fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61
  62 #ifdef CONFIG_64BIT
  63 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  64  * structures are incorrect, as the timespec structure from userspace
  65  * is 4 bytes too small. We define these alternatives here to teach
  66  * the kernel about the 32-bit struct packing.
  67  */
  68 struct btrfs_ioctl_timespec_32 {
  69         __u64 sec;
  70         __u32 nsec;
  71 } __attribute__ ((__packed__));
  72
  73 struct btrfs_ioctl_received_subvol_args_32 {
  74         char    uuid[BTRFS_UUID_SIZE];  /* in */
  75         __u64   stransid;               /* in */
  76         __u64   rtransid;               /* out */
  77         struct btrfs_ioctl_timespec_32 stime; /* in */
  78         struct btrfs_ioctl_timespec_32 rtime; /* out */
  79         __u64   flags;                  /* in */
  80         __u64   reserved[16];           /* in */
  81 } __attribute__ ((__packed__));
  82
  83 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  84                                 struct btrfs_ioctl_received_subvol_args_32)
  85 #endif
  86
  87
  88 static int btrfs_clone(struct inode *src, struct inode *inode,
  89                        u64 off, u64 olen, u64 olen_aligned, u64 destoff);
  90
  91 /* Mask out flags that are inappropriate for the given type of inode. */
  92 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  93 {
  94         if (S_ISDIR(mode))
  95                 return flags;
  96         else if (S_ISREG(mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 104  */
 105 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 106 {
 107         unsigned int iflags = 0;
 108
 109         if (flags & BTRFS_INODE_SYNC)
 110                 iflags |= FS_SYNC_FL;
 111         if (flags & BTRFS_INODE_IMMUTABLE)
 112                 iflags |= FS_IMMUTABLE_FL;
 113         if (flags & BTRFS_INODE_APPEND)
 114                 iflags |= FS_APPEND_FL;
 115         if (flags & BTRFS_INODE_NODUMP)
 116                 iflags |= FS_NODUMP_FL;
 117         if (flags & BTRFS_INODE_NOATIME)
 118                 iflags |= FS_NOATIME_FL;
 119         if (flags & BTRFS_INODE_DIRSYNC)
 120                 iflags |= FS_DIRSYNC_FL;
 121         if (flags & BTRFS_INODE_NODATACOW)
 122                 iflags |= FS_NOCOW_FL;
 123
 124         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 125                 iflags |= FS_COMPR_FL;
 126         else if (flags & BTRFS_INODE_NOCOMPRESS)
 127                 iflags |= FS_NOCOMP_FL;
 128
 129         return iflags;
 130 }
 131
 132 /*
 133  * Update inode->i_flags based on the btrfs internal flags.
 134  */
 135 void btrfs_update_iflags(struct inode *inode)
 136 {
 137         struct btrfs_inode *ip = BTRFS_I(inode);
 138
 139         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 140
 141         if (ip->flags & BTRFS_INODE_SYNC)
 142                 inode->i_flags |= S_SYNC;
 143         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 144                 inode->i_flags |= S_IMMUTABLE;
 145         if (ip->flags & BTRFS_INODE_APPEND)
 146                 inode->i_flags |= S_APPEND;
 147         if (ip->flags & BTRFS_INODE_NOATIME)
 148                 inode->i_flags |= S_NOATIME;
 149         if (ip->flags & BTRFS_INODE_DIRSYNC)
 150                 inode->i_flags |= S_DIRSYNC;
 151 }
 152
 153 /*
 154  * Inherit flags from the parent inode.
 155  *
 156  * Currently only the compression flags and the cow flags are inherited.
 157  */
 158 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 159 {
 160         unsigned int flags;
 161
 162         if (!dir)
 163                 return;
 164
 165         flags = BTRFS_I(dir)->flags;
 166
 167         if (flags & BTRFS_INODE_NOCOMPRESS) {
 168                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 169                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 170         } else if (flags & BTRFS_INODE_COMPRESS) {
 171                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 172                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 173         }
 174
 175         if (flags & BTRFS_INODE_NODATACOW) {
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 177                 if (S_ISREG(inode->i_mode))
 178                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 179         }
 180
 181         btrfs_update_iflags(inode);
 182 }
 183
 184 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 185 {
 186         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 187         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 188
 189         if (copy_to_user(arg, &flags, sizeof(flags)))
 190                 return -EFAULT;
 191         return 0;
 192 }
 193
 194 static int check_flags(unsigned int flags)
 195 {
 196         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 197                       FS_NOATIME_FL | FS_NODUMP_FL | \
 198                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 199                       FS_NOCOMP_FL | FS_COMPR_FL |
 200                       FS_NOCOW_FL))
 201                 return -EOPNOTSUPP;
 202
 203         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 204                 return -EINVAL;
 205
 206         return 0;
 207 }
 208
 209 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 210 {
 211         struct inode *inode = file_inode(file);
 212         struct btrfs_inode *ip = BTRFS_I(inode);
 213         struct btrfs_root *root = ip->root;
 214         struct btrfs_trans_handle *trans;
 215         unsigned int flags, oldflags;
 216         int ret;
 217         u64 ip_oldflags;
 218         unsigned int i_oldflags;
 219         umode_t mode;
 220
 221         if (!inode_owner_or_capable(inode))
 222                 return -EPERM;
 223
 224         if (btrfs_root_readonly(root))
 225                 return -EROFS;
 226
 227         if (copy_from_user(&flags, arg, sizeof(flags)))
 228                 return -EFAULT;
 229
 230         ret = check_flags(flags);
 231         if (ret)
 232                 return ret;
 233
 234         ret = mnt_want_write_file(file);
 235         if (ret)
 236                 return ret;
 237
 238         mutex_lock(&inode->i_mutex);
 239
 240         ip_oldflags = ip->flags;
 241         i_oldflags = inode->i_flags;
 242         mode = inode->i_mode;
 243
 244         flags = btrfs_mask_flags(inode->i_mode, flags);
 245         oldflags = btrfs_flags_to_ioctl(ip->flags);
 246         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 247                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 248                         ret = -EPERM;
 249                         goto out_unlock;
 250                 }
 251         }
 252
 253         if (flags & FS_SYNC_FL)
 254                 ip->flags |= BTRFS_INODE_SYNC;
 255         else
 256                 ip->flags &= ~BTRFS_INODE_SYNC;
 257         if (flags & FS_IMMUTABLE_FL)
 258                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 259         else
 260                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 261         if (flags & FS_APPEND_FL)
 262                 ip->flags |= BTRFS_INODE_APPEND;
 263         else
 264                 ip->flags &= ~BTRFS_INODE_APPEND;
 265         if (flags & FS_NODUMP_FL)
 266                 ip->flags |= BTRFS_INODE_NODUMP;
 267         else
 268                 ip->flags &= ~BTRFS_INODE_NODUMP;
 269         if (flags & FS_NOATIME_FL)
 270                 ip->flags |= BTRFS_INODE_NOATIME;
 271         else
 272                 ip->flags &= ~BTRFS_INODE_NOATIME;
 273         if (flags & FS_DIRSYNC_FL)
 274                 ip->flags |= BTRFS_INODE_DIRSYNC;
 275         else
 276                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 277         if (flags & FS_NOCOW_FL) {
 278                 if (S_ISREG(mode)) {
 279                         /*
 280                          * It's safe to turn csums off here, no extents exist.
 281                          * Otherwise we want the flag to reflect the real COW
 282                          * status of the file and will not set it.
 283                          */
 284                         if (inode->i_size == 0)
 285                                 ip->flags |= BTRFS_INODE_NODATACOW
 286                                            | BTRFS_INODE_NODATASUM;
 287                 } else {
 288                         ip->flags |= BTRFS_INODE_NODATACOW;
 289                 }
 290         } else {
 291                 /*
 292                  * Revert back under same assuptions as above
 293                  */
 294                 if (S_ISREG(mode)) {
 295                         if (inode->i_size == 0)
 296                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 297                                              | BTRFS_INODE_NODATASUM);
 298                 } else {
 299                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 300                 }
 301         }
 302
 303         /*
 304          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 305          * flag may be changed automatically if compression code won't make
 306          * things smaller.
 307          */
 308         if (flags & FS_NOCOMP_FL) {
 309                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 310                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 311
 312                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 313                 if (ret && ret != -ENODATA)
 314                         goto out_drop;
 315         } else if (flags & FS_COMPR_FL) {
 316                 const char *comp;
 317
 318                 ip->flags |= BTRFS_INODE_COMPRESS;
 319                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 320
 321                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 322                         comp = "lzo";
 323                 else
 324                         comp = "zlib";
 325                 ret = btrfs_set_prop(inode, "btrfs.compression",
 326                                      comp, strlen(comp), 0);
 327                 if (ret)
 328                         goto out_drop;
 329
 330         } else {
 331                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 332         }
 333
 334         trans = btrfs_start_transaction(root, 1);
 335         if (IS_ERR(trans)) {
 336                 ret = PTR_ERR(trans);
 337                 goto out_drop;
 338         }
 339
 340         btrfs_update_iflags(inode);
 341         inode_inc_iversion(inode);
 342         inode->i_ctime = CURRENT_TIME;
 343         ret = btrfs_update_inode(trans, root, inode);
 344
 345         btrfs_end_transaction(trans, root);
 346  out_drop:
 347         if (ret) {
 348                 ip->flags = ip_oldflags;
 349                 inode->i_flags = i_oldflags;
 350         }
 351
 352  out_unlock:
 353         mutex_unlock(&inode->i_mutex);
 354         mnt_drop_write_file(file);
 355         return ret;
 356 }
 357
 358 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 359 {
 360         struct inode *inode = file_inode(file);
 361
 362         return put_user(inode->i_generation, arg);
 363 }
 364
 365 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 366 {
 367         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 368         struct btrfs_device *device;
 369         struct request_queue *q;
 370         struct fstrim_range range;
 371         u64 minlen = ULLONG_MAX;
 372         u64 num_devices = 0;
 373         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 374         int ret;
 375
 376         if (!capable(CAP_SYS_ADMIN))
 377                 return -EPERM;
 378
 379         rcu_read_lock();
 380         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 381                                 dev_list) {
 382                 if (!device->bdev)
 383                         continue;
 384                 q = bdev_get_queue(device->bdev);
 385                 if (blk_queue_discard(q)) {
 386                         num_devices++;
 387                         minlen = min((u64)q->limits.discard_granularity,
 388                                      minlen);
 389                 }
 390         }
 391         rcu_read_unlock();
 392
 393         if (!num_devices)
 394                 return -EOPNOTSUPP;
 395         if (copy_from_user(&range, arg, sizeof(range)))
 396                 return -EFAULT;
 397         if (range.start > total_bytes ||
 398             range.len < fs_info->sb->s_blocksize)
 399                 return -EINVAL;
 400
 401         range.len = min(range.len, total_bytes - range.start);
 402         range.minlen = max(range.minlen, minlen);
 403         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 404         if (ret < 0)
 405                 return ret;
 406
 407         if (copy_to_user(arg, &range, sizeof(range)))
 408                 return -EFAULT;
 409
 410         return 0;
 411 }
 412
 413 int btrfs_is_empty_uuid(u8 *uuid)
 414 {
 415         int i;
 416
 417         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 418                 if (uuid[i])
 419                         return 0;
 420         }
 421         return 1;
 422 }
 423
 424 static noinline int create_subvol(struct inode *dir,
 425                                   struct dentry *dentry,
 426                                   char *name, int namelen,
 427                                   u64 *async_transid,
 428                                   struct btrfs_qgroup_inherit *inherit)
 429 {
 430         struct btrfs_trans_handle *trans;
 431         struct btrfs_key key;
 432         struct btrfs_root_item root_item;
 433         struct btrfs_inode_item *inode_item;
 434         struct extent_buffer *leaf;
 435         struct btrfs_root *root = BTRFS_I(dir)->root;
 436         struct btrfs_root *new_root;
 437         struct btrfs_block_rsv block_rsv;
 438         struct timespec cur_time = CURRENT_TIME;
 439         struct inode *inode;
 440         int ret;
 441         int err;
 442         u64 objectid;
 443         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 444         u64 index = 0;
 445         u64 qgroup_reserved;
 446         uuid_le new_uuid;
 447
 448         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 449         if (ret)
 450                 return ret;
 451
 452         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 453         /*
 454          * The same as the snapshot creation, please see the comment
 455          * of create_snapshot().
 456          */
 457         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 458                                                8, &qgroup_reserved, false);
 459         if (ret)
 460                 return ret;
 461
 462         trans = btrfs_start_transaction(root, 0);
 463         if (IS_ERR(trans)) {
 464                 ret = PTR_ERR(trans);
 465                 btrfs_subvolume_release_metadata(root, &block_rsv,
 466                                                  qgroup_reserved);
 467                 return ret;
 468         }
 469         trans->block_rsv = &block_rsv;
 470         trans->bytes_reserved = block_rsv.size;
 471
 472         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 473         if (ret)
 474                 goto fail;
 475
 476         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 477                                       0, objectid, NULL, 0, 0, 0);
 478         if (IS_ERR(leaf)) {
 479                 ret = PTR_ERR(leaf);
 480                 goto fail;
 481         }
 482
 483         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 484         btrfs_set_header_bytenr(leaf, leaf->start);
 485         btrfs_set_header_generation(leaf, trans->transid);
 486         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 487         btrfs_set_header_owner(leaf, objectid);
 488
 489         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 490                             BTRFS_FSID_SIZE);
 491         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 492                             btrfs_header_chunk_tree_uuid(leaf),
 493                             BTRFS_UUID_SIZE);
 494         btrfs_mark_buffer_dirty(leaf);
 495
 496         memset(&root_item, 0, sizeof(root_item));
 497
 498         inode_item = &root_item.inode;
 499         btrfs_set_stack_inode_generation(inode_item, 1);
 500         btrfs_set_stack_inode_size(inode_item, 3);
 501         btrfs_set_stack_inode_nlink(inode_item, 1);
 502         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
 503         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 504
 505         btrfs_set_root_flags(&root_item, 0);
 506         btrfs_set_root_limit(&root_item, 0);
 507         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 508
 509         btrfs_set_root_bytenr(&root_item, leaf->start);
 510         btrfs_set_root_generation(&root_item, trans->transid);
 511         btrfs_set_root_level(&root_item, 0);
 512         btrfs_set_root_refs(&root_item, 1);
 513         btrfs_set_root_used(&root_item, leaf->len);
 514         btrfs_set_root_last_snapshot(&root_item, 0);
 515
 516         btrfs_set_root_generation_v2(&root_item,
 517                         btrfs_root_generation(&root_item));
 518         uuid_le_gen(&new_uuid);
 519         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 520         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 521         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 522         root_item.ctime = root_item.otime;
 523         btrfs_set_root_ctransid(&root_item, trans->transid);
 524         btrfs_set_root_otransid(&root_item, trans->transid);
 525
 526         btrfs_tree_unlock(leaf);
 527         free_extent_buffer(leaf);
 528         leaf = NULL;
 529
 530         btrfs_set_root_dirid(&root_item, new_dirid);
 531
 532         key.objectid = objectid;
 533         key.offset = 0;
 534         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 535         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 536                                 &root_item);
 537         if (ret)
 538                 goto fail;
 539
 540         key.offset = (u64)-1;
 541         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 542         if (IS_ERR(new_root)) {
 543                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 544                 ret = PTR_ERR(new_root);
 545                 goto fail;
 546         }
 547
 548         btrfs_record_root_in_trans(trans, new_root);
 549
 550         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 551         if (ret) {
 552                 /* We potentially lose an unused inode item here */
 553                 btrfs_abort_transaction(trans, root, ret);
 554                 goto fail;
 555         }
 556
 557         /*
 558          * insert the directory item
 559          */
 560         ret = btrfs_set_inode_index(dir, &index);
 561         if (ret) {
 562                 btrfs_abort_transaction(trans, root, ret);
 563                 goto fail;
 564         }
 565
 566         ret = btrfs_insert_dir_item(trans, root,
 567                                     name, namelen, dir, &key,
 568                                     BTRFS_FT_DIR, index);
 569         if (ret) {
 570                 btrfs_abort_transaction(trans, root, ret);
 571                 goto fail;
 572         }
 573
 574         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 575         ret = btrfs_update_inode(trans, root, dir);
 576         BUG_ON(ret);
 577
 578         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 579                                  objectid, root->root_key.objectid,
 580                                  btrfs_ino(dir), index, name, namelen);
 581         BUG_ON(ret);
 582
 583         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 584                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 585                                   objectid);
 586         if (ret)
 587                 btrfs_abort_transaction(trans, root, ret);
 588
 589 fail:
 590         trans->block_rsv = NULL;
 591         trans->bytes_reserved = 0;
 592         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 593
 594         if (async_transid) {
 595                 *async_transid = trans->transid;
 596                 err = btrfs_commit_transaction_async(trans, root, 1);
 597                 if (err)
 598                         err = btrfs_commit_transaction(trans, root);
 599         } else {
 600                 err = btrfs_commit_transaction(trans, root);
 601         }
 602         if (err && !ret)
 603                 ret = err;
 604
 605         if (!ret) {
 606                 inode = btrfs_lookup_dentry(dir, dentry);
 607                 if (IS_ERR(inode))
 608                         return PTR_ERR(inode);
 609                 d_instantiate(dentry, inode);
 610         }
 611         return ret;
 612 }
 613
 614 static void btrfs_wait_nocow_write(struct btrfs_root *root)
 615 {
 616         s64 writers;
 617         DEFINE_WAIT(wait);
 618
 619         do {
 620                 prepare_to_wait(&root->subv_writers->wait, &wait,
 621                                 TASK_UNINTERRUPTIBLE);
 622
 623                 writers = percpu_counter_sum(&root->subv_writers->counter);
 624                 if (writers)
 625                         schedule();
 626
 627                 finish_wait(&root->subv_writers->wait, &wait);
 628         } while (writers);
 629 }
 630
 631 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 632                            struct dentry *dentry, char *name, int namelen,
 633                            u64 *async_transid, bool readonly,
 634                            struct btrfs_qgroup_inherit *inherit)
 635 {
 636         struct inode *inode;
 637         struct btrfs_pending_snapshot *pending_snapshot;
 638         struct btrfs_trans_handle *trans;
 639         int ret;
 640
 641         if (!root->ref_cows)
 642                 return -EINVAL;
 643
 644         atomic_inc(&root->will_be_snapshoted);
 645         smp_mb__after_atomic_inc();
 646         btrfs_wait_nocow_write(root);
 647
 648         ret = btrfs_start_delalloc_inodes(root, 0);
 649         if (ret)
 650                 goto out;
 651
 652         btrfs_wait_ordered_extents(root, -1);
 653
 654         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 655         if (!pending_snapshot) {
 656                 ret = -ENOMEM;
 657                 goto out;
 658         }
 659
 660         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 661                              BTRFS_BLOCK_RSV_TEMP);
 662         /*
 663          * 1 - parent dir inode
 664          * 2 - dir entries
 665          * 1 - root item
 666          * 2 - root ref/backref
 667          * 1 - root of snapshot
 668          * 1 - UUID item
 669          */
 670         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 671                                         &pending_snapshot->block_rsv, 8,
 672                                         &pending_snapshot->qgroup_reserved,
 673                                         false);
 674         if (ret)
 675                 goto free;
 676
 677         pending_snapshot->dentry = dentry;
 678         pending_snapshot->root = root;
 679         pending_snapshot->readonly = readonly;
 680         pending_snapshot->dir = dir;
 681         pending_snapshot->inherit = inherit;
 682
 683         trans = btrfs_start_transaction(root, 0);
 684         if (IS_ERR(trans)) {
 685                 ret = PTR_ERR(trans);
 686                 goto fail;
 687         }
 688
 689         spin_lock(&root->fs_info->trans_lock);
 690         list_add(&pending_snapshot->list,
 691                  &trans->transaction->pending_snapshots);
 692         spin_unlock(&root->fs_info->trans_lock);
 693         if (async_transid) {
 694                 *async_transid = trans->transid;
 695                 ret = btrfs_commit_transaction_async(trans,
 696                                      root->fs_info->extent_root, 1);
 697                 if (ret)
 698                         ret = btrfs_commit_transaction(trans, root);
 699         } else {
 700                 ret = btrfs_commit_transaction(trans,
 701                                                root->fs_info->extent_root);
 702         }
 703         if (ret)
 704                 goto fail;
 705
 706         ret = pending_snapshot->error;
 707         if (ret)
 708                 goto fail;
 709
 710         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 711         if (ret)
 712                 goto fail;
 713
 714         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 715         if (IS_ERR(inode)) {
 716                 ret = PTR_ERR(inode);
 717                 goto fail;
 718         }
 719
 720         d_instantiate(dentry, inode);
 721         ret = 0;
 722 fail:
 723         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 724                                          &pending_snapshot->block_rsv,
 725                                          pending_snapshot->qgroup_reserved);
 726 free:
 727         kfree(pending_snapshot);
 728 out:
 729         atomic_dec(&root->will_be_snapshoted);
 730         return ret;
 731 }
 732
 733 /*  copy of check_sticky in fs/namei.c()
 734 * It's inline, so penalty for filesystems that don't use sticky bit is
 735 * minimal.
 736 */
 737 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 738 {
 739         kuid_t fsuid = current_fsuid();
 740
 741         if (!(dir->i_mode & S_ISVTX))
 742                 return 0;
 743         if (uid_eq(inode->i_uid, fsuid))
 744                 return 0;
 745         if (uid_eq(dir->i_uid, fsuid))
 746                 return 0;
 747         return !capable(CAP_FOWNER);
 748 }
 749
 750 /*  copy of may_delete in fs/namei.c()
 751  *      Check whether we can remove a link victim from directory dir, check
 752  *  whether the type of victim is right.
 753  *  1. We can't do it if dir is read-only (done in permission())
 754  *  2. We should have write and exec permissions on dir
 755  *  3. We can't remove anything from append-only dir
 756  *  4. We can't do anything with immutable dir (done in permission())
 757  *  5. If the sticky bit on dir is set we should either
 758  *      a. be owner of dir, or
 759  *      b. be owner of victim, or
 760  *      c. have CAP_FOWNER capability
 761  *  6. If the victim is append-only or immutable we can't do antyhing with
 762  *     links pointing to it.
 763  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 764  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 765  *  9. We can't remove a root or mountpoint.
 766  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 767  *     nfs_async_unlink().
 768  */
 769
 770 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 771 {
 772         int error;
 773
 774         if (!victim->d_inode)
 775                 return -ENOENT;
 776
 777         BUG_ON(victim->d_parent->d_inode != dir);
 778         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 779
 780         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 781         if (error)
 782                 return error;
 783         if (IS_APPEND(dir))
 784                 return -EPERM;
 785         if (btrfs_check_sticky(dir, victim->d_inode)||
 786                 IS_APPEND(victim->d_inode)||
 787             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 788                 return -EPERM;
 789         if (isdir) {
 790                 if (!S_ISDIR(victim->d_inode->i_mode))
 791                         return -ENOTDIR;
 792                 if (IS_ROOT(victim))
 793                         return -EBUSY;
 794         } else if (S_ISDIR(victim->d_inode->i_mode))
 795                 return -EISDIR;
 796         if (IS_DEADDIR(dir))
 797                 return -ENOENT;
 798         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 799                 return -EBUSY;
 800         return 0;
 801 }
 802
 803 /* copy of may_create in fs/namei.c() */
 804 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 805 {
 806         if (child->d_inode)
 807                 return -EEXIST;
 808         if (IS_DEADDIR(dir))
 809                 return -ENOENT;
 810         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 811 }
 812
 813 /*
 814  * Create a new subvolume below @parent.  This is largely modeled after
 815  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 816  * inside this filesystem so it's quite a bit simpler.
 817  */
 818 static noinline int btrfs_mksubvol(struct path *parent,
 819                                    char *name, int namelen,
 820                                    struct btrfs_root *snap_src,
 821                                    u64 *async_transid, bool readonly,
 822                                    struct btrfs_qgroup_inherit *inherit)
 823 {
 824         struct inode *dir  = parent->dentry->d_inode;
 825         struct dentry *dentry;
 826         int error;
 827
 828         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 829         if (error == -EINTR)
 830                 return error;
 831
 832         dentry = lookup_one_len(name, parent->dentry, namelen);
 833         error = PTR_ERR(dentry);
 834         if (IS_ERR(dentry))
 835                 goto out_unlock;
 836
 837         error = -EEXIST;
 838         if (dentry->d_inode)
 839                 goto out_dput;
 840
 841         error = btrfs_may_create(dir, dentry);
 842         if (error)
 843                 goto out_dput;
 844
 845         /*
 846          * even if this name doesn't exist, we may get hash collisions.
 847          * check for them now when we can safely fail
 848          */
 849         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 850                                                dir->i_ino, name,
 851                                                namelen);
 852         if (error)
 853                 goto out_dput;
 854
 855         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 856
 857         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 858                 goto out_up_read;
 859
 860         if (snap_src) {
 861                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 862                                         async_transid, readonly, inherit);
 863         } else {
 864                 error = create_subvol(dir, dentry, name, namelen,
 865                                       async_transid, inherit);
 866         }
 867         if (!error)
 868                 fsnotify_mkdir(dir, dentry);
 869 out_up_read:
 870         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 871 out_dput:
 872         dput(dentry);
 873 out_unlock:
 874         mutex_unlock(&dir->i_mutex);
 875         return error;
 876 }
 877
 878 /*
 879  * When we're defragging a range, we don't want to kick it off again
 880  * if it is really just waiting for delalloc to send it down.
 881  * If we find a nice big extent or delalloc range for the bytes in the
 882  * file you want to defrag, we return 0 to let you know to skip this
 883  * part of the file
 884  */
 885 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 886 {
 887         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 888         struct extent_map *em = NULL;
 889         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 890         u64 end;
 891
 892         read_lock(&em_tree->lock);
 893         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 894         read_unlock(&em_tree->lock);
 895
 896         if (em) {
 897                 end = extent_map_end(em);
 898                 free_extent_map(em);
 899                 if (end - offset > thresh)
 900                         return 0;
 901         }
 902         /* if we already have a nice delalloc here, just stop */
 903         thresh /= 2;
 904         end = count_range_bits(io_tree, &offset, offset + thresh,
 905                                thresh, EXTENT_DELALLOC, 1);
 906         if (end >= thresh)
 907                 return 0;
 908         return 1;
 909 }
 910
 911 /*
 912  * helper function to walk through a file and find extents
 913  * newer than a specific transid, and smaller than thresh.
 914  *
 915  * This is used by the defragging code to find new and small
 916  * extents
 917  */
 918 static int find_new_extents(struct btrfs_root *root,
 919                             struct inode *inode, u64 newer_than,
 920                             u64 *off, int thresh)
 921 {
 922         struct btrfs_path *path;
 923         struct btrfs_key min_key;
 924         struct extent_buffer *leaf;
 925         struct btrfs_file_extent_item *extent;
 926         int type;
 927         int ret;
 928         u64 ino = btrfs_ino(inode);
 929
 930         path = btrfs_alloc_path();
 931         if (!path)
 932                 return -ENOMEM;
 933
 934         min_key.objectid = ino;
 935         min_key.type = BTRFS_EXTENT_DATA_KEY;
 936         min_key.offset = *off;
 937
 938         while (1) {
 939                 path->keep_locks = 1;
 940                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 941                 if (ret != 0)
 942                         goto none;
 943                 path->keep_locks = 0;
 944                 btrfs_unlock_up_safe(path, 1);
 945 process_slot:
 946                 if (min_key.objectid != ino)
 947                         goto none;
 948                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 949                         goto none;
 950
 951                 leaf = path->nodes[0];
 952                 extent = btrfs_item_ptr(leaf, path->slots[0],
 953                                         struct btrfs_file_extent_item);
 954
 955                 type = btrfs_file_extent_type(leaf, extent);
 956                 if (type == BTRFS_FILE_EXTENT_REG &&
 957                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 958                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 959                         *off = min_key.offset;
 960                         btrfs_free_path(path);
 961                         return 0;
 962                 }
 963
 964                 path->slots[0]++;
 965                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 966                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 967                         goto process_slot;
 968                 }
 969
 970                 if (min_key.offset == (u64)-1)
 971                         goto none;
 972
 973                 min_key.offset++;
 974                 btrfs_release_path(path);
 975         }
 976 none:
 977         btrfs_free_path(path);
 978         return -ENOENT;
 979 }
 980
 981 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 982 {
 983         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 984         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 985         struct extent_map *em;
 986         u64 len = PAGE_CACHE_SIZE;
 987
 988         /*
 989          * hopefully we have this extent in the tree already, try without
 990          * the full extent lock
 991          */
 992         read_lock(&em_tree->lock);
 993         em = lookup_extent_mapping(em_tree, start, len);
 994         read_unlock(&em_tree->lock);
 995
 996         if (!em) {
 997                 struct extent_state *cached = NULL;
 998                 u64 end = start + len - 1;
 999
1000                 /* get the big lock and read metadata off disk */
1001                 lock_extent_bits(io_tree, start, end, 0, &cached);
1002                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1003                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1004
1005                 if (IS_ERR(em))
1006                         return NULL;
1007         }
1008
1009         return em;
1010 }
1011
1012 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1013 {
1014         struct extent_map *next;
1015         bool ret = true;
1016
1017         /* this is the last extent */
1018         if (em->start + em->len >= i_size_read(inode))
1019                 return false;
1020
1021         next = defrag_lookup_extent(inode, em->start + em->len);
1022         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE ||
1023             (em->block_start + em->block_len == next->block_start))
1024                 ret = false;
1025
1026         free_extent_map(next);
1027         return ret;
1028 }
1029
1030 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
1031                                u64 *last_len, u64 *skip, u64 *defrag_end,
1032                                int compress)
1033 {
1034         struct extent_map *em;
1035         int ret = 1;
1036         bool next_mergeable = true;
1037
1038         /*
1039          * make sure that once we start defragging an extent, we keep on
1040          * defragging it
1041          */
1042         if (start < *defrag_end)
1043                 return 1;
1044
1045         *skip = 0;
1046
1047         em = defrag_lookup_extent(inode, start);
1048         if (!em)
1049                 return 0;
1050
1051         /* this will cover holes, and inline extents */
1052         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1053                 ret = 0;
1054                 goto out;
1055         }
1056
1057         next_mergeable = defrag_check_next_extent(inode, em);
1058
1059         /*
1060          * we hit a real extent, if it is big or the next extent is not a
1061          * real extent, don't bother defragging it
1062          */
1063         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1064             (em->len >= thresh || !next_mergeable))
1065                 ret = 0;
1066 out:
1067         /*
1068          * last_len ends up being a counter of how many bytes we've defragged.
1069          * every time we choose not to defrag an extent, we reset *last_len
1070          * so that the next tiny extent will force a defrag.
1071          *
1072          * The end result of this is that tiny extents before a single big
1073          * extent will force at least part of that big extent to be defragged.
1074          */
1075         if (ret) {
1076                 *defrag_end = extent_map_end(em);
1077         } else {
1078                 *last_len = 0;
1079                 *skip = extent_map_end(em);
1080                 *defrag_end = 0;
1081         }
1082
1083         free_extent_map(em);
1084         return ret;
1085 }
1086
1087 /*
1088  * it doesn't do much good to defrag one or two pages
1089  * at a time.  This pulls in a nice chunk of pages
1090  * to COW and defrag.
1091  *
1092  * It also makes sure the delalloc code has enough
1093  * dirty data to avoid making new small extents as part
1094  * of the defrag
1095  *
1096  * It's a good idea to start RA on this range
1097  * before calling this.
1098  */
1099 static int cluster_pages_for_defrag(struct inode *inode,
1100                                     struct page **pages,
1101                                     unsigned long start_index,
1102                                     unsigned long num_pages)
1103 {
1104         unsigned long file_end;
1105         u64 isize = i_size_read(inode);
1106         u64 page_start;
1107         u64 page_end;
1108         u64 page_cnt;
1109         int ret;
1110         int i;
1111         int i_done;
1112         struct btrfs_ordered_extent *ordered;
1113         struct extent_state *cached_state = NULL;
1114         struct extent_io_tree *tree;
1115         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1116
1117         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1118         if (!isize || start_index > file_end)
1119                 return 0;
1120
1121         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1122
1123         ret = btrfs_delalloc_reserve_space(inode,
1124                                            page_cnt << PAGE_CACHE_SHIFT);
1125         if (ret)
1126                 return ret;
1127         i_done = 0;
1128         tree = &BTRFS_I(inode)->io_tree;
1129
1130         /* step one, lock all the pages */
1131         for (i = 0; i < page_cnt; i++) {
1132                 struct page *page;
1133 again:
1134                 page = find_or_create_page(inode->i_mapping,
1135                                            start_index + i, mask);
1136                 if (!page)
1137                         break;
1138
1139                 page_start = page_offset(page);
1140                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1141                 while (1) {
1142                         lock_extent_bits(tree, page_start, page_end,
1143                                          0, &cached_state);
1144                         ordered = btrfs_lookup_ordered_extent(inode,
1145                                                               page_start);
1146                         unlock_extent_cached(tree, page_start, page_end,
1147                                              &cached_state, GFP_NOFS);
1148                         if (!ordered)
1149                                 break;
1150
1151                         unlock_page(page);
1152                         btrfs_start_ordered_extent(inode, ordered, 1);
1153                         btrfs_put_ordered_extent(ordered);
1154                         lock_page(page);
1155                         /*
1156                          * we unlocked the page above, so we need check if
1157                          * it was released or not.
1158                          */
1159                         if (page->mapping != inode->i_mapping) {
1160                                 unlock_page(page);
1161                                 page_cache_release(page);
1162                                 goto again;
1163                         }
1164                 }
1165
1166                 if (!PageUptodate(page)) {
1167                         btrfs_readpage(NULL, page);
1168                         lock_page(page);
1169                         if (!PageUptodate(page)) {
1170                                 unlock_page(page);
1171                                 page_cache_release(page);
1172                                 ret = -EIO;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (page->mapping != inode->i_mapping) {
1178                         unlock_page(page);
1179                         page_cache_release(page);
1180                         goto again;
1181                 }
1182
1183                 pages[i] = page;
1184                 i_done++;
1185         }
1186         if (!i_done || ret)
1187                 goto out;
1188
1189         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1190                 goto out;
1191
1192         /*
1193          * so now we have a nice long stream of locked
1194          * and up to date pages, lets wait on them
1195          */
1196         for (i = 0; i < i_done; i++)
1197                 wait_on_page_writeback(pages[i]);
1198
1199         page_start = page_offset(pages[0]);
1200         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1201
1202         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1203                          page_start, page_end - 1, 0, &cached_state);
1204         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1205                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1207                           &cached_state, GFP_NOFS);
1208
1209         if (i_done != page_cnt) {
1210                 spin_lock(&BTRFS_I(inode)->lock);
1211                 BTRFS_I(inode)->outstanding_extents++;
1212                 spin_unlock(&BTRFS_I(inode)->lock);
1213                 btrfs_delalloc_release_space(inode,
1214                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1215         }
1216
1217
1218         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1219                           &cached_state, GFP_NOFS);
1220
1221         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1222                              page_start, page_end - 1, &cached_state,
1223                              GFP_NOFS);
1224
1225         for (i = 0; i < i_done; i++) {
1226                 clear_page_dirty_for_io(pages[i]);
1227                 ClearPageChecked(pages[i]);
1228                 set_page_extent_mapped(pages[i]);
1229                 set_page_dirty(pages[i]);
1230                 unlock_page(pages[i]);
1231                 page_cache_release(pages[i]);
1232         }
1233         return i_done;
1234 out:
1235         for (i = 0; i < i_done; i++) {
1236                 unlock_page(pages[i]);
1237                 page_cache_release(pages[i]);
1238         }
1239         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1240         return ret;
1241
1242 }
1243
1244 int btrfs_defrag_file(struct inode *inode, struct file *file,
1245                       struct btrfs_ioctl_defrag_range_args *range,
1246                       u64 newer_than, unsigned long max_to_defrag)
1247 {
1248         struct btrfs_root *root = BTRFS_I(inode)->root;
1249         struct file_ra_state *ra = NULL;
1250         unsigned long last_index;
1251         u64 isize = i_size_read(inode);
1252         u64 last_len = 0;
1253         u64 skip = 0;
1254         u64 defrag_end = 0;
1255         u64 newer_off = range->start;
1256         unsigned long i;
1257         unsigned long ra_index = 0;
1258         int ret;
1259         int defrag_count = 0;
1260         int compress_type = BTRFS_COMPRESS_ZLIB;
1261         int extent_thresh = range->extent_thresh;
1262         unsigned long max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1263         unsigned long cluster = max_cluster;
1264         u64 new_align = ~((u64)128 * 1024 - 1);
1265         struct page **pages = NULL;
1266
1267         if (isize == 0)
1268                 return 0;
1269
1270         if (range->start >= isize)
1271                 return -EINVAL;
1272
1273         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1274                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1275                         return -EINVAL;
1276                 if (range->compress_type)
1277                         compress_type = range->compress_type;
1278         }
1279
1280         if (extent_thresh == 0)
1281                 extent_thresh = 256 * 1024;
1282
1283         /*
1284          * if we were not given a file, allocate a readahead
1285          * context
1286          */
1287         if (!file) {
1288                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1289                 if (!ra)
1290                         return -ENOMEM;
1291                 file_ra_state_init(ra, inode->i_mapping);
1292         } else {
1293                 ra = &file->f_ra;
1294         }
1295
1296         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1297                         GFP_NOFS);
1298         if (!pages) {
1299                 ret = -ENOMEM;
1300                 goto out_ra;
1301         }
1302
1303         /* find the last page to defrag */
1304         if (range->start + range->len > range->start) {
1305                 last_index = min_t(u64, isize - 1,
1306                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1307         } else {
1308                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1309         }
1310
1311         if (newer_than) {
1312                 ret = find_new_extents(root, inode, newer_than,
1313                                        &newer_off, 64 * 1024);
1314                 if (!ret) {
1315                         range->start = newer_off;
1316                         /*
1317                          * we always align our defrag to help keep
1318                          * the extents in the file evenly spaced
1319                          */
1320                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1321                 } else
1322                         goto out_ra;
1323         } else {
1324                 i = range->start >> PAGE_CACHE_SHIFT;
1325         }
1326         if (!max_to_defrag)
1327                 max_to_defrag = last_index + 1;
1328
1329         /*
1330          * make writeback starts from i, so the defrag range can be
1331          * written sequentially.
1332          */
1333         if (i < inode->i_mapping->writeback_index)
1334                 inode->i_mapping->writeback_index = i;
1335
1336         while (i <= last_index && defrag_count < max_to_defrag &&
1337                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1338                 PAGE_CACHE_SHIFT)) {
1339                 /*
1340                  * make sure we stop running if someone unmounts
1341                  * the FS
1342                  */
1343                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1344                         break;
1345
1346                 if (btrfs_defrag_cancelled(root->fs_info)) {
1347                         printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n");
1348                         ret = -EAGAIN;
1349                         break;
1350                 }
1351
1352                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1353                                          extent_thresh, &last_len, &skip,
1354                                          &defrag_end, range->flags &
1355                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1356                         unsigned long next;
1357                         /*
1358                          * the should_defrag function tells us how much to skip
1359                          * bump our counter by the suggested amount
1360                          */
1361                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1362                         i = max(i + 1, next);
1363                         continue;
1364                 }
1365
1366                 if (!newer_than) {
1367                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1368                                    PAGE_CACHE_SHIFT) - i;
1369                         cluster = min(cluster, max_cluster);
1370                 } else {
1371                         cluster = max_cluster;
1372                 }
1373
1374                 if (i + cluster > ra_index) {
1375                         ra_index = max(i, ra_index);
1376                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1377                                        cluster);
1378                         ra_index += max_cluster;
1379                 }
1380
1381                 mutex_lock(&inode->i_mutex);
1382                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1383                         BTRFS_I(inode)->force_compress = compress_type;
1384                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1385                 if (ret < 0) {
1386                         mutex_unlock(&inode->i_mutex);
1387                         goto out_ra;
1388                 }
1389
1390                 defrag_count += ret;
1391                 balance_dirty_pages_ratelimited(inode->i_mapping);
1392                 mutex_unlock(&inode->i_mutex);
1393
1394                 if (newer_than) {
1395                         if (newer_off == (u64)-1)
1396                                 break;
1397
1398                         if (ret > 0)
1399                                 i += ret;
1400
1401                         newer_off = max(newer_off + 1,
1402                                         (u64)i << PAGE_CACHE_SHIFT);
1403
1404                         ret = find_new_extents(root, inode,
1405                                                newer_than, &newer_off,
1406                                                64 * 1024);
1407                         if (!ret) {
1408                                 range->start = newer_off;
1409                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1410                         } else {
1411                                 break;
1412                         }
1413                 } else {
1414                         if (ret > 0) {
1415                                 i += ret;
1416                                 last_len += ret << PAGE_CACHE_SHIFT;
1417                         } else {
1418                                 i++;
1419                                 last_len = 0;
1420                         }
1421                 }
1422         }
1423
1424         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1425                 filemap_flush(inode->i_mapping);
1426                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1427                              &BTRFS_I(inode)->runtime_flags))
1428                         filemap_flush(inode->i_mapping);
1429         }
1430
1431         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1432                 /* the filemap_flush will queue IO into the worker threads, but
1433                  * we have to make sure the IO is actually started and that
1434                  * ordered extents get created before we return
1435                  */
1436                 atomic_inc(&root->fs_info->async_submit_draining);
1437                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1438                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1439                         wait_event(root->fs_info->async_submit_wait,
1440                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1441                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1442                 }
1443                 atomic_dec(&root->fs_info->async_submit_draining);
1444         }
1445
1446         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1447                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1448         }
1449
1450         ret = defrag_count;
1451
1452 out_ra:
1453         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1454                 mutex_lock(&inode->i_mutex);
1455                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1456                 mutex_unlock(&inode->i_mutex);
1457         }
1458         if (!file)
1459                 kfree(ra);
1460         kfree(pages);
1461         return ret;
1462 }
1463
1464 static noinline int btrfs_ioctl_resize(struct file *file,
1465                                         void __user *arg)
1466 {
1467         u64 new_size;
1468         u64 old_size;
1469         u64 devid = 1;
1470         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1471         struct btrfs_ioctl_vol_args *vol_args;
1472         struct btrfs_trans_handle *trans;
1473         struct btrfs_device *device = NULL;
1474         char *sizestr;
1475         char *retptr;
1476         char *devstr = NULL;
1477         int ret = 0;
1478         int mod = 0;
1479
1480         if (!capable(CAP_SYS_ADMIN))
1481                 return -EPERM;
1482
1483         ret = mnt_want_write_file(file);
1484         if (ret)
1485                 return ret;
1486
1487         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1488                         1)) {
1489                 mnt_drop_write_file(file);
1490                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1491         }
1492
1493         mutex_lock(&root->fs_info->volume_mutex);
1494         vol_args = memdup_user(arg, sizeof(*vol_args));
1495         if (IS_ERR(vol_args)) {
1496                 ret = PTR_ERR(vol_args);
1497                 goto out;
1498         }
1499
1500         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1501
1502         sizestr = vol_args->name;
1503         devstr = strchr(sizestr, ':');
1504         if (devstr) {
1505                 char *end;
1506                 sizestr = devstr + 1;
1507                 *devstr = '\0';
1508                 devstr = vol_args->name;
1509                 devid = simple_strtoull(devstr, &end, 10);
1510                 if (!devid) {
1511                         ret = -EINVAL;
1512                         goto out_free;
1513                 }
1514                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1515         }
1516
1517         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1518         if (!device) {
1519                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1520                        devid);
1521                 ret = -ENODEV;
1522                 goto out_free;
1523         }
1524
1525         if (!device->writeable) {
1526                 btrfs_info(root->fs_info,
1527                            "resizer unable to apply on readonly device %llu",
1528                        devid);
1529                 ret = -EPERM;
1530                 goto out_free;
1531         }
1532
1533         if (!strcmp(sizestr, "max"))
1534                 new_size = device->bdev->bd_inode->i_size;
1535         else {
1536                 if (sizestr[0] == '-') {
1537                         mod = -1;
1538                         sizestr++;
1539                 } else if (sizestr[0] == '+') {
1540                         mod = 1;
1541                         sizestr++;
1542                 }
1543                 new_size = memparse(sizestr, &retptr);
1544                 if (*retptr != '\0' || new_size == 0) {
1545                         ret = -EINVAL;
1546                         goto out_free;
1547                 }
1548         }
1549
1550         if (device->is_tgtdev_for_dev_replace) {
1551                 ret = -EPERM;
1552                 goto out_free;
1553         }
1554
1555         old_size = device->total_bytes;
1556
1557         if (mod < 0) {
1558                 if (new_size > old_size) {
1559                         ret = -EINVAL;
1560                         goto out_free;
1561                 }
1562                 new_size = old_size - new_size;
1563         } else if (mod > 0) {
1564                 if (new_size > ULLONG_MAX - old_size) {
1565                         ret = -EINVAL;
1566                         goto out_free;
1567                 }
1568                 new_size = old_size + new_size;
1569         }
1570
1571         if (new_size < 256 * 1024 * 1024) {
1572                 ret = -EINVAL;
1573                 goto out_free;
1574         }
1575         if (new_size > device->bdev->bd_inode->i_size) {
1576                 ret = -EFBIG;
1577                 goto out_free;
1578         }
1579
1580         do_div(new_size, root->sectorsize);
1581         new_size *= root->sectorsize;
1582
1583         printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n",
1584                       rcu_str_deref(device->name), new_size);
1585
1586         if (new_size > old_size) {
1587                 trans = btrfs_start_transaction(root, 0);
1588                 if (IS_ERR(trans)) {
1589                         ret = PTR_ERR(trans);
1590                         goto out_free;
1591                 }
1592                 ret = btrfs_grow_device(trans, device, new_size);
1593                 btrfs_commit_transaction(trans, root);
1594         } else if (new_size < old_size) {
1595                 ret = btrfs_shrink_device(device, new_size);
1596         } /* equal, nothing need to do */
1597
1598 out_free:
1599         kfree(vol_args);
1600 out:
1601         mutex_unlock(&root->fs_info->volume_mutex);
1602         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1603         mnt_drop_write_file(file);
1604         return ret;
1605 }
1606
1607 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1608                                 char *name, unsigned long fd, int subvol,
1609                                 u64 *transid, bool readonly,
1610                                 struct btrfs_qgroup_inherit *inherit)
1611 {
1612         int namelen;
1613         int ret = 0;
1614
1615         ret = mnt_want_write_file(file);
1616         if (ret)
1617                 goto out;
1618
1619         namelen = strlen(name);
1620         if (strchr(name, '/')) {
1621                 ret = -EINVAL;
1622                 goto out_drop_write;
1623         }
1624
1625         if (name[0] == '.' &&
1626            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1627                 ret = -EEXIST;
1628                 goto out_drop_write;
1629         }
1630
1631         if (subvol) {
1632                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1633                                      NULL, transid, readonly, inherit);
1634         } else {
1635                 struct fd src = fdget(fd);
1636                 struct inode *src_inode;
1637                 if (!src.file) {
1638                         ret = -EINVAL;
1639                         goto out_drop_write;
1640                 }
1641
1642                 src_inode = file_inode(src.file);
1643                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1644                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1645                                    "Snapshot src from another FS");
1646                         ret = -EXDEV;
1647                 } else if (!inode_owner_or_capable(src_inode)) {
1648                         /*
1649                          * Subvolume creation is not restricted, but snapshots
1650                          * are limited to own subvolumes only
1651                          */
1652                         ret = -EPERM;
1653                 } else {
1654                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1655                                              BTRFS_I(src_inode)->root,
1656                                              transid, readonly, inherit);
1657                 }
1658                 fdput(src);
1659         }
1660 out_drop_write:
1661         mnt_drop_write_file(file);
1662 out:
1663         return ret;
1664 }
1665
1666 static noinline int btrfs_ioctl_snap_create(struct file *file,
1667                                             void __user *arg, int subvol)
1668 {
1669         struct btrfs_ioctl_vol_args *vol_args;
1670         int ret;
1671
1672         vol_args = memdup_user(arg, sizeof(*vol_args));
1673         if (IS_ERR(vol_args))
1674                 return PTR_ERR(vol_args);
1675         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1676
1677         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1678                                               vol_args->fd, subvol,
1679                                               NULL, false, NULL);
1680
1681         kfree(vol_args);
1682         return ret;
1683 }
1684
1685 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1686                                                void __user *arg, int subvol)
1687 {
1688         struct btrfs_ioctl_vol_args_v2 *vol_args;
1689         int ret;
1690         u64 transid = 0;
1691         u64 *ptr = NULL;
1692         bool readonly = false;
1693         struct btrfs_qgroup_inherit *inherit = NULL;
1694
1695         vol_args = memdup_user(arg, sizeof(*vol_args));
1696         if (IS_ERR(vol_args))
1697                 return PTR_ERR(vol_args);
1698         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1699
1700         if (vol_args->flags &
1701             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1702               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1703                 ret = -EOPNOTSUPP;
1704                 goto out;
1705         }
1706
1707         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1708                 ptr = &transid;
1709         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1710                 readonly = true;
1711         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1712                 if (vol_args->size > PAGE_CACHE_SIZE) {
1713                         ret = -EINVAL;
1714                         goto out;
1715                 }
1716                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1717                 if (IS_ERR(inherit)) {
1718                         ret = PTR_ERR(inherit);
1719                         goto out;
1720                 }
1721         }
1722
1723         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1724                                               vol_args->fd, subvol, ptr,
1725                                               readonly, inherit);
1726
1727         if (ret == 0 && ptr &&
1728             copy_to_user(arg +
1729                          offsetof(struct btrfs_ioctl_vol_args_v2,
1730                                   transid), ptr, sizeof(*ptr)))
1731                 ret = -EFAULT;
1732 out:
1733         kfree(vol_args);
1734         kfree(inherit);
1735         return ret;
1736 }
1737
1738 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1739                                                 void __user *arg)
1740 {
1741         struct inode *inode = file_inode(file);
1742         struct btrfs_root *root = BTRFS_I(inode)->root;
1743         int ret = 0;
1744         u64 flags = 0;
1745
1746         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1747                 return -EINVAL;
1748
1749         down_read(&root->fs_info->subvol_sem);
1750         if (btrfs_root_readonly(root))
1751                 flags |= BTRFS_SUBVOL_RDONLY;
1752         up_read(&root->fs_info->subvol_sem);
1753
1754         if (copy_to_user(arg, &flags, sizeof(flags)))
1755                 ret = -EFAULT;
1756
1757         return ret;
1758 }
1759
1760 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1761                                               void __user *arg)
1762 {
1763         struct inode *inode = file_inode(file);
1764         struct btrfs_root *root = BTRFS_I(inode)->root;
1765         struct btrfs_trans_handle *trans;
1766         u64 root_flags;
1767         u64 flags;
1768         int ret = 0;
1769
1770         if (!inode_owner_or_capable(inode))
1771                 return -EPERM;
1772
1773         ret = mnt_want_write_file(file);
1774         if (ret)
1775                 goto out;
1776
1777         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1778                 ret = -EINVAL;
1779                 goto out_drop_write;
1780         }
1781
1782         if (copy_from_user(&flags, arg, sizeof(flags))) {
1783                 ret = -EFAULT;
1784                 goto out_drop_write;
1785         }
1786
1787         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1788                 ret = -EINVAL;
1789                 goto out_drop_write;
1790         }
1791
1792         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1793                 ret = -EOPNOTSUPP;
1794                 goto out_drop_write;
1795         }
1796
1797         down_write(&root->fs_info->subvol_sem);
1798
1799         /* nothing to do */
1800         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1801                 goto out_drop_sem;
1802
1803         root_flags = btrfs_root_flags(&root->root_item);
1804         if (flags & BTRFS_SUBVOL_RDONLY) {
1805                 btrfs_set_root_flags(&root->root_item,
1806                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1807         } else {
1808                 /*
1809                  * Block RO -> RW transition if this subvolume is involved in
1810                  * send
1811                  */
1812                 spin_lock(&root->root_item_lock);
1813                 if (root->send_in_progress == 0) {
1814                         btrfs_set_root_flags(&root->root_item,
1815                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1816                         spin_unlock(&root->root_item_lock);
1817                 } else {
1818                         spin_unlock(&root->root_item_lock);
1819                         btrfs_warn(root->fs_info,
1820                         "Attempt to set subvolume %llu read-write during send",
1821                                         root->root_key.objectid);
1822                         ret = -EPERM;
1823                         goto out_drop_sem;
1824                 }
1825         }
1826
1827         trans = btrfs_start_transaction(root, 1);
1828         if (IS_ERR(trans)) {
1829                 ret = PTR_ERR(trans);
1830                 goto out_reset;
1831         }
1832
1833         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1834                                 &root->root_key, &root->root_item);
1835
1836         btrfs_commit_transaction(trans, root);
1837 out_reset:
1838         if (ret)
1839                 btrfs_set_root_flags(&root->root_item, root_flags);
1840 out_drop_sem:
1841         up_write(&root->fs_info->subvol_sem);
1842 out_drop_write:
1843         mnt_drop_write_file(file);
1844 out:
1845         return ret;
1846 }
1847
1848 /*
1849  * helper to check if the subvolume references other subvolumes
1850  */
1851 static noinline int may_destroy_subvol(struct btrfs_root *root)
1852 {
1853         struct btrfs_path *path;
1854         struct btrfs_dir_item *di;
1855         struct btrfs_key key;
1856         u64 dir_id;
1857         int ret;
1858
1859         path = btrfs_alloc_path();
1860         if (!path)
1861                 return -ENOMEM;
1862
1863         /* Make sure this root isn't set as the default subvol */
1864         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1865         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1866                                    dir_id, "default", 7, 0);
1867         if (di && !IS_ERR(di)) {
1868                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1869                 if (key.objectid == root->root_key.objectid) {
1870                         ret = -EPERM;
1871                         btrfs_err(root->fs_info, "deleting default subvolume "
1872                                   "%llu is not allowed", key.objectid);
1873                         goto out;
1874                 }
1875                 btrfs_release_path(path);
1876         }
1877
1878         key.objectid = root->root_key.objectid;
1879         key.type = BTRFS_ROOT_REF_KEY;
1880         key.offset = (u64)-1;
1881
1882         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1883                                 &key, path, 0, 0);
1884         if (ret < 0)
1885                 goto out;
1886         BUG_ON(ret == 0);
1887
1888         ret = 0;
1889         if (path->slots[0] > 0) {
1890                 path->slots[0]--;
1891                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1892                 if (key.objectid == root->root_key.objectid &&
1893                     key.type == BTRFS_ROOT_REF_KEY)
1894                         ret = -ENOTEMPTY;
1895         }
1896 out:
1897         btrfs_free_path(path);
1898         return ret;
1899 }
1900
1901 static noinline int key_in_sk(struct btrfs_key *key,
1902                               struct btrfs_ioctl_search_key *sk)
1903 {
1904         struct btrfs_key test;
1905         int ret;
1906
1907         test.objectid = sk->min_objectid;
1908         test.type = sk->min_type;
1909         test.offset = sk->min_offset;
1910
1911         ret = btrfs_comp_cpu_keys(key, &test);
1912         if (ret < 0)
1913                 return 0;
1914
1915         test.objectid = sk->max_objectid;
1916         test.type = sk->max_type;
1917         test.offset = sk->max_offset;
1918
1919         ret = btrfs_comp_cpu_keys(key, &test);
1920         if (ret > 0)
1921                 return 0;
1922         return 1;
1923 }
1924
1925 static noinline int copy_to_sk(struct btrfs_root *root,
1926                                struct btrfs_path *path,
1927                                struct btrfs_key *key,
1928                                struct btrfs_ioctl_search_key *sk,
1929                                char *buf,
1930                                unsigned long *sk_offset,
1931                                int *num_found)
1932 {
1933         u64 found_transid;
1934         struct extent_buffer *leaf;
1935         struct btrfs_ioctl_search_header sh;
1936         unsigned long item_off;
1937         unsigned long item_len;
1938         int nritems;
1939         int i;
1940         int slot;
1941         int ret = 0;
1942
1943         leaf = path->nodes[0];
1944         slot = path->slots[0];
1945         nritems = btrfs_header_nritems(leaf);
1946
1947         if (btrfs_header_generation(leaf) > sk->max_transid) {
1948                 i = nritems;
1949                 goto advance_key;
1950         }
1951         found_transid = btrfs_header_generation(leaf);
1952
1953         for (i = slot; i < nritems; i++) {
1954                 item_off = btrfs_item_ptr_offset(leaf, i);
1955                 item_len = btrfs_item_size_nr(leaf, i);
1956
1957                 btrfs_item_key_to_cpu(leaf, key, i);
1958                 if (!key_in_sk(key, sk))
1959                         continue;
1960
1961                 if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1962                         item_len = 0;
1963
1964                 if (sizeof(sh) + item_len + *sk_offset >
1965                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1966                         ret = 1;
1967                         goto overflow;
1968                 }
1969
1970                 sh.objectid = key->objectid;
1971                 sh.offset = key->offset;
1972                 sh.type = key->type;
1973                 sh.len = item_len;
1974                 sh.transid = found_transid;
1975
1976                 /* copy search result header */
1977                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1978                 *sk_offset += sizeof(sh);
1979
1980                 if (item_len) {
1981                         char *p = buf + *sk_offset;
1982                         /* copy the item */
1983                         read_extent_buffer(leaf, p,
1984                                            item_off, item_len);
1985                         *sk_offset += item_len;
1986                 }
1987                 (*num_found)++;
1988
1989                 if (*num_found >= sk->nr_items)
1990                         break;
1991         }
1992 advance_key:
1993         ret = 0;
1994         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1995                 key->offset++;
1996         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1997                 key->offset = 0;
1998                 key->type++;
1999         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
2000                 key->offset = 0;
2001                 key->type = 0;
2002                 key->objectid++;
2003         } else
2004                 ret = 1;
2005 overflow:
2006         return ret;
2007 }
2008
2009 static noinline int search_ioctl(struct inode *inode,
2010                                  struct btrfs_ioctl_search_args *args)
2011 {
2012         struct btrfs_root *root;
2013         struct btrfs_key key;
2014         struct btrfs_path *path;
2015         struct btrfs_ioctl_search_key *sk = &args->key;
2016         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2017         int ret;
2018         int num_found = 0;
2019         unsigned long sk_offset = 0;
2020
2021         path = btrfs_alloc_path();
2022         if (!path)
2023                 return -ENOMEM;
2024
2025         if (sk->tree_id == 0) {
2026                 /* search the root of the inode that was passed */
2027                 root = BTRFS_I(inode)->root;
2028         } else {
2029                 key.objectid = sk->tree_id;
2030                 key.type = BTRFS_ROOT_ITEM_KEY;
2031                 key.offset = (u64)-1;
2032                 root = btrfs_read_fs_root_no_name(info, &key);
2033                 if (IS_ERR(root)) {
2034                         printk(KERN_ERR "BTRFS: could not find root %llu\n",
2035                                sk->tree_id);
2036                         btrfs_free_path(path);
2037                         return -ENOENT;
2038                 }
2039         }
2040
2041         key.objectid = sk->min_objectid;
2042         key.type = sk->min_type;
2043         key.offset = sk->min_offset;
2044
2045         path->keep_locks = 1;
2046
2047         while (1) {
2048                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2049                 if (ret != 0) {
2050                         if (ret > 0)
2051                                 ret = 0;
2052                         goto err;
2053                 }
2054                 ret = copy_to_sk(root, path, &key, sk, args->buf,
2055                                  &sk_offset, &num_found);
2056                 btrfs_release_path(path);
2057                 if (ret || num_found >= sk->nr_items)
2058                         break;
2059
2060         }
2061         ret = 0;
2062 err:
2063         sk->nr_items = num_found;
2064         btrfs_free_path(path);
2065         return ret;
2066 }
2067
2068 static noinline int btrfs_ioctl_tree_search(struct file *file,
2069                                            void __user *argp)
2070 {
2071          struct btrfs_ioctl_search_args *args;
2072          struct inode *inode;
2073          int ret;
2074
2075         if (!capable(CAP_SYS_ADMIN))
2076                 return -EPERM;
2077
2078         args = memdup_user(argp, sizeof(*args));
2079         if (IS_ERR(args))
2080                 return PTR_ERR(args);
2081
2082         inode = file_inode(file);
2083         ret = search_ioctl(inode, args);
2084         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2085                 ret = -EFAULT;
2086         kfree(args);
2087         return ret;
2088 }
2089
2090 /*
2091  * Search INODE_REFs to identify path name of 'dirid' directory
2092  * in a 'tree_id' tree. and sets path name to 'name'.
2093  */
2094 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2095                                 u64 tree_id, u64 dirid, char *name)
2096 {
2097         struct btrfs_root *root;
2098         struct btrfs_key key;
2099         char *ptr;
2100         int ret = -1;
2101         int slot;
2102         int len;
2103         int total_len = 0;
2104         struct btrfs_inode_ref *iref;
2105         struct extent_buffer *l;
2106         struct btrfs_path *path;
2107
2108         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2109                 name[0]='\0';
2110                 return 0;
2111         }
2112
2113         path = btrfs_alloc_path();
2114         if (!path)
2115                 return -ENOMEM;
2116
2117         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2118
2119         key.objectid = tree_id;
2120         key.type = BTRFS_ROOT_ITEM_KEY;
2121         key.offset = (u64)-1;
2122         root = btrfs_read_fs_root_no_name(info, &key);
2123         if (IS_ERR(root)) {
2124                 printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id);
2125                 ret = -ENOENT;
2126                 goto out;
2127         }
2128
2129         key.objectid = dirid;
2130         key.type = BTRFS_INODE_REF_KEY;
2131         key.offset = (u64)-1;
2132
2133         while (1) {
2134                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2135                 if (ret < 0)
2136                         goto out;
2137                 else if (ret > 0) {
2138                         ret = btrfs_previous_item(root, path, dirid,
2139                                                   BTRFS_INODE_REF_KEY);
2140                         if (ret < 0)
2141                                 goto out;
2142                         else if (ret > 0) {
2143                                 ret = -ENOENT;
2144                                 goto out;
2145                         }
2146                 }
2147
2148                 l = path->nodes[0];
2149                 slot = path->slots[0];
2150                 btrfs_item_key_to_cpu(l, &key, slot);
2151
2152                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2153                 len = btrfs_inode_ref_name_len(l, iref);
2154                 ptr -= len + 1;
2155                 total_len += len + 1;
2156                 if (ptr < name) {
2157                         ret = -ENAMETOOLONG;
2158                         goto out;
2159                 }
2160
2161                 *(ptr + len) = '/';
2162                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2163
2164                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2165                         break;
2166
2167                 btrfs_release_path(path);
2168                 key.objectid = key.offset;
2169                 key.offset = (u64)-1;
2170                 dirid = key.objectid;
2171         }
2172         memmove(name, ptr, total_len);
2173         name[total_len] = '\0';
2174         ret = 0;
2175 out:
2176         btrfs_free_path(path);
2177         return ret;
2178 }
2179
2180 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2181                                            void __user *argp)
2182 {
2183          struct btrfs_ioctl_ino_lookup_args *args;
2184          struct inode *inode;
2185          int ret;
2186
2187         if (!capable(CAP_SYS_ADMIN))
2188                 return -EPERM;
2189
2190         args = memdup_user(argp, sizeof(*args));
2191         if (IS_ERR(args))
2192                 return PTR_ERR(args);
2193
2194         inode = file_inode(file);
2195
2196         if (args->treeid == 0)
2197                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2198
2199         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2200                                         args->treeid, args->objectid,
2201                                         args->name);
2202
2203         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2204                 ret = -EFAULT;
2205
2206         kfree(args);
2207         return ret;
2208 }
2209
2210 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2211                                              void __user *arg)
2212 {
2213         struct dentry *parent = file->f_path.dentry;
2214         struct dentry *dentry;
2215         struct inode *dir = parent->d_inode;
2216         struct inode *inode;
2217         struct btrfs_root *root = BTRFS_I(dir)->root;
2218         struct btrfs_root *dest = NULL;
2219         struct btrfs_ioctl_vol_args *vol_args;
2220         struct btrfs_trans_handle *trans;
2221         struct btrfs_block_rsv block_rsv;
2222         u64 qgroup_reserved;
2223         int namelen;
2224         int ret;
2225         int err = 0;
2226
2227         vol_args = memdup_user(arg, sizeof(*vol_args));
2228         if (IS_ERR(vol_args))
2229                 return PTR_ERR(vol_args);
2230
2231         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2232         namelen = strlen(vol_args->name);
2233         if (strchr(vol_args->name, '/') ||
2234             strncmp(vol_args->name, "..", namelen) == 0) {
2235                 err = -EINVAL;
2236                 goto out;
2237         }
2238
2239         err = mnt_want_write_file(file);
2240         if (err)
2241                 goto out;
2242
2243         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2244         if (err == -EINTR)
2245                 goto out_drop_write;
2246         dentry = lookup_one_len(vol_args->name, parent, namelen);
2247         if (IS_ERR(dentry)) {
2248                 err = PTR_ERR(dentry);
2249                 goto out_unlock_dir;
2250         }
2251
2252         if (!dentry->d_inode) {
2253                 err = -ENOENT;
2254                 goto out_dput;
2255         }
2256
2257         inode = dentry->d_inode;
2258         dest = BTRFS_I(inode)->root;
2259         if (!capable(CAP_SYS_ADMIN)) {
2260                 /*
2261                  * Regular user.  Only allow this with a special mount
2262                  * option, when the user has write+exec access to the
2263                  * subvol root, and when rmdir(2) would have been
2264                  * allowed.
2265                  *
2266                  * Note that this is _not_ check that the subvol is
2267                  * empty or doesn't contain data that we wouldn't
2268                  * otherwise be able to delete.
2269                  *
2270                  * Users who want to delete empty subvols should try
2271                  * rmdir(2).
2272                  */
2273                 err = -EPERM;
2274                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2275                         goto out_dput;
2276
2277                 /*
2278                  * Do not allow deletion if the parent dir is the same
2279                  * as the dir to be deleted.  That means the ioctl
2280                  * must be called on the dentry referencing the root
2281                  * of the subvol, not a random directory contained
2282                  * within it.
2283                  */
2284                 err = -EINVAL;
2285                 if (root == dest)
2286                         goto out_dput;
2287
2288                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2289                 if (err)
2290                         goto out_dput;
2291         }
2292
2293         /* check if subvolume may be deleted by a user */
2294         err = btrfs_may_delete(dir, dentry, 1);
2295         if (err)
2296                 goto out_dput;
2297
2298         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2299                 err = -EINVAL;
2300                 goto out_dput;
2301         }
2302
2303         mutex_lock(&inode->i_mutex);
2304         err = d_invalidate(dentry);
2305         if (err)
2306                 goto out_unlock;
2307
2308         down_write(&root->fs_info->subvol_sem);
2309
2310         err = may_destroy_subvol(dest);
2311         if (err)
2312                 goto out_up_write;
2313
2314         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2315         /*
2316          * One for dir inode, two for dir entries, two for root
2317          * ref/backref.
2318          */
2319         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2320                                                5, &qgroup_reserved, true);
2321         if (err)
2322                 goto out_up_write;
2323
2324         trans = btrfs_start_transaction(root, 0);
2325         if (IS_ERR(trans)) {
2326                 err = PTR_ERR(trans);
2327                 goto out_release;
2328         }
2329         trans->block_rsv = &block_rsv;
2330         trans->bytes_reserved = block_rsv.size;
2331
2332         ret = btrfs_unlink_subvol(trans, root, dir,
2333                                 dest->root_key.objectid,
2334                                 dentry->d_name.name,
2335                                 dentry->d_name.len);
2336         if (ret) {
2337                 err = ret;
2338                 btrfs_abort_transaction(trans, root, ret);
2339                 goto out_end_trans;
2340         }
2341
2342         btrfs_record_root_in_trans(trans, dest);
2343
2344         memset(&dest->root_item.drop_progress, 0,
2345                 sizeof(dest->root_item.drop_progress));
2346         dest->root_item.drop_level = 0;
2347         btrfs_set_root_refs(&dest->root_item, 0);
2348
2349         if (!xchg(&dest->orphan_item_inserted, 1)) {
2350                 ret = btrfs_insert_orphan_item(trans,
2351                                         root->fs_info->tree_root,
2352                                         dest->root_key.objectid);
2353                 if (ret) {
2354                         btrfs_abort_transaction(trans, root, ret);
2355                         err = ret;
2356                         goto out_end_trans;
2357                 }
2358         }
2359
2360         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2361                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2362                                   dest->root_key.objectid);
2363         if (ret && ret != -ENOENT) {
2364                 btrfs_abort_transaction(trans, root, ret);
2365                 err = ret;
2366                 goto out_end_trans;
2367         }
2368         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2369                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2370                                           dest->root_item.received_uuid,
2371                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2372                                           dest->root_key.objectid);
2373                 if (ret && ret != -ENOENT) {
2374                         btrfs_abort_transaction(trans, root, ret);
2375                         err = ret;
2376                         goto out_end_trans;
2377                 }
2378         }
2379
2380 out_end_trans:
2381         trans->block_rsv = NULL;
2382         trans->bytes_reserved = 0;
2383         ret = btrfs_end_transaction(trans, root);
2384         if (ret && !err)
2385                 err = ret;
2386         inode->i_flags |= S_DEAD;
2387 out_release:
2388         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2389 out_up_write:
2390         up_write(&root->fs_info->subvol_sem);
2391 out_unlock:
2392         mutex_unlock(&inode->i_mutex);
2393         if (!err) {
2394                 shrink_dcache_sb(root->fs_info->sb);
2395                 btrfs_invalidate_inodes(dest);
2396                 d_delete(dentry);
2397
2398                 /* the last ref */
2399                 if (dest->cache_inode) {
2400                         iput(dest->cache_inode);
2401                         dest->cache_inode = NULL;
2402                 }
2403         }
2404 out_dput:
2405         dput(dentry);
2406 out_unlock_dir:
2407         mutex_unlock(&dir->i_mutex);
2408 out_drop_write:
2409         mnt_drop_write_file(file);
2410 out:
2411         kfree(vol_args);
2412         return err;
2413 }
2414
2415 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2416 {
2417         struct inode *inode = file_inode(file);
2418         struct btrfs_root *root = BTRFS_I(inode)->root;
2419         struct btrfs_ioctl_defrag_range_args *range;
2420         int ret;
2421
2422         ret = mnt_want_write_file(file);
2423         if (ret)
2424                 return ret;
2425
2426         if (btrfs_root_readonly(root)) {
2427                 ret = -EROFS;
2428                 goto out;
2429         }
2430
2431         switch (inode->i_mode & S_IFMT) {
2432         case S_IFDIR:
2433                 if (!capable(CAP_SYS_ADMIN)) {
2434                         ret = -EPERM;
2435                         goto out;
2436                 }
2437                 ret = btrfs_defrag_root(root);
2438                 if (ret)
2439                         goto out;
2440                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2441                 break;
2442         case S_IFREG:
2443                 if (!(file->f_mode & FMODE_WRITE)) {
2444                         ret = -EINVAL;
2445                         goto out;
2446                 }
2447
2448                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2449                 if (!range) {
2450                         ret = -ENOMEM;
2451                         goto out;
2452                 }
2453
2454                 if (argp) {
2455                         if (copy_from_user(range, argp,
2456                                            sizeof(*range))) {
2457                                 ret = -EFAULT;
2458                                 kfree(range);
2459                                 goto out;
2460                         }
2461                         /* compression requires us to start the IO */
2462                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2463                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2464                                 range->extent_thresh = (u32)-1;
2465                         }
2466                 } else {
2467                         /* the rest are all set to zero by kzalloc */
2468                         range->len = (u64)-1;
2469                 }
2470                 ret = btrfs_defrag_file(file_inode(file), file,
2471                                         range, 0, 0);
2472                 if (ret > 0)
2473                         ret = 0;
2474                 kfree(range);
2475                 break;
2476         default:
2477                 ret = -EINVAL;
2478         }
2479 out:
2480         mnt_drop_write_file(file);
2481         return ret;
2482 }
2483
2484 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2485 {
2486         struct btrfs_ioctl_vol_args *vol_args;
2487         int ret;
2488
2489         if (!capable(CAP_SYS_ADMIN))
2490                 return -EPERM;
2491
2492         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2493                         1)) {
2494                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2495         }
2496
2497         mutex_lock(&root->fs_info->volume_mutex);
2498         vol_args = memdup_user(arg, sizeof(*vol_args));
2499         if (IS_ERR(vol_args)) {
2500                 ret = PTR_ERR(vol_args);
2501                 goto out;
2502         }
2503
2504         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2505         ret = btrfs_init_new_device(root, vol_args->name);
2506
2507         kfree(vol_args);
2508 out:
2509         mutex_unlock(&root->fs_info->volume_mutex);
2510         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2511         return ret;
2512 }
2513
2514 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2515 {
2516         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2517         struct btrfs_ioctl_vol_args *vol_args;
2518         int ret;
2519
2520         if (!capable(CAP_SYS_ADMIN))
2521                 return -EPERM;
2522
2523         ret = mnt_want_write_file(file);
2524         if (ret)
2525                 return ret;
2526
2527         vol_args = memdup_user(arg, sizeof(*vol_args));
2528         if (IS_ERR(vol_args)) {
2529                 ret = PTR_ERR(vol_args);
2530                 goto out;
2531         }
2532
2533         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2534
2535         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2536                         1)) {
2537                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2538                 goto out;
2539         }
2540
2541         mutex_lock(&root->fs_info->volume_mutex);
2542         ret = btrfs_rm_device(root, vol_args->name);
2543         mutex_unlock(&root->fs_info->volume_mutex);
2544         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2545
2546 out:
2547         kfree(vol_args);
2548         mnt_drop_write_file(file);
2549         return ret;
2550 }
2551
2552 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2553 {
2554         struct btrfs_ioctl_fs_info_args *fi_args;
2555         struct btrfs_device *device;
2556         struct btrfs_device *next;
2557         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2558         int ret = 0;
2559
2560         if (!capable(CAP_SYS_ADMIN))
2561                 return -EPERM;
2562
2563         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2564         if (!fi_args)
2565                 return -ENOMEM;
2566
2567         mutex_lock(&fs_devices->device_list_mutex);
2568         fi_args->num_devices = fs_devices->num_devices;
2569         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2570
2571         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2572                 if (device->devid > fi_args->max_id)
2573                         fi_args->max_id = device->devid;
2574         }
2575         mutex_unlock(&fs_devices->device_list_mutex);
2576
2577         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2578                 ret = -EFAULT;
2579
2580         kfree(fi_args);
2581         return ret;
2582 }
2583
2584 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2585 {
2586         struct btrfs_ioctl_dev_info_args *di_args;
2587         struct btrfs_device *dev;
2588         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2589         int ret = 0;
2590         char *s_uuid = NULL;
2591
2592         if (!capable(CAP_SYS_ADMIN))
2593                 return -EPERM;
2594
2595         di_args = memdup_user(arg, sizeof(*di_args));
2596         if (IS_ERR(di_args))
2597                 return PTR_ERR(di_args);
2598
2599         if (!btrfs_is_empty_uuid(di_args->uuid))
2600                 s_uuid = di_args->uuid;
2601
2602         mutex_lock(&fs_devices->device_list_mutex);
2603         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2604
2605         if (!dev) {
2606                 ret = -ENODEV;
2607                 goto out;
2608         }
2609
2610         di_args->devid = dev->devid;
2611         di_args->bytes_used = dev->bytes_used;
2612         di_args->total_bytes = dev->total_bytes;
2613         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2614         if (dev->name) {
2615                 struct rcu_string *name;
2616
2617                 rcu_read_lock();
2618                 name = rcu_dereference(dev->name);
2619                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2620                 rcu_read_unlock();
2621                 di_args->path[sizeof(di_args->path) - 1] = 0;
2622         } else {
2623                 di_args->path[0] = '\0';
2624         }
2625
2626 out:
2627         mutex_unlock(&fs_devices->device_list_mutex);
2628         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2629                 ret = -EFAULT;
2630
2631         kfree(di_args);
2632         return ret;
2633 }
2634
2635 static struct page *extent_same_get_page(struct inode *inode, u64 off)
2636 {
2637         struct page *page;
2638         pgoff_t index;
2639         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2640
2641         index = off >> PAGE_CACHE_SHIFT;
2642
2643         page = grab_cache_page(inode->i_mapping, index);
2644         if (!page)
2645                 return NULL;
2646
2647         if (!PageUptodate(page)) {
2648                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2649                                                  0))
2650                         return NULL;
2651                 lock_page(page);
2652                 if (!PageUptodate(page)) {
2653                         unlock_page(page);
2654                         page_cache_release(page);
2655                         return NULL;
2656                 }
2657         }
2658         unlock_page(page);
2659
2660         return page;
2661 }
2662
2663 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2664 {
2665         /* do any pending delalloc/csum calc on src, one way or
2666            another, and lock file content */
2667         while (1) {
2668                 struct btrfs_ordered_extent *ordered;
2669                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2670                 ordered = btrfs_lookup_first_ordered_extent(inode,
2671                                                             off + len - 1);
2672                 if (!ordered &&
2673                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2674                                     off + len - 1, EXTENT_DELALLOC, 0, NULL))
2675                         break;
2676                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2677                 if (ordered)
2678                         btrfs_put_ordered_extent(ordered);
2679                 btrfs_wait_ordered_range(inode, off, len);
2680         }
2681 }
2682
2683 static void btrfs_double_unlock(struct inode *inode1, u64 loff1,
2684                                 struct inode *inode2, u64 loff2, u64 len)
2685 {
2686         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2687         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2688
2689         mutex_unlock(&inode1->i_mutex);
2690         mutex_unlock(&inode2->i_mutex);
2691 }
2692
2693 static void btrfs_double_lock(struct inode *inode1, u64 loff1,
2694                               struct inode *inode2, u64 loff2, u64 len)
2695 {
2696         if (inode1 < inode2) {
2697                 swap(inode1, inode2);
2698                 swap(loff1, loff2);
2699         }
2700
2701         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2702         lock_extent_range(inode1, loff1, len);
2703         if (inode1 != inode2) {
2704                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2705                 lock_extent_range(inode2, loff2, len);
2706         }
2707 }
2708
2709 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2710                           u64 dst_loff, u64 len)
2711 {
2712         int ret = 0;
2713         struct page *src_page, *dst_page;
2714         unsigned int cmp_len = PAGE_CACHE_SIZE;
2715         void *addr, *dst_addr;
2716
2717         while (len) {
2718                 if (len < PAGE_CACHE_SIZE)
2719                         cmp_len = len;
2720
2721                 src_page = extent_same_get_page(src, loff);
2722                 if (!src_page)
2723                         return -EINVAL;
2724                 dst_page = extent_same_get_page(dst, dst_loff);
2725                 if (!dst_page) {
2726                         page_cache_release(src_page);
2727                         return -EINVAL;
2728                 }
2729                 addr = kmap_atomic(src_page);
2730                 dst_addr = kmap_atomic(dst_page);
2731
2732                 flush_dcache_page(src_page);
2733                 flush_dcache_page(dst_page);
2734
2735                 if (memcmp(addr, dst_addr, cmp_len))
2736                         ret = BTRFS_SAME_DATA_DIFFERS;
2737
2738                 kunmap_atomic(addr);
2739                 kunmap_atomic(dst_addr);
2740                 page_cache_release(src_page);
2741                 page_cache_release(dst_page);
2742
2743                 if (ret)
2744                         break;
2745
2746                 loff += cmp_len;
2747                 dst_loff += cmp_len;
2748                 len -= cmp_len;
2749         }
2750
2751         return ret;
2752 }
2753
2754 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len)
2755 {
2756         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2757
2758         if (off + len > inode->i_size || off + len < off)
2759                 return -EINVAL;
2760         /* Check that we are block aligned - btrfs_clone() requires this */
2761         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
2762                 return -EINVAL;
2763
2764         return 0;
2765 }
2766
2767 static int btrfs_extent_same(struct inode *src, u64 loff, u64 len,
2768                              struct inode *dst, u64 dst_loff)
2769 {
2770         int ret;
2771
2772         /*
2773          * btrfs_clone() can't handle extents in the same file
2774          * yet. Once that works, we can drop this check and replace it
2775          * with a check for the same inode, but overlapping extents.
2776          */
2777         if (src == dst)
2778                 return -EINVAL;
2779
2780         btrfs_double_lock(src, loff, dst, dst_loff, len);
2781
2782         ret = extent_same_check_offsets(src, loff, len);
2783         if (ret)
2784                 goto out_unlock;
2785
2786         ret = extent_same_check_offsets(dst, dst_loff, len);
2787         if (ret)
2788                 goto out_unlock;
2789
2790         /* don't make the dst file partly checksummed */
2791         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2792             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
2793                 ret = -EINVAL;
2794                 goto out_unlock;
2795         }
2796
2797         ret = btrfs_cmp_data(src, loff, dst, dst_loff, len);
2798         if (ret == 0)
2799                 ret = btrfs_clone(src, dst, loff, len, len, dst_loff);
2800
2801 out_unlock:
2802         btrfs_double_unlock(src, loff, dst, dst_loff, len);
2803
2804         return ret;
2805 }
2806
2807 #define BTRFS_MAX_DEDUPE_LEN    (16 * 1024 * 1024)
2808
2809 static long btrfs_ioctl_file_extent_same(struct file *file,
2810                         struct btrfs_ioctl_same_args __user *argp)
2811 {
2812         struct btrfs_ioctl_same_args *same;
2813         struct btrfs_ioctl_same_extent_info *info;
2814         struct inode *src = file_inode(file);
2815         u64 off;
2816         u64 len;
2817         int i;
2818         int ret;
2819         unsigned long size;
2820         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
2821         bool is_admin = capable(CAP_SYS_ADMIN);
2822         u16 count;
2823
2824         if (!(file->f_mode & FMODE_READ))
2825                 return -EINVAL;
2826
2827         ret = mnt_want_write_file(file);
2828         if (ret)
2829                 return ret;
2830
2831         if (get_user(count, &argp->dest_count)) {
2832                 ret = -EFAULT;
2833                 goto out;
2834         }
2835
2836         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
2837
2838         same = memdup_user(argp, size);
2839
2840         if (IS_ERR(same)) {
2841                 ret = PTR_ERR(same);
2842                 goto out;
2843         }
2844
2845         off = same->logical_offset;
2846         len = same->length;
2847
2848         /*
2849          * Limit the total length we will dedupe for each operation.
2850          * This is intended to bound the total time spent in this
2851          * ioctl to something sane.
2852          */
2853         if (len > BTRFS_MAX_DEDUPE_LEN)
2854                 len = BTRFS_MAX_DEDUPE_LEN;
2855
2856         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
2857                 /*
2858                  * Btrfs does not support blocksize < page_size. As a
2859                  * result, btrfs_cmp_data() won't correctly handle
2860                  * this situation without an update.
2861                  */
2862                 ret = -EINVAL;
2863                 goto out;
2864         }
2865
2866         ret = -EISDIR;
2867         if (S_ISDIR(src->i_mode))
2868                 goto out;
2869
2870         ret = -EACCES;
2871         if (!S_ISREG(src->i_mode))
2872                 goto out;
2873
2874         /* pre-format output fields to sane values */
2875         for (i = 0; i < count; i++) {
2876                 same->info[i].bytes_deduped = 0ULL;
2877                 same->info[i].status = 0;
2878         }
2879
2880         for (i = 0, info = same->info; i < count; i++, info++) {
2881                 struct inode *dst;
2882                 struct fd dst_file = fdget(info->fd);
2883                 if (!dst_file.file) {
2884                         info->status = -EBADF;
2885                         continue;
2886                 }
2887                 dst = file_inode(dst_file.file);
2888
2889                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
2890                         info->status = -EINVAL;
2891                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
2892                         info->status = -EXDEV;
2893                 } else if (S_ISDIR(dst->i_mode)) {
2894                         info->status = -EISDIR;
2895                 } else if (!S_ISREG(dst->i_mode)) {
2896                         info->status = -EACCES;
2897                 } else {
2898                         info->status = btrfs_extent_same(src, off, len, dst,
2899                                                         info->logical_offset);
2900                         if (info->status == 0)
2901                                 info->bytes_deduped += len;
2902                 }
2903                 fdput(dst_file);
2904         }
2905
2906         ret = copy_to_user(argp, same, size);
2907         if (ret)
2908                 ret = -EFAULT;
2909
2910 out:
2911         mnt_drop_write_file(file);
2912         return ret;
2913 }
2914
2915 /**
2916  * btrfs_clone() - clone a range from inode file to another
2917  *
2918  * @src: Inode to clone from
2919  * @inode: Inode to clone to
2920  * @off: Offset within source to start clone from
2921  * @olen: Original length, passed by user, of range to clone
2922  * @olen_aligned: Block-aligned value of olen, extent_same uses
2923  *               identical values here
2924  * @destoff: Offset within @inode to start clone
2925  */
2926 static int btrfs_clone(struct inode *src, struct inode *inode,
2927                        u64 off, u64 olen, u64 olen_aligned, u64 destoff)
2928 {
2929         struct btrfs_root *root = BTRFS_I(inode)->root;
2930         struct btrfs_path *path = NULL;
2931         struct extent_buffer *leaf;
2932         struct btrfs_trans_handle *trans;
2933         char *buf = NULL;
2934         struct btrfs_key key;
2935         u32 nritems;
2936         int slot;
2937         int ret;
2938         u64 len = olen_aligned;
2939
2940         ret = -ENOMEM;
2941         buf = vmalloc(btrfs_level_size(root, 0));
2942         if (!buf)
2943                 return ret;
2944
2945         path = btrfs_alloc_path();
2946         if (!path) {
2947                 vfree(buf);
2948                 return ret;
2949         }
2950
2951         path->reada = 2;
2952         /* clone data */
2953         key.objectid = btrfs_ino(src);
2954         key.type = BTRFS_EXTENT_DATA_KEY;
2955         key.offset = 0;
2956
2957         while (1) {
2958                 /*
2959                  * note the key will change type as we walk through the
2960                  * tree.
2961                  */
2962                 path->leave_spinning = 1;
2963                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2964                                 0, 0);
2965                 if (ret < 0)
2966                         goto out;
2967
2968                 nritems = btrfs_header_nritems(path->nodes[0]);
2969 process_slot:
2970                 if (path->slots[0] >= nritems) {
2971                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2972                         if (ret < 0)
2973                                 goto out;
2974                         if (ret > 0)
2975                                 break;
2976                         nritems = btrfs_header_nritems(path->nodes[0]);
2977                 }
2978                 leaf = path->nodes[0];
2979                 slot = path->slots[0];
2980
2981                 btrfs_item_key_to_cpu(leaf, &key, slot);
2982                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2983                     key.objectid != btrfs_ino(src))
2984                         break;
2985
2986                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2987                         struct btrfs_file_extent_item *extent;
2988                         int type;
2989                         u32 size;
2990                         struct btrfs_key new_key;
2991                         u64 disko = 0, diskl = 0;
2992                         u64 datao = 0, datal = 0;
2993                         u8 comp;
2994                         u64 endoff;
2995
2996                         extent = btrfs_item_ptr(leaf, slot,
2997                                                 struct btrfs_file_extent_item);
2998                         comp = btrfs_file_extent_compression(leaf, extent);
2999                         type = btrfs_file_extent_type(leaf, extent);
3000                         if (type == BTRFS_FILE_EXTENT_REG ||
3001                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3002                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3003                                                                       extent);
3004                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3005                                                                  extent);
3006                                 datao = btrfs_file_extent_offset(leaf, extent);
3007                                 datal = btrfs_file_extent_num_bytes(leaf,
3008                                                                     extent);
3009                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3010                                 /* take upper bound, may be compressed */
3011                                 datal = btrfs_file_extent_ram_bytes(leaf,
3012                                                                     extent);
3013                         }
3014
3015                         if (key.offset + datal <= off ||
3016                             key.offset >= off + len - 1) {
3017                                 path->slots[0]++;
3018                                 goto process_slot;
3019                         }
3020
3021                         size = btrfs_item_size_nr(leaf, slot);
3022                         read_extent_buffer(leaf, buf,
3023                                            btrfs_item_ptr_offset(leaf, slot),
3024                                            size);
3025
3026                         btrfs_release_path(path);
3027                         path->leave_spinning = 0;
3028
3029                         memcpy(&new_key, &key, sizeof(new_key));
3030                         new_key.objectid = btrfs_ino(inode);
3031                         if (off <= key.offset)
3032                                 new_key.offset = key.offset + destoff - off;
3033                         else
3034                                 new_key.offset = destoff;
3035
3036                         /*
3037                          * 1 - adjusting old extent (we may have to split it)
3038                          * 1 - add new extent
3039                          * 1 - inode update
3040                          */
3041                         trans = btrfs_start_transaction(root, 3);
3042                         if (IS_ERR(trans)) {
3043                                 ret = PTR_ERR(trans);
3044                                 goto out;
3045                         }
3046
3047                         if (type == BTRFS_FILE_EXTENT_REG ||
3048                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3049                                 /*
3050                                  *    a  | --- range to clone ---|  b
3051                                  * | ------------- extent ------------- |
3052                                  */
3053
3054                                 /* substract range b */
3055                                 if (key.offset + datal > off + len)
3056                                         datal = off + len - key.offset;
3057
3058                                 /* substract range a */
3059                                 if (off > key.offset) {
3060                                         datao += off - key.offset;
3061                                         datal -= off - key.offset;
3062                                 }
3063
3064                                 ret = btrfs_drop_extents(trans, root, inode,
3065                                                          new_key.offset,
3066                                                          new_key.offset + datal,
3067                                                          1);
3068                                 if (ret) {
3069                                         if (ret != -EOPNOTSUPP)
3070                                                 btrfs_abort_transaction(trans,
3071                                                                 root, ret);
3072                                         btrfs_end_transaction(trans, root);
3073                                         goto out;
3074                                 }
3075
3076                                 ret = btrfs_insert_empty_item(trans, root, path,
3077                                                               &new_key, size);
3078                                 if (ret) {
3079                                         btrfs_abort_transaction(trans, root,
3080                                                                 ret);
3081                                         btrfs_end_transaction(trans, root);
3082                                         goto out;
3083                                 }
3084
3085                                 leaf = path->nodes[0];
3086                                 slot = path->slots[0];
3087                                 write_extent_buffer(leaf, buf,
3088                                             btrfs_item_ptr_offset(leaf, slot),
3089                                             size);
3090
3091                                 extent = btrfs_item_ptr(leaf, slot,
3092                                                 struct btrfs_file_extent_item);
3093
3094                                 /* disko == 0 means it's a hole */
3095                                 if (!disko)
3096                                         datao = 0;
3097
3098                                 btrfs_set_file_extent_offset(leaf, extent,
3099                                                              datao);
3100                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3101                                                                 datal);
3102                                 if (disko) {
3103                                         inode_add_bytes(inode, datal);
3104                                         ret = btrfs_inc_extent_ref(trans, root,
3105                                                         disko, diskl, 0,
3106                                                         root->root_key.objectid,
3107                                                         btrfs_ino(inode),
3108                                                         new_key.offset - datao,
3109                                                         0);
3110                                         if (ret) {
3111                                                 btrfs_abort_transaction(trans,
3112                                                                         root,
3113                                                                         ret);
3114                                                 btrfs_end_transaction(trans,
3115                                                                       root);
3116                                                 goto out;
3117
3118                                         }
3119                                 }
3120                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3121                                 u64 skip = 0;
3122                                 u64 trim = 0;
3123                                 u64 aligned_end = 0;
3124
3125                                 if (off > key.offset) {
3126                                         skip = off - key.offset;
3127                                         new_key.offset += skip;
3128                                 }
3129
3130                                 if (key.offset + datal > off + len)
3131                                         trim = key.offset + datal - (off + len);
3132
3133                                 if (comp && (skip || trim)) {
3134                                         ret = -EINVAL;
3135                                         btrfs_end_transaction(trans, root);
3136                                         goto out;
3137                                 }
3138                                 size -= skip + trim;
3139                                 datal -= skip + trim;
3140
3141                                 aligned_end = ALIGN(new_key.offset + datal,
3142                                                     root->sectorsize);
3143                                 ret = btrfs_drop_extents(trans, root, inode,
3144                                                          new_key.offset,
3145                                                          aligned_end,
3146                                                          1);
3147                                 if (ret) {
3148                                         if (ret != -EOPNOTSUPP)
3149                                                 btrfs_abort_transaction(trans,
3150                                                         root, 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          *
3235          * - split destination inode's inline extents.  The inline extents can
3236          *   be either compressed or non-compressed.
3237          */
3238
3239         /* the destination must be opened for writing */
3240         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3241                 return -EINVAL;
3242
3243         if (btrfs_root_readonly(root))
3244                 return -EROFS;
3245
3246         ret = mnt_want_write_file(file);
3247         if (ret)
3248                 return ret;
3249
3250         src_file = fdget(srcfd);
3251         if (!src_file.file) {
3252                 ret = -EBADF;
3253                 goto out_drop_write;
3254         }
3255
3256         ret = -EXDEV;
3257         if (src_file.file->f_path.mnt != file->f_path.mnt)
3258                 goto out_fput;
3259
3260         src = file_inode(src_file.file);
3261
3262         ret = -EINVAL;
3263         if (src == inode)
3264                 same_inode = 1;
3265
3266         /* the src must be open for reading */
3267         if (!(src_file.file->f_mode & FMODE_READ))
3268                 goto out_fput;
3269
3270         /* don't make the dst file partly checksummed */
3271         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3272             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3273                 goto out_fput;
3274
3275         ret = -EISDIR;
3276         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3277                 goto out_fput;
3278
3279         ret = -EXDEV;
3280         if (src->i_sb != inode->i_sb)
3281                 goto out_fput;
3282
3283         if (!same_inode) {
3284                 if (inode < src) {
3285                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
3286                         mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
3287                 } else {
3288                         mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
3289                         mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
3290                 }
3291         } else {
3292                 mutex_lock(&src->i_mutex);
3293         }
3294
3295         /* determine range to clone */
3296         ret = -EINVAL;
3297         if (off + len > src->i_size || off + len < off)
3298                 goto out_unlock;
3299         if (len == 0)
3300                 olen = len = src->i_size - off;
3301         /* if we extend to eof, continue to block boundary */
3302         if (off + len == src->i_size)
3303                 len = ALIGN(src->i_size, bs) - off;
3304
3305         /* verify the end result is block aligned */
3306         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3307             !IS_ALIGNED(destoff, bs))
3308                 goto out_unlock;
3309
3310         /* verify if ranges are overlapped within the same file */
3311         if (same_inode) {
3312                 if (destoff + len > off && destoff < off + len)
3313                         goto out_unlock;
3314         }
3315
3316         if (destoff > inode->i_size) {
3317                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3318                 if (ret)
3319                         goto out_unlock;
3320         }
3321
3322         /* truncate page cache pages from target inode range */
3323         truncate_inode_pages_range(&inode->i_data, destoff,
3324                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3325
3326         lock_extent_range(src, off, len);
3327
3328         ret = btrfs_clone(src, inode, off, olen, len, destoff);
3329
3330         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
3331 out_unlock:
3332         if (!same_inode) {
3333                 if (inode < src) {
3334                         mutex_unlock(&src->i_mutex);
3335                         mutex_unlock(&inode->i_mutex);
3336                 } else {
3337                         mutex_unlock(&inode->i_mutex);
3338                         mutex_unlock(&src->i_mutex);
3339                 }
3340         } else {
3341                 mutex_unlock(&src->i_mutex);
3342         }
3343 out_fput:
3344         fdput(src_file);
3345 out_drop_write:
3346         mnt_drop_write_file(file);
3347         return ret;
3348 }
3349
3350 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3351 {
3352         struct btrfs_ioctl_clone_range_args args;
3353
3354         if (copy_from_user(&args, argp, sizeof(args)))
3355                 return -EFAULT;
3356         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3357                                  args.src_length, args.dest_offset);
3358 }
3359
3360 /*
3361  * there are many ways the trans_start and trans_end ioctls can lead
3362  * to deadlocks.  They should only be used by applications that
3363  * basically own the machine, and have a very in depth understanding
3364  * of all the possible deadlocks and enospc problems.
3365  */
3366 static long btrfs_ioctl_trans_start(struct file *file)
3367 {
3368         struct inode *inode = file_inode(file);
3369         struct btrfs_root *root = BTRFS_I(inode)->root;
3370         struct btrfs_trans_handle *trans;
3371         int ret;
3372
3373         ret = -EPERM;
3374         if (!capable(CAP_SYS_ADMIN))
3375                 goto out;
3376
3377         ret = -EINPROGRESS;
3378         if (file->private_data)
3379                 goto out;
3380
3381         ret = -EROFS;
3382         if (btrfs_root_readonly(root))
3383                 goto out;
3384
3385         ret = mnt_want_write_file(file);
3386         if (ret)
3387                 goto out;
3388
3389         atomic_inc(&root->fs_info->open_ioctl_trans);
3390
3391         ret = -ENOMEM;
3392         trans = btrfs_start_ioctl_transaction(root);
3393         if (IS_ERR(trans))
3394                 goto out_drop;
3395
3396         file->private_data = trans;
3397         return 0;
3398
3399 out_drop:
3400         atomic_dec(&root->fs_info->open_ioctl_trans);
3401         mnt_drop_write_file(file);
3402 out:
3403         return ret;
3404 }
3405
3406 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3407 {
3408         struct inode *inode = file_inode(file);
3409         struct btrfs_root *root = BTRFS_I(inode)->root;
3410         struct btrfs_root *new_root;
3411         struct btrfs_dir_item *di;
3412         struct btrfs_trans_handle *trans;
3413         struct btrfs_path *path;
3414         struct btrfs_key location;
3415         struct btrfs_disk_key disk_key;
3416         u64 objectid = 0;
3417         u64 dir_id;
3418         int ret;
3419
3420         if (!capable(CAP_SYS_ADMIN))
3421                 return -EPERM;
3422
3423         ret = mnt_want_write_file(file);
3424         if (ret)
3425                 return ret;
3426
3427         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3428                 ret = -EFAULT;
3429                 goto out;
3430         }
3431
3432         if (!objectid)
3433                 objectid = BTRFS_FS_TREE_OBJECTID;
3434
3435         location.objectid = objectid;
3436         location.type = BTRFS_ROOT_ITEM_KEY;
3437         location.offset = (u64)-1;
3438
3439         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
3440         if (IS_ERR(new_root)) {
3441                 ret = PTR_ERR(new_root);
3442                 goto out;
3443         }
3444
3445         path = btrfs_alloc_path();
3446         if (!path) {
3447                 ret = -ENOMEM;
3448                 goto out;
3449         }
3450         path->leave_spinning = 1;
3451
3452         trans = btrfs_start_transaction(root, 1);
3453         if (IS_ERR(trans)) {
3454                 btrfs_free_path(path);
3455                 ret = PTR_ERR(trans);
3456                 goto out;
3457         }
3458
3459         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
3460         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
3461                                    dir_id, "default", 7, 1);
3462         if (IS_ERR_OR_NULL(di)) {
3463                 btrfs_free_path(path);
3464                 btrfs_end_transaction(trans, root);
3465                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
3466                            "item, this isn't going to work");
3467                 ret = -ENOENT;
3468                 goto out;
3469         }
3470
3471         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3472         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3473         btrfs_mark_buffer_dirty(path->nodes[0]);
3474         btrfs_free_path(path);
3475
3476         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
3477         btrfs_end_transaction(trans, root);
3478 out:
3479         mnt_drop_write_file(file);
3480         return ret;
3481 }
3482
3483 void btrfs_get_block_group_info(struct list_head *groups_list,
3484                                 struct btrfs_ioctl_space_info *space)
3485 {
3486         struct btrfs_block_group_cache *block_group;
3487
3488         space->total_bytes = 0;
3489         space->used_bytes = 0;
3490         space->flags = 0;
3491         list_for_each_entry(block_group, groups_list, list) {
3492                 space->flags = block_group->flags;
3493                 space->total_bytes += block_group->key.offset;
3494                 space->used_bytes +=
3495                         btrfs_block_group_used(&block_group->item);
3496         }
3497 }
3498
3499 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
3500 {
3501         struct btrfs_ioctl_space_args space_args;
3502         struct btrfs_ioctl_space_info space;
3503         struct btrfs_ioctl_space_info *dest;
3504         struct btrfs_ioctl_space_info *dest_orig;
3505         struct btrfs_ioctl_space_info __user *user_dest;
3506         struct btrfs_space_info *info;
3507         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3508                        BTRFS_BLOCK_GROUP_SYSTEM,
3509                        BTRFS_BLOCK_GROUP_METADATA,
3510                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3511         int num_types = 4;
3512         int alloc_size;
3513         int ret = 0;
3514         u64 slot_count = 0;
3515         int i, c;
3516
3517         if (copy_from_user(&space_args,
3518                            (struct btrfs_ioctl_space_args __user *)arg,
3519                            sizeof(space_args)))
3520                 return -EFAULT;
3521
3522         for (i = 0; i < num_types; i++) {
3523                 struct btrfs_space_info *tmp;
3524
3525                 info = NULL;
3526                 rcu_read_lock();
3527                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3528                                         list) {
3529                         if (tmp->flags == types[i]) {
3530                                 info = tmp;
3531                                 break;
3532                         }
3533                 }
3534                 rcu_read_unlock();
3535
3536                 if (!info)
3537                         continue;
3538
3539                 down_read(&info->groups_sem);
3540                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3541                         if (!list_empty(&info->block_groups[c]))
3542                                 slot_count++;
3543                 }
3544                 up_read(&info->groups_sem);
3545         }
3546
3547         /*
3548          * Global block reserve, exported as a space_info
3549          */
3550         slot_count++;
3551
3552         /* space_slots == 0 means they are asking for a count */
3553         if (space_args.space_slots == 0) {
3554                 space_args.total_spaces = slot_count;
3555                 goto out;
3556         }
3557
3558         slot_count = min_t(u64, space_args.space_slots, slot_count);
3559
3560         alloc_size = sizeof(*dest) * slot_count;
3561
3562         /* we generally have at most 6 or so space infos, one for each raid
3563          * level.  So, a whole page should be more than enough for everyone
3564          */
3565         if (alloc_size > PAGE_CACHE_SIZE)
3566                 return -ENOMEM;
3567
3568         space_args.total_spaces = 0;
3569         dest = kmalloc(alloc_size, GFP_NOFS);
3570         if (!dest)
3571                 return -ENOMEM;
3572         dest_orig = dest;
3573
3574         /* now we have a buffer to copy into */
3575         for (i = 0; i < num_types; i++) {
3576                 struct btrfs_space_info *tmp;
3577
3578                 if (!slot_count)
3579                         break;
3580
3581                 info = NULL;
3582                 rcu_read_lock();
3583                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3584                                         list) {
3585                         if (tmp->flags == types[i]) {
3586                                 info = tmp;
3587                                 break;
3588                         }
3589                 }
3590                 rcu_read_unlock();
3591
3592                 if (!info)
3593                         continue;
3594                 down_read(&info->groups_sem);
3595                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3596                         if (!list_empty(&info->block_groups[c])) {
3597                                 btrfs_get_block_group_info(
3598                                         &info->block_groups[c], &space);
3599                                 memcpy(dest, &space, sizeof(space));
3600                                 dest++;
3601                                 space_args.total_spaces++;
3602                                 slot_count--;
3603                         }
3604                         if (!slot_count)
3605                                 break;
3606                 }
3607                 up_read(&info->groups_sem);
3608         }
3609
3610         /*
3611          * Add global block reserve
3612          */
3613         if (slot_count) {
3614                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
3615
3616                 spin_lock(&block_rsv->lock);
3617                 space.total_bytes = block_rsv->size;
3618                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3619                 spin_unlock(&block_rsv->lock);
3620                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3621                 memcpy(dest, &space, sizeof(space));
3622                 space_args.total_spaces++;
3623         }
3624
3625         user_dest = (struct btrfs_ioctl_space_info __user *)
3626                 (arg + sizeof(struct btrfs_ioctl_space_args));
3627
3628         if (copy_to_user(user_dest, dest_orig, alloc_size))
3629                 ret = -EFAULT;
3630
3631         kfree(dest_orig);
3632 out:
3633         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3634                 ret = -EFAULT;
3635
3636         return ret;
3637 }
3638
3639 /*
3640  * there are many ways the trans_start and trans_end ioctls can lead
3641  * to deadlocks.  They should only be used by applications that
3642  * basically own the machine, and have a very in depth understanding
3643  * of all the possible deadlocks and enospc problems.
3644  */
3645 long btrfs_ioctl_trans_end(struct file *file)
3646 {
3647         struct inode *inode = file_inode(file);
3648         struct btrfs_root *root = BTRFS_I(inode)->root;
3649         struct btrfs_trans_handle *trans;
3650
3651         trans = file->private_data;
3652         if (!trans)
3653                 return -EINVAL;
3654         file->private_data = NULL;
3655
3656         btrfs_end_transaction(trans, root);
3657
3658         atomic_dec(&root->fs_info->open_ioctl_trans);
3659
3660         mnt_drop_write_file(file);
3661         return 0;
3662 }
3663
3664 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3665                                             void __user *argp)
3666 {
3667         struct btrfs_trans_handle *trans;
3668         u64 transid;
3669         int ret;
3670
3671         trans = btrfs_attach_transaction_barrier(root);
3672         if (IS_ERR(trans)) {
3673                 if (PTR_ERR(trans) != -ENOENT)
3674                         return PTR_ERR(trans);
3675
3676                 /* No running transaction, don't bother */
3677                 transid = root->fs_info->last_trans_committed;
3678                 goto out;
3679         }
3680         transid = trans->transid;
3681         ret = btrfs_commit_transaction_async(trans, root, 0);
3682         if (ret) {
3683                 btrfs_end_transaction(trans, root);
3684                 return ret;
3685         }
3686 out:
3687         if (argp)
3688                 if (copy_to_user(argp, &transid, sizeof(transid)))
3689                         return -EFAULT;
3690         return 0;
3691 }
3692
3693 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3694                                            void __user *argp)
3695 {
3696         u64 transid;
3697
3698         if (argp) {
3699                 if (copy_from_user(&transid, argp, sizeof(transid)))
3700                         return -EFAULT;
3701         } else {
3702                 transid = 0;  /* current trans */
3703         }
3704         return btrfs_wait_for_commit(root, transid);
3705 }
3706
3707 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3708 {
3709         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3710         struct btrfs_ioctl_scrub_args *sa;
3711         int ret;
3712
3713         if (!capable(CAP_SYS_ADMIN))
3714                 return -EPERM;
3715
3716         sa = memdup_user(arg, sizeof(*sa));
3717         if (IS_ERR(sa))
3718                 return PTR_ERR(sa);
3719
3720         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3721                 ret = mnt_want_write_file(file);
3722                 if (ret)
3723                         goto out;
3724         }
3725
3726         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3727                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3728                               0);
3729
3730         if (copy_to_user(arg, sa, sizeof(*sa)))
3731                 ret = -EFAULT;
3732
3733         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3734                 mnt_drop_write_file(file);
3735 out:
3736         kfree(sa);
3737         return ret;
3738 }
3739
3740 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3741 {
3742         if (!capable(CAP_SYS_ADMIN))
3743                 return -EPERM;
3744
3745         return btrfs_scrub_cancel(root->fs_info);
3746 }
3747
3748 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3749                                        void __user *arg)
3750 {
3751         struct btrfs_ioctl_scrub_args *sa;
3752         int ret;
3753
3754         if (!capable(CAP_SYS_ADMIN))
3755                 return -EPERM;
3756
3757         sa = memdup_user(arg, sizeof(*sa));
3758         if (IS_ERR(sa))
3759                 return PTR_ERR(sa);
3760
3761         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3762
3763         if (copy_to_user(arg, sa, sizeof(*sa)))
3764                 ret = -EFAULT;
3765
3766         kfree(sa);
3767         return ret;
3768 }
3769
3770 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3771                                       void __user *arg)
3772 {
3773         struct btrfs_ioctl_get_dev_stats *sa;
3774         int ret;
3775
3776         sa = memdup_user(arg, sizeof(*sa));
3777         if (IS_ERR(sa))
3778                 return PTR_ERR(sa);
3779
3780         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3781                 kfree(sa);
3782                 return -EPERM;
3783         }
3784
3785         ret = btrfs_get_dev_stats(root, sa);
3786
3787         if (copy_to_user(arg, sa, sizeof(*sa)))
3788                 ret = -EFAULT;
3789
3790         kfree(sa);
3791         return ret;
3792 }
3793
3794 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3795 {
3796         struct btrfs_ioctl_dev_replace_args *p;
3797         int ret;
3798
3799         if (!capable(CAP_SYS_ADMIN))
3800                 return -EPERM;
3801
3802         p = memdup_user(arg, sizeof(*p));
3803         if (IS_ERR(p))
3804                 return PTR_ERR(p);
3805
3806         switch (p->cmd) {
3807         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3808                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
3809                         ret = -EROFS;
3810                         goto out;
3811                 }
3812                 if (atomic_xchg(
3813                         &root->fs_info->mutually_exclusive_operation_running,
3814                         1)) {
3815                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3816                 } else {
3817                         ret = btrfs_dev_replace_start(root, p);
3818                         atomic_set(
3819                          &root->fs_info->mutually_exclusive_operation_running,
3820                          0);
3821                 }
3822                 break;
3823         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3824                 btrfs_dev_replace_status(root->fs_info, p);
3825                 ret = 0;
3826                 break;
3827         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3828                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3829                 break;
3830         default:
3831                 ret = -EINVAL;
3832                 break;
3833         }
3834
3835         if (copy_to_user(arg, p, sizeof(*p)))
3836                 ret = -EFAULT;
3837 out:
3838         kfree(p);
3839         return ret;
3840 }
3841
3842 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3843 {
3844         int ret = 0;
3845         int i;
3846         u64 rel_ptr;
3847         int size;
3848         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3849         struct inode_fs_paths *ipath = NULL;
3850         struct btrfs_path *path;
3851
3852         if (!capable(CAP_DAC_READ_SEARCH))
3853                 return -EPERM;
3854
3855         path = btrfs_alloc_path();
3856         if (!path) {
3857                 ret = -ENOMEM;
3858                 goto out;
3859         }
3860
3861         ipa = memdup_user(arg, sizeof(*ipa));
3862         if (IS_ERR(ipa)) {
3863                 ret = PTR_ERR(ipa);
3864                 ipa = NULL;
3865                 goto out;
3866         }
3867
3868         size = min_t(u32, ipa->size, 4096);
3869         ipath = init_ipath(size, root, path);
3870         if (IS_ERR(ipath)) {
3871                 ret = PTR_ERR(ipath);
3872                 ipath = NULL;
3873                 goto out;
3874         }
3875
3876         ret = paths_from_inode(ipa->inum, ipath);
3877         if (ret < 0)
3878                 goto out;
3879
3880         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3881                 rel_ptr = ipath->fspath->val[i] -
3882                           (u64)(unsigned long)ipath->fspath->val;
3883                 ipath->fspath->val[i] = rel_ptr;
3884         }
3885
3886         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3887                            (void *)(unsigned long)ipath->fspath, size);
3888         if (ret) {
3889                 ret = -EFAULT;
3890                 goto out;
3891         }
3892
3893 out:
3894         btrfs_free_path(path);
3895         free_ipath(ipath);
3896         kfree(ipa);
3897
3898         return ret;
3899 }
3900
3901 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3902 {
3903         struct btrfs_data_container *inodes = ctx;
3904         const size_t c = 3 * sizeof(u64);
3905
3906         if (inodes->bytes_left >= c) {
3907                 inodes->bytes_left -= c;
3908                 inodes->val[inodes->elem_cnt] = inum;
3909                 inodes->val[inodes->elem_cnt + 1] = offset;
3910                 inodes->val[inodes->elem_cnt + 2] = root;
3911                 inodes->elem_cnt += 3;
3912         } else {
3913                 inodes->bytes_missing += c - inodes->bytes_left;
3914                 inodes->bytes_left = 0;
3915                 inodes->elem_missed += 3;
3916         }
3917
3918         return 0;
3919 }
3920
3921 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3922                                         void __user *arg)
3923 {
3924         int ret = 0;
3925         int size;
3926         struct btrfs_ioctl_logical_ino_args *loi;
3927         struct btrfs_data_container *inodes = NULL;
3928         struct btrfs_path *path = NULL;
3929
3930         if (!capable(CAP_SYS_ADMIN))
3931                 return -EPERM;
3932
3933         loi = memdup_user(arg, sizeof(*loi));
3934         if (IS_ERR(loi)) {
3935                 ret = PTR_ERR(loi);
3936                 loi = NULL;
3937                 goto out;
3938         }
3939
3940         path = btrfs_alloc_path();
3941         if (!path) {
3942                 ret = -ENOMEM;
3943                 goto out;
3944         }
3945
3946         size = min_t(u32, loi->size, 64 * 1024);
3947         inodes = init_data_container(size);
3948         if (IS_ERR(inodes)) {
3949                 ret = PTR_ERR(inodes);
3950                 inodes = NULL;
3951                 goto out;
3952         }
3953
3954         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3955                                           build_ino_list, inodes);
3956         if (ret == -EINVAL)
3957                 ret = -ENOENT;
3958         if (ret < 0)
3959                 goto out;
3960
3961         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3962                            (void *)(unsigned long)inodes, size);
3963         if (ret)
3964                 ret = -EFAULT;
3965
3966 out:
3967         btrfs_free_path(path);
3968         vfree(inodes);
3969         kfree(loi);
3970
3971         return ret;
3972 }
3973
3974 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3975                                struct btrfs_ioctl_balance_args *bargs)
3976 {
3977         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3978
3979         bargs->flags = bctl->flags;
3980
3981         if (atomic_read(&fs_info->balance_running))
3982                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3983         if (atomic_read(&fs_info->balance_pause_req))
3984                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3985         if (atomic_read(&fs_info->balance_cancel_req))
3986                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3987
3988         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3989         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3990         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3991
3992         if (lock) {
3993                 spin_lock(&fs_info->balance_lock);
3994                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3995                 spin_unlock(&fs_info->balance_lock);
3996         } else {
3997                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3998         }
3999 }
4000
4001 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4002 {
4003         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4004         struct btrfs_fs_info *fs_info = root->fs_info;
4005         struct btrfs_ioctl_balance_args *bargs;
4006         struct btrfs_balance_control *bctl;
4007         bool need_unlock; /* for mut. excl. ops lock */
4008         int ret;
4009
4010         if (!capable(CAP_SYS_ADMIN))
4011                 return -EPERM;
4012
4013         ret = mnt_want_write_file(file);
4014         if (ret)
4015                 return ret;
4016
4017 again:
4018         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4019                 mutex_lock(&fs_info->volume_mutex);
4020                 mutex_lock(&fs_info->balance_mutex);
4021                 need_unlock = true;
4022                 goto locked;
4023         }
4024
4025         /*
4026          * mut. excl. ops lock is locked.  Three possibilites:
4027          *   (1) some other op is running
4028          *   (2) balance is running
4029          *   (3) balance is paused -- special case (think resume)
4030          */
4031         mutex_lock(&fs_info->balance_mutex);
4032         if (fs_info->balance_ctl) {
4033                 /* this is either (2) or (3) */
4034                 if (!atomic_read(&fs_info->balance_running)) {
4035                         mutex_unlock(&fs_info->balance_mutex);
4036                         if (!mutex_trylock(&fs_info->volume_mutex))
4037                                 goto again;
4038                         mutex_lock(&fs_info->balance_mutex);
4039
4040                         if (fs_info->balance_ctl &&
4041                             !atomic_read(&fs_info->balance_running)) {
4042                                 /* this is (3) */
4043                                 need_unlock = false;
4044                                 goto locked;
4045                         }
4046
4047                         mutex_unlock(&fs_info->balance_mutex);
4048                         mutex_unlock(&fs_info->volume_mutex);
4049                         goto again;
4050                 } else {
4051                         /* this is (2) */
4052                         mutex_unlock(&fs_info->balance_mutex);
4053                         ret = -EINPROGRESS;
4054                         goto out;
4055                 }
4056         } else {
4057                 /* this is (1) */
4058                 mutex_unlock(&fs_info->balance_mutex);
4059                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4060                 goto out;
4061         }
4062
4063 locked:
4064         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4065
4066         if (arg) {
4067                 bargs = memdup_user(arg, sizeof(*bargs));
4068                 if (IS_ERR(bargs)) {
4069                         ret = PTR_ERR(bargs);
4070                         goto out_unlock;
4071                 }
4072
4073                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4074                         if (!fs_info->balance_ctl) {
4075                                 ret = -ENOTCONN;
4076                                 goto out_bargs;
4077                         }
4078
4079                         bctl = fs_info->balance_ctl;
4080                         spin_lock(&fs_info->balance_lock);
4081                         bctl->flags |= BTRFS_BALANCE_RESUME;
4082                         spin_unlock(&fs_info->balance_lock);
4083
4084                         goto do_balance;
4085                 }
4086         } else {
4087                 bargs = NULL;
4088         }
4089
4090         if (fs_info->balance_ctl) {
4091                 ret = -EINPROGRESS;
4092                 goto out_bargs;
4093         }
4094
4095         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4096         if (!bctl) {
4097                 ret = -ENOMEM;
4098                 goto out_bargs;
4099         }
4100
4101         bctl->fs_info = fs_info;
4102         if (arg) {
4103                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4104                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4105                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4106
4107                 bctl->flags = bargs->flags;
4108         } else {
4109                 /* balance everything - no filters */
4110                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4111         }
4112
4113 do_balance:
4114         /*
4115          * Ownership of bctl and mutually_exclusive_operation_running
4116          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4117          * or, if restriper was paused all the way until unmount, in
4118          * free_fs_info.  mutually_exclusive_operation_running is
4119          * cleared in __cancel_balance.
4120          */
4121         need_unlock = false;
4122
4123         ret = btrfs_balance(bctl, bargs);
4124
4125         if (arg) {
4126                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4127                         ret = -EFAULT;
4128         }
4129
4130 out_bargs:
4131         kfree(bargs);
4132 out_unlock:
4133         mutex_unlock(&fs_info->balance_mutex);
4134         mutex_unlock(&fs_info->volume_mutex);
4135         if (need_unlock)
4136                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4137 out:
4138         mnt_drop_write_file(file);
4139         return ret;
4140 }
4141
4142 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4143 {
4144         if (!capable(CAP_SYS_ADMIN))
4145                 return -EPERM;
4146
4147         switch (cmd) {
4148         case BTRFS_BALANCE_CTL_PAUSE:
4149                 return btrfs_pause_balance(root->fs_info);
4150         case BTRFS_BALANCE_CTL_CANCEL:
4151                 return btrfs_cancel_balance(root->fs_info);
4152         }
4153
4154         return -EINVAL;
4155 }
4156
4157 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4158                                          void __user *arg)
4159 {
4160         struct btrfs_fs_info *fs_info = root->fs_info;
4161         struct btrfs_ioctl_balance_args *bargs;
4162         int ret = 0;
4163
4164         if (!capable(CAP_SYS_ADMIN))
4165                 return -EPERM;
4166
4167         mutex_lock(&fs_info->balance_mutex);
4168         if (!fs_info->balance_ctl) {
4169                 ret = -ENOTCONN;
4170                 goto out;
4171         }
4172
4173         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
4174         if (!bargs) {
4175                 ret = -ENOMEM;
4176                 goto out;
4177         }
4178
4179         update_ioctl_balance_args(fs_info, 1, bargs);
4180
4181         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4182                 ret = -EFAULT;
4183
4184         kfree(bargs);
4185 out:
4186         mutex_unlock(&fs_info->balance_mutex);
4187         return ret;
4188 }
4189
4190 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4191 {
4192         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4193         struct btrfs_ioctl_quota_ctl_args *sa;
4194         struct btrfs_trans_handle *trans = NULL;
4195         int ret;
4196         int err;
4197
4198         if (!capable(CAP_SYS_ADMIN))
4199                 return -EPERM;
4200
4201         ret = mnt_want_write_file(file);
4202         if (ret)
4203                 return ret;
4204
4205         sa = memdup_user(arg, sizeof(*sa));
4206         if (IS_ERR(sa)) {
4207                 ret = PTR_ERR(sa);
4208                 goto drop_write;
4209         }
4210
4211         down_write(&root->fs_info->subvol_sem);
4212         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4213         if (IS_ERR(trans)) {
4214                 ret = PTR_ERR(trans);
4215                 goto out;
4216         }
4217
4218         switch (sa->cmd) {
4219         case BTRFS_QUOTA_CTL_ENABLE:
4220                 ret = btrfs_quota_enable(trans, root->fs_info);
4221                 break;
4222         case BTRFS_QUOTA_CTL_DISABLE:
4223                 ret = btrfs_quota_disable(trans, root->fs_info);
4224                 break;
4225         default:
4226                 ret = -EINVAL;
4227                 break;
4228         }
4229
4230         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4231         if (err && !ret)
4232                 ret = err;
4233 out:
4234         kfree(sa);
4235         up_write(&root->fs_info->subvol_sem);
4236 drop_write:
4237         mnt_drop_write_file(file);
4238         return ret;
4239 }
4240
4241 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4242 {
4243         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4244         struct btrfs_ioctl_qgroup_assign_args *sa;
4245         struct btrfs_trans_handle *trans;
4246         int ret;
4247         int err;
4248
4249         if (!capable(CAP_SYS_ADMIN))
4250                 return -EPERM;
4251
4252         ret = mnt_want_write_file(file);
4253         if (ret)
4254                 return ret;
4255
4256         sa = memdup_user(arg, sizeof(*sa));
4257         if (IS_ERR(sa)) {
4258                 ret = PTR_ERR(sa);
4259                 goto drop_write;
4260         }
4261
4262         trans = btrfs_join_transaction(root);
4263         if (IS_ERR(trans)) {
4264                 ret = PTR_ERR(trans);
4265                 goto out;
4266         }
4267
4268         /* FIXME: check if the IDs really exist */
4269         if (sa->assign) {
4270                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4271                                                 sa->src, sa->dst);
4272         } else {
4273                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4274                                                 sa->src, sa->dst);
4275         }
4276
4277         err = btrfs_end_transaction(trans, root);
4278         if (err && !ret)
4279                 ret = err;
4280
4281 out:
4282         kfree(sa);
4283 drop_write:
4284         mnt_drop_write_file(file);
4285         return ret;
4286 }
4287
4288 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4289 {
4290         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4291         struct btrfs_ioctl_qgroup_create_args *sa;
4292         struct btrfs_trans_handle *trans;
4293         int ret;
4294         int err;
4295
4296         if (!capable(CAP_SYS_ADMIN))
4297                 return -EPERM;
4298
4299         ret = mnt_want_write_file(file);
4300         if (ret)
4301                 return ret;
4302
4303         sa = memdup_user(arg, sizeof(*sa));
4304         if (IS_ERR(sa)) {
4305                 ret = PTR_ERR(sa);
4306                 goto drop_write;
4307         }
4308
4309         if (!sa->qgroupid) {
4310                 ret = -EINVAL;
4311                 goto out;
4312         }
4313
4314         trans = btrfs_join_transaction(root);
4315         if (IS_ERR(trans)) {
4316                 ret = PTR_ERR(trans);
4317                 goto out;
4318         }
4319
4320         /* FIXME: check if the IDs really exist */
4321         if (sa->create) {
4322                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
4323                                           NULL);
4324         } else {
4325                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4326         }
4327
4328         err = btrfs_end_transaction(trans, root);
4329         if (err && !ret)
4330                 ret = err;
4331
4332 out:
4333         kfree(sa);
4334 drop_write:
4335         mnt_drop_write_file(file);
4336         return ret;
4337 }
4338
4339 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4340 {
4341         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4342         struct btrfs_ioctl_qgroup_limit_args *sa;
4343         struct btrfs_trans_handle *trans;
4344         int ret;
4345         int err;
4346         u64 qgroupid;
4347
4348         if (!capable(CAP_SYS_ADMIN))
4349                 return -EPERM;
4350
4351         ret = mnt_want_write_file(file);
4352         if (ret)
4353                 return ret;
4354
4355         sa = memdup_user(arg, sizeof(*sa));
4356         if (IS_ERR(sa)) {
4357                 ret = PTR_ERR(sa);
4358                 goto drop_write;
4359         }
4360
4361         trans = btrfs_join_transaction(root);
4362         if (IS_ERR(trans)) {
4363                 ret = PTR_ERR(trans);
4364                 goto out;
4365         }
4366
4367         qgroupid = sa->qgroupid;
4368         if (!qgroupid) {
4369                 /* take the current subvol as qgroup */
4370                 qgroupid = root->root_key.objectid;
4371         }
4372
4373         /* FIXME: check if the IDs really exist */
4374         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4375
4376         err = btrfs_end_transaction(trans, root);
4377         if (err && !ret)
4378                 ret = err;
4379
4380 out:
4381         kfree(sa);
4382 drop_write:
4383         mnt_drop_write_file(file);
4384         return ret;
4385 }
4386
4387 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4388 {
4389         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4390         struct btrfs_ioctl_quota_rescan_args *qsa;
4391         int ret;
4392
4393         if (!capable(CAP_SYS_ADMIN))
4394                 return -EPERM;
4395
4396         ret = mnt_want_write_file(file);
4397         if (ret)
4398                 return ret;
4399
4400         qsa = memdup_user(arg, sizeof(*qsa));
4401         if (IS_ERR(qsa)) {
4402                 ret = PTR_ERR(qsa);
4403                 goto drop_write;
4404         }
4405
4406         if (qsa->flags) {
4407                 ret = -EINVAL;
4408                 goto out;
4409         }
4410
4411         ret = btrfs_qgroup_rescan(root->fs_info);
4412
4413 out:
4414         kfree(qsa);
4415 drop_write:
4416         mnt_drop_write_file(file);
4417         return ret;
4418 }
4419
4420 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4421 {
4422         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4423         struct btrfs_ioctl_quota_rescan_args *qsa;
4424         int ret = 0;
4425
4426         if (!capable(CAP_SYS_ADMIN))
4427                 return -EPERM;
4428
4429         qsa = kzalloc(sizeof(*qsa), GFP_NOFS);
4430         if (!qsa)
4431                 return -ENOMEM;
4432
4433         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4434                 qsa->flags = 1;
4435                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
4436         }
4437
4438         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4439                 ret = -EFAULT;
4440
4441         kfree(qsa);
4442         return ret;
4443 }
4444
4445 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4446 {
4447         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4448
4449         if (!capable(CAP_SYS_ADMIN))
4450                 return -EPERM;
4451
4452         return btrfs_qgroup_wait_for_completion(root->fs_info);
4453 }
4454
4455 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4456                                             struct btrfs_ioctl_received_subvol_args *sa)
4457 {
4458         struct inode *inode = file_inode(file);
4459         struct btrfs_root *root = BTRFS_I(inode)->root;
4460         struct btrfs_root_item *root_item = &root->root_item;
4461         struct btrfs_trans_handle *trans;
4462         struct timespec ct = CURRENT_TIME;
4463         int ret = 0;
4464         int received_uuid_changed;
4465
4466         if (!inode_owner_or_capable(inode))
4467                 return -EPERM;
4468
4469         ret = mnt_want_write_file(file);
4470         if (ret < 0)
4471                 return ret;
4472
4473         down_write(&root->fs_info->subvol_sem);
4474
4475         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
4476                 ret = -EINVAL;
4477                 goto out;
4478         }
4479
4480         if (btrfs_root_readonly(root)) {
4481                 ret = -EROFS;
4482                 goto out;
4483         }
4484
4485         /*
4486          * 1 - root item
4487          * 2 - uuid items (received uuid + subvol uuid)
4488          */
4489         trans = btrfs_start_transaction(root, 3);
4490         if (IS_ERR(trans)) {
4491                 ret = PTR_ERR(trans);
4492                 trans = NULL;
4493                 goto out;
4494         }
4495
4496         sa->rtransid = trans->transid;
4497         sa->rtime.sec = ct.tv_sec;
4498         sa->rtime.nsec = ct.tv_nsec;
4499
4500         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4501                                        BTRFS_UUID_SIZE);
4502         if (received_uuid_changed &&
4503             !btrfs_is_empty_uuid(root_item->received_uuid))
4504                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
4505                                     root_item->received_uuid,
4506                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4507                                     root->root_key.objectid);
4508         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4509         btrfs_set_root_stransid(root_item, sa->stransid);
4510         btrfs_set_root_rtransid(root_item, sa->rtransid);
4511         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4512         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4513         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4514         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4515
4516         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4517                                 &root->root_key, &root->root_item);
4518         if (ret < 0) {
4519                 btrfs_end_transaction(trans, root);
4520                 goto out;
4521         }
4522         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4523                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
4524                                           sa->uuid,
4525                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4526                                           root->root_key.objectid);
4527                 if (ret < 0 && ret != -EEXIST) {
4528                         btrfs_abort_transaction(trans, root, ret);
4529                         goto out;
4530                 }
4531         }
4532         ret = btrfs_commit_transaction(trans, root);
4533         if (ret < 0) {
4534                 btrfs_abort_transaction(trans, root, ret);
4535                 goto out;
4536         }
4537
4538 out:
4539         up_write(&root->fs_info->subvol_sem);
4540         mnt_drop_write_file(file);
4541         return ret;
4542 }
4543
4544 #ifdef CONFIG_64BIT
4545 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4546                                                 void __user *arg)
4547 {
4548         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4549         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4550         int ret = 0;
4551
4552         args32 = memdup_user(arg, sizeof(*args32));
4553         if (IS_ERR(args32)) {
4554                 ret = PTR_ERR(args32);
4555                 args32 = NULL;
4556                 goto out;
4557         }
4558
4559         args64 = kmalloc(sizeof(*args64), GFP_NOFS);
4560         if (!args64) {
4561                 ret = -ENOMEM;
4562                 goto out;
4563         }
4564
4565         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4566         args64->stransid = args32->stransid;
4567         args64->rtransid = args32->rtransid;
4568         args64->stime.sec = args32->stime.sec;
4569         args64->stime.nsec = args32->stime.nsec;
4570         args64->rtime.sec = args32->rtime.sec;
4571         args64->rtime.nsec = args32->rtime.nsec;
4572         args64->flags = args32->flags;
4573
4574         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4575         if (ret)
4576                 goto out;
4577
4578         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4579         args32->stransid = args64->stransid;
4580         args32->rtransid = args64->rtransid;
4581         args32->stime.sec = args64->stime.sec;
4582         args32->stime.nsec = args64->stime.nsec;
4583         args32->rtime.sec = args64->rtime.sec;
4584         args32->rtime.nsec = args64->rtime.nsec;
4585         args32->flags = args64->flags;
4586
4587         ret = copy_to_user(arg, args32, sizeof(*args32));
4588         if (ret)
4589                 ret = -EFAULT;
4590
4591 out:
4592         kfree(args32);
4593         kfree(args64);
4594         return ret;
4595 }
4596 #endif
4597
4598 static long btrfs_ioctl_set_received_subvol(struct file *file,
4599                                             void __user *arg)
4600 {
4601         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4602         int ret = 0;
4603
4604         sa = memdup_user(arg, sizeof(*sa));
4605         if (IS_ERR(sa)) {
4606                 ret = PTR_ERR(sa);
4607                 sa = NULL;
4608                 goto out;
4609         }
4610
4611         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4612
4613         if (ret)
4614                 goto out;
4615
4616         ret = copy_to_user(arg, sa, sizeof(*sa));
4617         if (ret)
4618                 ret = -EFAULT;
4619
4620 out:
4621         kfree(sa);
4622         return ret;
4623 }
4624
4625 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
4626 {
4627         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4628         size_t len;
4629         int ret;
4630         char label[BTRFS_LABEL_SIZE];
4631
4632         spin_lock(&root->fs_info->super_lock);
4633         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4634         spin_unlock(&root->fs_info->super_lock);
4635
4636         len = strnlen(label, BTRFS_LABEL_SIZE);
4637
4638         if (len == BTRFS_LABEL_SIZE) {
4639                 btrfs_warn(root->fs_info,
4640                         "label is too long, return the first %zu bytes", --len);
4641         }
4642
4643         ret = copy_to_user(arg, label, len);
4644
4645         return ret ? -EFAULT : 0;
4646 }
4647
4648 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4649 {
4650         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4651         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4652         struct btrfs_trans_handle *trans;
4653         char label[BTRFS_LABEL_SIZE];
4654         int ret;
4655
4656         if (!capable(CAP_SYS_ADMIN))
4657                 return -EPERM;
4658
4659         if (copy_from_user(label, arg, sizeof(label)))
4660                 return -EFAULT;
4661
4662         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4663                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
4664                        BTRFS_LABEL_SIZE - 1);
4665                 return -EINVAL;
4666         }
4667
4668         ret = mnt_want_write_file(file);
4669         if (ret)
4670                 return ret;
4671
4672         trans = btrfs_start_transaction(root, 0);
4673         if (IS_ERR(trans)) {
4674                 ret = PTR_ERR(trans);
4675                 goto out_unlock;
4676         }
4677
4678         spin_lock(&root->fs_info->super_lock);
4679         strcpy(super_block->label, label);
4680         spin_unlock(&root->fs_info->super_lock);
4681         ret = btrfs_commit_transaction(trans, root);
4682
4683 out_unlock:
4684         mnt_drop_write_file(file);
4685         return ret;
4686 }
4687
4688 #define INIT_FEATURE_FLAGS(suffix) \
4689         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4690           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4691           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4692
4693 static int btrfs_ioctl_get_supported_features(struct file *file,
4694                                               void __user *arg)
4695 {
4696         static struct btrfs_ioctl_feature_flags features[3] = {
4697                 INIT_FEATURE_FLAGS(SUPP),
4698                 INIT_FEATURE_FLAGS(SAFE_SET),
4699                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4700         };
4701
4702         if (copy_to_user(arg, &features, sizeof(features)))
4703                 return -EFAULT;
4704
4705         return 0;
4706 }
4707
4708 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
4709 {
4710         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4711         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4712         struct btrfs_ioctl_feature_flags features;
4713
4714         features.compat_flags = btrfs_super_compat_flags(super_block);
4715         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4716         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4717
4718         if (copy_to_user(arg, &features, sizeof(features)))
4719                 return -EFAULT;
4720
4721         return 0;
4722 }
4723
4724 static int check_feature_bits(struct btrfs_root *root,
4725                               enum btrfs_feature_set set,
4726                               u64 change_mask, u64 flags, u64 supported_flags,
4727                               u64 safe_set, u64 safe_clear)
4728 {
4729         const char *type = btrfs_feature_set_names[set];
4730         char *names;
4731         u64 disallowed, unsupported;
4732         u64 set_mask = flags & change_mask;
4733         u64 clear_mask = ~flags & change_mask;
4734
4735         unsupported = set_mask & ~supported_flags;
4736         if (unsupported) {
4737                 names = btrfs_printable_features(set, unsupported);
4738                 if (names) {
4739                         btrfs_warn(root->fs_info,
4740                            "this kernel does not support the %s feature bit%s",
4741                            names, strchr(names, ',') ? "s" : "");
4742                         kfree(names);
4743                 } else
4744                         btrfs_warn(root->fs_info,
4745                            "this kernel does not support %s bits 0x%llx",
4746                            type, unsupported);
4747                 return -EOPNOTSUPP;
4748         }
4749
4750         disallowed = set_mask & ~safe_set;
4751         if (disallowed) {
4752                 names = btrfs_printable_features(set, disallowed);
4753                 if (names) {
4754                         btrfs_warn(root->fs_info,
4755                            "can't set the %s feature bit%s while mounted",
4756                            names, strchr(names, ',') ? "s" : "");
4757                         kfree(names);
4758                 } else
4759                         btrfs_warn(root->fs_info,
4760                            "can't set %s bits 0x%llx while mounted",
4761                            type, disallowed);
4762                 return -EPERM;
4763         }
4764
4765         disallowed = clear_mask & ~safe_clear;
4766         if (disallowed) {
4767                 names = btrfs_printable_features(set, disallowed);
4768                 if (names) {
4769                         btrfs_warn(root->fs_info,
4770                            "can't clear the %s feature bit%s while mounted",
4771                            names, strchr(names, ',') ? "s" : "");
4772                         kfree(names);
4773                 } else
4774                         btrfs_warn(root->fs_info,
4775                            "can't clear %s bits 0x%llx while mounted",
4776                            type, disallowed);
4777                 return -EPERM;
4778         }
4779
4780         return 0;
4781 }
4782
4783 #define check_feature(root, change_mask, flags, mask_base)      \
4784 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
4785                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4786                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4787                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4788
4789 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4790 {
4791         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4792         struct btrfs_super_block *super_block = root->fs_info->super_copy;
4793         struct btrfs_ioctl_feature_flags flags[2];
4794         struct btrfs_trans_handle *trans;
4795         u64 newflags;
4796         int ret;
4797
4798         if (!capable(CAP_SYS_ADMIN))
4799                 return -EPERM;
4800
4801         if (copy_from_user(flags, arg, sizeof(flags)))
4802                 return -EFAULT;
4803
4804         /* Nothing to do */
4805         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4806             !flags[0].incompat_flags)
4807                 return 0;
4808
4809         ret = check_feature(root, flags[0].compat_flags,
4810                             flags[1].compat_flags, COMPAT);
4811         if (ret)
4812                 return ret;
4813
4814         ret = check_feature(root, flags[0].compat_ro_flags,
4815                             flags[1].compat_ro_flags, COMPAT_RO);
4816         if (ret)
4817                 return ret;
4818
4819         ret = check_feature(root, flags[0].incompat_flags,
4820                             flags[1].incompat_flags, INCOMPAT);
4821         if (ret)
4822                 return ret;
4823
4824         trans = btrfs_start_transaction(root, 0);
4825         if (IS_ERR(trans))
4826                 return PTR_ERR(trans);
4827
4828         spin_lock(&root->fs_info->super_lock);
4829         newflags = btrfs_super_compat_flags(super_block);
4830         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4831         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4832         btrfs_set_super_compat_flags(super_block, newflags);
4833
4834         newflags = btrfs_super_compat_ro_flags(super_block);
4835         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4836         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4837         btrfs_set_super_compat_ro_flags(super_block, newflags);
4838
4839         newflags = btrfs_super_incompat_flags(super_block);
4840         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4841         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4842         btrfs_set_super_incompat_flags(super_block, newflags);
4843         spin_unlock(&root->fs_info->super_lock);
4844
4845         return btrfs_commit_transaction(trans, root);
4846 }
4847
4848 long btrfs_ioctl(struct file *file, unsigned int
4849                 cmd, unsigned long arg)
4850 {
4851         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4852         void __user *argp = (void __user *)arg;
4853
4854         switch (cmd) {
4855         case FS_IOC_GETFLAGS:
4856                 return btrfs_ioctl_getflags(file, argp);
4857         case FS_IOC_SETFLAGS:
4858                 return btrfs_ioctl_setflags(file, argp);
4859         case FS_IOC_GETVERSION:
4860                 return btrfs_ioctl_getversion(file, argp);
4861         case FITRIM:
4862                 return btrfs_ioctl_fitrim(file, argp);
4863         case BTRFS_IOC_SNAP_CREATE:
4864                 return btrfs_ioctl_snap_create(file, argp, 0);
4865         case BTRFS_IOC_SNAP_CREATE_V2:
4866                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4867         case BTRFS_IOC_SUBVOL_CREATE:
4868                 return btrfs_ioctl_snap_create(file, argp, 1);
4869         case BTRFS_IOC_SUBVOL_CREATE_V2:
4870                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4871         case BTRFS_IOC_SNAP_DESTROY:
4872                 return btrfs_ioctl_snap_destroy(file, argp);
4873         case BTRFS_IOC_SUBVOL_GETFLAGS:
4874                 return btrfs_ioctl_subvol_getflags(file, argp);
4875         case BTRFS_IOC_SUBVOL_SETFLAGS:
4876                 return btrfs_ioctl_subvol_setflags(file, argp);
4877         case BTRFS_IOC_DEFAULT_SUBVOL:
4878                 return btrfs_ioctl_default_subvol(file, argp);
4879         case BTRFS_IOC_DEFRAG:
4880                 return btrfs_ioctl_defrag(file, NULL);
4881         case BTRFS_IOC_DEFRAG_RANGE:
4882                 return btrfs_ioctl_defrag(file, argp);
4883         case BTRFS_IOC_RESIZE:
4884                 return btrfs_ioctl_resize(file, argp);
4885         case BTRFS_IOC_ADD_DEV:
4886                 return btrfs_ioctl_add_dev(root, argp);
4887         case BTRFS_IOC_RM_DEV:
4888                 return btrfs_ioctl_rm_dev(file, argp);
4889         case BTRFS_IOC_FS_INFO:
4890                 return btrfs_ioctl_fs_info(root, argp);
4891         case BTRFS_IOC_DEV_INFO:
4892                 return btrfs_ioctl_dev_info(root, argp);
4893         case BTRFS_IOC_BALANCE:
4894                 return btrfs_ioctl_balance(file, NULL);
4895         case BTRFS_IOC_CLONE:
4896                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
4897         case BTRFS_IOC_CLONE_RANGE:
4898                 return btrfs_ioctl_clone_range(file, argp);
4899         case BTRFS_IOC_TRANS_START:
4900                 return btrfs_ioctl_trans_start(file);
4901         case BTRFS_IOC_TRANS_END:
4902                 return btrfs_ioctl_trans_end(file);
4903         case BTRFS_IOC_TREE_SEARCH:
4904                 return btrfs_ioctl_tree_search(file, argp);
4905         case BTRFS_IOC_INO_LOOKUP:
4906                 return btrfs_ioctl_ino_lookup(file, argp);
4907         case BTRFS_IOC_INO_PATHS:
4908                 return btrfs_ioctl_ino_to_path(root, argp);
4909         case BTRFS_IOC_LOGICAL_INO:
4910                 return btrfs_ioctl_logical_to_ino(root, argp);
4911         case BTRFS_IOC_SPACE_INFO:
4912                 return btrfs_ioctl_space_info(root, argp);
4913         case BTRFS_IOC_SYNC: {
4914                 int ret;
4915
4916                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
4917                 if (ret)
4918                         return ret;
4919                 ret = btrfs_sync_fs(file->f_dentry->d_sb, 1);
4920                 return ret;
4921         }
4922         case BTRFS_IOC_START_SYNC:
4923                 return btrfs_ioctl_start_sync(root, argp);
4924         case BTRFS_IOC_WAIT_SYNC:
4925                 return btrfs_ioctl_wait_sync(root, argp);
4926         case BTRFS_IOC_SCRUB:
4927                 return btrfs_ioctl_scrub(file, argp);
4928         case BTRFS_IOC_SCRUB_CANCEL:
4929                 return btrfs_ioctl_scrub_cancel(root, argp);
4930         case BTRFS_IOC_SCRUB_PROGRESS:
4931                 return btrfs_ioctl_scrub_progress(root, argp);
4932         case BTRFS_IOC_BALANCE_V2:
4933                 return btrfs_ioctl_balance(file, argp);
4934         case BTRFS_IOC_BALANCE_CTL:
4935                 return btrfs_ioctl_balance_ctl(root, arg);
4936         case BTRFS_IOC_BALANCE_PROGRESS:
4937                 return btrfs_ioctl_balance_progress(root, argp);
4938         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4939                 return btrfs_ioctl_set_received_subvol(file, argp);
4940 #ifdef CONFIG_64BIT
4941         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4942                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4943 #endif
4944         case BTRFS_IOC_SEND:
4945                 return btrfs_ioctl_send(file, argp);
4946         case BTRFS_IOC_GET_DEV_STATS:
4947                 return btrfs_ioctl_get_dev_stats(root, argp);
4948         case BTRFS_IOC_QUOTA_CTL:
4949                 return btrfs_ioctl_quota_ctl(file, argp);
4950         case BTRFS_IOC_QGROUP_ASSIGN:
4951                 return btrfs_ioctl_qgroup_assign(file, argp);
4952         case BTRFS_IOC_QGROUP_CREATE:
4953                 return btrfs_ioctl_qgroup_create(file, argp);
4954         case BTRFS_IOC_QGROUP_LIMIT:
4955                 return btrfs_ioctl_qgroup_limit(file, argp);
4956         case BTRFS_IOC_QUOTA_RESCAN:
4957                 return btrfs_ioctl_quota_rescan(file, argp);
4958         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4959                 return btrfs_ioctl_quota_rescan_status(file, argp);
4960         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4961                 return btrfs_ioctl_quota_rescan_wait(file, argp);
4962         case BTRFS_IOC_DEV_REPLACE:
4963                 return btrfs_ioctl_dev_replace(root, argp);
4964         case BTRFS_IOC_GET_FSLABEL:
4965                 return btrfs_ioctl_get_fslabel(file, argp);
4966         case BTRFS_IOC_SET_FSLABEL:
4967                 return btrfs_ioctl_set_fslabel(file, argp);
4968         case BTRFS_IOC_FILE_EXTENT_SAME:
4969                 return btrfs_ioctl_file_extent_same(file, argp);
4970         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4971                 return btrfs_ioctl_get_supported_features(file, argp);
4972         case BTRFS_IOC_GET_FEATURES:
4973                 return btrfs_ioctl_get_features(file, argp);
4974         case BTRFS_IOC_SET_FEATURES:
4975                 return btrfs_ioctl_set_features(file, argp);
4976         }
4977
4978         return -ENOTTY;
4979 }