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