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_flags(umode_t mode, unsigned int flags)
  97 {
  98         if (S_ISDIR(mode))
  99                 return flags;
 100         else if (S_ISREG(mode))
 101                 return flags & ~FS_DIRSYNC_FL;
 102         else
 103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 104 }
 105
 106 /*
 107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 108  */
 109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_update_iflags(struct inode *inode)
 140 {
 141         struct btrfs_inode *ip = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (ip->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (ip->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (ip->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (ip->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 161 {
 162         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 163         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 164
 165         if (copy_to_user(arg, &flags, sizeof(flags)))
 166                 return -EFAULT;
 167         return 0;
 168 }
 169
 170 static int check_flags(unsigned int flags)
 171 {
 172         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 173                       FS_NOATIME_FL | FS_NODUMP_FL | \
 174                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 175                       FS_NOCOMP_FL | FS_COMPR_FL |
 176                       FS_NOCOW_FL))
 177                 return -EOPNOTSUPP;
 178
 179         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 180                 return -EINVAL;
 181
 182         return 0;
 183 }
 184
 185 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 186 {
 187         struct inode *inode = file_inode(file);
 188         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 189         struct btrfs_inode *ip = BTRFS_I(inode);
 190         struct btrfs_root *root = ip->root;
 191         struct btrfs_trans_handle *trans;
 192         unsigned int flags, oldflags;
 193         int ret;
 194         u64 ip_oldflags;
 195         unsigned int i_oldflags;
 196         umode_t mode;
 197
 198         if (!inode_owner_or_capable(inode))
 199                 return -EPERM;
 200
 201         if (btrfs_root_readonly(root))
 202                 return -EROFS;
 203
 204         if (copy_from_user(&flags, arg, sizeof(flags)))
 205                 return -EFAULT;
 206
 207         ret = check_flags(flags);
 208         if (ret)
 209                 return ret;
 210
 211         ret = mnt_want_write_file(file);
 212         if (ret)
 213                 return ret;
 214
 215         inode_lock(inode);
 216
 217         ip_oldflags = ip->flags;
 218         i_oldflags = inode->i_flags;
 219         mode = inode->i_mode;
 220
 221         flags = btrfs_mask_flags(inode->i_mode, flags);
 222         oldflags = btrfs_flags_to_ioctl(ip->flags);
 223         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 224                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 225                         ret = -EPERM;
 226                         goto out_unlock;
 227                 }
 228         }
 229
 230         if (flags & FS_SYNC_FL)
 231                 ip->flags |= BTRFS_INODE_SYNC;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_SYNC;
 234         if (flags & FS_IMMUTABLE_FL)
 235                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 238         if (flags & FS_APPEND_FL)
 239                 ip->flags |= BTRFS_INODE_APPEND;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_APPEND;
 242         if (flags & FS_NODUMP_FL)
 243                 ip->flags |= BTRFS_INODE_NODUMP;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_NODUMP;
 246         if (flags & FS_NOATIME_FL)
 247                 ip->flags |= BTRFS_INODE_NOATIME;
 248         else
 249                 ip->flags &= ~BTRFS_INODE_NOATIME;
 250         if (flags & FS_DIRSYNC_FL)
 251                 ip->flags |= BTRFS_INODE_DIRSYNC;
 252         else
 253                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 254         if (flags & FS_NOCOW_FL) {
 255                 if (S_ISREG(mode)) {
 256                         /*
 257                          * It's safe to turn csums off here, no extents exist.
 258                          * Otherwise we want the flag to reflect the real COW
 259                          * status of the file and will not set it.
 260                          */
 261                         if (inode->i_size == 0)
 262                                 ip->flags |= BTRFS_INODE_NODATACOW
 263                                            | BTRFS_INODE_NODATASUM;
 264                 } else {
 265                         ip->flags |= BTRFS_INODE_NODATACOW;
 266                 }
 267         } else {
 268                 /*
 269                  * Revert back under same assumptions as above
 270                  */
 271                 if (S_ISREG(mode)) {
 272                         if (inode->i_size == 0)
 273                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 274                                              | BTRFS_INODE_NODATASUM);
 275                 } else {
 276                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 277                 }
 278         }
 279
 280         /*
 281          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 282          * flag may be changed automatically if compression code won't make
 283          * things smaller.
 284          */
 285         if (flags & FS_NOCOMP_FL) {
 286                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 287                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 288
 289                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 290                 if (ret && ret != -ENODATA)
 291                         goto out_drop;
 292         } else if (flags & FS_COMPR_FL) {
 293                 const char *comp;
 294
 295                 ip->flags |= BTRFS_INODE_COMPRESS;
 296                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 297
 298                 comp = btrfs_compress_type2str(fs_info->compress_type);
 299                 if (!comp || comp[0] == 0)
 300                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 301
 302                 ret = btrfs_set_prop(inode, "btrfs.compression",
 303                                      comp, strlen(comp), 0);
 304                 if (ret)
 305                         goto out_drop;
 306
 307         } else {
 308                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 309                 if (ret && ret != -ENODATA)
 310                         goto out_drop;
 311                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 312         }
 313
 314         trans = btrfs_start_transaction(root, 1);
 315         if (IS_ERR(trans)) {
 316                 ret = PTR_ERR(trans);
 317                 goto out_drop;
 318         }
 319
 320         btrfs_update_iflags(inode);
 321         inode_inc_iversion(inode);
 322         inode->i_ctime = current_time(inode);
 323         ret = btrfs_update_inode(trans, root, inode);
 324
 325         btrfs_end_transaction(trans);
 326  out_drop:
 327         if (ret) {
 328                 ip->flags = ip_oldflags;
 329                 inode->i_flags = i_oldflags;
 330         }
 331
 332  out_unlock:
 333         inode_unlock(inode);
 334         mnt_drop_write_file(file);
 335         return ret;
 336 }
 337
 338 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 339 {
 340         struct inode *inode = file_inode(file);
 341
 342         return put_user(inode->i_generation, arg);
 343 }
 344
 345 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 346 {
 347         struct inode *inode = file_inode(file);
 348         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 349         struct btrfs_device *device;
 350         struct request_queue *q;
 351         struct fstrim_range range;
 352         u64 minlen = ULLONG_MAX;
 353         u64 num_devices = 0;
 354         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 355         int ret;
 356
 357         if (!capable(CAP_SYS_ADMIN))
 358                 return -EPERM;
 359
 360         rcu_read_lock();
 361         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 362                                 dev_list) {
 363                 if (!device->bdev)
 364                         continue;
 365                 q = bdev_get_queue(device->bdev);
 366                 if (blk_queue_discard(q)) {
 367                         num_devices++;
 368                         minlen = min_t(u64, q->limits.discard_granularity,
 369                                      minlen);
 370                 }
 371         }
 372         rcu_read_unlock();
 373
 374         if (!num_devices)
 375                 return -EOPNOTSUPP;
 376         if (copy_from_user(&range, arg, sizeof(range)))
 377                 return -EFAULT;
 378         if (range.start > total_bytes ||
 379             range.len < fs_info->sb->s_blocksize)
 380                 return -EINVAL;
 381
 382         range.len = min(range.len, total_bytes - range.start);
 383         range.minlen = max(range.minlen, minlen);
 384         ret = btrfs_trim_fs(fs_info, &range);
 385         if (ret < 0)
 386                 return ret;
 387
 388         if (copy_to_user(arg, &range, sizeof(range)))
 389                 return -EFAULT;
 390
 391         return 0;
 392 }
 393
 394 int btrfs_is_empty_uuid(u8 *uuid)
 395 {
 396         int i;
 397
 398         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 399                 if (uuid[i])
 400                         return 0;
 401         }
 402         return 1;
 403 }
 404
 405 static noinline int create_subvol(struct inode *dir,
 406                                   struct dentry *dentry,
 407                                   const char *name, int namelen,
 408                                   u64 *async_transid,
 409                                   struct btrfs_qgroup_inherit *inherit)
 410 {
 411         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 412         struct btrfs_trans_handle *trans;
 413         struct btrfs_key key;
 414         struct btrfs_root_item *root_item;
 415         struct btrfs_inode_item *inode_item;
 416         struct extent_buffer *leaf;
 417         struct btrfs_root *root = BTRFS_I(dir)->root;
 418         struct btrfs_root *new_root;
 419         struct btrfs_block_rsv block_rsv;
 420         struct timespec cur_time = current_time(dir);
 421         struct inode *inode;
 422         int ret;
 423         int err;
 424         u64 objectid;
 425         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 426         u64 index = 0;
 427         u64 qgroup_reserved;
 428         uuid_le new_uuid;
 429
 430         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 431         if (!root_item)
 432                 return -ENOMEM;
 433
 434         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 435         if (ret)
 436                 goto fail_free;
 437
 438         /*
 439          * Don't create subvolume whose level is not zero. Or qgroup will be
 440          * screwed up since it assumes subvolume qgroup's level to be 0.
 441          */
 442         if (btrfs_qgroup_level(objectid)) {
 443                 ret = -ENOSPC;
 444                 goto fail_free;
 445         }
 446
 447         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 448         /*
 449          * The same as the snapshot creation, please see the comment
 450          * of create_snapshot().
 451          */
 452         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 453                                                8, &qgroup_reserved, false);
 454         if (ret)
 455                 goto fail_free;
 456
 457         trans = btrfs_start_transaction(root, 0);
 458         if (IS_ERR(trans)) {
 459                 ret = PTR_ERR(trans);
 460                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 461                 goto fail_free;
 462         }
 463         trans->block_rsv = &block_rsv;
 464         trans->bytes_reserved = block_rsv.size;
 465
 466         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 467         if (ret)
 468                 goto fail;
 469
 470         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 471         if (IS_ERR(leaf)) {
 472                 ret = PTR_ERR(leaf);
 473                 goto fail;
 474         }
 475
 476         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 477         btrfs_set_header_bytenr(leaf, leaf->start);
 478         btrfs_set_header_generation(leaf, trans->transid);
 479         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 480         btrfs_set_header_owner(leaf, objectid);
 481
 482         write_extent_buffer_fsid(leaf, fs_info->fsid);
 483         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 484         btrfs_mark_buffer_dirty(leaf);
 485
 486         inode_item = &root_item->inode;
 487         btrfs_set_stack_inode_generation(inode_item, 1);
 488         btrfs_set_stack_inode_size(inode_item, 3);
 489         btrfs_set_stack_inode_nlink(inode_item, 1);
 490         btrfs_set_stack_inode_nbytes(inode_item,
 491                                      fs_info->nodesize);
 492         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 493
 494         btrfs_set_root_flags(root_item, 0);
 495         btrfs_set_root_limit(root_item, 0);
 496         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 497
 498         btrfs_set_root_bytenr(root_item, leaf->start);
 499         btrfs_set_root_generation(root_item, trans->transid);
 500         btrfs_set_root_level(root_item, 0);
 501         btrfs_set_root_refs(root_item, 1);
 502         btrfs_set_root_used(root_item, leaf->len);
 503         btrfs_set_root_last_snapshot(root_item, 0);
 504
 505         btrfs_set_root_generation_v2(root_item,
 506                         btrfs_root_generation(root_item));
 507         uuid_le_gen(&new_uuid);
 508         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 509         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 510         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 511         root_item->ctime = root_item->otime;
 512         btrfs_set_root_ctransid(root_item, trans->transid);
 513         btrfs_set_root_otransid(root_item, trans->transid);
 514
 515         btrfs_tree_unlock(leaf);
 516         free_extent_buffer(leaf);
 517         leaf = NULL;
 518
 519         btrfs_set_root_dirid(root_item, new_dirid);
 520
 521         key.objectid = objectid;
 522         key.offset = 0;
 523         key.type = BTRFS_ROOT_ITEM_KEY;
 524         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 525                                 root_item);
 526         if (ret)
 527                 goto fail;
 528
 529         key.offset = (u64)-1;
 530         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 531         if (IS_ERR(new_root)) {
 532                 ret = PTR_ERR(new_root);
 533                 btrfs_abort_transaction(trans, ret);
 534                 goto fail;
 535         }
 536
 537         btrfs_record_root_in_trans(trans, new_root);
 538
 539         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 540         if (ret) {
 541                 /* We potentially lose an unused inode item here */
 542                 btrfs_abort_transaction(trans, ret);
 543                 goto fail;
 544         }
 545
 546         mutex_lock(&new_root->objectid_mutex);
 547         new_root->highest_objectid = new_dirid;
 548         mutex_unlock(&new_root->objectid_mutex);
 549
 550         /*
 551          * insert the directory item
 552          */
 553         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 554         if (ret) {
 555                 btrfs_abort_transaction(trans, ret);
 556                 goto fail;
 557         }
 558
 559         ret = btrfs_insert_dir_item(trans, root,
 560                                     name, namelen, BTRFS_I(dir), &key,
 561                                     BTRFS_FT_DIR, index);
 562         if (ret) {
 563                 btrfs_abort_transaction(trans, ret);
 564                 goto fail;
 565         }
 566
 567         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 568         ret = btrfs_update_inode(trans, root, dir);
 569         BUG_ON(ret);
 570
 571         ret = btrfs_add_root_ref(trans, fs_info,
 572                                  objectid, root->root_key.objectid,
 573                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 574         BUG_ON(ret);
 575
 576         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 577                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 578         if (ret)
 579                 btrfs_abort_transaction(trans, ret);
 580
 581 fail:
 582         kfree(root_item);
 583         trans->block_rsv = NULL;
 584         trans->bytes_reserved = 0;
 585         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 586
 587         if (async_transid) {
 588                 *async_transid = trans->transid;
 589                 err = btrfs_commit_transaction_async(trans, 1);
 590                 if (err)
 591                         err = btrfs_commit_transaction(trans);
 592         } else {
 593                 err = btrfs_commit_transaction(trans);
 594         }
 595         if (err && !ret)
 596                 ret = err;
 597
 598         if (!ret) {
 599                 inode = btrfs_lookup_dentry(dir, dentry);
 600                 if (IS_ERR(inode))
 601                         return PTR_ERR(inode);
 602                 d_instantiate(dentry, inode);
 603         }
 604         return ret;
 605
 606 fail_free:
 607         kfree(root_item);
 608         return ret;
 609 }
 610
 611 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 612                            struct dentry *dentry,
 613                            u64 *async_transid, bool readonly,
 614                            struct btrfs_qgroup_inherit *inherit)
 615 {
 616         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 617         struct inode *inode;
 618         struct btrfs_pending_snapshot *pending_snapshot;
 619         struct btrfs_trans_handle *trans;
 620         int ret;
 621
 622         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 623                 return -EINVAL;
 624
 625         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 626         if (!pending_snapshot)
 627                 return -ENOMEM;
 628
 629         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 630                         GFP_KERNEL);
 631         pending_snapshot->path = btrfs_alloc_path();
 632         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 633                 ret = -ENOMEM;
 634                 goto free_pending;
 635         }
 636
 637         atomic_inc(&root->will_be_snapshotted);
 638         smp_mb__after_atomic();
 639         /* wait for no snapshot writes */
 640         wait_event(root->subv_writers->wait,
 641                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 642
 643         ret = btrfs_start_delalloc_inodes(root, 0);
 644         if (ret)
 645                 goto dec_and_free;
 646
 647         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 648
 649         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 650                              BTRFS_BLOCK_RSV_TEMP);
 651         /*
 652          * 1 - parent dir inode
 653          * 2 - dir entries
 654          * 1 - root item
 655          * 2 - root ref/backref
 656          * 1 - root of snapshot
 657          * 1 - UUID item
 658          */
 659         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 660                                         &pending_snapshot->block_rsv, 8,
 661                                         &pending_snapshot->qgroup_reserved,
 662                                         false);
 663         if (ret)
 664                 goto dec_and_free;
 665
 666         pending_snapshot->dentry = dentry;
 667         pending_snapshot->root = root;
 668         pending_snapshot->readonly = readonly;
 669         pending_snapshot->dir = dir;
 670         pending_snapshot->inherit = inherit;
 671
 672         trans = btrfs_start_transaction(root, 0);
 673         if (IS_ERR(trans)) {
 674                 ret = PTR_ERR(trans);
 675                 goto fail;
 676         }
 677
 678         spin_lock(&fs_info->trans_lock);
 679         list_add(&pending_snapshot->list,
 680                  &trans->transaction->pending_snapshots);
 681         spin_unlock(&fs_info->trans_lock);
 682         if (async_transid) {
 683                 *async_transid = trans->transid;
 684                 ret = btrfs_commit_transaction_async(trans, 1);
 685                 if (ret)
 686                         ret = btrfs_commit_transaction(trans);
 687         } else {
 688                 ret = btrfs_commit_transaction(trans);
 689         }
 690         if (ret)
 691                 goto fail;
 692
 693         ret = pending_snapshot->error;
 694         if (ret)
 695                 goto fail;
 696
 697         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 698         if (ret)
 699                 goto fail;
 700
 701         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 702         if (IS_ERR(inode)) {
 703                 ret = PTR_ERR(inode);
 704                 goto fail;
 705         }
 706
 707         d_instantiate(dentry, inode);
 708         ret = 0;
 709 fail:
 710         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 711 dec_and_free:
 712         if (atomic_dec_and_test(&root->will_be_snapshotted))
 713                 wake_up_var(&root->will_be_snapshotted);
 714 free_pending:
 715         kfree(pending_snapshot->root_item);
 716         btrfs_free_path(pending_snapshot->path);
 717         kfree(pending_snapshot);
 718
 719         return ret;
 720 }
 721
 722 /*  copy of may_delete in fs/namei.c()
 723  *      Check whether we can remove a link victim from directory dir, check
 724  *  whether the type of victim is right.
 725  *  1. We can't do it if dir is read-only (done in permission())
 726  *  2. We should have write and exec permissions on dir
 727  *  3. We can't remove anything from append-only dir
 728  *  4. We can't do anything with immutable dir (done in permission())
 729  *  5. If the sticky bit on dir is set we should either
 730  *      a. be owner of dir, or
 731  *      b. be owner of victim, or
 732  *      c. have CAP_FOWNER capability
 733  *  6. If the victim is append-only or immutable we can't do anything with
 734  *     links pointing to it.
 735  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 736  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 737  *  9. We can't remove a root or mountpoint.
 738  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 739  *     nfs_async_unlink().
 740  */
 741
 742 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 743 {
 744         int error;
 745
 746         if (d_really_is_negative(victim))
 747                 return -ENOENT;
 748
 749         BUG_ON(d_inode(victim->d_parent) != dir);
 750         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 751
 752         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 753         if (error)
 754                 return error;
 755         if (IS_APPEND(dir))
 756                 return -EPERM;
 757         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 758             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 759                 return -EPERM;
 760         if (isdir) {
 761                 if (!d_is_dir(victim))
 762                         return -ENOTDIR;
 763                 if (IS_ROOT(victim))
 764                         return -EBUSY;
 765         } else if (d_is_dir(victim))
 766                 return -EISDIR;
 767         if (IS_DEADDIR(dir))
 768                 return -ENOENT;
 769         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 770                 return -EBUSY;
 771         return 0;
 772 }
 773
 774 /* copy of may_create in fs/namei.c() */
 775 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 776 {
 777         if (d_really_is_positive(child))
 778                 return -EEXIST;
 779         if (IS_DEADDIR(dir))
 780                 return -ENOENT;
 781         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 782 }
 783
 784 /*
 785  * Create a new subvolume below @parent.  This is largely modeled after
 786  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 787  * inside this filesystem so it's quite a bit simpler.
 788  */
 789 static noinline int btrfs_mksubvol(const struct path *parent,
 790                                    const char *name, int namelen,
 791                                    struct btrfs_root *snap_src,
 792                                    u64 *async_transid, bool readonly,
 793                                    struct btrfs_qgroup_inherit *inherit)
 794 {
 795         struct inode *dir = d_inode(parent->dentry);
 796         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 797         struct dentry *dentry;
 798         int error;
 799
 800         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 801         if (error == -EINTR)
 802                 return error;
 803
 804         dentry = lookup_one_len(name, parent->dentry, namelen);
 805         error = PTR_ERR(dentry);
 806         if (IS_ERR(dentry))
 807                 goto out_unlock;
 808
 809         error = btrfs_may_create(dir, dentry);
 810         if (error)
 811                 goto out_dput;
 812
 813         /*
 814          * even if this name doesn't exist, we may get hash collisions.
 815          * check for them now when we can safely fail
 816          */
 817         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 818                                                dir->i_ino, name,
 819                                                namelen);
 820         if (error)
 821                 goto out_dput;
 822
 823         down_read(&fs_info->subvol_sem);
 824
 825         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 826                 goto out_up_read;
 827
 828         if (snap_src) {
 829                 error = create_snapshot(snap_src, dir, dentry,
 830                                         async_transid, readonly, inherit);
 831         } else {
 832                 error = create_subvol(dir, dentry, name, namelen,
 833                                       async_transid, inherit);
 834         }
 835         if (!error)
 836                 fsnotify_mkdir(dir, dentry);
 837 out_up_read:
 838         up_read(&fs_info->subvol_sem);
 839 out_dput:
 840         dput(dentry);
 841 out_unlock:
 842         inode_unlock(dir);
 843         return error;
 844 }
 845
 846 /*
 847  * When we're defragging a range, we don't want to kick it off again
 848  * if it is really just waiting for delalloc to send it down.
 849  * If we find a nice big extent or delalloc range for the bytes in the
 850  * file you want to defrag, we return 0 to let you know to skip this
 851  * part of the file
 852  */
 853 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 854 {
 855         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 856         struct extent_map *em = NULL;
 857         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 858         u64 end;
 859
 860         read_lock(&em_tree->lock);
 861         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 862         read_unlock(&em_tree->lock);
 863
 864         if (em) {
 865                 end = extent_map_end(em);
 866                 free_extent_map(em);
 867                 if (end - offset > thresh)
 868                         return 0;
 869         }
 870         /* if we already have a nice delalloc here, just stop */
 871         thresh /= 2;
 872         end = count_range_bits(io_tree, &offset, offset + thresh,
 873                                thresh, EXTENT_DELALLOC, 1);
 874         if (end >= thresh)
 875                 return 0;
 876         return 1;
 877 }
 878
 879 /*
 880  * helper function to walk through a file and find extents
 881  * newer than a specific transid, and smaller than thresh.
 882  *
 883  * This is used by the defragging code to find new and small
 884  * extents
 885  */
 886 static int find_new_extents(struct btrfs_root *root,
 887                             struct inode *inode, u64 newer_than,
 888                             u64 *off, u32 thresh)
 889 {
 890         struct btrfs_path *path;
 891         struct btrfs_key min_key;
 892         struct extent_buffer *leaf;
 893         struct btrfs_file_extent_item *extent;
 894         int type;
 895         int ret;
 896         u64 ino = btrfs_ino(BTRFS_I(inode));
 897
 898         path = btrfs_alloc_path();
 899         if (!path)
 900                 return -ENOMEM;
 901
 902         min_key.objectid = ino;
 903         min_key.type = BTRFS_EXTENT_DATA_KEY;
 904         min_key.offset = *off;
 905
 906         while (1) {
 907                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 908                 if (ret != 0)
 909                         goto none;
 910 process_slot:
 911                 if (min_key.objectid != ino)
 912                         goto none;
 913                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 914                         goto none;
 915
 916                 leaf = path->nodes[0];
 917                 extent = btrfs_item_ptr(leaf, path->slots[0],
 918                                         struct btrfs_file_extent_item);
 919
 920                 type = btrfs_file_extent_type(leaf, extent);
 921                 if (type == BTRFS_FILE_EXTENT_REG &&
 922                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 923                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 924                         *off = min_key.offset;
 925                         btrfs_free_path(path);
 926                         return 0;
 927                 }
 928
 929                 path->slots[0]++;
 930                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 931                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 932                         goto process_slot;
 933                 }
 934
 935                 if (min_key.offset == (u64)-1)
 936                         goto none;
 937
 938                 min_key.offset++;
 939                 btrfs_release_path(path);
 940         }
 941 none:
 942         btrfs_free_path(path);
 943         return -ENOENT;
 944 }
 945
 946 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 947 {
 948         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 949         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 950         struct extent_map *em;
 951         u64 len = PAGE_SIZE;
 952
 953         /*
 954          * hopefully we have this extent in the tree already, try without
 955          * the full extent lock
 956          */
 957         read_lock(&em_tree->lock);
 958         em = lookup_extent_mapping(em_tree, start, len);
 959         read_unlock(&em_tree->lock);
 960
 961         if (!em) {
 962                 struct extent_state *cached = NULL;
 963                 u64 end = start + len - 1;
 964
 965                 /* get the big lock and read metadata off disk */
 966                 lock_extent_bits(io_tree, start, end, &cached);
 967                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 968                 unlock_extent_cached(io_tree, start, end, &cached);
 969
 970                 if (IS_ERR(em))
 971                         return NULL;
 972         }
 973
 974         return em;
 975 }
 976
 977 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 978 {
 979         struct extent_map *next;
 980         bool ret = true;
 981
 982         /* this is the last extent */
 983         if (em->start + em->len >= i_size_read(inode))
 984                 return false;
 985
 986         next = defrag_lookup_extent(inode, em->start + em->len);
 987         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 988                 ret = false;
 989         else if ((em->block_start + em->block_len == next->block_start) &&
 990                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
 991                 ret = false;
 992
 993         free_extent_map(next);
 994         return ret;
 995 }
 996
 997 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 998                                u64 *last_len, u64 *skip, u64 *defrag_end,
 999                                int compress)
1000 {
1001         struct extent_map *em;
1002         int ret = 1;
1003         bool next_mergeable = true;
1004         bool prev_mergeable = true;
1005
1006         /*
1007          * make sure that once we start defragging an extent, we keep on
1008          * defragging it
1009          */
1010         if (start < *defrag_end)
1011                 return 1;
1012
1013         *skip = 0;
1014
1015         em = defrag_lookup_extent(inode, start);
1016         if (!em)
1017                 return 0;
1018
1019         /* this will cover holes, and inline extents */
1020         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1021                 ret = 0;
1022                 goto out;
1023         }
1024
1025         if (!*defrag_end)
1026                 prev_mergeable = false;
1027
1028         next_mergeable = defrag_check_next_extent(inode, em);
1029         /*
1030          * we hit a real extent, if it is big or the next extent is not a
1031          * real extent, don't bother defragging it
1032          */
1033         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1034             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1035                 ret = 0;
1036 out:
1037         /*
1038          * last_len ends up being a counter of how many bytes we've defragged.
1039          * every time we choose not to defrag an extent, we reset *last_len
1040          * so that the next tiny extent will force a defrag.
1041          *
1042          * The end result of this is that tiny extents before a single big
1043          * extent will force at least part of that big extent to be defragged.
1044          */
1045         if (ret) {
1046                 *defrag_end = extent_map_end(em);
1047         } else {
1048                 *last_len = 0;
1049                 *skip = extent_map_end(em);
1050                 *defrag_end = 0;
1051         }
1052
1053         free_extent_map(em);
1054         return ret;
1055 }
1056
1057 /*
1058  * it doesn't do much good to defrag one or two pages
1059  * at a time.  This pulls in a nice chunk of pages
1060  * to COW and defrag.
1061  *
1062  * It also makes sure the delalloc code has enough
1063  * dirty data to avoid making new small extents as part
1064  * of the defrag
1065  *
1066  * It's a good idea to start RA on this range
1067  * before calling this.
1068  */
1069 static int cluster_pages_for_defrag(struct inode *inode,
1070                                     struct page **pages,
1071                                     unsigned long start_index,
1072                                     unsigned long num_pages)
1073 {
1074         unsigned long file_end;
1075         u64 isize = i_size_read(inode);
1076         u64 page_start;
1077         u64 page_end;
1078         u64 page_cnt;
1079         int ret;
1080         int i;
1081         int i_done;
1082         struct btrfs_ordered_extent *ordered;
1083         struct extent_state *cached_state = NULL;
1084         struct extent_io_tree *tree;
1085         struct extent_changeset *data_reserved = NULL;
1086         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1087
1088         file_end = (isize - 1) >> PAGE_SHIFT;
1089         if (!isize || start_index > file_end)
1090                 return 0;
1091
1092         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1093
1094         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1095                         start_index << PAGE_SHIFT,
1096                         page_cnt << PAGE_SHIFT);
1097         if (ret)
1098                 return ret;
1099         i_done = 0;
1100         tree = &BTRFS_I(inode)->io_tree;
1101
1102         /* step one, lock all the pages */
1103         for (i = 0; i < page_cnt; i++) {
1104                 struct page *page;
1105 again:
1106                 page = find_or_create_page(inode->i_mapping,
1107                                            start_index + i, mask);
1108                 if (!page)
1109                         break;
1110
1111                 page_start = page_offset(page);
1112                 page_end = page_start + PAGE_SIZE - 1;
1113                 while (1) {
1114                         lock_extent_bits(tree, page_start, page_end,
1115                                          &cached_state);
1116                         ordered = btrfs_lookup_ordered_extent(inode,
1117                                                               page_start);
1118                         unlock_extent_cached(tree, page_start, page_end,
1119                                              &cached_state);
1120                         if (!ordered)
1121                                 break;
1122
1123                         unlock_page(page);
1124                         btrfs_start_ordered_extent(inode, ordered, 1);
1125                         btrfs_put_ordered_extent(ordered);
1126                         lock_page(page);
1127                         /*
1128                          * we unlocked the page above, so we need check if
1129                          * it was released or not.
1130                          */
1131                         if (page->mapping != inode->i_mapping) {
1132                                 unlock_page(page);
1133                                 put_page(page);
1134                                 goto again;
1135                         }
1136                 }
1137
1138                 if (!PageUptodate(page)) {
1139                         btrfs_readpage(NULL, page);
1140                         lock_page(page);
1141                         if (!PageUptodate(page)) {
1142                                 unlock_page(page);
1143                                 put_page(page);
1144                                 ret = -EIO;
1145                                 break;
1146                         }
1147                 }
1148
1149                 if (page->mapping != inode->i_mapping) {
1150                         unlock_page(page);
1151                         put_page(page);
1152                         goto again;
1153                 }
1154
1155                 pages[i] = page;
1156                 i_done++;
1157         }
1158         if (!i_done || ret)
1159                 goto out;
1160
1161         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1162                 goto out;
1163
1164         /*
1165          * so now we have a nice long stream of locked
1166          * and up to date pages, lets wait on them
1167          */
1168         for (i = 0; i < i_done; i++)
1169                 wait_on_page_writeback(pages[i]);
1170
1171         page_start = page_offset(pages[0]);
1172         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1173
1174         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1175                          page_start, page_end - 1, &cached_state);
1176         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1177                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1178                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1179                           &cached_state);
1180
1181         if (i_done != page_cnt) {
1182                 spin_lock(&BTRFS_I(inode)->lock);
1183                 BTRFS_I(inode)->outstanding_extents++;
1184                 spin_unlock(&BTRFS_I(inode)->lock);
1185                 btrfs_delalloc_release_space(inode, data_reserved,
1186                                 start_index << PAGE_SHIFT,
1187                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1188         }
1189
1190
1191         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1192                           &cached_state);
1193
1194         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1195                              page_start, page_end - 1, &cached_state);
1196
1197         for (i = 0; i < i_done; i++) {
1198                 clear_page_dirty_for_io(pages[i]);
1199                 ClearPageChecked(pages[i]);
1200                 set_page_extent_mapped(pages[i]);
1201                 set_page_dirty(pages[i]);
1202                 unlock_page(pages[i]);
1203                 put_page(pages[i]);
1204         }
1205         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1206                                        false);
1207         extent_changeset_free(data_reserved);
1208         return i_done;
1209 out:
1210         for (i = 0; i < i_done; i++) {
1211                 unlock_page(pages[i]);
1212                 put_page(pages[i]);
1213         }
1214         btrfs_delalloc_release_space(inode, data_reserved,
1215                         start_index << PAGE_SHIFT,
1216                         page_cnt << PAGE_SHIFT, true);
1217         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1218                                        true);
1219         extent_changeset_free(data_reserved);
1220         return ret;
1221
1222 }
1223
1224 int btrfs_defrag_file(struct inode *inode, struct file *file,
1225                       struct btrfs_ioctl_defrag_range_args *range,
1226                       u64 newer_than, unsigned long max_to_defrag)
1227 {
1228         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1229         struct btrfs_root *root = BTRFS_I(inode)->root;
1230         struct file_ra_state *ra = NULL;
1231         unsigned long last_index;
1232         u64 isize = i_size_read(inode);
1233         u64 last_len = 0;
1234         u64 skip = 0;
1235         u64 defrag_end = 0;
1236         u64 newer_off = range->start;
1237         unsigned long i;
1238         unsigned long ra_index = 0;
1239         int ret;
1240         int defrag_count = 0;
1241         int compress_type = BTRFS_COMPRESS_ZLIB;
1242         u32 extent_thresh = range->extent_thresh;
1243         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1244         unsigned long cluster = max_cluster;
1245         u64 new_align = ~((u64)SZ_128K - 1);
1246         struct page **pages = NULL;
1247         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1248
1249         if (isize == 0)
1250                 return 0;
1251
1252         if (range->start >= isize)
1253                 return -EINVAL;
1254
1255         if (do_compress) {
1256                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1257                         return -EINVAL;
1258                 if (range->compress_type)
1259                         compress_type = range->compress_type;
1260         }
1261
1262         if (extent_thresh == 0)
1263                 extent_thresh = SZ_256K;
1264
1265         /*
1266          * If we were not given a file, allocate a readahead context. As
1267          * readahead is just an optimization, defrag will work without it so
1268          * we don't error out.
1269          */
1270         if (!file) {
1271                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1272                 if (ra)
1273                         file_ra_state_init(ra, inode->i_mapping);
1274         } else {
1275                 ra = &file->f_ra;
1276         }
1277
1278         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1279         if (!pages) {
1280                 ret = -ENOMEM;
1281                 goto out_ra;
1282         }
1283
1284         /* find the last page to defrag */
1285         if (range->start + range->len > range->start) {
1286                 last_index = min_t(u64, isize - 1,
1287                          range->start + range->len - 1) >> PAGE_SHIFT;
1288         } else {
1289                 last_index = (isize - 1) >> PAGE_SHIFT;
1290         }
1291
1292         if (newer_than) {
1293                 ret = find_new_extents(root, inode, newer_than,
1294                                        &newer_off, SZ_64K);
1295                 if (!ret) {
1296                         range->start = newer_off;
1297                         /*
1298                          * we always align our defrag to help keep
1299                          * the extents in the file evenly spaced
1300                          */
1301                         i = (newer_off & new_align) >> PAGE_SHIFT;
1302                 } else
1303                         goto out_ra;
1304         } else {
1305                 i = range->start >> PAGE_SHIFT;
1306         }
1307         if (!max_to_defrag)
1308                 max_to_defrag = last_index - i + 1;
1309
1310         /*
1311          * make writeback starts from i, so the defrag range can be
1312          * written sequentially.
1313          */
1314         if (i < inode->i_mapping->writeback_index)
1315                 inode->i_mapping->writeback_index = i;
1316
1317         while (i <= last_index && defrag_count < max_to_defrag &&
1318                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1319                 /*
1320                  * make sure we stop running if someone unmounts
1321                  * the FS
1322                  */
1323                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1324                         break;
1325
1326                 if (btrfs_defrag_cancelled(fs_info)) {
1327                         btrfs_debug(fs_info, "defrag_file cancelled");
1328                         ret = -EAGAIN;
1329                         break;
1330                 }
1331
1332                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1333                                          extent_thresh, &last_len, &skip,
1334                                          &defrag_end, do_compress)){
1335                         unsigned long next;
1336                         /*
1337                          * the should_defrag function tells us how much to skip
1338                          * bump our counter by the suggested amount
1339                          */
1340                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1341                         i = max(i + 1, next);
1342                         continue;
1343                 }
1344
1345                 if (!newer_than) {
1346                         cluster = (PAGE_ALIGN(defrag_end) >>
1347                                    PAGE_SHIFT) - i;
1348                         cluster = min(cluster, max_cluster);
1349                 } else {
1350                         cluster = max_cluster;
1351                 }
1352
1353                 if (i + cluster > ra_index) {
1354                         ra_index = max(i, ra_index);
1355                         if (ra)
1356                                 page_cache_sync_readahead(inode->i_mapping, ra,
1357                                                 file, ra_index, cluster);
1358                         ra_index += cluster;
1359                 }
1360
1361                 inode_lock(inode);
1362                 if (do_compress)
1363                         BTRFS_I(inode)->defrag_compress = compress_type;
1364                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1365                 if (ret < 0) {
1366                         inode_unlock(inode);
1367                         goto out_ra;
1368                 }
1369
1370                 defrag_count += ret;
1371                 balance_dirty_pages_ratelimited(inode->i_mapping);
1372                 inode_unlock(inode);
1373
1374                 if (newer_than) {
1375                         if (newer_off == (u64)-1)
1376                                 break;
1377
1378                         if (ret > 0)
1379                                 i += ret;
1380
1381                         newer_off = max(newer_off + 1,
1382                                         (u64)i << PAGE_SHIFT);
1383
1384                         ret = find_new_extents(root, inode, newer_than,
1385                                                &newer_off, SZ_64K);
1386                         if (!ret) {
1387                                 range->start = newer_off;
1388                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1389                         } else {
1390                                 break;
1391                         }
1392                 } else {
1393                         if (ret > 0) {
1394                                 i += ret;
1395                                 last_len += ret << PAGE_SHIFT;
1396                         } else {
1397                                 i++;
1398                                 last_len = 0;
1399                         }
1400                 }
1401         }
1402
1403         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1404                 filemap_flush(inode->i_mapping);
1405                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1406                              &BTRFS_I(inode)->runtime_flags))
1407                         filemap_flush(inode->i_mapping);
1408         }
1409
1410         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1411                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1412         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1413                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1414         }
1415
1416         ret = defrag_count;
1417
1418 out_ra:
1419         if (do_compress) {
1420                 inode_lock(inode);
1421                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1422                 inode_unlock(inode);
1423         }
1424         if (!file)
1425                 kfree(ra);
1426         kfree(pages);
1427         return ret;
1428 }
1429
1430 static noinline int btrfs_ioctl_resize(struct file *file,
1431                                         void __user *arg)
1432 {
1433         struct inode *inode = file_inode(file);
1434         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1435         u64 new_size;
1436         u64 old_size;
1437         u64 devid = 1;
1438         struct btrfs_root *root = BTRFS_I(inode)->root;
1439         struct btrfs_ioctl_vol_args *vol_args;
1440         struct btrfs_trans_handle *trans;
1441         struct btrfs_device *device = NULL;
1442         char *sizestr;
1443         char *retptr;
1444         char *devstr = NULL;
1445         int ret = 0;
1446         int mod = 0;
1447
1448         if (!capable(CAP_SYS_ADMIN))
1449                 return -EPERM;
1450
1451         ret = mnt_want_write_file(file);
1452         if (ret)
1453                 return ret;
1454
1455         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1456                 mnt_drop_write_file(file);
1457                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1458         }
1459
1460         mutex_lock(&fs_info->volume_mutex);
1461         vol_args = memdup_user(arg, sizeof(*vol_args));
1462         if (IS_ERR(vol_args)) {
1463                 ret = PTR_ERR(vol_args);
1464                 goto out;
1465         }
1466
1467         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1468
1469         sizestr = vol_args->name;
1470         devstr = strchr(sizestr, ':');
1471         if (devstr) {
1472                 sizestr = devstr + 1;
1473                 *devstr = '\0';
1474                 devstr = vol_args->name;
1475                 ret = kstrtoull(devstr, 10, &devid);
1476                 if (ret)
1477                         goto out_free;
1478                 if (!devid) {
1479                         ret = -EINVAL;
1480                         goto out_free;
1481                 }
1482                 btrfs_info(fs_info, "resizing devid %llu", devid);
1483         }
1484
1485         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1486         if (!device) {
1487                 btrfs_info(fs_info, "resizer unable to find device %llu",
1488                            devid);
1489                 ret = -ENODEV;
1490                 goto out_free;
1491         }
1492
1493         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1494                 btrfs_info(fs_info,
1495                            "resizer unable to apply on readonly device %llu",
1496                        devid);
1497                 ret = -EPERM;
1498                 goto out_free;
1499         }
1500
1501         if (!strcmp(sizestr, "max"))
1502                 new_size = device->bdev->bd_inode->i_size;
1503         else {
1504                 if (sizestr[0] == '-') {
1505                         mod = -1;
1506                         sizestr++;
1507                 } else if (sizestr[0] == '+') {
1508                         mod = 1;
1509                         sizestr++;
1510                 }
1511                 new_size = memparse(sizestr, &retptr);
1512                 if (*retptr != '\0' || new_size == 0) {
1513                         ret = -EINVAL;
1514                         goto out_free;
1515                 }
1516         }
1517
1518         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1519                 ret = -EPERM;
1520                 goto out_free;
1521         }
1522
1523         old_size = btrfs_device_get_total_bytes(device);
1524
1525         if (mod < 0) {
1526                 if (new_size > old_size) {
1527                         ret = -EINVAL;
1528                         goto out_free;
1529                 }
1530                 new_size = old_size - new_size;
1531         } else if (mod > 0) {
1532                 if (new_size > ULLONG_MAX - old_size) {
1533                         ret = -ERANGE;
1534                         goto out_free;
1535                 }
1536                 new_size = old_size + new_size;
1537         }
1538
1539         if (new_size < SZ_256M) {
1540                 ret = -EINVAL;
1541                 goto out_free;
1542         }
1543         if (new_size > device->bdev->bd_inode->i_size) {
1544                 ret = -EFBIG;
1545                 goto out_free;
1546         }
1547
1548         new_size = round_down(new_size, fs_info->sectorsize);
1549
1550         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1551                           rcu_str_deref(device->name), new_size);
1552
1553         if (new_size > old_size) {
1554                 trans = btrfs_start_transaction(root, 0);
1555                 if (IS_ERR(trans)) {
1556                         ret = PTR_ERR(trans);
1557                         goto out_free;
1558                 }
1559                 ret = btrfs_grow_device(trans, device, new_size);
1560                 btrfs_commit_transaction(trans);
1561         } else if (new_size < old_size) {
1562                 ret = btrfs_shrink_device(device, new_size);
1563         } /* equal, nothing need to do */
1564
1565 out_free:
1566         kfree(vol_args);
1567 out:
1568         mutex_unlock(&fs_info->volume_mutex);
1569         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1570         mnt_drop_write_file(file);
1571         return ret;
1572 }
1573
1574 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1575                                 const char *name, unsigned long fd, int subvol,
1576                                 u64 *transid, bool readonly,
1577                                 struct btrfs_qgroup_inherit *inherit)
1578 {
1579         int namelen;
1580         int ret = 0;
1581
1582         if (!S_ISDIR(file_inode(file)->i_mode))
1583                 return -ENOTDIR;
1584
1585         ret = mnt_want_write_file(file);
1586         if (ret)
1587                 goto out;
1588
1589         namelen = strlen(name);
1590         if (strchr(name, '/')) {
1591                 ret = -EINVAL;
1592                 goto out_drop_write;
1593         }
1594
1595         if (name[0] == '.' &&
1596            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1597                 ret = -EEXIST;
1598                 goto out_drop_write;
1599         }
1600
1601         if (subvol) {
1602                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1603                                      NULL, transid, readonly, inherit);
1604         } else {
1605                 struct fd src = fdget(fd);
1606                 struct inode *src_inode;
1607                 if (!src.file) {
1608                         ret = -EINVAL;
1609                         goto out_drop_write;
1610                 }
1611
1612                 src_inode = file_inode(src.file);
1613                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1614                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1615                                    "Snapshot src from another FS");
1616                         ret = -EXDEV;
1617                 } else if (!inode_owner_or_capable(src_inode)) {
1618                         /*
1619                          * Subvolume creation is not restricted, but snapshots
1620                          * are limited to own subvolumes only
1621                          */
1622                         ret = -EPERM;
1623                 } else {
1624                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1625                                              BTRFS_I(src_inode)->root,
1626                                              transid, readonly, inherit);
1627                 }
1628                 fdput(src);
1629         }
1630 out_drop_write:
1631         mnt_drop_write_file(file);
1632 out:
1633         return ret;
1634 }
1635
1636 static noinline int btrfs_ioctl_snap_create(struct file *file,
1637                                             void __user *arg, int subvol)
1638 {
1639         struct btrfs_ioctl_vol_args *vol_args;
1640         int ret;
1641
1642         if (!S_ISDIR(file_inode(file)->i_mode))
1643                 return -ENOTDIR;
1644
1645         vol_args = memdup_user(arg, sizeof(*vol_args));
1646         if (IS_ERR(vol_args))
1647                 return PTR_ERR(vol_args);
1648         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1649
1650         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1651                                               vol_args->fd, subvol,
1652                                               NULL, false, NULL);
1653
1654         kfree(vol_args);
1655         return ret;
1656 }
1657
1658 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1659                                                void __user *arg, int subvol)
1660 {
1661         struct btrfs_ioctl_vol_args_v2 *vol_args;
1662         int ret;
1663         u64 transid = 0;
1664         u64 *ptr = NULL;
1665         bool readonly = false;
1666         struct btrfs_qgroup_inherit *inherit = NULL;
1667
1668         if (!S_ISDIR(file_inode(file)->i_mode))
1669                 return -ENOTDIR;
1670
1671         vol_args = memdup_user(arg, sizeof(*vol_args));
1672         if (IS_ERR(vol_args))
1673                 return PTR_ERR(vol_args);
1674         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1675
1676         if (vol_args->flags &
1677             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1678               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1679                 ret = -EOPNOTSUPP;
1680                 goto free_args;
1681         }
1682
1683         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1684                 ptr = &transid;
1685         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1686                 readonly = true;
1687         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1688                 if (vol_args->size > PAGE_SIZE) {
1689                         ret = -EINVAL;
1690                         goto free_args;
1691                 }
1692                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1693                 if (IS_ERR(inherit)) {
1694                         ret = PTR_ERR(inherit);
1695                         goto free_args;
1696                 }
1697         }
1698
1699         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1700                                               vol_args->fd, subvol, ptr,
1701                                               readonly, inherit);
1702         if (ret)
1703                 goto free_inherit;
1704
1705         if (ptr && copy_to_user(arg +
1706                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1707                                         transid),
1708                                 ptr, sizeof(*ptr)))
1709                 ret = -EFAULT;
1710
1711 free_inherit:
1712         kfree(inherit);
1713 free_args:
1714         kfree(vol_args);
1715         return ret;
1716 }
1717
1718 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1719                                                 void __user *arg)
1720 {
1721         struct inode *inode = file_inode(file);
1722         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1723         struct btrfs_root *root = BTRFS_I(inode)->root;
1724         int ret = 0;
1725         u64 flags = 0;
1726
1727         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1728                 return -EINVAL;
1729
1730         down_read(&fs_info->subvol_sem);
1731         if (btrfs_root_readonly(root))
1732                 flags |= BTRFS_SUBVOL_RDONLY;
1733         up_read(&fs_info->subvol_sem);
1734
1735         if (copy_to_user(arg, &flags, sizeof(flags)))
1736                 ret = -EFAULT;
1737
1738         return ret;
1739 }
1740
1741 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1742                                               void __user *arg)
1743 {
1744         struct inode *inode = file_inode(file);
1745         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1746         struct btrfs_root *root = BTRFS_I(inode)->root;
1747         struct btrfs_trans_handle *trans;
1748         u64 root_flags;
1749         u64 flags;
1750         int ret = 0;
1751
1752         if (!inode_owner_or_capable(inode))
1753                 return -EPERM;
1754
1755         ret = mnt_want_write_file(file);
1756         if (ret)
1757                 goto out;
1758
1759         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1760                 ret = -EINVAL;
1761                 goto out_drop_write;
1762         }
1763
1764         if (copy_from_user(&flags, arg, sizeof(flags))) {
1765                 ret = -EFAULT;
1766                 goto out_drop_write;
1767         }
1768
1769         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1770                 ret = -EINVAL;
1771                 goto out_drop_write;
1772         }
1773
1774         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1775                 ret = -EOPNOTSUPP;
1776                 goto out_drop_write;
1777         }
1778
1779         down_write(&fs_info->subvol_sem);
1780
1781         /* nothing to do */
1782         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1783                 goto out_drop_sem;
1784
1785         root_flags = btrfs_root_flags(&root->root_item);
1786         if (flags & BTRFS_SUBVOL_RDONLY) {
1787                 btrfs_set_root_flags(&root->root_item,
1788                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1789         } else {
1790                 /*
1791                  * Block RO -> RW transition if this subvolume is involved in
1792                  * send
1793                  */
1794                 spin_lock(&root->root_item_lock);
1795                 if (root->send_in_progress == 0) {
1796                         btrfs_set_root_flags(&root->root_item,
1797                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1798                         spin_unlock(&root->root_item_lock);
1799                 } else {
1800                         spin_unlock(&root->root_item_lock);
1801                         btrfs_warn(fs_info,
1802                                    "Attempt to set subvolume %llu read-write during send",
1803                                    root->root_key.objectid);
1804                         ret = -EPERM;
1805                         goto out_drop_sem;
1806                 }
1807         }
1808
1809         trans = btrfs_start_transaction(root, 1);
1810         if (IS_ERR(trans)) {
1811                 ret = PTR_ERR(trans);
1812                 goto out_reset;
1813         }
1814
1815         ret = btrfs_update_root(trans, fs_info->tree_root,
1816                                 &root->root_key, &root->root_item);
1817         if (ret < 0) {
1818                 btrfs_end_transaction(trans);
1819                 goto out_reset;
1820         }
1821
1822         ret = btrfs_commit_transaction(trans);
1823
1824 out_reset:
1825         if (ret)
1826                 btrfs_set_root_flags(&root->root_item, root_flags);
1827 out_drop_sem:
1828         up_write(&fs_info->subvol_sem);
1829 out_drop_write:
1830         mnt_drop_write_file(file);
1831 out:
1832         return ret;
1833 }
1834
1835 /*
1836  * helper to check if the subvolume references other subvolumes
1837  */
1838 static noinline int may_destroy_subvol(struct btrfs_root *root)
1839 {
1840         struct btrfs_fs_info *fs_info = root->fs_info;
1841         struct btrfs_path *path;
1842         struct btrfs_dir_item *di;
1843         struct btrfs_key key;
1844         u64 dir_id;
1845         int ret;
1846
1847         path = btrfs_alloc_path();
1848         if (!path)
1849                 return -ENOMEM;
1850
1851         /* Make sure this root isn't set as the default subvol */
1852         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1853         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1854                                    dir_id, "default", 7, 0);
1855         if (di && !IS_ERR(di)) {
1856                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1857                 if (key.objectid == root->root_key.objectid) {
1858                         ret = -EPERM;
1859                         btrfs_err(fs_info,
1860                                   "deleting default subvolume %llu is not allowed",
1861                                   key.objectid);
1862                         goto out;
1863                 }
1864                 btrfs_release_path(path);
1865         }
1866
1867         key.objectid = root->root_key.objectid;
1868         key.type = BTRFS_ROOT_REF_KEY;
1869         key.offset = (u64)-1;
1870
1871         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1872         if (ret < 0)
1873                 goto out;
1874         BUG_ON(ret == 0);
1875
1876         ret = 0;
1877         if (path->slots[0] > 0) {
1878                 path->slots[0]--;
1879                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1880                 if (key.objectid == root->root_key.objectid &&
1881                     key.type == BTRFS_ROOT_REF_KEY)
1882                         ret = -ENOTEMPTY;
1883         }
1884 out:
1885         btrfs_free_path(path);
1886         return ret;
1887 }
1888
1889 static noinline int key_in_sk(struct btrfs_key *key,
1890                               struct btrfs_ioctl_search_key *sk)
1891 {
1892         struct btrfs_key test;
1893         int ret;
1894
1895         test.objectid = sk->min_objectid;
1896         test.type = sk->min_type;
1897         test.offset = sk->min_offset;
1898
1899         ret = btrfs_comp_cpu_keys(key, &test);
1900         if (ret < 0)
1901                 return 0;
1902
1903         test.objectid = sk->max_objectid;
1904         test.type = sk->max_type;
1905         test.offset = sk->max_offset;
1906
1907         ret = btrfs_comp_cpu_keys(key, &test);
1908         if (ret > 0)
1909                 return 0;
1910         return 1;
1911 }
1912
1913 static noinline int copy_to_sk(struct btrfs_path *path,
1914                                struct btrfs_key *key,
1915                                struct btrfs_ioctl_search_key *sk,
1916                                size_t *buf_size,
1917                                char __user *ubuf,
1918                                unsigned long *sk_offset,
1919                                int *num_found)
1920 {
1921         u64 found_transid;
1922         struct extent_buffer *leaf;
1923         struct btrfs_ioctl_search_header sh;
1924         struct btrfs_key test;
1925         unsigned long item_off;
1926         unsigned long item_len;
1927         int nritems;
1928         int i;
1929         int slot;
1930         int ret = 0;
1931
1932         leaf = path->nodes[0];
1933         slot = path->slots[0];
1934         nritems = btrfs_header_nritems(leaf);
1935
1936         if (btrfs_header_generation(leaf) > sk->max_transid) {
1937                 i = nritems;
1938                 goto advance_key;
1939         }
1940         found_transid = btrfs_header_generation(leaf);
1941
1942         for (i = slot; i < nritems; i++) {
1943                 item_off = btrfs_item_ptr_offset(leaf, i);
1944                 item_len = btrfs_item_size_nr(leaf, i);
1945
1946                 btrfs_item_key_to_cpu(leaf, key, i);
1947                 if (!key_in_sk(key, sk))
1948                         continue;
1949
1950                 if (sizeof(sh) + item_len > *buf_size) {
1951                         if (*num_found) {
1952                                 ret = 1;
1953                                 goto out;
1954                         }
1955
1956                         /*
1957                          * return one empty item back for v1, which does not
1958                          * handle -EOVERFLOW
1959                          */
1960
1961                         *buf_size = sizeof(sh) + item_len;
1962                         item_len = 0;
1963                         ret = -EOVERFLOW;
1964                 }
1965
1966                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1967                         ret = 1;
1968                         goto out;
1969                 }
1970
1971                 sh.objectid = key->objectid;
1972                 sh.offset = key->offset;
1973                 sh.type = key->type;
1974                 sh.len = item_len;
1975                 sh.transid = found_transid;
1976
1977                 /* copy search result header */
1978                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1979                         ret = -EFAULT;
1980                         goto out;
1981                 }
1982
1983                 *sk_offset += sizeof(sh);
1984
1985                 if (item_len) {
1986                         char __user *up = ubuf + *sk_offset;
1987                         /* copy the item */
1988                         if (read_extent_buffer_to_user(leaf, up,
1989                                                        item_off, item_len)) {
1990                                 ret = -EFAULT;
1991                                 goto out;
1992                         }
1993
1994                         *sk_offset += item_len;
1995                 }
1996                 (*num_found)++;
1997
1998                 if (ret) /* -EOVERFLOW from above */
1999                         goto out;
2000
2001                 if (*num_found >= sk->nr_items) {
2002                         ret = 1;
2003                         goto out;
2004                 }
2005         }
2006 advance_key:
2007         ret = 0;
2008         test.objectid = sk->max_objectid;
2009         test.type = sk->max_type;
2010         test.offset = sk->max_offset;
2011         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2012                 ret = 1;
2013         else if (key->offset < (u64)-1)
2014                 key->offset++;
2015         else if (key->type < (u8)-1) {
2016                 key->offset = 0;
2017                 key->type++;
2018         } else if (key->objectid < (u64)-1) {
2019                 key->offset = 0;
2020                 key->type = 0;
2021                 key->objectid++;
2022         } else
2023                 ret = 1;
2024 out:
2025         /*
2026          *  0: all items from this leaf copied, continue with next
2027          *  1: * more items can be copied, but unused buffer is too small
2028          *     * all items were found
2029          *     Either way, it will stops the loop which iterates to the next
2030          *     leaf
2031          *  -EOVERFLOW: item was to large for buffer
2032          *  -EFAULT: could not copy extent buffer back to userspace
2033          */
2034         return ret;
2035 }
2036
2037 static noinline int search_ioctl(struct inode *inode,
2038                                  struct btrfs_ioctl_search_key *sk,
2039                                  size_t *buf_size,
2040                                  char __user *ubuf)
2041 {
2042         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2043         struct btrfs_root *root;
2044         struct btrfs_key key;
2045         struct btrfs_path *path;
2046         int ret;
2047         int num_found = 0;
2048         unsigned long sk_offset = 0;
2049
2050         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2051                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2052                 return -EOVERFLOW;
2053         }
2054
2055         path = btrfs_alloc_path();
2056         if (!path)
2057                 return -ENOMEM;
2058
2059         if (sk->tree_id == 0) {
2060                 /* search the root of the inode that was passed */
2061                 root = BTRFS_I(inode)->root;
2062         } else {
2063                 key.objectid = sk->tree_id;
2064                 key.type = BTRFS_ROOT_ITEM_KEY;
2065                 key.offset = (u64)-1;
2066                 root = btrfs_read_fs_root_no_name(info, &key);
2067                 if (IS_ERR(root)) {
2068                         btrfs_free_path(path);
2069                         return -ENOENT;
2070                 }
2071         }
2072
2073         key.objectid = sk->min_objectid;
2074         key.type = sk->min_type;
2075         key.offset = sk->min_offset;
2076
2077         while (1) {
2078                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2079                 if (ret != 0) {
2080                         if (ret > 0)
2081                                 ret = 0;
2082                         goto err;
2083                 }
2084                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2085                                  &sk_offset, &num_found);
2086                 btrfs_release_path(path);
2087                 if (ret)
2088                         break;
2089
2090         }
2091         if (ret > 0)
2092                 ret = 0;
2093 err:
2094         sk->nr_items = num_found;
2095         btrfs_free_path(path);
2096         return ret;
2097 }
2098
2099 static noinline int btrfs_ioctl_tree_search(struct file *file,
2100                                            void __user *argp)
2101 {
2102         struct btrfs_ioctl_search_args __user *uargs;
2103         struct btrfs_ioctl_search_key sk;
2104         struct inode *inode;
2105         int ret;
2106         size_t buf_size;
2107
2108         if (!capable(CAP_SYS_ADMIN))
2109                 return -EPERM;
2110
2111         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2112
2113         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2114                 return -EFAULT;
2115
2116         buf_size = sizeof(uargs->buf);
2117
2118         inode = file_inode(file);
2119         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2120
2121         /*
2122          * In the origin implementation an overflow is handled by returning a
2123          * search header with a len of zero, so reset ret.
2124          */
2125         if (ret == -EOVERFLOW)
2126                 ret = 0;
2127
2128         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2129                 ret = -EFAULT;
2130         return ret;
2131 }
2132
2133 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2134                                                void __user *argp)
2135 {
2136         struct btrfs_ioctl_search_args_v2 __user *uarg;
2137         struct btrfs_ioctl_search_args_v2 args;
2138         struct inode *inode;
2139         int ret;
2140         size_t buf_size;
2141         const size_t buf_limit = SZ_16M;
2142
2143         if (!capable(CAP_SYS_ADMIN))
2144                 return -EPERM;
2145
2146         /* copy search header and buffer size */
2147         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2148         if (copy_from_user(&args, uarg, sizeof(args)))
2149                 return -EFAULT;
2150
2151         buf_size = args.buf_size;
2152
2153         /* limit result size to 16MB */
2154         if (buf_size > buf_limit)
2155                 buf_size = buf_limit;
2156
2157         inode = file_inode(file);
2158         ret = search_ioctl(inode, &args.key, &buf_size,
2159                            (char __user *)(&uarg->buf[0]));
2160         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2161                 ret = -EFAULT;
2162         else if (ret == -EOVERFLOW &&
2163                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2164                 ret = -EFAULT;
2165
2166         return ret;
2167 }
2168
2169 /*
2170  * Search INODE_REFs to identify path name of 'dirid' directory
2171  * in a 'tree_id' tree. and sets path name to 'name'.
2172  */
2173 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2174                                 u64 tree_id, u64 dirid, char *name)
2175 {
2176         struct btrfs_root *root;
2177         struct btrfs_key key;
2178         char *ptr;
2179         int ret = -1;
2180         int slot;
2181         int len;
2182         int total_len = 0;
2183         struct btrfs_inode_ref *iref;
2184         struct extent_buffer *l;
2185         struct btrfs_path *path;
2186
2187         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2188                 name[0]='\0';
2189                 return 0;
2190         }
2191
2192         path = btrfs_alloc_path();
2193         if (!path)
2194                 return -ENOMEM;
2195
2196         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2197
2198         key.objectid = tree_id;
2199         key.type = BTRFS_ROOT_ITEM_KEY;
2200         key.offset = (u64)-1;
2201         root = btrfs_read_fs_root_no_name(info, &key);
2202         if (IS_ERR(root)) {
2203                 btrfs_err(info, "could not find root %llu", tree_id);
2204                 ret = -ENOENT;
2205                 goto out;
2206         }
2207
2208         key.objectid = dirid;
2209         key.type = BTRFS_INODE_REF_KEY;
2210         key.offset = (u64)-1;
2211
2212         while (1) {
2213                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2214                 if (ret < 0)
2215                         goto out;
2216                 else if (ret > 0) {
2217                         ret = btrfs_previous_item(root, path, dirid,
2218                                                   BTRFS_INODE_REF_KEY);
2219                         if (ret < 0)
2220                                 goto out;
2221                         else if (ret > 0) {
2222                                 ret = -ENOENT;
2223                                 goto out;
2224                         }
2225                 }
2226
2227                 l = path->nodes[0];
2228                 slot = path->slots[0];
2229                 btrfs_item_key_to_cpu(l, &key, slot);
2230
2231                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2232                 len = btrfs_inode_ref_name_len(l, iref);
2233                 ptr -= len + 1;
2234                 total_len += len + 1;
2235                 if (ptr < name) {
2236                         ret = -ENAMETOOLONG;
2237                         goto out;
2238                 }
2239
2240                 *(ptr + len) = '/';
2241                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2242
2243                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2244                         break;
2245
2246                 btrfs_release_path(path);
2247                 key.objectid = key.offset;
2248                 key.offset = (u64)-1;
2249                 dirid = key.objectid;
2250         }
2251         memmove(name, ptr, total_len);
2252         name[total_len] = '\0';
2253         ret = 0;
2254 out:
2255         btrfs_free_path(path);
2256         return ret;
2257 }
2258
2259 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2260                                            void __user *argp)
2261 {
2262          struct btrfs_ioctl_ino_lookup_args *args;
2263          struct inode *inode;
2264         int ret = 0;
2265
2266         args = memdup_user(argp, sizeof(*args));
2267         if (IS_ERR(args))
2268                 return PTR_ERR(args);
2269
2270         inode = file_inode(file);
2271
2272         /*
2273          * Unprivileged query to obtain the containing subvolume root id. The
2274          * path is reset so it's consistent with btrfs_search_path_in_tree.
2275          */
2276         if (args->treeid == 0)
2277                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2278
2279         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2280                 args->name[0] = 0;
2281                 goto out;
2282         }
2283
2284         if (!capable(CAP_SYS_ADMIN)) {
2285                 ret = -EPERM;
2286                 goto out;
2287         }
2288
2289         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2290                                         args->treeid, args->objectid,
2291                                         args->name);
2292
2293 out:
2294         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2295                 ret = -EFAULT;
2296
2297         kfree(args);
2298         return ret;
2299 }
2300
2301 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2302                                              void __user *arg)
2303 {
2304         struct dentry *parent = file->f_path.dentry;
2305         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2306         struct dentry *dentry;
2307         struct inode *dir = d_inode(parent);
2308         struct inode *inode;
2309         struct btrfs_root *root = BTRFS_I(dir)->root;
2310         struct btrfs_root *dest = NULL;
2311         struct btrfs_ioctl_vol_args *vol_args;
2312         struct btrfs_trans_handle *trans;
2313         struct btrfs_block_rsv block_rsv;
2314         u64 root_flags;
2315         u64 qgroup_reserved;
2316         int namelen;
2317         int ret;
2318         int err = 0;
2319
2320         if (!S_ISDIR(dir->i_mode))
2321                 return -ENOTDIR;
2322
2323         vol_args = memdup_user(arg, sizeof(*vol_args));
2324         if (IS_ERR(vol_args))
2325                 return PTR_ERR(vol_args);
2326
2327         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2328         namelen = strlen(vol_args->name);
2329         if (strchr(vol_args->name, '/') ||
2330             strncmp(vol_args->name, "..", namelen) == 0) {
2331                 err = -EINVAL;
2332                 goto out;
2333         }
2334
2335         err = mnt_want_write_file(file);
2336         if (err)
2337                 goto out;
2338
2339
2340         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2341         if (err == -EINTR)
2342                 goto out_drop_write;
2343         dentry = lookup_one_len(vol_args->name, parent, namelen);
2344         if (IS_ERR(dentry)) {
2345                 err = PTR_ERR(dentry);
2346                 goto out_unlock_dir;
2347         }
2348
2349         if (d_really_is_negative(dentry)) {
2350                 err = -ENOENT;
2351                 goto out_dput;
2352         }
2353
2354         inode = d_inode(dentry);
2355         dest = BTRFS_I(inode)->root;
2356         if (!capable(CAP_SYS_ADMIN)) {
2357                 /*
2358                  * Regular user.  Only allow this with a special mount
2359                  * option, when the user has write+exec access to the
2360                  * subvol root, and when rmdir(2) would have been
2361                  * allowed.
2362                  *
2363                  * Note that this is _not_ check that the subvol is
2364                  * empty or doesn't contain data that we wouldn't
2365                  * otherwise be able to delete.
2366                  *
2367                  * Users who want to delete empty subvols should try
2368                  * rmdir(2).
2369                  */
2370                 err = -EPERM;
2371                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2372                         goto out_dput;
2373
2374                 /*
2375                  * Do not allow deletion if the parent dir is the same
2376                  * as the dir to be deleted.  That means the ioctl
2377                  * must be called on the dentry referencing the root
2378                  * of the subvol, not a random directory contained
2379                  * within it.
2380                  */
2381                 err = -EINVAL;
2382                 if (root == dest)
2383                         goto out_dput;
2384
2385                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2386                 if (err)
2387                         goto out_dput;
2388         }
2389
2390         /* check if subvolume may be deleted by a user */
2391         err = btrfs_may_delete(dir, dentry, 1);
2392         if (err)
2393                 goto out_dput;
2394
2395         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2396                 err = -EINVAL;
2397                 goto out_dput;
2398         }
2399
2400         inode_lock(inode);
2401
2402         /*
2403          * Don't allow to delete a subvolume with send in progress. This is
2404          * inside the i_mutex so the error handling that has to drop the bit
2405          * again is not run concurrently.
2406          */
2407         spin_lock(&dest->root_item_lock);
2408         root_flags = btrfs_root_flags(&dest->root_item);
2409         if (dest->send_in_progress == 0) {
2410                 btrfs_set_root_flags(&dest->root_item,
2411                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2412                 spin_unlock(&dest->root_item_lock);
2413         } else {
2414                 spin_unlock(&dest->root_item_lock);
2415                 btrfs_warn(fs_info,
2416                            "Attempt to delete subvolume %llu during send",
2417                            dest->root_key.objectid);
2418                 err = -EPERM;
2419                 goto out_unlock_inode;
2420         }
2421
2422         down_write(&fs_info->subvol_sem);
2423
2424         err = may_destroy_subvol(dest);
2425         if (err)
2426                 goto out_up_write;
2427
2428         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2429         /*
2430          * One for dir inode, two for dir entries, two for root
2431          * ref/backref.
2432          */
2433         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2434                                                5, &qgroup_reserved, true);
2435         if (err)
2436                 goto out_up_write;
2437
2438         trans = btrfs_start_transaction(root, 0);
2439         if (IS_ERR(trans)) {
2440                 err = PTR_ERR(trans);
2441                 goto out_release;
2442         }
2443         trans->block_rsv = &block_rsv;
2444         trans->bytes_reserved = block_rsv.size;
2445
2446         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2447
2448         ret = btrfs_unlink_subvol(trans, root, dir,
2449                                 dest->root_key.objectid,
2450                                 dentry->d_name.name,
2451                                 dentry->d_name.len);
2452         if (ret) {
2453                 err = ret;
2454                 btrfs_abort_transaction(trans, ret);
2455                 goto out_end_trans;
2456         }
2457
2458         btrfs_record_root_in_trans(trans, dest);
2459
2460         memset(&dest->root_item.drop_progress, 0,
2461                 sizeof(dest->root_item.drop_progress));
2462         dest->root_item.drop_level = 0;
2463         btrfs_set_root_refs(&dest->root_item, 0);
2464
2465         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2466                 ret = btrfs_insert_orphan_item(trans,
2467                                         fs_info->tree_root,
2468                                         dest->root_key.objectid);
2469                 if (ret) {
2470                         btrfs_abort_transaction(trans, ret);
2471                         err = ret;
2472                         goto out_end_trans;
2473                 }
2474         }
2475
2476         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2477                                   BTRFS_UUID_KEY_SUBVOL,
2478                                   dest->root_key.objectid);
2479         if (ret && ret != -ENOENT) {
2480                 btrfs_abort_transaction(trans, ret);
2481                 err = ret;
2482                 goto out_end_trans;
2483         }
2484         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2485                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2486                                           dest->root_item.received_uuid,
2487                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2488                                           dest->root_key.objectid);
2489                 if (ret && ret != -ENOENT) {
2490                         btrfs_abort_transaction(trans, ret);
2491                         err = ret;
2492                         goto out_end_trans;
2493                 }
2494         }
2495
2496 out_end_trans:
2497         trans->block_rsv = NULL;
2498         trans->bytes_reserved = 0;
2499         ret = btrfs_end_transaction(trans);
2500         if (ret && !err)
2501                 err = ret;
2502         inode->i_flags |= S_DEAD;
2503 out_release:
2504         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2505 out_up_write:
2506         up_write(&fs_info->subvol_sem);
2507         if (err) {
2508                 spin_lock(&dest->root_item_lock);
2509                 root_flags = btrfs_root_flags(&dest->root_item);
2510                 btrfs_set_root_flags(&dest->root_item,
2511                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2512                 spin_unlock(&dest->root_item_lock);
2513         }
2514 out_unlock_inode:
2515         inode_unlock(inode);
2516         if (!err) {
2517                 d_invalidate(dentry);
2518                 btrfs_invalidate_inodes(dest);
2519                 d_delete(dentry);
2520                 ASSERT(dest->send_in_progress == 0);
2521
2522                 /* the last ref */
2523                 if (dest->ino_cache_inode) {
2524                         iput(dest->ino_cache_inode);
2525                         dest->ino_cache_inode = NULL;
2526                 }
2527         }
2528 out_dput:
2529         dput(dentry);
2530 out_unlock_dir:
2531         inode_unlock(dir);
2532 out_drop_write:
2533         mnt_drop_write_file(file);
2534 out:
2535         kfree(vol_args);
2536         return err;
2537 }
2538
2539 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2540 {
2541         struct inode *inode = file_inode(file);
2542         struct btrfs_root *root = BTRFS_I(inode)->root;
2543         struct btrfs_ioctl_defrag_range_args *range;
2544         int ret;
2545
2546         ret = mnt_want_write_file(file);
2547         if (ret)
2548                 return ret;
2549
2550         if (btrfs_root_readonly(root)) {
2551                 ret = -EROFS;
2552                 goto out;
2553         }
2554
2555         switch (inode->i_mode & S_IFMT) {
2556         case S_IFDIR:
2557                 if (!capable(CAP_SYS_ADMIN)) {
2558                         ret = -EPERM;
2559                         goto out;
2560                 }
2561                 ret = btrfs_defrag_root(root);
2562                 break;
2563         case S_IFREG:
2564                 if (!(file->f_mode & FMODE_WRITE)) {
2565                         ret = -EINVAL;
2566                         goto out;
2567                 }
2568
2569                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2570                 if (!range) {
2571                         ret = -ENOMEM;
2572                         goto out;
2573                 }
2574
2575                 if (argp) {
2576                         if (copy_from_user(range, argp,
2577                                            sizeof(*range))) {
2578                                 ret = -EFAULT;
2579                                 kfree(range);
2580                                 goto out;
2581                         }
2582                         /* compression requires us to start the IO */
2583                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2584                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2585                                 range->extent_thresh = (u32)-1;
2586                         }
2587                 } else {
2588                         /* the rest are all set to zero by kzalloc */
2589                         range->len = (u64)-1;
2590                 }
2591                 ret = btrfs_defrag_file(file_inode(file), file,
2592                                         range, BTRFS_OLDEST_GENERATION, 0);
2593                 if (ret > 0)
2594                         ret = 0;
2595                 kfree(range);
2596                 break;
2597         default:
2598                 ret = -EINVAL;
2599         }
2600 out:
2601         mnt_drop_write_file(file);
2602         return ret;
2603 }
2604
2605 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2606 {
2607         struct btrfs_ioctl_vol_args *vol_args;
2608         int ret;
2609
2610         if (!capable(CAP_SYS_ADMIN))
2611                 return -EPERM;
2612
2613         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2614                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2615
2616         mutex_lock(&fs_info->volume_mutex);
2617         vol_args = memdup_user(arg, sizeof(*vol_args));
2618         if (IS_ERR(vol_args)) {
2619                 ret = PTR_ERR(vol_args);
2620                 goto out;
2621         }
2622
2623         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2624         ret = btrfs_init_new_device(fs_info, vol_args->name);
2625
2626         if (!ret)
2627                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2628
2629         kfree(vol_args);
2630 out:
2631         mutex_unlock(&fs_info->volume_mutex);
2632         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2633         return ret;
2634 }
2635
2636 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2637 {
2638         struct inode *inode = file_inode(file);
2639         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2640         struct btrfs_ioctl_vol_args_v2 *vol_args;
2641         int ret;
2642
2643         if (!capable(CAP_SYS_ADMIN))
2644                 return -EPERM;
2645
2646         ret = mnt_want_write_file(file);
2647         if (ret)
2648                 return ret;
2649
2650         vol_args = memdup_user(arg, sizeof(*vol_args));
2651         if (IS_ERR(vol_args)) {
2652                 ret = PTR_ERR(vol_args);
2653                 goto err_drop;
2654         }
2655
2656         /* Check for compatibility reject unknown flags */
2657         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2658                 return -EOPNOTSUPP;
2659
2660         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2661                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2662                 goto out;
2663         }
2664
2665         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2666                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2667         } else {
2668                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2669                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2670         }
2671         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2672
2673         if (!ret) {
2674                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2675                         btrfs_info(fs_info, "device deleted: id %llu",
2676                                         vol_args->devid);
2677                 else
2678                         btrfs_info(fs_info, "device deleted: %s",
2679                                         vol_args->name);
2680         }
2681 out:
2682         kfree(vol_args);
2683 err_drop:
2684         mnt_drop_write_file(file);
2685         return ret;
2686 }
2687
2688 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2689 {
2690         struct inode *inode = file_inode(file);
2691         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2692         struct btrfs_ioctl_vol_args *vol_args;
2693         int ret;
2694
2695         if (!capable(CAP_SYS_ADMIN))
2696                 return -EPERM;
2697
2698         ret = mnt_want_write_file(file);
2699         if (ret)
2700                 return ret;
2701
2702         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2703                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2704                 goto out_drop_write;
2705         }
2706
2707         vol_args = memdup_user(arg, sizeof(*vol_args));
2708         if (IS_ERR(vol_args)) {
2709                 ret = PTR_ERR(vol_args);
2710                 goto out;
2711         }
2712
2713         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2714         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2715
2716         if (!ret)
2717                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2718         kfree(vol_args);
2719 out:
2720         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2721 out_drop_write:
2722         mnt_drop_write_file(file);
2723
2724         return ret;
2725 }
2726
2727 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2728                                 void __user *arg)
2729 {
2730         struct btrfs_ioctl_fs_info_args *fi_args;
2731         struct btrfs_device *device;
2732         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2733         int ret = 0;
2734
2735         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2736         if (!fi_args)
2737                 return -ENOMEM;
2738
2739         rcu_read_lock();
2740         fi_args->num_devices = fs_devices->num_devices;
2741
2742         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2743                 if (device->devid > fi_args->max_id)
2744                         fi_args->max_id = device->devid;
2745         }
2746         rcu_read_unlock();
2747
2748         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2749         fi_args->nodesize = fs_info->nodesize;
2750         fi_args->sectorsize = fs_info->sectorsize;
2751         fi_args->clone_alignment = fs_info->sectorsize;
2752
2753         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2754                 ret = -EFAULT;
2755
2756         kfree(fi_args);
2757         return ret;
2758 }
2759
2760 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2761                                  void __user *arg)
2762 {
2763         struct btrfs_ioctl_dev_info_args *di_args;
2764         struct btrfs_device *dev;
2765         int ret = 0;
2766         char *s_uuid = NULL;
2767
2768         di_args = memdup_user(arg, sizeof(*di_args));
2769         if (IS_ERR(di_args))
2770                 return PTR_ERR(di_args);
2771
2772         if (!btrfs_is_empty_uuid(di_args->uuid))
2773                 s_uuid = di_args->uuid;
2774
2775         rcu_read_lock();
2776         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2777
2778         if (!dev) {
2779                 ret = -ENODEV;
2780                 goto out;
2781         }
2782
2783         di_args->devid = dev->devid;
2784         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2785         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2786         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2787         if (dev->name) {
2788                 struct rcu_string *name;
2789
2790                 name = rcu_dereference(dev->name);
2791                 strncpy(di_args->path, name->str, sizeof(di_args->path) - 1);
2792                 di_args->path[sizeof(di_args->path) - 1] = 0;
2793         } else {
2794                 di_args->path[0] = '\0';
2795         }
2796
2797 out:
2798         rcu_read_unlock();
2799         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2800                 ret = -EFAULT;
2801
2802         kfree(di_args);
2803         return ret;
2804 }
2805
2806 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2807 {
2808         struct page *page;
2809
2810         page = grab_cache_page(inode->i_mapping, index);
2811         if (!page)
2812                 return ERR_PTR(-ENOMEM);
2813
2814         if (!PageUptodate(page)) {
2815                 int ret;
2816
2817                 ret = btrfs_readpage(NULL, page);
2818                 if (ret)
2819                         return ERR_PTR(ret);
2820                 lock_page(page);
2821                 if (!PageUptodate(page)) {
2822                         unlock_page(page);
2823                         put_page(page);
2824                         return ERR_PTR(-EIO);
2825                 }
2826                 if (page->mapping != inode->i_mapping) {
2827                         unlock_page(page);
2828                         put_page(page);
2829                         return ERR_PTR(-EAGAIN);
2830                 }
2831         }
2832
2833         return page;
2834 }
2835
2836 static int gather_extent_pages(struct inode *inode, struct page **pages,
2837                                int num_pages, u64 off)
2838 {
2839         int i;
2840         pgoff_t index = off >> PAGE_SHIFT;
2841
2842         for (i = 0; i < num_pages; i++) {
2843 again:
2844                 pages[i] = extent_same_get_page(inode, index + i);
2845                 if (IS_ERR(pages[i])) {
2846                         int err = PTR_ERR(pages[i]);
2847
2848                         if (err == -EAGAIN)
2849                                 goto again;
2850                         pages[i] = NULL;
2851                         return err;
2852                 }
2853         }
2854         return 0;
2855 }
2856
2857 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2858                              bool retry_range_locking)
2859 {
2860         /*
2861          * Do any pending delalloc/csum calculations on inode, one way or
2862          * another, and lock file content.
2863          * The locking order is:
2864          *
2865          *   1) pages
2866          *   2) range in the inode's io tree
2867          */
2868         while (1) {
2869                 struct btrfs_ordered_extent *ordered;
2870                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2871                 ordered = btrfs_lookup_first_ordered_extent(inode,
2872                                                             off + len - 1);
2873                 if ((!ordered ||
2874                      ordered->file_offset + ordered->len <= off ||
2875                      ordered->file_offset >= off + len) &&
2876                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2877                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2878                         if (ordered)
2879                                 btrfs_put_ordered_extent(ordered);
2880                         break;
2881                 }
2882                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2883                 if (ordered)
2884                         btrfs_put_ordered_extent(ordered);
2885                 if (!retry_range_locking)
2886                         return -EAGAIN;
2887                 btrfs_wait_ordered_range(inode, off, len);
2888         }
2889         return 0;
2890 }
2891
2892 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2893 {
2894         inode_unlock(inode1);
2895         inode_unlock(inode2);
2896 }
2897
2898 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2899 {
2900         if (inode1 < inode2)
2901                 swap(inode1, inode2);
2902
2903         inode_lock_nested(inode1, I_MUTEX_PARENT);
2904         inode_lock_nested(inode2, I_MUTEX_CHILD);
2905 }
2906
2907 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2908                                       struct inode *inode2, u64 loff2, u64 len)
2909 {
2910         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2911         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2912 }
2913
2914 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2915                                     struct inode *inode2, u64 loff2, u64 len,
2916                                     bool retry_range_locking)
2917 {
2918         int ret;
2919
2920         if (inode1 < inode2) {
2921                 swap(inode1, inode2);
2922                 swap(loff1, loff2);
2923         }
2924         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2925         if (ret)
2926                 return ret;
2927         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2928         if (ret)
2929                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2930                               loff1 + len - 1);
2931         return ret;
2932 }
2933
2934 struct cmp_pages {
2935         int             num_pages;
2936         struct page     **src_pages;
2937         struct page     **dst_pages;
2938 };
2939
2940 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2941 {
2942         int i;
2943         struct page *pg;
2944
2945         for (i = 0; i < cmp->num_pages; i++) {
2946                 pg = cmp->src_pages[i];
2947                 if (pg) {
2948                         unlock_page(pg);
2949                         put_page(pg);
2950                 }
2951                 pg = cmp->dst_pages[i];
2952                 if (pg) {
2953                         unlock_page(pg);
2954                         put_page(pg);
2955                 }
2956         }
2957         kfree(cmp->src_pages);
2958         kfree(cmp->dst_pages);
2959 }
2960
2961 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2962                                   struct inode *dst, u64 dst_loff,
2963                                   u64 len, struct cmp_pages *cmp)
2964 {
2965         int ret;
2966         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2967         struct page **src_pgarr, **dst_pgarr;
2968
2969         /*
2970          * We must gather up all the pages before we initiate our
2971          * extent locking. We use an array for the page pointers. Size
2972          * of the array is bounded by len, which is in turn bounded by
2973          * BTRFS_MAX_DEDUPE_LEN.
2974          */
2975         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2976         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2977         if (!src_pgarr || !dst_pgarr) {
2978                 kfree(src_pgarr);
2979                 kfree(dst_pgarr);
2980                 return -ENOMEM;
2981         }
2982         cmp->num_pages = num_pages;
2983         cmp->src_pages = src_pgarr;
2984         cmp->dst_pages = dst_pgarr;
2985
2986         /*
2987          * If deduping ranges in the same inode, locking rules make it mandatory
2988          * to always lock pages in ascending order to avoid deadlocks with
2989          * concurrent tasks (such as starting writeback/delalloc).
2990          */
2991         if (src == dst && dst_loff < loff) {
2992                 swap(src_pgarr, dst_pgarr);
2993                 swap(loff, dst_loff);
2994         }
2995
2996         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
2997         if (ret)
2998                 goto out;
2999
3000         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3001
3002 out:
3003         if (ret)
3004                 btrfs_cmp_data_free(cmp);
3005         return ret;
3006 }
3007
3008 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3009 {
3010         int ret = 0;
3011         int i;
3012         struct page *src_page, *dst_page;
3013         unsigned int cmp_len = PAGE_SIZE;
3014         void *addr, *dst_addr;
3015
3016         i = 0;
3017         while (len) {
3018                 if (len < PAGE_SIZE)
3019                         cmp_len = len;
3020
3021                 BUG_ON(i >= cmp->num_pages);
3022
3023                 src_page = cmp->src_pages[i];
3024                 dst_page = cmp->dst_pages[i];
3025                 ASSERT(PageLocked(src_page));
3026                 ASSERT(PageLocked(dst_page));
3027
3028                 addr = kmap_atomic(src_page);
3029                 dst_addr = kmap_atomic(dst_page);
3030
3031                 flush_dcache_page(src_page);
3032                 flush_dcache_page(dst_page);
3033
3034                 if (memcmp(addr, dst_addr, cmp_len))
3035                         ret = -EBADE;
3036
3037                 kunmap_atomic(addr);
3038                 kunmap_atomic(dst_addr);
3039
3040                 if (ret)
3041                         break;
3042
3043                 len -= cmp_len;
3044                 i++;
3045         }
3046
3047         return ret;
3048 }
3049
3050 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3051                                      u64 olen)
3052 {
3053         u64 len = *plen;
3054         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3055
3056         if (off + olen > inode->i_size || off + olen < off)
3057                 return -EINVAL;
3058
3059         /* if we extend to eof, continue to block boundary */
3060         if (off + len == inode->i_size)
3061                 *plen = len = ALIGN(inode->i_size, bs) - off;
3062
3063         /* Check that we are block aligned - btrfs_clone() requires this */
3064         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3065                 return -EINVAL;
3066
3067         return 0;
3068 }
3069
3070 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3071                              struct inode *dst, u64 dst_loff)
3072 {
3073         int ret;
3074         u64 len = olen;
3075         struct cmp_pages cmp;
3076         bool same_inode = (src == dst);
3077         u64 same_lock_start = 0;
3078         u64 same_lock_len = 0;
3079
3080         if (len == 0)
3081                 return 0;
3082
3083         if (same_inode)
3084                 inode_lock(src);
3085         else
3086                 btrfs_double_inode_lock(src, dst);
3087
3088         ret = extent_same_check_offsets(src, loff, &len, olen);
3089         if (ret)
3090                 goto out_unlock;
3091
3092         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3093         if (ret)
3094                 goto out_unlock;
3095
3096         if (same_inode) {
3097                 /*
3098                  * Single inode case wants the same checks, except we
3099                  * don't want our length pushed out past i_size as
3100                  * comparing that data range makes no sense.
3101                  *
3102                  * extent_same_check_offsets() will do this for an
3103                  * unaligned length at i_size, so catch it here and
3104                  * reject the request.
3105                  *
3106                  * This effectively means we require aligned extents
3107                  * for the single-inode case, whereas the other cases
3108                  * allow an unaligned length so long as it ends at
3109                  * i_size.
3110                  */
3111                 if (len != olen) {
3112                         ret = -EINVAL;
3113                         goto out_unlock;
3114                 }
3115
3116                 /* Check for overlapping ranges */
3117                 if (dst_loff + len > loff && dst_loff < loff + len) {
3118                         ret = -EINVAL;
3119                         goto out_unlock;
3120                 }
3121
3122                 same_lock_start = min_t(u64, loff, dst_loff);
3123                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3124         }
3125
3126         /* don't make the dst file partly checksummed */
3127         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3128             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3129                 ret = -EINVAL;
3130                 goto out_unlock;
3131         }
3132
3133 again:
3134         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3135         if (ret)
3136                 goto out_unlock;
3137
3138         if (same_inode)
3139                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3140                                         false);
3141         else
3142                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3143                                                false);
3144         /*
3145          * If one of the inodes has dirty pages in the respective range or
3146          * ordered extents, we need to flush dellaloc and wait for all ordered
3147          * extents in the range. We must unlock the pages and the ranges in the
3148          * io trees to avoid deadlocks when flushing delalloc (requires locking
3149          * pages) and when waiting for ordered extents to complete (they require
3150          * range locking).
3151          */
3152         if (ret == -EAGAIN) {
3153                 /*
3154                  * Ranges in the io trees already unlocked. Now unlock all
3155                  * pages before waiting for all IO to complete.
3156                  */
3157                 btrfs_cmp_data_free(&cmp);
3158                 if (same_inode) {
3159                         btrfs_wait_ordered_range(src, same_lock_start,
3160                                                  same_lock_len);
3161                 } else {
3162                         btrfs_wait_ordered_range(src, loff, len);
3163                         btrfs_wait_ordered_range(dst, dst_loff, len);
3164                 }
3165                 goto again;
3166         }
3167         ASSERT(ret == 0);
3168         if (WARN_ON(ret)) {
3169                 /* ranges in the io trees already unlocked */
3170                 btrfs_cmp_data_free(&cmp);
3171                 return ret;
3172         }
3173
3174         /* pass original length for comparison so we stay within i_size */
3175         ret = btrfs_cmp_data(olen, &cmp);
3176         if (ret == 0)
3177                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3178
3179         if (same_inode)
3180                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3181                               same_lock_start + same_lock_len - 1);
3182         else
3183                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3184
3185         btrfs_cmp_data_free(&cmp);
3186 out_unlock:
3187         if (same_inode)
3188                 inode_unlock(src);
3189         else
3190                 btrfs_double_inode_unlock(src, dst);
3191
3192         return ret;
3193 }
3194
3195 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3196
3197 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3198                                 struct file *dst_file, u64 dst_loff)
3199 {
3200         struct inode *src = file_inode(src_file);
3201         struct inode *dst = file_inode(dst_file);
3202         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3203         ssize_t res;
3204
3205         if (olen > BTRFS_MAX_DEDUPE_LEN)
3206                 olen = BTRFS_MAX_DEDUPE_LEN;
3207
3208         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3209                 /*
3210                  * Btrfs does not support blocksize < page_size. As a
3211                  * result, btrfs_cmp_data() won't correctly handle
3212                  * this situation without an update.
3213                  */
3214                 return -EINVAL;
3215         }
3216
3217         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3218         if (res)
3219                 return res;
3220         return olen;
3221 }
3222
3223 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3224                                      struct inode *inode,
3225                                      u64 endoff,
3226                                      const u64 destoff,
3227                                      const u64 olen,
3228                                      int no_time_update)
3229 {
3230         struct btrfs_root *root = BTRFS_I(inode)->root;
3231         int ret;
3232
3233         inode_inc_iversion(inode);
3234         if (!no_time_update)
3235                 inode->i_mtime = inode->i_ctime = current_time(inode);
3236         /*
3237          * We round up to the block size at eof when determining which
3238          * extents to clone above, but shouldn't round up the file size.
3239          */
3240         if (endoff > destoff + olen)
3241                 endoff = destoff + olen;
3242         if (endoff > inode->i_size)
3243                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3244
3245         ret = btrfs_update_inode(trans, root, inode);
3246         if (ret) {
3247                 btrfs_abort_transaction(trans, ret);
3248                 btrfs_end_transaction(trans);
3249                 goto out;
3250         }
3251         ret = btrfs_end_transaction(trans);
3252 out:
3253         return ret;
3254 }
3255
3256 static void clone_update_extent_map(struct btrfs_inode *inode,
3257                                     const struct btrfs_trans_handle *trans,
3258                                     const struct btrfs_path *path,
3259                                     const u64 hole_offset,
3260                                     const u64 hole_len)
3261 {
3262         struct extent_map_tree *em_tree = &inode->extent_tree;
3263         struct extent_map *em;
3264         int ret;
3265
3266         em = alloc_extent_map();
3267         if (!em) {
3268                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3269                 return;
3270         }
3271
3272         if (path) {
3273                 struct btrfs_file_extent_item *fi;
3274
3275                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3276                                     struct btrfs_file_extent_item);
3277                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3278                 em->generation = -1;
3279                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3280                     BTRFS_FILE_EXTENT_INLINE)
3281                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3282                                         &inode->runtime_flags);
3283         } else {
3284                 em->start = hole_offset;
3285                 em->len = hole_len;
3286                 em->ram_bytes = em->len;
3287                 em->orig_start = hole_offset;
3288                 em->block_start = EXTENT_MAP_HOLE;
3289                 em->block_len = 0;
3290                 em->orig_block_len = 0;
3291                 em->compress_type = BTRFS_COMPRESS_NONE;
3292                 em->generation = trans->transid;
3293         }
3294
3295         while (1) {
3296                 write_lock(&em_tree->lock);
3297                 ret = add_extent_mapping(em_tree, em, 1);
3298                 write_unlock(&em_tree->lock);
3299                 if (ret != -EEXIST) {
3300                         free_extent_map(em);
3301                         break;
3302                 }
3303                 btrfs_drop_extent_cache(inode, em->start,
3304                                         em->start + em->len - 1, 0);
3305         }
3306
3307         if (ret)
3308                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3309 }
3310
3311 /*
3312  * Make sure we do not end up inserting an inline extent into a file that has
3313  * already other (non-inline) extents. If a file has an inline extent it can
3314  * not have any other extents and the (single) inline extent must start at the
3315  * file offset 0. Failing to respect these rules will lead to file corruption,
3316  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3317  *
3318  * We can have extents that have been already written to disk or we can have
3319  * dirty ranges still in delalloc, in which case the extent maps and items are
3320  * created only when we run delalloc, and the delalloc ranges might fall outside
3321  * the range we are currently locking in the inode's io tree. So we check the
3322  * inode's i_size because of that (i_size updates are done while holding the
3323  * i_mutex, which we are holding here).
3324  * We also check to see if the inode has a size not greater than "datal" but has
3325  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3326  * protected against such concurrent fallocate calls by the i_mutex).
3327  *
3328  * If the file has no extents but a size greater than datal, do not allow the
3329  * copy because we would need turn the inline extent into a non-inline one (even
3330  * with NO_HOLES enabled). If we find our destination inode only has one inline
3331  * extent, just overwrite it with the source inline extent if its size is less
3332  * than the source extent's size, or we could copy the source inline extent's
3333  * data into the destination inode's inline extent if the later is greater then
3334  * the former.
3335  */
3336 static int clone_copy_inline_extent(struct inode *dst,
3337                                     struct btrfs_trans_handle *trans,
3338                                     struct btrfs_path *path,
3339                                     struct btrfs_key *new_key,
3340                                     const u64 drop_start,
3341                                     const u64 datal,
3342                                     const u64 skip,
3343                                     const u64 size,
3344                                     char *inline_data)
3345 {
3346         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3347         struct btrfs_root *root = BTRFS_I(dst)->root;
3348         const u64 aligned_end = ALIGN(new_key->offset + datal,
3349                                       fs_info->sectorsize);
3350         int ret;
3351         struct btrfs_key key;
3352
3353         if (new_key->offset > 0)
3354                 return -EOPNOTSUPP;
3355
3356         key.objectid = btrfs_ino(BTRFS_I(dst));
3357         key.type = BTRFS_EXTENT_DATA_KEY;
3358         key.offset = 0;
3359         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3360         if (ret < 0) {
3361                 return ret;
3362         } else if (ret > 0) {
3363                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3364                         ret = btrfs_next_leaf(root, path);
3365                         if (ret < 0)
3366                                 return ret;
3367                         else if (ret > 0)
3368                                 goto copy_inline_extent;
3369                 }
3370                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3371                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3372                     key.type == BTRFS_EXTENT_DATA_KEY) {
3373                         ASSERT(key.offset > 0);
3374                         return -EOPNOTSUPP;
3375                 }
3376         } else if (i_size_read(dst) <= datal) {
3377                 struct btrfs_file_extent_item *ei;
3378                 u64 ext_len;
3379
3380                 /*
3381                  * If the file size is <= datal, make sure there are no other
3382                  * extents following (can happen do to an fallocate call with
3383                  * the flag FALLOC_FL_KEEP_SIZE).
3384                  */
3385                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3386                                     struct btrfs_file_extent_item);
3387                 /*
3388                  * If it's an inline extent, it can not have other extents
3389                  * following it.
3390                  */
3391                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3392                     BTRFS_FILE_EXTENT_INLINE)
3393                         goto copy_inline_extent;
3394
3395                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3396                 if (ext_len > aligned_end)
3397                         return -EOPNOTSUPP;
3398
3399                 ret = btrfs_next_item(root, path);
3400                 if (ret < 0) {
3401                         return ret;
3402                 } else if (ret == 0) {
3403                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3404                                               path->slots[0]);
3405                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3406                             key.type == BTRFS_EXTENT_DATA_KEY)
3407                                 return -EOPNOTSUPP;
3408                 }
3409         }
3410
3411 copy_inline_extent:
3412         /*
3413          * We have no extent items, or we have an extent at offset 0 which may
3414          * or may not be inlined. All these cases are dealt the same way.
3415          */
3416         if (i_size_read(dst) > datal) {
3417                 /*
3418                  * If the destination inode has an inline extent...
3419                  * This would require copying the data from the source inline
3420                  * extent into the beginning of the destination's inline extent.
3421                  * But this is really complex, both extents can be compressed
3422                  * or just one of them, which would require decompressing and
3423                  * re-compressing data (which could increase the new compressed
3424                  * size, not allowing the compressed data to fit anymore in an
3425                  * inline extent).
3426                  * So just don't support this case for now (it should be rare,
3427                  * we are not really saving space when cloning inline extents).
3428                  */
3429                 return -EOPNOTSUPP;
3430         }
3431
3432         btrfs_release_path(path);
3433         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3434         if (ret)
3435                 return ret;
3436         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3437         if (ret)
3438                 return ret;
3439
3440         if (skip) {
3441                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3442
3443                 memmove(inline_data + start, inline_data + start + skip, datal);
3444         }
3445
3446         write_extent_buffer(path->nodes[0], inline_data,
3447                             btrfs_item_ptr_offset(path->nodes[0],
3448                                                   path->slots[0]),
3449                             size);
3450         inode_add_bytes(dst, datal);
3451
3452         return 0;
3453 }
3454
3455 /**
3456  * btrfs_clone() - clone a range from inode file to another
3457  *
3458  * @src: Inode to clone from
3459  * @inode: Inode to clone to
3460  * @off: Offset within source to start clone from
3461  * @olen: Original length, passed by user, of range to clone
3462  * @olen_aligned: Block-aligned value of olen
3463  * @destoff: Offset within @inode to start clone
3464  * @no_time_update: Whether to update mtime/ctime on the target inode
3465  */
3466 static int btrfs_clone(struct inode *src, struct inode *inode,
3467                        const u64 off, const u64 olen, const u64 olen_aligned,
3468                        const u64 destoff, int no_time_update)
3469 {
3470         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3471         struct btrfs_root *root = BTRFS_I(inode)->root;
3472         struct btrfs_path *path = NULL;
3473         struct extent_buffer *leaf;
3474         struct btrfs_trans_handle *trans;
3475         char *buf = NULL;
3476         struct btrfs_key key;
3477         u32 nritems;
3478         int slot;
3479         int ret;
3480         const u64 len = olen_aligned;
3481         u64 last_dest_end = destoff;
3482
3483         ret = -ENOMEM;
3484         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3485         if (!buf)
3486                 return ret;
3487
3488         path = btrfs_alloc_path();
3489         if (!path) {
3490                 kvfree(buf);
3491                 return ret;
3492         }
3493
3494         path->reada = READA_FORWARD;
3495         /* clone data */
3496         key.objectid = btrfs_ino(BTRFS_I(src));
3497         key.type = BTRFS_EXTENT_DATA_KEY;
3498         key.offset = off;
3499
3500         while (1) {
3501                 u64 next_key_min_offset = key.offset + 1;
3502
3503                 /*
3504                  * note the key will change type as we walk through the
3505                  * tree.
3506                  */
3507                 path->leave_spinning = 1;
3508                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3509                                 0, 0);
3510                 if (ret < 0)
3511                         goto out;
3512                 /*
3513                  * First search, if no extent item that starts at offset off was
3514                  * found but the previous item is an extent item, it's possible
3515                  * it might overlap our target range, therefore process it.
3516                  */
3517                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3518                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3519                                               path->slots[0] - 1);
3520                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3521                                 path->slots[0]--;
3522                 }
3523
3524                 nritems = btrfs_header_nritems(path->nodes[0]);
3525 process_slot:
3526                 if (path->slots[0] >= nritems) {
3527                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3528                         if (ret < 0)
3529                                 goto out;
3530                         if (ret > 0)
3531                                 break;
3532                         nritems = btrfs_header_nritems(path->nodes[0]);
3533                 }
3534                 leaf = path->nodes[0];
3535                 slot = path->slots[0];
3536
3537                 btrfs_item_key_to_cpu(leaf, &key, slot);
3538                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3539                     key.objectid != btrfs_ino(BTRFS_I(src)))
3540                         break;
3541
3542                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3543                         struct btrfs_file_extent_item *extent;
3544                         int type;
3545                         u32 size;
3546                         struct btrfs_key new_key;
3547                         u64 disko = 0, diskl = 0;
3548                         u64 datao = 0, datal = 0;
3549                         u8 comp;
3550                         u64 drop_start;
3551
3552                         extent = btrfs_item_ptr(leaf, slot,
3553                                                 struct btrfs_file_extent_item);
3554                         comp = btrfs_file_extent_compression(leaf, extent);
3555                         type = btrfs_file_extent_type(leaf, extent);
3556                         if (type == BTRFS_FILE_EXTENT_REG ||
3557                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3558                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3559                                                                       extent);
3560                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3561                                                                  extent);
3562                                 datao = btrfs_file_extent_offset(leaf, extent);
3563                                 datal = btrfs_file_extent_num_bytes(leaf,
3564                                                                     extent);
3565                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3566                                 /* take upper bound, may be compressed */
3567                                 datal = btrfs_file_extent_ram_bytes(leaf,
3568                                                                     extent);
3569                         }
3570
3571                         /*
3572                          * The first search might have left us at an extent
3573                          * item that ends before our target range's start, can
3574                          * happen if we have holes and NO_HOLES feature enabled.
3575                          */
3576                         if (key.offset + datal <= off) {
3577                                 path->slots[0]++;
3578                                 goto process_slot;
3579                         } else if (key.offset >= off + len) {
3580                                 break;
3581                         }
3582                         next_key_min_offset = key.offset + datal;
3583                         size = btrfs_item_size_nr(leaf, slot);
3584                         read_extent_buffer(leaf, buf,
3585                                            btrfs_item_ptr_offset(leaf, slot),
3586                                            size);
3587
3588                         btrfs_release_path(path);
3589                         path->leave_spinning = 0;
3590
3591                         memcpy(&new_key, &key, sizeof(new_key));
3592                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3593                         if (off <= key.offset)
3594                                 new_key.offset = key.offset + destoff - off;
3595                         else
3596                                 new_key.offset = destoff;
3597
3598                         /*
3599                          * Deal with a hole that doesn't have an extent item
3600                          * that represents it (NO_HOLES feature enabled).
3601                          * This hole is either in the middle of the cloning
3602                          * range or at the beginning (fully overlaps it or
3603                          * partially overlaps it).
3604                          */
3605                         if (new_key.offset != last_dest_end)
3606                                 drop_start = last_dest_end;
3607                         else
3608                                 drop_start = new_key.offset;
3609
3610                         /*
3611                          * 1 - adjusting old extent (we may have to split it)
3612                          * 1 - add new extent
3613                          * 1 - inode update
3614                          */
3615                         trans = btrfs_start_transaction(root, 3);
3616                         if (IS_ERR(trans)) {
3617                                 ret = PTR_ERR(trans);
3618                                 goto out;
3619                         }
3620
3621                         if (type == BTRFS_FILE_EXTENT_REG ||
3622                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3623                                 /*
3624                                  *    a  | --- range to clone ---|  b
3625                                  * | ------------- extent ------------- |
3626                                  */
3627
3628                                 /* subtract range b */
3629                                 if (key.offset + datal > off + len)
3630                                         datal = off + len - key.offset;
3631
3632                                 /* subtract range a */
3633                                 if (off > key.offset) {
3634                                         datao += off - key.offset;
3635                                         datal -= off - key.offset;
3636                                 }
3637
3638                                 ret = btrfs_drop_extents(trans, root, inode,
3639                                                          drop_start,
3640                                                          new_key.offset + datal,
3641                                                          1);
3642                                 if (ret) {
3643                                         if (ret != -EOPNOTSUPP)
3644                                                 btrfs_abort_transaction(trans,
3645                                                                         ret);
3646                                         btrfs_end_transaction(trans);
3647                                         goto out;
3648                                 }
3649
3650                                 ret = btrfs_insert_empty_item(trans, root, path,
3651                                                               &new_key, size);
3652                                 if (ret) {
3653                                         btrfs_abort_transaction(trans, ret);
3654                                         btrfs_end_transaction(trans);
3655                                         goto out;
3656                                 }
3657
3658                                 leaf = path->nodes[0];
3659                                 slot = path->slots[0];
3660                                 write_extent_buffer(leaf, buf,
3661                                             btrfs_item_ptr_offset(leaf, slot),
3662                                             size);
3663
3664                                 extent = btrfs_item_ptr(leaf, slot,
3665                                                 struct btrfs_file_extent_item);
3666
3667                                 /* disko == 0 means it's a hole */
3668                                 if (!disko)
3669                                         datao = 0;
3670
3671                                 btrfs_set_file_extent_offset(leaf, extent,
3672                                                              datao);
3673                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3674                                                                 datal);
3675
3676                                 if (disko) {
3677                                         inode_add_bytes(inode, datal);
3678                                         ret = btrfs_inc_extent_ref(trans,
3679                                                         root,
3680                                                         disko, diskl, 0,
3681                                                         root->root_key.objectid,
3682                                                         btrfs_ino(BTRFS_I(inode)),
3683                                                         new_key.offset - datao);
3684                                         if (ret) {
3685                                                 btrfs_abort_transaction(trans,
3686                                                                         ret);
3687                                                 btrfs_end_transaction(trans);
3688                                                 goto out;
3689
3690                                         }
3691                                 }
3692                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3693                                 u64 skip = 0;
3694                                 u64 trim = 0;
3695
3696                                 if (off > key.offset) {
3697                                         skip = off - key.offset;
3698                                         new_key.offset += skip;
3699                                 }
3700
3701                                 if (key.offset + datal > off + len)
3702                                         trim = key.offset + datal - (off + len);
3703
3704                                 if (comp && (skip || trim)) {
3705                                         ret = -EINVAL;
3706                                         btrfs_end_transaction(trans);
3707                                         goto out;
3708                                 }
3709                                 size -= skip + trim;
3710                                 datal -= skip + trim;
3711
3712                                 ret = clone_copy_inline_extent(inode,
3713                                                                trans, path,
3714                                                                &new_key,
3715                                                                drop_start,
3716                                                                datal,
3717                                                                skip, size, buf);
3718                                 if (ret) {
3719                                         if (ret != -EOPNOTSUPP)
3720                                                 btrfs_abort_transaction(trans,
3721                                                                         ret);
3722                                         btrfs_end_transaction(trans);
3723                                         goto out;
3724                                 }
3725                                 leaf = path->nodes[0];
3726                                 slot = path->slots[0];
3727                         }
3728
3729                         /* If we have an implicit hole (NO_HOLES feature). */
3730                         if (drop_start < new_key.offset)
3731                                 clone_update_extent_map(BTRFS_I(inode), trans,
3732                                                 NULL, drop_start,
3733                                                 new_key.offset - drop_start);
3734
3735                         clone_update_extent_map(BTRFS_I(inode), trans,
3736                                         path, 0, 0);
3737
3738                         btrfs_mark_buffer_dirty(leaf);
3739                         btrfs_release_path(path);
3740
3741                         last_dest_end = ALIGN(new_key.offset + datal,
3742                                               fs_info->sectorsize);
3743                         ret = clone_finish_inode_update(trans, inode,
3744                                                         last_dest_end,
3745                                                         destoff, olen,
3746                                                         no_time_update);
3747                         if (ret)
3748                                 goto out;
3749                         if (new_key.offset + datal >= destoff + len)
3750                                 break;
3751                 }
3752                 btrfs_release_path(path);
3753                 key.offset = next_key_min_offset;
3754
3755                 if (fatal_signal_pending(current)) {
3756                         ret = -EINTR;
3757                         goto out;
3758                 }
3759         }
3760         ret = 0;
3761
3762         if (last_dest_end < destoff + len) {
3763                 /*
3764                  * We have an implicit hole (NO_HOLES feature is enabled) that
3765                  * fully or partially overlaps our cloning range at its end.
3766                  */
3767                 btrfs_release_path(path);
3768
3769                 /*
3770                  * 1 - remove extent(s)
3771                  * 1 - inode update
3772                  */
3773                 trans = btrfs_start_transaction(root, 2);
3774                 if (IS_ERR(trans)) {
3775                         ret = PTR_ERR(trans);
3776                         goto out;
3777                 }
3778                 ret = btrfs_drop_extents(trans, root, inode,
3779                                          last_dest_end, destoff + len, 1);
3780                 if (ret) {
3781                         if (ret != -EOPNOTSUPP)
3782                                 btrfs_abort_transaction(trans, ret);
3783                         btrfs_end_transaction(trans);
3784                         goto out;
3785                 }
3786                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3787                                 last_dest_end,
3788                                 destoff + len - last_dest_end);
3789                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3790                                                 destoff, olen, no_time_update);
3791         }
3792
3793 out:
3794         btrfs_free_path(path);
3795         kvfree(buf);
3796         return ret;
3797 }
3798
3799 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3800                                         u64 off, u64 olen, u64 destoff)
3801 {
3802         struct inode *inode = file_inode(file);
3803         struct inode *src = file_inode(file_src);
3804         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3805         struct btrfs_root *root = BTRFS_I(inode)->root;
3806         int ret;
3807         u64 len = olen;
3808         u64 bs = fs_info->sb->s_blocksize;
3809         int same_inode = src == inode;
3810
3811         /*
3812          * TODO:
3813          * - split compressed inline extents.  annoying: we need to
3814          *   decompress into destination's address_space (the file offset
3815          *   may change, so source mapping won't do), then recompress (or
3816          *   otherwise reinsert) a subrange.
3817          *
3818          * - split destination inode's inline extents.  The inline extents can
3819          *   be either compressed or non-compressed.
3820          */
3821
3822         if (btrfs_root_readonly(root))
3823                 return -EROFS;
3824
3825         if (file_src->f_path.mnt != file->f_path.mnt ||
3826             src->i_sb != inode->i_sb)
3827                 return -EXDEV;
3828
3829         /* don't make the dst file partly checksummed */
3830         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3831             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3832                 return -EINVAL;
3833
3834         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3835                 return -EISDIR;
3836
3837         if (!same_inode) {
3838                 btrfs_double_inode_lock(src, inode);
3839         } else {
3840                 inode_lock(src);
3841         }
3842
3843         /* determine range to clone */
3844         ret = -EINVAL;
3845         if (off + len > src->i_size || off + len < off)
3846                 goto out_unlock;
3847         if (len == 0)
3848                 olen = len = src->i_size - off;
3849         /* if we extend to eof, continue to block boundary */
3850         if (off + len == src->i_size)
3851                 len = ALIGN(src->i_size, bs) - off;
3852
3853         if (len == 0) {
3854                 ret = 0;
3855                 goto out_unlock;
3856         }
3857
3858         /* verify the end result is block aligned */
3859         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3860             !IS_ALIGNED(destoff, bs))
3861                 goto out_unlock;
3862
3863         /* verify if ranges are overlapped within the same file */
3864         if (same_inode) {
3865                 if (destoff + len > off && destoff < off + len)
3866                         goto out_unlock;
3867         }
3868
3869         if (destoff > inode->i_size) {
3870                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3871                 if (ret)
3872                         goto out_unlock;
3873         }
3874
3875         /*
3876          * Lock the target range too. Right after we replace the file extent
3877          * items in the fs tree (which now point to the cloned data), we might
3878          * have a worker replace them with extent items relative to a write
3879          * operation that was issued before this clone operation (i.e. confront
3880          * with inode.c:btrfs_finish_ordered_io).
3881          */
3882         if (same_inode) {
3883                 u64 lock_start = min_t(u64, off, destoff);
3884                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3885
3886                 ret = lock_extent_range(src, lock_start, lock_len, true);
3887         } else {
3888                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3889                                                true);
3890         }
3891         ASSERT(ret == 0);
3892         if (WARN_ON(ret)) {
3893                 /* ranges in the io trees already unlocked */
3894                 goto out_unlock;
3895         }
3896
3897         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3898
3899         if (same_inode) {
3900                 u64 lock_start = min_t(u64, off, destoff);
3901                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3902
3903                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3904         } else {
3905                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3906         }
3907         /*
3908          * Truncate page cache pages so that future reads will see the cloned
3909          * data immediately and not the previous data.
3910          */
3911         truncate_inode_pages_range(&inode->i_data,
3912                                 round_down(destoff, PAGE_SIZE),
3913                                 round_up(destoff + len, PAGE_SIZE) - 1);
3914 out_unlock:
3915         if (!same_inode)
3916                 btrfs_double_inode_unlock(src, inode);
3917         else
3918                 inode_unlock(src);
3919         return ret;
3920 }
3921
3922 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3923                 struct file *dst_file, loff_t destoff, u64 len)
3924 {
3925         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3926 }
3927
3928 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3929 {
3930         struct inode *inode = file_inode(file);
3931         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3932         struct btrfs_root *root = BTRFS_I(inode)->root;
3933         struct btrfs_root *new_root;
3934         struct btrfs_dir_item *di;
3935         struct btrfs_trans_handle *trans;
3936         struct btrfs_path *path;
3937         struct btrfs_key location;
3938         struct btrfs_disk_key disk_key;
3939         u64 objectid = 0;
3940         u64 dir_id;
3941         int ret;
3942
3943         if (!capable(CAP_SYS_ADMIN))
3944                 return -EPERM;
3945
3946         ret = mnt_want_write_file(file);
3947         if (ret)
3948                 return ret;
3949
3950         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3951                 ret = -EFAULT;
3952                 goto out;
3953         }
3954
3955         if (!objectid)
3956                 objectid = BTRFS_FS_TREE_OBJECTID;
3957
3958         location.objectid = objectid;
3959         location.type = BTRFS_ROOT_ITEM_KEY;
3960         location.offset = (u64)-1;
3961
3962         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
3963         if (IS_ERR(new_root)) {
3964                 ret = PTR_ERR(new_root);
3965                 goto out;
3966         }
3967         if (!is_fstree(new_root->objectid)) {
3968                 ret = -ENOENT;
3969                 goto out;
3970         }
3971
3972         path = btrfs_alloc_path();
3973         if (!path) {
3974                 ret = -ENOMEM;
3975                 goto out;
3976         }
3977         path->leave_spinning = 1;
3978
3979         trans = btrfs_start_transaction(root, 1);
3980         if (IS_ERR(trans)) {
3981                 btrfs_free_path(path);
3982                 ret = PTR_ERR(trans);
3983                 goto out;
3984         }
3985
3986         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3987         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3988                                    dir_id, "default", 7, 1);
3989         if (IS_ERR_OR_NULL(di)) {
3990                 btrfs_free_path(path);
3991                 btrfs_end_transaction(trans);
3992                 btrfs_err(fs_info,
3993                           "Umm, you don't have the default diritem, this isn't going to work");
3994                 ret = -ENOENT;
3995                 goto out;
3996         }
3997
3998         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3999         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4000         btrfs_mark_buffer_dirty(path->nodes[0]);
4001         btrfs_free_path(path);
4002
4003         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4004         btrfs_end_transaction(trans);
4005 out:
4006         mnt_drop_write_file(file);
4007         return ret;
4008 }
4009
4010 void btrfs_get_block_group_info(struct list_head *groups_list,
4011                                 struct btrfs_ioctl_space_info *space)
4012 {
4013         struct btrfs_block_group_cache *block_group;
4014
4015         space->total_bytes = 0;
4016         space->used_bytes = 0;
4017         space->flags = 0;
4018         list_for_each_entry(block_group, groups_list, list) {
4019                 space->flags = block_group->flags;
4020                 space->total_bytes += block_group->key.offset;
4021                 space->used_bytes +=
4022                         btrfs_block_group_used(&block_group->item);
4023         }
4024 }
4025
4026 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4027                                    void __user *arg)
4028 {
4029         struct btrfs_ioctl_space_args space_args;
4030         struct btrfs_ioctl_space_info space;
4031         struct btrfs_ioctl_space_info *dest;
4032         struct btrfs_ioctl_space_info *dest_orig;
4033         struct btrfs_ioctl_space_info __user *user_dest;
4034         struct btrfs_space_info *info;
4035         static const u64 types[] = {
4036                 BTRFS_BLOCK_GROUP_DATA,
4037                 BTRFS_BLOCK_GROUP_SYSTEM,
4038                 BTRFS_BLOCK_GROUP_METADATA,
4039                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4040         };
4041         int num_types = 4;
4042         int alloc_size;
4043         int ret = 0;
4044         u64 slot_count = 0;
4045         int i, c;
4046
4047         if (copy_from_user(&space_args,
4048                            (struct btrfs_ioctl_space_args __user *)arg,
4049                            sizeof(space_args)))
4050                 return -EFAULT;
4051
4052         for (i = 0; i < num_types; i++) {
4053                 struct btrfs_space_info *tmp;
4054
4055                 info = NULL;
4056                 rcu_read_lock();
4057                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4058                                         list) {
4059                         if (tmp->flags == types[i]) {
4060                                 info = tmp;
4061                                 break;
4062                         }
4063                 }
4064                 rcu_read_unlock();
4065
4066                 if (!info)
4067                         continue;
4068
4069                 down_read(&info->groups_sem);
4070                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4071                         if (!list_empty(&info->block_groups[c]))
4072                                 slot_count++;
4073                 }
4074                 up_read(&info->groups_sem);
4075         }
4076
4077         /*
4078          * Global block reserve, exported as a space_info
4079          */
4080         slot_count++;
4081
4082         /* space_slots == 0 means they are asking for a count */
4083         if (space_args.space_slots == 0) {
4084                 space_args.total_spaces = slot_count;
4085                 goto out;
4086         }
4087
4088         slot_count = min_t(u64, space_args.space_slots, slot_count);
4089
4090         alloc_size = sizeof(*dest) * slot_count;
4091
4092         /* we generally have at most 6 or so space infos, one for each raid
4093          * level.  So, a whole page should be more than enough for everyone
4094          */
4095         if (alloc_size > PAGE_SIZE)
4096                 return -ENOMEM;
4097
4098         space_args.total_spaces = 0;
4099         dest = kmalloc(alloc_size, GFP_KERNEL);
4100         if (!dest)
4101                 return -ENOMEM;
4102         dest_orig = dest;
4103
4104         /* now we have a buffer to copy into */
4105         for (i = 0; i < num_types; i++) {
4106                 struct btrfs_space_info *tmp;
4107
4108                 if (!slot_count)
4109                         break;
4110
4111                 info = NULL;
4112                 rcu_read_lock();
4113                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4114                                         list) {
4115                         if (tmp->flags == types[i]) {
4116                                 info = tmp;
4117                                 break;
4118                         }
4119                 }
4120                 rcu_read_unlock();
4121
4122                 if (!info)
4123                         continue;
4124                 down_read(&info->groups_sem);
4125                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4126                         if (!list_empty(&info->block_groups[c])) {
4127                                 btrfs_get_block_group_info(
4128                                         &info->block_groups[c], &space);
4129                                 memcpy(dest, &space, sizeof(space));
4130                                 dest++;
4131                                 space_args.total_spaces++;
4132                                 slot_count--;
4133                         }
4134                         if (!slot_count)
4135                                 break;
4136                 }
4137                 up_read(&info->groups_sem);
4138         }
4139
4140         /*
4141          * Add global block reserve
4142          */
4143         if (slot_count) {
4144                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4145
4146                 spin_lock(&block_rsv->lock);
4147                 space.total_bytes = block_rsv->size;
4148                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4149                 spin_unlock(&block_rsv->lock);
4150                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4151                 memcpy(dest, &space, sizeof(space));
4152                 space_args.total_spaces++;
4153         }
4154
4155         user_dest = (struct btrfs_ioctl_space_info __user *)
4156                 (arg + sizeof(struct btrfs_ioctl_space_args));
4157
4158         if (copy_to_user(user_dest, dest_orig, alloc_size))
4159                 ret = -EFAULT;
4160
4161         kfree(dest_orig);
4162 out:
4163         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4164                 ret = -EFAULT;
4165
4166         return ret;
4167 }
4168
4169 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4170                                             void __user *argp)
4171 {
4172         struct btrfs_trans_handle *trans;
4173         u64 transid;
4174         int ret;
4175
4176         trans = btrfs_attach_transaction_barrier(root);
4177         if (IS_ERR(trans)) {
4178                 if (PTR_ERR(trans) != -ENOENT)
4179                         return PTR_ERR(trans);
4180
4181                 /* No running transaction, don't bother */
4182                 transid = root->fs_info->last_trans_committed;
4183                 goto out;
4184         }
4185         transid = trans->transid;
4186         ret = btrfs_commit_transaction_async(trans, 0);
4187         if (ret) {
4188                 btrfs_end_transaction(trans);
4189                 return ret;
4190         }
4191 out:
4192         if (argp)
4193                 if (copy_to_user(argp, &transid, sizeof(transid)))
4194                         return -EFAULT;
4195         return 0;
4196 }
4197
4198 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4199                                            void __user *argp)
4200 {
4201         u64 transid;
4202
4203         if (argp) {
4204                 if (copy_from_user(&transid, argp, sizeof(transid)))
4205                         return -EFAULT;
4206         } else {
4207                 transid = 0;  /* current trans */
4208         }
4209         return btrfs_wait_for_commit(fs_info, transid);
4210 }
4211
4212 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4213 {
4214         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4215         struct btrfs_ioctl_scrub_args *sa;
4216         int ret;
4217
4218         if (!capable(CAP_SYS_ADMIN))
4219                 return -EPERM;
4220
4221         sa = memdup_user(arg, sizeof(*sa));
4222         if (IS_ERR(sa))
4223                 return PTR_ERR(sa);
4224
4225         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4226                 ret = mnt_want_write_file(file);
4227                 if (ret)
4228                         goto out;
4229         }
4230
4231         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4232                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4233                               0);
4234
4235         if (copy_to_user(arg, sa, sizeof(*sa)))
4236                 ret = -EFAULT;
4237
4238         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4239                 mnt_drop_write_file(file);
4240 out:
4241         kfree(sa);
4242         return ret;
4243 }
4244
4245 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4246 {
4247         if (!capable(CAP_SYS_ADMIN))
4248                 return -EPERM;
4249
4250         return btrfs_scrub_cancel(fs_info);
4251 }
4252
4253 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4254                                        void __user *arg)
4255 {
4256         struct btrfs_ioctl_scrub_args *sa;
4257         int ret;
4258
4259         if (!capable(CAP_SYS_ADMIN))
4260                 return -EPERM;
4261
4262         sa = memdup_user(arg, sizeof(*sa));
4263         if (IS_ERR(sa))
4264                 return PTR_ERR(sa);
4265
4266         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4267
4268         if (copy_to_user(arg, sa, sizeof(*sa)))
4269                 ret = -EFAULT;
4270
4271         kfree(sa);
4272         return ret;
4273 }
4274
4275 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4276                                       void __user *arg)
4277 {
4278         struct btrfs_ioctl_get_dev_stats *sa;
4279         int ret;
4280
4281         sa = memdup_user(arg, sizeof(*sa));
4282         if (IS_ERR(sa))
4283                 return PTR_ERR(sa);
4284
4285         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4286                 kfree(sa);
4287                 return -EPERM;
4288         }
4289
4290         ret = btrfs_get_dev_stats(fs_info, sa);
4291
4292         if (copy_to_user(arg, sa, sizeof(*sa)))
4293                 ret = -EFAULT;
4294
4295         kfree(sa);
4296         return ret;
4297 }
4298
4299 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4300                                     void __user *arg)
4301 {
4302         struct btrfs_ioctl_dev_replace_args *p;
4303         int ret;
4304
4305         if (!capable(CAP_SYS_ADMIN))
4306                 return -EPERM;
4307
4308         p = memdup_user(arg, sizeof(*p));
4309         if (IS_ERR(p))
4310                 return PTR_ERR(p);
4311
4312         switch (p->cmd) {
4313         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4314                 if (sb_rdonly(fs_info->sb)) {
4315                         ret = -EROFS;
4316                         goto out;
4317                 }
4318                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4319                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4320                 } else {
4321                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4322                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4323                 }
4324                 break;
4325         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4326                 btrfs_dev_replace_status(fs_info, p);
4327                 ret = 0;
4328                 break;
4329         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4330                 p->result = btrfs_dev_replace_cancel(fs_info);
4331                 ret = 0;
4332                 break;
4333         default:
4334                 ret = -EINVAL;
4335                 break;
4336         }
4337
4338         if (copy_to_user(arg, p, sizeof(*p)))
4339                 ret = -EFAULT;
4340 out:
4341         kfree(p);
4342         return ret;
4343 }
4344
4345 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4346 {
4347         int ret = 0;
4348         int i;
4349         u64 rel_ptr;
4350         int size;
4351         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4352         struct inode_fs_paths *ipath = NULL;
4353         struct btrfs_path *path;
4354
4355         if (!capable(CAP_DAC_READ_SEARCH))
4356                 return -EPERM;
4357
4358         path = btrfs_alloc_path();
4359         if (!path) {
4360                 ret = -ENOMEM;
4361                 goto out;
4362         }
4363
4364         ipa = memdup_user(arg, sizeof(*ipa));
4365         if (IS_ERR(ipa)) {
4366                 ret = PTR_ERR(ipa);
4367                 ipa = NULL;
4368                 goto out;
4369         }
4370
4371         size = min_t(u32, ipa->size, 4096);
4372         ipath = init_ipath(size, root, path);
4373         if (IS_ERR(ipath)) {
4374                 ret = PTR_ERR(ipath);
4375                 ipath = NULL;
4376                 goto out;
4377         }
4378
4379         ret = paths_from_inode(ipa->inum, ipath);
4380         if (ret < 0)
4381                 goto out;
4382
4383         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4384                 rel_ptr = ipath->fspath->val[i] -
4385                           (u64)(unsigned long)ipath->fspath->val;
4386                 ipath->fspath->val[i] = rel_ptr;
4387         }
4388
4389         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4390                            ipath->fspath, size);
4391         if (ret) {
4392                 ret = -EFAULT;
4393                 goto out;
4394         }
4395
4396 out:
4397         btrfs_free_path(path);
4398         free_ipath(ipath);
4399         kfree(ipa);
4400
4401         return ret;
4402 }
4403
4404 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4405 {
4406         struct btrfs_data_container *inodes = ctx;
4407         const size_t c = 3 * sizeof(u64);
4408
4409         if (inodes->bytes_left >= c) {
4410                 inodes->bytes_left -= c;
4411                 inodes->val[inodes->elem_cnt] = inum;
4412                 inodes->val[inodes->elem_cnt + 1] = offset;
4413                 inodes->val[inodes->elem_cnt + 2] = root;
4414                 inodes->elem_cnt += 3;
4415         } else {
4416                 inodes->bytes_missing += c - inodes->bytes_left;
4417                 inodes->bytes_left = 0;
4418                 inodes->elem_missed += 3;
4419         }
4420
4421         return 0;
4422 }
4423
4424 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4425                                         void __user *arg, int version)
4426 {
4427         int ret = 0;
4428         int size;
4429         struct btrfs_ioctl_logical_ino_args *loi;
4430         struct btrfs_data_container *inodes = NULL;
4431         struct btrfs_path *path = NULL;
4432         bool ignore_offset;
4433
4434         if (!capable(CAP_SYS_ADMIN))
4435                 return -EPERM;
4436
4437         loi = memdup_user(arg, sizeof(*loi));
4438         if (IS_ERR(loi))
4439                 return PTR_ERR(loi);
4440
4441         if (version == 1) {
4442                 ignore_offset = false;
4443                 size = min_t(u32, loi->size, SZ_64K);
4444         } else {
4445                 /* All reserved bits must be 0 for now */
4446                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4447                         ret = -EINVAL;
4448                         goto out_loi;
4449                 }
4450                 /* Only accept flags we have defined so far */
4451                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4452                         ret = -EINVAL;
4453                         goto out_loi;
4454                 }
4455                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4456                 size = min_t(u32, loi->size, SZ_16M);
4457         }
4458
4459         path = btrfs_alloc_path();
4460         if (!path) {
4461                 ret = -ENOMEM;
4462                 goto out;
4463         }
4464
4465         inodes = init_data_container(size);
4466         if (IS_ERR(inodes)) {
4467                 ret = PTR_ERR(inodes);
4468                 inodes = NULL;
4469                 goto out;
4470         }
4471
4472         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4473                                           build_ino_list, inodes, ignore_offset);
4474         if (ret == -EINVAL)
4475                 ret = -ENOENT;
4476         if (ret < 0)
4477                 goto out;
4478
4479         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4480                            size);
4481         if (ret)
4482                 ret = -EFAULT;
4483
4484 out:
4485         btrfs_free_path(path);
4486         kvfree(inodes);
4487 out_loi:
4488         kfree(loi);
4489
4490         return ret;
4491 }
4492
4493 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4494                                struct btrfs_ioctl_balance_args *bargs)
4495 {
4496         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4497
4498         bargs->flags = bctl->flags;
4499
4500         if (atomic_read(&fs_info->balance_running))
4501                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4502         if (atomic_read(&fs_info->balance_pause_req))
4503                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4504         if (atomic_read(&fs_info->balance_cancel_req))
4505                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4506
4507         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4508         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4509         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4510
4511         if (lock) {
4512                 spin_lock(&fs_info->balance_lock);
4513                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4514                 spin_unlock(&fs_info->balance_lock);
4515         } else {
4516                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4517         }
4518 }
4519
4520 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4521 {
4522         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4523         struct btrfs_fs_info *fs_info = root->fs_info;
4524         struct btrfs_ioctl_balance_args *bargs;
4525         struct btrfs_balance_control *bctl;
4526         bool need_unlock; /* for mut. excl. ops lock */
4527         int ret;
4528
4529         if (!capable(CAP_SYS_ADMIN))
4530                 return -EPERM;
4531
4532         ret = mnt_want_write_file(file);
4533         if (ret)
4534                 return ret;
4535
4536 again:
4537         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4538                 mutex_lock(&fs_info->volume_mutex);
4539                 mutex_lock(&fs_info->balance_mutex);
4540                 need_unlock = true;
4541                 goto locked;
4542         }
4543
4544         /*
4545          * mut. excl. ops lock is locked.  Three possibilities:
4546          *   (1) some other op is running
4547          *   (2) balance is running
4548          *   (3) balance is paused -- special case (think resume)
4549          */
4550         mutex_lock(&fs_info->balance_mutex);
4551         if (fs_info->balance_ctl) {
4552                 /* this is either (2) or (3) */
4553                 if (!atomic_read(&fs_info->balance_running)) {
4554                         mutex_unlock(&fs_info->balance_mutex);
4555                         if (!mutex_trylock(&fs_info->volume_mutex))
4556                                 goto again;
4557                         mutex_lock(&fs_info->balance_mutex);
4558
4559                         if (fs_info->balance_ctl &&
4560                             !atomic_read(&fs_info->balance_running)) {
4561                                 /* this is (3) */
4562                                 need_unlock = false;
4563                                 goto locked;
4564                         }
4565
4566                         mutex_unlock(&fs_info->balance_mutex);
4567                         mutex_unlock(&fs_info->volume_mutex);
4568                         goto again;
4569                 } else {
4570                         /* this is (2) */
4571                         mutex_unlock(&fs_info->balance_mutex);
4572                         ret = -EINPROGRESS;
4573                         goto out;
4574                 }
4575         } else {
4576                 /* this is (1) */
4577                 mutex_unlock(&fs_info->balance_mutex);
4578                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4579                 goto out;
4580         }
4581
4582 locked:
4583         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4584
4585         if (arg) {
4586                 bargs = memdup_user(arg, sizeof(*bargs));
4587                 if (IS_ERR(bargs)) {
4588                         ret = PTR_ERR(bargs);
4589                         goto out_unlock;
4590                 }
4591
4592                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4593                         if (!fs_info->balance_ctl) {
4594                                 ret = -ENOTCONN;
4595                                 goto out_bargs;
4596                         }
4597
4598                         bctl = fs_info->balance_ctl;
4599                         spin_lock(&fs_info->balance_lock);
4600                         bctl->flags |= BTRFS_BALANCE_RESUME;
4601                         spin_unlock(&fs_info->balance_lock);
4602
4603                         goto do_balance;
4604                 }
4605         } else {
4606                 bargs = NULL;
4607         }
4608
4609         if (fs_info->balance_ctl) {
4610                 ret = -EINPROGRESS;
4611                 goto out_bargs;
4612         }
4613
4614         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4615         if (!bctl) {
4616                 ret = -ENOMEM;
4617                 goto out_bargs;
4618         }
4619
4620         bctl->fs_info = fs_info;
4621         if (arg) {
4622                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4623                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4624                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4625
4626                 bctl->flags = bargs->flags;
4627         } else {
4628                 /* balance everything - no filters */
4629                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4630         }
4631
4632         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4633                 ret = -EINVAL;
4634                 goto out_bctl;
4635         }
4636
4637 do_balance:
4638         /*
4639          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4640          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4641          * or, if restriper was paused all the way until unmount, in
4642          * free_fs_info.  The flag is cleared in __cancel_balance.
4643          */
4644         need_unlock = false;
4645
4646         ret = btrfs_balance(bctl, bargs);
4647         bctl = NULL;
4648
4649         if (arg) {
4650                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4651                         ret = -EFAULT;
4652         }
4653
4654 out_bctl:
4655         kfree(bctl);
4656 out_bargs:
4657         kfree(bargs);
4658 out_unlock:
4659         mutex_unlock(&fs_info->balance_mutex);
4660         mutex_unlock(&fs_info->volume_mutex);
4661         if (need_unlock)
4662                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4663 out:
4664         mnt_drop_write_file(file);
4665         return ret;
4666 }
4667
4668 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4669 {
4670         if (!capable(CAP_SYS_ADMIN))
4671                 return -EPERM;
4672
4673         switch (cmd) {
4674         case BTRFS_BALANCE_CTL_PAUSE:
4675                 return btrfs_pause_balance(fs_info);
4676         case BTRFS_BALANCE_CTL_CANCEL:
4677                 return btrfs_cancel_balance(fs_info);
4678         }
4679
4680         return -EINVAL;
4681 }
4682
4683 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4684                                          void __user *arg)
4685 {
4686         struct btrfs_ioctl_balance_args *bargs;
4687         int ret = 0;
4688
4689         if (!capable(CAP_SYS_ADMIN))
4690                 return -EPERM;
4691
4692         mutex_lock(&fs_info->balance_mutex);
4693         if (!fs_info->balance_ctl) {
4694                 ret = -ENOTCONN;
4695                 goto out;
4696         }
4697
4698         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4699         if (!bargs) {
4700                 ret = -ENOMEM;
4701                 goto out;
4702         }
4703
4704         update_ioctl_balance_args(fs_info, 1, bargs);
4705
4706         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4707                 ret = -EFAULT;
4708
4709         kfree(bargs);
4710 out:
4711         mutex_unlock(&fs_info->balance_mutex);
4712         return ret;
4713 }
4714
4715 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4716 {
4717         struct inode *inode = file_inode(file);
4718         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4719         struct btrfs_ioctl_quota_ctl_args *sa;
4720         struct btrfs_trans_handle *trans = NULL;
4721         int ret;
4722         int err;
4723
4724         if (!capable(CAP_SYS_ADMIN))
4725                 return -EPERM;
4726
4727         ret = mnt_want_write_file(file);
4728         if (ret)
4729                 return ret;
4730
4731         sa = memdup_user(arg, sizeof(*sa));
4732         if (IS_ERR(sa)) {
4733                 ret = PTR_ERR(sa);
4734                 goto drop_write;
4735         }
4736
4737         down_write(&fs_info->subvol_sem);
4738         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4739         if (IS_ERR(trans)) {
4740                 ret = PTR_ERR(trans);
4741                 goto out;
4742         }
4743
4744         switch (sa->cmd) {
4745         case BTRFS_QUOTA_CTL_ENABLE:
4746                 ret = btrfs_quota_enable(trans, fs_info);
4747                 break;
4748         case BTRFS_QUOTA_CTL_DISABLE:
4749                 ret = btrfs_quota_disable(trans, fs_info);
4750                 break;
4751         default:
4752                 ret = -EINVAL;
4753                 break;
4754         }
4755
4756         err = btrfs_commit_transaction(trans);
4757         if (err && !ret)
4758                 ret = err;
4759 out:
4760         kfree(sa);
4761         up_write(&fs_info->subvol_sem);
4762 drop_write:
4763         mnt_drop_write_file(file);
4764         return ret;
4765 }
4766
4767 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4768 {
4769         struct inode *inode = file_inode(file);
4770         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4771         struct btrfs_root *root = BTRFS_I(inode)->root;
4772         struct btrfs_ioctl_qgroup_assign_args *sa;
4773         struct btrfs_trans_handle *trans;
4774         int ret;
4775         int err;
4776
4777         if (!capable(CAP_SYS_ADMIN))
4778                 return -EPERM;
4779
4780         ret = mnt_want_write_file(file);
4781         if (ret)
4782                 return ret;
4783
4784         sa = memdup_user(arg, sizeof(*sa));
4785         if (IS_ERR(sa)) {
4786                 ret = PTR_ERR(sa);
4787                 goto drop_write;
4788         }
4789
4790         trans = btrfs_join_transaction(root);
4791         if (IS_ERR(trans)) {
4792                 ret = PTR_ERR(trans);
4793                 goto out;
4794         }
4795
4796         if (sa->assign) {
4797                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4798                                                 sa->src, sa->dst);
4799         } else {
4800                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4801                                                 sa->src, sa->dst);
4802         }
4803
4804         /* update qgroup status and info */
4805         err = btrfs_run_qgroups(trans, fs_info);
4806         if (err < 0)
4807                 btrfs_handle_fs_error(fs_info, err,
4808                                       "failed to update qgroup status and info");
4809         err = btrfs_end_transaction(trans);
4810         if (err && !ret)
4811                 ret = err;
4812
4813 out:
4814         kfree(sa);
4815 drop_write:
4816         mnt_drop_write_file(file);
4817         return ret;
4818 }
4819
4820 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4821 {
4822         struct inode *inode = file_inode(file);
4823         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4824         struct btrfs_root *root = BTRFS_I(inode)->root;
4825         struct btrfs_ioctl_qgroup_create_args *sa;
4826         struct btrfs_trans_handle *trans;
4827         int ret;
4828         int err;
4829
4830         if (!capable(CAP_SYS_ADMIN))
4831                 return -EPERM;
4832
4833         ret = mnt_want_write_file(file);
4834         if (ret)
4835                 return ret;
4836
4837         sa = memdup_user(arg, sizeof(*sa));
4838         if (IS_ERR(sa)) {
4839                 ret = PTR_ERR(sa);
4840                 goto drop_write;
4841         }
4842
4843         if (!sa->qgroupid) {
4844                 ret = -EINVAL;
4845                 goto out;
4846         }
4847
4848         trans = btrfs_join_transaction(root);
4849         if (IS_ERR(trans)) {
4850                 ret = PTR_ERR(trans);
4851                 goto out;
4852         }
4853
4854         if (sa->create) {
4855                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4856         } else {
4857                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4858         }
4859
4860         err = btrfs_end_transaction(trans);
4861         if (err && !ret)
4862                 ret = err;
4863
4864 out:
4865         kfree(sa);
4866 drop_write:
4867         mnt_drop_write_file(file);
4868         return ret;
4869 }
4870
4871 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4872 {
4873         struct inode *inode = file_inode(file);
4874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4875         struct btrfs_root *root = BTRFS_I(inode)->root;
4876         struct btrfs_ioctl_qgroup_limit_args *sa;
4877         struct btrfs_trans_handle *trans;
4878         int ret;
4879         int err;
4880         u64 qgroupid;
4881
4882         if (!capable(CAP_SYS_ADMIN))
4883                 return -EPERM;
4884
4885         ret = mnt_want_write_file(file);
4886         if (ret)
4887                 return ret;
4888
4889         sa = memdup_user(arg, sizeof(*sa));
4890         if (IS_ERR(sa)) {
4891                 ret = PTR_ERR(sa);
4892                 goto drop_write;
4893         }
4894
4895         trans = btrfs_join_transaction(root);
4896         if (IS_ERR(trans)) {
4897                 ret = PTR_ERR(trans);
4898                 goto out;
4899         }
4900
4901         qgroupid = sa->qgroupid;
4902         if (!qgroupid) {
4903                 /* take the current subvol as qgroup */
4904                 qgroupid = root->root_key.objectid;
4905         }
4906
4907         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
4908
4909         err = btrfs_end_transaction(trans);
4910         if (err && !ret)
4911                 ret = err;
4912
4913 out:
4914         kfree(sa);
4915 drop_write:
4916         mnt_drop_write_file(file);
4917         return ret;
4918 }
4919
4920 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4921 {
4922         struct inode *inode = file_inode(file);
4923         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4924         struct btrfs_ioctl_quota_rescan_args *qsa;
4925         int ret;
4926
4927         if (!capable(CAP_SYS_ADMIN))
4928                 return -EPERM;
4929
4930         ret = mnt_want_write_file(file);
4931         if (ret)
4932                 return ret;
4933
4934         qsa = memdup_user(arg, sizeof(*qsa));
4935         if (IS_ERR(qsa)) {
4936                 ret = PTR_ERR(qsa);
4937                 goto drop_write;
4938         }
4939
4940         if (qsa->flags) {
4941                 ret = -EINVAL;
4942                 goto out;
4943         }
4944
4945         ret = btrfs_qgroup_rescan(fs_info);
4946
4947 out:
4948         kfree(qsa);
4949 drop_write:
4950         mnt_drop_write_file(file);
4951         return ret;
4952 }
4953
4954 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4955 {
4956         struct inode *inode = file_inode(file);
4957         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4958         struct btrfs_ioctl_quota_rescan_args *qsa;
4959         int ret = 0;
4960
4961         if (!capable(CAP_SYS_ADMIN))
4962                 return -EPERM;
4963
4964         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4965         if (!qsa)
4966                 return -ENOMEM;
4967
4968         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4969                 qsa->flags = 1;
4970                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4971         }
4972
4973         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4974                 ret = -EFAULT;
4975
4976         kfree(qsa);
4977         return ret;
4978 }
4979
4980 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4981 {
4982         struct inode *inode = file_inode(file);
4983         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4984
4985         if (!capable(CAP_SYS_ADMIN))
4986                 return -EPERM;
4987
4988         return btrfs_qgroup_wait_for_completion(fs_info, true);
4989 }
4990
4991 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4992                                             struct btrfs_ioctl_received_subvol_args *sa)
4993 {
4994         struct inode *inode = file_inode(file);
4995         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4996         struct btrfs_root *root = BTRFS_I(inode)->root;
4997         struct btrfs_root_item *root_item = &root->root_item;
4998         struct btrfs_trans_handle *trans;
4999         struct timespec ct = current_time(inode);
5000         int ret = 0;
5001         int received_uuid_changed;
5002
5003         if (!inode_owner_or_capable(inode))
5004                 return -EPERM;
5005
5006         ret = mnt_want_write_file(file);
5007         if (ret < 0)
5008                 return ret;
5009
5010         down_write(&fs_info->subvol_sem);
5011
5012         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5013                 ret = -EINVAL;
5014                 goto out;
5015         }
5016
5017         if (btrfs_root_readonly(root)) {
5018                 ret = -EROFS;
5019                 goto out;
5020         }
5021
5022         /*
5023          * 1 - root item
5024          * 2 - uuid items (received uuid + subvol uuid)
5025          */
5026         trans = btrfs_start_transaction(root, 3);
5027         if (IS_ERR(trans)) {
5028                 ret = PTR_ERR(trans);
5029                 trans = NULL;
5030                 goto out;
5031         }
5032
5033         sa->rtransid = trans->transid;
5034         sa->rtime.sec = ct.tv_sec;
5035         sa->rtime.nsec = ct.tv_nsec;
5036
5037         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5038                                        BTRFS_UUID_SIZE);
5039         if (received_uuid_changed &&
5040             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5041                 ret = btrfs_uuid_tree_rem(trans, fs_info,
5042                                           root_item->received_uuid,
5043                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5044                                           root->root_key.objectid);
5045                 if (ret && ret != -ENOENT) {
5046                         btrfs_abort_transaction(trans, ret);
5047                         btrfs_end_transaction(trans);
5048                         goto out;
5049                 }
5050         }
5051         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5052         btrfs_set_root_stransid(root_item, sa->stransid);
5053         btrfs_set_root_rtransid(root_item, sa->rtransid);
5054         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5055         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5056         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5057         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5058
5059         ret = btrfs_update_root(trans, fs_info->tree_root,
5060                                 &root->root_key, &root->root_item);
5061         if (ret < 0) {
5062                 btrfs_end_transaction(trans);
5063                 goto out;
5064         }
5065         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5066                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5067                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5068                                           root->root_key.objectid);
5069                 if (ret < 0 && ret != -EEXIST) {
5070                         btrfs_abort_transaction(trans, ret);
5071                         btrfs_end_transaction(trans);
5072                         goto out;
5073                 }
5074         }
5075         ret = btrfs_commit_transaction(trans);
5076 out:
5077         up_write(&fs_info->subvol_sem);
5078         mnt_drop_write_file(file);
5079         return ret;
5080 }
5081
5082 #ifdef CONFIG_64BIT
5083 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5084                                                 void __user *arg)
5085 {
5086         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5087         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5088         int ret = 0;
5089
5090         args32 = memdup_user(arg, sizeof(*args32));
5091         if (IS_ERR(args32))
5092                 return PTR_ERR(args32);
5093
5094         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5095         if (!args64) {
5096                 ret = -ENOMEM;
5097                 goto out;
5098         }
5099
5100         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5101         args64->stransid = args32->stransid;
5102         args64->rtransid = args32->rtransid;
5103         args64->stime.sec = args32->stime.sec;
5104         args64->stime.nsec = args32->stime.nsec;
5105         args64->rtime.sec = args32->rtime.sec;
5106         args64->rtime.nsec = args32->rtime.nsec;
5107         args64->flags = args32->flags;
5108
5109         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5110         if (ret)
5111                 goto out;
5112
5113         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5114         args32->stransid = args64->stransid;
5115         args32->rtransid = args64->rtransid;
5116         args32->stime.sec = args64->stime.sec;
5117         args32->stime.nsec = args64->stime.nsec;
5118         args32->rtime.sec = args64->rtime.sec;
5119         args32->rtime.nsec = args64->rtime.nsec;
5120         args32->flags = args64->flags;
5121
5122         ret = copy_to_user(arg, args32, sizeof(*args32));
5123         if (ret)
5124                 ret = -EFAULT;
5125
5126 out:
5127         kfree(args32);
5128         kfree(args64);
5129         return ret;
5130 }
5131 #endif
5132
5133 static long btrfs_ioctl_set_received_subvol(struct file *file,
5134                                             void __user *arg)
5135 {
5136         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5137         int ret = 0;
5138
5139         sa = memdup_user(arg, sizeof(*sa));
5140         if (IS_ERR(sa))
5141                 return PTR_ERR(sa);
5142
5143         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5144
5145         if (ret)
5146                 goto out;
5147
5148         ret = copy_to_user(arg, sa, sizeof(*sa));
5149         if (ret)
5150                 ret = -EFAULT;
5151
5152 out:
5153         kfree(sa);
5154         return ret;
5155 }
5156
5157 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5158 {
5159         struct inode *inode = file_inode(file);
5160         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5161         size_t len;
5162         int ret;
5163         char label[BTRFS_LABEL_SIZE];
5164
5165         spin_lock(&fs_info->super_lock);
5166         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5167         spin_unlock(&fs_info->super_lock);
5168
5169         len = strnlen(label, BTRFS_LABEL_SIZE);
5170
5171         if (len == BTRFS_LABEL_SIZE) {
5172                 btrfs_warn(fs_info,
5173                            "label is too long, return the first %zu bytes",
5174                            --len);
5175         }
5176
5177         ret = copy_to_user(arg, label, len);
5178
5179         return ret ? -EFAULT : 0;
5180 }
5181
5182 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5183 {
5184         struct inode *inode = file_inode(file);
5185         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5186         struct btrfs_root *root = BTRFS_I(inode)->root;
5187         struct btrfs_super_block *super_block = fs_info->super_copy;
5188         struct btrfs_trans_handle *trans;
5189         char label[BTRFS_LABEL_SIZE];
5190         int ret;
5191
5192         if (!capable(CAP_SYS_ADMIN))
5193                 return -EPERM;
5194
5195         if (copy_from_user(label, arg, sizeof(label)))
5196                 return -EFAULT;
5197
5198         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5199                 btrfs_err(fs_info,
5200                           "unable to set label with more than %d bytes",
5201                           BTRFS_LABEL_SIZE - 1);
5202                 return -EINVAL;
5203         }
5204
5205         ret = mnt_want_write_file(file);
5206         if (ret)
5207                 return ret;
5208
5209         trans = btrfs_start_transaction(root, 0);
5210         if (IS_ERR(trans)) {
5211                 ret = PTR_ERR(trans);
5212                 goto out_unlock;
5213         }
5214
5215         spin_lock(&fs_info->super_lock);
5216         strcpy(super_block->label, label);
5217         spin_unlock(&fs_info->super_lock);
5218         ret = btrfs_commit_transaction(trans);
5219
5220 out_unlock:
5221         mnt_drop_write_file(file);
5222         return ret;
5223 }
5224
5225 #define INIT_FEATURE_FLAGS(suffix) \
5226         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5227           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5228           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5229
5230 int btrfs_ioctl_get_supported_features(void __user *arg)
5231 {
5232         static const struct btrfs_ioctl_feature_flags features[3] = {
5233                 INIT_FEATURE_FLAGS(SUPP),
5234                 INIT_FEATURE_FLAGS(SAFE_SET),
5235                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5236         };
5237
5238         if (copy_to_user(arg, &features, sizeof(features)))
5239                 return -EFAULT;
5240
5241         return 0;
5242 }
5243
5244 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5245 {
5246         struct inode *inode = file_inode(file);
5247         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5248         struct btrfs_super_block *super_block = fs_info->super_copy;
5249         struct btrfs_ioctl_feature_flags features;
5250
5251         features.compat_flags = btrfs_super_compat_flags(super_block);
5252         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5253         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5254
5255         if (copy_to_user(arg, &features, sizeof(features)))
5256                 return -EFAULT;
5257
5258         return 0;
5259 }
5260
5261 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5262                               enum btrfs_feature_set set,
5263                               u64 change_mask, u64 flags, u64 supported_flags,
5264                               u64 safe_set, u64 safe_clear)
5265 {
5266         const char *type = btrfs_feature_set_names[set];
5267         char *names;
5268         u64 disallowed, unsupported;
5269         u64 set_mask = flags & change_mask;
5270         u64 clear_mask = ~flags & change_mask;
5271
5272         unsupported = set_mask & ~supported_flags;
5273         if (unsupported) {
5274                 names = btrfs_printable_features(set, unsupported);
5275                 if (names) {
5276                         btrfs_warn(fs_info,
5277                                    "this kernel does not support the %s feature bit%s",
5278                                    names, strchr(names, ',') ? "s" : "");
5279                         kfree(names);
5280                 } else
5281                         btrfs_warn(fs_info,
5282                                    "this kernel does not support %s bits 0x%llx",
5283                                    type, unsupported);
5284                 return -EOPNOTSUPP;
5285         }
5286
5287         disallowed = set_mask & ~safe_set;
5288         if (disallowed) {
5289                 names = btrfs_printable_features(set, disallowed);
5290                 if (names) {
5291                         btrfs_warn(fs_info,
5292                                    "can't set the %s feature bit%s while mounted",
5293                                    names, strchr(names, ',') ? "s" : "");
5294                         kfree(names);
5295                 } else
5296                         btrfs_warn(fs_info,
5297                                    "can't set %s bits 0x%llx while mounted",
5298                                    type, disallowed);
5299                 return -EPERM;
5300         }
5301
5302         disallowed = clear_mask & ~safe_clear;
5303         if (disallowed) {
5304                 names = btrfs_printable_features(set, disallowed);
5305                 if (names) {
5306                         btrfs_warn(fs_info,
5307                                    "can't clear the %s feature bit%s while mounted",
5308                                    names, strchr(names, ',') ? "s" : "");
5309                         kfree(names);
5310                 } else
5311                         btrfs_warn(fs_info,
5312                                    "can't clear %s bits 0x%llx while mounted",
5313                                    type, disallowed);
5314                 return -EPERM;
5315         }
5316
5317         return 0;
5318 }
5319
5320 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5321 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5322                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5323                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5324                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5325
5326 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5327 {
5328         struct inode *inode = file_inode(file);
5329         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5330         struct btrfs_root *root = BTRFS_I(inode)->root;
5331         struct btrfs_super_block *super_block = fs_info->super_copy;
5332         struct btrfs_ioctl_feature_flags flags[2];
5333         struct btrfs_trans_handle *trans;
5334         u64 newflags;
5335         int ret;
5336
5337         if (!capable(CAP_SYS_ADMIN))
5338                 return -EPERM;
5339
5340         if (copy_from_user(flags, arg, sizeof(flags)))
5341                 return -EFAULT;
5342
5343         /* Nothing to do */
5344         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5345             !flags[0].incompat_flags)
5346                 return 0;
5347
5348         ret = check_feature(fs_info, flags[0].compat_flags,
5349                             flags[1].compat_flags, COMPAT);
5350         if (ret)
5351                 return ret;
5352
5353         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5354                             flags[1].compat_ro_flags, COMPAT_RO);
5355         if (ret)
5356                 return ret;
5357
5358         ret = check_feature(fs_info, flags[0].incompat_flags,
5359                             flags[1].incompat_flags, INCOMPAT);
5360         if (ret)
5361                 return ret;
5362
5363         ret = mnt_want_write_file(file);
5364         if (ret)
5365                 return ret;
5366
5367         trans = btrfs_start_transaction(root, 0);
5368         if (IS_ERR(trans)) {
5369                 ret = PTR_ERR(trans);
5370                 goto out_drop_write;
5371         }
5372
5373         spin_lock(&fs_info->super_lock);
5374         newflags = btrfs_super_compat_flags(super_block);
5375         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5376         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5377         btrfs_set_super_compat_flags(super_block, newflags);
5378
5379         newflags = btrfs_super_compat_ro_flags(super_block);
5380         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5381         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5382         btrfs_set_super_compat_ro_flags(super_block, newflags);
5383
5384         newflags = btrfs_super_incompat_flags(super_block);
5385         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5386         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5387         btrfs_set_super_incompat_flags(super_block, newflags);
5388         spin_unlock(&fs_info->super_lock);
5389
5390         ret = btrfs_commit_transaction(trans);
5391 out_drop_write:
5392         mnt_drop_write_file(file);
5393
5394         return ret;
5395 }
5396
5397 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5398 {
5399         struct btrfs_ioctl_send_args *arg;
5400         int ret;
5401
5402         if (compat) {
5403 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5404                 struct btrfs_ioctl_send_args_32 args32;
5405
5406                 ret = copy_from_user(&args32, argp, sizeof(args32));
5407                 if (ret)
5408                         return -EFAULT;
5409                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5410                 if (!arg)
5411                         return -ENOMEM;
5412                 arg->send_fd = args32.send_fd;
5413                 arg->clone_sources_count = args32.clone_sources_count;
5414                 arg->clone_sources = compat_ptr(args32.clone_sources);
5415                 arg->parent_root = args32.parent_root;
5416                 arg->flags = args32.flags;
5417                 memcpy(arg->reserved, args32.reserved,
5418                        sizeof(args32.reserved));
5419 #else
5420                 return -ENOTTY;
5421 #endif
5422         } else {
5423                 arg = memdup_user(argp, sizeof(*arg));
5424                 if (IS_ERR(arg))
5425                         return PTR_ERR(arg);
5426         }
5427         ret = btrfs_ioctl_send(file, arg);
5428         kfree(arg);
5429         return ret;
5430 }
5431
5432 long btrfs_ioctl(struct file *file, unsigned int
5433                 cmd, unsigned long arg)
5434 {
5435         struct inode *inode = file_inode(file);
5436         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5437         struct btrfs_root *root = BTRFS_I(inode)->root;
5438         void __user *argp = (void __user *)arg;
5439
5440         switch (cmd) {
5441         case FS_IOC_GETFLAGS:
5442                 return btrfs_ioctl_getflags(file, argp);
5443         case FS_IOC_SETFLAGS:
5444                 return btrfs_ioctl_setflags(file, argp);
5445         case FS_IOC_GETVERSION:
5446                 return btrfs_ioctl_getversion(file, argp);
5447         case FITRIM:
5448                 return btrfs_ioctl_fitrim(file, argp);
5449         case BTRFS_IOC_SNAP_CREATE:
5450                 return btrfs_ioctl_snap_create(file, argp, 0);
5451         case BTRFS_IOC_SNAP_CREATE_V2:
5452                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5453         case BTRFS_IOC_SUBVOL_CREATE:
5454                 return btrfs_ioctl_snap_create(file, argp, 1);
5455         case BTRFS_IOC_SUBVOL_CREATE_V2:
5456                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5457         case BTRFS_IOC_SNAP_DESTROY:
5458                 return btrfs_ioctl_snap_destroy(file, argp);
5459         case BTRFS_IOC_SUBVOL_GETFLAGS:
5460                 return btrfs_ioctl_subvol_getflags(file, argp);
5461         case BTRFS_IOC_SUBVOL_SETFLAGS:
5462                 return btrfs_ioctl_subvol_setflags(file, argp);
5463         case BTRFS_IOC_DEFAULT_SUBVOL:
5464                 return btrfs_ioctl_default_subvol(file, argp);
5465         case BTRFS_IOC_DEFRAG:
5466                 return btrfs_ioctl_defrag(file, NULL);
5467         case BTRFS_IOC_DEFRAG_RANGE:
5468                 return btrfs_ioctl_defrag(file, argp);
5469         case BTRFS_IOC_RESIZE:
5470                 return btrfs_ioctl_resize(file, argp);
5471         case BTRFS_IOC_ADD_DEV:
5472                 return btrfs_ioctl_add_dev(fs_info, argp);
5473         case BTRFS_IOC_RM_DEV:
5474                 return btrfs_ioctl_rm_dev(file, argp);
5475         case BTRFS_IOC_RM_DEV_V2:
5476                 return btrfs_ioctl_rm_dev_v2(file, argp);
5477         case BTRFS_IOC_FS_INFO:
5478                 return btrfs_ioctl_fs_info(fs_info, argp);
5479         case BTRFS_IOC_DEV_INFO:
5480                 return btrfs_ioctl_dev_info(fs_info, argp);
5481         case BTRFS_IOC_BALANCE:
5482                 return btrfs_ioctl_balance(file, NULL);
5483         case BTRFS_IOC_TREE_SEARCH:
5484                 return btrfs_ioctl_tree_search(file, argp);
5485         case BTRFS_IOC_TREE_SEARCH_V2:
5486                 return btrfs_ioctl_tree_search_v2(file, argp);
5487         case BTRFS_IOC_INO_LOOKUP:
5488                 return btrfs_ioctl_ino_lookup(file, argp);
5489         case BTRFS_IOC_INO_PATHS:
5490                 return btrfs_ioctl_ino_to_path(root, argp);
5491         case BTRFS_IOC_LOGICAL_INO:
5492                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5493         case BTRFS_IOC_LOGICAL_INO_V2:
5494                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5495         case BTRFS_IOC_SPACE_INFO:
5496                 return btrfs_ioctl_space_info(fs_info, argp);
5497         case BTRFS_IOC_SYNC: {
5498                 int ret;
5499
5500                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5501                 if (ret)
5502                         return ret;
5503                 ret = btrfs_sync_fs(inode->i_sb, 1);
5504                 /*
5505                  * The transaction thread may want to do more work,
5506                  * namely it pokes the cleaner kthread that will start
5507                  * processing uncleaned subvols.
5508                  */
5509                 wake_up_process(fs_info->transaction_kthread);
5510                 return ret;
5511         }
5512         case BTRFS_IOC_START_SYNC:
5513                 return btrfs_ioctl_start_sync(root, argp);
5514         case BTRFS_IOC_WAIT_SYNC:
5515                 return btrfs_ioctl_wait_sync(fs_info, argp);
5516         case BTRFS_IOC_SCRUB:
5517                 return btrfs_ioctl_scrub(file, argp);
5518         case BTRFS_IOC_SCRUB_CANCEL:
5519                 return btrfs_ioctl_scrub_cancel(fs_info);
5520         case BTRFS_IOC_SCRUB_PROGRESS:
5521                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5522         case BTRFS_IOC_BALANCE_V2:
5523                 return btrfs_ioctl_balance(file, argp);
5524         case BTRFS_IOC_BALANCE_CTL:
5525                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5526         case BTRFS_IOC_BALANCE_PROGRESS:
5527                 return btrfs_ioctl_balance_progress(fs_info, argp);
5528         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5529                 return btrfs_ioctl_set_received_subvol(file, argp);
5530 #ifdef CONFIG_64BIT
5531         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5532                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5533 #endif
5534         case BTRFS_IOC_SEND:
5535                 return _btrfs_ioctl_send(file, argp, false);
5536 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5537         case BTRFS_IOC_SEND_32:
5538                 return _btrfs_ioctl_send(file, argp, true);
5539 #endif
5540         case BTRFS_IOC_GET_DEV_STATS:
5541                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5542         case BTRFS_IOC_QUOTA_CTL:
5543                 return btrfs_ioctl_quota_ctl(file, argp);
5544         case BTRFS_IOC_QGROUP_ASSIGN:
5545                 return btrfs_ioctl_qgroup_assign(file, argp);
5546         case BTRFS_IOC_QGROUP_CREATE:
5547                 return btrfs_ioctl_qgroup_create(file, argp);
5548         case BTRFS_IOC_QGROUP_LIMIT:
5549                 return btrfs_ioctl_qgroup_limit(file, argp);
5550         case BTRFS_IOC_QUOTA_RESCAN:
5551                 return btrfs_ioctl_quota_rescan(file, argp);
5552         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5553                 return btrfs_ioctl_quota_rescan_status(file, argp);
5554         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5555                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5556         case BTRFS_IOC_DEV_REPLACE:
5557                 return btrfs_ioctl_dev_replace(fs_info, argp);
5558         case BTRFS_IOC_GET_FSLABEL:
5559                 return btrfs_ioctl_get_fslabel(file, argp);
5560         case BTRFS_IOC_SET_FSLABEL:
5561                 return btrfs_ioctl_set_fslabel(file, argp);
5562         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5563                 return btrfs_ioctl_get_supported_features(argp);
5564         case BTRFS_IOC_GET_FEATURES:
5565                 return btrfs_ioctl_get_features(file, argp);
5566         case BTRFS_IOC_SET_FEATURES:
5567                 return btrfs_ioctl_set_features(file, argp);
5568         }
5569
5570         return -ENOTTY;
5571 }
5572
5573 #ifdef CONFIG_COMPAT
5574 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5575 {
5576         /*
5577          * These all access 32-bit values anyway so no further
5578          * handling is necessary.
5579          */
5580         switch (cmd) {
5581         case FS_IOC32_GETFLAGS:
5582                 cmd = FS_IOC_GETFLAGS;
5583                 break;
5584         case FS_IOC32_SETFLAGS:
5585                 cmd = FS_IOC_SETFLAGS;
5586                 break;
5587         case FS_IOC32_GETVERSION:
5588                 cmd = FS_IOC_GETVERSION;
5589                 break;
5590         }
5591
5592         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5593 }
5594 #endif