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