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 #include "tree-log.h"
  63 #include "compression.h"
  64
  65 #ifdef CONFIG_64BIT
  66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  67  * structures are incorrect, as the timespec structure from userspace
  68  * is 4 bytes too small. We define these alternatives here to teach
  69  * the kernel about the 32-bit struct packing.
  70  */
  71 struct btrfs_ioctl_timespec_32 {
  72         __u64 sec;
  73         __u32 nsec;
  74 } __attribute__ ((__packed__));
  75
  76 struct btrfs_ioctl_received_subvol_args_32 {
  77         char    uuid[BTRFS_UUID_SIZE];  /* in */
  78         __u64   stransid;               /* in */
  79         __u64   rtransid;               /* out */
  80         struct btrfs_ioctl_timespec_32 stime; /* in */
  81         struct btrfs_ioctl_timespec_32 rtime; /* out */
  82         __u64   flags;                  /* in */
  83         __u64   reserved[16];           /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  87                                 struct btrfs_ioctl_received_subvol_args_32)
  88 #endif
  89
  90
  91 static int btrfs_clone(struct inode *src, struct inode *inode,
  92                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  93                        int no_time_update);
  94
  95 /* Mask out flags that are inappropriate for the given type of inode. */
  96 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  97 {
  98         if (S_ISDIR(mode))
  99                 return flags;
 100         else if (S_ISREG(mode))
 101                 return flags & ~FS_DIRSYNC_FL;
 102         else
 103                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 104 }
 105
 106 /*
 107  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 108  */
 109 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_update_iflags(struct inode *inode)
 140 {
 141         struct btrfs_inode *ip = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (ip->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (ip->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (ip->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (ip->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 /*
 161  * Inherit flags from the parent inode.
 162  *
 163  * Currently only the compression flags and the cow flags are inherited.
 164  */
 165 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 166 {
 167         unsigned int flags;
 168
 169         if (!dir)
 170                 return;
 171
 172         flags = BTRFS_I(dir)->flags;
 173
 174         if (flags & BTRFS_INODE_NOCOMPRESS) {
 175                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 176                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 177         } else if (flags & BTRFS_INODE_COMPRESS) {
 178                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 179                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 180         }
 181
 182         if (flags & BTRFS_INODE_NODATACOW) {
 183                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 184                 if (S_ISREG(inode->i_mode))
 185                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
 186         }
 187
 188         btrfs_update_iflags(inode);
 189 }
 190
 191 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 192 {
 193         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 194         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 195
 196         if (copy_to_user(arg, &flags, sizeof(flags)))
 197                 return -EFAULT;
 198         return 0;
 199 }
 200
 201 static int check_flags(unsigned int flags)
 202 {
 203         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 204                       FS_NOATIME_FL | FS_NODUMP_FL | \
 205                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 206                       FS_NOCOMP_FL | FS_COMPR_FL |
 207                       FS_NOCOW_FL))
 208                 return -EOPNOTSUPP;
 209
 210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 211                 return -EINVAL;
 212
 213         return 0;
 214 }
 215
 216 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 217 {
 218         struct inode *inode = file_inode(file);
 219         struct btrfs_inode *ip = BTRFS_I(inode);
 220         struct btrfs_root *root = ip->root;
 221         struct btrfs_trans_handle *trans;
 222         unsigned int flags, oldflags;
 223         int ret;
 224         u64 ip_oldflags;
 225         unsigned int i_oldflags;
 226         umode_t mode;
 227
 228         if (!inode_owner_or_capable(inode))
 229                 return -EPERM;
 230
 231         if (btrfs_root_readonly(root))
 232                 return -EROFS;
 233
 234         if (copy_from_user(&flags, arg, sizeof(flags)))
 235                 return -EFAULT;
 236
 237         ret = check_flags(flags);
 238         if (ret)
 239                 return ret;
 240
 241         ret = mnt_want_write_file(file);
 242         if (ret)
 243                 return ret;
 244
 245         inode_lock(inode);
 246
 247         ip_oldflags = ip->flags;
 248         i_oldflags = inode->i_flags;
 249         mode = inode->i_mode;
 250
 251         flags = btrfs_mask_flags(inode->i_mode, flags);
 252         oldflags = btrfs_flags_to_ioctl(ip->flags);
 253         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 254                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 255                         ret = -EPERM;
 256                         goto out_unlock;
 257                 }
 258         }
 259
 260         if (flags & FS_SYNC_FL)
 261                 ip->flags |= BTRFS_INODE_SYNC;
 262         else
 263                 ip->flags &= ~BTRFS_INODE_SYNC;
 264         if (flags & FS_IMMUTABLE_FL)
 265                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 266         else
 267                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 268         if (flags & FS_APPEND_FL)
 269                 ip->flags |= BTRFS_INODE_APPEND;
 270         else
 271                 ip->flags &= ~BTRFS_INODE_APPEND;
 272         if (flags & FS_NODUMP_FL)
 273                 ip->flags |= BTRFS_INODE_NODUMP;
 274         else
 275                 ip->flags &= ~BTRFS_INODE_NODUMP;
 276         if (flags & FS_NOATIME_FL)
 277                 ip->flags |= BTRFS_INODE_NOATIME;
 278         else
 279                 ip->flags &= ~BTRFS_INODE_NOATIME;
 280         if (flags & FS_DIRSYNC_FL)
 281                 ip->flags |= BTRFS_INODE_DIRSYNC;
 282         else
 283                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 284         if (flags & FS_NOCOW_FL) {
 285                 if (S_ISREG(mode)) {
 286                         /*
 287                          * It's safe to turn csums off here, no extents exist.
 288                          * Otherwise we want the flag to reflect the real COW
 289                          * status of the file and will not set it.
 290                          */
 291                         if (inode->i_size == 0)
 292                                 ip->flags |= BTRFS_INODE_NODATACOW
 293                                            | BTRFS_INODE_NODATASUM;
 294                 } else {
 295                         ip->flags |= BTRFS_INODE_NODATACOW;
 296                 }
 297         } else {
 298                 /*
 299                  * Revert back under same assumptions as above
 300                  */
 301                 if (S_ISREG(mode)) {
 302                         if (inode->i_size == 0)
 303                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 304                                              | BTRFS_INODE_NODATASUM);
 305                 } else {
 306                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 307                 }
 308         }
 309
 310         /*
 311          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 312          * flag may be changed automatically if compression code won't make
 313          * things smaller.
 314          */
 315         if (flags & FS_NOCOMP_FL) {
 316                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 317                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 318
 319                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 320                 if (ret && ret != -ENODATA)
 321                         goto out_drop;
 322         } else if (flags & FS_COMPR_FL) {
 323                 const char *comp;
 324
 325                 ip->flags |= BTRFS_INODE_COMPRESS;
 326                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 327
 328                 if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
 329                         comp = "lzo";
 330                 else
 331                         comp = "zlib";
 332                 ret = btrfs_set_prop(inode, "btrfs.compression",
 333                                      comp, strlen(comp), 0);
 334                 if (ret)
 335                         goto out_drop;
 336
 337         } else {
 338                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 339                 if (ret && ret != -ENODATA)
 340                         goto out_drop;
 341                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 342         }
 343
 344         trans = btrfs_start_transaction(root, 1);
 345         if (IS_ERR(trans)) {
 346                 ret = PTR_ERR(trans);
 347                 goto out_drop;
 348         }
 349
 350         btrfs_update_iflags(inode);
 351         inode_inc_iversion(inode);
 352         inode->i_ctime = current_time(inode);
 353         ret = btrfs_update_inode(trans, root, inode);
 354
 355         btrfs_end_transaction(trans, root);
 356  out_drop:
 357         if (ret) {
 358                 ip->flags = ip_oldflags;
 359                 inode->i_flags = i_oldflags;
 360         }
 361
 362  out_unlock:
 363         inode_unlock(inode);
 364         mnt_drop_write_file(file);
 365         return ret;
 366 }
 367
 368 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 369 {
 370         struct inode *inode = file_inode(file);
 371
 372         return put_user(inode->i_generation, arg);
 373 }
 374
 375 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 376 {
 377         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
 378         struct btrfs_device *device;
 379         struct request_queue *q;
 380         struct fstrim_range range;
 381         u64 minlen = ULLONG_MAX;
 382         u64 num_devices = 0;
 383         int ret;
 384
 385         if (!capable(CAP_SYS_ADMIN))
 386                 return -EPERM;
 387
 388         /*
 389          * If the fs is mounted with nologreplay, which requires it to be
 390          * mounted in RO mode as well, we can not allow discard on free space
 391          * inside block groups, because log trees refer to extents that are not
 392          * pinned in a block group's free space cache (pinning the extents is
 393          * precisely the first phase of replaying a log tree).
 394          */
 395         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 396                 return -EROFS;
 397
 398         rcu_read_lock();
 399         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 400                                 dev_list) {
 401                 if (!device->bdev)
 402                         continue;
 403                 q = bdev_get_queue(device->bdev);
 404                 if (blk_queue_discard(q)) {
 405                         num_devices++;
 406                         minlen = min((u64)q->limits.discard_granularity,
 407                                      minlen);
 408                 }
 409         }
 410         rcu_read_unlock();
 411
 412         if (!num_devices)
 413                 return -EOPNOTSUPP;
 414         if (copy_from_user(&range, arg, sizeof(range)))
 415                 return -EFAULT;
 416
 417         /*
 418          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 419          * block group is in the logical address space, which can be any
 420          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 421          */
 422         if (range.len < fs_info->sb->s_blocksize)
 423                 return -EINVAL;
 424
 425         range.minlen = max(range.minlen, minlen);
 426         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 427         if (ret < 0)
 428                 return ret;
 429
 430         if (copy_to_user(arg, &range, sizeof(range)))
 431                 return -EFAULT;
 432
 433         return 0;
 434 }
 435
 436 int btrfs_is_empty_uuid(u8 *uuid)
 437 {
 438         int i;
 439
 440         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 441                 if (uuid[i])
 442                         return 0;
 443         }
 444         return 1;
 445 }
 446
 447 static noinline int create_subvol(struct inode *dir,
 448                                   struct dentry *dentry,
 449                                   char *name, int namelen,
 450                                   u64 *async_transid,
 451                                   struct btrfs_qgroup_inherit *inherit)
 452 {
 453         struct btrfs_trans_handle *trans;
 454         struct btrfs_key key;
 455         struct btrfs_root_item *root_item;
 456         struct btrfs_inode_item *inode_item;
 457         struct extent_buffer *leaf;
 458         struct btrfs_root *root = BTRFS_I(dir)->root;
 459         struct btrfs_root *new_root;
 460         struct btrfs_block_rsv block_rsv;
 461         struct timespec cur_time = current_time(dir);
 462         struct inode *inode;
 463         int ret;
 464         int err;
 465         u64 objectid;
 466         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 467         u64 index = 0;
 468         u64 qgroup_reserved;
 469         uuid_le new_uuid;
 470
 471         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 472         if (!root_item)
 473                 return -ENOMEM;
 474
 475         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 476         if (ret)
 477                 goto fail_free;
 478
 479         /*
 480          * Don't create subvolume whose level is not zero. Or qgroup will be
 481          * screwed up since it assumes subvolume qgroup's level to be 0.
 482          */
 483         if (btrfs_qgroup_level(objectid)) {
 484                 ret = -ENOSPC;
 485                 goto fail_free;
 486         }
 487
 488         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 489         /*
 490          * The same as the snapshot creation, please see the comment
 491          * of create_snapshot().
 492          */
 493         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 494                                                8, &qgroup_reserved, false);
 495         if (ret)
 496                 goto fail_free;
 497
 498         trans = btrfs_start_transaction(root, 0);
 499         if (IS_ERR(trans)) {
 500                 ret = PTR_ERR(trans);
 501                 btrfs_subvolume_release_metadata(root, &block_rsv,
 502                                                  qgroup_reserved);
 503                 goto fail_free;
 504         }
 505         trans->block_rsv = &block_rsv;
 506         trans->bytes_reserved = block_rsv.size;
 507
 508         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit);
 509         if (ret)
 510                 goto fail;
 511
 512         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 513         if (IS_ERR(leaf)) {
 514                 ret = PTR_ERR(leaf);
 515                 goto fail;
 516         }
 517
 518         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 519         btrfs_set_header_bytenr(leaf, leaf->start);
 520         btrfs_set_header_generation(leaf, trans->transid);
 521         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 522         btrfs_set_header_owner(leaf, objectid);
 523
 524         write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(),
 525                             BTRFS_FSID_SIZE);
 526         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 527                             btrfs_header_chunk_tree_uuid(leaf),
 528                             BTRFS_UUID_SIZE);
 529         btrfs_mark_buffer_dirty(leaf);
 530
 531         inode_item = &root_item->inode;
 532         btrfs_set_stack_inode_generation(inode_item, 1);
 533         btrfs_set_stack_inode_size(inode_item, 3);
 534         btrfs_set_stack_inode_nlink(inode_item, 1);
 535         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
 536         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 537
 538         btrfs_set_root_flags(root_item, 0);
 539         btrfs_set_root_limit(root_item, 0);
 540         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 541
 542         btrfs_set_root_bytenr(root_item, leaf->start);
 543         btrfs_set_root_generation(root_item, trans->transid);
 544         btrfs_set_root_level(root_item, 0);
 545         btrfs_set_root_refs(root_item, 1);
 546         btrfs_set_root_used(root_item, leaf->len);
 547         btrfs_set_root_last_snapshot(root_item, 0);
 548
 549         btrfs_set_root_generation_v2(root_item,
 550                         btrfs_root_generation(root_item));
 551         uuid_le_gen(&new_uuid);
 552         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 553         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 554         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 555         root_item->ctime = root_item->otime;
 556         btrfs_set_root_ctransid(root_item, trans->transid);
 557         btrfs_set_root_otransid(root_item, trans->transid);
 558
 559         btrfs_tree_unlock(leaf);
 560
 561         btrfs_set_root_dirid(root_item, new_dirid);
 562
 563         key.objectid = objectid;
 564         key.offset = 0;
 565         key.type = BTRFS_ROOT_ITEM_KEY;
 566         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 567                                 root_item);
 568         if (ret) {
 569                 /*
 570                  * Since we don't abort the transaction in this case, free the
 571                  * tree block so that we don't leak space and leave the
 572                  * filesystem in an inconsistent state (an extent item in the
 573                  * extent tree without backreferences). Also no need to have
 574                  * the tree block locked since it is not in any tree at this
 575                  * point, so no other task can find it and use it.
 576                  */
 577                 btrfs_free_tree_block(trans, root, leaf, 0, 1);
 578                 free_extent_buffer(leaf);
 579                 goto fail;
 580         }
 581
 582         free_extent_buffer(leaf);
 583         leaf = NULL;
 584
 585         key.offset = (u64)-1;
 586         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 587         if (IS_ERR(new_root)) {
 588                 ret = PTR_ERR(new_root);
 589                 btrfs_abort_transaction(trans, ret);
 590                 goto fail;
 591         }
 592
 593         btrfs_record_root_in_trans(trans, new_root);
 594
 595         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 596         if (ret) {
 597                 /* We potentially lose an unused inode item here */
 598                 btrfs_abort_transaction(trans, ret);
 599                 goto fail;
 600         }
 601
 602         mutex_lock(&new_root->objectid_mutex);
 603         new_root->highest_objectid = new_dirid;
 604         mutex_unlock(&new_root->objectid_mutex);
 605
 606         /*
 607          * insert the directory item
 608          */
 609         ret = btrfs_set_inode_index(dir, &index);
 610         if (ret) {
 611                 btrfs_abort_transaction(trans, ret);
 612                 goto fail;
 613         }
 614
 615         ret = btrfs_insert_dir_item(trans, root,
 616                                     name, namelen, dir, &key,
 617                                     BTRFS_FT_DIR, index);
 618         if (ret) {
 619                 btrfs_abort_transaction(trans, ret);
 620                 goto fail;
 621         }
 622
 623         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 624         ret = btrfs_update_inode(trans, root, dir);
 625         if (ret) {
 626                 btrfs_abort_transaction(trans, ret);
 627                 goto fail;
 628         }
 629
 630         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 631                                  objectid, root->root_key.objectid,
 632                                  btrfs_ino(dir), index, name, namelen);
 633         if (ret) {
 634                 btrfs_abort_transaction(trans, ret);
 635                 goto fail;
 636         }
 637
 638         ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
 639                                   root_item->uuid, BTRFS_UUID_KEY_SUBVOL,
 640                                   objectid);
 641         if (ret)
 642                 btrfs_abort_transaction(trans, ret);
 643
 644 fail:
 645         kfree(root_item);
 646         trans->block_rsv = NULL;
 647         trans->bytes_reserved = 0;
 648         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
 649
 650         if (async_transid) {
 651                 *async_transid = trans->transid;
 652                 err = btrfs_commit_transaction_async(trans, root, 1);
 653                 if (err)
 654                         err = btrfs_commit_transaction(trans, root);
 655         } else {
 656                 err = btrfs_commit_transaction(trans, root);
 657         }
 658         if (err && !ret)
 659                 ret = err;
 660
 661         if (!ret) {
 662                 inode = btrfs_lookup_dentry(dir, dentry);
 663                 if (IS_ERR(inode))
 664                         return PTR_ERR(inode);
 665                 d_instantiate(dentry, inode);
 666         }
 667         return ret;
 668
 669 fail_free:
 670         kfree(root_item);
 671         return ret;
 672 }
 673
 674 static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root)
 675 {
 676         s64 writers;
 677         DEFINE_WAIT(wait);
 678
 679         do {
 680                 prepare_to_wait(&root->subv_writers->wait, &wait,
 681                                 TASK_UNINTERRUPTIBLE);
 682
 683                 writers = percpu_counter_sum(&root->subv_writers->counter);
 684                 if (writers)
 685                         schedule();
 686
 687                 finish_wait(&root->subv_writers->wait, &wait);
 688         } while (writers);
 689 }
 690
 691 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 692                            struct dentry *dentry, char *name, int namelen,
 693                            u64 *async_transid, bool readonly,
 694                            struct btrfs_qgroup_inherit *inherit)
 695 {
 696         struct inode *inode;
 697         struct btrfs_pending_snapshot *pending_snapshot;
 698         struct btrfs_trans_handle *trans;
 699         int ret;
 700
 701         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 702                 return -EINVAL;
 703
 704         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 705         if (!pending_snapshot)
 706                 return -ENOMEM;
 707
 708         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 709                         GFP_NOFS);
 710         pending_snapshot->path = btrfs_alloc_path();
 711         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 712                 ret = -ENOMEM;
 713                 goto free_pending;
 714         }
 715
 716         atomic_inc(&root->will_be_snapshoted);
 717         smp_mb__after_atomic();
 718         btrfs_wait_for_no_snapshoting_writes(root);
 719
 720         ret = btrfs_start_delalloc_inodes(root, 0);
 721         if (ret)
 722                 goto dec_and_free;
 723
 724         btrfs_wait_ordered_extents(root, -1, 0, (u64)-1);
 725
 726         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 727                              BTRFS_BLOCK_RSV_TEMP);
 728         /*
 729          * 1 - parent dir inode
 730          * 2 - dir entries
 731          * 1 - root item
 732          * 2 - root ref/backref
 733          * 1 - root of snapshot
 734          * 1 - UUID item
 735          */
 736         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 737                                         &pending_snapshot->block_rsv, 8,
 738                                         &pending_snapshot->qgroup_reserved,
 739                                         false);
 740         if (ret)
 741                 goto dec_and_free;
 742
 743         pending_snapshot->dentry = dentry;
 744         pending_snapshot->root = root;
 745         pending_snapshot->readonly = readonly;
 746         pending_snapshot->dir = dir;
 747         pending_snapshot->inherit = inherit;
 748
 749         trans = btrfs_start_transaction(root, 0);
 750         if (IS_ERR(trans)) {
 751                 ret = PTR_ERR(trans);
 752                 goto fail;
 753         }
 754
 755         spin_lock(&root->fs_info->trans_lock);
 756         list_add(&pending_snapshot->list,
 757                  &trans->transaction->pending_snapshots);
 758         spin_unlock(&root->fs_info->trans_lock);
 759         if (async_transid) {
 760                 *async_transid = trans->transid;
 761                 ret = btrfs_commit_transaction_async(trans,
 762                                      root->fs_info->extent_root, 1);
 763                 if (ret)
 764                         ret = btrfs_commit_transaction(trans, root);
 765         } else {
 766                 ret = btrfs_commit_transaction(trans,
 767                                                root->fs_info->extent_root);
 768         }
 769         if (ret)
 770                 goto fail;
 771
 772         ret = pending_snapshot->error;
 773         if (ret)
 774                 goto fail;
 775
 776         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 777         if (ret)
 778                 goto fail;
 779
 780         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 781         if (IS_ERR(inode)) {
 782                 ret = PTR_ERR(inode);
 783                 goto fail;
 784         }
 785
 786         d_instantiate(dentry, inode);
 787         ret = 0;
 788 fail:
 789         btrfs_subvolume_release_metadata(BTRFS_I(dir)->root,
 790                                          &pending_snapshot->block_rsv,
 791                                          pending_snapshot->qgroup_reserved);
 792 dec_and_free:
 793         if (atomic_dec_and_test(&root->will_be_snapshoted))
 794                 wake_up_atomic_t(&root->will_be_snapshoted);
 795 free_pending:
 796         kfree(pending_snapshot->root_item);
 797         btrfs_free_path(pending_snapshot->path);
 798         kfree(pending_snapshot);
 799
 800         return ret;
 801 }
 802
 803 /*  copy of may_delete in fs/namei.c()
 804  *      Check whether we can remove a link victim from directory dir, check
 805  *  whether the type of victim is right.
 806  *  1. We can't do it if dir is read-only (done in permission())
 807  *  2. We should have write and exec permissions on dir
 808  *  3. We can't remove anything from append-only dir
 809  *  4. We can't do anything with immutable dir (done in permission())
 810  *  5. If the sticky bit on dir is set we should either
 811  *      a. be owner of dir, or
 812  *      b. be owner of victim, or
 813  *      c. have CAP_FOWNER capability
 814  *  6. If the victim is append-only or immutable we can't do anything with
 815  *     links pointing to it.
 816  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 817  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 818  *  9. We can't remove a root or mountpoint.
 819  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 820  *     nfs_async_unlink().
 821  */
 822
 823 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 824 {
 825         int error;
 826
 827         if (d_really_is_negative(victim))
 828                 return -ENOENT;
 829
 830         BUG_ON(d_inode(victim->d_parent) != dir);
 831         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 832
 833         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 834         if (error)
 835                 return error;
 836         if (IS_APPEND(dir))
 837                 return -EPERM;
 838         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 839             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 840                 return -EPERM;
 841         if (isdir) {
 842                 if (!d_is_dir(victim))
 843                         return -ENOTDIR;
 844                 if (IS_ROOT(victim))
 845                         return -EBUSY;
 846         } else if (d_is_dir(victim))
 847                 return -EISDIR;
 848         if (IS_DEADDIR(dir))
 849                 return -ENOENT;
 850         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 851                 return -EBUSY;
 852         return 0;
 853 }
 854
 855 /* copy of may_create in fs/namei.c() */
 856 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 857 {
 858         if (d_really_is_positive(child))
 859                 return -EEXIST;
 860         if (IS_DEADDIR(dir))
 861                 return -ENOENT;
 862         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 863 }
 864
 865 /*
 866  * Create a new subvolume below @parent.  This is largely modeled after
 867  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 868  * inside this filesystem so it's quite a bit simpler.
 869  */
 870 static noinline int btrfs_mksubvol(struct path *parent,
 871                                    char *name, int namelen,
 872                                    struct btrfs_root *snap_src,
 873                                    u64 *async_transid, bool readonly,
 874                                    struct btrfs_qgroup_inherit *inherit)
 875 {
 876         struct inode *dir  = d_inode(parent->dentry);
 877         struct dentry *dentry;
 878         int error;
 879
 880         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 881         if (error == -EINTR)
 882                 return error;
 883
 884         dentry = lookup_one_len(name, parent->dentry, namelen);
 885         error = PTR_ERR(dentry);
 886         if (IS_ERR(dentry))
 887                 goto out_unlock;
 888
 889         error = btrfs_may_create(dir, dentry);
 890         if (error)
 891                 goto out_dput;
 892
 893         /*
 894          * even if this name doesn't exist, we may get hash collisions.
 895          * check for them now when we can safely fail
 896          */
 897         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 898                                                dir->i_ino, name,
 899                                                namelen);
 900         if (error)
 901                 goto out_dput;
 902
 903         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 904
 905         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 906                 goto out_up_read;
 907
 908         if (snap_src) {
 909                 error = create_snapshot(snap_src, dir, dentry, name, namelen,
 910                                         async_transid, readonly, inherit);
 911         } else {
 912                 error = create_subvol(dir, dentry, name, namelen,
 913                                       async_transid, inherit);
 914         }
 915         if (!error)
 916                 fsnotify_mkdir(dir, dentry);
 917 out_up_read:
 918         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 919 out_dput:
 920         dput(dentry);
 921 out_unlock:
 922         inode_unlock(dir);
 923         return error;
 924 }
 925
 926 /*
 927  * When we're defragging a range, we don't want to kick it off again
 928  * if it is really just waiting for delalloc to send it down.
 929  * If we find a nice big extent or delalloc range for the bytes in the
 930  * file you want to defrag, we return 0 to let you know to skip this
 931  * part of the file
 932  */
 933 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 934 {
 935         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 936         struct extent_map *em = NULL;
 937         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 938         u64 end;
 939
 940         read_lock(&em_tree->lock);
 941         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 942         read_unlock(&em_tree->lock);
 943
 944         if (em) {
 945                 end = extent_map_end(em);
 946                 free_extent_map(em);
 947                 if (end - offset > thresh)
 948                         return 0;
 949         }
 950         /* if we already have a nice delalloc here, just stop */
 951         thresh /= 2;
 952         end = count_range_bits(io_tree, &offset, offset + thresh,
 953                                thresh, EXTENT_DELALLOC, 1);
 954         if (end >= thresh)
 955                 return 0;
 956         return 1;
 957 }
 958
 959 /*
 960  * helper function to walk through a file and find extents
 961  * newer than a specific transid, and smaller than thresh.
 962  *
 963  * This is used by the defragging code to find new and small
 964  * extents
 965  */
 966 static int find_new_extents(struct btrfs_root *root,
 967                             struct inode *inode, u64 newer_than,
 968                             u64 *off, u32 thresh)
 969 {
 970         struct btrfs_path *path;
 971         struct btrfs_key min_key;
 972         struct extent_buffer *leaf;
 973         struct btrfs_file_extent_item *extent;
 974         int type;
 975         int ret;
 976         u64 ino = btrfs_ino(inode);
 977
 978         path = btrfs_alloc_path();
 979         if (!path)
 980                 return -ENOMEM;
 981
 982         min_key.objectid = ino;
 983         min_key.type = BTRFS_EXTENT_DATA_KEY;
 984         min_key.offset = *off;
 985
 986         while (1) {
 987                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 988                 if (ret != 0)
 989                         goto none;
 990 process_slot:
 991                 if (min_key.objectid != ino)
 992                         goto none;
 993                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 994                         goto none;
 995
 996                 leaf = path->nodes[0];
 997                 extent = btrfs_item_ptr(leaf, path->slots[0],
 998                                         struct btrfs_file_extent_item);
 999
1000                 type = btrfs_file_extent_type(leaf, extent);
1001                 if (type == BTRFS_FILE_EXTENT_REG &&
1002                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1003                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1004                         *off = min_key.offset;
1005                         btrfs_free_path(path);
1006                         return 0;
1007                 }
1008
1009                 path->slots[0]++;
1010                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1011                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1012                         goto process_slot;
1013                 }
1014
1015                 if (min_key.offset == (u64)-1)
1016                         goto none;
1017
1018                 min_key.offset++;
1019                 btrfs_release_path(path);
1020         }
1021 none:
1022         btrfs_free_path(path);
1023         return -ENOENT;
1024 }
1025
1026 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1027 {
1028         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1029         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1030         struct extent_map *em;
1031         u64 len = PAGE_SIZE;
1032
1033         /*
1034          * hopefully we have this extent in the tree already, try without
1035          * the full extent lock
1036          */
1037         read_lock(&em_tree->lock);
1038         em = lookup_extent_mapping(em_tree, start, len);
1039         read_unlock(&em_tree->lock);
1040
1041         if (!em) {
1042                 struct extent_state *cached = NULL;
1043                 u64 end = start + len - 1;
1044
1045                 /* get the big lock and read metadata off disk */
1046                 lock_extent_bits(io_tree, start, end, &cached);
1047                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
1048                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
1049
1050                 if (IS_ERR(em))
1051                         return NULL;
1052         }
1053
1054         return em;
1055 }
1056
1057 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1058 {
1059         struct extent_map *next;
1060         bool ret = true;
1061
1062         /* this is the last extent */
1063         if (em->start + em->len >= i_size_read(inode))
1064                 return false;
1065
1066         next = defrag_lookup_extent(inode, em->start + em->len);
1067         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1068                 ret = false;
1069         else if ((em->block_start + em->block_len == next->block_start) &&
1070                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1071                 ret = false;
1072
1073         free_extent_map(next);
1074         return ret;
1075 }
1076
1077 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1078                                u64 *last_len, u64 *skip, u64 *defrag_end,
1079                                int compress)
1080 {
1081         struct extent_map *em;
1082         int ret = 1;
1083         bool next_mergeable = true;
1084         bool prev_mergeable = true;
1085
1086         /*
1087          * make sure that once we start defragging an extent, we keep on
1088          * defragging it
1089          */
1090         if (start < *defrag_end)
1091                 return 1;
1092
1093         *skip = 0;
1094
1095         em = defrag_lookup_extent(inode, start);
1096         if (!em)
1097                 return 0;
1098
1099         /* this will cover holes, and inline extents */
1100         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1101                 ret = 0;
1102                 goto out;
1103         }
1104
1105         if (!*defrag_end)
1106                 prev_mergeable = false;
1107
1108         next_mergeable = defrag_check_next_extent(inode, em);
1109         /*
1110          * we hit a real extent, if it is big or the next extent is not a
1111          * real extent, don't bother defragging it
1112          */
1113         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1114             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1115                 ret = 0;
1116 out:
1117         /*
1118          * last_len ends up being a counter of how many bytes we've defragged.
1119          * every time we choose not to defrag an extent, we reset *last_len
1120          * so that the next tiny extent will force a defrag.
1121          *
1122          * The end result of this is that tiny extents before a single big
1123          * extent will force at least part of that big extent to be defragged.
1124          */
1125         if (ret) {
1126                 *defrag_end = extent_map_end(em);
1127         } else {
1128                 *last_len = 0;
1129                 *skip = extent_map_end(em);
1130                 *defrag_end = 0;
1131         }
1132
1133         free_extent_map(em);
1134         return ret;
1135 }
1136
1137 /*
1138  * it doesn't do much good to defrag one or two pages
1139  * at a time.  This pulls in a nice chunk of pages
1140  * to COW and defrag.
1141  *
1142  * It also makes sure the delalloc code has enough
1143  * dirty data to avoid making new small extents as part
1144  * of the defrag
1145  *
1146  * It's a good idea to start RA on this range
1147  * before calling this.
1148  */
1149 static int cluster_pages_for_defrag(struct inode *inode,
1150                                     struct page **pages,
1151                                     unsigned long start_index,
1152                                     unsigned long num_pages)
1153 {
1154         unsigned long file_end;
1155         u64 isize = i_size_read(inode);
1156         u64 page_start;
1157         u64 page_end;
1158         u64 page_cnt;
1159         int ret;
1160         int i;
1161         int i_done;
1162         struct btrfs_ordered_extent *ordered;
1163         struct extent_state *cached_state = NULL;
1164         struct extent_io_tree *tree;
1165         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1166
1167         file_end = (isize - 1) >> PAGE_SHIFT;
1168         if (!isize || start_index > file_end)
1169                 return 0;
1170
1171         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1172
1173         ret = btrfs_delalloc_reserve_space(inode,
1174                         start_index << PAGE_SHIFT,
1175                         page_cnt << PAGE_SHIFT);
1176         if (ret)
1177                 return ret;
1178         i_done = 0;
1179         tree = &BTRFS_I(inode)->io_tree;
1180
1181         /* step one, lock all the pages */
1182         for (i = 0; i < page_cnt; i++) {
1183                 struct page *page;
1184 again:
1185                 page = find_or_create_page(inode->i_mapping,
1186                                            start_index + i, mask);
1187                 if (!page)
1188                         break;
1189
1190                 page_start = page_offset(page);
1191                 page_end = page_start + PAGE_SIZE - 1;
1192                 while (1) {
1193                         lock_extent_bits(tree, page_start, page_end,
1194                                          &cached_state);
1195                         ordered = btrfs_lookup_ordered_extent(inode,
1196                                                               page_start);
1197                         unlock_extent_cached(tree, page_start, page_end,
1198                                              &cached_state, GFP_NOFS);
1199                         if (!ordered)
1200                                 break;
1201
1202                         unlock_page(page);
1203                         btrfs_start_ordered_extent(inode, ordered, 1);
1204                         btrfs_put_ordered_extent(ordered);
1205                         lock_page(page);
1206                         /*
1207                          * we unlocked the page above, so we need check if
1208                          * it was released or not.
1209                          */
1210                         if (page->mapping != inode->i_mapping) {
1211                                 unlock_page(page);
1212                                 put_page(page);
1213                                 goto again;
1214                         }
1215                 }
1216
1217                 if (!PageUptodate(page)) {
1218                         btrfs_readpage(NULL, page);
1219                         lock_page(page);
1220                         if (!PageUptodate(page)) {
1221                                 unlock_page(page);
1222                                 put_page(page);
1223                                 ret = -EIO;
1224                                 break;
1225                         }
1226                 }
1227
1228                 if (page->mapping != inode->i_mapping) {
1229                         unlock_page(page);
1230                         put_page(page);
1231                         goto again;
1232                 }
1233
1234                 pages[i] = page;
1235                 i_done++;
1236         }
1237         if (!i_done || ret)
1238                 goto out;
1239
1240         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1241                 goto out;
1242
1243         /*
1244          * so now we have a nice long stream of locked
1245          * and up to date pages, lets wait on them
1246          */
1247         for (i = 0; i < i_done; i++)
1248                 wait_on_page_writeback(pages[i]);
1249
1250         page_start = page_offset(pages[0]);
1251         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1252
1253         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1254                          page_start, page_end - 1, &cached_state);
1255         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1256                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1257                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1258                           &cached_state, GFP_NOFS);
1259
1260         if (i_done != page_cnt) {
1261                 spin_lock(&BTRFS_I(inode)->lock);
1262                 BTRFS_I(inode)->outstanding_extents++;
1263                 spin_unlock(&BTRFS_I(inode)->lock);
1264                 btrfs_delalloc_release_space(inode,
1265                                 start_index << PAGE_SHIFT,
1266                                 (page_cnt - i_done) << PAGE_SHIFT);
1267         }
1268
1269
1270         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1271                           &cached_state);
1272
1273         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1274                              page_start, page_end - 1, &cached_state,
1275                              GFP_NOFS);
1276
1277         for (i = 0; i < i_done; i++) {
1278                 clear_page_dirty_for_io(pages[i]);
1279                 ClearPageChecked(pages[i]);
1280                 set_page_extent_mapped(pages[i]);
1281                 set_page_dirty(pages[i]);
1282                 unlock_page(pages[i]);
1283                 put_page(pages[i]);
1284         }
1285         return i_done;
1286 out:
1287         for (i = 0; i < i_done; i++) {
1288                 unlock_page(pages[i]);
1289                 put_page(pages[i]);
1290         }
1291         btrfs_delalloc_release_space(inode,
1292                         start_index << PAGE_SHIFT,
1293                         page_cnt << PAGE_SHIFT);
1294         return ret;
1295
1296 }
1297
1298 int btrfs_defrag_file(struct inode *inode, struct file *file,
1299                       struct btrfs_ioctl_defrag_range_args *range,
1300                       u64 newer_than, unsigned long max_to_defrag)
1301 {
1302         struct btrfs_root *root = BTRFS_I(inode)->root;
1303         struct file_ra_state *ra = NULL;
1304         unsigned long last_index;
1305         u64 isize = i_size_read(inode);
1306         u64 last_len = 0;
1307         u64 skip = 0;
1308         u64 defrag_end = 0;
1309         u64 newer_off = range->start;
1310         unsigned long i;
1311         unsigned long ra_index = 0;
1312         int ret;
1313         int defrag_count = 0;
1314         int compress_type = BTRFS_COMPRESS_ZLIB;
1315         u32 extent_thresh = range->extent_thresh;
1316         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1317         unsigned long cluster = max_cluster;
1318         u64 new_align = ~((u64)SZ_128K - 1);
1319         struct page **pages = NULL;
1320
1321         if (isize == 0)
1322                 return 0;
1323
1324         if (range->start >= isize)
1325                 return -EINVAL;
1326
1327         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1328                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1329                         return -EINVAL;
1330                 if (range->compress_type)
1331                         compress_type = range->compress_type;
1332         }
1333
1334         if (extent_thresh == 0)
1335                 extent_thresh = SZ_256K;
1336
1337         /*
1338          * if we were not given a file, allocate a readahead
1339          * context
1340          */
1341         if (!file) {
1342                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1343                 if (!ra)
1344                         return -ENOMEM;
1345                 file_ra_state_init(ra, inode->i_mapping);
1346         } else {
1347                 ra = &file->f_ra;
1348         }
1349
1350         pages = kmalloc_array(max_cluster, sizeof(struct page *),
1351                         GFP_NOFS);
1352         if (!pages) {
1353                 ret = -ENOMEM;
1354                 goto out_ra;
1355         }
1356
1357         /* find the last page to defrag */
1358         if (range->start + range->len > range->start) {
1359                 last_index = min_t(u64, isize - 1,
1360                          range->start + range->len - 1) >> PAGE_SHIFT;
1361         } else {
1362                 last_index = (isize - 1) >> PAGE_SHIFT;
1363         }
1364
1365         if (newer_than) {
1366                 ret = find_new_extents(root, inode, newer_than,
1367                                        &newer_off, SZ_64K);
1368                 if (!ret) {
1369                         range->start = newer_off;
1370                         /*
1371                          * we always align our defrag to help keep
1372                          * the extents in the file evenly spaced
1373                          */
1374                         i = (newer_off & new_align) >> PAGE_SHIFT;
1375                 } else
1376                         goto out_ra;
1377         } else {
1378                 i = range->start >> PAGE_SHIFT;
1379         }
1380         if (!max_to_defrag)
1381                 max_to_defrag = last_index - i + 1;
1382
1383         /*
1384          * make writeback starts from i, so the defrag range can be
1385          * written sequentially.
1386          */
1387         if (i < inode->i_mapping->writeback_index)
1388                 inode->i_mapping->writeback_index = i;
1389
1390         while (i <= last_index && defrag_count < max_to_defrag &&
1391                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1392                 /*
1393                  * make sure we stop running if someone unmounts
1394                  * the FS
1395                  */
1396                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1397                         break;
1398
1399                 if (btrfs_defrag_cancelled(root->fs_info)) {
1400                         btrfs_debug(root->fs_info, "defrag_file cancelled");
1401                         ret = -EAGAIN;
1402                         break;
1403                 }
1404
1405                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1406                                          extent_thresh, &last_len, &skip,
1407                                          &defrag_end, range->flags &
1408                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1409                         unsigned long next;
1410                         /*
1411                          * the should_defrag function tells us how much to skip
1412                          * bump our counter by the suggested amount
1413                          */
1414                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1415                         i = max(i + 1, next);
1416                         continue;
1417                 }
1418
1419                 if (!newer_than) {
1420                         cluster = (PAGE_ALIGN(defrag_end) >>
1421                                    PAGE_SHIFT) - i;
1422                         cluster = min(cluster, max_cluster);
1423                 } else {
1424                         cluster = max_cluster;
1425                 }
1426
1427                 if (i + cluster > ra_index) {
1428                         ra_index = max(i, ra_index);
1429                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1430                                        cluster);
1431                         ra_index += cluster;
1432                 }
1433
1434                 inode_lock(inode);
1435                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1436                         BTRFS_I(inode)->force_compress = compress_type;
1437                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1438                 if (ret < 0) {
1439                         inode_unlock(inode);
1440                         goto out_ra;
1441                 }
1442
1443                 defrag_count += ret;
1444                 balance_dirty_pages_ratelimited(inode->i_mapping);
1445                 inode_unlock(inode);
1446
1447                 if (newer_than) {
1448                         if (newer_off == (u64)-1)
1449                                 break;
1450
1451                         if (ret > 0)
1452                                 i += ret;
1453
1454                         newer_off = max(newer_off + 1,
1455                                         (u64)i << PAGE_SHIFT);
1456
1457                         ret = find_new_extents(root, inode, newer_than,
1458                                                &newer_off, SZ_64K);
1459                         if (!ret) {
1460                                 range->start = newer_off;
1461                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1462                         } else {
1463                                 break;
1464                         }
1465                 } else {
1466                         if (ret > 0) {
1467                                 i += ret;
1468                                 last_len += ret << PAGE_SHIFT;
1469                         } else {
1470                                 i++;
1471                                 last_len = 0;
1472                         }
1473                 }
1474         }
1475
1476         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1477                 filemap_flush(inode->i_mapping);
1478                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1479                              &BTRFS_I(inode)->runtime_flags))
1480                         filemap_flush(inode->i_mapping);
1481         }
1482
1483         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1484                 /* the filemap_flush will queue IO into the worker threads, but
1485                  * we have to make sure the IO is actually started and that
1486                  * ordered extents get created before we return
1487                  */
1488                 atomic_inc(&root->fs_info->async_submit_draining);
1489                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1490                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1491                         wait_event(root->fs_info->async_submit_wait,
1492                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1493                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1494                 }
1495                 atomic_dec(&root->fs_info->async_submit_draining);
1496         }
1497
1498         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1499                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1500         }
1501
1502         ret = defrag_count;
1503
1504 out_ra:
1505         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1506                 inode_lock(inode);
1507                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1508                 inode_unlock(inode);
1509         }
1510         if (!file)
1511                 kfree(ra);
1512         kfree(pages);
1513         return ret;
1514 }
1515
1516 static noinline int btrfs_ioctl_resize(struct file *file,
1517                                         void __user *arg)
1518 {
1519         u64 new_size;
1520         u64 old_size;
1521         u64 devid = 1;
1522         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
1523         struct btrfs_ioctl_vol_args *vol_args;
1524         struct btrfs_trans_handle *trans;
1525         struct btrfs_device *device = NULL;
1526         char *sizestr;
1527         char *retptr;
1528         char *devstr = NULL;
1529         int ret = 0;
1530         int mod = 0;
1531
1532         if (!capable(CAP_SYS_ADMIN))
1533                 return -EPERM;
1534
1535         ret = mnt_want_write_file(file);
1536         if (ret)
1537                 return ret;
1538
1539         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1540                         1)) {
1541                 mnt_drop_write_file(file);
1542                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1543         }
1544
1545         mutex_lock(&root->fs_info->volume_mutex);
1546         vol_args = memdup_user(arg, sizeof(*vol_args));
1547         if (IS_ERR(vol_args)) {
1548                 ret = PTR_ERR(vol_args);
1549                 goto out;
1550         }
1551
1552         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1553
1554         sizestr = vol_args->name;
1555         devstr = strchr(sizestr, ':');
1556         if (devstr) {
1557                 sizestr = devstr + 1;
1558                 *devstr = '\0';
1559                 devstr = vol_args->name;
1560                 ret = kstrtoull(devstr, 10, &devid);
1561                 if (ret)
1562                         goto out_free;
1563                 if (!devid) {
1564                         ret = -EINVAL;
1565                         goto out_free;
1566                 }
1567                 btrfs_info(root->fs_info, "resizing devid %llu", devid);
1568         }
1569
1570         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1571         if (!device) {
1572                 btrfs_info(root->fs_info, "resizer unable to find device %llu",
1573                        devid);
1574                 ret = -ENODEV;
1575                 goto out_free;
1576         }
1577
1578         if (!device->writeable) {
1579                 btrfs_info(root->fs_info,
1580                            "resizer unable to apply on readonly device %llu",
1581                        devid);
1582                 ret = -EPERM;
1583                 goto out_free;
1584         }
1585
1586         if (!strcmp(sizestr, "max"))
1587                 new_size = device->bdev->bd_inode->i_size;
1588         else {
1589                 if (sizestr[0] == '-') {
1590                         mod = -1;
1591                         sizestr++;
1592                 } else if (sizestr[0] == '+') {
1593                         mod = 1;
1594                         sizestr++;
1595                 }
1596                 new_size = memparse(sizestr, &retptr);
1597                 if (*retptr != '\0' || new_size == 0) {
1598                         ret = -EINVAL;
1599                         goto out_free;
1600                 }
1601         }
1602
1603         if (device->is_tgtdev_for_dev_replace) {
1604                 ret = -EPERM;
1605                 goto out_free;
1606         }
1607
1608         old_size = btrfs_device_get_total_bytes(device);
1609
1610         if (mod < 0) {
1611                 if (new_size > old_size) {
1612                         ret = -EINVAL;
1613                         goto out_free;
1614                 }
1615                 new_size = old_size - new_size;
1616         } else if (mod > 0) {
1617                 if (new_size > ULLONG_MAX - old_size) {
1618                         ret = -ERANGE;
1619                         goto out_free;
1620                 }
1621                 new_size = old_size + new_size;
1622         }
1623
1624         if (new_size < SZ_256M) {
1625                 ret = -EINVAL;
1626                 goto out_free;
1627         }
1628         if (new_size > device->bdev->bd_inode->i_size) {
1629                 ret = -EFBIG;
1630                 goto out_free;
1631         }
1632
1633         new_size = div_u64(new_size, root->sectorsize);
1634         new_size *= root->sectorsize;
1635
1636         btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu",
1637                       rcu_str_deref(device->name), new_size);
1638
1639         if (new_size > old_size) {
1640                 trans = btrfs_start_transaction(root, 0);
1641                 if (IS_ERR(trans)) {
1642                         ret = PTR_ERR(trans);
1643                         goto out_free;
1644                 }
1645                 ret = btrfs_grow_device(trans, device, new_size);
1646                 btrfs_commit_transaction(trans, root);
1647         } else if (new_size < old_size) {
1648                 ret = btrfs_shrink_device(device, new_size);
1649         } /* equal, nothing need to do */
1650
1651 out_free:
1652         kfree(vol_args);
1653 out:
1654         mutex_unlock(&root->fs_info->volume_mutex);
1655         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1656         mnt_drop_write_file(file);
1657         return ret;
1658 }
1659
1660 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1661                                 char *name, unsigned long fd, int subvol,
1662                                 u64 *transid, bool readonly,
1663                                 struct btrfs_qgroup_inherit *inherit)
1664 {
1665         int namelen;
1666         int ret = 0;
1667
1668         if (!S_ISDIR(file_inode(file)->i_mode))
1669                 return -ENOTDIR;
1670
1671         ret = mnt_want_write_file(file);
1672         if (ret)
1673                 goto out;
1674
1675         namelen = strlen(name);
1676         if (strchr(name, '/')) {
1677                 ret = -EINVAL;
1678                 goto out_drop_write;
1679         }
1680
1681         if (name[0] == '.' &&
1682            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1683                 ret = -EEXIST;
1684                 goto out_drop_write;
1685         }
1686
1687         if (subvol) {
1688                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1689                                      NULL, transid, readonly, inherit);
1690         } else {
1691                 struct fd src = fdget(fd);
1692                 struct inode *src_inode;
1693                 if (!src.file) {
1694                         ret = -EINVAL;
1695                         goto out_drop_write;
1696                 }
1697
1698                 src_inode = file_inode(src.file);
1699                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1700                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1701                                    "Snapshot src from another FS");
1702                         ret = -EXDEV;
1703                 } else if (!inode_owner_or_capable(src_inode)) {
1704                         /*
1705                          * Subvolume creation is not restricted, but snapshots
1706                          * are limited to own subvolumes only
1707                          */
1708                         ret = -EPERM;
1709                 } else {
1710                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1711                                              BTRFS_I(src_inode)->root,
1712                                              transid, readonly, inherit);
1713                 }
1714                 fdput(src);
1715         }
1716 out_drop_write:
1717         mnt_drop_write_file(file);
1718 out:
1719         return ret;
1720 }
1721
1722 static noinline int btrfs_ioctl_snap_create(struct file *file,
1723                                             void __user *arg, int subvol)
1724 {
1725         struct btrfs_ioctl_vol_args *vol_args;
1726         int ret;
1727
1728         if (!S_ISDIR(file_inode(file)->i_mode))
1729                 return -ENOTDIR;
1730
1731         vol_args = memdup_user(arg, sizeof(*vol_args));
1732         if (IS_ERR(vol_args))
1733                 return PTR_ERR(vol_args);
1734         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1735
1736         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1737                                               vol_args->fd, subvol,
1738                                               NULL, false, NULL);
1739
1740         kfree(vol_args);
1741         return ret;
1742 }
1743
1744 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1745                                                void __user *arg, int subvol)
1746 {
1747         struct btrfs_ioctl_vol_args_v2 *vol_args;
1748         int ret;
1749         u64 transid = 0;
1750         u64 *ptr = NULL;
1751         bool readonly = false;
1752         struct btrfs_qgroup_inherit *inherit = NULL;
1753
1754         if (!S_ISDIR(file_inode(file)->i_mode))
1755                 return -ENOTDIR;
1756
1757         vol_args = memdup_user(arg, sizeof(*vol_args));
1758         if (IS_ERR(vol_args))
1759                 return PTR_ERR(vol_args);
1760         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1761
1762         if (vol_args->flags &
1763             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1764               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1765                 ret = -EOPNOTSUPP;
1766                 goto free_args;
1767         }
1768
1769         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1770                 ptr = &transid;
1771         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1772                 readonly = true;
1773         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1774                 if (vol_args->size > PAGE_SIZE) {
1775                         ret = -EINVAL;
1776                         goto free_args;
1777                 }
1778                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1779                 if (IS_ERR(inherit)) {
1780                         ret = PTR_ERR(inherit);
1781                         goto free_args;
1782                 }
1783         }
1784
1785         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1786                                               vol_args->fd, subvol, ptr,
1787                                               readonly, inherit);
1788         if (ret)
1789                 goto free_inherit;
1790
1791         if (ptr && copy_to_user(arg +
1792                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1793                                         transid),
1794                                 ptr, sizeof(*ptr)))
1795                 ret = -EFAULT;
1796
1797 free_inherit:
1798         kfree(inherit);
1799 free_args:
1800         kfree(vol_args);
1801         return ret;
1802 }
1803
1804 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1805                                                 void __user *arg)
1806 {
1807         struct inode *inode = file_inode(file);
1808         struct btrfs_root *root = BTRFS_I(inode)->root;
1809         int ret = 0;
1810         u64 flags = 0;
1811
1812         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1813                 return -EINVAL;
1814
1815         down_read(&root->fs_info->subvol_sem);
1816         if (btrfs_root_readonly(root))
1817                 flags |= BTRFS_SUBVOL_RDONLY;
1818         up_read(&root->fs_info->subvol_sem);
1819
1820         if (copy_to_user(arg, &flags, sizeof(flags)))
1821                 ret = -EFAULT;
1822
1823         return ret;
1824 }
1825
1826 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1827                                               void __user *arg)
1828 {
1829         struct inode *inode = file_inode(file);
1830         struct btrfs_root *root = BTRFS_I(inode)->root;
1831         struct btrfs_trans_handle *trans;
1832         u64 root_flags;
1833         u64 flags;
1834         int ret = 0;
1835
1836         if (!inode_owner_or_capable(inode))
1837                 return -EPERM;
1838
1839         ret = mnt_want_write_file(file);
1840         if (ret)
1841                 goto out;
1842
1843         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1844                 ret = -EINVAL;
1845                 goto out_drop_write;
1846         }
1847
1848         if (copy_from_user(&flags, arg, sizeof(flags))) {
1849                 ret = -EFAULT;
1850                 goto out_drop_write;
1851         }
1852
1853         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1854                 ret = -EINVAL;
1855                 goto out_drop_write;
1856         }
1857
1858         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1859                 ret = -EOPNOTSUPP;
1860                 goto out_drop_write;
1861         }
1862
1863         down_write(&root->fs_info->subvol_sem);
1864
1865         /* nothing to do */
1866         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1867                 goto out_drop_sem;
1868
1869         root_flags = btrfs_root_flags(&root->root_item);
1870         if (flags & BTRFS_SUBVOL_RDONLY) {
1871                 btrfs_set_root_flags(&root->root_item,
1872                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1873         } else {
1874                 /*
1875                  * Block RO -> RW transition if this subvolume is involved in
1876                  * send
1877                  */
1878                 spin_lock(&root->root_item_lock);
1879                 if (root->send_in_progress == 0) {
1880                         btrfs_set_root_flags(&root->root_item,
1881                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1882                         spin_unlock(&root->root_item_lock);
1883                 } else {
1884                         spin_unlock(&root->root_item_lock);
1885                         btrfs_warn(root->fs_info,
1886                         "Attempt to set subvolume %llu read-write during send",
1887                                         root->root_key.objectid);
1888                         ret = -EPERM;
1889                         goto out_drop_sem;
1890                 }
1891         }
1892
1893         trans = btrfs_start_transaction(root, 1);
1894         if (IS_ERR(trans)) {
1895                 ret = PTR_ERR(trans);
1896                 goto out_reset;
1897         }
1898
1899         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1900                                 &root->root_key, &root->root_item);
1901
1902         btrfs_commit_transaction(trans, root);
1903 out_reset:
1904         if (ret)
1905                 btrfs_set_root_flags(&root->root_item, root_flags);
1906 out_drop_sem:
1907         up_write(&root->fs_info->subvol_sem);
1908 out_drop_write:
1909         mnt_drop_write_file(file);
1910 out:
1911         return ret;
1912 }
1913
1914 /*
1915  * helper to check if the subvolume references other subvolumes
1916  */
1917 static noinline int may_destroy_subvol(struct btrfs_root *root)
1918 {
1919         struct btrfs_path *path;
1920         struct btrfs_dir_item *di;
1921         struct btrfs_key key;
1922         u64 dir_id;
1923         int ret;
1924
1925         path = btrfs_alloc_path();
1926         if (!path)
1927                 return -ENOMEM;
1928
1929         /* Make sure this root isn't set as the default subvol */
1930         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
1931         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path,
1932                                    dir_id, "default", 7, 0);
1933         if (di && !IS_ERR(di)) {
1934                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1935                 if (key.objectid == root->root_key.objectid) {
1936                         ret = -EPERM;
1937                         btrfs_err(root->fs_info,
1938                                   "deleting default subvolume %llu is not allowed",
1939                                   key.objectid);
1940                         goto out;
1941                 }
1942                 btrfs_release_path(path);
1943         }
1944
1945         key.objectid = root->root_key.objectid;
1946         key.type = BTRFS_ROOT_REF_KEY;
1947         key.offset = (u64)-1;
1948
1949         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1950                                 &key, path, 0, 0);
1951         if (ret < 0)
1952                 goto out;
1953         BUG_ON(ret == 0);
1954
1955         ret = 0;
1956         if (path->slots[0] > 0) {
1957                 path->slots[0]--;
1958                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1959                 if (key.objectid == root->root_key.objectid &&
1960                     key.type == BTRFS_ROOT_REF_KEY)
1961                         ret = -ENOTEMPTY;
1962         }
1963 out:
1964         btrfs_free_path(path);
1965         return ret;
1966 }
1967
1968 static noinline int key_in_sk(struct btrfs_key *key,
1969                               struct btrfs_ioctl_search_key *sk)
1970 {
1971         struct btrfs_key test;
1972         int ret;
1973
1974         test.objectid = sk->min_objectid;
1975         test.type = sk->min_type;
1976         test.offset = sk->min_offset;
1977
1978         ret = btrfs_comp_cpu_keys(key, &test);
1979         if (ret < 0)
1980                 return 0;
1981
1982         test.objectid = sk->max_objectid;
1983         test.type = sk->max_type;
1984         test.offset = sk->max_offset;
1985
1986         ret = btrfs_comp_cpu_keys(key, &test);
1987         if (ret > 0)
1988                 return 0;
1989         return 1;
1990 }
1991
1992 static noinline int copy_to_sk(struct btrfs_path *path,
1993                                struct btrfs_key *key,
1994                                struct btrfs_ioctl_search_key *sk,
1995                                size_t *buf_size,
1996                                char __user *ubuf,
1997                                unsigned long *sk_offset,
1998                                int *num_found)
1999 {
2000         u64 found_transid;
2001         struct extent_buffer *leaf;
2002         struct btrfs_ioctl_search_header sh;
2003         struct btrfs_key test;
2004         unsigned long item_off;
2005         unsigned long item_len;
2006         int nritems;
2007         int i;
2008         int slot;
2009         int ret = 0;
2010
2011         leaf = path->nodes[0];
2012         slot = path->slots[0];
2013         nritems = btrfs_header_nritems(leaf);
2014
2015         if (btrfs_header_generation(leaf) > sk->max_transid) {
2016                 i = nritems;
2017                 goto advance_key;
2018         }
2019         found_transid = btrfs_header_generation(leaf);
2020
2021         for (i = slot; i < nritems; i++) {
2022                 item_off = btrfs_item_ptr_offset(leaf, i);
2023                 item_len = btrfs_item_size_nr(leaf, i);
2024
2025                 btrfs_item_key_to_cpu(leaf, key, i);
2026                 if (!key_in_sk(key, sk))
2027                         continue;
2028
2029                 if (sizeof(sh) + item_len > *buf_size) {
2030                         if (*num_found) {
2031                                 ret = 1;
2032                                 goto out;
2033                         }
2034
2035                         /*
2036                          * return one empty item back for v1, which does not
2037                          * handle -EOVERFLOW
2038                          */
2039
2040                         *buf_size = sizeof(sh) + item_len;
2041                         item_len = 0;
2042                         ret = -EOVERFLOW;
2043                 }
2044
2045                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2046                         ret = 1;
2047                         goto out;
2048                 }
2049
2050                 sh.objectid = key->objectid;
2051                 sh.offset = key->offset;
2052                 sh.type = key->type;
2053                 sh.len = item_len;
2054                 sh.transid = found_transid;
2055
2056                 /*
2057                  * Copy search result header. If we fault then loop again so we
2058                  * can fault in the pages and -EFAULT there if there's a
2059                  * problem. Otherwise we'll fault and then copy the buffer in
2060                  * properly this next time through
2061                  */
2062                 if (probe_user_write(ubuf + *sk_offset, &sh, sizeof(sh))) {
2063                         ret = 0;
2064                         goto out;
2065                 }
2066
2067                 *sk_offset += sizeof(sh);
2068
2069                 if (item_len) {
2070                         char __user *up = ubuf + *sk_offset;
2071                         /*
2072                          * Copy the item, same behavior as above, but reset the
2073                          * * sk_offset so we copy the full thing again.
2074                          */
2075                         if (read_extent_buffer_to_user_nofault(leaf, up,
2076                                                 item_off, item_len)) {
2077                                 ret = 0;
2078                                 *sk_offset -= sizeof(sh);
2079                                 goto out;
2080                         }
2081
2082                         *sk_offset += item_len;
2083                 }
2084                 (*num_found)++;
2085
2086                 if (ret) /* -EOVERFLOW from above */
2087                         goto out;
2088
2089                 if (*num_found >= sk->nr_items) {
2090                         ret = 1;
2091                         goto out;
2092                 }
2093         }
2094 advance_key:
2095         ret = 0;
2096         test.objectid = sk->max_objectid;
2097         test.type = sk->max_type;
2098         test.offset = sk->max_offset;
2099         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2100                 ret = 1;
2101         else if (key->offset < (u64)-1)
2102                 key->offset++;
2103         else if (key->type < (u8)-1) {
2104                 key->offset = 0;
2105                 key->type++;
2106         } else if (key->objectid < (u64)-1) {
2107                 key->offset = 0;
2108                 key->type = 0;
2109                 key->objectid++;
2110         } else
2111                 ret = 1;
2112 out:
2113         /*
2114          *  0: all items from this leaf copied, continue with next
2115          *  1: * more items can be copied, but unused buffer is too small
2116          *     * all items were found
2117          *     Either way, it will stops the loop which iterates to the next
2118          *     leaf
2119          *  -EOVERFLOW: item was to large for buffer
2120          *  -EFAULT: could not copy extent buffer back to userspace
2121          */
2122         return ret;
2123 }
2124
2125 static noinline int search_ioctl(struct inode *inode,
2126                                  struct btrfs_ioctl_search_key *sk,
2127                                  size_t *buf_size,
2128                                  char __user *ubuf)
2129 {
2130         struct btrfs_root *root;
2131         struct btrfs_key key;
2132         struct btrfs_path *path;
2133         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
2134         int ret;
2135         int num_found = 0;
2136         unsigned long sk_offset = 0;
2137
2138         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2139                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2140                 return -EOVERFLOW;
2141         }
2142
2143         path = btrfs_alloc_path();
2144         if (!path)
2145                 return -ENOMEM;
2146
2147         if (sk->tree_id == 0) {
2148                 /* search the root of the inode that was passed */
2149                 root = BTRFS_I(inode)->root;
2150         } else {
2151                 key.objectid = sk->tree_id;
2152                 key.type = BTRFS_ROOT_ITEM_KEY;
2153                 key.offset = (u64)-1;
2154                 root = btrfs_read_fs_root_no_name(info, &key);
2155                 if (IS_ERR(root)) {
2156                         btrfs_free_path(path);
2157                         return -ENOENT;
2158                 }
2159         }
2160
2161         key.objectid = sk->min_objectid;
2162         key.type = sk->min_type;
2163         key.offset = sk->min_offset;
2164
2165         while (1) {
2166                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2167                                                *buf_size - sk_offset);
2168                 if (ret)
2169                         break;
2170
2171                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2172                 if (ret != 0) {
2173                         if (ret > 0)
2174                                 ret = 0;
2175                         goto err;
2176                 }
2177                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2178                                  &sk_offset, &num_found);
2179                 btrfs_release_path(path);
2180                 if (ret)
2181                         break;
2182
2183         }
2184         if (ret > 0)
2185                 ret = 0;
2186 err:
2187         sk->nr_items = num_found;
2188         btrfs_free_path(path);
2189         return ret;
2190 }
2191
2192 static noinline int btrfs_ioctl_tree_search(struct file *file,
2193                                            void __user *argp)
2194 {
2195         struct btrfs_ioctl_search_args __user *uargs;
2196         struct btrfs_ioctl_search_key sk;
2197         struct inode *inode;
2198         int ret;
2199         size_t buf_size;
2200
2201         if (!capable(CAP_SYS_ADMIN))
2202                 return -EPERM;
2203
2204         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2205
2206         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2207                 return -EFAULT;
2208
2209         buf_size = sizeof(uargs->buf);
2210
2211         inode = file_inode(file);
2212         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2213
2214         /*
2215          * In the origin implementation an overflow is handled by returning a
2216          * search header with a len of zero, so reset ret.
2217          */
2218         if (ret == -EOVERFLOW)
2219                 ret = 0;
2220
2221         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2222                 ret = -EFAULT;
2223         return ret;
2224 }
2225
2226 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2227                                                void __user *argp)
2228 {
2229         struct btrfs_ioctl_search_args_v2 __user *uarg;
2230         struct btrfs_ioctl_search_args_v2 args;
2231         struct inode *inode;
2232         int ret;
2233         size_t buf_size;
2234         const size_t buf_limit = SZ_16M;
2235
2236         if (!capable(CAP_SYS_ADMIN))
2237                 return -EPERM;
2238
2239         /* copy search header and buffer size */
2240         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2241         if (copy_from_user(&args, uarg, sizeof(args)))
2242                 return -EFAULT;
2243
2244         buf_size = args.buf_size;
2245
2246         if (buf_size < sizeof(struct btrfs_ioctl_search_header))
2247                 return -EOVERFLOW;
2248
2249         /* limit result size to 16MB */
2250         if (buf_size > buf_limit)
2251                 buf_size = buf_limit;
2252
2253         inode = file_inode(file);
2254         ret = search_ioctl(inode, &args.key, &buf_size,
2255                            (char *)(&uarg->buf[0]));
2256         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2257                 ret = -EFAULT;
2258         else if (ret == -EOVERFLOW &&
2259                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2260                 ret = -EFAULT;
2261
2262         return ret;
2263 }
2264
2265 /*
2266  * Search INODE_REFs to identify path name of 'dirid' directory
2267  * in a 'tree_id' tree. and sets path name to 'name'.
2268  */
2269 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2270                                 u64 tree_id, u64 dirid, char *name)
2271 {
2272         struct btrfs_root *root;
2273         struct btrfs_key key;
2274         char *ptr;
2275         int ret = -1;
2276         int slot;
2277         int len;
2278         int total_len = 0;
2279         struct btrfs_inode_ref *iref;
2280         struct extent_buffer *l;
2281         struct btrfs_path *path;
2282
2283         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2284                 name[0]='\0';
2285                 return 0;
2286         }
2287
2288         path = btrfs_alloc_path();
2289         if (!path)
2290                 return -ENOMEM;
2291
2292         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2293
2294         key.objectid = tree_id;
2295         key.type = BTRFS_ROOT_ITEM_KEY;
2296         key.offset = (u64)-1;
2297         root = btrfs_read_fs_root_no_name(info, &key);
2298         if (IS_ERR(root)) {
2299                 btrfs_err(info, "could not find root %llu", tree_id);
2300                 ret = -ENOENT;
2301                 goto out;
2302         }
2303
2304         key.objectid = dirid;
2305         key.type = BTRFS_INODE_REF_KEY;
2306         key.offset = (u64)-1;
2307
2308         while (1) {
2309                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2310                 if (ret < 0)
2311                         goto out;
2312                 else if (ret > 0) {
2313                         ret = btrfs_previous_item(root, path, dirid,
2314                                                   BTRFS_INODE_REF_KEY);
2315                         if (ret < 0)
2316                                 goto out;
2317                         else if (ret > 0) {
2318                                 ret = -ENOENT;
2319                                 goto out;
2320                         }
2321                 }
2322
2323                 l = path->nodes[0];
2324                 slot = path->slots[0];
2325                 btrfs_item_key_to_cpu(l, &key, slot);
2326
2327                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2328                 len = btrfs_inode_ref_name_len(l, iref);
2329                 ptr -= len + 1;
2330                 total_len += len + 1;
2331                 if (ptr < name) {
2332                         ret = -ENAMETOOLONG;
2333                         goto out;
2334                 }
2335
2336                 *(ptr + len) = '/';
2337                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2338
2339                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2340                         break;
2341
2342                 btrfs_release_path(path);
2343                 key.objectid = key.offset;
2344                 key.offset = (u64)-1;
2345                 dirid = key.objectid;
2346         }
2347         memmove(name, ptr, total_len);
2348         name[total_len] = '\0';
2349         ret = 0;
2350 out:
2351         btrfs_free_path(path);
2352         return ret;
2353 }
2354
2355 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2356                                            void __user *argp)
2357 {
2358          struct btrfs_ioctl_ino_lookup_args *args;
2359          struct inode *inode;
2360         int ret = 0;
2361
2362         args = memdup_user(argp, sizeof(*args));
2363         if (IS_ERR(args))
2364                 return PTR_ERR(args);
2365
2366         inode = file_inode(file);
2367
2368         /*
2369          * Unprivileged query to obtain the containing subvolume root id. The
2370          * path is reset so it's consistent with btrfs_search_path_in_tree.
2371          */
2372         if (args->treeid == 0)
2373                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2374
2375         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2376                 args->name[0] = 0;
2377                 goto out;
2378         }
2379
2380         if (!capable(CAP_SYS_ADMIN)) {
2381                 ret = -EPERM;
2382                 goto out;
2383         }
2384
2385         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2386                                         args->treeid, args->objectid,
2387                                         args->name);
2388
2389 out:
2390         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2391                 ret = -EFAULT;
2392
2393         kfree(args);
2394         return ret;
2395 }
2396
2397 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2398                                              void __user *arg)
2399 {
2400         struct dentry *parent = file->f_path.dentry;
2401         struct dentry *dentry;
2402         struct inode *dir = d_inode(parent);
2403         struct inode *inode;
2404         struct btrfs_root *root = BTRFS_I(dir)->root;
2405         struct btrfs_root *dest = NULL;
2406         struct btrfs_ioctl_vol_args *vol_args;
2407         struct btrfs_trans_handle *trans;
2408         struct btrfs_block_rsv block_rsv;
2409         u64 root_flags;
2410         u64 qgroup_reserved;
2411         int namelen;
2412         int ret;
2413         int err = 0;
2414
2415         if (!S_ISDIR(dir->i_mode))
2416                 return -ENOTDIR;
2417
2418         vol_args = memdup_user(arg, sizeof(*vol_args));
2419         if (IS_ERR(vol_args))
2420                 return PTR_ERR(vol_args);
2421
2422         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2423         namelen = strlen(vol_args->name);
2424         if (strchr(vol_args->name, '/') ||
2425             strncmp(vol_args->name, "..", namelen) == 0) {
2426                 err = -EINVAL;
2427                 goto out;
2428         }
2429
2430         err = mnt_want_write_file(file);
2431         if (err)
2432                 goto out;
2433
2434
2435         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2436         if (err == -EINTR)
2437                 goto out_drop_write;
2438         dentry = lookup_one_len(vol_args->name, parent, namelen);
2439         if (IS_ERR(dentry)) {
2440                 err = PTR_ERR(dentry);
2441                 goto out_unlock_dir;
2442         }
2443
2444         if (d_really_is_negative(dentry)) {
2445                 err = -ENOENT;
2446                 goto out_dput;
2447         }
2448
2449         inode = d_inode(dentry);
2450         dest = BTRFS_I(inode)->root;
2451         if (!capable(CAP_SYS_ADMIN)) {
2452                 /*
2453                  * Regular user.  Only allow this with a special mount
2454                  * option, when the user has write+exec access to the
2455                  * subvol root, and when rmdir(2) would have been
2456                  * allowed.
2457                  *
2458                  * Note that this is _not_ check that the subvol is
2459                  * empty or doesn't contain data that we wouldn't
2460                  * otherwise be able to delete.
2461                  *
2462                  * Users who want to delete empty subvols should try
2463                  * rmdir(2).
2464                  */
2465                 err = -EPERM;
2466                 if (!btrfs_test_opt(root->fs_info, USER_SUBVOL_RM_ALLOWED))
2467                         goto out_dput;
2468
2469                 /*
2470                  * Do not allow deletion if the parent dir is the same
2471                  * as the dir to be deleted.  That means the ioctl
2472                  * must be called on the dentry referencing the root
2473                  * of the subvol, not a random directory contained
2474                  * within it.
2475                  */
2476                 err = -EINVAL;
2477                 if (root == dest)
2478                         goto out_dput;
2479
2480                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2481                 if (err)
2482                         goto out_dput;
2483         }
2484
2485         /* check if subvolume may be deleted by a user */
2486         err = btrfs_may_delete(dir, dentry, 1);
2487         if (err)
2488                 goto out_dput;
2489
2490         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2491                 err = -EINVAL;
2492                 goto out_dput;
2493         }
2494
2495         inode_lock(inode);
2496
2497         /*
2498          * Don't allow to delete a subvolume with send in progress. This is
2499          * inside the i_mutex so the error handling that has to drop the bit
2500          * again is not run concurrently.
2501          */
2502         spin_lock(&dest->root_item_lock);
2503         root_flags = btrfs_root_flags(&dest->root_item);
2504         if (dest->send_in_progress == 0) {
2505                 btrfs_set_root_flags(&dest->root_item,
2506                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2507                 spin_unlock(&dest->root_item_lock);
2508         } else {
2509                 spin_unlock(&dest->root_item_lock);
2510                 btrfs_warn(root->fs_info,
2511                         "Attempt to delete subvolume %llu during send",
2512                         dest->root_key.objectid);
2513                 err = -EPERM;
2514                 goto out_unlock_inode;
2515         }
2516
2517         down_write(&root->fs_info->subvol_sem);
2518
2519         err = may_destroy_subvol(dest);
2520         if (err)
2521                 goto out_up_write;
2522
2523         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2524         /*
2525          * One for dir inode, two for dir entries, two for root
2526          * ref/backref.
2527          */
2528         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2529                                                5, &qgroup_reserved, true);
2530         if (err)
2531                 goto out_up_write;
2532
2533         trans = btrfs_start_transaction(root, 0);
2534         if (IS_ERR(trans)) {
2535                 err = PTR_ERR(trans);
2536                 goto out_release;
2537         }
2538         trans->block_rsv = &block_rsv;
2539         trans->bytes_reserved = block_rsv.size;
2540
2541         btrfs_record_snapshot_destroy(trans, dir);
2542
2543         ret = btrfs_unlink_subvol(trans, root, dir,
2544                                 dest->root_key.objectid,
2545                                 dentry->d_name.name,
2546                                 dentry->d_name.len);
2547         if (ret) {
2548                 err = ret;
2549                 btrfs_abort_transaction(trans, ret);
2550                 goto out_end_trans;
2551         }
2552
2553         btrfs_record_root_in_trans(trans, dest);
2554
2555         memset(&dest->root_item.drop_progress, 0,
2556                 sizeof(dest->root_item.drop_progress));
2557         dest->root_item.drop_level = 0;
2558         btrfs_set_root_refs(&dest->root_item, 0);
2559
2560         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2561                 ret = btrfs_insert_orphan_item(trans,
2562                                         root->fs_info->tree_root,
2563                                         dest->root_key.objectid);
2564                 if (ret) {
2565                         btrfs_abort_transaction(trans, ret);
2566                         err = ret;
2567                         goto out_end_trans;
2568                 }
2569         }
2570
2571         ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2572                                   dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL,
2573                                   dest->root_key.objectid);
2574         if (ret && ret != -ENOENT) {
2575                 btrfs_abort_transaction(trans, ret);
2576                 err = ret;
2577                 goto out_end_trans;
2578         }
2579         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2580                 ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
2581                                           dest->root_item.received_uuid,
2582                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2583                                           dest->root_key.objectid);
2584                 if (ret && ret != -ENOENT) {
2585                         btrfs_abort_transaction(trans, ret);
2586                         err = ret;
2587                         goto out_end_trans;
2588                 }
2589         }
2590
2591 out_end_trans:
2592         trans->block_rsv = NULL;
2593         trans->bytes_reserved = 0;
2594         ret = btrfs_end_transaction(trans, root);
2595         if (ret && !err)
2596                 err = ret;
2597         inode->i_flags |= S_DEAD;
2598 out_release:
2599         btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved);
2600 out_up_write:
2601         up_write(&root->fs_info->subvol_sem);
2602         if (err) {
2603                 spin_lock(&dest->root_item_lock);
2604                 root_flags = btrfs_root_flags(&dest->root_item);
2605                 btrfs_set_root_flags(&dest->root_item,
2606                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2607                 spin_unlock(&dest->root_item_lock);
2608         }
2609 out_unlock_inode:
2610         inode_unlock(inode);
2611         if (!err) {
2612                 d_invalidate(dentry);
2613                 btrfs_invalidate_inodes(dest);
2614                 d_delete(dentry);
2615                 ASSERT(dest->send_in_progress == 0);
2616
2617                 /* the last ref */
2618                 if (dest->ino_cache_inode) {
2619                         iput(dest->ino_cache_inode);
2620                         dest->ino_cache_inode = NULL;
2621                 }
2622         }
2623 out_dput:
2624         dput(dentry);
2625 out_unlock_dir:
2626         inode_unlock(dir);
2627 out_drop_write:
2628         mnt_drop_write_file(file);
2629 out:
2630         kfree(vol_args);
2631         return err;
2632 }
2633
2634 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2635 {
2636         struct inode *inode = file_inode(file);
2637         struct btrfs_root *root = BTRFS_I(inode)->root;
2638         struct btrfs_ioctl_defrag_range_args *range;
2639         int ret;
2640
2641         ret = mnt_want_write_file(file);
2642         if (ret)
2643                 return ret;
2644
2645         if (btrfs_root_readonly(root)) {
2646                 ret = -EROFS;
2647                 goto out;
2648         }
2649
2650         switch (inode->i_mode & S_IFMT) {
2651         case S_IFDIR:
2652                 if (!capable(CAP_SYS_ADMIN)) {
2653                         ret = -EPERM;
2654                         goto out;
2655                 }
2656                 ret = btrfs_defrag_root(root);
2657                 if (ret)
2658                         goto out;
2659                 ret = btrfs_defrag_root(root->fs_info->extent_root);
2660                 break;
2661         case S_IFREG:
2662                 if (!(file->f_mode & FMODE_WRITE)) {
2663                         ret = -EINVAL;
2664                         goto out;
2665                 }
2666
2667                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2668                 if (!range) {
2669                         ret = -ENOMEM;
2670                         goto out;
2671                 }
2672
2673                 if (argp) {
2674                         if (copy_from_user(range, argp,
2675                                            sizeof(*range))) {
2676                                 ret = -EFAULT;
2677                                 kfree(range);
2678                                 goto out;
2679                         }
2680                         /* compression requires us to start the IO */
2681                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2682                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2683                                 range->extent_thresh = (u32)-1;
2684                         }
2685                 } else {
2686                         /* the rest are all set to zero by kzalloc */
2687                         range->len = (u64)-1;
2688                 }
2689                 ret = btrfs_defrag_file(file_inode(file), file,
2690                                         range, 0, 0);
2691                 if (ret > 0)
2692                         ret = 0;
2693                 kfree(range);
2694                 break;
2695         default:
2696                 ret = -EINVAL;
2697         }
2698 out:
2699         mnt_drop_write_file(file);
2700         return ret;
2701 }
2702
2703 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2704 {
2705         struct btrfs_ioctl_vol_args *vol_args;
2706         int ret;
2707
2708         if (!capable(CAP_SYS_ADMIN))
2709                 return -EPERM;
2710
2711         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2712                         1)) {
2713                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2714         }
2715
2716         mutex_lock(&root->fs_info->volume_mutex);
2717         vol_args = memdup_user(arg, sizeof(*vol_args));
2718         if (IS_ERR(vol_args)) {
2719                 ret = PTR_ERR(vol_args);
2720                 goto out;
2721         }
2722
2723         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2724         ret = btrfs_init_new_device(root, vol_args->name);
2725
2726         if (!ret)
2727                 btrfs_info(root->fs_info, "disk added %s",vol_args->name);
2728
2729         kfree(vol_args);
2730 out:
2731         mutex_unlock(&root->fs_info->volume_mutex);
2732         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2733         return ret;
2734 }
2735
2736 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2737 {
2738         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2739         struct btrfs_ioctl_vol_args_v2 *vol_args;
2740         int ret;
2741
2742         if (!capable(CAP_SYS_ADMIN))
2743                 return -EPERM;
2744
2745         ret = mnt_want_write_file(file);
2746         if (ret)
2747                 return ret;
2748
2749         vol_args = memdup_user(arg, sizeof(*vol_args));
2750         if (IS_ERR(vol_args)) {
2751                 ret = PTR_ERR(vol_args);
2752                 goto err_drop;
2753         }
2754
2755         /* Check for compatibility reject unknown flags */
2756         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
2757                 ret = -EOPNOTSUPP;
2758                 goto out;
2759         }
2760
2761         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2762                         1)) {
2763                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2764                 goto out;
2765         }
2766
2767         mutex_lock(&root->fs_info->volume_mutex);
2768         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2769                 ret = btrfs_rm_device(root, NULL, vol_args->devid);
2770         } else {
2771                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2772                 ret = btrfs_rm_device(root, vol_args->name, 0);
2773         }
2774         mutex_unlock(&root->fs_info->volume_mutex);
2775         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2776
2777         if (!ret) {
2778                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2779                         btrfs_info(root->fs_info, "device deleted: id %llu",
2780                                         vol_args->devid);
2781                 else
2782                         btrfs_info(root->fs_info, "device deleted: %s",
2783                                         vol_args->name);
2784         }
2785 out:
2786         kfree(vol_args);
2787 err_drop:
2788         mnt_drop_write_file(file);
2789         return ret;
2790 }
2791
2792 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2793 {
2794         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
2795         struct btrfs_ioctl_vol_args *vol_args;
2796         int ret;
2797
2798         if (!capable(CAP_SYS_ADMIN))
2799                 return -EPERM;
2800
2801         ret = mnt_want_write_file(file);
2802         if (ret)
2803                 return ret;
2804
2805         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2806                         1)) {
2807                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2808                 goto out_drop_write;
2809         }
2810
2811         vol_args = memdup_user(arg, sizeof(*vol_args));
2812         if (IS_ERR(vol_args)) {
2813                 ret = PTR_ERR(vol_args);
2814                 goto out;
2815         }
2816
2817         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2818         mutex_lock(&root->fs_info->volume_mutex);
2819         ret = btrfs_rm_device(root, vol_args->name, 0);
2820         mutex_unlock(&root->fs_info->volume_mutex);
2821
2822         if (!ret)
2823                 btrfs_info(root->fs_info, "disk deleted %s",vol_args->name);
2824         kfree(vol_args);
2825 out:
2826         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2827 out_drop_write:
2828         mnt_drop_write_file(file);
2829
2830         return ret;
2831 }
2832
2833 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2834 {
2835         struct btrfs_ioctl_fs_info_args *fi_args;
2836         struct btrfs_device *device;
2837         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2838         int ret = 0;
2839
2840         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2841         if (!fi_args)
2842                 return -ENOMEM;
2843
2844         mutex_lock(&fs_devices->device_list_mutex);
2845         fi_args->num_devices = fs_devices->num_devices;
2846         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2847
2848         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2849                 if (device->devid > fi_args->max_id)
2850                         fi_args->max_id = device->devid;
2851         }
2852         mutex_unlock(&fs_devices->device_list_mutex);
2853
2854         fi_args->nodesize = root->fs_info->super_copy->nodesize;
2855         fi_args->sectorsize = root->fs_info->super_copy->sectorsize;
2856         fi_args->clone_alignment = root->fs_info->super_copy->sectorsize;
2857
2858         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2859                 ret = -EFAULT;
2860
2861         kfree(fi_args);
2862         return ret;
2863 }
2864
2865 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2866 {
2867         struct btrfs_ioctl_dev_info_args *di_args;
2868         struct btrfs_device *dev;
2869         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2870         int ret = 0;
2871         char *s_uuid = NULL;
2872
2873         di_args = memdup_user(arg, sizeof(*di_args));
2874         if (IS_ERR(di_args))
2875                 return PTR_ERR(di_args);
2876
2877         if (!btrfs_is_empty_uuid(di_args->uuid))
2878                 s_uuid = di_args->uuid;
2879
2880         mutex_lock(&fs_devices->device_list_mutex);
2881         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2882
2883         if (!dev) {
2884                 ret = -ENODEV;
2885                 goto out;
2886         }
2887
2888         di_args->devid = dev->devid;
2889         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2890         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2891         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2892         if (dev->name) {
2893                 struct rcu_string *name;
2894
2895                 rcu_read_lock();
2896                 name = rcu_dereference(dev->name);
2897                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2898                 rcu_read_unlock();
2899                 di_args->path[sizeof(di_args->path) - 1] = 0;
2900         } else {
2901                 di_args->path[0] = '\0';
2902         }
2903
2904 out:
2905         mutex_unlock(&fs_devices->device_list_mutex);
2906         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2907                 ret = -EFAULT;
2908
2909         kfree(di_args);
2910         return ret;
2911 }
2912
2913 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2914 {
2915         struct page *page;
2916
2917         page = grab_cache_page(inode->i_mapping, index);
2918         if (!page)
2919                 return ERR_PTR(-ENOMEM);
2920
2921         if (!PageUptodate(page)) {
2922                 int ret;
2923
2924                 ret = btrfs_readpage(NULL, page);
2925                 if (ret)
2926                         return ERR_PTR(ret);
2927                 lock_page(page);
2928                 if (!PageUptodate(page)) {
2929                         unlock_page(page);
2930                         put_page(page);
2931                         return ERR_PTR(-EIO);
2932                 }
2933                 if (page->mapping != inode->i_mapping) {
2934                         unlock_page(page);
2935                         put_page(page);
2936                         return ERR_PTR(-EAGAIN);
2937                 }
2938         }
2939
2940         return page;
2941 }
2942
2943 static int gather_extent_pages(struct inode *inode, struct page **pages,
2944                                int num_pages, u64 off)
2945 {
2946         int i;
2947         pgoff_t index = off >> PAGE_SHIFT;
2948
2949         for (i = 0; i < num_pages; i++) {
2950 again:
2951                 pages[i] = extent_same_get_page(inode, index + i);
2952                 if (IS_ERR(pages[i])) {
2953                         int err = PTR_ERR(pages[i]);
2954
2955                         if (err == -EAGAIN)
2956                                 goto again;
2957                         pages[i] = NULL;
2958                         return err;
2959                 }
2960         }
2961         return 0;
2962 }
2963
2964 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2965                              bool retry_range_locking)
2966 {
2967         /*
2968          * Do any pending delalloc/csum calculations on inode, one way or
2969          * another, and lock file content.
2970          * The locking order is:
2971          *
2972          *   1) pages
2973          *   2) range in the inode's io tree
2974          */
2975         while (1) {
2976                 struct btrfs_ordered_extent *ordered;
2977                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2978                 ordered = btrfs_lookup_first_ordered_extent(inode,
2979                                                             off + len - 1);
2980                 if ((!ordered ||
2981                      ordered->file_offset + ordered->len <= off ||
2982                      ordered->file_offset >= off + len) &&
2983                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2984                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2985                         if (ordered)
2986                                 btrfs_put_ordered_extent(ordered);
2987                         break;
2988                 }
2989                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2990                 if (ordered)
2991                         btrfs_put_ordered_extent(ordered);
2992                 if (!retry_range_locking)
2993                         return -EAGAIN;
2994                 btrfs_wait_ordered_range(inode, off, len);
2995         }
2996         return 0;
2997 }
2998
2999 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3000 {
3001         inode_unlock(inode1);
3002         inode_unlock(inode2);
3003 }
3004
3005 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3006 {
3007         if (inode1 < inode2)
3008                 swap(inode1, inode2);
3009
3010         inode_lock_nested(inode1, I_MUTEX_PARENT);
3011         inode_lock_nested(inode2, I_MUTEX_CHILD);
3012 }
3013
3014 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3015                                       struct inode *inode2, u64 loff2, u64 len)
3016 {
3017         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3018         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3019 }
3020
3021 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3022                                     struct inode *inode2, u64 loff2, u64 len,
3023                                     bool retry_range_locking)
3024 {
3025         int ret;
3026
3027         if (inode1 < inode2) {
3028                 swap(inode1, inode2);
3029                 swap(loff1, loff2);
3030         }
3031         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3032         if (ret)
3033                 return ret;
3034         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3035         if (ret)
3036                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3037                               loff1 + len - 1);
3038         return ret;
3039 }
3040
3041 struct cmp_pages {
3042         int             num_pages;
3043         struct page     **src_pages;
3044         struct page     **dst_pages;
3045 };
3046
3047 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3048 {
3049         int i;
3050         struct page *pg;
3051
3052         for (i = 0; i < cmp->num_pages; i++) {
3053                 pg = cmp->src_pages[i];
3054                 if (pg) {
3055                         unlock_page(pg);
3056                         put_page(pg);
3057                 }
3058                 pg = cmp->dst_pages[i];
3059                 if (pg) {
3060                         unlock_page(pg);
3061                         put_page(pg);
3062                 }
3063         }
3064         kfree(cmp->src_pages);
3065         kfree(cmp->dst_pages);
3066 }
3067
3068 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3069                                   struct inode *dst, u64 dst_loff,
3070                                   u64 len, struct cmp_pages *cmp)
3071 {
3072         int ret;
3073         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3074         struct page **src_pgarr, **dst_pgarr;
3075
3076         /*
3077          * We must gather up all the pages before we initiate our
3078          * extent locking. We use an array for the page pointers. Size
3079          * of the array is bounded by len, which is in turn bounded by
3080          * BTRFS_MAX_DEDUPE_LEN.
3081          */
3082         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3083         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3084         if (!src_pgarr || !dst_pgarr) {
3085                 kfree(src_pgarr);
3086                 kfree(dst_pgarr);
3087                 return -ENOMEM;
3088         }
3089         cmp->num_pages = num_pages;
3090         cmp->src_pages = src_pgarr;
3091         cmp->dst_pages = dst_pgarr;
3092
3093         ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff);
3094         if (ret)
3095                 goto out;
3096
3097         ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff);
3098
3099 out:
3100         if (ret)
3101                 btrfs_cmp_data_free(cmp);
3102         return ret;
3103 }
3104
3105 static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst,
3106                           u64 dst_loff, u64 len, struct cmp_pages *cmp)
3107 {
3108         int ret = 0;
3109         int i;
3110         struct page *src_page, *dst_page;
3111         unsigned int cmp_len = PAGE_SIZE;
3112         void *addr, *dst_addr;
3113
3114         i = 0;
3115         while (len) {
3116                 if (len < PAGE_SIZE)
3117                         cmp_len = len;
3118
3119                 BUG_ON(i >= cmp->num_pages);
3120
3121                 src_page = cmp->src_pages[i];
3122                 dst_page = cmp->dst_pages[i];
3123                 ASSERT(PageLocked(src_page));
3124                 ASSERT(PageLocked(dst_page));
3125
3126                 addr = kmap_atomic(src_page);
3127                 dst_addr = kmap_atomic(dst_page);
3128
3129                 flush_dcache_page(src_page);
3130                 flush_dcache_page(dst_page);
3131
3132                 if (memcmp(addr, dst_addr, cmp_len))
3133                         ret = -EBADE;
3134
3135                 kunmap_atomic(addr);
3136                 kunmap_atomic(dst_addr);
3137
3138                 if (ret)
3139                         break;
3140
3141                 len -= cmp_len;
3142                 i++;
3143         }
3144
3145         return ret;
3146 }
3147
3148 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3149                                      u64 olen)
3150 {
3151         u64 len = *plen;
3152         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3153
3154         if (off + olen > inode->i_size || off + olen < off)
3155                 return -EINVAL;
3156
3157         /* if we extend to eof, continue to block boundary */
3158         if (off + len == inode->i_size)
3159                 *plen = len = ALIGN(inode->i_size, bs) - off;
3160
3161         /* Check that we are block aligned - btrfs_clone() requires this */
3162         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3163                 return -EINVAL;
3164
3165         return 0;
3166 }
3167
3168 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3169                              struct inode *dst, u64 dst_loff)
3170 {
3171         int ret;
3172         u64 len = olen;
3173         struct cmp_pages cmp;
3174         int same_inode = 0;
3175         u64 same_lock_start = 0;
3176         u64 same_lock_len = 0;
3177
3178         if (src == dst)
3179                 same_inode = 1;
3180
3181         if (len == 0)
3182                 return 0;
3183
3184         if (same_inode) {
3185                 inode_lock(src);
3186
3187                 ret = extent_same_check_offsets(src, loff, &len, olen);
3188                 if (ret)
3189                         goto out_unlock;
3190                 ret = extent_same_check_offsets(src, dst_loff, &len, olen);
3191                 if (ret)
3192                         goto out_unlock;
3193
3194                 /*
3195                  * Single inode case wants the same checks, except we
3196                  * don't want our length pushed out past i_size as
3197                  * comparing that data range makes no sense.
3198                  *
3199                  * extent_same_check_offsets() will do this for an
3200                  * unaligned length at i_size, so catch it here and
3201                  * reject the request.
3202                  *
3203                  * This effectively means we require aligned extents
3204                  * for the single-inode case, whereas the other cases
3205                  * allow an unaligned length so long as it ends at
3206                  * i_size.
3207                  */
3208                 if (len != olen) {
3209                         ret = -EINVAL;
3210                         goto out_unlock;
3211                 }
3212
3213                 /* Check for overlapping ranges */
3214                 if (dst_loff + len > loff && dst_loff < loff + len) {
3215                         ret = -EINVAL;
3216                         goto out_unlock;
3217                 }
3218
3219                 same_lock_start = min_t(u64, loff, dst_loff);
3220                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3221         } else {
3222                 btrfs_double_inode_lock(src, dst);
3223
3224                 ret = extent_same_check_offsets(src, loff, &len, olen);
3225                 if (ret)
3226                         goto out_unlock;
3227
3228                 ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3229                 if (ret)
3230                         goto out_unlock;
3231         }
3232
3233         /* don't make the dst file partly checksummed */
3234         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3235             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3236                 ret = -EINVAL;
3237                 goto out_unlock;
3238         }
3239
3240 again:
3241         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3242         if (ret)
3243                 goto out_unlock;
3244
3245         if (same_inode)
3246                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3247                                         false);
3248         else
3249                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3250                                                false);
3251         /*
3252          * If one of the inodes has dirty pages in the respective range or
3253          * ordered extents, we need to flush dellaloc and wait for all ordered
3254          * extents in the range. We must unlock the pages and the ranges in the
3255          * io trees to avoid deadlocks when flushing delalloc (requires locking
3256          * pages) and when waiting for ordered extents to complete (they require
3257          * range locking).
3258          */
3259         if (ret == -EAGAIN) {
3260                 /*
3261                  * Ranges in the io trees already unlocked. Now unlock all
3262                  * pages before waiting for all IO to complete.
3263                  */
3264                 btrfs_cmp_data_free(&cmp);
3265                 if (same_inode) {
3266                         btrfs_wait_ordered_range(src, same_lock_start,
3267                                                  same_lock_len);
3268                 } else {
3269                         btrfs_wait_ordered_range(src, loff, len);
3270                         btrfs_wait_ordered_range(dst, dst_loff, len);
3271                 }
3272                 goto again;
3273         }
3274         ASSERT(ret == 0);
3275         if (WARN_ON(ret)) {
3276                 /* ranges in the io trees already unlocked */
3277                 btrfs_cmp_data_free(&cmp);
3278                 return ret;
3279         }
3280
3281         /* pass original length for comparison so we stay within i_size */
3282         ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp);
3283         if (ret == 0)
3284                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3285
3286         if (same_inode)
3287                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3288                               same_lock_start + same_lock_len - 1);
3289         else
3290                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3291
3292         btrfs_cmp_data_free(&cmp);
3293 out_unlock:
3294         if (same_inode)
3295                 inode_unlock(src);
3296         else
3297                 btrfs_double_inode_unlock(src, dst);
3298
3299         return ret;
3300 }
3301
3302 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3303
3304 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3305                                 struct file *dst_file, u64 dst_loff)
3306 {
3307         struct inode *src = file_inode(src_file);
3308         struct inode *dst = file_inode(dst_file);
3309         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3310         ssize_t res;
3311
3312         if (olen > BTRFS_MAX_DEDUPE_LEN)
3313                 olen = BTRFS_MAX_DEDUPE_LEN;
3314
3315         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3316                 /*
3317                  * Btrfs does not support blocksize < page_size. As a
3318                  * result, btrfs_cmp_data() won't correctly handle
3319                  * this situation without an update.
3320                  */
3321                 return -EINVAL;
3322         }
3323
3324         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3325         if (res)
3326                 return res;
3327         return olen;
3328 }
3329
3330 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3331                                      struct inode *inode,
3332                                      u64 endoff,
3333                                      const u64 destoff,
3334                                      const u64 olen,
3335                                      int no_time_update)
3336 {
3337         struct btrfs_root *root = BTRFS_I(inode)->root;
3338         int ret;
3339
3340         inode_inc_iversion(inode);
3341         if (!no_time_update)
3342                 inode->i_mtime = inode->i_ctime = current_time(inode);
3343         /*
3344          * We round up to the block size at eof when determining which
3345          * extents to clone above, but shouldn't round up the file size.
3346          */
3347         if (endoff > destoff + olen)
3348                 endoff = destoff + olen;
3349         if (endoff > inode->i_size)
3350                 btrfs_i_size_write(inode, endoff);
3351
3352         ret = btrfs_update_inode(trans, root, inode);
3353         if (ret) {
3354                 btrfs_abort_transaction(trans, ret);
3355                 btrfs_end_transaction(trans, root);
3356                 goto out;
3357         }
3358         ret = btrfs_end_transaction(trans, root);
3359 out:
3360         return ret;
3361 }
3362
3363 static void clone_update_extent_map(struct inode *inode,
3364                                     const struct btrfs_trans_handle *trans,
3365                                     const struct btrfs_path *path,
3366                                     const u64 hole_offset,
3367                                     const u64 hole_len)
3368 {
3369         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3370         struct extent_map *em;
3371         int ret;
3372
3373         em = alloc_extent_map();
3374         if (!em) {
3375                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3376                         &BTRFS_I(inode)->runtime_flags);
3377                 return;
3378         }
3379
3380         if (path) {
3381                 struct btrfs_file_extent_item *fi;
3382
3383                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3384                                     struct btrfs_file_extent_item);
3385                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3386                 em->generation = -1;
3387                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3388                     BTRFS_FILE_EXTENT_INLINE)
3389                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3390                                 &BTRFS_I(inode)->runtime_flags);
3391         } else {
3392                 em->start = hole_offset;
3393                 em->len = hole_len;
3394                 em->ram_bytes = em->len;
3395                 em->orig_start = hole_offset;
3396                 em->block_start = EXTENT_MAP_HOLE;
3397                 em->block_len = 0;
3398                 em->orig_block_len = 0;
3399                 em->compress_type = BTRFS_COMPRESS_NONE;
3400                 em->generation = trans->transid;
3401         }
3402
3403         while (1) {
3404                 write_lock(&em_tree->lock);
3405                 ret = add_extent_mapping(em_tree, em, 1);
3406                 write_unlock(&em_tree->lock);
3407                 if (ret != -EEXIST) {
3408                         free_extent_map(em);
3409                         break;
3410                 }
3411                 btrfs_drop_extent_cache(inode, em->start,
3412                                         em->start + em->len - 1, 0);
3413         }
3414
3415         if (ret)
3416                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3417                         &BTRFS_I(inode)->runtime_flags);
3418 }
3419
3420 /*
3421  * Make sure we do not end up inserting an inline extent into a file that has
3422  * already other (non-inline) extents. If a file has an inline extent it can
3423  * not have any other extents and the (single) inline extent must start at the
3424  * file offset 0. Failing to respect these rules will lead to file corruption,
3425  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3426  *
3427  * We can have extents that have been already written to disk or we can have
3428  * dirty ranges still in delalloc, in which case the extent maps and items are
3429  * created only when we run delalloc, and the delalloc ranges might fall outside
3430  * the range we are currently locking in the inode's io tree. So we check the
3431  * inode's i_size because of that (i_size updates are done while holding the
3432  * i_mutex, which we are holding here).
3433  * We also check to see if the inode has a size not greater than "datal" but has
3434  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3435  * protected against such concurrent fallocate calls by the i_mutex).
3436  *
3437  * If the file has no extents but a size greater than datal, do not allow the
3438  * copy because we would need turn the inline extent into a non-inline one (even
3439  * with NO_HOLES enabled). If we find our destination inode only has one inline
3440  * extent, just overwrite it with the source inline extent if its size is less
3441  * than the source extent's size, or we could copy the source inline extent's
3442  * data into the destination inode's inline extent if the later is greater then
3443  * the former.
3444  */
3445 static int clone_copy_inline_extent(struct inode *src,
3446                                     struct inode *dst,
3447                                     struct btrfs_trans_handle *trans,
3448                                     struct btrfs_path *path,
3449                                     struct btrfs_key *new_key,
3450                                     const u64 drop_start,
3451                                     const u64 datal,
3452                                     const u64 skip,
3453                                     const u64 size,
3454                                     char *inline_data)
3455 {
3456         struct btrfs_root *root = BTRFS_I(dst)->root;
3457         const u64 aligned_end = ALIGN(new_key->offset + datal,
3458                                       root->sectorsize);
3459         int ret;
3460         struct btrfs_key key;
3461
3462         if (new_key->offset > 0)
3463                 return -EOPNOTSUPP;
3464
3465         key.objectid = btrfs_ino(dst);
3466         key.type = BTRFS_EXTENT_DATA_KEY;
3467         key.offset = 0;
3468         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3469         if (ret < 0) {
3470                 return ret;
3471         } else if (ret > 0) {
3472                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3473                         ret = btrfs_next_leaf(root, path);
3474                         if (ret < 0)
3475                                 return ret;
3476                         else if (ret > 0)
3477                                 goto copy_inline_extent;
3478                 }
3479                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3480                 if (key.objectid == btrfs_ino(dst) &&
3481                     key.type == BTRFS_EXTENT_DATA_KEY) {
3482                         ASSERT(key.offset > 0);
3483                         return -EOPNOTSUPP;
3484                 }
3485         } else if (i_size_read(dst) <= datal) {
3486                 struct btrfs_file_extent_item *ei;
3487                 u64 ext_len;
3488
3489                 /*
3490                  * If the file size is <= datal, make sure there are no other
3491                  * extents following (can happen do to an fallocate call with
3492                  * the flag FALLOC_FL_KEEP_SIZE).
3493                  */
3494                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3495                                     struct btrfs_file_extent_item);
3496                 /*
3497                  * If it's an inline extent, it can not have other extents
3498                  * following it.
3499                  */
3500                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3501                     BTRFS_FILE_EXTENT_INLINE)
3502                         goto copy_inline_extent;
3503
3504                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3505                 if (ext_len > aligned_end)
3506                         return -EOPNOTSUPP;
3507
3508                 ret = btrfs_next_item(root, path);
3509                 if (ret < 0) {
3510                         return ret;
3511                 } else if (ret == 0) {
3512                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3513                                               path->slots[0]);
3514                         if (key.objectid == btrfs_ino(dst) &&
3515                             key.type == BTRFS_EXTENT_DATA_KEY)
3516                                 return -EOPNOTSUPP;
3517                 }
3518         }
3519
3520 copy_inline_extent:
3521         /*
3522          * We have no extent items, or we have an extent at offset 0 which may
3523          * or may not be inlined. All these cases are dealt the same way.
3524          */
3525         if (i_size_read(dst) > datal) {
3526                 /*
3527                  * If the destination inode has an inline extent...
3528                  * This would require copying the data from the source inline
3529                  * extent into the beginning of the destination's inline extent.
3530                  * But this is really complex, both extents can be compressed
3531                  * or just one of them, which would require decompressing and
3532                  * re-compressing data (which could increase the new compressed
3533                  * size, not allowing the compressed data to fit anymore in an
3534                  * inline extent).
3535                  * So just don't support this case for now (it should be rare,
3536                  * we are not really saving space when cloning inline extents).
3537                  */
3538                 return -EOPNOTSUPP;
3539         }
3540
3541         btrfs_release_path(path);
3542         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3543         if (ret)
3544                 return ret;
3545         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3546         if (ret)
3547                 return ret;
3548
3549         if (skip) {
3550                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3551
3552                 memmove(inline_data + start, inline_data + start + skip, datal);
3553         }
3554
3555         write_extent_buffer(path->nodes[0], inline_data,
3556                             btrfs_item_ptr_offset(path->nodes[0],
3557                                                   path->slots[0]),
3558                             size);
3559         inode_add_bytes(dst, datal);
3560
3561         return 0;
3562 }
3563
3564 /**
3565  * btrfs_clone() - clone a range from inode file to another
3566  *
3567  * @src: Inode to clone from
3568  * @inode: Inode to clone to
3569  * @off: Offset within source to start clone from
3570  * @olen: Original length, passed by user, of range to clone
3571  * @olen_aligned: Block-aligned value of olen
3572  * @destoff: Offset within @inode to start clone
3573  * @no_time_update: Whether to update mtime/ctime on the target inode
3574  */
3575 static int btrfs_clone(struct inode *src, struct inode *inode,
3576                        const u64 off, const u64 olen, const u64 olen_aligned,
3577                        const u64 destoff, int no_time_update)
3578 {
3579         struct btrfs_root *root = BTRFS_I(inode)->root;
3580         struct btrfs_path *path = NULL;
3581         struct extent_buffer *leaf;
3582         struct btrfs_trans_handle *trans;
3583         char *buf = NULL;
3584         struct btrfs_key key;
3585         u32 nritems;
3586         int slot;
3587         int ret;
3588         const u64 len = olen_aligned;
3589         u64 last_dest_end = destoff;
3590
3591         ret = -ENOMEM;
3592         buf = kmalloc(root->nodesize, GFP_KERNEL | __GFP_NOWARN);
3593         if (!buf) {
3594                 buf = vmalloc(root->nodesize);
3595                 if (!buf)
3596                         return ret;
3597         }
3598
3599         path = btrfs_alloc_path();
3600         if (!path) {
3601                 kvfree(buf);
3602                 return ret;
3603         }
3604
3605         path->reada = READA_FORWARD;
3606         /* clone data */
3607         key.objectid = btrfs_ino(src);
3608         key.type = BTRFS_EXTENT_DATA_KEY;
3609         key.offset = off;
3610
3611         while (1) {
3612                 u64 next_key_min_offset = key.offset + 1;
3613
3614                 /*
3615                  * note the key will change type as we walk through the
3616                  * tree.
3617                  */
3618                 path->leave_spinning = 1;
3619                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3620                                 0, 0);
3621                 if (ret < 0)
3622                         goto out;
3623                 /*
3624                  * First search, if no extent item that starts at offset off was
3625                  * found but the previous item is an extent item, it's possible
3626                  * it might overlap our target range, therefore process it.
3627                  */
3628                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3629                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3630                                               path->slots[0] - 1);
3631                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3632                                 path->slots[0]--;
3633                 }
3634
3635                 nritems = btrfs_header_nritems(path->nodes[0]);
3636 process_slot:
3637                 if (path->slots[0] >= nritems) {
3638                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3639                         if (ret < 0)
3640                                 goto out;
3641                         if (ret > 0)
3642                                 break;
3643                         nritems = btrfs_header_nritems(path->nodes[0]);
3644                 }
3645                 leaf = path->nodes[0];
3646                 slot = path->slots[0];
3647
3648                 btrfs_item_key_to_cpu(leaf, &key, slot);
3649                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3650                     key.objectid != btrfs_ino(src))
3651                         break;
3652
3653                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3654                         struct btrfs_file_extent_item *extent;
3655                         int type;
3656                         u32 size;
3657                         struct btrfs_key new_key;
3658                         u64 disko = 0, diskl = 0;
3659                         u64 datao = 0, datal = 0;
3660                         u8 comp;
3661                         u64 drop_start;
3662
3663                         extent = btrfs_item_ptr(leaf, slot,
3664                                                 struct btrfs_file_extent_item);
3665                         comp = btrfs_file_extent_compression(leaf, extent);
3666                         type = btrfs_file_extent_type(leaf, extent);
3667                         if (type == BTRFS_FILE_EXTENT_REG ||
3668                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3669                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3670                                                                       extent);
3671                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3672                                                                  extent);
3673                                 datao = btrfs_file_extent_offset(leaf, extent);
3674                                 datal = btrfs_file_extent_num_bytes(leaf,
3675                                                                     extent);
3676                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3677                                 /* take upper bound, may be compressed */
3678                                 datal = btrfs_file_extent_ram_bytes(leaf,
3679                                                                     extent);
3680                         }
3681
3682                         /*
3683                          * The first search might have left us at an extent
3684                          * item that ends before our target range's start, can
3685                          * happen if we have holes and NO_HOLES feature enabled.
3686                          */
3687                         if (key.offset + datal <= off) {
3688                                 path->slots[0]++;
3689                                 goto process_slot;
3690                         } else if (key.offset >= off + len) {
3691                                 break;
3692                         }
3693                         next_key_min_offset = key.offset + datal;
3694                         size = btrfs_item_size_nr(leaf, slot);
3695                         read_extent_buffer(leaf, buf,
3696                                            btrfs_item_ptr_offset(leaf, slot),
3697                                            size);
3698
3699                         btrfs_release_path(path);
3700                         path->leave_spinning = 0;
3701
3702                         memcpy(&new_key, &key, sizeof(new_key));
3703                         new_key.objectid = btrfs_ino(inode);
3704                         if (off <= key.offset)
3705                                 new_key.offset = key.offset + destoff - off;
3706                         else
3707                                 new_key.offset = destoff;
3708
3709                         /*
3710                          * Deal with a hole that doesn't have an extent item
3711                          * that represents it (NO_HOLES feature enabled).
3712                          * This hole is either in the middle of the cloning
3713                          * range or at the beginning (fully overlaps it or
3714                          * partially overlaps it).
3715                          */
3716                         if (new_key.offset != last_dest_end)
3717                                 drop_start = last_dest_end;
3718                         else
3719                                 drop_start = new_key.offset;
3720
3721                         /*
3722                          * 1 - adjusting old extent (we may have to split it)
3723                          * 1 - add new extent
3724                          * 1 - inode update
3725                          */
3726                         trans = btrfs_start_transaction(root, 3);
3727                         if (IS_ERR(trans)) {
3728                                 ret = PTR_ERR(trans);
3729                                 goto out;
3730                         }
3731
3732                         if (type == BTRFS_FILE_EXTENT_REG ||
3733                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3734                                 /*
3735                                  *    a  | --- range to clone ---|  b
3736                                  * | ------------- extent ------------- |
3737                                  */
3738
3739                                 /* subtract range b */
3740                                 if (key.offset + datal > off + len)
3741                                         datal = off + len - key.offset;
3742
3743                                 /* subtract range a */
3744                                 if (off > key.offset) {
3745                                         datao += off - key.offset;
3746                                         datal -= off - key.offset;
3747                                 }
3748
3749                                 ret = btrfs_drop_extents(trans, root, inode,
3750                                                          drop_start,
3751                                                          new_key.offset + datal,
3752                                                          1);
3753                                 if (ret) {
3754                                         if (ret != -EOPNOTSUPP)
3755                                                 btrfs_abort_transaction(trans,
3756                                                                         ret);
3757                                         btrfs_end_transaction(trans, root);
3758                                         goto out;
3759                                 }
3760
3761                                 ret = btrfs_insert_empty_item(trans, root, path,
3762                                                               &new_key, size);
3763                                 if (ret) {
3764                                         btrfs_abort_transaction(trans, ret);
3765                                         btrfs_end_transaction(trans, root);
3766                                         goto out;
3767                                 }
3768
3769                                 leaf = path->nodes[0];
3770                                 slot = path->slots[0];
3771                                 write_extent_buffer(leaf, buf,
3772                                             btrfs_item_ptr_offset(leaf, slot),
3773                                             size);
3774
3775                                 extent = btrfs_item_ptr(leaf, slot,
3776                                                 struct btrfs_file_extent_item);
3777
3778                                 /* disko == 0 means it's a hole */
3779                                 if (!disko)
3780                                         datao = 0;
3781
3782                                 btrfs_set_file_extent_offset(leaf, extent,
3783                                                              datao);
3784                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3785                                                                 datal);
3786
3787                                 if (disko) {
3788                                         inode_add_bytes(inode, datal);
3789                                         ret = btrfs_inc_extent_ref(trans, root,
3790                                                         disko, diskl, 0,
3791                                                         root->root_key.objectid,
3792                                                         btrfs_ino(inode),
3793                                                         new_key.offset - datao);
3794                                         if (ret) {
3795                                                 btrfs_abort_transaction(trans,
3796                                                                         ret);
3797                                                 btrfs_end_transaction(trans,
3798                                                                       root);
3799                                                 goto out;
3800
3801                                         }
3802                                 }
3803                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3804                                 u64 skip = 0;
3805                                 u64 trim = 0;
3806
3807                                 if (off > key.offset) {
3808                                         skip = off - key.offset;
3809                                         new_key.offset += skip;
3810                                 }
3811
3812                                 if (key.offset + datal > off + len)
3813                                         trim = key.offset + datal - (off + len);
3814
3815                                 if (comp && (skip || trim)) {
3816                                         ret = -EINVAL;
3817                                         btrfs_end_transaction(trans, root);
3818                                         goto out;
3819                                 }
3820                                 size -= skip + trim;
3821                                 datal -= skip + trim;
3822
3823                                 ret = clone_copy_inline_extent(src, inode,
3824                                                                trans, path,
3825                                                                &new_key,
3826                                                                drop_start,
3827                                                                datal,
3828                                                                skip, size, buf);
3829                                 if (ret) {
3830                                         if (ret != -EOPNOTSUPP)
3831                                                 btrfs_abort_transaction(trans,
3832                                                                         ret);
3833                                         btrfs_end_transaction(trans, root);
3834                                         goto out;
3835                                 }
3836                                 leaf = path->nodes[0];
3837                                 slot = path->slots[0];
3838                         }
3839
3840                         /* If we have an implicit hole (NO_HOLES feature). */
3841                         if (drop_start < new_key.offset)
3842                                 clone_update_extent_map(inode, trans,
3843                                                 NULL, drop_start,
3844                                                 new_key.offset - drop_start);
3845
3846                         clone_update_extent_map(inode, trans, path, 0, 0);
3847
3848                         btrfs_mark_buffer_dirty(leaf);
3849                         btrfs_release_path(path);
3850
3851                         last_dest_end = ALIGN(new_key.offset + datal,
3852                                               root->sectorsize);
3853                         ret = clone_finish_inode_update(trans, inode,
3854                                                         last_dest_end,
3855                                                         destoff, olen,
3856                                                         no_time_update);
3857                         if (ret)
3858                                 goto out;
3859                         if (new_key.offset + datal >= destoff + len)
3860                                 break;
3861                 }
3862                 btrfs_release_path(path);
3863                 key.offset = next_key_min_offset;
3864
3865                 if (fatal_signal_pending(current)) {
3866                         ret = -EINTR;
3867                         goto out;
3868                 }
3869
3870                 cond_resched();
3871         }
3872         ret = 0;
3873
3874         if (last_dest_end < destoff + len) {
3875                 /*
3876                  * We have an implicit hole (NO_HOLES feature is enabled) that
3877                  * fully or partially overlaps our cloning range at its end.
3878                  */
3879                 btrfs_release_path(path);
3880
3881                 /*
3882                  * 1 - remove extent(s)
3883                  * 1 - inode update
3884                  */
3885                 trans = btrfs_start_transaction(root, 2);
3886                 if (IS_ERR(trans)) {
3887                         ret = PTR_ERR(trans);
3888                         goto out;
3889                 }
3890                 ret = btrfs_drop_extents(trans, root, inode,
3891                                          last_dest_end, destoff + len, 1);
3892                 if (ret) {
3893                         if (ret != -EOPNOTSUPP)
3894                                 btrfs_abort_transaction(trans, ret);
3895                         btrfs_end_transaction(trans, root);
3896                         goto out;
3897                 }
3898                 clone_update_extent_map(inode, trans, NULL, last_dest_end,
3899                                         destoff + len - last_dest_end);
3900                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3901                                                 destoff, olen, no_time_update);
3902         }
3903
3904 out:
3905         btrfs_free_path(path);
3906         kvfree(buf);
3907         return ret;
3908 }
3909
3910 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3911                                         u64 off, u64 olen, u64 destoff)
3912 {
3913         struct inode *inode = file_inode(file);
3914         struct inode *src = file_inode(file_src);
3915         struct btrfs_root *root = BTRFS_I(inode)->root;
3916         int ret;
3917         u64 len = olen;
3918         u64 bs = root->fs_info->sb->s_blocksize;
3919         int same_inode = src == inode;
3920
3921         /*
3922          * TODO:
3923          * - split compressed inline extents.  annoying: we need to
3924          *   decompress into destination's address_space (the file offset
3925          *   may change, so source mapping won't do), then recompress (or
3926          *   otherwise reinsert) a subrange.
3927          *
3928          * - split destination inode's inline extents.  The inline extents can
3929          *   be either compressed or non-compressed.
3930          */
3931
3932         if (btrfs_root_readonly(root))
3933                 return -EROFS;
3934
3935         if (file_src->f_path.mnt != file->f_path.mnt ||
3936             src->i_sb != inode->i_sb)
3937                 return -EXDEV;
3938
3939         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3940                 return -EISDIR;
3941
3942         if (!same_inode) {
3943                 btrfs_double_inode_lock(src, inode);
3944         } else {
3945                 inode_lock(src);
3946         }
3947
3948         /* don't make the dst file partly checksummed */
3949         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3950             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
3951                 ret = -EINVAL;
3952                 goto out_unlock;
3953         }
3954
3955         /* determine range to clone */
3956         ret = -EINVAL;
3957         if (off + len > src->i_size || off + len < off)
3958                 goto out_unlock;
3959         if (len == 0)
3960                 olen = len = src->i_size - off;
3961         /*
3962          * If we extend to eof, continue to block boundary if and only if the
3963          * destination end offset matches the destination file's size, otherwise
3964          * we would be corrupting data by placing the eof block into the middle
3965          * of a file.
3966          */
3967         if (off + len == src->i_size) {
3968                 if (!IS_ALIGNED(len, bs) && destoff + len < inode->i_size)
3969                         goto out_unlock;
3970                 len = ALIGN(src->i_size, bs) - off;
3971         }
3972
3973         if (len == 0) {
3974                 ret = 0;
3975                 goto out_unlock;
3976         }
3977
3978         /* verify the end result is block aligned */
3979         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3980             !IS_ALIGNED(destoff, bs))
3981                 goto out_unlock;
3982
3983         /* verify if ranges are overlapped within the same file */
3984         if (same_inode) {
3985                 if (destoff + len > off && destoff < off + len)
3986                         goto out_unlock;
3987         }
3988
3989         if (destoff > inode->i_size) {
3990                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3991                 if (ret)
3992                         goto out_unlock;
3993         }
3994
3995         /*
3996          * Lock the target range too. Right after we replace the file extent
3997          * items in the fs tree (which now point to the cloned data), we might
3998          * have a worker replace them with extent items relative to a write
3999          * operation that was issued before this clone operation (i.e. confront
4000          * with inode.c:btrfs_finish_ordered_io).
4001          */
4002         if (same_inode) {
4003                 u64 lock_start = min_t(u64, off, destoff);
4004                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
4005
4006                 ret = lock_extent_range(src, lock_start, lock_len, true);
4007         } else {
4008                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
4009                                                true);
4010         }
4011         ASSERT(ret == 0);
4012         if (WARN_ON(ret)) {
4013                 /* ranges in the io trees already unlocked */
4014                 goto out_unlock;
4015         }
4016
4017         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
4018
4019         if (same_inode) {
4020                 u64 lock_start = min_t(u64, off, destoff);
4021                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
4022
4023                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
4024         } else {
4025                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
4026         }
4027         /*
4028          * Truncate page cache pages so that future reads will see the cloned
4029          * data immediately and not the previous data.
4030          */
4031         truncate_inode_pages_range(&inode->i_data,
4032                                 round_down(destoff, PAGE_SIZE),
4033                                 round_up(destoff + len, PAGE_SIZE) - 1);
4034 out_unlock:
4035         if (!same_inode)
4036                 btrfs_double_inode_unlock(src, inode);
4037         else
4038                 inode_unlock(src);
4039         return ret;
4040 }
4041
4042 ssize_t btrfs_copy_file_range(struct file *file_in, loff_t pos_in,
4043                               struct file *file_out, loff_t pos_out,
4044                               size_t len, unsigned int flags)
4045 {
4046         ssize_t ret;
4047
4048         ret = btrfs_clone_files(file_out, file_in, pos_in, len, pos_out);
4049         if (ret == 0)
4050                 ret = len;
4051         return ret;
4052 }
4053
4054 int btrfs_clone_file_range(struct file *src_file, loff_t off,
4055                 struct file *dst_file, loff_t destoff, u64 len)
4056 {
4057         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
4058 }
4059
4060 /*
4061  * there are many ways the trans_start and trans_end ioctls can lead
4062  * to deadlocks.  They should only be used by applications that
4063  * basically own the machine, and have a very in depth understanding
4064  * of all the possible deadlocks and enospc problems.
4065  */
4066 static long btrfs_ioctl_trans_start(struct file *file)
4067 {
4068         struct inode *inode = file_inode(file);
4069         struct btrfs_root *root = BTRFS_I(inode)->root;
4070         struct btrfs_trans_handle *trans;
4071         int ret;
4072
4073         ret = -EPERM;
4074         if (!capable(CAP_SYS_ADMIN))
4075                 goto out;
4076
4077         ret = -EINPROGRESS;
4078         if (file->private_data)
4079                 goto out;
4080
4081         ret = -EROFS;
4082         if (btrfs_root_readonly(root))
4083                 goto out;
4084
4085         ret = mnt_want_write_file(file);
4086         if (ret)
4087                 goto out;
4088
4089         atomic_inc(&root->fs_info->open_ioctl_trans);
4090
4091         ret = -ENOMEM;
4092         trans = btrfs_start_ioctl_transaction(root);
4093         if (IS_ERR(trans))
4094                 goto out_drop;
4095
4096         file->private_data = trans;
4097         return 0;
4098
4099 out_drop:
4100         atomic_dec(&root->fs_info->open_ioctl_trans);
4101         mnt_drop_write_file(file);
4102 out:
4103         return ret;
4104 }
4105
4106 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4107 {
4108         struct inode *inode = file_inode(file);
4109         struct btrfs_root *root = BTRFS_I(inode)->root;
4110         struct btrfs_root *new_root;
4111         struct btrfs_dir_item *di;
4112         struct btrfs_trans_handle *trans;
4113         struct btrfs_path *path;
4114         struct btrfs_key location;
4115         struct btrfs_disk_key disk_key;
4116         u64 objectid = 0;
4117         u64 dir_id;
4118         int ret;
4119
4120         if (!capable(CAP_SYS_ADMIN))
4121                 return -EPERM;
4122
4123         ret = mnt_want_write_file(file);
4124         if (ret)
4125                 return ret;
4126
4127         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4128                 ret = -EFAULT;
4129                 goto out;
4130         }
4131
4132         if (!objectid)
4133                 objectid = BTRFS_FS_TREE_OBJECTID;
4134
4135         location.objectid = objectid;
4136         location.type = BTRFS_ROOT_ITEM_KEY;
4137         location.offset = (u64)-1;
4138
4139         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4140         if (IS_ERR(new_root)) {
4141                 ret = PTR_ERR(new_root);
4142                 goto out;
4143         }
4144         if (!is_fstree(new_root->objectid)) {
4145                 ret = -ENOENT;
4146                 goto out;
4147         }
4148
4149         path = btrfs_alloc_path();
4150         if (!path) {
4151                 ret = -ENOMEM;
4152                 goto out;
4153         }
4154         path->leave_spinning = 1;
4155
4156         trans = btrfs_start_transaction(root, 1);
4157         if (IS_ERR(trans)) {
4158                 btrfs_free_path(path);
4159                 ret = PTR_ERR(trans);
4160                 goto out;
4161         }
4162
4163         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
4164         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
4165                                    dir_id, "default", 7, 1);
4166         if (IS_ERR_OR_NULL(di)) {
4167                 btrfs_free_path(path);
4168                 btrfs_end_transaction(trans, root);
4169                 btrfs_err(new_root->fs_info,
4170                           "Umm, you don't have the default diritem, this isn't going to work");
4171                 ret = -ENOENT;
4172                 goto out;
4173         }
4174
4175         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4176         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4177         btrfs_mark_buffer_dirty(path->nodes[0]);
4178         btrfs_free_path(path);
4179
4180         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
4181         btrfs_end_transaction(trans, root);
4182 out:
4183         mnt_drop_write_file(file);
4184         return ret;
4185 }
4186
4187 void btrfs_get_block_group_info(struct list_head *groups_list,
4188                                 struct btrfs_ioctl_space_info *space)
4189 {
4190         struct btrfs_block_group_cache *block_group;
4191
4192         space->total_bytes = 0;
4193         space->used_bytes = 0;
4194         space->flags = 0;
4195         list_for_each_entry(block_group, groups_list, list) {
4196                 space->flags = block_group->flags;
4197                 space->total_bytes += block_group->key.offset;
4198                 space->used_bytes +=
4199                         btrfs_block_group_used(&block_group->item);
4200         }
4201 }
4202
4203 static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
4204 {
4205         struct btrfs_ioctl_space_args space_args;
4206         struct btrfs_ioctl_space_info space;
4207         struct btrfs_ioctl_space_info *dest;
4208         struct btrfs_ioctl_space_info *dest_orig;
4209         struct btrfs_ioctl_space_info __user *user_dest;
4210         struct btrfs_space_info *info;
4211         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4212                        BTRFS_BLOCK_GROUP_SYSTEM,
4213                        BTRFS_BLOCK_GROUP_METADATA,
4214                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4215         int num_types = 4;
4216         int alloc_size;
4217         int ret = 0;
4218         u64 slot_count = 0;
4219         int i, c;
4220
4221         if (copy_from_user(&space_args,
4222                            (struct btrfs_ioctl_space_args __user *)arg,
4223                            sizeof(space_args)))
4224                 return -EFAULT;
4225
4226         for (i = 0; i < num_types; i++) {
4227                 struct btrfs_space_info *tmp;
4228
4229                 info = NULL;
4230                 rcu_read_lock();
4231                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4232                                         list) {
4233                         if (tmp->flags == types[i]) {
4234                                 info = tmp;
4235                                 break;
4236                         }
4237                 }
4238                 rcu_read_unlock();
4239
4240                 if (!info)
4241                         continue;
4242
4243                 down_read(&info->groups_sem);
4244                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4245                         if (!list_empty(&info->block_groups[c]))
4246                                 slot_count++;
4247                 }
4248                 up_read(&info->groups_sem);
4249         }
4250
4251         /*
4252          * Global block reserve, exported as a space_info
4253          */
4254         slot_count++;
4255
4256         /* space_slots == 0 means they are asking for a count */
4257         if (space_args.space_slots == 0) {
4258                 space_args.total_spaces = slot_count;
4259                 goto out;
4260         }
4261
4262         slot_count = min_t(u64, space_args.space_slots, slot_count);
4263
4264         alloc_size = sizeof(*dest) * slot_count;
4265
4266         /* we generally have at most 6 or so space infos, one for each raid
4267          * level.  So, a whole page should be more than enough for everyone
4268          */
4269         if (alloc_size > PAGE_SIZE)
4270                 return -ENOMEM;
4271
4272         space_args.total_spaces = 0;
4273         dest = kmalloc(alloc_size, GFP_KERNEL);
4274         if (!dest)
4275                 return -ENOMEM;
4276         dest_orig = dest;
4277
4278         /* now we have a buffer to copy into */
4279         for (i = 0; i < num_types; i++) {
4280                 struct btrfs_space_info *tmp;
4281
4282                 if (!slot_count)
4283                         break;
4284
4285                 info = NULL;
4286                 rcu_read_lock();
4287                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
4288                                         list) {
4289                         if (tmp->flags == types[i]) {
4290                                 info = tmp;
4291                                 break;
4292                         }
4293                 }
4294                 rcu_read_unlock();
4295
4296                 if (!info)
4297                         continue;
4298                 down_read(&info->groups_sem);
4299                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4300                         if (!list_empty(&info->block_groups[c])) {
4301                                 btrfs_get_block_group_info(
4302                                         &info->block_groups[c], &space);
4303                                 memcpy(dest, &space, sizeof(space));
4304                                 dest++;
4305                                 space_args.total_spaces++;
4306                                 slot_count--;
4307                         }
4308                         if (!slot_count)
4309                                 break;
4310                 }
4311                 up_read(&info->groups_sem);
4312         }
4313
4314         /*
4315          * Add global block reserve
4316          */
4317         if (slot_count) {
4318                 struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv;
4319
4320                 spin_lock(&block_rsv->lock);
4321                 space.total_bytes = block_rsv->size;
4322                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4323                 spin_unlock(&block_rsv->lock);
4324                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4325                 memcpy(dest, &space, sizeof(space));
4326                 space_args.total_spaces++;
4327         }
4328
4329         user_dest = (struct btrfs_ioctl_space_info __user *)
4330                 (arg + sizeof(struct btrfs_ioctl_space_args));
4331
4332         if (copy_to_user(user_dest, dest_orig, alloc_size))
4333                 ret = -EFAULT;
4334
4335         kfree(dest_orig);
4336 out:
4337         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4338                 ret = -EFAULT;
4339
4340         return ret;
4341 }
4342
4343 /*
4344  * there are many ways the trans_start and trans_end ioctls can lead
4345  * to deadlocks.  They should only be used by applications that
4346  * basically own the machine, and have a very in depth understanding
4347  * of all the possible deadlocks and enospc problems.
4348  */
4349 long btrfs_ioctl_trans_end(struct file *file)
4350 {
4351         struct inode *inode = file_inode(file);
4352         struct btrfs_root *root = BTRFS_I(inode)->root;
4353         struct btrfs_trans_handle *trans;
4354
4355         trans = file->private_data;
4356         if (!trans)
4357                 return -EINVAL;
4358         file->private_data = NULL;
4359
4360         btrfs_end_transaction(trans, root);
4361
4362         atomic_dec(&root->fs_info->open_ioctl_trans);
4363
4364         mnt_drop_write_file(file);
4365         return 0;
4366 }
4367
4368 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4369                                             void __user *argp)
4370 {
4371         struct btrfs_trans_handle *trans;
4372         u64 transid;
4373         int ret;
4374
4375         trans = btrfs_attach_transaction_barrier(root);
4376         if (IS_ERR(trans)) {
4377                 if (PTR_ERR(trans) != -ENOENT)
4378                         return PTR_ERR(trans);
4379
4380                 /* No running transaction, don't bother */
4381                 transid = root->fs_info->last_trans_committed;
4382                 goto out;
4383         }
4384         transid = trans->transid;
4385         ret = btrfs_commit_transaction_async(trans, root, 0);
4386         if (ret) {
4387                 btrfs_end_transaction(trans, root);
4388                 return ret;
4389         }
4390 out:
4391         if (argp)
4392                 if (copy_to_user(argp, &transid, sizeof(transid)))
4393                         return -EFAULT;
4394         return 0;
4395 }
4396
4397 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
4398                                            void __user *argp)
4399 {
4400         u64 transid;
4401
4402         if (argp) {
4403                 if (copy_from_user(&transid, argp, sizeof(transid)))
4404                         return -EFAULT;
4405         } else {
4406                 transid = 0;  /* current trans */
4407         }
4408         return btrfs_wait_for_commit(root, transid);
4409 }
4410
4411 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4412 {
4413         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4414         struct btrfs_ioctl_scrub_args *sa;
4415         int ret;
4416
4417         if (!capable(CAP_SYS_ADMIN))
4418                 return -EPERM;
4419
4420         sa = memdup_user(arg, sizeof(*sa));
4421         if (IS_ERR(sa))
4422                 return PTR_ERR(sa);
4423
4424         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4425                 ret = mnt_want_write_file(file);
4426                 if (ret)
4427                         goto out;
4428         }
4429
4430         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
4431                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4432                               0);
4433
4434         if (copy_to_user(arg, sa, sizeof(*sa)))
4435                 ret = -EFAULT;
4436
4437         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4438                 mnt_drop_write_file(file);
4439 out:
4440         kfree(sa);
4441         return ret;
4442 }
4443
4444 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
4445 {
4446         if (!capable(CAP_SYS_ADMIN))
4447                 return -EPERM;
4448
4449         return btrfs_scrub_cancel(root->fs_info);
4450 }
4451
4452 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
4453                                        void __user *arg)
4454 {
4455         struct btrfs_ioctl_scrub_args *sa;
4456         int ret;
4457
4458         if (!capable(CAP_SYS_ADMIN))
4459                 return -EPERM;
4460
4461         sa = memdup_user(arg, sizeof(*sa));
4462         if (IS_ERR(sa))
4463                 return PTR_ERR(sa);
4464
4465         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
4466
4467         if (copy_to_user(arg, sa, sizeof(*sa)))
4468                 ret = -EFAULT;
4469
4470         kfree(sa);
4471         return ret;
4472 }
4473
4474 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
4475                                       void __user *arg)
4476 {
4477         struct btrfs_ioctl_get_dev_stats *sa;
4478         int ret;
4479
4480         sa = memdup_user(arg, sizeof(*sa));
4481         if (IS_ERR(sa))
4482                 return PTR_ERR(sa);
4483
4484         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4485                 kfree(sa);
4486                 return -EPERM;
4487         }
4488
4489         ret = btrfs_get_dev_stats(root, sa);
4490
4491         if (copy_to_user(arg, sa, sizeof(*sa)))
4492                 ret = -EFAULT;
4493
4494         kfree(sa);
4495         return ret;
4496 }
4497
4498 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
4499 {
4500         struct btrfs_ioctl_dev_replace_args *p;
4501         int ret;
4502
4503         if (!capable(CAP_SYS_ADMIN))
4504                 return -EPERM;
4505
4506         p = memdup_user(arg, sizeof(*p));
4507         if (IS_ERR(p))
4508                 return PTR_ERR(p);
4509
4510         switch (p->cmd) {
4511         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4512                 if (root->fs_info->sb->s_flags & MS_RDONLY) {
4513                         ret = -EROFS;
4514                         goto out;
4515                 }
4516                 if (atomic_xchg(
4517                         &root->fs_info->mutually_exclusive_operation_running,
4518                         1)) {
4519                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4520                 } else {
4521                         ret = btrfs_dev_replace_by_ioctl(root, p);
4522                         atomic_set(
4523                          &root->fs_info->mutually_exclusive_operation_running,
4524                          0);
4525                 }
4526                 break;
4527         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4528                 btrfs_dev_replace_status(root->fs_info, p);
4529                 ret = 0;
4530                 break;
4531         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4532                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
4533                 break;
4534         default:
4535                 ret = -EINVAL;
4536                 break;
4537         }
4538
4539         if (copy_to_user(arg, p, sizeof(*p)))
4540                 ret = -EFAULT;
4541 out:
4542         kfree(p);
4543         return ret;
4544 }
4545
4546 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4547 {
4548         int ret = 0;
4549         int i;
4550         u64 rel_ptr;
4551         int size;
4552         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4553         struct inode_fs_paths *ipath = NULL;
4554         struct btrfs_path *path;
4555
4556         if (!capable(CAP_DAC_READ_SEARCH))
4557                 return -EPERM;
4558
4559         path = btrfs_alloc_path();
4560         if (!path) {
4561                 ret = -ENOMEM;
4562                 goto out;
4563         }
4564
4565         ipa = memdup_user(arg, sizeof(*ipa));
4566         if (IS_ERR(ipa)) {
4567                 ret = PTR_ERR(ipa);
4568                 ipa = NULL;
4569                 goto out;
4570         }
4571
4572         size = min_t(u32, ipa->size, 4096);
4573         ipath = init_ipath(size, root, path);
4574         if (IS_ERR(ipath)) {
4575                 ret = PTR_ERR(ipath);
4576                 ipath = NULL;
4577                 goto out;
4578         }
4579
4580         ret = paths_from_inode(ipa->inum, ipath);
4581         if (ret < 0)
4582                 goto out;
4583
4584         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4585                 rel_ptr = ipath->fspath->val[i] -
4586                           (u64)(unsigned long)ipath->fspath->val;
4587                 ipath->fspath->val[i] = rel_ptr;
4588         }
4589
4590         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4591                            (void *)(unsigned long)ipath->fspath, size);
4592         if (ret) {
4593                 ret = -EFAULT;
4594                 goto out;
4595         }
4596
4597 out:
4598         btrfs_free_path(path);
4599         free_ipath(ipath);
4600         kfree(ipa);
4601
4602         return ret;
4603 }
4604
4605 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4606 {
4607         struct btrfs_data_container *inodes = ctx;
4608         const size_t c = 3 * sizeof(u64);
4609
4610         if (inodes->bytes_left >= c) {
4611                 inodes->bytes_left -= c;
4612                 inodes->val[inodes->elem_cnt] = inum;
4613                 inodes->val[inodes->elem_cnt + 1] = offset;
4614                 inodes->val[inodes->elem_cnt + 2] = root;
4615                 inodes->elem_cnt += 3;
4616         } else {
4617                 inodes->bytes_missing += c - inodes->bytes_left;
4618                 inodes->bytes_left = 0;
4619                 inodes->elem_missed += 3;
4620         }
4621
4622         return 0;
4623 }
4624
4625 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
4626                                         void __user *arg)
4627 {
4628         int ret = 0;
4629         int size;
4630         struct btrfs_ioctl_logical_ino_args *loi;
4631         struct btrfs_data_container *inodes = NULL;
4632         struct btrfs_path *path = NULL;
4633
4634         if (!capable(CAP_SYS_ADMIN))
4635                 return -EPERM;
4636
4637         loi = memdup_user(arg, sizeof(*loi));
4638         if (IS_ERR(loi)) {
4639                 ret = PTR_ERR(loi);
4640                 loi = NULL;
4641                 goto out;
4642         }
4643
4644         path = btrfs_alloc_path();
4645         if (!path) {
4646                 ret = -ENOMEM;
4647                 goto out;
4648         }
4649
4650         size = min_t(u32, loi->size, SZ_64K);
4651         inodes = init_data_container(size);
4652         if (IS_ERR(inodes)) {
4653                 ret = PTR_ERR(inodes);
4654                 inodes = NULL;
4655                 goto out;
4656         }
4657
4658         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
4659                                           build_ino_list, inodes);
4660         if (ret == -EINVAL)
4661                 ret = -ENOENT;
4662         if (ret < 0)
4663                 goto out;
4664
4665         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4666                            (void *)(unsigned long)inodes, size);
4667         if (ret)
4668                 ret = -EFAULT;
4669
4670 out:
4671         btrfs_free_path(path);
4672         vfree(inodes);
4673         kfree(loi);
4674
4675         return ret;
4676 }
4677
4678 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4679                                struct btrfs_ioctl_balance_args *bargs)
4680 {
4681         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4682
4683         bargs->flags = bctl->flags;
4684
4685         if (atomic_read(&fs_info->balance_running))
4686                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4687         if (atomic_read(&fs_info->balance_pause_req))
4688                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4689         if (atomic_read(&fs_info->balance_cancel_req))
4690                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4691
4692         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4693         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4694         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4695
4696         if (lock) {
4697                 spin_lock(&fs_info->balance_lock);
4698                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4699                 spin_unlock(&fs_info->balance_lock);
4700         } else {
4701                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4702         }
4703 }
4704
4705 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4706 {
4707         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4708         struct btrfs_fs_info *fs_info = root->fs_info;
4709         struct btrfs_ioctl_balance_args *bargs;
4710         struct btrfs_balance_control *bctl;
4711         bool need_unlock; /* for mut. excl. ops lock */
4712         int ret;
4713
4714         if (!capable(CAP_SYS_ADMIN))
4715                 return -EPERM;
4716
4717         ret = mnt_want_write_file(file);
4718         if (ret)
4719                 return ret;
4720
4721 again:
4722         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
4723                 mutex_lock(&fs_info->volume_mutex);
4724                 mutex_lock(&fs_info->balance_mutex);
4725                 need_unlock = true;
4726                 goto locked;
4727         }
4728
4729         /*
4730          * mut. excl. ops lock is locked.  Three possibilities:
4731          *   (1) some other op is running
4732          *   (2) balance is running
4733          *   (3) balance is paused -- special case (think resume)
4734          */
4735         mutex_lock(&fs_info->balance_mutex);
4736         if (fs_info->balance_ctl) {
4737                 /* this is either (2) or (3) */
4738                 if (!atomic_read(&fs_info->balance_running)) {
4739                         mutex_unlock(&fs_info->balance_mutex);
4740                         if (!mutex_trylock(&fs_info->volume_mutex))
4741                                 goto again;
4742                         mutex_lock(&fs_info->balance_mutex);
4743
4744                         if (fs_info->balance_ctl &&
4745                             !atomic_read(&fs_info->balance_running)) {
4746                                 /* this is (3) */
4747                                 need_unlock = false;
4748                                 goto locked;
4749                         }
4750
4751                         mutex_unlock(&fs_info->balance_mutex);
4752                         mutex_unlock(&fs_info->volume_mutex);
4753                         goto again;
4754                 } else {
4755                         /* this is (2) */
4756                         mutex_unlock(&fs_info->balance_mutex);
4757                         ret = -EINPROGRESS;
4758                         goto out;
4759                 }
4760         } else {
4761                 /* this is (1) */
4762                 mutex_unlock(&fs_info->balance_mutex);
4763                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4764                 goto out;
4765         }
4766
4767 locked:
4768         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
4769
4770         if (arg) {
4771                 bargs = memdup_user(arg, sizeof(*bargs));
4772                 if (IS_ERR(bargs)) {
4773                         ret = PTR_ERR(bargs);
4774                         goto out_unlock;
4775                 }
4776
4777                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4778                         if (!fs_info->balance_ctl) {
4779                                 ret = -ENOTCONN;
4780                                 goto out_bargs;
4781                         }
4782
4783                         bctl = fs_info->balance_ctl;
4784                         spin_lock(&fs_info->balance_lock);
4785                         bctl->flags |= BTRFS_BALANCE_RESUME;
4786                         spin_unlock(&fs_info->balance_lock);
4787
4788                         goto do_balance;
4789                 }
4790         } else {
4791                 bargs = NULL;
4792         }
4793
4794         if (fs_info->balance_ctl) {
4795                 ret = -EINPROGRESS;
4796                 goto out_bargs;
4797         }
4798
4799         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4800         if (!bctl) {
4801                 ret = -ENOMEM;
4802                 goto out_bargs;
4803         }
4804
4805         bctl->fs_info = fs_info;
4806         if (arg) {
4807                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4808                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4809                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4810
4811                 bctl->flags = bargs->flags;
4812         } else {
4813                 /* balance everything - no filters */
4814                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4815         }
4816
4817         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4818                 ret = -EINVAL;
4819                 goto out_bctl;
4820         }
4821
4822 do_balance:
4823         /*
4824          * Ownership of bctl and mutually_exclusive_operation_running
4825          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4826          * or, if restriper was paused all the way until unmount, in
4827          * free_fs_info.  mutually_exclusive_operation_running is
4828          * cleared in __cancel_balance.
4829          */
4830         need_unlock = false;
4831
4832         ret = btrfs_balance(bctl, bargs);
4833         bctl = NULL;
4834
4835         if (arg) {
4836                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4837                         ret = -EFAULT;
4838         }
4839
4840 out_bctl:
4841         kfree(bctl);
4842 out_bargs:
4843         kfree(bargs);
4844 out_unlock:
4845         mutex_unlock(&fs_info->balance_mutex);
4846         mutex_unlock(&fs_info->volume_mutex);
4847         if (need_unlock)
4848                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
4849 out:
4850         mnt_drop_write_file(file);
4851         return ret;
4852 }
4853
4854 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
4855 {
4856         if (!capable(CAP_SYS_ADMIN))
4857                 return -EPERM;
4858
4859         switch (cmd) {
4860         case BTRFS_BALANCE_CTL_PAUSE:
4861                 return btrfs_pause_balance(root->fs_info);
4862         case BTRFS_BALANCE_CTL_CANCEL:
4863                 return btrfs_cancel_balance(root->fs_info);
4864         }
4865
4866         return -EINVAL;
4867 }
4868
4869 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
4870                                          void __user *arg)
4871 {
4872         struct btrfs_fs_info *fs_info = root->fs_info;
4873         struct btrfs_ioctl_balance_args *bargs;
4874         int ret = 0;
4875
4876         if (!capable(CAP_SYS_ADMIN))
4877                 return -EPERM;
4878
4879         mutex_lock(&fs_info->balance_mutex);
4880         if (!fs_info->balance_ctl) {
4881                 ret = -ENOTCONN;
4882                 goto out;
4883         }
4884
4885         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4886         if (!bargs) {
4887                 ret = -ENOMEM;
4888                 goto out;
4889         }
4890
4891         update_ioctl_balance_args(fs_info, 1, bargs);
4892
4893         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4894                 ret = -EFAULT;
4895
4896         kfree(bargs);
4897 out:
4898         mutex_unlock(&fs_info->balance_mutex);
4899         return ret;
4900 }
4901
4902 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4903 {
4904         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4905         struct btrfs_ioctl_quota_ctl_args *sa;
4906         struct btrfs_trans_handle *trans = NULL;
4907         int ret;
4908         int err;
4909
4910         if (!capable(CAP_SYS_ADMIN))
4911                 return -EPERM;
4912
4913         ret = mnt_want_write_file(file);
4914         if (ret)
4915                 return ret;
4916
4917         sa = memdup_user(arg, sizeof(*sa));
4918         if (IS_ERR(sa)) {
4919                 ret = PTR_ERR(sa);
4920                 goto drop_write;
4921         }
4922
4923         down_write(&root->fs_info->subvol_sem);
4924         trans = btrfs_start_transaction(root->fs_info->tree_root, 2);
4925         if (IS_ERR(trans)) {
4926                 ret = PTR_ERR(trans);
4927                 goto out;
4928         }
4929
4930         switch (sa->cmd) {
4931         case BTRFS_QUOTA_CTL_ENABLE:
4932                 ret = btrfs_quota_enable(trans, root->fs_info);
4933                 break;
4934         case BTRFS_QUOTA_CTL_DISABLE:
4935                 ret = btrfs_quota_disable(trans, root->fs_info);
4936                 break;
4937         default:
4938                 ret = -EINVAL;
4939                 break;
4940         }
4941
4942         err = btrfs_commit_transaction(trans, root->fs_info->tree_root);
4943         if (err && !ret)
4944                 ret = err;
4945 out:
4946         kfree(sa);
4947         up_write(&root->fs_info->subvol_sem);
4948 drop_write:
4949         mnt_drop_write_file(file);
4950         return ret;
4951 }
4952
4953 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4954 {
4955         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4956         struct btrfs_ioctl_qgroup_assign_args *sa;
4957         struct btrfs_trans_handle *trans;
4958         int ret;
4959         int err;
4960
4961         if (!capable(CAP_SYS_ADMIN))
4962                 return -EPERM;
4963
4964         ret = mnt_want_write_file(file);
4965         if (ret)
4966                 return ret;
4967
4968         sa = memdup_user(arg, sizeof(*sa));
4969         if (IS_ERR(sa)) {
4970                 ret = PTR_ERR(sa);
4971                 goto drop_write;
4972         }
4973
4974         trans = btrfs_join_transaction(root);
4975         if (IS_ERR(trans)) {
4976                 ret = PTR_ERR(trans);
4977                 goto out;
4978         }
4979
4980         /* FIXME: check if the IDs really exist */
4981         if (sa->assign) {
4982                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
4983                                                 sa->src, sa->dst);
4984         } else {
4985                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
4986                                                 sa->src, sa->dst);
4987         }
4988
4989         /* update qgroup status and info */
4990         err = btrfs_run_qgroups(trans, root->fs_info);
4991         if (err < 0)
4992                 btrfs_handle_fs_error(root->fs_info, err,
4993                             "failed to update qgroup status and info");
4994         err = btrfs_end_transaction(trans, root);
4995         if (err && !ret)
4996                 ret = err;
4997
4998 out:
4999         kfree(sa);
5000 drop_write:
5001         mnt_drop_write_file(file);
5002         return ret;
5003 }
5004
5005 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
5006 {
5007         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5008         struct btrfs_ioctl_qgroup_create_args *sa;
5009         struct btrfs_trans_handle *trans;
5010         int ret;
5011         int err;
5012
5013         if (!capable(CAP_SYS_ADMIN))
5014                 return -EPERM;
5015
5016         ret = mnt_want_write_file(file);
5017         if (ret)
5018                 return ret;
5019
5020         sa = memdup_user(arg, sizeof(*sa));
5021         if (IS_ERR(sa)) {
5022                 ret = PTR_ERR(sa);
5023                 goto drop_write;
5024         }
5025
5026         if (!sa->qgroupid) {
5027                 ret = -EINVAL;
5028                 goto out;
5029         }
5030
5031         trans = btrfs_join_transaction(root);
5032         if (IS_ERR(trans)) {
5033                 ret = PTR_ERR(trans);
5034                 goto out;
5035         }
5036
5037         /* FIXME: check if the IDs really exist */
5038         if (sa->create) {
5039                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid);
5040         } else {
5041                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
5042         }
5043
5044         err = btrfs_end_transaction(trans, root);
5045         if (err && !ret)
5046                 ret = err;
5047
5048 out:
5049         kfree(sa);
5050 drop_write:
5051         mnt_drop_write_file(file);
5052         return ret;
5053 }
5054
5055 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5056 {
5057         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5058         struct btrfs_ioctl_qgroup_limit_args *sa;
5059         struct btrfs_trans_handle *trans;
5060         int ret;
5061         int err;
5062         u64 qgroupid;
5063
5064         if (!capable(CAP_SYS_ADMIN))
5065                 return -EPERM;
5066
5067         ret = mnt_want_write_file(file);
5068         if (ret)
5069                 return ret;
5070
5071         sa = memdup_user(arg, sizeof(*sa));
5072         if (IS_ERR(sa)) {
5073                 ret = PTR_ERR(sa);
5074                 goto drop_write;
5075         }
5076
5077         trans = btrfs_join_transaction(root);
5078         if (IS_ERR(trans)) {
5079                 ret = PTR_ERR(trans);
5080                 goto out;
5081         }
5082
5083         qgroupid = sa->qgroupid;
5084         if (!qgroupid) {
5085                 /* take the current subvol as qgroup */
5086                 qgroupid = root->root_key.objectid;
5087         }
5088
5089         /* FIXME: check if the IDs really exist */
5090         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
5091
5092         err = btrfs_end_transaction(trans, root);
5093         if (err && !ret)
5094                 ret = err;
5095
5096 out:
5097         kfree(sa);
5098 drop_write:
5099         mnt_drop_write_file(file);
5100         return ret;
5101 }
5102
5103 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5104 {
5105         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5106         struct btrfs_ioctl_quota_rescan_args *qsa;
5107         int ret;
5108
5109         if (!capable(CAP_SYS_ADMIN))
5110                 return -EPERM;
5111
5112         ret = mnt_want_write_file(file);
5113         if (ret)
5114                 return ret;
5115
5116         qsa = memdup_user(arg, sizeof(*qsa));
5117         if (IS_ERR(qsa)) {
5118                 ret = PTR_ERR(qsa);
5119                 goto drop_write;
5120         }
5121
5122         if (qsa->flags) {
5123                 ret = -EINVAL;
5124                 goto out;
5125         }
5126
5127         ret = btrfs_qgroup_rescan(root->fs_info);
5128
5129 out:
5130         kfree(qsa);
5131 drop_write:
5132         mnt_drop_write_file(file);
5133         return ret;
5134 }
5135
5136 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5137 {
5138         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5139         struct btrfs_ioctl_quota_rescan_args *qsa;
5140         int ret = 0;
5141
5142         if (!capable(CAP_SYS_ADMIN))
5143                 return -EPERM;
5144
5145         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5146         if (!qsa)
5147                 return -ENOMEM;
5148
5149         if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5150                 qsa->flags = 1;
5151                 qsa->progress = root->fs_info->qgroup_rescan_progress.objectid;
5152         }
5153
5154         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5155                 ret = -EFAULT;
5156
5157         kfree(qsa);
5158         return ret;
5159 }
5160
5161 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5162 {
5163         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5164
5165         if (!capable(CAP_SYS_ADMIN))
5166                 return -EPERM;
5167
5168         return btrfs_qgroup_wait_for_completion(root->fs_info, true);
5169 }
5170
5171 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5172                                             struct btrfs_ioctl_received_subvol_args *sa)
5173 {
5174         struct inode *inode = file_inode(file);
5175         struct btrfs_root *root = BTRFS_I(inode)->root;
5176         struct btrfs_root_item *root_item = &root->root_item;
5177         struct btrfs_trans_handle *trans;
5178         struct timespec ct = current_time(inode);
5179         int ret = 0;
5180         int received_uuid_changed;
5181
5182         if (!inode_owner_or_capable(inode))
5183                 return -EPERM;
5184
5185         ret = mnt_want_write_file(file);
5186         if (ret < 0)
5187                 return ret;
5188
5189         down_write(&root->fs_info->subvol_sem);
5190
5191         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
5192                 ret = -EINVAL;
5193                 goto out;
5194         }
5195
5196         if (btrfs_root_readonly(root)) {
5197                 ret = -EROFS;
5198                 goto out;
5199         }
5200
5201         /*
5202          * 1 - root item
5203          * 2 - uuid items (received uuid + subvol uuid)
5204          */
5205         trans = btrfs_start_transaction(root, 3);
5206         if (IS_ERR(trans)) {
5207                 ret = PTR_ERR(trans);
5208                 trans = NULL;
5209                 goto out;
5210         }
5211
5212         sa->rtransid = trans->transid;
5213         sa->rtime.sec = ct.tv_sec;
5214         sa->rtime.nsec = ct.tv_nsec;
5215
5216         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5217                                        BTRFS_UUID_SIZE);
5218         if (received_uuid_changed &&
5219             !btrfs_is_empty_uuid(root_item->received_uuid))
5220                 btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root,
5221                                     root_item->received_uuid,
5222                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5223                                     root->root_key.objectid);
5224         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5225         btrfs_set_root_stransid(root_item, sa->stransid);
5226         btrfs_set_root_rtransid(root_item, sa->rtransid);
5227         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5228         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5229         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5230         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5231
5232         ret = btrfs_update_root(trans, root->fs_info->tree_root,
5233                                 &root->root_key, &root->root_item);
5234         if (ret < 0) {
5235                 btrfs_end_transaction(trans, root);
5236                 goto out;
5237         }
5238         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5239                 ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root,
5240                                           sa->uuid,
5241                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5242                                           root->root_key.objectid);
5243                 if (ret < 0 && ret != -EEXIST) {
5244                         btrfs_abort_transaction(trans, ret);
5245                         goto out;
5246                 }
5247         }
5248         ret = btrfs_commit_transaction(trans, root);
5249         if (ret < 0) {
5250                 btrfs_abort_transaction(trans, ret);
5251                 goto out;
5252         }
5253
5254 out:
5255         up_write(&root->fs_info->subvol_sem);
5256         mnt_drop_write_file(file);
5257         return ret;
5258 }
5259
5260 #ifdef CONFIG_64BIT
5261 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5262                                                 void __user *arg)
5263 {
5264         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5265         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5266         int ret = 0;
5267
5268         args32 = memdup_user(arg, sizeof(*args32));
5269         if (IS_ERR(args32)) {
5270                 ret = PTR_ERR(args32);
5271                 args32 = NULL;
5272                 goto out;
5273         }
5274
5275         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5276         if (!args64) {
5277                 ret = -ENOMEM;
5278                 goto out;
5279         }
5280
5281         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5282         args64->stransid = args32->stransid;
5283         args64->rtransid = args32->rtransid;
5284         args64->stime.sec = args32->stime.sec;
5285         args64->stime.nsec = args32->stime.nsec;
5286         args64->rtime.sec = args32->rtime.sec;
5287         args64->rtime.nsec = args32->rtime.nsec;
5288         args64->flags = args32->flags;
5289
5290         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5291         if (ret)
5292                 goto out;
5293
5294         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5295         args32->stransid = args64->stransid;
5296         args32->rtransid = args64->rtransid;
5297         args32->stime.sec = args64->stime.sec;
5298         args32->stime.nsec = args64->stime.nsec;
5299         args32->rtime.sec = args64->rtime.sec;
5300         args32->rtime.nsec = args64->rtime.nsec;
5301         args32->flags = args64->flags;
5302
5303         ret = copy_to_user(arg, args32, sizeof(*args32));
5304         if (ret)
5305                 ret = -EFAULT;
5306
5307 out:
5308         kfree(args32);
5309         kfree(args64);
5310         return ret;
5311 }
5312 #endif
5313
5314 static long btrfs_ioctl_set_received_subvol(struct file *file,
5315                                             void __user *arg)
5316 {
5317         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5318         int ret = 0;
5319
5320         sa = memdup_user(arg, sizeof(*sa));
5321         if (IS_ERR(sa)) {
5322                 ret = PTR_ERR(sa);
5323                 sa = NULL;
5324                 goto out;
5325         }
5326
5327         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5328
5329         if (ret)
5330                 goto out;
5331
5332         ret = copy_to_user(arg, sa, sizeof(*sa));
5333         if (ret)
5334                 ret = -EFAULT;
5335
5336 out:
5337         kfree(sa);
5338         return ret;
5339 }
5340
5341 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5342 {
5343         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5344         size_t len;
5345         int ret;
5346         char label[BTRFS_LABEL_SIZE];
5347
5348         spin_lock(&root->fs_info->super_lock);
5349         memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5350         spin_unlock(&root->fs_info->super_lock);
5351
5352         len = strnlen(label, BTRFS_LABEL_SIZE);
5353
5354         if (len == BTRFS_LABEL_SIZE) {
5355                 btrfs_warn(root->fs_info,
5356                         "label is too long, return the first %zu bytes", --len);
5357         }
5358
5359         ret = copy_to_user(arg, label, len);
5360
5361         return ret ? -EFAULT : 0;
5362 }
5363
5364 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5365 {
5366         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5367         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5368         struct btrfs_trans_handle *trans;
5369         char label[BTRFS_LABEL_SIZE];
5370         int ret;
5371
5372         if (!capable(CAP_SYS_ADMIN))
5373                 return -EPERM;
5374
5375         if (copy_from_user(label, arg, sizeof(label)))
5376                 return -EFAULT;
5377
5378         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5379                 btrfs_err(root->fs_info,
5380                           "unable to set label with more than %d bytes",
5381                           BTRFS_LABEL_SIZE - 1);
5382                 return -EINVAL;
5383         }
5384
5385         ret = mnt_want_write_file(file);
5386         if (ret)
5387                 return ret;
5388
5389         trans = btrfs_start_transaction(root, 0);
5390         if (IS_ERR(trans)) {
5391                 ret = PTR_ERR(trans);
5392                 goto out_unlock;
5393         }
5394
5395         spin_lock(&root->fs_info->super_lock);
5396         strcpy(super_block->label, label);
5397         spin_unlock(&root->fs_info->super_lock);
5398         ret = btrfs_commit_transaction(trans, root);
5399
5400 out_unlock:
5401         mnt_drop_write_file(file);
5402         return ret;
5403 }
5404
5405 #define INIT_FEATURE_FLAGS(suffix) \
5406         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5407           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5408           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5409
5410 int btrfs_ioctl_get_supported_features(void __user *arg)
5411 {
5412         static const struct btrfs_ioctl_feature_flags features[3] = {
5413                 INIT_FEATURE_FLAGS(SUPP),
5414                 INIT_FEATURE_FLAGS(SAFE_SET),
5415                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5416         };
5417
5418         if (copy_to_user(arg, &features, sizeof(features)))
5419                 return -EFAULT;
5420
5421         return 0;
5422 }
5423
5424 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5425 {
5426         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5427         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5428         struct btrfs_ioctl_feature_flags features;
5429
5430         features.compat_flags = btrfs_super_compat_flags(super_block);
5431         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5432         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5433
5434         if (copy_to_user(arg, &features, sizeof(features)))
5435                 return -EFAULT;
5436
5437         return 0;
5438 }
5439
5440 static int check_feature_bits(struct btrfs_root *root,
5441                               enum btrfs_feature_set set,
5442                               u64 change_mask, u64 flags, u64 supported_flags,
5443                               u64 safe_set, u64 safe_clear)
5444 {
5445         const char *type = btrfs_feature_set_names[set];
5446         char *names;
5447         u64 disallowed, unsupported;
5448         u64 set_mask = flags & change_mask;
5449         u64 clear_mask = ~flags & change_mask;
5450
5451         unsupported = set_mask & ~supported_flags;
5452         if (unsupported) {
5453                 names = btrfs_printable_features(set, unsupported);
5454                 if (names) {
5455                         btrfs_warn(root->fs_info,
5456                            "this kernel does not support the %s feature bit%s",
5457                            names, strchr(names, ',') ? "s" : "");
5458                         kfree(names);
5459                 } else
5460                         btrfs_warn(root->fs_info,
5461                            "this kernel does not support %s bits 0x%llx",
5462                            type, unsupported);
5463                 return -EOPNOTSUPP;
5464         }
5465
5466         disallowed = set_mask & ~safe_set;
5467         if (disallowed) {
5468                 names = btrfs_printable_features(set, disallowed);
5469                 if (names) {
5470                         btrfs_warn(root->fs_info,
5471                            "can't set the %s feature bit%s while mounted",
5472                            names, strchr(names, ',') ? "s" : "");
5473                         kfree(names);
5474                 } else
5475                         btrfs_warn(root->fs_info,
5476                            "can't set %s bits 0x%llx while mounted",
5477                            type, disallowed);
5478                 return -EPERM;
5479         }
5480
5481         disallowed = clear_mask & ~safe_clear;
5482         if (disallowed) {
5483                 names = btrfs_printable_features(set, disallowed);
5484                 if (names) {
5485                         btrfs_warn(root->fs_info,
5486                            "can't clear the %s feature bit%s while mounted",
5487                            names, strchr(names, ',') ? "s" : "");
5488                         kfree(names);
5489                 } else
5490                         btrfs_warn(root->fs_info,
5491                            "can't clear %s bits 0x%llx while mounted",
5492                            type, disallowed);
5493                 return -EPERM;
5494         }
5495
5496         return 0;
5497 }
5498
5499 #define check_feature(root, change_mask, flags, mask_base)      \
5500 check_feature_bits(root, FEAT_##mask_base, change_mask, flags,  \
5501                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5502                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5503                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5504
5505 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5506 {
5507         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5508         struct btrfs_super_block *super_block = root->fs_info->super_copy;
5509         struct btrfs_ioctl_feature_flags flags[2];
5510         struct btrfs_trans_handle *trans;
5511         u64 newflags;
5512         int ret;
5513
5514         if (!capable(CAP_SYS_ADMIN))
5515                 return -EPERM;
5516
5517         if (copy_from_user(flags, arg, sizeof(flags)))
5518                 return -EFAULT;
5519
5520         /* Nothing to do */
5521         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5522             !flags[0].incompat_flags)
5523                 return 0;
5524
5525         ret = check_feature(root, flags[0].compat_flags,
5526                             flags[1].compat_flags, COMPAT);
5527         if (ret)
5528                 return ret;
5529
5530         ret = check_feature(root, flags[0].compat_ro_flags,
5531                             flags[1].compat_ro_flags, COMPAT_RO);
5532         if (ret)
5533                 return ret;
5534
5535         ret = check_feature(root, flags[0].incompat_flags,
5536                             flags[1].incompat_flags, INCOMPAT);
5537         if (ret)
5538                 return ret;
5539
5540         ret = mnt_want_write_file(file);
5541         if (ret)
5542                 return ret;
5543
5544         trans = btrfs_start_transaction(root, 0);
5545         if (IS_ERR(trans)) {
5546                 ret = PTR_ERR(trans);
5547                 goto out_drop_write;
5548         }
5549
5550         spin_lock(&root->fs_info->super_lock);
5551         newflags = btrfs_super_compat_flags(super_block);
5552         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5553         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5554         btrfs_set_super_compat_flags(super_block, newflags);
5555
5556         newflags = btrfs_super_compat_ro_flags(super_block);
5557         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5558         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5559         btrfs_set_super_compat_ro_flags(super_block, newflags);
5560
5561         newflags = btrfs_super_incompat_flags(super_block);
5562         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5563         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5564         btrfs_set_super_incompat_flags(super_block, newflags);
5565         spin_unlock(&root->fs_info->super_lock);
5566
5567         ret = btrfs_commit_transaction(trans, root);
5568 out_drop_write:
5569         mnt_drop_write_file(file);
5570
5571         return ret;
5572 }
5573
5574 long btrfs_ioctl(struct file *file, unsigned int
5575                 cmd, unsigned long arg)
5576 {
5577         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5578         void __user *argp = (void __user *)arg;
5579
5580         switch (cmd) {
5581         case FS_IOC_GETFLAGS:
5582                 return btrfs_ioctl_getflags(file, argp);
5583         case FS_IOC_SETFLAGS:
5584                 return btrfs_ioctl_setflags(file, argp);
5585         case FS_IOC_GETVERSION:
5586                 return btrfs_ioctl_getversion(file, argp);
5587         case FITRIM:
5588                 return btrfs_ioctl_fitrim(file, argp);
5589         case BTRFS_IOC_SNAP_CREATE:
5590                 return btrfs_ioctl_snap_create(file, argp, 0);
5591         case BTRFS_IOC_SNAP_CREATE_V2:
5592                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5593         case BTRFS_IOC_SUBVOL_CREATE:
5594                 return btrfs_ioctl_snap_create(file, argp, 1);
5595         case BTRFS_IOC_SUBVOL_CREATE_V2:
5596                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5597         case BTRFS_IOC_SNAP_DESTROY:
5598                 return btrfs_ioctl_snap_destroy(file, argp);
5599         case BTRFS_IOC_SUBVOL_GETFLAGS:
5600                 return btrfs_ioctl_subvol_getflags(file, argp);
5601         case BTRFS_IOC_SUBVOL_SETFLAGS:
5602                 return btrfs_ioctl_subvol_setflags(file, argp);
5603         case BTRFS_IOC_DEFAULT_SUBVOL:
5604                 return btrfs_ioctl_default_subvol(file, argp);
5605         case BTRFS_IOC_DEFRAG:
5606                 return btrfs_ioctl_defrag(file, NULL);
5607         case BTRFS_IOC_DEFRAG_RANGE:
5608                 return btrfs_ioctl_defrag(file, argp);
5609         case BTRFS_IOC_RESIZE:
5610                 return btrfs_ioctl_resize(file, argp);
5611         case BTRFS_IOC_ADD_DEV:
5612                 return btrfs_ioctl_add_dev(root, argp);
5613         case BTRFS_IOC_RM_DEV:
5614                 return btrfs_ioctl_rm_dev(file, argp);
5615         case BTRFS_IOC_RM_DEV_V2:
5616                 return btrfs_ioctl_rm_dev_v2(file, argp);
5617         case BTRFS_IOC_FS_INFO:
5618                 return btrfs_ioctl_fs_info(root, argp);
5619         case BTRFS_IOC_DEV_INFO:
5620                 return btrfs_ioctl_dev_info(root, argp);
5621         case BTRFS_IOC_BALANCE:
5622                 return btrfs_ioctl_balance(file, NULL);
5623         case BTRFS_IOC_TRANS_START:
5624                 return btrfs_ioctl_trans_start(file);
5625         case BTRFS_IOC_TRANS_END:
5626                 return btrfs_ioctl_trans_end(file);
5627         case BTRFS_IOC_TREE_SEARCH:
5628                 return btrfs_ioctl_tree_search(file, argp);
5629         case BTRFS_IOC_TREE_SEARCH_V2:
5630                 return btrfs_ioctl_tree_search_v2(file, argp);
5631         case BTRFS_IOC_INO_LOOKUP:
5632                 return btrfs_ioctl_ino_lookup(file, argp);
5633         case BTRFS_IOC_INO_PATHS:
5634                 return btrfs_ioctl_ino_to_path(root, argp);
5635         case BTRFS_IOC_LOGICAL_INO:
5636                 return btrfs_ioctl_logical_to_ino(root, argp);
5637         case BTRFS_IOC_SPACE_INFO:
5638                 return btrfs_ioctl_space_info(root, argp);
5639         case BTRFS_IOC_SYNC: {
5640                 int ret;
5641
5642                 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
5643                 if (ret)
5644                         return ret;
5645                 ret = btrfs_sync_fs(file_inode(file)->i_sb, 1);
5646                 /*
5647                  * The transaction thread may want to do more work,
5648                  * namely it pokes the cleaner kthread that will start
5649                  * processing uncleaned subvols.
5650                  */
5651                 wake_up_process(root->fs_info->transaction_kthread);
5652                 return ret;
5653         }
5654         case BTRFS_IOC_START_SYNC:
5655                 return btrfs_ioctl_start_sync(root, argp);
5656         case BTRFS_IOC_WAIT_SYNC:
5657                 return btrfs_ioctl_wait_sync(root, argp);
5658         case BTRFS_IOC_SCRUB:
5659                 return btrfs_ioctl_scrub(file, argp);
5660         case BTRFS_IOC_SCRUB_CANCEL:
5661                 return btrfs_ioctl_scrub_cancel(root, argp);
5662         case BTRFS_IOC_SCRUB_PROGRESS:
5663                 return btrfs_ioctl_scrub_progress(root, argp);
5664         case BTRFS_IOC_BALANCE_V2:
5665                 return btrfs_ioctl_balance(file, argp);
5666         case BTRFS_IOC_BALANCE_CTL:
5667                 return btrfs_ioctl_balance_ctl(root, arg);
5668         case BTRFS_IOC_BALANCE_PROGRESS:
5669                 return btrfs_ioctl_balance_progress(root, argp);
5670         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5671                 return btrfs_ioctl_set_received_subvol(file, argp);
5672 #ifdef CONFIG_64BIT
5673         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5674                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5675 #endif
5676         case BTRFS_IOC_SEND:
5677                 return btrfs_ioctl_send(file, argp);
5678         case BTRFS_IOC_GET_DEV_STATS:
5679                 return btrfs_ioctl_get_dev_stats(root, argp);
5680         case BTRFS_IOC_QUOTA_CTL:
5681                 return btrfs_ioctl_quota_ctl(file, argp);
5682         case BTRFS_IOC_QGROUP_ASSIGN:
5683                 return btrfs_ioctl_qgroup_assign(file, argp);
5684         case BTRFS_IOC_QGROUP_CREATE:
5685                 return btrfs_ioctl_qgroup_create(file, argp);
5686         case BTRFS_IOC_QGROUP_LIMIT:
5687                 return btrfs_ioctl_qgroup_limit(file, argp);
5688         case BTRFS_IOC_QUOTA_RESCAN:
5689                 return btrfs_ioctl_quota_rescan(file, argp);
5690         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5691                 return btrfs_ioctl_quota_rescan_status(file, argp);
5692         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5693                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5694         case BTRFS_IOC_DEV_REPLACE:
5695                 return btrfs_ioctl_dev_replace(root, argp);
5696         case BTRFS_IOC_GET_FSLABEL:
5697                 return btrfs_ioctl_get_fslabel(file, argp);
5698         case BTRFS_IOC_SET_FSLABEL:
5699                 return btrfs_ioctl_set_fslabel(file, argp);
5700         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5701                 return btrfs_ioctl_get_supported_features(argp);
5702         case BTRFS_IOC_GET_FEATURES:
5703                 return btrfs_ioctl_get_features(file, argp);
5704         case BTRFS_IOC_SET_FEATURES:
5705                 return btrfs_ioctl_set_features(file, argp);
5706         }
5707
5708         return -ENOTTY;
5709 }
5710
5711 #ifdef CONFIG_COMPAT
5712 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5713 {
5714         /*
5715          * These all access 32-bit values anyway so no further
5716          * handling is necessary.
5717          */
5718         switch (cmd) {
5719         case FS_IOC32_GETFLAGS:
5720                 cmd = FS_IOC_GETFLAGS;
5721                 break;
5722         case FS_IOC32_SETFLAGS:
5723                 cmd = FS_IOC_SETFLAGS;
5724                 break;
5725         case FS_IOC32_GETVERSION:
5726                 cmd = FS_IOC_GETVERSION;
5727                 break;
5728         }
5729
5730         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5731 }
5732 #endif