fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  91                        int no_time_update);
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  95 {
  96         if (S_ISDIR(mode))
  97                 return flags;
  98         else if (S_ISREG(mode))
  99                 return flags & ~FS_DIRSYNC_FL;
 100         else
 101                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 102 }
 103
 104 /*
 105  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 106  */
 107 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 108 {
 109         unsigned int iflags = 0;
 110
 111         if (flags & BTRFS_INODE_SYNC)
 112                 iflags |= FS_SYNC_FL;
 113         if (flags & BTRFS_INODE_IMMUTABLE)
 114                 iflags |= FS_IMMUTABLE_FL;
 115         if (flags & BTRFS_INODE_APPEND)
 116                 iflags |= FS_APPEND_FL;
 117         if (flags & BTRFS_INODE_NODUMP)
 118                 iflags |= FS_NODUMP_FL;
 119         if (flags & BTRFS_INODE_NOATIME)
 120                 iflags |= FS_NOATIME_FL;
 121         if (flags & BTRFS_INODE_DIRSYNC)
 122                 iflags |= FS_DIRSYNC_FL;
 123         if (flags & BTRFS_INODE_NODATACOW)
 124                 iflags |= FS_NOCOW_FL;
 125
 126         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 127                 iflags |= FS_COMPR_FL;
 128         else if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130
 131         return iflags;
 132 }
 133
 134 /*
 135  * Update inode->i_flags based on the btrfs internal flags.
 136  */
 137 void btrfs_update_iflags(struct inode *inode)
 138 {
 139         struct btrfs_inode *ip = BTRFS_I(inode);
 140         unsigned int new_fl = 0;
 141
 142         if (ip->flags & BTRFS_INODE_SYNC)
 143                 new_fl |= S_SYNC;
 144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 145                 new_fl |= S_IMMUTABLE;
 146         if (ip->flags & BTRFS_INODE_APPEND)
 147                 new_fl |= S_APPEND;
 148         if (ip->flags & BTRFS_INODE_NOATIME)
 149                 new_fl |= S_NOATIME;
 150         if (ip->flags & BTRFS_INODE_DIRSYNC)
 151                 new_fl |= S_DIRSYNC;
 152
 153         set_mask_bits(&inode->i_flags,
 154                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 155                       new_fl);
 156 }
 157
 158 /*
 159  * Inherit flags from the parent inode.
 160  *
 161  * Currently only the compression flags and the cow flags are inherited.
 162  */
 163 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 164 {
 165         unsigned int flags;
 166
 167         if (!dir)
 168                 return;
 169
 170         flags = BTRFS_I(dir)->flags;
 171
 172         if (flags & BTRFS_INODE_NOCOMPRESS) {
 173                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 174                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 175         } else if (flags & BTRFS_INODE_COMPRESS) {
 176                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 177                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 178         }
 179
 180         if (flags & BTRFS_INODE_NODATACOW) {
 181                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 182                 if (S_ISREG(inode->i_mode))
 183                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 184         }
 185
 186         btrfs_update_iflags(inode);
 187 }
 188
 189 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 190 {
 191         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 192         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 193
 194         if (copy_to_user(arg, &flags, sizeof(flags)))
 195                 return -EFAULT;
 196         return 0;
 197 }
 198
 199 static int check_flags(unsigned int flags)
 200 {
 201         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 202                       FS_NOATIME_FL | FS_NODUMP_FL | \
 203                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 204                       FS_NOCOMP_FL | FS_COMPR_FL |
 205                       FS_NOCOW_FL))
 206                 return -EOPNOTSUPP;
 207
 208         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 215 {
 216         struct inode *inode = file_inode(file);
 217         struct btrfs_inode *ip = BTRFS_I(inode);
 218         struct btrfs_root *root = ip->root;
 219         struct btrfs_trans_handle *trans;
 220         unsigned int flags, oldflags;
 221         int ret;
 222         u64 ip_oldflags;
 223         unsigned int i_oldflags;
 224         umode_t mode;
 225
 226         if (!inode_owner_or_capable(inode))
 227                 return -EPERM;
 228
 229         if (btrfs_root_readonly(root))
 230                 return -EROFS;
 231
 232         if (copy_from_user(&flags, arg, sizeof(flags)))
 233                 return -EFAULT;
 234
 235         ret = check_flags(flags);
 236         if (ret)
 237                 return ret;
 238
 239         ret = mnt_want_write_file(file);
 240         if (ret)
 241                 return ret;
 242
 243         mutex_lock(&inode->i_mutex);
 244
 245         ip_oldflags = ip->flags;
 246         i_oldflags = inode->i_flags;
 247         mode = inode->i_mode;
 248
 249         flags = btrfs_mask_flags(inode->i_mode, flags);
 250         oldflags = btrfs_flags_to_ioctl(ip->flags);
 251         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 252                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 253                         ret = -EPERM;
 254                         goto out_unlock;
 255                 }
 256         }
 257
 258         if (flags & FS_SYNC_FL)
 259                 ip->flags |= BTRFS_INODE_SYNC;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_SYNC;
 262         if (flags & FS_IMMUTABLE_FL)
 263                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 266         if (flags & FS_APPEND_FL)
 267                 ip->flags |= BTRFS_INODE_APPEND;
 268         else
 269                 ip->flags &= ~BTRFS_INODE_APPEND;
 270         if (flags & FS_NODUMP_FL)
 271                 ip->flags |= BTRFS_INODE_NODUMP;
 272         else
 273                 ip->flags &= ~BTRFS_INODE_NODUMP;
 274         if (flags & FS_NOATIME_FL)
 275                 ip->flags |= BTRFS_INODE_NOATIME;
 276         else
 277                 ip->flags &= ~BTRFS_INODE_NOATIME;
 278         if (flags & FS_DIRSYNC_FL)
 279                 ip->flags |= BTRFS_INODE_DIRSYNC;
 280         else
 281                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 282         if (flags & FS_NOCOW_FL) {
 283                 if (S_ISREG(mode)) {
 284                         /*
 285                          * It's safe to turn csums off here, no extents exist.
 286                          * Otherwise we want the flag to reflect the real COW
 287                          * status of the file and will not set it.
 288                          */
 289                         if (inode->i_size == 0)
 290                                 ip->flags |= BTRFS_INODE_NODATACOW
 291                                            | BTRFS_INODE_NODATASUM;
 292                 } else {
 293                         ip->flags |= BTRFS_INODE_NODATACOW;
 294                 }
 295         } else {
 296                 /*
 297                  * Revert back under same assuptions as above
 298                  */
 299                 if (S_ISREG(mode)) {
 300                         if (inode->i_size == 0)
 301                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 302                                              | BTRFS_INODE_NODATASUM);
 303                 } else {
 304                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 305                 }
 306         }
 307
 308         /*
 309          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 310          * flag may be changed automatically if compression code won't make
 311          * things smaller.
 312          */
 313         if (flags & FS_NOCOMP_FL) {
 314                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 315                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 316
 317                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 318                 if (ret && ret != -ENODATA)
 319                         goto out_drop;
 320         } else if (flags & FS_COMPR_FL) {
 321                 const char *comp;
 322
 323                 ip->flags |= BTRFS_INODE_COMPRESS;
 324                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 325
 326                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 327                         comp = "lzo";
 328                 else
 329                         comp = "zlib";
 330                 ret = btrfs_set_prop(inode, "btrfs.compression",
 331                                      comp, strlen(comp), 0);
 332                 if (ret)
 333                         goto out_drop;
 334
 335         } else {
 336                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 337                 if (ret && ret != -ENODATA)
 338                         goto out_drop;
 339                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 340         }
 341
 342         trans = btrfs_start_transaction(root, 1);
 343         if (IS_ERR(trans)) {
 344                 ret = PTR_ERR(trans);
 345                 goto out_drop;
 346         }
 347
 348         btrfs_update_iflags(inode);
 349         inode_inc_iversion(inode);
 350         inode->i_ctime = CURRENT_TIME;
 351         ret = btrfs_update_inode(trans, root, inode);
 352
 353         btrfs_end_transaction(trans, root);
 354  out_drop:
 355         if (ret) {
 356                 ip->flags = ip_oldflags;
 357                 inode->i_flags = i_oldflags;
 358         }
 359
 360  out_unlock:
 361         mutex_unlock(&inode->i_mutex);
 362         mnt_drop_write_file(file);
 363         return ret;
 364 }
 365
 366 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 367 {
 368         struct inode *inode = file_inode(file);
 369
 370         return put_user(inode->i_generation, arg);
 371 }
 372
 373 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 374 {
 375         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 376         struct btrfs_device *device;
 377         struct request_queue *q;
 378         struct fstrim_range range;
 379         u64 minlen = ULLONG_MAX;
 380         u64 num_devices = 0;
 381         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 382         int ret;
 383
 384         if (!capable(CAP_SYS_ADMIN))
 385                 return -EPERM;
 386
 387         rcu_read_lock();
 388         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 389                                 dev_list) {
 390                 if (!device->bdev)
 391                         continue;
 392                 q = bdev_get_queue(device->bdev);
 393                 if (blk_queue_discard(q)) {
 394                         num_devices++;
 395                         minlen = min((u64)q->limits.discard_granularity,
 396                                      minlen);
 397                 }
 398         }
 399         rcu_read_unlock();
 400
 401         if (!num_devices)
 402                 return -EOPNOTSUPP;
 403         if (copy_from_user(&range, arg, sizeof(range)))
 404                 return -EFAULT;
 405         if (range.start > total_bytes ||
 406             range.len < fs_info->sb->s_blocksize)
 407                 return -EINVAL;
 408
 409         range.len = min(range.len, total_bytes - range.start);
 410         range.minlen = max(range.minlen, minlen);
 411         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 412         if (ret < 0)
 413                 return ret;
 414
 415         if (copy_to_user(arg, &range, sizeof(range)))
 416                 return -EFAULT;
 417
 418         return 0;
 419 }
 420
 421 int btrfs_is_empty_uuid(u8 *uuid)
 422 {
 423         int i;
 424
 425         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 426                 if (uuid[i])
 427                         return 0;
 428         }
 429         return 1;
 430 }
 431
 432 static noinline int create_subvol(struct inode *dir,
 433                                   struct dentry *dentry,
 434                                   char *name, int namelen,
 435                                   u64 *async_transid,
 436                                   struct btrfs_qgroup_inherit *inherit)
 437 {
 438         struct btrfs_trans_handle *trans;
 439         struct btrfs_key key;
 440         struct btrfs_root_item root_item;
 441         struct btrfs_inode_item *inode_item;
 442         struct extent_buffer *leaf;
 443         struct btrfs_root *root = BTRFS_I(dir)->root;
 444         struct btrfs_root *new_root;
 445         struct btrfs_block_rsv block_rsv;
 446         struct timespec cur_time = CURRENT_TIME;
 447         struct inode *inode;
 448         int ret;
 449         int err;
 450         u64 objectid;
 451         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 452         u64 index = 0;
 453         u64 qgroup_reserved;
 454         uuid_le new_uuid;
 455
 456         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 457         if (ret)
 458                 return ret;
 459
 460         /*
 461          * Don't create subvolume whose level is not zero. Or qgroup will be
 462          * screwed up since it assume subvolme qgroup's level to be 0.
 463          */
 464         if (btrfs_qgroup_level(objectid))
 465                 return -ENOSPC;
 466
 467         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 468         /*
 469          * The same as the snapshot creation, please see the comment
 470          * of create_snapshot().
 471          */
 472         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 473                                                8, &qgroup_reserved, false);
 474         if (ret)
 475                 return ret;
 476
 477         trans = btrfs_start_transaction(root, 0);
 478         if (IS_ERR(trans)) {
 479                 ret = PTR_ERR(trans);
 480                 btrfs_subvolume_release_metadata(root, &block_rsv,
 481                                                  qgroup_reserved);
 482                 return ret;
 483         }
 484         trans->block_rsv = &block_rsv;
 485         trans->bytes_reserved = block_rsv.size;
 486
 487         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 488         if (ret)
 489                 goto fail;
 490
 491         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 492         if (IS_ERR(leaf)) {
 493                 ret = PTR_ERR(leaf);
 494                 goto fail;
 495         }
 496
 497         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 498         btrfs_set_header_bytenr(leaf, leaf->start);
 499         btrfs_set_header_generation(leaf, trans->transid);
 500         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 501         btrfs_set_header_owner(leaf, objectid);
 502
 503         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 504                             BTRFS_FSID_SIZE);
 505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 506                             btrfs_header_chunk_tree_uuid(leaf),
 507                             BTRFS_UUID_SIZE);
 508         btrfs_mark_buffer_dirty(leaf);
 509
 510         memset(&root_item, 0, sizeof(root_item));
 511
 512         inode_item = &root_item.inode;
 513         btrfs_set_stack_inode_generation(inode_item, 1);
 514         btrfs_set_stack_inode_size(inode_item, 3);
 515         btrfs_set_stack_inode_nlink(inode_item, 1);
 516         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 517         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 518
 519         btrfs_set_root_flags(&root_item, 0);
 520         btrfs_set_root_limit(&root_item, 0);
 521         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 522
 523         btrfs_set_root_bytenr(&root_item, leaf->start);
 524         btrfs_set_root_generation(&root_item, trans->transid);
 525         btrfs_set_root_level(&root_item, 0);
 526         btrfs_set_root_refs(&root_item, 1);
 527         btrfs_set_root_used(&root_item, leaf->len);
 528         btrfs_set_root_last_snapshot(&root_item, 0);
 529
 530         btrfs_set_root_generation_v2(&root_item,
 531                         btrfs_root_generation(&root_item));
 532         uuid_le_gen(&new_uuid);
 533         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 534         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 535         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 536         root_item.ctime = root_item.otime;
 537         btrfs_set_root_ctransid(&root_item, trans->transid);
 538         btrfs_set_root_otransid(&root_item, trans->transid);
 539
 540         btrfs_tree_unlock(leaf);
 541         free_extent_buffer(leaf);
 542         leaf = NULL;
 543
 544         btrfs_set_root_dirid(&root_item, new_dirid);
 545
 546         key.objectid = objectid;
 547         key.offset = 0;
 548         key.type = BTRFS_ROOT_ITEM_KEY;
 549         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 550                                 &root_item);
 551         if (ret)
 552                 goto fail;
 553
 554         key.offset = (u64)-1;
 555         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 556         if (IS_ERR(new_root)) {
 557                 ret = PTR_ERR(new_root);
 558                 btrfs_abort_transaction(trans, root, ret);
 559                 goto fail;
 560         }
 561
 562         btrfs_record_root_in_trans(trans, new_root);
 563
 564         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 565         if (ret) {
 566                 /* We potentially lose an unused inode item here */
 567                 btrfs_abort_transaction(trans, root, ret);
 568                 goto fail;
 569         }
 570
 571         /*
 572          * insert the directory item
 573          */
 574         ret = btrfs_set_inode_index(dir, &index);
 575         if (ret) {
 576                 btrfs_abort_transaction(trans, root, ret);
 577                 goto fail;
 578         }
 579
 580         ret = btrfs_insert_dir_item(trans, root,
 581                                     name, namelen, dir, &key,
 582                                     BTRFS_FT_DIR, index);
 583         if (ret) {
 584                 btrfs_abort_transaction(trans, root, ret);
 585                 goto fail;
 586         }
 587
 588         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 589         ret = btrfs_update_inode(trans, root, dir);
 590         BUG_ON(ret);
 591
 592         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 593                                  objectid, root->root_key.objectid,
 594                                  btrfs_ino(dir), index, name, namelen);
 595         BUG_ON(ret);
 596
 597         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 598                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 599                                   objectid);
 600         if (ret)
 601                 btrfs_abort_transaction(trans, root, ret);
 602
 603 fail:
 604         trans->block_rsv = NULL;
 605         trans->bytes_reserved = 0;
 606         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 607
 608         if (async_transid) {
 609                 *async_transid = trans->transid;
 610                 err = btrfs_commit_transaction_async(trans, root, 1);
 611                 if (err)
 612                         err = btrfs_commit_transaction(trans, root);
 613         } else {
 614                 err = btrfs_commit_transaction(trans, root);
 615         }
 616         if (err && !ret)
 617                 ret = err;
 618
 619         if (!ret) {
 620                 inode = btrfs_lookup_dentry(dir, dentry);
 621                 if (IS_ERR(inode))
 622                         return PTR_ERR(inode);
 623                 d_instantiate(dentry, inode);
 624         }
 625         return ret;
 626 }
 627
 628 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 629 {
 630         s64 writers;
 631         DEFINE_WAIT(wait);
 632
 633         do {
 634                 prepare_to_wait(&root->subv_writers->wait, &wait,
 635                                 TASK_UNINTERRUPTIBLE);
 636
 637                 writers = percpu_counter_sum(&root->subv_writers->counter);
 638                 if (writers)
 639                         schedule();
 640
 641                 finish_wait(&root->subv_writers->wait, &wait);
 642         } while (writers);
 643 }
 644
 645 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 646                            struct dentry *dentry, char *name, int namelen,
 647                            u64 *async_transid, bool readonly,
 648                            struct btrfs_qgroup_inherit *inherit)
 649 {
 650         struct inode *inode;
 651         struct btrfs_pending_snapshot *pending_snapshot;
 652         struct btrfs_trans_handle *trans;
 653         int ret;
 654
 655         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 656                 return -EINVAL;
 657
 658         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 659         if (!pending_snapshot)
 660                 return -ENOMEM;
 661
 662         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 663                         GFP_NOFS);
 664         pending_snapshot->path = btrfs_alloc_path();
 665         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 666                 ret = -ENOMEM;
 667                 goto free_pending;
 668         }
 669
 670         atomic_inc(&root->will_be_snapshoted);
 671         smp_mb__after_atomic();
 672         btrfs_wait_for_no_snapshoting_writes(root);
 673
 674         ret = btrfs_start_delalloc_inodes(root, 0);
 675         if (ret)
 676                 goto dec_and_free;
 677
 678         btrfs_wait_ordered_extents(root, -1);
 679
 680         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 681                              BTRFS_BLOCK_RSV_TEMP);
 682         /*
 683          * 1 - parent dir inode
 684          * 2 - dir entries
 685          * 1 - root item
 686          * 2 - root ref/backref
 687          * 1 - root of snapshot
 688          * 1 - UUID item
 689          */
 690         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 691                                         &pending_snapshot->block_rsv, 8,
 692                                         &pending_snapshot->qgroup_reserved,
 693                                         false);
 694         if (ret)
 695                 goto dec_and_free;
 696
 697         pending_snapshot->dentry = dentry;
 698         pending_snapshot->root = root;
 699         pending_snapshot->readonly = readonly;
 700         pending_snapshot->dir = dir;
 701         pending_snapshot->inherit = inherit;
 702
 703         trans = btrfs_start_transaction(root, 0);
 704         if (IS_ERR(trans)) {
 705                 ret = PTR_ERR(trans);
 706                 goto fail;
 707         }
 708
 709         spin_lock(&root->fs_info->trans_lock);
 710         list_add(&pending_snapshot->list,
 711                  &trans->transaction->pending_snapshots);
 712         spin_unlock(&root->fs_info->trans_lock);
 713         if (async_transid) {
 714                 *async_transid = trans->transid;
 715                 ret = btrfs_commit_transaction_async(trans,
 716                                      root->fs_info->extent_root, 1);
 717                 if (ret)
 718                         ret = btrfs_commit_transaction(trans, root);
 719         } else {
 720                 ret = btrfs_commit_transaction(trans,
 721                                                root->fs_info->extent_root);
 722         }
 723         if (ret)
 724                 goto fail;
 725
 726         ret = pending_snapshot->error;
 727         if (ret)
 728                 goto fail;
 729
 730         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 731         if (ret)
 732                 goto fail;
 733
 734         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 735         if (IS_ERR(inode)) {
 736                 ret = PTR_ERR(inode);
 737                 goto fail;
 738         }
 739
 740         d_instantiate(dentry, inode);
 741         ret = 0;
 742 fail:
 743         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 744                                          &pending_snapshot->block_rsv,
 745                                          pending_snapshot->qgroup_reserved);
 746 dec_and_free:
 747         if (atomic_dec_and_test(&root->will_be_snapshoted))
 748                 wake_up_atomic_t(&root->will_be_snapshoted);
 749 free_pending:
 750         kfree(pending_snapshot->root_item);
 751         btrfs_free_path(pending_snapshot->path);
 752         kfree(pending_snapshot);
 753
 754         return ret;
 755 }
 756
 757 /*  copy of may_delete in fs/namei.c()
 758  *      Check whether we can remove a link victim from directory dir, check
 759  *  whether the type of victim is right.
 760  *  1. We can't do it if dir is read-only (done in permission())
 761  *  2. We should have write and exec permissions on dir
 762  *  3. We can't remove anything from append-only dir
 763  *  4. We can't do anything with immutable dir (done in permission())
 764  *  5. If the sticky bit on dir is set we should either
 765  *      a. be owner of dir, or
 766  *      b. be owner of victim, or
 767  *      c. have CAP_FOWNER capability
 768  *  6. If the victim is append-only or immutable we can't do antyhing with
 769  *     links pointing to it.
 770  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 771  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 772  *  9. We can't remove a root or mountpoint.
 773  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 774  *     nfs_async_unlink().
 775  */
 776
 777 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 778 {
 779         int error;
 780
 781         if (d_really_is_negative(victim))
 782                 return -ENOENT;
 783
 784         BUG_ON(d_inode(victim->d_parent) != dir);
 785         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 786
 787         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 788         if (error)
 789                 return error;
 790         if (IS_APPEND(dir))
 791                 return -EPERM;
 792         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 793             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 794                 return -EPERM;
 795         if (isdir) {
 796                 if (!d_is_dir(victim))
 797                         return -ENOTDIR;
 798                 if (IS_ROOT(victim))
 799                         return -EBUSY;
 800         } else if (d_is_dir(victim))
 801                 return -EISDIR;
 802         if (IS_DEADDIR(dir))
 803                 return -ENOENT;
 804         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 805                 return -EBUSY;
 806         return 0;
 807 }
 808
 809 /* copy of may_create in fs/namei.c() */
 810 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 811 {
 812         if (d_really_is_positive(child))
 813                 return -EEXIST;
 814         if (IS_DEADDIR(dir))
 815                 return -ENOENT;
 816         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 817 }
 818
 819 /*
 820  * Create a new subvolume below @parent.  This is largely modeled after
 821  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 822  * inside this filesystem so it's quite a bit simpler.
 823  */
 824 static noinline int btrfs_mksubvol(struct path *parent,
 825                                    char *name, int namelen,
 826                                    struct btrfs_root *snap_src,
 827                                    u64 *async_transid, bool readonly,
 828                                    struct btrfs_qgroup_inherit *inherit)
 829 {
 830         struct inode *dir  = d_inode(parent->dentry);
 831         struct dentry *dentry;
 832         int error;
 833
 834         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 835         if (error == -EINTR)
 836                 return error;
 837
 838         dentry = lookup_one_len(name, parent->dentry, namelen);
 839         error = PTR_ERR(dentry);
 840         if (IS_ERR(dentry))
 841                 goto out_unlock;
 842
 843         error = -EEXIST;
 844         if (d_really_is_positive(dentry))
 845                 goto out_dput;
 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         mutex_unlock(&dir->i_mutex);
 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                 mutex_lock(&inode->i_mutex);
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                         mutex_unlock(&inode->i_mutex);
1398                         goto out_ra;
1399                 }
1400
1401                 defrag_count += ret;
1402                 balance_dirty_pages_ratelimited(inode->i_mapping);
1403                 mutex_unlock(&inode->i_mutex);
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                 mutex_lock(&inode->i_mutex);
1465                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1466                 mutex_unlock(&inode->i_mutex);
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_err(info, "could not find root %llu",
2097                                sk->tree_id);
2098                         btrfs_free_path(path);
2099                         return -ENOENT;
2100                 }
2101         }
2102
2103         key.objectid = sk->min_objectid;
2104         key.type = sk->min_type;
2105         key.offset = sk->min_offset;
2106
2107         while (1) {
2108                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2109                 if (ret != 0) {
2110                         if (ret > 0)
2111                                 ret = 0;
2112                         goto err;
2113                 }
2114                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2115                                  &sk_offset, &num_found);
2116                 btrfs_release_path(path);
2117                 if (ret)
2118                         break;
2119
2120         }
2121         if (ret > 0)
2122                 ret = 0;
2123 err:
2124         sk->nr_items = num_found;
2125         btrfs_free_path(path);
2126         return ret;
2127 }
2128
2129 static noinline int btrfs_ioctl_tree_search(struct file *file,
2130                                            void __user *argp)
2131 {
2132         struct btrfs_ioctl_search_args __user *uargs;
2133         struct btrfs_ioctl_search_key sk;
2134         struct inode *inode;
2135         int ret;
2136         size_t buf_size;
2137
2138         if (!capable(CAP_SYS_ADMIN))
2139                 return -EPERM;
2140
2141         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2142
2143         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2144                 return -EFAULT;
2145
2146         buf_size = sizeof(uargs->buf);
2147
2148         inode = file_inode(file);
2149         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2150
2151         /*
2152          * In the origin implementation an overflow is handled by returning a
2153          * search header with a len of zero, so reset ret.
2154          */
2155         if (ret == -EOVERFLOW)
2156                 ret = 0;
2157
2158         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2159                 ret = -EFAULT;
2160         return ret;
2161 }
2162
2163 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2164                                                void __user *argp)
2165 {
2166         struct btrfs_ioctl_search_args_v2 __user *uarg;
2167         struct btrfs_ioctl_search_args_v2 args;
2168         struct inode *inode;
2169         int ret;
2170         size_t buf_size;
2171         const size_t buf_limit = SZ_16M;
2172
2173         if (!capable(CAP_SYS_ADMIN))
2174                 return -EPERM;
2175
2176         /* copy search header and buffer size */
2177         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2178         if (copy_from_user(&args, uarg, sizeof(args)))
2179                 return -EFAULT;
2180
2181         buf_size = args.buf_size;
2182
2183         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2184                 return -EOVERFLOW;
2185
2186         /* limit result size to 16MB */
2187         if (buf_size > buf_limit)
2188                 buf_size = buf_limit;
2189
2190         inode = file_inode(file);
2191         ret = search_ioctl(inode, &args.key, &buf_size,
2192                            (char *)(&uarg->buf[0]));
2193         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2194                 ret = -EFAULT;
2195         else if (ret == -EOVERFLOW &&
2196                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2197                 ret = -EFAULT;
2198
2199         return ret;
2200 }
2201
2202 /*
2203  * Search INODE_REFs to identify path name of 'dirid' directory
2204  * in a 'tree_id' tree. and sets path name to 'name'.
2205  */
2206 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2207                                 u64 tree_id, u64 dirid, char *name)
2208 {
2209         struct btrfs_root *root;
2210         struct btrfs_key key;
2211         char *ptr;
2212         int ret = -1;
2213         int slot;
2214         int len;
2215         int total_len = 0;
2216         struct btrfs_inode_ref *iref;
2217         struct extent_buffer *l;
2218         struct btrfs_path *path;
2219
2220         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2221                 name[0]='\0';
2222                 return 0;
2223         }
2224
2225         path = btrfs_alloc_path();
2226         if (!path)
2227                 return -ENOMEM;
2228
2229         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2230
2231         key.objectid = tree_id;
2232         key.type = BTRFS_ROOT_ITEM_KEY;
2233         key.offset = (u64)-1;
2234         root = btrfs_read_fs_root_no_name(info, &key);
2235         if (IS_ERR(root)) {
2236                 btrfs_err(info, "could not find root %llu", tree_id);
2237                 ret = -ENOENT;
2238                 goto out;
2239         }
2240
2241         key.objectid = dirid;
2242         key.type = BTRFS_INODE_REF_KEY;
2243         key.offset = (u64)-1;
2244
2245         while (1) {
2246                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2247                 if (ret < 0)
2248                         goto out;
2249                 else if (ret > 0) {
2250                         ret = btrfs_previous_item(root, path, dirid,
2251                                                   BTRFS_INODE_REF_KEY);
2252                         if (ret < 0)
2253                                 goto out;
2254                         else if (ret > 0) {
2255                                 ret = -ENOENT;
2256                                 goto out;
2257                         }
2258                 }
2259
2260                 l = path->nodes[0];
2261                 slot = path->slots[0];
2262                 btrfs_item_key_to_cpu(l, &key, slot);
2263
2264                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2265                 len = btrfs_inode_ref_name_len(l, iref);
2266                 ptr -= len + 1;
2267                 total_len += len + 1;
2268                 if (ptr < name) {
2269                         ret = -ENAMETOOLONG;
2270                         goto out;
2271                 }
2272
2273                 *(ptr + len) = '/';
2274                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2275
2276                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2277                         break;
2278
2279                 btrfs_release_path(path);
2280                 key.objectid = key.offset;
2281                 key.offset = (u64)-1;
2282                 dirid = key.objectid;
2283         }
2284         memmove(name, ptr, total_len);
2285         name[total_len] = '\0';
2286         ret = 0;
2287 out:
2288         btrfs_free_path(path);
2289         return ret;
2290 }
2291
2292 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2293                                            void __user *argp)
2294 {
2295          struct btrfs_ioctl_ino_lookup_args *args;
2296          struct inode *inode;
2297         int ret = 0;
2298
2299         args = memdup_user(argp, sizeof(*args));
2300         if (IS_ERR(args))
2301                 return PTR_ERR(args);
2302
2303         inode = file_inode(file);
2304
2305         /*
2306          * Unprivileged query to obtain the containing subvolume root id. The
2307          * path is reset so it's consistent with btrfs_search_path_in_tree.
2308          */
2309         if (args->treeid == 0)
2310                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2311
2312         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2313                 args->name[0] = 0;
2314                 goto out;
2315         }
2316
2317         if (!capable(CAP_SYS_ADMIN)) {
2318                 ret = -EPERM;
2319                 goto out;
2320         }
2321
2322         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2323                                         args->treeid, args->objectid,
2324                                         args->name);
2325
2326 out:
2327         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2328                 ret = -EFAULT;
2329
2330         kfree(args);
2331         return ret;
2332 }
2333
2334 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2335                                              void __user *arg)
2336 {
2337         struct dentry *parent = file->f_path.dentry;
2338         struct dentry *dentry;
2339         struct inode *dir = d_inode(parent);
2340         struct inode *inode;
2341         struct btrfs_root *root = BTRFS_I(dir)->root;
2342         struct btrfs_root *dest = NULL;
2343         struct btrfs_ioctl_vol_args *vol_args;
2344         struct btrfs_trans_handle *trans;
2345         struct btrfs_block_rsv block_rsv;
2346         u64 root_flags;
2347         u64 qgroup_reserved;
2348         int namelen;
2349         int ret;
2350         int err = 0;
2351
2352         vol_args = memdup_user(arg, sizeof(*vol_args));
2353         if (IS_ERR(vol_args))
2354                 return PTR_ERR(vol_args);
2355
2356         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2357         namelen = strlen(vol_args->name);
2358         if (strchr(vol_args->name, '/') ||
2359             strncmp(vol_args->name, "..", namelen) == 0) {
2360                 err = -EINVAL;
2361                 goto out;
2362         }
2363
2364         err = mnt_want_write_file(file);
2365         if (err)
2366                 goto out;
2367
2368
2369         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2370         if (err == -EINTR)
2371                 goto out_drop_write;
2372         dentry = lookup_one_len(vol_args->name, parent, namelen);
2373         if (IS_ERR(dentry)) {
2374                 err = PTR_ERR(dentry);
2375                 goto out_unlock_dir;
2376         }
2377
2378         if (d_really_is_negative(dentry)) {
2379                 err = -ENOENT;
2380                 goto out_dput;
2381         }
2382
2383         inode = d_inode(dentry);
2384         dest = BTRFS_I(inode)->root;
2385         if (!capable(CAP_SYS_ADMIN)) {
2386                 /*
2387                  * Regular user.  Only allow this with a special mount
2388                  * option, when the user has write+exec access to the
2389                  * subvol root, and when rmdir(2) would have been
2390                  * allowed.
2391                  *
2392                  * Note that this is _not_ check that the subvol is
2393                  * empty or doesn't contain data that we wouldn't
2394                  * otherwise be able to delete.
2395                  *
2396                  * Users who want to delete empty subvols should try
2397                  * rmdir(2).
2398                  */
2399                 err = -EPERM;
2400                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2401                         goto out_dput;
2402
2403                 /*
2404                  * Do not allow deletion if the parent dir is the same
2405                  * as the dir to be deleted.  That means the ioctl
2406                  * must be called on the dentry referencing the root
2407                  * of the subvol, not a random directory contained
2408                  * within it.
2409                  */
2410                 err = -EINVAL;
2411                 if (root == dest)
2412                         goto out_dput;
2413
2414                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2415                 if (err)
2416                         goto out_dput;
2417         }
2418
2419         /* check if subvolume may be deleted by a user */
2420         err = btrfs_may_delete(dir, dentry, 1);
2421         if (err)
2422                 goto out_dput;
2423
2424         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2425                 err = -EINVAL;
2426                 goto out_dput;
2427         }
2428
2429         mutex_lock(&inode->i_mutex);
2430
2431         /*
2432          * Don't allow to delete a subvolume with send in progress. This is
2433          * inside the i_mutex so the error handling that has to drop the bit
2434          * again is not run concurrently.
2435          */
2436         spin_lock(&dest->root_item_lock);
2437         root_flags = btrfs_root_flags(&dest->root_item);
2438         if (dest->send_in_progress == 0) {
2439                 btrfs_set_root_flags(&dest->root_item,
2440                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2441                 spin_unlock(&dest->root_item_lock);
2442         } else {
2443                 spin_unlock(&dest->root_item_lock);
2444                 btrfs_warn(root->fs_info,
2445                         "Attempt to delete subvolume %llu during send",
2446                         dest->root_key.objectid);
2447                 err = -EPERM;
2448                 goto out_unlock_inode;
2449         }
2450
2451         down_write(&root->fs_info->subvol_sem);
2452
2453         err = may_destroy_subvol(dest);
2454         if (err)
2455                 goto out_up_write;
2456
2457         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2458         /*
2459          * One for dir inode, two for dir entries, two for root
2460          * ref/backref.
2461          */
2462         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2463                                                5, &qgroup_reserved, true);
2464         if (err)
2465                 goto out_up_write;
2466
2467         trans = btrfs_start_transaction(root, 0);
2468         if (IS_ERR(trans)) {
2469                 err = PTR_ERR(trans);
2470                 goto out_release;
2471         }
2472         trans->block_rsv = &block_rsv;
2473         trans->bytes_reserved = block_rsv.size;
2474
2475         ret = btrfs_unlink_subvol(trans, root, dir,
2476                                 dest->root_key.objectid,
2477                                 dentry->d_name.name,
2478                                 dentry->d_name.len);
2479         if (ret) {
2480                 err = ret;
2481                 btrfs_abort_transaction(trans, root, ret);
2482                 goto out_end_trans;
2483         }
2484
2485         btrfs_record_root_in_trans(trans, dest);
2486
2487         memset(&dest->root_item.drop_progress, 0,
2488                 sizeof(dest->root_item.drop_progress));
2489         dest->root_item.drop_level = 0;
2490         btrfs_set_root_refs(&dest->root_item, 0);
2491
2492         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2493                 ret = btrfs_insert_orphan_item(trans,
2494                                         root->fs_info->tree_root,
2495                                         dest->root_key.objectid);
2496                 if (ret) {
2497                         btrfs_abort_transaction(trans, root, ret);
2498                         err = ret;
2499                         goto out_end_trans;
2500                 }
2501         }
2502
2503         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2504                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2505                                   dest->root_key.objectid);
2506         if (ret && ret != -ENOENT) {
2507                 btrfs_abort_transaction(trans, root, ret);
2508                 err = ret;
2509                 goto out_end_trans;
2510         }
2511         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2512                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2513                                           dest->root_item.received_uuid,
2514                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2515                                           dest->root_key.objectid);
2516                 if (ret && ret != -ENOENT) {
2517                         btrfs_abort_transaction(trans, root, ret);
2518                         err = ret;
2519                         goto out_end_trans;
2520                 }
2521         }
2522
2523 out_end_trans:
2524         trans->block_rsv = NULL;
2525         trans->bytes_reserved = 0;
2526         ret = btrfs_end_transaction(trans, root);
2527         if (ret && !err)
2528                 err = ret;
2529         inode->i_flags |= S_DEAD;
2530 out_release:
2531         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2532 out_up_write:
2533         up_write(&root->fs_info->subvol_sem);
2534         if (err) {
2535                 spin_lock(&dest->root_item_lock);
2536                 root_flags = btrfs_root_flags(&dest->root_item);
2537                 btrfs_set_root_flags(&dest->root_item,
2538                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2539                 spin_unlock(&dest->root_item_lock);
2540         }
2541 out_unlock_inode:
2542         mutex_unlock(&inode->i_mutex);
2543         if (!err) {
2544                 d_invalidate(dentry);
2545                 btrfs_invalidate_inodes(dest);
2546                 d_delete(dentry);
2547                 ASSERT(dest->send_in_progress == 0);
2548
2549                 /* the last ref */
2550                 if (dest->ino_cache_inode) {
2551                         iput(dest->ino_cache_inode);
2552                         dest->ino_cache_inode = NULL;
2553                 }
2554         }
2555 out_dput:
2556         dput(dentry);
2557 out_unlock_dir:
2558         mutex_unlock(&dir->i_mutex);
2559 out_drop_write:
2560         mnt_drop_write_file(file);
2561 out:
2562         kfree(vol_args);
2563         return err;
2564 }
2565
2566 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2567 {
2568         struct inode *inode = file_inode(file);
2569         struct btrfs_root *root = BTRFS_I(inode)->root;
2570         struct btrfs_ioctl_defrag_range_args *range;
2571         int ret;
2572
2573         ret = mnt_want_write_file(file);
2574         if (ret)
2575                 return ret;
2576
2577         if (btrfs_root_readonly(root)) {
2578                 ret = -EROFS;
2579                 goto out;
2580         }
2581
2582         switch (inode->i_mode & S_IFMT) {
2583         case S_IFDIR:
2584                 if (!capable(CAP_SYS_ADMIN)) {
2585                         ret = -EPERM;
2586                         goto out;
2587                 }
2588                 ret = btrfs_defrag_root(root);
2589                 if (ret)
2590                         goto out;
2591                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2592                 break;
2593         case S_IFREG:
2594                 if (!(file->f_mode & FMODE_WRITE)) {
2595                         ret = -EINVAL;
2596                         goto out;
2597                 }
2598
2599                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2600                 if (!range) {
2601                         ret = -ENOMEM;
2602                         goto out;
2603                 }
2604
2605                 if (argp) {
2606                         if (copy_from_user(range, argp,
2607                                            sizeof(*range))) {
2608                                 ret = -EFAULT;
2609                                 kfree(range);
2610                                 goto out;
2611                         }
2612                         /* compression requires us to start the IO */
2613                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2614                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2615                                 range->extent_thresh = (u32)-1;
2616                         }
2617                 } else {
2618                         /* the rest are all set to zero by kzalloc */
2619                         range->len = (u64)-1;
2620                 }
2621                 ret = btrfs_defrag_file(file_inode(file), file,
2622                                         range, 0, 0);
2623                 if (ret > 0)
2624                         ret = 0;
2625                 kfree(range);
2626                 break;
2627         default:
2628                 ret = -EINVAL;
2629         }
2630 out:
2631         mnt_drop_write_file(file);
2632         return ret;
2633 }
2634
2635 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2636 {
2637         struct btrfs_ioctl_vol_args *vol_args;
2638         int ret;
2639
2640         if (!capable(CAP_SYS_ADMIN))
2641                 return -EPERM;
2642
2643         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2644                         1)) {
2645                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2646         }
2647
2648         mutex_lock(&root->fs_info->volume_mutex);
2649         vol_args = memdup_user(arg, sizeof(*vol_args));
2650         if (IS_ERR(vol_args)) {
2651                 ret = PTR_ERR(vol_args);
2652                 goto out;
2653         }
2654
2655         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2656         ret = btrfs_init_new_device(root, vol_args->name);
2657
2658         if (!ret)
2659                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2660
2661         kfree(vol_args);
2662 out:
2663         mutex_unlock(&root->fs_info->volume_mutex);
2664         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2665         return ret;
2666 }
2667
2668 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2669 {
2670         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2671         struct btrfs_ioctl_vol_args *vol_args;
2672         int ret;
2673
2674         if (!capable(CAP_SYS_ADMIN))
2675                 return -EPERM;
2676
2677         ret = mnt_want_write_file(file);
2678         if (ret)
2679                 return ret;
2680
2681         vol_args = memdup_user(arg, sizeof(*vol_args));
2682         if (IS_ERR(vol_args)) {
2683                 ret = PTR_ERR(vol_args);
2684                 goto err_drop;
2685         }
2686
2687         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2688
2689         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2690                         1)) {
2691                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2692                 goto out;
2693         }
2694
2695         mutex_lock(&root->fs_info->volume_mutex);
2696         ret = btrfs_rm_device(root, vol_args->name);
2697         mutex_unlock(&root->fs_info->volume_mutex);
2698         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2699
2700         if (!ret)
2701                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2702
2703 out:
2704         kfree(vol_args);
2705 err_drop:
2706         mnt_drop_write_file(file);
2707         return ret;
2708 }
2709
2710 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2711 {
2712         struct btrfs_ioctl_fs_info_args *fi_args;
2713         struct btrfs_device *device;
2714         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2715         int ret = 0;
2716
2717         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2718         if (!fi_args)
2719                 return -ENOMEM;
2720
2721         mutex_lock(&fs_devices->device_list_mutex);
2722         fi_args->num_devices = fs_devices->num_devices;
2723         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2724
2725         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2726                 if (device->devid > fi_args->max_id)
2727                         fi_args->max_id = device->devid;
2728         }
2729         mutex_unlock(&fs_devices->device_list_mutex);
2730
2731         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2732         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2733         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2734
2735         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2736                 ret = -EFAULT;
2737
2738         kfree(fi_args);
2739         return ret;
2740 }
2741
2742 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2743 {
2744         struct btrfs_ioctl_dev_info_args *di_args;
2745         struct btrfs_device *dev;
2746         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2747         int ret = 0;
2748         char *s_uuid = NULL;
2749
2750         di_args = memdup_user(arg, sizeof(*di_args));
2751         if (IS_ERR(di_args))
2752                 return PTR_ERR(di_args);
2753
2754         if (!btrfs_is_empty_uuid(di_args->uuid))
2755                 s_uuid = di_args->uuid;
2756
2757         mutex_lock(&fs_devices->device_list_mutex);
2758         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2759
2760         if (!dev) {
2761                 ret = -ENODEV;
2762                 goto out;
2763         }
2764
2765         di_args->devid = dev->devid;
2766         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2767         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2768         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2769         if (dev->name) {
2770                 struct rcu_string *name;
2771
2772                 rcu_read_lock();
2773                 name = rcu_dereference(dev->name);
2774                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2775                 rcu_read_unlock();
2776                 di_args->path[sizeof(di_args->path) - 1] = 0;
2777         } else {
2778                 di_args->path[0] = '\0';
2779         }
2780
2781 out:
2782         mutex_unlock(&fs_devices->device_list_mutex);
2783         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2784                 ret = -EFAULT;
2785
2786         kfree(di_args);
2787         return ret;
2788 }
2789
2790 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2791 {
2792         struct page *page;
2793         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2794
2795         page = grab_cache_page(inode->i_mapping, index);
2796         if (!page)
2797                 return NULL;
2798
2799         if (!PageUptodate(page)) {
2800                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2801                                                  0))
2802                         return NULL;
2803                 lock_page(page);
2804                 if (!PageUptodate(page)) {
2805                         unlock_page(page);
2806                         page_cache_release(page);
2807                         return NULL;
2808                 }
2809         }
2810         unlock_page(page);
2811
2812         return page;
2813 }
2814
2815 static int gather_extent_pages(struct inode *inode, struct page **pages,
2816                                int num_pages, u64 off)
2817 {
2818         int i;
2819         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2820
2821         for (i = 0; i < num_pages; i++) {
2822                 pages[i] = extent_same_get_page(inode, index + i);
2823                 if (!pages[i])
2824                         return -ENOMEM;
2825         }
2826         return 0;
2827 }
2828
2829 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2830 {
2831         /* do any pending delalloc/csum calc on src, one way or
2832            another, and lock file content */
2833         while (1) {
2834                 struct btrfs_ordered_extent *ordered;
2835                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2836                 ordered = btrfs_lookup_first_ordered_extent(inode,
2837                                                             off + len - 1);
2838                 if ((!ordered ||
2839                      ordered->file_offset + ordered->len <= off ||
2840                      ordered->file_offset >= off + len) &&
2841                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2842                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2843                         if (ordered)
2844                                 btrfs_put_ordered_extent(ordered);
2845                         break;
2846                 }
2847                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2848                 if (ordered)
2849                         btrfs_put_ordered_extent(ordered);
2850                 btrfs_wait_ordered_range(inode, off, len);
2851         }
2852 }
2853
2854 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2855 {
2856         mutex_unlock(&inode1->i_mutex);
2857         mutex_unlock(&inode2->i_mutex);
2858 }
2859
2860 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2861 {
2862         if (inode1 < inode2)
2863                 swap(inode1, inode2);
2864
2865         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2866         mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2867 }
2868
2869 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2870                                       struct inode *inode2, u64 loff2, u64 len)
2871 {
2872         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2873         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2874 }
2875
2876 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2877                                      struct inode *inode2, u64 loff2, u64 len)
2878 {
2879         if (inode1 < inode2) {
2880                 swap(inode1, inode2);
2881                 swap(loff1, loff2);
2882         }
2883         lock_extent_range(inode1, loff1, len);
2884         lock_extent_range(inode2, loff2, len);
2885 }
2886
2887 struct cmp_pages {
2888         int             num_pages;
2889         struct page     **src_pages;
2890         struct page     **dst_pages;
2891 };
2892
2893 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2894 {
2895         int i;
2896         struct page *pg;
2897
2898         for (i = 0; i < cmp->num_pages; i++) {
2899                 pg = cmp->src_pages[i];
2900                 if (pg)
2901                         page_cache_release(pg);
2902                 pg = cmp->dst_pages[i];
2903                 if (pg)
2904                         page_cache_release(pg);
2905         }
2906         kfree(cmp->src_pages);
2907         kfree(cmp->dst_pages);
2908 }
2909
2910 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2911                                   struct inode *dst, u64 dst_loff,
2912                                   u64 len, struct cmp_pages *cmp)
2913 {
2914         int ret;
2915         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2916         struct page **src_pgarr, **dst_pgarr;
2917
2918         /*
2919          * We must gather up all the pages before we initiate our
2920          * extent locking. We use an array for the page pointers. Size
2921          * of the array is bounded by len, which is in turn bounded by
2922          * BTRFS_MAX_DEDUPE_LEN.
2923          */
2924         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2925         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2926         if (!src_pgarr || !dst_pgarr) {
2927                 kfree(src_pgarr);
2928                 kfree(dst_pgarr);
2929                 return -ENOMEM;
2930         }
2931         cmp->num_pages = num_pages;
2932         cmp->src_pages = src_pgarr;
2933         cmp->dst_pages = dst_pgarr;
2934
2935         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2936         if (ret)
2937                 goto out;
2938
2939         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2940
2941 out:
2942         if (ret)
2943                 btrfs_cmp_data_free(cmp);
2944         return 0;
2945 }
2946
2947 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2948                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2949 {
2950         int ret = 0;
2951         int i;
2952         struct page *src_page, *dst_page;
2953         unsigned int cmp_len = PAGE_CACHE_SIZE;
2954         void *addr, *dst_addr;
2955
2956         i = 0;
2957         while (len) {
2958                 if (len < PAGE_CACHE_SIZE)
2959                         cmp_len = len;
2960
2961                 BUG_ON(i >= cmp->num_pages);
2962
2963                 src_page = cmp->src_pages[i];
2964                 dst_page = cmp->dst_pages[i];
2965
2966                 addr = kmap_atomic(src_page);
2967                 dst_addr = kmap_atomic(dst_page);
2968
2969                 flush_dcache_page(src_page);
2970                 flush_dcache_page(dst_page);
2971
2972                 if (memcmp(addr, dst_addr, cmp_len))
2973                         ret = BTRFS_SAME_DATA_DIFFERS;
2974
2975                 kunmap_atomic(addr);
2976                 kunmap_atomic(dst_addr);
2977
2978                 if (ret)
2979                         break;
2980
2981                 len -= cmp_len;
2982                 i++;
2983         }
2984
2985         return ret;
2986 }
2987
2988 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
2989                                      u64 olen)
2990 {
2991         u64 len = *plen;
2992         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2993
2994         if (off + olen > inode->i_size || off + olen < off)
2995                 return -EINVAL;
2996
2997         /* if we extend to eof, continue to block boundary */
2998         if (off + len == inode->i_size)
2999                 *plen = len = ALIGN(inode->i_size, bs) - off;
3000
3001         /* Check that we are block aligned - btrfs_clone() requires this */
3002         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3003                 return -EINVAL;
3004
3005         return 0;
3006 }
3007
3008 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3009                              struct inode *dst, u64 dst_loff)
3010 {
3011         int ret;
3012         u64 len = olen;
3013         struct cmp_pages cmp;
3014         int same_inode = 0;
3015         u64 same_lock_start = 0;
3016         u64 same_lock_len = 0;
3017
3018         if (src == dst)
3019                 same_inode = 1;
3020
3021         if (len == 0)
3022                 return 0;
3023
3024         if (same_inode) {
3025                 mutex_lock(&src->i_mutex);
3026
3027                 ret = extent_same_check_offsets(src, loff, &len, olen);
3028                 if (ret)
3029                         goto out_unlock;
3030
3031                 /*
3032                  * Single inode case wants the same checks, except we
3033                  * don't want our length pushed out past i_size as
3034                  * comparing that data range makes no sense.
3035                  *
3036                  * extent_same_check_offsets() will do this for an
3037                  * unaligned length at i_size, so catch it here and
3038                  * reject the request.
3039                  *
3040                  * This effectively means we require aligned extents
3041                  * for the single-inode case, whereas the other cases
3042                  * allow an unaligned length so long as it ends at
3043                  * i_size.
3044                  */
3045                 if (len != olen) {
3046                         ret = -EINVAL;
3047                         goto out_unlock;
3048                 }
3049
3050                 /* Check for overlapping ranges */
3051                 if (dst_loff + len > loff && dst_loff < loff + len) {
3052                         ret = -EINVAL;
3053                         goto out_unlock;
3054                 }
3055
3056                 same_lock_start = min_t(u64, loff, dst_loff);
3057                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3058         } else {
3059                 btrfs_double_inode_lock(src, dst);
3060
3061                 ret = extent_same_check_offsets(src, loff, &len, olen);
3062                 if (ret)
3063                         goto out_unlock;
3064
3065                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3066                 if (ret)
3067                         goto out_unlock;
3068         }
3069
3070         /* don't make the dst file partly checksummed */
3071         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3072             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3073                 ret = -EINVAL;
3074                 goto out_unlock;
3075         }
3076
3077         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3078         if (ret)
3079                 goto out_unlock;
3080
3081         if (same_inode)
3082                 lock_extent_range(src, same_lock_start, same_lock_len);
3083         else
3084                 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3085
3086         /* pass original length for comparison so we stay within i_size */
3087         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3088         if (ret == 0)
3089                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3090
3091         if (same_inode)
3092                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3093                               same_lock_start + same_lock_len - 1);
3094         else
3095                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3096
3097         btrfs_cmp_data_free(&cmp);
3098 out_unlock:
3099         if (same_inode)
3100                 mutex_unlock(&src->i_mutex);
3101         else
3102                 btrfs_double_inode_unlock(src, dst);
3103
3104         return ret;
3105 }
3106
3107 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3108
3109 static long btrfs_ioctl_file_extent_same(struct file *file,
3110                         struct btrfs_ioctl_same_args __user *argp)
3111 {
3112         struct btrfs_ioctl_same_args *same = NULL;
3113         struct btrfs_ioctl_same_extent_info *info;
3114         struct inode *src = file_inode(file);
3115         u64 off;
3116         u64 len;
3117         int i;
3118         int ret;
3119         unsigned long size;
3120         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3121         bool is_admin = capable(CAP_SYS_ADMIN);
3122         u16 count;
3123
3124         if (!(file->f_mode & FMODE_READ))
3125                 return -EINVAL;
3126
3127         ret = mnt_want_write_file(file);
3128         if (ret)
3129                 return ret;
3130
3131         if (get_user(count, &argp->dest_count)) {
3132                 ret = -EFAULT;
3133                 goto out;
3134         }
3135
3136         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3137
3138         same = memdup_user(argp, size);
3139
3140         if (IS_ERR(same)) {
3141                 ret = PTR_ERR(same);
3142                 same = NULL;
3143                 goto out;
3144         }
3145
3146         off = same->logical_offset;
3147         len = same->length;
3148
3149         /*
3150          * Limit the total length we will dedupe for each operation.
3151          * This is intended to bound the total time spent in this
3152          * ioctl to something sane.
3153          */
3154         if (len > BTRFS_MAX_DEDUPE_LEN)
3155                 len = BTRFS_MAX_DEDUPE_LEN;
3156
3157         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3158                 /*
3159                  * Btrfs does not support blocksize < page_size. As a
3160                  * result, btrfs_cmp_data() won't correctly handle
3161                  * this situation without an update.
3162                  */
3163                 ret = -EINVAL;
3164                 goto out;
3165         }
3166
3167         ret = -EISDIR;
3168         if (S_ISDIR(src->i_mode))
3169                 goto out;
3170
3171         ret = -EACCES;
3172         if (!S_ISREG(src->i_mode))
3173                 goto out;
3174
3175         /* pre-format output fields to sane values */
3176         for (i = 0; i < count; i++) {
3177                 same->info[i].bytes_deduped = 0ULL;
3178                 same->info[i].status = 0;
3179         }
3180
3181         for (i = 0, info = same->info; i < count; i++, info++) {
3182                 struct inode *dst;
3183                 struct fd dst_file = fdget(info->fd);
3184                 if (!dst_file.file) {
3185                         info->status = -EBADF;
3186                         continue;
3187                 }
3188                 dst = file_inode(dst_file.file);
3189
3190                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3191                         info->status = -EINVAL;
3192                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3193                         info->status = -EXDEV;
3194                 } else if (S_ISDIR(dst->i_mode)) {
3195                         info->status = -EISDIR;
3196                 } else if (!S_ISREG(dst->i_mode)) {
3197                         info->status = -EACCES;
3198                 } else {
3199                         info->status = btrfs_extent_same(src, off, len, dst,
3200                                                         info->logical_offset);
3201                         if (info->status == 0)
3202                                 info->bytes_deduped += len;
3203                 }
3204                 fdput(dst_file);
3205         }
3206
3207         ret = copy_to_user(argp, same, size);
3208         if (ret)
3209                 ret = -EFAULT;
3210
3211 out:
3212         mnt_drop_write_file(file);
3213         kfree(same);
3214         return ret;
3215 }
3216
3217 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3218                                      struct inode *inode,
3219                                      u64 endoff,
3220                                      const u64 destoff,
3221                                      const u64 olen,
3222                                      int no_time_update)
3223 {
3224         struct btrfs_root *root = BTRFS_I(inode)->root;
3225         int ret;
3226
3227         inode_inc_iversion(inode);
3228         if (!no_time_update)
3229                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3230         /*
3231          * We round up to the block size at eof when determining which
3232          * extents to clone above, but shouldn't round up the file size.
3233          */
3234         if (endoff > destoff + olen)
3235                 endoff = destoff + olen;
3236         if (endoff > inode->i_size)
3237                 btrfs_i_size_write(inode, endoff);
3238
3239         ret = btrfs_update_inode(trans, root, inode);
3240         if (ret) {
3241                 btrfs_abort_transaction(trans, root, ret);
3242                 btrfs_end_transaction(trans, root);
3243                 goto out;
3244         }
3245         ret = btrfs_end_transaction(trans, root);
3246 out:
3247         return ret;
3248 }
3249
3250 static void clone_update_extent_map(struct inode *inode,
3251                                     const struct btrfs_trans_handle *trans,
3252                                     const struct btrfs_path *path,
3253                                     const u64 hole_offset,
3254                                     const u64 hole_len)
3255 {
3256         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3257         struct extent_map *em;
3258         int ret;
3259
3260         em = alloc_extent_map();
3261         if (!em) {
3262                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3263                         &BTRFS_I(inode)->runtime_flags);
3264                 return;
3265         }
3266
3267         if (path) {
3268                 struct btrfs_file_extent_item *fi;
3269
3270                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3271                                     struct btrfs_file_extent_item);
3272                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3273                 em->generation = -1;
3274                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3275                     BTRFS_FILE_EXTENT_INLINE)
3276                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3277                                 &BTRFS_I(inode)->runtime_flags);
3278         } else {
3279                 em->start = hole_offset;
3280                 em->len = hole_len;
3281                 em->ram_bytes = em->len;
3282                 em->orig_start = hole_offset;
3283                 em->block_start = EXTENT_MAP_HOLE;
3284                 em->block_len = 0;
3285                 em->orig_block_len = 0;
3286                 em->compress_type = BTRFS_COMPRESS_NONE;
3287                 em->generation = trans->transid;
3288         }
3289
3290         while (1) {
3291                 write_lock(&em_tree->lock);
3292                 ret = add_extent_mapping(em_tree, em, 1);
3293                 write_unlock(&em_tree->lock);
3294                 if (ret != -EEXIST) {
3295                         free_extent_map(em);
3296                         break;
3297                 }
3298                 btrfs_drop_extent_cache(inode, em->start,
3299                                         em->start + em->len - 1, 0);
3300         }
3301
3302         if (ret)
3303                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3304                         &BTRFS_I(inode)->runtime_flags);
3305 }
3306
3307 /*
3308  * Make sure we do not end up inserting an inline extent into a file that has
3309  * already other (non-inline) extents. If a file has an inline extent it can
3310  * not have any other extents and the (single) inline extent must start at the
3311  * file offset 0. Failing to respect these rules will lead to file corruption,
3312  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3313  *
3314  * We can have extents that have been already written to disk or we can have
3315  * dirty ranges still in delalloc, in which case the extent maps and items are
3316  * created only when we run delalloc, and the delalloc ranges might fall outside
3317  * the range we are currently locking in the inode's io tree. So we check the
3318  * inode's i_size because of that (i_size updates are done while holding the
3319  * i_mutex, which we are holding here).
3320  * We also check to see if the inode has a size not greater than "datal" but has
3321  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3322  * protected against such concurrent fallocate calls by the i_mutex).
3323  *
3324  * If the file has no extents but a size greater than datal, do not allow the
3325  * copy because we would need turn the inline extent into a non-inline one (even
3326  * with NO_HOLES enabled). If we find our destination inode only has one inline
3327  * extent, just overwrite it with the source inline extent if its size is less
3328  * than the source extent's size, or we could copy the source inline extent's
3329  * data into the destination inode's inline extent if the later is greater then
3330  * the former.
3331  */
3332 static int clone_copy_inline_extent(struct inode *src,
3333                                     struct inode *dst,
3334                                     struct btrfs_trans_handle *trans,
3335                                     struct btrfs_path *path,
3336                                     struct btrfs_key *new_key,
3337                                     const u64 drop_start,
3338                                     const u64 datal,
3339                                     const u64 skip,
3340                                     const u64 size,
3341                                     char *inline_data)
3342 {
3343         struct btrfs_root *root = BTRFS_I(dst)->root;
3344         const u64 aligned_end = ALIGN(new_key->offset + datal,
3345                                       root->sectorsize);
3346         int ret;
3347         struct btrfs_key key;
3348
3349         if (new_key->offset > 0)
3350                 return -EOPNOTSUPP;
3351
3352         key.objectid = btrfs_ino(dst);
3353         key.type = BTRFS_EXTENT_DATA_KEY;
3354         key.offset = 0;
3355         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3356         if (ret < 0) {
3357                 return ret;
3358         } else if (ret > 0) {
3359                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3360                         ret = btrfs_next_leaf(root, path);
3361                         if (ret < 0)
3362                                 return ret;
3363                         else if (ret > 0)
3364                                 goto copy_inline_extent;
3365                 }
3366                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3367                 if (key.objectid == btrfs_ino(dst) &&
3368                     key.type == BTRFS_EXTENT_DATA_KEY) {
3369                         ASSERT(key.offset > 0);
3370                         return -EOPNOTSUPP;
3371                 }
3372         } else if (i_size_read(dst) <= datal) {
3373                 struct btrfs_file_extent_item *ei;
3374                 u64 ext_len;
3375
3376                 /*
3377                  * If the file size is <= datal, make sure there are no other
3378                  * extents following (can happen do to an fallocate call with
3379                  * the flag FALLOC_FL_KEEP_SIZE).
3380                  */
3381                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3382                                     struct btrfs_file_extent_item);
3383                 /*
3384                  * If it's an inline extent, it can not have other extents
3385                  * following it.
3386                  */
3387                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3388                     BTRFS_FILE_EXTENT_INLINE)
3389                         goto copy_inline_extent;
3390
3391                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3392                 if (ext_len > aligned_end)
3393                         return -EOPNOTSUPP;
3394
3395                 ret = btrfs_next_item(root, path);
3396                 if (ret < 0) {
3397                         return ret;
3398                 } else if (ret == 0) {
3399                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3400                                               path->slots[0]);
3401                         if (key.objectid == btrfs_ino(dst) &&
3402                             key.type == BTRFS_EXTENT_DATA_KEY)
3403                                 return -EOPNOTSUPP;
3404                 }
3405         }
3406
3407 copy_inline_extent:
3408         /*
3409          * We have no extent items, or we have an extent at offset 0 which may
3410          * or may not be inlined. All these cases are dealt the same way.
3411          */
3412         if (i_size_read(dst) > datal) {
3413                 /*
3414                  * If the destination inode has an inline extent...
3415                  * This would require copying the data from the source inline
3416                  * extent into the beginning of the destination's inline extent.
3417                  * But this is really complex, both extents can be compressed
3418                  * or just one of them, which would require decompressing and
3419                  * re-compressing data (which could increase the new compressed
3420                  * size, not allowing the compressed data to fit anymore in an
3421                  * inline extent).
3422                  * So just don't support this case for now (it should be rare,
3423                  * we are not really saving space when cloning inline extents).
3424                  */
3425                 return -EOPNOTSUPP;
3426         }
3427
3428         btrfs_release_path(path);
3429         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3430         if (ret)
3431                 return ret;
3432         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3433         if (ret)
3434                 return ret;
3435
3436         if (skip) {
3437                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3438
3439                 memmove(inline_data + start, inline_data + start + skip, datal);
3440         }
3441
3442         write_extent_buffer(path->nodes[0], inline_data,
3443                             btrfs_item_ptr_offset(path->nodes[0],
3444                                                   path->slots[0]),
3445                             size);
3446         inode_add_bytes(dst, datal);
3447
3448         return 0;
3449 }
3450
3451 /**
3452  * btrfs_clone() - clone a range from inode file to another
3453  *
3454  * @src: Inode to clone from
3455  * @inode: Inode to clone to
3456  * @off: Offset within source to start clone from
3457  * @olen: Original length, passed by user, of range to clone
3458  * @olen_aligned: Block-aligned value of olen
3459  * @destoff: Offset within @inode to start clone
3460  * @no_time_update: Whether to update mtime/ctime on the target inode
3461  */
3462 static int btrfs_clone(struct inode *src, struct inode *inode,
3463                        const u64 off, const u64 olen, const u64 olen_aligned,
3464                        const u64 destoff, int no_time_update)
3465 {
3466         struct btrfs_root *root = BTRFS_I(inode)->root;
3467         struct btrfs_path *path = NULL;
3468         struct extent_buffer *leaf;
3469         struct btrfs_trans_handle *trans;
3470         char *buf = NULL;
3471         struct btrfs_key key;
3472         u32 nritems;
3473         int slot;
3474         int ret;
3475         const u64 len = olen_aligned;
3476         u64 last_dest_end = destoff;
3477
3478         ret = -ENOMEM;
3479         buf = vmalloc(root->nodesize);
3480         if (!buf)
3481                 return ret;
3482
3483         path = btrfs_alloc_path();
3484         if (!path) {
3485                 vfree(buf);
3486                 return ret;
3487         }
3488
3489         path->reada = READA_FORWARD;
3490         /* clone data */
3491         key.objectid = btrfs_ino(src);
3492         key.type = BTRFS_EXTENT_DATA_KEY;
3493         key.offset = off;
3494
3495         while (1) {
3496                 u64 next_key_min_offset = key.offset + 1;
3497
3498                 /*
3499                  * note the key will change type as we walk through the
3500                  * tree.
3501                  */
3502                 path->leave_spinning = 1;
3503                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3504                                 0, 0);
3505                 if (ret < 0)
3506                         goto out;
3507                 /*
3508                  * First search, if no extent item that starts at offset off was
3509                  * found but the previous item is an extent item, it's possible
3510                  * it might overlap our target range, therefore process it.
3511                  */
3512                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3513                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3514                                               path->slots[0] - 1);
3515                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3516                                 path->slots[0]--;
3517                 }
3518
3519                 nritems = btrfs_header_nritems(path->nodes[0]);
3520 process_slot:
3521                 if (path->slots[0] >= nritems) {
3522                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3523                         if (ret < 0)
3524                                 goto out;
3525                         if (ret > 0)
3526                                 break;
3527                         nritems = btrfs_header_nritems(path->nodes[0]);
3528                 }
3529                 leaf = path->nodes[0];
3530                 slot = path->slots[0];
3531
3532                 btrfs_item_key_to_cpu(leaf, &key, slot);
3533                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3534                     key.objectid != btrfs_ino(src))
3535                         break;
3536
3537                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3538                         struct btrfs_file_extent_item *extent;
3539                         int type;
3540                         u32 size;
3541                         struct btrfs_key new_key;
3542                         u64 disko = 0, diskl = 0;
3543                         u64 datao = 0, datal = 0;
3544                         u8 comp;
3545                         u64 drop_start;
3546
3547                         extent = btrfs_item_ptr(leaf, slot,
3548                                                 struct btrfs_file_extent_item);
3549                         comp = btrfs_file_extent_compression(leaf, extent);
3550                         type = btrfs_file_extent_type(leaf, extent);
3551                         if (type == BTRFS_FILE_EXTENT_REG ||
3552                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3553                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3554                                                                       extent);
3555                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3556                                                                  extent);
3557                                 datao = btrfs_file_extent_offset(leaf, extent);
3558                                 datal = btrfs_file_extent_num_bytes(leaf,
3559                                                                     extent);
3560                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3561                                 /* take upper bound, may be compressed */
3562                                 datal = btrfs_file_extent_ram_bytes(leaf,
3563                                                                     extent);
3564                         }
3565
3566                         /*
3567                          * The first search might have left us at an extent
3568                          * item that ends before our target range's start, can
3569                          * happen if we have holes and NO_HOLES feature enabled.
3570                          */
3571                         if (key.offset + datal <= off) {
3572                                 path->slots[0]++;
3573                                 goto process_slot;
3574                         } else if (key.offset >= off + len) {
3575                                 break;
3576                         }
3577                         next_key_min_offset = key.offset + datal;
3578                         size = btrfs_item_size_nr(leaf, slot);
3579                         read_extent_buffer(leaf, buf,
3580                                            btrfs_item_ptr_offset(leaf, slot),
3581                                            size);
3582
3583                         btrfs_release_path(path);
3584                         path->leave_spinning = 0;
3585
3586                         memcpy(&new_key, &key, sizeof(new_key));
3587                         new_key.objectid = btrfs_ino(inode);
3588                         if (off <= key.offset)
3589                                 new_key.offset = key.offset + destoff - off;
3590                         else
3591                                 new_key.offset = destoff;
3592
3593                         /*
3594                          * Deal with a hole that doesn't have an extent item
3595                          * that represents it (NO_HOLES feature enabled).
3596                          * This hole is either in the middle of the cloning
3597                          * range or at the beginning (fully overlaps it or
3598                          * partially overlaps it).
3599                          */
3600                         if (new_key.offset != last_dest_end)
3601                                 drop_start = last_dest_end;
3602                         else
3603                                 drop_start = new_key.offset;
3604
3605                         /*
3606                          * 1 - adjusting old extent (we may have to split it)
3607                          * 1 - add new extent
3608                          * 1 - inode update
3609                          */
3610                         trans = btrfs_start_transaction(root, 3);
3611                         if (IS_ERR(trans)) {
3612                                 ret = PTR_ERR(trans);
3613                                 goto out;
3614                         }
3615
3616                         if (type == BTRFS_FILE_EXTENT_REG ||
3617                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3618                                 /*
3619                                  *    a  | --- range to clone ---|  b
3620                                  * | ------------- extent ------------- |
3621                                  */
3622
3623                                 /* subtract range b */
3624                                 if (key.offset + datal > off + len)
3625                                         datal = off + len - key.offset;
3626
3627                                 /* subtract range a */
3628                                 if (off > key.offset) {
3629                                         datao += off - key.offset;
3630                                         datal -= off - key.offset;
3631                                 }
3632
3633                                 ret = btrfs_drop_extents(trans, root, inode,
3634                                                          drop_start,
3635                                                          new_key.offset + datal,
3636                                                          1);
3637                                 if (ret) {
3638                                         if (ret != -EOPNOTSUPP)
3639                                                 btrfs_abort_transaction(trans,
3640                                                                 root, ret);
3641                                         btrfs_end_transaction(trans, root);
3642                                         goto out;
3643                                 }
3644
3645                                 ret = btrfs_insert_empty_item(trans, root, path,
3646                                                               &new_key, size);
3647                                 if (ret) {
3648                                         btrfs_abort_transaction(trans, root,
3649                                                                 ret);
3650                                         btrfs_end_transaction(trans, root);
3651                                         goto out;
3652                                 }
3653
3654                                 leaf = path->nodes[0];
3655                                 slot = path->slots[0];
3656                                 write_extent_buffer(leaf, buf,
3657                                             btrfs_item_ptr_offset(leaf, slot),
3658                                             size);
3659
3660                                 extent = btrfs_item_ptr(leaf, slot,
3661                                                 struct btrfs_file_extent_item);
3662
3663                                 /* disko == 0 means it's a hole */
3664                                 if (!disko)
3665                                         datao = 0;
3666
3667                                 btrfs_set_file_extent_offset(leaf, extent,
3668                                                              datao);
3669                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3670                                                                 datal);
3671
3672                                 if (disko) {
3673                                         inode_add_bytes(inode, datal);
3674                                         ret = btrfs_inc_extent_ref(trans, root,
3675                                                         disko, diskl, 0,
3676                                                         root->root_key.objectid,
3677                                                         btrfs_ino(inode),
3678                                                         new_key.offset - datao);
3679                                         if (ret) {
3680                                                 btrfs_abort_transaction(trans,
3681                                                                         root,
3682                                                                         ret);
3683                                                 btrfs_end_transaction(trans,
3684                                                                       root);
3685                                                 goto out;
3686
3687                                         }
3688                                 }
3689                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3690                                 u64 skip = 0;
3691                                 u64 trim = 0;
3692
3693                                 if (off > key.offset) {
3694                                         skip = off - key.offset;
3695                                         new_key.offset += skip;
3696                                 }
3697
3698                                 if (key.offset + datal > off + len)
3699                                         trim = key.offset + datal - (off + len);
3700
3701                                 if (comp && (skip || trim)) {
3702                                         ret = -EINVAL;
3703                                         btrfs_end_transaction(trans, root);
3704                                         goto out;
3705                                 }
3706                                 size -= skip + trim;
3707                                 datal -= skip + trim;
3708
3709                                 ret = clone_copy_inline_extent(src, inode,
3710                                                                trans, path,
3711                                                                &new_key,
3712                                                                drop_start,
3713                                                                datal,
3714                                                                skip, size, buf);
3715                                 if (ret) {
3716                                         if (ret != -EOPNOTSUPP)
3717                                                 btrfs_abort_transaction(trans,
3718                                                                         root,
3719                                                                         ret);
3720                                         btrfs_end_transaction(trans, root);
3721                                         goto out;
3722                                 }
3723                                 leaf = path->nodes[0];
3724                                 slot = path->slots[0];
3725                         }
3726
3727                         /* If we have an implicit hole (NO_HOLES feature). */
3728                         if (drop_start < new_key.offset)
3729                                 clone_update_extent_map(inode, trans,
3730                                                 NULL, drop_start,
3731                                                 new_key.offset - drop_start);
3732
3733                         clone_update_extent_map(inode, trans, path, 0, 0);
3734
3735                         btrfs_mark_buffer_dirty(leaf);
3736                         btrfs_release_path(path);
3737
3738                         last_dest_end = ALIGN(new_key.offset + datal,
3739                                               root->sectorsize);
3740                         ret = clone_finish_inode_update(trans, inode,
3741                                                         last_dest_end,
3742                                                         destoff, olen,
3743                                                         no_time_update);
3744                         if (ret)
3745                                 goto out;
3746                         if (new_key.offset + datal >= destoff + len)
3747                                 break;
3748                 }
3749                 btrfs_release_path(path);
3750                 key.offset = next_key_min_offset;
3751         }
3752         ret = 0;
3753
3754         if (last_dest_end < destoff + len) {
3755                 /*
3756                  * We have an implicit hole (NO_HOLES feature is enabled) that
3757                  * fully or partially overlaps our cloning range at its end.
3758                  */
3759                 btrfs_release_path(path);
3760
3761                 /*
3762                  * 1 - remove extent(s)
3763                  * 1 - inode update
3764                  */
3765                 trans = btrfs_start_transaction(root, 2);
3766                 if (IS_ERR(trans)) {
3767                         ret = PTR_ERR(trans);
3768                         goto out;
3769                 }
3770                 ret = btrfs_drop_extents(trans, root, inode,
3771                                          last_dest_end, destoff + len, 1);
3772                 if (ret) {
3773                         if (ret != -EOPNOTSUPP)
3774                                 btrfs_abort_transaction(trans, root, ret);
3775                         btrfs_end_transaction(trans, root);
3776                         goto out;
3777                 }
3778                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3779                                         destoff + len - last_dest_end);
3780                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3781                                                 destoff, olen, no_time_update);
3782         }
3783
3784 out:
3785         btrfs_free_path(path);
3786         vfree(buf);
3787         return ret;
3788 }
3789
3790 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3791                                        u64 off, u64 olen, u64 destoff)
3792 {
3793         struct inode *inode = file_inode(file);
3794         struct btrfs_root *root = BTRFS_I(inode)->root;
3795         struct fd src_file;
3796         struct inode *src;
3797         int ret;
3798         u64 len = olen;
3799         u64 bs = root->fs_info->sb->s_blocksize;
3800         int same_inode = 0;
3801
3802         /*
3803          * TODO:
3804          * - split compressed inline extents.  annoying: we need to
3805          *   decompress into destination's address_space (the file offset
3806          *   may change, so source mapping won't do), then recompress (or
3807          *   otherwise reinsert) a subrange.
3808          *
3809          * - split destination inode's inline extents.  The inline extents can
3810          *   be either compressed or non-compressed.
3811          */
3812
3813         /* the destination must be opened for writing */
3814         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3815                 return -EINVAL;
3816
3817         if (btrfs_root_readonly(root))
3818                 return -EROFS;
3819
3820         ret = mnt_want_write_file(file);
3821         if (ret)
3822                 return ret;
3823
3824         src_file = fdget(srcfd);
3825         if (!src_file.file) {
3826                 ret = -EBADF;
3827                 goto out_drop_write;
3828         }
3829
3830         ret = -EXDEV;
3831         if (src_file.file->f_path.mnt != file->f_path.mnt)
3832                 goto out_fput;
3833
3834         src = file_inode(src_file.file);
3835
3836         ret = -EINVAL;
3837         if (src == inode)
3838                 same_inode = 1;
3839
3840         /* the src must be open for reading */
3841         if (!(src_file.file->f_mode & FMODE_READ))
3842                 goto out_fput;
3843
3844         /* don't make the dst file partly checksummed */
3845         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3846             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3847                 goto out_fput;
3848
3849         ret = -EISDIR;
3850         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3851                 goto out_fput;
3852
3853         ret = -EXDEV;
3854         if (src->i_sb != inode->i_sb)
3855                 goto out_fput;
3856
3857         if (!same_inode) {
3858                 btrfs_double_inode_lock(src, inode);
3859         } else {
3860                 mutex_lock(&src->i_mutex);
3861         }
3862
3863         /* determine range to clone */
3864         ret = -EINVAL;
3865         if (off + len > src->i_size || off + len < off)
3866                 goto out_unlock;
3867         if (len == 0)
3868                 olen = len = src->i_size - off;
3869         /* if we extend to eof, continue to block boundary */
3870         if (off + len == src->i_size)
3871                 len = ALIGN(src->i_size, bs) - off;
3872
3873         if (len == 0) {
3874                 ret = 0;
3875                 goto out_unlock;
3876         }
3877
3878         /* verify the end result is block aligned */
3879         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3880             !IS_ALIGNED(destoff, bs))
3881                 goto out_unlock;
3882
3883         /* verify if ranges are overlapped within the same file */
3884         if (same_inode) {
3885                 if (destoff + len > off && destoff < off + len)
3886                         goto out_unlock;
3887         }
3888
3889         if (destoff > inode->i_size) {
3890                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3891                 if (ret)
3892                         goto out_unlock;
3893         }
3894
3895         /*
3896          * Lock the target range too. Right after we replace the file extent
3897          * items in the fs tree (which now point to the cloned data), we might
3898          * have a worker replace them with extent items relative to a write
3899          * operation that was issued before this clone operation (i.e. confront
3900          * with inode.c:btrfs_finish_ordered_io).
3901          */
3902         if (same_inode) {
3903                 u64 lock_start = min_t(u64, off, destoff);
3904                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3905
3906                 lock_extent_range(src, lock_start, lock_len);
3907         } else {
3908                 btrfs_double_extent_lock(src, off, inode, destoff, len);
3909         }
3910
3911         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3912
3913         if (same_inode) {
3914                 u64 lock_start = min_t(u64, off, destoff);
3915                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3916
3917                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3918         } else {
3919                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3920         }
3921         /*
3922          * Truncate page cache pages so that future reads will see the cloned
3923          * data immediately and not the previous data.
3924          */
3925         truncate_inode_pages_range(&inode->i_data, destoff,
3926                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3927 out_unlock:
3928         if (!same_inode)
3929                 btrfs_double_inode_unlock(src, inode);
3930         else
3931                 mutex_unlock(&src->i_mutex);
3932 out_fput:
3933         fdput(src_file);
3934 out_drop_write:
3935         mnt_drop_write_file(file);
3936         return ret;
3937 }
3938
3939 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3940 {
3941         struct btrfs_ioctl_clone_range_args args;
3942
3943         if (copy_from_user(&args, argp, sizeof(args)))
3944                 return -EFAULT;
3945         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3946                                  args.src_length, args.dest_offset);
3947 }
3948
3949 /*
3950  * there are many ways the trans_start and trans_end ioctls can lead
3951  * to deadlocks.  They should only be used by applications that
3952  * basically own the machine, and have a very in depth understanding
3953  * of all the possible deadlocks and enospc problems.
3954  */
3955 static long btrfs_ioctl_trans_start(struct file *file)
3956 {
3957         struct inode *inode = file_inode(file);
3958         struct btrfs_root *root = BTRFS_I(inode)->root;
3959         struct btrfs_trans_handle *trans;
3960         int ret;
3961
3962         ret = -EPERM;
3963         if (!capable(CAP_SYS_ADMIN))
3964                 goto out;
3965
3966         ret = -EINPROGRESS;
3967         if (file->private_data)
3968                 goto out;
3969
3970         ret = -EROFS;
3971         if (btrfs_root_readonly(root))
3972                 goto out;
3973
3974         ret = mnt_want_write_file(file);
3975         if (ret)
3976                 goto out;
3977
3978         atomic_inc(&root->fs_info->open_ioctl_trans);
3979
3980         ret = -ENOMEM;
3981         trans = btrfs_start_ioctl_transaction(root);
3982         if (IS_ERR(trans))
3983                 goto out_drop;
3984
3985         file->private_data = trans;
3986         return 0;
3987
3988 out_drop:
3989         atomic_dec(&root->fs_info->open_ioctl_trans);
3990         mnt_drop_write_file(file);
3991 out:
3992         return ret;
3993 }
3994
3995 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3996 {
3997         struct inode *inode = file_inode(file);
3998         struct btrfs_root *root = BTRFS_I(inode)->root;
3999         struct btrfs_root *new_root;
4000         struct btrfs_dir_item *di;
4001         struct btrfs_trans_handle *trans;
4002         struct btrfs_path *path;
4003         struct btrfs_key location;
4004         struct btrfs_disk_key disk_key;
4005         u64 objectid = 0;
4006         u64 dir_id;
4007         int ret;
4008
4009         if (!capable(CAP_SYS_ADMIN))
4010                 return -EPERM;
4011
4012         ret = mnt_want_write_file(file);
4013         if (ret)
4014                 return ret;
4015
4016         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4017                 ret = -EFAULT;
4018                 goto out;
4019         }
4020
4021         if (!objectid)
4022                 objectid = BTRFS_FS_TREE_OBJECTID;
4023
4024         location.objectid = objectid;
4025         location.type = BTRFS_ROOT_ITEM_KEY;
4026         location.offset = (u64)-1;
4027
4028         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4029         if (IS_ERR(new_root)) {
4030                 ret = PTR_ERR(new_root);
4031                 goto out;
4032         }
4033
4034         path = btrfs_alloc_path();
4035         if (!path) {
4036                 ret = -ENOMEM;
4037                 goto out;
4038         }
4039         path->leave_spinning = 1;
4040
4041         trans = btrfs_start_transaction(root, 1);
4042         if (IS_ERR(trans)) {
4043                 btrfs_free_path(path);
4044                 ret = PTR_ERR(trans);
4045                 goto out;
4046         }
4047
4048         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4049         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4050                                    dir_id, "default", 7, 1);
4051         if (IS_ERR_OR_NULL(di)) {
4052                 btrfs_free_path(path);
4053                 btrfs_end_transaction(trans, root);
4054                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4055                            "item, this isn't going to work");
4056                 ret = -ENOENT;
4057                 goto out;
4058         }
4059
4060         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4061         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4062         btrfs_mark_buffer_dirty(path->nodes[0]);
4063         btrfs_free_path(path);
4064
4065         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4066         btrfs_end_transaction(trans, root);
4067 out:
4068         mnt_drop_write_file(file);
4069         return ret;
4070 }
4071
4072 void btrfs_get_block_group_info(struct list_head *groups_list,
4073                                 struct btrfs_ioctl_space_info *space)
4074 {
4075         struct btrfs_block_group_cache *block_group;
4076
4077         space->total_bytes = 0;
4078         space->used_bytes = 0;
4079         space->flags = 0;
4080         list_for_each_entry(block_group, groups_list, list) {
4081                 space->flags = block_group->flags;
4082                 space->total_bytes += block_group->key.offset;
4083                 space->used_bytes +=
4084                         btrfs_block_group_used(&block_group->item);
4085         }
4086 }
4087
4088 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4089 {
4090         struct btrfs_ioctl_space_args space_args;
4091         struct btrfs_ioctl_space_info space;
4092         struct btrfs_ioctl_space_info *dest;
4093         struct btrfs_ioctl_space_info *dest_orig;
4094         struct btrfs_ioctl_space_info __user *user_dest;
4095         struct btrfs_space_info *info;
4096         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4097                        BTRFS_BLOCK_GROUP_SYSTEM,
4098                        BTRFS_BLOCK_GROUP_METADATA,
4099                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4100         int num_types = 4;
4101         int alloc_size;
4102         int ret = 0;
4103         u64 slot_count = 0;
4104         int i, c;
4105
4106         if (copy_from_user(&space_args,
4107                            (struct btrfs_ioctl_space_args __user *)arg,
4108                            sizeof(space_args)))
4109                 return -EFAULT;
4110
4111         for (i = 0; i < num_types; i++) {
4112                 struct btrfs_space_info *tmp;
4113
4114                 info = NULL;
4115                 rcu_read_lock();
4116                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4117                                         list) {
4118                         if (tmp->flags == types[i]) {
4119                                 info = tmp;
4120                                 break;
4121                         }
4122                 }
4123                 rcu_read_unlock();
4124
4125                 if (!info)
4126                         continue;
4127
4128                 down_read(&info->groups_sem);
4129                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4130                         if (!list_empty(&info->block_groups[c]))
4131                                 slot_count++;
4132                 }
4133                 up_read(&info->groups_sem);
4134         }
4135
4136         /*
4137          * Global block reserve, exported as a space_info
4138          */
4139         slot_count++;
4140
4141         /* space_slots == 0 means they are asking for a count */
4142         if (space_args.space_slots == 0) {
4143                 space_args.total_spaces = slot_count;
4144                 goto out;
4145         }
4146
4147         slot_count = min_t(u64, space_args.space_slots, slot_count);
4148
4149         alloc_size = sizeof(*dest) * slot_count;
4150
4151         /* we generally have at most 6 or so space infos, one for each raid
4152          * level.  So, a whole page should be more than enough for everyone
4153          */
4154         if (alloc_size > PAGE_CACHE_SIZE)
4155                 return -ENOMEM;
4156
4157         space_args.total_spaces = 0;
4158         dest = kmalloc(alloc_size, GFP_KERNEL);
4159         if (!dest)
4160                 return -ENOMEM;
4161         dest_orig = dest;
4162
4163         /* now we have a buffer to copy into */
4164         for (i = 0; i < num_types; i++) {
4165                 struct btrfs_space_info *tmp;
4166
4167                 if (!slot_count)
4168                         break;
4169
4170                 info = NULL;
4171                 rcu_read_lock();
4172                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4173                                         list) {
4174                         if (tmp->flags == types[i]) {
4175                                 info = tmp;
4176                                 break;
4177                         }
4178                 }
4179                 rcu_read_unlock();
4180
4181                 if (!info)
4182                         continue;
4183                 down_read(&info->groups_sem);
4184                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4185                         if (!list_empty(&info->block_groups[c])) {
4186                                 btrfs_get_block_group_info(
4187                                         &info->block_groups[c], &space);
4188                                 memcpy(dest, &space, sizeof(space));
4189                                 dest++;
4190                                 space_args.total_spaces++;
4191                                 slot_count--;
4192                         }
4193                         if (!slot_count)
4194                                 break;
4195                 }
4196                 up_read(&info->groups_sem);
4197         }
4198
4199         /*
4200          * Add global block reserve
4201          */
4202         if (slot_count) {
4203                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4204
4205                 spin_lock(&block_rsv->lock);
4206                 space.total_bytes = block_rsv->size;
4207                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4208                 spin_unlock(&block_rsv->lock);
4209                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4210                 memcpy(dest, &space, sizeof(space));
4211                 space_args.total_spaces++;
4212         }
4213
4214         user_dest = (struct btrfs_ioctl_space_info __user *)
4215                 (arg + sizeof(struct btrfs_ioctl_space_args));
4216
4217         if (copy_to_user(user_dest, dest_orig, alloc_size))
4218                 ret = -EFAULT;
4219
4220         kfree(dest_orig);
4221 out:
4222         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4223                 ret = -EFAULT;
4224
4225         return ret;
4226 }
4227
4228 /*
4229  * there are many ways the trans_start and trans_end ioctls can lead
4230  * to deadlocks.  They should only be used by applications that
4231  * basically own the machine, and have a very in depth understanding
4232  * of all the possible deadlocks and enospc problems.
4233  */
4234 long btrfs_ioctl_trans_end(struct file *file)
4235 {
4236         struct inode *inode = file_inode(file);
4237         struct btrfs_root *root = BTRFS_I(inode)->root;
4238         struct btrfs_trans_handle *trans;
4239
4240         trans = file->private_data;
4241         if (!trans)
4242                 return -EINVAL;
4243         file->private_data = NULL;
4244
4245         btrfs_end_transaction(trans, root);
4246
4247         atomic_dec(&root->fs_info->open_ioctl_trans);
4248
4249         mnt_drop_write_file(file);
4250         return 0;
4251 }
4252
4253 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4254                                             void __user *argp)
4255 {
4256         struct btrfs_trans_handle *trans;
4257         u64 transid;
4258         int ret;
4259
4260         trans = btrfs_attach_transaction_barrier(root);
4261         if (IS_ERR(trans)) {
4262                 if (PTR_ERR(trans) != -ENOENT)
4263                         return PTR_ERR(trans);
4264
4265                 /* No running transaction, don't bother */
4266                 transid = root->fs_info->last_trans_committed;
4267                 goto out;
4268         }
4269         transid = trans->transid;
4270         ret = btrfs_commit_transaction_async(trans, root, 0);
4271         if (ret) {
4272                 btrfs_end_transaction(trans, root);
4273                 return ret;
4274         }
4275 out:
4276         if (argp)
4277                 if (copy_to_user(argp, &transid, sizeof(transid)))
4278                         return -EFAULT;
4279         return 0;
4280 }
4281
4282 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4283                                            void __user *argp)
4284 {
4285         u64 transid;
4286
4287         if (argp) {
4288                 if (copy_from_user(&transid, argp, sizeof(transid)))
4289                         return -EFAULT;
4290         } else {
4291                 transid = 0;  /* current trans */
4292         }
4293         return btrfs_wait_for_commit(root, transid);
4294 }
4295
4296 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4297 {
4298         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4299         struct btrfs_ioctl_scrub_args *sa;
4300         int ret;
4301
4302         if (!capable(CAP_SYS_ADMIN))
4303                 return -EPERM;
4304
4305         sa = memdup_user(arg, sizeof(*sa));
4306         if (IS_ERR(sa))
4307                 return PTR_ERR(sa);
4308
4309         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4310                 ret = mnt_want_write_file(file);
4311                 if (ret)
4312                         goto out;
4313         }
4314
4315         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4316                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4317                               0);
4318
4319         if (copy_to_user(arg, sa, sizeof(*sa)))
4320                 ret = -EFAULT;
4321
4322         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4323                 mnt_drop_write_file(file);
4324 out:
4325         kfree(sa);
4326         return ret;
4327 }
4328
4329 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4330 {
4331         if (!capable(CAP_SYS_ADMIN))
4332                 return -EPERM;
4333
4334         return btrfs_scrub_cancel(root->fs_info);
4335 }
4336
4337 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4338                                        void __user *arg)
4339 {
4340         struct btrfs_ioctl_scrub_args *sa;
4341         int ret;
4342
4343         if (!capable(CAP_SYS_ADMIN))
4344                 return -EPERM;
4345
4346         sa = memdup_user(arg, sizeof(*sa));
4347         if (IS_ERR(sa))
4348                 return PTR_ERR(sa);
4349
4350         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4351
4352         if (copy_to_user(arg, sa, sizeof(*sa)))
4353                 ret = -EFAULT;
4354
4355         kfree(sa);
4356         return ret;
4357 }
4358
4359 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4360                                       void __user *arg)
4361 {
4362         struct btrfs_ioctl_get_dev_stats *sa;
4363         int ret;
4364
4365         sa = memdup_user(arg, sizeof(*sa));
4366         if (IS_ERR(sa))
4367                 return PTR_ERR(sa);
4368
4369         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4370                 kfree(sa);
4371                 return -EPERM;
4372         }
4373
4374         ret = btrfs_get_dev_stats(root, sa);
4375
4376         if (copy_to_user(arg, sa, sizeof(*sa)))
4377                 ret = -EFAULT;
4378
4379         kfree(sa);
4380         return ret;
4381 }
4382
4383 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4384 {
4385         struct btrfs_ioctl_dev_replace_args *p;
4386         int ret;
4387
4388         if (!capable(CAP_SYS_ADMIN))
4389                 return -EPERM;
4390
4391         p = memdup_user(arg, sizeof(*p));
4392         if (IS_ERR(p))
4393                 return PTR_ERR(p);
4394
4395         switch (p->cmd) {
4396         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4397                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4398                         ret = -EROFS;
4399                         goto out;
4400                 }
4401                 if (atomic_xchg(
4402                         &root->fs_info->mutually_exclusive_operation_running,
4403                         1)) {
4404                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4405                 } else {
4406                         ret = btrfs_dev_replace_start(root, p);
4407                         atomic_set(
4408                          &root->fs_info->mutually_exclusive_operation_running,
4409                          0);
4410                 }
4411                 break;
4412         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4413                 btrfs_dev_replace_status(root->fs_info, p);
4414                 ret = 0;
4415                 break;
4416         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4417                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4418                 break;
4419         default:
4420                 ret = -EINVAL;
4421                 break;
4422         }
4423
4424         if (copy_to_user(arg, p, sizeof(*p)))
4425                 ret = -EFAULT;
4426 out:
4427         kfree(p);
4428         return ret;
4429 }
4430
4431 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4432 {
4433         int ret = 0;
4434         int i;
4435         u64 rel_ptr;
4436         int size;
4437         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4438         struct inode_fs_paths *ipath = NULL;
4439         struct btrfs_path *path;
4440
4441         if (!capable(CAP_DAC_READ_SEARCH))
4442                 return -EPERM;
4443
4444         path = btrfs_alloc_path();
4445         if (!path) {
4446                 ret = -ENOMEM;
4447                 goto out;
4448         }
4449
4450         ipa = memdup_user(arg, sizeof(*ipa));
4451         if (IS_ERR(ipa)) {
4452                 ret = PTR_ERR(ipa);
4453                 ipa = NULL;
4454                 goto out;
4455         }
4456
4457         size = min_t(u32, ipa->size, 4096);
4458         ipath = init_ipath(size, root, path);
4459         if (IS_ERR(ipath)) {
4460                 ret = PTR_ERR(ipath);
4461                 ipath = NULL;
4462                 goto out;
4463         }
4464
4465         ret = paths_from_inode(ipa->inum, ipath);
4466         if (ret < 0)
4467                 goto out;
4468
4469         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4470                 rel_ptr = ipath->fspath->val[i] -
4471                           (u64)(unsigned long)ipath->fspath->val;
4472                 ipath->fspath->val[i] = rel_ptr;
4473         }
4474
4475         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4476                            (void *)(unsigned long)ipath->fspath, size);
4477         if (ret) {
4478                 ret = -EFAULT;
4479                 goto out;
4480         }
4481
4482 out:
4483         btrfs_free_path(path);
4484         free_ipath(ipath);
4485         kfree(ipa);
4486
4487         return ret;
4488 }
4489
4490 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4491 {
4492         struct btrfs_data_container *inodes = ctx;
4493         const size_t c = 3 * sizeof(u64);
4494
4495         if (inodes->bytes_left >= c) {
4496                 inodes->bytes_left -= c;
4497                 inodes->val[inodes->elem_cnt] = inum;
4498                 inodes->val[inodes->elem_cnt + 1] = offset;
4499                 inodes->val[inodes->elem_cnt + 2] = root;
4500                 inodes->elem_cnt += 3;
4501         } else {
4502                 inodes->bytes_missing += c - inodes->bytes_left;
4503                 inodes->bytes_left = 0;
4504                 inodes->elem_missed += 3;
4505         }
4506
4507         return 0;
4508 }
4509
4510 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4511                                         void __user *arg)
4512 {
4513         int ret = 0;
4514         int size;
4515         struct btrfs_ioctl_logical_ino_args *loi;
4516         struct btrfs_data_container *inodes = NULL;
4517         struct btrfs_path *path = NULL;
4518
4519         if (!capable(CAP_SYS_ADMIN))
4520                 return -EPERM;
4521
4522         loi = memdup_user(arg, sizeof(*loi));
4523         if (IS_ERR(loi)) {
4524                 ret = PTR_ERR(loi);
4525                 loi = NULL;
4526                 goto out;
4527         }
4528
4529         path = btrfs_alloc_path();
4530         if (!path) {
4531                 ret = -ENOMEM;
4532                 goto out;
4533         }
4534
4535         size = min_t(u32, loi->size, SZ_64K);
4536         inodes = init_data_container(size);
4537         if (IS_ERR(inodes)) {
4538                 ret = PTR_ERR(inodes);
4539                 inodes = NULL;
4540                 goto out;
4541         }
4542
4543         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4544                                           build_ino_list, inodes);
4545         if (ret == -EINVAL)
4546                 ret = -ENOENT;
4547         if (ret < 0)
4548                 goto out;
4549
4550         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4551                            (void *)(unsigned long)inodes, size);
4552         if (ret)
4553                 ret = -EFAULT;
4554
4555 out:
4556         btrfs_free_path(path);
4557         vfree(inodes);
4558         kfree(loi);
4559
4560         return ret;
4561 }
4562
4563 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4564                                struct btrfs_ioctl_balance_args *bargs)
4565 {
4566         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4567
4568         bargs->flags = bctl->flags;
4569
4570         if (atomic_read(&fs_info->balance_running))
4571                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4572         if (atomic_read(&fs_info->balance_pause_req))
4573                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4574         if (atomic_read(&fs_info->balance_cancel_req))
4575                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4576
4577         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4578         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4579         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4580
4581         if (lock) {
4582                 spin_lock(&fs_info->balance_lock);
4583                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4584                 spin_unlock(&fs_info->balance_lock);
4585         } else {
4586                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4587         }
4588 }
4589
4590 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4591 {
4592         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4593         struct btrfs_fs_info *fs_info = root->fs_info;
4594         struct btrfs_ioctl_balance_args *bargs;
4595         struct btrfs_balance_control *bctl;
4596         bool need_unlock; /* for mut. excl. ops lock */
4597         int ret;
4598
4599         if (!capable(CAP_SYS_ADMIN))
4600                 return -EPERM;
4601
4602         ret = mnt_want_write_file(file);
4603         if (ret)
4604                 return ret;
4605
4606 again:
4607         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4608                 mutex_lock(&fs_info->volume_mutex);
4609                 mutex_lock(&fs_info->balance_mutex);
4610                 need_unlock = true;
4611                 goto locked;
4612         }
4613
4614         /*
4615          * mut. excl. ops lock is locked.  Three possibilites:
4616          *   (1) some other op is running
4617          *   (2) balance is running
4618          *   (3) balance is paused -- special case (think resume)
4619          */
4620         mutex_lock(&fs_info->balance_mutex);
4621         if (fs_info->balance_ctl) {
4622                 /* this is either (2) or (3) */
4623                 if (!atomic_read(&fs_info->balance_running)) {
4624                         mutex_unlock(&fs_info->balance_mutex);
4625                         if (!mutex_trylock(&fs_info->volume_mutex))
4626                                 goto again;
4627                         mutex_lock(&fs_info->balance_mutex);
4628
4629                         if (fs_info->balance_ctl &&
4630                             !atomic_read(&fs_info->balance_running)) {
4631                                 /* this is (3) */
4632                                 need_unlock = false;
4633                                 goto locked;
4634                         }
4635
4636                         mutex_unlock(&fs_info->balance_mutex);
4637                         mutex_unlock(&fs_info->volume_mutex);
4638                         goto again;
4639                 } else {
4640                         /* this is (2) */
4641                         mutex_unlock(&fs_info->balance_mutex);
4642                         ret = -EINPROGRESS;
4643                         goto out;
4644                 }
4645         } else {
4646                 /* this is (1) */
4647                 mutex_unlock(&fs_info->balance_mutex);
4648                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4649                 goto out;
4650         }
4651
4652 locked:
4653         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4654
4655         if (arg) {
4656                 bargs = memdup_user(arg, sizeof(*bargs));
4657                 if (IS_ERR(bargs)) {
4658                         ret = PTR_ERR(bargs);
4659                         goto out_unlock;
4660                 }
4661
4662                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4663                         if (!fs_info->balance_ctl) {
4664                                 ret = -ENOTCONN;
4665                                 goto out_bargs;
4666                         }
4667
4668                         bctl = fs_info->balance_ctl;
4669                         spin_lock(&fs_info->balance_lock);
4670                         bctl->flags |= BTRFS_BALANCE_RESUME;
4671                         spin_unlock(&fs_info->balance_lock);
4672
4673                         goto do_balance;
4674                 }
4675         } else {
4676                 bargs = NULL;
4677         }
4678
4679         if (fs_info->balance_ctl) {
4680                 ret = -EINPROGRESS;
4681                 goto out_bargs;
4682         }
4683
4684         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4685         if (!bctl) {
4686                 ret = -ENOMEM;
4687                 goto out_bargs;
4688         }
4689
4690         bctl->fs_info = fs_info;
4691         if (arg) {
4692                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4693                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4694                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4695
4696                 bctl->flags = bargs->flags;
4697         } else {
4698                 /* balance everything - no filters */
4699                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4700         }
4701
4702         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4703                 ret = -EINVAL;
4704                 goto out_bctl;
4705         }
4706
4707 do_balance:
4708         /*
4709          * Ownership of bctl and mutually_exclusive_operation_running
4710          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4711          * or, if restriper was paused all the way until unmount, in
4712          * free_fs_info.  mutually_exclusive_operation_running is
4713          * cleared in __cancel_balance.
4714          */
4715         need_unlock = false;
4716
4717         ret = btrfs_balance(bctl, bargs);
4718         bctl = NULL;
4719
4720         if (arg) {
4721                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4722                         ret = -EFAULT;
4723         }
4724
4725 out_bctl:
4726         kfree(bctl);
4727 out_bargs:
4728         kfree(bargs);
4729 out_unlock:
4730         mutex_unlock(&fs_info->balance_mutex);
4731         mutex_unlock(&fs_info->volume_mutex);
4732         if (need_unlock)
4733                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4734 out:
4735         mnt_drop_write_file(file);
4736         return ret;
4737 }
4738
4739 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4740 {
4741         if (!capable(CAP_SYS_ADMIN))
4742                 return -EPERM;
4743
4744         switch (cmd) {
4745         case BTRFS_BALANCE_CTL_PAUSE:
4746                 return btrfs_pause_balance(root->fs_info);
4747         case BTRFS_BALANCE_CTL_CANCEL:
4748                 return btrfs_cancel_balance(root->fs_info);
4749         }
4750
4751         return -EINVAL;
4752 }
4753
4754 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4755                                          void __user *arg)
4756 {
4757         struct btrfs_fs_info *fs_info = root->fs_info;
4758         struct btrfs_ioctl_balance_args *bargs;
4759         int ret = 0;
4760
4761         if (!capable(CAP_SYS_ADMIN))
4762                 return -EPERM;
4763
4764         mutex_lock(&fs_info->balance_mutex);
4765         if (!fs_info->balance_ctl) {
4766                 ret = -ENOTCONN;
4767                 goto out;
4768         }
4769
4770         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4771         if (!bargs) {
4772                 ret = -ENOMEM;
4773                 goto out;
4774         }
4775
4776         update_ioctl_balance_args(fs_info, 1, bargs);
4777
4778         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4779                 ret = -EFAULT;
4780
4781         kfree(bargs);
4782 out:
4783         mutex_unlock(&fs_info->balance_mutex);
4784         return ret;
4785 }
4786
4787 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4788 {
4789         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4790         struct btrfs_ioctl_quota_ctl_args *sa;
4791         struct btrfs_trans_handle *trans = NULL;
4792         int ret;
4793         int err;
4794
4795         if (!capable(CAP_SYS_ADMIN))
4796                 return -EPERM;
4797
4798         ret = mnt_want_write_file(file);
4799         if (ret)
4800                 return ret;
4801
4802         sa = memdup_user(arg, sizeof(*sa));
4803         if (IS_ERR(sa)) {
4804                 ret = PTR_ERR(sa);
4805                 goto drop_write;
4806         }
4807
4808         down_write(&root->fs_info->subvol_sem);
4809         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4810         if (IS_ERR(trans)) {
4811                 ret = PTR_ERR(trans);
4812                 goto out;
4813         }
4814
4815         switch (sa->cmd) {
4816         case BTRFS_QUOTA_CTL_ENABLE:
4817                 ret = btrfs_quota_enable(trans, root->fs_info);
4818                 break;
4819         case BTRFS_QUOTA_CTL_DISABLE:
4820                 ret = btrfs_quota_disable(trans, root->fs_info);
4821                 break;
4822         default:
4823                 ret = -EINVAL;
4824                 break;
4825         }
4826
4827         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4828         if (err && !ret)
4829                 ret = err;
4830 out:
4831         kfree(sa);
4832         up_write(&root->fs_info->subvol_sem);
4833 drop_write:
4834         mnt_drop_write_file(file);
4835         return ret;
4836 }
4837
4838 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4839 {
4840         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4841         struct btrfs_ioctl_qgroup_assign_args *sa;
4842         struct btrfs_trans_handle *trans;
4843         int ret;
4844         int err;
4845
4846         if (!capable(CAP_SYS_ADMIN))
4847                 return -EPERM;
4848
4849         ret = mnt_want_write_file(file);
4850         if (ret)
4851                 return ret;
4852
4853         sa = memdup_user(arg, sizeof(*sa));
4854         if (IS_ERR(sa)) {
4855                 ret = PTR_ERR(sa);
4856                 goto drop_write;
4857         }
4858
4859         trans = btrfs_join_transaction(root);
4860         if (IS_ERR(trans)) {
4861                 ret = PTR_ERR(trans);
4862                 goto out;
4863         }
4864
4865         /* FIXME: check if the IDs really exist */
4866         if (sa->assign) {
4867                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4868                                                 sa->src, sa->dst);
4869         } else {
4870                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4871                                                 sa->src, sa->dst);
4872         }
4873
4874         /* update qgroup status and info */
4875         err = btrfs_run_qgroups(trans, root->fs_info);
4876         if (err < 0)
4877                 btrfs_std_error(root->fs_info, ret,
4878                             "failed to update qgroup status and info\n");
4879         err = btrfs_end_transaction(trans, root);
4880         if (err && !ret)
4881                 ret = err;
4882
4883 out:
4884         kfree(sa);
4885 drop_write:
4886         mnt_drop_write_file(file);
4887         return ret;
4888 }
4889
4890 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4891 {
4892         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4893         struct btrfs_ioctl_qgroup_create_args *sa;
4894         struct btrfs_trans_handle *trans;
4895         int ret;
4896         int err;
4897
4898         if (!capable(CAP_SYS_ADMIN))
4899                 return -EPERM;
4900
4901         ret = mnt_want_write_file(file);
4902         if (ret)
4903                 return ret;
4904
4905         sa = memdup_user(arg, sizeof(*sa));
4906         if (IS_ERR(sa)) {
4907                 ret = PTR_ERR(sa);
4908                 goto drop_write;
4909         }
4910
4911         if (!sa->qgroupid) {
4912                 ret = -EINVAL;
4913                 goto out;
4914         }
4915
4916         trans = btrfs_join_transaction(root);
4917         if (IS_ERR(trans)) {
4918                 ret = PTR_ERR(trans);
4919                 goto out;
4920         }
4921
4922         /* FIXME: check if the IDs really exist */
4923         if (sa->create) {
4924                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4925         } else {
4926                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4927         }
4928
4929         err = btrfs_end_transaction(trans, root);
4930         if (err && !ret)
4931                 ret = err;
4932
4933 out:
4934         kfree(sa);
4935 drop_write:
4936         mnt_drop_write_file(file);
4937         return ret;
4938 }
4939
4940 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4941 {
4942         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4943         struct btrfs_ioctl_qgroup_limit_args *sa;
4944         struct btrfs_trans_handle *trans;
4945         int ret;
4946         int err;
4947         u64 qgroupid;
4948
4949         if (!capable(CAP_SYS_ADMIN))
4950                 return -EPERM;
4951
4952         ret = mnt_want_write_file(file);
4953         if (ret)
4954                 return ret;
4955
4956         sa = memdup_user(arg, sizeof(*sa));
4957         if (IS_ERR(sa)) {
4958                 ret = PTR_ERR(sa);
4959                 goto drop_write;
4960         }
4961
4962         trans = btrfs_join_transaction(root);
4963         if (IS_ERR(trans)) {
4964                 ret = PTR_ERR(trans);
4965                 goto out;
4966         }
4967
4968         qgroupid = sa->qgroupid;
4969         if (!qgroupid) {
4970                 /* take the current subvol as qgroup */
4971                 qgroupid = root->root_key.objectid;
4972         }
4973
4974         /* FIXME: check if the IDs really exist */
4975         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4976
4977         err = btrfs_end_transaction(trans, root);
4978         if (err && !ret)
4979                 ret = err;
4980
4981 out:
4982         kfree(sa);
4983 drop_write:
4984         mnt_drop_write_file(file);
4985         return ret;
4986 }
4987
4988 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4989 {
4990         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4991         struct btrfs_ioctl_quota_rescan_args *qsa;
4992         int ret;
4993
4994         if (!capable(CAP_SYS_ADMIN))
4995                 return -EPERM;
4996
4997         ret = mnt_want_write_file(file);
4998         if (ret)
4999                 return ret;
5000
5001         qsa = memdup_user(arg, sizeof(*qsa));
5002         if (IS_ERR(qsa)) {
5003                 ret = PTR_ERR(qsa);
5004                 goto drop_write;
5005         }
5006
5007         if (qsa->flags) {
5008                 ret = -EINVAL;
5009                 goto out;
5010         }
5011
5012         ret = btrfs_qgroup_rescan(root->fs_info);
5013
5014 out:
5015         kfree(qsa);
5016 drop_write:
5017         mnt_drop_write_file(file);
5018         return ret;
5019 }
5020
5021 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5022 {
5023         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5024         struct btrfs_ioctl_quota_rescan_args *qsa;
5025         int ret = 0;
5026
5027         if (!capable(CAP_SYS_ADMIN))
5028                 return -EPERM;
5029
5030         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5031         if (!qsa)
5032                 return -ENOMEM;
5033
5034         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5035                 qsa->flags = 1;
5036                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5037         }
5038
5039         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5040                 ret = -EFAULT;
5041
5042         kfree(qsa);
5043         return ret;
5044 }
5045
5046 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5047 {
5048         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5049
5050         if (!capable(CAP_SYS_ADMIN))
5051                 return -EPERM;
5052
5053         return btrfs_qgroup_wait_for_completion(root->fs_info);
5054 }
5055
5056 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5057                                             struct btrfs_ioctl_received_subvol_args *sa)
5058 {
5059         struct inode *inode = file_inode(file);
5060         struct btrfs_root *root = BTRFS_I(inode)->root;
5061         struct btrfs_root_item *root_item = &root->root_item;
5062         struct btrfs_trans_handle *trans;
5063         struct timespec ct = CURRENT_TIME;
5064         int ret = 0;
5065         int received_uuid_changed;
5066
5067         if (!inode_owner_or_capable(inode))
5068                 return -EPERM;
5069
5070         ret = mnt_want_write_file(file);
5071         if (ret < 0)
5072                 return ret;
5073
5074         down_write(&root->fs_info->subvol_sem);
5075
5076         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5077                 ret = -EINVAL;
5078                 goto out;
5079         }
5080
5081         if (btrfs_root_readonly(root)) {
5082                 ret = -EROFS;
5083                 goto out;
5084         }
5085
5086         /*
5087          * 1 - root item
5088          * 2 - uuid items (received uuid + subvol uuid)
5089          */
5090         trans = btrfs_start_transaction(root, 3);
5091         if (IS_ERR(trans)) {
5092                 ret = PTR_ERR(trans);
5093                 trans = NULL;
5094                 goto out;
5095         }
5096
5097         sa->rtransid = trans->transid;
5098         sa->rtime.sec = ct.tv_sec;
5099         sa->rtime.nsec = ct.tv_nsec;
5100
5101         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5102                                        BTRFS_UUID_SIZE);
5103         if (received_uuid_changed &&
5104             !btrfs_is_empty_uuid(root_item->received_uuid))
5105                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5106                                     root_item->received_uuid,
5107                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5108                                     root->root_key.objectid);
5109         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5110         btrfs_set_root_stransid(root_item, sa->stransid);
5111         btrfs_set_root_rtransid(root_item, sa->rtransid);
5112         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5113         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5114         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5115         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5116
5117         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5118                                 &root->root_key, &root->root_item);
5119         if (ret < 0) {
5120                 btrfs_end_transaction(trans, root);
5121                 goto out;
5122         }
5123         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5124                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5125                                           sa->uuid,
5126                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5127                                           root->root_key.objectid);
5128                 if (ret < 0 && ret != -EEXIST) {
5129                         btrfs_abort_transaction(trans, root, ret);
5130                         goto out;
5131                 }
5132         }
5133         ret = btrfs_commit_transaction(trans, root);
5134         if (ret < 0) {
5135                 btrfs_abort_transaction(trans, root, ret);
5136                 goto out;
5137         }
5138
5139 out:
5140         up_write(&root->fs_info->subvol_sem);
5141         mnt_drop_write_file(file);
5142         return ret;
5143 }
5144
5145 #ifdef CONFIG_64BIT
5146 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5147                                                 void __user *arg)
5148 {
5149         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5150         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5151         int ret = 0;
5152
5153         args32 = memdup_user(arg, sizeof(*args32));
5154         if (IS_ERR(args32)) {
5155                 ret = PTR_ERR(args32);
5156                 args32 = NULL;
5157                 goto out;
5158         }
5159
5160         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5161         if (!args64) {
5162                 ret = -ENOMEM;
5163                 goto out;
5164         }
5165
5166         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5167         args64->stransid = args32->stransid;
5168         args64->rtransid = args32->rtransid;
5169         args64->stime.sec = args32->stime.sec;
5170         args64->stime.nsec = args32->stime.nsec;
5171         args64->rtime.sec = args32->rtime.sec;
5172         args64->rtime.nsec = args32->rtime.nsec;
5173         args64->flags = args32->flags;
5174
5175         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5176         if (ret)
5177                 goto out;
5178
5179         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5180         args32->stransid = args64->stransid;
5181         args32->rtransid = args64->rtransid;
5182         args32->stime.sec = args64->stime.sec;
5183         args32->stime.nsec = args64->stime.nsec;
5184         args32->rtime.sec = args64->rtime.sec;
5185         args32->rtime.nsec = args64->rtime.nsec;
5186         args32->flags = args64->flags;
5187
5188         ret = copy_to_user(arg, args32, sizeof(*args32));
5189         if (ret)
5190                 ret = -EFAULT;
5191
5192 out:
5193         kfree(args32);
5194         kfree(args64);
5195         return ret;
5196 }
5197 #endif
5198
5199 static long btrfs_ioctl_set_received_subvol(struct file *file,
5200                                             void __user *arg)
5201 {
5202         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5203         int ret = 0;
5204
5205         sa = memdup_user(arg, sizeof(*sa));
5206         if (IS_ERR(sa)) {
5207                 ret = PTR_ERR(sa);
5208                 sa = NULL;
5209                 goto out;
5210         }
5211
5212         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5213
5214         if (ret)
5215                 goto out;
5216
5217         ret = copy_to_user(arg, sa, sizeof(*sa));
5218         if (ret)
5219                 ret = -EFAULT;
5220
5221 out:
5222         kfree(sa);
5223         return ret;
5224 }
5225
5226 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5227 {
5228         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5229         size_t len;
5230         int ret;
5231         char label[BTRFS_LABEL_SIZE];
5232
5233         spin_lock(&root->fs_info->super_lock);
5234         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5235         spin_unlock(&root->fs_info->super_lock);
5236
5237         len = strnlen(label, BTRFS_LABEL_SIZE);
5238
5239         if (len == BTRFS_LABEL_SIZE) {
5240                 btrfs_warn(root->fs_info,
5241                         "label is too long, return the first %zu bytes", --len);
5242         }
5243
5244         ret = copy_to_user(arg, label, len);
5245
5246         return ret ? -EFAULT : 0;
5247 }
5248
5249 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5250 {
5251         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5252         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5253         struct btrfs_trans_handle *trans;
5254         char label[BTRFS_LABEL_SIZE];
5255         int ret;
5256
5257         if (!capable(CAP_SYS_ADMIN))
5258                 return -EPERM;
5259
5260         if (copy_from_user(label, arg, sizeof(label)))
5261                 return -EFAULT;
5262
5263         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5264                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5265                        BTRFS_LABEL_SIZE - 1);
5266                 return -EINVAL;
5267         }
5268
5269         ret = mnt_want_write_file(file);
5270         if (ret)
5271                 return ret;
5272
5273         trans = btrfs_start_transaction(root, 0);
5274         if (IS_ERR(trans)) {
5275                 ret = PTR_ERR(trans);
5276                 goto out_unlock;
5277         }
5278
5279         spin_lock(&root->fs_info->super_lock);
5280         strcpy(super_block->label, label);
5281         spin_unlock(&root->fs_info->super_lock);
5282         ret = btrfs_commit_transaction(trans, root);
5283
5284 out_unlock:
5285         mnt_drop_write_file(file);
5286         return ret;
5287 }
5288
5289 #define INIT_FEATURE_FLAGS(suffix) \
5290         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5291           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5292           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5293
5294 static int btrfs_ioctl_get_supported_features(struct file *file,
5295                                               void __user *arg)
5296 {
5297         static const struct btrfs_ioctl_feature_flags features[3] = {
5298                 INIT_FEATURE_FLAGS(SUPP),
5299                 INIT_FEATURE_FLAGS(SAFE_SET),
5300                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5301         };
5302
5303         if (copy_to_user(arg, &features, sizeof(features)))
5304                 return -EFAULT;
5305
5306         return 0;
5307 }
5308
5309 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5310 {
5311         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5312         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5313         struct btrfs_ioctl_feature_flags features;
5314
5315         features.compat_flags = btrfs_super_compat_flags(super_block);
5316         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5317         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5318
5319         if (copy_to_user(arg, &features, sizeof(features)))
5320                 return -EFAULT;
5321
5322         return 0;
5323 }
5324
5325 static int check_feature_bits(struct btrfs_root *root,
5326                               enum btrfs_feature_set set,
5327                               u64 change_mask, u64 flags, u64 supported_flags,
5328                               u64 safe_set, u64 safe_clear)
5329 {
5330         const char *type = btrfs_feature_set_names[set];
5331         char *names;
5332         u64 disallowed, unsupported;
5333         u64 set_mask = flags & change_mask;
5334         u64 clear_mask = ~flags & change_mask;
5335
5336         unsupported = set_mask & ~supported_flags;
5337         if (unsupported) {
5338                 names = btrfs_printable_features(set, unsupported);
5339                 if (names) {
5340                         btrfs_warn(root->fs_info,
5341                            "this kernel does not support the %s feature bit%s",
5342                            names, strchr(names, ',') ? "s" : "");
5343                         kfree(names);
5344                 } else
5345                         btrfs_warn(root->fs_info,
5346                            "this kernel does not support %s bits 0x%llx",
5347                            type, unsupported);
5348                 return -EOPNOTSUPP;
5349         }
5350
5351         disallowed = set_mask & ~safe_set;
5352         if (disallowed) {
5353                 names = btrfs_printable_features(set, disallowed);
5354                 if (names) {
5355                         btrfs_warn(root->fs_info,
5356                            "can't set the %s feature bit%s while mounted",
5357                            names, strchr(names, ',') ? "s" : "");
5358                         kfree(names);
5359                 } else
5360                         btrfs_warn(root->fs_info,
5361                            "can't set %s bits 0x%llx while mounted",
5362                            type, disallowed);
5363                 return -EPERM;
5364         }
5365
5366         disallowed = clear_mask & ~safe_clear;
5367         if (disallowed) {
5368                 names = btrfs_printable_features(set, disallowed);
5369                 if (names) {
5370                         btrfs_warn(root->fs_info,
5371                            "can't clear the %s feature bit%s while mounted",
5372                            names, strchr(names, ',') ? "s" : "");
5373                         kfree(names);
5374                 } else
5375                         btrfs_warn(root->fs_info,
5376                            "can't clear %s bits 0x%llx while mounted",
5377                            type, disallowed);
5378                 return -EPERM;
5379         }
5380
5381         return 0;
5382 }
5383
5384 #define check_feature(root, change_mask, flags, mask_base)      \
5385 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5386                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5387                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5388                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5389
5390 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5391 {
5392         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5393         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5394         struct btrfs_ioctl_feature_flags flags[2];
5395         struct btrfs_trans_handle *trans;
5396         u64 newflags;
5397         int ret;
5398
5399         if (!capable(CAP_SYS_ADMIN))
5400                 return -EPERM;
5401
5402         if (copy_from_user(flags, arg, sizeof(flags)))
5403                 return -EFAULT;
5404
5405         /* Nothing to do */
5406         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5407             !flags[0].incompat_flags)
5408                 return 0;
5409
5410         ret = check_feature(root, flags[0].compat_flags,
5411                             flags[1].compat_flags, COMPAT);
5412         if (ret)
5413                 return ret;
5414
5415         ret = check_feature(root, flags[0].compat_ro_flags,
5416                             flags[1].compat_ro_flags, COMPAT_RO);
5417         if (ret)
5418                 return ret;
5419
5420         ret = check_feature(root, flags[0].incompat_flags,
5421                             flags[1].incompat_flags, INCOMPAT);
5422         if (ret)
5423                 return ret;
5424
5425         trans = btrfs_start_transaction(root, 0);
5426         if (IS_ERR(trans))
5427                 return PTR_ERR(trans);
5428
5429         spin_lock(&root->fs_info->super_lock);
5430         newflags = btrfs_super_compat_flags(super_block);
5431         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5432         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5433         btrfs_set_super_compat_flags(super_block, newflags);
5434
5435         newflags = btrfs_super_compat_ro_flags(super_block);
5436         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5437         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5438         btrfs_set_super_compat_ro_flags(super_block, newflags);
5439
5440         newflags = btrfs_super_incompat_flags(super_block);
5441         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5442         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5443         btrfs_set_super_incompat_flags(super_block, newflags);
5444         spin_unlock(&root->fs_info->super_lock);
5445
5446         return btrfs_commit_transaction(trans, root);
5447 }
5448
5449 long btrfs_ioctl(struct file *file, unsigned int
5450                 cmd, unsigned long arg)
5451 {
5452         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5453         void __user *argp = (void __user *)arg;
5454
5455         switch (cmd) {
5456         case FS_IOC_GETFLAGS:
5457                 return btrfs_ioctl_getflags(file, argp);
5458         case FS_IOC_SETFLAGS:
5459                 return btrfs_ioctl_setflags(file, argp);
5460         case FS_IOC_GETVERSION:
5461                 return btrfs_ioctl_getversion(file, argp);
5462         case FITRIM:
5463                 return btrfs_ioctl_fitrim(file, argp);
5464         case BTRFS_IOC_SNAP_CREATE:
5465                 return btrfs_ioctl_snap_create(file, argp, 0);
5466         case BTRFS_IOC_SNAP_CREATE_V2:
5467                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5468         case BTRFS_IOC_SUBVOL_CREATE:
5469                 return btrfs_ioctl_snap_create(file, argp, 1);
5470         case BTRFS_IOC_SUBVOL_CREATE_V2:
5471                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5472         case BTRFS_IOC_SNAP_DESTROY:
5473                 return btrfs_ioctl_snap_destroy(file, argp);
5474         case BTRFS_IOC_SUBVOL_GETFLAGS:
5475                 return btrfs_ioctl_subvol_getflags(file, argp);
5476         case BTRFS_IOC_SUBVOL_SETFLAGS:
5477                 return btrfs_ioctl_subvol_setflags(file, argp);
5478         case BTRFS_IOC_DEFAULT_SUBVOL:
5479                 return btrfs_ioctl_default_subvol(file, argp);
5480         case BTRFS_IOC_DEFRAG:
5481                 return btrfs_ioctl_defrag(file, NULL);
5482         case BTRFS_IOC_DEFRAG_RANGE:
5483                 return btrfs_ioctl_defrag(file, argp);
5484         case BTRFS_IOC_RESIZE:
5485                 return btrfs_ioctl_resize(file, argp);
5486         case BTRFS_IOC_ADD_DEV:
5487                 return btrfs_ioctl_add_dev(root, argp);
5488         case BTRFS_IOC_RM_DEV:
5489                 return btrfs_ioctl_rm_dev(file, argp);
5490         case BTRFS_IOC_FS_INFO:
5491                 return btrfs_ioctl_fs_info(root, argp);
5492         case BTRFS_IOC_DEV_INFO:
5493                 return btrfs_ioctl_dev_info(root, argp);
5494         case BTRFS_IOC_BALANCE:
5495                 return btrfs_ioctl_balance(file, NULL);
5496         case BTRFS_IOC_CLONE:
5497                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5498         case BTRFS_IOC_CLONE_RANGE:
5499                 return btrfs_ioctl_clone_range(file, argp);
5500         case BTRFS_IOC_TRANS_START:
5501                 return btrfs_ioctl_trans_start(file);
5502         case BTRFS_IOC_TRANS_END:
5503                 return btrfs_ioctl_trans_end(file);
5504         case BTRFS_IOC_TREE_SEARCH:
5505                 return btrfs_ioctl_tree_search(file, argp);
5506         case BTRFS_IOC_TREE_SEARCH_V2:
5507                 return btrfs_ioctl_tree_search_v2(file, argp);
5508         case BTRFS_IOC_INO_LOOKUP:
5509                 return btrfs_ioctl_ino_lookup(file, argp);
5510         case BTRFS_IOC_INO_PATHS:
5511                 return btrfs_ioctl_ino_to_path(root, argp);
5512         case BTRFS_IOC_LOGICAL_INO:
5513                 return btrfs_ioctl_logical_to_ino(root, argp);
5514         case BTRFS_IOC_SPACE_INFO:
5515                 return btrfs_ioctl_space_info(root, argp);
5516         case BTRFS_IOC_SYNC: {
5517                 int ret;
5518
5519                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5520                 if (ret)
5521                         return ret;
5522                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5523                 /*
5524                  * The transaction thread may want to do more work,
5525                  * namely it pokes the cleaner ktread that will start
5526                  * processing uncleaned subvols.
5527                  */
5528                 wake_up_process(root->fs_info->transaction_kthread);
5529                 return ret;
5530         }
5531         case BTRFS_IOC_START_SYNC:
5532                 return btrfs_ioctl_start_sync(root, argp);
5533         case BTRFS_IOC_WAIT_SYNC:
5534                 return btrfs_ioctl_wait_sync(root, argp);
5535         case BTRFS_IOC_SCRUB:
5536                 return btrfs_ioctl_scrub(file, argp);
5537         case BTRFS_IOC_SCRUB_CANCEL:
5538                 return btrfs_ioctl_scrub_cancel(root, argp);
5539         case BTRFS_IOC_SCRUB_PROGRESS:
5540                 return btrfs_ioctl_scrub_progress(root, argp);
5541         case BTRFS_IOC_BALANCE_V2:
5542                 return btrfs_ioctl_balance(file, argp);
5543         case BTRFS_IOC_BALANCE_CTL:
5544                 return btrfs_ioctl_balance_ctl(root, arg);
5545         case BTRFS_IOC_BALANCE_PROGRESS:
5546                 return btrfs_ioctl_balance_progress(root, argp);
5547         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5548                 return btrfs_ioctl_set_received_subvol(file, argp);
5549 #ifdef CONFIG_64BIT
5550         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5551                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5552 #endif
5553         case BTRFS_IOC_SEND:
5554                 return btrfs_ioctl_send(file, argp);
5555         case BTRFS_IOC_GET_DEV_STATS:
5556                 return btrfs_ioctl_get_dev_stats(root, argp);
5557         case BTRFS_IOC_QUOTA_CTL:
5558                 return btrfs_ioctl_quota_ctl(file, argp);
5559         case BTRFS_IOC_QGROUP_ASSIGN:
5560                 return btrfs_ioctl_qgroup_assign(file, argp);
5561         case BTRFS_IOC_QGROUP_CREATE:
5562                 return btrfs_ioctl_qgroup_create(file, argp);
5563         case BTRFS_IOC_QGROUP_LIMIT:
5564                 return btrfs_ioctl_qgroup_limit(file, argp);
5565         case BTRFS_IOC_QUOTA_RESCAN:
5566                 return btrfs_ioctl_quota_rescan(file, argp);
5567         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5568                 return btrfs_ioctl_quota_rescan_status(file, argp);
5569         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5570                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5571         case BTRFS_IOC_DEV_REPLACE:
5572                 return btrfs_ioctl_dev_replace(root, argp);
5573         case BTRFS_IOC_GET_FSLABEL:
5574                 return btrfs_ioctl_get_fslabel(file, argp);
5575         case BTRFS_IOC_SET_FSLABEL:
5576                 return btrfs_ioctl_set_fslabel(file, argp);
5577         case BTRFS_IOC_FILE_EXTENT_SAME:
5578                 return btrfs_ioctl_file_extent_same(file, argp);
5579         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5580                 return btrfs_ioctl_get_supported_features(file, argp);
5581         case BTRFS_IOC_GET_FEATURES:
5582                 return btrfs_ioctl_get_features(file, argp);
5583         case BTRFS_IOC_SET_FEATURES:
5584                 return btrfs_ioctl_set_features(file, argp);
5585         }
5586
5587         return -ENOTTY;
5588 }