fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  91
  92 /* Mask out flags that are inappropriate for the given type of inode. */
  93 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  94 {
  95         if (S_ISDIR(mode))
  96                 return flags;
  97         else if (S_ISREG(mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 105  */
 106 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 126                 iflags |= FS_COMPR_FL;
 127         else if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_update_iflags(struct inode *inode)
 137 {
 138         struct btrfs_inode *ip = BTRFS_I(inode);
 139
 140         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 141
 142         if (ip->flags & BTRFS_INODE_SYNC)
 143                 inode->i_flags |= S_SYNC;
 144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 145                 inode->i_flags |= S_IMMUTABLE;
 146         if (ip->flags & BTRFS_INODE_APPEND)
 147                 inode->i_flags |= S_APPEND;
 148         if (ip->flags & BTRFS_INODE_NOATIME)
 149                 inode->i_flags |= S_NOATIME;
 150         if (ip->flags & BTRFS_INODE_DIRSYNC)
 151                 inode->i_flags |= S_DIRSYNC;
 152 }
 153
 154 /*
 155  * Inherit flags from the parent inode.
 156  *
 157  * Currently only the compression flags and the cow flags are inherited.
 158  */
 159 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 160 {
 161         unsigned int flags;
 162
 163         if (!dir)
 164                 return;
 165
 166         flags = BTRFS_I(dir)->flags;
 167
 168         if (flags & BTRFS_INODE_NOCOMPRESS) {
 169                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 170                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 171         } else if (flags & BTRFS_INODE_COMPRESS) {
 172                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 173                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 174         }
 175
 176         if (flags & BTRFS_INODE_NODATACOW) {
 177                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 178                 if (S_ISREG(inode->i_mode))
 179                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 180         }
 181
 182         btrfs_update_iflags(inode);
 183 }
 184
 185 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 186 {
 187         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 188         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 189
 190         if (copy_to_user(arg, &flags, sizeof(flags)))
 191                 return -EFAULT;
 192         return 0;
 193 }
 194
 195 static int check_flags(unsigned int flags)
 196 {
 197         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 198                       FS_NOATIME_FL | FS_NODUMP_FL | \
 199                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 200                       FS_NOCOMP_FL | FS_COMPR_FL |
 201                       FS_NOCOW_FL))
 202                 return -EOPNOTSUPP;
 203
 204         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 205                 return -EINVAL;
 206
 207         return 0;
 208 }
 209
 210 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 211 {
 212         struct inode *inode = file_inode(file);
 213         struct btrfs_inode *ip = BTRFS_I(inode);
 214         struct btrfs_root *root = ip->root;
 215         struct btrfs_trans_handle *trans;
 216         unsigned int flags, oldflags;
 217         int ret;
 218         u64 ip_oldflags;
 219         unsigned int i_oldflags;
 220         umode_t mode;
 221
 222         if (!inode_owner_or_capable(inode))
 223                 return -EPERM;
 224
 225         if (btrfs_root_readonly(root))
 226                 return -EROFS;
 227
 228         if (copy_from_user(&flags, arg, sizeof(flags)))
 229                 return -EFAULT;
 230
 231         ret = check_flags(flags);
 232         if (ret)
 233                 return ret;
 234
 235         ret = mnt_want_write_file(file);
 236         if (ret)
 237                 return ret;
 238
 239         mutex_lock(&inode->i_mutex);
 240
 241         ip_oldflags = ip->flags;
 242         i_oldflags = inode->i_flags;
 243         mode = inode->i_mode;
 244
 245         flags = btrfs_mask_flags(inode->i_mode, flags);
 246         oldflags = btrfs_flags_to_ioctl(ip->flags);
 247         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 248                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 249                         ret = -EPERM;
 250                         goto out_unlock;
 251                 }
 252         }
 253
 254         if (flags & FS_SYNC_FL)
 255                 ip->flags |= BTRFS_INODE_SYNC;
 256         else
 257                 ip->flags &= ~BTRFS_INODE_SYNC;
 258         if (flags & FS_IMMUTABLE_FL)
 259                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 262         if (flags & FS_APPEND_FL)
 263                 ip->flags |= BTRFS_INODE_APPEND;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_APPEND;
 266         if (flags & FS_NODUMP_FL)
 267                 ip->flags |= BTRFS_INODE_NODUMP;
 268         else
 269                 ip->flags &= ~BTRFS_INODE_NODUMP;
 270         if (flags & FS_NOATIME_FL)
 271                 ip->flags |= BTRFS_INODE_NOATIME;
 272         else
 273                 ip->flags &= ~BTRFS_INODE_NOATIME;
 274         if (flags & FS_DIRSYNC_FL)
 275                 ip->flags |= BTRFS_INODE_DIRSYNC;
 276         else
 277                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 278         if (flags & FS_NOCOW_FL) {
 279                 if (S_ISREG(mode)) {
 280                         /*
 281                          * It's safe to turn csums off here, no extents exist.
 282                          * Otherwise we want the flag to reflect the real COW
 283                          * status of the file and will not set it.
 284                          */
 285                         if (inode->i_size == 0)
 286                                 ip->flags |= BTRFS_INODE_NODATACOW
 287                                            | BTRFS_INODE_NODATASUM;
 288                 } else {
 289                         ip->flags |= BTRFS_INODE_NODATACOW;
 290                 }
 291         } else {
 292                 /*
 293                  * Revert back under same assuptions as above
 294                  */
 295                 if (S_ISREG(mode)) {
 296                         if (inode->i_size == 0)
 297                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 298                                              | BTRFS_INODE_NODATASUM);
 299                 } else {
 300                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 301                 }
 302         }
 303
 304         /*
 305          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 306          * flag may be changed automatically if compression code won't make
 307          * things smaller.
 308          */
 309         if (flags & FS_NOCOMP_FL) {
 310                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 311                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 312
 313                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 314                 if (ret && ret != -ENODATA)
 315                         goto out_drop;
 316         } else if (flags & FS_COMPR_FL) {
 317                 const char *comp;
 318
 319                 ip->flags |= BTRFS_INODE_COMPRESS;
 320                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 321
 322                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 323                         comp = "lzo";
 324                 else
 325                         comp = "zlib";
 326                 ret = btrfs_set_prop(inode, "btrfs.compression",
 327                                      comp, strlen(comp), 0);
 328                 if (ret)
 329                         goto out_drop;
 330
 331         } else {
 332                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 333         }
 334
 335         trans = btrfs_start_transaction(root, 1);
 336         if (IS_ERR(trans)) {
 337                 ret = PTR_ERR(trans);
 338                 goto out_drop;
 339         }
 340
 341         btrfs_update_iflags(inode);
 342         inode_inc_iversion(inode);
 343         inode->i_ctime = CURRENT_TIME;
 344         ret = btrfs_update_inode(trans, root, inode);
 345
 346         btrfs_end_transaction(trans, root);
 347  out_drop:
 348         if (ret) {
 349                 ip->flags = ip_oldflags;
 350                 inode->i_flags = i_oldflags;
 351         }
 352
 353  out_unlock:
 354         mutex_unlock(&inode->i_mutex);
 355         mnt_drop_write_file(file);
 356         return ret;
 357 }
 358
 359 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 360 {
 361         struct inode *inode = file_inode(file);
 362
 363         return put_user(inode->i_generation, arg);
 364 }
 365
 366 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 367 {
 368         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 369         struct btrfs_device *device;
 370         struct request_queue *q;
 371         struct fstrim_range range;
 372         u64 minlen = ULLONG_MAX;
 373         u64 num_devices = 0;
 374         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 375         int ret;
 376
 377         if (!capable(CAP_SYS_ADMIN))
 378                 return -EPERM;
 379
 380         rcu_read_lock();
 381         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 382                                 dev_list) {
 383                 if (!device->bdev)
 384                         continue;
 385                 q = bdev_get_queue(device->bdev);
 386                 if (blk_queue_discard(q)) {
 387                         num_devices++;
 388                         minlen = min((u64)q->limits.discard_granularity,
 389                                      minlen);
 390                 }
 391         }
 392         rcu_read_unlock();
 393
 394         if (!num_devices)
 395                 return -EOPNOTSUPP;
 396         if (copy_from_user(&range, arg, sizeof(range)))
 397                 return -EFAULT;
 398         if (range.start > total_bytes ||
 399             range.len < fs_info->sb->s_blocksize)
 400                 return -EINVAL;
 401
 402         range.len = min(range.len, total_bytes - range.start);
 403         range.minlen = max(range.minlen, minlen);
 404         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 405         if (ret < 0)
 406                 return ret;
 407
 408         if (copy_to_user(arg, &range, sizeof(range)))
 409                 return -EFAULT;
 410
 411         return 0;
 412 }
 413
 414 int btrfs_is_empty_uuid(u8 *uuid)
 415 {
 416         int i;
 417
 418         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 419                 if (uuid[i])
 420                         return 0;
 421         }
 422         return 1;
 423 }
 424
 425 static noinline int create_subvol(struct inode *dir,
 426                                   struct dentry *dentry,
 427                                   char *name, int namelen,
 428                                   u64 *async_transid,
 429                                   struct btrfs_qgroup_inherit *inherit)
 430 {
 431         struct btrfs_trans_handle *trans;
 432         struct btrfs_key key;
 433         struct btrfs_root_item root_item;
 434         struct btrfs_inode_item *inode_item;
 435         struct extent_buffer *leaf;
 436         struct btrfs_root *root = BTRFS_I(dir)->root;
 437         struct btrfs_root *new_root;
 438         struct btrfs_block_rsv block_rsv;
 439         struct timespec cur_time = CURRENT_TIME;
 440         struct inode *inode;
 441         int ret;
 442         int err;
 443         u64 objectid;
 444         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 445         u64 index = 0;
 446         u64 qgroup_reserved;
 447         uuid_le new_uuid;
 448
 449         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 450         if (ret)
 451                 return ret;
 452
 453         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 454         /*
 455          * The same as the snapshot creation, please see the comment
 456          * of create_snapshot().
 457          */
 458         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 459                                                8, &qgroup_reserved, false);
 460         if (ret)
 461                 return ret;
 462
 463         trans = btrfs_start_transaction(root, 0);
 464         if (IS_ERR(trans)) {
 465                 ret = PTR_ERR(trans);
 466                 btrfs_subvolume_release_metadata(root, &block_rsv,
 467                                                  qgroup_reserved);
 468                 return ret;
 469         }
 470         trans->block_rsv = &block_rsv;
 471         trans->bytes_reserved = block_rsv.size;
 472
 473         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 474         if (ret)
 475                 goto fail;
 476
 477         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 478                                       0, objectid, NULL, 0, 0, 0);
 479         if (IS_ERR(leaf)) {
 480                 ret = PTR_ERR(leaf);
 481                 goto fail;
 482         }
 483
 484         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 485         btrfs_set_header_bytenr(leaf, leaf->start);
 486         btrfs_set_header_generation(leaf, trans->transid);
 487         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 488         btrfs_set_header_owner(leaf, objectid);
 489
 490         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 491                             BTRFS_FSID_SIZE);
 492         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 493                             btrfs_header_chunk_tree_uuid(leaf),
 494                             BTRFS_UUID_SIZE);
 495         btrfs_mark_buffer_dirty(leaf);
 496
 497         memset(&root_item, 0, sizeof(root_item));
 498
 499         inode_item = &root_item.inode;
 500         btrfs_set_stack_inode_generation(inode_item, 1);
 501         btrfs_set_stack_inode_size(inode_item, 3);
 502         btrfs_set_stack_inode_nlink(inode_item, 1);
 503         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 504         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 505
 506         btrfs_set_root_flags(&root_item, 0);
 507         btrfs_set_root_limit(&root_item, 0);
 508         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 509
 510         btrfs_set_root_bytenr(&root_item, leaf->start);
 511         btrfs_set_root_generation(&root_item, trans->transid);
 512         btrfs_set_root_level(&root_item, 0);
 513         btrfs_set_root_refs(&root_item, 1);
 514         btrfs_set_root_used(&root_item, leaf->len);
 515         btrfs_set_root_last_snapshot(&root_item, 0);
 516
 517         btrfs_set_root_generation_v2(&root_item,
 518                         btrfs_root_generation(&root_item));
 519         uuid_le_gen(&new_uuid);
 520         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 521         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 522         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 523         root_item.ctime = root_item.otime;
 524         btrfs_set_root_ctransid(&root_item, trans->transid);
 525         btrfs_set_root_otransid(&root_item, trans->transid);
 526
 527         btrfs_tree_unlock(leaf);
 528         free_extent_buffer(leaf);
 529         leaf = NULL;
 530
 531         btrfs_set_root_dirid(&root_item, new_dirid);
 532
 533         key.objectid = objectid;
 534         key.offset = 0;
 535         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 536         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 537                                 &root_item);
 538         if (ret)
 539                 goto fail;
 540
 541         key.offset = (u64)-1;
 542         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 543         if (IS_ERR(new_root)) {
 544                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 545                 ret = PTR_ERR(new_root);
 546                 goto fail;
 547         }
 548
 549         btrfs_record_root_in_trans(trans, new_root);
 550
 551         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 552         if (ret) {
 553                 /* We potentially lose an unused inode item here */
 554                 btrfs_abort_transaction(trans, root, ret);
 555                 goto fail;
 556         }
 557
 558         /*
 559          * insert the directory item
 560          */
 561         ret = btrfs_set_inode_index(dir, &index);
 562         if (ret) {
 563                 btrfs_abort_transaction(trans, root, ret);
 564                 goto fail;
 565         }
 566
 567         ret = btrfs_insert_dir_item(trans, root,
 568                                     name, namelen, dir, &key,
 569                                     BTRFS_FT_DIR, index);
 570         if (ret) {
 571                 btrfs_abort_transaction(trans, root, ret);
 572                 goto fail;
 573         }
 574
 575         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 576         ret = btrfs_update_inode(trans, root, dir);
 577         BUG_ON(ret);
 578
 579         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 580                                  objectid, root->root_key.objectid,
 581                                  btrfs_ino(dir), index, name, namelen);
 582         BUG_ON(ret);
 583
 584         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 585                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 586                                   objectid);
 587         if (ret)
 588                 btrfs_abort_transaction(trans, root, ret);
 589
 590 fail:
 591         trans->block_rsv = NULL;
 592         trans->bytes_reserved = 0;
 593         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 594
 595         if (async_transid) {
 596                 *async_transid = trans->transid;
 597                 err = btrfs_commit_transaction_async(trans, root, 1);
 598                 if (err)
 599                         err = btrfs_commit_transaction(trans, root);
 600         } else {
 601                 err = btrfs_commit_transaction(trans, root);
 602         }
 603         if (err && !ret)
 604                 ret = err;
 605
 606         if (!ret) {
 607                 inode = btrfs_lookup_dentry(dir, dentry);
 608                 if (IS_ERR(inode))
 609                         return PTR_ERR(inode);
 610                 d_instantiate(dentry, inode);
 611         }
 612         return ret;
 613 }
 614
 615 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 616 {
 617         s64 writers;
 618         DEFINE_WAIT(wait);
 619
 620         do {
 621                 prepare_to_wait(&root->subv_writers->wait, &wait,
 622                                 TASK_UNINTERRUPTIBLE);
 623
 624                 writers = percpu_counter_sum(&root->subv_writers->counter);
 625                 if (writers)
 626                         schedule();
 627
 628                 finish_wait(&root->subv_writers->wait, &wait);
 629         } while (writers);
 630 }
 631
 632 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 633                            struct dentry *dentry, char *name, int namelen,
 634                            u64 *async_transid, bool readonly,
 635                            struct btrfs_qgroup_inherit *inherit)
 636 {
 637         struct inode *inode;
 638         struct btrfs_pending_snapshot *pending_snapshot;
 639         struct btrfs_trans_handle *trans;
 640         int ret;
 641
 642         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 643                 return -EINVAL;
 644
 645         atomic_inc(&root->will_be_snapshoted);
 646         smp_mb__after_atomic_inc();
 647         btrfs_wait_nocow_write(root);
 648
 649         ret = btrfs_start_delalloc_inodes(root, 0);
 650         if (ret)
 651                 goto out;
 652
 653         btrfs_wait_ordered_extents(root, -1);
 654
 655         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 656         if (!pending_snapshot) {
 657                 ret = -ENOMEM;
 658                 goto out;
 659         }
 660
 661         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 662                              BTRFS_BLOCK_RSV_TEMP);
 663         /*
 664          * 1 - parent dir inode
 665          * 2 - dir entries
 666          * 1 - root item
 667          * 2 - root ref/backref
 668          * 1 - root of snapshot
 669          * 1 - UUID item
 670          */
 671         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 672                                         &pending_snapshot->block_rsv, 8,
 673                                         &pending_snapshot->qgroup_reserved,
 674                                         false);
 675         if (ret)
 676                 goto free;
 677
 678         pending_snapshot->dentry = dentry;
 679         pending_snapshot->root = root;
 680         pending_snapshot->readonly = readonly;
 681         pending_snapshot->dir = dir;
 682         pending_snapshot->inherit = inherit;
 683
 684         trans = btrfs_start_transaction(root, 0);
 685         if (IS_ERR(trans)) {
 686                 ret = PTR_ERR(trans);
 687                 goto fail;
 688         }
 689
 690         spin_lock(&root->fs_info->trans_lock);
 691         list_add(&pending_snapshot->list,
 692                  &trans->transaction->pending_snapshots);
 693         spin_unlock(&root->fs_info->trans_lock);
 694         if (async_transid) {
 695                 *async_transid = trans->transid;
 696                 ret = btrfs_commit_transaction_async(trans,
 697                                      root->fs_info->extent_root, 1);
 698                 if (ret)
 699                         ret = btrfs_commit_transaction(trans, root);
 700         } else {
 701                 ret = btrfs_commit_transaction(trans,
 702                                                root->fs_info->extent_root);
 703         }
 704         if (ret)
 705                 goto fail;
 706
 707         ret = pending_snapshot->error;
 708         if (ret)
 709                 goto fail;
 710
 711         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 712         if (ret)
 713                 goto fail;
 714
 715         /*
 716          * If orphan cleanup did remove any orphans, it means the tree was
 717          * modified and therefore the commit root is not the same as the
 718          * current root anymore. This is a problem, because send uses the
 719          * commit root and therefore can see inode items that don't exist
 720          * in the current root anymore, and for example make calls to
 721          * btrfs_iget, which will do tree lookups based on the current root
 722          * and not on the commit root. Those lookups will fail, returning a
 723          * -ESTALE error, and making send fail with that error. So make sure
 724          * a send does not see any orphans we have just removed, and that it
 725          * will see the same inodes regardless of whether a transaction
 726          * commit happened before it started (meaning that the commit root
 727          * will be the same as the current root) or not.
 728          */
 729         if (readonly && pending_snapshot->snap->node !=
 730             pending_snapshot->snap->commit_root) {
 731                 trans = btrfs_join_transaction(pending_snapshot->snap);
 732                 if (IS_ERR(trans) && PTR_ERR(trans) != -ENOENT) {
 733                         ret = PTR_ERR(trans);
 734                         goto fail;
 735                 }
 736                 if (!IS_ERR(trans)) {
 737                         ret = btrfs_commit_transaction(trans,
 738                                                        pending_snapshot->snap);
 739                         if (ret)
 740                                 goto fail;
 741                 }
 742         }
 743
 744         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 745         if (IS_ERR(inode)) {
 746                 ret = PTR_ERR(inode);
 747                 goto fail;
 748         }
 749
 750         d_instantiate(dentry, inode);
 751         ret = 0;
 752 fail:
 753         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 754                                          &pending_snapshot->block_rsv,
 755                                          pending_snapshot->qgroup_reserved);
 756 free:
 757         kfree(pending_snapshot);
 758 out:
 759         atomic_dec(&root->will_be_snapshoted);
 760         return ret;
 761 }
 762
 763 /*  copy of check_sticky in fs/namei.c()
 764 * It's inline, so penalty for filesystems that don't use sticky bit is
 765 * minimal.
 766 */
 767 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 768 {
 769         kuid_t fsuid = current_fsuid();
 770
 771         if (!(dir->i_mode & S_ISVTX))
 772                 return 0;
 773         if (uid_eq(inode->i_uid, fsuid))
 774                 return 0;
 775         if (uid_eq(dir->i_uid, fsuid))
 776                 return 0;
 777         return !capable(CAP_FOWNER);
 778 }
 779
 780 /*  copy of may_delete in fs/namei.c()
 781  *      Check whether we can remove a link victim from directory dir, check
 782  *  whether the type of victim is right.
 783  *  1. We can't do it if dir is read-only (done in permission())
 784  *  2. We should have write and exec permissions on dir
 785  *  3. We can't remove anything from append-only dir
 786  *  4. We can't do anything with immutable dir (done in permission())
 787  *  5. If the sticky bit on dir is set we should either
 788  *      a. be owner of dir, or
 789  *      b. be owner of victim, or
 790  *      c. have CAP_FOWNER capability
 791  *  6. If the victim is append-only or immutable we can't do antyhing with
 792  *     links pointing to it.
 793  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 794  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 795  *  9. We can't remove a root or mountpoint.
 796  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 797  *     nfs_async_unlink().
 798  */
 799
 800 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 801 {
 802         int error;
 803
 804         if (!victim->d_inode)
 805                 return -ENOENT;
 806
 807         BUG_ON(victim->d_parent->d_inode != dir);
 808         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 809
 810         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 811         if (error)
 812                 return error;
 813         if (IS_APPEND(dir))
 814                 return -EPERM;
 815         if (btrfs_check_sticky(dir, victim->d_inode)||
 816                 IS_APPEND(victim->d_inode)||
 817             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 818                 return -EPERM;
 819         if (isdir) {
 820                 if (!S_ISDIR(victim->d_inode->i_mode))
 821                         return -ENOTDIR;
 822                 if (IS_ROOT(victim))
 823                         return -EBUSY;
 824         } else if (S_ISDIR(victim->d_inode->i_mode))
 825                 return -EISDIR;
 826         if (IS_DEADDIR(dir))
 827                 return -ENOENT;
 828         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 829                 return -EBUSY;
 830         return 0;
 831 }
 832
 833 /* copy of may_create in fs/namei.c() */
 834 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 835 {
 836         if (child->d_inode)
 837                 return -EEXIST;
 838         if (IS_DEADDIR(dir))
 839                 return -ENOENT;
 840         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 841 }
 842
 843 /*
 844  * Create a new subvolume below @parent.  This is largely modeled after
 845  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 846  * inside this filesystem so it's quite a bit simpler.
 847  */
 848 static noinline int btrfs_mksubvol(struct path *parent,
 849                                    char *name, int namelen,
 850                                    struct btrfs_root *snap_src,
 851                                    u64 *async_transid, bool readonly,
 852                                    struct btrfs_qgroup_inherit *inherit)
 853 {
 854         struct inode *dir  = parent->dentry->d_inode;
 855         struct dentry *dentry;
 856         int error;
 857
 858         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 859         if (error == -EINTR)
 860                 return error;
 861
 862         dentry = lookup_one_len(name, parent->dentry, namelen);
 863         error = PTR_ERR(dentry);
 864         if (IS_ERR(dentry))
 865                 goto out_unlock;
 866
 867         error = -EEXIST;
 868         if (dentry->d_inode)
 869                 goto out_dput;
 870
 871         error = btrfs_may_create(dir, dentry);
 872         if (error)
 873                 goto out_dput;
 874
 875         /*
 876          * even if this name doesn't exist, we may get hash collisions.
 877          * check for them now when we can safely fail
 878          */
 879         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 880                                                dir->i_ino, name,
 881                                                namelen);
 882         if (error)
 883                 goto out_dput;
 884
 885         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 886
 887         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 888                 goto out_up_read;
 889
 890         if (snap_src) {
 891                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 892                                         async_transid, readonly, inherit);
 893         } else {
 894                 error = create_subvol(dir, dentry, name, namelen,
 895                                       async_transid, inherit);
 896         }
 897         if (!error)
 898                 fsnotify_mkdir(dir, dentry);
 899 out_up_read:
 900         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 901 out_dput:
 902         dput(dentry);
 903 out_unlock:
 904         mutex_unlock(&dir->i_mutex);
 905         return error;
 906 }
 907
 908 /*
 909  * When we're defragging a range, we don't want to kick it off again
 910  * if it is really just waiting for delalloc to send it down.
 911  * If we find a nice big extent or delalloc range for the bytes in the
 912  * file you want to defrag, we return 0 to let you know to skip this
 913  * part of the file
 914  */
 915 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 916 {
 917         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 918         struct extent_map *em = NULL;
 919         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 920         u64 end;
 921
 922         read_lock(&em_tree->lock);
 923         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 924         read_unlock(&em_tree->lock);
 925
 926         if (em) {
 927                 end = extent_map_end(em);
 928                 free_extent_map(em);
 929                 if (end - offset > thresh)
 930                         return 0;
 931         }
 932         /* if we already have a nice delalloc here, just stop */
 933         thresh /= 2;
 934         end = count_range_bits(io_tree, &offset, offset + thresh,
 935                                thresh, EXTENT_DELALLOC, 1);
 936         if (end >= thresh)
 937                 return 0;
 938         return 1;
 939 }
 940
 941 /*
 942  * helper function to walk through a file and find extents
 943  * newer than a specific transid, and smaller than thresh.
 944  *
 945  * This is used by the defragging code to find new and small
 946  * extents
 947  */
 948 static int find_new_extents(struct btrfs_root *root,
 949                             struct inode *inode, u64 newer_than,
 950                             u64 *off, int thresh)
 951 {
 952         struct btrfs_path *path;
 953         struct btrfs_key min_key;
 954         struct extent_buffer *leaf;
 955         struct btrfs_file_extent_item *extent;
 956         int type;
 957         int ret;
 958         u64 ino = btrfs_ino(inode);
 959
 960         path = btrfs_alloc_path();
 961         if (!path)
 962                 return -ENOMEM;
 963
 964         min_key.objectid = ino;
 965         min_key.type = BTRFS_EXTENT_DATA_KEY;
 966         min_key.offset = *off;
 967
 968         while (1) {
 969                 path->keep_locks = 1;
 970                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 971                 if (ret != 0)
 972                         goto none;
 973                 path->keep_locks = 0;
 974                 btrfs_unlock_up_safe(path, 1);
 975 process_slot:
 976                 if (min_key.objectid != ino)
 977                         goto none;
 978                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 979                         goto none;
 980
 981                 leaf = path->nodes[0];
 982                 extent = btrfs_item_ptr(leaf, path->slots[0],
 983                                         struct btrfs_file_extent_item);
 984
 985                 type = btrfs_file_extent_type(leaf, extent);
 986                 if (type == BTRFS_FILE_EXTENT_REG &&
 987                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 988                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 989                         *off = min_key.offset;
 990                         btrfs_free_path(path);
 991                         return 0;
 992                 }
 993
 994                 path->slots[0]++;
 995                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 996                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 997                         goto process_slot;
 998                 }
 999
1000                 if (min_key.offset == (u64)-1)
1001                         goto none;
1002
1003                 min_key.offset++;
1004                 btrfs_release_path(path);
1005         }
1006 none:
1007         btrfs_free_path(path);
1008         return -ENOENT;
1009 }
1010
1011 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1012 {
1013         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1014         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1015         struct extent_map *em;
1016         u64 len = PAGE_CACHE_SIZE;
1017
1018         /*
1019          * hopefully we have this extent in the tree already, try without
1020          * the full extent lock
1021          */
1022         read_lock(&em_tree->lock);
1023         em = lookup_extent_mapping(em_tree, start, len);
1024         read_unlock(&em_tree->lock);
1025
1026         if (!em) {
1027                 struct extent_state *cached = NULL;
1028                 u64 end = start + len - 1;
1029
1030                 /* get the big lock and read metadata off disk */
1031                 lock_extent_bits(io_tree, start, end, 0, &cached);
1032                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1033                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1034
1035                 if (IS_ERR(em))
1036                         return NULL;
1037         }
1038
1039         return em;
1040 }
1041
1042 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1043 {
1044         struct extent_map *next;
1045         bool ret = true;
1046
1047         /* this is the last extent */
1048         if (em->start + em->len >= i_size_read(inode))
1049                 return false;
1050
1051         next = defrag_lookup_extent(inode, em->start + em->len);
1052         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1053             (em->block_start + em->block_len == next->block_start))
1054                 ret = false;
1055
1056         free_extent_map(next);
1057         return ret;
1058 }
1059
1060 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1061                                u64 *last_len, u64 *skip, u64 *defrag_end,
1062                                int compress)
1063 {
1064         struct extent_map *em;
1065         int ret = 1;
1066         bool next_mergeable = true;
1067
1068         /*
1069          * make sure that once we start defragging an extent, we keep on
1070          * defragging it
1071          */
1072         if (start < *defrag_end)
1073                 return 1;
1074
1075         *skip = 0;
1076
1077         em = defrag_lookup_extent(inode, start);
1078         if (!em)
1079                 return 0;
1080
1081         /* this will cover holes, and inline extents */
1082         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1083                 ret = 0;
1084                 goto out;
1085         }
1086
1087         next_mergeable = defrag_check_next_extent(inode, em);
1088
1089         /*
1090          * we hit a real extent, if it is big or the next extent is not a
1091          * real extent, don't bother defragging it
1092          */
1093         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1094             (em->len >= thresh || !next_mergeable))
1095                 ret = 0;
1096 out:
1097         /*
1098          * last_len ends up being a counter of how many bytes we've defragged.
1099          * every time we choose not to defrag an extent, we reset *last_len
1100          * so that the next tiny extent will force a defrag.
1101          *
1102          * The end result of this is that tiny extents before a single big
1103          * extent will force at least part of that big extent to be defragged.
1104          */
1105         if (ret) {
1106                 *defrag_end = extent_map_end(em);
1107         } else {
1108                 *last_len = 0;
1109                 *skip = extent_map_end(em);
1110                 *defrag_end = 0;
1111         }
1112
1113         free_extent_map(em);
1114         return ret;
1115 }
1116
1117 /*
1118  * it doesn't do much good to defrag one or two pages
1119  * at a time.  This pulls in a nice chunk of pages
1120  * to COW and defrag.
1121  *
1122  * It also makes sure the delalloc code has enough
1123  * dirty data to avoid making new small extents as part
1124  * of the defrag
1125  *
1126  * It's a good idea to start RA on this range
1127  * before calling this.
1128  */
1129 static int cluster_pages_for_defrag(struct inode *inode,
1130                                     struct page **pages,
1131                                     unsigned long start_index,
1132                                     unsigned long num_pages)
1133 {
1134         unsigned long file_end;
1135         u64 isize = i_size_read(inode);
1136         u64 page_start;
1137         u64 page_end;
1138         u64 page_cnt;
1139         int ret;
1140         int i;
1141         int i_done;
1142         struct btrfs_ordered_extent *ordered;
1143         struct extent_state *cached_state = NULL;
1144         struct extent_io_tree *tree;
1145         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1146
1147         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1148         if (!isize || start_index > file_end)
1149                 return 0;
1150
1151         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1152
1153         ret = btrfs_delalloc_reserve_space(inode,
1154                                            page_cnt << PAGE_CACHE_SHIFT);
1155         if (ret)
1156                 return ret;
1157         i_done = 0;
1158         tree = &BTRFS_I(inode)->io_tree;
1159
1160         /* step one, lock all the pages */
1161         for (i = 0; i < page_cnt; i++) {
1162                 struct page *page;
1163 again:
1164                 page = find_or_create_page(inode->i_mapping,
1165                                            start_index + i, mask);
1166                 if (!page)
1167                         break;
1168
1169                 page_start = page_offset(page);
1170                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1171                 while (1) {
1172                         lock_extent_bits(tree, page_start, page_end,
1173                                          0, &cached_state);
1174                         ordered = btrfs_lookup_ordered_extent(inode,
1175                                                               page_start);
1176                         unlock_extent_cached(tree, page_start, page_end,
1177                                              &cached_state, GFP_NOFS);
1178                         if (!ordered)
1179                                 break;
1180
1181                         unlock_page(page);
1182                         btrfs_start_ordered_extent(inode, ordered, 1);
1183                         btrfs_put_ordered_extent(ordered);
1184                         lock_page(page);
1185                         /*
1186                          * we unlocked the page above, so we need check if
1187                          * it was released or not.
1188                          */
1189                         if (page->mapping != inode->i_mapping) {
1190                                 unlock_page(page);
1191                                 page_cache_release(page);
1192                                 goto again;
1193                         }
1194                 }
1195
1196                 if (!PageUptodate(page)) {
1197                         btrfs_readpage(NULL, page);
1198                         lock_page(page);
1199                         if (!PageUptodate(page)) {
1200                                 unlock_page(page);
1201                                 page_cache_release(page);
1202                                 ret = -EIO;
1203                                 break;
1204                         }
1205                 }
1206
1207                 if (page->mapping != inode->i_mapping) {
1208                         unlock_page(page);
1209                         page_cache_release(page);
1210                         goto again;
1211                 }
1212
1213                 pages[i] = page;
1214                 i_done++;
1215         }
1216         if (!i_done || ret)
1217                 goto out;
1218
1219         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1220                 goto out;
1221
1222         /*
1223          * so now we have a nice long stream of locked
1224          * and up to date pages, lets wait on them
1225          */
1226         for (i = 0; i < i_done; i++)
1227                 wait_on_page_writeback(pages[i]);
1228
1229         page_start = page_offset(pages[0]);
1230         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1231
1232         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1233                          page_start, page_end - 1, 0, &cached_state);
1234         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1235                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1236                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1237                           &cached_state, GFP_NOFS);
1238
1239         if (i_done != page_cnt) {
1240                 spin_lock(&BTRFS_I(inode)->lock);
1241                 BTRFS_I(inode)->outstanding_extents++;
1242                 spin_unlock(&BTRFS_I(inode)->lock);
1243                 btrfs_delalloc_release_space(inode,
1244                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1245         }
1246
1247
1248         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1249                           &cached_state, GFP_NOFS);
1250
1251         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1252                              page_start, page_end - 1, &cached_state,
1253                              GFP_NOFS);
1254
1255         for (i = 0; i < i_done; i++) {
1256                 clear_page_dirty_for_io(pages[i]);
1257                 ClearPageChecked(pages[i]);
1258                 set_page_extent_mapped(pages[i]);
1259                 set_page_dirty(pages[i]);
1260                 unlock_page(pages[i]);
1261                 page_cache_release(pages[i]);
1262         }
1263         return i_done;
1264 out:
1265         for (i = 0; i < i_done; i++) {
1266                 unlock_page(pages[i]);
1267                 page_cache_release(pages[i]);
1268         }
1269         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1270         return ret;
1271
1272 }
1273
1274 int btrfs_defrag_file(struct inode *inode, struct file *file,
1275                       struct btrfs_ioctl_defrag_range_args *range,
1276                       u64 newer_than, unsigned long max_to_defrag)
1277 {
1278         struct btrfs_root *root = BTRFS_I(inode)->root;
1279         struct file_ra_state *ra = NULL;
1280         unsigned long last_index;
1281         u64 isize = i_size_read(inode);
1282         u64 last_len = 0;
1283         u64 skip = 0;
1284         u64 defrag_end = 0;
1285         u64 newer_off = range->start;
1286         unsigned long i;
1287         unsigned long ra_index = 0;
1288         int ret;
1289         int defrag_count = 0;
1290         int compress_type = BTRFS_COMPRESS_ZLIB;
1291         int extent_thresh = range->extent_thresh;
1292         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1293         unsigned long cluster = max_cluster;
1294         u64 new_align = ~((u64)128 * 1024 - 1);
1295         struct page **pages = NULL;
1296
1297         if (isize == 0)
1298                 return 0;
1299
1300         if (range->start >= isize)
1301                 return -EINVAL;
1302
1303         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1304                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1305                         return -EINVAL;
1306                 if (range->compress_type)
1307                         compress_type = range->compress_type;
1308         }
1309
1310         if (extent_thresh == 0)
1311                 extent_thresh = 256 * 1024;
1312
1313         /*
1314          * if we were not given a file, allocate a readahead
1315          * context
1316          */
1317         if (!file) {
1318                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1319                 if (!ra)
1320                         return -ENOMEM;
1321                 file_ra_state_init(ra, inode->i_mapping);
1322         } else {
1323                 ra = &file->f_ra;
1324         }
1325
1326         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1327                         GFP_NOFS);
1328         if (!pages) {
1329                 ret = -ENOMEM;
1330                 goto out_ra;
1331         }
1332
1333         /* find the last page to defrag */
1334         if (range->start + range->len > range->start) {
1335                 last_index = min_t(u64, isize - 1,
1336                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1337         } else {
1338                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1339         }
1340
1341         if (newer_than) {
1342                 ret = find_new_extents(root, inode, newer_than,
1343                                        &newer_off, 64 * 1024);
1344                 if (!ret) {
1345                         range->start = newer_off;
1346                         /*
1347                          * we always align our defrag to help keep
1348                          * the extents in the file evenly spaced
1349                          */
1350                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1351                 } else
1352                         goto out_ra;
1353         } else {
1354                 i = range->start >> PAGE_CACHE_SHIFT;
1355         }
1356         if (!max_to_defrag)
1357                 max_to_defrag = last_index + 1;
1358
1359         /*
1360          * make writeback starts from i, so the defrag range can be
1361          * written sequentially.
1362          */
1363         if (i < inode->i_mapping->writeback_index)
1364                 inode->i_mapping->writeback_index = i;
1365
1366         while (i <= last_index && defrag_count < max_to_defrag &&
1367                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1368                 PAGE_CACHE_SHIFT)) {
1369                 /*
1370                  * make sure we stop running if someone unmounts
1371                  * the FS
1372                  */
1373                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1374                         break;
1375
1376                 if (btrfs_defrag_cancelled(root->fs_info)) {
1377                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1378                         ret = -EAGAIN;
1379                         break;
1380                 }
1381
1382                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1383                                          extent_thresh, &last_len, &skip,
1384                                          &defrag_end, range->flags &
1385                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1386                         unsigned long next;
1387                         /*
1388                          * the should_defrag function tells us how much to skip
1389                          * bump our counter by the suggested amount
1390                          */
1391                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1392                         i = max(i + 1, next);
1393                         continue;
1394                 }
1395
1396                 if (!newer_than) {
1397                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1398                                    PAGE_CACHE_SHIFT) - i;
1399                         cluster = min(cluster, max_cluster);
1400                 } else {
1401                         cluster = max_cluster;
1402                 }
1403
1404                 if (i + cluster > ra_index) {
1405                         ra_index = max(i, ra_index);
1406                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1407                                        cluster);
1408                         ra_index += max_cluster;
1409                 }
1410
1411                 mutex_lock(&inode->i_mutex);
1412                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1413                         BTRFS_I(inode)->force_compress = compress_type;
1414                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1415                 if (ret < 0) {
1416                         mutex_unlock(&inode->i_mutex);
1417                         goto out_ra;
1418                 }
1419
1420                 defrag_count += ret;
1421                 balance_dirty_pages_ratelimited(inode->i_mapping);
1422                 mutex_unlock(&inode->i_mutex);
1423
1424                 if (newer_than) {
1425                         if (newer_off == (u64)-1)
1426                                 break;
1427
1428                         if (ret > 0)
1429                                 i += ret;
1430
1431                         newer_off = max(newer_off + 1,
1432                                         (u64)i << PAGE_CACHE_SHIFT);
1433
1434                         ret = find_new_extents(root, inode,
1435                                                newer_than, &newer_off,
1436                                                64 * 1024);
1437                         if (!ret) {
1438                                 range->start = newer_off;
1439                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1440                         } else {
1441                                 break;
1442                         }
1443                 } else {
1444                         if (ret > 0) {
1445                                 i += ret;
1446                                 last_len += ret << PAGE_CACHE_SHIFT;
1447                         } else {
1448                                 i++;
1449                                 last_len = 0;
1450                         }
1451                 }
1452         }
1453
1454         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1455                 filemap_flush(inode->i_mapping);
1456                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1457                              &BTRFS_I(inode)->runtime_flags))
1458                         filemap_flush(inode->i_mapping);
1459         }
1460
1461         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1462                 /* the filemap_flush will queue IO into the worker threads, but
1463                  * we have to make sure the IO is actually started and that
1464                  * ordered extents get created before we return
1465                  */
1466                 atomic_inc(&root->fs_info->async_submit_draining);
1467                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1468                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1469                         wait_event(root->fs_info->async_submit_wait,
1470                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1471                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1472                 }
1473                 atomic_dec(&root->fs_info->async_submit_draining);
1474         }
1475
1476         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1477                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1478         }
1479
1480         ret = defrag_count;
1481
1482 out_ra:
1483         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1484                 mutex_lock(&inode->i_mutex);
1485                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1486                 mutex_unlock(&inode->i_mutex);
1487         }
1488         if (!file)
1489                 kfree(ra);
1490         kfree(pages);
1491         return ret;
1492 }
1493
1494 static noinline int btrfs_ioctl_resize(struct file *file,
1495                                         void __user *arg)
1496 {
1497         u64 new_size;
1498         u64 old_size;
1499         u64 devid = 1;
1500         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1501         struct btrfs_ioctl_vol_args *vol_args;
1502         struct btrfs_trans_handle *trans;
1503         struct btrfs_device *device = NULL;
1504         char *sizestr;
1505         char *retptr;
1506         char *devstr = NULL;
1507         int ret = 0;
1508         int mod = 0;
1509
1510         if (!capable(CAP_SYS_ADMIN))
1511                 return -EPERM;
1512
1513         ret = mnt_want_write_file(file);
1514         if (ret)
1515                 return ret;
1516
1517         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1518                         1)) {
1519                 mnt_drop_write_file(file);
1520                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1521         }
1522
1523         mutex_lock(&root->fs_info->volume_mutex);
1524         vol_args = memdup_user(arg, sizeof(*vol_args));
1525         if (IS_ERR(vol_args)) {
1526                 ret = PTR_ERR(vol_args);
1527                 goto out;
1528         }
1529
1530         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1531
1532         sizestr = vol_args->name;
1533         devstr = strchr(sizestr, ':');
1534         if (devstr) {
1535                 sizestr = devstr + 1;
1536                 *devstr = '\0';
1537                 devstr = vol_args->name;
1538                 ret = kstrtoull(devstr, 10, &devid);
1539                 if (ret)
1540                         goto out_free;
1541                 if (!devid) {
1542                         ret = -EINVAL;
1543                         goto out_free;
1544                 }
1545                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1546         }
1547
1548         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1549         if (!device) {
1550                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1551                        devid);
1552                 ret = -ENODEV;
1553                 goto out_free;
1554         }
1555
1556         if (!device->writeable) {
1557                 btrfs_info(root->fs_info,
1558                            "resizer unable to apply on readonly device %llu",
1559                        devid);
1560                 ret = -EPERM;
1561                 goto out_free;
1562         }
1563
1564         if (!strcmp(sizestr, "max"))
1565                 new_size = device->bdev->bd_inode->i_size;
1566         else {
1567                 if (sizestr[0] == '-') {
1568                         mod = -1;
1569                         sizestr++;
1570                 } else if (sizestr[0] == '+') {
1571                         mod = 1;
1572                         sizestr++;
1573                 }
1574                 new_size = memparse(sizestr, &retptr);
1575                 if (*retptr != '\0' || new_size == 0) {
1576                         ret = -EINVAL;
1577                         goto out_free;
1578                 }
1579         }
1580
1581         if (device->is_tgtdev_for_dev_replace) {
1582                 ret = -EPERM;
1583                 goto out_free;
1584         }
1585
1586         old_size = device->total_bytes;
1587
1588         if (mod < 0) {
1589                 if (new_size > old_size) {
1590                         ret = -EINVAL;
1591                         goto out_free;
1592                 }
1593                 new_size = old_size - new_size;
1594         } else if (mod > 0) {
1595                 if (new_size > ULLONG_MAX - old_size) {
1596                         ret = -ERANGE;
1597                         goto out_free;
1598                 }
1599                 new_size = old_size + new_size;
1600         }
1601
1602         if (new_size < 256 * 1024 * 1024) {
1603                 ret = -EINVAL;
1604                 goto out_free;
1605         }
1606         if (new_size > device->bdev->bd_inode->i_size) {
1607                 ret = -EFBIG;
1608                 goto out_free;
1609         }
1610
1611         do_div(new_size, root->sectorsize);
1612         new_size *= root->sectorsize;
1613
1614         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1615                       rcu_str_deref(device->name), new_size);
1616
1617         if (new_size > old_size) {
1618                 trans = btrfs_start_transaction(root, 0);
1619                 if (IS_ERR(trans)) {
1620                         ret = PTR_ERR(trans);
1621                         goto out_free;
1622                 }
1623                 ret = btrfs_grow_device(trans, device, new_size);
1624                 btrfs_commit_transaction(trans, root);
1625         } else if (new_size < old_size) {
1626                 ret = btrfs_shrink_device(device, new_size);
1627         } /* equal, nothing need to do */
1628
1629 out_free:
1630         kfree(vol_args);
1631 out:
1632         mutex_unlock(&root->fs_info->volume_mutex);
1633         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1634         mnt_drop_write_file(file);
1635         return ret;
1636 }
1637
1638 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1639                                 char *name, unsigned long fd, int subvol,
1640                                 u64 *transid, bool readonly,
1641                                 struct btrfs_qgroup_inherit *inherit)
1642 {
1643         int namelen;
1644         int ret = 0;
1645
1646         ret = mnt_want_write_file(file);
1647         if (ret)
1648                 goto out;
1649
1650         namelen = strlen(name);
1651         if (strchr(name, '/')) {
1652                 ret = -EINVAL;
1653                 goto out_drop_write;
1654         }
1655
1656         if (name[0] == '.' &&
1657            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1658                 ret = -EEXIST;
1659                 goto out_drop_write;
1660         }
1661
1662         if (subvol) {
1663                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1664                                      NULL, transid, readonly, inherit);
1665         } else {
1666                 struct fd src = fdget(fd);
1667                 struct inode *src_inode;
1668                 if (!src.file) {
1669                         ret = -EINVAL;
1670                         goto out_drop_write;
1671                 }
1672
1673                 src_inode = file_inode(src.file);
1674                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1675                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1676                                    "Snapshot src from another FS");
1677                         ret = -EXDEV;
1678                 } else if (!inode_owner_or_capable(src_inode)) {
1679                         /*
1680                          * Subvolume creation is not restricted, but snapshots
1681                          * are limited to own subvolumes only
1682                          */
1683                         ret = -EPERM;
1684                 } else {
1685                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1686                                              BTRFS_I(src_inode)->root,
1687                                              transid, readonly, inherit);
1688                 }
1689                 fdput(src);
1690         }
1691 out_drop_write:
1692         mnt_drop_write_file(file);
1693 out:
1694         return ret;
1695 }
1696
1697 static noinline int btrfs_ioctl_snap_create(struct file *file,
1698                                             void __user *arg, int subvol)
1699 {
1700         struct btrfs_ioctl_vol_args *vol_args;
1701         int ret;
1702
1703         vol_args = memdup_user(arg, sizeof(*vol_args));
1704         if (IS_ERR(vol_args))
1705                 return PTR_ERR(vol_args);
1706         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1707
1708         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1709                                               vol_args->fd, subvol,
1710                                               NULL, false, NULL);
1711
1712         kfree(vol_args);
1713         return ret;
1714 }
1715
1716 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1717                                                void __user *arg, int subvol)
1718 {
1719         struct btrfs_ioctl_vol_args_v2 *vol_args;
1720         int ret;
1721         u64 transid = 0;
1722         u64 *ptr = NULL;
1723         bool readonly = false;
1724         struct btrfs_qgroup_inherit *inherit = NULL;
1725
1726         vol_args = memdup_user(arg, sizeof(*vol_args));
1727         if (IS_ERR(vol_args))
1728                 return PTR_ERR(vol_args);
1729         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1730
1731         if (vol_args->flags &
1732             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1733               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1734                 ret = -EOPNOTSUPP;
1735                 goto out;
1736         }
1737
1738         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1739                 ptr = &transid;
1740         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1741                 readonly = true;
1742         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1743                 if (vol_args->size > PAGE_CACHE_SIZE) {
1744                         ret = -EINVAL;
1745                         goto out;
1746                 }
1747                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1748                 if (IS_ERR(inherit)) {
1749                         ret = PTR_ERR(inherit);
1750                         goto out;
1751                 }
1752         }
1753
1754         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1755                                               vol_args->fd, subvol, ptr,
1756                                               readonly, inherit);
1757
1758         if (ret == 0 && ptr &&
1759             copy_to_user(arg +
1760                          offsetof(struct btrfs_ioctl_vol_args_v2,
1761                                   transid), ptr, sizeof(*ptr)))
1762                 ret = -EFAULT;
1763 out:
1764         kfree(vol_args);
1765         kfree(inherit);
1766         return ret;
1767 }
1768
1769 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1770                                                 void __user *arg)
1771 {
1772         struct inode *inode = file_inode(file);
1773         struct btrfs_root *root = BTRFS_I(inode)->root;
1774         int ret = 0;
1775         u64 flags = 0;
1776
1777         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1778                 return -EINVAL;
1779
1780         down_read(&root->fs_info->subvol_sem);
1781         if (btrfs_root_readonly(root))
1782                 flags |= BTRFS_SUBVOL_RDONLY;
1783         up_read(&root->fs_info->subvol_sem);
1784
1785         if (copy_to_user(arg, &flags, sizeof(flags)))
1786                 ret = -EFAULT;
1787
1788         return ret;
1789 }
1790
1791 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1792                                               void __user *arg)
1793 {
1794         struct inode *inode = file_inode(file);
1795         struct btrfs_root *root = BTRFS_I(inode)->root;
1796         struct btrfs_trans_handle *trans;
1797         u64 root_flags;
1798         u64 flags;
1799         int ret = 0;
1800
1801         if (!inode_owner_or_capable(inode))
1802                 return -EPERM;
1803
1804         ret = mnt_want_write_file(file);
1805         if (ret)
1806                 goto out;
1807
1808         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1809                 ret = -EINVAL;
1810                 goto out_drop_write;
1811         }
1812
1813         if (copy_from_user(&flags, arg, sizeof(flags))) {
1814                 ret = -EFAULT;
1815                 goto out_drop_write;
1816         }
1817
1818         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1819                 ret = -EINVAL;
1820                 goto out_drop_write;
1821         }
1822
1823         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1824                 ret = -EOPNOTSUPP;
1825                 goto out_drop_write;
1826         }
1827
1828         down_write(&root->fs_info->subvol_sem);
1829
1830         /* nothing to do */
1831         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1832                 goto out_drop_sem;
1833
1834         root_flags = btrfs_root_flags(&root->root_item);
1835         if (flags & BTRFS_SUBVOL_RDONLY) {
1836                 btrfs_set_root_flags(&root->root_item,
1837                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1838         } else {
1839                 /*
1840                  * Block RO -> RW transition if this subvolume is involved in
1841                  * send
1842                  */
1843                 spin_lock(&root->root_item_lock);
1844                 if (root->send_in_progress == 0) {
1845                         btrfs_set_root_flags(&root->root_item,
1846                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1847                         spin_unlock(&root->root_item_lock);
1848                 } else {
1849                         spin_unlock(&root->root_item_lock);
1850                         btrfs_warn(root->fs_info,
1851                         "Attempt to set subvolume %llu read-write during send",
1852                                         root->root_key.objectid);
1853                         ret = -EPERM;
1854                         goto out_drop_sem;
1855                 }
1856         }
1857
1858         trans = btrfs_start_transaction(root, 1);
1859         if (IS_ERR(trans)) {
1860                 ret = PTR_ERR(trans);
1861                 goto out_reset;
1862         }
1863
1864         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1865                                 &root->root_key, &root->root_item);
1866
1867         btrfs_commit_transaction(trans, root);
1868 out_reset:
1869         if (ret)
1870                 btrfs_set_root_flags(&root->root_item, root_flags);
1871 out_drop_sem:
1872         up_write(&root->fs_info->subvol_sem);
1873 out_drop_write:
1874         mnt_drop_write_file(file);
1875 out:
1876         return ret;
1877 }
1878
1879 /*
1880  * helper to check if the subvolume references other subvolumes
1881  */
1882 static noinline int may_destroy_subvol(struct btrfs_root *root)
1883 {
1884         struct btrfs_path *path;
1885         struct btrfs_dir_item *di;
1886         struct btrfs_key key;
1887         u64 dir_id;
1888         int ret;
1889
1890         path = btrfs_alloc_path();
1891         if (!path)
1892                 return -ENOMEM;
1893
1894         /* Make sure this root isn't set as the default subvol */
1895         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1896         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1897                                    dir_id, "default", 7, 0);
1898         if (di && !IS_ERR(di)) {
1899                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1900                 if (key.objectid == root->root_key.objectid) {
1901                         ret = -EPERM;
1902                         btrfs_err(root->fs_info, "deleting default subvolume "
1903                                   "%llu is not allowed", key.objectid);
1904                         goto out;
1905                 }
1906                 btrfs_release_path(path);
1907         }
1908
1909         key.objectid = root->root_key.objectid;
1910         key.type = BTRFS_ROOT_REF_KEY;
1911         key.offset = (u64)-1;
1912
1913         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1914                                 &key, path, 0, 0);
1915         if (ret < 0)
1916                 goto out;
1917         BUG_ON(ret == 0);
1918
1919         ret = 0;
1920         if (path->slots[0] > 0) {
1921                 path->slots[0]--;
1922                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1923                 if (key.objectid == root->root_key.objectid &&
1924                     key.type == BTRFS_ROOT_REF_KEY)
1925                         ret = -ENOTEMPTY;
1926         }
1927 out:
1928         btrfs_free_path(path);
1929         return ret;
1930 }
1931
1932 static noinline int key_in_sk(struct btrfs_key *key,
1933                               struct btrfs_ioctl_search_key *sk)
1934 {
1935         struct btrfs_key test;
1936         int ret;
1937
1938         test.objectid = sk->min_objectid;
1939         test.type = sk->min_type;
1940         test.offset = sk->min_offset;
1941
1942         ret = btrfs_comp_cpu_keys(key, &test);
1943         if (ret < 0)
1944                 return 0;
1945
1946         test.objectid = sk->max_objectid;
1947         test.type = sk->max_type;
1948         test.offset = sk->max_offset;
1949
1950         ret = btrfs_comp_cpu_keys(key, &test);
1951         if (ret > 0)
1952                 return 0;
1953         return 1;
1954 }
1955
1956 static noinline int copy_to_sk(struct btrfs_root *root,
1957                                struct btrfs_path *path,
1958                                struct btrfs_key *key,
1959                                struct btrfs_ioctl_search_key *sk,
1960                                char *buf,
1961                                unsigned long *sk_offset,
1962                                int *num_found)
1963 {
1964         u64 found_transid;
1965         struct extent_buffer *leaf;
1966         struct btrfs_ioctl_search_header sh;
1967         unsigned long item_off;
1968         unsigned long item_len;
1969         int nritems;
1970         int i;
1971         int slot;
1972         int ret = 0;
1973
1974         leaf = path->nodes[0];
1975         slot = path->slots[0];
1976         nritems = btrfs_header_nritems(leaf);
1977
1978         if (btrfs_header_generation(leaf) > sk->max_transid) {
1979                 i = nritems;
1980                 goto advance_key;
1981         }
1982         found_transid = btrfs_header_generation(leaf);
1983
1984         for (i = slot; i < nritems; i++) {
1985                 item_off = btrfs_item_ptr_offset(leaf, i);
1986                 item_len = btrfs_item_size_nr(leaf, i);
1987
1988                 btrfs_item_key_to_cpu(leaf, key, i);
1989                 if (!key_in_sk(key, sk))
1990                         continue;
1991
1992                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1993                         item_len = 0;
1994
1995                 if (sizeof(sh) + item_len + *sk_offset >
1996                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1997                         ret = 1;
1998                         goto overflow;
1999                 }
2000
2001                 sh.objectid = key->objectid;
2002                 sh.offset = key->offset;
2003                 sh.type = key->type;
2004                 sh.len = item_len;
2005                 sh.transid = found_transid;
2006
2007                 /* copy search result header */
2008                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
2009                 *sk_offset += sizeof(sh);
2010
2011                 if (item_len) {
2012                         char *p = buf + *sk_offset;
2013                         /* copy the item */
2014                         read_extent_buffer(leaf, p,
2015                                            item_off, item_len);
2016                         *sk_offset += item_len;
2017                 }
2018                 (*num_found)++;
2019
2020                 if (*num_found >= sk->nr_items)
2021                         break;
2022         }
2023 advance_key:
2024         ret = 0;
2025         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
2026                 key->offset++;
2027         else if (key->type < (u8)-1 && key->type < sk->max_type) {
2028                 key->offset = 0;
2029                 key->type++;
2030         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2031                 key->offset = 0;
2032                 key->type = 0;
2033                 key->objectid++;
2034         } else
2035                 ret = 1;
2036 overflow:
2037         return ret;
2038 }
2039
2040 static noinline int search_ioctl(struct inode *inode,
2041                                  struct btrfs_ioctl_search_args *args)
2042 {
2043         struct btrfs_root *root;
2044         struct btrfs_key key;
2045         struct btrfs_path *path;
2046         struct btrfs_ioctl_search_key *sk = &args->key;
2047         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2048         int ret;
2049         int num_found = 0;
2050         unsigned long sk_offset = 0;
2051
2052         path = btrfs_alloc_path();
2053         if (!path)
2054                 return -ENOMEM;
2055
2056         if (sk->tree_id == 0) {
2057                 /* search the root of the inode that was passed */
2058                 root = BTRFS_I(inode)->root;
2059         } else {
2060                 key.objectid = sk->tree_id;
2061                 key.type = BTRFS_ROOT_ITEM_KEY;
2062                 key.offset = (u64)-1;
2063                 root = btrfs_read_fs_root_no_name(info, &key);
2064                 if (IS_ERR(root)) {
2065                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2066                                sk->tree_id);
2067                         btrfs_free_path(path);
2068                         return -ENOENT;
2069                 }
2070         }
2071
2072         key.objectid = sk->min_objectid;
2073         key.type = sk->min_type;
2074         key.offset = sk->min_offset;
2075
2076         path->keep_locks = 1;
2077
2078         while (1) {
2079                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2080                 if (ret != 0) {
2081                         if (ret > 0)
2082                                 ret = 0;
2083                         goto err;
2084                 }
2085                 ret = copy_to_sk(root, path, &key, sk, args->buf,
2086                                  &sk_offset, &num_found);
2087                 btrfs_release_path(path);
2088                 if (ret || num_found >= sk->nr_items)
2089                         break;
2090
2091         }
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 *args;
2103          struct inode *inode;
2104          int ret;
2105
2106         if (!capable(CAP_SYS_ADMIN))
2107                 return -EPERM;
2108
2109         args = memdup_user(argp, sizeof(*args));
2110         if (IS_ERR(args))
2111                 return PTR_ERR(args);
2112
2113         inode = file_inode(file);
2114         ret = search_ioctl(inode, args);
2115         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2116                 ret = -EFAULT;
2117         kfree(args);
2118         return ret;
2119 }
2120
2121 /*
2122  * Search INODE_REFs to identify path name of 'dirid' directory
2123  * in a 'tree_id' tree. and sets path name to 'name'.
2124  */
2125 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2126                                 u64 tree_id, u64 dirid, char *name)
2127 {
2128         struct btrfs_root *root;
2129         struct btrfs_key key;
2130         char *ptr;
2131         int ret = -1;
2132         int slot;
2133         int len;
2134         int total_len = 0;
2135         struct btrfs_inode_ref *iref;
2136         struct extent_buffer *l;
2137         struct btrfs_path *path;
2138
2139         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2140                 name[0]='\0';
2141                 return 0;
2142         }
2143
2144         path = btrfs_alloc_path();
2145         if (!path)
2146                 return -ENOMEM;
2147
2148         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2149
2150         key.objectid = tree_id;
2151         key.type = BTRFS_ROOT_ITEM_KEY;
2152         key.offset = (u64)-1;
2153         root = btrfs_read_fs_root_no_name(info, &key);
2154         if (IS_ERR(root)) {
2155                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2156                 ret = -ENOENT;
2157                 goto out;
2158         }
2159
2160         key.objectid = dirid;
2161         key.type = BTRFS_INODE_REF_KEY;
2162         key.offset = (u64)-1;
2163
2164         while (1) {
2165                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2166                 if (ret < 0)
2167                         goto out;
2168                 else if (ret > 0) {
2169                         ret = btrfs_previous_item(root, path, dirid,
2170                                                   BTRFS_INODE_REF_KEY);
2171                         if (ret < 0)
2172                                 goto out;
2173                         else if (ret > 0) {
2174                                 ret = -ENOENT;
2175                                 goto out;
2176                         }
2177                 }
2178
2179                 l = path->nodes[0];
2180                 slot = path->slots[0];
2181                 btrfs_item_key_to_cpu(l, &key, slot);
2182
2183                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2184                 len = btrfs_inode_ref_name_len(l, iref);
2185                 ptr -= len + 1;
2186                 total_len += len + 1;
2187                 if (ptr < name) {
2188                         ret = -ENAMETOOLONG;
2189                         goto out;
2190                 }
2191
2192                 *(ptr + len) = '/';
2193                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2194
2195                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2196                         break;
2197
2198                 btrfs_release_path(path);
2199                 key.objectid = key.offset;
2200                 key.offset = (u64)-1;
2201                 dirid = key.objectid;
2202         }
2203         memmove(name, ptr, total_len);
2204         name[total_len] = '\0';
2205         ret = 0;
2206 out:
2207         btrfs_free_path(path);
2208         return ret;
2209 }
2210
2211 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2212                                            void __user *argp)
2213 {
2214          struct btrfs_ioctl_ino_lookup_args *args;
2215          struct inode *inode;
2216          int ret;
2217
2218         if (!capable(CAP_SYS_ADMIN))
2219                 return -EPERM;
2220
2221         args = memdup_user(argp, sizeof(*args));
2222         if (IS_ERR(args))
2223                 return PTR_ERR(args);
2224
2225         inode = file_inode(file);
2226
2227         if (args->treeid == 0)
2228                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2229
2230         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2231                                         args->treeid, args->objectid,
2232                                         args->name);
2233
2234         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2235                 ret = -EFAULT;
2236
2237         kfree(args);
2238         return ret;
2239 }
2240
2241 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2242                                              void __user *arg)
2243 {
2244         struct dentry *parent = file->f_path.dentry;
2245         struct dentry *dentry;
2246         struct inode *dir = parent->d_inode;
2247         struct inode *inode;
2248         struct btrfs_root *root = BTRFS_I(dir)->root;
2249         struct btrfs_root *dest = NULL;
2250         struct btrfs_ioctl_vol_args *vol_args;
2251         struct btrfs_trans_handle *trans;
2252         struct btrfs_block_rsv block_rsv;
2253         u64 root_flags;
2254         u64 qgroup_reserved;
2255         int namelen;
2256         int ret;
2257         int err = 0;
2258
2259         vol_args = memdup_user(arg, sizeof(*vol_args));
2260         if (IS_ERR(vol_args))
2261                 return PTR_ERR(vol_args);
2262
2263         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2264         namelen = strlen(vol_args->name);
2265         if (strchr(vol_args->name, '/') ||
2266             strncmp(vol_args->name, "..", namelen) == 0) {
2267                 err = -EINVAL;
2268                 goto out;
2269         }
2270
2271         err = mnt_want_write_file(file);
2272         if (err)
2273                 goto out;
2274
2275
2276         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2277         if (err == -EINTR)
2278                 goto out_drop_write;
2279         dentry = lookup_one_len(vol_args->name, parent, namelen);
2280         if (IS_ERR(dentry)) {
2281                 err = PTR_ERR(dentry);
2282                 goto out_unlock_dir;
2283         }
2284
2285         if (!dentry->d_inode) {
2286                 err = -ENOENT;
2287                 goto out_dput;
2288         }
2289
2290         inode = dentry->d_inode;
2291         dest = BTRFS_I(inode)->root;
2292         if (!capable(CAP_SYS_ADMIN)) {
2293                 /*
2294                  * Regular user.  Only allow this with a special mount
2295                  * option, when the user has write+exec access to the
2296                  * subvol root, and when rmdir(2) would have been
2297                  * allowed.
2298                  *
2299                  * Note that this is _not_ check that the subvol is
2300                  * empty or doesn't contain data that we wouldn't
2301                  * otherwise be able to delete.
2302                  *
2303                  * Users who want to delete empty subvols should try
2304                  * rmdir(2).
2305                  */
2306                 err = -EPERM;
2307                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2308                         goto out_dput;
2309
2310                 /*
2311                  * Do not allow deletion if the parent dir is the same
2312                  * as the dir to be deleted.  That means the ioctl
2313                  * must be called on the dentry referencing the root
2314                  * of the subvol, not a random directory contained
2315                  * within it.
2316                  */
2317                 err = -EINVAL;
2318                 if (root == dest)
2319                         goto out_dput;
2320
2321                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2322                 if (err)
2323                         goto out_dput;
2324         }
2325
2326         /* check if subvolume may be deleted by a user */
2327         err = btrfs_may_delete(dir, dentry, 1);
2328         if (err)
2329                 goto out_dput;
2330
2331         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2332                 err = -EINVAL;
2333                 goto out_dput;
2334         }
2335
2336         mutex_lock(&inode->i_mutex);
2337
2338         /*
2339          * Don't allow to delete a subvolume with send in progress. This is
2340          * inside the i_mutex so the error handling that has to drop the bit
2341          * again is not run concurrently.
2342          */
2343         spin_lock(&dest->root_item_lock);
2344         root_flags = btrfs_root_flags(&dest->root_item);
2345         if (dest->send_in_progress == 0) {
2346                 btrfs_set_root_flags(&dest->root_item,
2347                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2348                 spin_unlock(&dest->root_item_lock);
2349         } else {
2350                 spin_unlock(&dest->root_item_lock);
2351                 btrfs_warn(root->fs_info,
2352                         "Attempt to delete subvolume %llu during send",
2353                         dest->root_key.objectid);
2354                 err = -EPERM;
2355                 goto out_dput;
2356         }
2357
2358         err = d_invalidate(dentry);
2359         if (err)
2360                 goto out_unlock;
2361
2362         down_write(&root->fs_info->subvol_sem);
2363
2364         err = may_destroy_subvol(dest);
2365         if (err)
2366                 goto out_up_write;
2367
2368         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2369         /*
2370          * One for dir inode, two for dir entries, two for root
2371          * ref/backref.
2372          */
2373         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2374                                                5, &qgroup_reserved, true);
2375         if (err)
2376                 goto out_up_write;
2377
2378         trans = btrfs_start_transaction(root, 0);
2379         if (IS_ERR(trans)) {
2380                 err = PTR_ERR(trans);
2381                 goto out_release;
2382         }
2383         trans->block_rsv = &block_rsv;
2384         trans->bytes_reserved = block_rsv.size;
2385
2386         ret = btrfs_unlink_subvol(trans, root, dir,
2387                                 dest->root_key.objectid,
2388                                 dentry->d_name.name,
2389                                 dentry->d_name.len);
2390         if (ret) {
2391                 err = ret;
2392                 btrfs_abort_transaction(trans, root, ret);
2393                 goto out_end_trans;
2394         }
2395
2396         btrfs_record_root_in_trans(trans, dest);
2397
2398         memset(&dest->root_item.drop_progress, 0,
2399                 sizeof(dest->root_item.drop_progress));
2400         dest->root_item.drop_level = 0;
2401         btrfs_set_root_refs(&dest->root_item, 0);
2402
2403         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2404                 ret = btrfs_insert_orphan_item(trans,
2405                                         root->fs_info->tree_root,
2406                                         dest->root_key.objectid);
2407                 if (ret) {
2408                         btrfs_abort_transaction(trans, root, ret);
2409                         err = ret;
2410                         goto out_end_trans;
2411                 }
2412         }
2413
2414         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2415                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2416                                   dest->root_key.objectid);
2417         if (ret && ret != -ENOENT) {
2418                 btrfs_abort_transaction(trans, root, ret);
2419                 err = ret;
2420                 goto out_end_trans;
2421         }
2422         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2423                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2424                                           dest->root_item.received_uuid,
2425                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2426                                           dest->root_key.objectid);
2427                 if (ret && ret != -ENOENT) {
2428                         btrfs_abort_transaction(trans, root, ret);
2429                         err = ret;
2430                         goto out_end_trans;
2431                 }
2432         }
2433
2434 out_end_trans:
2435         trans->block_rsv = NULL;
2436         trans->bytes_reserved = 0;
2437         ret = btrfs_end_transaction(trans, root);
2438         if (ret && !err)
2439                 err = ret;
2440         inode->i_flags |= S_DEAD;
2441 out_release:
2442         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2443 out_up_write:
2444         up_write(&root->fs_info->subvol_sem);
2445 out_unlock:
2446         if (err) {
2447                 spin_lock(&dest->root_item_lock);
2448                 root_flags = btrfs_root_flags(&dest->root_item);
2449                 btrfs_set_root_flags(&dest->root_item,
2450                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2451                 spin_unlock(&dest->root_item_lock);
2452         }
2453         mutex_unlock(&inode->i_mutex);
2454         if (!err) {
2455                 shrink_dcache_sb(root->fs_info->sb);
2456                 btrfs_invalidate_inodes(dest);
2457                 d_delete(dentry);
2458                 ASSERT(dest->send_in_progress == 0);
2459
2460                 /* the last ref */
2461                 if (dest->cache_inode) {
2462                         iput(dest->cache_inode);
2463                         dest->cache_inode = NULL;
2464                 }
2465         }
2466 out_dput:
2467         dput(dentry);
2468 out_unlock_dir:
2469         mutex_unlock(&dir->i_mutex);
2470 out_drop_write:
2471         mnt_drop_write_file(file);
2472 out:
2473         kfree(vol_args);
2474         return err;
2475 }
2476
2477 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2478 {
2479         struct inode *inode = file_inode(file);
2480         struct btrfs_root *root = BTRFS_I(inode)->root;
2481         struct btrfs_ioctl_defrag_range_args *range;
2482         int ret;
2483
2484         ret = mnt_want_write_file(file);
2485         if (ret)
2486                 return ret;
2487
2488         if (btrfs_root_readonly(root)) {
2489                 ret = -EROFS;
2490                 goto out;
2491         }
2492
2493         switch (inode->i_mode & S_IFMT) {
2494         case S_IFDIR:
2495                 if (!capable(CAP_SYS_ADMIN)) {
2496                         ret = -EPERM;
2497                         goto out;
2498                 }
2499                 ret = btrfs_defrag_root(root);
2500                 if (ret)
2501                         goto out;
2502                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2503                 break;
2504         case S_IFREG:
2505                 if (!(file->f_mode & FMODE_WRITE)) {
2506                         ret = -EINVAL;
2507                         goto out;
2508                 }
2509
2510                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2511                 if (!range) {
2512                         ret = -ENOMEM;
2513                         goto out;
2514                 }
2515
2516                 if (argp) {
2517                         if (copy_from_user(range, argp,
2518                                            sizeof(*range))) {
2519                                 ret = -EFAULT;
2520                                 kfree(range);
2521                                 goto out;
2522                         }
2523                         /* compression requires us to start the IO */
2524                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2525                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2526                                 range->extent_thresh = (u32)-1;
2527                         }
2528                 } else {
2529                         /* the rest are all set to zero by kzalloc */
2530                         range->len = (u64)-1;
2531                 }
2532                 ret = btrfs_defrag_file(file_inode(file), file,
2533                                         range, 0, 0);
2534                 if (ret > 0)
2535                         ret = 0;
2536                 kfree(range);
2537                 break;
2538         default:
2539                 ret = -EINVAL;
2540         }
2541 out:
2542         mnt_drop_write_file(file);
2543         return ret;
2544 }
2545
2546 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2547 {
2548         struct btrfs_ioctl_vol_args *vol_args;
2549         int ret;
2550
2551         if (!capable(CAP_SYS_ADMIN))
2552                 return -EPERM;
2553
2554         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2555                         1)) {
2556                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2557         }
2558
2559         mutex_lock(&root->fs_info->volume_mutex);
2560         vol_args = memdup_user(arg, sizeof(*vol_args));
2561         if (IS_ERR(vol_args)) {
2562                 ret = PTR_ERR(vol_args);
2563                 goto out;
2564         }
2565
2566         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2567         ret = btrfs_init_new_device(root, vol_args->name);
2568
2569         kfree(vol_args);
2570 out:
2571         mutex_unlock(&root->fs_info->volume_mutex);
2572         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2573         return ret;
2574 }
2575
2576 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2577 {
2578         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2579         struct btrfs_ioctl_vol_args *vol_args;
2580         int ret;
2581
2582         if (!capable(CAP_SYS_ADMIN))
2583                 return -EPERM;
2584
2585         ret = mnt_want_write_file(file);
2586         if (ret)
2587                 return ret;
2588
2589         vol_args = memdup_user(arg, sizeof(*vol_args));
2590         if (IS_ERR(vol_args)) {
2591                 ret = PTR_ERR(vol_args);
2592                 goto out;
2593         }
2594
2595         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2596
2597         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2598                         1)) {
2599                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2600                 goto out;
2601         }
2602
2603         mutex_lock(&root->fs_info->volume_mutex);
2604         ret = btrfs_rm_device(root, vol_args->name);
2605         mutex_unlock(&root->fs_info->volume_mutex);
2606         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2607
2608 out:
2609         kfree(vol_args);
2610         mnt_drop_write_file(file);
2611         return ret;
2612 }
2613
2614 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2615 {
2616         struct btrfs_ioctl_fs_info_args *fi_args;
2617         struct btrfs_device *device;
2618         struct btrfs_device *next;
2619         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2620         int ret = 0;
2621
2622         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2623         if (!fi_args)
2624                 return -ENOMEM;
2625
2626         mutex_lock(&fs_devices->device_list_mutex);
2627         fi_args->num_devices = fs_devices->num_devices;
2628         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2629
2630         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2631                 if (device->devid > fi_args->max_id)
2632                         fi_args->max_id = device->devid;
2633         }
2634         mutex_unlock(&fs_devices->device_list_mutex);
2635
2636         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2637         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2638         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2639
2640         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2641                 ret = -EFAULT;
2642
2643         kfree(fi_args);
2644         return ret;
2645 }
2646
2647 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2648 {
2649         struct btrfs_ioctl_dev_info_args *di_args;
2650         struct btrfs_device *dev;
2651         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2652         int ret = 0;
2653         char *s_uuid = NULL;
2654
2655         di_args = memdup_user(arg, sizeof(*di_args));
2656         if (IS_ERR(di_args))
2657                 return PTR_ERR(di_args);
2658
2659         if (!btrfs_is_empty_uuid(di_args->uuid))
2660                 s_uuid = di_args->uuid;
2661
2662         mutex_lock(&fs_devices->device_list_mutex);
2663         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2664
2665         if (!dev) {
2666                 ret = -ENODEV;
2667                 goto out;
2668         }
2669
2670         di_args->devid = dev->devid;
2671         di_args->bytes_used = dev->bytes_used;
2672         di_args->total_bytes = dev->total_bytes;
2673         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2674         if (dev->name) {
2675                 struct rcu_string *name;
2676
2677                 rcu_read_lock();
2678                 name = rcu_dereference(dev->name);
2679                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2680                 rcu_read_unlock();
2681                 di_args->path[sizeof(di_args->path) - 1] = 0;
2682         } else {
2683                 di_args->path[0] = '\0';
2684         }
2685
2686 out:
2687         mutex_unlock(&fs_devices->device_list_mutex);
2688         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2689                 ret = -EFAULT;
2690
2691         kfree(di_args);
2692         return ret;
2693 }
2694
2695 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2696 {
2697         struct page *page;
2698         pgoff_t index;
2699         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2700
2701         index = off >> PAGE_CACHE_SHIFT;
2702
2703         page = grab_cache_page(inode->i_mapping, index);
2704         if (!page)
2705                 return NULL;
2706
2707         if (!PageUptodate(page)) {
2708                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2709                                                  0))
2710                         return NULL;
2711                 lock_page(page);
2712                 if (!PageUptodate(page)) {
2713                         unlock_page(page);
2714                         page_cache_release(page);
2715                         return NULL;
2716                 }
2717         }
2718         unlock_page(page);
2719
2720         return page;
2721 }
2722
2723 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2724 {
2725         /* do any pending delalloc/csum calc on src, one way or
2726            another, and lock file content */
2727         while (1) {
2728                 struct btrfs_ordered_extent *ordered;
2729                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2730                 ordered = btrfs_lookup_first_ordered_extent(inode,
2731                                                             off + len - 1);
2732                 if ((!ordered ||
2733                      ordered->file_offset + ordered->len <= off ||
2734                      ordered->file_offset >= off + len) &&
2735                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2736                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2737                         if (ordered)
2738                                 btrfs_put_ordered_extent(ordered);
2739                         break;
2740                 }
2741                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2742                 if (ordered)
2743                         btrfs_put_ordered_extent(ordered);
2744                 btrfs_wait_ordered_range(inode, off, len);
2745         }
2746 }
2747
2748 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2749                                 struct inode *inode2, u64 loff2, u64 len)
2750 {
2751         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2752         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2753
2754         mutex_unlock(&inode1->i_mutex);
2755         mutex_unlock(&inode2->i_mutex);
2756 }
2757
2758 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2759                               struct inode *inode2, u64 loff2, u64 len)
2760 {
2761         if (inode1 < inode2) {
2762                 swap(inode1, inode2);
2763                 swap(loff1, loff2);
2764         }
2765
2766         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2767         lock_extent_range(inode1, loff1, len);
2768         if (inode1 != inode2) {
2769                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2770                 lock_extent_range(inode2, loff2, len);
2771         }
2772 }
2773
2774 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2775                           u64 dst_loff, u64 len)
2776 {
2777         int ret = 0;
2778         struct page *src_page, *dst_page;
2779         unsigned int cmp_len = PAGE_CACHE_SIZE;
2780         void *addr, *dst_addr;
2781
2782         while (len) {
2783                 if (len < PAGE_CACHE_SIZE)
2784                         cmp_len = len;
2785
2786                 src_page = extent_same_get_page(src, loff);
2787                 if (!src_page)
2788                         return -EINVAL;
2789                 dst_page = extent_same_get_page(dst, dst_loff);
2790                 if (!dst_page) {
2791                         page_cache_release(src_page);
2792                         return -EINVAL;
2793                 }
2794                 addr = kmap_atomic(src_page);
2795                 dst_addr = kmap_atomic(dst_page);
2796
2797                 flush_dcache_page(src_page);
2798                 flush_dcache_page(dst_page);
2799
2800                 if (memcmp(addr, dst_addr, cmp_len))
2801                         ret = BTRFS_SAME_DATA_DIFFERS;
2802
2803                 kunmap_atomic(addr);
2804                 kunmap_atomic(dst_addr);
2805                 page_cache_release(src_page);
2806                 page_cache_release(dst_page);
2807
2808                 if (ret)
2809                         break;
2810
2811                 loff += cmp_len;
2812                 dst_loff += cmp_len;
2813                 len -= cmp_len;
2814         }
2815
2816         return ret;
2817 }
2818
2819 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2820 {
2821         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2822
2823         if (off + len > inode->i_size || off + len < off)
2824                 return -EINVAL;
2825         /* Check that we are block aligned - btrfs_clone() requires this */
2826         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2827                 return -EINVAL;
2828
2829         return 0;
2830 }
2831
2832 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2833                              struct inode *dst, u64 dst_loff)
2834 {
2835         int ret;
2836
2837         /*
2838          * btrfs_clone() can't handle extents in the same file
2839          * yet. Once that works, we can drop this check and replace it
2840          * with a check for the same inode, but overlapping extents.
2841          */
2842         if (src == dst)
2843                 return -EINVAL;
2844
2845         btrfs_double_lock(src, loff, dst, dst_loff, len);
2846
2847         ret = extent_same_check_offsets(src, loff, len);
2848         if (ret)
2849                 goto out_unlock;
2850
2851         ret = extent_same_check_offsets(dst, dst_loff, len);
2852         if (ret)
2853                 goto out_unlock;
2854
2855         /* don't make the dst file partly checksummed */
2856         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2857             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2858                 ret = -EINVAL;
2859                 goto out_unlock;
2860         }
2861
2862         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2863         if (ret == 0)
2864                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2865
2866 out_unlock:
2867         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2868
2869         return ret;
2870 }
2871
2872 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2873
2874 static long btrfs_ioctl_file_extent_same(struct file *file,
2875                         struct btrfs_ioctl_same_args __user *argp)
2876 {
2877         struct btrfs_ioctl_same_args *same;
2878         struct btrfs_ioctl_same_extent_info *info;
2879         struct inode *src = file_inode(file);
2880         u64 off;
2881         u64 len;
2882         int i;
2883         int ret;
2884         unsigned long size;
2885         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2886         bool is_admin = capable(CAP_SYS_ADMIN);
2887         u16 count;
2888
2889         if (!(file->f_mode & FMODE_READ))
2890                 return -EINVAL;
2891
2892         ret = mnt_want_write_file(file);
2893         if (ret)
2894                 return ret;
2895
2896         if (get_user(count, &argp->dest_count)) {
2897                 ret = -EFAULT;
2898                 goto out;
2899         }
2900
2901         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2902
2903         same = memdup_user(argp, size);
2904
2905         if (IS_ERR(same)) {
2906                 ret = PTR_ERR(same);
2907                 goto out;
2908         }
2909
2910         off = same->logical_offset;
2911         len = same->length;
2912
2913         /*
2914          * Limit the total length we will dedupe for each operation.
2915          * This is intended to bound the total time spent in this
2916          * ioctl to something sane.
2917          */
2918         if (len > BTRFS_MAX_DEDUPE_LEN)
2919                 len = BTRFS_MAX_DEDUPE_LEN;
2920
2921         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2922                 /*
2923                  * Btrfs does not support blocksize < page_size. As a
2924                  * result, btrfs_cmp_data() won't correctly handle
2925                  * this situation without an update.
2926                  */
2927                 ret = -EINVAL;
2928                 goto out;
2929         }
2930
2931         ret = -EISDIR;
2932         if (S_ISDIR(src->i_mode))
2933                 goto out;
2934
2935         ret = -EACCES;
2936         if (!S_ISREG(src->i_mode))
2937                 goto out;
2938
2939         /* pre-format output fields to sane values */
2940         for (i = 0; i < count; i++) {
2941                 same->info[i].bytes_deduped = 0ULL;
2942                 same->info[i].status = 0;
2943         }
2944
2945         for (i = 0, info = same->info; i < count; i++, info++) {
2946                 struct inode *dst;
2947                 struct fd dst_file = fdget(info->fd);
2948                 if (!dst_file.file) {
2949                         info->status = -EBADF;
2950                         continue;
2951                 }
2952                 dst = file_inode(dst_file.file);
2953
2954                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
2955                         info->status = -EINVAL;
2956                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
2957                         info->status = -EXDEV;
2958                 } else if (S_ISDIR(dst->i_mode)) {
2959                         info->status = -EISDIR;
2960                 } else if (!S_ISREG(dst->i_mode)) {
2961                         info->status = -EACCES;
2962                 } else {
2963                         info->status = btrfs_extent_same(src, off, len, dst,
2964                                                         info->logical_offset);
2965                         if (info->status == 0)
2966                                 info->bytes_deduped += len;
2967                 }
2968                 fdput(dst_file);
2969         }
2970
2971         ret = copy_to_user(argp, same, size);
2972         if (ret)
2973                 ret = -EFAULT;
2974
2975 out:
2976         mnt_drop_write_file(file);
2977         return ret;
2978 }
2979
2980 /* Helper to check and see if this root currently has a ref on the given disk
2981  * bytenr.  If it does then we need to update the quota for this root.  This
2982  * doesn't do anything if quotas aren't enabled.
2983  */
2984 static int check_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2985                      u64 disko)
2986 {
2987         struct seq_list tree_mod_seq_elem = {};
2988         struct ulist *roots;
2989         struct ulist_iterator uiter;
2990         struct ulist_node *root_node = NULL;
2991         int ret;
2992
2993         if (!root->fs_info->quota_enabled)
2994                 return 1;
2995
2996         btrfs_get_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
2997         ret = btrfs_find_all_roots(trans, root->fs_info, disko,
2998                                    tree_mod_seq_elem.seq, &roots);
2999         if (ret < 0)
3000                 goto out;
3001         ret = 0;
3002         ULIST_ITER_INIT(&uiter);
3003         while ((root_node = ulist_next(roots, &uiter))) {
3004                 if (root_node->val == root->objectid) {
3005                         ret = 1;
3006                         break;
3007                 }
3008         }
3009         ulist_free(roots);
3010 out:
3011         btrfs_put_tree_mod_seq(root->fs_info, &tree_mod_seq_elem);
3012         return ret;
3013 }
3014
3015 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3016                                      struct inode *inode,
3017                                      u64 endoff,
3018                                      const u64 destoff,
3019                                      const u64 olen)
3020 {
3021         struct btrfs_root *root = BTRFS_I(inode)->root;
3022         int ret;
3023
3024         inode_inc_iversion(inode);
3025         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3026         /*
3027          * We round up to the block size at eof when determining which
3028          * extents to clone above, but shouldn't round up the file size.
3029          */
3030         if (endoff > destoff + olen)
3031                 endoff = destoff + olen;
3032         if (endoff > inode->i_size)
3033                 btrfs_i_size_write(inode, endoff);
3034
3035         ret = btrfs_update_inode(trans, root, inode);
3036         if (ret) {
3037                 btrfs_abort_transaction(trans, root, ret);
3038                 btrfs_end_transaction(trans, root);
3039                 goto out;
3040         }
3041         ret = btrfs_end_transaction(trans, root);
3042 out:
3043         return ret;
3044 }
3045
3046 /**
3047  * btrfs_clone() - clone a range from inode file to another
3048  *
3049  * @src: Inode to clone from
3050  * @inode: Inode to clone to
3051  * @off: Offset within source to start clone from
3052  * @olen: Original length, passed by user, of range to clone
3053  * @olen_aligned: Block-aligned value of olen, extent_same uses
3054  *               identical values here
3055  * @destoff: Offset within @inode to start clone
3056  */
3057 static int btrfs_clone(struct inode *src, struct inode *inode,
3058                        const u64 off, const u64 olen, const u64 olen_aligned,
3059                        const u64 destoff)
3060 {
3061         struct btrfs_root *root = BTRFS_I(inode)->root;
3062         struct btrfs_path *path = NULL;
3063         struct extent_buffer *leaf;
3064         struct btrfs_trans_handle *trans;
3065         char *buf = NULL;
3066         struct btrfs_key key;
3067         u32 nritems;
3068         int slot;
3069         int ret;
3070         int no_quota;
3071         const u64 len = olen_aligned;
3072         u64 last_disko = 0;
3073         u64 last_dest_end = destoff;
3074
3075         ret = -ENOMEM;
3076         buf = vmalloc(btrfs_level_size(root, 0));
3077         if (!buf)
3078                 return ret;
3079
3080         path = btrfs_alloc_path();
3081         if (!path) {
3082                 vfree(buf);
3083                 return ret;
3084         }
3085
3086         path->reada = 2;
3087         /* clone data */
3088         key.objectid = btrfs_ino(src);
3089         key.type = BTRFS_EXTENT_DATA_KEY;
3090         key.offset = off;
3091
3092         while (1) {
3093                 /*
3094                  * note the key will change type as we walk through the
3095                  * tree.
3096                  */
3097                 path->leave_spinning = 1;
3098                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3099                                 0, 0);
3100                 if (ret < 0)
3101                         goto out;
3102                 /*
3103                  * First search, if no extent item that starts at offset off was
3104                  * found but the previous item is an extent item, it's possible
3105                  * it might overlap our target range, therefore process it.
3106                  */
3107                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3108                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3109                                               path->slots[0] - 1);
3110                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3111                                 path->slots[0]--;
3112                 }
3113
3114                 nritems = btrfs_header_nritems(path->nodes[0]);
3115 process_slot:
3116                 no_quota = 1;
3117                 if (path->slots[0] >= nritems) {
3118                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3119                         if (ret < 0)
3120                                 goto out;
3121                         if (ret > 0)
3122                                 break;
3123                         nritems = btrfs_header_nritems(path->nodes[0]);
3124                 }
3125                 leaf = path->nodes[0];
3126                 slot = path->slots[0];
3127
3128                 btrfs_item_key_to_cpu(leaf, &key, slot);
3129                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
3130                     key.objectid != btrfs_ino(src))
3131                         break;
3132
3133                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3134                         struct btrfs_file_extent_item *extent;
3135                         int type;
3136                         u32 size;
3137                         struct btrfs_key new_key;
3138                         u64 disko = 0, diskl = 0;
3139                         u64 datao = 0, datal = 0;
3140                         u8 comp;
3141                         u64 drop_start;
3142
3143                         extent = btrfs_item_ptr(leaf, slot,
3144                                                 struct btrfs_file_extent_item);
3145                         comp = btrfs_file_extent_compression(leaf, extent);
3146                         type = btrfs_file_extent_type(leaf, extent);
3147                         if (type == BTRFS_FILE_EXTENT_REG ||
3148                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3149                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3150                                                                       extent);
3151                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3152                                                                  extent);
3153                                 datao = btrfs_file_extent_offset(leaf, extent);
3154                                 datal = btrfs_file_extent_num_bytes(leaf,
3155                                                                     extent);
3156                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3157                                 /* take upper bound, may be compressed */
3158                                 datal = btrfs_file_extent_ram_bytes(leaf,
3159                                                                     extent);
3160                         }
3161
3162                         /*
3163                          * The first search might have left us at an extent
3164                          * item that ends before our target range's start, can
3165                          * happen if we have holes and NO_HOLES feature enabled.
3166                          */
3167                         if (key.offset + datal <= off) {
3168                                 path->slots[0]++;
3169                                 goto process_slot;
3170                         } else if (key.offset >= off + len) {
3171                                 break;
3172                         }
3173
3174                         size = btrfs_item_size_nr(leaf, slot);
3175                         read_extent_buffer(leaf, buf,
3176                                            btrfs_item_ptr_offset(leaf, slot),
3177                                            size);
3178
3179                         btrfs_release_path(path);
3180                         path->leave_spinning = 0;
3181
3182                         memcpy(&new_key, &key, sizeof(new_key));
3183                         new_key.objectid = btrfs_ino(inode);
3184                         if (off <= key.offset)
3185                                 new_key.offset = key.offset + destoff - off;
3186                         else
3187                                 new_key.offset = destoff;
3188
3189                         /*
3190                          * Deal with a hole that doesn't have an extent item
3191                          * that represents it (NO_HOLES feature enabled).
3192                          * This hole is either in the middle of the cloning
3193                          * range or at the beginning (fully overlaps it or
3194                          * partially overlaps it).
3195                          */
3196                         if (new_key.offset != last_dest_end)
3197                                 drop_start = last_dest_end;
3198                         else
3199                                 drop_start = new_key.offset;
3200
3201                         /*
3202                          * 1 - adjusting old extent (we may have to split it)
3203                          * 1 - add new extent
3204                          * 1 - inode update
3205                          */
3206                         trans = btrfs_start_transaction(root, 3);
3207                         if (IS_ERR(trans)) {
3208                                 ret = PTR_ERR(trans);
3209                                 goto out;
3210                         }
3211
3212                         if (type == BTRFS_FILE_EXTENT_REG ||
3213                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3214                                 /*
3215                                  *    a  | --- range to clone ---|  b
3216                                  * | ------------- extent ------------- |
3217                                  */
3218
3219                                 /* substract range b */
3220                                 if (key.offset + datal > off + len)
3221                                         datal = off + len - key.offset;
3222
3223                                 /* substract range a */
3224                                 if (off > key.offset) {
3225                                         datao += off - key.offset;
3226                                         datal -= off - key.offset;
3227                                 }
3228
3229                                 ret = btrfs_drop_extents(trans, root, inode,
3230                                                          drop_start,
3231                                                          new_key.offset + datal,
3232                                                          1);
3233                                 if (ret) {
3234                                         if (ret != -EOPNOTSUPP)
3235                                                 btrfs_abort_transaction(trans,
3236                                                                 root, ret);
3237                                         btrfs_end_transaction(trans, root);
3238                                         goto out;
3239                                 }
3240
3241                                 ret = btrfs_insert_empty_item(trans, root, path,
3242                                                               &new_key, size);
3243                                 if (ret) {
3244                                         btrfs_abort_transaction(trans, root,
3245                                                                 ret);
3246                                         btrfs_end_transaction(trans, root);
3247                                         goto out;
3248                                 }
3249
3250                                 leaf = path->nodes[0];
3251                                 slot = path->slots[0];
3252                                 write_extent_buffer(leaf, buf,
3253                                             btrfs_item_ptr_offset(leaf, slot),
3254                                             size);
3255
3256                                 extent = btrfs_item_ptr(leaf, slot,
3257                                                 struct btrfs_file_extent_item);
3258
3259                                 /* disko == 0 means it's a hole */
3260                                 if (!disko)
3261                                         datao = 0;
3262
3263                                 btrfs_set_file_extent_offset(leaf, extent,
3264                                                              datao);
3265                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3266                                                                 datal);
3267
3268                                 /*
3269                                  * We need to look up the roots that point at
3270                                  * this bytenr and see if the new root does.  If
3271                                  * it does not we need to make sure we update
3272                                  * quotas appropriately.
3273                                  */
3274                                 if (disko && root != BTRFS_I(src)->root &&
3275                                     disko != last_disko) {
3276                                         no_quota = check_ref(trans, root,
3277                                                              disko);
3278                                         if (no_quota < 0) {
3279                                                 btrfs_abort_transaction(trans,
3280                                                                         root,
3281                                                                         ret);
3282                                                 btrfs_end_transaction(trans,
3283                                                                       root);
3284                                                 ret = no_quota;
3285                                                 goto out;
3286                                         }
3287                                 }
3288
3289                                 if (disko) {
3290                                         inode_add_bytes(inode, datal);
3291                                         ret = btrfs_inc_extent_ref(trans, root,
3292                                                         disko, diskl, 0,
3293                                                         root->root_key.objectid,
3294                                                         btrfs_ino(inode),
3295                                                         new_key.offset - datao,
3296                                                         no_quota);
3297                                         if (ret) {
3298                                                 btrfs_abort_transaction(trans,
3299                                                                         root,
3300                                                                         ret);
3301                                                 btrfs_end_transaction(trans,
3302                                                                       root);
3303                                                 goto out;
3304
3305                                         }
3306                                 }
3307                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3308                                 u64 skip = 0;
3309                                 u64 trim = 0;
3310                                 u64 aligned_end = 0;
3311
3312                                 if (off > key.offset) {
3313                                         skip = off - key.offset;
3314                                         new_key.offset += skip;
3315                                 }
3316
3317                                 if (key.offset + datal > off + len)
3318                                         trim = key.offset + datal - (off + len);
3319
3320                                 if (comp && (skip || trim)) {
3321                                         ret = -EINVAL;
3322                                         btrfs_end_transaction(trans, root);
3323                                         goto out;
3324                                 }
3325                                 size -= skip + trim;
3326                                 datal -= skip + trim;
3327
3328                                 aligned_end = ALIGN(new_key.offset + datal,
3329                                                     root->sectorsize);
3330                                 ret = btrfs_drop_extents(trans, root, inode,
3331                                                          drop_start,
3332                                                          aligned_end,
3333                                                          1);
3334                                 if (ret) {
3335                                         if (ret != -EOPNOTSUPP)
3336                                                 btrfs_abort_transaction(trans,
3337                                                         root, ret);
3338                                         btrfs_end_transaction(trans, root);
3339                                         goto out;
3340                                 }
3341
3342                                 ret = btrfs_insert_empty_item(trans, root, path,
3343                                                               &new_key, size);
3344                                 if (ret) {
3345                                         btrfs_abort_transaction(trans, root,
3346                                                                 ret);
3347                                         btrfs_end_transaction(trans, root);
3348                                         goto out;
3349                                 }
3350
3351                                 if (skip) {
3352                                         u32 start =
3353                                           btrfs_file_extent_calc_inline_size(0);
3354                                         memmove(buf+start, buf+start+skip,
3355                                                 datal);
3356                                 }
3357
3358                                 leaf = path->nodes[0];
3359                                 slot = path->slots[0];
3360                                 write_extent_buffer(leaf, buf,
3361                                             btrfs_item_ptr_offset(leaf, slot),
3362                                             size);
3363                                 inode_add_bytes(inode, datal);
3364                         }
3365
3366                         btrfs_mark_buffer_dirty(leaf);
3367                         btrfs_release_path(path);
3368
3369                         last_dest_end = new_key.offset + datal;
3370                         ret = clone_finish_inode_update(trans, inode,
3371                                                         last_dest_end,
3372                                                         destoff, olen);
3373                         if (ret)
3374                                 goto out;
3375                         if (new_key.offset + datal >= destoff + len)
3376                                 break;
3377                 }
3378                 btrfs_release_path(path);
3379                 key.offset++;
3380         }
3381         ret = 0;
3382
3383         if (last_dest_end < destoff + len) {
3384                 /*
3385                  * We have an implicit hole (NO_HOLES feature is enabled) that
3386                  * fully or partially overlaps our cloning range at its end.
3387                  */
3388                 btrfs_release_path(path);
3389
3390                 /*
3391                  * 1 - remove extent(s)
3392                  * 1 - inode update
3393                  */
3394                 trans = btrfs_start_transaction(root, 2);
3395                 if (IS_ERR(trans)) {
3396                         ret = PTR_ERR(trans);
3397                         goto out;
3398                 }
3399                 ret = btrfs_drop_extents(trans, root, inode,
3400                                          last_dest_end, destoff + len, 1);
3401                 if (ret) {
3402                         if (ret != -EOPNOTSUPP)
3403                                 btrfs_abort_transaction(trans, root, ret);
3404                         btrfs_end_transaction(trans, root);
3405                         goto out;
3406                 }
3407                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3408                                                 destoff, olen);
3409         }
3410
3411 out:
3412         btrfs_free_path(path);
3413         vfree(buf);
3414         return ret;
3415 }
3416
3417 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3418                                        u64 off, u64 olen, u64 destoff)
3419 {
3420         struct inode *inode = file_inode(file);
3421         struct btrfs_root *root = BTRFS_I(inode)->root;
3422         struct fd src_file;
3423         struct inode *src;
3424         int ret;
3425         u64 len = olen;
3426         u64 bs = root->fs_info->sb->s_blocksize;
3427         int same_inode = 0;
3428
3429         /*
3430          * TODO:
3431          * - split compressed inline extents.  annoying: we need to
3432          *   decompress into destination's address_space (the file offset
3433          *   may change, so source mapping won't do), then recompress (or
3434          *   otherwise reinsert) a subrange.
3435          *
3436          * - split destination inode's inline extents.  The inline extents can
3437          *   be either compressed or non-compressed.
3438          */
3439
3440         /* the destination must be opened for writing */
3441         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3442                 return -EINVAL;
3443
3444         if (btrfs_root_readonly(root))
3445                 return -EROFS;
3446
3447         ret = mnt_want_write_file(file);
3448         if (ret)
3449                 return ret;
3450
3451         src_file = fdget(srcfd);
3452         if (!src_file.file) {
3453                 ret = -EBADF;
3454                 goto out_drop_write;
3455         }
3456
3457         ret = -EXDEV;
3458         if (src_file.file->f_path.mnt != file->f_path.mnt)
3459                 goto out_fput;
3460
3461         src = file_inode(src_file.file);
3462
3463         ret = -EINVAL;
3464         if (src == inode)
3465                 same_inode = 1;
3466
3467         /* the src must be open for reading */
3468         if (!(src_file.file->f_mode & FMODE_READ))
3469                 goto out_fput;
3470
3471         /* don't make the dst file partly checksummed */
3472         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3473             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3474                 goto out_fput;
3475
3476         ret = -EISDIR;
3477         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3478                 goto out_fput;
3479
3480         ret = -EXDEV;
3481         if (src->i_sb != inode->i_sb)
3482                 goto out_fput;
3483
3484         if (!same_inode) {
3485                 if (inode < src) {
3486                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3487                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3488                 } else {
3489                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3490                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3491                 }
3492         } else {
3493                 mutex_lock(&src->i_mutex);
3494         }
3495
3496         /* determine range to clone */
3497         ret = -EINVAL;
3498         if (off + len > src->i_size || off + len < off)
3499                 goto out_unlock;
3500         if (len == 0)
3501                 olen = len = src->i_size - off;
3502         /* if we extend to eof, continue to block boundary */
3503         if (off + len == src->i_size)
3504                 len = ALIGN(src->i_size, bs) - off;
3505
3506         /* verify the end result is block aligned */
3507         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3508             !IS_ALIGNED(destoff, bs))
3509                 goto out_unlock;
3510
3511         /* verify if ranges are overlapped within the same file */
3512         if (same_inode) {
3513                 if (destoff + len > off && destoff < off + len)
3514                         goto out_unlock;
3515         }
3516
3517         if (destoff > inode->i_size) {
3518                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3519                 if (ret)
3520                         goto out_unlock;
3521         }
3522
3523         /*
3524          * Lock the target range too. Right after we replace the file extent
3525          * items in the fs tree (which now point to the cloned data), we might
3526          * have a worker replace them with extent items relative to a write
3527          * operation that was issued before this clone operation (i.e. confront
3528          * with inode.c:btrfs_finish_ordered_io).
3529          */
3530         if (same_inode) {
3531                 u64 lock_start = min_t(u64, off, destoff);
3532                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3533
3534                 lock_extent_range(src, lock_start, lock_len);
3535         } else {
3536                 lock_extent_range(src, off, len);
3537                 lock_extent_range(inode, destoff, len);
3538         }
3539
3540         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3541
3542         if (same_inode) {
3543                 u64 lock_start = min_t(u64, off, destoff);
3544                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3545
3546                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3547         } else {
3548                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3549                 unlock_extent(&BTRFS_I(inode)->io_tree, destoff,
3550                               destoff + len - 1);
3551         }
3552         /*
3553          * Truncate page cache pages so that future reads will see the cloned
3554          * data immediately and not the previous data.
3555          */
3556         truncate_inode_pages_range(&inode->i_data, destoff,
3557                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3558 out_unlock:
3559         if (!same_inode) {
3560                 if (inode < src) {
3561                         mutex_unlock(&src->i_mutex);
3562                         mutex_unlock(&inode->i_mutex);
3563                 } else {
3564                         mutex_unlock(&inode->i_mutex);
3565                         mutex_unlock(&src->i_mutex);
3566                 }
3567         } else {
3568                 mutex_unlock(&src->i_mutex);
3569         }
3570 out_fput:
3571         fdput(src_file);
3572 out_drop_write:
3573         mnt_drop_write_file(file);
3574         return ret;
3575 }
3576
3577 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3578 {
3579         struct btrfs_ioctl_clone_range_args args;
3580
3581         if (copy_from_user(&args, argp, sizeof(args)))
3582                 return -EFAULT;
3583         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3584                                  args.src_length, args.dest_offset);
3585 }
3586
3587 /*
3588  * there are many ways the trans_start and trans_end ioctls can lead
3589  * to deadlocks.  They should only be used by applications that
3590  * basically own the machine, and have a very in depth understanding
3591  * of all the possible deadlocks and enospc problems.
3592  */
3593 static long btrfs_ioctl_trans_start(struct file *file)
3594 {
3595         struct inode *inode = file_inode(file);
3596         struct btrfs_root *root = BTRFS_I(inode)->root;
3597         struct btrfs_trans_handle *trans;
3598         int ret;
3599
3600         ret = -EPERM;
3601         if (!capable(CAP_SYS_ADMIN))
3602                 goto out;
3603
3604         ret = -EINPROGRESS;
3605         if (file->private_data)
3606                 goto out;
3607
3608         ret = -EROFS;
3609         if (btrfs_root_readonly(root))
3610                 goto out;
3611
3612         ret = mnt_want_write_file(file);
3613         if (ret)
3614                 goto out;
3615
3616         atomic_inc(&root->fs_info->open_ioctl_trans);
3617
3618         ret = -ENOMEM;
3619         trans = btrfs_start_ioctl_transaction(root);
3620         if (IS_ERR(trans))
3621                 goto out_drop;
3622
3623         file->private_data = trans;
3624         return 0;
3625
3626 out_drop:
3627         atomic_dec(&root->fs_info->open_ioctl_trans);
3628         mnt_drop_write_file(file);
3629 out:
3630         return ret;
3631 }
3632
3633 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3634 {
3635         struct inode *inode = file_inode(file);
3636         struct btrfs_root *root = BTRFS_I(inode)->root;
3637         struct btrfs_root *new_root;
3638         struct btrfs_dir_item *di;
3639         struct btrfs_trans_handle *trans;
3640         struct btrfs_path *path;
3641         struct btrfs_key location;
3642         struct btrfs_disk_key disk_key;
3643         u64 objectid = 0;
3644         u64 dir_id;
3645         int ret;
3646
3647         if (!capable(CAP_SYS_ADMIN))
3648                 return -EPERM;
3649
3650         ret = mnt_want_write_file(file);
3651         if (ret)
3652                 return ret;
3653
3654         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3655                 ret = -EFAULT;
3656                 goto out;
3657         }
3658
3659         if (!objectid)
3660                 objectid = BTRFS_FS_TREE_OBJECTID;
3661
3662         location.objectid = objectid;
3663         location.type = BTRFS_ROOT_ITEM_KEY;
3664         location.offset = (u64)-1;
3665
3666         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3667         if (IS_ERR(new_root)) {
3668                 ret = PTR_ERR(new_root);
3669                 goto out;
3670         }
3671
3672         path = btrfs_alloc_path();
3673         if (!path) {
3674                 ret = -ENOMEM;
3675                 goto out;
3676         }
3677         path->leave_spinning = 1;
3678
3679         trans = btrfs_start_transaction(root, 1);
3680         if (IS_ERR(trans)) {
3681                 btrfs_free_path(path);
3682                 ret = PTR_ERR(trans);
3683                 goto out;
3684         }
3685
3686         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3687         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3688                                    dir_id, "default", 7, 1);
3689         if (IS_ERR_OR_NULL(di)) {
3690                 btrfs_free_path(path);
3691                 btrfs_end_transaction(trans, root);
3692                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3693                            "item, this isn't going to work");
3694                 ret = -ENOENT;
3695                 goto out;
3696         }
3697
3698         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3699         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3700         btrfs_mark_buffer_dirty(path->nodes[0]);
3701         btrfs_free_path(path);
3702
3703         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3704         btrfs_end_transaction(trans, root);
3705 out:
3706         mnt_drop_write_file(file);
3707         return ret;
3708 }
3709
3710 void btrfs_get_block_group_info(struct list_head *groups_list,
3711                                 struct btrfs_ioctl_space_info *space)
3712 {
3713         struct btrfs_block_group_cache *block_group;
3714
3715         space->total_bytes = 0;
3716         space->used_bytes = 0;
3717         space->flags = 0;
3718         list_for_each_entry(block_group, groups_list, list) {
3719                 space->flags = block_group->flags;
3720                 space->total_bytes += block_group->key.offset;
3721                 space->used_bytes +=
3722                         btrfs_block_group_used(&block_group->item);
3723         }
3724 }
3725
3726 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3727 {
3728         struct btrfs_ioctl_space_args space_args;
3729         struct btrfs_ioctl_space_info space;
3730         struct btrfs_ioctl_space_info *dest;
3731         struct btrfs_ioctl_space_info *dest_orig;
3732         struct btrfs_ioctl_space_info __user *user_dest;
3733         struct btrfs_space_info *info;
3734         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3735                        BTRFS_BLOCK_GROUP_SYSTEM,
3736                        BTRFS_BLOCK_GROUP_METADATA,
3737                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3738         int num_types = 4;
3739         int alloc_size;
3740         int ret = 0;
3741         u64 slot_count = 0;
3742         int i, c;
3743
3744         if (copy_from_user(&space_args,
3745                            (struct btrfs_ioctl_space_args __user *)arg,
3746                            sizeof(space_args)))
3747                 return -EFAULT;
3748
3749         for (i = 0; i < num_types; i++) {
3750                 struct btrfs_space_info *tmp;
3751
3752                 info = NULL;
3753                 rcu_read_lock();
3754                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3755                                         list) {
3756                         if (tmp->flags == types[i]) {
3757                                 info = tmp;
3758                                 break;
3759                         }
3760                 }
3761                 rcu_read_unlock();
3762
3763                 if (!info)
3764                         continue;
3765
3766                 down_read(&info->groups_sem);
3767                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3768                         if (!list_empty(&info->block_groups[c]))
3769                                 slot_count++;
3770                 }
3771                 up_read(&info->groups_sem);
3772         }
3773
3774         /*
3775          * Global block reserve, exported as a space_info
3776          */
3777         slot_count++;
3778
3779         /* space_slots == 0 means they are asking for a count */
3780         if (space_args.space_slots == 0) {
3781                 space_args.total_spaces = slot_count;
3782                 goto out;
3783         }
3784
3785         slot_count = min_t(u64, space_args.space_slots, slot_count);
3786
3787         alloc_size = sizeof(*dest) * slot_count;
3788
3789         /* we generally have at most 6 or so space infos, one for each raid
3790          * level.  So, a whole page should be more than enough for everyone
3791          */
3792         if (alloc_size > PAGE_CACHE_SIZE)
3793                 return -ENOMEM;
3794
3795         space_args.total_spaces = 0;
3796         dest = kmalloc(alloc_size, GFP_NOFS);
3797         if (!dest)
3798                 return -ENOMEM;
3799         dest_orig = dest;
3800
3801         /* now we have a buffer to copy into */
3802         for (i = 0; i < num_types; i++) {
3803                 struct btrfs_space_info *tmp;
3804
3805                 if (!slot_count)
3806                         break;
3807
3808                 info = NULL;
3809                 rcu_read_lock();
3810                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3811                                         list) {
3812                         if (tmp->flags == types[i]) {
3813                                 info = tmp;
3814                                 break;
3815                         }
3816                 }
3817                 rcu_read_unlock();
3818
3819                 if (!info)
3820                         continue;
3821                 down_read(&info->groups_sem);
3822                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3823                         if (!list_empty(&info->block_groups[c])) {
3824                                 btrfs_get_block_group_info(
3825                                         &info->block_groups[c], &space);
3826                                 memcpy(dest, &space, sizeof(space));
3827                                 dest++;
3828                                 space_args.total_spaces++;
3829                                 slot_count--;
3830                         }
3831                         if (!slot_count)
3832                                 break;
3833                 }
3834                 up_read(&info->groups_sem);
3835         }
3836
3837         /*
3838          * Add global block reserve
3839          */
3840         if (slot_count) {
3841                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3842
3843                 spin_lock(&block_rsv->lock);
3844                 space.total_bytes = block_rsv->size;
3845                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3846                 spin_unlock(&block_rsv->lock);
3847                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3848                 memcpy(dest, &space, sizeof(space));
3849                 space_args.total_spaces++;
3850         }
3851
3852         user_dest = (struct btrfs_ioctl_space_info __user *)
3853                 (arg + sizeof(struct btrfs_ioctl_space_args));
3854
3855         if (copy_to_user(user_dest, dest_orig, alloc_size))
3856                 ret = -EFAULT;
3857
3858         kfree(dest_orig);
3859 out:
3860         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3861                 ret = -EFAULT;
3862
3863         return ret;
3864 }
3865
3866 /*
3867  * there are many ways the trans_start and trans_end ioctls can lead
3868  * to deadlocks.  They should only be used by applications that
3869  * basically own the machine, and have a very in depth understanding
3870  * of all the possible deadlocks and enospc problems.
3871  */
3872 long btrfs_ioctl_trans_end(struct file *file)
3873 {
3874         struct inode *inode = file_inode(file);
3875         struct btrfs_root *root = BTRFS_I(inode)->root;
3876         struct btrfs_trans_handle *trans;
3877
3878         trans = file->private_data;
3879         if (!trans)
3880                 return -EINVAL;
3881         file->private_data = NULL;
3882
3883         btrfs_end_transaction(trans, root);
3884
3885         atomic_dec(&root->fs_info->open_ioctl_trans);
3886
3887         mnt_drop_write_file(file);
3888         return 0;
3889 }
3890
3891 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3892                                             void __user *argp)
3893 {
3894         struct btrfs_trans_handle *trans;
3895         u64 transid;
3896         int ret;
3897
3898         trans = btrfs_attach_transaction_barrier(root);
3899         if (IS_ERR(trans)) {
3900                 if (PTR_ERR(trans) != -ENOENT)
3901                         return PTR_ERR(trans);
3902
3903                 /* No running transaction, don't bother */
3904                 transid = root->fs_info->last_trans_committed;
3905                 goto out;
3906         }
3907         transid = trans->transid;
3908         ret = btrfs_commit_transaction_async(trans, root, 0);
3909         if (ret) {
3910                 btrfs_end_transaction(trans, root);
3911                 return ret;
3912         }
3913 out:
3914         if (argp)
3915                 if (copy_to_user(argp, &transid, sizeof(transid)))
3916                         return -EFAULT;
3917         return 0;
3918 }
3919
3920 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3921                                            void __user *argp)
3922 {
3923         u64 transid;
3924
3925         if (argp) {
3926                 if (copy_from_user(&transid, argp, sizeof(transid)))
3927                         return -EFAULT;
3928         } else {
3929                 transid = 0;  /* current trans */
3930         }
3931         return btrfs_wait_for_commit(root, transid);
3932 }
3933
3934 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3935 {
3936         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3937         struct btrfs_ioctl_scrub_args *sa;
3938         int ret;
3939
3940         if (!capable(CAP_SYS_ADMIN))
3941                 return -EPERM;
3942
3943         sa = memdup_user(arg, sizeof(*sa));
3944         if (IS_ERR(sa))
3945                 return PTR_ERR(sa);
3946
3947         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3948                 ret = mnt_want_write_file(file);
3949                 if (ret)
3950                         goto out;
3951         }
3952
3953         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3954                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3955                               0);
3956
3957         if (copy_to_user(arg, sa, sizeof(*sa)))
3958                 ret = -EFAULT;
3959
3960         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3961                 mnt_drop_write_file(file);
3962 out:
3963         kfree(sa);
3964         return ret;
3965 }
3966
3967 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3968 {
3969         if (!capable(CAP_SYS_ADMIN))
3970                 return -EPERM;
3971
3972         return btrfs_scrub_cancel(root->fs_info);
3973 }
3974
3975 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3976                                        void __user *arg)
3977 {
3978         struct btrfs_ioctl_scrub_args *sa;
3979         int ret;
3980
3981         if (!capable(CAP_SYS_ADMIN))
3982                 return -EPERM;
3983
3984         sa = memdup_user(arg, sizeof(*sa));
3985         if (IS_ERR(sa))
3986                 return PTR_ERR(sa);
3987
3988         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3989
3990         if (copy_to_user(arg, sa, sizeof(*sa)))
3991                 ret = -EFAULT;
3992
3993         kfree(sa);
3994         return ret;
3995 }
3996
3997 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3998                                       void __user *arg)
3999 {
4000         struct btrfs_ioctl_get_dev_stats *sa;
4001         int ret;
4002
4003         sa = memdup_user(arg, sizeof(*sa));
4004         if (IS_ERR(sa))
4005                 return PTR_ERR(sa);
4006
4007         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4008                 kfree(sa);
4009                 return -EPERM;
4010         }
4011
4012         ret = btrfs_get_dev_stats(root, sa);
4013
4014         if (copy_to_user(arg, sa, sizeof(*sa)))
4015                 ret = -EFAULT;
4016
4017         kfree(sa);
4018         return ret;
4019 }
4020
4021 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4022 {
4023         struct btrfs_ioctl_dev_replace_args *p;
4024         int ret;
4025
4026         if (!capable(CAP_SYS_ADMIN))
4027                 return -EPERM;
4028
4029         p = memdup_user(arg, sizeof(*p));
4030         if (IS_ERR(p))
4031                 return PTR_ERR(p);
4032
4033         switch (p->cmd) {
4034         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4035                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4036                         ret = -EROFS;
4037                         goto out;
4038                 }
4039                 if (atomic_xchg(
4040                         &root->fs_info->mutually_exclusive_operation_running,
4041                         1)) {
4042                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4043                 } else {
4044                         ret = btrfs_dev_replace_start(root, p);
4045                         atomic_set(
4046                          &root->fs_info->mutually_exclusive_operation_running,
4047                          0);
4048                 }
4049                 break;
4050         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4051                 btrfs_dev_replace_status(root->fs_info, p);
4052                 ret = 0;
4053                 break;
4054         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4055                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4056                 break;
4057         default:
4058                 ret = -EINVAL;
4059                 break;
4060         }
4061
4062         if (copy_to_user(arg, p, sizeof(*p)))
4063                 ret = -EFAULT;
4064 out:
4065         kfree(p);
4066         return ret;
4067 }
4068
4069 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4070 {
4071         int ret = 0;
4072         int i;
4073         u64 rel_ptr;
4074         int size;
4075         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4076         struct inode_fs_paths *ipath = NULL;
4077         struct btrfs_path *path;
4078
4079         if (!capable(CAP_DAC_READ_SEARCH))
4080                 return -EPERM;
4081
4082         path = btrfs_alloc_path();
4083         if (!path) {
4084                 ret = -ENOMEM;
4085                 goto out;
4086         }
4087
4088         ipa = memdup_user(arg, sizeof(*ipa));
4089         if (IS_ERR(ipa)) {
4090                 ret = PTR_ERR(ipa);
4091                 ipa = NULL;
4092                 goto out;
4093         }
4094
4095         size = min_t(u32, ipa->size, 4096);
4096         ipath = init_ipath(size, root, path);
4097         if (IS_ERR(ipath)) {
4098                 ret = PTR_ERR(ipath);
4099                 ipath = NULL;
4100                 goto out;
4101         }
4102
4103         ret = paths_from_inode(ipa->inum, ipath);
4104         if (ret < 0)
4105                 goto out;
4106
4107         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4108                 rel_ptr = ipath->fspath->val[i] -
4109                           (u64)(unsigned long)ipath->fspath->val;
4110                 ipath->fspath->val[i] = rel_ptr;
4111         }
4112
4113         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4114                            (void *)(unsigned long)ipath->fspath, size);
4115         if (ret) {
4116                 ret = -EFAULT;
4117                 goto out;
4118         }
4119
4120 out:
4121         btrfs_free_path(path);
4122         free_ipath(ipath);
4123         kfree(ipa);
4124
4125         return ret;
4126 }
4127
4128 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4129 {
4130         struct btrfs_data_container *inodes = ctx;
4131         const size_t c = 3 * sizeof(u64);
4132
4133         if (inodes->bytes_left >= c) {
4134                 inodes->bytes_left -= c;
4135                 inodes->val[inodes->elem_cnt] = inum;
4136                 inodes->val[inodes->elem_cnt + 1] = offset;
4137                 inodes->val[inodes->elem_cnt + 2] = root;
4138                 inodes->elem_cnt += 3;
4139         } else {
4140                 inodes->bytes_missing += c - inodes->bytes_left;
4141                 inodes->bytes_left = 0;
4142                 inodes->elem_missed += 3;
4143         }
4144
4145         return 0;
4146 }
4147
4148 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4149                                         void __user *arg)
4150 {
4151         int ret = 0;
4152         int size;
4153         struct btrfs_ioctl_logical_ino_args *loi;
4154         struct btrfs_data_container *inodes = NULL;
4155         struct btrfs_path *path = NULL;
4156
4157         if (!capable(CAP_SYS_ADMIN))
4158                 return -EPERM;
4159
4160         loi = memdup_user(arg, sizeof(*loi));
4161         if (IS_ERR(loi)) {
4162                 ret = PTR_ERR(loi);
4163                 loi = NULL;
4164                 goto out;
4165         }
4166
4167         path = btrfs_alloc_path();
4168         if (!path) {
4169                 ret = -ENOMEM;
4170                 goto out;
4171         }
4172
4173         size = min_t(u32, loi->size, 64 * 1024);
4174         inodes = init_data_container(size);
4175         if (IS_ERR(inodes)) {
4176                 ret = PTR_ERR(inodes);
4177                 inodes = NULL;
4178                 goto out;
4179         }
4180
4181         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4182                                           build_ino_list, inodes);
4183         if (ret == -EINVAL)
4184                 ret = -ENOENT;
4185         if (ret < 0)
4186                 goto out;
4187
4188         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4189                            (void *)(unsigned long)inodes, size);
4190         if (ret)
4191                 ret = -EFAULT;
4192
4193 out:
4194         btrfs_free_path(path);
4195         vfree(inodes);
4196         kfree(loi);
4197
4198         return ret;
4199 }
4200
4201 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4202                                struct btrfs_ioctl_balance_args *bargs)
4203 {
4204         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4205
4206         bargs->flags = bctl->flags;
4207
4208         if (atomic_read(&fs_info->balance_running))
4209                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4210         if (atomic_read(&fs_info->balance_pause_req))
4211                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4212         if (atomic_read(&fs_info->balance_cancel_req))
4213                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4214
4215         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4216         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4217         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4218
4219         if (lock) {
4220                 spin_lock(&fs_info->balance_lock);
4221                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4222                 spin_unlock(&fs_info->balance_lock);
4223         } else {
4224                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4225         }
4226 }
4227
4228 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4229 {
4230         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4231         struct btrfs_fs_info *fs_info = root->fs_info;
4232         struct btrfs_ioctl_balance_args *bargs;
4233         struct btrfs_balance_control *bctl;
4234         bool need_unlock; /* for mut. excl. ops lock */
4235         int ret;
4236
4237         if (!capable(CAP_SYS_ADMIN))
4238                 return -EPERM;
4239
4240         ret = mnt_want_write_file(file);
4241         if (ret)
4242                 return ret;
4243
4244 again:
4245         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4246                 mutex_lock(&fs_info->volume_mutex);
4247                 mutex_lock(&fs_info->balance_mutex);
4248                 need_unlock = true;
4249                 goto locked;
4250         }
4251
4252         /*
4253          * mut. excl. ops lock is locked.  Three possibilites:
4254          *   (1) some other op is running
4255          *   (2) balance is running
4256          *   (3) balance is paused -- special case (think resume)
4257          */
4258         mutex_lock(&fs_info->balance_mutex);
4259         if (fs_info->balance_ctl) {
4260                 /* this is either (2) or (3) */
4261                 if (!atomic_read(&fs_info->balance_running)) {
4262                         mutex_unlock(&fs_info->balance_mutex);
4263                         if (!mutex_trylock(&fs_info->volume_mutex))
4264                                 goto again;
4265                         mutex_lock(&fs_info->balance_mutex);
4266
4267                         if (fs_info->balance_ctl &&
4268                             !atomic_read(&fs_info->balance_running)) {
4269                                 /* this is (3) */
4270                                 need_unlock = false;
4271                                 goto locked;
4272                         }
4273
4274                         mutex_unlock(&fs_info->balance_mutex);
4275                         mutex_unlock(&fs_info->volume_mutex);
4276                         goto again;
4277                 } else {
4278                         /* this is (2) */
4279                         mutex_unlock(&fs_info->balance_mutex);
4280                         ret = -EINPROGRESS;
4281                         goto out;
4282                 }
4283         } else {
4284                 /* this is (1) */
4285                 mutex_unlock(&fs_info->balance_mutex);
4286                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4287                 goto out;
4288         }
4289
4290 locked:
4291         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4292
4293         if (arg) {
4294                 bargs = memdup_user(arg, sizeof(*bargs));
4295                 if (IS_ERR(bargs)) {
4296                         ret = PTR_ERR(bargs);
4297                         goto out_unlock;
4298                 }
4299
4300                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4301                         if (!fs_info->balance_ctl) {
4302                                 ret = -ENOTCONN;
4303                                 goto out_bargs;
4304                         }
4305
4306                         bctl = fs_info->balance_ctl;
4307                         spin_lock(&fs_info->balance_lock);
4308                         bctl->flags |= BTRFS_BALANCE_RESUME;
4309                         spin_unlock(&fs_info->balance_lock);
4310
4311                         goto do_balance;
4312                 }
4313         } else {
4314                 bargs = NULL;
4315         }
4316
4317         if (fs_info->balance_ctl) {
4318                 ret = -EINPROGRESS;
4319                 goto out_bargs;
4320         }
4321
4322         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4323         if (!bctl) {
4324                 ret = -ENOMEM;
4325                 goto out_bargs;
4326         }
4327
4328         bctl->fs_info = fs_info;
4329         if (arg) {
4330                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4331                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4332                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4333
4334                 bctl->flags = bargs->flags;
4335         } else {
4336                 /* balance everything - no filters */
4337                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4338         }
4339
4340 do_balance:
4341         /*
4342          * Ownership of bctl and mutually_exclusive_operation_running
4343          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4344          * or, if restriper was paused all the way until unmount, in
4345          * free_fs_info.  mutually_exclusive_operation_running is
4346          * cleared in __cancel_balance.
4347          */
4348         need_unlock = false;
4349
4350         ret = btrfs_balance(bctl, bargs);
4351
4352         if (arg) {
4353                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4354                         ret = -EFAULT;
4355         }
4356
4357 out_bargs:
4358         kfree(bargs);
4359 out_unlock:
4360         mutex_unlock(&fs_info->balance_mutex);
4361         mutex_unlock(&fs_info->volume_mutex);
4362         if (need_unlock)
4363                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4364 out:
4365         mnt_drop_write_file(file);
4366         return ret;
4367 }
4368
4369 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4370 {
4371         if (!capable(CAP_SYS_ADMIN))
4372                 return -EPERM;
4373
4374         switch (cmd) {
4375         case BTRFS_BALANCE_CTL_PAUSE:
4376                 return btrfs_pause_balance(root->fs_info);
4377         case BTRFS_BALANCE_CTL_CANCEL:
4378                 return btrfs_cancel_balance(root->fs_info);
4379         }
4380
4381         return -EINVAL;
4382 }
4383
4384 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4385                                          void __user *arg)
4386 {
4387         struct btrfs_fs_info *fs_info = root->fs_info;
4388         struct btrfs_ioctl_balance_args *bargs;
4389         int ret = 0;
4390
4391         if (!capable(CAP_SYS_ADMIN))
4392                 return -EPERM;
4393
4394         mutex_lock(&fs_info->balance_mutex);
4395         if (!fs_info->balance_ctl) {
4396                 ret = -ENOTCONN;
4397                 goto out;
4398         }
4399
4400         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4401         if (!bargs) {
4402                 ret = -ENOMEM;
4403                 goto out;
4404         }
4405
4406         update_ioctl_balance_args(fs_info, 1, bargs);
4407
4408         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4409                 ret = -EFAULT;
4410
4411         kfree(bargs);
4412 out:
4413         mutex_unlock(&fs_info->balance_mutex);
4414         return ret;
4415 }
4416
4417 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4418 {
4419         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4420         struct btrfs_ioctl_quota_ctl_args *sa;
4421         struct btrfs_trans_handle *trans = NULL;
4422         int ret;
4423         int err;
4424
4425         if (!capable(CAP_SYS_ADMIN))
4426                 return -EPERM;
4427
4428         ret = mnt_want_write_file(file);
4429         if (ret)
4430                 return ret;
4431
4432         sa = memdup_user(arg, sizeof(*sa));
4433         if (IS_ERR(sa)) {
4434                 ret = PTR_ERR(sa);
4435                 goto drop_write;
4436         }
4437
4438         down_write(&root->fs_info->subvol_sem);
4439         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4440         if (IS_ERR(trans)) {
4441                 ret = PTR_ERR(trans);
4442                 goto out;
4443         }
4444
4445         switch (sa->cmd) {
4446         case BTRFS_QUOTA_CTL_ENABLE:
4447                 ret = btrfs_quota_enable(trans, root->fs_info);
4448                 break;
4449         case BTRFS_QUOTA_CTL_DISABLE:
4450                 ret = btrfs_quota_disable(trans, root->fs_info);
4451                 break;
4452         default:
4453                 ret = -EINVAL;
4454                 break;
4455         }
4456
4457         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4458         if (err && !ret)
4459                 ret = err;
4460 out:
4461         kfree(sa);
4462         up_write(&root->fs_info->subvol_sem);
4463 drop_write:
4464         mnt_drop_write_file(file);
4465         return ret;
4466 }
4467
4468 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4469 {
4470         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4471         struct btrfs_ioctl_qgroup_assign_args *sa;
4472         struct btrfs_trans_handle *trans;
4473         int ret;
4474         int err;
4475
4476         if (!capable(CAP_SYS_ADMIN))
4477                 return -EPERM;
4478
4479         ret = mnt_want_write_file(file);
4480         if (ret)
4481                 return ret;
4482
4483         sa = memdup_user(arg, sizeof(*sa));
4484         if (IS_ERR(sa)) {
4485                 ret = PTR_ERR(sa);
4486                 goto drop_write;
4487         }
4488
4489         trans = btrfs_join_transaction(root);
4490         if (IS_ERR(trans)) {
4491                 ret = PTR_ERR(trans);
4492                 goto out;
4493         }
4494
4495         /* FIXME: check if the IDs really exist */
4496         if (sa->assign) {
4497                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4498                                                 sa->src, sa->dst);
4499         } else {
4500                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4501                                                 sa->src, sa->dst);
4502         }
4503
4504         err = btrfs_end_transaction(trans, root);
4505         if (err && !ret)
4506                 ret = err;
4507
4508 out:
4509         kfree(sa);
4510 drop_write:
4511         mnt_drop_write_file(file);
4512         return ret;
4513 }
4514
4515 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4516 {
4517         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4518         struct btrfs_ioctl_qgroup_create_args *sa;
4519         struct btrfs_trans_handle *trans;
4520         int ret;
4521         int err;
4522
4523         if (!capable(CAP_SYS_ADMIN))
4524                 return -EPERM;
4525
4526         ret = mnt_want_write_file(file);
4527         if (ret)
4528                 return ret;
4529
4530         sa = memdup_user(arg, sizeof(*sa));
4531         if (IS_ERR(sa)) {
4532                 ret = PTR_ERR(sa);
4533                 goto drop_write;
4534         }
4535
4536         if (!sa->qgroupid) {
4537                 ret = -EINVAL;
4538                 goto out;
4539         }
4540
4541         trans = btrfs_join_transaction(root);
4542         if (IS_ERR(trans)) {
4543                 ret = PTR_ERR(trans);
4544                 goto out;
4545         }
4546
4547         /* FIXME: check if the IDs really exist */
4548         if (sa->create) {
4549                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4550                                           NULL);
4551         } else {
4552                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4553         }
4554
4555         err = btrfs_end_transaction(trans, root);
4556         if (err && !ret)
4557                 ret = err;
4558
4559 out:
4560         kfree(sa);
4561 drop_write:
4562         mnt_drop_write_file(file);
4563         return ret;
4564 }
4565
4566 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4567 {
4568         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4569         struct btrfs_ioctl_qgroup_limit_args *sa;
4570         struct btrfs_trans_handle *trans;
4571         int ret;
4572         int err;
4573         u64 qgroupid;
4574
4575         if (!capable(CAP_SYS_ADMIN))
4576                 return -EPERM;
4577
4578         ret = mnt_want_write_file(file);
4579         if (ret)
4580                 return ret;
4581
4582         sa = memdup_user(arg, sizeof(*sa));
4583         if (IS_ERR(sa)) {
4584                 ret = PTR_ERR(sa);
4585                 goto drop_write;
4586         }
4587
4588         trans = btrfs_join_transaction(root);
4589         if (IS_ERR(trans)) {
4590                 ret = PTR_ERR(trans);
4591                 goto out;
4592         }
4593
4594         qgroupid = sa->qgroupid;
4595         if (!qgroupid) {
4596                 /* take the current subvol as qgroup */
4597                 qgroupid = root->root_key.objectid;
4598         }
4599
4600         /* FIXME: check if the IDs really exist */
4601         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4602
4603         err = btrfs_end_transaction(trans, root);
4604         if (err && !ret)
4605                 ret = err;
4606
4607 out:
4608         kfree(sa);
4609 drop_write:
4610         mnt_drop_write_file(file);
4611         return ret;
4612 }
4613
4614 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4615 {
4616         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4617         struct btrfs_ioctl_quota_rescan_args *qsa;
4618         int ret;
4619
4620         if (!capable(CAP_SYS_ADMIN))
4621                 return -EPERM;
4622
4623         ret = mnt_want_write_file(file);
4624         if (ret)
4625                 return ret;
4626
4627         qsa = memdup_user(arg, sizeof(*qsa));
4628         if (IS_ERR(qsa)) {
4629                 ret = PTR_ERR(qsa);
4630                 goto drop_write;
4631         }
4632
4633         if (qsa->flags) {
4634                 ret = -EINVAL;
4635                 goto out;
4636         }
4637
4638         ret = btrfs_qgroup_rescan(root->fs_info);
4639
4640 out:
4641         kfree(qsa);
4642 drop_write:
4643         mnt_drop_write_file(file);
4644         return ret;
4645 }
4646
4647 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4648 {
4649         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4650         struct btrfs_ioctl_quota_rescan_args *qsa;
4651         int ret = 0;
4652
4653         if (!capable(CAP_SYS_ADMIN))
4654                 return -EPERM;
4655
4656         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4657         if (!qsa)
4658                 return -ENOMEM;
4659
4660         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4661                 qsa->flags = 1;
4662                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4663         }
4664
4665         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4666                 ret = -EFAULT;
4667
4668         kfree(qsa);
4669         return ret;
4670 }
4671
4672 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4673 {
4674         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4675
4676         if (!capable(CAP_SYS_ADMIN))
4677                 return -EPERM;
4678
4679         return btrfs_qgroup_wait_for_completion(root->fs_info);
4680 }
4681
4682 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4683                                             struct btrfs_ioctl_received_subvol_args *sa)
4684 {
4685         struct inode *inode = file_inode(file);
4686         struct btrfs_root *root = BTRFS_I(inode)->root;
4687         struct btrfs_root_item *root_item = &root->root_item;
4688         struct btrfs_trans_handle *trans;
4689         struct timespec ct = CURRENT_TIME;
4690         int ret = 0;
4691         int received_uuid_changed;
4692
4693         if (!inode_owner_or_capable(inode))
4694                 return -EPERM;
4695
4696         ret = mnt_want_write_file(file);
4697         if (ret < 0)
4698                 return ret;
4699
4700         down_write(&root->fs_info->subvol_sem);
4701
4702         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4703                 ret = -EINVAL;
4704                 goto out;
4705         }
4706
4707         if (btrfs_root_readonly(root)) {
4708                 ret = -EROFS;
4709                 goto out;
4710         }
4711
4712         /*
4713          * 1 - root item
4714          * 2 - uuid items (received uuid + subvol uuid)
4715          */
4716         trans = btrfs_start_transaction(root, 3);
4717         if (IS_ERR(trans)) {
4718                 ret = PTR_ERR(trans);
4719                 trans = NULL;
4720                 goto out;
4721         }
4722
4723         sa->rtransid = trans->transid;
4724         sa->rtime.sec = ct.tv_sec;
4725         sa->rtime.nsec = ct.tv_nsec;
4726
4727         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4728                                        BTRFS_UUID_SIZE);
4729         if (received_uuid_changed &&
4730             !btrfs_is_empty_uuid(root_item->received_uuid))
4731                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4732                                     root_item->received_uuid,
4733                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4734                                     root->root_key.objectid);
4735         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4736         btrfs_set_root_stransid(root_item, sa->stransid);
4737         btrfs_set_root_rtransid(root_item, sa->rtransid);
4738         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4739         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4740         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4741         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4742
4743         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4744                                 &root->root_key, &root->root_item);
4745         if (ret < 0) {
4746                 btrfs_end_transaction(trans, root);
4747                 goto out;
4748         }
4749         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4750                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4751                                           sa->uuid,
4752                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4753                                           root->root_key.objectid);
4754                 if (ret < 0 && ret != -EEXIST) {
4755                         btrfs_abort_transaction(trans, root, ret);
4756                         goto out;
4757                 }
4758         }
4759         ret = btrfs_commit_transaction(trans, root);
4760         if (ret < 0) {
4761                 btrfs_abort_transaction(trans, root, ret);
4762                 goto out;
4763         }
4764
4765 out:
4766         up_write(&root->fs_info->subvol_sem);
4767         mnt_drop_write_file(file);
4768         return ret;
4769 }
4770
4771 #ifdef CONFIG_64BIT
4772 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4773                                                 void __user *arg)
4774 {
4775         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4776         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4777         int ret = 0;
4778
4779         args32 = memdup_user(arg, sizeof(*args32));
4780         if (IS_ERR(args32)) {
4781                 ret = PTR_ERR(args32);
4782                 args32 = NULL;
4783                 goto out;
4784         }
4785
4786         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4787         if (!args64) {
4788                 ret = -ENOMEM;
4789                 goto out;
4790         }
4791
4792         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4793         args64->stransid = args32->stransid;
4794         args64->rtransid = args32->rtransid;
4795         args64->stime.sec = args32->stime.sec;
4796         args64->stime.nsec = args32->stime.nsec;
4797         args64->rtime.sec = args32->rtime.sec;
4798         args64->rtime.nsec = args32->rtime.nsec;
4799         args64->flags = args32->flags;
4800
4801         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4802         if (ret)
4803                 goto out;
4804
4805         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4806         args32->stransid = args64->stransid;
4807         args32->rtransid = args64->rtransid;
4808         args32->stime.sec = args64->stime.sec;
4809         args32->stime.nsec = args64->stime.nsec;
4810         args32->rtime.sec = args64->rtime.sec;
4811         args32->rtime.nsec = args64->rtime.nsec;
4812         args32->flags = args64->flags;
4813
4814         ret = copy_to_user(arg, args32, sizeof(*args32));
4815         if (ret)
4816                 ret = -EFAULT;
4817
4818 out:
4819         kfree(args32);
4820         kfree(args64);
4821         return ret;
4822 }
4823 #endif
4824
4825 static long btrfs_ioctl_set_received_subvol(struct file *file,
4826                                             void __user *arg)
4827 {
4828         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4829         int ret = 0;
4830
4831         sa = memdup_user(arg, sizeof(*sa));
4832         if (IS_ERR(sa)) {
4833                 ret = PTR_ERR(sa);
4834                 sa = NULL;
4835                 goto out;
4836         }
4837
4838         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4839
4840         if (ret)
4841                 goto out;
4842
4843         ret = copy_to_user(arg, sa, sizeof(*sa));
4844         if (ret)
4845                 ret = -EFAULT;
4846
4847 out:
4848         kfree(sa);
4849         return ret;
4850 }
4851
4852 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4853 {
4854         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4855         size_t len;
4856         int ret;
4857         char label[BTRFS_LABEL_SIZE];
4858
4859         spin_lock(&root->fs_info->super_lock);
4860         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4861         spin_unlock(&root->fs_info->super_lock);
4862
4863         len = strnlen(label, BTRFS_LABEL_SIZE);
4864
4865         if (len == BTRFS_LABEL_SIZE) {
4866                 btrfs_warn(root->fs_info,
4867                         "label is too long, return the first %zu bytes", --len);
4868         }
4869
4870         ret = copy_to_user(arg, label, len);
4871
4872         return ret ? -EFAULT : 0;
4873 }
4874
4875 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4876 {
4877         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4878         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4879         struct btrfs_trans_handle *trans;
4880         char label[BTRFS_LABEL_SIZE];
4881         int ret;
4882
4883         if (!capable(CAP_SYS_ADMIN))
4884                 return -EPERM;
4885
4886         if (copy_from_user(label, arg, sizeof(label)))
4887                 return -EFAULT;
4888
4889         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4890                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4891                        BTRFS_LABEL_SIZE - 1);
4892                 return -EINVAL;
4893         }
4894
4895         ret = mnt_want_write_file(file);
4896         if (ret)
4897                 return ret;
4898
4899         trans = btrfs_start_transaction(root, 0);
4900         if (IS_ERR(trans)) {
4901                 ret = PTR_ERR(trans);
4902                 goto out_unlock;
4903         }
4904
4905         spin_lock(&root->fs_info->super_lock);
4906         strcpy(super_block->label, label);
4907         spin_unlock(&root->fs_info->super_lock);
4908         ret = btrfs_commit_transaction(trans, root);
4909
4910 out_unlock:
4911         mnt_drop_write_file(file);
4912         return ret;
4913 }
4914
4915 #define INIT_FEATURE_FLAGS(suffix) \
4916         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4917           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4918           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4919
4920 static int btrfs_ioctl_get_supported_features(struct file *file,
4921                                               void __user *arg)
4922 {
4923         static struct btrfs_ioctl_feature_flags features[3] = {
4924                 INIT_FEATURE_FLAGS(SUPP),
4925                 INIT_FEATURE_FLAGS(SAFE_SET),
4926                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4927         };
4928
4929         if (copy_to_user(arg, &features, sizeof(features)))
4930                 return -EFAULT;
4931
4932         return 0;
4933 }
4934
4935 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4936 {
4937         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4938         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4939         struct btrfs_ioctl_feature_flags features;
4940
4941         features.compat_flags = btrfs_super_compat_flags(super_block);
4942         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4943         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4944
4945         if (copy_to_user(arg, &features, sizeof(features)))
4946                 return -EFAULT;
4947
4948         return 0;
4949 }
4950
4951 static int check_feature_bits(struct btrfs_root *root,
4952                               enum btrfs_feature_set set,
4953                               u64 change_mask, u64 flags, u64 supported_flags,
4954                               u64 safe_set, u64 safe_clear)
4955 {
4956         const char *type = btrfs_feature_set_names[set];
4957         char *names;
4958         u64 disallowed, unsupported;
4959         u64 set_mask = flags & change_mask;
4960         u64 clear_mask = ~flags & change_mask;
4961
4962         unsupported = set_mask & ~supported_flags;
4963         if (unsupported) {
4964                 names = btrfs_printable_features(set, unsupported);
4965                 if (names) {
4966                         btrfs_warn(root->fs_info,
4967                            "this kernel does not support the %s feature bit%s",
4968                            names, strchr(names, ',') ? "s" : "");
4969                         kfree(names);
4970                 } else
4971                         btrfs_warn(root->fs_info,
4972                            "this kernel does not support %s bits 0x%llx",
4973                            type, unsupported);
4974                 return -EOPNOTSUPP;
4975         }
4976
4977         disallowed = set_mask & ~safe_set;
4978         if (disallowed) {
4979                 names = btrfs_printable_features(set, disallowed);
4980                 if (names) {
4981                         btrfs_warn(root->fs_info,
4982                            "can't set the %s feature bit%s while mounted",
4983                            names, strchr(names, ',') ? "s" : "");
4984                         kfree(names);
4985                 } else
4986                         btrfs_warn(root->fs_info,
4987                            "can't set %s bits 0x%llx while mounted",
4988                            type, disallowed);
4989                 return -EPERM;
4990         }
4991
4992         disallowed = clear_mask & ~safe_clear;
4993         if (disallowed) {
4994                 names = btrfs_printable_features(set, disallowed);
4995                 if (names) {
4996                         btrfs_warn(root->fs_info,
4997                            "can't clear the %s feature bit%s while mounted",
4998                            names, strchr(names, ',') ? "s" : "");
4999                         kfree(names);
5000                 } else
5001                         btrfs_warn(root->fs_info,
5002                            "can't clear %s bits 0x%llx while mounted",
5003                            type, disallowed);
5004                 return -EPERM;
5005         }
5006
5007         return 0;
5008 }
5009
5010 #define check_feature(root, change_mask, flags, mask_base)      \
5011 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5012                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5013                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5014                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5015
5016 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5017 {
5018         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5019         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5020         struct btrfs_ioctl_feature_flags flags[2];
5021         struct btrfs_trans_handle *trans;
5022         u64 newflags;
5023         int ret;
5024
5025         if (!capable(CAP_SYS_ADMIN))
5026                 return -EPERM;
5027
5028         if (copy_from_user(flags, arg, sizeof(flags)))
5029                 return -EFAULT;
5030
5031         /* Nothing to do */
5032         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5033             !flags[0].incompat_flags)
5034                 return 0;
5035
5036         ret = check_feature(root, flags[0].compat_flags,
5037                             flags[1].compat_flags, COMPAT);
5038         if (ret)
5039                 return ret;
5040
5041         ret = check_feature(root, flags[0].compat_ro_flags,
5042                             flags[1].compat_ro_flags, COMPAT_RO);
5043         if (ret)
5044                 return ret;
5045
5046         ret = check_feature(root, flags[0].incompat_flags,
5047                             flags[1].incompat_flags, INCOMPAT);
5048         if (ret)
5049                 return ret;
5050
5051         trans = btrfs_start_transaction(root, 0);
5052         if (IS_ERR(trans))
5053                 return PTR_ERR(trans);
5054
5055         spin_lock(&root->fs_info->super_lock);
5056         newflags = btrfs_super_compat_flags(super_block);
5057         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5058         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5059         btrfs_set_super_compat_flags(super_block, newflags);
5060
5061         newflags = btrfs_super_compat_ro_flags(super_block);
5062         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5063         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5064         btrfs_set_super_compat_ro_flags(super_block, newflags);
5065
5066         newflags = btrfs_super_incompat_flags(super_block);
5067         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5068         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5069         btrfs_set_super_incompat_flags(super_block, newflags);
5070         spin_unlock(&root->fs_info->super_lock);
5071
5072         return btrfs_commit_transaction(trans, root);
5073 }
5074
5075 long btrfs_ioctl(struct file *file, unsigned int
5076                 cmd, unsigned long arg)
5077 {
5078         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5079         void __user *argp = (void __user *)arg;
5080
5081         switch (cmd) {
5082         case FS_IOC_GETFLAGS:
5083                 return btrfs_ioctl_getflags(file, argp);
5084         case FS_IOC_SETFLAGS:
5085                 return btrfs_ioctl_setflags(file, argp);
5086         case FS_IOC_GETVERSION:
5087                 return btrfs_ioctl_getversion(file, argp);
5088         case FITRIM:
5089                 return btrfs_ioctl_fitrim(file, argp);
5090         case BTRFS_IOC_SNAP_CREATE:
5091                 return btrfs_ioctl_snap_create(file, argp, 0);
5092         case BTRFS_IOC_SNAP_CREATE_V2:
5093                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5094         case BTRFS_IOC_SUBVOL_CREATE:
5095                 return btrfs_ioctl_snap_create(file, argp, 1);
5096         case BTRFS_IOC_SUBVOL_CREATE_V2:
5097                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5098         case BTRFS_IOC_SNAP_DESTROY:
5099                 return btrfs_ioctl_snap_destroy(file, argp);
5100         case BTRFS_IOC_SUBVOL_GETFLAGS:
5101                 return btrfs_ioctl_subvol_getflags(file, argp);
5102         case BTRFS_IOC_SUBVOL_SETFLAGS:
5103                 return btrfs_ioctl_subvol_setflags(file, argp);
5104         case BTRFS_IOC_DEFAULT_SUBVOL:
5105                 return btrfs_ioctl_default_subvol(file, argp);
5106         case BTRFS_IOC_DEFRAG:
5107                 return btrfs_ioctl_defrag(file, NULL);
5108         case BTRFS_IOC_DEFRAG_RANGE:
5109                 return btrfs_ioctl_defrag(file, argp);
5110         case BTRFS_IOC_RESIZE:
5111                 return btrfs_ioctl_resize(file, argp);
5112         case BTRFS_IOC_ADD_DEV:
5113                 return btrfs_ioctl_add_dev(root, argp);
5114         case BTRFS_IOC_RM_DEV:
5115                 return btrfs_ioctl_rm_dev(file, argp);
5116         case BTRFS_IOC_FS_INFO:
5117                 return btrfs_ioctl_fs_info(root, argp);
5118         case BTRFS_IOC_DEV_INFO:
5119                 return btrfs_ioctl_dev_info(root, argp);
5120         case BTRFS_IOC_BALANCE:
5121                 return btrfs_ioctl_balance(file, NULL);
5122         case BTRFS_IOC_CLONE:
5123                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5124         case BTRFS_IOC_CLONE_RANGE:
5125                 return btrfs_ioctl_clone_range(file, argp);
5126         case BTRFS_IOC_TRANS_START:
5127                 return btrfs_ioctl_trans_start(file);
5128         case BTRFS_IOC_TRANS_END:
5129                 return btrfs_ioctl_trans_end(file);
5130         case BTRFS_IOC_TREE_SEARCH:
5131                 return btrfs_ioctl_tree_search(file, argp);
5132         case BTRFS_IOC_INO_LOOKUP:
5133                 return btrfs_ioctl_ino_lookup(file, argp);
5134         case BTRFS_IOC_INO_PATHS:
5135                 return btrfs_ioctl_ino_to_path(root, argp);
5136         case BTRFS_IOC_LOGICAL_INO:
5137                 return btrfs_ioctl_logical_to_ino(root, argp);
5138         case BTRFS_IOC_SPACE_INFO:
5139                 return btrfs_ioctl_space_info(root, argp);
5140         case BTRFS_IOC_SYNC: {
5141                 int ret;
5142
5143                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5144                 if (ret)
5145                         return ret;
5146                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
5147                 return ret;
5148         }
5149         case BTRFS_IOC_START_SYNC:
5150                 return btrfs_ioctl_start_sync(root, argp);
5151         case BTRFS_IOC_WAIT_SYNC:
5152                 return btrfs_ioctl_wait_sync(root, argp);
5153         case BTRFS_IOC_SCRUB:
5154                 return btrfs_ioctl_scrub(file, argp);
5155         case BTRFS_IOC_SCRUB_CANCEL:
5156                 return btrfs_ioctl_scrub_cancel(root, argp);
5157         case BTRFS_IOC_SCRUB_PROGRESS:
5158                 return btrfs_ioctl_scrub_progress(root, argp);
5159         case BTRFS_IOC_BALANCE_V2:
5160                 return btrfs_ioctl_balance(file, argp);
5161         case BTRFS_IOC_BALANCE_CTL:
5162                 return btrfs_ioctl_balance_ctl(root, arg);
5163         case BTRFS_IOC_BALANCE_PROGRESS:
5164                 return btrfs_ioctl_balance_progress(root, argp);
5165         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5166                 return btrfs_ioctl_set_received_subvol(file, argp);
5167 #ifdef CONFIG_64BIT
5168         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5169                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5170 #endif
5171         case BTRFS_IOC_SEND:
5172                 return btrfs_ioctl_send(file, argp);
5173         case BTRFS_IOC_GET_DEV_STATS:
5174                 return btrfs_ioctl_get_dev_stats(root, argp);
5175         case BTRFS_IOC_QUOTA_CTL:
5176                 return btrfs_ioctl_quota_ctl(file, argp);
5177         case BTRFS_IOC_QGROUP_ASSIGN:
5178                 return btrfs_ioctl_qgroup_assign(file, argp);
5179         case BTRFS_IOC_QGROUP_CREATE:
5180                 return btrfs_ioctl_qgroup_create(file, argp);
5181         case BTRFS_IOC_QGROUP_LIMIT:
5182                 return btrfs_ioctl_qgroup_limit(file, argp);
5183         case BTRFS_IOC_QUOTA_RESCAN:
5184                 return btrfs_ioctl_quota_rescan(file, argp);
5185         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5186                 return btrfs_ioctl_quota_rescan_status(file, argp);
5187         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5188                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5189         case BTRFS_IOC_DEV_REPLACE:
5190                 return btrfs_ioctl_dev_replace(root, argp);
5191         case BTRFS_IOC_GET_FSLABEL:
5192                 return btrfs_ioctl_get_fslabel(file, argp);
5193         case BTRFS_IOC_SET_FSLABEL:
5194                 return btrfs_ioctl_set_fslabel(file, argp);
5195         case BTRFS_IOC_FILE_EXTENT_SAME:
5196                 return btrfs_ioctl_file_extent_same(file, argp);
5197         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5198                 return btrfs_ioctl_get_supported_features(file, argp);
5199         case BTRFS_IOC_GET_FEATURES:
5200                 return btrfs_ioctl_get_features(file, argp);
5201         case BTRFS_IOC_SET_FEATURES:
5202                 return btrfs_ioctl_set_features(file, argp);
5203         }
5204
5205         return -ENOTTY;
5206 }