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