fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/uuid.h>
  45 #include <linux/btrfs.h>
  46 #include <linux/uaccess.h>
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "volumes.h"
  53 #include "locking.h"
  54 #include "inode-map.h"
  55 #include "backref.h"
  56 #include "rcu-string.h"
  57 #include "send.h"
  58 #include "dev-replace.h"
  59 #include "props.h"
  60 #include "sysfs.h"
  61 #include "qgroup.h"
  62
  63 #ifdef CONFIG_64BIT
  64 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  65  * structures are incorrect, as the timespec structure from userspace
  66  * is 4 bytes too small. We define these alternatives here to teach
  67  * the kernel about the 32-bit struct packing.
  68  */
  69 struct btrfs_ioctl_timespec_32 {
  70         __u64 sec;
  71         __u32 nsec;
  72 } __attribute__ ((__packed__));
  73
  74 struct btrfs_ioctl_received_subvol_args_32 {
  75         char    uuid[BTRFS_UUID_SIZE];  /* in */
  76         __u64   stransid;               /* in */
  77         __u64   rtransid;               /* out */
  78         struct btrfs_ioctl_timespec_32 stime; /* in */
  79         struct btrfs_ioctl_timespec_32 rtime; /* out */
  80         __u64   flags;                  /* in */
  81         __u64   reserved[16];           /* in */
  82 } __attribute__ ((__packed__));
  83
  84 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  85                                 struct btrfs_ioctl_received_subvol_args_32)
  86 #endif
  87
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  91                        int no_time_update);
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  95 {
  96         if (S_ISDIR(mode))
  97                 return flags;
  98         else if (S_ISREG(mode))
  99                 return flags & ~FS_DIRSYNC_FL;
 100         else
 101                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 102 }
 103
 104 /*
 105  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 106  */
 107 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 108 {
 109         unsigned int iflags = 0;
 110
 111         if (flags & BTRFS_INODE_SYNC)
 112                 iflags |= FS_SYNC_FL;
 113         if (flags & BTRFS_INODE_IMMUTABLE)
 114                 iflags |= FS_IMMUTABLE_FL;
 115         if (flags & BTRFS_INODE_APPEND)
 116                 iflags |= FS_APPEND_FL;
 117         if (flags & BTRFS_INODE_NODUMP)
 118                 iflags |= FS_NODUMP_FL;
 119         if (flags & BTRFS_INODE_NOATIME)
 120                 iflags |= FS_NOATIME_FL;
 121         if (flags & BTRFS_INODE_DIRSYNC)
 122                 iflags |= FS_DIRSYNC_FL;
 123         if (flags & BTRFS_INODE_NODATACOW)
 124                 iflags |= FS_NOCOW_FL;
 125
 126         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
 127                 iflags |= FS_COMPR_FL;
 128         else if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130
 131         return iflags;
 132 }
 133
 134 /*
 135  * Update inode->i_flags based on the btrfs internal flags.
 136  */
 137 void btrfs_update_iflags(struct inode *inode)
 138 {
 139         struct btrfs_inode *ip = BTRFS_I(inode);
 140         unsigned int new_fl = 0;
 141
 142         if (ip->flags & BTRFS_INODE_SYNC)
 143                 new_fl |= S_SYNC;
 144         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 145                 new_fl |= S_IMMUTABLE;
 146         if (ip->flags & BTRFS_INODE_APPEND)
 147                 new_fl |= S_APPEND;
 148         if (ip->flags & BTRFS_INODE_NOATIME)
 149                 new_fl |= S_NOATIME;
 150         if (ip->flags & BTRFS_INODE_DIRSYNC)
 151                 new_fl |= S_DIRSYNC;
 152
 153         set_mask_bits(&inode->i_flags,
 154                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 155                       new_fl);
 156 }
 157
 158 /*
 159  * Inherit flags from the parent inode.
 160  *
 161  * Currently only the compression flags and the cow flags are inherited.
 162  */
 163 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 164 {
 165         unsigned int flags;
 166
 167         if (!dir)
 168                 return;
 169
 170         flags = BTRFS_I(dir)->flags;
 171
 172         if (flags & BTRFS_INODE_NOCOMPRESS) {
 173                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 174                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 175         } else if (flags & BTRFS_INODE_COMPRESS) {
 176                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 177                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 178         }
 179
 180         if (flags & BTRFS_INODE_NODATACOW) {
 181                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 182                 if (S_ISREG(inode->i_mode))
 183                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 184         }
 185
 186         btrfs_update_iflags(inode);
 187 }
 188
 189 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 190 {
 191         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 192         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 193
 194         if (copy_to_user(arg, &flags, sizeof(flags)))
 195                 return -EFAULT;
 196         return 0;
 197 }
 198
 199 static int check_flags(unsigned int flags)
 200 {
 201         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 202                       FS_NOATIME_FL | FS_NODUMP_FL | \
 203                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 204                       FS_NOCOMP_FL | FS_COMPR_FL |
 205                       FS_NOCOW_FL))
 206                 return -EOPNOTSUPP;
 207
 208         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 215 {
 216         struct inode *inode = file_inode(file);
 217         struct btrfs_inode *ip = BTRFS_I(inode);
 218         struct btrfs_root *root = ip->root;
 219         struct btrfs_trans_handle *trans;
 220         unsigned int flags, oldflags;
 221         int ret;
 222         u64 ip_oldflags;
 223         unsigned int i_oldflags;
 224         umode_t mode;
 225
 226         if (!inode_owner_or_capable(inode))
 227                 return -EPERM;
 228
 229         if (btrfs_root_readonly(root))
 230                 return -EROFS;
 231
 232         if (copy_from_user(&flags, arg, sizeof(flags)))
 233                 return -EFAULT;
 234
 235         ret = check_flags(flags);
 236         if (ret)
 237                 return ret;
 238
 239         ret = mnt_want_write_file(file);
 240         if (ret)
 241                 return ret;
 242
 243         mutex_lock(&inode->i_mutex);
 244
 245         ip_oldflags = ip->flags;
 246         i_oldflags = inode->i_flags;
 247         mode = inode->i_mode;
 248
 249         flags = btrfs_mask_flags(inode->i_mode, flags);
 250         oldflags = btrfs_flags_to_ioctl(ip->flags);
 251         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 252                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 253                         ret = -EPERM;
 254                         goto out_unlock;
 255                 }
 256         }
 257
 258         if (flags & FS_SYNC_FL)
 259                 ip->flags |= BTRFS_INODE_SYNC;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_SYNC;
 262         if (flags & FS_IMMUTABLE_FL)
 263                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 266         if (flags & FS_APPEND_FL)
 267                 ip->flags |= BTRFS_INODE_APPEND;
 268         else
 269                 ip->flags &= ~BTRFS_INODE_APPEND;
 270         if (flags & FS_NODUMP_FL)
 271                 ip->flags |= BTRFS_INODE_NODUMP;
 272         else
 273                 ip->flags &= ~BTRFS_INODE_NODUMP;
 274         if (flags & FS_NOATIME_FL)
 275                 ip->flags |= BTRFS_INODE_NOATIME;
 276         else
 277                 ip->flags &= ~BTRFS_INODE_NOATIME;
 278         if (flags & FS_DIRSYNC_FL)
 279                 ip->flags |= BTRFS_INODE_DIRSYNC;
 280         else
 281                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 282         if (flags & FS_NOCOW_FL) {
 283                 if (S_ISREG(mode)) {
 284                         /*
 285                          * It's safe to turn csums off here, no extents exist.
 286                          * Otherwise we want the flag to reflect the real COW
 287                          * status of the file and will not set it.
 288                          */
 289                         if (inode->i_size == 0)
 290                                 ip->flags |= BTRFS_INODE_NODATACOW
 291                                            | BTRFS_INODE_NODATASUM;
 292                 } else {
 293                         ip->flags |= BTRFS_INODE_NODATACOW;
 294                 }
 295         } else {
 296                 /*
 297                  * Revert back under same assuptions as above
 298                  */
 299                 if (S_ISREG(mode)) {
 300                         if (inode->i_size == 0)
 301                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 302                                              | BTRFS_INODE_NODATASUM);
 303                 } else {
 304                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 305                 }
 306         }
 307
 308         /*
 309          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 310          * flag may be changed automatically if compression code won't make
 311          * things smaller.
 312          */
 313         if (flags & FS_NOCOMP_FL) {
 314                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 315                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 316
 317                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 318                 if (ret && ret != -ENODATA)
 319                         goto out_drop;
 320         } else if (flags & FS_COMPR_FL) {
 321                 const char *comp;
 322
 323                 ip->flags |= BTRFS_INODE_COMPRESS;
 324                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 325
 326                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 327                         comp = "lzo";
 328                 else
 329                         comp = "zlib";
 330                 ret = btrfs_set_prop(inode, "btrfs.compression",
 331                                      comp, strlen(comp), 0);
 332                 if (ret)
 333                         goto out_drop;
 334
 335         } else {
 336                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 337                 if (ret && ret != -ENODATA)
 338                         goto out_drop;
 339                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 340         }
 341
 342         trans = btrfs_start_transaction(root, 1);
 343         if (IS_ERR(trans)) {
 344                 ret = PTR_ERR(trans);
 345                 goto out_drop;
 346         }
 347
 348         btrfs_update_iflags(inode);
 349         inode_inc_iversion(inode);
 350         inode->i_ctime = CURRENT_TIME;
 351         ret = btrfs_update_inode(trans, root, inode);
 352
 353         btrfs_end_transaction(trans, root);
 354  out_drop:
 355         if (ret) {
 356                 ip->flags = ip_oldflags;
 357                 inode->i_flags = i_oldflags;
 358         }
 359
 360  out_unlock:
 361         mutex_unlock(&inode->i_mutex);
 362         mnt_drop_write_file(file);
 363         return ret;
 364 }
 365
 366 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 367 {
 368         struct inode *inode = file_inode(file);
 369
 370         return put_user(inode->i_generation, arg);
 371 }
 372
 373 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 374 {
 375         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 376         struct btrfs_device *device;
 377         struct request_queue *q;
 378         struct fstrim_range range;
 379         u64 minlen = ULLONG_MAX;
 380         u64 num_devices = 0;
 381         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 382         int ret;
 383
 384         if (!capable(CAP_SYS_ADMIN))
 385                 return -EPERM;
 386
 387         rcu_read_lock();
 388         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 389                                 dev_list) {
 390                 if (!device->bdev)
 391                         continue;
 392                 q = bdev_get_queue(device->bdev);
 393                 if (blk_queue_discard(q)) {
 394                         num_devices++;
 395                         minlen = min((u64)q->limits.discard_granularity,
 396                                      minlen);
 397                 }
 398         }
 399         rcu_read_unlock();
 400
 401         if (!num_devices)
 402                 return -EOPNOTSUPP;
 403         if (copy_from_user(&range, arg, sizeof(range)))
 404                 return -EFAULT;
 405         if (range.start > total_bytes ||
 406             range.len < fs_info->sb->s_blocksize)
 407                 return -EINVAL;
 408
 409         range.len = min(range.len, total_bytes - range.start);
 410         range.minlen = max(range.minlen, minlen);
 411         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 412         if (ret < 0)
 413                 return ret;
 414
 415         if (copy_to_user(arg, &range, sizeof(range)))
 416                 return -EFAULT;
 417
 418         return 0;
 419 }
 420
 421 int btrfs_is_empty_uuid(u8 *uuid)
 422 {
 423         int i;
 424
 425         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 426                 if (uuid[i])
 427                         return 0;
 428         }
 429         return 1;
 430 }
 431
 432 static noinline int create_subvol(struct inode *dir,
 433                                   struct dentry *dentry,
 434                                   char *name, int namelen,
 435                                   u64 *async_transid,
 436                                   struct btrfs_qgroup_inherit *inherit)
 437 {
 438         struct btrfs_trans_handle *trans;
 439         struct btrfs_key key;
 440         struct btrfs_root_item root_item;
 441         struct btrfs_inode_item *inode_item;
 442         struct extent_buffer *leaf;
 443         struct btrfs_root *root = BTRFS_I(dir)->root;
 444         struct btrfs_root *new_root;
 445         struct btrfs_block_rsv block_rsv;
 446         struct timespec cur_time = CURRENT_TIME;
 447         struct inode *inode;
 448         int ret;
 449         int err;
 450         u64 objectid;
 451         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 452         u64 index = 0;
 453         u64 qgroup_reserved;
 454         uuid_le new_uuid;
 455
 456         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 457         if (ret)
 458                 return ret;
 459
 460         /*
 461          * Don't create subvolume whose level is not zero. Or qgroup will be
 462          * screwed up since it assume subvolme qgroup's level to be 0.
 463          */
 464         if (btrfs_qgroup_level(objectid))
 465                 return -ENOSPC;
 466
 467         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 468         /*
 469          * The same as the snapshot creation, please see the comment
 470          * of create_snapshot().
 471          */
 472         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 473                                                8, &qgroup_reserved, false);
 474         if (ret)
 475                 return ret;
 476
 477         trans = btrfs_start_transaction(root, 0);
 478         if (IS_ERR(trans)) {
 479                 ret = PTR_ERR(trans);
 480                 btrfs_subvolume_release_metadata(root, &block_rsv,
 481                                                  qgroup_reserved);
 482                 return ret;
 483         }
 484         trans->block_rsv = &block_rsv;
 485         trans->bytes_reserved = block_rsv.size;
 486
 487         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 488         if (ret)
 489                 goto fail;
 490
 491         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 492         if (IS_ERR(leaf)) {
 493                 ret = PTR_ERR(leaf);
 494                 goto fail;
 495         }
 496
 497         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 498         btrfs_set_header_bytenr(leaf, leaf->start);
 499         btrfs_set_header_generation(leaf, trans->transid);
 500         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 501         btrfs_set_header_owner(leaf, objectid);
 502
 503         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 504                             BTRFS_FSID_SIZE);
 505         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 506                             btrfs_header_chunk_tree_uuid(leaf),
 507                             BTRFS_UUID_SIZE);
 508         btrfs_mark_buffer_dirty(leaf);
 509
 510         memset(&root_item, 0, sizeof(root_item));
 511
 512         inode_item = &root_item.inode;
 513         btrfs_set_stack_inode_generation(inode_item, 1);
 514         btrfs_set_stack_inode_size(inode_item, 3);
 515         btrfs_set_stack_inode_nlink(inode_item, 1);
 516         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 517         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 518
 519         btrfs_set_root_flags(&root_item, 0);
 520         btrfs_set_root_limit(&root_item, 0);
 521         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 522
 523         btrfs_set_root_bytenr(&root_item, leaf->start);
 524         btrfs_set_root_generation(&root_item, trans->transid);
 525         btrfs_set_root_level(&root_item, 0);
 526         btrfs_set_root_refs(&root_item, 1);
 527         btrfs_set_root_used(&root_item, leaf->len);
 528         btrfs_set_root_last_snapshot(&root_item, 0);
 529
 530         btrfs_set_root_generation_v2(&root_item,
 531                         btrfs_root_generation(&root_item));
 532         uuid_le_gen(&new_uuid);
 533         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
 534         btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec);
 535         btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec);
 536         root_item.ctime = root_item.otime;
 537         btrfs_set_root_ctransid(&root_item, trans->transid);
 538         btrfs_set_root_otransid(&root_item, trans->transid);
 539
 540         btrfs_tree_unlock(leaf);
 541         free_extent_buffer(leaf);
 542         leaf = NULL;
 543
 544         btrfs_set_root_dirid(&root_item, new_dirid);
 545
 546         key.objectid = objectid;
 547         key.offset = 0;
 548         key.type = BTRFS_ROOT_ITEM_KEY;
 549         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 550                                 &root_item);
 551         if (ret)
 552                 goto fail;
 553
 554         key.offset = (u64)-1;
 555         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 556         if (IS_ERR(new_root)) {
 557                 ret = PTR_ERR(new_root);
 558                 btrfs_abort_transaction(trans, root, ret);
 559                 goto fail;
 560         }
 561
 562         btrfs_record_root_in_trans(trans, new_root);
 563
 564         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 565         if (ret) {
 566                 /* We potentially lose an unused inode item here */
 567                 btrfs_abort_transaction(trans, root, ret);
 568                 goto fail;
 569         }
 570
 571         mutex_lock(&new_root->objectid_mutex);
 572         new_root->highest_objectid = new_dirid;
 573         mutex_unlock(&new_root->objectid_mutex);
 574
 575         /*
 576          * insert the directory item
 577          */
 578         ret = btrfs_set_inode_index(dir, &index);
 579         if (ret) {
 580                 btrfs_abort_transaction(trans, root, ret);
 581                 goto fail;
 582         }
 583
 584         ret = btrfs_insert_dir_item(trans, root,
 585                                     name, namelen, dir, &key,
 586                                     BTRFS_FT_DIR, index);
 587         if (ret) {
 588                 btrfs_abort_transaction(trans, root, ret);
 589                 goto fail;
 590         }
 591
 592         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 593         ret = btrfs_update_inode(trans, root, dir);
 594         BUG_ON(ret);
 595
 596         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 597                                  objectid, root->root_key.objectid,
 598                                  btrfs_ino(dir), index, name, namelen);
 599         BUG_ON(ret);
 600
 601         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 602                                   root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
 603                                   objectid);
 604         if (ret)
 605                 btrfs_abort_transaction(trans, root, ret);
 606
 607 fail:
 608         trans->block_rsv = NULL;
 609         trans->bytes_reserved = 0;
 610         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 611
 612         if (async_transid) {
 613                 *async_transid = trans->transid;
 614                 err = btrfs_commit_transaction_async(trans, root, 1);
 615                 if (err)
 616                         err = btrfs_commit_transaction(trans, root);
 617         } else {
 618                 err = btrfs_commit_transaction(trans, root);
 619         }
 620         if (err && !ret)
 621                 ret = err;
 622
 623         if (!ret) {
 624                 inode = btrfs_lookup_dentry(dir, dentry);
 625                 if (IS_ERR(inode))
 626                         return PTR_ERR(inode);
 627                 d_instantiate(dentry, inode);
 628         }
 629         return ret;
 630 }
 631
 632 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 633 {
 634         s64 writers;
 635         DEFINE_WAIT(wait);
 636
 637         do {
 638                 prepare_to_wait(&root->subv_writers->wait, &wait,
 639                                 TASK_UNINTERRUPTIBLE);
 640
 641                 writers = percpu_counter_sum(&root->subv_writers->counter);
 642                 if (writers)
 643                         schedule();
 644
 645                 finish_wait(&root->subv_writers->wait, &wait);
 646         } while (writers);
 647 }
 648
 649 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 650                            struct dentry *dentry, char *name, int namelen,
 651                            u64 *async_transid, bool readonly,
 652                            struct btrfs_qgroup_inherit *inherit)
 653 {
 654         struct inode *inode;
 655         struct btrfs_pending_snapshot *pending_snapshot;
 656         struct btrfs_trans_handle *trans;
 657         int ret;
 658
 659         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 660                 return -EINVAL;
 661
 662         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 663         if (!pending_snapshot)
 664                 return -ENOMEM;
 665
 666         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 667                         GFP_NOFS);
 668         pending_snapshot->path = btrfs_alloc_path();
 669         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 670                 ret = -ENOMEM;
 671                 goto free_pending;
 672         }
 673
 674         atomic_inc(&root->will_be_snapshoted);
 675         smp_mb__after_atomic();
 676         btrfs_wait_for_no_snapshoting_writes(root);
 677
 678         ret = btrfs_start_delalloc_inodes(root, 0);
 679         if (ret)
 680                 goto dec_and_free;
 681
 682         btrfs_wait_ordered_extents(root, -1);
 683
 684         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 685                              BTRFS_BLOCK_RSV_TEMP);
 686         /*
 687          * 1 - parent dir inode
 688          * 2 - dir entries
 689          * 1 - root item
 690          * 2 - root ref/backref
 691          * 1 - root of snapshot
 692          * 1 - UUID item
 693          */
 694         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 695                                         &pending_snapshot->block_rsv, 8,
 696                                         &pending_snapshot->qgroup_reserved,
 697                                         false);
 698         if (ret)
 699                 goto dec_and_free;
 700
 701         pending_snapshot->dentry = dentry;
 702         pending_snapshot->root = root;
 703         pending_snapshot->readonly = readonly;
 704         pending_snapshot->dir = dir;
 705         pending_snapshot->inherit = inherit;
 706
 707         trans = btrfs_start_transaction(root, 0);
 708         if (IS_ERR(trans)) {
 709                 ret = PTR_ERR(trans);
 710                 goto fail;
 711         }
 712
 713         spin_lock(&root->fs_info->trans_lock);
 714         list_add(&pending_snapshot->list,
 715                  &trans->transaction->pending_snapshots);
 716         spin_unlock(&root->fs_info->trans_lock);
 717         if (async_transid) {
 718                 *async_transid = trans->transid;
 719                 ret = btrfs_commit_transaction_async(trans,
 720                                      root->fs_info->extent_root, 1);
 721                 if (ret)
 722                         ret = btrfs_commit_transaction(trans, root);
 723         } else {
 724                 ret = btrfs_commit_transaction(trans,
 725                                                root->fs_info->extent_root);
 726         }
 727         if (ret)
 728                 goto fail;
 729
 730         ret = pending_snapshot->error;
 731         if (ret)
 732                 goto fail;
 733
 734         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 735         if (ret)
 736                 goto fail;
 737
 738         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 739         if (IS_ERR(inode)) {
 740                 ret = PTR_ERR(inode);
 741                 goto fail;
 742         }
 743
 744         d_instantiate(dentry, inode);
 745         ret = 0;
 746 fail:
 747         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 748                                          &pending_snapshot->block_rsv,
 749                                          pending_snapshot->qgroup_reserved);
 750 dec_and_free:
 751         if (atomic_dec_and_test(&root->will_be_snapshoted))
 752                 wake_up_atomic_t(&root->will_be_snapshoted);
 753 free_pending:
 754         kfree(pending_snapshot->root_item);
 755         btrfs_free_path(pending_snapshot->path);
 756         kfree(pending_snapshot);
 757
 758         return ret;
 759 }
 760
 761 /*  copy of may_delete in fs/namei.c()
 762  *      Check whether we can remove a link victim from directory dir, check
 763  *  whether the type of victim is right.
 764  *  1. We can't do it if dir is read-only (done in permission())
 765  *  2. We should have write and exec permissions on dir
 766  *  3. We can't remove anything from append-only dir
 767  *  4. We can't do anything with immutable dir (done in permission())
 768  *  5. If the sticky bit on dir is set we should either
 769  *      a. be owner of dir, or
 770  *      b. be owner of victim, or
 771  *      c. have CAP_FOWNER capability
 772  *  6. If the victim is append-only or immutable we can't do antyhing with
 773  *     links pointing to it.
 774  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 775  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 776  *  9. We can't remove a root or mountpoint.
 777  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 778  *     nfs_async_unlink().
 779  */
 780
 781 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 782 {
 783         int error;
 784
 785         if (d_really_is_negative(victim))
 786                 return -ENOENT;
 787
 788         BUG_ON(d_inode(victim->d_parent) != dir);
 789         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 790
 791         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 792         if (error)
 793                 return error;
 794         if (IS_APPEND(dir))
 795                 return -EPERM;
 796         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 797             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 798                 return -EPERM;
 799         if (isdir) {
 800                 if (!d_is_dir(victim))
 801                         return -ENOTDIR;
 802                 if (IS_ROOT(victim))
 803                         return -EBUSY;
 804         } else if (d_is_dir(victim))
 805                 return -EISDIR;
 806         if (IS_DEADDIR(dir))
 807                 return -ENOENT;
 808         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 809                 return -EBUSY;
 810         return 0;
 811 }
 812
 813 /* copy of may_create in fs/namei.c() */
 814 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 815 {
 816         if (d_really_is_positive(child))
 817                 return -EEXIST;
 818         if (IS_DEADDIR(dir))
 819                 return -ENOENT;
 820         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 821 }
 822
 823 /*
 824  * Create a new subvolume below @parent.  This is largely modeled after
 825  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 826  * inside this filesystem so it's quite a bit simpler.
 827  */
 828 static noinline int btrfs_mksubvol(struct path *parent,
 829                                    char *name, int namelen,
 830                                    struct btrfs_root *snap_src,
 831                                    u64 *async_transid, bool readonly,
 832                                    struct btrfs_qgroup_inherit *inherit)
 833 {
 834         struct inode *dir  = d_inode(parent->dentry);
 835         struct dentry *dentry;
 836         int error;
 837
 838         error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
 839         if (error == -EINTR)
 840                 return error;
 841
 842         dentry = lookup_one_len(name, parent->dentry, namelen);
 843         error = PTR_ERR(dentry);
 844         if (IS_ERR(dentry))
 845                 goto out_unlock;
 846
 847         error = -EEXIST;
 848         if (d_really_is_positive(dentry))
 849                 goto out_dput;
 850
 851         error = btrfs_may_create(dir, dentry);
 852         if (error)
 853                 goto out_dput;
 854
 855         /*
 856          * even if this name doesn't exist, we may get hash collisions.
 857          * check for them now when we can safely fail
 858          */
 859         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 860                                                dir->i_ino, name,
 861                                                namelen);
 862         if (error)
 863                 goto out_dput;
 864
 865         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 866
 867         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 868                 goto out_up_read;
 869
 870         if (snap_src) {
 871                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 872                                         async_transid, readonly, inherit);
 873         } else {
 874                 error = create_subvol(dir, dentry, name, namelen,
 875                                       async_transid, inherit);
 876         }
 877         if (!error)
 878                 fsnotify_mkdir(dir, dentry);
 879 out_up_read:
 880         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 881 out_dput:
 882         dput(dentry);
 883 out_unlock:
 884         mutex_unlock(&dir->i_mutex);
 885         return error;
 886 }
 887
 888 /*
 889  * When we're defragging a range, we don't want to kick it off again
 890  * if it is really just waiting for delalloc to send it down.
 891  * If we find a nice big extent or delalloc range for the bytes in the
 892  * file you want to defrag, we return 0 to let you know to skip this
 893  * part of the file
 894  */
 895 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 896 {
 897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 898         struct extent_map *em = NULL;
 899         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 900         u64 end;
 901
 902         read_lock(&em_tree->lock);
 903         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 904         read_unlock(&em_tree->lock);
 905
 906         if (em) {
 907                 end = extent_map_end(em);
 908                 free_extent_map(em);
 909                 if (end - offset > thresh)
 910                         return 0;
 911         }
 912         /* if we already have a nice delalloc here, just stop */
 913         thresh /= 2;
 914         end = count_range_bits(io_tree, &offset, offset + thresh,
 915                                thresh, EXTENT_DELALLOC, 1);
 916         if (end >= thresh)
 917                 return 0;
 918         return 1;
 919 }
 920
 921 /*
 922  * helper function to walk through a file and find extents
 923  * newer than a specific transid, and smaller than thresh.
 924  *
 925  * This is used by the defragging code to find new and small
 926  * extents
 927  */
 928 static int find_new_extents(struct btrfs_root *root,
 929                             struct inode *inode, u64 newer_than,
 930                             u64 *off, u32 thresh)
 931 {
 932         struct btrfs_path *path;
 933         struct btrfs_key min_key;
 934         struct extent_buffer *leaf;
 935         struct btrfs_file_extent_item *extent;
 936         int type;
 937         int ret;
 938         u64 ino = btrfs_ino(inode);
 939
 940         path = btrfs_alloc_path();
 941         if (!path)
 942                 return -ENOMEM;
 943
 944         min_key.objectid = ino;
 945         min_key.type = BTRFS_EXTENT_DATA_KEY;
 946         min_key.offset = *off;
 947
 948         while (1) {
 949                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 950                 if (ret != 0)
 951                         goto none;
 952 process_slot:
 953                 if (min_key.objectid != ino)
 954                         goto none;
 955                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 956                         goto none;
 957
 958                 leaf = path->nodes[0];
 959                 extent = btrfs_item_ptr(leaf, path->slots[0],
 960                                         struct btrfs_file_extent_item);
 961
 962                 type = btrfs_file_extent_type(leaf, extent);
 963                 if (type == BTRFS_FILE_EXTENT_REG &&
 964                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 965                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 966                         *off = min_key.offset;
 967                         btrfs_free_path(path);
 968                         return 0;
 969                 }
 970
 971                 path->slots[0]++;
 972                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 973                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 974                         goto process_slot;
 975                 }
 976
 977                 if (min_key.offset == (u64)-1)
 978                         goto none;
 979
 980                 min_key.offset++;
 981                 btrfs_release_path(path);
 982         }
 983 none:
 984         btrfs_free_path(path);
 985         return -ENOENT;
 986 }
 987
 988 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 989 {
 990         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 991         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 992         struct extent_map *em;
 993         u64 len = PAGE_CACHE_SIZE;
 994
 995         /*
 996          * hopefully we have this extent in the tree already, try without
 997          * the full extent lock
 998          */
 999         read_lock(&em_tree->lock);
1000         em = lookup_extent_mapping(em_tree, start, len);
1001         read_unlock(&em_tree->lock);
1002
1003         if (!em) {
1004                 struct extent_state *cached = NULL;
1005                 u64 end = start + len - 1;
1006
1007                 /* get the big lock and read metadata off disk */
1008                 lock_extent_bits(io_tree, start, end, &cached);
1009                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1010                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1011
1012                 if (IS_ERR(em))
1013                         return NULL;
1014         }
1015
1016         return em;
1017 }
1018
1019 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1020 {
1021         struct extent_map *next;
1022         bool ret = true;
1023
1024         /* this is the last extent */
1025         if (em->start + em->len >= i_size_read(inode))
1026                 return false;
1027
1028         next = defrag_lookup_extent(inode, em->start + em->len);
1029         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1030                 ret = false;
1031         else if ((em->block_start + em->block_len == next->block_start) &&
1032                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1033                 ret = false;
1034
1035         free_extent_map(next);
1036         return ret;
1037 }
1038
1039 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1040                                u64 *last_len, u64 *skip, u64 *defrag_end,
1041                                int compress)
1042 {
1043         struct extent_map *em;
1044         int ret = 1;
1045         bool next_mergeable = true;
1046         bool prev_mergeable = true;
1047
1048         /*
1049          * make sure that once we start defragging an extent, we keep on
1050          * defragging it
1051          */
1052         if (start < *defrag_end)
1053                 return 1;
1054
1055         *skip = 0;
1056
1057         em = defrag_lookup_extent(inode, start);
1058         if (!em)
1059                 return 0;
1060
1061         /* this will cover holes, and inline extents */
1062         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1063                 ret = 0;
1064                 goto out;
1065         }
1066
1067         if (!*defrag_end)
1068                 prev_mergeable = false;
1069
1070         next_mergeable = defrag_check_next_extent(inode, em);
1071         /*
1072          * we hit a real extent, if it is big or the next extent is not a
1073          * real extent, don't bother defragging it
1074          */
1075         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1076             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1077                 ret = 0;
1078 out:
1079         /*
1080          * last_len ends up being a counter of how many bytes we've defragged.
1081          * every time we choose not to defrag an extent, we reset *last_len
1082          * so that the next tiny extent will force a defrag.
1083          *
1084          * The end result of this is that tiny extents before a single big
1085          * extent will force at least part of that big extent to be defragged.
1086          */
1087         if (ret) {
1088                 *defrag_end = extent_map_end(em);
1089         } else {
1090                 *last_len = 0;
1091                 *skip = extent_map_end(em);
1092                 *defrag_end = 0;
1093         }
1094
1095         free_extent_map(em);
1096         return ret;
1097 }
1098
1099 /*
1100  * it doesn't do much good to defrag one or two pages
1101  * at a time.  This pulls in a nice chunk of pages
1102  * to COW and defrag.
1103  *
1104  * It also makes sure the delalloc code has enough
1105  * dirty data to avoid making new small extents as part
1106  * of the defrag
1107  *
1108  * It's a good idea to start RA on this range
1109  * before calling this.
1110  */
1111 static int cluster_pages_for_defrag(struct inode *inode,
1112                                     struct page **pages,
1113                                     unsigned long start_index,
1114                                     unsigned long num_pages)
1115 {
1116         unsigned long file_end;
1117         u64 isize = i_size_read(inode);
1118         u64 page_start;
1119         u64 page_end;
1120         u64 page_cnt;
1121         int ret;
1122         int i;
1123         int i_done;
1124         struct btrfs_ordered_extent *ordered;
1125         struct extent_state *cached_state = NULL;
1126         struct extent_io_tree *tree;
1127         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1128
1129         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
1130         if (!isize || start_index > file_end)
1131                 return 0;
1132
1133         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1134
1135         ret = btrfs_delalloc_reserve_space(inode,
1136                         start_index << PAGE_CACHE_SHIFT,
1137                         page_cnt << PAGE_CACHE_SHIFT);
1138         if (ret)
1139                 return ret;
1140         i_done = 0;
1141         tree = &BTRFS_I(inode)->io_tree;
1142
1143         /* step one, lock all the pages */
1144         for (i = 0; i < page_cnt; i++) {
1145                 struct page *page;
1146 again:
1147                 page = find_or_create_page(inode->i_mapping,
1148                                            start_index + i, mask);
1149                 if (!page)
1150                         break;
1151
1152                 page_start = page_offset(page);
1153                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1154                 while (1) {
1155                         lock_extent_bits(tree, page_start, page_end,
1156                                          &cached_state);
1157                         ordered = btrfs_lookup_ordered_extent(inode,
1158                                                               page_start);
1159                         unlock_extent_cached(tree, page_start, page_end,
1160                                              &cached_state, GFP_NOFS);
1161                         if (!ordered)
1162                                 break;
1163
1164                         unlock_page(page);
1165                         btrfs_start_ordered_extent(inode, ordered, 1);
1166                         btrfs_put_ordered_extent(ordered);
1167                         lock_page(page);
1168                         /*
1169                          * we unlocked the page above, so we need check if
1170                          * it was released or not.
1171                          */
1172                         if (page->mapping != inode->i_mapping) {
1173                                 unlock_page(page);
1174                                 page_cache_release(page);
1175                                 goto again;
1176                         }
1177                 }
1178
1179                 if (!PageUptodate(page)) {
1180                         btrfs_readpage(NULL, page);
1181                         lock_page(page);
1182                         if (!PageUptodate(page)) {
1183                                 unlock_page(page);
1184                                 page_cache_release(page);
1185                                 ret = -EIO;
1186                                 break;
1187                         }
1188                 }
1189
1190                 if (page->mapping != inode->i_mapping) {
1191                         unlock_page(page);
1192                         page_cache_release(page);
1193                         goto again;
1194                 }
1195
1196                 pages[i] = page;
1197                 i_done++;
1198         }
1199         if (!i_done || ret)
1200                 goto out;
1201
1202         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1203                 goto out;
1204
1205         /*
1206          * so now we have a nice long stream of locked
1207          * and up to date pages, lets wait on them
1208          */
1209         for (i = 0; i < i_done; i++)
1210                 wait_on_page_writeback(pages[i]);
1211
1212         page_start = page_offset(pages[0]);
1213         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1214
1215         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1216                          page_start, page_end - 1, &cached_state);
1217         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1218                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1219                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1220                           &cached_state, GFP_NOFS);
1221
1222         if (i_done != page_cnt) {
1223                 spin_lock(&BTRFS_I(inode)->lock);
1224                 BTRFS_I(inode)->outstanding_extents++;
1225                 spin_unlock(&BTRFS_I(inode)->lock);
1226                 btrfs_delalloc_release_space(inode,
1227                                 start_index << PAGE_CACHE_SHIFT,
1228                                 (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1229         }
1230
1231
1232         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1233                           &cached_state, GFP_NOFS);
1234
1235         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1236                              page_start, page_end - 1, &cached_state,
1237                              GFP_NOFS);
1238
1239         for (i = 0; i < i_done; i++) {
1240                 clear_page_dirty_for_io(pages[i]);
1241                 ClearPageChecked(pages[i]);
1242                 set_page_extent_mapped(pages[i]);
1243                 set_page_dirty(pages[i]);
1244                 unlock_page(pages[i]);
1245                 page_cache_release(pages[i]);
1246         }
1247         return i_done;
1248 out:
1249         for (i = 0; i < i_done; i++) {
1250                 unlock_page(pages[i]);
1251                 page_cache_release(pages[i]);
1252         }
1253         btrfs_delalloc_release_space(inode,
1254                         start_index << PAGE_CACHE_SHIFT,
1255                         page_cnt << PAGE_CACHE_SHIFT);
1256         return ret;
1257
1258 }
1259
1260 int btrfs_defrag_file(struct inode *inode, struct file *file,
1261                       struct btrfs_ioctl_defrag_range_args *range,
1262                       u64 newer_than, unsigned long max_to_defrag)
1263 {
1264         struct btrfs_root *root = BTRFS_I(inode)->root;
1265         struct file_ra_state *ra = NULL;
1266         unsigned long last_index;
1267         u64 isize = i_size_read(inode);
1268         u64 last_len = 0;
1269         u64 skip = 0;
1270         u64 defrag_end = 0;
1271         u64 newer_off = range->start;
1272         unsigned long i;
1273         unsigned long ra_index = 0;
1274         int ret;
1275         int defrag_count = 0;
1276         int compress_type = BTRFS_COMPRESS_ZLIB;
1277         u32 extent_thresh = range->extent_thresh;
1278         unsigned long max_cluster = SZ_256K >> PAGE_CACHE_SHIFT;
1279         unsigned long cluster = max_cluster;
1280         u64 new_align = ~((u64)SZ_128K - 1);
1281         struct page **pages = NULL;
1282
1283         if (isize == 0)
1284                 return 0;
1285
1286         if (range->start >= isize)
1287                 return -EINVAL;
1288
1289         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1290                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1291                         return -EINVAL;
1292                 if (range->compress_type)
1293                         compress_type = range->compress_type;
1294         }
1295
1296         if (extent_thresh == 0)
1297                 extent_thresh = SZ_256K;
1298
1299         /*
1300          * if we were not given a file, allocate a readahead
1301          * context
1302          */
1303         if (!file) {
1304                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1305                 if (!ra)
1306                         return -ENOMEM;
1307                 file_ra_state_init(ra, inode->i_mapping);
1308         } else {
1309                 ra = &file->f_ra;
1310         }
1311
1312         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1313                         GFP_NOFS);
1314         if (!pages) {
1315                 ret = -ENOMEM;
1316                 goto out_ra;
1317         }
1318
1319         /* find the last page to defrag */
1320         if (range->start + range->len > range->start) {
1321                 last_index = min_t(u64, isize - 1,
1322                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1323         } else {
1324                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1325         }
1326
1327         if (newer_than) {
1328                 ret = find_new_extents(root, inode, newer_than,
1329                                        &newer_off, SZ_64K);
1330                 if (!ret) {
1331                         range->start = newer_off;
1332                         /*
1333                          * we always align our defrag to help keep
1334                          * the extents in the file evenly spaced
1335                          */
1336                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1337                 } else
1338                         goto out_ra;
1339         } else {
1340                 i = range->start >> PAGE_CACHE_SHIFT;
1341         }
1342         if (!max_to_defrag)
1343                 max_to_defrag = last_index - i + 1;
1344
1345         /*
1346          * make writeback starts from i, so the defrag range can be
1347          * written sequentially.
1348          */
1349         if (i < inode->i_mapping->writeback_index)
1350                 inode->i_mapping->writeback_index = i;
1351
1352         while (i <= last_index && defrag_count < max_to_defrag &&
1353                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE))) {
1354                 /*
1355                  * make sure we stop running if someone unmounts
1356                  * the FS
1357                  */
1358                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1359                         break;
1360
1361                 if (btrfs_defrag_cancelled(root->fs_info)) {
1362                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1363                         ret = -EAGAIN;
1364                         break;
1365                 }
1366
1367                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1368                                          extent_thresh, &last_len, &skip,
1369                                          &defrag_end, range->flags &
1370                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1371                         unsigned long next;
1372                         /*
1373                          * the should_defrag function tells us how much to skip
1374                          * bump our counter by the suggested amount
1375                          */
1376                         next = DIV_ROUND_UP(skip, PAGE_CACHE_SIZE);
1377                         i = max(i + 1, next);
1378                         continue;
1379                 }
1380
1381                 if (!newer_than) {
1382                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1383                                    PAGE_CACHE_SHIFT) - i;
1384                         cluster = min(cluster, max_cluster);
1385                 } else {
1386                         cluster = max_cluster;
1387                 }
1388
1389                 if (i + cluster > ra_index) {
1390                         ra_index = max(i, ra_index);
1391                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1392                                        cluster);
1393                         ra_index += cluster;
1394                 }
1395
1396                 mutex_lock(&inode->i_mutex);
1397                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1398                         BTRFS_I(inode)->force_compress = compress_type;
1399                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1400                 if (ret < 0) {
1401                         mutex_unlock(&inode->i_mutex);
1402                         goto out_ra;
1403                 }
1404
1405                 defrag_count += ret;
1406                 balance_dirty_pages_ratelimited(inode->i_mapping);
1407                 mutex_unlock(&inode->i_mutex);
1408
1409                 if (newer_than) {
1410                         if (newer_off == (u64)-1)
1411                                 break;
1412
1413                         if (ret > 0)
1414                                 i += ret;
1415
1416                         newer_off = max(newer_off + 1,
1417                                         (u64)i << PAGE_CACHE_SHIFT);
1418
1419                         ret = find_new_extents(root, inode, newer_than,
1420                                                &newer_off, SZ_64K);
1421                         if (!ret) {
1422                                 range->start = newer_off;
1423                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1424                         } else {
1425                                 break;
1426                         }
1427                 } else {
1428                         if (ret > 0) {
1429                                 i += ret;
1430                                 last_len += ret << PAGE_CACHE_SHIFT;
1431                         } else {
1432                                 i++;
1433                                 last_len = 0;
1434                         }
1435                 }
1436         }
1437
1438         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1439                 filemap_flush(inode->i_mapping);
1440                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1441                              &BTRFS_I(inode)->runtime_flags))
1442                         filemap_flush(inode->i_mapping);
1443         }
1444
1445         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1446                 /* the filemap_flush will queue IO into the worker threads, but
1447                  * we have to make sure the IO is actually started and that
1448                  * ordered extents get created before we return
1449                  */
1450                 atomic_inc(&root->fs_info->async_submit_draining);
1451                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1452                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1453                         wait_event(root->fs_info->async_submit_wait,
1454                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1455                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1456                 }
1457                 atomic_dec(&root->fs_info->async_submit_draining);
1458         }
1459
1460         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1461                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1462                 btrfs_sysfs_feature_update(root->fs_info,
1463                         BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO, FEAT_INCOMPAT);
1464         }
1465
1466         ret = defrag_count;
1467
1468 out_ra:
1469         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1470                 mutex_lock(&inode->i_mutex);
1471                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1472                 mutex_unlock(&inode->i_mutex);
1473         }
1474         if (!file)
1475                 kfree(ra);
1476         kfree(pages);
1477         return ret;
1478 }
1479
1480 static noinline int btrfs_ioctl_resize(struct file *file,
1481                                         void __user *arg)
1482 {
1483         u64 new_size;
1484         u64 old_size;
1485         u64 devid = 1;
1486         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1487         struct btrfs_ioctl_vol_args *vol_args;
1488         struct btrfs_trans_handle *trans;
1489         struct btrfs_device *device = NULL;
1490         char *sizestr;
1491         char *retptr;
1492         char *devstr = NULL;
1493         int ret = 0;
1494         int mod = 0;
1495
1496         if (!capable(CAP_SYS_ADMIN))
1497                 return -EPERM;
1498
1499         ret = mnt_want_write_file(file);
1500         if (ret)
1501                 return ret;
1502
1503         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1504                         1)) {
1505                 mnt_drop_write_file(file);
1506                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1507         }
1508
1509         mutex_lock(&root->fs_info->volume_mutex);
1510         vol_args = memdup_user(arg, sizeof(*vol_args));
1511         if (IS_ERR(vol_args)) {
1512                 ret = PTR_ERR(vol_args);
1513                 goto out;
1514         }
1515
1516         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1517
1518         sizestr = vol_args->name;
1519         devstr = strchr(sizestr, ':');
1520         if (devstr) {
1521                 sizestr = devstr + 1;
1522                 *devstr = '\0';
1523                 devstr = vol_args->name;
1524                 ret = kstrtoull(devstr, 10, &devid);
1525                 if (ret)
1526                         goto out_free;
1527                 if (!devid) {
1528                         ret = -EINVAL;
1529                         goto out_free;
1530                 }
1531                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1532         }
1533
1534         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1535         if (!device) {
1536                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1537                        devid);
1538                 ret = -ENODEV;
1539                 goto out_free;
1540         }
1541
1542         if (!device->writeable) {
1543                 btrfs_info(root->fs_info,
1544                            "resizer unable to apply on readonly device %llu",
1545                        devid);
1546                 ret = -EPERM;
1547                 goto out_free;
1548         }
1549
1550         if (!strcmp(sizestr, "max"))
1551                 new_size = device->bdev->bd_inode->i_size;
1552         else {
1553                 if (sizestr[0] == '-') {
1554                         mod = -1;
1555                         sizestr++;
1556                 } else if (sizestr[0] == '+') {
1557                         mod = 1;
1558                         sizestr++;
1559                 }
1560                 new_size = memparse(sizestr, &retptr);
1561                 if (*retptr != '\0' || new_size == 0) {
1562                         ret = -EINVAL;
1563                         goto out_free;
1564                 }
1565         }
1566
1567         if (device->is_tgtdev_for_dev_replace) {
1568                 ret = -EPERM;
1569                 goto out_free;
1570         }
1571
1572         old_size = btrfs_device_get_total_bytes(device);
1573
1574         if (mod < 0) {
1575                 if (new_size > old_size) {
1576                         ret = -EINVAL;
1577                         goto out_free;
1578                 }
1579                 new_size = old_size - new_size;
1580         } else if (mod > 0) {
1581                 if (new_size > ULLONG_MAX - old_size) {
1582                         ret = -ERANGE;
1583                         goto out_free;
1584                 }
1585                 new_size = old_size + new_size;
1586         }
1587
1588         if (new_size < SZ_256M) {
1589                 ret = -EINVAL;
1590                 goto out_free;
1591         }
1592         if (new_size > device->bdev->bd_inode->i_size) {
1593                 ret = -EFBIG;
1594                 goto out_free;
1595         }
1596
1597         new_size = div_u64(new_size, root->sectorsize);
1598         new_size *= root->sectorsize;
1599
1600         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1601                       rcu_str_deref(device->name), new_size);
1602
1603         if (new_size > old_size) {
1604                 trans = btrfs_start_transaction(root, 0);
1605                 if (IS_ERR(trans)) {
1606                         ret = PTR_ERR(trans);
1607                         goto out_free;
1608                 }
1609                 ret = btrfs_grow_device(trans, device, new_size);
1610                 btrfs_commit_transaction(trans, root);
1611         } else if (new_size < old_size) {
1612                 ret = btrfs_shrink_device(device, new_size);
1613         } /* equal, nothing need to do */
1614
1615 out_free:
1616         kfree(vol_args);
1617 out:
1618         mutex_unlock(&root->fs_info->volume_mutex);
1619         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1620         mnt_drop_write_file(file);
1621         return ret;
1622 }
1623
1624 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1625                                 char *name, unsigned long fd, int subvol,
1626                                 u64 *transid, bool readonly,
1627                                 struct btrfs_qgroup_inherit *inherit)
1628 {
1629         int namelen;
1630         int ret = 0;
1631
1632         ret = mnt_want_write_file(file);
1633         if (ret)
1634                 goto out;
1635
1636         namelen = strlen(name);
1637         if (strchr(name, '/')) {
1638                 ret = -EINVAL;
1639                 goto out_drop_write;
1640         }
1641
1642         if (name[0] == '.' &&
1643            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1644                 ret = -EEXIST;
1645                 goto out_drop_write;
1646         }
1647
1648         if (subvol) {
1649                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1650                                      NULL, transid, readonly, inherit);
1651         } else {
1652                 struct fd src = fdget(fd);
1653                 struct inode *src_inode;
1654                 if (!src.file) {
1655                         ret = -EINVAL;
1656                         goto out_drop_write;
1657                 }
1658
1659                 src_inode = file_inode(src.file);
1660                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1661                         btrfs_info(BTRFS_I(src_inode)->root->fs_info,
1662                                    "Snapshot src from another FS");
1663                         ret = -EXDEV;
1664                 } else if (!inode_owner_or_capable(src_inode)) {
1665                         /*
1666                          * Subvolume creation is not restricted, but snapshots
1667                          * are limited to own subvolumes only
1668                          */
1669                         ret = -EPERM;
1670                 } else {
1671                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1672                                              BTRFS_I(src_inode)->root,
1673                                              transid, readonly, inherit);
1674                 }
1675                 fdput(src);
1676         }
1677 out_drop_write:
1678         mnt_drop_write_file(file);
1679 out:
1680         return ret;
1681 }
1682
1683 static noinline int btrfs_ioctl_snap_create(struct file *file,
1684                                             void __user *arg, int subvol)
1685 {
1686         struct btrfs_ioctl_vol_args *vol_args;
1687         int ret;
1688
1689         vol_args = memdup_user(arg, sizeof(*vol_args));
1690         if (IS_ERR(vol_args))
1691                 return PTR_ERR(vol_args);
1692         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1693
1694         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1695                                               vol_args->fd, subvol,
1696                                               NULL, false, NULL);
1697
1698         kfree(vol_args);
1699         return ret;
1700 }
1701
1702 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1703                                                void __user *arg, int subvol)
1704 {
1705         struct btrfs_ioctl_vol_args_v2 *vol_args;
1706         int ret;
1707         u64 transid = 0;
1708         u64 *ptr = NULL;
1709         bool readonly = false;
1710         struct btrfs_qgroup_inherit *inherit = NULL;
1711
1712         vol_args = memdup_user(arg, sizeof(*vol_args));
1713         if (IS_ERR(vol_args))
1714                 return PTR_ERR(vol_args);
1715         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1716
1717         if (vol_args->flags &
1718             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1719               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1720                 ret = -EOPNOTSUPP;
1721                 goto free_args;
1722         }
1723
1724         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1725                 ptr = &transid;
1726         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1727                 readonly = true;
1728         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1729                 if (vol_args->size > PAGE_CACHE_SIZE) {
1730                         ret = -EINVAL;
1731                         goto free_args;
1732                 }
1733                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1734                 if (IS_ERR(inherit)) {
1735                         ret = PTR_ERR(inherit);
1736                         goto free_args;
1737                 }
1738         }
1739
1740         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1741                                               vol_args->fd, subvol, ptr,
1742                                               readonly, inherit);
1743         if (ret)
1744                 goto free_inherit;
1745
1746         if (ptr && copy_to_user(arg +
1747                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1748                                         transid),
1749                                 ptr, sizeof(*ptr)))
1750                 ret = -EFAULT;
1751
1752 free_inherit:
1753         kfree(inherit);
1754 free_args:
1755         kfree(vol_args);
1756         return ret;
1757 }
1758
1759 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1760                                                 void __user *arg)
1761 {
1762         struct inode *inode = file_inode(file);
1763         struct btrfs_root *root = BTRFS_I(inode)->root;
1764         int ret = 0;
1765         u64 flags = 0;
1766
1767         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1768                 return -EINVAL;
1769
1770         down_read(&root->fs_info->subvol_sem);
1771         if (btrfs_root_readonly(root))
1772                 flags |= BTRFS_SUBVOL_RDONLY;
1773         up_read(&root->fs_info->subvol_sem);
1774
1775         if (copy_to_user(arg, &flags, sizeof(flags)))
1776                 ret = -EFAULT;
1777
1778         return ret;
1779 }
1780
1781 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1782                                               void __user *arg)
1783 {
1784         struct inode *inode = file_inode(file);
1785         struct btrfs_root *root = BTRFS_I(inode)->root;
1786         struct btrfs_trans_handle *trans;
1787         u64 root_flags;
1788         u64 flags;
1789         int ret = 0;
1790
1791         if (!inode_owner_or_capable(inode))
1792                 return -EPERM;
1793
1794         ret = mnt_want_write_file(file);
1795         if (ret)
1796                 goto out;
1797
1798         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1799                 ret = -EINVAL;
1800                 goto out_drop_write;
1801         }
1802
1803         if (copy_from_user(&flags, arg, sizeof(flags))) {
1804                 ret = -EFAULT;
1805                 goto out_drop_write;
1806         }
1807
1808         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1809                 ret = -EINVAL;
1810                 goto out_drop_write;
1811         }
1812
1813         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1814                 ret = -EOPNOTSUPP;
1815                 goto out_drop_write;
1816         }
1817
1818         down_write(&root->fs_info->subvol_sem);
1819
1820         /* nothing to do */
1821         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1822                 goto out_drop_sem;
1823
1824         root_flags = btrfs_root_flags(&root->root_item);
1825         if (flags & BTRFS_SUBVOL_RDONLY) {
1826                 btrfs_set_root_flags(&root->root_item,
1827                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1828         } else {
1829                 /*
1830                  * Block RO -> RW transition if this subvolume is involved in
1831                  * send
1832                  */
1833                 spin_lock(&root->root_item_lock);
1834                 if (root->send_in_progress == 0) {
1835                         btrfs_set_root_flags(&root->root_item,
1836                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1837                         spin_unlock(&root->root_item_lock);
1838                 } else {
1839                         spin_unlock(&root->root_item_lock);
1840                         btrfs_warn(root->fs_info,
1841                         "Attempt to set subvolume %llu read-write during send",
1842                                         root->root_key.objectid);
1843                         ret = -EPERM;
1844                         goto out_drop_sem;
1845                 }
1846         }
1847
1848         trans = btrfs_start_transaction(root, 1);
1849         if (IS_ERR(trans)) {
1850                 ret = PTR_ERR(trans);
1851                 goto out_reset;
1852         }
1853
1854         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1855                                 &root->root_key, &root->root_item);
1856
1857         btrfs_commit_transaction(trans, root);
1858 out_reset:
1859         if (ret)
1860                 btrfs_set_root_flags(&root->root_item, root_flags);
1861 out_drop_sem:
1862         up_write(&root->fs_info->subvol_sem);
1863 out_drop_write:
1864         mnt_drop_write_file(file);
1865 out:
1866         return ret;
1867 }
1868
1869 /*
1870  * helper to check if the subvolume references other subvolumes
1871  */
1872 static noinline int may_destroy_subvol(struct btrfs_root *root)
1873 {
1874         struct btrfs_path *path;
1875         struct btrfs_dir_item *di;
1876         struct btrfs_key key;
1877         u64 dir_id;
1878         int ret;
1879
1880         path = btrfs_alloc_path();
1881         if (!path)
1882                 return -ENOMEM;
1883
1884         /* Make sure this root isn't set as the default subvol */
1885         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1886         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1887                                    dir_id, "default", 7, 0);
1888         if (di && !IS_ERR(di)) {
1889                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1890                 if (key.objectid == root->root_key.objectid) {
1891                         ret = -EPERM;
1892                         btrfs_err(root->fs_info, "deleting default subvolume "
1893                                   "%llu is not allowed", key.objectid);
1894                         goto out;
1895                 }
1896                 btrfs_release_path(path);
1897         }
1898
1899         key.objectid = root->root_key.objectid;
1900         key.type = BTRFS_ROOT_REF_KEY;
1901         key.offset = (u64)-1;
1902
1903         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1904                                 &key, path, 0, 0);
1905         if (ret < 0)
1906                 goto out;
1907         BUG_ON(ret == 0);
1908
1909         ret = 0;
1910         if (path->slots[0] > 0) {
1911                 path->slots[0]--;
1912                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1913                 if (key.objectid == root->root_key.objectid &&
1914                     key.type == BTRFS_ROOT_REF_KEY)
1915                         ret = -ENOTEMPTY;
1916         }
1917 out:
1918         btrfs_free_path(path);
1919         return ret;
1920 }
1921
1922 static noinline int key_in_sk(struct btrfs_key *key,
1923                               struct btrfs_ioctl_search_key *sk)
1924 {
1925         struct btrfs_key test;
1926         int ret;
1927
1928         test.objectid = sk->min_objectid;
1929         test.type = sk->min_type;
1930         test.offset = sk->min_offset;
1931
1932         ret = btrfs_comp_cpu_keys(key, &test);
1933         if (ret < 0)
1934                 return 0;
1935
1936         test.objectid = sk->max_objectid;
1937         test.type = sk->max_type;
1938         test.offset = sk->max_offset;
1939
1940         ret = btrfs_comp_cpu_keys(key, &test);
1941         if (ret > 0)
1942                 return 0;
1943         return 1;
1944 }
1945
1946 static noinline int copy_to_sk(struct btrfs_root *root,
1947                                struct btrfs_path *path,
1948                                struct btrfs_key *key,
1949                                struct btrfs_ioctl_search_key *sk,
1950                                size_t *buf_size,
1951                                char __user *ubuf,
1952                                unsigned long *sk_offset,
1953                                int *num_found)
1954 {
1955         u64 found_transid;
1956         struct extent_buffer *leaf;
1957         struct btrfs_ioctl_search_header sh;
1958         struct btrfs_key test;
1959         unsigned long item_off;
1960         unsigned long item_len;
1961         int nritems;
1962         int i;
1963         int slot;
1964         int ret = 0;
1965
1966         leaf = path->nodes[0];
1967         slot = path->slots[0];
1968         nritems = btrfs_header_nritems(leaf);
1969
1970         if (btrfs_header_generation(leaf) > sk->max_transid) {
1971                 i = nritems;
1972                 goto advance_key;
1973         }
1974         found_transid = btrfs_header_generation(leaf);
1975
1976         for (i = slot; i < nritems; i++) {
1977                 item_off = btrfs_item_ptr_offset(leaf, i);
1978                 item_len = btrfs_item_size_nr(leaf, i);
1979
1980                 btrfs_item_key_to_cpu(leaf, key, i);
1981                 if (!key_in_sk(key, sk))
1982                         continue;
1983
1984                 if (sizeof(sh) + item_len > *buf_size) {
1985                         if (*num_found) {
1986                                 ret = 1;
1987                                 goto out;
1988                         }
1989
1990                         /*
1991                          * return one empty item back for v1, which does not
1992                          * handle -EOVERFLOW
1993                          */
1994
1995                         *buf_size = sizeof(sh) + item_len;
1996                         item_len = 0;
1997                         ret = -EOVERFLOW;
1998                 }
1999
2000                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2001                         ret = 1;
2002                         goto out;
2003                 }
2004
2005                 sh.objectid = key->objectid;
2006                 sh.offset = key->offset;
2007                 sh.type = key->type;
2008                 sh.len = item_len;
2009                 sh.transid = found_transid;
2010
2011                 /* copy search result header */
2012                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2013                         ret = -EFAULT;
2014                         goto out;
2015                 }
2016
2017                 *sk_offset += sizeof(sh);
2018
2019                 if (item_len) {
2020                         char __user *up = ubuf + *sk_offset;
2021                         /* copy the item */
2022                         if (read_extent_buffer_to_user(leaf, up,
2023                                                        item_off, item_len)) {
2024                                 ret = -EFAULT;
2025                                 goto out;
2026                         }
2027
2028                         *sk_offset += item_len;
2029                 }
2030                 (*num_found)++;
2031
2032                 if (ret) /* -EOVERFLOW from above */
2033                         goto out;
2034
2035                 if (*num_found >= sk->nr_items) {
2036                         ret = 1;
2037                         goto out;
2038                 }
2039         }
2040 advance_key:
2041         ret = 0;
2042         test.objectid = sk->max_objectid;
2043         test.type = sk->max_type;
2044         test.offset = sk->max_offset;
2045         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2046                 ret = 1;
2047         else if (key->offset < (u64)-1)
2048                 key->offset++;
2049         else if (key->type < (u8)-1) {
2050                 key->offset = 0;
2051                 key->type++;
2052         } else if (key->objectid < (u64)-1) {
2053                 key->offset = 0;
2054                 key->type = 0;
2055                 key->objectid++;
2056         } else
2057                 ret = 1;
2058 out:
2059         /*
2060          *  0: all items from this leaf copied, continue with next
2061          *  1: * more items can be copied, but unused buffer is too small
2062          *     * all items were found
2063          *     Either way, it will stops the loop which iterates to the next
2064          *     leaf
2065          *  -EOVERFLOW: item was to large for buffer
2066          *  -EFAULT: could not copy extent buffer back to userspace
2067          */
2068         return ret;
2069 }
2070
2071 static noinline int search_ioctl(struct inode *inode,
2072                                  struct btrfs_ioctl_search_key *sk,
2073                                  size_t *buf_size,
2074                                  char __user *ubuf)
2075 {
2076         struct btrfs_root *root;
2077         struct btrfs_key key;
2078         struct btrfs_path *path;
2079         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2080         int ret;
2081         int num_found = 0;
2082         unsigned long sk_offset = 0;
2083
2084         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2085                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2086                 return -EOVERFLOW;
2087         }
2088
2089         path = btrfs_alloc_path();
2090         if (!path)
2091                 return -ENOMEM;
2092
2093         if (sk->tree_id == 0) {
2094                 /* search the root of the inode that was passed */
2095                 root = BTRFS_I(inode)->root;
2096         } else {
2097                 key.objectid = sk->tree_id;
2098                 key.type = BTRFS_ROOT_ITEM_KEY;
2099                 key.offset = (u64)-1;
2100                 root = btrfs_read_fs_root_no_name(info, &key);
2101                 if (IS_ERR(root)) {
2102                         btrfs_err(info, "could not find root %llu",
2103                                sk->tree_id);
2104                         btrfs_free_path(path);
2105                         return -ENOENT;
2106                 }
2107         }
2108
2109         key.objectid = sk->min_objectid;
2110         key.type = sk->min_type;
2111         key.offset = sk->min_offset;
2112
2113         while (1) {
2114                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2115                 if (ret != 0) {
2116                         if (ret > 0)
2117                                 ret = 0;
2118                         goto err;
2119                 }
2120                 ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf,
2121                                  &sk_offset, &num_found);
2122                 btrfs_release_path(path);
2123                 if (ret)
2124                         break;
2125
2126         }
2127         if (ret > 0)
2128                 ret = 0;
2129 err:
2130         sk->nr_items = num_found;
2131         btrfs_free_path(path);
2132         return ret;
2133 }
2134
2135 static noinline int btrfs_ioctl_tree_search(struct file *file,
2136                                            void __user *argp)
2137 {
2138         struct btrfs_ioctl_search_args __user *uargs;
2139         struct btrfs_ioctl_search_key sk;
2140         struct inode *inode;
2141         int ret;
2142         size_t buf_size;
2143
2144         if (!capable(CAP_SYS_ADMIN))
2145                 return -EPERM;
2146
2147         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2148
2149         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2150                 return -EFAULT;
2151
2152         buf_size = sizeof(uargs->buf);
2153
2154         inode = file_inode(file);
2155         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2156
2157         /*
2158          * In the origin implementation an overflow is handled by returning a
2159          * search header with a len of zero, so reset ret.
2160          */
2161         if (ret == -EOVERFLOW)
2162                 ret = 0;
2163
2164         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2165                 ret = -EFAULT;
2166         return ret;
2167 }
2168
2169 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2170                                                void __user *argp)
2171 {
2172         struct btrfs_ioctl_search_args_v2 __user *uarg;
2173         struct btrfs_ioctl_search_args_v2 args;
2174         struct inode *inode;
2175         int ret;
2176         size_t buf_size;
2177         const size_t buf_limit = SZ_16M;
2178
2179         if (!capable(CAP_SYS_ADMIN))
2180                 return -EPERM;
2181
2182         /* copy search header and buffer size */
2183         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2184         if (copy_from_user(&args, uarg, sizeof(args)))
2185                 return -EFAULT;
2186
2187         buf_size = args.buf_size;
2188
2189         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2190                 return -EOVERFLOW;
2191
2192         /* limit result size to 16MB */
2193         if (buf_size > buf_limit)
2194                 buf_size = buf_limit;
2195
2196         inode = file_inode(file);
2197         ret = search_ioctl(inode, &args.key, &buf_size,
2198                            (char *)(&uarg->buf[0]));
2199         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2200                 ret = -EFAULT;
2201         else if (ret == -EOVERFLOW &&
2202                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2203                 ret = -EFAULT;
2204
2205         return ret;
2206 }
2207
2208 /*
2209  * Search INODE_REFs to identify path name of 'dirid' directory
2210  * in a 'tree_id' tree. and sets path name to 'name'.
2211  */
2212 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2213                                 u64 tree_id, u64 dirid, char *name)
2214 {
2215         struct btrfs_root *root;
2216         struct btrfs_key key;
2217         char *ptr;
2218         int ret = -1;
2219         int slot;
2220         int len;
2221         int total_len = 0;
2222         struct btrfs_inode_ref *iref;
2223         struct extent_buffer *l;
2224         struct btrfs_path *path;
2225
2226         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2227                 name[0]='\0';
2228                 return 0;
2229         }
2230
2231         path = btrfs_alloc_path();
2232         if (!path)
2233                 return -ENOMEM;
2234
2235         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2236
2237         key.objectid = tree_id;
2238         key.type = BTRFS_ROOT_ITEM_KEY;
2239         key.offset = (u64)-1;
2240         root = btrfs_read_fs_root_no_name(info, &key);
2241         if (IS_ERR(root)) {
2242                 btrfs_err(info, "could not find root %llu", tree_id);
2243                 ret = -ENOENT;
2244                 goto out;
2245         }
2246
2247         key.objectid = dirid;
2248         key.type = BTRFS_INODE_REF_KEY;
2249         key.offset = (u64)-1;
2250
2251         while (1) {
2252                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2253                 if (ret < 0)
2254                         goto out;
2255                 else if (ret > 0) {
2256                         ret = btrfs_previous_item(root, path, dirid,
2257                                                   BTRFS_INODE_REF_KEY);
2258                         if (ret < 0)
2259                                 goto out;
2260                         else if (ret > 0) {
2261                                 ret = -ENOENT;
2262                                 goto out;
2263                         }
2264                 }
2265
2266                 l = path->nodes[0];
2267                 slot = path->slots[0];
2268                 btrfs_item_key_to_cpu(l, &key, slot);
2269
2270                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2271                 len = btrfs_inode_ref_name_len(l, iref);
2272                 ptr -= len + 1;
2273                 total_len += len + 1;
2274                 if (ptr < name) {
2275                         ret = -ENAMETOOLONG;
2276                         goto out;
2277                 }
2278
2279                 *(ptr + len) = '/';
2280                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2281
2282                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2283                         break;
2284
2285                 btrfs_release_path(path);
2286                 key.objectid = key.offset;
2287                 key.offset = (u64)-1;
2288                 dirid = key.objectid;
2289         }
2290         memmove(name, ptr, total_len);
2291         name[total_len] = '\0';
2292         ret = 0;
2293 out:
2294         btrfs_free_path(path);
2295         return ret;
2296 }
2297
2298 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2299                                            void __user *argp)
2300 {
2301          struct btrfs_ioctl_ino_lookup_args *args;
2302          struct inode *inode;
2303         int ret = 0;
2304
2305         args = memdup_user(argp, sizeof(*args));
2306         if (IS_ERR(args))
2307                 return PTR_ERR(args);
2308
2309         inode = file_inode(file);
2310
2311         /*
2312          * Unprivileged query to obtain the containing subvolume root id. The
2313          * path is reset so it's consistent with btrfs_search_path_in_tree.
2314          */
2315         if (args->treeid == 0)
2316                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2317
2318         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2319                 args->name[0] = 0;
2320                 goto out;
2321         }
2322
2323         if (!capable(CAP_SYS_ADMIN)) {
2324                 ret = -EPERM;
2325                 goto out;
2326         }
2327
2328         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2329                                         args->treeid, args->objectid,
2330                                         args->name);
2331
2332 out:
2333         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2334                 ret = -EFAULT;
2335
2336         kfree(args);
2337         return ret;
2338 }
2339
2340 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2341                                              void __user *arg)
2342 {
2343         struct dentry *parent = file->f_path.dentry;
2344         struct dentry *dentry;
2345         struct inode *dir = d_inode(parent);
2346         struct inode *inode;
2347         struct btrfs_root *root = BTRFS_I(dir)->root;
2348         struct btrfs_root *dest = NULL;
2349         struct btrfs_ioctl_vol_args *vol_args;
2350         struct btrfs_trans_handle *trans;
2351         struct btrfs_block_rsv block_rsv;
2352         u64 root_flags;
2353         u64 qgroup_reserved;
2354         int namelen;
2355         int ret;
2356         int err = 0;
2357
2358         vol_args = memdup_user(arg, sizeof(*vol_args));
2359         if (IS_ERR(vol_args))
2360                 return PTR_ERR(vol_args);
2361
2362         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2363         namelen = strlen(vol_args->name);
2364         if (strchr(vol_args->name, '/') ||
2365             strncmp(vol_args->name, "..", namelen) == 0) {
2366                 err = -EINVAL;
2367                 goto out;
2368         }
2369
2370         err = mnt_want_write_file(file);
2371         if (err)
2372                 goto out;
2373
2374
2375         err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT);
2376         if (err == -EINTR)
2377                 goto out_drop_write;
2378         dentry = lookup_one_len(vol_args->name, parent, namelen);
2379         if (IS_ERR(dentry)) {
2380                 err = PTR_ERR(dentry);
2381                 goto out_unlock_dir;
2382         }
2383
2384         if (d_really_is_negative(dentry)) {
2385                 err = -ENOENT;
2386                 goto out_dput;
2387         }
2388
2389         inode = d_inode(dentry);
2390         dest = BTRFS_I(inode)->root;
2391         if (!capable(CAP_SYS_ADMIN)) {
2392                 /*
2393                  * Regular user.  Only allow this with a special mount
2394                  * option, when the user has write+exec access to the
2395                  * subvol root, and when rmdir(2) would have been
2396                  * allowed.
2397                  *
2398                  * Note that this is _not_ check that the subvol is
2399                  * empty or doesn't contain data that we wouldn't
2400                  * otherwise be able to delete.
2401                  *
2402                  * Users who want to delete empty subvols should try
2403                  * rmdir(2).
2404                  */
2405                 err = -EPERM;
2406                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2407                         goto out_dput;
2408
2409                 /*
2410                  * Do not allow deletion if the parent dir is the same
2411                  * as the dir to be deleted.  That means the ioctl
2412                  * must be called on the dentry referencing the root
2413                  * of the subvol, not a random directory contained
2414                  * within it.
2415                  */
2416                 err = -EINVAL;
2417                 if (root == dest)
2418                         goto out_dput;
2419
2420                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2421                 if (err)
2422                         goto out_dput;
2423         }
2424
2425         /* check if subvolume may be deleted by a user */
2426         err = btrfs_may_delete(dir, dentry, 1);
2427         if (err)
2428                 goto out_dput;
2429
2430         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2431                 err = -EINVAL;
2432                 goto out_dput;
2433         }
2434
2435         mutex_lock(&inode->i_mutex);
2436
2437         /*
2438          * Don't allow to delete a subvolume with send in progress. This is
2439          * inside the i_mutex so the error handling that has to drop the bit
2440          * again is not run concurrently.
2441          */
2442         spin_lock(&dest->root_item_lock);
2443         root_flags = btrfs_root_flags(&dest->root_item);
2444         if (dest->send_in_progress == 0) {
2445                 btrfs_set_root_flags(&dest->root_item,
2446                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2447                 spin_unlock(&dest->root_item_lock);
2448         } else {
2449                 spin_unlock(&dest->root_item_lock);
2450                 btrfs_warn(root->fs_info,
2451                         "Attempt to delete subvolume %llu during send",
2452                         dest->root_key.objectid);
2453                 err = -EPERM;
2454                 goto out_unlock_inode;
2455         }
2456
2457         down_write(&root->fs_info->subvol_sem);
2458
2459         err = may_destroy_subvol(dest);
2460         if (err)
2461                 goto out_up_write;
2462
2463         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2464         /*
2465          * One for dir inode, two for dir entries, two for root
2466          * ref/backref.
2467          */
2468         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2469                                                5, &qgroup_reserved, true);
2470         if (err)
2471                 goto out_up_write;
2472
2473         trans = btrfs_start_transaction(root, 0);
2474         if (IS_ERR(trans)) {
2475                 err = PTR_ERR(trans);
2476                 goto out_release;
2477         }
2478         trans->block_rsv = &block_rsv;
2479         trans->bytes_reserved = block_rsv.size;
2480
2481         ret = btrfs_unlink_subvol(trans, root, dir,
2482                                 dest->root_key.objectid,
2483                                 dentry->d_name.name,
2484                                 dentry->d_name.len);
2485         if (ret) {
2486                 err = ret;
2487                 btrfs_abort_transaction(trans, root, ret);
2488                 goto out_end_trans;
2489         }
2490
2491         btrfs_record_root_in_trans(trans, dest);
2492
2493         memset(&dest->root_item.drop_progress, 0,
2494                 sizeof(dest->root_item.drop_progress));
2495         dest->root_item.drop_level = 0;
2496         btrfs_set_root_refs(&dest->root_item, 0);
2497
2498         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2499                 ret = btrfs_insert_orphan_item(trans,
2500                                         root->fs_info->tree_root,
2501                                         dest->root_key.objectid);
2502                 if (ret) {
2503                         btrfs_abort_transaction(trans, root, ret);
2504                         err = ret;
2505                         goto out_end_trans;
2506                 }
2507         }
2508
2509         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2510                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2511                                   dest->root_key.objectid);
2512         if (ret && ret != -ENOENT) {
2513                 btrfs_abort_transaction(trans, root, ret);
2514                 err = ret;
2515                 goto out_end_trans;
2516         }
2517         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2518                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2519                                           dest->root_item.received_uuid,
2520                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2521                                           dest->root_key.objectid);
2522                 if (ret && ret != -ENOENT) {
2523                         btrfs_abort_transaction(trans, root, ret);
2524                         err = ret;
2525                         goto out_end_trans;
2526                 }
2527         }
2528
2529 out_end_trans:
2530         trans->block_rsv = NULL;
2531         trans->bytes_reserved = 0;
2532         ret = btrfs_end_transaction(trans, root);
2533         if (ret && !err)
2534                 err = ret;
2535         inode->i_flags |= S_DEAD;
2536 out_release:
2537         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2538 out_up_write:
2539         up_write(&root->fs_info->subvol_sem);
2540         if (err) {
2541                 spin_lock(&dest->root_item_lock);
2542                 root_flags = btrfs_root_flags(&dest->root_item);
2543                 btrfs_set_root_flags(&dest->root_item,
2544                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2545                 spin_unlock(&dest->root_item_lock);
2546         }
2547 out_unlock_inode:
2548         mutex_unlock(&inode->i_mutex);
2549         if (!err) {
2550                 d_invalidate(dentry);
2551                 btrfs_invalidate_inodes(dest);
2552                 d_delete(dentry);
2553                 ASSERT(dest->send_in_progress == 0);
2554
2555                 /* the last ref */
2556                 if (dest->ino_cache_inode) {
2557                         iput(dest->ino_cache_inode);
2558                         dest->ino_cache_inode = NULL;
2559                 }
2560         }
2561 out_dput:
2562         dput(dentry);
2563 out_unlock_dir:
2564         mutex_unlock(&dir->i_mutex);
2565 out_drop_write:
2566         mnt_drop_write_file(file);
2567 out:
2568         kfree(vol_args);
2569         return err;
2570 }
2571
2572 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2573 {
2574         struct inode *inode = file_inode(file);
2575         struct btrfs_root *root = BTRFS_I(inode)->root;
2576         struct btrfs_ioctl_defrag_range_args *range;
2577         int ret;
2578
2579         ret = mnt_want_write_file(file);
2580         if (ret)
2581                 return ret;
2582
2583         if (btrfs_root_readonly(root)) {
2584                 ret = -EROFS;
2585                 goto out;
2586         }
2587
2588         switch (inode->i_mode & S_IFMT) {
2589         case S_IFDIR:
2590                 if (!capable(CAP_SYS_ADMIN)) {
2591                         ret = -EPERM;
2592                         goto out;
2593                 }
2594                 ret = btrfs_defrag_root(root);
2595                 if (ret)
2596                         goto out;
2597                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2598                 break;
2599         case S_IFREG:
2600                 if (!(file->f_mode & FMODE_WRITE)) {
2601                         ret = -EINVAL;
2602                         goto out;
2603                 }
2604
2605                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2606                 if (!range) {
2607                         ret = -ENOMEM;
2608                         goto out;
2609                 }
2610
2611                 if (argp) {
2612                         if (copy_from_user(range, argp,
2613                                            sizeof(*range))) {
2614                                 ret = -EFAULT;
2615                                 kfree(range);
2616                                 goto out;
2617                         }
2618                         /* compression requires us to start the IO */
2619                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2620                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2621                                 range->extent_thresh = (u32)-1;
2622                         }
2623                 } else {
2624                         /* the rest are all set to zero by kzalloc */
2625                         range->len = (u64)-1;
2626                 }
2627                 ret = btrfs_defrag_file(file_inode(file), file,
2628                                         range, 0, 0);
2629                 if (ret > 0)
2630                         ret = 0;
2631                 kfree(range);
2632                 break;
2633         default:
2634                 ret = -EINVAL;
2635         }
2636 out:
2637         mnt_drop_write_file(file);
2638         return ret;
2639 }
2640
2641 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2642 {
2643         struct btrfs_ioctl_vol_args *vol_args;
2644         int ret;
2645
2646         if (!capable(CAP_SYS_ADMIN))
2647                 return -EPERM;
2648
2649         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2650                         1)) {
2651                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2652         }
2653
2654         mutex_lock(&root->fs_info->volume_mutex);
2655         vol_args = memdup_user(arg, sizeof(*vol_args));
2656         if (IS_ERR(vol_args)) {
2657                 ret = PTR_ERR(vol_args);
2658                 goto out;
2659         }
2660
2661         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2662         ret = btrfs_init_new_device(root, vol_args->name);
2663
2664         if (!ret)
2665                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2666
2667         kfree(vol_args);
2668 out:
2669         mutex_unlock(&root->fs_info->volume_mutex);
2670         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2671         return ret;
2672 }
2673
2674 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2675 {
2676         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2677         struct btrfs_ioctl_vol_args *vol_args;
2678         int ret;
2679
2680         if (!capable(CAP_SYS_ADMIN))
2681                 return -EPERM;
2682
2683         ret = mnt_want_write_file(file);
2684         if (ret)
2685                 return ret;
2686
2687         vol_args = memdup_user(arg, sizeof(*vol_args));
2688         if (IS_ERR(vol_args)) {
2689                 ret = PTR_ERR(vol_args);
2690                 goto err_drop;
2691         }
2692
2693         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2694
2695         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2696                         1)) {
2697                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2698                 goto out;
2699         }
2700
2701         mutex_lock(&root->fs_info->volume_mutex);
2702         ret = btrfs_rm_device(root, vol_args->name);
2703         mutex_unlock(&root->fs_info->volume_mutex);
2704         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2705
2706         if (!ret)
2707                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2708
2709 out:
2710         kfree(vol_args);
2711 err_drop:
2712         mnt_drop_write_file(file);
2713         return ret;
2714 }
2715
2716 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2717 {
2718         struct btrfs_ioctl_fs_info_args *fi_args;
2719         struct btrfs_device *device;
2720         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2721         int ret = 0;
2722
2723         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2724         if (!fi_args)
2725                 return -ENOMEM;
2726
2727         mutex_lock(&fs_devices->device_list_mutex);
2728         fi_args->num_devices = fs_devices->num_devices;
2729         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2730
2731         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2732                 if (device->devid > fi_args->max_id)
2733                         fi_args->max_id = device->devid;
2734         }
2735         mutex_unlock(&fs_devices->device_list_mutex);
2736
2737         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2738         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2739         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2740
2741         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2742                 ret = -EFAULT;
2743
2744         kfree(fi_args);
2745         return ret;
2746 }
2747
2748 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2749 {
2750         struct btrfs_ioctl_dev_info_args *di_args;
2751         struct btrfs_device *dev;
2752         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2753         int ret = 0;
2754         char *s_uuid = NULL;
2755
2756         di_args = memdup_user(arg, sizeof(*di_args));
2757         if (IS_ERR(di_args))
2758                 return PTR_ERR(di_args);
2759
2760         if (!btrfs_is_empty_uuid(di_args->uuid))
2761                 s_uuid = di_args->uuid;
2762
2763         mutex_lock(&fs_devices->device_list_mutex);
2764         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2765
2766         if (!dev) {
2767                 ret = -ENODEV;
2768                 goto out;
2769         }
2770
2771         di_args->devid = dev->devid;
2772         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2773         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2774         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2775         if (dev->name) {
2776                 struct rcu_string *name;
2777
2778                 rcu_read_lock();
2779                 name = rcu_dereference(dev->name);
2780                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2781                 rcu_read_unlock();
2782                 di_args->path[sizeof(di_args->path) - 1] = 0;
2783         } else {
2784                 di_args->path[0] = '\0';
2785         }
2786
2787 out:
2788         mutex_unlock(&fs_devices->device_list_mutex);
2789         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2790                 ret = -EFAULT;
2791
2792         kfree(di_args);
2793         return ret;
2794 }
2795
2796 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2797 {
2798         struct page *page;
2799         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2800
2801         page = grab_cache_page(inode->i_mapping, index);
2802         if (!page)
2803                 return NULL;
2804
2805         if (!PageUptodate(page)) {
2806                 if (extent_read_full_page_nolock(tree, page, btrfs_get_extent,
2807                                                  0))
2808                         return NULL;
2809                 lock_page(page);
2810                 if (!PageUptodate(page)) {
2811                         unlock_page(page);
2812                         page_cache_release(page);
2813                         return NULL;
2814                 }
2815         }
2816         unlock_page(page);
2817
2818         return page;
2819 }
2820
2821 static int gather_extent_pages(struct inode *inode, struct page **pages,
2822                                int num_pages, u64 off)
2823 {
2824         int i;
2825         pgoff_t index = off >> PAGE_CACHE_SHIFT;
2826
2827         for (i = 0; i < num_pages; i++) {
2828                 pages[i] = extent_same_get_page(inode, index + i);
2829                 if (!pages[i])
2830                         return -ENOMEM;
2831         }
2832         return 0;
2833 }
2834
2835 static inline void lock_extent_range(struct inode *inode, u64 off, u64 len)
2836 {
2837         /* do any pending delalloc/csum calc on src, one way or
2838            another, and lock file content */
2839         while (1) {
2840                 struct btrfs_ordered_extent *ordered;
2841                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2842                 ordered = btrfs_lookup_first_ordered_extent(inode,
2843                                                             off + len - 1);
2844                 if ((!ordered ||
2845                      ordered->file_offset + ordered->len <= off ||
2846                      ordered->file_offset >= off + len) &&
2847                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2848                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2849                         if (ordered)
2850                                 btrfs_put_ordered_extent(ordered);
2851                         break;
2852                 }
2853                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2854                 if (ordered)
2855                         btrfs_put_ordered_extent(ordered);
2856                 btrfs_wait_ordered_range(inode, off, len);
2857         }
2858 }
2859
2860 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2861 {
2862         mutex_unlock(&inode1->i_mutex);
2863         mutex_unlock(&inode2->i_mutex);
2864 }
2865
2866 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2867 {
2868         if (inode1 < inode2)
2869                 swap(inode1, inode2);
2870
2871         mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
2872         mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
2873 }
2874
2875 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2876                                       struct inode *inode2, u64 loff2, u64 len)
2877 {
2878         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2879         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2880 }
2881
2882 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2883                                      struct inode *inode2, u64 loff2, u64 len)
2884 {
2885         if (inode1 < inode2) {
2886                 swap(inode1, inode2);
2887                 swap(loff1, loff2);
2888         }
2889         lock_extent_range(inode1, loff1, len);
2890         lock_extent_range(inode2, loff2, len);
2891 }
2892
2893 struct cmp_pages {
2894         int             num_pages;
2895         struct page     **src_pages;
2896         struct page     **dst_pages;
2897 };
2898
2899 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2900 {
2901         int i;
2902         struct page *pg;
2903
2904         for (i = 0; i < cmp->num_pages; i++) {
2905                 pg = cmp->src_pages[i];
2906                 if (pg)
2907                         page_cache_release(pg);
2908                 pg = cmp->dst_pages[i];
2909                 if (pg)
2910                         page_cache_release(pg);
2911         }
2912         kfree(cmp->src_pages);
2913         kfree(cmp->dst_pages);
2914 }
2915
2916 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2917                                   struct inode *dst, u64 dst_loff,
2918                                   u64 len, struct cmp_pages *cmp)
2919 {
2920         int ret;
2921         int num_pages = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
2922         struct page **src_pgarr, **dst_pgarr;
2923
2924         /*
2925          * We must gather up all the pages before we initiate our
2926          * extent locking. We use an array for the page pointers. Size
2927          * of the array is bounded by len, which is in turn bounded by
2928          * BTRFS_MAX_DEDUPE_LEN.
2929          */
2930         src_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2931         dst_pgarr = kzalloc(num_pages * sizeof(struct page *), GFP_NOFS);
2932         if (!src_pgarr || !dst_pgarr) {
2933                 kfree(src_pgarr);
2934                 kfree(dst_pgarr);
2935                 return -ENOMEM;
2936         }
2937         cmp->num_pages = num_pages;
2938         cmp->src_pages = src_pgarr;
2939         cmp->dst_pages = dst_pgarr;
2940
2941         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
2942         if (ret)
2943                 goto out;
2944
2945         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
2946
2947 out:
2948         if (ret)
2949                 btrfs_cmp_data_free(cmp);
2950         return 0;
2951 }
2952
2953 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
2954                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
2955 {
2956         int ret = 0;
2957         int i;
2958         struct page *src_page, *dst_page;
2959         unsigned int cmp_len = PAGE_CACHE_SIZE;
2960         void *addr, *dst_addr;
2961
2962         i = 0;
2963         while (len) {
2964                 if (len < PAGE_CACHE_SIZE)
2965                         cmp_len = len;
2966
2967                 BUG_ON(i >= cmp->num_pages);
2968
2969                 src_page = cmp->src_pages[i];
2970                 dst_page = cmp->dst_pages[i];
2971
2972                 addr = kmap_atomic(src_page);
2973                 dst_addr = kmap_atomic(dst_page);
2974
2975                 flush_dcache_page(src_page);
2976                 flush_dcache_page(dst_page);
2977
2978                 if (memcmp(addr, dst_addr, cmp_len))
2979                         ret = BTRFS_SAME_DATA_DIFFERS;
2980
2981                 kunmap_atomic(addr);
2982                 kunmap_atomic(dst_addr);
2983
2984                 if (ret)
2985                         break;
2986
2987                 len -= cmp_len;
2988                 i++;
2989         }
2990
2991         return ret;
2992 }
2993
2994 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
2995                                      u64 olen)
2996 {
2997         u64 len = *plen;
2998         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
2999
3000         if (off + olen > inode->i_size || off + olen < off)
3001                 return -EINVAL;
3002
3003         /* if we extend to eof, continue to block boundary */
3004         if (off + len == inode->i_size)
3005                 *plen = len = ALIGN(inode->i_size, bs) - off;
3006
3007         /* Check that we are block aligned - btrfs_clone() requires this */
3008         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3009                 return -EINVAL;
3010
3011         return 0;
3012 }
3013
3014 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3015                              struct inode *dst, u64 dst_loff)
3016 {
3017         int ret;
3018         u64 len = olen;
3019         struct cmp_pages cmp;
3020         int same_inode = 0;
3021         u64 same_lock_start = 0;
3022         u64 same_lock_len = 0;
3023
3024         if (src == dst)
3025                 same_inode = 1;
3026
3027         if (len == 0)
3028                 return 0;
3029
3030         if (same_inode) {
3031                 mutex_lock(&src->i_mutex);
3032
3033                 ret = extent_same_check_offsets(src, loff, &len, olen);
3034                 if (ret)
3035                         goto out_unlock;
3036
3037                 /*
3038                  * Single inode case wants the same checks, except we
3039                  * don't want our length pushed out past i_size as
3040                  * comparing that data range makes no sense.
3041                  *
3042                  * extent_same_check_offsets() will do this for an
3043                  * unaligned length at i_size, so catch it here and
3044                  * reject the request.
3045                  *
3046                  * This effectively means we require aligned extents
3047                  * for the single-inode case, whereas the other cases
3048                  * allow an unaligned length so long as it ends at
3049                  * i_size.
3050                  */
3051                 if (len != olen) {
3052                         ret = -EINVAL;
3053                         goto out_unlock;
3054                 }
3055
3056                 /* Check for overlapping ranges */
3057                 if (dst_loff + len > loff && dst_loff < loff + len) {
3058                         ret = -EINVAL;
3059                         goto out_unlock;
3060                 }
3061
3062                 same_lock_start = min_t(u64, loff, dst_loff);
3063                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3064         } else {
3065                 btrfs_double_inode_lock(src, dst);
3066
3067                 ret = extent_same_check_offsets(src, loff, &len, olen);
3068                 if (ret)
3069                         goto out_unlock;
3070
3071                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3072                 if (ret)
3073                         goto out_unlock;
3074         }
3075
3076         /* don't make the dst file partly checksummed */
3077         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3078             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3079                 ret = -EINVAL;
3080                 goto out_unlock;
3081         }
3082
3083         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3084         if (ret)
3085                 goto out_unlock;
3086
3087         if (same_inode)
3088                 lock_extent_range(src, same_lock_start, same_lock_len);
3089         else
3090                 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3091
3092         /* pass original length for comparison so we stay within i_size */
3093         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3094         if (ret == 0)
3095                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3096
3097         if (same_inode)
3098                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3099                               same_lock_start + same_lock_len - 1);
3100         else
3101                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3102
3103         btrfs_cmp_data_free(&cmp);
3104 out_unlock:
3105         if (same_inode)
3106                 mutex_unlock(&src->i_mutex);
3107         else
3108                 btrfs_double_inode_unlock(src, dst);
3109
3110         return ret;
3111 }
3112
3113 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3114
3115 static long btrfs_ioctl_file_extent_same(struct file *file,
3116                         struct btrfs_ioctl_same_args __user *argp)
3117 {
3118         struct btrfs_ioctl_same_args *same = NULL;
3119         struct btrfs_ioctl_same_extent_info *info;
3120         struct inode *src = file_inode(file);
3121         u64 off;
3122         u64 len;
3123         int i;
3124         int ret;
3125         unsigned long size;
3126         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3127         bool is_admin = capable(CAP_SYS_ADMIN);
3128         u16 count;
3129
3130         if (!(file->f_mode & FMODE_READ))
3131                 return -EINVAL;
3132
3133         ret = mnt_want_write_file(file);
3134         if (ret)
3135                 return ret;
3136
3137         if (get_user(count, &argp->dest_count)) {
3138                 ret = -EFAULT;
3139                 goto out;
3140         }
3141
3142         size = offsetof(struct btrfs_ioctl_same_args __user, info[count]);
3143
3144         same = memdup_user(argp, size);
3145
3146         if (IS_ERR(same)) {
3147                 ret = PTR_ERR(same);
3148                 same = NULL;
3149                 goto out;
3150         }
3151
3152         off = same->logical_offset;
3153         len = same->length;
3154
3155         /*
3156          * Limit the total length we will dedupe for each operation.
3157          * This is intended to bound the total time spent in this
3158          * ioctl to something sane.
3159          */
3160         if (len > BTRFS_MAX_DEDUPE_LEN)
3161                 len = BTRFS_MAX_DEDUPE_LEN;
3162
3163         if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) {
3164                 /*
3165                  * Btrfs does not support blocksize < page_size. As a
3166                  * result, btrfs_cmp_data() won't correctly handle
3167                  * this situation without an update.
3168                  */
3169                 ret = -EINVAL;
3170                 goto out;
3171         }
3172
3173         ret = -EISDIR;
3174         if (S_ISDIR(src->i_mode))
3175                 goto out;
3176
3177         ret = -EACCES;
3178         if (!S_ISREG(src->i_mode))
3179                 goto out;
3180
3181         /* pre-format output fields to sane values */
3182         for (i = 0; i < count; i++) {
3183                 same->info[i].bytes_deduped = 0ULL;
3184                 same->info[i].status = 0;
3185         }
3186
3187         for (i = 0, info = same->info; i < count; i++, info++) {
3188                 struct inode *dst;
3189                 struct fd dst_file = fdget(info->fd);
3190                 if (!dst_file.file) {
3191                         info->status = -EBADF;
3192                         continue;
3193                 }
3194                 dst = file_inode(dst_file.file);
3195
3196                 if (!(is_admin || (dst_file.file->f_mode & FMODE_WRITE))) {
3197                         info->status = -EINVAL;
3198                 } else if (file->f_path.mnt != dst_file.file->f_path.mnt) {
3199                         info->status = -EXDEV;
3200                 } else if (S_ISDIR(dst->i_mode)) {
3201                         info->status = -EISDIR;
3202                 } else if (!S_ISREG(dst->i_mode)) {
3203                         info->status = -EACCES;
3204                 } else {
3205                         info->status = btrfs_extent_same(src, off, len, dst,
3206                                                         info->logical_offset);
3207                         if (info->status == 0)
3208                                 info->bytes_deduped += len;
3209                 }
3210                 fdput(dst_file);
3211         }
3212
3213         ret = copy_to_user(argp, same, size);
3214         if (ret)
3215                 ret = -EFAULT;
3216
3217 out:
3218         mnt_drop_write_file(file);
3219         kfree(same);
3220         return ret;
3221 }
3222
3223 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3224                                      struct inode *inode,
3225                                      u64 endoff,
3226                                      const u64 destoff,
3227                                      const u64 olen,
3228                                      int no_time_update)
3229 {
3230         struct btrfs_root *root = BTRFS_I(inode)->root;
3231         int ret;
3232
3233         inode_inc_iversion(inode);
3234         if (!no_time_update)
3235                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3236         /*
3237          * We round up to the block size at eof when determining which
3238          * extents to clone above, but shouldn't round up the file size.
3239          */
3240         if (endoff > destoff + olen)
3241                 endoff = destoff + olen;
3242         if (endoff > inode->i_size)
3243                 btrfs_i_size_write(inode, endoff);
3244
3245         ret = btrfs_update_inode(trans, root, inode);
3246         if (ret) {
3247                 btrfs_abort_transaction(trans, root, ret);
3248                 btrfs_end_transaction(trans, root);
3249                 goto out;
3250         }
3251         ret = btrfs_end_transaction(trans, root);
3252 out:
3253         return ret;
3254 }
3255
3256 static void clone_update_extent_map(struct inode *inode,
3257                                     const struct btrfs_trans_handle *trans,
3258                                     const struct btrfs_path *path,
3259                                     const u64 hole_offset,
3260                                     const u64 hole_len)
3261 {
3262         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3263         struct extent_map *em;
3264         int ret;
3265
3266         em = alloc_extent_map();
3267         if (!em) {
3268                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3269                         &BTRFS_I(inode)->runtime_flags);
3270                 return;
3271         }
3272
3273         if (path) {
3274                 struct btrfs_file_extent_item *fi;
3275
3276                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3277                                     struct btrfs_file_extent_item);
3278                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3279                 em->generation = -1;
3280                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3281                     BTRFS_FILE_EXTENT_INLINE)
3282                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3283                                 &BTRFS_I(inode)->runtime_flags);
3284         } else {
3285                 em->start = hole_offset;
3286                 em->len = hole_len;
3287                 em->ram_bytes = em->len;
3288                 em->orig_start = hole_offset;
3289                 em->block_start = EXTENT_MAP_HOLE;
3290                 em->block_len = 0;
3291                 em->orig_block_len = 0;
3292                 em->compress_type = BTRFS_COMPRESS_NONE;
3293                 em->generation = trans->transid;
3294         }
3295
3296         while (1) {
3297                 write_lock(&em_tree->lock);
3298                 ret = add_extent_mapping(em_tree, em, 1);
3299                 write_unlock(&em_tree->lock);
3300                 if (ret != -EEXIST) {
3301                         free_extent_map(em);
3302                         break;
3303                 }
3304                 btrfs_drop_extent_cache(inode, em->start,
3305                                         em->start + em->len - 1, 0);
3306         }
3307
3308         if (ret)
3309                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3310                         &BTRFS_I(inode)->runtime_flags);
3311 }
3312
3313 /*
3314  * Make sure we do not end up inserting an inline extent into a file that has
3315  * already other (non-inline) extents. If a file has an inline extent it can
3316  * not have any other extents and the (single) inline extent must start at the
3317  * file offset 0. Failing to respect these rules will lead to file corruption,
3318  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3319  *
3320  * We can have extents that have been already written to disk or we can have
3321  * dirty ranges still in delalloc, in which case the extent maps and items are
3322  * created only when we run delalloc, and the delalloc ranges might fall outside
3323  * the range we are currently locking in the inode's io tree. So we check the
3324  * inode's i_size because of that (i_size updates are done while holding the
3325  * i_mutex, which we are holding here).
3326  * We also check to see if the inode has a size not greater than "datal" but has
3327  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3328  * protected against such concurrent fallocate calls by the i_mutex).
3329  *
3330  * If the file has no extents but a size greater than datal, do not allow the
3331  * copy because we would need turn the inline extent into a non-inline one (even
3332  * with NO_HOLES enabled). If we find our destination inode only has one inline
3333  * extent, just overwrite it with the source inline extent if its size is less
3334  * than the source extent's size, or we could copy the source inline extent's
3335  * data into the destination inode's inline extent if the later is greater then
3336  * the former.
3337  */
3338 static int clone_copy_inline_extent(struct inode *src,
3339                                     struct inode *dst,
3340                                     struct btrfs_trans_handle *trans,
3341                                     struct btrfs_path *path,
3342                                     struct btrfs_key *new_key,
3343                                     const u64 drop_start,
3344                                     const u64 datal,
3345                                     const u64 skip,
3346                                     const u64 size,
3347                                     char *inline_data)
3348 {
3349         struct btrfs_root *root = BTRFS_I(dst)->root;
3350         const u64 aligned_end = ALIGN(new_key->offset + datal,
3351                                       root->sectorsize);
3352         int ret;
3353         struct btrfs_key key;
3354
3355         if (new_key->offset > 0)
3356                 return -EOPNOTSUPP;
3357
3358         key.objectid = btrfs_ino(dst);
3359         key.type = BTRFS_EXTENT_DATA_KEY;
3360         key.offset = 0;
3361         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3362         if (ret < 0) {
3363                 return ret;
3364         } else if (ret > 0) {
3365                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3366                         ret = btrfs_next_leaf(root, path);
3367                         if (ret < 0)
3368                                 return ret;
3369                         else if (ret > 0)
3370                                 goto copy_inline_extent;
3371                 }
3372                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3373                 if (key.objectid == btrfs_ino(dst) &&
3374                     key.type == BTRFS_EXTENT_DATA_KEY) {
3375                         ASSERT(key.offset > 0);
3376                         return -EOPNOTSUPP;
3377                 }
3378         } else if (i_size_read(dst) <= datal) {
3379                 struct btrfs_file_extent_item *ei;
3380                 u64 ext_len;
3381
3382                 /*
3383                  * If the file size is <= datal, make sure there are no other
3384                  * extents following (can happen do to an fallocate call with
3385                  * the flag FALLOC_FL_KEEP_SIZE).
3386                  */
3387                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3388                                     struct btrfs_file_extent_item);
3389                 /*
3390                  * If it's an inline extent, it can not have other extents
3391                  * following it.
3392                  */
3393                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3394                     BTRFS_FILE_EXTENT_INLINE)
3395                         goto copy_inline_extent;
3396
3397                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3398                 if (ext_len > aligned_end)
3399                         return -EOPNOTSUPP;
3400
3401                 ret = btrfs_next_item(root, path);
3402                 if (ret < 0) {
3403                         return ret;
3404                 } else if (ret == 0) {
3405                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3406                                               path->slots[0]);
3407                         if (key.objectid == btrfs_ino(dst) &&
3408                             key.type == BTRFS_EXTENT_DATA_KEY)
3409                                 return -EOPNOTSUPP;
3410                 }
3411         }
3412
3413 copy_inline_extent:
3414         /*
3415          * We have no extent items, or we have an extent at offset 0 which may
3416          * or may not be inlined. All these cases are dealt the same way.
3417          */
3418         if (i_size_read(dst) > datal) {
3419                 /*
3420                  * If the destination inode has an inline extent...
3421                  * This would require copying the data from the source inline
3422                  * extent into the beginning of the destination's inline extent.
3423                  * But this is really complex, both extents can be compressed
3424                  * or just one of them, which would require decompressing and
3425                  * re-compressing data (which could increase the new compressed
3426                  * size, not allowing the compressed data to fit anymore in an
3427                  * inline extent).
3428                  * So just don't support this case for now (it should be rare,
3429                  * we are not really saving space when cloning inline extents).
3430                  */
3431                 return -EOPNOTSUPP;
3432         }
3433
3434         btrfs_release_path(path);
3435         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3436         if (ret)
3437                 return ret;
3438         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3439         if (ret)
3440                 return ret;
3441
3442         if (skip) {
3443                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3444
3445                 memmove(inline_data + start, inline_data + start + skip, datal);
3446         }
3447
3448         write_extent_buffer(path->nodes[0], inline_data,
3449                             btrfs_item_ptr_offset(path->nodes[0],
3450                                                   path->slots[0]),
3451                             size);
3452         inode_add_bytes(dst, datal);
3453
3454         return 0;
3455 }
3456
3457 /**
3458  * btrfs_clone() - clone a range from inode file to another
3459  *
3460  * @src: Inode to clone from
3461  * @inode: Inode to clone to
3462  * @off: Offset within source to start clone from
3463  * @olen: Original length, passed by user, of range to clone
3464  * @olen_aligned: Block-aligned value of olen
3465  * @destoff: Offset within @inode to start clone
3466  * @no_time_update: Whether to update mtime/ctime on the target inode
3467  */
3468 static int btrfs_clone(struct inode *src, struct inode *inode,
3469                        const u64 off, const u64 olen, const u64 olen_aligned,
3470                        const u64 destoff, int no_time_update)
3471 {
3472         struct btrfs_root *root = BTRFS_I(inode)->root;
3473         struct btrfs_path *path = NULL;
3474         struct extent_buffer *leaf;
3475         struct btrfs_trans_handle *trans;
3476         char *buf = NULL;
3477         struct btrfs_key key;
3478         u32 nritems;
3479         int slot;
3480         int ret;
3481         const u64 len = olen_aligned;
3482         u64 last_dest_end = destoff;
3483
3484         ret = -ENOMEM;
3485         buf = vmalloc(root->nodesize);
3486         if (!buf)
3487                 return ret;
3488
3489         path = btrfs_alloc_path();
3490         if (!path) {
3491                 vfree(buf);
3492                 return ret;
3493         }
3494
3495         path->reada = READA_FORWARD;
3496         /* clone data */
3497         key.objectid = btrfs_ino(src);
3498         key.type = BTRFS_EXTENT_DATA_KEY;
3499         key.offset = off;
3500
3501         while (1) {
3502                 u64 next_key_min_offset = key.offset + 1;
3503
3504                 /*
3505                  * note the key will change type as we walk through the
3506                  * tree.
3507                  */
3508                 path->leave_spinning = 1;
3509                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3510                                 0, 0);
3511                 if (ret < 0)
3512                         goto out;
3513                 /*
3514                  * First search, if no extent item that starts at offset off was
3515                  * found but the previous item is an extent item, it's possible
3516                  * it might overlap our target range, therefore process it.
3517                  */
3518                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3519                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3520                                               path->slots[0] - 1);
3521                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3522                                 path->slots[0]--;
3523                 }
3524
3525                 nritems = btrfs_header_nritems(path->nodes[0]);
3526 process_slot:
3527                 if (path->slots[0] >= nritems) {
3528                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3529                         if (ret < 0)
3530                                 goto out;
3531                         if (ret > 0)
3532                                 break;
3533                         nritems = btrfs_header_nritems(path->nodes[0]);
3534                 }
3535                 leaf = path->nodes[0];
3536                 slot = path->slots[0];
3537
3538                 btrfs_item_key_to_cpu(leaf, &key, slot);
3539                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3540                     key.objectid != btrfs_ino(src))
3541                         break;
3542
3543                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3544                         struct btrfs_file_extent_item *extent;
3545                         int type;
3546                         u32 size;
3547                         struct btrfs_key new_key;
3548                         u64 disko = 0, diskl = 0;
3549                         u64 datao = 0, datal = 0;
3550                         u8 comp;
3551                         u64 drop_start;
3552
3553                         extent = btrfs_item_ptr(leaf, slot,
3554                                                 struct btrfs_file_extent_item);
3555                         comp = btrfs_file_extent_compression(leaf, extent);
3556                         type = btrfs_file_extent_type(leaf, extent);
3557                         if (type == BTRFS_FILE_EXTENT_REG ||
3558                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3559                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3560                                                                       extent);
3561                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3562                                                                  extent);
3563                                 datao = btrfs_file_extent_offset(leaf, extent);
3564                                 datal = btrfs_file_extent_num_bytes(leaf,
3565                                                                     extent);
3566                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3567                                 /* take upper bound, may be compressed */
3568                                 datal = btrfs_file_extent_ram_bytes(leaf,
3569                                                                     extent);
3570                         }
3571
3572                         /*
3573                          * The first search might have left us at an extent
3574                          * item that ends before our target range's start, can
3575                          * happen if we have holes and NO_HOLES feature enabled.
3576                          */
3577                         if (key.offset + datal <= off) {
3578                                 path->slots[0]++;
3579                                 goto process_slot;
3580                         } else if (key.offset >= off + len) {
3581                                 break;
3582                         }
3583                         next_key_min_offset = key.offset + datal;
3584                         size = btrfs_item_size_nr(leaf, slot);
3585                         read_extent_buffer(leaf, buf,
3586                                            btrfs_item_ptr_offset(leaf, slot),
3587                                            size);
3588
3589                         btrfs_release_path(path);
3590                         path->leave_spinning = 0;
3591
3592                         memcpy(&new_key, &key, sizeof(new_key));
3593                         new_key.objectid = btrfs_ino(inode);
3594                         if (off <= key.offset)
3595                                 new_key.offset = key.offset + destoff - off;
3596                         else
3597                                 new_key.offset = destoff;
3598
3599                         /*
3600                          * Deal with a hole that doesn't have an extent item
3601                          * that represents it (NO_HOLES feature enabled).
3602                          * This hole is either in the middle of the cloning
3603                          * range or at the beginning (fully overlaps it or
3604                          * partially overlaps it).
3605                          */
3606                         if (new_key.offset != last_dest_end)
3607                                 drop_start = last_dest_end;
3608                         else
3609                                 drop_start = new_key.offset;
3610
3611                         /*
3612                          * 1 - adjusting old extent (we may have to split it)
3613                          * 1 - add new extent
3614                          * 1 - inode update
3615                          */
3616                         trans = btrfs_start_transaction(root, 3);
3617                         if (IS_ERR(trans)) {
3618                                 ret = PTR_ERR(trans);
3619                                 goto out;
3620                         }
3621
3622                         if (type == BTRFS_FILE_EXTENT_REG ||
3623                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3624                                 /*
3625                                  *    a  | --- range to clone ---|  b
3626                                  * | ------------- extent ------------- |
3627                                  */
3628
3629                                 /* subtract range b */
3630                                 if (key.offset + datal > off + len)
3631                                         datal = off + len - key.offset;
3632
3633                                 /* subtract range a */
3634                                 if (off > key.offset) {
3635                                         datao += off - key.offset;
3636                                         datal -= off - key.offset;
3637                                 }
3638
3639                                 ret = btrfs_drop_extents(trans, root, inode,
3640                                                          drop_start,
3641                                                          new_key.offset + datal,
3642                                                          1);
3643                                 if (ret) {
3644                                         if (ret != -EOPNOTSUPP)
3645                                                 btrfs_abort_transaction(trans,
3646                                                                 root, ret);
3647                                         btrfs_end_transaction(trans, root);
3648                                         goto out;
3649                                 }
3650
3651                                 ret = btrfs_insert_empty_item(trans, root, path,
3652                                                               &new_key, size);
3653                                 if (ret) {
3654                                         btrfs_abort_transaction(trans, root,
3655                                                                 ret);
3656                                         btrfs_end_transaction(trans, root);
3657                                         goto out;
3658                                 }
3659
3660                                 leaf = path->nodes[0];
3661                                 slot = path->slots[0];
3662                                 write_extent_buffer(leaf, buf,
3663                                             btrfs_item_ptr_offset(leaf, slot),
3664                                             size);
3665
3666                                 extent = btrfs_item_ptr(leaf, slot,
3667                                                 struct btrfs_file_extent_item);
3668
3669                                 /* disko == 0 means it's a hole */
3670                                 if (!disko)
3671                                         datao = 0;
3672
3673                                 btrfs_set_file_extent_offset(leaf, extent,
3674                                                              datao);
3675                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3676                                                                 datal);
3677
3678                                 if (disko) {
3679                                         inode_add_bytes(inode, datal);
3680                                         ret = btrfs_inc_extent_ref(trans, root,
3681                                                         disko, diskl, 0,
3682                                                         root->root_key.objectid,
3683                                                         btrfs_ino(inode),
3684                                                         new_key.offset - datao);
3685                                         if (ret) {
3686                                                 btrfs_abort_transaction(trans,
3687                                                                         root,
3688                                                                         ret);
3689                                                 btrfs_end_transaction(trans,
3690                                                                       root);
3691                                                 goto out;
3692
3693                                         }
3694                                 }
3695                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3696                                 u64 skip = 0;
3697                                 u64 trim = 0;
3698
3699                                 if (off > key.offset) {
3700                                         skip = off - key.offset;
3701                                         new_key.offset += skip;
3702                                 }
3703
3704                                 if (key.offset + datal > off + len)
3705                                         trim = key.offset + datal - (off + len);
3706
3707                                 if (comp && (skip || trim)) {
3708                                         ret = -EINVAL;
3709                                         btrfs_end_transaction(trans, root);
3710                                         goto out;
3711                                 }
3712                                 size -= skip + trim;
3713                                 datal -= skip + trim;
3714
3715                                 ret = clone_copy_inline_extent(src, inode,
3716                                                                trans, path,
3717                                                                &new_key,
3718                                                                drop_start,
3719                                                                datal,
3720                                                                skip, size, buf);
3721                                 if (ret) {
3722                                         if (ret != -EOPNOTSUPP)
3723                                                 btrfs_abort_transaction(trans,
3724                                                                         root,
3725                                                                         ret);
3726                                         btrfs_end_transaction(trans, root);
3727                                         goto out;
3728                                 }
3729                                 leaf = path->nodes[0];
3730                                 slot = path->slots[0];
3731                         }
3732
3733                         /* If we have an implicit hole (NO_HOLES feature). */
3734                         if (drop_start < new_key.offset)
3735                                 clone_update_extent_map(inode, trans,
3736                                                 NULL, drop_start,
3737                                                 new_key.offset - drop_start);
3738
3739                         clone_update_extent_map(inode, trans, path, 0, 0);
3740
3741                         btrfs_mark_buffer_dirty(leaf);
3742                         btrfs_release_path(path);
3743
3744                         last_dest_end = ALIGN(new_key.offset + datal,
3745                                               root->sectorsize);
3746                         ret = clone_finish_inode_update(trans, inode,
3747                                                         last_dest_end,
3748                                                         destoff, olen,
3749                                                         no_time_update);
3750                         if (ret)
3751                                 goto out;
3752                         if (new_key.offset + datal >= destoff + len)
3753                                 break;
3754                 }
3755                 btrfs_release_path(path);
3756                 key.offset = next_key_min_offset;
3757         }
3758         ret = 0;
3759
3760         if (last_dest_end < destoff + len) {
3761                 /*
3762                  * We have an implicit hole (NO_HOLES feature is enabled) that
3763                  * fully or partially overlaps our cloning range at its end.
3764                  */
3765                 btrfs_release_path(path);
3766
3767                 /*
3768                  * 1 - remove extent(s)
3769                  * 1 - inode update
3770                  */
3771                 trans = btrfs_start_transaction(root, 2);
3772                 if (IS_ERR(trans)) {
3773                         ret = PTR_ERR(trans);
3774                         goto out;
3775                 }
3776                 ret = btrfs_drop_extents(trans, root, inode,
3777                                          last_dest_end, destoff + len, 1);
3778                 if (ret) {
3779                         if (ret != -EOPNOTSUPP)
3780                                 btrfs_abort_transaction(trans, root, ret);
3781                         btrfs_end_transaction(trans, root);
3782                         goto out;
3783                 }
3784                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3785                                         destoff + len - last_dest_end);
3786                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3787                                                 destoff, olen, no_time_update);
3788         }
3789
3790 out:
3791         btrfs_free_path(path);
3792         vfree(buf);
3793         return ret;
3794 }
3795
3796 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
3797                                        u64 off, u64 olen, u64 destoff)
3798 {
3799         struct inode *inode = file_inode(file);
3800         struct btrfs_root *root = BTRFS_I(inode)->root;
3801         struct fd src_file;
3802         struct inode *src;
3803         int ret;
3804         u64 len = olen;
3805         u64 bs = root->fs_info->sb->s_blocksize;
3806         int same_inode = 0;
3807
3808         /*
3809          * TODO:
3810          * - split compressed inline extents.  annoying: we need to
3811          *   decompress into destination's address_space (the file offset
3812          *   may change, so source mapping won't do), then recompress (or
3813          *   otherwise reinsert) a subrange.
3814          *
3815          * - split destination inode's inline extents.  The inline extents can
3816          *   be either compressed or non-compressed.
3817          */
3818
3819         /* the destination must be opened for writing */
3820         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
3821                 return -EINVAL;
3822
3823         if (btrfs_root_readonly(root))
3824                 return -EROFS;
3825
3826         ret = mnt_want_write_file(file);
3827         if (ret)
3828                 return ret;
3829
3830         src_file = fdget(srcfd);
3831         if (!src_file.file) {
3832                 ret = -EBADF;
3833                 goto out_drop_write;
3834         }
3835
3836         ret = -EXDEV;
3837         if (src_file.file->f_path.mnt != file->f_path.mnt)
3838                 goto out_fput;
3839
3840         src = file_inode(src_file.file);
3841
3842         ret = -EINVAL;
3843         if (src == inode)
3844                 same_inode = 1;
3845
3846         /* the src must be open for reading */
3847         if (!(src_file.file->f_mode & FMODE_READ))
3848                 goto out_fput;
3849
3850         /* don't make the dst file partly checksummed */
3851         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3852             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3853                 goto out_fput;
3854
3855         ret = -EISDIR;
3856         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3857                 goto out_fput;
3858
3859         ret = -EXDEV;
3860         if (src->i_sb != inode->i_sb)
3861                 goto out_fput;
3862
3863         if (!same_inode) {
3864                 btrfs_double_inode_lock(src, inode);
3865         } else {
3866                 mutex_lock(&src->i_mutex);
3867         }
3868
3869         /* determine range to clone */
3870         ret = -EINVAL;
3871         if (off + len > src->i_size || off + len < off)
3872                 goto out_unlock;
3873         if (len == 0)
3874                 olen = len = src->i_size - off;
3875         /* if we extend to eof, continue to block boundary */
3876         if (off + len == src->i_size)
3877                 len = ALIGN(src->i_size, bs) - off;
3878
3879         if (len == 0) {
3880                 ret = 0;
3881                 goto out_unlock;
3882         }
3883
3884         /* verify the end result is block aligned */
3885         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3886             !IS_ALIGNED(destoff, bs))
3887                 goto out_unlock;
3888
3889         /* verify if ranges are overlapped within the same file */
3890         if (same_inode) {
3891                 if (destoff + len > off && destoff < off + len)
3892                         goto out_unlock;
3893         }
3894
3895         if (destoff > inode->i_size) {
3896                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3897                 if (ret)
3898                         goto out_unlock;
3899         }
3900
3901         /*
3902          * Lock the target range too. Right after we replace the file extent
3903          * items in the fs tree (which now point to the cloned data), we might
3904          * have a worker replace them with extent items relative to a write
3905          * operation that was issued before this clone operation (i.e. confront
3906          * with inode.c:btrfs_finish_ordered_io).
3907          */
3908         if (same_inode) {
3909                 u64 lock_start = min_t(u64, off, destoff);
3910                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3911
3912                 lock_extent_range(src, lock_start, lock_len);
3913         } else {
3914                 btrfs_double_extent_lock(src, off, inode, destoff, len);
3915         }
3916
3917         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3918
3919         if (same_inode) {
3920                 u64 lock_start = min_t(u64, off, destoff);
3921                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3922
3923                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3924         } else {
3925                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3926         }
3927         /*
3928          * Truncate page cache pages so that future reads will see the cloned
3929          * data immediately and not the previous data.
3930          */
3931         truncate_inode_pages_range(&inode->i_data, destoff,
3932                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
3933 out_unlock:
3934         if (!same_inode)
3935                 btrfs_double_inode_unlock(src, inode);
3936         else
3937                 mutex_unlock(&src->i_mutex);
3938 out_fput:
3939         fdput(src_file);
3940 out_drop_write:
3941         mnt_drop_write_file(file);
3942         return ret;
3943 }
3944
3945 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
3946 {
3947         struct btrfs_ioctl_clone_range_args args;
3948
3949         if (copy_from_user(&args, argp, sizeof(args)))
3950                 return -EFAULT;
3951         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
3952                                  args.src_length, args.dest_offset);
3953 }
3954
3955 /*
3956  * there are many ways the trans_start and trans_end ioctls can lead
3957  * to deadlocks.  They should only be used by applications that
3958  * basically own the machine, and have a very in depth understanding
3959  * of all the possible deadlocks and enospc problems.
3960  */
3961 static long btrfs_ioctl_trans_start(struct file *file)
3962 {
3963         struct inode *inode = file_inode(file);
3964         struct btrfs_root *root = BTRFS_I(inode)->root;
3965         struct btrfs_trans_handle *trans;
3966         int ret;
3967
3968         ret = -EPERM;
3969         if (!capable(CAP_SYS_ADMIN))
3970                 goto out;
3971
3972         ret = -EINPROGRESS;
3973         if (file->private_data)
3974                 goto out;
3975
3976         ret = -EROFS;
3977         if (btrfs_root_readonly(root))
3978                 goto out;
3979
3980         ret = mnt_want_write_file(file);
3981         if (ret)
3982                 goto out;
3983
3984         atomic_inc(&root->fs_info->open_ioctl_trans);
3985
3986         ret = -ENOMEM;
3987         trans = btrfs_start_ioctl_transaction(root);
3988         if (IS_ERR(trans))
3989                 goto out_drop;
3990
3991         file->private_data = trans;
3992         return 0;
3993
3994 out_drop:
3995         atomic_dec(&root->fs_info->open_ioctl_trans);
3996         mnt_drop_write_file(file);
3997 out:
3998         return ret;
3999 }
4000
4001 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4002 {
4003         struct inode *inode = file_inode(file);
4004         struct btrfs_root *root = BTRFS_I(inode)->root;
4005         struct btrfs_root *new_root;
4006         struct btrfs_dir_item *di;
4007         struct btrfs_trans_handle *trans;
4008         struct btrfs_path *path;
4009         struct btrfs_key location;
4010         struct btrfs_disk_key disk_key;
4011         u64 objectid = 0;
4012         u64 dir_id;
4013         int ret;
4014
4015         if (!capable(CAP_SYS_ADMIN))
4016                 return -EPERM;
4017
4018         ret = mnt_want_write_file(file);
4019         if (ret)
4020                 return ret;
4021
4022         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4023                 ret = -EFAULT;
4024                 goto out;
4025         }
4026
4027         if (!objectid)
4028                 objectid = BTRFS_FS_TREE_OBJECTID;
4029
4030         location.objectid = objectid;
4031         location.type = BTRFS_ROOT_ITEM_KEY;
4032         location.offset = (u64)-1;
4033
4034         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4035         if (IS_ERR(new_root)) {
4036                 ret = PTR_ERR(new_root);
4037                 goto out;
4038         }
4039
4040         path = btrfs_alloc_path();
4041         if (!path) {
4042                 ret = -ENOMEM;
4043                 goto out;
4044         }
4045         path->leave_spinning = 1;
4046
4047         trans = btrfs_start_transaction(root, 1);
4048         if (IS_ERR(trans)) {
4049                 btrfs_free_path(path);
4050                 ret = PTR_ERR(trans);
4051                 goto out;
4052         }
4053
4054         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4055         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4056                                    dir_id, "default", 7, 1);
4057         if (IS_ERR_OR_NULL(di)) {
4058                 btrfs_free_path(path);
4059                 btrfs_end_transaction(trans, root);
4060                 btrfs_err(new_root->fs_info, "Umm, you don't have the default dir"
4061                            "item, this isn't going to work");
4062                 ret = -ENOENT;
4063                 goto out;
4064         }
4065
4066         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4067         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4068         btrfs_mark_buffer_dirty(path->nodes[0]);
4069         btrfs_free_path(path);
4070
4071         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4072         btrfs_sysfs_feature_update(root->fs_info,
4073                 BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL, FEAT_INCOMPAT);
4074         btrfs_end_transaction(trans, root);
4075 out:
4076         mnt_drop_write_file(file);
4077         return ret;
4078 }
4079
4080 void btrfs_get_block_group_info(struct list_head *groups_list,
4081                                 struct btrfs_ioctl_space_info *space)
4082 {
4083         struct btrfs_block_group_cache *block_group;
4084
4085         space->total_bytes = 0;
4086         space->used_bytes = 0;
4087         space->flags = 0;
4088         list_for_each_entry(block_group, groups_list, list) {
4089                 space->flags = block_group->flags;
4090                 space->total_bytes += block_group->key.offset;
4091                 space->used_bytes +=
4092                         btrfs_block_group_used(&block_group->item);
4093         }
4094 }
4095
4096 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4097 {
4098         struct btrfs_ioctl_space_args space_args;
4099         struct btrfs_ioctl_space_info space;
4100         struct btrfs_ioctl_space_info *dest;
4101         struct btrfs_ioctl_space_info *dest_orig;
4102         struct btrfs_ioctl_space_info __user *user_dest;
4103         struct btrfs_space_info *info;
4104         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4105                        BTRFS_BLOCK_GROUP_SYSTEM,
4106                        BTRFS_BLOCK_GROUP_METADATA,
4107                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4108         int num_types = 4;
4109         int alloc_size;
4110         int ret = 0;
4111         u64 slot_count = 0;
4112         int i, c;
4113
4114         if (copy_from_user(&space_args,
4115                            (struct btrfs_ioctl_space_args __user *)arg,
4116                            sizeof(space_args)))
4117                 return -EFAULT;
4118
4119         for (i = 0; i < num_types; i++) {
4120                 struct btrfs_space_info *tmp;
4121
4122                 info = NULL;
4123                 rcu_read_lock();
4124                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4125                                         list) {
4126                         if (tmp->flags == types[i]) {
4127                                 info = tmp;
4128                                 break;
4129                         }
4130                 }
4131                 rcu_read_unlock();
4132
4133                 if (!info)
4134                         continue;
4135
4136                 down_read(&info->groups_sem);
4137                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4138                         if (!list_empty(&info->block_groups[c]))
4139                                 slot_count++;
4140                 }
4141                 up_read(&info->groups_sem);
4142         }
4143
4144         /*
4145          * Global block reserve, exported as a space_info
4146          */
4147         slot_count++;
4148
4149         /* space_slots == 0 means they are asking for a count */
4150         if (space_args.space_slots == 0) {
4151                 space_args.total_spaces = slot_count;
4152                 goto out;
4153         }
4154
4155         slot_count = min_t(u64, space_args.space_slots, slot_count);
4156
4157         alloc_size = sizeof(*dest) * slot_count;
4158
4159         /* we generally have at most 6 or so space infos, one for each raid
4160          * level.  So, a whole page should be more than enough for everyone
4161          */
4162         if (alloc_size > PAGE_CACHE_SIZE)
4163                 return -ENOMEM;
4164
4165         space_args.total_spaces = 0;
4166         dest = kmalloc(alloc_size, GFP_KERNEL);
4167         if (!dest)
4168                 return -ENOMEM;
4169         dest_orig = dest;
4170
4171         /* now we have a buffer to copy into */
4172         for (i = 0; i < num_types; i++) {
4173                 struct btrfs_space_info *tmp;
4174
4175                 if (!slot_count)
4176                         break;
4177
4178                 info = NULL;
4179                 rcu_read_lock();
4180                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4181                                         list) {
4182                         if (tmp->flags == types[i]) {
4183                                 info = tmp;
4184                                 break;
4185                         }
4186                 }
4187                 rcu_read_unlock();
4188
4189                 if (!info)
4190                         continue;
4191                 down_read(&info->groups_sem);
4192                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4193                         if (!list_empty(&info->block_groups[c])) {
4194                                 btrfs_get_block_group_info(
4195                                         &info->block_groups[c], &space);
4196                                 memcpy(dest, &space, sizeof(space));
4197                                 dest++;
4198                                 space_args.total_spaces++;
4199                                 slot_count--;
4200                         }
4201                         if (!slot_count)
4202                                 break;
4203                 }
4204                 up_read(&info->groups_sem);
4205         }
4206
4207         /*
4208          * Add global block reserve
4209          */
4210         if (slot_count) {
4211                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4212
4213                 spin_lock(&block_rsv->lock);
4214                 space.total_bytes = block_rsv->size;
4215                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4216                 spin_unlock(&block_rsv->lock);
4217                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4218                 memcpy(dest, &space, sizeof(space));
4219                 space_args.total_spaces++;
4220         }
4221
4222         user_dest = (struct btrfs_ioctl_space_info __user *)
4223                 (arg + sizeof(struct btrfs_ioctl_space_args));
4224
4225         if (copy_to_user(user_dest, dest_orig, alloc_size))
4226                 ret = -EFAULT;
4227
4228         kfree(dest_orig);
4229 out:
4230         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4231                 ret = -EFAULT;
4232
4233         return ret;
4234 }
4235
4236 /*
4237  * there are many ways the trans_start and trans_end ioctls can lead
4238  * to deadlocks.  They should only be used by applications that
4239  * basically own the machine, and have a very in depth understanding
4240  * of all the possible deadlocks and enospc problems.
4241  */
4242 long btrfs_ioctl_trans_end(struct file *file)
4243 {
4244         struct inode *inode = file_inode(file);
4245         struct btrfs_root *root = BTRFS_I(inode)->root;
4246         struct btrfs_trans_handle *trans;
4247
4248         trans = file->private_data;
4249         if (!trans)
4250                 return -EINVAL;
4251         file->private_data = NULL;
4252
4253         btrfs_end_transaction(trans, root);
4254
4255         atomic_dec(&root->fs_info->open_ioctl_trans);
4256
4257         mnt_drop_write_file(file);
4258         return 0;
4259 }
4260
4261 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4262                                             void __user *argp)
4263 {
4264         struct btrfs_trans_handle *trans;
4265         u64 transid;
4266         int ret;
4267
4268         trans = btrfs_attach_transaction_barrier(root);
4269         if (IS_ERR(trans)) {
4270                 if (PTR_ERR(trans) != -ENOENT)
4271                         return PTR_ERR(trans);
4272
4273                 /* No running transaction, don't bother */
4274                 transid = root->fs_info->last_trans_committed;
4275                 goto out;
4276         }
4277         transid = trans->transid;
4278         ret = btrfs_commit_transaction_async(trans, root, 0);
4279         if (ret) {
4280                 btrfs_end_transaction(trans, root);
4281                 return ret;
4282         }
4283 out:
4284         if (argp)
4285                 if (copy_to_user(argp, &transid, sizeof(transid)))
4286                         return -EFAULT;
4287         return 0;
4288 }
4289
4290 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4291                                            void __user *argp)
4292 {
4293         u64 transid;
4294
4295         if (argp) {
4296                 if (copy_from_user(&transid, argp, sizeof(transid)))
4297                         return -EFAULT;
4298         } else {
4299                 transid = 0;  /* current trans */
4300         }
4301         return btrfs_wait_for_commit(root, transid);
4302 }
4303
4304 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4305 {
4306         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4307         struct btrfs_ioctl_scrub_args *sa;
4308         int ret;
4309
4310         if (!capable(CAP_SYS_ADMIN))
4311                 return -EPERM;
4312
4313         sa = memdup_user(arg, sizeof(*sa));
4314         if (IS_ERR(sa))
4315                 return PTR_ERR(sa);
4316
4317         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4318                 ret = mnt_want_write_file(file);
4319                 if (ret)
4320                         goto out;
4321         }
4322
4323         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4324                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4325                               0);
4326
4327         if (copy_to_user(arg, sa, sizeof(*sa)))
4328                 ret = -EFAULT;
4329
4330         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4331                 mnt_drop_write_file(file);
4332 out:
4333         kfree(sa);
4334         return ret;
4335 }
4336
4337 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4338 {
4339         if (!capable(CAP_SYS_ADMIN))
4340                 return -EPERM;
4341
4342         return btrfs_scrub_cancel(root->fs_info);
4343 }
4344
4345 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4346                                        void __user *arg)
4347 {
4348         struct btrfs_ioctl_scrub_args *sa;
4349         int ret;
4350
4351         if (!capable(CAP_SYS_ADMIN))
4352                 return -EPERM;
4353
4354         sa = memdup_user(arg, sizeof(*sa));
4355         if (IS_ERR(sa))
4356                 return PTR_ERR(sa);
4357
4358         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4359
4360         if (copy_to_user(arg, sa, sizeof(*sa)))
4361                 ret = -EFAULT;
4362
4363         kfree(sa);
4364         return ret;
4365 }
4366
4367 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4368                                       void __user *arg)
4369 {
4370         struct btrfs_ioctl_get_dev_stats *sa;
4371         int ret;
4372
4373         sa = memdup_user(arg, sizeof(*sa));
4374         if (IS_ERR(sa))
4375                 return PTR_ERR(sa);
4376
4377         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4378                 kfree(sa);
4379                 return -EPERM;
4380         }
4381
4382         ret = btrfs_get_dev_stats(root, sa);
4383
4384         if (copy_to_user(arg, sa, sizeof(*sa)))
4385                 ret = -EFAULT;
4386
4387         kfree(sa);
4388         return ret;
4389 }
4390
4391 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4392 {
4393         struct btrfs_ioctl_dev_replace_args *p;
4394         int ret;
4395
4396         if (!capable(CAP_SYS_ADMIN))
4397                 return -EPERM;
4398
4399         p = memdup_user(arg, sizeof(*p));
4400         if (IS_ERR(p))
4401                 return PTR_ERR(p);
4402
4403         switch (p->cmd) {
4404         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4405                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4406                         ret = -EROFS;
4407                         goto out;
4408                 }
4409                 if (atomic_xchg(
4410                         &root->fs_info->mutually_exclusive_operation_running,
4411                         1)) {
4412                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4413                 } else {
4414                         ret = btrfs_dev_replace_start(root, p);
4415                         atomic_set(
4416                          &root->fs_info->mutually_exclusive_operation_running,
4417                          0);
4418                 }
4419                 break;
4420         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4421                 btrfs_dev_replace_status(root->fs_info, p);
4422                 ret = 0;
4423                 break;
4424         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4425                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4426                 break;
4427         default:
4428                 ret = -EINVAL;
4429                 break;
4430         }
4431
4432         if (copy_to_user(arg, p, sizeof(*p)))
4433                 ret = -EFAULT;
4434 out:
4435         kfree(p);
4436         return ret;
4437 }
4438
4439 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4440 {
4441         int ret = 0;
4442         int i;
4443         u64 rel_ptr;
4444         int size;
4445         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4446         struct inode_fs_paths *ipath = NULL;
4447         struct btrfs_path *path;
4448
4449         if (!capable(CAP_DAC_READ_SEARCH))
4450                 return -EPERM;
4451
4452         path = btrfs_alloc_path();
4453         if (!path) {
4454                 ret = -ENOMEM;
4455                 goto out;
4456         }
4457
4458         ipa = memdup_user(arg, sizeof(*ipa));
4459         if (IS_ERR(ipa)) {
4460                 ret = PTR_ERR(ipa);
4461                 ipa = NULL;
4462                 goto out;
4463         }
4464
4465         size = min_t(u32, ipa->size, 4096);
4466         ipath = init_ipath(size, root, path);
4467         if (IS_ERR(ipath)) {
4468                 ret = PTR_ERR(ipath);
4469                 ipath = NULL;
4470                 goto out;
4471         }
4472
4473         ret = paths_from_inode(ipa->inum, ipath);
4474         if (ret < 0)
4475                 goto out;
4476
4477         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4478                 rel_ptr = ipath->fspath->val[i] -
4479                           (u64)(unsigned long)ipath->fspath->val;
4480                 ipath->fspath->val[i] = rel_ptr;
4481         }
4482
4483         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4484                            (void *)(unsigned long)ipath->fspath, size);
4485         if (ret) {
4486                 ret = -EFAULT;
4487                 goto out;
4488         }
4489
4490 out:
4491         btrfs_free_path(path);
4492         free_ipath(ipath);
4493         kfree(ipa);
4494
4495         return ret;
4496 }
4497
4498 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4499 {
4500         struct btrfs_data_container *inodes = ctx;
4501         const size_t c = 3 * sizeof(u64);
4502
4503         if (inodes->bytes_left >= c) {
4504                 inodes->bytes_left -= c;
4505                 inodes->val[inodes->elem_cnt] = inum;
4506                 inodes->val[inodes->elem_cnt + 1] = offset;
4507                 inodes->val[inodes->elem_cnt + 2] = root;
4508                 inodes->elem_cnt += 3;
4509         } else {
4510                 inodes->bytes_missing += c - inodes->bytes_left;
4511                 inodes->bytes_left = 0;
4512                 inodes->elem_missed += 3;
4513         }
4514
4515         return 0;
4516 }
4517
4518 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4519                                         void __user *arg)
4520 {
4521         int ret = 0;
4522         int size;
4523         struct btrfs_ioctl_logical_ino_args *loi;
4524         struct btrfs_data_container *inodes = NULL;
4525         struct btrfs_path *path = NULL;
4526
4527         if (!capable(CAP_SYS_ADMIN))
4528                 return -EPERM;
4529
4530         loi = memdup_user(arg, sizeof(*loi));
4531         if (IS_ERR(loi)) {
4532                 ret = PTR_ERR(loi);
4533                 loi = NULL;
4534                 goto out;
4535         }
4536
4537         path = btrfs_alloc_path();
4538         if (!path) {
4539                 ret = -ENOMEM;
4540                 goto out;
4541         }
4542
4543         size = min_t(u32, loi->size, SZ_64K);
4544         inodes = init_data_container(size);
4545         if (IS_ERR(inodes)) {
4546                 ret = PTR_ERR(inodes);
4547                 inodes = NULL;
4548                 goto out;
4549         }
4550
4551         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4552                                           build_ino_list, inodes);
4553         if (ret == -EINVAL)
4554                 ret = -ENOENT;
4555         if (ret < 0)
4556                 goto out;
4557
4558         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4559                            (void *)(unsigned long)inodes, size);
4560         if (ret)
4561                 ret = -EFAULT;
4562
4563 out:
4564         btrfs_free_path(path);
4565         vfree(inodes);
4566         kfree(loi);
4567
4568         return ret;
4569 }
4570
4571 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4572                                struct btrfs_ioctl_balance_args *bargs)
4573 {
4574         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4575
4576         bargs->flags = bctl->flags;
4577
4578         if (atomic_read(&fs_info->balance_running))
4579                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4580         if (atomic_read(&fs_info->balance_pause_req))
4581                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4582         if (atomic_read(&fs_info->balance_cancel_req))
4583                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4584
4585         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4586         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4587         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4588
4589         if (lock) {
4590                 spin_lock(&fs_info->balance_lock);
4591                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4592                 spin_unlock(&fs_info->balance_lock);
4593         } else {
4594                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4595         }
4596 }
4597
4598 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4599 {
4600         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4601         struct btrfs_fs_info *fs_info = root->fs_info;
4602         struct btrfs_ioctl_balance_args *bargs;
4603         struct btrfs_balance_control *bctl;
4604         bool need_unlock; /* for mut. excl. ops lock */
4605         int ret;
4606
4607         if (!capable(CAP_SYS_ADMIN))
4608                 return -EPERM;
4609
4610         ret = mnt_want_write_file(file);
4611         if (ret)
4612                 return ret;
4613
4614 again:
4615         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4616                 mutex_lock(&fs_info->volume_mutex);
4617                 mutex_lock(&fs_info->balance_mutex);
4618                 need_unlock = true;
4619                 goto locked;
4620         }
4621
4622         /*
4623          * mut. excl. ops lock is locked.  Three possibilites:
4624          *   (1) some other op is running
4625          *   (2) balance is running
4626          *   (3) balance is paused -- special case (think resume)
4627          */
4628         mutex_lock(&fs_info->balance_mutex);
4629         if (fs_info->balance_ctl) {
4630                 /* this is either (2) or (3) */
4631                 if (!atomic_read(&fs_info->balance_running)) {
4632                         mutex_unlock(&fs_info->balance_mutex);
4633                         if (!mutex_trylock(&fs_info->volume_mutex))
4634                                 goto again;
4635                         mutex_lock(&fs_info->balance_mutex);
4636
4637                         if (fs_info->balance_ctl &&
4638                             !atomic_read(&fs_info->balance_running)) {
4639                                 /* this is (3) */
4640                                 need_unlock = false;
4641                                 goto locked;
4642                         }
4643
4644                         mutex_unlock(&fs_info->balance_mutex);
4645                         mutex_unlock(&fs_info->volume_mutex);
4646                         goto again;
4647                 } else {
4648                         /* this is (2) */
4649                         mutex_unlock(&fs_info->balance_mutex);
4650                         ret = -EINPROGRESS;
4651                         goto out;
4652                 }
4653         } else {
4654                 /* this is (1) */
4655                 mutex_unlock(&fs_info->balance_mutex);
4656                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4657                 goto out;
4658         }
4659
4660 locked:
4661         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4662
4663         if (arg) {
4664                 bargs = memdup_user(arg, sizeof(*bargs));
4665                 if (IS_ERR(bargs)) {
4666                         ret = PTR_ERR(bargs);
4667                         goto out_unlock;
4668                 }
4669
4670                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4671                         if (!fs_info->balance_ctl) {
4672                                 ret = -ENOTCONN;
4673                                 goto out_bargs;
4674                         }
4675
4676                         bctl = fs_info->balance_ctl;
4677                         spin_lock(&fs_info->balance_lock);
4678                         bctl->flags |= BTRFS_BALANCE_RESUME;
4679                         spin_unlock(&fs_info->balance_lock);
4680
4681                         goto do_balance;
4682                 }
4683         } else {
4684                 bargs = NULL;
4685         }
4686
4687         if (fs_info->balance_ctl) {
4688                 ret = -EINPROGRESS;
4689                 goto out_bargs;
4690         }
4691
4692         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4693         if (!bctl) {
4694                 ret = -ENOMEM;
4695                 goto out_bargs;
4696         }
4697
4698         bctl->fs_info = fs_info;
4699         if (arg) {
4700                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4701                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4702                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4703
4704                 bctl->flags = bargs->flags;
4705         } else {
4706                 /* balance everything - no filters */
4707                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4708         }
4709
4710         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4711                 ret = -EINVAL;
4712                 goto out_bctl;
4713         }
4714
4715 do_balance:
4716         /*
4717          * Ownership of bctl and mutually_exclusive_operation_running
4718          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4719          * or, if restriper was paused all the way until unmount, in
4720          * free_fs_info.  mutually_exclusive_operation_running is
4721          * cleared in __cancel_balance.
4722          */
4723         need_unlock = false;
4724
4725         ret = btrfs_balance(bctl, bargs);
4726         bctl = NULL;
4727
4728         if (arg) {
4729                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4730                         ret = -EFAULT;
4731         }
4732
4733 out_bctl:
4734         kfree(bctl);
4735 out_bargs:
4736         kfree(bargs);
4737 out_unlock:
4738         mutex_unlock(&fs_info->balance_mutex);
4739         mutex_unlock(&fs_info->volume_mutex);
4740         if (need_unlock)
4741                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4742 out:
4743         mnt_drop_write_file(file);
4744         return ret;
4745 }
4746
4747 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4748 {
4749         if (!capable(CAP_SYS_ADMIN))
4750                 return -EPERM;
4751
4752         switch (cmd) {
4753         case BTRFS_BALANCE_CTL_PAUSE:
4754                 return btrfs_pause_balance(root->fs_info);
4755         case BTRFS_BALANCE_CTL_CANCEL:
4756                 return btrfs_cancel_balance(root->fs_info);
4757         }
4758
4759         return -EINVAL;
4760 }
4761
4762 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4763                                          void __user *arg)
4764 {
4765         struct btrfs_fs_info *fs_info = root->fs_info;
4766         struct btrfs_ioctl_balance_args *bargs;
4767         int ret = 0;
4768
4769         if (!capable(CAP_SYS_ADMIN))
4770                 return -EPERM;
4771
4772         mutex_lock(&fs_info->balance_mutex);
4773         if (!fs_info->balance_ctl) {
4774                 ret = -ENOTCONN;
4775                 goto out;
4776         }
4777
4778         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4779         if (!bargs) {
4780                 ret = -ENOMEM;
4781                 goto out;
4782         }
4783
4784         update_ioctl_balance_args(fs_info, 1, bargs);
4785
4786         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4787                 ret = -EFAULT;
4788
4789         kfree(bargs);
4790 out:
4791         mutex_unlock(&fs_info->balance_mutex);
4792         return ret;
4793 }
4794
4795 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4796 {
4797         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4798         struct btrfs_ioctl_quota_ctl_args *sa;
4799         struct btrfs_trans_handle *trans = NULL;
4800         int ret;
4801         int err;
4802
4803         if (!capable(CAP_SYS_ADMIN))
4804                 return -EPERM;
4805
4806         ret = mnt_want_write_file(file);
4807         if (ret)
4808                 return ret;
4809
4810         sa = memdup_user(arg, sizeof(*sa));
4811         if (IS_ERR(sa)) {
4812                 ret = PTR_ERR(sa);
4813                 goto drop_write;
4814         }
4815
4816         down_write(&root->fs_info->subvol_sem);
4817         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4818         if (IS_ERR(trans)) {
4819                 ret = PTR_ERR(trans);
4820                 goto out;
4821         }
4822
4823         switch (sa->cmd) {
4824         case BTRFS_QUOTA_CTL_ENABLE:
4825                 ret = btrfs_quota_enable(trans, root->fs_info);
4826                 break;
4827         case BTRFS_QUOTA_CTL_DISABLE:
4828                 ret = btrfs_quota_disable(trans, root->fs_info);
4829                 break;
4830         default:
4831                 ret = -EINVAL;
4832                 break;
4833         }
4834
4835         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4836         if (err && !ret)
4837                 ret = err;
4838 out:
4839         kfree(sa);
4840         up_write(&root->fs_info->subvol_sem);
4841 drop_write:
4842         mnt_drop_write_file(file);
4843         return ret;
4844 }
4845
4846 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4847 {
4848         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4849         struct btrfs_ioctl_qgroup_assign_args *sa;
4850         struct btrfs_trans_handle *trans;
4851         int ret;
4852         int err;
4853
4854         if (!capable(CAP_SYS_ADMIN))
4855                 return -EPERM;
4856
4857         ret = mnt_want_write_file(file);
4858         if (ret)
4859                 return ret;
4860
4861         sa = memdup_user(arg, sizeof(*sa));
4862         if (IS_ERR(sa)) {
4863                 ret = PTR_ERR(sa);
4864                 goto drop_write;
4865         }
4866
4867         trans = btrfs_join_transaction(root);
4868         if (IS_ERR(trans)) {
4869                 ret = PTR_ERR(trans);
4870                 goto out;
4871         }
4872
4873         /* FIXME: check if the IDs really exist */
4874         if (sa->assign) {
4875                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4876                                                 sa->src, sa->dst);
4877         } else {
4878                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4879                                                 sa->src, sa->dst);
4880         }
4881
4882         /* update qgroup status and info */
4883         err = btrfs_run_qgroups(trans, root->fs_info);
4884         if (err < 0)
4885                 btrfs_std_error(root->fs_info, ret,
4886                             "failed to update qgroup status and info\n");
4887         err = btrfs_end_transaction(trans, root);
4888         if (err && !ret)
4889                 ret = err;
4890
4891 out:
4892         kfree(sa);
4893 drop_write:
4894         mnt_drop_write_file(file);
4895         return ret;
4896 }
4897
4898 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4899 {
4900         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4901         struct btrfs_ioctl_qgroup_create_args *sa;
4902         struct btrfs_trans_handle *trans;
4903         int ret;
4904         int err;
4905
4906         if (!capable(CAP_SYS_ADMIN))
4907                 return -EPERM;
4908
4909         ret = mnt_want_write_file(file);
4910         if (ret)
4911                 return ret;
4912
4913         sa = memdup_user(arg, sizeof(*sa));
4914         if (IS_ERR(sa)) {
4915                 ret = PTR_ERR(sa);
4916                 goto drop_write;
4917         }
4918
4919         if (!sa->qgroupid) {
4920                 ret = -EINVAL;
4921                 goto out;
4922         }
4923
4924         trans = btrfs_join_transaction(root);
4925         if (IS_ERR(trans)) {
4926                 ret = PTR_ERR(trans);
4927                 goto out;
4928         }
4929
4930         /* FIXME: check if the IDs really exist */
4931         if (sa->create) {
4932                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
4933         } else {
4934                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
4935         }
4936
4937         err = btrfs_end_transaction(trans, root);
4938         if (err && !ret)
4939                 ret = err;
4940
4941 out:
4942         kfree(sa);
4943 drop_write:
4944         mnt_drop_write_file(file);
4945         return ret;
4946 }
4947
4948 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4949 {
4950         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4951         struct btrfs_ioctl_qgroup_limit_args *sa;
4952         struct btrfs_trans_handle *trans;
4953         int ret;
4954         int err;
4955         u64 qgroupid;
4956
4957         if (!capable(CAP_SYS_ADMIN))
4958                 return -EPERM;
4959
4960         ret = mnt_want_write_file(file);
4961         if (ret)
4962                 return ret;
4963
4964         sa = memdup_user(arg, sizeof(*sa));
4965         if (IS_ERR(sa)) {
4966                 ret = PTR_ERR(sa);
4967                 goto drop_write;
4968         }
4969
4970         trans = btrfs_join_transaction(root);
4971         if (IS_ERR(trans)) {
4972                 ret = PTR_ERR(trans);
4973                 goto out;
4974         }
4975
4976         qgroupid = sa->qgroupid;
4977         if (!qgroupid) {
4978                 /* take the current subvol as qgroup */
4979                 qgroupid = root->root_key.objectid;
4980         }
4981
4982         /* FIXME: check if the IDs really exist */
4983         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
4984
4985         err = btrfs_end_transaction(trans, root);
4986         if (err && !ret)
4987                 ret = err;
4988
4989 out:
4990         kfree(sa);
4991 drop_write:
4992         mnt_drop_write_file(file);
4993         return ret;
4994 }
4995
4996 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4997 {
4998         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4999         struct btrfs_ioctl_quota_rescan_args *qsa;
5000         int ret;
5001
5002         if (!capable(CAP_SYS_ADMIN))
5003                 return -EPERM;
5004
5005         ret = mnt_want_write_file(file);
5006         if (ret)
5007                 return ret;
5008
5009         qsa = memdup_user(arg, sizeof(*qsa));
5010         if (IS_ERR(qsa)) {
5011                 ret = PTR_ERR(qsa);
5012                 goto drop_write;
5013         }
5014
5015         if (qsa->flags) {
5016                 ret = -EINVAL;
5017                 goto out;
5018         }
5019
5020         ret = btrfs_qgroup_rescan(root->fs_info);
5021
5022 out:
5023         kfree(qsa);
5024 drop_write:
5025         mnt_drop_write_file(file);
5026         return ret;
5027 }
5028
5029 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5030 {
5031         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5032         struct btrfs_ioctl_quota_rescan_args *qsa;
5033         int ret = 0;
5034
5035         if (!capable(CAP_SYS_ADMIN))
5036                 return -EPERM;
5037
5038         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5039         if (!qsa)
5040                 return -ENOMEM;
5041
5042         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5043                 qsa->flags = 1;
5044                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5045         }
5046
5047         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5048                 ret = -EFAULT;
5049
5050         kfree(qsa);
5051         return ret;
5052 }
5053
5054 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5055 {
5056         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5057
5058         if (!capable(CAP_SYS_ADMIN))
5059                 return -EPERM;
5060
5061         return btrfs_qgroup_wait_for_completion(root->fs_info);
5062 }
5063
5064 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5065                                             struct btrfs_ioctl_received_subvol_args *sa)
5066 {
5067         struct inode *inode = file_inode(file);
5068         struct btrfs_root *root = BTRFS_I(inode)->root;
5069         struct btrfs_root_item *root_item = &root->root_item;
5070         struct btrfs_trans_handle *trans;
5071         struct timespec ct = CURRENT_TIME;
5072         int ret = 0;
5073         int received_uuid_changed;
5074
5075         if (!inode_owner_or_capable(inode))
5076                 return -EPERM;
5077
5078         ret = mnt_want_write_file(file);
5079         if (ret < 0)
5080                 return ret;
5081
5082         down_write(&root->fs_info->subvol_sem);
5083
5084         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5085                 ret = -EINVAL;
5086                 goto out;
5087         }
5088
5089         if (btrfs_root_readonly(root)) {
5090                 ret = -EROFS;
5091                 goto out;
5092         }
5093
5094         /*
5095          * 1 - root item
5096          * 2 - uuid items (received uuid + subvol uuid)
5097          */
5098         trans = btrfs_start_transaction(root, 3);
5099         if (IS_ERR(trans)) {
5100                 ret = PTR_ERR(trans);
5101                 trans = NULL;
5102                 goto out;
5103         }
5104
5105         sa->rtransid = trans->transid;
5106         sa->rtime.sec = ct.tv_sec;
5107         sa->rtime.nsec = ct.tv_nsec;
5108
5109         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5110                                        BTRFS_UUID_SIZE);
5111         if (received_uuid_changed &&
5112             !btrfs_is_empty_uuid(root_item->received_uuid))
5113                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5114                                     root_item->received_uuid,
5115                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5116                                     root->root_key.objectid);
5117         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5118         btrfs_set_root_stransid(root_item, sa->stransid);
5119         btrfs_set_root_rtransid(root_item, sa->rtransid);
5120         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5121         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5122         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5123         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5124
5125         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5126                                 &root->root_key, &root->root_item);
5127         if (ret < 0) {
5128                 btrfs_end_transaction(trans, root);
5129                 goto out;
5130         }
5131         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5132                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5133                                           sa->uuid,
5134                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5135                                           root->root_key.objectid);
5136                 if (ret < 0 && ret != -EEXIST) {
5137                         btrfs_abort_transaction(trans, root, ret);
5138                         goto out;
5139                 }
5140         }
5141         ret = btrfs_commit_transaction(trans, root);
5142         if (ret < 0) {
5143                 btrfs_abort_transaction(trans, root, ret);
5144                 goto out;
5145         }
5146
5147 out:
5148         up_write(&root->fs_info->subvol_sem);
5149         mnt_drop_write_file(file);
5150         return ret;
5151 }
5152
5153 #ifdef CONFIG_64BIT
5154 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5155                                                 void __user *arg)
5156 {
5157         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5158         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5159         int ret = 0;
5160
5161         args32 = memdup_user(arg, sizeof(*args32));
5162         if (IS_ERR(args32)) {
5163                 ret = PTR_ERR(args32);
5164                 args32 = NULL;
5165                 goto out;
5166         }
5167
5168         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5169         if (!args64) {
5170                 ret = -ENOMEM;
5171                 goto out;
5172         }
5173
5174         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5175         args64->stransid = args32->stransid;
5176         args64->rtransid = args32->rtransid;
5177         args64->stime.sec = args32->stime.sec;
5178         args64->stime.nsec = args32->stime.nsec;
5179         args64->rtime.sec = args32->rtime.sec;
5180         args64->rtime.nsec = args32->rtime.nsec;
5181         args64->flags = args32->flags;
5182
5183         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5184         if (ret)
5185                 goto out;
5186
5187         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5188         args32->stransid = args64->stransid;
5189         args32->rtransid = args64->rtransid;
5190         args32->stime.sec = args64->stime.sec;
5191         args32->stime.nsec = args64->stime.nsec;
5192         args32->rtime.sec = args64->rtime.sec;
5193         args32->rtime.nsec = args64->rtime.nsec;
5194         args32->flags = args64->flags;
5195
5196         ret = copy_to_user(arg, args32, sizeof(*args32));
5197         if (ret)
5198                 ret = -EFAULT;
5199
5200 out:
5201         kfree(args32);
5202         kfree(args64);
5203         return ret;
5204 }
5205 #endif
5206
5207 static long btrfs_ioctl_set_received_subvol(struct file *file,
5208                                             void __user *arg)
5209 {
5210         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5211         int ret = 0;
5212
5213         sa = memdup_user(arg, sizeof(*sa));
5214         if (IS_ERR(sa)) {
5215                 ret = PTR_ERR(sa);
5216                 sa = NULL;
5217                 goto out;
5218         }
5219
5220         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5221
5222         if (ret)
5223                 goto out;
5224
5225         ret = copy_to_user(arg, sa, sizeof(*sa));
5226         if (ret)
5227                 ret = -EFAULT;
5228
5229 out:
5230         kfree(sa);
5231         return ret;
5232 }
5233
5234 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5235 {
5236         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5237         size_t len;
5238         int ret;
5239         char label[BTRFS_LABEL_SIZE];
5240
5241         spin_lock(&root->fs_info->super_lock);
5242         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5243         spin_unlock(&root->fs_info->super_lock);
5244
5245         len = strnlen(label, BTRFS_LABEL_SIZE);
5246
5247         if (len == BTRFS_LABEL_SIZE) {
5248                 btrfs_warn(root->fs_info,
5249                         "label is too long, return the first %zu bytes", --len);
5250         }
5251
5252         ret = copy_to_user(arg, label, len);
5253
5254         return ret ? -EFAULT : 0;
5255 }
5256
5257 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5258 {
5259         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5260         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5261         struct btrfs_trans_handle *trans;
5262         char label[BTRFS_LABEL_SIZE];
5263         int ret;
5264
5265         if (!capable(CAP_SYS_ADMIN))
5266                 return -EPERM;
5267
5268         if (copy_from_user(label, arg, sizeof(label)))
5269                 return -EFAULT;
5270
5271         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5272                 btrfs_err(root->fs_info, "unable to set label with more than %d bytes",
5273                        BTRFS_LABEL_SIZE - 1);
5274                 return -EINVAL;
5275         }
5276
5277         ret = mnt_want_write_file(file);
5278         if (ret)
5279                 return ret;
5280
5281         trans = btrfs_start_transaction(root, 0);
5282         if (IS_ERR(trans)) {
5283                 ret = PTR_ERR(trans);
5284                 goto out_unlock;
5285         }
5286
5287         spin_lock(&root->fs_info->super_lock);
5288         strcpy(super_block->label, label);
5289         spin_unlock(&root->fs_info->super_lock);
5290         ret = btrfs_commit_transaction(trans, root);
5291
5292 out_unlock:
5293         mnt_drop_write_file(file);
5294         return ret;
5295 }
5296
5297 #define INIT_FEATURE_FLAGS(suffix) \
5298         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5299           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5300           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5301
5302 static int btrfs_ioctl_get_supported_features(struct file *file,
5303                                               void __user *arg)
5304 {
5305         static const struct btrfs_ioctl_feature_flags features[3] = {
5306                 INIT_FEATURE_FLAGS(SUPP),
5307                 INIT_FEATURE_FLAGS(SAFE_SET),
5308                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5309         };
5310
5311         if (copy_to_user(arg, &features, sizeof(features)))
5312                 return -EFAULT;
5313
5314         return 0;
5315 }
5316
5317 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5318 {
5319         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5320         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5321         struct btrfs_ioctl_feature_flags features;
5322
5323         features.compat_flags = btrfs_super_compat_flags(super_block);
5324         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5325         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5326
5327         if (copy_to_user(arg, &features, sizeof(features)))
5328                 return -EFAULT;
5329
5330         return 0;
5331 }
5332
5333 static int check_feature_bits(struct btrfs_root *root,
5334                               enum btrfs_feature_set set,
5335                               u64 change_mask, u64 flags, u64 supported_flags,
5336                               u64 safe_set, u64 safe_clear)
5337 {
5338         const char *type = btrfs_feature_set_names[set];
5339         char *names;
5340         u64 disallowed, unsupported;
5341         u64 set_mask = flags & change_mask;
5342         u64 clear_mask = ~flags & change_mask;
5343
5344         unsupported = set_mask & ~supported_flags;
5345         if (unsupported) {
5346                 names = btrfs_printable_features(set, unsupported);
5347                 if (names) {
5348                         btrfs_warn(root->fs_info,
5349                            "this kernel does not support the %s feature bit%s",
5350                            names, strchr(names, ',') ? "s" : "");
5351                         kfree(names);
5352                 } else
5353                         btrfs_warn(root->fs_info,
5354                            "this kernel does not support %s bits 0x%llx",
5355                            type, unsupported);
5356                 return -EOPNOTSUPP;
5357         }
5358
5359         disallowed = set_mask & ~safe_set;
5360         if (disallowed) {
5361                 names = btrfs_printable_features(set, disallowed);
5362                 if (names) {
5363                         btrfs_warn(root->fs_info,
5364                            "can't set the %s feature bit%s while mounted",
5365                            names, strchr(names, ',') ? "s" : "");
5366                         kfree(names);
5367                 } else
5368                         btrfs_warn(root->fs_info,
5369                            "can't set %s bits 0x%llx while mounted",
5370                            type, disallowed);
5371                 return -EPERM;
5372         }
5373
5374         disallowed = clear_mask & ~safe_clear;
5375         if (disallowed) {
5376                 names = btrfs_printable_features(set, disallowed);
5377                 if (names) {
5378                         btrfs_warn(root->fs_info,
5379                            "can't clear the %s feature bit%s while mounted",
5380                            names, strchr(names, ',') ? "s" : "");
5381                         kfree(names);
5382                 } else
5383                         btrfs_warn(root->fs_info,
5384                            "can't clear %s bits 0x%llx while mounted",
5385                            type, disallowed);
5386                 return -EPERM;
5387         }
5388
5389         return 0;
5390 }
5391
5392 #define check_feature(root, change_mask, flags, mask_base)      \
5393 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5394                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5395                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5396                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5397
5398 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5399 {
5400         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5401         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5402         struct btrfs_ioctl_feature_flags flags[2];
5403         struct btrfs_trans_handle *trans;
5404         u64 newflags;
5405         int ret;
5406
5407         if (!capable(CAP_SYS_ADMIN))
5408                 return -EPERM;
5409
5410         if (copy_from_user(flags, arg, sizeof(flags)))
5411                 return -EFAULT;
5412
5413         /* Nothing to do */
5414         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5415             !flags[0].incompat_flags)
5416                 return 0;
5417
5418         ret = check_feature(root, flags[0].compat_flags,
5419                             flags[1].compat_flags, COMPAT);
5420         if (ret)
5421                 return ret;
5422
5423         ret = check_feature(root, flags[0].compat_ro_flags,
5424                             flags[1].compat_ro_flags, COMPAT_RO);
5425         if (ret)
5426                 return ret;
5427
5428         ret = check_feature(root, flags[0].incompat_flags,
5429                             flags[1].incompat_flags, INCOMPAT);
5430         if (ret)
5431                 return ret;
5432
5433         trans = btrfs_start_transaction(root, 0);
5434         if (IS_ERR(trans))
5435                 return PTR_ERR(trans);
5436
5437         spin_lock(&root->fs_info->super_lock);
5438         newflags = btrfs_super_compat_flags(super_block);
5439         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5440         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5441         btrfs_set_super_compat_flags(super_block, newflags);
5442
5443         newflags = btrfs_super_compat_ro_flags(super_block);
5444         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5445         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5446         btrfs_set_super_compat_ro_flags(super_block, newflags);
5447
5448         newflags = btrfs_super_incompat_flags(super_block);
5449         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5450         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5451         btrfs_set_super_incompat_flags(super_block, newflags);
5452         spin_unlock(&root->fs_info->super_lock);
5453
5454         return btrfs_commit_transaction(trans, root);
5455 }
5456
5457 long btrfs_ioctl(struct file *file, unsigned int
5458                 cmd, unsigned long arg)
5459 {
5460         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5461         void __user *argp = (void __user *)arg;
5462
5463         switch (cmd) {
5464         case FS_IOC_GETFLAGS:
5465                 return btrfs_ioctl_getflags(file, argp);
5466         case FS_IOC_SETFLAGS:
5467                 return btrfs_ioctl_setflags(file, argp);
5468         case FS_IOC_GETVERSION:
5469                 return btrfs_ioctl_getversion(file, argp);
5470         case FITRIM:
5471                 return btrfs_ioctl_fitrim(file, argp);
5472         case BTRFS_IOC_SNAP_CREATE:
5473                 return btrfs_ioctl_snap_create(file, argp, 0);
5474         case BTRFS_IOC_SNAP_CREATE_V2:
5475                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5476         case BTRFS_IOC_SUBVOL_CREATE:
5477                 return btrfs_ioctl_snap_create(file, argp, 1);
5478         case BTRFS_IOC_SUBVOL_CREATE_V2:
5479                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5480         case BTRFS_IOC_SNAP_DESTROY:
5481                 return btrfs_ioctl_snap_destroy(file, argp);
5482         case BTRFS_IOC_SUBVOL_GETFLAGS:
5483                 return btrfs_ioctl_subvol_getflags(file, argp);
5484         case BTRFS_IOC_SUBVOL_SETFLAGS:
5485                 return btrfs_ioctl_subvol_setflags(file, argp);
5486         case BTRFS_IOC_DEFAULT_SUBVOL:
5487                 return btrfs_ioctl_default_subvol(file, argp);
5488         case BTRFS_IOC_DEFRAG:
5489                 return btrfs_ioctl_defrag(file, NULL);
5490         case BTRFS_IOC_DEFRAG_RANGE:
5491                 return btrfs_ioctl_defrag(file, argp);
5492         case BTRFS_IOC_RESIZE:
5493                 return btrfs_ioctl_resize(file, argp);
5494         case BTRFS_IOC_ADD_DEV:
5495                 return btrfs_ioctl_add_dev(root, argp);
5496         case BTRFS_IOC_RM_DEV:
5497                 return btrfs_ioctl_rm_dev(file, argp);
5498         case BTRFS_IOC_FS_INFO:
5499                 return btrfs_ioctl_fs_info(root, argp);
5500         case BTRFS_IOC_DEV_INFO:
5501                 return btrfs_ioctl_dev_info(root, argp);
5502         case BTRFS_IOC_BALANCE:
5503                 return btrfs_ioctl_balance(file, NULL);
5504         case BTRFS_IOC_CLONE:
5505                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
5506         case BTRFS_IOC_CLONE_RANGE:
5507                 return btrfs_ioctl_clone_range(file, argp);
5508         case BTRFS_IOC_TRANS_START:
5509                 return btrfs_ioctl_trans_start(file);
5510         case BTRFS_IOC_TRANS_END:
5511                 return btrfs_ioctl_trans_end(file);
5512         case BTRFS_IOC_TREE_SEARCH:
5513                 return btrfs_ioctl_tree_search(file, argp);
5514         case BTRFS_IOC_TREE_SEARCH_V2:
5515                 return btrfs_ioctl_tree_search_v2(file, argp);
5516         case BTRFS_IOC_INO_LOOKUP:
5517                 return btrfs_ioctl_ino_lookup(file, argp);
5518         case BTRFS_IOC_INO_PATHS:
5519                 return btrfs_ioctl_ino_to_path(root, argp);
5520         case BTRFS_IOC_LOGICAL_INO:
5521                 return btrfs_ioctl_logical_to_ino(root, argp);
5522         case BTRFS_IOC_SPACE_INFO:
5523                 return btrfs_ioctl_space_info(root, argp);
5524         case BTRFS_IOC_SYNC: {
5525                 int ret;
5526
5527                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5528                 if (ret)
5529                         return ret;
5530                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5531                 /*
5532                  * The transaction thread may want to do more work,
5533                  * namely it pokes the cleaner ktread that will start
5534                  * processing uncleaned subvols.
5535                  */
5536                 wake_up_process(root->fs_info->transaction_kthread);
5537                 return ret;
5538         }
5539         case BTRFS_IOC_START_SYNC:
5540                 return btrfs_ioctl_start_sync(root, argp);
5541         case BTRFS_IOC_WAIT_SYNC:
5542                 return btrfs_ioctl_wait_sync(root, argp);
5543         case BTRFS_IOC_SCRUB:
5544                 return btrfs_ioctl_scrub(file, argp);
5545         case BTRFS_IOC_SCRUB_CANCEL:
5546                 return btrfs_ioctl_scrub_cancel(root, argp);
5547         case BTRFS_IOC_SCRUB_PROGRESS:
5548                 return btrfs_ioctl_scrub_progress(root, argp);
5549         case BTRFS_IOC_BALANCE_V2:
5550                 return btrfs_ioctl_balance(file, argp);
5551         case BTRFS_IOC_BALANCE_CTL:
5552                 return btrfs_ioctl_balance_ctl(root, arg);
5553         case BTRFS_IOC_BALANCE_PROGRESS:
5554                 return btrfs_ioctl_balance_progress(root, argp);
5555         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5556                 return btrfs_ioctl_set_received_subvol(file, argp);
5557 #ifdef CONFIG_64BIT
5558         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5559                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5560 #endif
5561         case BTRFS_IOC_SEND:
5562                 return btrfs_ioctl_send(file, argp);
5563         case BTRFS_IOC_GET_DEV_STATS:
5564                 return btrfs_ioctl_get_dev_stats(root, argp);
5565         case BTRFS_IOC_QUOTA_CTL:
5566                 return btrfs_ioctl_quota_ctl(file, argp);
5567         case BTRFS_IOC_QGROUP_ASSIGN:
5568                 return btrfs_ioctl_qgroup_assign(file, argp);
5569         case BTRFS_IOC_QGROUP_CREATE:
5570                 return btrfs_ioctl_qgroup_create(file, argp);
5571         case BTRFS_IOC_QGROUP_LIMIT:
5572                 return btrfs_ioctl_qgroup_limit(file, argp);
5573         case BTRFS_IOC_QUOTA_RESCAN:
5574                 return btrfs_ioctl_quota_rescan(file, argp);
5575         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5576                 return btrfs_ioctl_quota_rescan_status(file, argp);
5577         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5578                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5579         case BTRFS_IOC_DEV_REPLACE:
5580                 return btrfs_ioctl_dev_replace(root, argp);
5581         case BTRFS_IOC_GET_FSLABEL:
5582                 return btrfs_ioctl_get_fslabel(file, argp);
5583         case BTRFS_IOC_SET_FSLABEL:
5584                 return btrfs_ioctl_set_fslabel(file, argp);
5585         case BTRFS_IOC_FILE_EXTENT_SAME:
5586                 return btrfs_ioctl_file_extent_same(file, argp);
5587         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5588                 return btrfs_ioctl_get_supported_features(file, argp);
5589         case BTRFS_IOC_GET_FEATURES:
5590                 return btrfs_ioctl_get_features(file, argp);
5591         case BTRFS_IOC_SET_FEATURES:
5592                 return btrfs_ioctl_set_features(file, argp);
5593         }
5594
5595         return -ENOTTY;
5596 }