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                        int no_time_update);
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  95 {
  96         if (S_ISDIR(mode))
  97                 return flags;
  98         else if (S_ISREG(mode))
  99                 return flags & ~FS_DIRSYNC_FL;
 100         else
 101                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 102 }
 103
 104 /*
 105  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 106  */
 107 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 108 {
 109         unsigned int iflags = 0;
 110
 111         if (flags & BTRFS_INODE_SYNC)
 112                 iflags |= FS_SYNC_FL;
 113         if (flags & BTRFS_INODE_IMMUTABLE)
 114                 iflags |= FS_IMMUTABLE_FL;
 115         if (flags & BTRFS_INODE_APPEND)
 116                 iflags |= FS_APPEND_FL;
 117         if (flags & BTRFS_INODE_NODUMP)
 118                 iflags |= FS_NODUMP_FL;
 119         if (flags & BTRFS_INODE_NOATIME)
 120                 iflags |= FS_NOATIME_FL;
 121         if (flags & BTRFS_INODE_DIRSYNC)
 122                 iflags |= FS_DIRSYNC_FL;
 123         if (flags & BTRFS_INODE_NODATACOW)
 124                 iflags |= FS_NOCOW_FL;
 125
 126         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 127                 iflags |= FS_COMPR_FL;
 128         else if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130
 131         return iflags;
 132 }
 133
 134 /*
 135  * Update inode->i_flags based on the btrfs internal flags.
 136  */
 137 void btrfs_update_iflags(struct inode *inode)
 138 {
 139         struct btrfs_inode *ip = BTRFS_I(inode);
 140         unsigned int new_fl = 0;
 141
 142         if (ip->flags & BTRFS_INODE_SYNC)
 143                 new_fl |= S_SYNC;
 144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 145                 new_fl |= S_IMMUTABLE;
 146         if (ip->flags & BTRFS_INODE_APPEND)
 147                 new_fl |= S_APPEND;
 148         if (ip->flags & BTRFS_INODE_NOATIME)
 149                 new_fl |= S_NOATIME;
 150         if (ip->flags & BTRFS_INODE_DIRSYNC)
 151                 new_fl |= S_DIRSYNC;
 152
 153         set_mask_bits(&inode->i_flags,
 154                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 155                       new_fl);
 156 }
 157
 158 /*
 159  * Inherit flags from the parent inode.
 160  *
 161  * Currently only the compression flags and the cow flags are inherited.
 162  */
 163 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 164 {
 165         unsigned int flags;
 166
 167         if (!dir)
 168                 return;
 169
 170         flags = BTRFS_I(dir)->flags;
 171
 172         if (flags & BTRFS_INODE_NOCOMPRESS) {
 173                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 174                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 175         } else if (flags & BTRFS_INODE_COMPRESS) {
 176                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 177                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 178         }
 179
 180         if (flags & BTRFS_INODE_NODATACOW) {
 181                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 182                 if (S_ISREG(inode->i_mode))
 183                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 184         }
 185
 186         btrfs_update_iflags(inode);
 187 }
 188
 189 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 190 {
 191         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 192         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 193
 194         if (copy_to_user(arg, &flags, sizeof(flags)))
 195                 return -EFAULT;
 196         return 0;
 197 }
 198
 199 static int check_flags(unsigned int flags)
 200 {
 201         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 202                       FS_NOATIME_FL | FS_NODUMP_FL | \
 203                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 204                       FS_NOCOMP_FL | FS_COMPR_FL |
 205                       FS_NOCOW_FL))
 206                 return -EOPNOTSUPP;
 207
 208         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 215 {
 216         struct inode *inode = file_inode(file);
 217         struct btrfs_inode *ip = BTRFS_I(inode);
 218         struct btrfs_root *root = ip->root;
 219         struct btrfs_trans_handle *trans;
 220         unsigned int flags, oldflags;
 221         int ret;
 222         u64 ip_oldflags;
 223         unsigned int i_oldflags;
 224         umode_t mode;
 225
 226         if (!inode_owner_or_capable(inode))
 227                 return -EPERM;
 228
 229         if (btrfs_root_readonly(root))
 230                 return -EROFS;
 231
 232         if (copy_from_user(&flags, arg, sizeof(flags)))
 233                 return -EFAULT;
 234
 235         ret = check_flags(flags);
 236         if (ret)
 237                 return ret;
 238
 239         ret = mnt_want_write_file(file);
 240         if (ret)
 241                 return ret;
 242
 243         mutex_lock(&inode->i_mutex);
 244
 245         ip_oldflags = ip->flags;
 246         i_oldflags = inode->i_flags;
 247         mode = inode->i_mode;
 248
 249         flags = btrfs_mask_flags(inode->i_mode, flags);
 250         oldflags = btrfs_flags_to_ioctl(ip->flags);
 251         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 252                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 253                         ret = -EPERM;
 254                         goto out_unlock;
 255                 }
 256         }
 257
 258         if (flags & FS_SYNC_FL)
 259                 ip->flags |= BTRFS_INODE_SYNC;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_SYNC;
 262         if (flags & FS_IMMUTABLE_FL)
 263                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 266         if (flags & FS_APPEND_FL)
 267                 ip->flags |= BTRFS_INODE_APPEND;
 268         else
 269                 ip->flags &= ~BTRFS_INODE_APPEND;
 270         if (flags & FS_NODUMP_FL)
 271                 ip->flags |= BTRFS_INODE_NODUMP;
 272         else
 273                 ip->flags &= ~BTRFS_INODE_NODUMP;
 274         if (flags & FS_NOATIME_FL)
 275                 ip->flags |= BTRFS_INODE_NOATIME;
 276         else
 277                 ip->flags &= ~BTRFS_INODE_NOATIME;
 278         if (flags & FS_DIRSYNC_FL)
 279                 ip->flags |= BTRFS_INODE_DIRSYNC;
 280         else
 281                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 282         if (flags & FS_NOCOW_FL) {
 283                 if (S_ISREG(mode)) {
 284                         /*
 285                          * It's safe to turn csums off here, no extents exist.
 286                          * Otherwise we want the flag to reflect the real COW
 287                          * status of the file and will not set it.
 288                          */
 289                         if (inode->i_size == 0)
 290                                 ip->flags |= BTRFS_INODE_NODATACOW
 291                                            | BTRFS_INODE_NODATASUM;
 292                 } else {
 293                         ip->flags |= BTRFS_INODE_NODATACOW;
 294                 }
 295         } else {
 296                 /*
 297                  * Revert back under same assuptions as above
 298                  */
 299                 if (S_ISREG(mode)) {
 300                         if (inode->i_size == 0)
 301                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 302                                              | BTRFS_INODE_NODATASUM);
 303                 } else {
 304                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 305                 }
 306         }
 307
 308         /*
 309          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 310          * flag may be changed automatically if compression code won't make
 311          * things smaller.
 312          */
 313         if (flags & FS_NOCOMP_FL) {
 314                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 315                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 316
 317                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 318                 if (ret && ret != -ENODATA)
 319                         goto out_drop;
 320         } else if (flags & FS_COMPR_FL) {
 321                 const char *comp;
 322
 323                 ip->flags |= BTRFS_INODE_COMPRESS;
 324                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 325
 326                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 327                         comp = "lzo";
 328                 else
 329                         comp = "zlib";
 330                 ret = btrfs_set_prop(inode, "btrfs.compression",
 331                                      comp, strlen(comp), 0);
 332                 if (ret)
 333                         goto out_drop;
 334
 335         } else {
 336                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 337                 if (ret && ret != -ENODATA)
 338                         goto out_drop;
 339                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 340         }
 341
 342         trans = btrfs_start_transaction(root, 1);
 343         if (IS_ERR(trans)) {
 344                 ret = PTR_ERR(trans);
 345                 goto out_drop;
 346         }
 347
 348         btrfs_update_iflags(inode);
 349         inode_inc_iversion(inode);
 350         inode->i_ctime = CURRENT_TIME;
 351         ret = btrfs_update_inode(trans, root, inode);
 352
 353         btrfs_end_transaction(trans, root);
 354  out_drop:
 355         if (ret) {
 356                 ip->flags = ip_oldflags;
 357                 inode->i_flags = i_oldflags;
 358         }
 359
 360  out_unlock:
 361         mutex_unlock(&inode->i_mutex);
 362         mnt_drop_write_file(file);
 363         return ret;
 364 }
 365
 366 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 367 {
 368         struct inode *inode = file_inode(file);
 369
 370         return put_user(inode->i_generation, arg);
 371 }
 372
 373 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 374 {
 375         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 376         struct btrfs_device *device;
 377         struct request_queue *q;
 378         struct fstrim_range range;
 379         u64 minlen = ULLONG_MAX;
 380         u64 num_devices = 0;
 381         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 382         int ret;
 383
 384         if (!capable(CAP_SYS_ADMIN))
 385                 return -EPERM;
 386
 387         rcu_read_lock();
 388         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 389                                 dev_list) {
 390                 if (!device->bdev)
 391                         continue;
 392                 q = bdev_get_queue(device->bdev);
 393                 if (blk_queue_discard(q)) {
 394                         num_devices++;
 395                         minlen = min((u64)q->limits.discard_granularity,
 396                                      minlen);
 397                 }
 398         }
 399         rcu_read_unlock();
 400
 401         if (!num_devices)
 402                 return -EOPNOTSUPP;
 403         if (copy_from_user(&range, arg, sizeof(range)))
 404                 return -EFAULT;
 405         if (range.start > total_bytes ||
 406             range.len < fs_info->sb->s_blocksize)
 407                 return -EINVAL;
 408
 409         range.len = min(range.len, total_bytes - range.start);
 410         range.minlen = max(range.minlen, minlen);
 411         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 412         if (ret < 0)
 413                 return ret;
 414
 415         if (copy_to_user(arg, &range, sizeof(range)))
 416                 return -EFAULT;
 417
 418         return 0;
 419 }
 420
 421 int btrfs_is_empty_uuid(u8 *uuid)
 422 {
 423         int i;
 424
 425         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 426                 if (uuid[i])
 427                         return 0;
 428         }
 429         return 1;
 430 }
 431
 432 static noinline int create_subvol(struct inode *dir,
 433                                   struct dentry *dentry,
 434                                   char *name, int namelen,
 435                                   u64 *async_transid,
 436                                   struct btrfs_qgroup_inherit *inherit)
 437 {
 438         struct btrfs_trans_handle *trans;
 439         struct btrfs_key key;
 440         struct btrfs_root_item root_item;
 441         struct btrfs_inode_item *inode_item;
 442         struct extent_buffer *leaf;
 443         struct btrfs_root *root = BTRFS_I(dir)->root;
 444         struct btrfs_root *new_root;
 445         struct btrfs_block_rsv block_rsv;
 446         struct timespec cur_time = CURRENT_TIME;
 447         struct inode *inode;
 448         int ret;
 449         int err;
 450         u64 objectid;
 451         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 452         u64 index = 0;
 453         u64 qgroup_reserved;
 454         uuid_le new_uuid;
 455
 456         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 457         if (ret)
 458                 return ret;
 459
 460         /*
 461          * Don't create subvolume whose level is not zero. Or qgroup will be
 462          * screwed up since it assume subvolme qgroup's level to be 0.
 463          */
 464         if (btrfs_qgroup_level(objectid))
 465                 return -ENOSPC;
 466
 467         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 468         /*
 469          * The same as the snapshot creation, please see the comment
 470          * of create_snapshot().
 471          */
 472         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 473                                                8, &qgroup_reserved, false);
 474         if (ret)
 475                 return ret;
 476
 477         trans = btrfs_start_transaction(root, 0);
 478         if (IS_ERR(trans)) {
 479                 ret = PTR_ERR(trans);
 480                 btrfs_subvolume_release_metadata(root, &block_rsv,
 481                                                  qgroup_reserved);
 482                 return ret;
 483         }
 484         trans->block_rsv = &block_rsv;
 485         trans->bytes_reserved = block_rsv.size;
 486
 487         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 488         if (ret)
 489                 goto fail;
 490
 491         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 492         if (IS_ERR(leaf)) {
 493                 ret = PTR_ERR(leaf);
 494                 goto fail;
 495         }
 496
 497         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 498         btrfs_set_header_bytenr(leaf, leaf->start);
 499         btrfs_set_header_generation(leaf, trans->transid);
 500         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 501         btrfs_set_header_owner(leaf, objectid);
 502
 503         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 504                             BTRFS_FSID_SIZE);
 505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 506                             btrfs_header_chunk_tree_uuid(leaf),
 507                             BTRFS_UUID_SIZE);
 508         btrfs_mark_buffer_dirty(leaf);
 509
 510         memset(&root_item, 0, sizeof(root_item));
 511
 512         inode_item = &root_item.inode;
 513         btrfs_set_stack_inode_generation(inode_item, 1);
 514         btrfs_set_stack_inode_size(inode_item, 3);
 515         btrfs_set_stack_inode_nlink(inode_item, 1);
 516         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 517         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 518
 519         btrfs_set_root_flags(&root_item, 0);
 520         btrfs_set_root_limit(&root_item, 0);
 521         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 522
 523         btrfs_set_root_bytenr(&root_item, leaf->start);
 524         btrfs_set_root_generation(&root_item, trans->transid);
 525         btrfs_set_root_level(&root_item, 0);
 526         btrfs_set_root_refs(&root_item, 1);
 527         btrfs_set_root_used(&root_item, leaf->len);
 528         btrfs_set_root_last_snapshot(&root_item, 0);
 529
 530         btrfs_set_root_generation_v2(&root_item,
 531                         btrfs_root_generation(&root_item));
 532         uuid_le_gen(&new_uuid);
 533         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 534         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 535         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 536         root_item.ctime = root_item.otime;
 537         btrfs_set_root_ctransid(&root_item, trans->transid);
 538         btrfs_set_root_otransid(&root_item, trans->transid);
 539
 540         btrfs_tree_unlock(leaf);
 541         free_extent_buffer(leaf);
 542         leaf = NULL;
 543
 544         btrfs_set_root_dirid(&root_item, new_dirid);
 545
 546         key.objectid = objectid;
 547         key.offset = 0;
 548         key.type = BTRFS_ROOT_ITEM_KEY;
 549         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 550                                 &root_item);
 551         if (ret)
 552                 goto fail;
 553
 554         key.offset = (u64)-1;
 555         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 556         if (IS_ERR(new_root)) {
 557                 ret = PTR_ERR(new_root);
 558                 btrfs_abort_transaction(trans, root, ret);
 559                 goto fail;
 560         }
 561
 562         btrfs_record_root_in_trans(trans, new_root);
 563
 564         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 565         if (ret) {
 566                 /* We potentially lose an unused inode item here */
 567                 btrfs_abort_transaction(trans, root, ret);
 568                 goto fail;
 569         }
 570
 571         mutex_lock(&new_root->objectid_mutex);
 572         new_root->highest_objectid = new_dirid;
 573         mutex_unlock(&new_root->objectid_mutex);
 574
 575         /*
 576          * insert the directory item
 577          */
 578         ret = btrfs_set_inode_index(dir, &index);
 579         if (ret) {
 580                 btrfs_abort_transaction(trans, root, ret);
 581                 goto fail;
 582         }
 583
 584         ret = btrfs_insert_dir_item(trans, root,
 585                                     name, namelen, dir, &key,
 586                                     BTRFS_FT_DIR, index);
 587         if (ret) {
 588                 btrfs_abort_transaction(trans, root, ret);
 589                 goto fail;
 590         }
 591
 592         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 593         ret = btrfs_update_inode(trans, root, dir);
 594         BUG_ON(ret);
 595
 596         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 597                                  objectid, root->root_key.objectid,
 598                                  btrfs_ino(dir), index, name, namelen);
 599         BUG_ON(ret);
 600
 601         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 602                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 603                                   objectid);
 604         if (ret)
 605                 btrfs_abort_transaction(trans, root, ret);
 606
 607 fail:
 608         trans->block_rsv = NULL;
 609         trans->bytes_reserved = 0;
 610         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 611
 612         if (async_transid) {
 613                 *async_transid = trans->transid;
 614                 err = btrfs_commit_transaction_async(trans, root, 1);
 615                 if (err)
 616                         err = btrfs_commit_transaction(trans, root);
 617         } else {
 618                 err = btrfs_commit_transaction(trans, root);
 619         }
 620         if (err && !ret)
 621                 ret = err;
 622
 623         if (!ret) {
 624                 inode = btrfs_lookup_dentry(dir, dentry);
 625                 if (IS_ERR(inode))
 626                         return PTR_ERR(inode);
 627                 d_instantiate(dentry, inode);
 628         }
 629         return ret;
 630 }
 631
 632 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 633 {
 634         s64 writers;
 635         DEFINE_WAIT(wait);
 636
 637         do {
 638                 prepare_to_wait(&root->subv_writers->wait, &wait,
 639                                 TASK_UNINTERRUPTIBLE);
 640
 641                 writers = percpu_counter_sum(&root->subv_writers->counter);
 642                 if (writers)
 643                         schedule();
 644
 645                 finish_wait(&root->subv_writers->wait, &wait);
 646         } while (writers);
 647 }
 648
 649 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 650                            struct dentry *dentry, char *name, int namelen,
 651                            u64 *async_transid, bool readonly,
 652                            struct btrfs_qgroup_inherit *inherit)
 653 {
 654         struct inode *inode;
 655         struct btrfs_pending_snapshot *pending_snapshot;
 656         struct btrfs_trans_handle *trans;
 657         int ret;
 658
 659         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 660                 return -EINVAL;
 661
 662         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 663         if (!pending_snapshot)
 664                 return -ENOMEM;
 665
 666         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 667                         GFP_NOFS);
 668         pending_snapshot->path = btrfs_alloc_path();
 669         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 670                 ret = -ENOMEM;
 671                 goto free_pending;
 672         }
 673
 674         atomic_inc(&root->will_be_snapshoted);
 675         smp_mb__after_atomic();
 676         btrfs_wait_for_no_snapshoting_writes(root);
 677
 678         ret = btrfs_start_delalloc_inodes(root, 0);
 679         if (ret)
 680                 goto dec_and_free;
 681
 682         btrfs_wait_ordered_extents(root, -1);
 683
 684         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 685                              BTRFS_BLOCK_RSV_TEMP);
 686         /*
 687          * 1 - parent dir inode
 688          * 2 - dir entries
 689          * 1 - root item
 690          * 2 - root ref/backref
 691          * 1 - root of snapshot
 692          * 1 - UUID item
 693          */
 694         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 695                                         &pending_snapshot->block_rsv, 8,
 696                                         &pending_snapshot->qgroup_reserved,
 697                                         false);
 698         if (ret)
 699                 goto dec_and_free;
 700
 701         pending_snapshot->dentry = dentry;
 702         pending_snapshot->root = root;
 703         pending_snapshot->readonly = readonly;
 704         pending_snapshot->dir = dir;
 705         pending_snapshot->inherit = inherit;
 706
 707         trans = btrfs_start_transaction(root, 0);
 708         if (IS_ERR(trans)) {
 709                 ret = PTR_ERR(trans);
 710                 goto fail;
 711         }
 712
 713         spin_lock(&root->fs_info->trans_lock);
 714         list_add(&pending_snapshot->list,
 715                  &trans->transaction->pending_snapshots);
 716         spin_unlock(&root->fs_info->trans_lock);
 717         if (async_transid) {
 718                 *async_transid = trans->transid;
 719                 ret = btrfs_commit_transaction_async(trans,
 720                                      root->fs_info->extent_root, 1);
 721                 if (ret)
 722                         ret = btrfs_commit_transaction(trans, root);
 723         } else {
 724                 ret = btrfs_commit_transaction(trans,
 725                                                root->fs_info->extent_root);
 726         }
 727         if (ret)
 728                 goto fail;
 729
 730         ret = pending_snapshot->error;
 731         if (ret)
 732                 goto fail;
 733
 734         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 735         if (ret)
 736                 goto fail;
 737
 738         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 739         if (IS_ERR(inode)) {
 740                 ret = PTR_ERR(inode);
 741                 goto fail;
 742         }
 743
 744         d_instantiate(dentry, inode);
 745         ret = 0;
 746 fail:
 747         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 748                                          &pending_snapshot->block_rsv,
 749                                          pending_snapshot->qgroup_reserved);
 750 dec_and_free:
 751         if (atomic_dec_and_test(&root->will_be_snapshoted))
 752                 wake_up_atomic_t(&root->will_be_snapshoted);
 753 free_pending:
 754         kfree(pending_snapshot->root_item);
 755         btrfs_free_path(pending_snapshot->path);
 756         kfree(pending_snapshot);
 757
 758         return ret;
 759 }
 760
 761 /*  copy of may_delete in fs/namei.c()
 762  *      Check whether we can remove a link victim from directory dir, check
 763  *  whether the type of victim is right.
 764  *  1. We can't do it if dir is read-only (done in permission())
 765  *  2. We should have write and exec permissions on dir
 766  *  3. We can't remove anything from append-only dir
 767  *  4. We can't do anything with immutable dir (done in permission())
 768  *  5. If the sticky bit on dir is set we should either
 769  *      a. be owner of dir, or
 770  *      b. be owner of victim, or
 771  *      c. have CAP_FOWNER capability
 772  *  6. If the victim is append-only or immutable we can't do antyhing with
 773  *     links pointing to it.
 774  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 775  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 776  *  9. We can't remove a root or mountpoint.
 777  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 778  *     nfs_async_unlink().
 779  */
 780
 781 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 782 {
 783         int error;
 784
 785         if (d_really_is_negative(victim))
 786                 return -ENOENT;
 787
 788         BUG_ON(d_inode(victim->d_parent) != dir);
 789         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 790
 791         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 792         if (error)
 793                 return error;
 794         if (IS_APPEND(dir))
 795                 return -EPERM;
 796         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 797             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 798                 return -EPERM;
 799         if (isdir) {
 800                 if (!d_is_dir(victim))
 801                         return -ENOTDIR;
 802                 if (IS_ROOT(victim))
 803                         return -EBUSY;
 804         } else if (d_is_dir(victim))
 805                 return -EISDIR;
 806         if (IS_DEADDIR(dir))
 807                 return -ENOENT;
 808         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 809                 return -EBUSY;
 810         return 0;
 811 }
 812
 813 /* copy of may_create in fs/namei.c() */
 814 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 815 {
 816         if (d_really_is_positive(child))
 817                 return -EEXIST;
 818         if (IS_DEADDIR(dir))
 819                 return -ENOENT;
 820         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 821 }
 822
 823 /*
 824  * Create a new subvolume below @parent.  This is largely modeled after
 825  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 826  * inside this filesystem so it's quite a bit simpler.
 827  */
 828 static noinline int btrfs_mksubvol(struct path *parent,
 829                                    char *name, int namelen,
 830                                    struct btrfs_root *snap_src,
 831                                    u64 *async_transid, bool readonly,
 832                                    struct btrfs_qgroup_inherit *inherit)
 833 {
 834         struct inode *dir  = d_inode(parent->dentry);
 835         struct dentry *dentry;
 836         int error;
 837
 838         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 839         if (error == -EINTR)
 840                 return error;
 841
 842         dentry = lookup_one_len(name, parent->dentry, namelen);
 843         error = PTR_ERR(dentry);
 844         if (IS_ERR(dentry))
 845                 goto out_unlock;
 846
 847         error = -EEXIST;
 848         if (d_really_is_positive(dentry))
 849                 goto out_dput;
 850
 851         error = btrfs_may_create(dir, dentry);
 852         if (error)
 853                 goto out_dput;
 854
 855         /*
 856          * even if this name doesn't exist, we may get hash collisions.
 857          * check for them now when we can safely fail
 858          */
 859         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 860                                                dir->i_ino, name,
 861                                                namelen);
 862         if (error)
 863                 goto out_dput;
 864
 865         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 866
 867         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 868                 goto out_up_read;
 869
 870         if (snap_src) {
 871                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 872                                         async_transid, readonly, inherit);
 873         } else {
 874                 error = create_subvol(dir, dentry, name, namelen,
 875                                       async_transid, inherit);
 876         }
 877         if (!error)
 878                 fsnotify_mkdir(dir, dentry);
 879 out_up_read:
 880         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 881 out_dput:
 882         dput(dentry);
 883 out_unlock:
 884         mutex_unlock(&dir->i_mutex);
 885         return error;
 886 }
 887
 888 /*
 889  * When we're defragging a range, we don't want to kick it off again
 890  * if it is really just waiting for delalloc to send it down.
 891  * If we find a nice big extent or delalloc range for the bytes in the
 892  * file you want to defrag, we return 0 to let you know to skip this
 893  * part of the file
 894  */
 895 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 896 {
 897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 898         struct extent_map *em = NULL;
 899         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 900         u64 end;
 901
 902         read_lock(&em_tree->lock);
 903         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 904         read_unlock(&em_tree->lock);
 905
 906         if (em) {
 907                 end = extent_map_end(em);
 908                 free_extent_map(em);
 909                 if (end - offset > thresh)
 910                         return 0;
 911         }
 912         /* if we already have a nice delalloc here, just stop */
 913         thresh /= 2;
 914         end = count_range_bits(io_tree, &offset, offset + thresh,
 915                                thresh, EXTENT_DELALLOC, 1);
 916         if (end >= thresh)
 917                 return 0;
 918         return 1;
 919 }
 920
 921 /*
 922  * helper function to walk through a file and find extents
 923  * newer than a specific transid, and smaller than thresh.
 924  *
 925  * This is used by the defragging code to find new and small
 926  * extents
 927  */
 928 static int find_new_extents(struct btrfs_root *root,
 929                             struct inode *inode, u64 newer_than,
 930                             u64 *off, u32 thresh)
 931 {
 932         struct btrfs_path *path;
 933         struct btrfs_key min_key;
 934         struct extent_buffer *leaf;
 935         struct btrfs_file_extent_item *extent;
 936         int type;
 937         int ret;
 938         u64 ino = btrfs_ino(inode);
 939
 940         path = btrfs_alloc_path();
 941         if (!path)
 942                 return -ENOMEM;
 943
 944         min_key.objectid = ino;
 945         min_key.type = BTRFS_EXTENT_DATA_KEY;
 946         min_key.offset = *off;
 947
 948         while (1) {
 949                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 950                 if (ret != 0)
 951                         goto none;
 952 process_slot:
 953                 if (min_key.objectid != ino)
 954                         goto none;
 955                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 956                         goto none;
 957
 958                 leaf = path->nodes[0];
 959                 extent = btrfs_item_ptr(leaf, path->slots[0],
 960                                         struct btrfs_file_extent_item);
 961
 962                 type = btrfs_file_extent_type(leaf, extent);
 963                 if (type == BTRFS_FILE_EXTENT_REG &&
 964                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 965                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 966                         *off = min_key.offset;
 967                         btrfs_free_path(path);
 968                         return 0;
 969                 }
 970
 971                 path->slots[0]++;
 972                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 973                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 974                         goto process_slot;
 975                 }
 976
 977                 if (min_key.offset == (u64)-1)
 978                         goto none;
 979
 980                 min_key.offset++;
 981                 btrfs_release_path(path);
 982         }
 983 none:
 984         btrfs_free_path(path);
 985         return -ENOENT;
 986 }
 987
 988 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 989 {
 990         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 991         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 992         struct extent_map *em;
 993         u64 len = PAGE_CACHE_SIZE;
 994
 995         /*
 996          * hopefully we have this extent in the tree already, try without
 997          * the full extent lock
 998          */
 999         read_lock(&em_tree->lock);
1000         em = lookup_extent_mapping(em_tree, start, len);
1001         read_unlock(&em_tree->lock);
1002
1003         if (!em) {
1004                 struct extent_state *cached = NULL;
1005                 u64 end = start + len - 1;
1006
1007                 /* get the big lock and read metadata off disk */
1008                 lock_extent_bits(io_tree, start, end, &cached);
1009                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1010                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1011
1012                 if (IS_ERR(em))
1013                         return NULL;
1014         }
1015
1016         return em;
1017 }
1018
1019 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1020 {
1021         struct extent_map *next;
1022         bool ret = true;
1023
1024         /* this is the last extent */
1025         if (em->start + em->len >= i_size_read(inode))
1026                 return false;
1027
1028         next = defrag_lookup_extent(inode, em->start + em->len);
1029         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1030                 ret = false;
1031         else if ((em->block_start + em->block_len == next->block_start) &&
1032                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1033                 ret = false;
1034
1035         free_extent_map(next);
1036         return ret;
1037 }
1038
1039 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1040                                u64 *last_len, u64 *skip, u64 *defrag_end,
1041                                int compress)
1042 {
1043         struct extent_map *em;
1044         int ret = 1;
1045         bool next_mergeable = true;
1046         bool prev_mergeable = true;
1047
1048         /*
1049          * make sure that once we start defragging an extent, we keep on
1050          * defragging it
1051          */
1052         if (start < *defrag_end)
1053                 return 1;
1054
1055         *skip = 0;
1056
1057         em = defrag_lookup_extent(inode, start);
1058         if (!em)
1059                 return 0;
1060
1061         /* this will cover holes, and inline extents */
1062         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1063                 ret = 0;
1064                 goto out;
1065         }
1066
1067         if (!*defrag_end)
1068                 prev_mergeable = false;
1069
1070         next_mergeable = defrag_check_next_extent(inode, em);
1071         /*
1072          * we hit a real extent, if it is big or the next extent is not a
1073          * real extent, don't bother defragging it
1074          */
1075         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1076             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1077                 ret = 0;
1078 out:
1079         /*
1080          * last_len ends up being a counter of how many bytes we've defragged.
1081          * every time we choose not to defrag an extent, we reset *last_len
1082          * so that the next tiny extent will force a defrag.
1083          *
1084          * The end result of this is that tiny extents before a single big
1085          * extent will force at least part of that big extent to be defragged.
1086          */
1087         if (ret) {
1088                 *defrag_end = extent_map_end(em);
1089         } else {
1090                 *last_len = 0;
1091                 *skip = extent_map_end(em);
1092                 *defrag_end = 0;
1093         }
1094
1095         free_extent_map(em);
1096         return ret;
1097 }
1098
1099 /*
1100  * it doesn't do much good to defrag one or two pages
1101  * at a time.  This pulls in a nice chunk of pages
1102  * to COW and defrag.
1103  *
1104  * It also makes sure the delalloc code has enough
1105  * dirty data to avoid making new small extents as part
1106  * of the defrag
1107  *
1108  * It's a good idea to start RA on this range
1109  * before calling this.
1110  */
1111 static int cluster_pages_for_defrag(struct inode *inode,
1112                                     struct page **pages,
1113                                     unsigned long start_index,
1114                                     unsigned long num_pages)
1115 {
1116         unsigned long file_end;
1117         u64 isize = i_size_read(inode);
1118         u64 page_start;
1119         u64 page_end;
1120         u64 page_cnt;
1121         int ret;
1122         int i;
1123         int i_done;
1124         struct btrfs_ordered_extent *ordered;
1125         struct extent_state *cached_state = NULL;
1126         struct extent_io_tree *tree;
1127         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1128
1129         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1130         if (!isize || start_index > file_end)
1131                 return 0;
1132
1133         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1134
1135         ret = btrfs_delalloc_reserve_space(inode,
1136                         start_index << PAGE_CACHE_SHIFT,
1137                         page_cnt << PAGE_CACHE_SHIFT);
1138         if (ret)
1139                 return ret;
1140         i_done = 0;
1141         tree = &BTRFS_I(inode)->io_tree;
1142
1143         /* step one, lock all the pages */
1144         for (i = 0; i < page_cnt; i++) {
1145                 struct page *page;
1146 again:
1147                 page = find_or_create_page(inode->i_mapping,
1148                                            start_index + i, mask);
1149                 if (!page)
1150                         break;
1151
1152                 page_start = page_offset(page);
1153                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1154                 while (1) {
1155                         lock_extent_bits(tree, page_start, page_end,
1156                                          &cached_state);
1157                         ordered = btrfs_lookup_ordered_extent(inode,
1158                                                               page_start);
1159                         unlock_extent_cached(tree, page_start, page_end,
1160                                              &cached_state, GFP_NOFS);
1161                         if (!ordered)
1162                                 break;
1163
1164                         unlock_page(page);
1165                         btrfs_start_ordered_extent(inode, ordered, 1);
1166                         btrfs_put_ordered_extent(ordered);
1167                         lock_page(page);
1168                         /*
1169                          * we unlocked the page above, so we need check if
1170                          * it was released or not.
1171                          */
1172                         if (page->mapping != inode->i_mapping) {
1173                                 unlock_page(page);
1174                                 page_cache_release(page);
1175                                 goto again;
1176                         }
1177                 }
1178
1179                 if (!PageUptodate(page)) {
1180                         btrfs_readpage(NULL, page);
1181                         lock_page(page);
1182                         if (!PageUptodate(page)) {
1183                                 unlock_page(page);
1184                                 page_cache_release(page);
1185                                 ret = -EIO;
1186                                 break;
1187                         }
1188                 }
1189
1190                 if (page->mapping != inode->i_mapping) {
1191                         unlock_page(page);
1192                         page_cache_release(page);
1193                         goto again;
1194                 }
1195
1196                 pages[i] = page;
1197                 i_done++;
1198         }
1199         if (!i_done || ret)
1200                 goto out;
1201
1202         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1203                 goto out;
1204
1205         /*
1206          * so now we have a nice long stream of locked
1207          * and up to date pages, lets wait on them
1208          */
1209         for (i = 0; i < i_done; i++)
1210                 wait_on_page_writeback(pages[i]);
1211
1212         page_start = page_offset(pages[0]);
1213         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1214
1215         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1216                          page_start, page_end - 1, &cached_state);
1217         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1218                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1219                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1220                           &cached_state, GFP_NOFS);
1221
1222         if (i_done != page_cnt) {
1223                 spin_lock(&BTRFS_I(inode)->lock);
1224                 BTRFS_I(inode)->outstanding_extents++;
1225                 spin_unlock(&BTRFS_I(inode)->lock);
1226                 btrfs_delalloc_release_space(inode,
1227                                 start_index << PAGE_CACHE_SHIFT,
1228                                 (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1229         }
1230
1231
1232         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1233                           &cached_state, GFP_NOFS);
1234
1235         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1236                              page_start, page_end - 1, &cached_state,
1237                              GFP_NOFS);
1238
1239         for (i = 0; i < i_done; i++) {
1240                 clear_page_dirty_for_io(pages[i]);
1241                 ClearPageChecked(pages[i]);
1242                 set_page_extent_mapped(pages[i]);
1243                 set_page_dirty(pages[i]);
1244                 unlock_page(pages[i]);
1245                 page_cache_release(pages[i]);
1246         }
1247         return i_done;
1248 out:
1249         for (i = 0; i < i_done; i++) {
1250                 unlock_page(pages[i]);
1251                 page_cache_release(pages[i]);
1252         }
1253         btrfs_delalloc_release_space(inode,
1254                         start_index << PAGE_CACHE_SHIFT,
1255                         page_cnt << PAGE_CACHE_SHIFT);
1256         return ret;
1257
1258 }
1259
1260 int btrfs_defrag_file(struct inode *inode, struct file *file,
1261                       struct btrfs_ioctl_defrag_range_args *range,
1262                       u64 newer_than, unsigned long max_to_defrag)
1263 {
1264         struct btrfs_root *root = BTRFS_I(inode)->root;
1265         struct file_ra_state *ra = NULL;
1266         unsigned long last_index;
1267         u64 isize = i_size_read(inode);
1268         u64 last_len = 0;
1269         u64 skip = 0;
1270         u64 defrag_end = 0;
1271         u64 newer_off = range->start;
1272         unsigned long i;
1273         unsigned long ra_index = 0;
1274         int ret;
1275         int defrag_count = 0;
1276         int compress_type = BTRFS_COMPRESS_ZLIB;
1277         u32 extent_thresh = range->extent_thresh;
1278         unsigned long max_cluster = SZ_256K >> PAGE_CACHE_SHIFT;
1279         unsigned long cluster = max_cluster;
1280         u64 new_align = ~((u64)SZ_128K - 1);
1281         struct page **pages = NULL;
1282
1283         if (isize == 0)
1284                 return 0;
1285
1286         if (range->start >= isize)
1287                 return -EINVAL;
1288
1289         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1290                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1291                         return -EINVAL;
1292                 if (range->compress_type)
1293                         compress_type = range->compress_type;
1294         }
1295
1296         if (extent_thresh == 0)
1297                 extent_thresh = SZ_256K;
1298
1299         /*
1300          * if we were not given a file, allocate a readahead
1301          * context
1302          */
1303         if (!file) {
1304                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1305                 if (!ra)
1306                         return -ENOMEM;
1307                 file_ra_state_init(ra, inode->i_mapping);
1308         } else {
1309                 ra = &file->f_ra;
1310         }
1311
1312         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1313                         GFP_NOFS);
1314         if (!pages) {
1315                 ret = -ENOMEM;
1316                 goto out_ra;
1317         }
1318
1319         /* find the last page to defrag */
1320         if (range->start + range->len > range->start) {
1321                 last_index = min_t(u64, isize - 1,
1322                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1323         } else {
1324                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1325         }
1326
1327         if (newer_than) {
1328                 ret = find_new_extents(root, inode, newer_than,
1329                                        &newer_off, SZ_64K);
1330                 if (!ret) {
1331                         range->start = newer_off;
1332                         /*
1333                          * we always align our defrag to help keep
1334                          * the extents in the file evenly spaced
1335                          */
1336                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1337                 } else
1338                         goto out_ra;
1339         } else {
1340                 i = range->start >> PAGE_CACHE_SHIFT;
1341         }
1342         if (!max_to_defrag)
1343                 max_to_defrag = last_index - i + 1;
1344
1345         /*
1346          * make writeback starts from i, so the defrag range can be
1347          * written sequentially.
1348          */
1349         if (i < inode->i_mapping->writeback_index)
1350                 inode->i_mapping->writeback_index = i;
1351
1352         while (i <= last_index && defrag_count < max_to_defrag &&
1353                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1354                 /*
1355                  * make sure we stop running if someone unmounts
1356                  * the FS
1357                  */
1358                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1359                         break;
1360
1361                 if (btrfs_defrag_cancelled(root->fs_info)) {
1362                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1363                         ret = -EAGAIN;
1364                         break;
1365                 }
1366
1367                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1368                                          extent_thresh, &last_len, &skip,
1369                                          &defrag_end, range->flags &
1370                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1371                         unsigned long next;
1372                         /*
1373                          * the should_defrag function tells us how much to skip
1374                          * bump our counter by the suggested amount
1375                          */
1376                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1377                         i = max(i + 1, next);
1378                         continue;
1379                 }
1380
1381                 if (!newer_than) {
1382                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1383                                    PAGE_CACHE_SHIFT) - i;
1384                         cluster = min(cluster, max_cluster);
1385                 } else {
1386                         cluster = max_cluster;
1387                 }
1388
1389                 if (i + cluster > ra_index) {
1390                         ra_index = max(i, ra_index);
1391                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1392                                        cluster);
1393                         ra_index += cluster;
1394                 }
1395
1396                 mutex_lock(&inode->i_mutex);
1397                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1398                         BTRFS_I(inode)->force_compress = compress_type;
1399                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1400                 if (ret < 0) {
1401                         mutex_unlock(&inode->i_mutex);
1402                         goto out_ra;
1403                 }
1404
1405                 defrag_count += ret;
1406                 balance_dirty_pages_ratelimited(inode->i_mapping);
1407                 mutex_unlock(&inode->i_mutex);
1408
1409                 if (newer_than) {
1410                         if (newer_off == (u64)-1)
1411                                 break;
1412
1413                         if (ret > 0)
1414                                 i += ret;
1415
1416                         newer_off = max(newer_off + 1,
1417                                         (u64)i << PAGE_CACHE_SHIFT);
1418
1419                         ret = find_new_extents(root, inode, newer_than,
1420                                                &newer_off, SZ_64K);
1421                         if (!ret) {
1422                                 range->start = newer_off;
1423                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1424                         } else {
1425                                 break;
1426                         }
1427                 } else {
1428                         if (ret > 0) {
1429                                 i += ret;
1430                                 last_len += ret << PAGE_CACHE_SHIFT;
1431                         } else {
1432                                 i++;
1433                                 last_len = 0;
1434                         }
1435                 }
1436         }
1437
1438         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1439                 filemap_flush(inode->i_mapping);
1440                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1441                              &BTRFS_I(inode)->runtime_flags))
1442                         filemap_flush(inode->i_mapping);
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1446                 /* the filemap_flush will queue IO into the worker threads, but
1447                  * we have to make sure the IO is actually started and that
1448                  * ordered extents get created before we return
1449                  */
1450                 atomic_inc(&root->fs_info->async_submit_draining);
1451                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1452                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1453                         wait_event(root->fs_info->async_submit_wait,
1454                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1455                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1456                 }
1457                 atomic_dec(&root->fs_info->async_submit_draining);
1458         }
1459
1460         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1461                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1462         }
1463
1464         ret = defrag_count;
1465
1466 out_ra:
1467         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1468                 mutex_lock(&inode->i_mutex);
1469                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1470                 mutex_unlock(&inode->i_mutex);
1471         }
1472         if (!file)
1473                 kfree(ra);
1474         kfree(pages);
1475         return ret;
1476 }
1477
1478 static noinline int btrfs_ioctl_resize(struct file *file,
1479                                         void __user *arg)
1480 {
1481         u64 new_size;
1482         u64 old_size;
1483         u64 devid = 1;
1484         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1485         struct btrfs_ioctl_vol_args *vol_args;
1486         struct btrfs_trans_handle *trans;
1487         struct btrfs_device *device = NULL;
1488         char *sizestr;
1489         char *retptr;
1490         char *devstr = NULL;
1491         int ret = 0;
1492         int mod = 0;
1493
1494         if (!capable(CAP_SYS_ADMIN))
1495                 return -EPERM;
1496
1497         ret = mnt_want_write_file(file);
1498         if (ret)
1499                 return ret;
1500
1501         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1502                         1)) {
1503                 mnt_drop_write_file(file);
1504                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1505         }
1506
1507         mutex_lock(&root->fs_info->volume_mutex);
1508         vol_args = memdup_user(arg, sizeof(*vol_args));
1509         if (IS_ERR(vol_args)) {
1510                 ret = PTR_ERR(vol_args);
1511                 goto out;
1512         }
1513
1514         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1515
1516         sizestr = vol_args->name;
1517         devstr = strchr(sizestr, ':');
1518         if (devstr) {
1519                 sizestr = devstr + 1;
1520                 *devstr = '\0';
1521                 devstr = vol_args->name;
1522                 ret = kstrtoull(devstr, 10, &devid);
1523                 if (ret)
1524                         goto out_free;
1525                 if (!devid) {
1526                         ret = -EINVAL;
1527                         goto out_free;
1528                 }
1529                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1530         }
1531
1532         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1533         if (!device) {
1534                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1535                        devid);
1536                 ret = -ENODEV;
1537                 goto out_free;
1538         }
1539
1540         if (!device->writeable) {
1541                 btrfs_info(root->fs_info,
1542                            "resizer unable to apply on readonly device %llu",
1543                        devid);
1544                 ret = -EPERM;
1545                 goto out_free;
1546         }
1547
1548         if (!strcmp(sizestr, "max"))
1549                 new_size = device->bdev->bd_inode->i_size;
1550         else {
1551                 if (sizestr[0] == '-') {
1552                         mod = -1;
1553                         sizestr++;
1554                 } else if (sizestr[0] == '+') {
1555                         mod = 1;
1556                         sizestr++;
1557                 }
1558                 new_size = memparse(sizestr, &retptr);
1559                 if (*retptr != '\0' || new_size == 0) {
1560                         ret = -EINVAL;
1561                         goto out_free;
1562                 }
1563         }
1564
1565         if (device->is_tgtdev_for_dev_replace) {
1566                 ret = -EPERM;
1567                 goto out_free;
1568         }
1569
1570         old_size = btrfs_device_get_total_bytes(device);
1571
1572         if (mod < 0) {
1573                 if (new_size > old_size) {
1574                         ret = -EINVAL;
1575                         goto out_free;
1576                 }
1577                 new_size = old_size - new_size;
1578         } else if (mod > 0) {
1579                 if (new_size > ULLONG_MAX - old_size) {
1580                         ret = -ERANGE;
1581                         goto out_free;
1582                 }
1583                 new_size = old_size + new_size;
1584         }
1585
1586         if (new_size < SZ_256M) {
1587                 ret = -EINVAL;
1588                 goto out_free;
1589         }
1590         if (new_size > device->bdev->bd_inode->i_size) {
1591                 ret = -EFBIG;
1592                 goto out_free;
1593         }
1594
1595         new_size = div_u64(new_size, root->sectorsize);
1596         new_size *= root->sectorsize;
1597
1598         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1599                       rcu_str_deref(device->name), new_size);
1600
1601         if (new_size > old_size) {
1602                 trans = btrfs_start_transaction(root, 0);
1603                 if (IS_ERR(trans)) {
1604                         ret = PTR_ERR(trans);
1605                         goto out_free;
1606                 }
1607                 ret = btrfs_grow_device(trans, device, new_size);
1608                 btrfs_commit_transaction(trans, root);
1609         } else if (new_size < old_size) {
1610                 ret = btrfs_shrink_device(device, new_size);
1611         } /* equal, nothing need to do */
1612
1613 out_free:
1614         kfree(vol_args);
1615 out:
1616         mutex_unlock(&root->fs_info->volume_mutex);
1617         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1618         mnt_drop_write_file(file);
1619         return ret;
1620 }
1621
1622 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1623                                 char *name, unsigned long fd, int subvol,
1624                                 u64 *transid, bool readonly,
1625                                 struct btrfs_qgroup_inherit *inherit)
1626 {
1627         int namelen;
1628         int ret = 0;
1629
1630         ret = mnt_want_write_file(file);
1631         if (ret)
1632                 goto out;
1633
1634         namelen = strlen(name);
1635         if (strchr(name, '/')) {
1636                 ret = -EINVAL;
1637                 goto out_drop_write;
1638         }
1639
1640         if (name[0] == '.' &&
1641            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1642                 ret = -EEXIST;
1643                 goto out_drop_write;
1644         }
1645
1646         if (subvol) {
1647                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1648                                      NULL, transid, readonly, inherit);
1649         } else {
1650                 struct fd src = fdget(fd);
1651                 struct inode *src_inode;
1652                 if (!src.file) {
1653                         ret = -EINVAL;
1654                         goto out_drop_write;
1655                 }
1656
1657                 src_inode = file_inode(src.file);
1658                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1659                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1660                                    "Snapshot src from another FS");
1661                         ret = -EXDEV;
1662                 } else if (!inode_owner_or_capable(src_inode)) {
1663                         /*
1664                          * Subvolume creation is not restricted, but snapshots
1665                          * are limited to own subvolumes only
1666                          */
1667                         ret = -EPERM;
1668                 } else {
1669                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1670                                              BTRFS_I(src_inode)->root,
1671                                              transid, readonly, inherit);
1672                 }
1673                 fdput(src);
1674         }
1675 out_drop_write:
1676         mnt_drop_write_file(file);
1677 out:
1678         return ret;
1679 }
1680
1681 static noinline int btrfs_ioctl_snap_create(struct file *file,
1682                                             void __user *arg, int subvol)
1683 {
1684         struct btrfs_ioctl_vol_args *vol_args;
1685         int ret;
1686
1687         vol_args = memdup_user(arg, sizeof(*vol_args));
1688         if (IS_ERR(vol_args))
1689                 return PTR_ERR(vol_args);
1690         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1691
1692         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1693                                               vol_args->fd, subvol,
1694                                               NULL, false, NULL);
1695
1696         kfree(vol_args);
1697         return ret;
1698 }
1699
1700 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1701                                                void __user *arg, int subvol)
1702 {
1703         struct btrfs_ioctl_vol_args_v2 *vol_args;
1704         int ret;
1705         u64 transid = 0;
1706         u64 *ptr = NULL;
1707         bool readonly = false;
1708         struct btrfs_qgroup_inherit *inherit = NULL;
1709
1710         vol_args = memdup_user(arg, sizeof(*vol_args));
1711         if (IS_ERR(vol_args))
1712                 return PTR_ERR(vol_args);
1713         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1714
1715         if (vol_args->flags &
1716             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1717               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1718                 ret = -EOPNOTSUPP;
1719                 goto free_args;
1720         }
1721
1722         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1723                 ptr = &transid;
1724         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1725                 readonly = true;
1726         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1727                 if (vol_args->size > PAGE_CACHE_SIZE) {
1728                         ret = -EINVAL;
1729                         goto free_args;
1730                 }
1731                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1732                 if (IS_ERR(inherit)) {
1733                         ret = PTR_ERR(inherit);
1734                         goto free_args;
1735                 }
1736         }
1737
1738         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1739                                               vol_args->fd, subvol, ptr,
1740                                               readonly, inherit);
1741         if (ret)
1742                 goto free_inherit;
1743
1744         if (ptr && copy_to_user(arg +
1745                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1746                                         transid),
1747                                 ptr, sizeof(*ptr)))
1748                 ret = -EFAULT;
1749
1750 free_inherit:
1751         kfree(inherit);
1752 free_args:
1753         kfree(vol_args);
1754         return ret;
1755 }
1756
1757 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1758                                                 void __user *arg)
1759 {
1760         struct inode *inode = file_inode(file);
1761         struct btrfs_root *root = BTRFS_I(inode)->root;
1762         int ret = 0;
1763         u64 flags = 0;
1764
1765         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1766                 return -EINVAL;
1767
1768         down_read(&root->fs_info->subvol_sem);
1769         if (btrfs_root_readonly(root))
1770                 flags |= BTRFS_SUBVOL_RDONLY;
1771         up_read(&root->fs_info->subvol_sem);
1772
1773         if (copy_to_user(arg, &flags, sizeof(flags)))
1774                 ret = -EFAULT;
1775
1776         return ret;
1777 }
1778
1779 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1780                                               void __user *arg)
1781 {
1782         struct inode *inode = file_inode(file);
1783         struct btrfs_root *root = BTRFS_I(inode)->root;
1784         struct btrfs_trans_handle *trans;
1785         u64 root_flags;
1786         u64 flags;
1787         int ret = 0;
1788
1789         if (!inode_owner_or_capable(inode))
1790                 return -EPERM;
1791
1792         ret = mnt_want_write_file(file);
1793         if (ret)
1794                 goto out;
1795
1796         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1797                 ret = -EINVAL;
1798                 goto out_drop_write;
1799         }
1800
1801         if (copy_from_user(&flags, arg, sizeof(flags))) {
1802                 ret = -EFAULT;
1803                 goto out_drop_write;
1804         }
1805
1806         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1807                 ret = -EINVAL;
1808                 goto out_drop_write;
1809         }
1810
1811         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1812                 ret = -EOPNOTSUPP;
1813                 goto out_drop_write;
1814         }
1815
1816         down_write(&root->fs_info->subvol_sem);
1817
1818         /* nothing to do */
1819         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1820                 goto out_drop_sem;
1821
1822         root_flags = btrfs_root_flags(&root->root_item);
1823         if (flags & BTRFS_SUBVOL_RDONLY) {
1824                 btrfs_set_root_flags(&root->root_item,
1825                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1826         } else {
1827                 /*
1828                  * Block RO -> RW transition if this subvolume is involved in
1829                  * send
1830                  */
1831                 spin_lock(&root->root_item_lock);
1832                 if (root->send_in_progress == 0) {
1833                         btrfs_set_root_flags(&root->root_item,
1834                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1835                         spin_unlock(&root->root_item_lock);
1836                 } else {
1837                         spin_unlock(&root->root_item_lock);
1838                         btrfs_warn(root->fs_info,
1839                         "Attempt to set subvolume %llu read-write during send",
1840                                         root->root_key.objectid);
1841                         ret = -EPERM;
1842                         goto out_drop_sem;
1843                 }
1844         }
1845
1846         trans = btrfs_start_transaction(root, 1);
1847         if (IS_ERR(trans)) {
1848                 ret = PTR_ERR(trans);
1849                 goto out_reset;
1850         }
1851
1852         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1853                                 &root->root_key, &root->root_item);
1854
1855         btrfs_commit_transaction(trans, root);
1856 out_reset:
1857         if (ret)
1858                 btrfs_set_root_flags(&root->root_item, root_flags);
1859 out_drop_sem:
1860         up_write(&root->fs_info->subvol_sem);
1861 out_drop_write:
1862         mnt_drop_write_file(file);
1863 out:
1864         return ret;
1865 }
1866
1867 /*
1868  * helper to check if the subvolume references other subvolumes
1869  */
1870 static noinline int may_destroy_subvol(struct btrfs_root *root)
1871 {
1872         struct btrfs_path *path;
1873         struct btrfs_dir_item *di;
1874         struct btrfs_key key;
1875         u64 dir_id;
1876         int ret;
1877
1878         path = btrfs_alloc_path();
1879         if (!path)
1880                 return -ENOMEM;
1881
1882         /* Make sure this root isn't set as the default subvol */
1883         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1884         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1885                                    dir_id, "default", 7, 0);
1886         if (di && !IS_ERR(di)) {
1887                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1888                 if (key.objectid == root->root_key.objectid) {
1889                         ret = -EPERM;
1890                         btrfs_err(root->fs_info, "deleting default subvolume "
1891                                   "%llu is not allowed", key.objectid);
1892                         goto out;
1893                 }
1894                 btrfs_release_path(path);
1895         }
1896
1897         key.objectid = root->root_key.objectid;
1898         key.type = BTRFS_ROOT_REF_KEY;
1899         key.offset = (u64)-1;
1900
1901         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1902                                 &key, path, 0, 0);
1903         if (ret < 0)
1904                 goto out;
1905         BUG_ON(ret == 0);
1906
1907         ret = 0;
1908         if (path->slots[0] > 0) {
1909                 path->slots[0]--;
1910                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1911                 if (key.objectid == root->root_key.objectid &&
1912                     key.type == BTRFS_ROOT_REF_KEY)
1913                         ret = -ENOTEMPTY;
1914         }
1915 out:
1916         btrfs_free_path(path);
1917         return ret;
1918 }
1919
1920 static noinline int key_in_sk(struct btrfs_key *key,
1921                               struct btrfs_ioctl_search_key *sk)
1922 {
1923         struct btrfs_key test;
1924         int ret;
1925
1926         test.objectid = sk->min_objectid;
1927         test.type = sk->min_type;
1928         test.offset = sk->min_offset;
1929
1930         ret = btrfs_comp_cpu_keys(key, &test);
1931         if (ret < 0)
1932                 return 0;
1933
1934         test.objectid = sk->max_objectid;
1935         test.type = sk->max_type;
1936         test.offset = sk->max_offset;
1937
1938         ret = btrfs_comp_cpu_keys(key, &test);
1939         if (ret > 0)
1940                 return 0;
1941         return 1;
1942 }
1943
1944 static noinline int copy_to_sk(struct btrfs_root *root,
1945                                struct btrfs_path *path,
1946                                struct btrfs_key *key,
1947                                struct btrfs_ioctl_search_key *sk,
1948                                size_t *buf_size,
1949                                char __user *ubuf,
1950                                unsigned long *sk_offset,
1951                                int *num_found)
1952 {
1953         u64 found_transid;
1954         struct extent_buffer *leaf;
1955         struct btrfs_ioctl_search_header sh;
1956         struct btrfs_key test;
1957         unsigned long item_off;
1958         unsigned long item_len;
1959         int nritems;
1960         int i;
1961         int slot;
1962         int ret = 0;
1963
1964         leaf = path->nodes[0];
1965         slot = path->slots[0];
1966         nritems = btrfs_header_nritems(leaf);
1967
1968         if (btrfs_header_generation(leaf) > sk->max_transid) {
1969                 i = nritems;
1970                 goto advance_key;
1971         }
1972         found_transid = btrfs_header_generation(leaf);
1973
1974         for (i = slot; i < nritems; i++) {
1975                 item_off = btrfs_item_ptr_offset(leaf, i);
1976                 item_len = btrfs_item_size_nr(leaf, i);
1977
1978                 btrfs_item_key_to_cpu(leaf, key, i);
1979                 if (!key_in_sk(key, sk))
1980                         continue;
1981
1982                 if (sizeof(sh) + item_len > *buf_size) {
1983                         if (*num_found) {
1984                                 ret = 1;
1985                                 goto out;
1986                         }
1987
1988                         /*
1989                          * return one empty item back for v1, which does not
1990                          * handle -EOVERFLOW
1991                          */
1992
1993                         *buf_size = sizeof(sh) + item_len;
1994                         item_len = 0;
1995                         ret = -EOVERFLOW;
1996                 }
1997
1998                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1999                         ret = 1;
2000                         goto out;
2001                 }
2002
2003                 sh.objectid = key->objectid;
2004                 sh.offset = key->offset;
2005                 sh.type = key->type;
2006                 sh.len = item_len;
2007                 sh.transid = found_transid;
2008
2009                 /* copy search result header */
2010                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2011                         ret = -EFAULT;
2012                         goto out;
2013                 }
2014
2015                 *sk_offset += sizeof(sh);
2016
2017                 if (item_len) {
2018                         char __user *up = ubuf + *sk_offset;
2019                         /* copy the item */
2020                         if (read_extent_buffer_to_user(leaf, up,
2021                                                        item_off, item_len)) {
2022                                 ret = -EFAULT;
2023                                 goto out;
2024                         }
2025
2026                         *sk_offset += item_len;
2027                 }
2028                 (*num_found)++;
2029
2030                 if (ret) /* -EOVERFLOW from above */
2031                         goto out;
2032
2033                 if (*num_found >= sk->nr_items) {
2034                         ret = 1;
2035                         goto out;
2036                 }
2037         }
2038 advance_key:
2039         ret = 0;
2040         test.objectid = sk->max_objectid;
2041         test.type = sk->max_type;
2042         test.offset = sk->max_offset;
2043         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2044                 ret = 1;
2045         else if (key->offset < (u64)-1)
2046                 key->offset++;
2047         else if (key->type < (u8)-1) {
2048                 key->offset = 0;
2049                 key->type++;
2050         } else if (key->objectid < (u64)-1) {
2051                 key->offset = 0;
2052                 key->type = 0;
2053                 key->objectid++;
2054         } else
2055                 ret = 1;
2056 out:
2057         /*
2058          *  0: all items from this leaf copied, continue with next
2059          *  1: * more items can be copied, but unused buffer is too small
2060          *     * all items were found
2061          *     Either way, it will stops the loop which iterates to the next
2062          *     leaf
2063          *  -EOVERFLOW: item was to large for buffer
2064          *  -EFAULT: could not copy extent buffer back to userspace
2065          */
2066         return ret;
2067 }
2068
2069 static noinline int search_ioctl(struct inode *inode,
2070                                  struct btrfs_ioctl_search_key *sk,
2071                                  size_t *buf_size,
2072                                  char __user *ubuf)
2073 {
2074         struct btrfs_root *root;
2075         struct btrfs_key key;
2076         struct btrfs_path *path;
2077         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2078         int ret;
2079         int num_found = 0;
2080         unsigned long sk_offset = 0;
2081
2082         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2083                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2084                 return -EOVERFLOW;
2085         }
2086
2087         path = btrfs_alloc_path();
2088         if (!path)
2089                 return -ENOMEM;
2090
2091         if (sk->tree_id == 0) {
2092                 /* search the root of the inode that was passed */
2093                 root = BTRFS_I(inode)->root;
2094         } else {
2095                 key.objectid = sk->tree_id;
2096                 key.type = BTRFS_ROOT_ITEM_KEY;
2097                 key.offset = (u64)-1;
2098                 root = btrfs_read_fs_root_no_name(info, &key);
2099                 if (IS_ERR(root)) {
2100                         btrfs_err(info, "could not find root %llu",
2101                                sk->tree_id);
2102                         btrfs_free_path(path);
2103                         return -ENOENT;
2104                 }
2105         }
2106
2107         key.objectid = sk->min_objectid;
2108         key.type = sk->min_type;
2109         key.offset = sk->min_offset;
2110
2111         while (1) {
2112                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2113                 if (ret != 0) {
2114                         if (ret > 0)
2115                                 ret = 0;
2116                         goto err;
2117                 }
2118                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2119                                  &sk_offset, &num_found);
2120                 btrfs_release_path(path);
2121                 if (ret)
2122                         break;
2123
2124         }
2125         if (ret > 0)
2126                 ret = 0;
2127 err:
2128         sk->nr_items = num_found;
2129         btrfs_free_path(path);
2130         return ret;
2131 }
2132
2133 static noinline int btrfs_ioctl_tree_search(struct file *file,
2134                                            void __user *argp)
2135 {
2136         struct btrfs_ioctl_search_args __user *uargs;
2137         struct btrfs_ioctl_search_key sk;
2138         struct inode *inode;
2139         int ret;
2140         size_t buf_size;
2141
2142         if (!capable(CAP_SYS_ADMIN))
2143                 return -EPERM;
2144
2145         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2146
2147         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2148                 return -EFAULT;
2149
2150         buf_size = sizeof(uargs->buf);
2151
2152         inode = file_inode(file);
2153         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2154
2155         /*
2156          * In the origin implementation an overflow is handled by returning a
2157          * search header with a len of zero, so reset ret.
2158          */
2159         if (ret == -EOVERFLOW)
2160                 ret = 0;
2161
2162         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2163                 ret = -EFAULT;
2164         return ret;
2165 }
2166
2167 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2168                                                void __user *argp)
2169 {
2170         struct btrfs_ioctl_search_args_v2 __user *uarg;
2171         struct btrfs_ioctl_search_args_v2 args;
2172         struct inode *inode;
2173         int ret;
2174         size_t buf_size;
2175         const size_t buf_limit = SZ_16M;
2176
2177         if (!capable(CAP_SYS_ADMIN))
2178                 return -EPERM;
2179
2180         /* copy search header and buffer size */
2181         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2182         if (copy_from_user(&args, uarg, sizeof(args)))
2183                 return -EFAULT;
2184
2185         buf_size = args.buf_size;
2186
2187         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2188                 return -EOVERFLOW;
2189
2190         /* limit result size to 16MB */
2191         if (buf_size > buf_limit)
2192                 buf_size = buf_limit;
2193
2194         inode = file_inode(file);
2195         ret = search_ioctl(inode, &args.key, &buf_size,
2196                            (char *)(&uarg->buf[0]));
2197         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2198                 ret = -EFAULT;
2199         else if (ret == -EOVERFLOW &&
2200                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2201                 ret = -EFAULT;
2202
2203         return ret;
2204 }
2205
2206 /*
2207  * Search INODE_REFs to identify path name of 'dirid' directory
2208  * in a 'tree_id' tree. and sets path name to 'name'.
2209  */
2210 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2211                                 u64 tree_id, u64 dirid, char *name)
2212 {
2213         struct btrfs_root *root;
2214         struct btrfs_key key;
2215         char *ptr;
2216         int ret = -1;
2217         int slot;
2218         int len;
2219         int total_len = 0;
2220         struct btrfs_inode_ref *iref;
2221         struct extent_buffer *l;
2222         struct btrfs_path *path;
2223
2224         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2225                 name[0]='\0';
2226                 return 0;
2227         }
2228
2229         path = btrfs_alloc_path();
2230         if (!path)
2231                 return -ENOMEM;
2232
2233         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2234
2235         key.objectid = tree_id;
2236         key.type = BTRFS_ROOT_ITEM_KEY;
2237         key.offset = (u64)-1;
2238         root = btrfs_read_fs_root_no_name(info, &key);
2239         if (IS_ERR(root)) {
2240                 btrfs_err(info, "could not find root %llu", tree_id);
2241                 ret = -ENOENT;
2242                 goto out;
2243         }
2244
2245         key.objectid = dirid;
2246         key.type = BTRFS_INODE_REF_KEY;
2247         key.offset = (u64)-1;
2248
2249         while (1) {
2250                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2251                 if (ret < 0)
2252                         goto out;
2253                 else if (ret > 0) {
2254                         ret = btrfs_previous_item(root, path, dirid,
2255                                                   BTRFS_INODE_REF_KEY);
2256                         if (ret < 0)
2257                                 goto out;
2258                         else if (ret > 0) {
2259                                 ret = -ENOENT;
2260                                 goto out;
2261                         }
2262                 }
2263
2264                 l = path->nodes[0];
2265                 slot = path->slots[0];
2266                 btrfs_item_key_to_cpu(l, &key, slot);
2267
2268                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2269                 len = btrfs_inode_ref_name_len(l, iref);
2270                 ptr -= len + 1;
2271                 total_len += len + 1;
2272                 if (ptr < name) {
2273                         ret = -ENAMETOOLONG;
2274                         goto out;
2275                 }
2276
2277                 *(ptr + len) = '/';
2278                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2279
2280                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2281                         break;
2282
2283                 btrfs_release_path(path);
2284                 key.objectid = key.offset;
2285                 key.offset = (u64)-1;
2286                 dirid = key.objectid;
2287         }
2288         memmove(name, ptr, total_len);
2289         name[total_len] = '\0';
2290         ret = 0;
2291 out:
2292         btrfs_free_path(path);
2293         return ret;
2294 }
2295
2296 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2297                                            void __user *argp)
2298 {
2299          struct btrfs_ioctl_ino_lookup_args *args;
2300          struct inode *inode;
2301         int ret = 0;
2302
2303         args = memdup_user(argp, sizeof(*args));
2304         if (IS_ERR(args))
2305                 return PTR_ERR(args);
2306
2307         inode = file_inode(file);
2308
2309         /*
2310          * Unprivileged query to obtain the containing subvolume root id. The
2311          * path is reset so it's consistent with btrfs_search_path_in_tree.
2312          */
2313         if (args->treeid == 0)
2314                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2315
2316         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2317                 args->name[0] = 0;
2318                 goto out;
2319         }
2320
2321         if (!capable(CAP_SYS_ADMIN)) {
2322                 ret = -EPERM;
2323                 goto out;
2324         }
2325
2326         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2327                                         args->treeid, args->objectid,
2328                                         args->name);
2329
2330 out:
2331         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2332                 ret = -EFAULT;
2333
2334         kfree(args);
2335         return ret;
2336 }
2337
2338 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2339                                              void __user *arg)
2340 {
2341         struct dentry *parent = file->f_path.dentry;
2342         struct dentry *dentry;
2343         struct inode *dir = d_inode(parent);
2344         struct inode *inode;
2345         struct btrfs_root *root = BTRFS_I(dir)->root;
2346         struct btrfs_root *dest = NULL;
2347         struct btrfs_ioctl_vol_args *vol_args;
2348         struct btrfs_trans_handle *trans;
2349         struct btrfs_block_rsv block_rsv;
2350         u64 root_flags;
2351         u64 qgroup_reserved;
2352         int namelen;
2353         int ret;
2354         int err = 0;
2355
2356         vol_args = memdup_user(arg, sizeof(*vol_args));
2357         if (IS_ERR(vol_args))
2358                 return PTR_ERR(vol_args);
2359
2360         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2361         namelen = strlen(vol_args->name);
2362         if (strchr(vol_args->name, '/') ||
2363             strncmp(vol_args->name, "..", namelen) == 0) {
2364                 err = -EINVAL;
2365                 goto out;
2366         }
2367
2368         err = mnt_want_write_file(file);
2369         if (err)
2370                 goto out;
2371
2372
2373         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2374         if (err == -EINTR)
2375                 goto out_drop_write;
2376         dentry = lookup_one_len(vol_args->name, parent, namelen);
2377         if (IS_ERR(dentry)) {
2378                 err = PTR_ERR(dentry);
2379                 goto out_unlock_dir;
2380         }
2381
2382         if (d_really_is_negative(dentry)) {
2383                 err = -ENOENT;
2384                 goto out_dput;
2385         }
2386
2387         inode = d_inode(dentry);
2388         dest = BTRFS_I(inode)->root;
2389         if (!capable(CAP_SYS_ADMIN)) {
2390                 /*
2391                  * Regular user.  Only allow this with a special mount
2392                  * option, when the user has write+exec access to the
2393                  * subvol root, and when rmdir(2) would have been
2394                  * allowed.
2395                  *
2396                  * Note that this is _not_ check that the subvol is
2397                  * empty or doesn't contain data that we wouldn't
2398                  * otherwise be able to delete.
2399                  *
2400                  * Users who want to delete empty subvols should try
2401                  * rmdir(2).
2402                  */
2403                 err = -EPERM;
2404                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2405                         goto out_dput;
2406
2407                 /*
2408                  * Do not allow deletion if the parent dir is the same
2409                  * as the dir to be deleted.  That means the ioctl
2410                  * must be called on the dentry referencing the root
2411                  * of the subvol, not a random directory contained
2412                  * within it.
2413                  */
2414                 err = -EINVAL;
2415                 if (root == dest)
2416                         goto out_dput;
2417
2418                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2419                 if (err)
2420                         goto out_dput;
2421         }
2422
2423         /* check if subvolume may be deleted by a user */
2424         err = btrfs_may_delete(dir, dentry, 1);
2425         if (err)
2426                 goto out_dput;
2427
2428         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2429                 err = -EINVAL;
2430                 goto out_dput;
2431         }
2432
2433         mutex_lock(&inode->i_mutex);
2434
2435         /*
2436          * Don't allow to delete a subvolume with send in progress. This is
2437          * inside the i_mutex so the error handling that has to drop the bit
2438          * again is not run concurrently.
2439          */
2440         spin_lock(&dest->root_item_lock);
2441         root_flags = btrfs_root_flags(&dest->root_item);
2442         if (dest->send_in_progress == 0) {
2443                 btrfs_set_root_flags(&dest->root_item,
2444                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2445                 spin_unlock(&dest->root_item_lock);
2446         } else {
2447                 spin_unlock(&dest->root_item_lock);
2448                 btrfs_warn(root->fs_info,
2449                         "Attempt to delete subvolume %llu during send",
2450                         dest->root_key.objectid);
2451                 err = -EPERM;
2452                 goto out_unlock_inode;
2453         }
2454
2455         down_write(&root->fs_info->subvol_sem);
2456
2457         err = may_destroy_subvol(dest);
2458         if (err)
2459                 goto out_up_write;
2460
2461         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2462         /*
2463          * One for dir inode, two for dir entries, two for root
2464          * ref/backref.
2465          */
2466         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2467                                                5, &qgroup_reserved, true);
2468         if (err)
2469                 goto out_up_write;
2470
2471         trans = btrfs_start_transaction(root, 0);
2472         if (IS_ERR(trans)) {
2473                 err = PTR_ERR(trans);
2474                 goto out_release;
2475         }
2476         trans->block_rsv = &block_rsv;
2477         trans->bytes_reserved = block_rsv.size;
2478
2479         ret = btrfs_unlink_subvol(trans, root, dir,
2480                                 dest->root_key.objectid,
2481                                 dentry->d_name.name,
2482                                 dentry->d_name.len);
2483         if (ret) {
2484                 err = ret;
2485                 btrfs_abort_transaction(trans, root, ret);
2486                 goto out_end_trans;
2487         }
2488
2489         btrfs_record_root_in_trans(trans, dest);
2490
2491         memset(&dest->root_item.drop_progress, 0,
2492                 sizeof(dest->root_item.drop_progress));
2493         dest->root_item.drop_level = 0;
2494         btrfs_set_root_refs(&dest->root_item, 0);
2495
2496         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2497                 ret = btrfs_insert_orphan_item(trans,
2498                                         root->fs_info->tree_root,
2499                                         dest->root_key.objectid);
2500                 if (ret) {
2501                         btrfs_abort_transaction(trans, root, ret);
2502                         err = ret;
2503                         goto out_end_trans;
2504                 }
2505         }
2506
2507         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2508                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2509                                   dest->root_key.objectid);
2510         if (ret && ret != -ENOENT) {
2511                 btrfs_abort_transaction(trans, root, ret);
2512                 err = ret;
2513                 goto out_end_trans;
2514         }
2515         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2516                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2517                                           dest->root_item.received_uuid,
2518                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2519                                           dest->root_key.objectid);
2520                 if (ret && ret != -ENOENT) {
2521                         btrfs_abort_transaction(trans, root, ret);
2522                         err = ret;
2523                         goto out_end_trans;
2524                 }
2525         }
2526
2527 out_end_trans:
2528         trans->block_rsv = NULL;
2529         trans->bytes_reserved = 0;
2530         ret = btrfs_end_transaction(trans, root);
2531         if (ret && !err)
2532                 err = ret;
2533         inode->i_flags |= S_DEAD;
2534 out_release:
2535         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2536 out_up_write:
2537         up_write(&root->fs_info->subvol_sem);
2538         if (err) {
2539                 spin_lock(&dest->root_item_lock);
2540                 root_flags = btrfs_root_flags(&dest->root_item);
2541                 btrfs_set_root_flags(&dest->root_item,
2542                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2543                 spin_unlock(&dest->root_item_lock);
2544         }
2545 out_unlock_inode:
2546         mutex_unlock(&inode->i_mutex);
2547         if (!err) {
2548                 d_invalidate(dentry);
2549                 btrfs_invalidate_inodes(dest);
2550                 d_delete(dentry);
2551                 ASSERT(dest->send_in_progress == 0);
2552
2553                 /* the last ref */
2554                 if (dest->ino_cache_inode) {
2555                         iput(dest->ino_cache_inode);
2556                         dest->ino_cache_inode = NULL;
2557                 }
2558         }
2559 out_dput:
2560         dput(dentry);
2561 out_unlock_dir:
2562         mutex_unlock(&dir->i_mutex);
2563 out_drop_write:
2564         mnt_drop_write_file(file);
2565 out:
2566         kfree(vol_args);
2567         return err;
2568 }
2569
2570 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2571 {
2572         struct inode *inode = file_inode(file);
2573         struct btrfs_root *root = BTRFS_I(inode)->root;
2574         struct btrfs_ioctl_defrag_range_args *range;
2575         int ret;
2576
2577         ret = mnt_want_write_file(file);
2578         if (ret)
2579                 return ret;
2580
2581         if (btrfs_root_readonly(root)) {
2582                 ret = -EROFS;
2583                 goto out;
2584         }
2585
2586         switch (inode->i_mode & S_IFMT) {
2587         case S_IFDIR:
2588                 if (!capable(CAP_SYS_ADMIN)) {
2589                         ret = -EPERM;
2590                         goto out;
2591                 }
2592                 ret = btrfs_defrag_root(root);
2593                 if (ret)
2594                         goto out;
2595                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2596                 break;
2597         case S_IFREG:
2598                 if (!(file->f_mode & FMODE_WRITE)) {
2599                         ret = -EINVAL;
2600                         goto out;
2601                 }
2602
2603                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2604                 if (!range) {
2605                         ret = -ENOMEM;
2606                         goto out;
2607                 }
2608
2609                 if (argp) {
2610                         if (copy_from_user(range, argp,
2611                                            sizeof(*range))) {
2612                                 ret = -EFAULT;
2613                                 kfree(range);
2614                                 goto out;
2615                         }
2616                         /* compression requires us to start the IO */
2617                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2618                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2619                                 range->extent_thresh = (u32)-1;
2620                         }
2621                 } else {
2622                         /* the rest are all set to zero by kzalloc */
2623                         range->len = (u64)-1;
2624                 }
2625                 ret = btrfs_defrag_file(file_inode(file), file,
2626                                         range, 0, 0);
2627                 if (ret > 0)
2628                         ret = 0;
2629                 kfree(range);
2630                 break;
2631         default:
2632                 ret = -EINVAL;
2633         }
2634 out:
2635         mnt_drop_write_file(file);
2636         return ret;
2637 }
2638
2639 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2640 {
2641         struct btrfs_ioctl_vol_args *vol_args;
2642         int ret;
2643
2644         if (!capable(CAP_SYS_ADMIN))
2645                 return -EPERM;
2646
2647         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2648                         1)) {
2649                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2650         }
2651
2652         mutex_lock(&root->fs_info->volume_mutex);
2653         vol_args = memdup_user(arg, sizeof(*vol_args));
2654         if (IS_ERR(vol_args)) {
2655                 ret = PTR_ERR(vol_args);
2656                 goto out;
2657         }
2658
2659         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2660         ret = btrfs_init_new_device(root, vol_args->name);
2661
2662         if (!ret)
2663                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2664
2665         kfree(vol_args);
2666 out:
2667         mutex_unlock(&root->fs_info->volume_mutex);
2668         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2669         return ret;
2670 }
2671
2672 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2673 {
2674         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2675         struct btrfs_ioctl_vol_args *vol_args;
2676         int ret;
2677
2678         if (!capable(CAP_SYS_ADMIN))
2679                 return -EPERM;
2680
2681         ret = mnt_want_write_file(file);
2682         if (ret)
2683                 return ret;
2684
2685         vol_args = memdup_user(arg, sizeof(*vol_args));
2686         if (IS_ERR(vol_args)) {
2687                 ret = PTR_ERR(vol_args);
2688                 goto err_drop;
2689         }
2690
2691         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2692
2693         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2694                         1)) {
2695                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2696                 goto out;
2697         }
2698
2699         mutex_lock(&root->fs_info->volume_mutex);
2700         ret = btrfs_rm_device(root, vol_args->name);
2701         mutex_unlock(&root->fs_info->volume_mutex);
2702         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2703
2704         if (!ret)
2705                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2706
2707 out:
2708         kfree(vol_args);
2709 err_drop:
2710         mnt_drop_write_file(file);
2711         return ret;
2712 }
2713
2714 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2715 {
2716         struct btrfs_ioctl_fs_info_args *fi_args;
2717         struct btrfs_device *device;
2718         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2719         int ret = 0;
2720
2721         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2722         if (!fi_args)
2723                 return -ENOMEM;
2724
2725         mutex_lock(&fs_devices->device_list_mutex);
2726         fi_args->num_devices = fs_devices->num_devices;
2727         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2728
2729         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2730                 if (device->devid > fi_args->max_id)
2731                         fi_args->max_id = device->devid;
2732         }
2733         mutex_unlock(&fs_devices->device_list_mutex);
2734
2735         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2736         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2737         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2738
2739         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2740                 ret = -EFAULT;
2741
2742         kfree(fi_args);
2743         return ret;
2744 }
2745
2746 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2747 {
2748         struct btrfs_ioctl_dev_info_args *di_args;
2749         struct btrfs_device *dev;
2750         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2751         int ret = 0;
2752         char *s_uuid = NULL;
2753
2754         di_args = memdup_user(arg, sizeof(*di_args));
2755         if (IS_ERR(di_args))
2756                 return PTR_ERR(di_args);
2757
2758         if (!btrfs_is_empty_uuid(di_args->uuid))
2759                 s_uuid = di_args->uuid;
2760
2761         mutex_lock(&fs_devices->device_list_mutex);
2762         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2763
2764         if (!dev) {
2765                 ret = -ENODEV;
2766                 goto out;
2767         }
2768
2769         di_args->devid = dev->devid;
2770         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2771         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2772         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2773         if (dev->name) {
2774                 struct rcu_string *name;
2775
2776                 rcu_read_lock();
2777                 name = rcu_dereference(dev->name);
2778                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2779                 rcu_read_unlock();
2780                 di_args->path[sizeof(di_args->path) - 1] = 0;
2781         } else {
2782                 di_args->path[0] = '\0';
2783         }
2784
2785 out:
2786         mutex_unlock(&fs_devices->device_list_mutex);
2787         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2788                 ret = -EFAULT;
2789
2790         kfree(di_args);
2791         return ret;
2792 }
2793
2794 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2795 {
2796         struct page *page;
2797         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2798
2799         page = grab_cache_page(inode->i_mapping, index);
2800         if (!page)
2801                 return NULL;
2802
2803         if (!PageUptodate(page)) {
2804                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2805                                                  0))
2806                         return NULL;
2807                 lock_page(page);
2808                 if (!PageUptodate(page)) {
2809                         unlock_page(page);
2810                         page_cache_release(page);
2811                         return NULL;
2812                 }
2813         }
2814
2815         return page;
2816 }
2817
2818 static int gather_extent_pages(struct inode *inode, struct page **pages,
2819                                int num_pages, u64 off)
2820 {
2821         int i;
2822         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2823
2824         for (i = 0; i < num_pages; i++) {
2825                 pages[i] = extent_same_get_page(inode, index + i);
2826                 if (!pages[i])
2827                         return -ENOMEM;
2828         }
2829         return 0;
2830 }
2831
2832 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2833                              bool retry_range_locking)
2834 {
2835         /*
2836          * Do any pending delalloc/csum calculations on inode, one way or
2837          * another, and lock file content.
2838          * The locking order is:
2839          *
2840          *   1) pages
2841          *   2) range in the inode's io tree
2842          */
2843         while (1) {
2844                 struct btrfs_ordered_extent *ordered;
2845                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2846                 ordered = btrfs_lookup_first_ordered_extent(inode,
2847                                                             off + len - 1);
2848                 if ((!ordered ||
2849                      ordered->file_offset + ordered->len <= off ||
2850                      ordered->file_offset >= off + len) &&
2851                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2852                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2853                         if (ordered)
2854                                 btrfs_put_ordered_extent(ordered);
2855                         break;
2856                 }
2857                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2858                 if (ordered)
2859                         btrfs_put_ordered_extent(ordered);
2860                 if (!retry_range_locking)
2861                         return -EAGAIN;
2862                 btrfs_wait_ordered_range(inode, off, len);
2863         }
2864         return 0;
2865 }
2866
2867 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2868 {
2869         mutex_unlock(&inode1->i_mutex);
2870         mutex_unlock(&inode2->i_mutex);
2871 }
2872
2873 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2874 {
2875         if (inode1 < inode2)
2876                 swap(inode1, inode2);
2877
2878         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2879         mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2880 }
2881
2882 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2883                                       struct inode *inode2, u64 loff2, u64 len)
2884 {
2885         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2886         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2887 }
2888
2889 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2890                                     struct inode *inode2, u64 loff2, u64 len,
2891                                     bool retry_range_locking)
2892 {
2893         int ret;
2894
2895         if (inode1 < inode2) {
2896                 swap(inode1, inode2);
2897                 swap(loff1, loff2);
2898         }
2899         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2900         if (ret)
2901                 return ret;
2902         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2903         if (ret)
2904                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2905                               loff1 + len - 1);
2906         return ret;
2907 }
2908
2909 struct cmp_pages {
2910         int             num_pages;
2911         struct page     **src_pages;
2912         struct page     **dst_pages;
2913 };
2914
2915 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2916 {
2917         int i;
2918         struct page *pg;
2919
2920         for (i = 0; i < cmp->num_pages; i++) {
2921                 pg = cmp->src_pages[i];
2922                 if (pg) {
2923                         unlock_page(pg);
2924                         page_cache_release(pg);
2925                 }
2926                 pg = cmp->dst_pages[i];
2927                 if (pg) {
2928                         unlock_page(pg);
2929                         page_cache_release(pg);
2930                 }
2931         }
2932         kfree(cmp->src_pages);
2933         kfree(cmp->dst_pages);
2934 }
2935
2936 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2937                                   struct inode *dst, u64 dst_loff,
2938                                   u64 len, struct cmp_pages *cmp)
2939 {
2940         int ret;
2941         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2942         struct page **src_pgarr, **dst_pgarr;
2943
2944         /*
2945          * We must gather up all the pages before we initiate our
2946          * extent locking. We use an array for the page pointers. Size
2947          * of the array is bounded by len, which is in turn bounded by
2948          * BTRFS_MAX_DEDUPE_LEN.
2949          */
2950         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2951         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2952         if (!src_pgarr || !dst_pgarr) {
2953                 kfree(src_pgarr);
2954                 kfree(dst_pgarr);
2955                 return -ENOMEM;
2956         }
2957         cmp->num_pages = num_pages;
2958         cmp->src_pages = src_pgarr;
2959         cmp->dst_pages = dst_pgarr;
2960
2961         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2962         if (ret)
2963                 goto out;
2964
2965         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2966
2967 out:
2968         if (ret)
2969                 btrfs_cmp_data_free(cmp);
2970         return 0;
2971 }
2972
2973 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2974                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2975 {
2976         int ret = 0;
2977         int i;
2978         struct page *src_page, *dst_page;
2979         unsigned int cmp_len = PAGE_CACHE_SIZE;
2980         void *addr, *dst_addr;
2981
2982         i = 0;
2983         while (len) {
2984                 if (len < PAGE_CACHE_SIZE)
2985                         cmp_len = len;
2986
2987                 BUG_ON(i >= cmp->num_pages);
2988
2989                 src_page = cmp->src_pages[i];
2990                 dst_page = cmp->dst_pages[i];
2991                 ASSERT(PageLocked(src_page));
2992                 ASSERT(PageLocked(dst_page));
2993
2994                 addr = kmap_atomic(src_page);
2995                 dst_addr = kmap_atomic(dst_page);
2996
2997                 flush_dcache_page(src_page);
2998                 flush_dcache_page(dst_page);
2999
3000                 if (memcmp(addr, dst_addr, cmp_len))
3001                         ret = BTRFS_SAME_DATA_DIFFERS;
3002
3003                 kunmap_atomic(addr);
3004                 kunmap_atomic(dst_addr);
3005
3006                 if (ret)
3007                         break;
3008
3009                 len -= cmp_len;
3010                 i++;
3011         }
3012
3013         return ret;
3014 }
3015
3016 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3017                                      u64 olen)
3018 {
3019         u64 len = *plen;
3020         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3021
3022         if (off + olen > inode->i_size || off + olen < off)
3023                 return -EINVAL;
3024
3025         /* if we extend to eof, continue to block boundary */
3026         if (off + len == inode->i_size)
3027                 *plen = len = ALIGN(inode->i_size, bs) - off;
3028
3029         /* Check that we are block aligned - btrfs_clone() requires this */
3030         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3031                 return -EINVAL;
3032
3033         return 0;
3034 }
3035
3036 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3037                              struct inode *dst, u64 dst_loff)
3038 {
3039         int ret;
3040         u64 len = olen;
3041         struct cmp_pages cmp;
3042         int same_inode = 0;
3043         u64 same_lock_start = 0;
3044         u64 same_lock_len = 0;
3045
3046         if (src == dst)
3047                 same_inode = 1;
3048
3049         if (len == 0)
3050                 return 0;
3051
3052         if (same_inode) {
3053                 mutex_lock(&src->i_mutex);
3054
3055                 ret = extent_same_check_offsets(src, loff, &len, olen);
3056                 if (ret)
3057                         goto out_unlock;
3058
3059                 /*
3060                  * Single inode case wants the same checks, except we
3061                  * don't want our length pushed out past i_size as
3062                  * comparing that data range makes no sense.
3063                  *
3064                  * extent_same_check_offsets() will do this for an
3065                  * unaligned length at i_size, so catch it here and
3066                  * reject the request.
3067                  *
3068                  * This effectively means we require aligned extents
3069                  * for the single-inode case, whereas the other cases
3070                  * allow an unaligned length so long as it ends at
3071                  * i_size.
3072                  */
3073                 if (len != olen) {
3074                         ret = -EINVAL;
3075                         goto out_unlock;
3076                 }
3077
3078                 /* Check for overlapping ranges */
3079                 if (dst_loff + len > loff && dst_loff < loff + len) {
3080                         ret = -EINVAL;
3081                         goto out_unlock;
3082                 }
3083
3084                 same_lock_start = min_t(u64, loff, dst_loff);
3085                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3086         } else {
3087                 btrfs_double_inode_lock(src, dst);
3088
3089                 ret = extent_same_check_offsets(src, loff, &len, olen);
3090                 if (ret)
3091                         goto out_unlock;
3092
3093                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3094                 if (ret)
3095                         goto out_unlock;
3096         }
3097
3098         /* don't make the dst file partly checksummed */
3099         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3100             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3101                 ret = -EINVAL;
3102                 goto out_unlock;
3103         }
3104
3105 again:
3106         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3107         if (ret)
3108                 goto out_unlock;
3109
3110         if (same_inode)
3111                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3112                                         false);
3113         else
3114                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3115                                                false);
3116         /*
3117          * If one of the inodes has dirty pages in the respective range or
3118          * ordered extents, we need to flush dellaloc and wait for all ordered
3119          * extents in the range. We must unlock the pages and the ranges in the
3120          * io trees to avoid deadlocks when flushing delalloc (requires locking
3121          * pages) and when waiting for ordered extents to complete (they require
3122          * range locking).
3123          */
3124         if (ret == -EAGAIN) {
3125                 /*
3126                  * Ranges in the io trees already unlocked. Now unlock all
3127                  * pages before waiting for all IO to complete.
3128                  */
3129                 btrfs_cmp_data_free(&cmp);
3130                 if (same_inode) {
3131                         btrfs_wait_ordered_range(src, same_lock_start,
3132                                                  same_lock_len);
3133                 } else {
3134                         btrfs_wait_ordered_range(src, loff, len);
3135                         btrfs_wait_ordered_range(dst, dst_loff, len);
3136                 }
3137                 goto again;
3138         }
3139         ASSERT(ret == 0);
3140         if (WARN_ON(ret)) {
3141                 /* ranges in the io trees already unlocked */
3142                 btrfs_cmp_data_free(&cmp);
3143                 return ret;
3144         }
3145
3146         /* pass original length for comparison so we stay within i_size */
3147         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3148         if (ret == 0)
3149                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3150
3151         if (same_inode)
3152                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3153                               same_lock_start + same_lock_len - 1);
3154         else
3155                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3156
3157         btrfs_cmp_data_free(&cmp);
3158 out_unlock:
3159         if (same_inode)
3160                 mutex_unlock(&src->i_mutex);
3161         else
3162                 btrfs_double_inode_unlock(src, dst);
3163
3164         return ret;
3165 }
3166
3167 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3168
3169 static long btrfs_ioctl_file_extent_same(struct file *file,
3170                         struct btrfs_ioctl_same_args __user *argp)
3171 {
3172         struct btrfs_ioctl_same_args *same = NULL;
3173         struct btrfs_ioctl_same_extent_info *info;
3174         struct inode *src = file_inode(file);
3175         u64 off;
3176         u64 len;
3177         int i;
3178         int ret;
3179         unsigned long size;
3180         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3181         bool is_admin = capable(CAP_SYS_ADMIN);
3182         u16 count;
3183
3184         if (!(file->f_mode & FMODE_READ))
3185                 return -EINVAL;
3186
3187         ret = mnt_want_write_file(file);
3188         if (ret)
3189                 return ret;
3190
3191         if (get_user(count, &argp->dest_count)) {
3192                 ret = -EFAULT;
3193                 goto out;
3194         }
3195
3196         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3197
3198         same = memdup_user(argp, size);
3199
3200         if (IS_ERR(same)) {
3201                 ret = PTR_ERR(same);
3202                 same = NULL;
3203                 goto out;
3204         }
3205
3206         off = same->logical_offset;
3207         len = same->length;
3208
3209         /*
3210          * Limit the total length we will dedupe for each operation.
3211          * This is intended to bound the total time spent in this
3212          * ioctl to something sane.
3213          */
3214         if (len > BTRFS_MAX_DEDUPE_LEN)
3215                 len = BTRFS_MAX_DEDUPE_LEN;
3216
3217         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3218                 /*
3219                  * Btrfs does not support blocksize < page_size. As a
3220                  * result, btrfs_cmp_data() won't correctly handle
3221                  * this situation without an update.
3222                  */
3223                 ret = -EINVAL;
3224                 goto out;
3225         }
3226
3227         ret = -EISDIR;
3228         if (S_ISDIR(src->i_mode))
3229                 goto out;
3230
3231         ret = -EACCES;
3232         if (!S_ISREG(src->i_mode))
3233                 goto out;
3234
3235         /* pre-format output fields to sane values */
3236         for (i = 0; i < count; i++) {
3237                 same->info[i].bytes_deduped = 0ULL;
3238                 same->info[i].status = 0;
3239         }
3240
3241         for (i = 0, info = same->info; i < count; i++, info++) {
3242                 struct inode *dst;
3243                 struct fd dst_file = fdget(info->fd);
3244                 if (!dst_file.file) {
3245                         info->status = -EBADF;
3246                         continue;
3247                 }
3248                 dst = file_inode(dst_file.file);
3249
3250                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3251                         info->status = -EINVAL;
3252                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3253                         info->status = -EXDEV;
3254                 } else if (S_ISDIR(dst->i_mode)) {
3255                         info->status = -EISDIR;
3256                 } else if (!S_ISREG(dst->i_mode)) {
3257                         info->status = -EACCES;
3258                 } else {
3259                         info->status = btrfs_extent_same(src, off, len, dst,
3260                                                         info->logical_offset);
3261                         if (info->status == 0)
3262                                 info->bytes_deduped += len;
3263                 }
3264                 fdput(dst_file);
3265         }
3266
3267         ret = copy_to_user(argp, same, size);
3268         if (ret)
3269                 ret = -EFAULT;
3270
3271 out:
3272         mnt_drop_write_file(file);
3273         kfree(same);
3274         return ret;
3275 }
3276
3277 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3278                                      struct inode *inode,
3279                                      u64 endoff,
3280                                      const u64 destoff,
3281                                      const u64 olen,
3282                                      int no_time_update)
3283 {
3284         struct btrfs_root *root = BTRFS_I(inode)->root;
3285         int ret;
3286
3287         inode_inc_iversion(inode);
3288         if (!no_time_update)
3289                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3290         /*
3291          * We round up to the block size at eof when determining which
3292          * extents to clone above, but shouldn't round up the file size.
3293          */
3294         if (endoff > destoff + olen)
3295                 endoff = destoff + olen;
3296         if (endoff > inode->i_size)
3297                 btrfs_i_size_write(inode, endoff);
3298
3299         ret = btrfs_update_inode(trans, root, inode);
3300         if (ret) {
3301                 btrfs_abort_transaction(trans, root, ret);
3302                 btrfs_end_transaction(trans, root);
3303                 goto out;
3304         }
3305         ret = btrfs_end_transaction(trans, root);
3306 out:
3307         return ret;
3308 }
3309
3310 static void clone_update_extent_map(struct inode *inode,
3311                                     const struct btrfs_trans_handle *trans,
3312                                     const struct btrfs_path *path,
3313                                     const u64 hole_offset,
3314                                     const u64 hole_len)
3315 {
3316         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3317         struct extent_map *em;
3318         int ret;
3319
3320         em = alloc_extent_map();
3321         if (!em) {
3322                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3323                         &BTRFS_I(inode)->runtime_flags);
3324                 return;
3325         }
3326
3327         if (path) {
3328                 struct btrfs_file_extent_item *fi;
3329
3330                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3331                                     struct btrfs_file_extent_item);
3332                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3333                 em->generation = -1;
3334                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3335                     BTRFS_FILE_EXTENT_INLINE)
3336                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3337                                 &BTRFS_I(inode)->runtime_flags);
3338         } else {
3339                 em->start = hole_offset;
3340                 em->len = hole_len;
3341                 em->ram_bytes = em->len;
3342                 em->orig_start = hole_offset;
3343                 em->block_start = EXTENT_MAP_HOLE;
3344                 em->block_len = 0;
3345                 em->orig_block_len = 0;
3346                 em->compress_type = BTRFS_COMPRESS_NONE;
3347                 em->generation = trans->transid;
3348         }
3349
3350         while (1) {
3351                 write_lock(&em_tree->lock);
3352                 ret = add_extent_mapping(em_tree, em, 1);
3353                 write_unlock(&em_tree->lock);
3354                 if (ret != -EEXIST) {
3355                         free_extent_map(em);
3356                         break;
3357                 }
3358                 btrfs_drop_extent_cache(inode, em->start,
3359                                         em->start + em->len - 1, 0);
3360         }
3361
3362         if (ret)
3363                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3364                         &BTRFS_I(inode)->runtime_flags);
3365 }
3366
3367 /*
3368  * Make sure we do not end up inserting an inline extent into a file that has
3369  * already other (non-inline) extents. If a file has an inline extent it can
3370  * not have any other extents and the (single) inline extent must start at the
3371  * file offset 0. Failing to respect these rules will lead to file corruption,
3372  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3373  *
3374  * We can have extents that have been already written to disk or we can have
3375  * dirty ranges still in delalloc, in which case the extent maps and items are
3376  * created only when we run delalloc, and the delalloc ranges might fall outside
3377  * the range we are currently locking in the inode's io tree. So we check the
3378  * inode's i_size because of that (i_size updates are done while holding the
3379  * i_mutex, which we are holding here).
3380  * We also check to see if the inode has a size not greater than "datal" but has
3381  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3382  * protected against such concurrent fallocate calls by the i_mutex).
3383  *
3384  * If the file has no extents but a size greater than datal, do not allow the
3385  * copy because we would need turn the inline extent into a non-inline one (even
3386  * with NO_HOLES enabled). If we find our destination inode only has one inline
3387  * extent, just overwrite it with the source inline extent if its size is less
3388  * than the source extent's size, or we could copy the source inline extent's
3389  * data into the destination inode's inline extent if the later is greater then
3390  * the former.
3391  */
3392 static int clone_copy_inline_extent(struct inode *src,
3393                                     struct inode *dst,
3394                                     struct btrfs_trans_handle *trans,
3395                                     struct btrfs_path *path,
3396                                     struct btrfs_key *new_key,
3397                                     const u64 drop_start,
3398                                     const u64 datal,
3399                                     const u64 skip,
3400                                     const u64 size,
3401                                     char *inline_data)
3402 {
3403         struct btrfs_root *root = BTRFS_I(dst)->root;
3404         const u64 aligned_end = ALIGN(new_key->offset + datal,
3405                                       root->sectorsize);
3406         int ret;
3407         struct btrfs_key key;
3408
3409         if (new_key->offset > 0)
3410                 return -EOPNOTSUPP;
3411
3412         key.objectid = btrfs_ino(dst);
3413         key.type = BTRFS_EXTENT_DATA_KEY;
3414         key.offset = 0;
3415         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3416         if (ret < 0) {
3417                 return ret;
3418         } else if (ret > 0) {
3419                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3420                         ret = btrfs_next_leaf(root, path);
3421                         if (ret < 0)
3422                                 return ret;
3423                         else if (ret > 0)
3424                                 goto copy_inline_extent;
3425                 }
3426                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3427                 if (key.objectid == btrfs_ino(dst) &&
3428                     key.type == BTRFS_EXTENT_DATA_KEY) {
3429                         ASSERT(key.offset > 0);
3430                         return -EOPNOTSUPP;
3431                 }
3432         } else if (i_size_read(dst) <= datal) {
3433                 struct btrfs_file_extent_item *ei;
3434                 u64 ext_len;
3435
3436                 /*
3437                  * If the file size is <= datal, make sure there are no other
3438                  * extents following (can happen do to an fallocate call with
3439                  * the flag FALLOC_FL_KEEP_SIZE).
3440                  */
3441                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3442                                     struct btrfs_file_extent_item);
3443                 /*
3444                  * If it's an inline extent, it can not have other extents
3445                  * following it.
3446                  */
3447                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3448                     BTRFS_FILE_EXTENT_INLINE)
3449                         goto copy_inline_extent;
3450
3451                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3452                 if (ext_len > aligned_end)
3453                         return -EOPNOTSUPP;
3454
3455                 ret = btrfs_next_item(root, path);
3456                 if (ret < 0) {
3457                         return ret;
3458                 } else if (ret == 0) {
3459                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3460                                               path->slots[0]);
3461                         if (key.objectid == btrfs_ino(dst) &&
3462                             key.type == BTRFS_EXTENT_DATA_KEY)
3463                                 return -EOPNOTSUPP;
3464                 }
3465         }
3466
3467 copy_inline_extent:
3468         /*
3469          * We have no extent items, or we have an extent at offset 0 which may
3470          * or may not be inlined. All these cases are dealt the same way.
3471          */
3472         if (i_size_read(dst) > datal) {
3473                 /*
3474                  * If the destination inode has an inline extent...
3475                  * This would require copying the data from the source inline
3476                  * extent into the beginning of the destination's inline extent.
3477                  * But this is really complex, both extents can be compressed
3478                  * or just one of them, which would require decompressing and
3479                  * re-compressing data (which could increase the new compressed
3480                  * size, not allowing the compressed data to fit anymore in an
3481                  * inline extent).
3482                  * So just don't support this case for now (it should be rare,
3483                  * we are not really saving space when cloning inline extents).
3484                  */
3485                 return -EOPNOTSUPP;
3486         }
3487
3488         btrfs_release_path(path);
3489         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3490         if (ret)
3491                 return ret;
3492         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3493         if (ret)
3494                 return ret;
3495
3496         if (skip) {
3497                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3498
3499                 memmove(inline_data + start, inline_data + start + skip, datal);
3500         }
3501
3502         write_extent_buffer(path->nodes[0], inline_data,
3503                             btrfs_item_ptr_offset(path->nodes[0],
3504                                                   path->slots[0]),
3505                             size);
3506         inode_add_bytes(dst, datal);
3507
3508         return 0;
3509 }
3510
3511 /**
3512  * btrfs_clone() - clone a range from inode file to another
3513  *
3514  * @src: Inode to clone from
3515  * @inode: Inode to clone to
3516  * @off: Offset within source to start clone from
3517  * @olen: Original length, passed by user, of range to clone
3518  * @olen_aligned: Block-aligned value of olen
3519  * @destoff: Offset within @inode to start clone
3520  * @no_time_update: Whether to update mtime/ctime on the target inode
3521  */
3522 static int btrfs_clone(struct inode *src, struct inode *inode,
3523                        const u64 off, const u64 olen, const u64 olen_aligned,
3524                        const u64 destoff, int no_time_update)
3525 {
3526         struct btrfs_root *root = BTRFS_I(inode)->root;
3527         struct btrfs_path *path = NULL;
3528         struct extent_buffer *leaf;
3529         struct btrfs_trans_handle *trans;
3530         char *buf = NULL;
3531         struct btrfs_key key;
3532         u32 nritems;
3533         int slot;
3534         int ret;
3535         const u64 len = olen_aligned;
3536         u64 last_dest_end = destoff;
3537
3538         ret = -ENOMEM;
3539         buf = vmalloc(root->nodesize);
3540         if (!buf)
3541                 return ret;
3542
3543         path = btrfs_alloc_path();
3544         if (!path) {
3545                 vfree(buf);
3546                 return ret;
3547         }
3548
3549         path->reada = READA_FORWARD;
3550         /* clone data */
3551         key.objectid = btrfs_ino(src);
3552         key.type = BTRFS_EXTENT_DATA_KEY;
3553         key.offset = off;
3554
3555         while (1) {
3556                 u64 next_key_min_offset = key.offset + 1;
3557
3558                 /*
3559                  * note the key will change type as we walk through the
3560                  * tree.
3561                  */
3562                 path->leave_spinning = 1;
3563                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3564                                 0, 0);
3565                 if (ret < 0)
3566                         goto out;
3567                 /*
3568                  * First search, if no extent item that starts at offset off was
3569                  * found but the previous item is an extent item, it's possible
3570                  * it might overlap our target range, therefore process it.
3571                  */
3572                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3573                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3574                                               path->slots[0] - 1);
3575                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3576                                 path->slots[0]--;
3577                 }
3578
3579                 nritems = btrfs_header_nritems(path->nodes[0]);
3580 process_slot:
3581                 if (path->slots[0] >= nritems) {
3582                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3583                         if (ret < 0)
3584                                 goto out;
3585                         if (ret > 0)
3586                                 break;
3587                         nritems = btrfs_header_nritems(path->nodes[0]);
3588                 }
3589                 leaf = path->nodes[0];
3590                 slot = path->slots[0];
3591
3592                 btrfs_item_key_to_cpu(leaf, &key, slot);
3593                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3594                     key.objectid != btrfs_ino(src))
3595                         break;
3596
3597                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3598                         struct btrfs_file_extent_item *extent;
3599                         int type;
3600                         u32 size;
3601                         struct btrfs_key new_key;
3602                         u64 disko = 0, diskl = 0;
3603                         u64 datao = 0, datal = 0;
3604                         u8 comp;
3605                         u64 drop_start;
3606
3607                         extent = btrfs_item_ptr(leaf, slot,
3608                                                 struct btrfs_file_extent_item);
3609                         comp = btrfs_file_extent_compression(leaf, extent);
3610                         type = btrfs_file_extent_type(leaf, extent);
3611                         if (type == BTRFS_FILE_EXTENT_REG ||
3612                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3613                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3614                                                                       extent);
3615                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3616                                                                  extent);
3617                                 datao = btrfs_file_extent_offset(leaf, extent);
3618                                 datal = btrfs_file_extent_num_bytes(leaf,
3619                                                                     extent);
3620                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3621                                 /* take upper bound, may be compressed */
3622                                 datal = btrfs_file_extent_ram_bytes(leaf,
3623                                                                     extent);
3624                         }
3625
3626                         /*
3627                          * The first search might have left us at an extent
3628                          * item that ends before our target range's start, can
3629                          * happen if we have holes and NO_HOLES feature enabled.
3630                          */
3631                         if (key.offset + datal <= off) {
3632                                 path->slots[0]++;
3633                                 goto process_slot;
3634                         } else if (key.offset >= off + len) {
3635                                 break;
3636                         }
3637                         next_key_min_offset = key.offset + datal;
3638                         size = btrfs_item_size_nr(leaf, slot);
3639                         read_extent_buffer(leaf, buf,
3640                                            btrfs_item_ptr_offset(leaf, slot),
3641                                            size);
3642
3643                         btrfs_release_path(path);
3644                         path->leave_spinning = 0;
3645
3646                         memcpy(&new_key, &key, sizeof(new_key));
3647                         new_key.objectid = btrfs_ino(inode);
3648                         if (off <= key.offset)
3649                                 new_key.offset = key.offset + destoff - off;
3650                         else
3651                                 new_key.offset = destoff;
3652
3653                         /*
3654                          * Deal with a hole that doesn't have an extent item
3655                          * that represents it (NO_HOLES feature enabled).
3656                          * This hole is either in the middle of the cloning
3657                          * range or at the beginning (fully overlaps it or
3658                          * partially overlaps it).
3659                          */
3660                         if (new_key.offset != last_dest_end)
3661                                 drop_start = last_dest_end;
3662                         else
3663                                 drop_start = new_key.offset;
3664
3665                         /*
3666                          * 1 - adjusting old extent (we may have to split it)
3667                          * 1 - add new extent
3668                          * 1 - inode update
3669                          */
3670                         trans = btrfs_start_transaction(root, 3);
3671                         if (IS_ERR(trans)) {
3672                                 ret = PTR_ERR(trans);
3673                                 goto out;
3674                         }
3675
3676                         if (type == BTRFS_FILE_EXTENT_REG ||
3677                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3678                                 /*
3679                                  *    a  | --- range to clone ---|  b
3680                                  * | ------------- extent ------------- |
3681                                  */
3682
3683                                 /* subtract range b */
3684                                 if (key.offset + datal > off + len)
3685                                         datal = off + len - key.offset;
3686
3687                                 /* subtract range a */
3688                                 if (off > key.offset) {
3689                                         datao += off - key.offset;
3690                                         datal -= off - key.offset;
3691                                 }
3692
3693                                 ret = btrfs_drop_extents(trans, root, inode,
3694                                                          drop_start,
3695                                                          new_key.offset + datal,
3696                                                          1);
3697                                 if (ret) {
3698                                         if (ret != -EOPNOTSUPP)
3699                                                 btrfs_abort_transaction(trans,
3700                                                                 root, ret);
3701                                         btrfs_end_transaction(trans, root);
3702                                         goto out;
3703                                 }
3704
3705                                 ret = btrfs_insert_empty_item(trans, root, path,
3706                                                               &new_key, size);
3707                                 if (ret) {
3708                                         btrfs_abort_transaction(trans, root,
3709                                                                 ret);
3710                                         btrfs_end_transaction(trans, root);
3711                                         goto out;
3712                                 }
3713
3714                                 leaf = path->nodes[0];
3715                                 slot = path->slots[0];
3716                                 write_extent_buffer(leaf, buf,
3717                                             btrfs_item_ptr_offset(leaf, slot),
3718                                             size);
3719
3720                                 extent = btrfs_item_ptr(leaf, slot,
3721                                                 struct btrfs_file_extent_item);
3722
3723                                 /* disko == 0 means it's a hole */
3724                                 if (!disko)
3725                                         datao = 0;
3726
3727                                 btrfs_set_file_extent_offset(leaf, extent,
3728                                                              datao);
3729                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3730                                                                 datal);
3731
3732                                 if (disko) {
3733                                         inode_add_bytes(inode, datal);
3734                                         ret = btrfs_inc_extent_ref(trans, root,
3735                                                         disko, diskl, 0,
3736                                                         root->root_key.objectid,
3737                                                         btrfs_ino(inode),
3738                                                         new_key.offset - datao);
3739                                         if (ret) {
3740                                                 btrfs_abort_transaction(trans,
3741                                                                         root,
3742                                                                         ret);
3743                                                 btrfs_end_transaction(trans,
3744                                                                       root);
3745                                                 goto out;
3746
3747                                         }
3748                                 }
3749                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3750                                 u64 skip = 0;
3751                                 u64 trim = 0;
3752
3753                                 if (off > key.offset) {
3754                                         skip = off - key.offset;
3755                                         new_key.offset += skip;
3756                                 }
3757
3758                                 if (key.offset + datal > off + len)
3759                                         trim = key.offset + datal - (off + len);
3760
3761                                 if (comp && (skip || trim)) {
3762                                         ret = -EINVAL;
3763                                         btrfs_end_transaction(trans, root);
3764                                         goto out;
3765                                 }
3766                                 size -= skip + trim;
3767                                 datal -= skip + trim;
3768
3769                                 ret = clone_copy_inline_extent(src, inode,
3770                                                                trans, path,
3771                                                                &new_key,
3772                                                                drop_start,
3773                                                                datal,
3774                                                                skip, size, buf);
3775                                 if (ret) {
3776                                         if (ret != -EOPNOTSUPP)
3777                                                 btrfs_abort_transaction(trans,
3778                                                                         root,
3779                                                                         ret);
3780                                         btrfs_end_transaction(trans, root);
3781                                         goto out;
3782                                 }
3783                                 leaf = path->nodes[0];
3784                                 slot = path->slots[0];
3785                         }
3786
3787                         /* If we have an implicit hole (NO_HOLES feature). */
3788                         if (drop_start < new_key.offset)
3789                                 clone_update_extent_map(inode, trans,
3790                                                 NULL, drop_start,
3791                                                 new_key.offset - drop_start);
3792
3793                         clone_update_extent_map(inode, trans, path, 0, 0);
3794
3795                         btrfs_mark_buffer_dirty(leaf);
3796                         btrfs_release_path(path);
3797
3798                         last_dest_end = ALIGN(new_key.offset + datal,
3799                                               root->sectorsize);
3800                         ret = clone_finish_inode_update(trans, inode,
3801                                                         last_dest_end,
3802                                                         destoff, olen,
3803                                                         no_time_update);
3804                         if (ret)
3805                                 goto out;
3806                         if (new_key.offset + datal >= destoff + len)
3807                                 break;
3808                 }
3809                 btrfs_release_path(path);
3810                 key.offset = next_key_min_offset;
3811         }
3812         ret = 0;
3813
3814         if (last_dest_end < destoff + len) {
3815                 /*
3816                  * We have an implicit hole (NO_HOLES feature is enabled) that
3817                  * fully or partially overlaps our cloning range at its end.
3818                  */
3819                 btrfs_release_path(path);
3820
3821                 /*
3822                  * 1 - remove extent(s)
3823                  * 1 - inode update
3824                  */
3825                 trans = btrfs_start_transaction(root, 2);
3826                 if (IS_ERR(trans)) {
3827                         ret = PTR_ERR(trans);
3828                         goto out;
3829                 }
3830                 ret = btrfs_drop_extents(trans, root, inode,
3831                                          last_dest_end, destoff + len, 1);
3832                 if (ret) {
3833                         if (ret != -EOPNOTSUPP)
3834                                 btrfs_abort_transaction(trans, root, ret);
3835                         btrfs_end_transaction(trans, root);
3836                         goto out;
3837                 }
3838                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3839                                         destoff + len - last_dest_end);
3840                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3841                                                 destoff, olen, no_time_update);
3842         }
3843
3844 out:
3845         btrfs_free_path(path);
3846         vfree(buf);
3847         return ret;
3848 }
3849
3850 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3851                                        u64 off, u64 olen, u64 destoff)
3852 {
3853         struct inode *inode = file_inode(file);
3854         struct btrfs_root *root = BTRFS_I(inode)->root;
3855         struct fd src_file;
3856         struct inode *src;
3857         int ret;
3858         u64 len = olen;
3859         u64 bs = root->fs_info->sb->s_blocksize;
3860         int same_inode = 0;
3861
3862         /*
3863          * TODO:
3864          * - split compressed inline extents.  annoying: we need to
3865          *   decompress into destination's address_space (the file offset
3866          *   may change, so source mapping won't do), then recompress (or
3867          *   otherwise reinsert) a subrange.
3868          *
3869          * - split destination inode's inline extents.  The inline extents can
3870          *   be either compressed or non-compressed.
3871          */
3872
3873         /* the destination must be opened for writing */
3874         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3875                 return -EINVAL;
3876
3877         if (btrfs_root_readonly(root))
3878                 return -EROFS;
3879
3880         ret = mnt_want_write_file(file);
3881         if (ret)
3882                 return ret;
3883
3884         src_file = fdget(srcfd);
3885         if (!src_file.file) {
3886                 ret = -EBADF;
3887                 goto out_drop_write;
3888         }
3889
3890         ret = -EXDEV;
3891         if (src_file.file->f_path.mnt != file->f_path.mnt)
3892                 goto out_fput;
3893
3894         src = file_inode(src_file.file);
3895
3896         ret = -EINVAL;
3897         if (src == inode)
3898                 same_inode = 1;
3899
3900         /* the src must be open for reading */
3901         if (!(src_file.file->f_mode & FMODE_READ))
3902                 goto out_fput;
3903
3904         /* don't make the dst file partly checksummed */
3905         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3906             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3907                 goto out_fput;
3908
3909         ret = -EISDIR;
3910         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3911                 goto out_fput;
3912
3913         ret = -EXDEV;
3914         if (src->i_sb != inode->i_sb)
3915                 goto out_fput;
3916
3917         if (!same_inode) {
3918                 btrfs_double_inode_lock(src, inode);
3919         } else {
3920                 mutex_lock(&src->i_mutex);
3921         }
3922
3923         /* determine range to clone */
3924         ret = -EINVAL;
3925         if (off + len > src->i_size || off + len < off)
3926                 goto out_unlock;
3927         if (len == 0)
3928                 olen = len = src->i_size - off;
3929         /* if we extend to eof, continue to block boundary */
3930         if (off + len == src->i_size)
3931                 len = ALIGN(src->i_size, bs) - off;
3932
3933         if (len == 0) {
3934                 ret = 0;
3935                 goto out_unlock;
3936         }
3937
3938         /* verify the end result is block aligned */
3939         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3940             !IS_ALIGNED(destoff, bs))
3941                 goto out_unlock;
3942
3943         /* verify if ranges are overlapped within the same file */
3944         if (same_inode) {
3945                 if (destoff + len > off && destoff < off + len)
3946                         goto out_unlock;
3947         }
3948
3949         if (destoff > inode->i_size) {
3950                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3951                 if (ret)
3952                         goto out_unlock;
3953         }
3954
3955         /*
3956          * Lock the target range too. Right after we replace the file extent
3957          * items in the fs tree (which now point to the cloned data), we might
3958          * have a worker replace them with extent items relative to a write
3959          * operation that was issued before this clone operation (i.e. confront
3960          * with inode.c:btrfs_finish_ordered_io).
3961          */
3962         if (same_inode) {
3963                 u64 lock_start = min_t(u64, off, destoff);
3964                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3965
3966                 ret = lock_extent_range(src, lock_start, lock_len, true);
3967         } else {
3968                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3969                                                true);
3970         }
3971         ASSERT(ret == 0);
3972         if (WARN_ON(ret)) {
3973                 /* ranges in the io trees already unlocked */
3974                 goto out_unlock;
3975         }
3976
3977         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3978
3979         if (same_inode) {
3980                 u64 lock_start = min_t(u64, off, destoff);
3981                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3982
3983                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3984         } else {
3985                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3986         }
3987         /*
3988          * Truncate page cache pages so that future reads will see the cloned
3989          * data immediately and not the previous data.
3990          */
3991         truncate_inode_pages_range(&inode->i_data, destoff,
3992                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3993 out_unlock:
3994         if (!same_inode)
3995                 btrfs_double_inode_unlock(src, inode);
3996         else
3997                 mutex_unlock(&src->i_mutex);
3998 out_fput:
3999         fdput(src_file);
4000 out_drop_write:
4001         mnt_drop_write_file(file);
4002         return ret;
4003 }
4004
4005 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
4006 {
4007         struct btrfs_ioctl_clone_range_args args;
4008
4009         if (copy_from_user(&args, argp, sizeof(args)))
4010                 return -EFAULT;
4011         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
4012                                  args.src_length, args.dest_offset);
4013 }
4014
4015 /*
4016  * there are many ways the trans_start and trans_end ioctls can lead
4017  * to deadlocks.  They should only be used by applications that
4018  * basically own the machine, and have a very in depth understanding
4019  * of all the possible deadlocks and enospc problems.
4020  */
4021 static long btrfs_ioctl_trans_start(struct file *file)
4022 {
4023         struct inode *inode = file_inode(file);
4024         struct btrfs_root *root = BTRFS_I(inode)->root;
4025         struct btrfs_trans_handle *trans;
4026         int ret;
4027
4028         ret = -EPERM;
4029         if (!capable(CAP_SYS_ADMIN))
4030                 goto out;
4031
4032         ret = -EINPROGRESS;
4033         if (file->private_data)
4034                 goto out;
4035
4036         ret = -EROFS;
4037         if (btrfs_root_readonly(root))
4038                 goto out;
4039
4040         ret = mnt_want_write_file(file);
4041         if (ret)
4042                 goto out;
4043
4044         atomic_inc(&root->fs_info->open_ioctl_trans);
4045
4046         ret = -ENOMEM;
4047         trans = btrfs_start_ioctl_transaction(root);
4048         if (IS_ERR(trans))
4049                 goto out_drop;
4050
4051         file->private_data = trans;
4052         return 0;
4053
4054 out_drop:
4055         atomic_dec(&root->fs_info->open_ioctl_trans);
4056         mnt_drop_write_file(file);
4057 out:
4058         return ret;
4059 }
4060
4061 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4062 {
4063         struct inode *inode = file_inode(file);
4064         struct btrfs_root *root = BTRFS_I(inode)->root;
4065         struct btrfs_root *new_root;
4066         struct btrfs_dir_item *di;
4067         struct btrfs_trans_handle *trans;
4068         struct btrfs_path *path;
4069         struct btrfs_key location;
4070         struct btrfs_disk_key disk_key;
4071         u64 objectid = 0;
4072         u64 dir_id;
4073         int ret;
4074
4075         if (!capable(CAP_SYS_ADMIN))
4076                 return -EPERM;
4077
4078         ret = mnt_want_write_file(file);
4079         if (ret)
4080                 return ret;
4081
4082         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4083                 ret = -EFAULT;
4084                 goto out;
4085         }
4086
4087         if (!objectid)
4088                 objectid = BTRFS_FS_TREE_OBJECTID;
4089
4090         location.objectid = objectid;
4091         location.type = BTRFS_ROOT_ITEM_KEY;
4092         location.offset = (u64)-1;
4093
4094         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4095         if (IS_ERR(new_root)) {
4096                 ret = PTR_ERR(new_root);
4097                 goto out;
4098         }
4099
4100         path = btrfs_alloc_path();
4101         if (!path) {
4102                 ret = -ENOMEM;
4103                 goto out;
4104         }
4105         path->leave_spinning = 1;
4106
4107         trans = btrfs_start_transaction(root, 1);
4108         if (IS_ERR(trans)) {
4109                 btrfs_free_path(path);
4110                 ret = PTR_ERR(trans);
4111                 goto out;
4112         }
4113
4114         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4115         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4116                                    dir_id, "default", 7, 1);
4117         if (IS_ERR_OR_NULL(di)) {
4118                 btrfs_free_path(path);
4119                 btrfs_end_transaction(trans, root);
4120                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4121                            "item, this isn't going to work");
4122                 ret = -ENOENT;
4123                 goto out;
4124         }
4125
4126         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4127         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4128         btrfs_mark_buffer_dirty(path->nodes[0]);
4129         btrfs_free_path(path);
4130
4131         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4132         btrfs_end_transaction(trans, root);
4133 out:
4134         mnt_drop_write_file(file);
4135         return ret;
4136 }
4137
4138 void btrfs_get_block_group_info(struct list_head *groups_list,
4139                                 struct btrfs_ioctl_space_info *space)
4140 {
4141         struct btrfs_block_group_cache *block_group;
4142
4143         space->total_bytes = 0;
4144         space->used_bytes = 0;
4145         space->flags = 0;
4146         list_for_each_entry(block_group, groups_list, list) {
4147                 space->flags = block_group->flags;
4148                 space->total_bytes += block_group->key.offset;
4149                 space->used_bytes +=
4150                         btrfs_block_group_used(&block_group->item);
4151         }
4152 }
4153
4154 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4155 {
4156         struct btrfs_ioctl_space_args space_args;
4157         struct btrfs_ioctl_space_info space;
4158         struct btrfs_ioctl_space_info *dest;
4159         struct btrfs_ioctl_space_info *dest_orig;
4160         struct btrfs_ioctl_space_info __user *user_dest;
4161         struct btrfs_space_info *info;
4162         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4163                        BTRFS_BLOCK_GROUP_SYSTEM,
4164                        BTRFS_BLOCK_GROUP_METADATA,
4165                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4166         int num_types = 4;
4167         int alloc_size;
4168         int ret = 0;
4169         u64 slot_count = 0;
4170         int i, c;
4171
4172         if (copy_from_user(&space_args,
4173                            (struct btrfs_ioctl_space_args __user *)arg,
4174                            sizeof(space_args)))
4175                 return -EFAULT;
4176
4177         for (i = 0; i < num_types; i++) {
4178                 struct btrfs_space_info *tmp;
4179
4180                 info = NULL;
4181                 rcu_read_lock();
4182                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4183                                         list) {
4184                         if (tmp->flags == types[i]) {
4185                                 info = tmp;
4186                                 break;
4187                         }
4188                 }
4189                 rcu_read_unlock();
4190
4191                 if (!info)
4192                         continue;
4193
4194                 down_read(&info->groups_sem);
4195                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4196                         if (!list_empty(&info->block_groups[c]))
4197                                 slot_count++;
4198                 }
4199                 up_read(&info->groups_sem);
4200         }
4201
4202         /*
4203          * Global block reserve, exported as a space_info
4204          */
4205         slot_count++;
4206
4207         /* space_slots == 0 means they are asking for a count */
4208         if (space_args.space_slots == 0) {
4209                 space_args.total_spaces = slot_count;
4210                 goto out;
4211         }
4212
4213         slot_count = min_t(u64, space_args.space_slots, slot_count);
4214
4215         alloc_size = sizeof(*dest) * slot_count;
4216
4217         /* we generally have at most 6 or so space infos, one for each raid
4218          * level.  So, a whole page should be more than enough for everyone
4219          */
4220         if (alloc_size > PAGE_CACHE_SIZE)
4221                 return -ENOMEM;
4222
4223         space_args.total_spaces = 0;
4224         dest = kmalloc(alloc_size, GFP_KERNEL);
4225         if (!dest)
4226                 return -ENOMEM;
4227         dest_orig = dest;
4228
4229         /* now we have a buffer to copy into */
4230         for (i = 0; i < num_types; i++) {
4231                 struct btrfs_space_info *tmp;
4232
4233                 if (!slot_count)
4234                         break;
4235
4236                 info = NULL;
4237                 rcu_read_lock();
4238                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4239                                         list) {
4240                         if (tmp->flags == types[i]) {
4241                                 info = tmp;
4242                                 break;
4243                         }
4244                 }
4245                 rcu_read_unlock();
4246
4247                 if (!info)
4248                         continue;
4249                 down_read(&info->groups_sem);
4250                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4251                         if (!list_empty(&info->block_groups[c])) {
4252                                 btrfs_get_block_group_info(
4253                                         &info->block_groups[c], &space);
4254                                 memcpy(dest, &space, sizeof(space));
4255                                 dest++;
4256                                 space_args.total_spaces++;
4257                                 slot_count--;
4258                         }
4259                         if (!slot_count)
4260                                 break;
4261                 }
4262                 up_read(&info->groups_sem);
4263         }
4264
4265         /*
4266          * Add global block reserve
4267          */
4268         if (slot_count) {
4269                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4270
4271                 spin_lock(&block_rsv->lock);
4272                 space.total_bytes = block_rsv->size;
4273                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4274                 spin_unlock(&block_rsv->lock);
4275                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4276                 memcpy(dest, &space, sizeof(space));
4277                 space_args.total_spaces++;
4278         }
4279
4280         user_dest = (struct btrfs_ioctl_space_info __user *)
4281                 (arg + sizeof(struct btrfs_ioctl_space_args));
4282
4283         if (copy_to_user(user_dest, dest_orig, alloc_size))
4284                 ret = -EFAULT;
4285
4286         kfree(dest_orig);
4287 out:
4288         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4289                 ret = -EFAULT;
4290
4291         return ret;
4292 }
4293
4294 /*
4295  * there are many ways the trans_start and trans_end ioctls can lead
4296  * to deadlocks.  They should only be used by applications that
4297  * basically own the machine, and have a very in depth understanding
4298  * of all the possible deadlocks and enospc problems.
4299  */
4300 long btrfs_ioctl_trans_end(struct file *file)
4301 {
4302         struct inode *inode = file_inode(file);
4303         struct btrfs_root *root = BTRFS_I(inode)->root;
4304         struct btrfs_trans_handle *trans;
4305
4306         trans = file->private_data;
4307         if (!trans)
4308                 return -EINVAL;
4309         file->private_data = NULL;
4310
4311         btrfs_end_transaction(trans, root);
4312
4313         atomic_dec(&root->fs_info->open_ioctl_trans);
4314
4315         mnt_drop_write_file(file);
4316         return 0;
4317 }
4318
4319 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4320                                             void __user *argp)
4321 {
4322         struct btrfs_trans_handle *trans;
4323         u64 transid;
4324         int ret;
4325
4326         trans = btrfs_attach_transaction_barrier(root);
4327         if (IS_ERR(trans)) {
4328                 if (PTR_ERR(trans) != -ENOENT)
4329                         return PTR_ERR(trans);
4330
4331                 /* No running transaction, don't bother */
4332                 transid = root->fs_info->last_trans_committed;
4333                 goto out;
4334         }
4335         transid = trans->transid;
4336         ret = btrfs_commit_transaction_async(trans, root, 0);
4337         if (ret) {
4338                 btrfs_end_transaction(trans, root);
4339                 return ret;
4340         }
4341 out:
4342         if (argp)
4343                 if (copy_to_user(argp, &transid, sizeof(transid)))
4344                         return -EFAULT;
4345         return 0;
4346 }
4347
4348 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4349                                            void __user *argp)
4350 {
4351         u64 transid;
4352
4353         if (argp) {
4354                 if (copy_from_user(&transid, argp, sizeof(transid)))
4355                         return -EFAULT;
4356         } else {
4357                 transid = 0;  /* current trans */
4358         }
4359         return btrfs_wait_for_commit(root, transid);
4360 }
4361
4362 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4363 {
4364         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4365         struct btrfs_ioctl_scrub_args *sa;
4366         int ret;
4367
4368         if (!capable(CAP_SYS_ADMIN))
4369                 return -EPERM;
4370
4371         sa = memdup_user(arg, sizeof(*sa));
4372         if (IS_ERR(sa))
4373                 return PTR_ERR(sa);
4374
4375         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4376                 ret = mnt_want_write_file(file);
4377                 if (ret)
4378                         goto out;
4379         }
4380
4381         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4382                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4383                               0);
4384
4385         if (copy_to_user(arg, sa, sizeof(*sa)))
4386                 ret = -EFAULT;
4387
4388         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4389                 mnt_drop_write_file(file);
4390 out:
4391         kfree(sa);
4392         return ret;
4393 }
4394
4395 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4396 {
4397         if (!capable(CAP_SYS_ADMIN))
4398                 return -EPERM;
4399
4400         return btrfs_scrub_cancel(root->fs_info);
4401 }
4402
4403 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4404                                        void __user *arg)
4405 {
4406         struct btrfs_ioctl_scrub_args *sa;
4407         int ret;
4408
4409         if (!capable(CAP_SYS_ADMIN))
4410                 return -EPERM;
4411
4412         sa = memdup_user(arg, sizeof(*sa));
4413         if (IS_ERR(sa))
4414                 return PTR_ERR(sa);
4415
4416         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4417
4418         if (copy_to_user(arg, sa, sizeof(*sa)))
4419                 ret = -EFAULT;
4420
4421         kfree(sa);
4422         return ret;
4423 }
4424
4425 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4426                                       void __user *arg)
4427 {
4428         struct btrfs_ioctl_get_dev_stats *sa;
4429         int ret;
4430
4431         sa = memdup_user(arg, sizeof(*sa));
4432         if (IS_ERR(sa))
4433                 return PTR_ERR(sa);
4434
4435         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4436                 kfree(sa);
4437                 return -EPERM;
4438         }
4439
4440         ret = btrfs_get_dev_stats(root, sa);
4441
4442         if (copy_to_user(arg, sa, sizeof(*sa)))
4443                 ret = -EFAULT;
4444
4445         kfree(sa);
4446         return ret;
4447 }
4448
4449 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4450 {
4451         struct btrfs_ioctl_dev_replace_args *p;
4452         int ret;
4453
4454         if (!capable(CAP_SYS_ADMIN))
4455                 return -EPERM;
4456
4457         p = memdup_user(arg, sizeof(*p));
4458         if (IS_ERR(p))
4459                 return PTR_ERR(p);
4460
4461         switch (p->cmd) {
4462         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4463                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4464                         ret = -EROFS;
4465                         goto out;
4466                 }
4467                 if (atomic_xchg(
4468                         &root->fs_info->mutually_exclusive_operation_running,
4469                         1)) {
4470                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4471                 } else {
4472                         ret = btrfs_dev_replace_start(root, p);
4473                         atomic_set(
4474                          &root->fs_info->mutually_exclusive_operation_running,
4475                          0);
4476                 }
4477                 break;
4478         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4479                 btrfs_dev_replace_status(root->fs_info, p);
4480                 ret = 0;
4481                 break;
4482         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4483                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4484                 break;
4485         default:
4486                 ret = -EINVAL;
4487                 break;
4488         }
4489
4490         if (copy_to_user(arg, p, sizeof(*p)))
4491                 ret = -EFAULT;
4492 out:
4493         kfree(p);
4494         return ret;
4495 }
4496
4497 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4498 {
4499         int ret = 0;
4500         int i;
4501         u64 rel_ptr;
4502         int size;
4503         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4504         struct inode_fs_paths *ipath = NULL;
4505         struct btrfs_path *path;
4506
4507         if (!capable(CAP_DAC_READ_SEARCH))
4508                 return -EPERM;
4509
4510         path = btrfs_alloc_path();
4511         if (!path) {
4512                 ret = -ENOMEM;
4513                 goto out;
4514         }
4515
4516         ipa = memdup_user(arg, sizeof(*ipa));
4517         if (IS_ERR(ipa)) {
4518                 ret = PTR_ERR(ipa);
4519                 ipa = NULL;
4520                 goto out;
4521         }
4522
4523         size = min_t(u32, ipa->size, 4096);
4524         ipath = init_ipath(size, root, path);
4525         if (IS_ERR(ipath)) {
4526                 ret = PTR_ERR(ipath);
4527                 ipath = NULL;
4528                 goto out;
4529         }
4530
4531         ret = paths_from_inode(ipa->inum, ipath);
4532         if (ret < 0)
4533                 goto out;
4534
4535         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4536                 rel_ptr = ipath->fspath->val[i] -
4537                           (u64)(unsigned long)ipath->fspath->val;
4538                 ipath->fspath->val[i] = rel_ptr;
4539         }
4540
4541         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4542                            (void *)(unsigned long)ipath->fspath, size);
4543         if (ret) {
4544                 ret = -EFAULT;
4545                 goto out;
4546         }
4547
4548 out:
4549         btrfs_free_path(path);
4550         free_ipath(ipath);
4551         kfree(ipa);
4552
4553         return ret;
4554 }
4555
4556 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4557 {
4558         struct btrfs_data_container *inodes = ctx;
4559         const size_t c = 3 * sizeof(u64);
4560
4561         if (inodes->bytes_left >= c) {
4562                 inodes->bytes_left -= c;
4563                 inodes->val[inodes->elem_cnt] = inum;
4564                 inodes->val[inodes->elem_cnt + 1] = offset;
4565                 inodes->val[inodes->elem_cnt + 2] = root;
4566                 inodes->elem_cnt += 3;
4567         } else {
4568                 inodes->bytes_missing += c - inodes->bytes_left;
4569                 inodes->bytes_left = 0;
4570                 inodes->elem_missed += 3;
4571         }
4572
4573         return 0;
4574 }
4575
4576 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4577                                         void __user *arg)
4578 {
4579         int ret = 0;
4580         int size;
4581         struct btrfs_ioctl_logical_ino_args *loi;
4582         struct btrfs_data_container *inodes = NULL;
4583         struct btrfs_path *path = NULL;
4584
4585         if (!capable(CAP_SYS_ADMIN))
4586                 return -EPERM;
4587
4588         loi = memdup_user(arg, sizeof(*loi));
4589         if (IS_ERR(loi)) {
4590                 ret = PTR_ERR(loi);
4591                 loi = NULL;
4592                 goto out;
4593         }
4594
4595         path = btrfs_alloc_path();
4596         if (!path) {
4597                 ret = -ENOMEM;
4598                 goto out;
4599         }
4600
4601         size = min_t(u32, loi->size, SZ_64K);
4602         inodes = init_data_container(size);
4603         if (IS_ERR(inodes)) {
4604                 ret = PTR_ERR(inodes);
4605                 inodes = NULL;
4606                 goto out;
4607         }
4608
4609         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4610                                           build_ino_list, inodes);
4611         if (ret == -EINVAL)
4612                 ret = -ENOENT;
4613         if (ret < 0)
4614                 goto out;
4615
4616         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4617                            (void *)(unsigned long)inodes, size);
4618         if (ret)
4619                 ret = -EFAULT;
4620
4621 out:
4622         btrfs_free_path(path);
4623         vfree(inodes);
4624         kfree(loi);
4625
4626         return ret;
4627 }
4628
4629 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4630                                struct btrfs_ioctl_balance_args *bargs)
4631 {
4632         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4633
4634         bargs->flags = bctl->flags;
4635
4636         if (atomic_read(&fs_info->balance_running))
4637                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4638         if (atomic_read(&fs_info->balance_pause_req))
4639                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4640         if (atomic_read(&fs_info->balance_cancel_req))
4641                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4642
4643         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4644         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4645         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4646
4647         if (lock) {
4648                 spin_lock(&fs_info->balance_lock);
4649                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4650                 spin_unlock(&fs_info->balance_lock);
4651         } else {
4652                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4653         }
4654 }
4655
4656 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4657 {
4658         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4659         struct btrfs_fs_info *fs_info = root->fs_info;
4660         struct btrfs_ioctl_balance_args *bargs;
4661         struct btrfs_balance_control *bctl;
4662         bool need_unlock; /* for mut. excl. ops lock */
4663         int ret;
4664
4665         if (!capable(CAP_SYS_ADMIN))
4666                 return -EPERM;
4667
4668         ret = mnt_want_write_file(file);
4669         if (ret)
4670                 return ret;
4671
4672 again:
4673         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4674                 mutex_lock(&fs_info->volume_mutex);
4675                 mutex_lock(&fs_info->balance_mutex);
4676                 need_unlock = true;
4677                 goto locked;
4678         }
4679
4680         /*
4681          * mut. excl. ops lock is locked.  Three possibilites:
4682          *   (1) some other op is running
4683          *   (2) balance is running
4684          *   (3) balance is paused -- special case (think resume)
4685          */
4686         mutex_lock(&fs_info->balance_mutex);
4687         if (fs_info->balance_ctl) {
4688                 /* this is either (2) or (3) */
4689                 if (!atomic_read(&fs_info->balance_running)) {
4690                         mutex_unlock(&fs_info->balance_mutex);
4691                         if (!mutex_trylock(&fs_info->volume_mutex))
4692                                 goto again;
4693                         mutex_lock(&fs_info->balance_mutex);
4694
4695                         if (fs_info->balance_ctl &&
4696                             !atomic_read(&fs_info->balance_running)) {
4697                                 /* this is (3) */
4698                                 need_unlock = false;
4699                                 goto locked;
4700                         }
4701
4702                         mutex_unlock(&fs_info->balance_mutex);
4703                         mutex_unlock(&fs_info->volume_mutex);
4704                         goto again;
4705                 } else {
4706                         /* this is (2) */
4707                         mutex_unlock(&fs_info->balance_mutex);
4708                         ret = -EINPROGRESS;
4709                         goto out;
4710                 }
4711         } else {
4712                 /* this is (1) */
4713                 mutex_unlock(&fs_info->balance_mutex);
4714                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4715                 goto out;
4716         }
4717
4718 locked:
4719         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4720
4721         if (arg) {
4722                 bargs = memdup_user(arg, sizeof(*bargs));
4723                 if (IS_ERR(bargs)) {
4724                         ret = PTR_ERR(bargs);
4725                         goto out_unlock;
4726                 }
4727
4728                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4729                         if (!fs_info->balance_ctl) {
4730                                 ret = -ENOTCONN;
4731                                 goto out_bargs;
4732                         }
4733
4734                         bctl = fs_info->balance_ctl;
4735                         spin_lock(&fs_info->balance_lock);
4736                         bctl->flags |= BTRFS_BALANCE_RESUME;
4737                         spin_unlock(&fs_info->balance_lock);
4738
4739                         goto do_balance;
4740                 }
4741         } else {
4742                 bargs = NULL;
4743         }
4744
4745         if (fs_info->balance_ctl) {
4746                 ret = -EINPROGRESS;
4747                 goto out_bargs;
4748         }
4749
4750         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4751         if (!bctl) {
4752                 ret = -ENOMEM;
4753                 goto out_bargs;
4754         }
4755
4756         bctl->fs_info = fs_info;
4757         if (arg) {
4758                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4759                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4760                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4761
4762                 bctl->flags = bargs->flags;
4763         } else {
4764                 /* balance everything - no filters */
4765                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4766         }
4767
4768         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4769                 ret = -EINVAL;
4770                 goto out_bctl;
4771         }
4772
4773 do_balance:
4774         /*
4775          * Ownership of bctl and mutually_exclusive_operation_running
4776          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4777          * or, if restriper was paused all the way until unmount, in
4778          * free_fs_info.  mutually_exclusive_operation_running is
4779          * cleared in __cancel_balance.
4780          */
4781         need_unlock = false;
4782
4783         ret = btrfs_balance(bctl, bargs);
4784         bctl = NULL;
4785
4786         if (arg) {
4787                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4788                         ret = -EFAULT;
4789         }
4790
4791 out_bctl:
4792         kfree(bctl);
4793 out_bargs:
4794         kfree(bargs);
4795 out_unlock:
4796         mutex_unlock(&fs_info->balance_mutex);
4797         mutex_unlock(&fs_info->volume_mutex);
4798         if (need_unlock)
4799                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4800 out:
4801         mnt_drop_write_file(file);
4802         return ret;
4803 }
4804
4805 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4806 {
4807         if (!capable(CAP_SYS_ADMIN))
4808                 return -EPERM;
4809
4810         switch (cmd) {
4811         case BTRFS_BALANCE_CTL_PAUSE:
4812                 return btrfs_pause_balance(root->fs_info);
4813         case BTRFS_BALANCE_CTL_CANCEL:
4814                 return btrfs_cancel_balance(root->fs_info);
4815         }
4816
4817         return -EINVAL;
4818 }
4819
4820 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4821                                          void __user *arg)
4822 {
4823         struct btrfs_fs_info *fs_info = root->fs_info;
4824         struct btrfs_ioctl_balance_args *bargs;
4825         int ret = 0;
4826
4827         if (!capable(CAP_SYS_ADMIN))
4828                 return -EPERM;
4829
4830         mutex_lock(&fs_info->balance_mutex);
4831         if (!fs_info->balance_ctl) {
4832                 ret = -ENOTCONN;
4833                 goto out;
4834         }
4835
4836         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4837         if (!bargs) {
4838                 ret = -ENOMEM;
4839                 goto out;
4840         }
4841
4842         update_ioctl_balance_args(fs_info, 1, bargs);
4843
4844         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4845                 ret = -EFAULT;
4846
4847         kfree(bargs);
4848 out:
4849         mutex_unlock(&fs_info->balance_mutex);
4850         return ret;
4851 }
4852
4853 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4854 {
4855         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4856         struct btrfs_ioctl_quota_ctl_args *sa;
4857         struct btrfs_trans_handle *trans = NULL;
4858         int ret;
4859         int err;
4860
4861         if (!capable(CAP_SYS_ADMIN))
4862                 return -EPERM;
4863
4864         ret = mnt_want_write_file(file);
4865         if (ret)
4866                 return ret;
4867
4868         sa = memdup_user(arg, sizeof(*sa));
4869         if (IS_ERR(sa)) {
4870                 ret = PTR_ERR(sa);
4871                 goto drop_write;
4872         }
4873
4874         down_write(&root->fs_info->subvol_sem);
4875         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4876         if (IS_ERR(trans)) {
4877                 ret = PTR_ERR(trans);
4878                 goto out;
4879         }
4880
4881         switch (sa->cmd) {
4882         case BTRFS_QUOTA_CTL_ENABLE:
4883                 ret = btrfs_quota_enable(trans, root->fs_info);
4884                 break;
4885         case BTRFS_QUOTA_CTL_DISABLE:
4886                 ret = btrfs_quota_disable(trans, root->fs_info);
4887                 break;
4888         default:
4889                 ret = -EINVAL;
4890                 break;
4891         }
4892
4893         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4894         if (err && !ret)
4895                 ret = err;
4896 out:
4897         kfree(sa);
4898         up_write(&root->fs_info->subvol_sem);
4899 drop_write:
4900         mnt_drop_write_file(file);
4901         return ret;
4902 }
4903
4904 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4905 {
4906         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4907         struct btrfs_ioctl_qgroup_assign_args *sa;
4908         struct btrfs_trans_handle *trans;
4909         int ret;
4910         int err;
4911
4912         if (!capable(CAP_SYS_ADMIN))
4913                 return -EPERM;
4914
4915         ret = mnt_want_write_file(file);
4916         if (ret)
4917                 return ret;
4918
4919         sa = memdup_user(arg, sizeof(*sa));
4920         if (IS_ERR(sa)) {
4921                 ret = PTR_ERR(sa);
4922                 goto drop_write;
4923         }
4924
4925         trans = btrfs_join_transaction(root);
4926         if (IS_ERR(trans)) {
4927                 ret = PTR_ERR(trans);
4928                 goto out;
4929         }
4930
4931         /* FIXME: check if the IDs really exist */
4932         if (sa->assign) {
4933                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4934                                                 sa->src, sa->dst);
4935         } else {
4936                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4937                                                 sa->src, sa->dst);
4938         }
4939
4940         /* update qgroup status and info */
4941         err = btrfs_run_qgroups(trans, root->fs_info);
4942         if (err < 0)
4943                 btrfs_std_error(root->fs_info, ret,
4944                             "failed to update qgroup status and info\n");
4945         err = btrfs_end_transaction(trans, root);
4946         if (err && !ret)
4947                 ret = err;
4948
4949 out:
4950         kfree(sa);
4951 drop_write:
4952         mnt_drop_write_file(file);
4953         return ret;
4954 }
4955
4956 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4957 {
4958         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4959         struct btrfs_ioctl_qgroup_create_args *sa;
4960         struct btrfs_trans_handle *trans;
4961         int ret;
4962         int err;
4963
4964         if (!capable(CAP_SYS_ADMIN))
4965                 return -EPERM;
4966
4967         ret = mnt_want_write_file(file);
4968         if (ret)
4969                 return ret;
4970
4971         sa = memdup_user(arg, sizeof(*sa));
4972         if (IS_ERR(sa)) {
4973                 ret = PTR_ERR(sa);
4974                 goto drop_write;
4975         }
4976
4977         if (!sa->qgroupid) {
4978                 ret = -EINVAL;
4979                 goto out;
4980         }
4981
4982         trans = btrfs_join_transaction(root);
4983         if (IS_ERR(trans)) {
4984                 ret = PTR_ERR(trans);
4985                 goto out;
4986         }
4987
4988         /* FIXME: check if the IDs really exist */
4989         if (sa->create) {
4990                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4991         } else {
4992                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4993         }
4994
4995         err = btrfs_end_transaction(trans, root);
4996         if (err && !ret)
4997                 ret = err;
4998
4999 out:
5000         kfree(sa);
5001 drop_write:
5002         mnt_drop_write_file(file);
5003         return ret;
5004 }
5005
5006 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5007 {
5008         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5009         struct btrfs_ioctl_qgroup_limit_args *sa;
5010         struct btrfs_trans_handle *trans;
5011         int ret;
5012         int err;
5013         u64 qgroupid;
5014
5015         if (!capable(CAP_SYS_ADMIN))
5016                 return -EPERM;
5017
5018         ret = mnt_want_write_file(file);
5019         if (ret)
5020                 return ret;
5021
5022         sa = memdup_user(arg, sizeof(*sa));
5023         if (IS_ERR(sa)) {
5024                 ret = PTR_ERR(sa);
5025                 goto drop_write;
5026         }
5027
5028         trans = btrfs_join_transaction(root);
5029         if (IS_ERR(trans)) {
5030                 ret = PTR_ERR(trans);
5031                 goto out;
5032         }
5033
5034         qgroupid = sa->qgroupid;
5035         if (!qgroupid) {
5036                 /* take the current subvol as qgroup */
5037                 qgroupid = root->root_key.objectid;
5038         }
5039
5040         /* FIXME: check if the IDs really exist */
5041         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
5042
5043         err = btrfs_end_transaction(trans, root);
5044         if (err && !ret)
5045                 ret = err;
5046
5047 out:
5048         kfree(sa);
5049 drop_write:
5050         mnt_drop_write_file(file);
5051         return ret;
5052 }
5053
5054 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5055 {
5056         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5057         struct btrfs_ioctl_quota_rescan_args *qsa;
5058         int ret;
5059
5060         if (!capable(CAP_SYS_ADMIN))
5061                 return -EPERM;
5062
5063         ret = mnt_want_write_file(file);
5064         if (ret)
5065                 return ret;
5066
5067         qsa = memdup_user(arg, sizeof(*qsa));
5068         if (IS_ERR(qsa)) {
5069                 ret = PTR_ERR(qsa);
5070                 goto drop_write;
5071         }
5072
5073         if (qsa->flags) {
5074                 ret = -EINVAL;
5075                 goto out;
5076         }
5077
5078         ret = btrfs_qgroup_rescan(root->fs_info);
5079
5080 out:
5081         kfree(qsa);
5082 drop_write:
5083         mnt_drop_write_file(file);
5084         return ret;
5085 }
5086
5087 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5088 {
5089         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5090         struct btrfs_ioctl_quota_rescan_args *qsa;
5091         int ret = 0;
5092
5093         if (!capable(CAP_SYS_ADMIN))
5094                 return -EPERM;
5095
5096         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5097         if (!qsa)
5098                 return -ENOMEM;
5099
5100         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5101                 qsa->flags = 1;
5102                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5103         }
5104
5105         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5106                 ret = -EFAULT;
5107
5108         kfree(qsa);
5109         return ret;
5110 }
5111
5112 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5113 {
5114         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5115
5116         if (!capable(CAP_SYS_ADMIN))
5117                 return -EPERM;
5118
5119         return btrfs_qgroup_wait_for_completion(root->fs_info);
5120 }
5121
5122 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5123                                             struct btrfs_ioctl_received_subvol_args *sa)
5124 {
5125         struct inode *inode = file_inode(file);
5126         struct btrfs_root *root = BTRFS_I(inode)->root;
5127         struct btrfs_root_item *root_item = &root->root_item;
5128         struct btrfs_trans_handle *trans;
5129         struct timespec ct = CURRENT_TIME;
5130         int ret = 0;
5131         int received_uuid_changed;
5132
5133         if (!inode_owner_or_capable(inode))
5134                 return -EPERM;
5135
5136         ret = mnt_want_write_file(file);
5137         if (ret < 0)
5138                 return ret;
5139
5140         down_write(&root->fs_info->subvol_sem);
5141
5142         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5143                 ret = -EINVAL;
5144                 goto out;
5145         }
5146
5147         if (btrfs_root_readonly(root)) {
5148                 ret = -EROFS;
5149                 goto out;
5150         }
5151
5152         /*
5153          * 1 - root item
5154          * 2 - uuid items (received uuid + subvol uuid)
5155          */
5156         trans = btrfs_start_transaction(root, 3);
5157         if (IS_ERR(trans)) {
5158                 ret = PTR_ERR(trans);
5159                 trans = NULL;
5160                 goto out;
5161         }
5162
5163         sa->rtransid = trans->transid;
5164         sa->rtime.sec = ct.tv_sec;
5165         sa->rtime.nsec = ct.tv_nsec;
5166
5167         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5168                                        BTRFS_UUID_SIZE);
5169         if (received_uuid_changed &&
5170             !btrfs_is_empty_uuid(root_item->received_uuid))
5171                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5172                                     root_item->received_uuid,
5173                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5174                                     root->root_key.objectid);
5175         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5176         btrfs_set_root_stransid(root_item, sa->stransid);
5177         btrfs_set_root_rtransid(root_item, sa->rtransid);
5178         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5179         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5180         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5181         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5182
5183         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5184                                 &root->root_key, &root->root_item);
5185         if (ret < 0) {
5186                 btrfs_end_transaction(trans, root);
5187                 goto out;
5188         }
5189         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5190                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5191                                           sa->uuid,
5192                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5193                                           root->root_key.objectid);
5194                 if (ret < 0 && ret != -EEXIST) {
5195                         btrfs_abort_transaction(trans, root, ret);
5196                         goto out;
5197                 }
5198         }
5199         ret = btrfs_commit_transaction(trans, root);
5200         if (ret < 0) {
5201                 btrfs_abort_transaction(trans, root, ret);
5202                 goto out;
5203         }
5204
5205 out:
5206         up_write(&root->fs_info->subvol_sem);
5207         mnt_drop_write_file(file);
5208         return ret;
5209 }
5210
5211 #ifdef CONFIG_64BIT
5212 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5213                                                 void __user *arg)
5214 {
5215         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5216         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5217         int ret = 0;
5218
5219         args32 = memdup_user(arg, sizeof(*args32));
5220         if (IS_ERR(args32)) {
5221                 ret = PTR_ERR(args32);
5222                 args32 = NULL;
5223                 goto out;
5224         }
5225
5226         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5227         if (!args64) {
5228                 ret = -ENOMEM;
5229                 goto out;
5230         }
5231
5232         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5233         args64->stransid = args32->stransid;
5234         args64->rtransid = args32->rtransid;
5235         args64->stime.sec = args32->stime.sec;
5236         args64->stime.nsec = args32->stime.nsec;
5237         args64->rtime.sec = args32->rtime.sec;
5238         args64->rtime.nsec = args32->rtime.nsec;
5239         args64->flags = args32->flags;
5240
5241         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5242         if (ret)
5243                 goto out;
5244
5245         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5246         args32->stransid = args64->stransid;
5247         args32->rtransid = args64->rtransid;
5248         args32->stime.sec = args64->stime.sec;
5249         args32->stime.nsec = args64->stime.nsec;
5250         args32->rtime.sec = args64->rtime.sec;
5251         args32->rtime.nsec = args64->rtime.nsec;
5252         args32->flags = args64->flags;
5253
5254         ret = copy_to_user(arg, args32, sizeof(*args32));
5255         if (ret)
5256                 ret = -EFAULT;
5257
5258 out:
5259         kfree(args32);
5260         kfree(args64);
5261         return ret;
5262 }
5263 #endif
5264
5265 static long btrfs_ioctl_set_received_subvol(struct file *file,
5266                                             void __user *arg)
5267 {
5268         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5269         int ret = 0;
5270
5271         sa = memdup_user(arg, sizeof(*sa));
5272         if (IS_ERR(sa)) {
5273                 ret = PTR_ERR(sa);
5274                 sa = NULL;
5275                 goto out;
5276         }
5277
5278         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5279
5280         if (ret)
5281                 goto out;
5282
5283         ret = copy_to_user(arg, sa, sizeof(*sa));
5284         if (ret)
5285                 ret = -EFAULT;
5286
5287 out:
5288         kfree(sa);
5289         return ret;
5290 }
5291
5292 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5293 {
5294         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5295         size_t len;
5296         int ret;
5297         char label[BTRFS_LABEL_SIZE];
5298
5299         spin_lock(&root->fs_info->super_lock);
5300         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5301         spin_unlock(&root->fs_info->super_lock);
5302
5303         len = strnlen(label, BTRFS_LABEL_SIZE);
5304
5305         if (len == BTRFS_LABEL_SIZE) {
5306                 btrfs_warn(root->fs_info,
5307                         "label is too long, return the first %zu bytes", --len);
5308         }
5309
5310         ret = copy_to_user(arg, label, len);
5311
5312         return ret ? -EFAULT : 0;
5313 }
5314
5315 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5316 {
5317         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5318         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5319         struct btrfs_trans_handle *trans;
5320         char label[BTRFS_LABEL_SIZE];
5321         int ret;
5322
5323         if (!capable(CAP_SYS_ADMIN))
5324                 return -EPERM;
5325
5326         if (copy_from_user(label, arg, sizeof(label)))
5327                 return -EFAULT;
5328
5329         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5330                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5331                        BTRFS_LABEL_SIZE - 1);
5332                 return -EINVAL;
5333         }
5334
5335         ret = mnt_want_write_file(file);
5336         if (ret)
5337                 return ret;
5338
5339         trans = btrfs_start_transaction(root, 0);
5340         if (IS_ERR(trans)) {
5341                 ret = PTR_ERR(trans);
5342                 goto out_unlock;
5343         }
5344
5345         spin_lock(&root->fs_info->super_lock);
5346         strcpy(super_block->label, label);
5347         spin_unlock(&root->fs_info->super_lock);
5348         ret = btrfs_commit_transaction(trans, root);
5349
5350 out_unlock:
5351         mnt_drop_write_file(file);
5352         return ret;
5353 }
5354
5355 #define INIT_FEATURE_FLAGS(suffix) \
5356         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5357           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5358           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5359
5360 static int btrfs_ioctl_get_supported_features(struct file *file,
5361                                               void __user *arg)
5362 {
5363         static const struct btrfs_ioctl_feature_flags features[3] = {
5364                 INIT_FEATURE_FLAGS(SUPP),
5365                 INIT_FEATURE_FLAGS(SAFE_SET),
5366                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5367         };
5368
5369         if (copy_to_user(arg, &features, sizeof(features)))
5370                 return -EFAULT;
5371
5372         return 0;
5373 }
5374
5375 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5376 {
5377         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5378         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5379         struct btrfs_ioctl_feature_flags features;
5380
5381         features.compat_flags = btrfs_super_compat_flags(super_block);
5382         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5383         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5384
5385         if (copy_to_user(arg, &features, sizeof(features)))
5386                 return -EFAULT;
5387
5388         return 0;
5389 }
5390
5391 static int check_feature_bits(struct btrfs_root *root,
5392                               enum btrfs_feature_set set,
5393                               u64 change_mask, u64 flags, u64 supported_flags,
5394                               u64 safe_set, u64 safe_clear)
5395 {
5396         const char *type = btrfs_feature_set_names[set];
5397         char *names;
5398         u64 disallowed, unsupported;
5399         u64 set_mask = flags & change_mask;
5400         u64 clear_mask = ~flags & change_mask;
5401
5402         unsupported = set_mask & ~supported_flags;
5403         if (unsupported) {
5404                 names = btrfs_printable_features(set, unsupported);
5405                 if (names) {
5406                         btrfs_warn(root->fs_info,
5407                            "this kernel does not support the %s feature bit%s",
5408                            names, strchr(names, ',') ? "s" : "");
5409                         kfree(names);
5410                 } else
5411                         btrfs_warn(root->fs_info,
5412                            "this kernel does not support %s bits 0x%llx",
5413                            type, unsupported);
5414                 return -EOPNOTSUPP;
5415         }
5416
5417         disallowed = set_mask & ~safe_set;
5418         if (disallowed) {
5419                 names = btrfs_printable_features(set, disallowed);
5420                 if (names) {
5421                         btrfs_warn(root->fs_info,
5422                            "can't set the %s feature bit%s while mounted",
5423                            names, strchr(names, ',') ? "s" : "");
5424                         kfree(names);
5425                 } else
5426                         btrfs_warn(root->fs_info,
5427                            "can't set %s bits 0x%llx while mounted",
5428                            type, disallowed);
5429                 return -EPERM;
5430         }
5431
5432         disallowed = clear_mask & ~safe_clear;
5433         if (disallowed) {
5434                 names = btrfs_printable_features(set, disallowed);
5435                 if (names) {
5436                         btrfs_warn(root->fs_info,
5437                            "can't clear the %s feature bit%s while mounted",
5438                            names, strchr(names, ',') ? "s" : "");
5439                         kfree(names);
5440                 } else
5441                         btrfs_warn(root->fs_info,
5442                            "can't clear %s bits 0x%llx while mounted",
5443                            type, disallowed);
5444                 return -EPERM;
5445         }
5446
5447         return 0;
5448 }
5449
5450 #define check_feature(root, change_mask, flags, mask_base)      \
5451 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5452                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5453                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5454                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5455
5456 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5457 {
5458         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5459         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5460         struct btrfs_ioctl_feature_flags flags[2];
5461         struct btrfs_trans_handle *trans;
5462         u64 newflags;
5463         int ret;
5464
5465         if (!capable(CAP_SYS_ADMIN))
5466                 return -EPERM;
5467
5468         if (copy_from_user(flags, arg, sizeof(flags)))
5469                 return -EFAULT;
5470
5471         /* Nothing to do */
5472         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5473             !flags[0].incompat_flags)
5474                 return 0;
5475
5476         ret = check_feature(root, flags[0].compat_flags,
5477                             flags[1].compat_flags, COMPAT);
5478         if (ret)
5479                 return ret;
5480
5481         ret = check_feature(root, flags[0].compat_ro_flags,
5482                             flags[1].compat_ro_flags, COMPAT_RO);
5483         if (ret)
5484                 return ret;
5485
5486         ret = check_feature(root, flags[0].incompat_flags,
5487                             flags[1].incompat_flags, INCOMPAT);
5488         if (ret)
5489                 return ret;
5490
5491         trans = btrfs_start_transaction(root, 0);
5492         if (IS_ERR(trans))
5493                 return PTR_ERR(trans);
5494
5495         spin_lock(&root->fs_info->super_lock);
5496         newflags = btrfs_super_compat_flags(super_block);
5497         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5498         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5499         btrfs_set_super_compat_flags(super_block, newflags);
5500
5501         newflags = btrfs_super_compat_ro_flags(super_block);
5502         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5503         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5504         btrfs_set_super_compat_ro_flags(super_block, newflags);
5505
5506         newflags = btrfs_super_incompat_flags(super_block);
5507         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5508         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5509         btrfs_set_super_incompat_flags(super_block, newflags);
5510         spin_unlock(&root->fs_info->super_lock);
5511
5512         return btrfs_commit_transaction(trans, root);
5513 }
5514
5515 long btrfs_ioctl(struct file *file, unsigned int
5516                 cmd, unsigned long arg)
5517 {
5518         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5519         void __user *argp = (void __user *)arg;
5520
5521         switch (cmd) {
5522         case FS_IOC_GETFLAGS:
5523                 return btrfs_ioctl_getflags(file, argp);
5524         case FS_IOC_SETFLAGS:
5525                 return btrfs_ioctl_setflags(file, argp);
5526         case FS_IOC_GETVERSION:
5527                 return btrfs_ioctl_getversion(file, argp);
5528         case FITRIM:
5529                 return btrfs_ioctl_fitrim(file, argp);
5530         case BTRFS_IOC_SNAP_CREATE:
5531                 return btrfs_ioctl_snap_create(file, argp, 0);
5532         case BTRFS_IOC_SNAP_CREATE_V2:
5533                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5534         case BTRFS_IOC_SUBVOL_CREATE:
5535                 return btrfs_ioctl_snap_create(file, argp, 1);
5536         case BTRFS_IOC_SUBVOL_CREATE_V2:
5537                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5538         case BTRFS_IOC_SNAP_DESTROY:
5539                 return btrfs_ioctl_snap_destroy(file, argp);
5540         case BTRFS_IOC_SUBVOL_GETFLAGS:
5541                 return btrfs_ioctl_subvol_getflags(file, argp);
5542         case BTRFS_IOC_SUBVOL_SETFLAGS:
5543                 return btrfs_ioctl_subvol_setflags(file, argp);
5544         case BTRFS_IOC_DEFAULT_SUBVOL:
5545                 return btrfs_ioctl_default_subvol(file, argp);
5546         case BTRFS_IOC_DEFRAG:
5547                 return btrfs_ioctl_defrag(file, NULL);
5548         case BTRFS_IOC_DEFRAG_RANGE:
5549                 return btrfs_ioctl_defrag(file, argp);
5550         case BTRFS_IOC_RESIZE:
5551                 return btrfs_ioctl_resize(file, argp);
5552         case BTRFS_IOC_ADD_DEV:
5553                 return btrfs_ioctl_add_dev(root, argp);
5554         case BTRFS_IOC_RM_DEV:
5555                 return btrfs_ioctl_rm_dev(file, argp);
5556         case BTRFS_IOC_FS_INFO:
5557                 return btrfs_ioctl_fs_info(root, argp);
5558         case BTRFS_IOC_DEV_INFO:
5559                 return btrfs_ioctl_dev_info(root, argp);
5560         case BTRFS_IOC_BALANCE:
5561                 return btrfs_ioctl_balance(file, NULL);
5562         case BTRFS_IOC_CLONE:
5563                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5564         case BTRFS_IOC_CLONE_RANGE:
5565                 return btrfs_ioctl_clone_range(file, argp);
5566         case BTRFS_IOC_TRANS_START:
5567                 return btrfs_ioctl_trans_start(file);
5568         case BTRFS_IOC_TRANS_END:
5569                 return btrfs_ioctl_trans_end(file);
5570         case BTRFS_IOC_TREE_SEARCH:
5571                 return btrfs_ioctl_tree_search(file, argp);
5572         case BTRFS_IOC_TREE_SEARCH_V2:
5573                 return btrfs_ioctl_tree_search_v2(file, argp);
5574         case BTRFS_IOC_INO_LOOKUP:
5575                 return btrfs_ioctl_ino_lookup(file, argp);
5576         case BTRFS_IOC_INO_PATHS:
5577                 return btrfs_ioctl_ino_to_path(root, argp);
5578         case BTRFS_IOC_LOGICAL_INO:
5579                 return btrfs_ioctl_logical_to_ino(root, argp);
5580         case BTRFS_IOC_SPACE_INFO:
5581                 return btrfs_ioctl_space_info(root, argp);
5582         case BTRFS_IOC_SYNC: {
5583                 int ret;
5584
5585                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5586                 if (ret)
5587                         return ret;
5588                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5589                 /*
5590                  * The transaction thread may want to do more work,
5591                  * namely it pokes the cleaner ktread that will start
5592                  * processing uncleaned subvols.
5593                  */
5594                 wake_up_process(root->fs_info->transaction_kthread);
5595                 return ret;
5596         }
5597         case BTRFS_IOC_START_SYNC:
5598                 return btrfs_ioctl_start_sync(root, argp);
5599         case BTRFS_IOC_WAIT_SYNC:
5600                 return btrfs_ioctl_wait_sync(root, argp);
5601         case BTRFS_IOC_SCRUB:
5602                 return btrfs_ioctl_scrub(file, argp);
5603         case BTRFS_IOC_SCRUB_CANCEL:
5604                 return btrfs_ioctl_scrub_cancel(root, argp);
5605         case BTRFS_IOC_SCRUB_PROGRESS:
5606                 return btrfs_ioctl_scrub_progress(root, argp);
5607         case BTRFS_IOC_BALANCE_V2:
5608                 return btrfs_ioctl_balance(file, argp);
5609         case BTRFS_IOC_BALANCE_CTL:
5610                 return btrfs_ioctl_balance_ctl(root, arg);
5611         case BTRFS_IOC_BALANCE_PROGRESS:
5612                 return btrfs_ioctl_balance_progress(root, argp);
5613         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5614                 return btrfs_ioctl_set_received_subvol(file, argp);
5615 #ifdef CONFIG_64BIT
5616         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5617                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5618 #endif
5619         case BTRFS_IOC_SEND:
5620                 return btrfs_ioctl_send(file, argp);
5621         case BTRFS_IOC_GET_DEV_STATS:
5622                 return btrfs_ioctl_get_dev_stats(root, argp);
5623         case BTRFS_IOC_QUOTA_CTL:
5624                 return btrfs_ioctl_quota_ctl(file, argp);
5625         case BTRFS_IOC_QGROUP_ASSIGN:
5626                 return btrfs_ioctl_qgroup_assign(file, argp);
5627         case BTRFS_IOC_QGROUP_CREATE:
5628                 return btrfs_ioctl_qgroup_create(file, argp);
5629         case BTRFS_IOC_QGROUP_LIMIT:
5630                 return btrfs_ioctl_qgroup_limit(file, argp);
5631         case BTRFS_IOC_QUOTA_RESCAN:
5632                 return btrfs_ioctl_quota_rescan(file, argp);
5633         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5634                 return btrfs_ioctl_quota_rescan_status(file, argp);
5635         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5636                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5637         case BTRFS_IOC_DEV_REPLACE:
5638                 return btrfs_ioctl_dev_replace(root, argp);
5639         case BTRFS_IOC_GET_FSLABEL:
5640                 return btrfs_ioctl_get_fslabel(file, argp);
5641         case BTRFS_IOC_SET_FSLABEL:
5642                 return btrfs_ioctl_set_fslabel(file, argp);
5643         case BTRFS_IOC_FILE_EXTENT_SAME:
5644                 return btrfs_ioctl_file_extent_same(file, argp);
5645         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5646                 return btrfs_ioctl_get_supported_features(file, argp);
5647         case BTRFS_IOC_GET_FEATURES:
5648                 return btrfs_ioctl_get_features(file, argp);
5649         case BTRFS_IOC_SET_FEATURES:
5650                 return btrfs_ioctl_set_features(file, argp);
5651         }
5652
5653         return -ENOTTY;
5654 }