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
  62 #ifdef CONFIG_64BIT
  63 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  64  * structures are incorrect, as the timespec structure from userspace
  65  * is 4 bytes too small. We define these alternatives here to teach
  66  * the kernel about the 32-bit struct packing.
  67  */
  68 struct btrfs_ioctl_timespec_32 {
  69         __u64 sec;
  70         __u32 nsec;
  71 } __attribute__ ((__packed__));
  72
  73 struct btrfs_ioctl_received_subvol_args_32 {
  74         char    uuid[BTRFS_UUID_SIZE];  /* in */
  75         __u64   stransid;               /* in */
  76         __u64   rtransid;               /* out */
  77         struct btrfs_ioctl_timespec_32 stime; /* in */
  78         struct btrfs_ioctl_timespec_32 rtime; /* out */
  79         __u64   flags;                  /* in */
  80         __u64   reserved[16];           /* in */
  81 } __attribute__ ((__packed__));
  82
  83 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  84                                 struct btrfs_ioctl_received_subvol_args_32)
  85 #endif
  86
  87
  88 static int btrfs_clone(struct inode *src, struct inode *inode,
  89                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  90
  91 /* Mask out flags that are inappropriate for the given type of inode. */
  92 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  93 {
  94         if (S_ISDIR(mode))
  95                 return flags;
  96         else if (S_ISREG(mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 104  */
 105 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 106 {
 107         unsigned int iflags = 0;
 108
 109         if (flags & BTRFS_INODE_SYNC)
 110                 iflags |= FS_SYNC_FL;
 111         if (flags & BTRFS_INODE_IMMUTABLE)
 112                 iflags |= FS_IMMUTABLE_FL;
 113         if (flags & BTRFS_INODE_APPEND)
 114                 iflags |= FS_APPEND_FL;
 115         if (flags & BTRFS_INODE_NODUMP)
 116                 iflags |= FS_NODUMP_FL;
 117         if (flags & BTRFS_INODE_NOATIME)
 118                 iflags |= FS_NOATIME_FL;
 119         if (flags & BTRFS_INODE_DIRSYNC)
 120                 iflags |= FS_DIRSYNC_FL;
 121         if (flags & BTRFS_INODE_NODATACOW)
 122                 iflags |= FS_NOCOW_FL;
 123
 124         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 125                 iflags |= FS_COMPR_FL;
 126         else if (flags & BTRFS_INODE_NOCOMPRESS)
 127                 iflags |= FS_NOCOMP_FL;
 128
 129         return iflags;
 130 }
 131
 132 /*
 133  * Update inode->i_flags based on the btrfs internal flags.
 134  */
 135 void btrfs_update_iflags(struct inode *inode)
 136 {
 137         struct btrfs_inode *ip = BTRFS_I(inode);
 138
 139         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 inode->i_flags |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 inode->i_flags |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 inode->i_flags |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 inode->i_flags |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 inode->i_flags |= S_DIRSYNC;
 151 }
 152
 153 /*
 154  * Inherit flags from the parent inode.
 155  *
 156  * Currently only the compression flags and the cow flags are inherited.
 157  */
 158 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 159 {
 160         unsigned int flags;
 161
 162         if (!dir)
 163                 return;
 164
 165         flags = BTRFS_I(dir)->flags;
 166
 167         if (flags & BTRFS_INODE_NOCOMPRESS) {
 168                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 169                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 170         } else if (flags & BTRFS_INODE_COMPRESS) {
 171                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 172                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 173         }
 174
 175         if (flags & BTRFS_INODE_NODATACOW) {
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 177                 if (S_ISREG(inode->i_mode))
 178                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 179         }
 180
 181         btrfs_update_iflags(inode);
 182 }
 183
 184 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 185 {
 186         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 187         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 188
 189         if (copy_to_user(arg, &flags, sizeof(flags)))
 190                 return -EFAULT;
 191         return 0;
 192 }
 193
 194 static int check_flags(unsigned int flags)
 195 {
 196         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 197                       FS_NOATIME_FL | FS_NODUMP_FL | \
 198                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 199                       FS_NOCOMP_FL | FS_COMPR_FL |
 200                       FS_NOCOW_FL))
 201                 return -EOPNOTSUPP;
 202
 203         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 204                 return -EINVAL;
 205
 206         return 0;
 207 }
 208
 209 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 210 {
 211         struct inode *inode = file_inode(file);
 212         struct btrfs_inode *ip = BTRFS_I(inode);
 213         struct btrfs_root *root = ip->root;
 214         struct btrfs_trans_handle *trans;
 215         unsigned int flags, oldflags;
 216         int ret;
 217         u64 ip_oldflags;
 218         unsigned int i_oldflags;
 219         umode_t mode;
 220
 221         if (!inode_owner_or_capable(inode))
 222                 return -EPERM;
 223
 224         if (btrfs_root_readonly(root))
 225                 return -EROFS;
 226
 227         if (copy_from_user(&flags, arg, sizeof(flags)))
 228                 return -EFAULT;
 229
 230         ret = check_flags(flags);
 231         if (ret)
 232                 return ret;
 233
 234         ret = mnt_want_write_file(file);
 235         if (ret)
 236                 return ret;
 237
 238         mutex_lock(&inode->i_mutex);
 239
 240         ip_oldflags = ip->flags;
 241         i_oldflags = inode->i_flags;
 242         mode = inode->i_mode;
 243
 244         flags = btrfs_mask_flags(inode->i_mode, flags);
 245         oldflags = btrfs_flags_to_ioctl(ip->flags);
 246         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 247                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 248                         ret = -EPERM;
 249                         goto out_unlock;
 250                 }
 251         }
 252
 253         if (flags & FS_SYNC_FL)
 254                 ip->flags |= BTRFS_INODE_SYNC;
 255         else
 256                 ip->flags &= ~BTRFS_INODE_SYNC;
 257         if (flags & FS_IMMUTABLE_FL)
 258                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 261         if (flags & FS_APPEND_FL)
 262                 ip->flags |= BTRFS_INODE_APPEND;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_APPEND;
 265         if (flags & FS_NODUMP_FL)
 266                 ip->flags |= BTRFS_INODE_NODUMP;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_NODUMP;
 269         if (flags & FS_NOATIME_FL)
 270                 ip->flags |= BTRFS_INODE_NOATIME;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NOATIME;
 273         if (flags & FS_DIRSYNC_FL)
 274                 ip->flags |= BTRFS_INODE_DIRSYNC;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 277         if (flags & FS_NOCOW_FL) {
 278                 if (S_ISREG(mode)) {
 279                         /*
 280                          * It's safe to turn csums off here, no extents exist.
 281                          * Otherwise we want the flag to reflect the real COW
 282                          * status of the file and will not set it.
 283                          */
 284                         if (inode->i_size == 0)
 285                                 ip->flags |= BTRFS_INODE_NODATACOW
 286                                            | BTRFS_INODE_NODATASUM;
 287                 } else {
 288                         ip->flags |= BTRFS_INODE_NODATACOW;
 289                 }
 290         } else {
 291                 /*
 292                  * Revert back under same assuptions as above
 293                  */
 294                 if (S_ISREG(mode)) {
 295                         if (inode->i_size == 0)
 296                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 297                                              | BTRFS_INODE_NODATASUM);
 298                 } else {
 299                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 300                 }
 301         }
 302
 303         /*
 304          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 305          * flag may be changed automatically if compression code won't make
 306          * things smaller.
 307          */
 308         if (flags & FS_NOCOMP_FL) {
 309                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 310                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 311
 312                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 313                 if (ret && ret != -ENODATA)
 314                         goto out_drop;
 315         } else if (flags & FS_COMPR_FL) {
 316                 const char *comp;
 317
 318                 ip->flags |= BTRFS_INODE_COMPRESS;
 319                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 320
 321                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 322                         comp = "lzo";
 323                 else
 324                         comp = "zlib";
 325                 ret = btrfs_set_prop(inode, "btrfs.compression",
 326                                      comp, strlen(comp), 0);
 327                 if (ret)
 328                         goto out_drop;
 329
 330         } else {
 331                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 332         }
 333
 334         trans = btrfs_start_transaction(root, 1);
 335         if (IS_ERR(trans)) {
 336                 ret = PTR_ERR(trans);
 337                 goto out_drop;
 338         }
 339
 340         btrfs_update_iflags(inode);
 341         inode_inc_iversion(inode);
 342         inode->i_ctime = CURRENT_TIME;
 343         ret = btrfs_update_inode(trans, root, inode);
 344
 345         btrfs_end_transaction(trans, root);
 346  out_drop:
 347         if (ret) {
 348                 ip->flags = ip_oldflags;
 349                 inode->i_flags = i_oldflags;
 350         }
 351
 352  out_unlock:
 353         mutex_unlock(&inode->i_mutex);
 354         mnt_drop_write_file(file);
 355         return ret;
 356 }
 357
 358 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 359 {
 360         struct inode *inode = file_inode(file);
 361
 362         return put_user(inode->i_generation, arg);
 363 }
 364
 365 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 366 {
 367         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 368         struct btrfs_device *device;
 369         struct request_queue *q;
 370         struct fstrim_range range;
 371         u64 minlen = ULLONG_MAX;
 372         u64 num_devices = 0;
 373         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 374         int ret;
 375
 376         if (!capable(CAP_SYS_ADMIN))
 377                 return -EPERM;
 378
 379         rcu_read_lock();
 380         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 381                                 dev_list) {
 382                 if (!device->bdev)
 383                         continue;
 384                 q = bdev_get_queue(device->bdev);
 385                 if (blk_queue_discard(q)) {
 386                         num_devices++;
 387                         minlen = min((u64)q->limits.discard_granularity,
 388                                      minlen);
 389                 }
 390         }
 391         rcu_read_unlock();
 392
 393         if (!num_devices)
 394                 return -EOPNOTSUPP;
 395         if (copy_from_user(&range, arg, sizeof(range)))
 396                 return -EFAULT;
 397         if (range.start > total_bytes ||
 398             range.len < fs_info->sb->s_blocksize)
 399                 return -EINVAL;
 400
 401         range.len = min(range.len, total_bytes - range.start);
 402         range.minlen = max(range.minlen, minlen);
 403         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 404         if (ret < 0)
 405                 return ret;
 406
 407         if (copy_to_user(arg, &range, sizeof(range)))
 408                 return -EFAULT;
 409
 410         return 0;
 411 }
 412
 413 int btrfs_is_empty_uuid(u8 *uuid)
 414 {
 415         int i;
 416
 417         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 418                 if (uuid[i])
 419                         return 0;
 420         }
 421         return 1;
 422 }
 423
 424 static noinline int create_subvol(struct inode *dir,
 425                                   struct dentry *dentry,
 426                                   char *name, int namelen,
 427                                   u64 *async_transid,
 428                                   struct btrfs_qgroup_inherit *inherit)
 429 {
 430         struct btrfs_trans_handle *trans;
 431         struct btrfs_key key;
 432         struct btrfs_root_item root_item;
 433         struct btrfs_inode_item *inode_item;
 434         struct extent_buffer *leaf;
 435         struct btrfs_root *root = BTRFS_I(dir)->root;
 436         struct btrfs_root *new_root;
 437         struct btrfs_block_rsv block_rsv;
 438         struct timespec cur_time = CURRENT_TIME;
 439         struct inode *inode;
 440         int ret;
 441         int err;
 442         u64 objectid;
 443         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 444         u64 index = 0;
 445         u64 qgroup_reserved;
 446         uuid_le new_uuid;
 447
 448         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 449         if (ret)
 450                 return ret;
 451
 452         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 453         /*
 454          * The same as the snapshot creation, please see the comment
 455          * of create_snapshot().
 456          */
 457         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 458                                                8, &qgroup_reserved, false);
 459         if (ret)
 460                 return ret;
 461
 462         trans = btrfs_start_transaction(root, 0);
 463         if (IS_ERR(trans)) {
 464                 ret = PTR_ERR(trans);
 465                 btrfs_subvolume_release_metadata(root, &block_rsv,
 466                                                  qgroup_reserved);
 467                 return ret;
 468         }
 469         trans->block_rsv = &block_rsv;
 470         trans->bytes_reserved = block_rsv.size;
 471
 472         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 473         if (ret)
 474                 goto fail;
 475
 476         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 477                                       0, objectid, NULL, 0, 0, 0);
 478         if (IS_ERR(leaf)) {
 479                 ret = PTR_ERR(leaf);
 480                 goto fail;
 481         }
 482
 483         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 484         btrfs_set_header_bytenr(leaf, leaf->start);
 485         btrfs_set_header_generation(leaf, trans->transid);
 486         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 487         btrfs_set_header_owner(leaf, objectid);
 488
 489         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 490                             BTRFS_FSID_SIZE);
 491         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 492                             btrfs_header_chunk_tree_uuid(leaf),
 493                             BTRFS_UUID_SIZE);
 494         btrfs_mark_buffer_dirty(leaf);
 495
 496         memset(&root_item, 0, sizeof(root_item));
 497
 498         inode_item = &root_item.inode;
 499         btrfs_set_stack_inode_generation(inode_item, 1);
 500         btrfs_set_stack_inode_size(inode_item, 3);
 501         btrfs_set_stack_inode_nlink(inode_item, 1);
 502         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 503         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 504
 505         btrfs_set_root_flags(&root_item, 0);
 506         btrfs_set_root_limit(&root_item, 0);
 507         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 508
 509         btrfs_set_root_bytenr(&root_item, leaf->start);
 510         btrfs_set_root_generation(&root_item, trans->transid);
 511         btrfs_set_root_level(&root_item, 0);
 512         btrfs_set_root_refs(&root_item, 1);
 513         btrfs_set_root_used(&root_item, leaf->len);
 514         btrfs_set_root_last_snapshot(&root_item, 0);
 515
 516         btrfs_set_root_generation_v2(&root_item,
 517                         btrfs_root_generation(&root_item));
 518         uuid_le_gen(&new_uuid);
 519         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 520         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 521         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 522         root_item.ctime = root_item.otime;
 523         btrfs_set_root_ctransid(&root_item, trans->transid);
 524         btrfs_set_root_otransid(&root_item, trans->transid);
 525
 526         btrfs_tree_unlock(leaf);
 527         free_extent_buffer(leaf);
 528         leaf = NULL;
 529
 530         btrfs_set_root_dirid(&root_item, new_dirid);
 531
 532         key.objectid = objectid;
 533         key.offset = 0;
 534         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 535         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 536                                 &root_item);
 537         if (ret)
 538                 goto fail;
 539
 540         key.offset = (u64)-1;
 541         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 542         if (IS_ERR(new_root)) {
 543                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 544                 ret = PTR_ERR(new_root);
 545                 goto fail;
 546         }
 547
 548         btrfs_record_root_in_trans(trans, new_root);
 549
 550         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 551         if (ret) {
 552                 /* We potentially lose an unused inode item here */
 553                 btrfs_abort_transaction(trans, root, ret);
 554                 goto fail;
 555         }
 556
 557         /*
 558          * insert the directory item
 559          */
 560         ret = btrfs_set_inode_index(dir, &index);
 561         if (ret) {
 562                 btrfs_abort_transaction(trans, root, ret);
 563                 goto fail;
 564         }
 565
 566         ret = btrfs_insert_dir_item(trans, root,
 567                                     name, namelen, dir, &key,
 568                                     BTRFS_FT_DIR, index);
 569         if (ret) {
 570                 btrfs_abort_transaction(trans, root, ret);
 571                 goto fail;
 572         }
 573
 574         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 575         ret = btrfs_update_inode(trans, root, dir);
 576         BUG_ON(ret);
 577
 578         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 579                                  objectid, root->root_key.objectid,
 580                                  btrfs_ino(dir), index, name, namelen);
 581         BUG_ON(ret);
 582
 583         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 584                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 585                                   objectid);
 586         if (ret)
 587                 btrfs_abort_transaction(trans, root, ret);
 588
 589 fail:
 590         trans->block_rsv = NULL;
 591         trans->bytes_reserved = 0;
 592         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 593
 594         if (async_transid) {
 595                 *async_transid = trans->transid;
 596                 err = btrfs_commit_transaction_async(trans, root, 1);
 597                 if (err)
 598                         err = btrfs_commit_transaction(trans, root);
 599         } else {
 600                 err = btrfs_commit_transaction(trans, root);
 601         }
 602         if (err && !ret)
 603                 ret = err;
 604
 605         if (!ret) {
 606                 inode = btrfs_lookup_dentry(dir, dentry);
 607                 if (IS_ERR(inode))
 608                         return PTR_ERR(inode);
 609                 d_instantiate(dentry, inode);
 610         }
 611         return ret;
 612 }
 613
 614 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 615 {
 616         s64 writers;
 617         DEFINE_WAIT(wait);
 618
 619         do {
 620                 prepare_to_wait(&root->subv_writers->wait, &wait,
 621                                 TASK_UNINTERRUPTIBLE);
 622
 623                 writers = percpu_counter_sum(&root->subv_writers->counter);
 624                 if (writers)
 625                         schedule();
 626
 627                 finish_wait(&root->subv_writers->wait, &wait);
 628         } while (writers);
 629 }
 630
 631 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 632                            struct dentry *dentry, char *name, int namelen,
 633                            u64 *async_transid, bool readonly,
 634                            struct btrfs_qgroup_inherit *inherit)
 635 {
 636         struct inode *inode;
 637         struct btrfs_pending_snapshot *pending_snapshot;
 638         struct btrfs_trans_handle *trans;
 639         int ret;
 640
 641         if (!root->ref_cows)
 642                 return -EINVAL;
 643
 644         atomic_inc(&root->will_be_snapshoted);
 645         smp_mb__after_atomic_inc();
 646         btrfs_wait_nocow_write(root);
 647
 648         ret = btrfs_start_delalloc_inodes(root, 0);
 649         if (ret)
 650                 goto out;
 651
 652         btrfs_wait_ordered_extents(root, -1);
 653
 654         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 655         if (!pending_snapshot) {
 656                 ret = -ENOMEM;
 657                 goto out;
 658         }
 659
 660         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 661                              BTRFS_BLOCK_RSV_TEMP);
 662         /*
 663          * 1 - parent dir inode
 664          * 2 - dir entries
 665          * 1 - root item
 666          * 2 - root ref/backref
 667          * 1 - root of snapshot
 668          * 1 - UUID item
 669          */
 670         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 671                                         &pending_snapshot->block_rsv, 8,
 672                                         &pending_snapshot->qgroup_reserved,
 673                                         false);
 674         if (ret)
 675                 goto free;
 676
 677         pending_snapshot->dentry = dentry;
 678         pending_snapshot->root = root;
 679         pending_snapshot->readonly = readonly;
 680         pending_snapshot->dir = dir;
 681         pending_snapshot->inherit = inherit;
 682
 683         trans = btrfs_start_transaction(root, 0);
 684         if (IS_ERR(trans)) {
 685                 ret = PTR_ERR(trans);
 686                 goto fail;
 687         }
 688
 689         spin_lock(&root->fs_info->trans_lock);
 690         list_add(&pending_snapshot->list,
 691                  &trans->transaction->pending_snapshots);
 692         spin_unlock(&root->fs_info->trans_lock);
 693         if (async_transid) {
 694                 *async_transid = trans->transid;
 695                 ret = btrfs_commit_transaction_async(trans,
 696                                      root->fs_info->extent_root, 1);
 697                 if (ret)
 698                         ret = btrfs_commit_transaction(trans, root);
 699         } else {
 700                 ret = btrfs_commit_transaction(trans,
 701                                                root->fs_info->extent_root);
 702         }
 703         if (ret)
 704                 goto fail;
 705
 706         ret = pending_snapshot->error;
 707         if (ret)
 708                 goto fail;
 709
 710         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 711         if (ret)
 712                 goto fail;
 713
 714         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 715         if (IS_ERR(inode)) {
 716                 ret = PTR_ERR(inode);
 717                 goto fail;
 718         }
 719
 720         d_instantiate(dentry, inode);
 721         ret = 0;
 722 fail:
 723         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 724                                          &pending_snapshot->block_rsv,
 725                                          pending_snapshot->qgroup_reserved);
 726 free:
 727         kfree(pending_snapshot);
 728 out:
 729         atomic_dec(&root->will_be_snapshoted);
 730         return ret;
 731 }
 732
 733 /*  copy of check_sticky in fs/namei.c()
 734 * It's inline, so penalty for filesystems that don't use sticky bit is
 735 * minimal.
 736 */
 737 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 738 {
 739         kuid_t fsuid = current_fsuid();
 740
 741         if (!(dir->i_mode & S_ISVTX))
 742                 return 0;
 743         if (uid_eq(inode->i_uid, fsuid))
 744                 return 0;
 745         if (uid_eq(dir->i_uid, fsuid))
 746                 return 0;
 747         return !capable(CAP_FOWNER);
 748 }
 749
 750 /*  copy of may_delete in fs/namei.c()
 751  *      Check whether we can remove a link victim from directory dir, check
 752  *  whether the type of victim is right.
 753  *  1. We can't do it if dir is read-only (done in permission())
 754  *  2. We should have write and exec permissions on dir
 755  *  3. We can't remove anything from append-only dir
 756  *  4. We can't do anything with immutable dir (done in permission())
 757  *  5. If the sticky bit on dir is set we should either
 758  *      a. be owner of dir, or
 759  *      b. be owner of victim, or
 760  *      c. have CAP_FOWNER capability
 761  *  6. If the victim is append-only or immutable we can't do antyhing with
 762  *     links pointing to it.
 763  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 764  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 765  *  9. We can't remove a root or mountpoint.
 766  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 767  *     nfs_async_unlink().
 768  */
 769
 770 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 771 {
 772         int error;
 773
 774         if (!victim->d_inode)
 775                 return -ENOENT;
 776
 777         BUG_ON(victim->d_parent->d_inode != dir);
 778         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 779
 780         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 781         if (error)
 782                 return error;
 783         if (IS_APPEND(dir))
 784                 return -EPERM;
 785         if (btrfs_check_sticky(dir, victim->d_inode)||
 786                 IS_APPEND(victim->d_inode)||
 787             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 788                 return -EPERM;
 789         if (isdir) {
 790                 if (!S_ISDIR(victim->d_inode->i_mode))
 791                         return -ENOTDIR;
 792                 if (IS_ROOT(victim))
 793                         return -EBUSY;
 794         } else if (S_ISDIR(victim->d_inode->i_mode))
 795                 return -EISDIR;
 796         if (IS_DEADDIR(dir))
 797                 return -ENOENT;
 798         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 799                 return -EBUSY;
 800         return 0;
 801 }
 802
 803 /* copy of may_create in fs/namei.c() */
 804 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 805 {
 806         if (child->d_inode)
 807                 return -EEXIST;
 808         if (IS_DEADDIR(dir))
 809                 return -ENOENT;
 810         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 811 }
 812
 813 /*
 814  * Create a new subvolume below @parent.  This is largely modeled after
 815  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 816  * inside this filesystem so it's quite a bit simpler.
 817  */
 818 static noinline int btrfs_mksubvol(struct path *parent,
 819                                    char *name, int namelen,
 820                                    struct btrfs_root *snap_src,
 821                                    u64 *async_transid, bool readonly,
 822                                    struct btrfs_qgroup_inherit *inherit)
 823 {
 824         struct inode *dir  = parent->dentry->d_inode;
 825         struct dentry *dentry;
 826         int error;
 827
 828         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 829         if (error == -EINTR)
 830                 return error;
 831
 832         dentry = lookup_one_len(name, parent->dentry, namelen);
 833         error = PTR_ERR(dentry);
 834         if (IS_ERR(dentry))
 835                 goto out_unlock;
 836
 837         error = -EEXIST;
 838         if (dentry->d_inode)
 839                 goto out_dput;
 840
 841         error = btrfs_may_create(dir, dentry);
 842         if (error)
 843                 goto out_dput;
 844
 845         /*
 846          * even if this name doesn't exist, we may get hash collisions.
 847          * check for them now when we can safely fail
 848          */
 849         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 850                                                dir->i_ino, name,
 851                                                namelen);
 852         if (error)
 853                 goto out_dput;
 854
 855         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 856
 857         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 858                 goto out_up_read;
 859
 860         if (snap_src) {
 861                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 862                                         async_transid, readonly, inherit);
 863         } else {
 864                 error = create_subvol(dir, dentry, name, namelen,
 865                                       async_transid, inherit);
 866         }
 867         if (!error)
 868                 fsnotify_mkdir(dir, dentry);
 869 out_up_read:
 870         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 871 out_dput:
 872         dput(dentry);
 873 out_unlock:
 874         mutex_unlock(&dir->i_mutex);
 875         return error;
 876 }
 877
 878 /*
 879  * When we're defragging a range, we don't want to kick it off again
 880  * if it is really just waiting for delalloc to send it down.
 881  * If we find a nice big extent or delalloc range for the bytes in the
 882  * file you want to defrag, we return 0 to let you know to skip this
 883  * part of the file
 884  */
 885 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 886 {
 887         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 888         struct extent_map *em = NULL;
 889         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 890         u64 end;
 891
 892         read_lock(&em_tree->lock);
 893         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 894         read_unlock(&em_tree->lock);
 895
 896         if (em) {
 897                 end = extent_map_end(em);
 898                 free_extent_map(em);
 899                 if (end - offset > thresh)
 900                         return 0;
 901         }
 902         /* if we already have a nice delalloc here, just stop */
 903         thresh /= 2;
 904         end = count_range_bits(io_tree, &offset, offset + thresh,
 905                                thresh, EXTENT_DELALLOC, 1);
 906         if (end >= thresh)
 907                 return 0;
 908         return 1;
 909 }
 910
 911 /*
 912  * helper function to walk through a file and find extents
 913  * newer than a specific transid, and smaller than thresh.
 914  *
 915  * This is used by the defragging code to find new and small
 916  * extents
 917  */
 918 static int find_new_extents(struct btrfs_root *root,
 919                             struct inode *inode, u64 newer_than,
 920                             u64 *off, int thresh)
 921 {
 922         struct btrfs_path *path;
 923         struct btrfs_key min_key;
 924         struct extent_buffer *leaf;
 925         struct btrfs_file_extent_item *extent;
 926         int type;
 927         int ret;
 928         u64 ino = btrfs_ino(inode);
 929
 930         path = btrfs_alloc_path();
 931         if (!path)
 932                 return -ENOMEM;
 933
 934         min_key.objectid = ino;
 935         min_key.type = BTRFS_EXTENT_DATA_KEY;
 936         min_key.offset = *off;
 937
 938         while (1) {
 939                 path->keep_locks = 1;
 940                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 941                 if (ret != 0)
 942                         goto none;
 943                 path->keep_locks = 0;
 944                 btrfs_unlock_up_safe(path, 1);
 945 process_slot:
 946                 if (min_key.objectid != ino)
 947                         goto none;
 948                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 949                         goto none;
 950
 951                 leaf = path->nodes[0];
 952                 extent = btrfs_item_ptr(leaf, path->slots[0],
 953                                         struct btrfs_file_extent_item);
 954
 955                 type = btrfs_file_extent_type(leaf, extent);
 956                 if (type == BTRFS_FILE_EXTENT_REG &&
 957                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 958                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 959                         *off = min_key.offset;
 960                         btrfs_free_path(path);
 961                         return 0;
 962                 }
 963
 964                 path->slots[0]++;
 965                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 966                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 967                         goto process_slot;
 968                 }
 969
 970                 if (min_key.offset == (u64)-1)
 971                         goto none;
 972
 973                 min_key.offset++;
 974                 btrfs_release_path(path);
 975         }
 976 none:
 977         btrfs_free_path(path);
 978         return -ENOENT;
 979 }
 980
 981 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 982 {
 983         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 984         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 985         struct extent_map *em;
 986         u64 len = PAGE_CACHE_SIZE;
 987
 988         /*
 989          * hopefully we have this extent in the tree already, try without
 990          * the full extent lock
 991          */
 992         read_lock(&em_tree->lock);
 993         em = lookup_extent_mapping(em_tree, start, len);
 994         read_unlock(&em_tree->lock);
 995
 996         if (!em) {
 997                 struct extent_state *cached = NULL;
 998                 u64 end = start + len - 1;
 999
1000                 /* get the big lock and read metadata off disk */
1001                 lock_extent_bits(io_tree, start, end, 0, &cached);
1002                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1003                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1004
1005                 if (IS_ERR(em))
1006                         return NULL;
1007         }
1008
1009         return em;
1010 }
1011
1012 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1013 {
1014         struct extent_map *next;
1015         bool ret = true;
1016
1017         /* this is the last extent */
1018         if (em->start + em->len >= i_size_read(inode))
1019                 return false;
1020
1021         next = defrag_lookup_extent(inode, em->start + em->len);
1022         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1023             (em->block_start + em->block_len == next->block_start))
1024                 ret = false;
1025
1026         free_extent_map(next);
1027         return ret;
1028 }
1029
1030 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1031                                u64 *last_len, u64 *skip, u64 *defrag_end,
1032                                int compress)
1033 {
1034         struct extent_map *em;
1035         int ret = 1;
1036         bool next_mergeable = true;
1037
1038         /*
1039          * make sure that once we start defragging an extent, we keep on
1040          * defragging it
1041          */
1042         if (start < *defrag_end)
1043                 return 1;
1044
1045         *skip = 0;
1046
1047         em = defrag_lookup_extent(inode, start);
1048         if (!em)
1049                 return 0;
1050
1051         /* this will cover holes, and inline extents */
1052         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1053                 ret = 0;
1054                 goto out;
1055         }
1056
1057         next_mergeable = defrag_check_next_extent(inode, em);
1058
1059         /*
1060          * we hit a real extent, if it is big or the next extent is not a
1061          * real extent, don't bother defragging it
1062          */
1063         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1064             (em->len >= thresh || !next_mergeable))
1065                 ret = 0;
1066 out:
1067         /*
1068          * last_len ends up being a counter of how many bytes we've defragged.
1069          * every time we choose not to defrag an extent, we reset *last_len
1070          * so that the next tiny extent will force a defrag.
1071          *
1072          * The end result of this is that tiny extents before a single big
1073          * extent will force at least part of that big extent to be defragged.
1074          */
1075         if (ret) {
1076                 *defrag_end = extent_map_end(em);
1077         } else {
1078                 *last_len = 0;
1079                 *skip = extent_map_end(em);
1080                 *defrag_end = 0;
1081         }
1082
1083         free_extent_map(em);
1084         return ret;
1085 }
1086
1087 /*
1088  * it doesn't do much good to defrag one or two pages
1089  * at a time.  This pulls in a nice chunk of pages
1090  * to COW and defrag.
1091  *
1092  * It also makes sure the delalloc code has enough
1093  * dirty data to avoid making new small extents as part
1094  * of the defrag
1095  *
1096  * It's a good idea to start RA on this range
1097  * before calling this.
1098  */
1099 static int cluster_pages_for_defrag(struct inode *inode,
1100                                     struct page **pages,
1101                                     unsigned long start_index,
1102                                     unsigned long num_pages)
1103 {
1104         unsigned long file_end;
1105         u64 isize = i_size_read(inode);
1106         u64 page_start;
1107         u64 page_end;
1108         u64 page_cnt;
1109         int ret;
1110         int i;
1111         int i_done;
1112         struct btrfs_ordered_extent *ordered;
1113         struct extent_state *cached_state = NULL;
1114         struct extent_io_tree *tree;
1115         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1116
1117         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1118         if (!isize || start_index > file_end)
1119                 return 0;
1120
1121         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1122
1123         ret = btrfs_delalloc_reserve_space(inode,
1124                                            page_cnt << PAGE_CACHE_SHIFT);
1125         if (ret)
1126                 return ret;
1127         i_done = 0;
1128         tree = &BTRFS_I(inode)->io_tree;
1129
1130         /* step one, lock all the pages */
1131         for (i = 0; i < page_cnt; i++) {
1132                 struct page *page;
1133 again:
1134                 page = find_or_create_page(inode->i_mapping,
1135                                            start_index + i, mask);
1136                 if (!page)
1137                         break;
1138
1139                 page_start = page_offset(page);
1140                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1141                 while (1) {
1142                         lock_extent_bits(tree, page_start, page_end,
1143                                          0, &cached_state);
1144                         ordered = btrfs_lookup_ordered_extent(inode,
1145                                                               page_start);
1146                         unlock_extent_cached(tree, page_start, page_end,
1147                                              &cached_state, GFP_NOFS);
1148                         if (!ordered)
1149                                 break;
1150
1151                         unlock_page(page);
1152                         btrfs_start_ordered_extent(inode, ordered, 1);
1153                         btrfs_put_ordered_extent(ordered);
1154                         lock_page(page);
1155                         /*
1156                          * we unlocked the page above, so we need check if
1157                          * it was released or not.
1158                          */
1159                         if (page->mapping != inode->i_mapping) {
1160                                 unlock_page(page);
1161                                 page_cache_release(page);
1162                                 goto again;
1163                         }
1164                 }
1165
1166                 if (!PageUptodate(page)) {
1167                         btrfs_readpage(NULL, page);
1168                         lock_page(page);
1169                         if (!PageUptodate(page)) {
1170                                 unlock_page(page);
1171                                 page_cache_release(page);
1172                                 ret = -EIO;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (page->mapping != inode->i_mapping) {
1178                         unlock_page(page);
1179                         page_cache_release(page);
1180                         goto again;
1181                 }
1182
1183                 pages[i] = page;
1184                 i_done++;
1185         }
1186         if (!i_done || ret)
1187                 goto out;
1188
1189         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1190                 goto out;
1191
1192         /*
1193          * so now we have a nice long stream of locked
1194          * and up to date pages, lets wait on them
1195          */
1196         for (i = 0; i < i_done; i++)
1197                 wait_on_page_writeback(pages[i]);
1198
1199         page_start = page_offset(pages[0]);
1200         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1201
1202         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1203                          page_start, page_end - 1, 0, &cached_state);
1204         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1205                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1207                           &cached_state, GFP_NOFS);
1208
1209         if (i_done != page_cnt) {
1210                 spin_lock(&BTRFS_I(inode)->lock);
1211                 BTRFS_I(inode)->outstanding_extents++;
1212                 spin_unlock(&BTRFS_I(inode)->lock);
1213                 btrfs_delalloc_release_space(inode,
1214                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1215         }
1216
1217
1218         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1219                           &cached_state, GFP_NOFS);
1220
1221         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1222                              page_start, page_end - 1, &cached_state,
1223                              GFP_NOFS);
1224
1225         for (i = 0; i < i_done; i++) {
1226                 clear_page_dirty_for_io(pages[i]);
1227                 ClearPageChecked(pages[i]);
1228                 set_page_extent_mapped(pages[i]);
1229                 set_page_dirty(pages[i]);
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         return i_done;
1234 out:
1235         for (i = 0; i < i_done; i++) {
1236                 unlock_page(pages[i]);
1237                 page_cache_release(pages[i]);
1238         }
1239         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1240         return ret;
1241
1242 }
1243
1244 int btrfs_defrag_file(struct inode *inode, struct file *file,
1245                       struct btrfs_ioctl_defrag_range_args *range,
1246                       u64 newer_than, unsigned long max_to_defrag)
1247 {
1248         struct btrfs_root *root = BTRFS_I(inode)->root;
1249         struct file_ra_state *ra = NULL;
1250         unsigned long last_index;
1251         u64 isize = i_size_read(inode);
1252         u64 last_len = 0;
1253         u64 skip = 0;
1254         u64 defrag_end = 0;
1255         u64 newer_off = range->start;
1256         unsigned long i;
1257         unsigned long ra_index = 0;
1258         int ret;
1259         int defrag_count = 0;
1260         int compress_type = BTRFS_COMPRESS_ZLIB;
1261         int extent_thresh = range->extent_thresh;
1262         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1263         unsigned long cluster = max_cluster;
1264         u64 new_align = ~((u64)128 * 1024 - 1);
1265         struct page **pages = NULL;
1266
1267         if (isize == 0)
1268                 return 0;
1269
1270         if (range->start >= isize)
1271                 return -EINVAL;
1272
1273         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1274                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1275                         return -EINVAL;
1276                 if (range->compress_type)
1277                         compress_type = range->compress_type;
1278         }
1279
1280         if (extent_thresh == 0)
1281                 extent_thresh = 256 * 1024;
1282
1283         /*
1284          * if we were not given a file, allocate a readahead
1285          * context
1286          */
1287         if (!file) {
1288                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1289                 if (!ra)
1290                         return -ENOMEM;
1291                 file_ra_state_init(ra, inode->i_mapping);
1292         } else {
1293                 ra = &file->f_ra;
1294         }
1295
1296         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1297                         GFP_NOFS);
1298         if (!pages) {
1299                 ret = -ENOMEM;
1300                 goto out_ra;
1301         }
1302
1303         /* find the last page to defrag */
1304         if (range->start + range->len > range->start) {
1305                 last_index = min_t(u64, isize - 1,
1306                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1307         } else {
1308                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1309         }
1310
1311         if (newer_than) {
1312                 ret = find_new_extents(root, inode, newer_than,
1313                                        &newer_off, 64 * 1024);
1314                 if (!ret) {
1315                         range->start = newer_off;
1316                         /*
1317                          * we always align our defrag to help keep
1318                          * the extents in the file evenly spaced
1319                          */
1320                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1321                 } else
1322                         goto out_ra;
1323         } else {
1324                 i = range->start >> PAGE_CACHE_SHIFT;
1325         }
1326         if (!max_to_defrag)
1327                 max_to_defrag = last_index + 1;
1328
1329         /*
1330          * make writeback starts from i, so the defrag range can be
1331          * written sequentially.
1332          */
1333         if (i < inode->i_mapping->writeback_index)
1334                 inode->i_mapping->writeback_index = i;
1335
1336         while (i <= last_index && defrag_count < max_to_defrag &&
1337                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1338                 PAGE_CACHE_SHIFT)) {
1339                 /*
1340                  * make sure we stop running if someone unmounts
1341                  * the FS
1342                  */
1343                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1344                         break;
1345
1346                 if (btrfs_defrag_cancelled(root->fs_info)) {
1347                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1348                         ret = -EAGAIN;
1349                         break;
1350                 }
1351
1352                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1353                                          extent_thresh, &last_len, &skip,
1354                                          &defrag_end, range->flags &
1355                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1356                         unsigned long next;
1357                         /*
1358                          * the should_defrag function tells us how much to skip
1359                          * bump our counter by the suggested amount
1360                          */
1361                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1362                         i = max(i + 1, next);
1363                         continue;
1364                 }
1365
1366                 if (!newer_than) {
1367                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1368                                    PAGE_CACHE_SHIFT) - i;
1369                         cluster = min(cluster, max_cluster);
1370                 } else {
1371                         cluster = max_cluster;
1372                 }
1373
1374                 if (i + cluster > ra_index) {
1375                         ra_index = max(i, ra_index);
1376                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1377                                        cluster);
1378                         ra_index += max_cluster;
1379                 }
1380
1381                 mutex_lock(&inode->i_mutex);
1382                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1383                         BTRFS_I(inode)->force_compress = compress_type;
1384                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1385                 if (ret < 0) {
1386                         mutex_unlock(&inode->i_mutex);
1387                         goto out_ra;
1388                 }
1389
1390                 defrag_count += ret;
1391                 balance_dirty_pages_ratelimited(inode->i_mapping);
1392                 mutex_unlock(&inode->i_mutex);
1393
1394                 if (newer_than) {
1395                         if (newer_off == (u64)-1)
1396                                 break;
1397
1398                         if (ret > 0)
1399                                 i += ret;
1400
1401                         newer_off = max(newer_off + 1,
1402                                         (u64)i << PAGE_CACHE_SHIFT);
1403
1404                         ret = find_new_extents(root, inode,
1405                                                newer_than, &newer_off,
1406                                                64 * 1024);
1407                         if (!ret) {
1408                                 range->start = newer_off;
1409                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 break;
1412                         }
1413                 } else {
1414                         if (ret > 0) {
1415                                 i += ret;
1416                                 last_len += ret << PAGE_CACHE_SHIFT;
1417                         } else {
1418                                 i++;
1419                                 last_len = 0;
1420                         }
1421                 }
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1425                 filemap_flush(inode->i_mapping);
1426                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1427                              &BTRFS_I(inode)->runtime_flags))
1428                         filemap_flush(inode->i_mapping);
1429         }
1430
1431         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1432                 /* the filemap_flush will queue IO into the worker threads, but
1433                  * we have to make sure the IO is actually started and that
1434                  * ordered extents get created before we return
1435                  */
1436                 atomic_inc(&root->fs_info->async_submit_draining);
1437                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1438                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1439                         wait_event(root->fs_info->async_submit_wait,
1440                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1441                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1442                 }
1443                 atomic_dec(&root->fs_info->async_submit_draining);
1444         }
1445
1446         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1447                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1448         }
1449
1450         ret = defrag_count;
1451
1452 out_ra:
1453         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1454                 mutex_lock(&inode->i_mutex);
1455                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1456                 mutex_unlock(&inode->i_mutex);
1457         }
1458         if (!file)
1459                 kfree(ra);
1460         kfree(pages);
1461         return ret;
1462 }
1463
1464 static noinline int btrfs_ioctl_resize(struct file *file,
1465                                         void __user *arg)
1466 {
1467         u64 new_size;
1468         u64 old_size;
1469         u64 devid = 1;
1470         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1471         struct btrfs_ioctl_vol_args *vol_args;
1472         struct btrfs_trans_handle *trans;
1473         struct btrfs_device *device = NULL;
1474         char *sizestr;
1475         char *retptr;
1476         char *devstr = NULL;
1477         int ret = 0;
1478         int mod = 0;
1479
1480         if (!capable(CAP_SYS_ADMIN))
1481                 return -EPERM;
1482
1483         ret = mnt_want_write_file(file);
1484         if (ret)
1485                 return ret;
1486
1487         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1488                         1)) {
1489                 mnt_drop_write_file(file);
1490                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1491         }
1492
1493         mutex_lock(&root->fs_info->volume_mutex);
1494         vol_args = memdup_user(arg, sizeof(*vol_args));
1495         if (IS_ERR(vol_args)) {
1496                 ret = PTR_ERR(vol_args);
1497                 goto out;
1498         }
1499
1500         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1501
1502         sizestr = vol_args->name;
1503         devstr = strchr(sizestr, ':');
1504         if (devstr) {
1505                 char *end;
1506                 sizestr = devstr + 1;
1507                 *devstr = '\0';
1508                 devstr = vol_args->name;
1509                 devid = simple_strtoull(devstr, &end, 10);
1510                 if (!devid) {
1511                         ret = -EINVAL;
1512                         goto out_free;
1513                 }
1514                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1515         }
1516
1517         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1518         if (!device) {
1519                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1520                        devid);
1521                 ret = -ENODEV;
1522                 goto out_free;
1523         }
1524
1525         if (!device->writeable) {
1526                 btrfs_info(root->fs_info,
1527                            "resizer unable to apply on readonly device %llu",
1528                        devid);
1529                 ret = -EPERM;
1530                 goto out_free;
1531         }
1532
1533         if (!strcmp(sizestr, "max"))
1534                 new_size = device->bdev->bd_inode->i_size;
1535         else {
1536                 if (sizestr[0] == '-') {
1537                         mod = -1;
1538                         sizestr++;
1539                 } else if (sizestr[0] == '+') {
1540                         mod = 1;
1541                         sizestr++;
1542                 }
1543                 new_size = memparse(sizestr, &retptr);
1544                 if (*retptr != '\0' || new_size == 0) {
1545                         ret = -EINVAL;
1546                         goto out_free;
1547                 }
1548         }
1549
1550         if (device->is_tgtdev_for_dev_replace) {
1551                 ret = -EPERM;
1552                 goto out_free;
1553         }
1554
1555         old_size = device->total_bytes;
1556
1557         if (mod < 0) {
1558                 if (new_size > old_size) {
1559                         ret = -EINVAL;
1560                         goto out_free;
1561                 }
1562                 new_size = old_size - new_size;
1563         } else if (mod > 0) {
1564                 if (new_size > ULLONG_MAX - old_size) {
1565                         ret = -EINVAL;
1566                         goto out_free;
1567                 }
1568                 new_size = old_size + new_size;
1569         }
1570
1571         if (new_size < 256 * 1024 * 1024) {
1572                 ret = -EINVAL;
1573                 goto out_free;
1574         }
1575         if (new_size > device->bdev->bd_inode->i_size) {
1576                 ret = -EFBIG;
1577                 goto out_free;
1578         }
1579
1580         do_div(new_size, root->sectorsize);
1581         new_size *= root->sectorsize;
1582
1583         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1584                       rcu_str_deref(device->name), new_size);
1585
1586         if (new_size > old_size) {
1587                 trans = btrfs_start_transaction(root, 0);
1588                 if (IS_ERR(trans)) {
1589                         ret = PTR_ERR(trans);
1590                         goto out_free;
1591                 }
1592                 ret = btrfs_grow_device(trans, device, new_size);
1593                 btrfs_commit_transaction(trans, root);
1594         } else if (new_size < old_size) {
1595                 ret = btrfs_shrink_device(device, new_size);
1596         } /* equal, nothing need to do */
1597
1598 out_free:
1599         kfree(vol_args);
1600 out:
1601         mutex_unlock(&root->fs_info->volume_mutex);
1602         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1603         mnt_drop_write_file(file);
1604         return ret;
1605 }
1606
1607 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1608                                 char *name, unsigned long fd, int subvol,
1609                                 u64 *transid, bool readonly,
1610                                 struct btrfs_qgroup_inherit *inherit)
1611 {
1612         int namelen;
1613         int ret = 0;
1614
1615         ret = mnt_want_write_file(file);
1616         if (ret)
1617                 goto out;
1618
1619         namelen = strlen(name);
1620         if (strchr(name, '/')) {
1621                 ret = -EINVAL;
1622                 goto out_drop_write;
1623         }
1624
1625         if (name[0] == '.' &&
1626            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1627                 ret = -EEXIST;
1628                 goto out_drop_write;
1629         }
1630
1631         if (subvol) {
1632                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1633                                      NULL, transid, readonly, inherit);
1634         } else {
1635                 struct fd src = fdget(fd);
1636                 struct inode *src_inode;
1637                 if (!src.file) {
1638                         ret = -EINVAL;
1639                         goto out_drop_write;
1640                 }
1641
1642                 src_inode = file_inode(src.file);
1643                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1644                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1645                                    "Snapshot src from another FS");
1646                         ret = -EXDEV;
1647                 } else if (!inode_owner_or_capable(src_inode)) {
1648                         /*
1649                          * Subvolume creation is not restricted, but snapshots
1650                          * are limited to own subvolumes only
1651                          */
1652                         ret = -EPERM;
1653                 } else {
1654                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1655                                              BTRFS_I(src_inode)->root,
1656                                              transid, readonly, inherit);
1657                 }
1658                 fdput(src);
1659         }
1660 out_drop_write:
1661         mnt_drop_write_file(file);
1662 out:
1663         return ret;
1664 }
1665
1666 static noinline int btrfs_ioctl_snap_create(struct file *file,
1667                                             void __user *arg, int subvol)
1668 {
1669         struct btrfs_ioctl_vol_args *vol_args;
1670         int ret;
1671
1672         vol_args = memdup_user(arg, sizeof(*vol_args));
1673         if (IS_ERR(vol_args))
1674                 return PTR_ERR(vol_args);
1675         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1676
1677         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1678                                               vol_args->fd, subvol,
1679                                               NULL, false, NULL);
1680
1681         kfree(vol_args);
1682         return ret;
1683 }
1684
1685 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1686                                                void __user *arg, int subvol)
1687 {
1688         struct btrfs_ioctl_vol_args_v2 *vol_args;
1689         int ret;
1690         u64 transid = 0;
1691         u64 *ptr = NULL;
1692         bool readonly = false;
1693         struct btrfs_qgroup_inherit *inherit = NULL;
1694
1695         vol_args = memdup_user(arg, sizeof(*vol_args));
1696         if (IS_ERR(vol_args))
1697                 return PTR_ERR(vol_args);
1698         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1699
1700         if (vol_args->flags &
1701             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1702               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1703                 ret = -EOPNOTSUPP;
1704                 goto out;
1705         }
1706
1707         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1708                 ptr = &transid;
1709         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1710                 readonly = true;
1711         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1712                 if (vol_args->size > PAGE_CACHE_SIZE) {
1713                         ret = -EINVAL;
1714                         goto out;
1715                 }
1716                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1717                 if (IS_ERR(inherit)) {
1718                         ret = PTR_ERR(inherit);
1719                         goto out;
1720                 }
1721         }
1722
1723         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1724                                               vol_args->fd, subvol, ptr,
1725                                               readonly, inherit);
1726
1727         if (ret == 0 && ptr &&
1728             copy_to_user(arg +
1729                          offsetof(struct btrfs_ioctl_vol_args_v2,
1730                                   transid), ptr, sizeof(*ptr)))
1731                 ret = -EFAULT;
1732 out:
1733         kfree(vol_args);
1734         kfree(inherit);
1735         return ret;
1736 }
1737
1738 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1739                                                 void __user *arg)
1740 {
1741         struct inode *inode = file_inode(file);
1742         struct btrfs_root *root = BTRFS_I(inode)->root;
1743         int ret = 0;
1744         u64 flags = 0;
1745
1746         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1747                 return -EINVAL;
1748
1749         down_read(&root->fs_info->subvol_sem);
1750         if (btrfs_root_readonly(root))
1751                 flags |= BTRFS_SUBVOL_RDONLY;
1752         up_read(&root->fs_info->subvol_sem);
1753
1754         if (copy_to_user(arg, &flags, sizeof(flags)))
1755                 ret = -EFAULT;
1756
1757         return ret;
1758 }
1759
1760 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1761                                               void __user *arg)
1762 {
1763         struct inode *inode = file_inode(file);
1764         struct btrfs_root *root = BTRFS_I(inode)->root;
1765         struct btrfs_trans_handle *trans;
1766         u64 root_flags;
1767         u64 flags;
1768         int ret = 0;
1769
1770         if (!inode_owner_or_capable(inode))
1771                 return -EPERM;
1772
1773         ret = mnt_want_write_file(file);
1774         if (ret)
1775                 goto out;
1776
1777         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1778                 ret = -EINVAL;
1779                 goto out_drop_write;
1780         }
1781
1782         if (copy_from_user(&flags, arg, sizeof(flags))) {
1783                 ret = -EFAULT;
1784                 goto out_drop_write;
1785         }
1786
1787         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1788                 ret = -EINVAL;
1789                 goto out_drop_write;
1790         }
1791
1792         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1793                 ret = -EOPNOTSUPP;
1794                 goto out_drop_write;
1795         }
1796
1797         down_write(&root->fs_info->subvol_sem);
1798
1799         /* nothing to do */
1800         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1801                 goto out_drop_sem;
1802
1803         root_flags = btrfs_root_flags(&root->root_item);
1804         if (flags & BTRFS_SUBVOL_RDONLY) {
1805                 btrfs_set_root_flags(&root->root_item,
1806                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1807         } else {
1808                 /*
1809                  * Block RO -> RW transition if this subvolume is involved in
1810                  * send
1811                  */
1812                 spin_lock(&root->root_item_lock);
1813                 if (root->send_in_progress == 0) {
1814                         btrfs_set_root_flags(&root->root_item,
1815                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1816                         spin_unlock(&root->root_item_lock);
1817                 } else {
1818                         spin_unlock(&root->root_item_lock);
1819                         btrfs_warn(root->fs_info,
1820                         "Attempt to set subvolume %llu read-write during send",
1821                                         root->root_key.objectid);
1822                         ret = -EPERM;
1823                         goto out_drop_sem;
1824                 }
1825         }
1826
1827         trans = btrfs_start_transaction(root, 1);
1828         if (IS_ERR(trans)) {
1829                 ret = PTR_ERR(trans);
1830                 goto out_reset;
1831         }
1832
1833         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1834                                 &root->root_key, &root->root_item);
1835
1836         btrfs_commit_transaction(trans, root);
1837 out_reset:
1838         if (ret)
1839                 btrfs_set_root_flags(&root->root_item, root_flags);
1840 out_drop_sem:
1841         up_write(&root->fs_info->subvol_sem);
1842 out_drop_write:
1843         mnt_drop_write_file(file);
1844 out:
1845         return ret;
1846 }
1847
1848 /*
1849  * helper to check if the subvolume references other subvolumes
1850  */
1851 static noinline int may_destroy_subvol(struct btrfs_root *root)
1852 {
1853         struct btrfs_path *path;
1854         struct btrfs_dir_item *di;
1855         struct btrfs_key key;
1856         u64 dir_id;
1857         int ret;
1858
1859         path = btrfs_alloc_path();
1860         if (!path)
1861                 return -ENOMEM;
1862
1863         /* Make sure this root isn't set as the default subvol */
1864         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1865         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1866                                    dir_id, "default", 7, 0);
1867         if (di && !IS_ERR(di)) {
1868                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1869                 if (key.objectid == root->root_key.objectid) {
1870                         ret = -EPERM;
1871                         btrfs_err(root->fs_info, "deleting default subvolume "
1872                                   "%llu is not allowed", key.objectid);
1873                         goto out;
1874                 }
1875                 btrfs_release_path(path);
1876         }
1877
1878         key.objectid = root->root_key.objectid;
1879         key.type = BTRFS_ROOT_REF_KEY;
1880         key.offset = (u64)-1;
1881
1882         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1883                                 &key, path, 0, 0);
1884         if (ret < 0)
1885                 goto out;
1886         BUG_ON(ret == 0);
1887
1888         ret = 0;
1889         if (path->slots[0] > 0) {
1890                 path->slots[0]--;
1891                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1892                 if (key.objectid == root->root_key.objectid &&
1893                     key.type == BTRFS_ROOT_REF_KEY)
1894                         ret = -ENOTEMPTY;
1895         }
1896 out:
1897         btrfs_free_path(path);
1898         return ret;
1899 }
1900
1901 static noinline int key_in_sk(struct btrfs_key *key,
1902                               struct btrfs_ioctl_search_key *sk)
1903 {
1904         struct btrfs_key test;
1905         int ret;
1906
1907         test.objectid = sk->min_objectid;
1908         test.type = sk->min_type;
1909         test.offset = sk->min_offset;
1910
1911         ret = btrfs_comp_cpu_keys(key, &test);
1912         if (ret < 0)
1913                 return 0;
1914
1915         test.objectid = sk->max_objectid;
1916         test.type = sk->max_type;
1917         test.offset = sk->max_offset;
1918
1919         ret = btrfs_comp_cpu_keys(key, &test);
1920         if (ret > 0)
1921                 return 0;
1922         return 1;
1923 }
1924
1925 static noinline int copy_to_sk(struct btrfs_root *root,
1926                                struct btrfs_path *path,
1927                                struct btrfs_key *key,
1928                                struct btrfs_ioctl_search_key *sk,
1929                                char *buf,
1930                                unsigned long *sk_offset,
1931                                int *num_found)
1932 {
1933         u64 found_transid;
1934         struct extent_buffer *leaf;
1935         struct btrfs_ioctl_search_header sh;
1936         unsigned long item_off;
1937         unsigned long item_len;
1938         int nritems;
1939         int i;
1940         int slot;
1941         int ret = 0;
1942
1943         leaf = path->nodes[0];
1944         slot = path->slots[0];
1945         nritems = btrfs_header_nritems(leaf);
1946
1947         if (btrfs_header_generation(leaf) > sk->max_transid) {
1948                 i = nritems;
1949                 goto advance_key;
1950         }
1951         found_transid = btrfs_header_generation(leaf);
1952
1953         for (i = slot; i < nritems; i++) {
1954                 item_off = btrfs_item_ptr_offset(leaf, i);
1955                 item_len = btrfs_item_size_nr(leaf, i);
1956
1957                 btrfs_item_key_to_cpu(leaf, key, i);
1958                 if (!key_in_sk(key, sk))
1959                         continue;
1960
1961                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1962                         item_len = 0;
1963
1964                 if (sizeof(sh) + item_len + *sk_offset >
1965                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1966                         ret = 1;
1967                         goto overflow;
1968                 }
1969
1970                 sh.objectid = key->objectid;
1971                 sh.offset = key->offset;
1972                 sh.type = key->type;
1973                 sh.len = item_len;
1974                 sh.transid = found_transid;
1975
1976                 /* copy search result header */
1977                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1978                 *sk_offset += sizeof(sh);
1979
1980                 if (item_len) {
1981                         char *p = buf + *sk_offset;
1982                         /* copy the item */
1983                         read_extent_buffer(leaf, p,
1984                                            item_off, item_len);
1985                         *sk_offset += item_len;
1986                 }
1987                 (*num_found)++;
1988
1989                 if (*num_found >= sk->nr_items)
1990                         break;
1991         }
1992 advance_key:
1993         ret = 0;
1994         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1995                 key->offset++;
1996         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1997                 key->offset = 0;
1998                 key->type++;
1999         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2000                 key->offset = 0;
2001                 key->type = 0;
2002                 key->objectid++;
2003         } else
2004                 ret = 1;
2005 overflow:
2006         return ret;
2007 }
2008
2009 static noinline int search_ioctl(struct inode *inode,
2010                                  struct btrfs_ioctl_search_args *args)
2011 {
2012         struct btrfs_root *root;
2013         struct btrfs_key key;
2014         struct btrfs_path *path;
2015         struct btrfs_ioctl_search_key *sk = &args->key;
2016         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2017         int ret;
2018         int num_found = 0;
2019         unsigned long sk_offset = 0;
2020
2021         path = btrfs_alloc_path();
2022         if (!path)
2023                 return -ENOMEM;
2024
2025         if (sk->tree_id == 0) {
2026                 /* search the root of the inode that was passed */
2027                 root = BTRFS_I(inode)->root;
2028         } else {
2029                 key.objectid = sk->tree_id;
2030                 key.type = BTRFS_ROOT_ITEM_KEY;
2031                 key.offset = (u64)-1;
2032                 root = btrfs_read_fs_root_no_name(info, &key);
2033                 if (IS_ERR(root)) {
2034                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2035                                sk->tree_id);
2036                         btrfs_free_path(path);
2037                         return -ENOENT;
2038                 }
2039         }
2040
2041         key.objectid = sk->min_objectid;
2042         key.type = sk->min_type;
2043         key.offset = sk->min_offset;
2044
2045         path->keep_locks = 1;
2046
2047         while (1) {
2048                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2049                 if (ret != 0) {
2050                         if (ret > 0)
2051                                 ret = 0;
2052                         goto err;
2053                 }
2054                 ret = copy_to_sk(root, path, &key, sk, args->buf,
2055                                  &sk_offset, &num_found);
2056                 btrfs_release_path(path);
2057                 if (ret || num_found >= sk->nr_items)
2058                         break;
2059
2060         }
2061         ret = 0;
2062 err:
2063         sk->nr_items = num_found;
2064         btrfs_free_path(path);
2065         return ret;
2066 }
2067
2068 static noinline int btrfs_ioctl_tree_search(struct file *file,
2069                                            void __user *argp)
2070 {
2071          struct btrfs_ioctl_search_args *args;
2072          struct inode *inode;
2073          int ret;
2074
2075         if (!capable(CAP_SYS_ADMIN))
2076                 return -EPERM;
2077
2078         args = memdup_user(argp, sizeof(*args));
2079         if (IS_ERR(args))
2080                 return PTR_ERR(args);
2081
2082         inode = file_inode(file);
2083         ret = search_ioctl(inode, args);
2084         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2085                 ret = -EFAULT;
2086         kfree(args);
2087         return ret;
2088 }
2089
2090 /*
2091  * Search INODE_REFs to identify path name of 'dirid' directory
2092  * in a 'tree_id' tree. and sets path name to 'name'.
2093  */
2094 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2095                                 u64 tree_id, u64 dirid, char *name)
2096 {
2097         struct btrfs_root *root;
2098         struct btrfs_key key;
2099         char *ptr;
2100         int ret = -1;
2101         int slot;
2102         int len;
2103         int total_len = 0;
2104         struct btrfs_inode_ref *iref;
2105         struct extent_buffer *l;
2106         struct btrfs_path *path;
2107
2108         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2109                 name[0]='\0';
2110                 return 0;
2111         }
2112
2113         path = btrfs_alloc_path();
2114         if (!path)
2115                 return -ENOMEM;
2116
2117         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2118
2119         key.objectid = tree_id;
2120         key.type = BTRFS_ROOT_ITEM_KEY;
2121         key.offset = (u64)-1;
2122         root = btrfs_read_fs_root_no_name(info, &key);
2123         if (IS_ERR(root)) {
2124                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2125                 ret = -ENOENT;
2126                 goto out;
2127         }
2128
2129         key.objectid = dirid;
2130         key.type = BTRFS_INODE_REF_KEY;
2131         key.offset = (u64)-1;
2132
2133         while (1) {
2134                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2135                 if (ret < 0)
2136                         goto out;
2137                 else if (ret > 0) {
2138                         ret = btrfs_previous_item(root, path, dirid,
2139                                                   BTRFS_INODE_REF_KEY);
2140                         if (ret < 0)
2141                                 goto out;
2142                         else if (ret > 0) {
2143                                 ret = -ENOENT;
2144                                 goto out;
2145                         }
2146                 }
2147
2148                 l = path->nodes[0];
2149                 slot = path->slots[0];
2150                 btrfs_item_key_to_cpu(l, &key, slot);
2151
2152                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2153                 len = btrfs_inode_ref_name_len(l, iref);
2154                 ptr -= len + 1;
2155                 total_len += len + 1;
2156                 if (ptr < name) {
2157                         ret = -ENAMETOOLONG;
2158                         goto out;
2159                 }
2160
2161                 *(ptr + len) = '/';
2162                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2163
2164                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2165                         break;
2166
2167                 btrfs_release_path(path);
2168                 key.objectid = key.offset;
2169                 key.offset = (u64)-1;
2170                 dirid = key.objectid;
2171         }
2172         memmove(name, ptr, total_len);
2173         name[total_len] = '\0';
2174         ret = 0;
2175 out:
2176         btrfs_free_path(path);
2177         return ret;
2178 }
2179
2180 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2181                                            void __user *argp)
2182 {
2183          struct btrfs_ioctl_ino_lookup_args *args;
2184          struct inode *inode;
2185          int ret;
2186
2187         if (!capable(CAP_SYS_ADMIN))
2188                 return -EPERM;
2189
2190         args = memdup_user(argp, sizeof(*args));
2191         if (IS_ERR(args))
2192                 return PTR_ERR(args);
2193
2194         inode = file_inode(file);
2195
2196         if (args->treeid == 0)
2197                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2198
2199         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2200                                         args->treeid, args->objectid,
2201                                         args->name);
2202
2203         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2204                 ret = -EFAULT;
2205
2206         kfree(args);
2207         return ret;
2208 }
2209
2210 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2211                                              void __user *arg)
2212 {
2213         struct dentry *parent = file->f_path.dentry;
2214         struct dentry *dentry;
2215         struct inode *dir = parent->d_inode;
2216         struct inode *inode;
2217         struct btrfs_root *root = BTRFS_I(dir)->root;
2218         struct btrfs_root *dest = NULL;
2219         struct btrfs_ioctl_vol_args *vol_args;
2220         struct btrfs_trans_handle *trans;
2221         struct btrfs_block_rsv block_rsv;
2222         u64 root_flags;
2223         u64 qgroup_reserved;
2224         int namelen;
2225         int ret;
2226         int err = 0;
2227
2228         vol_args = memdup_user(arg, sizeof(*vol_args));
2229         if (IS_ERR(vol_args))
2230                 return PTR_ERR(vol_args);
2231
2232         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2233         namelen = strlen(vol_args->name);
2234         if (strchr(vol_args->name, '/') ||
2235             strncmp(vol_args->name, "..", namelen) == 0) {
2236                 err = -EINVAL;
2237                 goto out;
2238         }
2239
2240         err = mnt_want_write_file(file);
2241         if (err)
2242                 goto out;
2243
2244
2245         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2246         if (err == -EINTR)
2247                 goto out_drop_write;
2248         dentry = lookup_one_len(vol_args->name, parent, namelen);
2249         if (IS_ERR(dentry)) {
2250                 err = PTR_ERR(dentry);
2251                 goto out_unlock_dir;
2252         }
2253
2254         if (!dentry->d_inode) {
2255                 err = -ENOENT;
2256                 goto out_dput;
2257         }
2258
2259         inode = dentry->d_inode;
2260         dest = BTRFS_I(inode)->root;
2261         if (!capable(CAP_SYS_ADMIN)) {
2262                 /*
2263                  * Regular user.  Only allow this with a special mount
2264                  * option, when the user has write+exec access to the
2265                  * subvol root, and when rmdir(2) would have been
2266                  * allowed.
2267                  *
2268                  * Note that this is _not_ check that the subvol is
2269                  * empty or doesn't contain data that we wouldn't
2270                  * otherwise be able to delete.
2271                  *
2272                  * Users who want to delete empty subvols should try
2273                  * rmdir(2).
2274                  */
2275                 err = -EPERM;
2276                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2277                         goto out_dput;
2278
2279                 /*
2280                  * Do not allow deletion if the parent dir is the same
2281                  * as the dir to be deleted.  That means the ioctl
2282                  * must be called on the dentry referencing the root
2283                  * of the subvol, not a random directory contained
2284                  * within it.
2285                  */
2286                 err = -EINVAL;
2287                 if (root == dest)
2288                         goto out_dput;
2289
2290                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2291                 if (err)
2292                         goto out_dput;
2293         }
2294
2295         /* check if subvolume may be deleted by a user */
2296         err = btrfs_may_delete(dir, dentry, 1);
2297         if (err)
2298                 goto out_dput;
2299
2300         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2301                 err = -EINVAL;
2302                 goto out_dput;
2303         }
2304
2305         mutex_lock(&inode->i_mutex);
2306
2307         /*
2308          * Don't allow to delete a subvolume with send in progress. This is
2309          * inside the i_mutex so the error handling that has to drop the bit
2310          * again is not run concurrently.
2311          */
2312         spin_lock(&dest->root_item_lock);
2313         root_flags = btrfs_root_flags(&root->root_item);
2314         if (root->send_in_progress == 0) {
2315                 btrfs_set_root_flags(&root->root_item,
2316                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2317                 spin_unlock(&dest->root_item_lock);
2318         } else {
2319                 spin_unlock(&dest->root_item_lock);
2320                 btrfs_warn(root->fs_info,
2321                         "Attempt to delete subvolume %llu during send",
2322                         root->root_key.objectid);
2323                 err = -EPERM;
2324                 goto out_dput;
2325         }
2326
2327         err = d_invalidate(dentry);
2328         if (err)
2329                 goto out_unlock;
2330
2331         down_write(&root->fs_info->subvol_sem);
2332
2333         err = may_destroy_subvol(dest);
2334         if (err)
2335                 goto out_up_write;
2336
2337         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2338         /*
2339          * One for dir inode, two for dir entries, two for root
2340          * ref/backref.
2341          */
2342         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2343                                                5, &qgroup_reserved, true);
2344         if (err)
2345                 goto out_up_write;
2346
2347         trans = btrfs_start_transaction(root, 0);
2348         if (IS_ERR(trans)) {
2349                 err = PTR_ERR(trans);
2350                 goto out_release;
2351         }
2352         trans->block_rsv = &block_rsv;
2353         trans->bytes_reserved = block_rsv.size;
2354
2355         ret = btrfs_unlink_subvol(trans, root, dir,
2356                                 dest->root_key.objectid,
2357                                 dentry->d_name.name,
2358                                 dentry->d_name.len);
2359         if (ret) {
2360                 err = ret;
2361                 btrfs_abort_transaction(trans, root, ret);
2362                 goto out_end_trans;
2363         }
2364
2365         btrfs_record_root_in_trans(trans, dest);
2366
2367         memset(&dest->root_item.drop_progress, 0,
2368                 sizeof(dest->root_item.drop_progress));
2369         dest->root_item.drop_level = 0;
2370         btrfs_set_root_refs(&dest->root_item, 0);
2371
2372         if (!xchg(&dest->orphan_item_inserted, 1)) {
2373                 ret = btrfs_insert_orphan_item(trans,
2374                                         root->fs_info->tree_root,
2375                                         dest->root_key.objectid);
2376                 if (ret) {
2377                         btrfs_abort_transaction(trans, root, ret);
2378                         err = ret;
2379                         goto out_end_trans;
2380                 }
2381         }
2382
2383         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2384                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2385                                   dest->root_key.objectid);
2386         if (ret && ret != -ENOENT) {
2387                 btrfs_abort_transaction(trans, root, ret);
2388                 err = ret;
2389                 goto out_end_trans;
2390         }
2391         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2392                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2393                                           dest->root_item.received_uuid,
2394                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2395                                           dest->root_key.objectid);
2396                 if (ret && ret != -ENOENT) {
2397                         btrfs_abort_transaction(trans, root, ret);
2398                         err = ret;
2399                         goto out_end_trans;
2400                 }
2401         }
2402
2403 out_end_trans:
2404         trans->block_rsv = NULL;
2405         trans->bytes_reserved = 0;
2406         ret = btrfs_end_transaction(trans, root);
2407         if (ret && !err)
2408                 err = ret;
2409         inode->i_flags |= S_DEAD;
2410 out_release:
2411         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2412 out_up_write:
2413         up_write(&root->fs_info->subvol_sem);
2414 out_unlock:
2415         if (err) {
2416                 spin_lock(&dest->root_item_lock);
2417                 root_flags = btrfs_root_flags(&root->root_item);
2418                 btrfs_set_root_flags(&root->root_item,
2419                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2420                 spin_unlock(&dest->root_item_lock);
2421         }
2422         mutex_unlock(&inode->i_mutex);
2423         if (!err) {
2424                 shrink_dcache_sb(root->fs_info->sb);
2425                 btrfs_invalidate_inodes(dest);
2426                 d_delete(dentry);
2427                 ASSERT(dest->send_in_progress == 0);
2428
2429                 /* the last ref */
2430                 if (dest->cache_inode) {
2431                         iput(dest->cache_inode);
2432                         dest->cache_inode = NULL;
2433                 }
2434         }
2435 out_dput:
2436         dput(dentry);
2437 out_unlock_dir:
2438         mutex_unlock(&dir->i_mutex);
2439 out_drop_write:
2440         mnt_drop_write_file(file);
2441 out:
2442         kfree(vol_args);
2443         return err;
2444 }
2445
2446 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2447 {
2448         struct inode *inode = file_inode(file);
2449         struct btrfs_root *root = BTRFS_I(inode)->root;
2450         struct btrfs_ioctl_defrag_range_args *range;
2451         int ret;
2452
2453         ret = mnt_want_write_file(file);
2454         if (ret)
2455                 return ret;
2456
2457         if (btrfs_root_readonly(root)) {
2458                 ret = -EROFS;
2459                 goto out;
2460         }
2461
2462         switch (inode->i_mode & S_IFMT) {
2463         case S_IFDIR:
2464                 if (!capable(CAP_SYS_ADMIN)) {
2465                         ret = -EPERM;
2466                         goto out;
2467                 }
2468                 ret = btrfs_defrag_root(root);
2469                 if (ret)
2470                         goto out;
2471                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2472                 break;
2473         case S_IFREG:
2474                 if (!(file->f_mode & FMODE_WRITE)) {
2475                         ret = -EINVAL;
2476                         goto out;
2477                 }
2478
2479                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2480                 if (!range) {
2481                         ret = -ENOMEM;
2482                         goto out;
2483                 }
2484
2485                 if (argp) {
2486                         if (copy_from_user(range, argp,
2487                                            sizeof(*range))) {
2488                                 ret = -EFAULT;
2489                                 kfree(range);
2490                                 goto out;
2491                         }
2492                         /* compression requires us to start the IO */
2493                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2494                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2495                                 range->extent_thresh = (u32)-1;
2496                         }
2497                 } else {
2498                         /* the rest are all set to zero by kzalloc */
2499                         range->len = (u64)-1;
2500                 }
2501                 ret = btrfs_defrag_file(file_inode(file), file,
2502                                         range, 0, 0);
2503                 if (ret > 0)
2504                         ret = 0;
2505                 kfree(range);
2506                 break;
2507         default:
2508                 ret = -EINVAL;
2509         }
2510 out:
2511         mnt_drop_write_file(file);
2512         return ret;
2513 }
2514
2515 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2516 {
2517         struct btrfs_ioctl_vol_args *vol_args;
2518         int ret;
2519
2520         if (!capable(CAP_SYS_ADMIN))
2521                 return -EPERM;
2522
2523         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2524                         1)) {
2525                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2526         }
2527
2528         mutex_lock(&root->fs_info->volume_mutex);
2529         vol_args = memdup_user(arg, sizeof(*vol_args));
2530         if (IS_ERR(vol_args)) {
2531                 ret = PTR_ERR(vol_args);
2532                 goto out;
2533         }
2534
2535         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2536         ret = btrfs_init_new_device(root, vol_args->name);
2537
2538         kfree(vol_args);
2539 out:
2540         mutex_unlock(&root->fs_info->volume_mutex);
2541         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2542         return ret;
2543 }
2544
2545 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2546 {
2547         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2548         struct btrfs_ioctl_vol_args *vol_args;
2549         int ret;
2550
2551         if (!capable(CAP_SYS_ADMIN))
2552                 return -EPERM;
2553
2554         ret = mnt_want_write_file(file);
2555         if (ret)
2556                 return ret;
2557
2558         vol_args = memdup_user(arg, sizeof(*vol_args));
2559         if (IS_ERR(vol_args)) {
2560                 ret = PTR_ERR(vol_args);
2561                 goto out;
2562         }
2563
2564         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2565
2566         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2567                         1)) {
2568                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2569                 goto out;
2570         }
2571
2572         mutex_lock(&root->fs_info->volume_mutex);
2573         ret = btrfs_rm_device(root, vol_args->name);
2574         mutex_unlock(&root->fs_info->volume_mutex);
2575         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2576
2577 out:
2578         kfree(vol_args);
2579         mnt_drop_write_file(file);
2580         return ret;
2581 }
2582
2583 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2584 {
2585         struct btrfs_ioctl_fs_info_args *fi_args;
2586         struct btrfs_device *device;
2587         struct btrfs_device *next;
2588         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2589         int ret = 0;
2590
2591         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2592         if (!fi_args)
2593                 return -ENOMEM;
2594
2595         mutex_lock(&fs_devices->device_list_mutex);
2596         fi_args->num_devices = fs_devices->num_devices;
2597         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2598
2599         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2600                 if (device->devid > fi_args->max_id)
2601                         fi_args->max_id = device->devid;
2602         }
2603         mutex_unlock(&fs_devices->device_list_mutex);
2604
2605         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2606         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2607         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2608
2609         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2610                 ret = -EFAULT;
2611
2612         kfree(fi_args);
2613         return ret;
2614 }
2615
2616 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2617 {
2618         struct btrfs_ioctl_dev_info_args *di_args;
2619         struct btrfs_device *dev;
2620         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2621         int ret = 0;
2622         char *s_uuid = NULL;
2623
2624         di_args = memdup_user(arg, sizeof(*di_args));
2625         if (IS_ERR(di_args))
2626                 return PTR_ERR(di_args);
2627
2628         if (!btrfs_is_empty_uuid(di_args->uuid))
2629                 s_uuid = di_args->uuid;
2630
2631         mutex_lock(&fs_devices->device_list_mutex);
2632         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2633
2634         if (!dev) {
2635                 ret = -ENODEV;
2636                 goto out;
2637         }
2638
2639         di_args->devid = dev->devid;
2640         di_args->bytes_used = dev->bytes_used;
2641         di_args->total_bytes = dev->total_bytes;
2642         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2643         if (dev->name) {
2644                 struct rcu_string *name;
2645
2646                 rcu_read_lock();
2647                 name = rcu_dereference(dev->name);
2648                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2649                 rcu_read_unlock();
2650                 di_args->path[sizeof(di_args->path) - 1] = 0;
2651         } else {
2652                 di_args->path[0] = '\0';
2653         }
2654
2655 out:
2656         mutex_unlock(&fs_devices->device_list_mutex);
2657         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2658                 ret = -EFAULT;
2659
2660         kfree(di_args);
2661         return ret;
2662 }
2663
2664 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2665 {
2666         struct page *page;
2667         pgoff_t index;
2668         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2669
2670         index = off >> PAGE_CACHE_SHIFT;
2671
2672         page = grab_cache_page(inode->i_mapping, index);
2673         if (!page)
2674                 return NULL;
2675
2676         if (!PageUptodate(page)) {
2677                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2678                                                  0))
2679                         return NULL;
2680                 lock_page(page);
2681                 if (!PageUptodate(page)) {
2682                         unlock_page(page);
2683                         page_cache_release(page);
2684                         return NULL;
2685                 }
2686         }
2687         unlock_page(page);
2688
2689         return page;
2690 }
2691
2692 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2693 {
2694         /* do any pending delalloc/csum calc on src, one way or
2695            another, and lock file content */
2696         while (1) {
2697                 struct btrfs_ordered_extent *ordered;
2698                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2699                 ordered = btrfs_lookup_first_ordered_extent(inode,
2700                                                             off + len - 1);
2701                 if (!ordered &&
2702                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2703                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2704                         break;
2705                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2706                 if (ordered)
2707                         btrfs_put_ordered_extent(ordered);
2708                 btrfs_wait_ordered_range(inode, off, len);
2709         }
2710 }
2711
2712 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2713                                 struct inode *inode2, u64 loff2, u64 len)
2714 {
2715         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2716         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2717
2718         mutex_unlock(&inode1->i_mutex);
2719         mutex_unlock(&inode2->i_mutex);
2720 }
2721
2722 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2723                               struct inode *inode2, u64 loff2, u64 len)
2724 {
2725         if (inode1 < inode2) {
2726                 swap(inode1, inode2);
2727                 swap(loff1, loff2);
2728         }
2729
2730         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2731         lock_extent_range(inode1, loff1, len);
2732         if (inode1 != inode2) {
2733                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2734                 lock_extent_range(inode2, loff2, len);
2735         }
2736 }
2737
2738 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2739                           u64 dst_loff, u64 len)
2740 {
2741         int ret = 0;
2742         struct page *src_page, *dst_page;
2743         unsigned int cmp_len = PAGE_CACHE_SIZE;
2744         void *addr, *dst_addr;
2745
2746         while (len) {
2747                 if (len < PAGE_CACHE_SIZE)
2748                         cmp_len = len;
2749
2750                 src_page = extent_same_get_page(src, loff);
2751                 if (!src_page)
2752                         return -EINVAL;
2753                 dst_page = extent_same_get_page(dst, dst_loff);
2754                 if (!dst_page) {
2755                         page_cache_release(src_page);
2756                         return -EINVAL;
2757                 }
2758                 addr = kmap_atomic(src_page);
2759                 dst_addr = kmap_atomic(dst_page);
2760
2761                 flush_dcache_page(src_page);
2762                 flush_dcache_page(dst_page);
2763
2764                 if (memcmp(addr, dst_addr, cmp_len))
2765                         ret = BTRFS_SAME_DATA_DIFFERS;
2766
2767                 kunmap_atomic(addr);
2768                 kunmap_atomic(dst_addr);
2769                 page_cache_release(src_page);
2770                 page_cache_release(dst_page);
2771
2772                 if (ret)
2773                         break;
2774
2775                 loff += cmp_len;
2776                 dst_loff += cmp_len;
2777                 len -= cmp_len;
2778         }
2779
2780         return ret;
2781 }
2782
2783 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2784 {
2785         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2786
2787         if (off + len > inode->i_size || off + len < off)
2788                 return -EINVAL;
2789         /* Check that we are block aligned - btrfs_clone() requires this */
2790         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2791                 return -EINVAL;
2792
2793         return 0;
2794 }
2795
2796 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2797                              struct inode *dst, u64 dst_loff)
2798 {
2799         int ret;
2800
2801         /*
2802          * btrfs_clone() can't handle extents in the same file
2803          * yet. Once that works, we can drop this check and replace it
2804          * with a check for the same inode, but overlapping extents.
2805          */
2806         if (src == dst)
2807                 return -EINVAL;
2808
2809         btrfs_double_lock(src, loff, dst, dst_loff, len);
2810
2811         ret = extent_same_check_offsets(src, loff, len);
2812         if (ret)
2813                 goto out_unlock;
2814
2815         ret = extent_same_check_offsets(dst, dst_loff, len);
2816         if (ret)
2817                 goto out_unlock;
2818
2819         /* don't make the dst file partly checksummed */
2820         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2821             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2822                 ret = -EINVAL;
2823                 goto out_unlock;
2824         }
2825
2826         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2827         if (ret == 0)
2828                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2829
2830 out_unlock:
2831         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2832
2833         return ret;
2834 }
2835
2836 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2837
2838 static long btrfs_ioctl_file_extent_same(struct file *file,
2839                         struct btrfs_ioctl_same_args __user *argp)
2840 {
2841         struct btrfs_ioctl_same_args *same;
2842         struct btrfs_ioctl_same_extent_info *info;
2843         struct inode *src = file_inode(file);
2844         u64 off;
2845         u64 len;
2846         int i;
2847         int ret;
2848         unsigned long size;
2849         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2850         bool is_admin = capable(CAP_SYS_ADMIN);
2851         u16 count;
2852
2853         if (!(file->f_mode & FMODE_READ))
2854                 return -EINVAL;
2855
2856         ret = mnt_want_write_file(file);
2857         if (ret)
2858                 return ret;
2859
2860         if (get_user(count, &argp->dest_count)) {
2861                 ret = -EFAULT;
2862                 goto out;
2863         }
2864
2865         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2866
2867         same = memdup_user(argp, size);
2868
2869         if (IS_ERR(same)) {
2870                 ret = PTR_ERR(same);
2871                 goto out;
2872         }
2873
2874         off = same->logical_offset;
2875         len = same->length;
2876
2877         /*
2878          * Limit the total length we will dedupe for each operation.
2879          * This is intended to bound the total time spent in this
2880          * ioctl to something sane.
2881          */
2882         if (len > BTRFS_MAX_DEDUPE_LEN)
2883                 len = BTRFS_MAX_DEDUPE_LEN;
2884
2885         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2886                 /*
2887                  * Btrfs does not support blocksize < page_size. As a
2888                  * result, btrfs_cmp_data() won't correctly handle
2889                  * this situation without an update.
2890                  */
2891                 ret = -EINVAL;
2892                 goto out;
2893         }
2894
2895         ret = -EISDIR;
2896         if (S_ISDIR(src->i_mode))
2897                 goto out;
2898
2899         ret = -EACCES;
2900         if (!S_ISREG(src->i_mode))
2901                 goto out;
2902
2903         /* pre-format output fields to sane values */
2904         for (i = 0; i < count; i++) {
2905                 same->info[i].bytes_deduped = 0ULL;
2906                 same->info[i].status = 0;
2907         }
2908
2909         for (i = 0, info = same->info; i < count; i++, info++) {
2910                 struct inode *dst;
2911                 struct fd dst_file = fdget(info->fd);
2912                 if (!dst_file.file) {
2913                         info->status = -EBADF;
2914                         continue;
2915                 }
2916                 dst = file_inode(dst_file.file);
2917
2918                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
2919                         info->status = -EINVAL;
2920                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
2921                         info->status = -EXDEV;
2922                 } else if (S_ISDIR(dst->i_mode)) {
2923                         info->status = -EISDIR;
2924                 } else if (!S_ISREG(dst->i_mode)) {
2925                         info->status = -EACCES;
2926                 } else {
2927                         info->status = btrfs_extent_same(src, off, len, dst,
2928                                                         info->logical_offset);
2929                         if (info->status == 0)
2930                                 info->bytes_deduped += len;
2931                 }
2932                 fdput(dst_file);
2933         }
2934
2935         ret = copy_to_user(argp, same, size);
2936         if (ret)
2937                 ret = -EFAULT;
2938
2939 out:
2940         mnt_drop_write_file(file);
2941         return ret;
2942 }
2943
2944 /**
2945  * btrfs_clone() - clone a range from inode file to another
2946  *
2947  * @src: Inode to clone from
2948  * @inode: Inode to clone to
2949  * @off: Offset within source to start clone from
2950  * @olen: Original length, passed by user, of range to clone
2951  * @olen_aligned: Block-aligned value of olen, extent_same uses
2952  *               identical values here
2953  * @destoff: Offset within @inode to start clone
2954  */
2955 static int btrfs_clone(struct inode *src, struct inode *inode,
2956                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2957 {
2958         struct btrfs_root *root = BTRFS_I(inode)->root;
2959         struct btrfs_path *path = NULL;
2960         struct extent_buffer *leaf;
2961         struct btrfs_trans_handle *trans;
2962         char *buf = NULL;
2963         struct btrfs_key key;
2964         u32 nritems;
2965         int slot;
2966         int ret;
2967         u64 len = olen_aligned;
2968
2969         ret = -ENOMEM;
2970         buf = vmalloc(btrfs_level_size(root, 0));
2971         if (!buf)
2972                 return ret;
2973
2974         path = btrfs_alloc_path();
2975         if (!path) {
2976                 vfree(buf);
2977                 return ret;
2978         }
2979
2980         path->reada = 2;
2981         /* clone data */
2982         key.objectid = btrfs_ino(src);
2983         key.type = BTRFS_EXTENT_DATA_KEY;
2984         key.offset = 0;
2985
2986         while (1) {
2987                 /*
2988                  * note the key will change type as we walk through the
2989                  * tree.
2990                  */
2991                 path->leave_spinning = 1;
2992                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2993                                 0, 0);
2994                 if (ret < 0)
2995                         goto out;
2996
2997                 nritems = btrfs_header_nritems(path->nodes[0]);
2998 process_slot:
2999                 if (path->slots[0] >= nritems) {
3000                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3001                         if (ret < 0)
3002                                 goto out;
3003                         if (ret > 0)
3004                                 break;
3005                         nritems = btrfs_header_nritems(path->nodes[0]);
3006                 }
3007                 leaf = path->nodes[0];
3008                 slot = path->slots[0];
3009
3010                 btrfs_item_key_to_cpu(leaf, &key, slot);
3011                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
3012                     key.objectid != btrfs_ino(src))
3013                         break;
3014
3015                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3016                         struct btrfs_file_extent_item *extent;
3017                         int type;
3018                         u32 size;
3019                         struct btrfs_key new_key;
3020                         u64 disko = 0, diskl = 0;
3021                         u64 datao = 0, datal = 0;
3022                         u8 comp;
3023                         u64 endoff;
3024
3025                         extent = btrfs_item_ptr(leaf, slot,
3026                                                 struct btrfs_file_extent_item);
3027                         comp = btrfs_file_extent_compression(leaf, extent);
3028                         type = btrfs_file_extent_type(leaf, extent);
3029                         if (type == BTRFS_FILE_EXTENT_REG ||
3030                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3031                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3032                                                                       extent);
3033                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3034                                                                  extent);
3035                                 datao = btrfs_file_extent_offset(leaf, extent);
3036                                 datal = btrfs_file_extent_num_bytes(leaf,
3037                                                                     extent);
3038                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3039                                 /* take upper bound, may be compressed */
3040                                 datal = btrfs_file_extent_ram_bytes(leaf,
3041                                                                     extent);
3042                         }
3043
3044                         if (key.offset + datal <= off ||
3045                             key.offset >= off + len - 1) {
3046                                 path->slots[0]++;
3047                                 goto process_slot;
3048                         }
3049
3050                         size = btrfs_item_size_nr(leaf, slot);
3051                         read_extent_buffer(leaf, buf,
3052                                            btrfs_item_ptr_offset(leaf, slot),
3053                                            size);
3054
3055                         btrfs_release_path(path);
3056                         path->leave_spinning = 0;
3057
3058                         memcpy(&new_key, &key, sizeof(new_key));
3059                         new_key.objectid = btrfs_ino(inode);
3060                         if (off <= key.offset)
3061                                 new_key.offset = key.offset + destoff - off;
3062                         else
3063                                 new_key.offset = destoff;
3064
3065                         /*
3066                          * 1 - adjusting old extent (we may have to split it)
3067                          * 1 - add new extent
3068                          * 1 - inode update
3069                          */
3070                         trans = btrfs_start_transaction(root, 3);
3071                         if (IS_ERR(trans)) {
3072                                 ret = PTR_ERR(trans);
3073                                 goto out;
3074                         }
3075
3076                         if (type == BTRFS_FILE_EXTENT_REG ||
3077                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3078                                 /*
3079                                  *    a  | --- range to clone ---|  b
3080                                  * | ------------- extent ------------- |
3081                                  */
3082
3083                                 /* substract range b */
3084                                 if (key.offset + datal > off + len)
3085                                         datal = off + len - key.offset;
3086
3087                                 /* substract range a */
3088                                 if (off > key.offset) {
3089                                         datao += off - key.offset;
3090                                         datal -= off - key.offset;
3091                                 }
3092
3093                                 ret = btrfs_drop_extents(trans, root, inode,
3094                                                          new_key.offset,
3095                                                          new_key.offset + datal,
3096                                                          1);
3097                                 if (ret) {
3098                                         if (ret != -EOPNOTSUPP)
3099                                                 btrfs_abort_transaction(trans,
3100                                                                 root, ret);
3101                                         btrfs_end_transaction(trans, root);
3102                                         goto out;
3103                                 }
3104
3105                                 ret = btrfs_insert_empty_item(trans, root, path,
3106                                                               &new_key, size);
3107                                 if (ret) {
3108                                         btrfs_abort_transaction(trans, root,
3109                                                                 ret);
3110                                         btrfs_end_transaction(trans, root);
3111                                         goto out;
3112                                 }
3113
3114                                 leaf = path->nodes[0];
3115                                 slot = path->slots[0];
3116                                 write_extent_buffer(leaf, buf,
3117                                             btrfs_item_ptr_offset(leaf, slot),
3118                                             size);
3119
3120                                 extent = btrfs_item_ptr(leaf, slot,
3121                                                 struct btrfs_file_extent_item);
3122
3123                                 /* disko == 0 means it's a hole */
3124                                 if (!disko)
3125                                         datao = 0;
3126
3127                                 btrfs_set_file_extent_offset(leaf, extent,
3128                                                              datao);
3129                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3130                                                                 datal);
3131                                 if (disko) {
3132                                         inode_add_bytes(inode, datal);
3133                                         ret = btrfs_inc_extent_ref(trans, root,
3134                                                         disko, diskl, 0,
3135                                                         root->root_key.objectid,
3136                                                         btrfs_ino(inode),
3137                                                         new_key.offset - datao,
3138                                                         0);
3139                                         if (ret) {
3140                                                 btrfs_abort_transaction(trans,
3141                                                                         root,
3142                                                                         ret);
3143                                                 btrfs_end_transaction(trans,
3144                                                                       root);
3145                                                 goto out;
3146
3147                                         }
3148                                 }
3149                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3150                                 u64 skip = 0;
3151                                 u64 trim = 0;
3152                                 u64 aligned_end = 0;
3153
3154                                 if (off > key.offset) {
3155                                         skip = off - key.offset;
3156                                         new_key.offset += skip;
3157                                 }
3158
3159                                 if (key.offset + datal > off + len)
3160                                         trim = key.offset + datal - (off + len);
3161
3162                                 if (comp && (skip || trim)) {
3163                                         ret = -EINVAL;
3164                                         btrfs_end_transaction(trans, root);
3165                                         goto out;
3166                                 }
3167                                 size -= skip + trim;
3168                                 datal -= skip + trim;
3169
3170                                 aligned_end = ALIGN(new_key.offset + datal,
3171                                                     root->sectorsize);
3172                                 ret = btrfs_drop_extents(trans, root, inode,
3173                                                          new_key.offset,
3174                                                          aligned_end,
3175                                                          1);
3176                                 if (ret) {
3177                                         if (ret != -EOPNOTSUPP)
3178                                                 btrfs_abort_transaction(trans,
3179                                                         root, ret);
3180                                         btrfs_end_transaction(trans, root);
3181                                         goto out;
3182                                 }
3183
3184                                 ret = btrfs_insert_empty_item(trans, root, path,
3185                                                               &new_key, size);
3186                                 if (ret) {
3187                                         btrfs_abort_transaction(trans, root,
3188                                                                 ret);
3189                                         btrfs_end_transaction(trans, root);
3190                                         goto out;
3191                                 }
3192
3193                                 if (skip) {
3194                                         u32 start =
3195                                           btrfs_file_extent_calc_inline_size(0);
3196                                         memmove(buf+start, buf+start+skip,
3197                                                 datal);
3198                                 }
3199
3200                                 leaf = path->nodes[0];
3201                                 slot = path->slots[0];
3202                                 write_extent_buffer(leaf, buf,
3203                                             btrfs_item_ptr_offset(leaf, slot),
3204                                             size);
3205                                 inode_add_bytes(inode, datal);
3206                         }
3207
3208                         btrfs_mark_buffer_dirty(leaf);
3209                         btrfs_release_path(path);
3210
3211                         inode_inc_iversion(inode);
3212                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3213
3214                         /*
3215                          * we round up to the block size at eof when
3216                          * determining which extents to clone above,
3217                          * but shouldn't round up the file size
3218                          */
3219                         endoff = new_key.offset + datal;
3220                         if (endoff > destoff+olen)
3221                                 endoff = destoff+olen;
3222                         if (endoff > inode->i_size)
3223                                 btrfs_i_size_write(inode, endoff);
3224
3225                         ret = btrfs_update_inode(trans, root, inode);
3226                         if (ret) {
3227                                 btrfs_abort_transaction(trans, root, ret);
3228                                 btrfs_end_transaction(trans, root);
3229                                 goto out;
3230                         }
3231                         ret = btrfs_end_transaction(trans, root);
3232                 }
3233                 btrfs_release_path(path);
3234                 key.offset++;
3235         }
3236         ret = 0;
3237
3238 out:
3239         btrfs_release_path(path);
3240         btrfs_free_path(path);
3241         vfree(buf);
3242         return ret;
3243 }
3244
3245 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3246                                        u64 off, u64 olen, u64 destoff)
3247 {
3248         struct inode *inode = file_inode(file);
3249         struct btrfs_root *root = BTRFS_I(inode)->root;
3250         struct fd src_file;
3251         struct inode *src;
3252         int ret;
3253         u64 len = olen;
3254         u64 bs = root->fs_info->sb->s_blocksize;
3255         int same_inode = 0;
3256
3257         /*
3258          * TODO:
3259          * - split compressed inline extents.  annoying: we need to
3260          *   decompress into destination's address_space (the file offset
3261          *   may change, so source mapping won't do), then recompress (or
3262          *   otherwise reinsert) a subrange.
3263          *
3264          * - split destination inode's inline extents.  The inline extents can
3265          *   be either compressed or non-compressed.
3266          */
3267
3268         /* the destination must be opened for writing */
3269         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3270                 return -EINVAL;
3271
3272         if (btrfs_root_readonly(root))
3273                 return -EROFS;
3274
3275         ret = mnt_want_write_file(file);
3276         if (ret)
3277                 return ret;
3278
3279         src_file = fdget(srcfd);
3280         if (!src_file.file) {
3281                 ret = -EBADF;
3282                 goto out_drop_write;
3283         }
3284
3285         ret = -EXDEV;
3286         if (src_file.file->f_path.mnt != file->f_path.mnt)
3287                 goto out_fput;
3288
3289         src = file_inode(src_file.file);
3290
3291         ret = -EINVAL;
3292         if (src == inode)
3293                 same_inode = 1;
3294
3295         /* the src must be open for reading */
3296         if (!(src_file.file->f_mode & FMODE_READ))
3297                 goto out_fput;
3298
3299         /* don't make the dst file partly checksummed */
3300         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3301             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3302                 goto out_fput;
3303
3304         ret = -EISDIR;
3305         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3306                 goto out_fput;
3307
3308         ret = -EXDEV;
3309         if (src->i_sb != inode->i_sb)
3310                 goto out_fput;
3311
3312         if (!same_inode) {
3313                 if (inode < src) {
3314                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3315                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3316                 } else {
3317                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3318                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3319                 }
3320         } else {
3321                 mutex_lock(&src->i_mutex);
3322         }
3323
3324         /* determine range to clone */
3325         ret = -EINVAL;
3326         if (off + len > src->i_size || off + len < off)
3327                 goto out_unlock;
3328         if (len == 0)
3329                 olen = len = src->i_size - off;
3330         /* if we extend to eof, continue to block boundary */
3331         if (off + len == src->i_size)
3332                 len = ALIGN(src->i_size, bs) - off;
3333
3334         /* verify the end result is block aligned */
3335         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3336             !IS_ALIGNED(destoff, bs))
3337                 goto out_unlock;
3338
3339         /* verify if ranges are overlapped within the same file */
3340         if (same_inode) {
3341                 if (destoff + len > off && destoff < off + len)
3342                         goto out_unlock;
3343         }
3344
3345         if (destoff > inode->i_size) {
3346                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3347                 if (ret)
3348                         goto out_unlock;
3349         }
3350
3351         /* truncate page cache pages from target inode range */
3352         truncate_inode_pages_range(&inode->i_data, destoff,
3353                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3354
3355         lock_extent_range(src, off, len);
3356
3357         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3358
3359         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3360 out_unlock:
3361         if (!same_inode) {
3362                 if (inode < src) {
3363                         mutex_unlock(&src->i_mutex);
3364                         mutex_unlock(&inode->i_mutex);
3365                 } else {
3366                         mutex_unlock(&inode->i_mutex);
3367                         mutex_unlock(&src->i_mutex);
3368                 }
3369         } else {
3370                 mutex_unlock(&src->i_mutex);
3371         }
3372 out_fput:
3373         fdput(src_file);
3374 out_drop_write:
3375         mnt_drop_write_file(file);
3376         return ret;
3377 }
3378
3379 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3380 {
3381         struct btrfs_ioctl_clone_range_args args;
3382
3383         if (copy_from_user(&args, argp, sizeof(args)))
3384                 return -EFAULT;
3385         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3386                                  args.src_length, args.dest_offset);
3387 }
3388
3389 /*
3390  * there are many ways the trans_start and trans_end ioctls can lead
3391  * to deadlocks.  They should only be used by applications that
3392  * basically own the machine, and have a very in depth understanding
3393  * of all the possible deadlocks and enospc problems.
3394  */
3395 static long btrfs_ioctl_trans_start(struct file *file)
3396 {
3397         struct inode *inode = file_inode(file);
3398         struct btrfs_root *root = BTRFS_I(inode)->root;
3399         struct btrfs_trans_handle *trans;
3400         int ret;
3401
3402         ret = -EPERM;
3403         if (!capable(CAP_SYS_ADMIN))
3404                 goto out;
3405
3406         ret = -EINPROGRESS;
3407         if (file->private_data)
3408                 goto out;
3409
3410         ret = -EROFS;
3411         if (btrfs_root_readonly(root))
3412                 goto out;
3413
3414         ret = mnt_want_write_file(file);
3415         if (ret)
3416                 goto out;
3417
3418         atomic_inc(&root->fs_info->open_ioctl_trans);
3419
3420         ret = -ENOMEM;
3421         trans = btrfs_start_ioctl_transaction(root);
3422         if (IS_ERR(trans))
3423                 goto out_drop;
3424
3425         file->private_data = trans;
3426         return 0;
3427
3428 out_drop:
3429         atomic_dec(&root->fs_info->open_ioctl_trans);
3430         mnt_drop_write_file(file);
3431 out:
3432         return ret;
3433 }
3434
3435 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3436 {
3437         struct inode *inode = file_inode(file);
3438         struct btrfs_root *root = BTRFS_I(inode)->root;
3439         struct btrfs_root *new_root;
3440         struct btrfs_dir_item *di;
3441         struct btrfs_trans_handle *trans;
3442         struct btrfs_path *path;
3443         struct btrfs_key location;
3444         struct btrfs_disk_key disk_key;
3445         u64 objectid = 0;
3446         u64 dir_id;
3447         int ret;
3448
3449         if (!capable(CAP_SYS_ADMIN))
3450                 return -EPERM;
3451
3452         ret = mnt_want_write_file(file);
3453         if (ret)
3454                 return ret;
3455
3456         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3457                 ret = -EFAULT;
3458                 goto out;
3459         }
3460
3461         if (!objectid)
3462                 objectid = BTRFS_FS_TREE_OBJECTID;
3463
3464         location.objectid = objectid;
3465         location.type = BTRFS_ROOT_ITEM_KEY;
3466         location.offset = (u64)-1;
3467
3468         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3469         if (IS_ERR(new_root)) {
3470                 ret = PTR_ERR(new_root);
3471                 goto out;
3472         }
3473
3474         path = btrfs_alloc_path();
3475         if (!path) {
3476                 ret = -ENOMEM;
3477                 goto out;
3478         }
3479         path->leave_spinning = 1;
3480
3481         trans = btrfs_start_transaction(root, 1);
3482         if (IS_ERR(trans)) {
3483                 btrfs_free_path(path);
3484                 ret = PTR_ERR(trans);
3485                 goto out;
3486         }
3487
3488         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3489         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3490                                    dir_id, "default", 7, 1);
3491         if (IS_ERR_OR_NULL(di)) {
3492                 btrfs_free_path(path);
3493                 btrfs_end_transaction(trans, root);
3494                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3495                            "item, this isn't going to work");
3496                 ret = -ENOENT;
3497                 goto out;
3498         }
3499
3500         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3501         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3502         btrfs_mark_buffer_dirty(path->nodes[0]);
3503         btrfs_free_path(path);
3504
3505         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3506         btrfs_end_transaction(trans, root);
3507 out:
3508         mnt_drop_write_file(file);
3509         return ret;
3510 }
3511
3512 void btrfs_get_block_group_info(struct list_head *groups_list,
3513                                 struct btrfs_ioctl_space_info *space)
3514 {
3515         struct btrfs_block_group_cache *block_group;
3516
3517         space->total_bytes = 0;
3518         space->used_bytes = 0;
3519         space->flags = 0;
3520         list_for_each_entry(block_group, groups_list, list) {
3521                 space->flags = block_group->flags;
3522                 space->total_bytes += block_group->key.offset;
3523                 space->used_bytes +=
3524                         btrfs_block_group_used(&block_group->item);
3525         }
3526 }
3527
3528 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3529 {
3530         struct btrfs_ioctl_space_args space_args;
3531         struct btrfs_ioctl_space_info space;
3532         struct btrfs_ioctl_space_info *dest;
3533         struct btrfs_ioctl_space_info *dest_orig;
3534         struct btrfs_ioctl_space_info __user *user_dest;
3535         struct btrfs_space_info *info;
3536         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3537                        BTRFS_BLOCK_GROUP_SYSTEM,
3538                        BTRFS_BLOCK_GROUP_METADATA,
3539                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3540         int num_types = 4;
3541         int alloc_size;
3542         int ret = 0;
3543         u64 slot_count = 0;
3544         int i, c;
3545
3546         if (copy_from_user(&space_args,
3547                            (struct btrfs_ioctl_space_args __user *)arg,
3548                            sizeof(space_args)))
3549                 return -EFAULT;
3550
3551         for (i = 0; i < num_types; i++) {
3552                 struct btrfs_space_info *tmp;
3553
3554                 info = NULL;
3555                 rcu_read_lock();
3556                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3557                                         list) {
3558                         if (tmp->flags == types[i]) {
3559                                 info = tmp;
3560                                 break;
3561                         }
3562                 }
3563                 rcu_read_unlock();
3564
3565                 if (!info)
3566                         continue;
3567
3568                 down_read(&info->groups_sem);
3569                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3570                         if (!list_empty(&info->block_groups[c]))
3571                                 slot_count++;
3572                 }
3573                 up_read(&info->groups_sem);
3574         }
3575
3576         /*
3577          * Global block reserve, exported as a space_info
3578          */
3579         slot_count++;
3580
3581         /* space_slots == 0 means they are asking for a count */
3582         if (space_args.space_slots == 0) {
3583                 space_args.total_spaces = slot_count;
3584                 goto out;
3585         }
3586
3587         slot_count = min_t(u64, space_args.space_slots, slot_count);
3588
3589         alloc_size = sizeof(*dest) * slot_count;
3590
3591         /* we generally have at most 6 or so space infos, one for each raid
3592          * level.  So, a whole page should be more than enough for everyone
3593          */
3594         if (alloc_size > PAGE_CACHE_SIZE)
3595                 return -ENOMEM;
3596
3597         space_args.total_spaces = 0;
3598         dest = kmalloc(alloc_size, GFP_NOFS);
3599         if (!dest)
3600                 return -ENOMEM;
3601         dest_orig = dest;
3602
3603         /* now we have a buffer to copy into */
3604         for (i = 0; i < num_types; i++) {
3605                 struct btrfs_space_info *tmp;
3606
3607                 if (!slot_count)
3608                         break;
3609
3610                 info = NULL;
3611                 rcu_read_lock();
3612                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3613                                         list) {
3614                         if (tmp->flags == types[i]) {
3615                                 info = tmp;
3616                                 break;
3617                         }
3618                 }
3619                 rcu_read_unlock();
3620
3621                 if (!info)
3622                         continue;
3623                 down_read(&info->groups_sem);
3624                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3625                         if (!list_empty(&info->block_groups[c])) {
3626                                 btrfs_get_block_group_info(
3627                                         &info->block_groups[c], &space);
3628                                 memcpy(dest, &space, sizeof(space));
3629                                 dest++;
3630                                 space_args.total_spaces++;
3631                                 slot_count--;
3632                         }
3633                         if (!slot_count)
3634                                 break;
3635                 }
3636                 up_read(&info->groups_sem);
3637         }
3638
3639         /*
3640          * Add global block reserve
3641          */
3642         if (slot_count) {
3643                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3644
3645                 spin_lock(&block_rsv->lock);
3646                 space.total_bytes = block_rsv->size;
3647                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3648                 spin_unlock(&block_rsv->lock);
3649                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3650                 memcpy(dest, &space, sizeof(space));
3651                 space_args.total_spaces++;
3652         }
3653
3654         user_dest = (struct btrfs_ioctl_space_info __user *)
3655                 (arg + sizeof(struct btrfs_ioctl_space_args));
3656
3657         if (copy_to_user(user_dest, dest_orig, alloc_size))
3658                 ret = -EFAULT;
3659
3660         kfree(dest_orig);
3661 out:
3662         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3663                 ret = -EFAULT;
3664
3665         return ret;
3666 }
3667
3668 /*
3669  * there are many ways the trans_start and trans_end ioctls can lead
3670  * to deadlocks.  They should only be used by applications that
3671  * basically own the machine, and have a very in depth understanding
3672  * of all the possible deadlocks and enospc problems.
3673  */
3674 long btrfs_ioctl_trans_end(struct file *file)
3675 {
3676         struct inode *inode = file_inode(file);
3677         struct btrfs_root *root = BTRFS_I(inode)->root;
3678         struct btrfs_trans_handle *trans;
3679
3680         trans = file->private_data;
3681         if (!trans)
3682                 return -EINVAL;
3683         file->private_data = NULL;
3684
3685         btrfs_end_transaction(trans, root);
3686
3687         atomic_dec(&root->fs_info->open_ioctl_trans);
3688
3689         mnt_drop_write_file(file);
3690         return 0;
3691 }
3692
3693 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3694                                             void __user *argp)
3695 {
3696         struct btrfs_trans_handle *trans;
3697         u64 transid;
3698         int ret;
3699
3700         trans = btrfs_attach_transaction_barrier(root);
3701         if (IS_ERR(trans)) {
3702                 if (PTR_ERR(trans) != -ENOENT)
3703                         return PTR_ERR(trans);
3704
3705                 /* No running transaction, don't bother */
3706                 transid = root->fs_info->last_trans_committed;
3707                 goto out;
3708         }
3709         transid = trans->transid;
3710         ret = btrfs_commit_transaction_async(trans, root, 0);
3711         if (ret) {
3712                 btrfs_end_transaction(trans, root);
3713                 return ret;
3714         }
3715 out:
3716         if (argp)
3717                 if (copy_to_user(argp, &transid, sizeof(transid)))
3718                         return -EFAULT;
3719         return 0;
3720 }
3721
3722 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3723                                            void __user *argp)
3724 {
3725         u64 transid;
3726
3727         if (argp) {
3728                 if (copy_from_user(&transid, argp, sizeof(transid)))
3729                         return -EFAULT;
3730         } else {
3731                 transid = 0;  /* current trans */
3732         }
3733         return btrfs_wait_for_commit(root, transid);
3734 }
3735
3736 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3737 {
3738         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3739         struct btrfs_ioctl_scrub_args *sa;
3740         int ret;
3741
3742         if (!capable(CAP_SYS_ADMIN))
3743                 return -EPERM;
3744
3745         sa = memdup_user(arg, sizeof(*sa));
3746         if (IS_ERR(sa))
3747                 return PTR_ERR(sa);
3748
3749         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3750                 ret = mnt_want_write_file(file);
3751                 if (ret)
3752                         goto out;
3753         }
3754
3755         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3756                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3757                               0);
3758
3759         if (copy_to_user(arg, sa, sizeof(*sa)))
3760                 ret = -EFAULT;
3761
3762         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3763                 mnt_drop_write_file(file);
3764 out:
3765         kfree(sa);
3766         return ret;
3767 }
3768
3769 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3770 {
3771         if (!capable(CAP_SYS_ADMIN))
3772                 return -EPERM;
3773
3774         return btrfs_scrub_cancel(root->fs_info);
3775 }
3776
3777 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3778                                        void __user *arg)
3779 {
3780         struct btrfs_ioctl_scrub_args *sa;
3781         int ret;
3782
3783         if (!capable(CAP_SYS_ADMIN))
3784                 return -EPERM;
3785
3786         sa = memdup_user(arg, sizeof(*sa));
3787         if (IS_ERR(sa))
3788                 return PTR_ERR(sa);
3789
3790         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3791
3792         if (copy_to_user(arg, sa, sizeof(*sa)))
3793                 ret = -EFAULT;
3794
3795         kfree(sa);
3796         return ret;
3797 }
3798
3799 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3800                                       void __user *arg)
3801 {
3802         struct btrfs_ioctl_get_dev_stats *sa;
3803         int ret;
3804
3805         sa = memdup_user(arg, sizeof(*sa));
3806         if (IS_ERR(sa))
3807                 return PTR_ERR(sa);
3808
3809         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3810                 kfree(sa);
3811                 return -EPERM;
3812         }
3813
3814         ret = btrfs_get_dev_stats(root, sa);
3815
3816         if (copy_to_user(arg, sa, sizeof(*sa)))
3817                 ret = -EFAULT;
3818
3819         kfree(sa);
3820         return ret;
3821 }
3822
3823 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3824 {
3825         struct btrfs_ioctl_dev_replace_args *p;
3826         int ret;
3827
3828         if (!capable(CAP_SYS_ADMIN))
3829                 return -EPERM;
3830
3831         p = memdup_user(arg, sizeof(*p));
3832         if (IS_ERR(p))
3833                 return PTR_ERR(p);
3834
3835         switch (p->cmd) {
3836         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3837                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3838                         ret = -EROFS;
3839                         goto out;
3840                 }
3841                 if (atomic_xchg(
3842                         &root->fs_info->mutually_exclusive_operation_running,
3843                         1)) {
3844                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3845                 } else {
3846                         ret = btrfs_dev_replace_start(root, p);
3847                         atomic_set(
3848                          &root->fs_info->mutually_exclusive_operation_running,
3849                          0);
3850                 }
3851                 break;
3852         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3853                 btrfs_dev_replace_status(root->fs_info, p);
3854                 ret = 0;
3855                 break;
3856         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3857                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3858                 break;
3859         default:
3860                 ret = -EINVAL;
3861                 break;
3862         }
3863
3864         if (copy_to_user(arg, p, sizeof(*p)))
3865                 ret = -EFAULT;
3866 out:
3867         kfree(p);
3868         return ret;
3869 }
3870
3871 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3872 {
3873         int ret = 0;
3874         int i;
3875         u64 rel_ptr;
3876         int size;
3877         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3878         struct inode_fs_paths *ipath = NULL;
3879         struct btrfs_path *path;
3880
3881         if (!capable(CAP_DAC_READ_SEARCH))
3882                 return -EPERM;
3883
3884         path = btrfs_alloc_path();
3885         if (!path) {
3886                 ret = -ENOMEM;
3887                 goto out;
3888         }
3889
3890         ipa = memdup_user(arg, sizeof(*ipa));
3891         if (IS_ERR(ipa)) {
3892                 ret = PTR_ERR(ipa);
3893                 ipa = NULL;
3894                 goto out;
3895         }
3896
3897         size = min_t(u32, ipa->size, 4096);
3898         ipath = init_ipath(size, root, path);
3899         if (IS_ERR(ipath)) {
3900                 ret = PTR_ERR(ipath);
3901                 ipath = NULL;
3902                 goto out;
3903         }
3904
3905         ret = paths_from_inode(ipa->inum, ipath);
3906         if (ret < 0)
3907                 goto out;
3908
3909         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3910                 rel_ptr = ipath->fspath->val[i] -
3911                           (u64)(unsigned long)ipath->fspath->val;
3912                 ipath->fspath->val[i] = rel_ptr;
3913         }
3914
3915         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3916                            (void *)(unsigned long)ipath->fspath, size);
3917         if (ret) {
3918                 ret = -EFAULT;
3919                 goto out;
3920         }
3921
3922 out:
3923         btrfs_free_path(path);
3924         free_ipath(ipath);
3925         kfree(ipa);
3926
3927         return ret;
3928 }
3929
3930 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3931 {
3932         struct btrfs_data_container *inodes = ctx;
3933         const size_t c = 3 * sizeof(u64);
3934
3935         if (inodes->bytes_left >= c) {
3936                 inodes->bytes_left -= c;
3937                 inodes->val[inodes->elem_cnt] = inum;
3938                 inodes->val[inodes->elem_cnt + 1] = offset;
3939                 inodes->val[inodes->elem_cnt + 2] = root;
3940                 inodes->elem_cnt += 3;
3941         } else {
3942                 inodes->bytes_missing += c - inodes->bytes_left;
3943                 inodes->bytes_left = 0;
3944                 inodes->elem_missed += 3;
3945         }
3946
3947         return 0;
3948 }
3949
3950 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3951                                         void __user *arg)
3952 {
3953         int ret = 0;
3954         int size;
3955         struct btrfs_ioctl_logical_ino_args *loi;
3956         struct btrfs_data_container *inodes = NULL;
3957         struct btrfs_path *path = NULL;
3958
3959         if (!capable(CAP_SYS_ADMIN))
3960                 return -EPERM;
3961
3962         loi = memdup_user(arg, sizeof(*loi));
3963         if (IS_ERR(loi)) {
3964                 ret = PTR_ERR(loi);
3965                 loi = NULL;
3966                 goto out;
3967         }
3968
3969         path = btrfs_alloc_path();
3970         if (!path) {
3971                 ret = -ENOMEM;
3972                 goto out;
3973         }
3974
3975         size = min_t(u32, loi->size, 64 * 1024);
3976         inodes = init_data_container(size);
3977         if (IS_ERR(inodes)) {
3978                 ret = PTR_ERR(inodes);
3979                 inodes = NULL;
3980                 goto out;
3981         }
3982
3983         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3984                                           build_ino_list, inodes);
3985         if (ret == -EINVAL)
3986                 ret = -ENOENT;
3987         if (ret < 0)
3988                 goto out;
3989
3990         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3991                            (void *)(unsigned long)inodes, size);
3992         if (ret)
3993                 ret = -EFAULT;
3994
3995 out:
3996         btrfs_free_path(path);
3997         vfree(inodes);
3998         kfree(loi);
3999
4000         return ret;
4001 }
4002
4003 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4004                                struct btrfs_ioctl_balance_args *bargs)
4005 {
4006         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4007
4008         bargs->flags = bctl->flags;
4009
4010         if (atomic_read(&fs_info->balance_running))
4011                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4012         if (atomic_read(&fs_info->balance_pause_req))
4013                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4014         if (atomic_read(&fs_info->balance_cancel_req))
4015                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4016
4017         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4018         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4019         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4020
4021         if (lock) {
4022                 spin_lock(&fs_info->balance_lock);
4023                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4024                 spin_unlock(&fs_info->balance_lock);
4025         } else {
4026                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4027         }
4028 }
4029
4030 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4031 {
4032         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4033         struct btrfs_fs_info *fs_info = root->fs_info;
4034         struct btrfs_ioctl_balance_args *bargs;
4035         struct btrfs_balance_control *bctl;
4036         bool need_unlock; /* for mut. excl. ops lock */
4037         int ret;
4038
4039         if (!capable(CAP_SYS_ADMIN))
4040                 return -EPERM;
4041
4042         ret = mnt_want_write_file(file);
4043         if (ret)
4044                 return ret;
4045
4046 again:
4047         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4048                 mutex_lock(&fs_info->volume_mutex);
4049                 mutex_lock(&fs_info->balance_mutex);
4050                 need_unlock = true;
4051                 goto locked;
4052         }
4053
4054         /*
4055          * mut. excl. ops lock is locked.  Three possibilites:
4056          *   (1) some other op is running
4057          *   (2) balance is running
4058          *   (3) balance is paused -- special case (think resume)
4059          */
4060         mutex_lock(&fs_info->balance_mutex);
4061         if (fs_info->balance_ctl) {
4062                 /* this is either (2) or (3) */
4063                 if (!atomic_read(&fs_info->balance_running)) {
4064                         mutex_unlock(&fs_info->balance_mutex);
4065                         if (!mutex_trylock(&fs_info->volume_mutex))
4066                                 goto again;
4067                         mutex_lock(&fs_info->balance_mutex);
4068
4069                         if (fs_info->balance_ctl &&
4070                             !atomic_read(&fs_info->balance_running)) {
4071                                 /* this is (3) */
4072                                 need_unlock = false;
4073                                 goto locked;
4074                         }
4075
4076                         mutex_unlock(&fs_info->balance_mutex);
4077                         mutex_unlock(&fs_info->volume_mutex);
4078                         goto again;
4079                 } else {
4080                         /* this is (2) */
4081                         mutex_unlock(&fs_info->balance_mutex);
4082                         ret = -EINPROGRESS;
4083                         goto out;
4084                 }
4085         } else {
4086                 /* this is (1) */
4087                 mutex_unlock(&fs_info->balance_mutex);
4088                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4089                 goto out;
4090         }
4091
4092 locked:
4093         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4094
4095         if (arg) {
4096                 bargs = memdup_user(arg, sizeof(*bargs));
4097                 if (IS_ERR(bargs)) {
4098                         ret = PTR_ERR(bargs);
4099                         goto out_unlock;
4100                 }
4101
4102                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4103                         if (!fs_info->balance_ctl) {
4104                                 ret = -ENOTCONN;
4105                                 goto out_bargs;
4106                         }
4107
4108                         bctl = fs_info->balance_ctl;
4109                         spin_lock(&fs_info->balance_lock);
4110                         bctl->flags |= BTRFS_BALANCE_RESUME;
4111                         spin_unlock(&fs_info->balance_lock);
4112
4113                         goto do_balance;
4114                 }
4115         } else {
4116                 bargs = NULL;
4117         }
4118
4119         if (fs_info->balance_ctl) {
4120                 ret = -EINPROGRESS;
4121                 goto out_bargs;
4122         }
4123
4124         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4125         if (!bctl) {
4126                 ret = -ENOMEM;
4127                 goto out_bargs;
4128         }
4129
4130         bctl->fs_info = fs_info;
4131         if (arg) {
4132                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4133                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4134                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4135
4136                 bctl->flags = bargs->flags;
4137         } else {
4138                 /* balance everything - no filters */
4139                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4140         }
4141
4142 do_balance:
4143         /*
4144          * Ownership of bctl and mutually_exclusive_operation_running
4145          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4146          * or, if restriper was paused all the way until unmount, in
4147          * free_fs_info.  mutually_exclusive_operation_running is
4148          * cleared in __cancel_balance.
4149          */
4150         need_unlock = false;
4151
4152         ret = btrfs_balance(bctl, bargs);
4153
4154         if (arg) {
4155                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4156                         ret = -EFAULT;
4157         }
4158
4159 out_bargs:
4160         kfree(bargs);
4161 out_unlock:
4162         mutex_unlock(&fs_info->balance_mutex);
4163         mutex_unlock(&fs_info->volume_mutex);
4164         if (need_unlock)
4165                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4166 out:
4167         mnt_drop_write_file(file);
4168         return ret;
4169 }
4170
4171 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4172 {
4173         if (!capable(CAP_SYS_ADMIN))
4174                 return -EPERM;
4175
4176         switch (cmd) {
4177         case BTRFS_BALANCE_CTL_PAUSE:
4178                 return btrfs_pause_balance(root->fs_info);
4179         case BTRFS_BALANCE_CTL_CANCEL:
4180                 return btrfs_cancel_balance(root->fs_info);
4181         }
4182
4183         return -EINVAL;
4184 }
4185
4186 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4187                                          void __user *arg)
4188 {
4189         struct btrfs_fs_info *fs_info = root->fs_info;
4190         struct btrfs_ioctl_balance_args *bargs;
4191         int ret = 0;
4192
4193         if (!capable(CAP_SYS_ADMIN))
4194                 return -EPERM;
4195
4196         mutex_lock(&fs_info->balance_mutex);
4197         if (!fs_info->balance_ctl) {
4198                 ret = -ENOTCONN;
4199                 goto out;
4200         }
4201
4202         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4203         if (!bargs) {
4204                 ret = -ENOMEM;
4205                 goto out;
4206         }
4207
4208         update_ioctl_balance_args(fs_info, 1, bargs);
4209
4210         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4211                 ret = -EFAULT;
4212
4213         kfree(bargs);
4214 out:
4215         mutex_unlock(&fs_info->balance_mutex);
4216         return ret;
4217 }
4218
4219 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4220 {
4221         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4222         struct btrfs_ioctl_quota_ctl_args *sa;
4223         struct btrfs_trans_handle *trans = NULL;
4224         int ret;
4225         int err;
4226
4227         if (!capable(CAP_SYS_ADMIN))
4228                 return -EPERM;
4229
4230         ret = mnt_want_write_file(file);
4231         if (ret)
4232                 return ret;
4233
4234         sa = memdup_user(arg, sizeof(*sa));
4235         if (IS_ERR(sa)) {
4236                 ret = PTR_ERR(sa);
4237                 goto drop_write;
4238         }
4239
4240         down_write(&root->fs_info->subvol_sem);
4241         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4242         if (IS_ERR(trans)) {
4243                 ret = PTR_ERR(trans);
4244                 goto out;
4245         }
4246
4247         switch (sa->cmd) {
4248         case BTRFS_QUOTA_CTL_ENABLE:
4249                 ret = btrfs_quota_enable(trans, root->fs_info);
4250                 break;
4251         case BTRFS_QUOTA_CTL_DISABLE:
4252                 ret = btrfs_quota_disable(trans, root->fs_info);
4253                 break;
4254         default:
4255                 ret = -EINVAL;
4256                 break;
4257         }
4258
4259         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4260         if (err && !ret)
4261                 ret = err;
4262 out:
4263         kfree(sa);
4264         up_write(&root->fs_info->subvol_sem);
4265 drop_write:
4266         mnt_drop_write_file(file);
4267         return ret;
4268 }
4269
4270 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4271 {
4272         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4273         struct btrfs_ioctl_qgroup_assign_args *sa;
4274         struct btrfs_trans_handle *trans;
4275         int ret;
4276         int err;
4277
4278         if (!capable(CAP_SYS_ADMIN))
4279                 return -EPERM;
4280
4281         ret = mnt_want_write_file(file);
4282         if (ret)
4283                 return ret;
4284
4285         sa = memdup_user(arg, sizeof(*sa));
4286         if (IS_ERR(sa)) {
4287                 ret = PTR_ERR(sa);
4288                 goto drop_write;
4289         }
4290
4291         trans = btrfs_join_transaction(root);
4292         if (IS_ERR(trans)) {
4293                 ret = PTR_ERR(trans);
4294                 goto out;
4295         }
4296
4297         /* FIXME: check if the IDs really exist */
4298         if (sa->assign) {
4299                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4300                                                 sa->src, sa->dst);
4301         } else {
4302                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4303                                                 sa->src, sa->dst);
4304         }
4305
4306         err = btrfs_end_transaction(trans, root);
4307         if (err && !ret)
4308                 ret = err;
4309
4310 out:
4311         kfree(sa);
4312 drop_write:
4313         mnt_drop_write_file(file);
4314         return ret;
4315 }
4316
4317 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4318 {
4319         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4320         struct btrfs_ioctl_qgroup_create_args *sa;
4321         struct btrfs_trans_handle *trans;
4322         int ret;
4323         int err;
4324
4325         if (!capable(CAP_SYS_ADMIN))
4326                 return -EPERM;
4327
4328         ret = mnt_want_write_file(file);
4329         if (ret)
4330                 return ret;
4331
4332         sa = memdup_user(arg, sizeof(*sa));
4333         if (IS_ERR(sa)) {
4334                 ret = PTR_ERR(sa);
4335                 goto drop_write;
4336         }
4337
4338         if (!sa->qgroupid) {
4339                 ret = -EINVAL;
4340                 goto out;
4341         }
4342
4343         trans = btrfs_join_transaction(root);
4344         if (IS_ERR(trans)) {
4345                 ret = PTR_ERR(trans);
4346                 goto out;
4347         }
4348
4349         /* FIXME: check if the IDs really exist */
4350         if (sa->create) {
4351                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4352                                           NULL);
4353         } else {
4354                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4355         }
4356
4357         err = btrfs_end_transaction(trans, root);
4358         if (err && !ret)
4359                 ret = err;
4360
4361 out:
4362         kfree(sa);
4363 drop_write:
4364         mnt_drop_write_file(file);
4365         return ret;
4366 }
4367
4368 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4369 {
4370         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4371         struct btrfs_ioctl_qgroup_limit_args *sa;
4372         struct btrfs_trans_handle *trans;
4373         int ret;
4374         int err;
4375         u64 qgroupid;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         ret = mnt_want_write_file(file);
4381         if (ret)
4382                 return ret;
4383
4384         sa = memdup_user(arg, sizeof(*sa));
4385         if (IS_ERR(sa)) {
4386                 ret = PTR_ERR(sa);
4387                 goto drop_write;
4388         }
4389
4390         trans = btrfs_join_transaction(root);
4391         if (IS_ERR(trans)) {
4392                 ret = PTR_ERR(trans);
4393                 goto out;
4394         }
4395
4396         qgroupid = sa->qgroupid;
4397         if (!qgroupid) {
4398                 /* take the current subvol as qgroup */
4399                 qgroupid = root->root_key.objectid;
4400         }
4401
4402         /* FIXME: check if the IDs really exist */
4403         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4404
4405         err = btrfs_end_transaction(trans, root);
4406         if (err && !ret)
4407                 ret = err;
4408
4409 out:
4410         kfree(sa);
4411 drop_write:
4412         mnt_drop_write_file(file);
4413         return ret;
4414 }
4415
4416 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4417 {
4418         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4419         struct btrfs_ioctl_quota_rescan_args *qsa;
4420         int ret;
4421
4422         if (!capable(CAP_SYS_ADMIN))
4423                 return -EPERM;
4424
4425         ret = mnt_want_write_file(file);
4426         if (ret)
4427                 return ret;
4428
4429         qsa = memdup_user(arg, sizeof(*qsa));
4430         if (IS_ERR(qsa)) {
4431                 ret = PTR_ERR(qsa);
4432                 goto drop_write;
4433         }
4434
4435         if (qsa->flags) {
4436                 ret = -EINVAL;
4437                 goto out;
4438         }
4439
4440         ret = btrfs_qgroup_rescan(root->fs_info);
4441
4442 out:
4443         kfree(qsa);
4444 drop_write:
4445         mnt_drop_write_file(file);
4446         return ret;
4447 }
4448
4449 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4450 {
4451         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4452         struct btrfs_ioctl_quota_rescan_args *qsa;
4453         int ret = 0;
4454
4455         if (!capable(CAP_SYS_ADMIN))
4456                 return -EPERM;
4457
4458         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4459         if (!qsa)
4460                 return -ENOMEM;
4461
4462         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4463                 qsa->flags = 1;
4464                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4465         }
4466
4467         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4468                 ret = -EFAULT;
4469
4470         kfree(qsa);
4471         return ret;
4472 }
4473
4474 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4475 {
4476         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4477
4478         if (!capable(CAP_SYS_ADMIN))
4479                 return -EPERM;
4480
4481         return btrfs_qgroup_wait_for_completion(root->fs_info);
4482 }
4483
4484 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4485                                             struct btrfs_ioctl_received_subvol_args *sa)
4486 {
4487         struct inode *inode = file_inode(file);
4488         struct btrfs_root *root = BTRFS_I(inode)->root;
4489         struct btrfs_root_item *root_item = &root->root_item;
4490         struct btrfs_trans_handle *trans;
4491         struct timespec ct = CURRENT_TIME;
4492         int ret = 0;
4493         int received_uuid_changed;
4494
4495         if (!inode_owner_or_capable(inode))
4496                 return -EPERM;
4497
4498         ret = mnt_want_write_file(file);
4499         if (ret < 0)
4500                 return ret;
4501
4502         down_write(&root->fs_info->subvol_sem);
4503
4504         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4505                 ret = -EINVAL;
4506                 goto out;
4507         }
4508
4509         if (btrfs_root_readonly(root)) {
4510                 ret = -EROFS;
4511                 goto out;
4512         }
4513
4514         /*
4515          * 1 - root item
4516          * 2 - uuid items (received uuid + subvol uuid)
4517          */
4518         trans = btrfs_start_transaction(root, 3);
4519         if (IS_ERR(trans)) {
4520                 ret = PTR_ERR(trans);
4521                 trans = NULL;
4522                 goto out;
4523         }
4524
4525         sa->rtransid = trans->transid;
4526         sa->rtime.sec = ct.tv_sec;
4527         sa->rtime.nsec = ct.tv_nsec;
4528
4529         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4530                                        BTRFS_UUID_SIZE);
4531         if (received_uuid_changed &&
4532             !btrfs_is_empty_uuid(root_item->received_uuid))
4533                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4534                                     root_item->received_uuid,
4535                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4536                                     root->root_key.objectid);
4537         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4538         btrfs_set_root_stransid(root_item, sa->stransid);
4539         btrfs_set_root_rtransid(root_item, sa->rtransid);
4540         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4541         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4542         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4543         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4544
4545         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4546                                 &root->root_key, &root->root_item);
4547         if (ret < 0) {
4548                 btrfs_end_transaction(trans, root);
4549                 goto out;
4550         }
4551         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4552                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4553                                           sa->uuid,
4554                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4555                                           root->root_key.objectid);
4556                 if (ret < 0 && ret != -EEXIST) {
4557                         btrfs_abort_transaction(trans, root, ret);
4558                         goto out;
4559                 }
4560         }
4561         ret = btrfs_commit_transaction(trans, root);
4562         if (ret < 0) {
4563                 btrfs_abort_transaction(trans, root, ret);
4564                 goto out;
4565         }
4566
4567 out:
4568         up_write(&root->fs_info->subvol_sem);
4569         mnt_drop_write_file(file);
4570         return ret;
4571 }
4572
4573 #ifdef CONFIG_64BIT
4574 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4575                                                 void __user *arg)
4576 {
4577         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4578         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4579         int ret = 0;
4580
4581         args32 = memdup_user(arg, sizeof(*args32));
4582         if (IS_ERR(args32)) {
4583                 ret = PTR_ERR(args32);
4584                 args32 = NULL;
4585                 goto out;
4586         }
4587
4588         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4589         if (!args64) {
4590                 ret = -ENOMEM;
4591                 goto out;
4592         }
4593
4594         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4595         args64->stransid = args32->stransid;
4596         args64->rtransid = args32->rtransid;
4597         args64->stime.sec = args32->stime.sec;
4598         args64->stime.nsec = args32->stime.nsec;
4599         args64->rtime.sec = args32->rtime.sec;
4600         args64->rtime.nsec = args32->rtime.nsec;
4601         args64->flags = args32->flags;
4602
4603         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4604         if (ret)
4605                 goto out;
4606
4607         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4608         args32->stransid = args64->stransid;
4609         args32->rtransid = args64->rtransid;
4610         args32->stime.sec = args64->stime.sec;
4611         args32->stime.nsec = args64->stime.nsec;
4612         args32->rtime.sec = args64->rtime.sec;
4613         args32->rtime.nsec = args64->rtime.nsec;
4614         args32->flags = args64->flags;
4615
4616         ret = copy_to_user(arg, args32, sizeof(*args32));
4617         if (ret)
4618                 ret = -EFAULT;
4619
4620 out:
4621         kfree(args32);
4622         kfree(args64);
4623         return ret;
4624 }
4625 #endif
4626
4627 static long btrfs_ioctl_set_received_subvol(struct file *file,
4628                                             void __user *arg)
4629 {
4630         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4631         int ret = 0;
4632
4633         sa = memdup_user(arg, sizeof(*sa));
4634         if (IS_ERR(sa)) {
4635                 ret = PTR_ERR(sa);
4636                 sa = NULL;
4637                 goto out;
4638         }
4639
4640         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4641
4642         if (ret)
4643                 goto out;
4644
4645         ret = copy_to_user(arg, sa, sizeof(*sa));
4646         if (ret)
4647                 ret = -EFAULT;
4648
4649 out:
4650         kfree(sa);
4651         return ret;
4652 }
4653
4654 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4655 {
4656         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4657         size_t len;
4658         int ret;
4659         char label[BTRFS_LABEL_SIZE];
4660
4661         spin_lock(&root->fs_info->super_lock);
4662         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4663         spin_unlock(&root->fs_info->super_lock);
4664
4665         len = strnlen(label, BTRFS_LABEL_SIZE);
4666
4667         if (len == BTRFS_LABEL_SIZE) {
4668                 btrfs_warn(root->fs_info,
4669                         "label is too long, return the first %zu bytes", --len);
4670         }
4671
4672         ret = copy_to_user(arg, label, len);
4673
4674         return ret ? -EFAULT : 0;
4675 }
4676
4677 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4678 {
4679         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4680         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4681         struct btrfs_trans_handle *trans;
4682         char label[BTRFS_LABEL_SIZE];
4683         int ret;
4684
4685         if (!capable(CAP_SYS_ADMIN))
4686                 return -EPERM;
4687
4688         if (copy_from_user(label, arg, sizeof(label)))
4689                 return -EFAULT;
4690
4691         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4692                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4693                        BTRFS_LABEL_SIZE - 1);
4694                 return -EINVAL;
4695         }
4696
4697         ret = mnt_want_write_file(file);
4698         if (ret)
4699                 return ret;
4700
4701         trans = btrfs_start_transaction(root, 0);
4702         if (IS_ERR(trans)) {
4703                 ret = PTR_ERR(trans);
4704                 goto out_unlock;
4705         }
4706
4707         spin_lock(&root->fs_info->super_lock);
4708         strcpy(super_block->label, label);
4709         spin_unlock(&root->fs_info->super_lock);
4710         ret = btrfs_commit_transaction(trans, root);
4711
4712 out_unlock:
4713         mnt_drop_write_file(file);
4714         return ret;
4715 }
4716
4717 #define INIT_FEATURE_FLAGS(suffix) \
4718         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4719           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4720           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4721
4722 static int btrfs_ioctl_get_supported_features(struct file *file,
4723                                               void __user *arg)
4724 {
4725         static struct btrfs_ioctl_feature_flags features[3] = {
4726                 INIT_FEATURE_FLAGS(SUPP),
4727                 INIT_FEATURE_FLAGS(SAFE_SET),
4728                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4729         };
4730
4731         if (copy_to_user(arg, &features, sizeof(features)))
4732                 return -EFAULT;
4733
4734         return 0;
4735 }
4736
4737 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4738 {
4739         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4740         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4741         struct btrfs_ioctl_feature_flags features;
4742
4743         features.compat_flags = btrfs_super_compat_flags(super_block);
4744         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4745         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4746
4747         if (copy_to_user(arg, &features, sizeof(features)))
4748                 return -EFAULT;
4749
4750         return 0;
4751 }
4752
4753 static int check_feature_bits(struct btrfs_root *root,
4754                               enum btrfs_feature_set set,
4755                               u64 change_mask, u64 flags, u64 supported_flags,
4756                               u64 safe_set, u64 safe_clear)
4757 {
4758         const char *type = btrfs_feature_set_names[set];
4759         char *names;
4760         u64 disallowed, unsupported;
4761         u64 set_mask = flags & change_mask;
4762         u64 clear_mask = ~flags & change_mask;
4763
4764         unsupported = set_mask & ~supported_flags;
4765         if (unsupported) {
4766                 names = btrfs_printable_features(set, unsupported);
4767                 if (names) {
4768                         btrfs_warn(root->fs_info,
4769                            "this kernel does not support the %s feature bit%s",
4770                            names, strchr(names, ',') ? "s" : "");
4771                         kfree(names);
4772                 } else
4773                         btrfs_warn(root->fs_info,
4774                            "this kernel does not support %s bits 0x%llx",
4775                            type, unsupported);
4776                 return -EOPNOTSUPP;
4777         }
4778
4779         disallowed = set_mask & ~safe_set;
4780         if (disallowed) {
4781                 names = btrfs_printable_features(set, disallowed);
4782                 if (names) {
4783                         btrfs_warn(root->fs_info,
4784                            "can't set the %s feature bit%s while mounted",
4785                            names, strchr(names, ',') ? "s" : "");
4786                         kfree(names);
4787                 } else
4788                         btrfs_warn(root->fs_info,
4789                            "can't set %s bits 0x%llx while mounted",
4790                            type, disallowed);
4791                 return -EPERM;
4792         }
4793
4794         disallowed = clear_mask & ~safe_clear;
4795         if (disallowed) {
4796                 names = btrfs_printable_features(set, disallowed);
4797                 if (names) {
4798                         btrfs_warn(root->fs_info,
4799                            "can't clear the %s feature bit%s while mounted",
4800                            names, strchr(names, ',') ? "s" : "");
4801                         kfree(names);
4802                 } else
4803                         btrfs_warn(root->fs_info,
4804                            "can't clear %s bits 0x%llx while mounted",
4805                            type, disallowed);
4806                 return -EPERM;
4807         }
4808
4809         return 0;
4810 }
4811
4812 #define check_feature(root, change_mask, flags, mask_base)      \
4813 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
4814                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4815                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4816                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4817
4818 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4819 {
4820         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4821         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4822         struct btrfs_ioctl_feature_flags flags[2];
4823         struct btrfs_trans_handle *trans;
4824         u64 newflags;
4825         int ret;
4826
4827         if (!capable(CAP_SYS_ADMIN))
4828                 return -EPERM;
4829
4830         if (copy_from_user(flags, arg, sizeof(flags)))
4831                 return -EFAULT;
4832
4833         /* Nothing to do */
4834         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4835             !flags[0].incompat_flags)
4836                 return 0;
4837
4838         ret = check_feature(root, flags[0].compat_flags,
4839                             flags[1].compat_flags, COMPAT);
4840         if (ret)
4841                 return ret;
4842
4843         ret = check_feature(root, flags[0].compat_ro_flags,
4844                             flags[1].compat_ro_flags, COMPAT_RO);
4845         if (ret)
4846                 return ret;
4847
4848         ret = check_feature(root, flags[0].incompat_flags,
4849                             flags[1].incompat_flags, INCOMPAT);
4850         if (ret)
4851                 return ret;
4852
4853         trans = btrfs_start_transaction(root, 0);
4854         if (IS_ERR(trans))
4855                 return PTR_ERR(trans);
4856
4857         spin_lock(&root->fs_info->super_lock);
4858         newflags = btrfs_super_compat_flags(super_block);
4859         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4860         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4861         btrfs_set_super_compat_flags(super_block, newflags);
4862
4863         newflags = btrfs_super_compat_ro_flags(super_block);
4864         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4865         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4866         btrfs_set_super_compat_ro_flags(super_block, newflags);
4867
4868         newflags = btrfs_super_incompat_flags(super_block);
4869         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4870         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4871         btrfs_set_super_incompat_flags(super_block, newflags);
4872         spin_unlock(&root->fs_info->super_lock);
4873
4874         return btrfs_commit_transaction(trans, root);
4875 }
4876
4877 long btrfs_ioctl(struct file *file, unsigned int
4878                 cmd, unsigned long arg)
4879 {
4880         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4881         void __user *argp = (void __user *)arg;
4882
4883         switch (cmd) {
4884         case FS_IOC_GETFLAGS:
4885                 return btrfs_ioctl_getflags(file, argp);
4886         case FS_IOC_SETFLAGS:
4887                 return btrfs_ioctl_setflags(file, argp);
4888         case FS_IOC_GETVERSION:
4889                 return btrfs_ioctl_getversion(file, argp);
4890         case FITRIM:
4891                 return btrfs_ioctl_fitrim(file, argp);
4892         case BTRFS_IOC_SNAP_CREATE:
4893                 return btrfs_ioctl_snap_create(file, argp, 0);
4894         case BTRFS_IOC_SNAP_CREATE_V2:
4895                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4896         case BTRFS_IOC_SUBVOL_CREATE:
4897                 return btrfs_ioctl_snap_create(file, argp, 1);
4898         case BTRFS_IOC_SUBVOL_CREATE_V2:
4899                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4900         case BTRFS_IOC_SNAP_DESTROY:
4901                 return btrfs_ioctl_snap_destroy(file, argp);
4902         case BTRFS_IOC_SUBVOL_GETFLAGS:
4903                 return btrfs_ioctl_subvol_getflags(file, argp);
4904         case BTRFS_IOC_SUBVOL_SETFLAGS:
4905                 return btrfs_ioctl_subvol_setflags(file, argp);
4906         case BTRFS_IOC_DEFAULT_SUBVOL:
4907                 return btrfs_ioctl_default_subvol(file, argp);
4908         case BTRFS_IOC_DEFRAG:
4909                 return btrfs_ioctl_defrag(file, NULL);
4910         case BTRFS_IOC_DEFRAG_RANGE:
4911                 return btrfs_ioctl_defrag(file, argp);
4912         case BTRFS_IOC_RESIZE:
4913                 return btrfs_ioctl_resize(file, argp);
4914         case BTRFS_IOC_ADD_DEV:
4915                 return btrfs_ioctl_add_dev(root, argp);
4916         case BTRFS_IOC_RM_DEV:
4917                 return btrfs_ioctl_rm_dev(file, argp);
4918         case BTRFS_IOC_FS_INFO:
4919                 return btrfs_ioctl_fs_info(root, argp);
4920         case BTRFS_IOC_DEV_INFO:
4921                 return btrfs_ioctl_dev_info(root, argp);
4922         case BTRFS_IOC_BALANCE:
4923                 return btrfs_ioctl_balance(file, NULL);
4924         case BTRFS_IOC_CLONE:
4925                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4926         case BTRFS_IOC_CLONE_RANGE:
4927                 return btrfs_ioctl_clone_range(file, argp);
4928         case BTRFS_IOC_TRANS_START:
4929                 return btrfs_ioctl_trans_start(file);
4930         case BTRFS_IOC_TRANS_END:
4931                 return btrfs_ioctl_trans_end(file);
4932         case BTRFS_IOC_TREE_SEARCH:
4933                 return btrfs_ioctl_tree_search(file, argp);
4934         case BTRFS_IOC_INO_LOOKUP:
4935                 return btrfs_ioctl_ino_lookup(file, argp);
4936         case BTRFS_IOC_INO_PATHS:
4937                 return btrfs_ioctl_ino_to_path(root, argp);
4938         case BTRFS_IOC_LOGICAL_INO:
4939                 return btrfs_ioctl_logical_to_ino(root, argp);
4940         case BTRFS_IOC_SPACE_INFO:
4941                 return btrfs_ioctl_space_info(root, argp);
4942         case BTRFS_IOC_SYNC: {
4943                 int ret;
4944
4945                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
4946                 if (ret)
4947                         return ret;
4948                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4949                 return ret;
4950         }
4951         case BTRFS_IOC_START_SYNC:
4952                 return btrfs_ioctl_start_sync(root, argp);
4953         case BTRFS_IOC_WAIT_SYNC:
4954                 return btrfs_ioctl_wait_sync(root, argp);
4955         case BTRFS_IOC_SCRUB:
4956                 return btrfs_ioctl_scrub(file, argp);
4957         case BTRFS_IOC_SCRUB_CANCEL:
4958                 return btrfs_ioctl_scrub_cancel(root, argp);
4959         case BTRFS_IOC_SCRUB_PROGRESS:
4960                 return btrfs_ioctl_scrub_progress(root, argp);
4961         case BTRFS_IOC_BALANCE_V2:
4962                 return btrfs_ioctl_balance(file, argp);
4963         case BTRFS_IOC_BALANCE_CTL:
4964                 return btrfs_ioctl_balance_ctl(root, arg);
4965         case BTRFS_IOC_BALANCE_PROGRESS:
4966                 return btrfs_ioctl_balance_progress(root, argp);
4967         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4968                 return btrfs_ioctl_set_received_subvol(file, argp);
4969 #ifdef CONFIG_64BIT
4970         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4971                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4972 #endif
4973         case BTRFS_IOC_SEND:
4974                 return btrfs_ioctl_send(file, argp);
4975         case BTRFS_IOC_GET_DEV_STATS:
4976                 return btrfs_ioctl_get_dev_stats(root, argp);
4977         case BTRFS_IOC_QUOTA_CTL:
4978                 return btrfs_ioctl_quota_ctl(file, argp);
4979         case BTRFS_IOC_QGROUP_ASSIGN:
4980                 return btrfs_ioctl_qgroup_assign(file, argp);
4981         case BTRFS_IOC_QGROUP_CREATE:
4982                 return btrfs_ioctl_qgroup_create(file, argp);
4983         case BTRFS_IOC_QGROUP_LIMIT:
4984                 return btrfs_ioctl_qgroup_limit(file, argp);
4985         case BTRFS_IOC_QUOTA_RESCAN:
4986                 return btrfs_ioctl_quota_rescan(file, argp);
4987         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4988                 return btrfs_ioctl_quota_rescan_status(file, argp);
4989         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4990                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4991         case BTRFS_IOC_DEV_REPLACE:
4992                 return btrfs_ioctl_dev_replace(root, argp);
4993         case BTRFS_IOC_GET_FSLABEL:
4994                 return btrfs_ioctl_get_fslabel(file, argp);
4995         case BTRFS_IOC_SET_FSLABEL:
4996                 return btrfs_ioctl_set_fslabel(file, argp);
4997         case BTRFS_IOC_FILE_EXTENT_SAME:
4998                 return btrfs_ioctl_file_extent_same(file, argp);
4999         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5000                 return btrfs_ioctl_get_supported_features(file, argp);
5001         case BTRFS_IOC_GET_FEATURES:
5002                 return btrfs_ioctl_get_features(file, argp);
5003         case BTRFS_IOC_SET_FEATURES:
5004                 return btrfs_ioctl_set_features(file, argp);
5005         }
5006
5007         return -ENOTTY;
5008 }