EISDIR;
 909         if (IS_DEADDIR(dir))
 910                 return -ENOENT;
 911         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 912                 return -EBUSY;
 913         return 0;
 914 }
 915
 916 /* copy of may_create in fs/namei.c() */
 917 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 918 {
 919         if (d_really_is_positive(child))
 920                 return -EEXIST;
 921         if (IS_DEADDIR(dir))
 922                 return -ENOENT;
 923         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 924 }
 925
 926 /*
 927  * Create a new subvolume below @parent.  This is largely modeled after
 928  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 929  * inside this filesystem so it's quite a bit simpler.
 930  */
 931 static noinline int btrfs_mksubvol(const struct path *parent,
 932                                    const char *name, int namelen,
 933                                    struct btrfs_root *snap_src,
 934                                    u64 *async_transid, bool readonly,
 935                                    struct btrfs_qgroup_inherit *inherit)
 936 {
 937         struct inode *dir = d_inode(parent->dentry);
 938         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 939         struct dentry *dentry;
 940         int error;
 941
 942         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 943         if (error == -EINTR)
 944                 return error;
 945
 946         dentry = lookup_one_len(name, parent->dentry, namelen);
 947         error = PTR_ERR(dentry);
 948         if (IS_ERR(dentry))
 949                 goto out_unlock;
 950
 951         error = btrfs_may_create(dir, dentry);
 952         if (error)
 953                 goto out_dput;
 954
 955         /*
 956          * even if this name doesn't exist, we may get hash collisions.
 957          * check for them now when we can safely fail
 958          */
 959         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 960                                                dir->i_ino, name,
 961                                                namelen);
 962         if (error)
 963                 goto out_dput;
 964
 965         down_read(&fs_info->subvol_sem);
 966
 967         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 968                 goto out_up_read;
 969
 970         if (snap_src) {
 971                 error = create_snapshot(snap_src, dir, dentry,
 972                                         async_transid, readonly, inherit);
 973         } else {
 974                 error = create_subvol(dir, dentry, name, namelen,
 975                                       async_transid, inherit);
 976         }
 977         if (!error)
 978                 fsnotify_mkdir(dir, dentry);
 979 out_up_read:
 980         up_read(&fs_info->subvol_sem);
 981 out_dput:
 982         dput(dentry);
 983 out_unlock:
 984         inode_unlock(dir);
 985         return error;
 986 }
 987
 988 /*
 989  * When we're defragging a range, we don't want to kick it off again
 990  * if it is really just waiting for delalloc to send it down.
 991  * If we find a nice big extent or delalloc range for the bytes in the
 992  * file you want to defrag, we return 0 to let you know to skip this
 993  * part of the file
 994  */
 995 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 996 {
 997         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 998         struct extent_map *em = NULL;
 999         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1000         u64 end;
1001
1002         read_lock(&em_tree->lock);
1003         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1004         read_unlock(&em_tree->lock);
1005
1006         if (em) {
1007                 end = extent_map_end(em);
1008                 free_extent_map(em);
1009                 if (end - offset > thresh)
1010                         return 0;
1011         }
1012         /* if we already have a nice delalloc here, just stop */
1013         thresh /= 2;
1014         end = count_range_bits(io_tree, &offset, offset + thresh,
1015                                thresh, EXTENT_DELALLOC, 1);
1016         if (end >= thresh)
1017                 return 0;
1018         return 1;
1019 }
1020
1021 /*
1022  * helper function to walk through a file and find extents
1023  * newer than a specific transid, and smaller than thresh.
1024  *
1025  * This is used by the defragging code to find new and small
1026  * extents
1027  */
1028 static int find_new_extents(struct btrfs_root *root,
1029                             struct inode *inode, u64 newer_than,
1030                             u64 *off, u32 thresh)
1031 {
1032         struct btrfs_path *path;
1033         struct btrfs_key min_key;
1034         struct extent_buffer *leaf;
1035         struct btrfs_file_extent_item *extent;
1036         int type;
1037         int ret;
1038         u64 ino = btrfs_ino(BTRFS_I(inode));
1039
1040         path = btrfs_alloc_path();
1041         if (!path)
1042                 return -ENOMEM;
1043
1044         min_key.objectid = ino;
1045         min_key.type = BTRFS_EXTENT_DATA_KEY;
1046         min_key.offset = *off;
1047
1048         while (1) {
1049                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1050                 if (ret != 0)
1051                         goto none;
1052 process_slot:
1053                 if (min_key.objectid != ino)
1054                         goto none;
1055                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1056                         goto none;
1057
1058                 leaf = path->nodes[0];
1059                 extent = btrfs_item_ptr(leaf, path->slots[0],
1060                                         struct btrfs_file_extent_item);
1061
1062                 type = btrfs_file_extent_type(leaf, extent);
1063                 if (type == BTRFS_FILE_EXTENT_REG &&
1064                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1065                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1066                         *off = min_key.offset;
1067                         btrfs_free_path(path);
1068                         return 0;
1069                 }
1070
1071                 path->slots[0]++;
1072                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1073                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1074                         goto process_slot;
1075                 }
1076
1077                 if (min_key.offset == (u64)-1)
1078                         goto none;
1079
1080                 min_key.offset++;
1081                 btrfs_release_path(path);
1082         }
1083 none:
1084         btrfs_free_path(path);
1085         return -ENOENT;
1086 }
1087
1088 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1089 {
1090         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1091         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1092         struct extent_map *em;
1093         u64 len = PAGE_SIZE;
1094
1095         /*
1096          * hopefully we have this extent in the tree already, try without
1097          * the full extent lock
1098          */
1099         read_lock(&em_tree->lock);
1100         em = lookup_extent_mapping(em_tree, start, len);
1101         read_unlock(&em_tree->lock);
1102
1103         if (!em) {
1104                 struct extent_state *cached = NULL;
1105                 u64 end = start + len - 1;
1106
1107                 /* get the big lock and read metadata off disk */
1108                 lock_extent_bits(io_tree, start, end, &cached);
1109                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1110                 unlock_extent_cached(io_tree, start, end, &cached);
1111
1112                 if (IS_ERR(em))
1113                         return NULL;
1114         }
1115
1116         return em;
1117 }
1118
1119 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1120 {
1121         struct extent_map *next;
1122         bool ret = true;
1123
1124         /* this is the last extent */
1125         if (em->start + em->len >= i_size_read(inode))
1126                 return false;
1127
1128         next = defrag_lookup_extent(inode, em->start + em->len);
1129         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1130                 ret = false;
1131         else if ((em->block_start + em->block_len == next->block_start) &&
1132                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1133                 ret = false;
1134
1135         free_extent_map(next);
1136         return ret;
1137 }
1138
1139 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1140                                u64 *last_len, u64 *skip, u64 *defrag_end,
1141                                int compress)
1142 {
1143         struct extent_map *em;
1144         int ret = 1;
1145         bool next_mergeable = true;
1146         bool prev_mergeable = true;
1147
1148         /*
1149          * make sure that once we start defragging an extent, we keep on
1150          * defragging it
1151          */
1152         if (start < *defrag_end)
1153                 return 1;
1154
1155         *skip = 0;
1156
1157         em = defrag_lookup_extent(inode, start);
1158         if (!em)
1159                 return 0;
1160
1161         /* this will cover holes, and inline extents */
1162         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1163                 ret = 0;
1164                 goto out;
1165         }
1166
1167         if (!*defrag_end)
1168                 prev_mergeable = false;
1169
1170         next_mergeable = defrag_check_next_extent(inode, em);
1171         /*
1172          * we hit a real extent, if it is big or the next extent is not a
1173          * real extent, don't bother defragging it
1174          */
1175         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1176             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1177                 ret = 0;
1178 out:
1179         /*
1180          * last_len ends up being a counter of how many bytes we've defragged.
1181          * every time we choose not to defrag an extent, we reset *last_len
1182          * so that the next tiny extent will force a defrag.
1183          *
1184          * The end result of this is that tiny extents before a single big
1185          * extent will force at least part of that big extent to be defragged.
1186          */
1187         if (ret) {
1188                 *defrag_end = extent_map_end(em);
1189         } else {
1190                 *last_len = 0;
1191                 *skip = extent_map_end(em);
1192                 *defrag_end = 0;
1193         }
1194
1195         free_extent_map(em);
1196         return ret;
1197 }
1198
1199 /*
1200  * it doesn't do much good to defrag one or two pages
1201  * at a time.  This pulls in a nice chunk of pages
1202  * to COW and defrag.
1203  *
1204  * It also makes sure the delalloc code has enough
1205  * dirty data to avoid making new small extents as part
1206  * of the defrag
1207  *
1208  * It's a good idea to start RA on this range
1209  * before calling this.
1210  */
1211 static int cluster_pages_for_defrag(struct inode *inode,
1212                                     struct page **pages,
1213                                     unsigned long start_index,
1214                                     unsigned long num_pages)
1215 {
1216         unsigned long file_end;
1217         u64 isize = i_size_read(inode);
1218         u64 page_start;
1219         u64 page_end;
1220         u64 page_cnt;
1221         int ret;
1222         int i;
1223         int i_done;
1224         struct btrfs_ordered_extent *ordered;
1225         struct extent_state *cached_state = NULL;
1226         struct extent_io_tree *tree;
1227         struct extent_changeset *data_reserved = NULL;
1228         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1229
1230         file_end = (isize - 1) >> PAGE_SHIFT;
1231         if (!isize || start_index > file_end)
1232                 return 0;
1233
1234         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1235
1236         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1237                         start_index << PAGE_SHIFT,
1238                         page_cnt << PAGE_SHIFT);
1239         if (ret)
1240                 return ret;
1241         i_done = 0;
1242         tree = &BTRFS_I(inode)->io_tree;
1243
1244         /* step one, lock all the pages */
1245         for (i = 0; i < page_cnt; i++) {
1246                 struct page *page;
1247 again:
1248                 page = find_or_create_page(inode->i_mapping,
1249                                            start_index + i, mask);
1250                 if (!page)
1251                         break;
1252
1253                 page_start = page_offset(page);
1254                 page_end = page_start + PAGE_SIZE - 1;
1255                 while (1) {
1256                         lock_extent_bits(tree, page_start, page_end,
1257                                          &cached_state);
1258                         ordered = btrfs_lookup_ordered_extent(inode,
1259                                                               page_start);
1260                         unlock_extent_cached(tree, page_start, page_end,
1261                                              &cached_state);
1262                         if (!ordered)
1263                                 break;
1264
1265                         unlock_page(page);
1266                         btrfs_start_ordered_extent(inode, ordered, 1);
1267                         btrfs_put_ordered_extent(ordered);
1268                         lock_page(page);
1269                         /*
1270                          * we unlocked the page above, so we need check if
1271                          * it was released or not.
1272                          */
1273                         if (page->mapping != inode->i_mapping) {
1274                                 unlock_page(page);
1275                                 put_page(page);
1276                                 goto again;
1277                         }
1278                 }
1279
1280                 if (!PageUptodate(page)) {
1281                         btrfs_readpage(NULL, page);
1282                         lock_page(page);
1283                         if (!PageUptodate(page)) {
1284                                 unlock_page(page);
1285                                 put_page(page);
1286                                 ret = -EIO;
1287                                 break;
1288                         }
1289                 }
1290
1291                 if (page->mapping != inode->i_mapping) {
1292                         unlock_page(page);
1293                         put_page(page);
1294                         goto again;
1295                 }
1296
1297                 pages[i] = page;
1298                 i_done++;
1299         }
1300         if (!i_done || ret)
1301                 goto out;
1302
1303         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1304                 goto out;
1305
1306         /*
1307          * so now we have a nice long stream of locked
1308          * and up to date pages, lets wait on them
1309          */
1310         for (i = 0; i < i_done; i++)
1311                 wait_on_page_writeback(pages[i]);
1312
1313         page_start = page_offset(pages[0]);
1314         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1315
1316         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1317                          page_start, page_end - 1, &cached_state);
1318         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1319                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1320                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1321                           &cached_state);
1322
1323         if (i_done != page_cnt) {
1324                 spin_lock(&BTRFS_I(inode)->lock);
1325                 BTRFS_I(inode)->outstanding_extents++;
1326                 spin_unlock(&BTRFS_I(inode)->lock);
1327                 btrfs_delalloc_release_space(inode, data_reserved,
1328                                 start_index << PAGE_SHIFT,
1329                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1330         }
1331
1332
1333         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1334                           &cached_state);
1335
1336         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1337                              page_start, page_end - 1, &cached_state);
1338
1339         for (i = 0; i < i_done; i++) {
1340                 clear_page_dirty_for_io(pages[i]);
1341                 ClearPageChecked(pages[i]);
1342                 set_page_extent_mapped(pages[i]);
1343                 set_page_dirty(pages[i]);
1344                 unlock_page(pages[i]);
1345                 put_page(pages[i]);
1346         }
1347         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1348                                        false);
1349         extent_changeset_free(data_reserved);
1350         return i_done;
1351 out:
1352         for (i = 0; i < i_done; i++) {
1353                 unlock_page(pages[i]);
1354                 put_page(pages[i]);
1355         }
1356         btrfs_delalloc_release_space(inode, data_reserved,
1357                         start_index << PAGE_SHIFT,
1358                         page_cnt << PAGE_SHIFT, true);
1359         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1360                                        true);
1361         extent_changeset_free(data_reserved);
1362         return ret;
1363
1364 }
1365
1366 int btrfs_defrag_file(struct inode *inode, struct file *file,
1367                       struct btrfs_ioctl_defrag_range_args *range,
1368                       u64 newer_than, unsigned long max_to_defrag)
1369 {
1370         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1371         struct btrfs_root *root = BTRFS_I(inode)->root;
1372         struct file_ra_state *ra = NULL;
1373         unsigned long last_index;
1374         u64 isize = i_size_read(inode);
1375         u64 last_len = 0;
1376         u64 skip = 0;
1377         u64 defrag_end = 0;
1378         u64 newer_off = range->start;
1379         unsigned long i;
1380         unsigned long ra_index = 0;
1381         int ret;
1382         int defrag_count = 0;
1383         int compress_type = BTRFS_COMPRESS_ZLIB;
1384         u32 extent_thresh = range->extent_thresh;
1385         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1386         unsigned long cluster = max_cluster;
1387         u64 new_align = ~((u64)SZ_128K - 1);
1388         struct page **pages = NULL;
1389         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1390
1391         if (isize == 0)
1392                 return 0;
1393
1394         if (range->start >= isize)
1395                 return -EINVAL;
1396
1397         if (do_compress) {
1398                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1399                         return -EINVAL;
1400                 if (range->compress_type)
1401                         compress_type = range->compress_type;
1402         }
1403
1404         if (extent_thresh == 0)
1405                 extent_thresh = SZ_256K;
1406
1407         /*
1408          * If we were not given a file, allocate a readahead context. As
1409          * readahead is just an optimization, defrag will work without it so
1410          * we don't error out.
1411          */
1412         if (!file) {
1413                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1414                 if (ra)
1415                         file_ra_state_init(ra, inode->i_mapping);
1416         } else {
1417                 ra = &file->f_ra;
1418         }
1419
1420         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1421         if (!pages) {
1422                 ret = -ENOMEM;
1423                 goto out_ra;
1424         }
1425
1426         /* find the last page to defrag */
1427         if (range->start + range->len > range->start) {
1428                 last_index = min_t(u64, isize - 1,
1429                          range->start + range->len - 1) >> PAGE_SHIFT;
1430         } else {
1431                 last_index = (isize - 1) >> PAGE_SHIFT;
1432         }
1433
1434         if (newer_than) {
1435                 ret = find_new_extents(root, inode, newer_than,
1436                                        &newer_off, SZ_64K);
1437                 if (!ret) {
1438                         range->start = newer_off;
1439                         /*
1440                          * we always align our defrag to help keep
1441                          * the extents in the file evenly spaced
1442                          */
1443                         i = (newer_off & new_align) >> PAGE_SHIFT;
1444                 } else
1445                         goto out_ra;
1446         } else {
1447                 i = range->start >> PAGE_SHIFT;
1448         }
1449         if (!max_to_defrag)
1450                 max_to_defrag = last_index - i + 1;
1451
1452         /*
1453          * make writeback starts from i, so the defrag range can be
1454          * written sequentially.
1455          */
1456         if (i < inode->i_mapping->writeback_index)
1457                 inode->i_mapping->writeback_index = i;
1458
1459         while (i <= last_index && defrag_count < max_to_defrag &&
1460                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1461                 /*
1462                  * make sure we stop running if someone unmounts
1463                  * the FS
1464                  */
1465                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1466                         break;
1467
1468                 if (btrfs_defrag_cancelled(fs_info)) {
1469                         btrfs_debug(fs_info, "defrag_file cancelled");
1470                         ret = -EAGAIN;
1471                         break;
1472                 }
1473
1474                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1475                                          extent_thresh, &last_len, &skip,
1476                                          &defrag_end, do_compress)){
1477                         unsigned long next;
1478                         /*
1479                          * the should_defrag function tells us how much to skip
1480                          * bump our counter by the suggested amount
1481                          */
1482                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1483                         i = max(i + 1, next);
1484                         continue;
1485                 }
1486
1487                 if (!newer_than) {
1488                         cluster = (PAGE_ALIGN(defrag_end) >>
1489                                    PAGE_SHIFT) - i;
1490                         cluster = min(cluster, max_cluster);
1491                 } else {
1492                         cluster = max_cluster;
1493                 }
1494
1495                 if (i + cluster > ra_index) {
1496                         ra_index = max(i, ra_index);
1497                         if (ra)
1498                                 page_cache_sync_readahead(inode->i_mapping, ra,
1499                                                 file, ra_index, cluster);
1500                         ra_index += cluster;
1501                 }
1502
1503                 inode_lock(inode);
1504                 if (do_compress)
1505                         BTRFS_I(inode)->defrag_compress = compress_type;
1506                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1507                 if (ret < 0) {
1508                         inode_unlock(inode);
1509                         goto out_ra;
1510                 }
1511
1512                 defrag_count += ret;
1513                 balance_dirty_pages_ratelimited(inode->i_mapping);
1514                 inode_unlock(inode);
1515
1516                 if (newer_than) {
1517                         if (newer_off == (u64)-1)
1518                                 break;
1519
1520                         if (ret > 0)
1521                                 i += ret;
1522
1523                         newer_off = max(newer_off + 1,
1524                                         (u64)i << PAGE_SHIFT);
1525
1526                         ret = find_new_extents(root, inode, newer_than,
1527                                                &newer_off, SZ_64K);
1528                         if (!ret) {
1529                                 range->start = newer_off;
1530                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1531                         } else {
1532                                 break;
1533                         }
1534                 } else {
1535                         if (ret > 0) {
1536                                 i += ret;
1537                                 last_len += ret << PAGE_SHIFT;
1538                         } else {
1539                                 i++;
1540                                 last_len = 0;
1541                         }
1542                 }
1543         }
1544
1545         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1546                 filemap_flush(inode->i_mapping);
1547                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1548                              &BTRFS_I(inode)->runtime_flags))
1549                         filemap_flush(inode->i_mapping);
1550         }
1551
1552         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1553                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1554         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1555                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1556         }
1557
1558         ret = defrag_count;
1559
1560 out_ra:
1561         if (do_compress) {
1562                 inode_lock(inode);
1563                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1564                 inode_unlock(inode);
1565         }
1566         if (!file)
1567                 kfree(ra);
1568         kfree(pages);
1569         return ret;
1570 }
1571
1572 static noinline int btrfs_ioctl_resize(struct file *file,
1573                                         void __user *arg)
1574 {
1575         struct inode *inode = file_inode(file);
1576         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1577         u64 new_size;
1578         u64 old_size;
1579         u64 devid = 1;
1580         struct btrfs_root *root = BTRFS_I(inode)->root;
1581         struct btrfs_ioctl_vol_args *vol_args;
1582         struct btrfs_trans_handle *trans;
1583         struct btrfs_device *device = NULL;
1584         char *sizestr;
1585         char *retptr;
1586         char *devstr = NULL;
1587         int ret = 0;
1588         int mod = 0;
1589
1590         if (!capable(CAP_SYS_ADMIN))
1591                 return -EPERM;
1592
1593         ret = mnt_want_write_file(file);
1594         if (ret)
1595                 return ret;
1596
1597         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1598                 mnt_drop_write_file(file);
1599                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1600         }
1601
1602         vol_args = memdup_user(arg, sizeof(*vol_args));
1603         if (IS_ERR(vol_args)) {
1604                 ret = PTR_ERR(vol_args);
1605                 goto out;
1606         }
1607
1608         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1609
1610         sizestr = vol_args->name;
1611         devstr = strchr(sizestr, ':');
1612         if (devstr) {
1613                 sizestr = devstr + 1;
1614                 *devstr = '\0';
1615                 devstr = vol_args->name;
1616                 ret = kstrtoull(devstr, 10, &devid);
1617                 if (ret)
1618                         goto out_free;
1619                 if (!devid) {
1620                         ret = -EINVAL;
1621                         goto out_free;
1622                 }
1623                 btrfs_info(fs_info, "resizing devid %llu", devid);
1624         }
1625
1626         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1627         if (!device) {
1628                 btrfs_info(fs_info, "resizer unable to find device %llu",
1629                            devid);
1630                 ret = -ENODEV;
1631                 goto out_free;
1632         }
1633
1634         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1635                 btrfs_info(fs_info,
1636                            "resizer unable to apply on readonly device %llu",
1637                        devid);
1638                 ret = -EPERM;
1639                 goto out_free;
1640         }
1641
1642         if (!strcmp(sizestr, "max"))
1643                 new_size = device->bdev->bd_inode->i_size;
1644         else {
1645                 if (sizestr[0] == '-') {
1646                         mod = -1;
1647                         sizestr++;
1648                 } else if (sizestr[0] == '+') {
1649                         mod = 1;
1650                         sizestr++;
1651                 }
1652                 new_size = memparse(sizestr, &retptr);
1653                 if (*retptr != '\0' || new_size == 0) {
1654                         ret = -EINVAL;
1655                         goto out_free;
1656                 }
1657         }
1658
1659         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1660                 ret = -EPERM;
1661                 goto out_free;
1662         }
1663
1664         old_size = btrfs_device_get_total_bytes(device);
1665
1666         if (mod < 0) {
1667                 if (new_size > old_size) {
1668                         ret = -EINVAL;
1669                         goto out_free;
1670                 }
1671                 new_size = old_size - new_size;
1672         } else if (mod > 0) {
1673                 if (new_size > ULLONG_MAX - old_size) {
1674                         ret = -ERANGE;
1675                         goto out_free;
1676                 }
1677                 new_size = old_size + new_size;
1678         }
1679
1680         if (new_size < SZ_256M) {
1681                 ret = -EINVAL;
1682                 goto out_free;
1683         }
1684         if (new_size > device->bdev->bd_inode->i_size) {
1685                 ret = -EFBIG;
1686                 goto out_free;
1687         }
1688
1689         new_size = round_down(new_size, fs_info->sectorsize);
1690
1691         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1692                           rcu_str_deref(device->name), new_size);
1693
1694         if (new_size > old_size) {
1695                 trans = btrfs_start_transaction(root, 0);
1696                 if (IS_ERR(trans)) {
1697                         ret = PTR_ERR(trans);
1698                         goto out_free;
1699                 }
1700                 ret = btrfs_grow_device(trans, device, new_size);
1701                 btrfs_commit_transaction(trans);
1702         } else if (new_size < old_size) {
1703                 ret = btrfs_shrink_device(device, new_size);
1704         } /* equal, nothing need to do */
1705
1706 out_free:
1707         kfree(vol_args);
1708 out:
1709         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1710         mnt_drop_write_file(file);
1711         return ret;
1712 }
1713
1714 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1715                                 const char *name, unsigned long fd, int subvol,
1716                                 u64 *transid, bool readonly,
1717                                 struct btrfs_qgroup_inherit *inherit)
1718 {
1719         int namelen;
1720         int ret = 0;
1721
1722         if (!S_ISDIR(file_inode(file)->i_mode))
1723                 return -ENOTDIR;
1724
1725         ret = mnt_want_write_file(file);
1726         if (ret)
1727                 goto out;
1728
1729         namelen = strlen(name);
1730         if (strchr(name, '/')) {
1731                 ret = -EINVAL;
1732                 goto out_drop_write;
1733         }
1734
1735         if (name[0] == '.' &&
1736            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1737                 ret = -EEXIST;
1738                 goto out_drop_write;
1739         }
1740
1741         if (subvol) {
1742                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1743                                      NULL, transid, readonly, inherit);
1744         } else {
1745                 struct fd src = fdget(fd);
1746                 struct inode *src_inode;
1747                 if (!src.file) {
1748                         ret = -EINVAL;
1749                         goto out_drop_write;
1750                 }
1751
1752                 src_inode = file_inode(src.file);
1753                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1754                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1755                                    "Snapshot src from another FS");
1756                         ret = -EXDEV;
1757                 } else if (!inode_owner_or_capable(src_inode)) {
1758                         /*
1759                          * Subvolume creation is not restricted, but snapshots
1760                          * are limited to own subvolumes only
1761                          */
1762                         ret = -EPERM;
1763                 } else {
1764                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1765                                              BTRFS_I(src_inode)->root,
1766                                              transid, readonly, inherit);
1767                 }
1768                 fdput(src);
1769         }
1770 out_drop_write:
1771         mnt_drop_write_file(file);
1772 out:
1773         return ret;
1774 }
1775
1776 static noinline int btrfs_ioctl_snap_create(struct file *file,
1777                                             void __user *arg, int subvol)
1778 {
1779         struct btrfs_ioctl_vol_args *vol_args;
1780         int ret;
1781
1782         if (!S_ISDIR(file_inode(file)->i_mode))
1783                 return -ENOTDIR;
1784
1785         vol_args = memdup_user(arg, sizeof(*vol_args));
1786         if (IS_ERR(vol_args))
1787                 return PTR_ERR(vol_args);
1788         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1789
1790         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1791                                               vol_args->fd, subvol,
1792                                               NULL, false, NULL);
1793
1794         kfree(vol_args);
1795         return ret;
1796 }
1797
1798 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1799                                                void __user *arg, int subvol)
1800 {
1801         struct btrfs_ioctl_vol_args_v2 *vol_args;
1802         int ret;
1803         u64 transid = 0;
1804         u64 *ptr = NULL;
1805         bool readonly = false;
1806         struct btrfs_qgroup_inherit *inherit = NULL;
1807
1808         if (!S_ISDIR(file_inode(file)->i_mode))
1809                 return -ENOTDIR;
1810
1811         vol_args = memdup_user(arg, sizeof(*vol_args));
1812         if (IS_ERR(vol_args))
1813                 return PTR_ERR(vol_args);
1814         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1815
1816         if (vol_args->flags &
1817             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1818               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1819                 ret = -EOPNOTSUPP;
1820                 goto free_args;
1821         }
1822
1823         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1824                 ptr = &transid;
1825         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1826                 readonly = true;
1827         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1828                 if (vol_args->size > PAGE_SIZE) {
1829                         ret = -EINVAL;
1830                         goto free_args;
1831                 }
1832                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1833                 if (IS_ERR(inherit)) {
1834                         ret = PTR_ERR(inherit);
1835                         goto free_args;
1836                 }
1837         }
1838
1839         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1840                                               vol_args->fd, subvol, ptr,
1841                                               readonly, inherit);
1842         if (ret)
1843                 goto free_inherit;
1844
1845         if (ptr && copy_to_user(arg +
1846                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1847                                         transid),
1848                                 ptr, sizeof(*ptr)))
1849                 ret = -EFAULT;
1850
1851 free_inherit:
1852         kfree(inherit);
1853 free_args:
1854         kfree(vol_args);
1855         return ret;
1856 }
1857
1858 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1859                                                 void __user *arg)
1860 {
1861         struct inode *inode = file_inode(file);
1862         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1863         struct btrfs_root *root = BTRFS_I(inode)->root;
1864         int ret = 0;
1865         u64 flags = 0;
1866
1867         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1868                 return -EINVAL;
1869
1870         down_read(&fs_info->subvol_sem);
1871         if (btrfs_root_readonly(root))
1872                 flags |= BTRFS_SUBVOL_RDONLY;
1873         up_read(&fs_info->subvol_sem);
1874
1875         if (copy_to_user(arg, &flags, sizeof(flags)))
1876                 ret = -EFAULT;
1877
1878         return ret;
1879 }
1880
1881 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1882                                               void __user *arg)
1883 {
1884         struct inode *inode = file_inode(file);
1885         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1886         struct btrfs_root *root = BTRFS_I(inode)->root;
1887         struct btrfs_trans_handle *trans;
1888         u64 root_flags;
1889         u64 flags;
1890         int ret = 0;
1891
1892         if (!inode_owner_or_capable(inode))
1893                 return -EPERM;
1894
1895         ret = mnt_want_write_file(file);
1896         if (ret)
1897                 goto out;
1898
1899         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1900                 ret = -EINVAL;
1901                 goto out_drop_write;
1902         }
1903
1904         if (copy_from_user(&flags, arg, sizeof(flags))) {
1905                 ret = -EFAULT;
1906                 goto out_drop_write;
1907         }
1908
1909         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1910                 ret = -EINVAL;
1911                 goto out_drop_write;
1912         }
1913
1914         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1915                 ret = -EOPNOTSUPP;
1916                 goto out_drop_write;
1917         }
1918
1919         down_write(&fs_info->subvol_sem);
1920
1921         /* nothing to do */
1922         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1923                 goto out_drop_sem;
1924
1925         root_flags = btrfs_root_flags(&root->root_item);
1926         if (flags & BTRFS_SUBVOL_RDONLY) {
1927                 btrfs_set_root_flags(&root->root_item,
1928                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1929         } else {
1930                 /*
1931                  * Block RO -> RW transition if this subvolume is involved in
1932                  * send
1933                  */
1934                 spin_lock(&root->root_item_lock);
1935                 if (root->send_in_progress == 0) {
1936                         btrfs_set_root_flags(&root->root_item,
1937                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1938                         spin_unlock(&root->root_item_lock);
1939                 } else {
1940                         spin_unlock(&root->root_item_lock);
1941                         btrfs_warn(fs_info,
1942                                    "Attempt to set subvolume %llu read-write during send",
1943                                    root->root_key.objectid);
1944                         ret = -EPERM;
1945                         goto out_drop_sem;
1946                 }
1947         }
1948
1949         trans = btrfs_start_transaction(root, 1);
1950         if (IS_ERR(trans)) {
1951                 ret = PTR_ERR(trans);
1952                 goto out_reset;
1953         }
1954
1955         ret = btrfs_update_root(trans, fs_info->tree_root,
1956                                 &root->root_key, &root->root_item);
1957         if (ret < 0) {
1958                 btrfs_end_transaction(trans);
1959                 goto out_reset;
1960         }
1961
1962         ret = btrfs_commit_transaction(trans);
1963
1964 out_reset:
1965         if (ret)
1966                 btrfs_set_root_flags(&root->root_item, root_flags);
1967 out_drop_sem:
1968         up_write(&fs_info->subvol_sem);
1969 out_drop_write:
1970         mnt_drop_write_file(file);
1971 out:
1972         return ret;
1973 }
1974
1975 static noinline int key_in_sk(struct btrfs_key *key,
1976                               struct btrfs_ioctl_search_key *sk)
1977 {
1978         struct btrfs_key test;
1979         int ret;
1980
1981         test.objectid = sk->min_objectid;
1982         test.type = sk->min_type;
1983         test.offset = sk->min_offset;
1984
1985         ret = btrfs_comp_cpu_keys(key, &test);
1986         if (ret < 0)
1987                 return 0;
1988
1989         test.objectid = sk->max_objectid;
1990         test.type = sk->max_type;
1991         test.offset = sk->max_offset;
1992
1993         ret = btrfs_comp_cpu_keys(key, &test);
1994         if (ret > 0)
1995                 return 0;
1996         return 1;
1997 }
1998
1999 static noinline int copy_to_sk(struct btrfs_path *path,
2000                                struct btrfs_key *key,
2001                                struct btrfs_ioctl_search_key *sk,
2002                                size_t *buf_size,
2003                                char __user *ubuf,
2004                                unsigned long *sk_offset,
2005                                int *num_found)
2006 {
2007         u64 found_transid;
2008         struct extent_buffer *leaf;
2009         struct btrfs_ioctl_search_header sh;
2010         struct btrfs_key test;
2011         unsigned long item_off;
2012         unsigned long item_len;
2013         int nritems;
2014         int i;
2015         int slot;
2016         int ret = 0;
2017
2018         leaf = path->nodes[0];
2019         slot = path->slots[0];
2020         nritems = btrfs_header_nritems(leaf);
2021
2022         if (btrfs_header_generation(leaf) > sk->max_transid) {
2023                 i = nritems;
2024                 goto advance_key;
2025         }
2026         found_transid = btrfs_header_generation(leaf);
2027
2028         for (i = slot; i < nritems; i++) {
2029                 item_off = btrfs_item_ptr_offset(leaf, i);
2030                 item_len = btrfs_item_size_nr(leaf, i);
2031
2032                 btrfs_item_key_to_cpu(leaf, key, i);
2033                 if (!key_in_sk(key, sk))
2034                         continue;
2035
2036                 if (sizeof(sh) + item_len > *buf_size) {
2037                         if (*num_found) {
2038                                 ret = 1;
2039                                 goto out;
2040                         }
2041
2042                         /*
2043                          * return one empty item back for v1, which does not
2044                          * handle -EOVERFLOW
2045                          */
2046
2047                         *buf_size = sizeof(sh) + item_len;
2048                         item_len = 0;
2049                         ret = -EOVERFLOW;
2050                 }
2051
2052                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2053                         ret = 1;
2054                         goto out;
2055                 }
2056
2057                 sh.objectid = key->objectid;
2058                 sh.offset = key->offset;
2059                 sh.type = key->type;
2060                 sh.len = item_len;
2061                 sh.transid = found_transid;
2062
2063                 /* copy search result header */
2064                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2065                         ret = -EFAULT;
2066                         goto out;
2067                 }
2068
2069                 *sk_offset += sizeof(sh);
2070
2071                 if (item_len) {
2072                         char __user *up = ubuf + *sk_offset;
2073                         /* copy the item */
2074                         if (read_extent_buffer_to_user(leaf, up,
2075                                                        item_off, item_len)) {
2076                                 ret = -EFAULT;
2077                                 goto out;
2078                         }
2079
2080                         *sk_offset += item_len;
2081                 }
2082                 (*num_found)++;
2083
2084                 if (ret) /* -EOVERFLOW from above */
2085                         goto out;
2086
2087                 if (*num_found >= sk->nr_items) {
2088                         ret = 1;
2089                         goto out;
2090                 }
2091         }
2092 advance_key:
2093         ret = 0;
2094         test.objectid = sk->max_objectid;
2095         test.type = sk->max_type;
2096         test.offset = sk->max_offset;
2097         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2098                 ret = 1;
2099         else if (key->offset < (u64)-1)
2100                 key->offset++;
2101         else if (key->type < (u8)-1) {
2102                 key->offset = 0;
2103                 key->type++;
2104         } else if (key->objectid < (u64)-1) {
2105                 key->offset = 0;
2106                 key->type = 0;
2107                 key->objectid++;
2108         } else
2109                 ret = 1;
2110 out:
2111         /*
2112          *  0: all items from this leaf copied, continue with next
2113          *  1: * more items can be copied, but unused buffer is too small
2114          *     * all items were found
2115          *     Either way, it will stops the loop which iterates to the next
2116          *     leaf
2117          *  -EOVERFLOW: item was to large for buffer
2118          *  -EFAULT: could not copy extent buffer back to userspace
2119          */
2120         return ret;
2121 }
2122
2123 static noinline int search_ioctl(struct inode *inode,
2124                                  struct btrfs_ioctl_search_key *sk,
2125                                  size_t *buf_size,
2126                                  char __user *ubuf)
2127 {
2128         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2129         struct btrfs_root *root;
2130         struct btrfs_key key;
2131         struct btrfs_path *path;
2132         int ret;
2133         int num_found = 0;
2134         unsigned long sk_offset = 0;
2135
2136         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2137                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2138                 return -EOVERFLOW;
2139         }
2140
2141         path = btrfs_alloc_path();
2142         if (!path)
2143                 return -ENOMEM;
2144
2145         if (sk->tree_id == 0) {
2146                 /* search the root of the inode that was passed */
2147                 root = BTRFS_I(inode)->root;
2148         } else {
2149                 key.objectid = sk->tree_id;
2150                 key.type = BTRFS_ROOT_ITEM_KEY;
2151                 key.offset = (u64)-1;
2152                 root = btrfs_read_fs_root_no_name(info, &key);
2153                 if (IS_ERR(root)) {
2154                         btrfs_free_path(path);
2155                         return PTR_ERR(root);
2156                 }
2157         }
2158
2159         key.objectid = sk->min_objectid;
2160         key.type = sk->min_type;
2161         key.offset = sk->min_offset;
2162
2163         while (1) {
2164                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2165                 if (ret != 0) {
2166                         if (ret > 0)
2167                                 ret = 0;
2168                         goto err;
2169                 }
2170                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2171                                  &sk_offset, &num_found);
2172                 btrfs_release_path(path);
2173                 if (ret)
2174                         break;
2175
2176         }
2177         if (ret > 0)
2178                 ret = 0;
2179 err:
2180         sk->nr_items = num_found;
2181         btrfs_free_path(path);
2182         return ret;
2183 }
2184
2185 static noinline int btrfs_ioctl_tree_search(struct file *file,
2186                                            void __user *argp)
2187 {
2188         struct btrfs_ioctl_search_args __user *uargs;
2189         struct btrfs_ioctl_search_key sk;
2190         struct inode *inode;
2191         int ret;
2192         size_t buf_size;
2193
2194         if (!capable(CAP_SYS_ADMIN))
2195                 return -EPERM;
2196
2197         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2198
2199         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2200                 return -EFAULT;
2201
2202         buf_size = sizeof(uargs->buf);
2203
2204         inode = file_inode(file);
2205         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2206
2207         /*
2208          * In the origin implementation an overflow is handled by returning a
2209          * search header with a len of zero, so reset ret.
2210          */
2211         if (ret == -EOVERFLOW)
2212                 ret = 0;
2213
2214         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2215                 ret = -EFAULT;
2216         return ret;
2217 }
2218
2219 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2220                                                void __user *argp)
2221 {
2222         struct btrfs_ioctl_search_args_v2 __user *uarg;
2223         struct btrfs_ioctl_search_args_v2 args;
2224         struct inode *inode;
2225         int ret;
2226         size_t buf_size;
2227         const size_t buf_limit = SZ_16M;
2228
2229         if (!capable(CAP_SYS_ADMIN))
2230                 return -EPERM;
2231
2232         /* copy search header and buffer size */
2233         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2234         if (copy_from_user(&args, uarg, sizeof(args)))
2235                 return -EFAULT;
2236
2237         buf_size = args.buf_size;
2238
2239         /* limit result size to 16MB */
2240         if (buf_size > buf_limit)
2241                 buf_size = buf_limit;
2242
2243         inode = file_inode(file);
2244         ret = search_ioctl(inode, &args.key, &buf_size,
2245                            (char __user *)(&uarg->buf[0]));
2246         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2247                 ret = -EFAULT;
2248         else if (ret == -EOVERFLOW &&
2249                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2250                 ret = -EFAULT;
2251
2252         return ret;
2253 }
2254
2255 /*
2256  * Search INODE_REFs to identify path name of 'dirid' directory
2257  * in a 'tree_id' tree. and sets path name to 'name'.
2258  */
2259 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2260                                 u64 tree_id, u64 dirid, char *name)
2261 {
2262         struct btrfs_root *root;
2263         struct btrfs_key key;
2264         char *ptr;
2265         int ret = -1;
2266         int slot;
2267         int len;
2268         int total_len = 0;
2269         struct btrfs_inode_ref *iref;
2270         struct extent_buffer *l;
2271         struct btrfs_path *path;
2272
2273         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2274                 name[0]='\0';
2275                 return 0;
2276         }
2277
2278         path = btrfs_alloc_path();
2279         if (!path)
2280                 return -ENOMEM;
2281
2282         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2283
2284         key.objectid = tree_id;
2285         key.type = BTRFS_ROOT_ITEM_KEY;
2286         key.offset = (u64)-1;
2287         root = btrfs_read_fs_root_no_name(info, &key);
2288         if (IS_ERR(root)) {
2289                 ret = PTR_ERR(root);
2290                 goto out;
2291         }
2292
2293         key.objectid = dirid;
2294         key.type = BTRFS_INODE_REF_KEY;
2295         key.offset = (u64)-1;
2296
2297         while (1) {
2298                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2299                 if (ret < 0)
2300                         goto out;
2301                 else if (ret > 0) {
2302                         ret = btrfs_previous_item(root, path, dirid,
2303                                                   BTRFS_INODE_REF_KEY);
2304                         if (ret < 0)
2305                                 goto out;
2306                         else if (ret > 0) {
2307                                 ret = -ENOENT;
2308                                 goto out;
2309                         }
2310                 }
2311
2312                 l = path->nodes[0];
2313                 slot = path->slots[0];
2314                 btrfs_item_key_to_cpu(l, &key, slot);
2315
2316                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2317                 len = btrfs_inode_ref_name_len(l, iref);
2318                 ptr -= len + 1;
2319                 total_len += len + 1;
2320                 if (ptr < name) {
2321                         ret = -ENAMETOOLONG;
2322                         goto out;
2323                 }
2324
2325                 *(ptr + len) = '/';
2326                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2327
2328                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2329                         break;
2330
2331                 btrfs_release_path(path);
2332                 key.objectid = key.offset;
2333                 key.offset = (u64)-1;
2334                 dirid = key.objectid;
2335         }
2336         memmove(name, ptr, total_len);
2337         name[total_len] = '\0';
2338         ret = 0;
2339 out:
2340         btrfs_free_path(path);
2341         return ret;
2342 }
2343
2344 static int btrfs_search_path_in_tree_user(struct inode *inode,
2345                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2346 {
2347         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2348         struct super_block *sb = inode->i_sb;
2349         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2350         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2351         u64 dirid = args->dirid;
2352         unsigned long item_off;
2353         unsigned long item_len;
2354         struct btrfs_inode_ref *iref;
2355         struct btrfs_root_ref *rref;
2356         struct btrfs_root *root;
2357         struct btrfs_path *path;
2358         struct btrfs_key key, key2;
2359         struct extent_buffer *leaf;
2360         struct inode *temp_inode;
2361         char *ptr;
2362         int slot;
2363         int len;
2364         int total_len = 0;
2365         int ret;
2366
2367         path = btrfs_alloc_path();
2368         if (!path)
2369                 return -ENOMEM;
2370
2371         /*
2372          * If the bottom subvolume does not exist directly under upper_limit,
2373          * construct the path in from the bottom up.
2374          */
2375         if (dirid != upper_limit.objectid) {
2376                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2377
2378                 key.objectid = treeid;
2379                 key.type = BTRFS_ROOT_ITEM_KEY;
2380                 key.offset = (u64)-1;
2381                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2382                 if (IS_ERR(root)) {
2383                         ret = PTR_ERR(root);
2384                         goto out;
2385                 }
2386
2387                 key.objectid = dirid;
2388                 key.type = BTRFS_INODE_REF_KEY;
2389                 key.offset = (u64)-1;
2390                 while (1) {
2391                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2392                         if (ret < 0) {
2393                                 goto out;
2394                         } else if (ret > 0) {
2395                                 ret = btrfs_previous_item(root, path, dirid,
2396                                                           BTRFS_INODE_REF_KEY);
2397                                 if (ret < 0) {
2398                                         goto out;
2399                                 } else if (ret > 0) {
2400                                         ret = -ENOENT;
2401                                         goto out;
2402                                 }
2403                         }
2404
2405                         leaf = path->nodes[0];
2406                         slot = path->slots[0];
2407                         btrfs_item_key_to_cpu(leaf, &key, slot);
2408
2409                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2410                         len = btrfs_inode_ref_name_len(leaf, iref);
2411                         ptr -= len + 1;
2412                         total_len += len + 1;
2413                         if (ptr < args->path) {
2414                                 ret = -ENAMETOOLONG;
2415                                 goto out;
2416                         }
2417
2418                         *(ptr + len) = '/';
2419                         read_extent_buffer(leaf, ptr,
2420                                         (unsigned long)(iref + 1), len);
2421
2422                         /* Check the read+exec permission of this directory */
2423                         ret = btrfs_previous_item(root, path, dirid,
2424                                                   BTRFS_INODE_ITEM_KEY);
2425                         if (ret < 0) {
2426                                 goto out;
2427                         } else if (ret > 0) {
2428                                 ret = -ENOENT;
2429                                 goto out;
2430                         }
2431
2432                         leaf = path->nodes[0];
2433                         slot = path->slots[0];
2434                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2435                         if (key2.objectid != dirid) {
2436                                 ret = -ENOENT;
2437                                 goto out;
2438                         }
2439
2440                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2441                         if (IS_ERR(temp_inode)) {
2442                                 ret = PTR_ERR(temp_inode);
2443                                 goto out;
2444                         }
2445                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2446                         iput(temp_inode);
2447                         if (ret) {
2448                                 ret = -EACCES;
2449                                 goto out;
2450                         }
2451
2452                         if (key.offset == upper_limit.objectid)
2453                                 break;
2454                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2455                                 ret = -EACCES;
2456                                 goto out;
2457                         }
2458
2459                         btrfs_release_path(path);
2460                         key.objectid = key.offset;
2461                         key.offset = (u64)-1;
2462                         dirid = key.objectid;
2463                 }
2464
2465                 memmove(args->path, ptr, total_len);
2466                 args->path[total_len] = '\0';
2467                 btrfs_release_path(path);
2468         }
2469
2470         /* Get the bottom subvolume's name from ROOT_REF */
2471         root = fs_info->tree_root;
2472         key.objectid = treeid;
2473         key.type = BTRFS_ROOT_REF_KEY;
2474         key.offset = args->treeid;
2475         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2476         if (ret < 0) {
2477                 goto out;
2478         } else if (ret > 0) {
2479                 ret = -ENOENT;
2480                 goto out;
2481         }
2482
2483         leaf = path->nodes[0];
2484         slot = path->slots[0];
2485         btrfs_item_key_to_cpu(leaf, &key, slot);
2486
2487         item_off = btrfs_item_ptr_offset(leaf, slot);
2488         item_len = btrfs_item_size_nr(leaf, slot);
2489         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2490         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2491         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2492                 ret = -EINVAL;
2493                 goto out;
2494         }
2495
2496         /* Copy subvolume's name */
2497         item_off += sizeof(struct btrfs_root_ref);
2498         item_len -= sizeof(struct btrfs_root_ref);
2499         read_extent_buffer(leaf, args->name, item_off, item_len);
2500         args->name[item_len] = 0;
2501
2502 out:
2503         btrfs_free_path(path);
2504         return ret;
2505 }
2506
2507 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2508                                            void __user *argp)
2509 {
2510          struct btrfs_ioctl_ino_lookup_args *args;
2511          struct inode *inode;
2512         int ret = 0;
2513
2514         args = memdup_user(argp, sizeof(*args));
2515         if (IS_ERR(args))
2516                 return PTR_ERR(args);
2517
2518         inode = file_inode(file);
2519
2520         /*
2521          * Unprivileged query to obtain the containing subvolume root id. The
2522          * path is reset so it's consistent with btrfs_search_path_in_tree.
2523          */
2524         if (args->treeid == 0)
2525                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2526
2527         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2528                 args->name[0] = 0;
2529                 goto out;
2530         }
2531
2532         if (!capable(CAP_SYS_ADMIN)) {
2533                 ret = -EPERM;
2534                 goto out;
2535         }
2536
2537         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2538                                         args->treeid, args->objectid,
2539                                         args->name);
2540
2541 out:
2542         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2543                 ret = -EFAULT;
2544
2545         kfree(args);
2546         return ret;
2547 }
2548
2549 /*
2550  * Version of ino_lookup ioctl (unprivileged)
2551  *
2552  * The main differences from ino_lookup ioctl are:
2553  *
2554  *   1. Read + Exec permission will be checked using inode_permission() during
2555  *      path construction. -EACCES will be returned in case of failure.
2556  *   2. Path construction will be stopped at the inode number which corresponds
2557  *      to the fd with which this ioctl is called. If constructed path does not
2558  *      exist under fd's inode, -EACCES will be returned.
2559  *   3. The name of bottom subvolume is also searched and filled.
2560  */
2561 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2562 {
2563         struct btrfs_ioctl_ino_lookup_user_args *args;
2564         struct inode *inode;
2565         int ret;
2566
2567         args = memdup_user(argp, sizeof(*args));
2568         if (IS_ERR(args))
2569                 return PTR_ERR(args);
2570
2571         inode = file_inode(file);
2572
2573         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2574             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2575                 /*
2576                  * The subvolume does not exist under fd with which this is
2577                  * called
2578                  */
2579                 kfree(args);
2580                 return -EACCES;
2581         }
2582
2583         ret = btrfs_search_path_in_tree_user(inode, args);
2584
2585         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2586                 ret = -EFAULT;
2587
2588         kfree(args);
2589         return ret;
2590 }
2591
2592 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2593 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2594 {
2595         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2596         struct btrfs_fs_info *fs_info;
2597         struct btrfs_root *root;
2598         struct btrfs_path *path;
2599         struct btrfs_key key;
2600         struct btrfs_root_item *root_item;
2601         struct btrfs_root_ref *rref;
2602         struct extent_buffer *leaf;
2603         unsigned long item_off;
2604         unsigned long item_len;
2605         struct inode *inode;
2606         int slot;
2607         int ret = 0;
2608
2609         path = btrfs_alloc_path();
2610         if (!path)
2611                 return -ENOMEM;
2612
2613         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2614         if (!subvol_info) {
2615                 btrfs_free_path(path);
2616                 return -ENOMEM;
2617         }
2618
2619         inode = file_inode(file);
2620         fs_info = BTRFS_I(inode)->root->fs_info;
2621
2622         /* Get root_item of inode's subvolume */
2623         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2624         key.type = BTRFS_ROOT_ITEM_KEY;
2625         key.offset = (u64)-1;
2626         root = btrfs_read_fs_root_no_name(fs_info, &key);
2627         if (IS_ERR(root)) {
2628                 ret = PTR_ERR(root);
2629                 goto out;
2630         }
2631         root_item = &root->root_item;
2632
2633         subvol_info->treeid = key.objectid;
2634
2635         subvol_info->generation = btrfs_root_generation(root_item);
2636         subvol_info->flags = btrfs_root_flags(root_item);
2637
2638         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2639         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2640                                                     BTRFS_UUID_SIZE);
2641         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2642                                                     BTRFS_UUID_SIZE);
2643
2644         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2645         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2646         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2647
2648         subvol_info->otransid = btrfs_root_otransid(root_item);
2649         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2650         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2651
2652         subvol_info->stransid = btrfs_root_stransid(root_item);
2653         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2654         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2655
2656         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2657         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2658         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2659
2660         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2661                 /* Search root tree for ROOT_BACKREF of this subvolume */
2662                 root = fs_info->tree_root;
2663
2664                 key.type = BTRFS_ROOT_BACKREF_KEY;
2665                 key.offset = 0;
2666                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2667                 if (ret < 0) {
2668                         goto out;
2669                 } else if (path->slots[0] >=
2670                            btrfs_header_nritems(path->nodes[0])) {
2671                         ret = btrfs_next_leaf(root, path);
2672                         if (ret < 0) {
2673                                 goto out;
2674                         } else if (ret > 0) {
2675                                 ret = -EUCLEAN;
2676                                 goto out;
2677                         }
2678                 }
2679
2680                 leaf = path->nodes[0];
2681                 slot = path->slots[0];
2682                 btrfs_item_key_to_cpu(leaf, &key, slot);
2683                 if (key.objectid == subvol_info->treeid &&
2684                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2685                         subvol_info->parent_id = key.offset;
2686
2687                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2688                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2689
2690                         item_off = btrfs_item_ptr_offset(leaf, slot)
2691                                         + sizeof(struct btrfs_root_ref);
2692                         item_len = btrfs_item_size_nr(leaf, slot)
2693                                         - sizeof(struct btrfs_root_ref);
2694                         read_extent_buffer(leaf, subvol_info->name,
2695                                            item_off, item_len);
2696                 } else {
2697                         ret = -ENOENT;
2698                         goto out;
2699                 }
2700         }
2701
2702         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2703                 ret = -EFAULT;
2704
2705 out:
2706         btrfs_free_path(path);
2707         kzfree(subvol_info);
2708         return ret;
2709 }
2710
2711 /*
2712  * Return ROOT_REF information of the subvolume containing this inode
2713  * except the subvolume name.
2714  */
2715 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2716 {
2717         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2718         struct btrfs_root_ref *rref;
2719         struct btrfs_root *root;
2720         struct btrfs_path *path;
2721         struct btrfs_key key;
2722         struct extent_buffer *leaf;
2723         struct inode *inode;
2724         u64 objectid;
2725         int slot;
2726         int ret;
2727         u8 found;
2728
2729         path = btrfs_alloc_path();
2730         if (!path)
2731                 return -ENOMEM;
2732
2733         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2734         if (IS_ERR(rootrefs)) {
2735                 btrfs_free_path(path);
2736                 return PTR_ERR(rootrefs);
2737         }
2738
2739         inode = file_inode(file);
2740         root = BTRFS_I(inode)->root->fs_info->tree_root;
2741         objectid = BTRFS_I(inode)->root->root_key.objectid;
2742
2743         key.objectid = objectid;
2744         key.type = BTRFS_ROOT_REF_KEY;
2745         key.offset = rootrefs->min_treeid;
2746         found = 0;
2747
2748         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2749         if (ret < 0) {
2750                 goto out;
2751         } else if (path->slots[0] >=
2752                    btrfs_header_nritems(path->nodes[0])) {
2753                 ret = btrfs_next_leaf(root, path);
2754                 if (ret < 0) {
2755                         goto out;
2756                 } else if (ret > 0) {
2757                         ret = -EUCLEAN;
2758                         goto out;
2759                 }
2760         }
2761         while (1) {
2762                 leaf = path->nodes[0];
2763                 slot = path->slots[0];
2764
2765                 btrfs_item_key_to_cpu(leaf, &key, slot);
2766                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2767                         ret = 0;
2768                         goto out;
2769                 }
2770
2771                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2772                         ret = -EOVERFLOW;
2773                         goto out;
2774                 }
2775
2776                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2777                 rootrefs->rootref[found].treeid = key.offset;
2778                 rootrefs->rootref[found].dirid =
2779                                   btrfs_root_ref_dirid(leaf, rref);
2780                 found++;
2781
2782                 ret = btrfs_next_item(root, path);
2783                 if (ret < 0) {
2784                         goto out;
2785                 } else if (ret > 0) {
2786                         ret = -EUCLEAN;
2787                         goto out;
2788                 }
2789         }
2790
2791 out:
2792         if (!ret || ret == -EOVERFLOW) {
2793                 rootrefs->num_items = found;
2794                 /* update min_treeid for next search */
2795                 if (found)
2796                         rootrefs->min_treeid =
2797                                 rootrefs->rootref[found - 1].treeid + 1;
2798                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2799                         ret = -EFAULT;
2800         }
2801
2802         kfree(rootrefs);
2803         btrfs_free_path(path);
2804
2805         return ret;
2806 }
2807
2808 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2809                                              void __user *arg)
2810 {
2811         struct dentry *parent = file->f_path.dentry;
2812         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2813         struct dentry *dentry;
2814         struct inode *dir = d_inode(parent);
2815         struct inode *inode;
2816         struct btrfs_root *root = BTRFS_I(dir)->root;
2817         struct btrfs_root *dest = NULL;
2818         struct btrfs_ioctl_vol_args *vol_args;
2819         int namelen;
2820         int err = 0;
2821
2822         if (!S_ISDIR(dir->i_mode))
2823                 return -ENOTDIR;
2824
2825         vol_args = memdup_user(arg, sizeof(*vol_args));
2826         if (IS_ERR(vol_args))
2827                 return PTR_ERR(vol_args);
2828
2829         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2830         namelen = strlen(vol_args->name);
2831         if (strchr(vol_args->name, '/') ||
2832             strncmp(vol_args->name, "..", namelen) == 0) {
2833                 err = -EINVAL;
2834                 goto out;
2835         }
2836
2837         err = mnt_want_write_file(file);
2838         if (err)
2839                 goto out;
2840
2841
2842         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2843         if (err == -EINTR)
2844                 goto out_drop_write;
2845         dentry = lookup_one_len(vol_args->name, parent, namelen);
2846         if (IS_ERR(dentry)) {
2847                 err = PTR_ERR(dentry);
2848                 goto out_unlock_dir;
2849         }
2850
2851         if (d_really_is_negative(dentry)) {
2852                 err = -ENOENT;
2853                 goto out_dput;
2854         }
2855
2856         inode = d_inode(dentry);
2857         dest = BTRFS_I(inode)->root;
2858         if (!capable(CAP_SYS_ADMIN)) {
2859                 /*
2860                  * Regular user.  Only allow this with a special mount
2861                  * option, when the user has write+exec access to the
2862                  * subvol root, and when rmdir(2) would have been
2863                  * allowed.
2864                  *
2865                  * Note that this is _not_ check that the subvol is
2866                  * empty or doesn't contain data that we wouldn't
2867                  * otherwise be able to delete.
2868                  *
2869                  * Users who want to delete empty subvols should try
2870                  * rmdir(2).
2871                  */
2872                 err = -EPERM;
2873                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2874                         goto out_dput;
2875
2876                 /*
2877                  * Do not allow deletion if the parent dir is the same
2878                  * as the dir to be deleted.  That means the ioctl
2879                  * must be called on the dentry referencing the root
2880                  * of the subvol, not a random directory contained
2881                  * within it.
2882                  */
2883                 err = -EINVAL;
2884                 if (root == dest)
2885                         goto out_dput;
2886
2887                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2888                 if (err)
2889                         goto out_dput;
2890         }
2891
2892         /* check if subvolume may be deleted by a user */
2893         err = btrfs_may_delete(dir, dentry, 1);
2894         if (err)
2895                 goto out_dput;
2896
2897         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2898                 err = -EINVAL;
2899                 goto out_dput;
2900         }
2901
2902         inode_lock(inode);
2903         err = btrfs_delete_subvolume(dir, dentry);
2904         inode_unlock(inode);
2905         if (!err)
2906                 d_delete(dentry);
2907
2908 out_dput:
2909         dput(dentry);
2910 out_unlock_dir:
2911         inode_unlock(dir);
2912 out_drop_write:
2913         mnt_drop_write_file(file);
2914 out:
2915         kfree(vol_args);
2916         return err;
2917 }
2918
2919 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2920 {
2921         struct inode *inode = file_inode(file);
2922         struct btrfs_root *root = BTRFS_I(inode)->root;
2923         struct btrfs_ioctl_defrag_range_args *range;
2924         int ret;
2925
2926         ret = mnt_want_write_file(file);
2927         if (ret)
2928                 return ret;
2929
2930         if (btrfs_root_readonly(root)) {
2931                 ret = -EROFS;
2932                 goto out;
2933         }
2934
2935         switch (inode->i_mode & S_IFMT) {
2936         case S_IFDIR:
2937                 if (!capable(CAP_SYS_ADMIN)) {
2938                         ret = -EPERM;
2939                         goto out;
2940                 }
2941                 ret = btrfs_defrag_root(root);
2942                 break;
2943         case S_IFREG:
2944                 if (!(file->f_mode & FMODE_WRITE)) {
2945                         ret = -EINVAL;
2946                         goto out;
2947                 }
2948
2949                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2950                 if (!range) {
2951                         ret = -ENOMEM;
2952                         goto out;
2953                 }
2954
2955                 if (argp) {
2956                         if (copy_from_user(range, argp,
2957                                            sizeof(*range))) {
2958                                 ret = -EFAULT;
2959                                 kfree(range);
2960                                 goto out;
2961                         }
2962                         /* compression requires us to start the IO */
2963                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2964                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2965                                 range->extent_thresh = (u32)-1;
2966                         }
2967                 } else {
2968                         /* the rest are all set to zero by kzalloc */
2969                         range->len = (u64)-1;
2970                 }
2971                 ret = btrfs_defrag_file(file_inode(file), file,
2972                                         range, BTRFS_OLDEST_GENERATION, 0);
2973                 if (ret > 0)
2974                         ret = 0;
2975                 kfree(range);
2976                 break;
2977         default:
2978                 ret = -EINVAL;
2979         }
2980 out:
2981         mnt_drop_write_file(file);
2982         return ret;
2983 }
2984
2985 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2986 {
2987         struct btrfs_ioctl_vol_args *vol_args;
2988         int ret;
2989
2990         if (!capable(CAP_SYS_ADMIN))
2991                 return -EPERM;
2992
2993         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2994                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2995
2996         vol_args = memdup_user(arg, sizeof(*vol_args));
2997         if (IS_ERR(vol_args)) {
2998                 ret = PTR_ERR(vol_args);
2999                 goto out;
3000         }
3001
3002         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3003         ret = btrfs_init_new_device(fs_info, vol_args->name);
3004
3005         if (!ret)
3006                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3007
3008         kfree(vol_args);
3009 out:
3010         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3011         return ret;
3012 }
3013
3014 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3015 {
3016         struct inode *inode = file_inode(file);
3017         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3018         struct btrfs_ioctl_vol_args_v2 *vol_args;
3019         int ret;
3020
3021         if (!capable(CAP_SYS_ADMIN))
3022                 return -EPERM;
3023
3024         ret = mnt_want_write_file(file);
3025         if (ret)
3026                 return ret;
3027
3028         vol_args = memdup_user(arg, sizeof(*vol_args));
3029         if (IS_ERR(vol_args)) {
3030                 ret = PTR_ERR(vol_args);
3031                 goto err_drop;
3032         }
3033
3034         /* Check for compatibility reject unknown flags */
3035         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3036                 ret = -EOPNOTSUPP;
3037                 goto out;
3038         }
3039
3040         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3041                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3042                 goto out;
3043         }
3044
3045         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3046                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3047         } else {
3048                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3049                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3050         }
3051         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3052
3053         if (!ret) {
3054                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3055                         btrfs_info(fs_info, "device deleted: id %llu",
3056                                         vol_args->devid);
3057                 else
3058                         btrfs_info(fs_info, "device deleted: %s",
3059                                         vol_args->name);
3060         }
3061 out:
3062         kfree(vol_args);
3063 err_drop:
3064         mnt_drop_write_file(file);
3065         return ret;
3066 }
3067
3068 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3069 {
3070         struct inode *inode = file_inode(file);
3071         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3072         struct btrfs_ioctl_vol_args *vol_args;
3073         int ret;
3074
3075         if (!capable(CAP_SYS_ADMIN))
3076                 return -EPERM;
3077
3078         ret = mnt_want_write_file(file);
3079         if (ret)
3080                 return ret;
3081
3082         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3083                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3084                 goto out_drop_write;
3085         }
3086
3087         vol_args = memdup_user(arg, sizeof(*vol_args));
3088         if (IS_ERR(vol_args)) {
3089                 ret = PTR_ERR(vol_args);
3090                 goto out;
3091         }
3092
3093         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3094         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3095
3096         if (!ret)
3097                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3098         kfree(vol_args);
3099 out:
3100         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3101 out_drop_write:
3102         mnt_drop_write_file(file);
3103
3104         return ret;
3105 }
3106
3107 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3108                                 void __user *arg)
3109 {
3110         struct btrfs_ioctl_fs_info_args *fi_args;
3111         struct btrfs_device *device;
3112         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3113         int ret = 0;
3114
3115         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3116         if (!fi_args)
3117                 return -ENOMEM;
3118
3119         rcu_read_lock();
3120         fi_args->num_devices = fs_devices->num_devices;
3121
3122         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3123                 if (device->devid > fi_args->max_id)
3124                         fi_args->max_id = device->devid;
3125         }
3126         rcu_read_unlock();
3127
3128         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3129         fi_args->nodesize = fs_info->nodesize;
3130         fi_args->sectorsize = fs_info->sectorsize;
3131         fi_args->clone_alignment = fs_info->sectorsize;
3132
3133         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3134                 ret = -EFAULT;
3135
3136         kfree(fi_args);
3137         return ret;
3138 }
3139
3140 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3141                                  void __user *arg)
3142 {
3143         struct btrfs_ioctl_dev_info_args *di_args;
3144         struct btrfs_device *dev;
3145         int ret = 0;
3146         char *s_uuid = NULL;
3147
3148         di_args = memdup_user(arg, sizeof(*di_args));
3149         if (IS_ERR(di_args))
3150                 return PTR_ERR(di_args);
3151
3152         if (!btrfs_is_empty_uuid(di_args->uuid))
3153                 s_uuid = di_args->uuid;
3154
3155         rcu_read_lock();
3156         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
3157
3158         if (!dev) {
3159                 ret = -ENODEV;
3160                 goto out;
3161         }
3162
3163         di_args->devid = dev->devid;
3164         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3165         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3166         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3167         if (dev->name) {
3168                 struct rcu_string *name;
3169
3170                 name = rcu_dereference(dev->name);
3171                 strncpy(di_args->path, name->str, sizeof(di_args->path) - 1);
3172                 di_args->path[sizeof(di_args->path) - 1] = 0;
3173         } else {
3174                 di_args->path[0] = '\0';
3175         }
3176
3177 out:
3178         rcu_read_unlock();
3179         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3180                 ret = -EFAULT;
3181
3182         kfree(di_args);
3183         return ret;
3184 }
3185
3186 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3187 {
3188         struct page *page;
3189
3190         page = grab_cache_page(inode->i_mapping, index);
3191         if (!page)
3192                 return ERR_PTR(-ENOMEM);
3193
3194         if (!PageUptodate(page)) {
3195                 int ret;
3196
3197                 ret = btrfs_readpage(NULL, page);
3198                 if (ret)
3199                         return ERR_PTR(ret);
3200                 lock_page(page);
3201                 if (!PageUptodate(page)) {
3202                         unlock_page(page);
3203                         put_page(page);
3204                         return ERR_PTR(-EIO);
3205                 }
3206                 if (page->mapping != inode->i_mapping) {
3207                         unlock_page(page);
3208                         put_page(page);
3209                         return ERR_PTR(-EAGAIN);
3210                 }
3211         }
3212
3213         return page;
3214 }
3215
3216 static int gather_extent_pages(struct inode *inode, struct page **pages,
3217                                int num_pages, u64 off)
3218 {
3219         int i;
3220         pgoff_t index = off >> PAGE_SHIFT;
3221
3222         for (i = 0; i < num_pages; i++) {
3223 again:
3224                 pages[i] = extent_same_get_page(inode, index + i);
3225                 if (IS_ERR(pages[i])) {
3226                         int err = PTR_ERR(pages[i]);
3227
3228                         if (err == -EAGAIN)
3229                                 goto again;
3230                         pages[i] = NULL;
3231                         return err;
3232                 }
3233         }
3234         return 0;
3235 }
3236
3237 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3238                              bool retry_range_locking)
3239 {
3240         /*
3241          * Do any pending delalloc/csum calculations on inode, one way or
3242          * another, and lock file content.
3243          * The locking order is:
3244          *
3245          *   1) pages
3246          *   2) range in the inode's io tree
3247          */
3248         while (1) {
3249                 struct btrfs_ordered_extent *ordered;
3250                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3251                 ordered = btrfs_lookup_first_ordered_extent(inode,
3252                                                             off + len - 1);
3253                 if ((!ordered ||
3254                      ordered->file_offset + ordered->len <= off ||
3255                      ordered->file_offset >= off + len) &&
3256                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3257                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3258                         if (ordered)
3259                                 btrfs_put_ordered_extent(ordered);
3260                         break;
3261                 }
3262                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3263                 if (ordered)
3264                         btrfs_put_ordered_extent(ordered);
3265                 if (!retry_range_locking)
3266                         return -EAGAIN;
3267                 btrfs_wait_ordered_range(inode, off, len);
3268         }
3269         return 0;
3270 }
3271
3272 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3273 {
3274         inode_unlock(inode1);
3275         inode_unlock(inode2);
3276 }
3277
3278 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3279 {
3280         if (inode1 < inode2)
3281                 swap(inode1, inode2);
3282
3283         inode_lock_nested(inode1, I_MUTEX_PARENT);
3284         inode_lock_nested(inode2, I_MUTEX_CHILD);
3285 }
3286
3287 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3288                                       struct inode *inode2, u64 loff2, u64 len)
3289 {
3290         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3291         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3292 }
3293
3294 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3295                                     struct inode *inode2, u64 loff2, u64 len,
3296                                     bool retry_range_locking)
3297 {
3298         int ret;
3299
3300         if (inode1 < inode2) {
3301                 swap(inode1, inode2);
3302                 swap(loff1, loff2);
3303         }
3304         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3305         if (ret)
3306                 return ret;
3307         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3308         if (ret)
3309                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3310                               loff1 + len - 1);
3311         return ret;
3312 }
3313
3314 struct cmp_pages {
3315         int             num_pages;
3316         struct page     **src_pages;
3317         struct page     **dst_pages;
3318 };
3319
3320 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3321 {
3322         int i;
3323         struct page *pg;
3324
3325         for (i = 0; i < cmp->num_pages; i++) {
3326                 pg = cmp->src_pages[i];
3327                 if (pg) {
3328                         unlock_page(pg);
3329                         put_page(pg);
3330                 }
3331                 pg = cmp->dst_pages[i];
3332                 if (pg) {
3333                         unlock_page(pg);
3334                         put_page(pg);
3335                 }
3336         }
3337 }
3338
3339 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3340                                   struct inode *dst, u64 dst_loff,
3341                                   u64 len, struct cmp_pages *cmp)
3342 {
3343         int ret;
3344         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3345
3346         cmp->num_pages = num_pages;
3347
3348         ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3349         if (ret)
3350                 goto out;
3351
3352         ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3353
3354 out:
3355         if (ret)
3356                 btrfs_cmp_data_free(cmp);
3357         return ret;
3358 }
3359
3360 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3361 {
3362         int ret = 0;
3363         int i;
3364         struct page *src_page, *dst_page;
3365         unsigned int cmp_len = PAGE_SIZE;
3366         void *addr, *dst_addr;
3367
3368         i = 0;
3369         while (len) {
3370                 if (len < PAGE_SIZE)
3371                         cmp_len = len;
3372
3373                 BUG_ON(i >= cmp->num_pages);
3374
3375                 src_page = cmp->src_pages[i];
3376                 dst_page = cmp->dst_pages[i];
3377                 ASSERT(PageLocked(src_page));
3378                 ASSERT(PageLocked(dst_page));
3379
3380                 addr = kmap_atomic(src_page);
3381                 dst_addr = kmap_atomic(dst_page);
3382
3383                 flush_dcache_page(src_page);
3384                 flush_dcache_page(dst_page);
3385
3386                 if (memcmp(addr, dst_addr, cmp_len))
3387                         ret = -EBADE;
3388
3389                 kunmap_atomic(addr);
3390                 kunmap_atomic(dst_addr);
3391
3392                 if (ret)
3393                         break;
3394
3395                 len -= cmp_len;
3396                 i++;
3397         }
3398
3399         return ret;
3400 }
3401
3402 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3403                                      u64 olen)
3404 {
3405         u64 len = *plen;
3406         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3407
3408         if (off + olen > inode->i_size || off + olen < off)
3409                 return -EINVAL;
3410
3411         /* if we extend to eof, continue to block boundary */
3412         if (off + len == inode->i_size)
3413                 *plen = len = ALIGN(inode->i_size, bs) - off;
3414
3415         /* Check that we are block aligned - btrfs_clone() requires this */
3416         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3417                 return -EINVAL;
3418
3419         return 0;
3420 }
3421
3422 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3423                                    struct inode *dst, u64 dst_loff,
3424                                    struct cmp_pages *cmp)
3425 {
3426         int ret;
3427         u64 len = olen;
3428         bool same_inode = (src == dst);
3429         u64 same_lock_start = 0;
3430         u64 same_lock_len = 0;
3431
3432         ret = extent_same_check_offsets(src, loff, &len, olen);
3433         if (ret)
3434                 return ret;
3435
3436         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3437         if (ret)
3438                 return ret;
3439
3440         if (same_inode) {
3441                 /*
3442                  * Single inode case wants the same checks, except we
3443                  * don't want our length pushed out past i_size as
3444                  * comparing that data range makes no sense.
3445                  *
3446                  * extent_same_check_offsets() will do this for an
3447                  * unaligned length at i_size, so catch it here and
3448                  * reject the request.
3449                  *
3450                  * This effectively means we require aligned extents
3451                  * for the single-inode case, whereas the other cases
3452                  * allow an unaligned length so long as it ends at
3453                  * i_size.
3454                  */
3455                 if (len != olen)
3456                         return -EINVAL;
3457
3458                 /* Check for overlapping ranges */
3459                 if (dst_loff + len > loff && dst_loff < loff + len)
3460                         return -EINVAL;
3461
3462                 same_lock_start = min_t(u64, loff, dst_loff);
3463                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3464         }
3465
3466 again:
3467         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp);
3468         if (ret)
3469                 return ret;
3470
3471         if (same_inode)
3472                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3473                                         false);
3474         else
3475                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3476                                                false);
3477         /*
3478          * If one of the inodes has dirty pages in the respective range or
3479          * ordered extents, we need to flush dellaloc and wait for all ordered
3480          * extents in the range. We must unlock the pages and the ranges in the
3481          * io trees to avoid deadlocks when flushing delalloc (requires locking
3482          * pages) and when waiting for ordered extents to complete (they require
3483          * range locking).
3484          */
3485         if (ret == -EAGAIN) {
3486                 /*
3487                  * Ranges in the io trees already unlocked. Now unlock all
3488                  * pages before waiting for all IO to complete.
3489                  */
3490                 btrfs_cmp_data_free(cmp);
3491                 if (same_inode) {
3492                         btrfs_wait_ordered_range(src, same_lock_start,
3493                                                  same_lock_len);
3494                 } else {
3495                         btrfs_wait_ordered_range(src, loff, len);
3496                         btrfs_wait_ordered_range(dst, dst_loff, len);
3497                 }
3498                 goto again;
3499         }
3500         ASSERT(ret == 0);
3501         if (WARN_ON(ret)) {
3502                 /* ranges in the io trees already unlocked */
3503                 btrfs_cmp_data_free(cmp);
3504                 return ret;
3505         }
3506
3507         /* pass original length for comparison so we stay within i_size */
3508         ret = btrfs_cmp_data(olen, cmp);
3509         if (ret == 0)
3510                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3511
3512         if (same_inode)
3513                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3514                               same_lock_start + same_lock_len - 1);
3515         else
3516                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3517
3518         btrfs_cmp_data_free(cmp);
3519
3520         return ret;
3521 }
3522
3523 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3524
3525 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3526                              struct inode *dst, u64 dst_loff)
3527 {
3528         int ret;
3529         struct cmp_pages cmp;
3530         int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT;
3531         bool same_inode = (src == dst);
3532         u64 i, tail_len, chunk_count;
3533
3534         if (olen == 0)
3535                 return 0;
3536
3537         if (same_inode)
3538                 inode_lock(src);
3539         else
3540                 btrfs_double_inode_lock(src, dst);
3541
3542         /* don't make the dst file partly checksummed */
3543         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3544             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3545                 ret = -EINVAL;
3546                 goto out_unlock;
3547         }
3548
3549         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3550         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3551         if (chunk_count == 0)
3552                 num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT;
3553
3554         /*
3555          * If deduping ranges in the same inode, locking rules make it
3556          * mandatory to always lock pages in ascending order to avoid deadlocks
3557          * with concurrent tasks (such as starting writeback/delalloc).
3558          */
3559         if (same_inode && dst_loff < loff)
3560                 swap(loff, dst_loff);
3561
3562         /*
3563          * We must gather up all the pages before we initiate our extent
3564          * locking. We use an array for the page pointers. Size of the array is
3565          * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN.
3566          */
3567         cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3568                                        GFP_KERNEL | __GFP_ZERO);
3569         cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3570                                        GFP_KERNEL | __GFP_ZERO);
3571         if (!cmp.src_pages || !cmp.dst_pages) {
3572                 ret = -ENOMEM;
3573                 goto out_free;
3574         }
3575
3576         for (i = 0; i < chunk_count; i++) {
3577                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3578                                               dst, dst_loff, &cmp);
3579                 if (ret)
3580                         goto out_unlock;
3581
3582                 loff += BTRFS_MAX_DEDUPE_LEN;
3583                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3584         }
3585
3586         if (tail_len > 0)
3587                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3588                                               dst_loff, &cmp);
3589
3590 out_unlock:
3591         if (same_inode)
3592                 inode_unlock(src);
3593         else
3594                 btrfs_double_inode_unlock(src, dst);
3595
3596 out_free:
3597         kvfree(cmp.src_pages);
3598         kvfree(cmp.dst_pages);
3599
3600         return ret;
3601 }
3602
3603 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3604                                 struct file *dst_file, u64 dst_loff)
3605 {
3606         struct inode *src = file_inode(src_file);
3607         struct inode *dst = file_inode(dst_file);
3608         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3609         ssize_t res;
3610
3611         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3612                 /*
3613                  * Btrfs does not support blocksize < page_size. As a
3614                  * result, btrfs_cmp_data() won't correctly handle
3615                  * this situation without an update.
3616                  */
3617                 return -EINVAL;
3618         }
3619
3620         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3621         if (res)
3622                 return res;
3623         return olen;
3624 }
3625
3626 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3627                                      struct inode *inode,
3628                                      u64 endoff,
3629                                      const u64 destoff,
3630                                      const u64 olen,
3631                                      int no_time_update)
3632 {
3633         struct btrfs_root *root = BTRFS_I(inode)->root;
3634         int ret;
3635
3636         inode_inc_iversion(inode);
3637         if (!no_time_update)
3638                 inode->i_mtime = inode->i_ctime = current_time(inode);
3639         /*
3640          * We round up to the block size at eof when determining which
3641          * extents to clone above, but shouldn't round up the file size.
3642          */
3643         if (endoff > destoff + olen)
3644                 endoff = destoff + olen;
3645         if (endoff > inode->i_size)
3646                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3647
3648         ret = btrfs_update_inode(trans, root, inode);
3649         if (ret) {
3650                 btrfs_abort_transaction(trans, ret);
3651                 btrfs_end_transaction(trans);
3652                 goto out;
3653         }
3654         ret = btrfs_end_transaction(trans);
3655 out:
3656         return ret;
3657 }
3658
3659 static void clone_update_extent_map(struct btrfs_inode *inode,
3660                                     const struct btrfs_trans_handle *trans,
3661                                     const struct btrfs_path *path,
3662                                     const u64 hole_offset,
3663                                     const u64 hole_len)
3664 {
3665         struct extent_map_tree *em_tree = &inode->extent_tree;
3666         struct extent_map *em;
3667         int ret;
3668
3669         em = alloc_extent_map();
3670         if (!em) {
3671                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3672                 return;
3673         }
3674
3675         if (path) {
3676                 struct btrfs_file_extent_item *fi;
3677
3678                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3679                                     struct btrfs_file_extent_item);
3680                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3681                 em->generation = -1;
3682                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3683                     BTRFS_FILE_EXTENT_INLINE)
3684                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3685                                         &inode->runtime_flags);
3686         } else {
3687                 em->start = hole_offset;
3688                 em->len = hole_len;
3689                 em->ram_bytes = em->len;
3690                 em->orig_start = hole_offset;
3691                 em->block_start = EXTENT_MAP_HOLE;
3692                 em->block_len = 0;
3693                 em->orig_block_len = 0;
3694                 em->compress_type = BTRFS_COMPRESS_NONE;
3695                 em->generation = trans->transid;
3696         }
3697
3698         while (1) {
3699                 write_lock(&em_tree->lock);
3700                 ret = add_extent_mapping(em_tree, em, 1);
3701                 write_unlock(&em_tree->lock);
3702                 if (ret != -EEXIST) {
3703                         free_extent_map(em);
3704                         break;
3705                 }
3706                 btrfs_drop_extent_cache(inode, em->start,
3707                                         em->start + em->len - 1, 0);
3708         }
3709
3710         if (ret)
3711                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3712 }
3713
3714 /*
3715  * Make sure we do not end up inserting an inline extent into a file that has
3716  * already other (non-inline) extents. If a file has an inline extent it can
3717  * not have any other extents and the (single) inline extent must start at the
3718  * file offset 0. Failing to respect these rules will lead to file corruption,
3719  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3720  *
3721  * We can have extents that have been already written to disk or we can have
3722  * dirty ranges still in delalloc, in which case the extent maps and items are
3723  * created only when we run delalloc, and the delalloc ranges might fall outside
3724  * the range we are currently locking in the inode's io tree. So we check the
3725  * inode's i_size because of that (i_size updates are done while holding the
3726  * i_mutex, which we are holding here).
3727  * We also check to see if the inode has a size not greater than "datal" but has
3728  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3729  * protected against such concurrent fallocate calls by the i_mutex).
3730  *
3731  * If the file has no extents but a size greater than datal, do not allow the
3732  * copy because we would need turn the inline extent into a non-inline one (even
3733  * with NO_HOLES enabled). If we find our destination inode only has one inline
3734  * extent, just overwrite it with the source inline extent if its size is less
3735  * than the source extent's size, or we could copy the source inline extent's
3736  * data into the destination inode's inline extent if the later is greater then
3737  * the former.
3738  */
3739 static int clone_copy_inline_extent(struct inode *dst,
3740                                     struct btrfs_trans_handle *trans,
3741                                     struct btrfs_path *path,
3742                                     struct btrfs_key *new_key,
3743                                     const u64 drop_start,
3744                                     const u64 datal,
3745                                     const u64 skip,
3746                                     const u64 size,
3747                                     char *inline_data)
3748 {
3749         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3750         struct btrfs_root *root = BTRFS_I(dst)->root;
3751         const u64 aligned_end = ALIGN(new_key->offset + datal,
3752                                       fs_info->sectorsize);
3753         int ret;
3754         struct btrfs_key key;
3755
3756         if (new_key->offset > 0)
3757                 return -EOPNOTSUPP;
3758
3759         key.objectid = btrfs_ino(BTRFS_I(dst));
3760         key.type = BTRFS_EXTENT_DATA_KEY;
3761         key.offset = 0;
3762         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3763         if (ret < 0) {
3764                 return ret;
3765         } else if (ret > 0) {
3766                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3767                         ret = btrfs_next_leaf(root, path);
3768                         if (ret < 0)
3769                                 return ret;
3770                         else if (ret > 0)
3771