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         /*
 572          * insert the directory item
 573          */
 574         ret = btrfs_set_inode_index(dir, &index);
 575         if (ret) {
 576                 btrfs_abort_transaction(trans, root, ret);
 577                 goto fail;
 578         }
 579
 580         ret = btrfs_insert_dir_item(trans, root,
 581                                     name, namelen, dir, &key,
 582                                     BTRFS_FT_DIR, index);
 583         if (ret) {
 584                 btrfs_abort_transaction(trans, root, ret);
 585                 goto fail;
 586         }
 587
 588         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 589         ret = btrfs_update_inode(trans, root, dir);
 590         BUG_ON(ret);
 591
 592         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 593                                  objectid, root->root_key.objectid,
 594                                  btrfs_ino(dir), index, name, namelen);
 595         BUG_ON(ret);
 596
 597         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 598                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 599                                   objectid);
 600         if (ret)
 601                 btrfs_abort_transaction(trans, root, ret);
 602
 603 fail:
 604         trans->block_rsv = NULL;
 605         trans->bytes_reserved = 0;
 606         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 607
 608         if (async_transid) {
 609                 *async_transid = trans->transid;
 610                 err = btrfs_commit_transaction_async(trans, root, 1);
 611                 if (err)
 612                         err = btrfs_commit_transaction(trans, root);
 613         } else {
 614                 err = btrfs_commit_transaction(trans, root);
 615         }
 616         if (err && !ret)
 617                 ret = err;
 618
 619         if (!ret) {
 620                 inode = btrfs_lookup_dentry(dir, dentry);
 621                 if (IS_ERR(inode))
 622                         return PTR_ERR(inode);
 623                 d_instantiate(dentry, inode);
 624         }
 625         return ret;
 626 }
 627
 628 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 629 {
 630         s64 writers;
 631         DEFINE_WAIT(wait);
 632
 633         do {
 634                 prepare_to_wait(&root->subv_writers->wait, &wait,
 635                                 TASK_UNINTERRUPTIBLE);
 636
 637                 writers = percpu_counter_sum(&root->subv_writers->counter);
 638                 if (writers)
 639                         schedule();
 640
 641                 finish_wait(&root->subv_writers->wait, &wait);
 642         } while (writers);
 643 }
 644
 645 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 646                            struct dentry *dentry, char *name, int namelen,
 647                            u64 *async_transid, bool readonly,
 648                            struct btrfs_qgroup_inherit *inherit)
 649 {
 650         struct inode *inode;
 651         struct btrfs_pending_snapshot *pending_snapshot;
 652         struct btrfs_trans_handle *trans;
 653         int ret;
 654
 655         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 656                 return -EINVAL;
 657
 658         atomic_inc(&root->will_be_snapshoted);
 659         smp_mb__after_atomic();
 660         btrfs_wait_for_no_snapshoting_writes(root);
 661
 662         ret = btrfs_start_delalloc_inodes(root, 0);
 663         if (ret)
 664                 goto out;
 665
 666         btrfs_wait_ordered_extents(root, -1);
 667
 668         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 669         if (!pending_snapshot) {
 670                 ret = -ENOMEM;
 671                 goto out;
 672         }
 673
 674         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 675                              BTRFS_BLOCK_RSV_TEMP);
 676         /*
 677          * 1 - parent dir inode
 678          * 2 - dir entries
 679          * 1 - root item
 680          * 2 - root ref/backref
 681          * 1 - root of snapshot
 682          * 1 - UUID item
 683          */
 684         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 685                                         &pending_snapshot->block_rsv, 8,
 686                                         &pending_snapshot->qgroup_reserved,
 687                                         false);
 688         if (ret)
 689                 goto free;
 690
 691         pending_snapshot->dentry = dentry;
 692         pending_snapshot->root = root;
 693         pending_snapshot->readonly = readonly;
 694         pending_snapshot->dir = dir;
 695         pending_snapshot->inherit = inherit;
 696
 697         trans = btrfs_start_transaction(root, 0);
 698         if (IS_ERR(trans)) {
 699                 ret = PTR_ERR(trans);
 700                 goto fail;
 701         }
 702
 703         spin_lock(&root->fs_info->trans_lock);
 704         list_add(&pending_snapshot->list,
 705                  &trans->transaction->pending_snapshots);
 706         spin_unlock(&root->fs_info->trans_lock);
 707         if (async_transid) {
 708                 *async_transid = trans->transid;
 709                 ret = btrfs_commit_transaction_async(trans,
 710                                      root->fs_info->extent_root, 1);
 711                 if (ret)
 712                         ret = btrfs_commit_transaction(trans, root);
 713         } else {
 714                 ret = btrfs_commit_transaction(trans,
 715                                                root->fs_info->extent_root);
 716         }
 717         if (ret)
 718                 goto fail;
 719
 720         ret = pending_snapshot->error;
 721         if (ret)
 722                 goto fail;
 723
 724         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 725         if (ret)
 726                 goto fail;
 727
 728         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 729         if (IS_ERR(inode)) {
 730                 ret = PTR_ERR(inode);
 731                 goto fail;
 732         }
 733
 734         d_instantiate(dentry, inode);
 735         ret = 0;
 736 fail:
 737         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 738                                          &pending_snapshot->block_rsv,
 739                                          pending_snapshot->qgroup_reserved);
 740 free:
 741         kfree(pending_snapshot);
 742 out:
 743         if (atomic_dec_and_test(&root->will_be_snapshoted))
 744                 wake_up_atomic_t(&root->will_be_snapshoted);
 745         return ret;
 746 }
 747
 748 /*  copy of may_delete in fs/namei.c()
 749  *      Check whether we can remove a link victim from directory dir, check
 750  *  whether the type of victim is right.
 751  *  1. We can't do it if dir is read-only (done in permission())
 752  *  2. We should have write and exec permissions on dir
 753  *  3. We can't remove anything from append-only dir
 754  *  4. We can't do anything with immutable dir (done in permission())
 755  *  5. If the sticky bit on dir is set we should either
 756  *      a. be owner of dir, or
 757  *      b. be owner of victim, or
 758  *      c. have CAP_FOWNER capability
 759  *  6. If the victim is append-only or immutable we can't do antyhing with
 760  *     links pointing to it.
 761  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 762  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 763  *  9. We can't remove a root or mountpoint.
 764  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 765  *     nfs_async_unlink().
 766  */
 767
 768 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 769 {
 770         int error;
 771
 772         if (d_really_is_negative(victim))
 773                 return -ENOENT;
 774
 775         BUG_ON(d_inode(victim->d_parent) != dir);
 776         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 777
 778         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 779         if (error)
 780                 return error;
 781         if (IS_APPEND(dir))
 782                 return -EPERM;
 783         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 784             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 785                 return -EPERM;
 786         if (isdir) {
 787                 if (!d_is_dir(victim))
 788                         return -ENOTDIR;
 789                 if (IS_ROOT(victim))
 790                         return -EBUSY;
 791         } else if (d_is_dir(victim))
 792                 return -EISDIR;
 793         if (IS_DEADDIR(dir))
 794                 return -ENOENT;
 795         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 796                 return -EBUSY;
 797         return 0;
 798 }
 799
 800 /* copy of may_create in fs/namei.c() */
 801 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 802 {
 803         if (d_really_is_positive(child))
 804                 return -EEXIST;
 805         if (IS_DEADDIR(dir))
 806                 return -ENOENT;
 807         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 808 }
 809
 810 /*
 811  * Create a new subvolume below @parent.  This is largely modeled after
 812  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 813  * inside this filesystem so it's quite a bit simpler.
 814  */
 815 static noinline int btrfs_mksubvol(struct path *parent,
 816                                    char *name, int namelen,
 817                                    struct btrfs_root *snap_src,
 818                                    u64 *async_transid, bool readonly,
 819                                    struct btrfs_qgroup_inherit *inherit)
 820 {
 821         struct inode *dir  = d_inode(parent->dentry);
 822         struct dentry *dentry;
 823         int error;
 824
 825         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 826         if (error == -EINTR)
 827                 return error;
 828
 829         dentry = lookup_one_len(name, parent->dentry, namelen);
 830         error = PTR_ERR(dentry);
 831         if (IS_ERR(dentry))
 832                 goto out_unlock;
 833
 834         error = -EEXIST;
 835         if (d_really_is_positive(dentry))
 836                 goto out_dput;
 837
 838         error = btrfs_may_create(dir, dentry);
 839         if (error)
 840                 goto out_dput;
 841
 842         /*
 843          * even if this name doesn't exist, we may get hash collisions.
 844          * check for them now when we can safely fail
 845          */
 846         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 847                                                dir->i_ino, name,
 848                                                namelen);
 849         if (error)
 850                 goto out_dput;
 851
 852         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 853
 854         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 855                 goto out_up_read;
 856
 857         if (snap_src) {
 858                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 859                                         async_transid, readonly, inherit);
 860         } else {
 861                 error = create_subvol(dir, dentry, name, namelen,
 862                                       async_transid, inherit);
 863         }
 864         if (!error)
 865                 fsnotify_mkdir(dir, dentry);
 866 out_up_read:
 867         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 868 out_dput:
 869         dput(dentry);
 870 out_unlock:
 871         mutex_unlock(&dir->i_mutex);
 872         return error;
 873 }
 874
 875 /*
 876  * When we're defragging a range, we don't want to kick it off again
 877  * if it is really just waiting for delalloc to send it down.
 878  * If we find a nice big extent or delalloc range for the bytes in the
 879  * file you want to defrag, we return 0 to let you know to skip this
 880  * part of the file
 881  */
 882 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 883 {
 884         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 885         struct extent_map *em = NULL;
 886         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 887         u64 end;
 888
 889         read_lock(&em_tree->lock);
 890         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 891         read_unlock(&em_tree->lock);
 892
 893         if (em) {
 894                 end = extent_map_end(em);
 895                 free_extent_map(em);
 896                 if (end - offset > thresh)
 897                         return 0;
 898         }
 899         /* if we already have a nice delalloc here, just stop */
 900         thresh /= 2;
 901         end = count_range_bits(io_tree, &offset, offset + thresh,
 902                                thresh, EXTENT_DELALLOC, 1);
 903         if (end >= thresh)
 904                 return 0;
 905         return 1;
 906 }
 907
 908 /*
 909  * helper function to walk through a file and find extents
 910  * newer than a specific transid, and smaller than thresh.
 911  *
 912  * This is used by the defragging code to find new and small
 913  * extents
 914  */
 915 static int find_new_extents(struct btrfs_root *root,
 916                             struct inode *inode, u64 newer_than,
 917                             u64 *off, u32 thresh)
 918 {
 919         struct btrfs_path *path;
 920         struct btrfs_key min_key;
 921         struct extent_buffer *leaf;
 922         struct btrfs_file_extent_item *extent;
 923         int type;
 924         int ret;
 925         u64 ino = btrfs_ino(inode);
 926
 927         path = btrfs_alloc_path();
 928         if (!path)
 929                 return -ENOMEM;
 930
 931         min_key.objectid = ino;
 932         min_key.type = BTRFS_EXTENT_DATA_KEY;
 933         min_key.offset = *off;
 934
 935         while (1) {
 936                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 937                 if (ret != 0)
 938                         goto none;
 939 process_slot:
 940                 if (min_key.objectid != ino)
 941                         goto none;
 942                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 943                         goto none;
 944
 945                 leaf = path->nodes[0];
 946                 extent = btrfs_item_ptr(leaf, path->slots[0],
 947                                         struct btrfs_file_extent_item);
 948
 949                 type = btrfs_file_extent_type(leaf, extent);
 950                 if (type == BTRFS_FILE_EXTENT_REG &&
 951                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 952                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 953                         *off = min_key.offset;
 954                         btrfs_free_path(path);
 955                         return 0;
 956                 }
 957
 958                 path->slots[0]++;
 959                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 960                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 961                         goto process_slot;
 962                 }
 963
 964                 if (min_key.offset == (u64)-1)
 965                         goto none;
 966
 967                 min_key.offset++;
 968                 btrfs_release_path(path);
 969         }
 970 none:
 971         btrfs_free_path(path);
 972         return -ENOENT;
 973 }
 974
 975 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 976 {
 977         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 978         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 979         struct extent_map *em;
 980         u64 len = PAGE_CACHE_SIZE;
 981
 982         /*
 983          * hopefully we have this extent in the tree already, try without
 984          * the full extent lock
 985          */
 986         read_lock(&em_tree->lock);
 987         em = lookup_extent_mapping(em_tree, start, len);
 988         read_unlock(&em_tree->lock);
 989
 990         if (!em) {
 991                 struct extent_state *cached = NULL;
 992                 u64 end = start + len - 1;
 993
 994                 /* get the big lock and read metadata off disk */
 995                 lock_extent_bits(io_tree, start, end, 0, &cached);
 996                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 997                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 998
 999                 if (IS_ERR(em))
1000                         return NULL;
1001         }
1002
1003         return em;
1004 }
1005
1006 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1007 {
1008         struct extent_map *next;
1009         bool ret = true;
1010
1011         /* this is the last extent */
1012         if (em->start + em->len >= i_size_read(inode))
1013                 return false;
1014
1015         next = defrag_lookup_extent(inode, em->start + em->len);
1016         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1017                 ret = false;
1018         else if ((em->block_start + em->block_len == next->block_start) &&
1019                  (em->block_len > 128 * 1024 && next->block_len > 128 * 1024))
1020                 ret = false;
1021
1022         free_extent_map(next);
1023         return ret;
1024 }
1025
1026 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1027                                u64 *last_len, u64 *skip, u64 *defrag_end,
1028                                int compress)
1029 {
1030         struct extent_map *em;
1031         int ret = 1;
1032         bool next_mergeable = true;
1033         bool prev_mergeable = true;
1034
1035         /*
1036          * make sure that once we start defragging an extent, we keep on
1037          * defragging it
1038          */
1039         if (start < *defrag_end)
1040                 return 1;
1041
1042         *skip = 0;
1043
1044         em = defrag_lookup_extent(inode, start);
1045         if (!em)
1046                 return 0;
1047
1048         /* this will cover holes, and inline extents */
1049         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1050                 ret = 0;
1051                 goto out;
1052         }
1053
1054         if (!*defrag_end)
1055                 prev_mergeable = false;
1056
1057         next_mergeable = defrag_check_next_extent(inode, em);
1058         /*
1059          * we hit a real extent, if it is big or the next extent is not a
1060          * real extent, don't bother defragging it
1061          */
1062         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1063             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1064                 ret = 0;
1065 out:
1066         /*
1067          * last_len ends up being a counter of how many bytes we've defragged.
1068          * every time we choose not to defrag an extent, we reset *last_len
1069          * so that the next tiny extent will force a defrag.
1070          *
1071          * The end result of this is that tiny extents before a single big
1072          * extent will force at least part of that big extent to be defragged.
1073          */
1074         if (ret) {
1075                 *defrag_end = extent_map_end(em);
1076         } else {
1077                 *last_len = 0;
1078                 *skip = extent_map_end(em);
1079                 *defrag_end = 0;
1080         }
1081
1082         free_extent_map(em);
1083         return ret;
1084 }
1085
1086 /*
1087  * it doesn't do much good to defrag one or two pages
1088  * at a time.  This pulls in a nice chunk of pages
1089  * to COW and defrag.
1090  *
1091  * It also makes sure the delalloc code has enough
1092  * dirty data to avoid making new small extents as part
1093  * of the defrag
1094  *
1095  * It's a good idea to start RA on this range
1096  * before calling this.
1097  */
1098 static int cluster_pages_for_defrag(struct inode *inode,
1099                                     struct page **pages,
1100                                     unsigned long start_index,
1101                                     unsigned long num_pages)
1102 {
1103         unsigned long file_end;
1104         u64 isize = i_size_read(inode);
1105         u64 page_start;
1106         u64 page_end;
1107         u64 page_cnt;
1108         int ret;
1109         int i;
1110         int i_done;
1111         struct btrfs_ordered_extent *ordered;
1112         struct extent_state *cached_state = NULL;
1113         struct extent_io_tree *tree;
1114         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1115
1116         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1117         if (!isize || start_index > file_end)
1118                 return 0;
1119
1120         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1121
1122         ret = btrfs_delalloc_reserve_space(inode,
1123                         start_index << PAGE_CACHE_SHIFT,
1124                         page_cnt << PAGE_CACHE_SHIFT);
1125         if (ret)
1126                 return ret;
1127         i_done = 0;
1128         tree = &BTRFS_I(inode)->io_tree;
1129
1130         /* step one, lock all the pages */
1131         for (i = 0; i < page_cnt; i++) {
1132                 struct page *page;
1133 again:
1134                 page = find_or_create_page(inode->i_mapping,
1135                                            start_index + i, mask);
1136                 if (!page)
1137                         break;
1138
1139                 page_start = page_offset(page);
1140                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1141                 while (1) {
1142                         lock_extent_bits(tree, page_start, page_end,
1143                                          0, &cached_state);
1144                         ordered = btrfs_lookup_ordered_extent(inode,
1145                                                               page_start);
1146                         unlock_extent_cached(tree, page_start, page_end,
1147                                              &cached_state, GFP_NOFS);
1148                         if (!ordered)
1149                                 break;
1150
1151                         unlock_page(page);
1152                         btrfs_start_ordered_extent(inode, ordered, 1);
1153                         btrfs_put_ordered_extent(ordered);
1154                         lock_page(page);
1155                         /*
1156                          * we unlocked the page above, so we need check if
1157                          * it was released or not.
1158                          */
1159                         if (page->mapping != inode->i_mapping) {
1160                                 unlock_page(page);
1161                                 page_cache_release(page);
1162                                 goto again;
1163                         }
1164                 }
1165
1166                 if (!PageUptodate(page)) {
1167                         btrfs_readpage(NULL, page);
1168                         lock_page(page);
1169                         if (!PageUptodate(page)) {
1170                                 unlock_page(page);
1171                                 page_cache_release(page);
1172                                 ret = -EIO;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (page->mapping != inode->i_mapping) {
1178                         unlock_page(page);
1179                         page_cache_release(page);
1180                         goto again;
1181                 }
1182
1183                 pages[i] = page;
1184                 i_done++;
1185         }
1186         if (!i_done || ret)
1187                 goto out;
1188
1189         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1190                 goto out;
1191
1192         /*
1193          * so now we have a nice long stream of locked
1194          * and up to date pages, lets wait on them
1195          */
1196         for (i = 0; i < i_done; i++)
1197                 wait_on_page_writeback(pages[i]);
1198
1199         page_start = page_offset(pages[0]);
1200         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1201
1202         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1203                          page_start, page_end - 1, 0, &cached_state);
1204         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1205                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1207                           &cached_state, GFP_NOFS);
1208
1209         if (i_done != page_cnt) {
1210                 spin_lock(&BTRFS_I(inode)->lock);
1211                 BTRFS_I(inode)->outstanding_extents++;
1212                 spin_unlock(&BTRFS_I(inode)->lock);
1213                 btrfs_delalloc_release_space(inode,
1214                                 start_index << PAGE_CACHE_SHIFT,
1215                                 (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1216         }
1217
1218
1219         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1220                           &cached_state, GFP_NOFS);
1221
1222         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1223                              page_start, page_end - 1, &cached_state,
1224                              GFP_NOFS);
1225
1226         for (i = 0; i < i_done; i++) {
1227                 clear_page_dirty_for_io(pages[i]);
1228                 ClearPageChecked(pages[i]);
1229                 set_page_extent_mapped(pages[i]);
1230                 set_page_dirty(pages[i]);
1231                 unlock_page(pages[i]);
1232                 page_cache_release(pages[i]);
1233         }
1234         return i_done;
1235 out:
1236         for (i = 0; i < i_done; i++) {
1237                 unlock_page(pages[i]);
1238                 page_cache_release(pages[i]);
1239         }
1240         btrfs_delalloc_release_space(inode,
1241                         start_index << PAGE_CACHE_SHIFT,
1242                         page_cnt << PAGE_CACHE_SHIFT);
1243         return ret;
1244
1245 }
1246
1247 int btrfs_defrag_file(struct inode *inode, struct file *file,
1248                       struct btrfs_ioctl_defrag_range_args *range,
1249                       u64 newer_than, unsigned long max_to_defrag)
1250 {
1251         struct btrfs_root *root = BTRFS_I(inode)->root;
1252         struct file_ra_state *ra = NULL;
1253         unsigned long last_index;
1254         u64 isize = i_size_read(inode);
1255         u64 last_len = 0;
1256         u64 skip = 0;
1257         u64 defrag_end = 0;
1258         u64 newer_off = range->start;
1259         unsigned long i;
1260         unsigned long ra_index = 0;
1261         int ret;
1262         int defrag_count = 0;
1263         int compress_type = BTRFS_COMPRESS_ZLIB;
1264         u32 extent_thresh = range->extent_thresh;
1265         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1266         unsigned long cluster = max_cluster;
1267         u64 new_align = ~((u64)128 * 1024 - 1);
1268         struct page **pages = NULL;
1269
1270         if (isize == 0)
1271                 return 0;
1272
1273         if (range->start >= isize)
1274                 return -EINVAL;
1275
1276         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1277                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1278                         return -EINVAL;
1279                 if (range->compress_type)
1280                         compress_type = range->compress_type;
1281         }
1282
1283         if (extent_thresh == 0)
1284                 extent_thresh = 256 * 1024;
1285
1286         /*
1287          * if we were not given a file, allocate a readahead
1288          * context
1289          */
1290         if (!file) {
1291                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1292                 if (!ra)
1293                         return -ENOMEM;
1294                 file_ra_state_init(ra, inode->i_mapping);
1295         } else {
1296                 ra = &file->f_ra;
1297         }
1298
1299         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1300                         GFP_NOFS);
1301         if (!pages) {
1302                 ret = -ENOMEM;
1303                 goto out_ra;
1304         }
1305
1306         /* find the last page to defrag */
1307         if (range->start + range->len > range->start) {
1308                 last_index = min_t(u64, isize - 1,
1309                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1310         } else {
1311                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1312         }
1313
1314         if (newer_than) {
1315                 ret = find_new_extents(root, inode, newer_than,
1316                                        &newer_off, 64 * 1024);
1317                 if (!ret) {
1318                         range->start = newer_off;
1319                         /*
1320                          * we always align our defrag to help keep
1321                          * the extents in the file evenly spaced
1322                          */
1323                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1324                 } else
1325                         goto out_ra;
1326         } else {
1327                 i = range->start >> PAGE_CACHE_SHIFT;
1328         }
1329         if (!max_to_defrag)
1330                 max_to_defrag = last_index - i + 1;
1331
1332         /*
1333          * make writeback starts from i, so the defrag range can be
1334          * written sequentially.
1335          */
1336         if (i < inode->i_mapping->writeback_index)
1337                 inode->i_mapping->writeback_index = i;
1338
1339         while (i <= last_index && defrag_count < max_to_defrag &&
1340                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1341                 /*
1342                  * make sure we stop running if someone unmounts
1343                  * the FS
1344                  */
1345                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1346                         break;
1347
1348                 if (btrfs_defrag_cancelled(root->fs_info)) {
1349                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1350                         ret = -EAGAIN;
1351                         break;
1352                 }
1353
1354                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1355                                          extent_thresh, &last_len, &skip,
1356                                          &defrag_end, range->flags &
1357                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1358                         unsigned long next;
1359                         /*
1360                          * the should_defrag function tells us how much to skip
1361                          * bump our counter by the suggested amount
1362                          */
1363                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1364                         i = max(i + 1, next);
1365                         continue;
1366                 }
1367
1368                 if (!newer_than) {
1369                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1370                                    PAGE_CACHE_SHIFT) - i;
1371                         cluster = min(cluster, max_cluster);
1372                 } else {
1373                         cluster = max_cluster;
1374                 }
1375
1376                 if (i + cluster > ra_index) {
1377                         ra_index = max(i, ra_index);
1378                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1379                                        cluster);
1380                         ra_index += cluster;
1381                 }
1382
1383                 mutex_lock(&inode->i_mutex);
1384                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1385                         BTRFS_I(inode)->force_compress = compress_type;
1386                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1387                 if (ret < 0) {
1388                         mutex_unlock(&inode->i_mutex);
1389                         goto out_ra;
1390                 }
1391
1392                 defrag_count += ret;
1393                 balance_dirty_pages_ratelimited(inode->i_mapping);
1394                 mutex_unlock(&inode->i_mutex);
1395
1396                 if (newer_than) {
1397                         if (newer_off == (u64)-1)
1398                                 break;
1399
1400                         if (ret > 0)
1401                                 i += ret;
1402
1403                         newer_off = max(newer_off + 1,
1404                                         (u64)i << PAGE_CACHE_SHIFT);
1405
1406                         ret = find_new_extents(root, inode,
1407                                                newer_than, &newer_off,
1408                                                64 * 1024);
1409                         if (!ret) {
1410                                 range->start = newer_off;
1411                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1412                         } else {
1413                                 break;
1414                         }
1415                 } else {
1416                         if (ret > 0) {
1417                                 i += ret;
1418                                 last_len += ret << PAGE_CACHE_SHIFT;
1419                         } else {
1420                                 i++;
1421                                 last_len = 0;
1422                         }
1423                 }
1424         }
1425
1426         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1427                 filemap_flush(inode->i_mapping);
1428                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1429                              &BTRFS_I(inode)->runtime_flags))
1430                         filemap_flush(inode->i_mapping);
1431         }
1432
1433         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1434                 /* the filemap_flush will queue IO into the worker threads, but
1435                  * we have to make sure the IO is actually started and that
1436                  * ordered extents get created before we return
1437                  */
1438                 atomic_inc(&root->fs_info->async_submit_draining);
1439                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1440                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1441                         wait_event(root->fs_info->async_submit_wait,
1442                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1443                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1444                 }
1445                 atomic_dec(&root->fs_info->async_submit_draining);
1446         }
1447
1448         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1449                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1450         }
1451
1452         ret = defrag_count;
1453
1454 out_ra:
1455         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1456                 mutex_lock(&inode->i_mutex);
1457                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1458                 mutex_unlock(&inode->i_mutex);
1459         }
1460         if (!file)
1461                 kfree(ra);
1462         kfree(pages);
1463         return ret;
1464 }
1465
1466 static noinline int btrfs_ioctl_resize(struct file *file,
1467                                         void __user *arg)
1468 {
1469         u64 new_size;
1470         u64 old_size;
1471         u64 devid = 1;
1472         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1473         struct btrfs_ioctl_vol_args *vol_args;
1474         struct btrfs_trans_handle *trans;
1475         struct btrfs_device *device = NULL;
1476         char *sizestr;
1477         char *retptr;
1478         char *devstr = NULL;
1479         int ret = 0;
1480         int mod = 0;
1481
1482         if (!capable(CAP_SYS_ADMIN))
1483                 return -EPERM;
1484
1485         ret = mnt_want_write_file(file);
1486         if (ret)
1487                 return ret;
1488
1489         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1490                         1)) {
1491                 mnt_drop_write_file(file);
1492                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1493         }
1494
1495         mutex_lock(&root->fs_info->volume_mutex);
1496         vol_args = memdup_user(arg, sizeof(*vol_args));
1497         if (IS_ERR(vol_args)) {
1498                 ret = PTR_ERR(vol_args);
1499                 goto out;
1500         }
1501
1502         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1503
1504         sizestr = vol_args->name;
1505         devstr = strchr(sizestr, ':');
1506         if (devstr) {
1507                 sizestr = devstr + 1;
1508                 *devstr = '\0';
1509                 devstr = vol_args->name;
1510                 ret = kstrtoull(devstr, 10, &devid);
1511                 if (ret)
1512                         goto out_free;
1513                 if (!devid) {
1514                         ret = -EINVAL;
1515                         goto out_free;
1516                 }
1517                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1518         }
1519
1520         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1521         if (!device) {
1522                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1523                        devid);
1524                 ret = -ENODEV;
1525                 goto out_free;
1526         }
1527
1528         if (!device->writeable) {
1529                 btrfs_info(root->fs_info,
1530                            "resizer unable to apply on readonly device %llu",
1531                        devid);
1532                 ret = -EPERM;
1533                 goto out_free;
1534         }
1535
1536         if (!strcmp(sizestr, "max"))
1537                 new_size = device->bdev->bd_inode->i_size;
1538         else {
1539                 if (sizestr[0] == '-') {
1540                         mod = -1;
1541                         sizestr++;
1542                 } else if (sizestr[0] == '+') {
1543                         mod = 1;
1544                         sizestr++;
1545                 }
1546                 new_size = memparse(sizestr, &retptr);
1547                 if (*retptr != '\0' || new_size == 0) {
1548                         ret = -EINVAL;
1549                         goto out_free;
1550                 }
1551         }
1552
1553         if (device->is_tgtdev_for_dev_replace) {
1554                 ret = -EPERM;
1555                 goto out_free;
1556         }
1557
1558         old_size = btrfs_device_get_total_bytes(device);
1559
1560         if (mod < 0) {
1561                 if (new_size > old_size) {
1562                         ret = -EINVAL;
1563                         goto out_free;
1564                 }
1565                 new_size = old_size - new_size;
1566         } else if (mod > 0) {
1567                 if (new_size > ULLONG_MAX - old_size) {
1568                         ret = -ERANGE;
1569                         goto out_free;
1570                 }
1571                 new_size = old_size + new_size;
1572         }
1573
1574         if (new_size < 256 * 1024 * 1024) {
1575                 ret = -EINVAL;
1576                 goto out_free;
1577         }
1578         if (new_size > device->bdev->bd_inode->i_size) {
1579                 ret = -EFBIG;
1580                 goto out_free;
1581         }
1582
1583         new_size = div_u64(new_size, root->sectorsize);
1584         new_size *= root->sectorsize;
1585
1586         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1587                       rcu_str_deref(device->name), new_size);
1588
1589         if (new_size > old_size) {
1590                 trans = btrfs_start_transaction(root, 0);
1591                 if (IS_ERR(trans)) {
1592                         ret = PTR_ERR(trans);
1593                         goto out_free;
1594                 }
1595                 ret = btrfs_grow_device(trans, device, new_size);
1596                 btrfs_commit_transaction(trans, root);
1597         } else if (new_size < old_size) {
1598                 ret = btrfs_shrink_device(device, new_size);
1599         } /* equal, nothing need to do */
1600
1601 out_free:
1602         kfree(vol_args);
1603 out:
1604         mutex_unlock(&root->fs_info->volume_mutex);
1605         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1606         mnt_drop_write_file(file);
1607         return ret;
1608 }
1609
1610 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1611                                 char *name, unsigned long fd, int subvol,
1612                                 u64 *transid, bool readonly,
1613                                 struct btrfs_qgroup_inherit *inherit)
1614 {
1615         int namelen;
1616         int ret = 0;
1617
1618         ret = mnt_want_write_file(file);
1619         if (ret)
1620                 goto out;
1621
1622         namelen = strlen(name);
1623         if (strchr(name, '/')) {
1624                 ret = -EINVAL;
1625                 goto out_drop_write;
1626         }
1627
1628         if (name[0] == '.' &&
1629            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1630                 ret = -EEXIST;
1631                 goto out_drop_write;
1632         }
1633
1634         if (subvol) {
1635                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1636                                      NULL, transid, readonly, inherit);
1637         } else {
1638                 struct fd src = fdget(fd);
1639                 struct inode *src_inode;
1640                 if (!src.file) {
1641                         ret = -EINVAL;
1642                         goto out_drop_write;
1643                 }
1644
1645                 src_inode = file_inode(src.file);
1646                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1647                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1648                                    "Snapshot src from another FS");
1649                         ret = -EXDEV;
1650                 } else if (!inode_owner_or_capable(src_inode)) {
1651                         /*
1652                          * Subvolume creation is not restricted, but snapshots
1653                          * are limited to own subvolumes only
1654                          */
1655                         ret = -EPERM;
1656                 } else {
1657                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1658                                              BTRFS_I(src_inode)->root,
1659                                              transid, readonly, inherit);
1660                 }
1661                 fdput(src);
1662         }
1663 out_drop_write:
1664         mnt_drop_write_file(file);
1665 out:
1666         return ret;
1667 }
1668
1669 static noinline int btrfs_ioctl_snap_create(struct file *file,
1670                                             void __user *arg, int subvol)
1671 {
1672         struct btrfs_ioctl_vol_args *vol_args;
1673         int ret;
1674
1675         vol_args = memdup_user(arg, sizeof(*vol_args));
1676         if (IS_ERR(vol_args))
1677                 return PTR_ERR(vol_args);
1678         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1679
1680         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1681                                               vol_args->fd, subvol,
1682                                               NULL, false, NULL);
1683
1684         kfree(vol_args);
1685         return ret;
1686 }
1687
1688 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1689                                                void __user *arg, int subvol)
1690 {
1691         struct btrfs_ioctl_vol_args_v2 *vol_args;
1692         int ret;
1693         u64 transid = 0;
1694         u64 *ptr = NULL;
1695         bool readonly = false;
1696         struct btrfs_qgroup_inherit *inherit = NULL;
1697
1698         vol_args = memdup_user(arg, sizeof(*vol_args));
1699         if (IS_ERR(vol_args))
1700                 return PTR_ERR(vol_args);
1701         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1702
1703         if (vol_args->flags &
1704             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1705               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1706                 ret = -EOPNOTSUPP;
1707                 goto free_args;
1708         }
1709
1710         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1711                 ptr = &transid;
1712         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1713                 readonly = true;
1714         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1715                 if (vol_args->size > PAGE_CACHE_SIZE) {
1716                         ret = -EINVAL;
1717                         goto free_args;
1718                 }
1719                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1720                 if (IS_ERR(inherit)) {
1721                         ret = PTR_ERR(inherit);
1722                         goto free_args;
1723                 }
1724         }
1725
1726         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1727                                               vol_args->fd, subvol, ptr,
1728                                               readonly, inherit);
1729         if (ret)
1730                 goto free_inherit;
1731
1732         if (ptr && copy_to_user(arg +
1733                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1734                                         transid),
1735                                 ptr, sizeof(*ptr)))
1736                 ret = -EFAULT;
1737
1738 free_inherit:
1739         kfree(inherit);
1740 free_args:
1741         kfree(vol_args);
1742         return ret;
1743 }
1744
1745 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1746                                                 void __user *arg)
1747 {
1748         struct inode *inode = file_inode(file);
1749         struct btrfs_root *root = BTRFS_I(inode)->root;
1750         int ret = 0;
1751         u64 flags = 0;
1752
1753         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1754                 return -EINVAL;
1755
1756         down_read(&root->fs_info->subvol_sem);
1757         if (btrfs_root_readonly(root))
1758                 flags |= BTRFS_SUBVOL_RDONLY;
1759         up_read(&root->fs_info->subvol_sem);
1760
1761         if (copy_to_user(arg, &flags, sizeof(flags)))
1762                 ret = -EFAULT;
1763
1764         return ret;
1765 }
1766
1767 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1768                                               void __user *arg)
1769 {
1770         struct inode *inode = file_inode(file);
1771         struct btrfs_root *root = BTRFS_I(inode)->root;
1772         struct btrfs_trans_handle *trans;
1773         u64 root_flags;
1774         u64 flags;
1775         int ret = 0;
1776
1777         if (!inode_owner_or_capable(inode))
1778                 return -EPERM;
1779
1780         ret = mnt_want_write_file(file);
1781         if (ret)
1782                 goto out;
1783
1784         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1785                 ret = -EINVAL;
1786                 goto out_drop_write;
1787         }
1788
1789         if (copy_from_user(&flags, arg, sizeof(flags))) {
1790                 ret = -EFAULT;
1791                 goto out_drop_write;
1792         }
1793
1794         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1795                 ret = -EINVAL;
1796                 goto out_drop_write;
1797         }
1798
1799         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1800                 ret = -EOPNOTSUPP;
1801                 goto out_drop_write;
1802         }
1803
1804         down_write(&root->fs_info->subvol_sem);
1805
1806         /* nothing to do */
1807         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1808                 goto out_drop_sem;
1809
1810         root_flags = btrfs_root_flags(&root->root_item);
1811         if (flags & BTRFS_SUBVOL_RDONLY) {
1812                 btrfs_set_root_flags(&root->root_item,
1813                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1814         } else {
1815                 /*
1816                  * Block RO -> RW transition if this subvolume is involved in
1817                  * send
1818                  */
1819                 spin_lock(&root->root_item_lock);
1820                 if (root->send_in_progress == 0) {
1821                         btrfs_set_root_flags(&root->root_item,
1822                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1823                         spin_unlock(&root->root_item_lock);
1824                 } else {
1825                         spin_unlock(&root->root_item_lock);
1826                         btrfs_warn(root->fs_info,
1827                         "Attempt to set subvolume %llu read-write during send",
1828                                         root->root_key.objectid);
1829                         ret = -EPERM;
1830                         goto out_drop_sem;
1831                 }
1832         }
1833
1834         trans = btrfs_start_transaction(root, 1);
1835         if (IS_ERR(trans)) {
1836                 ret = PTR_ERR(trans);
1837                 goto out_reset;
1838         }
1839
1840         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1841                                 &root->root_key, &root->root_item);
1842
1843         btrfs_commit_transaction(trans, root);
1844 out_reset:
1845         if (ret)
1846                 btrfs_set_root_flags(&root->root_item, root_flags);
1847 out_drop_sem:
1848         up_write(&root->fs_info->subvol_sem);
1849 out_drop_write:
1850         mnt_drop_write_file(file);
1851 out:
1852         return ret;
1853 }
1854
1855 /*
1856  * helper to check if the subvolume references other subvolumes
1857  */
1858 static noinline int may_destroy_subvol(struct btrfs_root *root)
1859 {
1860         struct btrfs_path *path;
1861         struct btrfs_dir_item *di;
1862         struct btrfs_key key;
1863         u64 dir_id;
1864         int ret;
1865
1866         path = btrfs_alloc_path();
1867         if (!path)
1868                 return -ENOMEM;
1869
1870         /* Make sure this root isn't set as the default subvol */
1871         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1872         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1873                                    dir_id, "default", 7, 0);
1874         if (di && !IS_ERR(di)) {
1875                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1876                 if (key.objectid == root->root_key.objectid) {
1877                         ret = -EPERM;
1878                         btrfs_err(root->fs_info, "deleting default subvolume "
1879                                   "%llu is not allowed", key.objectid);
1880                         goto out;
1881                 }
1882                 btrfs_release_path(path);
1883         }
1884
1885         key.objectid = root->root_key.objectid;
1886         key.type = BTRFS_ROOT_REF_KEY;
1887         key.offset = (u64)-1;
1888
1889         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1890                                 &key, path, 0, 0);
1891         if (ret < 0)
1892                 goto out;
1893         BUG_ON(ret == 0);
1894
1895         ret = 0;
1896         if (path->slots[0] > 0) {
1897                 path->slots[0]--;
1898                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1899                 if (key.objectid == root->root_key.objectid &&
1900                     key.type == BTRFS_ROOT_REF_KEY)
1901                         ret = -ENOTEMPTY;
1902         }
1903 out:
1904         btrfs_free_path(path);
1905         return ret;
1906 }
1907
1908 static noinline int key_in_sk(struct btrfs_key *key,
1909                               struct btrfs_ioctl_search_key *sk)
1910 {
1911         struct btrfs_key test;
1912         int ret;
1913
1914         test.objectid = sk->min_objectid;
1915         test.type = sk->min_type;
1916         test.offset = sk->min_offset;
1917
1918         ret = btrfs_comp_cpu_keys(key, &test);
1919         if (ret < 0)
1920                 return 0;
1921
1922         test.objectid = sk->max_objectid;
1923         test.type = sk->max_type;
1924         test.offset = sk->max_offset;
1925
1926         ret = btrfs_comp_cpu_keys(key, &test);
1927         if (ret > 0)
1928                 return 0;
1929         return 1;
1930 }
1931
1932 static noinline int copy_to_sk(struct btrfs_root *root,
1933                                struct btrfs_path *path,
1934                                struct btrfs_key *key,
1935                                struct btrfs_ioctl_search_key *sk,
1936                                size_t *buf_size,
1937                                char __user *ubuf,
1938                                unsigned long *sk_offset,
1939                                int *num_found)
1940 {
1941         u64 found_transid;
1942         struct extent_buffer *leaf;
1943         struct btrfs_ioctl_search_header sh;
1944         struct btrfs_key test;
1945         unsigned long item_off;
1946         unsigned long item_len;
1947         int nritems;
1948         int i;
1949         int slot;
1950         int ret = 0;
1951
1952         leaf = path->nodes[0];
1953         slot = path->slots[0];
1954         nritems = btrfs_header_nritems(leaf);
1955
1956         if (btrfs_header_generation(leaf) > sk->max_transid) {
1957                 i = nritems;
1958                 goto advance_key;
1959         }
1960         found_transid = btrfs_header_generation(leaf);
1961
1962         for (i = slot; i < nritems; i++) {
1963                 item_off = btrfs_item_ptr_offset(leaf, i);
1964                 item_len = btrfs_item_size_nr(leaf, i);
1965
1966                 btrfs_item_key_to_cpu(leaf, key, i);
1967                 if (!key_in_sk(key, sk))
1968                         continue;
1969
1970                 if (sizeof(sh) + item_len > *buf_size) {
1971                         if (*num_found) {
1972                                 ret = 1;
1973                                 goto out;
1974                         }
1975
1976                         /*
1977                          * return one empty item back for v1, which does not
1978                          * handle -EOVERFLOW
1979                          */
1980
1981                         *buf_size = sizeof(sh) + item_len;
1982                         item_len = 0;
1983                         ret = -EOVERFLOW;
1984                 }
1985
1986                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1987                         ret = 1;
1988                         goto out;
1989                 }
1990
1991                 sh.objectid = key->objectid;
1992                 sh.offset = key->offset;
1993                 sh.type = key->type;
1994                 sh.len = item_len;
1995                 sh.transid = found_transid;
1996
1997                 /* copy search result header */
1998                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1999                         ret = -EFAULT;
2000                         goto out;
2001                 }
2002
2003                 *sk_offset += sizeof(sh);
2004
2005                 if (item_len) {
2006                         char __user *up = ubuf + *sk_offset;
2007                         /* copy the item */
2008                         if (read_extent_buffer_to_user(leaf, up,
2009                                                        item_off, item_len)) {
2010                                 ret = -EFAULT;
2011                                 goto out;
2012                         }
2013
2014                         *sk_offset += item_len;
2015                 }
2016                 (*num_found)++;
2017
2018                 if (ret) /* -EOVERFLOW from above */
2019                         goto out;
2020
2021                 if (*num_found >= sk->nr_items) {
2022                         ret = 1;
2023                         goto out;
2024                 }
2025         }
2026 advance_key:
2027         ret = 0;
2028         test.objectid = sk->max_objectid;
2029         test.type = sk->max_type;
2030         test.offset = sk->max_offset;
2031         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2032                 ret = 1;
2033         else if (key->offset < (u64)-1)
2034                 key->offset++;
2035         else if (key->type < (u8)-1) {
2036                 key->offset = 0;
2037                 key->type++;
2038         } else if (key->objectid < (u64)-1) {
2039                 key->offset = 0;
2040                 key->type = 0;
2041                 key->objectid++;
2042         } else
2043                 ret = 1;
2044 out:
2045         /*
2046          *  0: all items from this leaf copied, continue with next
2047          *  1: * more items can be copied, but unused buffer is too small
2048          *     * all items were found
2049          *     Either way, it will stops the loop which iterates to the next
2050          *     leaf
2051          *  -EOVERFLOW: item was to large for buffer
2052          *  -EFAULT: could not copy extent buffer back to userspace
2053          */
2054         return ret;
2055 }
2056
2057 static noinline int search_ioctl(struct inode *inode,
2058                                  struct btrfs_ioctl_search_key *sk,
2059                                  size_t *buf_size,
2060                                  char __user *ubuf)
2061 {
2062         struct btrfs_root *root;
2063         struct btrfs_key key;
2064         struct btrfs_path *path;
2065         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2066         int ret;
2067         int num_found = 0;
2068         unsigned long sk_offset = 0;
2069
2070         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2071                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2072                 return -EOVERFLOW;
2073         }
2074
2075         path = btrfs_alloc_path();
2076         if (!path)
2077                 return -ENOMEM;
2078
2079         if (sk->tree_id == 0) {
2080                 /* search the root of the inode that was passed */
2081                 root = BTRFS_I(inode)->root;
2082         } else {
2083                 key.objectid = sk->tree_id;
2084                 key.type = BTRFS_ROOT_ITEM_KEY;
2085                 key.offset = (u64)-1;
2086                 root = btrfs_read_fs_root_no_name(info, &key);
2087                 if (IS_ERR(root)) {
2088                         btrfs_err(info, "could not find root %llu",
2089                                sk->tree_id);
2090                         btrfs_free_path(path);
2091                         return -ENOENT;
2092                 }
2093         }
2094
2095         key.objectid = sk->min_objectid;
2096         key.type = sk->min_type;
2097         key.offset = sk->min_offset;
2098
2099         while (1) {
2100                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2101                 if (ret != 0) {
2102                         if (ret > 0)
2103                                 ret = 0;
2104                         goto err;
2105                 }
2106                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2107                                  &sk_offset, &num_found);
2108                 btrfs_release_path(path);
2109                 if (ret)
2110                         break;
2111
2112         }
2113         if (ret > 0)
2114                 ret = 0;
2115 err:
2116         sk->nr_items = num_found;
2117         btrfs_free_path(path);
2118         return ret;
2119 }
2120
2121 static noinline int btrfs_ioctl_tree_search(struct file *file,
2122                                            void __user *argp)
2123 {
2124         struct btrfs_ioctl_search_args __user *uargs;
2125         struct btrfs_ioctl_search_key sk;
2126         struct inode *inode;
2127         int ret;
2128         size_t buf_size;
2129
2130         if (!capable(CAP_SYS_ADMIN))
2131                 return -EPERM;
2132
2133         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2134
2135         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2136                 return -EFAULT;
2137
2138         buf_size = sizeof(uargs->buf);
2139
2140         inode = file_inode(file);
2141         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2142
2143         /*
2144          * In the origin implementation an overflow is handled by returning a
2145          * search header with a len of zero, so reset ret.
2146          */
2147         if (ret == -EOVERFLOW)
2148                 ret = 0;
2149
2150         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2151                 ret = -EFAULT;
2152         return ret;
2153 }
2154
2155 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2156                                                void __user *argp)
2157 {
2158         struct btrfs_ioctl_search_args_v2 __user *uarg;
2159         struct btrfs_ioctl_search_args_v2 args;
2160         struct inode *inode;
2161         int ret;
2162         size_t buf_size;
2163         const size_t buf_limit = 16 * 1024 * 1024;
2164
2165         if (!capable(CAP_SYS_ADMIN))
2166                 return -EPERM;
2167
2168         /* copy search header and buffer size */
2169         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2170         if (copy_from_user(&args, uarg, sizeof(args)))
2171                 return -EFAULT;
2172
2173         buf_size = args.buf_size;
2174
2175         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2176                 return -EOVERFLOW;
2177
2178         /* limit result size to 16MB */
2179         if (buf_size > buf_limit)
2180                 buf_size = buf_limit;
2181
2182         inode = file_inode(file);
2183         ret = search_ioctl(inode, &args.key, &buf_size,
2184                            (char *)(&uarg->buf[0]));
2185         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2186                 ret = -EFAULT;
2187         else if (ret == -EOVERFLOW &&
2188                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2189                 ret = -EFAULT;
2190
2191         return ret;
2192 }
2193
2194 /*
2195  * Search INODE_REFs to identify path name of 'dirid' directory
2196  * in a 'tree_id' tree. and sets path name to 'name'.
2197  */
2198 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2199                                 u64 tree_id, u64 dirid, char *name)
2200 {
2201         struct btrfs_root *root;
2202         struct btrfs_key key;
2203         char *ptr;
2204         int ret = -1;
2205         int slot;
2206         int len;
2207         int total_len = 0;
2208         struct btrfs_inode_ref *iref;
2209         struct extent_buffer *l;
2210         struct btrfs_path *path;
2211
2212         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2213                 name[0]='\0';
2214                 return 0;
2215         }
2216
2217         path = btrfs_alloc_path();
2218         if (!path)
2219                 return -ENOMEM;
2220
2221         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2222
2223         key.objectid = tree_id;
2224         key.type = BTRFS_ROOT_ITEM_KEY;
2225         key.offset = (u64)-1;
2226         root = btrfs_read_fs_root_no_name(info, &key);
2227         if (IS_ERR(root)) {
2228                 btrfs_err(info, "could not find root %llu", tree_id);
2229                 ret = -ENOENT;
2230                 goto out;
2231         }
2232
2233         key.objectid = dirid;
2234         key.type = BTRFS_INODE_REF_KEY;
2235         key.offset = (u64)-1;
2236
2237         while (1) {
2238                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2239                 if (ret < 0)
2240                         goto out;
2241                 else if (ret > 0) {
2242                         ret = btrfs_previous_item(root, path, dirid,
2243                                                   BTRFS_INODE_REF_KEY);
2244                         if (ret < 0)
2245                                 goto out;
2246                         else if (ret > 0) {
2247                                 ret = -ENOENT;
2248                                 goto out;
2249                         }
2250                 }
2251
2252                 l = path->nodes[0];
2253                 slot = path->slots[0];
2254                 btrfs_item_key_to_cpu(l, &key, slot);
2255
2256                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2257                 len = btrfs_inode_ref_name_len(l, iref);
2258                 ptr -= len + 1;
2259                 total_len += len + 1;
2260                 if (ptr < name) {
2261                         ret = -ENAMETOOLONG;
2262                         goto out;
2263                 }
2264
2265                 *(ptr + len) = '/';
2266                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2267
2268                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2269                         break;
2270
2271                 btrfs_release_path(path);
2272                 key.objectid = key.offset;
2273                 key.offset = (u64)-1;
2274                 dirid = key.objectid;
2275         }
2276         memmove(name, ptr, total_len);
2277         name[total_len] = '\0';
2278         ret = 0;
2279 out:
2280         btrfs_free_path(path);
2281         return ret;
2282 }
2283
2284 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2285                                            void __user *argp)
2286 {
2287          struct btrfs_ioctl_ino_lookup_args *args;
2288          struct inode *inode;
2289         int ret = 0;
2290
2291         args = memdup_user(argp, sizeof(*args));
2292         if (IS_ERR(args))
2293                 return PTR_ERR(args);
2294
2295         inode = file_inode(file);
2296
2297         /*
2298          * Unprivileged query to obtain the containing subvolume root id. The
2299          * path is reset so it's consistent with btrfs_search_path_in_tree.
2300          */
2301         if (args->treeid == 0)
2302                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2303
2304         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2305                 args->name[0] = 0;
2306                 goto out;
2307         }
2308
2309         if (!capable(CAP_SYS_ADMIN)) {
2310                 ret = -EPERM;
2311                 goto out;
2312         }
2313
2314         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2315                                         args->treeid, args->objectid,
2316                                         args->name);
2317
2318 out:
2319         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2320                 ret = -EFAULT;
2321
2322         kfree(args);
2323         return ret;
2324 }
2325
2326 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2327                                              void __user *arg)
2328 {
2329         struct dentry *parent = file->f_path.dentry;
2330         struct dentry *dentry;
2331         struct inode *dir = d_inode(parent);
2332         struct inode *inode;
2333         struct btrfs_root *root = BTRFS_I(dir)->root;
2334         struct btrfs_root *dest = NULL;
2335         struct btrfs_ioctl_vol_args *vol_args;
2336         struct btrfs_trans_handle *trans;
2337         struct btrfs_block_rsv block_rsv;
2338         u64 root_flags;
2339         u64 qgroup_reserved;
2340         int namelen;
2341         int ret;
2342         int err = 0;
2343
2344         vol_args = memdup_user(arg, sizeof(*vol_args));
2345         if (IS_ERR(vol_args))
2346                 return PTR_ERR(vol_args);
2347
2348         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2349         namelen = strlen(vol_args->name);
2350         if (strchr(vol_args->name, '/') ||
2351             strncmp(vol_args->name, "..", namelen) == 0) {
2352                 err = -EINVAL;
2353                 goto out;
2354         }
2355
2356         err = mnt_want_write_file(file);
2357         if (err)
2358                 goto out;
2359
2360
2361         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2362         if (err == -EINTR)
2363                 goto out_drop_write;
2364         dentry = lookup_one_len(vol_args->name, parent, namelen);
2365         if (IS_ERR(dentry)) {
2366                 err = PTR_ERR(dentry);
2367                 goto out_unlock_dir;
2368         }
2369
2370         if (d_really_is_negative(dentry)) {
2371                 err = -ENOENT;
2372                 goto out_dput;
2373         }
2374
2375         inode = d_inode(dentry);
2376         dest = BTRFS_I(inode)->root;
2377         if (!capable(CAP_SYS_ADMIN)) {
2378                 /*
2379                  * Regular user.  Only allow this with a special mount
2380                  * option, when the user has write+exec access to the
2381                  * subvol root, and when rmdir(2) would have been
2382                  * allowed.
2383                  *
2384                  * Note that this is _not_ check that the subvol is
2385                  * empty or doesn't contain data that we wouldn't
2386                  * otherwise be able to delete.
2387                  *
2388                  * Users who want to delete empty subvols should try
2389                  * rmdir(2).
2390                  */
2391                 err = -EPERM;
2392                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2393                         goto out_dput;
2394
2395                 /*
2396                  * Do not allow deletion if the parent dir is the same
2397                  * as the dir to be deleted.  That means the ioctl
2398                  * must be called on the dentry referencing the root
2399                  * of the subvol, not a random directory contained
2400                  * within it.
2401                  */
2402                 err = -EINVAL;
2403                 if (root == dest)
2404                         goto out_dput;
2405
2406                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2407                 if (err)
2408                         goto out_dput;
2409         }
2410
2411         /* check if subvolume may be deleted by a user */
2412         err = btrfs_may_delete(dir, dentry, 1);
2413         if (err)
2414                 goto out_dput;
2415
2416         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2417                 err = -EINVAL;
2418                 goto out_dput;
2419         }
2420
2421         mutex_lock(&inode->i_mutex);
2422
2423         /*
2424          * Don't allow to delete a subvolume with send in progress. This is
2425          * inside the i_mutex so the error handling that has to drop the bit
2426          * again is not run concurrently.
2427          */
2428         spin_lock(&dest->root_item_lock);
2429         root_flags = btrfs_root_flags(&dest->root_item);
2430         if (dest->send_in_progress == 0) {
2431                 btrfs_set_root_flags(&dest->root_item,
2432                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2433                 spin_unlock(&dest->root_item_lock);
2434         } else {
2435                 spin_unlock(&dest->root_item_lock);
2436                 btrfs_warn(root->fs_info,
2437                         "Attempt to delete subvolume %llu during send",
2438                         dest->root_key.objectid);
2439                 err = -EPERM;
2440                 goto out_unlock_inode;
2441         }
2442
2443         down_write(&root->fs_info->subvol_sem);
2444
2445         err = may_destroy_subvol(dest);
2446         if (err)
2447                 goto out_up_write;
2448
2449         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2450         /*
2451          * One for dir inode, two for dir entries, two for root
2452          * ref/backref.
2453          */
2454         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2455                                                5, &qgroup_reserved, true);
2456         if (err)
2457                 goto out_up_write;
2458
2459         trans = btrfs_start_transaction(root, 0);
2460         if (IS_ERR(trans)) {
2461                 err = PTR_ERR(trans);
2462                 goto out_release;
2463         }
2464         trans->block_rsv = &block_rsv;
2465         trans->bytes_reserved = block_rsv.size;
2466
2467         ret = btrfs_unlink_subvol(trans, root, dir,
2468                                 dest->root_key.objectid,
2469                                 dentry->d_name.name,
2470                                 dentry->d_name.len);
2471         if (ret) {
2472                 err = ret;
2473                 btrfs_abort_transaction(trans, root, ret);
2474                 goto out_end_trans;
2475         }
2476
2477         btrfs_record_root_in_trans(trans, dest);
2478
2479         memset(&dest->root_item.drop_progress, 0,
2480                 sizeof(dest->root_item.drop_progress));
2481         dest->root_item.drop_level = 0;
2482         btrfs_set_root_refs(&dest->root_item, 0);
2483
2484         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2485                 ret = btrfs_insert_orphan_item(trans,
2486                                         root->fs_info->tree_root,
2487                                         dest->root_key.objectid);
2488                 if (ret) {
2489                         btrfs_abort_transaction(trans, root, ret);
2490                         err = ret;
2491                         goto out_end_trans;
2492                 }
2493         }
2494
2495         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2496                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2497                                   dest->root_key.objectid);
2498         if (ret && ret != -ENOENT) {
2499                 btrfs_abort_transaction(trans, root, ret);
2500                 err = ret;
2501                 goto out_end_trans;
2502         }
2503         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2504                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2505                                           dest->root_item.received_uuid,
2506                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2507                                           dest->root_key.objectid);
2508                 if (ret && ret != -ENOENT) {
2509                         btrfs_abort_transaction(trans, root, ret);
2510                         err = ret;
2511                         goto out_end_trans;
2512                 }
2513         }
2514
2515 out_end_trans:
2516         trans->block_rsv = NULL;
2517         trans->bytes_reserved = 0;
2518         ret = btrfs_end_transaction(trans, root);
2519         if (ret && !err)
2520                 err = ret;
2521         inode->i_flags |= S_DEAD;
2522 out_release:
2523         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2524 out_up_write:
2525         up_write(&root->fs_info->subvol_sem);
2526         if (err) {
2527                 spin_lock(&dest->root_item_lock);
2528                 root_flags = btrfs_root_flags(&dest->root_item);
2529                 btrfs_set_root_flags(&dest->root_item,
2530                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2531                 spin_unlock(&dest->root_item_lock);
2532         }
2533 out_unlock_inode:
2534         mutex_unlock(&inode->i_mutex);
2535         if (!err) {
2536                 d_invalidate(dentry);
2537                 btrfs_invalidate_inodes(dest);
2538                 d_delete(dentry);
2539                 ASSERT(dest->send_in_progress == 0);
2540
2541                 /* the last ref */
2542                 if (dest->ino_cache_inode) {
2543                         iput(dest->ino_cache_inode);
2544                         dest->ino_cache_inode = NULL;
2545                 }
2546         }
2547 out_dput:
2548         dput(dentry);
2549 out_unlock_dir:
2550         mutex_unlock(&dir->i_mutex);
2551 out_drop_write:
2552         mnt_drop_write_file(file);
2553 out:
2554         kfree(vol_args);
2555         return err;
2556 }
2557
2558 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2559 {
2560         struct inode *inode = file_inode(file);
2561         struct btrfs_root *root = BTRFS_I(inode)->root;
2562         struct btrfs_ioctl_defrag_range_args *range;
2563         int ret;
2564
2565         ret = mnt_want_write_file(file);
2566         if (ret)
2567                 return ret;
2568
2569         if (btrfs_root_readonly(root)) {
2570                 ret = -EROFS;
2571                 goto out;
2572         }
2573
2574         switch (inode->i_mode & S_IFMT) {
2575         case S_IFDIR:
2576                 if (!capable(CAP_SYS_ADMIN)) {
2577                         ret = -EPERM;
2578                         goto out;
2579                 }
2580                 ret = btrfs_defrag_root(root);
2581                 if (ret)
2582                         goto out;
2583                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2584                 break;
2585         case S_IFREG:
2586                 if (!(file->f_mode & FMODE_WRITE)) {
2587                         ret = -EINVAL;
2588                         goto out;
2589                 }
2590
2591                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2592                 if (!range) {
2593                         ret = -ENOMEM;
2594                         goto out;
2595                 }
2596
2597                 if (argp) {
2598                         if (copy_from_user(range, argp,
2599                                            sizeof(*range))) {
2600                                 ret = -EFAULT;
2601                                 kfree(range);
2602                                 goto out;
2603                         }
2604                         /* compression requires us to start the IO */
2605                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2606                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2607                                 range->extent_thresh = (u32)-1;
2608                         }
2609                 } else {
2610                         /* the rest are all set to zero by kzalloc */
2611                         range->len = (u64)-1;
2612                 }
2613                 ret = btrfs_defrag_file(file_inode(file), file,
2614                                         range, 0, 0);
2615                 if (ret > 0)
2616                         ret = 0;
2617                 kfree(range);
2618                 break;
2619         default:
2620                 ret = -EINVAL;
2621         }
2622 out:
2623         mnt_drop_write_file(file);
2624         return ret;
2625 }
2626
2627 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2628 {
2629         struct btrfs_ioctl_vol_args *vol_args;
2630         int ret;
2631
2632         if (!capable(CAP_SYS_ADMIN))
2633                 return -EPERM;
2634
2635         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2636                         1)) {
2637                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2638         }
2639
2640         mutex_lock(&root->fs_info->volume_mutex);
2641         vol_args = memdup_user(arg, sizeof(*vol_args));
2642         if (IS_ERR(vol_args)) {
2643                 ret = PTR_ERR(vol_args);
2644                 goto out;
2645         }
2646
2647         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2648         ret = btrfs_init_new_device(root, vol_args->name);
2649
2650         if (!ret)
2651                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2652
2653         kfree(vol_args);
2654 out:
2655         mutex_unlock(&root->fs_info->volume_mutex);
2656         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2657         return ret;
2658 }
2659
2660 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2661 {
2662         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2663         struct btrfs_ioctl_vol_args *vol_args;
2664         int ret;
2665
2666         if (!capable(CAP_SYS_ADMIN))
2667                 return -EPERM;
2668
2669         ret = mnt_want_write_file(file);
2670         if (ret)
2671                 return ret;
2672
2673         vol_args = memdup_user(arg, sizeof(*vol_args));
2674         if (IS_ERR(vol_args)) {
2675                 ret = PTR_ERR(vol_args);
2676                 goto err_drop;
2677         }
2678
2679         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2680
2681         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2682                         1)) {
2683                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2684                 goto out;
2685         }
2686
2687         mutex_lock(&root->fs_info->volume_mutex);
2688         ret = btrfs_rm_device(root, vol_args->name);
2689         mutex_unlock(&root->fs_info->volume_mutex);
2690         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2691
2692         if (!ret)
2693                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2694
2695 out:
2696         kfree(vol_args);
2697 err_drop:
2698         mnt_drop_write_file(file);
2699         return ret;
2700 }
2701
2702 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2703 {
2704         struct btrfs_ioctl_fs_info_args *fi_args;
2705         struct btrfs_device *device;
2706         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2707         int ret = 0;
2708
2709         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2710         if (!fi_args)
2711                 return -ENOMEM;
2712
2713         mutex_lock(&fs_devices->device_list_mutex);
2714         fi_args->num_devices = fs_devices->num_devices;
2715         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2716
2717         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2718                 if (device->devid > fi_args->max_id)
2719                         fi_args->max_id = device->devid;
2720         }
2721         mutex_unlock(&fs_devices->device_list_mutex);
2722
2723         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2724         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2725         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2726
2727         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2728                 ret = -EFAULT;
2729
2730         kfree(fi_args);
2731         return ret;
2732 }
2733
2734 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2735 {
2736         struct btrfs_ioctl_dev_info_args *di_args;
2737         struct btrfs_device *dev;
2738         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2739         int ret = 0;
2740         char *s_uuid = NULL;
2741
2742         di_args = memdup_user(arg, sizeof(*di_args));
2743         if (IS_ERR(di_args))
2744                 return PTR_ERR(di_args);
2745
2746         if (!btrfs_is_empty_uuid(di_args->uuid))
2747                 s_uuid = di_args->uuid;
2748
2749         mutex_lock(&fs_devices->device_list_mutex);
2750         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2751
2752         if (!dev) {
2753                 ret = -ENODEV;
2754                 goto out;
2755         }
2756
2757         di_args->devid = dev->devid;
2758         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2759         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2760         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2761         if (dev->name) {
2762                 struct rcu_string *name;
2763
2764                 rcu_read_lock();
2765                 name = rcu_dereference(dev->name);
2766                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2767                 rcu_read_unlock();
2768                 di_args->path[sizeof(di_args->path) - 1] = 0;
2769         } else {
2770                 di_args->path[0] = '\0';
2771         }
2772
2773 out:
2774         mutex_unlock(&fs_devices->device_list_mutex);
2775         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2776                 ret = -EFAULT;
2777
2778         kfree(di_args);
2779         return ret;
2780 }
2781
2782 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2783 {
2784         struct page *page;
2785
2786         page = grab_cache_page(inode->i_mapping, index);
2787         if (!page)
2788                 return ERR_PTR(-ENOMEM);
2789
2790         if (!PageUptodate(page)) {
2791                 int ret;
2792
2793                 ret = btrfs_readpage(NULL, page);
2794                 if (ret)
2795                         return ERR_PTR(ret);
2796                 lock_page(page);
2797                 if (!PageUptodate(page)) {
2798                         unlock_page(page);
2799                         page_cache_release(page);
2800                         return ERR_PTR(-EIO);
2801                 }
2802                 if (page->mapping != inode->i_mapping) {
2803                         unlock_page(page);
2804                         page_cache_release(page);
2805                         return ERR_PTR(-EAGAIN);
2806                 }
2807         }
2808
2809         return page;
2810 }
2811
2812 static int gather_extent_pages(struct inode *inode, struct page **pages,
2813                                int num_pages, u64 off)
2814 {
2815         int i;
2816         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2817
2818         for (i = 0; i < num_pages; i++) {
2819 again:
2820                 pages[i] = extent_same_get_page(inode, index + i);
2821                 if (IS_ERR(pages[i])) {
2822                         int err = PTR_ERR(pages[i]);
2823
2824                         if (err == -EAGAIN)
2825                                 goto again;
2826                         pages[i] = NULL;
2827                         return err;
2828                 }
2829         }
2830         return 0;
2831 }
2832
2833 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2834                              bool retry_range_locking)
2835 {
2836         /*
2837          * Do any pending delalloc/csum calculations on inode, one way or
2838          * another, and lock file content.
2839          * The locking order is:
2840          *
2841          *   1) pages
2842          *   2) range in the inode's io tree
2843          */
2844         while (1) {
2845                 struct btrfs_ordered_extent *ordered;
2846                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2847                 ordered = btrfs_lookup_first_ordered_extent(inode,
2848                                                             off + len - 1);
2849                 if ((!ordered ||
2850                      ordered->file_offset + ordered->len <= off ||
2851                      ordered->file_offset >= off + len) &&
2852                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2853                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2854                         if (ordered)
2855                                 btrfs_put_ordered_extent(ordered);
2856                         break;
2857                 }
2858                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2859                 if (ordered)
2860                         btrfs_put_ordered_extent(ordered);
2861                 if (!retry_range_locking)
2862                         return -EAGAIN;
2863                 btrfs_wait_ordered_range(inode, off, len);
2864         }
2865         return 0;
2866 }
2867
2868 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2869 {
2870         mutex_unlock(&inode1->i_mutex);
2871         mutex_unlock(&inode2->i_mutex);
2872 }
2873
2874 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2875 {
2876         if (inode1 < inode2)
2877                 swap(inode1, inode2);
2878
2879         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2880         mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2881 }
2882
2883 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2884                                       struct inode *inode2, u64 loff2, u64 len)
2885 {
2886         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2887         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2888 }
2889
2890 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2891                                     struct inode *inode2, u64 loff2, u64 len,
2892                                     bool retry_range_locking)
2893 {
2894         int ret;
2895
2896         if (inode1 < inode2) {
2897                 swap(inode1, inode2);
2898                 swap(loff1, loff2);
2899         }
2900         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2901         if (ret)
2902                 return ret;
2903         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2904         if (ret)
2905                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2906                               loff1 + len - 1);
2907         return ret;
2908 }
2909
2910 struct cmp_pages {
2911         int             num_pages;
2912         struct page     **src_pages;
2913         struct page     **dst_pages;
2914 };
2915
2916 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2917 {
2918         int i;
2919         struct page *pg;
2920
2921         for (i = 0; i < cmp->num_pages; i++) {
2922                 pg = cmp->src_pages[i];
2923                 if (pg) {
2924                         unlock_page(pg);
2925                         page_cache_release(pg);
2926                 }
2927                 pg = cmp->dst_pages[i];
2928                 if (pg) {
2929                         unlock_page(pg);
2930                         page_cache_release(pg);
2931                 }
2932         }
2933         kfree(cmp->src_pages);
2934         kfree(cmp->dst_pages);
2935 }
2936
2937 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2938                                   struct inode *dst, u64 dst_loff,
2939                                   u64 len, struct cmp_pages *cmp)
2940 {
2941         int ret;
2942         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2943         struct page **src_pgarr, **dst_pgarr;
2944
2945         /*
2946          * We must gather up all the pages before we initiate our
2947          * extent locking. We use an array for the page pointers. Size
2948          * of the array is bounded by len, which is in turn bounded by
2949          * BTRFS_MAX_DEDUPE_LEN.
2950          */
2951         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2952         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2953         if (!src_pgarr || !dst_pgarr) {
2954                 kfree(src_pgarr);
2955                 kfree(dst_pgarr);
2956                 return -ENOMEM;
2957         }
2958         cmp->num_pages = num_pages;
2959         cmp->src_pages = src_pgarr;
2960         cmp->dst_pages = dst_pgarr;
2961
2962         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2963         if (ret)
2964                 goto out;
2965
2966         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2967
2968 out:
2969         if (ret)
2970                 btrfs_cmp_data_free(cmp);
2971         return 0;
2972 }
2973
2974 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2975                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2976 {
2977         int ret = 0;
2978         int i;
2979         struct page *src_page, *dst_page;
2980         unsigned int cmp_len = PAGE_CACHE_SIZE;
2981         void *addr, *dst_addr;
2982
2983         i = 0;
2984         while (len) {
2985                 if (len < PAGE_CACHE_SIZE)
2986                         cmp_len = len;
2987
2988                 BUG_ON(i >= cmp->num_pages);
2989
2990                 src_page = cmp->src_pages[i];
2991                 dst_page = cmp->dst_pages[i];
2992                 ASSERT(PageLocked(src_page));
2993                 ASSERT(PageLocked(dst_page));
2994
2995                 addr = kmap_atomic(src_page);
2996                 dst_addr = kmap_atomic(dst_page);
2997
2998                 flush_dcache_page(src_page);
2999                 flush_dcache_page(dst_page);
3000
3001                 if (memcmp(addr, dst_addr, cmp_len))
3002                         ret = BTRFS_SAME_DATA_DIFFERS;
3003
3004                 kunmap_atomic(addr);
3005                 kunmap_atomic(dst_addr);
3006
3007                 if (ret)
3008                         break;
3009
3010                 len -= cmp_len;
3011                 i++;
3012         }
3013
3014         return ret;
3015 }
3016
3017 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3018                                      u64 olen)
3019 {
3020         u64 len = *plen;
3021         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3022
3023         if (off + olen > inode->i_size || off + olen < off)
3024                 return -EINVAL;
3025
3026         /* if we extend to eof, continue to block boundary */
3027         if (off + len == inode->i_size)
3028                 *plen = len = ALIGN(inode->i_size, bs) - off;
3029
3030         /* Check that we are block aligned - btrfs_clone() requires this */
3031         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3032                 return -EINVAL;
3033
3034         return 0;
3035 }
3036
3037 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3038                              struct inode *dst, u64 dst_loff)
3039 {
3040         int ret;
3041         u64 len = olen;
3042         struct cmp_pages cmp;
3043         int same_inode = 0;
3044         u64 same_lock_start = 0;
3045         u64 same_lock_len = 0;
3046
3047         if (src == dst)
3048                 same_inode = 1;
3049
3050         if (len == 0)
3051                 return 0;
3052
3053         if (same_inode) {
3054                 mutex_lock(&src->i_mutex);
3055
3056                 ret = extent_same_check_offsets(src, loff, &len, olen);
3057                 if (ret)
3058                         goto out_unlock;
3059
3060                 /*
3061                  * Single inode case wants the same checks, except we
3062                  * don't want our length pushed out past i_size as
3063                  * comparing that data range makes no sense.
3064                  *
3065                  * extent_same_check_offsets() will do this for an
3066                  * unaligned length at i_size, so catch it here and
3067                  * reject the request.
3068                  *
3069                  * This effectively means we require aligned extents
3070                  * for the single-inode case, whereas the other cases
3071                  * allow an unaligned length so long as it ends at
3072                  * i_size.
3073                  */
3074                 if (len != olen) {
3075                         ret = -EINVAL;
3076                         goto out_unlock;
3077                 }
3078
3079                 /* Check for overlapping ranges */
3080                 if (dst_loff + len > loff && dst_loff < loff + len) {
3081                         ret = -EINVAL;
3082                         goto out_unlock;
3083                 }
3084
3085                 same_lock_start = min_t(u64, loff, dst_loff);
3086                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3087         } else {
3088                 btrfs_double_inode_lock(src, dst);
3089
3090                 ret = extent_same_check_offsets(src, loff, &len, olen);
3091                 if (ret)
3092                         goto out_unlock;
3093
3094                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3095                 if (ret)
3096                         goto out_unlock;
3097         }
3098
3099         /* don't make the dst file partly checksummed */
3100         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3101             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3102                 ret = -EINVAL;
3103                 goto out_unlock;
3104         }
3105
3106 again:
3107         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3108         if (ret)
3109                 goto out_unlock;
3110
3111         if (same_inode)
3112                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3113                                         false);
3114         else
3115                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3116                                                false);
3117         /*
3118          * If one of the inodes has dirty pages in the respective range or
3119          * ordered extents, we need to flush dellaloc and wait for all ordered
3120          * extents in the range. We must unlock the pages and the ranges in the
3121          * io trees to avoid deadlocks when flushing delalloc (requires locking
3122          * pages) and when waiting for ordered extents to complete (they require
3123          * range locking).
3124          */
3125         if (ret == -EAGAIN) {
3126                 /*
3127                  * Ranges in the io trees already unlocked. Now unlock all
3128                  * pages before waiting for all IO to complete.
3129                  */
3130                 btrfs_cmp_data_free(&cmp);
3131                 if (same_inode) {
3132                         btrfs_wait_ordered_range(src, same_lock_start,
3133                                                  same_lock_len);
3134                 } else {
3135                         btrfs_wait_ordered_range(src, loff, len);
3136                         btrfs_wait_ordered_range(dst, dst_loff, len);
3137                 }
3138                 goto again;
3139         }
3140         ASSERT(ret == 0);
3141         if (WARN_ON(ret)) {
3142                 /* ranges in the io trees already unlocked */
3143                 btrfs_cmp_data_free(&cmp);
3144                 return ret;
3145         }
3146
3147         /* pass original length for comparison so we stay within i_size */
3148         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3149         if (ret == 0)
3150                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3151
3152         if (same_inode)
3153                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3154                               same_lock_start + same_lock_len - 1);
3155         else
3156                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3157
3158         btrfs_cmp_data_free(&cmp);
3159 out_unlock:
3160         if (same_inode)
3161                 mutex_unlock(&src->i_mutex);
3162         else
3163                 btrfs_double_inode_unlock(src, dst);
3164
3165         return ret;
3166 }
3167
3168 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
3169
3170 static long btrfs_ioctl_file_extent_same(struct file *file,
3171                         struct btrfs_ioctl_same_args __user *argp)
3172 {
3173         struct btrfs_ioctl_same_args *same = NULL;
3174         struct btrfs_ioctl_same_extent_info *info;
3175         struct inode *src = file_inode(file);
3176         u64 off;
3177         u64 len;
3178         int i;
3179         int ret;
3180         unsigned long size;
3181         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3182         bool is_admin = capable(CAP_SYS_ADMIN);
3183         u16 count;
3184
3185         if (!(file->f_mode & FMODE_READ))
3186                 return -EINVAL;
3187
3188         ret = mnt_want_write_file(file);
3189         if (ret)
3190                 return ret;
3191
3192         if (get_user(count, &argp->dest_count)) {
3193                 ret = -EFAULT;
3194                 goto out;
3195         }
3196
3197         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3198
3199         same = memdup_user(argp, size);
3200
3201         if (IS_ERR(same)) {
3202                 ret = PTR_ERR(same);
3203                 same = NULL;
3204                 goto out;
3205         }
3206
3207         off = same->logical_offset;
3208         len = same->length;
3209
3210         /*
3211          * Limit the total length we will dedupe for each operation.
3212          * This is intended to bound the total time spent in this
3213          * ioctl to something sane.
3214          */
3215         if (len > BTRFS_MAX_DEDUPE_LEN)
3216                 len = BTRFS_MAX_DEDUPE_LEN;
3217
3218         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3219                 /*
3220                  * Btrfs does not support blocksize < page_size. As a
3221                  * result, btrfs_cmp_data() won't correctly handle
3222                  * this situation without an update.
3223                  */
3224                 ret = -EINVAL;
3225                 goto out;
3226         }
3227
3228         ret = -EISDIR;
3229         if (S_ISDIR(src->i_mode))
3230                 goto out;
3231
3232         ret = -EACCES;
3233         if (!S_ISREG(src->i_mode))
3234                 goto out;
3235
3236         /* pre-format output fields to sane values */
3237         for (i = 0; i < count; i++) {
3238                 same->info[i].bytes_deduped = 0ULL;
3239                 same->info[i].status = 0;
3240         }
3241
3242         for (i = 0, info = same->info; i < count; i++, info++) {
3243                 struct inode *dst;
3244                 struct fd dst_file = fdget(info->fd);
3245                 if (!dst_file.file) {
3246                         info->status = -EBADF;
3247                         continue;
3248                 }
3249                 dst = file_inode(dst_file.file);
3250
3251                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3252                         info->status = -EINVAL;
3253                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3254                         info->status = -EXDEV;
3255                 } else if (S_ISDIR(dst->i_mode)) {
3256                         info->status = -EISDIR;
3257                 } else if (!S_ISREG(dst->i_mode)) {
3258                         info->status = -EACCES;
3259                 } else {
3260                         info->status = btrfs_extent_same(src, off, len, dst,
3261                                                         info->logical_offset);
3262                         if (info->status == 0)
3263                                 info->bytes_deduped += len;
3264                 }
3265                 fdput(dst_file);
3266         }
3267
3268         ret = copy_to_user(argp, same, size);
3269         if (ret)
3270                 ret = -EFAULT;
3271
3272 out:
3273         mnt_drop_write_file(file);
3274         kfree(same);
3275         return ret;
3276 }
3277
3278 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3279                                      struct inode *inode,
3280                                      u64 endoff,
3281                                      const u64 destoff,
3282                                      const u64 olen,
3283                                      int no_time_update)
3284 {
3285         struct btrfs_root *root = BTRFS_I(inode)->root;
3286         int ret;
3287
3288         inode_inc_iversion(inode);
3289         if (!no_time_update)
3290                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3291         /*
3292          * We round up to the block size at eof when determining which
3293          * extents to clone above, but shouldn't round up the file size.
3294          */
3295         if (endoff > destoff + olen)
3296                 endoff = destoff + olen;
3297         if (endoff > inode->i_size)
3298                 btrfs_i_size_write(inode, endoff);
3299
3300         ret = btrfs_update_inode(trans, root, inode);
3301         if (ret) {
3302                 btrfs_abort_transaction(trans, root, ret);
3303                 btrfs_end_transaction(trans, root);
3304                 goto out;
3305         }
3306         ret = btrfs_end_transaction(trans, root);
3307 out:
3308         return ret;
3309 }
3310
3311 static void clone_update_extent_map(struct inode *inode,
3312                                     const struct btrfs_trans_handle *trans,
3313                                     const struct btrfs_path *path,
3314                                     const u64 hole_offset,
3315                                     const u64 hole_len)
3316 {
3317         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3318         struct extent_map *em;
3319         int ret;
3320
3321         em = alloc_extent_map();
3322         if (!em) {
3323                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3324                         &BTRFS_I(inode)->runtime_flags);
3325                 return;
3326         }
3327
3328         if (path) {
3329                 struct btrfs_file_extent_item *fi;
3330
3331                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3332                                     struct btrfs_file_extent_item);
3333                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3334                 em->generation = -1;
3335                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3336                     BTRFS_FILE_EXTENT_INLINE)
3337                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3338                                 &BTRFS_I(inode)->runtime_flags);
3339         } else {
3340                 em->start = hole_offset;
3341                 em->len = hole_len;
3342                 em->ram_bytes = em->len;
3343                 em->orig_start = hole_offset;
3344                 em->block_start = EXTENT_MAP_HOLE;
3345                 em->block_len = 0;
3346                 em->orig_block_len = 0;
3347                 em->compress_type = BTRFS_COMPRESS_NONE;
3348                 em->generation = trans->transid;
3349         }
3350
3351         while (1) {
3352                 write_lock(&em_tree->lock);
3353                 ret = add_extent_mapping(em_tree, em, 1);
3354                 write_unlock(&em_tree->lock);
3355                 if (ret != -EEXIST) {
3356                         free_extent_map(em);
3357                         break;
3358                 }
3359                 btrfs_drop_extent_cache(inode, em->start,
3360                                         em->start + em->len - 1, 0);
3361         }
3362
3363         if (ret)
3364                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3365                         &BTRFS_I(inode)->runtime_flags);
3366 }
3367
3368 /*
3369  * Make sure we do not end up inserting an inline extent into a file that has
3370  * already other (non-inline) extents. If a file has an inline extent it can
3371  * not have any other extents and the (single) inline extent must start at the
3372  * file offset 0. Failing to respect these rules will lead to file corruption,
3373  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3374  *
3375  * We can have extents that have been already written to disk or we can have
3376  * dirty ranges still in delalloc, in which case the extent maps and items are
3377  * created only when we run delalloc, and the delalloc ranges might fall outside
3378  * the range we are currently locking in the inode's io tree. So we check the
3379  * inode's i_size because of that (i_size updates are done while holding the
3380  * i_mutex, which we are holding here).
3381  * We also check to see if the inode has a size not greater than "datal" but has
3382  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3383  * protected against such concurrent fallocate calls by the i_mutex).
3384  *
3385  * If the file has no extents but a size greater than datal, do not allow the
3386  * copy because we would need turn the inline extent into a non-inline one (even
3387  * with NO_HOLES enabled). If we find our destination inode only has one inline
3388  * extent, just overwrite it with the source inline extent if its size is less
3389  * than the source extent's size, or we could copy the source inline extent's
3390  * data into the destination inode's inline extent if the later is greater then
3391  * the former.
3392  */
3393 static int clone_copy_inline_extent(struct inode *src,
3394                                     struct inode *dst,
3395                                     struct btrfs_trans_handle *trans,
3396                                     struct btrfs_path *path,
3397                                     struct btrfs_key *new_key,
3398                                     const u64 drop_start,
3399                                     const u64 datal,
3400                                     const u64 skip,
3401                                     const u64 size,
3402                                     char *inline_data)
3403 {
3404         struct btrfs_root *root = BTRFS_I(dst)->root;
3405         const u64 aligned_end = ALIGN(new_key->offset + datal,
3406                                       root->sectorsize);
3407         int ret;
3408         struct btrfs_key key;
3409
3410         if (new_key->offset > 0)
3411                 return -EOPNOTSUPP;
3412
3413         key.objectid = btrfs_ino(dst);
3414         key.type = BTRFS_EXTENT_DATA_KEY;
3415         key.offset = 0;
3416         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3417         if (ret < 0) {
3418                 return ret;
3419         } else if (ret > 0) {
3420                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3421                         ret = btrfs_next_leaf(root, path);
3422                         if (ret < 0)
3423                                 return ret;
3424                         else if (ret > 0)
3425                                 goto copy_inline_extent;
3426                 }
3427                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3428                 if (key.objectid == btrfs_ino(dst) &&
3429                     key.type == BTRFS_EXTENT_DATA_KEY) {
3430                         ASSERT(key.offset > 0);
3431                         return -EOPNOTSUPP;
3432                 }
3433         } else if (i_size_read(dst) <= datal) {
3434                 struct btrfs_file_extent_item *ei;
3435                 u64 ext_len;
3436
3437                 /*
3438                  * If the file size is <= datal, make sure there are no other
3439                  * extents following (can happen do to an fallocate call with
3440                  * the flag FALLOC_FL_KEEP_SIZE).
3441                  */
3442                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3443                                     struct btrfs_file_extent_item);
3444                 /*
3445                  * If it's an inline extent, it can not have other extents
3446                  * following it.
3447                  */
3448                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3449                     BTRFS_FILE_EXTENT_INLINE)
3450                         goto copy_inline_extent;
3451
3452                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3453                 if (ext_len > aligned_end)
3454                         return -EOPNOTSUPP;
3455
3456                 ret = btrfs_next_item(root, path);
3457                 if (ret < 0) {
3458                         return ret;
3459                 } else if (ret == 0) {
3460                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3461                                               path->slots[0]);
3462                         if (key.objectid == btrfs_ino(dst) &&
3463                             key.type == BTRFS_EXTENT_DATA_KEY)
3464                                 return -EOPNOTSUPP;
3465                 }
3466         }
3467
3468 copy_inline_extent:
3469         /*
3470          * We have no extent items, or we have an extent at offset 0 which may
3471          * or may not be inlined. All these cases are dealt the same way.
3472          */
3473         if (i_size_read(dst) > datal) {
3474                 /*
3475                  * If the destination inode has an inline extent...
3476                  * This would require copying the data from the source inline
3477                  * extent into the beginning of the destination's inline extent.
3478                  * But this is really complex, both extents can be compressed
3479                  * or just one of them, which would require decompressing and
3480                  * re-compressing data (which could increase the new compressed
3481                  * size, not allowing the compressed data to fit anymore in an
3482                  * inline extent).
3483                  * So just don't support this case for now (it should be rare,
3484                  * we are not really saving space when cloning inline extents).
3485                  */
3486                 return -EOPNOTSUPP;
3487         }
3488
3489         btrfs_release_path(path);
3490         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3491         if (ret)
3492                 return ret;
3493         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3494         if (ret)
3495                 return ret;
3496
3497         if (skip) {
3498                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3499
3500                 memmove(inline_data + start, inline_data + start + skip, datal);
3501         }
3502
3503         write_extent_buffer(path->nodes[0], inline_data,
3504                             btrfs_item_ptr_offset(path->nodes[0],
3505                                                   path->slots[0]),
3506                             size);
3507         inode_add_bytes(dst, datal);
3508
3509         return 0;
3510 }
3511
3512 /**
3513  * btrfs_clone() - clone a range from inode file to another
3514  *
3515  * @src: Inode to clone from
3516  * @inode: Inode to clone to
3517  * @off: Offset within source to start clone from
3518  * @olen: Original length, passed by user, of range to clone
3519  * @olen_aligned: Block-aligned value of olen
3520  * @destoff: Offset within @inode to start clone
3521  * @no_time_update: Whether to update mtime/ctime on the target inode
3522  */
3523 static int btrfs_clone(struct inode *src, struct inode *inode,
3524                        const u64 off, const u64 olen, const u64 olen_aligned,
3525                        const u64 destoff, int no_time_update)
3526 {
3527         struct btrfs_root *root = BTRFS_I(inode)->root;
3528         struct btrfs_path *path = NULL;
3529         struct extent_buffer *leaf;
3530         struct btrfs_trans_handle *trans;
3531         char *buf = NULL;
3532         struct btrfs_key key;
3533         u32 nritems;
3534         int slot;
3535         int ret;
3536         const u64 len = olen_aligned;
3537         u64 last_dest_end = destoff;
3538
3539         ret = -ENOMEM;
3540         buf = vmalloc(root->nodesize);
3541         if (!buf)
3542                 return ret;
3543
3544         path = btrfs_alloc_path();
3545         if (!path) {
3546                 vfree(buf);
3547                 return ret;
3548         }
3549
3550         path->reada = 2;
3551         /* clone data */
3552         key.objectid = btrfs_ino(src);
3553         key.type = BTRFS_EXTENT_DATA_KEY;
3554         key.offset = off;
3555
3556         while (1) {
3557                 u64 next_key_min_offset = key.offset + 1;
3558
3559                 /*
3560                  * note the key will change type as we walk through the
3561                  * tree.
3562                  */
3563                 path->leave_spinning = 1;
3564                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3565                                 0, 0);
3566                 if (ret < 0)
3567                         goto out;
3568                 /*
3569                  * First search, if no extent item that starts at offset off was
3570                  * found but the previous item is an extent item, it's possible
3571                  * it might overlap our target range, therefore process it.
3572                  */
3573                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3574                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3575                                               path->slots[0] - 1);
3576                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3577                                 path->slots[0]--;
3578                 }
3579
3580                 nritems = btrfs_header_nritems(path->nodes[0]);
3581 process_slot:
3582                 if (path->slots[0] >= nritems) {
3583                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3584                         if (ret < 0)
3585                                 goto out;
3586                         if (ret > 0)
3587                                 break;
3588                         nritems = btrfs_header_nritems(path->nodes[0]);
3589                 }
3590                 leaf = path->nodes[0];
3591                 slot = path->slots[0];
3592
3593                 btrfs_item_key_to_cpu(leaf, &key, slot);
3594                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3595                     key.objectid != btrfs_ino(src))
3596                         break;
3597
3598                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3599                         struct btrfs_file_extent_item *extent;
3600                         int type;
3601                         u32 size;
3602                         struct btrfs_key new_key;
3603                         u64 disko = 0, diskl = 0;
3604                         u64 datao = 0, datal = 0;
3605                         u8 comp;
3606                         u64 drop_start;
3607
3608                         extent = btrfs_item_ptr(leaf, slot,
3609                                                 struct btrfs_file_extent_item);
3610                         comp = btrfs_file_extent_compression(leaf, extent);
3611                         type = btrfs_file_extent_type(leaf, extent);
3612                         if (type == BTRFS_FILE_EXTENT_REG ||
3613                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3614                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3615                                                                       extent);
3616                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3617                                                                  extent);
3618                                 datao = btrfs_file_extent_offset(leaf, extent);
3619                                 datal = btrfs_file_extent_num_bytes(leaf,
3620                                                                     extent);
3621                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3622                                 /* take upper bound, may be compressed */
3623                                 datal = btrfs_file_extent_ram_bytes(leaf,
3624                                                                     extent);
3625                         }
3626
3627                         /*
3628                          * The first search might have left us at an extent
3629                          * item that ends before our target range's start, can
3630                          * happen if we have holes and NO_HOLES feature enabled.
3631                          */
3632                         if (key.offset + datal <= off) {
3633                                 path->slots[0]++;
3634                                 goto process_slot;
3635                         } else if (key.offset >= off + len) {
3636                                 break;
3637                         }
3638                         next_key_min_offset = key.offset + datal;
3639                         size = btrfs_item_size_nr(leaf, slot);
3640                         read_extent_buffer(leaf, buf,
3641                                            btrfs_item_ptr_offset(leaf, slot),
3642                                            size);
3643
3644                         btrfs_release_path(path);
3645                         path->leave_spinning = 0;
3646
3647                         memcpy(&new_key, &key, sizeof(new_key));
3648                         new_key.objectid = btrfs_ino(inode);
3649                         if (off <= key.offset)
3650                                 new_key.offset = key.offset + destoff - off;
3651                         else
3652                                 new_key.offset = destoff;
3653
3654                         /*
3655                          * Deal with a hole that doesn't have an extent item
3656                          * that represents it (NO_HOLES feature enabled).
3657                          * This hole is either in the middle of the cloning
3658                          * range or at the beginning (fully overlaps it or
3659                          * partially overlaps it).
3660                          */
3661                         if (new_key.offset != last_dest_end)
3662                                 drop_start = last_dest_end;
3663                         else
3664                                 drop_start = new_key.offset;
3665
3666                         /*
3667                          * 1 - adjusting old extent (we may have to split it)
3668                          * 1 - add new extent
3669                          * 1 - inode update
3670                          */
3671                         trans = btrfs_start_transaction(root, 3);
3672                         if (IS_ERR(trans)) {
3673                                 ret = PTR_ERR(trans);
3674                                 goto out;
3675                         }
3676
3677                         if (type == BTRFS_FILE_EXTENT_REG ||
3678                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3679                                 /*
3680                                  *    a  | --- range to clone ---|  b
3681                                  * | ------------- extent ------------- |
3682                                  */
3683
3684                                 /* subtract range b */
3685                                 if (key.offset + datal > off + len)
3686                                         datal = off + len - key.offset;
3687
3688                                 /* subtract range a */
3689                                 if (off > key.offset) {
3690                                         datao += off - key.offset;
3691                                         datal -= off - key.offset;
3692                                 }
3693
3694                                 ret = btrfs_drop_extents(trans, root, inode,
3695                                                          drop_start,
3696                                                          new_key.offset + datal,
3697                                                          1);
3698                                 if (ret) {
3699                                         if (ret != -EOPNOTSUPP)
3700                                                 btrfs_abort_transaction(trans,
3701                                                                 root, ret);
3702                                         btrfs_end_transaction(trans, root);
3703                                         goto out;
3704                                 }
3705
3706                                 ret = btrfs_insert_empty_item(trans, root, path,
3707                                                               &new_key, size);
3708                                 if (ret) {
3709                                         btrfs_abort_transaction(trans, root,
3710                                                                 ret);
3711                                         btrfs_end_transaction(trans, root);
3712                                         goto out;
3713                                 }
3714
3715                                 leaf = path->nodes[0];
3716                                 slot = path->slots[0];
3717                                 write_extent_buffer(leaf, buf,
3718                                             btrfs_item_ptr_offset(leaf, slot),
3719                                             size);
3720
3721                                 extent = btrfs_item_ptr(leaf, slot,
3722                                                 struct btrfs_file_extent_item);
3723
3724                                 /* disko == 0 means it's a hole */
3725                                 if (!disko)
3726                                         datao = 0;
3727
3728                                 btrfs_set_file_extent_offset(leaf, extent,
3729                                                              datao);
3730                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3731                                                                 datal);
3732
3733                                 if (disko) {
3734                                         inode_add_bytes(inode, datal);
3735                                         ret = btrfs_inc_extent_ref(trans, root,
3736                                                         disko, diskl, 0,
3737                                                         root->root_key.objectid,
3738                                                         btrfs_ino(inode),
3739                                                         new_key.offset - datao);
3740                                         if (ret) {
3741                                                 btrfs_abort_transaction(trans,
3742                                                                         root,
3743                                                                         ret);
3744                                                 btrfs_end_transaction(trans,
3745                                                                       root);
3746                                                 goto out;
3747
3748                                         }
3749                                 }
3750                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3751                                 u64 skip = 0;
3752                                 u64 trim = 0;
3753
3754                                 if (off > key.offset) {
3755                                         skip = off - key.offset;
3756                                         new_key.offset += skip;
3757                                 }
3758
3759                                 if (key.offset + datal > off + len)
3760                                         trim = key.offset + datal - (off + len);
3761
3762                                 if (comp && (skip || trim)) {
3763                                         ret = -EINVAL;
3764                                         btrfs_end_transaction(trans, root);
3765                                         goto out;
3766                                 }
3767                                 size -= skip + trim;
3768                                 datal -= skip + trim;
3769
3770                                 ret = clone_copy_inline_extent(src, inode,
3771                                                                trans, path,
3772                                                                &new_key,
3773                                                                drop_start,
3774                                                                datal,
3775                                                                skip, size, buf);
3776                                 if (ret) {
3777                                         if (ret != -EOPNOTSUPP)
3778                                                 btrfs_abort_transaction(trans,
3779                                                                         root,
3780                                                                         ret);
3781                                         btrfs_end_transaction(trans, root);
3782                                         goto out;
3783                                 }
3784                                 leaf = path->nodes[0];
3785                                 slot = path->slots[0];
3786                         }
3787
3788                         /* If we have an implicit hole (NO_HOLES feature). */
3789                         if (drop_start < new_key.offset)
3790                                 clone_update_extent_map(inode, trans,
3791                                                 NULL, drop_start,
3792                                                 new_key.offset - drop_start);
3793
3794                         clone_update_extent_map(inode, trans, path, 0, 0);
3795
3796                         btrfs_mark_buffer_dirty(leaf);
3797                         btrfs_release_path(path);
3798
3799                         last_dest_end = ALIGN(new_key.offset + datal,
3800                                               root->sectorsize);
3801                         ret = clone_finish_inode_update(trans, inode,
3802                                                         last_dest_end,
3803                                                         destoff, olen,
3804                                                         no_time_update);
3805                         if (ret)
3806                                 goto out;
3807                         if (new_key.offset + datal >= destoff + len)
3808                                 break;
3809                 }
3810                 btrfs_release_path(path);
3811                 key.offset = next_key_min_offset;
3812         }
3813         ret = 0;
3814
3815         if (last_dest_end < destoff + len) {
3816                 /*
3817                  * We have an implicit hole (NO_HOLES feature is enabled) that
3818                  * fully or partially overlaps our cloning range at its end.
3819                  */
3820                 btrfs_release_path(path);
3821
3822                 /*
3823                  * 1 - remove extent(s)
3824                  * 1 - inode update
3825                  */
3826                 trans = btrfs_start_transaction(root, 2);
3827                 if (IS_ERR(trans)) {
3828                         ret = PTR_ERR(trans);
3829                         goto out;
3830                 }
3831                 ret = btrfs_drop_extents(trans, root, inode,
3832                                          last_dest_end, destoff + len, 1);
3833                 if (ret) {
3834                         if (ret != -EOPNOTSUPP)
3835                                 btrfs_abort_transaction(trans, root, ret);
3836                         btrfs_end_transaction(trans, root);
3837                         goto out;
3838                 }
3839                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3840                                         destoff + len - last_dest_end);
3841                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3842                                                 destoff, olen, no_time_update);
3843         }
3844
3845 out:
3846         btrfs_free_path(path);
3847         vfree(buf);
3848         return ret;
3849 }
3850
3851 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3852                                        u64 off, u64 olen, u64 destoff)
3853 {
3854         struct inode *inode = file_inode(file);
3855         struct btrfs_root *root = BTRFS_I(inode)->root;
3856         struct fd src_file;
3857         struct inode *src;
3858         int ret;
3859         u64 len = olen;
3860         u64 bs = root->fs_info->sb->s_blocksize;
3861         int same_inode = 0;
3862
3863         /*
3864          * TODO:
3865          * - split compressed inline extents.  annoying: we need to
3866          *   decompress into destination's address_space (the file offset
3867          *   may change, so source mapping won't do), then recompress (or
3868          *   otherwise reinsert) a subrange.
3869          *
3870          * - split destination inode's inline extents.  The inline extents can
3871          *   be either compressed or non-compressed.
3872          */
3873
3874         /* the destination must be opened for writing */
3875         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3876                 return -EINVAL;
3877
3878         if (btrfs_root_readonly(root))
3879                 return -EROFS;
3880
3881         ret = mnt_want_write_file(file);
3882         if (ret)
3883                 return ret;
3884
3885         src_file = fdget(srcfd);
3886         if (!src_file.file) {
3887                 ret = -EBADF;
3888                 goto out_drop_write;
3889         }
3890
3891         ret = -EXDEV;
3892         if (src_file.file->f_path.mnt != file->f_path.mnt)
3893                 goto out_fput;
3894
3895         src = file_inode(src_file.file);
3896
3897         ret = -EINVAL;
3898         if (src == inode)
3899                 same_inode = 1;
3900
3901         /* the src must be open for reading */
3902         if (!(src_file.file->f_mode & FMODE_READ))
3903                 goto out_fput;
3904
3905         /* don't make the dst file partly checksummed */
3906         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3907             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3908                 goto out_fput;
3909
3910         ret = -EISDIR;
3911         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3912                 goto out_fput;
3913
3914         ret = -EXDEV;
3915         if (src->i_sb != inode->i_sb)
3916                 goto out_fput;
3917
3918         if (!same_inode) {
3919                 btrfs_double_inode_lock(src, inode);
3920         } else {
3921                 mutex_lock(&src->i_mutex);
3922         }
3923
3924         /* determine range to clone */
3925         ret = -EINVAL;
3926         if (off + len > src->i_size || off + len < off)
3927                 goto out_unlock;
3928         if (len == 0)
3929                 olen = len = src->i_size - off;
3930         /* if we extend to eof, continue to block boundary */
3931         if (off + len == src->i_size)
3932                 len = ALIGN(src->i_size, bs) - off;
3933
3934         if (len == 0) {
3935                 ret = 0;
3936                 goto out_unlock;
3937         }
3938
3939         /* verify the end result is block aligned */
3940         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3941             !IS_ALIGNED(destoff, bs))
3942                 goto out_unlock;
3943
3944         /* verify if ranges are overlapped within the same file */
3945         if (same_inode) {
3946                 if (destoff + len > off && destoff < off + len)
3947                         goto out_unlock;
3948         }
3949
3950         if (destoff > inode->i_size) {
3951                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3952                 if (ret)
3953                         goto out_unlock;
3954         }
3955
3956         /*
3957          * Lock the target range too. Right after we replace the file extent
3958          * items in the fs tree (which now point to the cloned data), we might
3959          * have a worker replace them with extent items relative to a write
3960          * operation that was issued before this clone operation (i.e. confront
3961          * with inode.c:btrfs_finish_ordered_io).
3962          */
3963         if (same_inode) {
3964                 u64 lock_start = min_t(u64, off, destoff);
3965                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3966
3967                 ret = lock_extent_range(src, lock_start, lock_len, true);
3968         } else {
3969                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3970                                                true);
3971         }
3972         ASSERT(ret == 0);
3973         if (WARN_ON(ret)) {
3974                 /* ranges in the io trees already unlocked */
3975                 goto out_unlock;
3976         }
3977
3978         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3979
3980         if (same_inode) {
3981                 u64 lock_start = min_t(u64, off, destoff);
3982                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3983
3984                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3985         } else {
3986                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3987         }
3988         /*
3989          * Truncate page cache pages so that future reads will see the cloned
3990          * data immediately and not the previous data.
3991          */
3992         truncate_inode_pages_range(&inode->i_data, destoff,
3993                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3994 out_unlock:
3995         if (!same_inode)
3996                 btrfs_double_inode_unlock(src, inode);
3997         else
3998                 mutex_unlock(&src->i_mutex);
3999 out_fput:
4000         fdput(src_file);
4001 out_drop_write:
4002         mnt_drop_write_file(file);
4003         return ret;
4004 }
4005
4006 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
4007 {
4008         struct btrfs_ioctl_clone_range_args args;
4009
4010         if (copy_from_user(&args, argp, sizeof(args)))
4011                 return -EFAULT;
4012         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
4013                                  args.src_length, args.dest_offset);
4014 }
4015
4016 /*
4017  * there are many ways the trans_start and trans_end ioctls can lead
4018  * to deadlocks.  They should only be used by applications that
4019  * basically own the machine, and have a very in depth understanding
4020  * of all the possible deadlocks and enospc problems.
4021  */
4022 static long btrfs_ioctl_trans_start(struct file *file)
4023 {
4024         struct inode *inode = file_inode(file);
4025         struct btrfs_root *root = BTRFS_I(inode)->root;
4026         struct btrfs_trans_handle *trans;
4027         int ret;
4028
4029         ret = -EPERM;
4030         if (!capable(CAP_SYS_ADMIN))
4031                 goto out;
4032
4033         ret = -EINPROGRESS;
4034         if (file->private_data)
4035                 goto out;
4036
4037         ret = -EROFS;
4038         if (btrfs_root_readonly(root))
4039                 goto out;
4040
4041         ret = mnt_want_write_file(file);
4042         if (ret)
4043                 goto out;
4044
4045         atomic_inc(&root->fs_info->open_ioctl_trans);
4046
4047         ret = -ENOMEM;
4048         trans = btrfs_start_ioctl_transaction(root);
4049         if (IS_ERR(trans))
4050                 goto out_drop;
4051
4052         file->private_data = trans;
4053         return 0;
4054
4055 out_drop:
4056         atomic_dec(&root->fs_info->open_ioctl_trans);
4057         mnt_drop_write_file(file);
4058 out:
4059         return ret;
4060 }
4061
4062 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4063 {
4064         struct inode *inode = file_inode(file);
4065         struct btrfs_root *root = BTRFS_I(inode)->root;
4066         struct btrfs_root *new_root;
4067         struct btrfs_dir_item *di;
4068         struct btrfs_trans_handle *trans;
4069         struct btrfs_path *path;
4070         struct btrfs_key location;
4071         struct btrfs_disk_key disk_key;
4072         u64 objectid = 0;
4073         u64 dir_id;
4074         int ret;
4075
4076         if (!capable(CAP_SYS_ADMIN))
4077                 return -EPERM;
4078
4079         ret = mnt_want_write_file(file);
4080         if (ret)
4081                 return ret;
4082
4083         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4084                 ret = -EFAULT;
4085                 goto out;
4086         }
4087
4088         if (!objectid)
4089                 objectid = BTRFS_FS_TREE_OBJECTID;
4090
4091         location.objectid = objectid;
4092         location.type = BTRFS_ROOT_ITEM_KEY;
4093         location.offset = (u64)-1;
4094
4095         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4096         if (IS_ERR(new_root)) {
4097                 ret = PTR_ERR(new_root);
4098                 goto out;
4099         }
4100
4101         path = btrfs_alloc_path();
4102         if (!path) {
4103                 ret = -ENOMEM;
4104                 goto out;
4105         }
4106         path->leave_spinning = 1;
4107
4108         trans = btrfs_start_transaction(root, 1);
4109         if (IS_ERR(trans)) {
4110                 btrfs_free_path(path);
4111                 ret = PTR_ERR(trans);
4112                 goto out;
4113         }
4114
4115         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4116         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4117                                    dir_id, "default", 7, 1);
4118         if (IS_ERR_OR_NULL(di)) {
4119                 btrfs_free_path(path);
4120                 btrfs_end_transaction(trans, root);
4121                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4122                            "item, this isn't going to work");
4123                 ret = -ENOENT;
4124                 goto out;
4125         }
4126
4127         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4128         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4129         btrfs_mark_buffer_dirty(path->nodes[0]);
4130         btrfs_free_path(path);
4131
4132         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4133         btrfs_end_transaction(trans, root);
4134 out:
4135         mnt_drop_write_file(file);
4136         return ret;
4137 }
4138
4139 void btrfs_get_block_group_info(struct list_head *groups_list,
4140                                 struct btrfs_ioctl_space_info *space)
4141 {
4142         struct btrfs_block_group_cache *block_group;
4143
4144         space->total_bytes = 0;
4145         space->used_bytes = 0;
4146         space->flags = 0;
4147         list_for_each_entry(block_group, groups_list, list) {
4148                 space->flags = block_group->flags;
4149                 space->total_bytes += block_group->key.offset;
4150                 space->used_bytes +=
4151                         btrfs_block_group_used(&block_group->item);
4152         }
4153 }
4154
4155 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4156 {
4157         struct btrfs_ioctl_space_args space_args;
4158         struct btrfs_ioctl_space_info space;
4159         struct btrfs_ioctl_space_info *dest;
4160         struct btrfs_ioctl_space_info *dest_orig;
4161         struct btrfs_ioctl_space_info __user *user_dest;
4162         struct btrfs_space_info *info;
4163         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4164                        BTRFS_BLOCK_GROUP_SYSTEM,
4165                        BTRFS_BLOCK_GROUP_METADATA,
4166                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4167         int num_types = 4;
4168         int alloc_size;
4169         int ret = 0;
4170         u64 slot_count = 0;
4171         int i, c;
4172
4173         if (copy_from_user(&space_args,
4174                            (struct btrfs_ioctl_space_args __user *)arg,
4175                            sizeof(space_args)))
4176                 return -EFAULT;
4177
4178         for (i = 0; i < num_types; i++) {
4179                 struct btrfs_space_info *tmp;
4180
4181                 info = NULL;
4182                 rcu_read_lock();
4183                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4184                                         list) {
4185                         if (tmp->flags == types[i]) {
4186                                 info = tmp;
4187                                 break;
4188                         }
4189                 }
4190                 rcu_read_unlock();
4191
4192                 if (!info)
4193                         continue;
4194
4195                 down_read(&info->groups_sem);
4196                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4197                         if (!list_empty(&info->block_groups[c]))
4198                                 slot_count++;
4199                 }
4200                 up_read(&info->groups_sem);
4201         }
4202
4203         /*
4204          * Global block reserve, exported as a space_info
4205          */
4206         slot_count++;
4207
4208         /* space_slots == 0 means they are asking for a count */
4209         if (space_args.space_slots == 0) {
4210                 space_args.total_spaces = slot_count;
4211                 goto out;
4212         }
4213
4214         slot_count = min_t(u64, space_args.space_slots, slot_count);
4215
4216         alloc_size = sizeof(*dest) * slot_count;
4217
4218         /* we generally have at most 6 or so space infos, one for each raid
4219          * level.  So, a whole page should be more than enough for everyone
4220          */
4221         if (alloc_size > PAGE_CACHE_SIZE)
4222                 return -ENOMEM;
4223
4224         space_args.total_spaces = 0;
4225         dest = kmalloc(alloc_size, GFP_NOFS);
4226         if (!dest)
4227                 return -ENOMEM;
4228         dest_orig = dest;
4229
4230         /* now we have a buffer to copy into */
4231         for (i = 0; i < num_types; i++) {
4232                 struct btrfs_space_info *tmp;
4233
4234                 if (!slot_count)
4235                         break;
4236
4237                 info = NULL;
4238                 rcu_read_lock();
4239                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4240                                         list) {
4241                         if (tmp->flags == types[i]) {
4242                                 info = tmp;
4243                                 break;
4244                         }
4245                 }
4246                 rcu_read_unlock();
4247
4248                 if (!info)
4249                         continue;
4250                 down_read(&info->groups_sem);
4251                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4252                         if (!list_empty(&info->block_groups[c])) {
4253                                 btrfs_get_block_group_info(
4254                                         &info->block_groups[c], &space);
4255                                 memcpy(dest, &space, sizeof(space));
4256                                 dest++;
4257                                 space_args.total_spaces++;
4258                                 slot_count--;
4259                         }
4260                         if (!slot_count)
4261                                 break;
4262                 }
4263                 up_read(&info->groups_sem);
4264         }
4265
4266         /*
4267          * Add global block reserve
4268          */
4269         if (slot_count) {
4270                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4271
4272                 spin_lock(&block_rsv->lock);
4273                 space.total_bytes = block_rsv->size;
4274                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4275                 spin_unlock(&block_rsv->lock);
4276                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4277                 memcpy(dest, &space, sizeof(space));
4278                 space_args.total_spaces++;
4279         }
4280
4281         user_dest = (struct btrfs_ioctl_space_info __user *)
4282                 (arg + sizeof(struct btrfs_ioctl_space_args));
4283
4284         if (copy_to_user(user_dest, dest_orig, alloc_size))
4285                 ret = -EFAULT;
4286
4287         kfree(dest_orig);
4288 out:
4289         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4290                 ret = -EFAULT;
4291
4292         return ret;
4293 }
4294
4295 /*
4296  * there are many ways the trans_start and trans_end ioctls can lead
4297  * to deadlocks.  They should only be used by applications that
4298  * basically own the machine, and have a very in depth understanding
4299  * of all the possible deadlocks and enospc problems.
4300  */
4301 long btrfs_ioctl_trans_end(struct file *file)
4302 {
4303         struct inode *inode = file_inode(file);
4304         struct btrfs_root *root = BTRFS_I(inode)->root;
4305         struct btrfs_trans_handle *trans;
4306
4307         trans = file->private_data;
4308         if (!trans)
4309                 return -EINVAL;
4310         file->private_data = NULL;
4311
4312         btrfs_end_transaction(trans, root);
4313
4314         atomic_dec(&root->fs_info->open_ioctl_trans);
4315
4316         mnt_drop_write_file(file);
4317         return 0;
4318 }
4319
4320 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4321                                             void __user *argp)
4322 {
4323         struct btrfs_trans_handle *trans;
4324         u64 transid;
4325         int ret;
4326
4327         trans = btrfs_attach_transaction_barrier(root);
4328         if (IS_ERR(trans)) {
4329                 if (PTR_ERR(trans) != -ENOENT)
4330                         return PTR_ERR(trans);
4331
4332                 /* No running transaction, don't bother */
4333                 transid = root->fs_info->last_trans_committed;
4334                 goto out;
4335         }
4336         transid = trans->transid;
4337         ret = btrfs_commit_transaction_async(trans, root, 0);
4338         if (ret) {
4339                 btrfs_end_transaction(trans, root);
4340                 return ret;
4341         }
4342 out:
4343         if (argp)
4344                 if (copy_to_user(argp, &transid, sizeof(transid)))
4345                         return -EFAULT;
4346         return 0;
4347 }
4348
4349 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4350                                            void __user *argp)
4351 {
4352         u64 transid;
4353
4354         if (argp) {
4355                 if (copy_from_user(&transid, argp, sizeof(transid)))
4356                         return -EFAULT;
4357         } else {
4358                 transid = 0;  /* current trans */
4359         }
4360         return btrfs_wait_for_commit(root, transid);
4361 }
4362
4363 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4364 {
4365         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4366         struct btrfs_ioctl_scrub_args *sa;
4367         int ret;
4368
4369         if (!capable(CAP_SYS_ADMIN))
4370                 return -EPERM;
4371
4372         sa = memdup_user(arg, sizeof(*sa));
4373         if (IS_ERR(sa))
4374                 return PTR_ERR(sa);
4375
4376         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4377                 ret = mnt_want_write_file(file);
4378                 if (ret)
4379                         goto out;
4380         }
4381
4382         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4383                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4384                               0);
4385
4386         if (copy_to_user(arg, sa, sizeof(*sa)))
4387                 ret = -EFAULT;
4388
4389         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4390                 mnt_drop_write_file(file);
4391 out:
4392         kfree(sa);
4393         return ret;
4394 }
4395
4396 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4397 {
4398         if (!capable(CAP_SYS_ADMIN))
4399                 return -EPERM;
4400
4401         return btrfs_scrub_cancel(root->fs_info);
4402 }
4403
4404 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4405                                        void __user *arg)
4406 {
4407         struct btrfs_ioctl_scrub_args *sa;
4408         int ret;
4409
4410         if (!capable(CAP_SYS_ADMIN))
4411                 return -EPERM;
4412
4413         sa = memdup_user(arg, sizeof(*sa));
4414         if (IS_ERR(sa))
4415                 return PTR_ERR(sa);
4416
4417         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4418
4419         if (copy_to_user(arg, sa, sizeof(*sa)))
4420                 ret = -EFAULT;
4421
4422         kfree(sa);
4423         return ret;
4424 }
4425
4426 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4427                                       void __user *arg)
4428 {
4429         struct btrfs_ioctl_get_dev_stats *sa;
4430         int ret;
4431
4432         sa = memdup_user(arg, sizeof(*sa));
4433         if (IS_ERR(sa))
4434                 return PTR_ERR(sa);
4435
4436         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4437                 kfree(sa);
4438                 return -EPERM;
4439         }
4440
4441         ret = btrfs_get_dev_stats(root, sa);
4442
4443         if (copy_to_user(arg, sa, sizeof(*sa)))
4444                 ret = -EFAULT;
4445
4446         kfree(sa);
4447         return ret;
4448 }
4449
4450 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4451 {
4452         struct btrfs_ioctl_dev_replace_args *p;
4453         int ret;
4454
4455         if (!capable(CAP_SYS_ADMIN))
4456                 return -EPERM;
4457
4458         p = memdup_user(arg, sizeof(*p));
4459         if (IS_ERR(p))
4460                 return PTR_ERR(p);
4461
4462         switch (p->cmd) {
4463         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4464                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4465                         ret = -EROFS;
4466                         goto out;
4467                 }
4468                 if (atomic_xchg(
4469                         &root->fs_info->mutually_exclusive_operation_running,
4470                         1)) {
4471                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4472                 } else {
4473                         ret = btrfs_dev_replace_start(root, p);
4474                         atomic_set(
4475                          &root->fs_info->mutually_exclusive_operation_running,
4476                          0);
4477                 }
4478                 break;
4479         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4480                 btrfs_dev_replace_status(root->fs_info, p);
4481                 ret = 0;
4482                 break;
4483         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4484                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4485                 break;
4486         default:
4487                 ret = -EINVAL;
4488                 break;
4489         }
4490
4491         if (copy_to_user(arg, p, sizeof(*p)))
4492                 ret = -EFAULT;
4493 out:
4494         kfree(p);
4495         return ret;
4496 }
4497
4498 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4499 {
4500         int ret = 0;
4501         int i;
4502         u64 rel_ptr;
4503         int size;
4504         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4505         struct inode_fs_paths *ipath = NULL;
4506         struct btrfs_path *path;
4507
4508         if (!capable(CAP_DAC_READ_SEARCH))
4509                 return -EPERM;
4510
4511         path = btrfs_alloc_path();
4512         if (!path) {
4513                 ret = -ENOMEM;
4514                 goto out;
4515         }
4516
4517         ipa = memdup_user(arg, sizeof(*ipa));
4518         if (IS_ERR(ipa)) {
4519                 ret = PTR_ERR(ipa);
4520                 ipa = NULL;
4521                 goto out;
4522         }
4523
4524         size = min_t(u32, ipa->size, 4096);
4525         ipath = init_ipath(size, root, path);
4526         if (IS_ERR(ipath)) {
4527                 ret = PTR_ERR(ipath);
4528                 ipath = NULL;
4529                 goto out;
4530         }
4531
4532         ret = paths_from_inode(ipa->inum, ipath);
4533         if (ret < 0)
4534                 goto out;
4535
4536         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4537                 rel_ptr = ipath->fspath->val[i] -
4538                           (u64)(unsigned long)ipath->fspath->val;
4539                 ipath->fspath->val[i] = rel_ptr;
4540         }
4541
4542         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4543                            (void *)(unsigned long)ipath->fspath, size);
4544         if (ret) {
4545                 ret = -EFAULT;
4546                 goto out;
4547         }
4548
4549 out:
4550         btrfs_free_path(path);
4551         free_ipath(ipath);
4552         kfree(ipa);
4553
4554         return ret;
4555 }
4556
4557 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4558 {
4559         struct btrfs_data_container *inodes = ctx;
4560         const size_t c = 3 * sizeof(u64);
4561
4562         if (inodes->bytes_left >= c) {
4563                 inodes->bytes_left -= c;
4564                 inodes->val[inodes->elem_cnt] = inum;
4565                 inodes->val[inodes->elem_cnt + 1] = offset;
4566                 inodes->val[inodes->elem_cnt + 2] = root;
4567                 inodes->elem_cnt += 3;
4568         } else {
4569                 inodes->bytes_missing += c - inodes->bytes_left;
4570                 inodes->bytes_left = 0;
4571                 inodes->elem_missed += 3;
4572         }
4573
4574         return 0;
4575 }
4576
4577 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4578                                         void __user *arg)
4579 {
4580         int ret = 0;
4581         int size;
4582         struct btrfs_ioctl_logical_ino_args *loi;
4583         struct btrfs_data_container *inodes = NULL;
4584         struct btrfs_path *path = NULL;
4585
4586         if (!capable(CAP_SYS_ADMIN))
4587                 return -EPERM;
4588
4589         loi = memdup_user(arg, sizeof(*loi));
4590         if (IS_ERR(loi)) {
4591                 ret = PTR_ERR(loi);
4592                 loi = NULL;
4593                 goto out;
4594         }
4595
4596         path = btrfs_alloc_path();
4597         if (!path) {
4598                 ret = -ENOMEM;
4599                 goto out;
4600         }
4601
4602         size = min_t(u32, loi->size, 64 * 1024);
4603         inodes = init_data_container(size);
4604         if (IS_ERR(inodes)) {
4605                 ret = PTR_ERR(inodes);
4606                 inodes = NULL;
4607                 goto out;
4608         }
4609
4610         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4611                                           build_ino_list, inodes);
4612         if (ret == -EINVAL)
4613                 ret = -ENOENT;
4614         if (ret < 0)
4615                 goto out;
4616
4617         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4618                            (void *)(unsigned long)inodes, size);
4619         if (ret)
4620                 ret = -EFAULT;
4621
4622 out:
4623         btrfs_free_path(path);
4624         vfree(inodes);
4625         kfree(loi);
4626
4627         return ret;
4628 }
4629
4630 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4631                                struct btrfs_ioctl_balance_args *bargs)
4632 {
4633         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4634
4635         bargs->flags = bctl->flags;
4636
4637         if (atomic_read(&fs_info->balance_running))
4638                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4639         if (atomic_read(&fs_info->balance_pause_req))
4640                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4641         if (atomic_read(&fs_info->balance_cancel_req))
4642                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4643
4644         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4645         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4646         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4647
4648         if (lock) {
4649                 spin_lock(&fs_info->balance_lock);
4650                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4651                 spin_unlock(&fs_info->balance_lock);
4652         } else {
4653                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4654         }
4655 }
4656
4657 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4658 {
4659         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4660         struct btrfs_fs_info *fs_info = root->fs_info;
4661         struct btrfs_ioctl_balance_args *bargs;
4662         struct btrfs_balance_control *bctl;
4663         bool need_unlock; /* for mut. excl. ops lock */
4664         int ret;
4665
4666         if (!capable(CAP_SYS_ADMIN))
4667                 return -EPERM;
4668
4669         ret = mnt_want_write_file(file);
4670         if (ret)
4671                 return ret;
4672
4673 again:
4674         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4675                 mutex_lock(&fs_info->volume_mutex);
4676                 mutex_lock(&fs_info->balance_mutex);
4677                 need_unlock = true;
4678                 goto locked;
4679         }
4680
4681         /*
4682          * mut. excl. ops lock is locked.  Three possibilites:
4683          *   (1) some other op is running
4684          *   (2) balance is running
4685          *   (3) balance is paused -- special case (think resume)
4686          */
4687         mutex_lock(&fs_info->balance_mutex);
4688         if (fs_info->balance_ctl) {
4689                 /* this is either (2) or (3) */
4690                 if (!atomic_read(&fs_info->balance_running)) {
4691                         mutex_unlock(&fs_info->balance_mutex);
4692                         if (!mutex_trylock(&fs_info->volume_mutex))
4693                                 goto again;
4694                         mutex_lock(&fs_info->balance_mutex);
4695
4696                         if (fs_info->balance_ctl &&
4697                             !atomic_read(&fs_info->balance_running)) {
4698                                 /* this is (3) */
4699                                 need_unlock = false;
4700                                 goto locked;
4701                         }
4702
4703                         mutex_unlock(&fs_info->balance_mutex);
4704                         mutex_unlock(&fs_info->volume_mutex);
4705                         goto again;
4706                 } else {
4707                         /* this is (2) */
4708                         mutex_unlock(&fs_info->balance_mutex);
4709                         ret = -EINPROGRESS;
4710                         goto out;
4711                 }
4712         } else {
4713                 /* this is (1) */
4714                 mutex_unlock(&fs_info->balance_mutex);
4715                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4716                 goto out;
4717         }
4718
4719 locked:
4720         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4721
4722         if (arg) {
4723                 bargs = memdup_user(arg, sizeof(*bargs));
4724                 if (IS_ERR(bargs)) {
4725                         ret = PTR_ERR(bargs);
4726                         goto out_unlock;
4727                 }
4728
4729                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4730                         if (!fs_info->balance_ctl) {
4731                                 ret = -ENOTCONN;
4732                                 goto out_bargs;
4733                         }
4734
4735                         bctl = fs_info->balance_ctl;
4736                         spin_lock(&fs_info->balance_lock);
4737                         bctl->flags |= BTRFS_BALANCE_RESUME;
4738                         spin_unlock(&fs_info->balance_lock);
4739
4740                         goto do_balance;
4741                 }
4742         } else {
4743                 bargs = NULL;
4744         }
4745
4746         if (fs_info->balance_ctl) {
4747                 ret = -EINPROGRESS;
4748                 goto out_bargs;
4749         }
4750
4751         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4752         if (!bctl) {
4753                 ret = -ENOMEM;
4754                 goto out_bargs;
4755         }
4756
4757         bctl->fs_info = fs_info;
4758         if (arg) {
4759                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4760                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4761                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4762
4763                 bctl->flags = bargs->flags;
4764         } else {
4765                 /* balance everything - no filters */
4766                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4767         }
4768
4769         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4770                 ret = -EINVAL;
4771                 goto out_bctl;
4772         }
4773
4774 do_balance:
4775         /*
4776          * Ownership of bctl and mutually_exclusive_operation_running
4777          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4778          * or, if restriper was paused all the way until unmount, in
4779          * free_fs_info.  mutually_exclusive_operation_running is
4780          * cleared in __cancel_balance.
4781          */
4782         need_unlock = false;
4783
4784         ret = btrfs_balance(bctl, bargs);
4785         bctl = NULL;
4786
4787         if (arg) {
4788                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4789                         ret = -EFAULT;
4790         }
4791
4792 out_bctl:
4793         kfree(bctl);
4794 out_bargs:
4795         kfree(bargs);
4796 out_unlock:
4797         mutex_unlock(&fs_info->balance_mutex);
4798         mutex_unlock(&fs_info->volume_mutex);
4799         if (need_unlock)
4800                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4801 out:
4802         mnt_drop_write_file(file);
4803         return ret;
4804 }
4805
4806 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4807 {
4808         if (!capable(CAP_SYS_ADMIN))
4809                 return -EPERM;
4810
4811         switch (cmd) {
4812         case BTRFS_BALANCE_CTL_PAUSE:
4813                 return btrfs_pause_balance(root->fs_info);
4814         case BTRFS_BALANCE_CTL_CANCEL:
4815                 return btrfs_cancel_balance(root->fs_info);
4816         }
4817
4818         return -EINVAL;
4819 }
4820
4821 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4822                                          void __user *arg)
4823 {
4824         struct btrfs_fs_info *fs_info = root->fs_info;
4825         struct btrfs_ioctl_balance_args *bargs;
4826         int ret = 0;
4827
4828         if (!capable(CAP_SYS_ADMIN))
4829                 return -EPERM;
4830
4831         mutex_lock(&fs_info->balance_mutex);
4832         if (!fs_info->balance_ctl) {
4833                 ret = -ENOTCONN;
4834                 goto out;
4835         }
4836
4837         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4838         if (!bargs) {
4839                 ret = -ENOMEM;
4840                 goto out;
4841         }
4842
4843         update_ioctl_balance_args(fs_info, 1, bargs);
4844
4845         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4846                 ret = -EFAULT;
4847
4848         kfree(bargs);
4849 out:
4850         mutex_unlock(&fs_info->balance_mutex);
4851         return ret;
4852 }
4853
4854 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4855 {
4856         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4857         struct btrfs_ioctl_quota_ctl_args *sa;
4858         struct btrfs_trans_handle *trans = NULL;
4859         int ret;
4860         int err;
4861
4862         if (!capable(CAP_SYS_ADMIN))
4863                 return -EPERM;
4864
4865         ret = mnt_want_write_file(file);
4866         if (ret)
4867                 return ret;
4868
4869         sa = memdup_user(arg, sizeof(*sa));
4870         if (IS_ERR(sa)) {
4871                 ret = PTR_ERR(sa);
4872                 goto drop_write;
4873         }
4874
4875         down_write(&root->fs_info->subvol_sem);
4876         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4877         if (IS_ERR(trans)) {
4878                 ret = PTR_ERR(trans);
4879                 goto out;
4880         }
4881
4882         switch (sa->cmd) {
4883         case BTRFS_QUOTA_CTL_ENABLE:
4884                 ret = btrfs_quota_enable(trans, root->fs_info);
4885                 break;
4886         case BTRFS_QUOTA_CTL_DISABLE:
4887                 ret = btrfs_quota_disable(trans, root->fs_info);
4888                 break;
4889         default:
4890                 ret = -EINVAL;
4891                 break;
4892         }
4893
4894         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4895         if (err && !ret)
4896                 ret = err;
4897 out:
4898         kfree(sa);
4899         up_write(&root->fs_info->subvol_sem);
4900 drop_write:
4901         mnt_drop_write_file(file);
4902         return ret;
4903 }
4904
4905 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4906 {
4907         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4908         struct btrfs_ioctl_qgroup_assign_args *sa;
4909         struct btrfs_trans_handle *trans;
4910         int ret;
4911         int err;
4912
4913         if (!capable(CAP_SYS_ADMIN))
4914                 return -EPERM;
4915
4916         ret = mnt_want_write_file(file);
4917         if (ret)
4918                 return ret;
4919
4920         sa = memdup_user(arg, sizeof(*sa));
4921         if (IS_ERR(sa)) {
4922                 ret = PTR_ERR(sa);
4923                 goto drop_write;
4924         }
4925
4926         trans = btrfs_join_transaction(root);
4927         if (IS_ERR(trans)) {
4928                 ret = PTR_ERR(trans);
4929                 goto out;
4930         }
4931
4932         /* FIXME: check if the IDs really exist */
4933         if (sa->assign) {
4934                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4935                                                 sa->src, sa->dst);
4936         } else {
4937                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4938                                                 sa->src, sa->dst);
4939         }
4940
4941         /* update qgroup status and info */
4942         err = btrfs_run_qgroups(trans, root->fs_info);
4943         if (err < 0)
4944                 btrfs_std_error(root->fs_info, ret,
4945                             "failed to update qgroup status and info\n");
4946         err = btrfs_end_transaction(trans, root);
4947         if (err && !ret)
4948                 ret = err;
4949
4950 out:
4951         kfree(sa);
4952 drop_write:
4953         mnt_drop_write_file(file);
4954         return ret;
4955 }
4956
4957 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4958 {
4959         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4960         struct btrfs_ioctl_qgroup_create_args *sa;
4961         struct btrfs_trans_handle *trans;
4962         int ret;
4963         int err;
4964
4965         if (!capable(CAP_SYS_ADMIN))
4966                 return -EPERM;
4967
4968         ret = mnt_want_write_file(file);
4969         if (ret)
4970                 return ret;
4971
4972         sa = memdup_user(arg, sizeof(*sa));
4973         if (IS_ERR(sa)) {
4974                 ret = PTR_ERR(sa);
4975                 goto drop_write;
4976         }
4977
4978         if (!sa->qgroupid) {
4979                 ret = -EINVAL;
4980                 goto out;
4981         }
4982
4983         trans = btrfs_join_transaction(root);
4984         if (IS_ERR(trans)) {
4985                 ret = PTR_ERR(trans);
4986                 goto out;
4987         }
4988
4989         /* FIXME: check if the IDs really exist */
4990         if (sa->create) {
4991                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4992         } else {
4993                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4994         }
4995
4996         err = btrfs_end_transaction(trans, root);
4997         if (err && !ret)
4998                 ret = err;
4999
5000 out:
5001         kfree(sa);
5002 drop_write:
5003         mnt_drop_write_file(file);
5004         return ret;
5005 }
5006
5007 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5008 {
5009         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5010         struct btrfs_ioctl_qgroup_limit_args *sa;
5011         struct btrfs_trans_handle *trans;
5012         int ret;
5013         int err;
5014         u64 qgroupid;
5015
5016         if (!capable(CAP_SYS_ADMIN))
5017                 return -EPERM;
5018
5019         ret = mnt_want_write_file(file);
5020         if (ret)
5021                 return ret;
5022
5023         sa = memdup_user(arg, sizeof(*sa));
5024         if (IS_ERR(sa)) {
5025                 ret = PTR_ERR(sa);
5026                 goto drop_write;
5027         }
5028
5029         trans = btrfs_join_transaction(root);
5030         if (IS_ERR(trans)) {
5031                 ret = PTR_ERR(trans);
5032                 goto out;
5033         }
5034
5035         qgroupid = sa->qgroupid;
5036         if (!qgroupid) {
5037                 /* take the current subvol as qgroup */
5038                 qgroupid = root->root_key.objectid;
5039         }
5040
5041         /* FIXME: check if the IDs really exist */
5042         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
5043
5044         err = btrfs_end_transaction(trans, root);
5045         if (err && !ret)
5046                 ret = err;
5047
5048 out:
5049         kfree(sa);
5050 drop_write:
5051         mnt_drop_write_file(file);
5052         return ret;
5053 }
5054
5055 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5056 {
5057         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5058         struct btrfs_ioctl_quota_rescan_args *qsa;
5059         int ret;
5060
5061         if (!capable(CAP_SYS_ADMIN))
5062                 return -EPERM;
5063
5064         ret = mnt_want_write_file(file);
5065         if (ret)
5066                 return ret;
5067
5068         qsa = memdup_user(arg, sizeof(*qsa));
5069         if (IS_ERR(qsa)) {
5070                 ret = PTR_ERR(qsa);
5071                 goto drop_write;
5072         }
5073
5074         if (qsa->flags) {
5075                 ret = -EINVAL;
5076                 goto out;
5077         }
5078
5079         ret = btrfs_qgroup_rescan(root->fs_info);
5080
5081 out:
5082         kfree(qsa);
5083 drop_write:
5084         mnt_drop_write_file(file);
5085         return ret;
5086 }
5087
5088 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5089 {
5090         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5091         struct btrfs_ioctl_quota_rescan_args *qsa;
5092         int ret = 0;
5093
5094         if (!capable(CAP_SYS_ADMIN))
5095                 return -EPERM;
5096
5097         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
5098         if (!qsa)
5099                 return -ENOMEM;
5100
5101         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5102                 qsa->flags = 1;
5103                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5104         }
5105
5106         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5107                 ret = -EFAULT;
5108
5109         kfree(qsa);
5110         return ret;
5111 }
5112
5113 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5114 {
5115         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5116
5117         if (!capable(CAP_SYS_ADMIN))
5118                 return -EPERM;
5119
5120         return btrfs_qgroup_wait_for_completion(root->fs_info);
5121 }
5122
5123 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5124                                             struct btrfs_ioctl_received_subvol_args *sa)
5125 {
5126         struct inode *inode = file_inode(file);
5127         struct btrfs_root *root = BTRFS_I(inode)->root;
5128         struct btrfs_root_item *root_item = &root->root_item;
5129         struct btrfs_trans_handle *trans;
5130         struct timespec ct = CURRENT_TIME;
5131         int ret = 0;
5132         int received_uuid_changed;
5133
5134         if (!inode_owner_or_capable(inode))
5135                 return -EPERM;
5136
5137         ret = mnt_want_write_file(file);
5138         if (ret < 0)
5139                 return ret;
5140
5141         down_write(&root->fs_info->subvol_sem);
5142
5143         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5144                 ret = -EINVAL;
5145                 goto out;
5146         }
5147
5148         if (btrfs_root_readonly(root)) {
5149                 ret = -EROFS;
5150                 goto out;
5151         }
5152
5153         /*
5154          * 1 - root item
5155          * 2 - uuid items (received uuid + subvol uuid)
5156          */
5157         trans = btrfs_start_transaction(root, 3);
5158         if (IS_ERR(trans)) {
5159                 ret = PTR_ERR(trans);
5160                 trans = NULL;
5161                 goto out;
5162         }
5163
5164         sa->rtransid = trans->transid;
5165         sa->rtime.sec = ct.tv_sec;
5166         sa->rtime.nsec = ct.tv_nsec;
5167
5168         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5169                                        BTRFS_UUID_SIZE);
5170         if (received_uuid_changed &&
5171             !btrfs_is_empty_uuid(root_item->received_uuid))
5172                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5173                                     root_item->received_uuid,
5174                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5175                                     root->root_key.objectid);
5176         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5177         btrfs_set_root_stransid(root_item, sa->stransid);
5178         btrfs_set_root_rtransid(root_item, sa->rtransid);
5179         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5180         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5181         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5182         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5183
5184         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5185                                 &root->root_key, &root->root_item);
5186         if (ret < 0) {
5187                 btrfs_end_transaction(trans, root);
5188                 goto out;
5189         }
5190         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5191                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5192                                           sa->uuid,
5193                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5194                                           root->root_key.objectid);
5195                 if (ret < 0 && ret != -EEXIST) {
5196                         btrfs_abort_transaction(trans, root, ret);
5197                         goto out;
5198                 }
5199         }
5200         ret = btrfs_commit_transaction(trans, root);
5201         if (ret < 0) {
5202                 btrfs_abort_transaction(trans, root, ret);
5203                 goto out;
5204         }
5205
5206 out:
5207         up_write(&root->fs_info->subvol_sem);
5208         mnt_drop_write_file(file);
5209         return ret;
5210 }
5211
5212 #ifdef CONFIG_64BIT
5213 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5214                                                 void __user *arg)
5215 {
5216         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5217         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5218         int ret = 0;
5219
5220         args32 = memdup_user(arg, sizeof(*args32));
5221         if (IS_ERR(args32)) {
5222                 ret = PTR_ERR(args32);
5223                 args32 = NULL;
5224                 goto out;
5225         }
5226
5227         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
5228         if (!args64) {
5229                 ret = -ENOMEM;
5230                 goto out;
5231         }
5232
5233         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5234         args64->stransid = args32->stransid;
5235         args64->rtransid = args32->rtransid;
5236         args64->stime.sec = args32->stime.sec;
5237         args64->stime.nsec = args32->stime.nsec;
5238         args64->rtime.sec = args32->rtime.sec;
5239         args64->rtime.nsec = args32->rtime.nsec;
5240         args64->flags = args32->flags;
5241
5242         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5243         if (ret)
5244                 goto out;
5245
5246         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5247         args32->stransid = args64->stransid;
5248         args32->rtransid = args64->rtransid;
5249         args32->stime.sec = args64->stime.sec;
5250         args32->stime.nsec = args64->stime.nsec;
5251         args32->rtime.sec = args64->rtime.sec;
5252         args32->rtime.nsec = args64->rtime.nsec;
5253         args32->flags = args64->flags;
5254
5255         ret = copy_to_user(arg, args32, sizeof(*args32));
5256         if (ret)
5257                 ret = -EFAULT;
5258
5259 out:
5260         kfree(args32);
5261         kfree(args64);
5262         return ret;
5263 }
5264 #endif
5265
5266 static long btrfs_ioctl_set_received_subvol(struct file *file,
5267                                             void __user *arg)
5268 {
5269         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5270         int ret = 0;
5271
5272         sa = memdup_user(arg, sizeof(*sa));
5273         if (IS_ERR(sa)) {
5274                 ret = PTR_ERR(sa);
5275                 sa = NULL;
5276                 goto out;
5277         }
5278
5279         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5280
5281         if (ret)
5282                 goto out;
5283
5284         ret = copy_to_user(arg, sa, sizeof(*sa));
5285         if (ret)
5286                 ret = -EFAULT;
5287
5288 out:
5289         kfree(sa);
5290         return ret;
5291 }
5292
5293 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5294 {
5295         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5296         size_t len;
5297         int ret;
5298         char label[BTRFS_LABEL_SIZE];
5299
5300         spin_lock(&root->fs_info->super_lock);
5301         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5302         spin_unlock(&root->fs_info->super_lock);
5303
5304         len = strnlen(label, BTRFS_LABEL_SIZE);
5305
5306         if (len == BTRFS_LABEL_SIZE) {
5307                 btrfs_warn(root->fs_info,
5308                         "label is too long, return the first %zu bytes", --len);
5309         }
5310
5311         ret = copy_to_user(arg, label, len);
5312
5313         return ret ? -EFAULT : 0;
5314 }
5315
5316 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5317 {
5318         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5319         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5320         struct btrfs_trans_handle *trans;
5321         char label[BTRFS_LABEL_SIZE];
5322         int ret;
5323
5324         if (!capable(CAP_SYS_ADMIN))
5325                 return -EPERM;
5326
5327         if (copy_from_user(label, arg, sizeof(label)))
5328                 return -EFAULT;
5329
5330         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5331                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5332                        BTRFS_LABEL_SIZE - 1);
5333                 return -EINVAL;
5334         }
5335
5336         ret = mnt_want_write_file(file);
5337         if (ret)
5338                 return ret;
5339
5340         trans = btrfs_start_transaction(root, 0);
5341         if (IS_ERR(trans)) {
5342                 ret = PTR_ERR(trans);
5343                 goto out_unlock;
5344         }
5345
5346         spin_lock(&root->fs_info->super_lock);
5347         strcpy(super_block->label, label);
5348         spin_unlock(&root->fs_info->super_lock);
5349         ret = btrfs_commit_transaction(trans, root);
5350
5351 out_unlock:
5352         mnt_drop_write_file(file);
5353         return ret;
5354 }
5355
5356 #define INIT_FEATURE_FLAGS(suffix) \
5357         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5358           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5359           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5360
5361 static int btrfs_ioctl_get_supported_features(struct file *file,
5362                                               void __user *arg)
5363 {
5364         static struct btrfs_ioctl_feature_flags features[3] = {
5365                 INIT_FEATURE_FLAGS(SUPP),
5366                 INIT_FEATURE_FLAGS(SAFE_SET),
5367                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5368         };
5369
5370         if (copy_to_user(arg, &features, sizeof(features)))
5371                 return -EFAULT;
5372
5373         return 0;
5374 }
5375
5376 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5377 {
5378         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5379         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5380         struct btrfs_ioctl_feature_flags features;
5381
5382         features.compat_flags = btrfs_super_compat_flags(super_block);
5383         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5384         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5385
5386         if (copy_to_user(arg, &features, sizeof(features)))
5387                 return -EFAULT;
5388
5389         return 0;
5390 }
5391
5392 static int check_feature_bits(struct btrfs_root *root,
5393                               enum btrfs_feature_set set,
5394                               u64 change_mask, u64 flags, u64 supported_flags,
5395                               u64 safe_set, u64 safe_clear)
5396 {
5397         const char *type = btrfs_feature_set_names[set];
5398         char *names;
5399         u64 disallowed, unsupported;
5400         u64 set_mask = flags & change_mask;
5401         u64 clear_mask = ~flags & change_mask;
5402
5403         unsupported = set_mask & ~supported_flags;
5404         if (unsupported) {
5405                 names = btrfs_printable_features(set, unsupported);
5406                 if (names) {
5407                         btrfs_warn(root->fs_info,
5408                            "this kernel does not support the %s feature bit%s",
5409                            names, strchr(names, ',') ? "s" : "");
5410                         kfree(names);
5411                 } else
5412                         btrfs_warn(root->fs_info,
5413                            "this kernel does not support %s bits 0x%llx",
5414                            type, unsupported);
5415                 return -EOPNOTSUPP;
5416         }
5417
5418         disallowed = set_mask & ~safe_set;
5419         if (disallowed) {
5420                 names = btrfs_printable_features(set, disallowed);
5421                 if (names) {
5422                         btrfs_warn(root->fs_info,
5423                            "can't set the %s feature bit%s while mounted",
5424                            names, strchr(names, ',') ? "s" : "");
5425                         kfree(names);
5426                 } else
5427                         btrfs_warn(root->fs_info,
5428                            "can't set %s bits 0x%llx while mounted",
5429                            type, disallowed);
5430                 return -EPERM;
5431         }
5432
5433         disallowed = clear_mask & ~safe_clear;
5434         if (disallowed) {
5435                 names = btrfs_printable_features(set, disallowed);
5436                 if (names) {
5437                         btrfs_warn(root->fs_info,
5438                            "can't clear the %s feature bit%s while mounted",
5439                            names, strchr(names, ',') ? "s" : "");
5440                         kfree(names);
5441                 } else
5442                         btrfs_warn(root->fs_info,
5443                            "can't clear %s bits 0x%llx while mounted",
5444                            type, disallowed);
5445                 return -EPERM;
5446         }
5447
5448         return 0;
5449 }
5450
5451 #define check_feature(root, change_mask, flags, mask_base)      \
5452 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5453                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5454                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5455                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5456
5457 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5458 {
5459         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5460         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5461         struct btrfs_ioctl_feature_flags flags[2];
5462         struct btrfs_trans_handle *trans;
5463         u64 newflags;
5464         int ret;
5465
5466         if (!capable(CAP_SYS_ADMIN))
5467                 return -EPERM;
5468
5469         if (copy_from_user(flags, arg, sizeof(flags)))
5470                 return -EFAULT;
5471
5472         /* Nothing to do */
5473         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5474             !flags[0].incompat_flags)
5475                 return 0;
5476
5477         ret = check_feature(root, flags[0].compat_flags,
5478                             flags[1].compat_flags, COMPAT);
5479         if (ret)
5480                 return ret;
5481
5482         ret = check_feature(root, flags[0].compat_ro_flags,
5483                             flags[1].compat_ro_flags, COMPAT_RO);
5484         if (ret)
5485                 return ret;
5486
5487         ret = check_feature(root, flags[0].incompat_flags,
5488                             flags[1].incompat_flags, INCOMPAT);
5489         if (ret)
5490                 return ret;
5491
5492         trans = btrfs_start_transaction(root, 0);
5493         if (IS_ERR(trans))
5494                 return PTR_ERR(trans);
5495
5496         spin_lock(&root->fs_info->super_lock);
5497         newflags = btrfs_super_compat_flags(super_block);
5498         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5499         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5500         btrfs_set_super_compat_flags(super_block, newflags);
5501
5502         newflags = btrfs_super_compat_ro_flags(super_block);
5503         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5504         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5505         btrfs_set_super_compat_ro_flags(super_block, newflags);
5506
5507         newflags = btrfs_super_incompat_flags(super_block);
5508         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5509         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5510         btrfs_set_super_incompat_flags(super_block, newflags);
5511         spin_unlock(&root->fs_info->super_lock);
5512
5513         return btrfs_commit_transaction(trans, root);
5514 }
5515
5516 long btrfs_ioctl(struct file *file, unsigned int
5517                 cmd, unsigned long arg)
5518 {
5519         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5520         void __user *argp = (void __user *)arg;
5521
5522         switch (cmd) {
5523         case FS_IOC_GETFLAGS:
5524                 return btrfs_ioctl_getflags(file, argp);
5525         case FS_IOC_SETFLAGS:
5526                 return btrfs_ioctl_setflags(file, argp);
5527         case FS_IOC_GETVERSION:
5528                 return btrfs_ioctl_getversion(file, argp);
5529         case FITRIM:
5530                 return btrfs_ioctl_fitrim(file, argp);
5531         case BTRFS_IOC_SNAP_CREATE:
5532                 return btrfs_ioctl_snap_create(file, argp, 0);
5533         case BTRFS_IOC_SNAP_CREATE_V2:
5534                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5535         case BTRFS_IOC_SUBVOL_CREATE:
5536                 return btrfs_ioctl_snap_create(file, argp, 1);
5537         case BTRFS_IOC_SUBVOL_CREATE_V2:
5538                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5539         case BTRFS_IOC_SNAP_DESTROY:
5540                 return btrfs_ioctl_snap_destroy(file, argp);
5541         case BTRFS_IOC_SUBVOL_GETFLAGS:
5542                 return btrfs_ioctl_subvol_getflags(file, argp);
5543         case BTRFS_IOC_SUBVOL_SETFLAGS:
5544                 return btrfs_ioctl_subvol_setflags(file, argp);
5545         case BTRFS_IOC_DEFAULT_SUBVOL:
5546                 return btrfs_ioctl_default_subvol(file, argp);
5547         case BTRFS_IOC_DEFRAG:
5548                 return btrfs_ioctl_defrag(file, NULL);
5549         case BTRFS_IOC_DEFRAG_RANGE:
5550                 return btrfs_ioctl_defrag(file, argp);
5551         case BTRFS_IOC_RESIZE:
5552                 return btrfs_ioctl_resize(file, argp);
5553         case BTRFS_IOC_ADD_DEV:
5554                 return btrfs_ioctl_add_dev(root, argp);
5555         case BTRFS_IOC_RM_DEV:
5556                 return btrfs_ioctl_rm_dev(file, argp);
5557         case BTRFS_IOC_FS_INFO:
5558                 return btrfs_ioctl_fs_info(root, argp);
5559         case BTRFS_IOC_DEV_INFO:
5560                 return btrfs_ioctl_dev_info(root, argp);
5561         case BTRFS_IOC_BALANCE:
5562                 return btrfs_ioctl_balance(file, NULL);
5563         case BTRFS_IOC_CLONE:
5564                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5565         case BTRFS_IOC_CLONE_RANGE:
5566                 return btrfs_ioctl_clone_range(file, argp);
5567         case BTRFS_IOC_TRANS_START:
5568                 return btrfs_ioctl_trans_start(file);
5569         case BTRFS_IOC_TRANS_END:
5570                 return btrfs_ioctl_trans_end(file);
5571         case BTRFS_IOC_TREE_SEARCH:
5572                 return btrfs_ioctl_tree_search(file, argp);
5573         case BTRFS_IOC_TREE_SEARCH_V2:
5574                 return btrfs_ioctl_tree_search_v2(file, argp);
5575         case BTRFS_IOC_INO_LOOKUP:
5576                 return btrfs_ioctl_ino_lookup(file, argp);
5577         case BTRFS_IOC_INO_PATHS:
5578                 return btrfs_ioctl_ino_to_path(root, argp);
5579         case BTRFS_IOC_LOGICAL_INO:
5580                 return btrfs_ioctl_logical_to_ino(root, argp);
5581         case BTRFS_IOC_SPACE_INFO:
5582                 return btrfs_ioctl_space_info(root, argp);
5583         case BTRFS_IOC_SYNC: {
5584                 int ret;
5585
5586                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5587                 if (ret)
5588                         return ret;
5589                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5590                 /*
5591                  * The transaction thread may want to do more work,
5592                  * namely it pokes the cleaner ktread that will start
5593                  * processing uncleaned subvols.
5594                  */
5595                 wake_up_process(root->fs_info->transaction_kthread);
5596                 return ret;
5597         }
5598         case BTRFS_IOC_START_SYNC:
5599                 return btrfs_ioctl_start_sync(root, argp);
5600         case BTRFS_IOC_WAIT_SYNC:
5601                 return btrfs_ioctl_wait_sync(root, argp);
5602         case BTRFS_IOC_SCRUB:
5603                 return btrfs_ioctl_scrub(file, argp);
5604         case BTRFS_IOC_SCRUB_CANCEL:
5605                 return btrfs_ioctl_scrub_cancel(root, argp);
5606         case BTRFS_IOC_SCRUB_PROGRESS:
5607                 return btrfs_ioctl_scrub_progress(root, argp);
5608         case BTRFS_IOC_BALANCE_V2:
5609                 return btrfs_ioctl_balance(file, argp);
5610         case BTRFS_IOC_BALANCE_CTL:
5611                 return btrfs_ioctl_balance_ctl(root, arg);
5612         case BTRFS_IOC_BALANCE_PROGRESS:
5613                 return btrfs_ioctl_balance_progress(root, argp);
5614         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5615                 return btrfs_ioctl_set_received_subvol(file, argp);
5616 #ifdef CONFIG_64BIT
5617         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5618                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5619 #endif
5620         case BTRFS_IOC_SEND:
5621                 return btrfs_ioctl_send(file, argp);
5622         case BTRFS_IOC_GET_DEV_STATS:
5623                 return btrfs_ioctl_get_dev_stats(root, argp);
5624         case BTRFS_IOC_QUOTA_CTL:
5625                 return btrfs_ioctl_quota_ctl(file, argp);
5626         case BTRFS_IOC_QGROUP_ASSIGN:
5627                 return btrfs_ioctl_qgroup_assign(file, argp);
5628         case BTRFS_IOC_QGROUP_CREATE:
5629                 return btrfs_ioctl_qgroup_create(file, argp);
5630         case BTRFS_IOC_QGROUP_LIMIT:
5631                 return btrfs_ioctl_qgroup_limit(file, argp);
5632         case BTRFS_IOC_QUOTA_RESCAN:
5633                 return btrfs_ioctl_quota_rescan(file, argp);
5634         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5635                 return btrfs_ioctl_quota_rescan_status(file, argp);
5636         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5637                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5638         case BTRFS_IOC_DEV_REPLACE:
5639                 return btrfs_ioctl_dev_replace(root, argp);
5640         case BTRFS_IOC_GET_FSLABEL:
5641                 return btrfs_ioctl_get_fslabel(file, argp);
5642         case BTRFS_IOC_SET_FSLABEL:
5643                 return btrfs_ioctl_set_fslabel(file, argp);
5644         case BTRFS_IOC_FILE_EXTENT_SAME:
5645                 return btrfs_ioctl_file_extent_same(file, argp);
5646         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5647                 return btrfs_ioctl_get_supported_features(file, argp);
5648         case BTRFS_IOC_GET_FEATURES:
5649                 return btrfs_ioctl_get_features(file, argp);
5650         case BTRFS_IOC_SET_FEATURES:
5651                 return btrfs_ioctl_set_features(file, argp);
5652         }
5653
5654         return -ENOTTY;
5655 }