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/mm.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 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 160 {
 161         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 162         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 163
 164         if (copy_to_user(arg, &flags, sizeof(flags)))
 165                 return -EFAULT;
 166         return 0;
 167 }
 168
 169 static int check_flags(unsigned int flags)
 170 {
 171         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 172                       FS_NOATIME_FL | FS_NODUMP_FL | \
 173                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 174                       FS_NOCOMP_FL | FS_COMPR_FL |
 175                       FS_NOCOW_FL))
 176                 return -EOPNOTSUPP;
 177
 178         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 179                 return -EINVAL;
 180
 181         return 0;
 182 }
 183
 184 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 185 {
 186         struct inode *inode = file_inode(file);
 187         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 188         struct btrfs_inode *ip = BTRFS_I(inode);
 189         struct btrfs_root *root = ip->root;
 190         struct btrfs_trans_handle *trans;
 191         unsigned int flags, oldflags;
 192         int ret;
 193         u64 ip_oldflags;
 194         unsigned int i_oldflags;
 195         umode_t mode;
 196
 197         if (!inode_owner_or_capable(inode))
 198                 return -EPERM;
 199
 200         if (btrfs_root_readonly(root))
 201                 return -EROFS;
 202
 203         if (copy_from_user(&flags, arg, sizeof(flags)))
 204                 return -EFAULT;
 205
 206         ret = check_flags(flags);
 207         if (ret)
 208                 return ret;
 209
 210         ret = mnt_want_write_file(file);
 211         if (ret)
 212                 return ret;
 213
 214         inode_lock(inode);
 215
 216         ip_oldflags = ip->flags;
 217         i_oldflags = inode->i_flags;
 218         mode = inode->i_mode;
 219
 220         flags = btrfs_mask_flags(inode->i_mode, flags);
 221         oldflags = btrfs_flags_to_ioctl(ip->flags);
 222         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 223                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 224                         ret = -EPERM;
 225                         goto out_unlock;
 226                 }
 227         }
 228
 229         if (flags & FS_SYNC_FL)
 230                 ip->flags |= BTRFS_INODE_SYNC;
 231         else
 232                 ip->flags &= ~BTRFS_INODE_SYNC;
 233         if (flags & FS_IMMUTABLE_FL)
 234                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 235         else
 236                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 237         if (flags & FS_APPEND_FL)
 238                 ip->flags |= BTRFS_INODE_APPEND;
 239         else
 240                 ip->flags &= ~BTRFS_INODE_APPEND;
 241         if (flags & FS_NODUMP_FL)
 242                 ip->flags |= BTRFS_INODE_NODUMP;
 243         else
 244                 ip->flags &= ~BTRFS_INODE_NODUMP;
 245         if (flags & FS_NOATIME_FL)
 246                 ip->flags |= BTRFS_INODE_NOATIME;
 247         else
 248                 ip->flags &= ~BTRFS_INODE_NOATIME;
 249         if (flags & FS_DIRSYNC_FL)
 250                 ip->flags |= BTRFS_INODE_DIRSYNC;
 251         else
 252                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 253         if (flags & FS_NOCOW_FL) {
 254                 if (S_ISREG(mode)) {
 255                         /*
 256                          * It's safe to turn csums off here, no extents exist.
 257                          * Otherwise we want the flag to reflect the real COW
 258                          * status of the file and will not set it.
 259                          */
 260                         if (inode->i_size == 0)
 261                                 ip->flags |= BTRFS_INODE_NODATACOW
 262                                            | BTRFS_INODE_NODATASUM;
 263                 } else {
 264                         ip->flags |= BTRFS_INODE_NODATACOW;
 265                 }
 266         } else {
 267                 /*
 268                  * Revert back under same assumptions as above
 269                  */
 270                 if (S_ISREG(mode)) {
 271                         if (inode->i_size == 0)
 272                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 273                                              | BTRFS_INODE_NODATASUM);
 274                 } else {
 275                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 276                 }
 277         }
 278
 279         /*
 280          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 281          * flag may be changed automatically if compression code won't make
 282          * things smaller.
 283          */
 284         if (flags & FS_NOCOMP_FL) {
 285                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 286                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 287
 288                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 289                 if (ret && ret != -ENODATA)
 290                         goto out_drop;
 291         } else if (flags & FS_COMPR_FL) {
 292                 const char *comp;
 293
 294                 ip->flags |= BTRFS_INODE_COMPRESS;
 295                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 296
 297                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
 298                         comp = "lzo";
 299                 else if (fs_info->compress_type == BTRFS_COMPRESS_ZLIB)
 300                         comp = "zlib";
 301                 else
 302                         comp = "zstd";
 303                 ret = btrfs_set_prop(inode, "btrfs.compression",
 304                                      comp, strlen(comp), 0);
 305                 if (ret)
 306                         goto out_drop;
 307
 308         } else {
 309                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 310                 if (ret && ret != -ENODATA)
 311                         goto out_drop;
 312                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 313         }
 314
 315         trans = btrfs_start_transaction(root, 1);
 316         if (IS_ERR(trans)) {
 317                 ret = PTR_ERR(trans);
 318                 goto out_drop;
 319         }
 320
 321         btrfs_update_iflags(inode);
 322         inode_inc_iversion(inode);
 323         inode->i_ctime = current_time(inode);
 324         ret = btrfs_update_inode(trans, root, inode);
 325
 326         btrfs_end_transaction(trans);
 327  out_drop:
 328         if (ret) {
 329                 ip->flags = ip_oldflags;
 330                 inode->i_flags = i_oldflags;
 331         }
 332
 333  out_unlock:
 334         inode_unlock(inode);
 335         mnt_drop_write_file(file);
 336         return ret;
 337 }
 338
 339 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 340 {
 341         struct inode *inode = file_inode(file);
 342
 343         return put_user(inode->i_generation, arg);
 344 }
 345
 346 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 347 {
 348         struct inode *inode = file_inode(file);
 349         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 350         struct btrfs_device *device;
 351         struct request_queue *q;
 352         struct fstrim_range range;
 353         u64 minlen = ULLONG_MAX;
 354         u64 num_devices = 0;
 355         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 356         int ret;
 357
 358         if (!capable(CAP_SYS_ADMIN))
 359                 return -EPERM;
 360
 361         rcu_read_lock();
 362         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 363                                 dev_list) {
 364                 if (!device->bdev)
 365                         continue;
 366                 q = bdev_get_queue(device->bdev);
 367                 if (blk_queue_discard(q)) {
 368                         num_devices++;
 369                         minlen = min_t(u64, q->limits.discard_granularity,
 370                                      minlen);
 371                 }
 372         }
 373         rcu_read_unlock();
 374
 375         if (!num_devices)
 376                 return -EOPNOTSUPP;
 377         if (copy_from_user(&range, arg, sizeof(range)))
 378                 return -EFAULT;
 379         if (range.start > total_bytes ||
 380             range.len < fs_info->sb->s_blocksize)
 381                 return -EINVAL;
 382
 383         range.len = min(range.len, total_bytes - range.start);
 384         range.minlen = max(range.minlen, minlen);
 385         ret = btrfs_trim_fs(fs_info, &range);
 386         if (ret < 0)
 387                 return ret;
 388
 389         if (copy_to_user(arg, &range, sizeof(range)))
 390                 return -EFAULT;
 391
 392         return 0;
 393 }
 394
 395 int btrfs_is_empty_uuid(u8 *uuid)
 396 {
 397         int i;
 398
 399         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 400                 if (uuid[i])
 401                         return 0;
 402         }
 403         return 1;
 404 }
 405
 406 static noinline int create_subvol(struct inode *dir,
 407                                   struct dentry *dentry,
 408                                   const char *name, int namelen,
 409                                   u64 *async_transid,
 410                                   struct btrfs_qgroup_inherit *inherit)
 411 {
 412         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 413         struct btrfs_trans_handle *trans;
 414         struct btrfs_key key;
 415         struct btrfs_root_item *root_item;
 416         struct btrfs_inode_item *inode_item;
 417         struct extent_buffer *leaf;
 418         struct btrfs_root *root = BTRFS_I(dir)->root;
 419         struct btrfs_root *new_root;
 420         struct btrfs_block_rsv block_rsv;
 421         struct timespec cur_time = current_time(dir);
 422         struct inode *inode;
 423         int ret;
 424         int err;
 425         u64 objectid;
 426         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 427         u64 index = 0;
 428         u64 qgroup_reserved;
 429         uuid_le new_uuid;
 430
 431         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 432         if (!root_item)
 433                 return -ENOMEM;
 434
 435         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 436         if (ret)
 437                 goto fail_free;
 438
 439         /*
 440          * Don't create subvolume whose level is not zero. Or qgroup will be
 441          * screwed up since it assumes subvolume qgroup's level to be 0.
 442          */
 443         if (btrfs_qgroup_level(objectid)) {
 444                 ret = -ENOSPC;
 445                 goto fail_free;
 446         }
 447
 448         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 449         /*
 450          * The same as the snapshot creation, please see the comment
 451          * of create_snapshot().
 452          */
 453         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 454                                                8, &qgroup_reserved, false);
 455         if (ret)
 456                 goto fail_free;
 457
 458         trans = btrfs_start_transaction(root, 0);
 459         if (IS_ERR(trans)) {
 460                 ret = PTR_ERR(trans);
 461                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 462                 goto fail_free;
 463         }
 464         trans->block_rsv = &block_rsv;
 465         trans->bytes_reserved = block_rsv.size;
 466
 467         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 468         if (ret)
 469                 goto fail;
 470
 471         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 472         if (IS_ERR(leaf)) {
 473                 ret = PTR_ERR(leaf);
 474                 goto fail;
 475         }
 476
 477         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 478         btrfs_set_header_bytenr(leaf, leaf->start);
 479         btrfs_set_header_generation(leaf, trans->transid);
 480         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 481         btrfs_set_header_owner(leaf, objectid);
 482
 483         write_extent_buffer_fsid(leaf, fs_info->fsid);
 484         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 485         btrfs_mark_buffer_dirty(leaf);
 486
 487         inode_item = &root_item->inode;
 488         btrfs_set_stack_inode_generation(inode_item, 1);
 489         btrfs_set_stack_inode_size(inode_item, 3);
 490         btrfs_set_stack_inode_nlink(inode_item, 1);
 491         btrfs_set_stack_inode_nbytes(inode_item,
 492                                      fs_info->nodesize);
 493         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 494
 495         btrfs_set_root_flags(root_item, 0);
 496         btrfs_set_root_limit(root_item, 0);
 497         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 498
 499         btrfs_set_root_bytenr(root_item, leaf->start);
 500         btrfs_set_root_generation(root_item, trans->transid);
 501         btrfs_set_root_level(root_item, 0);
 502         btrfs_set_root_refs(root_item, 1);
 503         btrfs_set_root_used(root_item, leaf->len);
 504         btrfs_set_root_last_snapshot(root_item, 0);
 505
 506         btrfs_set_root_generation_v2(root_item,
 507                         btrfs_root_generation(root_item));
 508         uuid_le_gen(&new_uuid);
 509         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 510         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 511         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 512         root_item->ctime = root_item->otime;
 513         btrfs_set_root_ctransid(root_item, trans->transid);
 514         btrfs_set_root_otransid(root_item, trans->transid);
 515
 516         btrfs_tree_unlock(leaf);
 517         free_extent_buffer(leaf);
 518         leaf = NULL;
 519
 520         btrfs_set_root_dirid(root_item, new_dirid);
 521
 522         key.objectid = objectid;
 523         key.offset = 0;
 524         key.type = BTRFS_ROOT_ITEM_KEY;
 525         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 526                                 root_item);
 527         if (ret)
 528                 goto fail;
 529
 530         key.offset = (u64)-1;
 531         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 532         if (IS_ERR(new_root)) {
 533                 ret = PTR_ERR(new_root);
 534                 btrfs_abort_transaction(trans, ret);
 535                 goto fail;
 536         }
 537
 538         btrfs_record_root_in_trans(trans, new_root);
 539
 540         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 541         if (ret) {
 542                 /* We potentially lose an unused inode item here */
 543                 btrfs_abort_transaction(trans, ret);
 544                 goto fail;
 545         }
 546
 547         mutex_lock(&new_root->objectid_mutex);
 548         new_root->highest_objectid = new_dirid;
 549         mutex_unlock(&new_root->objectid_mutex);
 550
 551         /*
 552          * insert the directory item
 553          */
 554         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 555         if (ret) {
 556                 btrfs_abort_transaction(trans, ret);
 557                 goto fail;
 558         }
 559
 560         ret = btrfs_insert_dir_item(trans, root,
 561                                     name, namelen, BTRFS_I(dir), &key,
 562                                     BTRFS_FT_DIR, index);
 563         if (ret) {
 564                 btrfs_abort_transaction(trans, ret);
 565                 goto fail;
 566         }
 567
 568         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 569         ret = btrfs_update_inode(trans, root, dir);
 570         BUG_ON(ret);
 571
 572         ret = btrfs_add_root_ref(trans, fs_info,
 573                                  objectid, root->root_key.objectid,
 574                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 575         BUG_ON(ret);
 576
 577         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 578                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 579         if (ret)
 580                 btrfs_abort_transaction(trans, ret);
 581
 582 fail:
 583         kfree(root_item);
 584         trans->block_rsv = NULL;
 585         trans->bytes_reserved = 0;
 586         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 587
 588         if (async_transid) {
 589                 *async_transid = trans->transid;
 590                 err = btrfs_commit_transaction_async(trans, 1);
 591                 if (err)
 592                         err = btrfs_commit_transaction(trans);
 593         } else {
 594                 err = btrfs_commit_transaction(trans);
 595         }
 596         if (err && !ret)
 597                 ret = err;
 598
 599         if (!ret) {
 600                 inode = btrfs_lookup_dentry(dir, dentry);
 601                 if (IS_ERR(inode))
 602                         return PTR_ERR(inode);
 603                 d_instantiate(dentry, inode);
 604         }
 605         return ret;
 606
 607 fail_free:
 608         kfree(root_item);
 609         return ret;
 610 }
 611
 612 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 613                            struct dentry *dentry,
 614                            u64 *async_transid, bool readonly,
 615                            struct btrfs_qgroup_inherit *inherit)
 616 {
 617         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 618         struct inode *inode;
 619         struct btrfs_pending_snapshot *pending_snapshot;
 620         struct btrfs_trans_handle *trans;
 621         int ret;
 622
 623         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 624                 return -EINVAL;
 625
 626         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 627         if (!pending_snapshot)
 628                 return -ENOMEM;
 629
 630         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 631                         GFP_KERNEL);
 632         pending_snapshot->path = btrfs_alloc_path();
 633         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 634                 ret = -ENOMEM;
 635                 goto free_pending;
 636         }
 637
 638         atomic_inc(&root->will_be_snapshotted);
 639         smp_mb__after_atomic();
 640         /* wait for no snapshot writes */
 641         wait_event(root->subv_writers->wait,
 642                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 643
 644         ret = btrfs_start_delalloc_inodes(root, 0);
 645         if (ret)
 646                 goto dec_and_free;
 647
 648         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 649
 650         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 651                              BTRFS_BLOCK_RSV_TEMP);
 652         /*
 653          * 1 - parent dir inode
 654          * 2 - dir entries
 655          * 1 - root item
 656          * 2 - root ref/backref
 657          * 1 - root of snapshot
 658          * 1 - UUID item
 659          */
 660         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 661                                         &pending_snapshot->block_rsv, 8,
 662                                         &pending_snapshot->qgroup_reserved,
 663                                         false);
 664         if (ret)
 665                 goto dec_and_free;
 666
 667         pending_snapshot->dentry = dentry;
 668         pending_snapshot->root = root;
 669         pending_snapshot->readonly = readonly;
 670         pending_snapshot->dir = dir;
 671         pending_snapshot->inherit = inherit;
 672
 673         trans = btrfs_start_transaction(root, 0);
 674         if (IS_ERR(trans)) {
 675                 ret = PTR_ERR(trans);
 676                 goto fail;
 677         }
 678
 679         spin_lock(&fs_info->trans_lock);
 680         list_add(&pending_snapshot->list,
 681                  &trans->transaction->pending_snapshots);
 682         spin_unlock(&fs_info->trans_lock);
 683         if (async_transid) {
 684                 *async_transid = trans->transid;
 685                 ret = btrfs_commit_transaction_async(trans, 1);
 686                 if (ret)
 687                         ret = btrfs_commit_transaction(trans);
 688         } else {
 689                 ret = btrfs_commit_transaction(trans);
 690         }
 691         if (ret)
 692                 goto fail;
 693
 694         ret = pending_snapshot->error;
 695         if (ret)
 696                 goto fail;
 697
 698         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 699         if (ret)
 700                 goto fail;
 701
 702         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 703         if (IS_ERR(inode)) {
 704                 ret = PTR_ERR(inode);
 705                 goto fail;
 706         }
 707
 708         d_instantiate(dentry, inode);
 709         ret = 0;
 710 fail:
 711         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 712 dec_and_free:
 713         if (atomic_dec_and_test(&root->will_be_snapshotted))
 714                 wake_up_atomic_t(&root->will_be_snapshotted);
 715 free_pending:
 716         kfree(pending_snapshot->root_item);
 717         btrfs_free_path(pending_snapshot->path);
 718         kfree(pending_snapshot);
 719
 720         return ret;
 721 }
 722
 723 /*  copy of may_delete in fs/namei.c()
 724  *      Check whether we can remove a link victim from directory dir, check
 725  *  whether the type of victim is right.
 726  *  1. We can't do it if dir is read-only (done in permission())
 727  *  2. We should have write and exec permissions on dir
 728  *  3. We can't remove anything from append-only dir
 729  *  4. We can't do anything with immutable dir (done in permission())
 730  *  5. If the sticky bit on dir is set we should either
 731  *      a. be owner of dir, or
 732  *      b. be owner of victim, or
 733  *      c. have CAP_FOWNER capability
 734  *  6. If the victim is append-only or immutable we can't do anything with
 735  *     links pointing to it.
 736  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 737  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 738  *  9. We can't remove a root or mountpoint.
 739  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 740  *     nfs_async_unlink().
 741  */
 742
 743 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 744 {
 745         int error;
 746
 747         if (d_really_is_negative(victim))
 748                 return -ENOENT;
 749
 750         BUG_ON(d_inode(victim->d_parent) != dir);
 751         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 752
 753         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 754         if (error)
 755                 return error;
 756         if (IS_APPEND(dir))
 757                 return -EPERM;
 758         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 759             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 760                 return -EPERM;
 761         if (isdir) {
 762                 if (!d_is_dir(victim))
 763                         return -ENOTDIR;
 764                 if (IS_ROOT(victim))
 765                         return -EBUSY;
 766         } else if (d_is_dir(victim))
 767                 return -EISDIR;
 768         if (IS_DEADDIR(dir))
 769                 return -ENOENT;
 770         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 771                 return -EBUSY;
 772         return 0;
 773 }
 774
 775 /* copy of may_create in fs/namei.c() */
 776 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 777 {
 778         if (d_really_is_positive(child))
 779                 return -EEXIST;
 780         if (IS_DEADDIR(dir))
 781                 return -ENOENT;
 782         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 783 }
 784
 785 /*
 786  * Create a new subvolume below @parent.  This is largely modeled after
 787  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 788  * inside this filesystem so it's quite a bit simpler.
 789  */
 790 static noinline int btrfs_mksubvol(const struct path *parent,
 791                                    const char *name, int namelen,
 792                                    struct btrfs_root *snap_src,
 793                                    u64 *async_transid, bool readonly,
 794                                    struct btrfs_qgroup_inherit *inherit)
 795 {
 796         struct inode *dir = d_inode(parent->dentry);
 797         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 798         struct dentry *dentry;
 799         int error;
 800
 801         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 802         if (error == -EINTR)
 803                 return error;
 804
 805         dentry = lookup_one_len(name, parent->dentry, namelen);
 806         error = PTR_ERR(dentry);
 807         if (IS_ERR(dentry))
 808                 goto out_unlock;
 809
 810         error = btrfs_may_create(dir, dentry);
 811         if (error)
 812                 goto out_dput;
 813
 814         /*
 815          * even if this name doesn't exist, we may get hash collisions.
 816          * check for them now when we can safely fail
 817          */
 818         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 819                                                dir->i_ino, name,
 820                                                namelen);
 821         if (error)
 822                 goto out_dput;
 823
 824         down_read(&fs_info->subvol_sem);
 825
 826         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 827                 goto out_up_read;
 828
 829         if (snap_src) {
 830                 error = create_snapshot(snap_src, dir, dentry,
 831                                         async_transid, readonly, inherit);
 832         } else {
 833                 error = create_subvol(dir, dentry, name, namelen,
 834                                       async_transid, inherit);
 835         }
 836         if (!error)
 837                 fsnotify_mkdir(dir, dentry);
 838 out_up_read:
 839         up_read(&fs_info->subvol_sem);
 840 out_dput:
 841         dput(dentry);
 842 out_unlock:
 843         inode_unlock(dir);
 844         return error;
 845 }
 846
 847 /*
 848  * When we're defragging a range, we don't want to kick it off again
 849  * if it is really just waiting for delalloc to send it down.
 850  * If we find a nice big extent or delalloc range for the bytes in the
 851  * file you want to defrag, we return 0 to let you know to skip this
 852  * part of the file
 853  */
 854 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 855 {
 856         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 857         struct extent_map *em = NULL;
 858         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 859         u64 end;
 860
 861         read_lock(&em_tree->lock);
 862         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 863         read_unlock(&em_tree->lock);
 864
 865         if (em) {
 866                 end = extent_map_end(em);
 867                 free_extent_map(em);
 868                 if (end - offset > thresh)
 869                         return 0;
 870         }
 871         /* if we already have a nice delalloc here, just stop */
 872         thresh /= 2;
 873         end = count_range_bits(io_tree, &offset, offset + thresh,
 874                                thresh, EXTENT_DELALLOC, 1);
 875         if (end >= thresh)
 876                 return 0;
 877         return 1;
 878 }
 879
 880 /*
 881  * helper function to walk through a file and find extents
 882  * newer than a specific transid, and smaller than thresh.
 883  *
 884  * This is used by the defragging code to find new and small
 885  * extents
 886  */
 887 static int find_new_extents(struct btrfs_root *root,
 888                             struct inode *inode, u64 newer_than,
 889                             u64 *off, u32 thresh)
 890 {
 891         struct btrfs_path *path;
 892         struct btrfs_key min_key;
 893         struct extent_buffer *leaf;
 894         struct btrfs_file_extent_item *extent;
 895         int type;
 896         int ret;
 897         u64 ino = btrfs_ino(BTRFS_I(inode));
 898
 899         path = btrfs_alloc_path();
 900         if (!path)
 901                 return -ENOMEM;
 902
 903         min_key.objectid = ino;
 904         min_key.type = BTRFS_EXTENT_DATA_KEY;
 905         min_key.offset = *off;
 906
 907         while (1) {
 908                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 909                 if (ret != 0)
 910                         goto none;
 911 process_slot:
 912                 if (min_key.objectid != ino)
 913                         goto none;
 914                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 915                         goto none;
 916
 917                 leaf = path->nodes[0];
 918                 extent = btrfs_item_ptr(leaf, path->slots[0],
 919                                         struct btrfs_file_extent_item);
 920
 921                 type = btrfs_file_extent_type(leaf, extent);
 922                 if (type == BTRFS_FILE_EXTENT_REG &&
 923                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 924                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 925                         *off = min_key.offset;
 926                         btrfs_free_path(path);
 927                         return 0;
 928                 }
 929
 930                 path->slots[0]++;
 931                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 932                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 933                         goto process_slot;
 934                 }
 935
 936                 if (min_key.offset == (u64)-1)
 937                         goto none;
 938
 939                 min_key.offset++;
 940                 btrfs_release_path(path);
 941         }
 942 none:
 943         btrfs_free_path(path);
 944         return -ENOENT;
 945 }
 946
 947 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 948 {
 949         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 950         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 951         struct extent_map *em;
 952         u64 len = PAGE_SIZE;
 953
 954         /*
 955          * hopefully we have this extent in the tree already, try without
 956          * the full extent lock
 957          */
 958         read_lock(&em_tree->lock);
 959         em = lookup_extent_mapping(em_tree, start, len);
 960         read_unlock(&em_tree->lock);
 961
 962         if (!em) {
 963                 struct extent_state *cached = NULL;
 964                 u64 end = start + len - 1;
 965
 966                 /* get the big lock and read metadata off disk */
 967                 lock_extent_bits(io_tree, start, end, &cached);
 968                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 969                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 970
 971                 if (IS_ERR(em))
 972                         return NULL;
 973         }
 974
 975         return em;
 976 }
 977
 978 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 979 {
 980         struct extent_map *next;
 981         bool ret = true;
 982
 983         /* this is the last extent */
 984         if (em->start + em->len >= i_size_read(inode))
 985                 return false;
 986
 987         next = defrag_lookup_extent(inode, em->start + em->len);
 988         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 989                 ret = false;
 990         else if ((em->block_start + em->block_len == next->block_start) &&
 991                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
 992                 ret = false;
 993
 994         free_extent_map(next);
 995         return ret;
 996 }
 997
 998 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 999                                u64 *last_len, u64 *skip, u64 *defrag_end,
1000                                int compress)
1001 {
1002         struct extent_map *em;
1003         int ret = 1;
1004         bool next_mergeable = true;
1005         bool prev_mergeable = true;
1006
1007         /*
1008          * make sure that once we start defragging an extent, we keep on
1009          * defragging it
1010          */
1011         if (start < *defrag_end)
1012                 return 1;
1013
1014         *skip = 0;
1015
1016         em = defrag_lookup_extent(inode, start);
1017         if (!em)
1018                 return 0;
1019
1020         /* this will cover holes, and inline extents */
1021         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1022                 ret = 0;
1023                 goto out;
1024         }
1025
1026         if (!*defrag_end)
1027                 prev_mergeable = false;
1028
1029         next_mergeable = defrag_check_next_extent(inode, em);
1030         /*
1031          * we hit a real extent, if it is big or the next extent is not a
1032          * real extent, don't bother defragging it
1033          */
1034         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1035             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1036                 ret = 0;
1037 out:
1038         /*
1039          * last_len ends up being a counter of how many bytes we've defragged.
1040          * every time we choose not to defrag an extent, we reset *last_len
1041          * so that the next tiny extent will force a defrag.
1042          *
1043          * The end result of this is that tiny extents before a single big
1044          * extent will force at least part of that big extent to be defragged.
1045          */
1046         if (ret) {
1047                 *defrag_end = extent_map_end(em);
1048         } else {
1049                 *last_len = 0;
1050                 *skip = extent_map_end(em);
1051                 *defrag_end = 0;
1052         }
1053
1054         free_extent_map(em);
1055         return ret;
1056 }
1057
1058 /*
1059  * it doesn't do much good to defrag one or two pages
1060  * at a time.  This pulls in a nice chunk of pages
1061  * to COW and defrag.
1062  *
1063  * It also makes sure the delalloc code has enough
1064  * dirty data to avoid making new small extents as part
1065  * of the defrag
1066  *
1067  * It's a good idea to start RA on this range
1068  * before calling this.
1069  */
1070 static int cluster_pages_for_defrag(struct inode *inode,
1071                                     struct page **pages,
1072                                     unsigned long start_index,
1073                                     unsigned long num_pages)
1074 {
1075         unsigned long file_end;
1076         u64 isize = i_size_read(inode);
1077         u64 page_start;
1078         u64 page_end;
1079         u64 page_cnt;
1080         int ret;
1081         int i;
1082         int i_done;
1083         struct btrfs_ordered_extent *ordered;
1084         struct extent_state *cached_state = NULL;
1085         struct extent_io_tree *tree;
1086         struct extent_changeset *data_reserved = NULL;
1087         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1088
1089         file_end = (isize - 1) >> PAGE_SHIFT;
1090         if (!isize || start_index > file_end)
1091                 return 0;
1092
1093         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1094
1095         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1096                         start_index << PAGE_SHIFT,
1097                         page_cnt << PAGE_SHIFT);
1098         if (ret)
1099                 return ret;
1100         i_done = 0;
1101         tree = &BTRFS_I(inode)->io_tree;
1102
1103         /* step one, lock all the pages */
1104         for (i = 0; i < page_cnt; i++) {
1105                 struct page *page;
1106 again:
1107                 page = find_or_create_page(inode->i_mapping,
1108                                            start_index + i, mask);
1109                 if (!page)
1110                         break;
1111
1112                 page_start = page_offset(page);
1113                 page_end = page_start + PAGE_SIZE - 1;
1114                 while (1) {
1115                         lock_extent_bits(tree, page_start, page_end,
1116                                          &cached_state);
1117                         ordered = btrfs_lookup_ordered_extent(inode,
1118                                                               page_start);
1119                         unlock_extent_cached(tree, page_start, page_end,
1120                                              &cached_state, GFP_NOFS);
1121                         if (!ordered)
1122                                 break;
1123
1124                         unlock_page(page);
1125                         btrfs_start_ordered_extent(inode, ordered, 1);
1126                         btrfs_put_ordered_extent(ordered);
1127                         lock_page(page);
1128                         /*
1129                          * we unlocked the page above, so we need check if
1130                          * it was released or not.
1131                          */
1132                         if (page->mapping != inode->i_mapping) {
1133                                 unlock_page(page);
1134                                 put_page(page);
1135                                 goto again;
1136                         }
1137                 }
1138
1139                 if (!PageUptodate(page)) {
1140                         btrfs_readpage(NULL, page);
1141                         lock_page(page);
1142                         if (!PageUptodate(page)) {
1143                                 unlock_page(page);
1144                                 put_page(page);
1145                                 ret = -EIO;
1146                                 break;
1147                         }
1148                 }
1149
1150                 if (page->mapping != inode->i_mapping) {
1151                         unlock_page(page);
1152                         put_page(page);
1153                         goto again;
1154                 }
1155
1156                 pages[i] = page;
1157                 i_done++;
1158         }
1159         if (!i_done || ret)
1160                 goto out;
1161
1162         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1163                 goto out;
1164
1165         /*
1166          * so now we have a nice long stream of locked
1167          * and up to date pages, lets wait on them
1168          */
1169         for (i = 0; i < i_done; i++)
1170                 wait_on_page_writeback(pages[i]);
1171
1172         page_start = page_offset(pages[0]);
1173         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1174
1175         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1176                          page_start, page_end - 1, &cached_state);
1177         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1178                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1179                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1180                           &cached_state, GFP_NOFS);
1181
1182         if (i_done != page_cnt) {
1183                 spin_lock(&BTRFS_I(inode)->lock);
1184                 BTRFS_I(inode)->outstanding_extents++;
1185                 spin_unlock(&BTRFS_I(inode)->lock);
1186                 btrfs_delalloc_release_space(inode, data_reserved,
1187                                 start_index << PAGE_SHIFT,
1188                                 (page_cnt - i_done) << PAGE_SHIFT);
1189         }
1190
1191
1192         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1193                           &cached_state);
1194
1195         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1196                              page_start, page_end - 1, &cached_state,
1197                              GFP_NOFS);
1198
1199         for (i = 0; i < i_done; i++) {
1200                 clear_page_dirty_for_io(pages[i]);
1201                 ClearPageChecked(pages[i]);
1202                 set_page_extent_mapped(pages[i]);
1203                 set_page_dirty(pages[i]);
1204                 unlock_page(pages[i]);
1205                 put_page(pages[i]);
1206         }
1207         extent_changeset_free(data_reserved);
1208         return i_done;
1209 out:
1210         for (i = 0; i < i_done; i++) {
1211                 unlock_page(pages[i]);
1212                 put_page(pages[i]);
1213         }
1214         btrfs_delalloc_release_space(inode, data_reserved,
1215                         start_index << PAGE_SHIFT,
1216                         page_cnt << PAGE_SHIFT);
1217         extent_changeset_free(data_reserved);
1218         return ret;
1219
1220 }
1221
1222 int btrfs_defrag_file(struct inode *inode, struct file *file,
1223                       struct btrfs_ioctl_defrag_range_args *range,
1224                       u64 newer_than, unsigned long max_to_defrag)
1225 {
1226         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1227         struct btrfs_root *root = BTRFS_I(inode)->root;
1228         struct file_ra_state *ra = NULL;
1229         unsigned long last_index;
1230         u64 isize = i_size_read(inode);
1231         u64 last_len = 0;
1232         u64 skip = 0;
1233         u64 defrag_end = 0;
1234         u64 newer_off = range->start;
1235         unsigned long i;
1236         unsigned long ra_index = 0;
1237         int ret;
1238         int defrag_count = 0;
1239         int compress_type = BTRFS_COMPRESS_ZLIB;
1240         u32 extent_thresh = range->extent_thresh;
1241         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1242         unsigned long cluster = max_cluster;
1243         u64 new_align = ~((u64)SZ_128K - 1);
1244         struct page **pages = NULL;
1245         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1246
1247         if (isize == 0)
1248                 return 0;
1249
1250         if (range->start >= isize)
1251                 return -EINVAL;
1252
1253         if (do_compress) {
1254                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1255                         return -EINVAL;
1256                 if (range->compress_type)
1257                         compress_type = range->compress_type;
1258         }
1259
1260         if (extent_thresh == 0)
1261                 extent_thresh = SZ_256K;
1262
1263         /*
1264          * If we were not given a file, allocate a readahead context. As
1265          * readahead is just an optimization, defrag will work without it so
1266          * we don't error out.
1267          */
1268         if (!file) {
1269                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1270                 if (ra)
1271                         file_ra_state_init(ra, inode->i_mapping);
1272         } else {
1273                 ra = &file->f_ra;
1274         }
1275
1276         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1277         if (!pages) {
1278                 ret = -ENOMEM;
1279                 goto out_ra;
1280         }
1281
1282         /* find the last page to defrag */
1283         if (range->start + range->len > range->start) {
1284                 last_index = min_t(u64, isize - 1,
1285                          range->start + range->len - 1) >> PAGE_SHIFT;
1286         } else {
1287                 last_index = (isize - 1) >> PAGE_SHIFT;
1288         }
1289
1290         if (newer_than) {
1291                 ret = find_new_extents(root, inode, newer_than,
1292                                        &newer_off, SZ_64K);
1293                 if (!ret) {
1294                         range->start = newer_off;
1295                         /*
1296                          * we always align our defrag to help keep
1297                          * the extents in the file evenly spaced
1298                          */
1299                         i = (newer_off & new_align) >> PAGE_SHIFT;
1300                 } else
1301                         goto out_ra;
1302         } else {
1303                 i = range->start >> PAGE_SHIFT;
1304         }
1305         if (!max_to_defrag)
1306                 max_to_defrag = last_index - i + 1;
1307
1308         /*
1309          * make writeback starts from i, so the defrag range can be
1310          * written sequentially.
1311          */
1312         if (i < inode->i_mapping->writeback_index)
1313                 inode->i_mapping->writeback_index = i;
1314
1315         while (i <= last_index && defrag_count < max_to_defrag &&
1316                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1317                 /*
1318                  * make sure we stop running if someone unmounts
1319                  * the FS
1320                  */
1321                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1322                         break;
1323
1324                 if (btrfs_defrag_cancelled(fs_info)) {
1325                         btrfs_debug(fs_info, "defrag_file cancelled");
1326                         ret = -EAGAIN;
1327                         break;
1328                 }
1329
1330                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1331                                          extent_thresh, &last_len, &skip,
1332                                          &defrag_end, do_compress)){
1333                         unsigned long next;
1334                         /*
1335                          * the should_defrag function tells us how much to skip
1336                          * bump our counter by the suggested amount
1337                          */
1338                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1339                         i = max(i + 1, next);
1340                         continue;
1341                 }
1342
1343                 if (!newer_than) {
1344                         cluster = (PAGE_ALIGN(defrag_end) >>
1345                                    PAGE_SHIFT) - i;
1346                         cluster = min(cluster, max_cluster);
1347                 } else {
1348                         cluster = max_cluster;
1349                 }
1350
1351                 if (i + cluster > ra_index) {
1352                         ra_index = max(i, ra_index);
1353                         if (ra)
1354                                 page_cache_sync_readahead(inode->i_mapping, ra,
1355                                                 file, ra_index, cluster);
1356                         ra_index += cluster;
1357                 }
1358
1359                 inode_lock(inode);
1360                 if (do_compress)
1361                         BTRFS_I(inode)->defrag_compress = compress_type;
1362                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1363                 if (ret < 0) {
1364                         inode_unlock(inode);
1365                         goto out_ra;
1366                 }
1367
1368                 defrag_count += ret;
1369                 balance_dirty_pages_ratelimited(inode->i_mapping);
1370                 inode_unlock(inode);
1371
1372                 if (newer_than) {
1373                         if (newer_off == (u64)-1)
1374                                 break;
1375
1376                         if (ret > 0)
1377                                 i += ret;
1378
1379                         newer_off = max(newer_off + 1,
1380                                         (u64)i << PAGE_SHIFT);
1381
1382                         ret = find_new_extents(root, inode, newer_than,
1383                                                &newer_off, SZ_64K);
1384                         if (!ret) {
1385                                 range->start = newer_off;
1386                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1387                         } else {
1388                                 break;
1389                         }
1390                 } else {
1391                         if (ret > 0) {
1392                                 i += ret;
1393                                 last_len += ret << PAGE_SHIFT;
1394                         } else {
1395                                 i++;
1396                                 last_len = 0;
1397                         }
1398                 }
1399         }
1400
1401         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1402                 filemap_flush(inode->i_mapping);
1403                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1404                              &BTRFS_I(inode)->runtime_flags))
1405                         filemap_flush(inode->i_mapping);
1406         }
1407
1408         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1409                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1410         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1411                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1412         }
1413
1414         ret = defrag_count;
1415
1416 out_ra:
1417         if (do_compress) {
1418                 inode_lock(inode);
1419                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1420                 inode_unlock(inode);
1421         }
1422         if (!file)
1423                 kfree(ra);
1424         kfree(pages);
1425         return ret;
1426 }
1427
1428 static noinline int btrfs_ioctl_resize(struct file *file,
1429                                         void __user *arg)
1430 {
1431         struct inode *inode = file_inode(file);
1432         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1433         u64 new_size;
1434         u64 old_size;
1435         u64 devid = 1;
1436         struct btrfs_root *root = BTRFS_I(inode)->root;
1437         struct btrfs_ioctl_vol_args *vol_args;
1438         struct btrfs_trans_handle *trans;
1439         struct btrfs_device *device = NULL;
1440         char *sizestr;
1441         char *retptr;
1442         char *devstr = NULL;
1443         int ret = 0;
1444         int mod = 0;
1445
1446         if (!capable(CAP_SYS_ADMIN))
1447                 return -EPERM;
1448
1449         ret = mnt_want_write_file(file);
1450         if (ret)
1451                 return ret;
1452
1453         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1454                 mnt_drop_write_file(file);
1455                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1456         }
1457
1458         mutex_lock(&fs_info->volume_mutex);
1459         vol_args = memdup_user(arg, sizeof(*vol_args));
1460         if (IS_ERR(vol_args)) {
1461                 ret = PTR_ERR(vol_args);
1462                 goto out;
1463         }
1464
1465         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1466
1467         sizestr = vol_args->name;
1468         devstr = strchr(sizestr, ':');
1469         if (devstr) {
1470                 sizestr = devstr + 1;
1471                 *devstr = '\0';
1472                 devstr = vol_args->name;
1473                 ret = kstrtoull(devstr, 10, &devid);
1474                 if (ret)
1475                         goto out_free;
1476                 if (!devid) {
1477                         ret = -EINVAL;
1478                         goto out_free;
1479                 }
1480                 btrfs_info(fs_info, "resizing devid %llu", devid);
1481         }
1482
1483         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1484         if (!device) {
1485                 btrfs_info(fs_info, "resizer unable to find device %llu",
1486                            devid);
1487                 ret = -ENODEV;
1488                 goto out_free;
1489         }
1490
1491         if (!device->writeable) {
1492                 btrfs_info(fs_info,
1493                            "resizer unable to apply on readonly device %llu",
1494                        devid);
1495                 ret = -EPERM;
1496                 goto out_free;
1497         }
1498
1499         if (!strcmp(sizestr, "max"))
1500                 new_size = device->bdev->bd_inode->i_size;
1501         else {
1502                 if (sizestr[0] == '-') {
1503                         mod = -1;
1504                         sizestr++;
1505                 } else if (sizestr[0] == '+') {
1506                         mod = 1;
1507                         sizestr++;
1508                 }
1509                 new_size = memparse(sizestr, &retptr);
1510                 if (*retptr != '\0' || new_size == 0) {
1511                         ret = -EINVAL;
1512                         goto out_free;
1513                 }
1514         }
1515
1516         if (device->is_tgtdev_for_dev_replace) {
1517                 ret = -EPERM;
1518                 goto out_free;
1519         }
1520
1521         old_size = btrfs_device_get_total_bytes(device);
1522
1523         if (mod < 0) {
1524                 if (new_size > old_size) {
1525                         ret = -EINVAL;
1526                         goto out_free;
1527                 }
1528                 new_size = old_size - new_size;
1529         } else if (mod > 0) {
1530                 if (new_size > ULLONG_MAX - old_size) {
1531                         ret = -ERANGE;
1532                         goto out_free;
1533                 }
1534                 new_size = old_size + new_size;
1535         }
1536
1537         if (new_size < SZ_256M) {
1538                 ret = -EINVAL;
1539                 goto out_free;
1540         }
1541         if (new_size > device->bdev->bd_inode->i_size) {
1542                 ret = -EFBIG;
1543                 goto out_free;
1544         }
1545
1546         new_size = round_down(new_size, fs_info->sectorsize);
1547
1548         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1549                           rcu_str_deref(device->name), new_size);
1550
1551         if (new_size > old_size) {
1552                 trans = btrfs_start_transaction(root, 0);
1553                 if (IS_ERR(trans)) {
1554                         ret = PTR_ERR(trans);
1555                         goto out_free;
1556                 }
1557                 ret = btrfs_grow_device(trans, device, new_size);
1558                 btrfs_commit_transaction(trans);
1559         } else if (new_size < old_size) {
1560                 ret = btrfs_shrink_device(device, new_size);
1561         } /* equal, nothing need to do */
1562
1563 out_free:
1564         kfree(vol_args);
1565 out:
1566         mutex_unlock(&fs_info->volume_mutex);
1567         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1568         mnt_drop_write_file(file);
1569         return ret;
1570 }
1571
1572 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1573                                 const char *name, unsigned long fd, int subvol,
1574                                 u64 *transid, bool readonly,
1575                                 struct btrfs_qgroup_inherit *inherit)
1576 {
1577         int namelen;
1578         int ret = 0;
1579
1580         if (!S_ISDIR(file_inode(file)->i_mode))
1581                 return -ENOTDIR;
1582
1583         ret = mnt_want_write_file(file);
1584         if (ret)
1585                 goto out;
1586
1587         namelen = strlen(name);
1588         if (strchr(name, '/')) {
1589                 ret = -EINVAL;
1590                 goto out_drop_write;
1591         }
1592
1593         if (name[0] == '.' &&
1594            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1595                 ret = -EEXIST;
1596                 goto out_drop_write;
1597         }
1598
1599         if (subvol) {
1600                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1601                                      NULL, transid, readonly, inherit);
1602         } else {
1603                 struct fd src = fdget(fd);
1604                 struct inode *src_inode;
1605                 if (!src.file) {
1606                         ret = -EINVAL;
1607                         goto out_drop_write;
1608                 }
1609
1610                 src_inode = file_inode(src.file);
1611                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1612                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1613                                    "Snapshot src from another FS");
1614                         ret = -EXDEV;
1615                 } else if (!inode_owner_or_capable(src_inode)) {
1616                         /*
1617                          * Subvolume creation is not restricted, but snapshots
1618                          * are limited to own subvolumes only
1619                          */
1620                         ret = -EPERM;
1621                 } else {
1622                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1623                                              BTRFS_I(src_inode)->root,
1624                                              transid, readonly, inherit);
1625                 }
1626                 fdput(src);
1627         }
1628 out_drop_write:
1629         mnt_drop_write_file(file);
1630 out:
1631         return ret;
1632 }
1633
1634 static noinline int btrfs_ioctl_snap_create(struct file *file,
1635                                             void __user *arg, int subvol)
1636 {
1637         struct btrfs_ioctl_vol_args *vol_args;
1638         int ret;
1639
1640         if (!S_ISDIR(file_inode(file)->i_mode))
1641                 return -ENOTDIR;
1642
1643         vol_args = memdup_user(arg, sizeof(*vol_args));
1644         if (IS_ERR(vol_args))
1645                 return PTR_ERR(vol_args);
1646         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1647
1648         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1649                                               vol_args->fd, subvol,
1650                                               NULL, false, NULL);
1651
1652         kfree(vol_args);
1653         return ret;
1654 }
1655
1656 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1657                                                void __user *arg, int subvol)
1658 {
1659         struct btrfs_ioctl_vol_args_v2 *vol_args;
1660         int ret;
1661         u64 transid = 0;
1662         u64 *ptr = NULL;
1663         bool readonly = false;
1664         struct btrfs_qgroup_inherit *inherit = NULL;
1665
1666         if (!S_ISDIR(file_inode(file)->i_mode))
1667                 return -ENOTDIR;
1668
1669         vol_args = memdup_user(arg, sizeof(*vol_args));
1670         if (IS_ERR(vol_args))
1671                 return PTR_ERR(vol_args);
1672         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1673
1674         if (vol_args->flags &
1675             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1676               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1677                 ret = -EOPNOTSUPP;
1678                 goto free_args;
1679         }
1680
1681         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1682                 ptr = &transid;
1683         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1684                 readonly = true;
1685         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1686                 if (vol_args->size > PAGE_SIZE) {
1687                         ret = -EINVAL;
1688                         goto free_args;
1689                 }
1690                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1691                 if (IS_ERR(inherit)) {
1692                         ret = PTR_ERR(inherit);
1693                         goto free_args;
1694                 }
1695         }
1696
1697         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1698                                               vol_args->fd, subvol, ptr,
1699                                               readonly, inherit);
1700         if (ret)
1701                 goto free_inherit;
1702
1703         if (ptr && copy_to_user(arg +
1704                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1705                                         transid),
1706                                 ptr, sizeof(*ptr)))
1707                 ret = -EFAULT;
1708
1709 free_inherit:
1710         kfree(inherit);
1711 free_args:
1712         kfree(vol_args);
1713         return ret;
1714 }
1715
1716 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1717                                                 void __user *arg)
1718 {
1719         struct inode *inode = file_inode(file);
1720         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1721         struct btrfs_root *root = BTRFS_I(inode)->root;
1722         int ret = 0;
1723         u64 flags = 0;
1724
1725         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1726                 return -EINVAL;
1727
1728         down_read(&fs_info->subvol_sem);
1729         if (btrfs_root_readonly(root))
1730                 flags |= BTRFS_SUBVOL_RDONLY;
1731         up_read(&fs_info->subvol_sem);
1732
1733         if (copy_to_user(arg, &flags, sizeof(flags)))
1734                 ret = -EFAULT;
1735
1736         return ret;
1737 }
1738
1739 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1740                                               void __user *arg)
1741 {
1742         struct inode *inode = file_inode(file);
1743         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1744         struct btrfs_root *root = BTRFS_I(inode)->root;
1745         struct btrfs_trans_handle *trans;
1746         u64 root_flags;
1747         u64 flags;
1748         int ret = 0;
1749
1750         if (!inode_owner_or_capable(inode))
1751                 return -EPERM;
1752
1753         ret = mnt_want_write_file(file);
1754         if (ret)
1755                 goto out;
1756
1757         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1758                 ret = -EINVAL;
1759                 goto out_drop_write;
1760         }
1761
1762         if (copy_from_user(&flags, arg, sizeof(flags))) {
1763                 ret = -EFAULT;
1764                 goto out_drop_write;
1765         }
1766
1767         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1768                 ret = -EINVAL;
1769                 goto out_drop_write;
1770         }
1771
1772         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1773                 ret = -EOPNOTSUPP;
1774                 goto out_drop_write;
1775         }
1776
1777         down_write(&fs_info->subvol_sem);
1778
1779         /* nothing to do */
1780         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1781                 goto out_drop_sem;
1782
1783         root_flags = btrfs_root_flags(&root->root_item);
1784         if (flags & BTRFS_SUBVOL_RDONLY) {
1785                 btrfs_set_root_flags(&root->root_item,
1786                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1787         } else {
1788                 /*
1789                  * Block RO -> RW transition if this subvolume is involved in
1790                  * send
1791                  */
1792                 spin_lock(&root->root_item_lock);
1793                 if (root->send_in_progress == 0) {
1794                         btrfs_set_root_flags(&root->root_item,
1795                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1796                         spin_unlock(&root->root_item_lock);
1797                 } else {
1798                         spin_unlock(&root->root_item_lock);
1799                         btrfs_warn(fs_info,
1800                                    "Attempt to set subvolume %llu read-write during send",
1801                                    root->root_key.objectid);
1802                         ret = -EPERM;
1803                         goto out_drop_sem;
1804                 }
1805         }
1806
1807         trans = btrfs_start_transaction(root, 1);
1808         if (IS_ERR(trans)) {
1809                 ret = PTR_ERR(trans);
1810                 goto out_reset;
1811         }
1812
1813         ret = btrfs_update_root(trans, fs_info->tree_root,
1814                                 &root->root_key, &root->root_item);
1815         if (ret < 0) {
1816                 btrfs_end_transaction(trans);
1817                 goto out_reset;
1818         }
1819
1820         ret = btrfs_commit_transaction(trans);
1821
1822 out_reset:
1823         if (ret)
1824                 btrfs_set_root_flags(&root->root_item, root_flags);
1825 out_drop_sem:
1826         up_write(&fs_info->subvol_sem);
1827 out_drop_write:
1828         mnt_drop_write_file(file);
1829 out:
1830         return ret;
1831 }
1832
1833 /*
1834  * helper to check if the subvolume references other subvolumes
1835  */
1836 static noinline int may_destroy_subvol(struct btrfs_root *root)
1837 {
1838         struct btrfs_fs_info *fs_info = root->fs_info;
1839         struct btrfs_path *path;
1840         struct btrfs_dir_item *di;
1841         struct btrfs_key key;
1842         u64 dir_id;
1843         int ret;
1844
1845         path = btrfs_alloc_path();
1846         if (!path)
1847                 return -ENOMEM;
1848
1849         /* Make sure this root isn't set as the default subvol */
1850         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1851         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1852                                    dir_id, "default", 7, 0);
1853         if (di && !IS_ERR(di)) {
1854                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1855                 if (key.objectid == root->root_key.objectid) {
1856                         ret = -EPERM;
1857                         btrfs_err(fs_info,
1858                                   "deleting default subvolume %llu is not allowed",
1859                                   key.objectid);
1860                         goto out;
1861                 }
1862                 btrfs_release_path(path);
1863         }
1864
1865         key.objectid = root->root_key.objectid;
1866         key.type = BTRFS_ROOT_REF_KEY;
1867         key.offset = (u64)-1;
1868
1869         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1870         if (ret < 0)
1871                 goto out;
1872         BUG_ON(ret == 0);
1873
1874         ret = 0;
1875         if (path->slots[0] > 0) {
1876                 path->slots[0]--;
1877                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1878                 if (key.objectid == root->root_key.objectid &&
1879                     key.type == BTRFS_ROOT_REF_KEY)
1880                         ret = -ENOTEMPTY;
1881         }
1882 out:
1883         btrfs_free_path(path);
1884         return ret;
1885 }
1886
1887 static noinline int key_in_sk(struct btrfs_key *key,
1888                               struct btrfs_ioctl_search_key *sk)
1889 {
1890         struct btrfs_key test;
1891         int ret;
1892
1893         test.objectid = sk->min_objectid;
1894         test.type = sk->min_type;
1895         test.offset = sk->min_offset;
1896
1897         ret = btrfs_comp_cpu_keys(key, &test);
1898         if (ret < 0)
1899                 return 0;
1900
1901         test.objectid = sk->max_objectid;
1902         test.type = sk->max_type;
1903         test.offset = sk->max_offset;
1904
1905         ret = btrfs_comp_cpu_keys(key, &test);
1906         if (ret > 0)
1907                 return 0;
1908         return 1;
1909 }
1910
1911 static noinline int copy_to_sk(struct btrfs_path *path,
1912                                struct btrfs_key *key,
1913                                struct btrfs_ioctl_search_key *sk,
1914                                size_t *buf_size,
1915                                char __user *ubuf,
1916                                unsigned long *sk_offset,
1917                                int *num_found)
1918 {
1919         u64 found_transid;
1920         struct extent_buffer *leaf;
1921         struct btrfs_ioctl_search_header sh;
1922         struct btrfs_key test;
1923         unsigned long item_off;
1924         unsigned long item_len;
1925         int nritems;
1926         int i;
1927         int slot;
1928         int ret = 0;
1929
1930         leaf = path->nodes[0];
1931         slot = path->slots[0];
1932         nritems = btrfs_header_nritems(leaf);
1933
1934         if (btrfs_header_generation(leaf) > sk->max_transid) {
1935                 i = nritems;
1936                 goto advance_key;
1937         }
1938         found_transid = btrfs_header_generation(leaf);
1939
1940         for (i = slot; i < nritems; i++) {
1941                 item_off = btrfs_item_ptr_offset(leaf, i);
1942                 item_len = btrfs_item_size_nr(leaf, i);
1943
1944                 btrfs_item_key_to_cpu(leaf, key, i);
1945                 if (!key_in_sk(key, sk))
1946                         continue;
1947
1948                 if (sizeof(sh) + item_len > *buf_size) {
1949                         if (*num_found) {
1950                                 ret = 1;
1951                                 goto out;
1952                         }
1953
1954                         /*
1955                          * return one empty item back for v1, which does not
1956                          * handle -EOVERFLOW
1957                          */
1958
1959                         *buf_size = sizeof(sh) + item_len;
1960                         item_len = 0;
1961                         ret = -EOVERFLOW;
1962                 }
1963
1964                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1965                         ret = 1;
1966                         goto out;
1967                 }
1968
1969                 sh.objectid = key->objectid;
1970                 sh.offset = key->offset;
1971                 sh.type = key->type;
1972                 sh.len = item_len;
1973                 sh.transid = found_transid;
1974
1975                 /* copy search result header */
1976                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1977                         ret = -EFAULT;
1978                         goto out;
1979                 }
1980
1981                 *sk_offset += sizeof(sh);
1982
1983                 if (item_len) {
1984                         char __user *up = ubuf + *sk_offset;
1985                         /* copy the item */
1986                         if (read_extent_buffer_to_user(leaf, up,
1987                                                        item_off, item_len)) {
1988                                 ret = -EFAULT;
1989                                 goto out;
1990                         }
1991
1992                         *sk_offset += item_len;
1993                 }
1994                 (*num_found)++;
1995
1996                 if (ret) /* -EOVERFLOW from above */
1997                         goto out;
1998
1999                 if (*num_found >= sk->nr_items) {
2000                         ret = 1;
2001                         goto out;
2002                 }
2003         }
2004 advance_key:
2005         ret = 0;
2006         test.objectid = sk->max_objectid;
2007         test.type = sk->max_type;
2008         test.offset = sk->max_offset;
2009         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2010                 ret = 1;
2011         else if (key->offset < (u64)-1)
2012                 key->offset++;
2013         else if (key->type < (u8)-1) {
2014                 key->offset = 0;
2015                 key->type++;
2016         } else if (key->objectid < (u64)-1) {
2017                 key->offset = 0;
2018                 key->type = 0;
2019                 key->objectid++;
2020         } else
2021                 ret = 1;
2022 out:
2023         /*
2024          *  0: all items from this leaf copied, continue with next
2025          *  1: * more items can be copied, but unused buffer is too small
2026          *     * all items were found
2027          *     Either way, it will stops the loop which iterates to the next
2028          *     leaf
2029          *  -EOVERFLOW: item was to large for buffer
2030          *  -EFAULT: could not copy extent buffer back to userspace
2031          */
2032         return ret;
2033 }
2034
2035 static noinline int search_ioctl(struct inode *inode,
2036                                  struct btrfs_ioctl_search_key *sk,
2037                                  size_t *buf_size,
2038                                  char __user *ubuf)
2039 {
2040         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2041         struct btrfs_root *root;
2042         struct btrfs_key key;
2043         struct btrfs_path *path;
2044         int ret;
2045         int num_found = 0;
2046         unsigned long sk_offset = 0;
2047
2048         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2049                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2050                 return -EOVERFLOW;
2051         }
2052
2053         path = btrfs_alloc_path();
2054         if (!path)
2055                 return -ENOMEM;
2056
2057         if (sk->tree_id == 0) {
2058                 /* search the root of the inode that was passed */
2059                 root = BTRFS_I(inode)->root;
2060         } else {
2061                 key.objectid = sk->tree_id;
2062                 key.type = BTRFS_ROOT_ITEM_KEY;
2063                 key.offset = (u64)-1;
2064                 root = btrfs_read_fs_root_no_name(info, &key);
2065                 if (IS_ERR(root)) {
2066                         btrfs_free_path(path);
2067                         return -ENOENT;
2068                 }
2069         }
2070
2071         key.objectid = sk->min_objectid;
2072         key.type = sk->min_type;
2073         key.offset = sk->min_offset;
2074
2075         while (1) {
2076                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2077                 if (ret != 0) {
2078                         if (ret > 0)
2079                                 ret = 0;
2080                         goto err;
2081                 }
2082                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2083                                  &sk_offset, &num_found);
2084                 btrfs_release_path(path);
2085                 if (ret)
2086                         break;
2087
2088         }
2089         if (ret > 0)
2090                 ret = 0;
2091 err:
2092         sk->nr_items = num_found;
2093         btrfs_free_path(path);
2094         return ret;
2095 }
2096
2097 static noinline int btrfs_ioctl_tree_search(struct file *file,
2098                                            void __user *argp)
2099 {
2100         struct btrfs_ioctl_search_args __user *uargs;
2101         struct btrfs_ioctl_search_key sk;
2102         struct inode *inode;
2103         int ret;
2104         size_t buf_size;
2105
2106         if (!capable(CAP_SYS_ADMIN))
2107                 return -EPERM;
2108
2109         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2110
2111         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2112                 return -EFAULT;
2113
2114         buf_size = sizeof(uargs->buf);
2115
2116         inode = file_inode(file);
2117         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2118
2119         /*
2120          * In the origin implementation an overflow is handled by returning a
2121          * search header with a len of zero, so reset ret.
2122          */
2123         if (ret == -EOVERFLOW)
2124                 ret = 0;
2125
2126         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2127                 ret = -EFAULT;
2128         return ret;
2129 }
2130
2131 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2132                                                void __user *argp)
2133 {
2134         struct btrfs_ioctl_search_args_v2 __user *uarg;
2135         struct btrfs_ioctl_search_args_v2 args;
2136         struct inode *inode;
2137         int ret;
2138         size_t buf_size;
2139         const size_t buf_limit = SZ_16M;
2140
2141         if (!capable(CAP_SYS_ADMIN))
2142                 return -EPERM;
2143
2144         /* copy search header and buffer size */
2145         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2146         if (copy_from_user(&args, uarg, sizeof(args)))
2147                 return -EFAULT;
2148
2149         buf_size = args.buf_size;
2150
2151         /* limit result size to 16MB */
2152         if (buf_size > buf_limit)
2153                 buf_size = buf_limit;
2154
2155         inode = file_inode(file);
2156         ret = search_ioctl(inode, &args.key, &buf_size,
2157                            (char __user *)(&uarg->buf[0]));
2158         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2159                 ret = -EFAULT;
2160         else if (ret == -EOVERFLOW &&
2161                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2162                 ret = -EFAULT;
2163
2164         return ret;
2165 }
2166
2167 /*
2168  * Search INODE_REFs to identify path name of 'dirid' directory
2169  * in a 'tree_id' tree. and sets path name to 'name'.
2170  */
2171 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2172                                 u64 tree_id, u64 dirid, char *name)
2173 {
2174         struct btrfs_root *root;
2175         struct btrfs_key key;
2176         char *ptr;
2177         int ret = -1;
2178         int slot;
2179         int len;
2180         int total_len = 0;
2181         struct btrfs_inode_ref *iref;
2182         struct extent_buffer *l;
2183         struct btrfs_path *path;
2184
2185         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2186                 name[0]='\0';
2187                 return 0;
2188         }
2189
2190         path = btrfs_alloc_path();
2191         if (!path)
2192                 return -ENOMEM;
2193
2194         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2195
2196         key.objectid = tree_id;
2197         key.type = BTRFS_ROOT_ITEM_KEY;
2198         key.offset = (u64)-1;
2199         root = btrfs_read_fs_root_no_name(info, &key);
2200         if (IS_ERR(root)) {
2201                 btrfs_err(info, "could not find root %llu", tree_id);
2202                 ret = -ENOENT;
2203                 goto out;
2204         }
2205
2206         key.objectid = dirid;
2207         key.type = BTRFS_INODE_REF_KEY;
2208         key.offset = (u64)-1;
2209
2210         while (1) {
2211                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2212                 if (ret < 0)
2213                         goto out;
2214                 else if (ret > 0) {
2215                         ret = btrfs_previous_item(root, path, dirid,
2216                                                   BTRFS_INODE_REF_KEY);
2217                         if (ret < 0)
2218                                 goto out;
2219                         else if (ret > 0) {
2220                                 ret = -ENOENT;
2221                                 goto out;
2222                         }
2223                 }
2224
2225                 l = path->nodes[0];
2226                 slot = path->slots[0];
2227                 btrfs_item_key_to_cpu(l, &key, slot);
2228
2229                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2230                 len = btrfs_inode_ref_name_len(l, iref);
2231                 ptr -= len + 1;
2232                 total_len += len + 1;
2233                 if (ptr < name) {
2234                         ret = -ENAMETOOLONG;
2235                         goto out;
2236                 }
2237
2238                 *(ptr + len) = '/';
2239                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2240
2241                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2242                         break;
2243
2244                 btrfs_release_path(path);
2245                 key.objectid = key.offset;
2246                 key.offset = (u64)-1;
2247                 dirid = key.objectid;
2248         }
2249         memmove(name, ptr, total_len);
2250         name[total_len] = '\0';
2251         ret = 0;
2252 out:
2253         btrfs_free_path(path);
2254         return ret;
2255 }
2256
2257 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2258                                            void __user *argp)
2259 {
2260          struct btrfs_ioctl_ino_lookup_args *args;
2261          struct inode *inode;
2262         int ret = 0;
2263
2264         args = memdup_user(argp, sizeof(*args));
2265         if (IS_ERR(args))
2266                 return PTR_ERR(args);
2267
2268         inode = file_inode(file);
2269
2270         /*
2271          * Unprivileged query to obtain the containing subvolume root id. The
2272          * path is reset so it's consistent with btrfs_search_path_in_tree.
2273          */
2274         if (args->treeid == 0)
2275                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2276
2277         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2278                 args->name[0] = 0;
2279                 goto out;
2280         }
2281
2282         if (!capable(CAP_SYS_ADMIN)) {
2283                 ret = -EPERM;
2284                 goto out;
2285         }
2286
2287         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2288                                         args->treeid, args->objectid,
2289                                         args->name);
2290
2291 out:
2292         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2293                 ret = -EFAULT;
2294
2295         kfree(args);
2296         return ret;
2297 }
2298
2299 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2300                                              void __user *arg)
2301 {
2302         struct dentry *parent = file->f_path.dentry;
2303         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2304         struct dentry *dentry;
2305         struct inode *dir = d_inode(parent);
2306         struct inode *inode;
2307         struct btrfs_root *root = BTRFS_I(dir)->root;
2308         struct btrfs_root *dest = NULL;
2309         struct btrfs_ioctl_vol_args *vol_args;
2310         struct btrfs_trans_handle *trans;
2311         struct btrfs_block_rsv block_rsv;
2312         u64 root_flags;
2313         u64 qgroup_reserved;
2314         int namelen;
2315         int ret;
2316         int err = 0;
2317
2318         if (!S_ISDIR(dir->i_mode))
2319                 return -ENOTDIR;
2320
2321         vol_args = memdup_user(arg, sizeof(*vol_args));
2322         if (IS_ERR(vol_args))
2323                 return PTR_ERR(vol_args);
2324
2325         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2326         namelen = strlen(vol_args->name);
2327         if (strchr(vol_args->name, '/') ||
2328             strncmp(vol_args->name, "..", namelen) == 0) {
2329                 err = -EINVAL;
2330                 goto out;
2331         }
2332
2333         err = mnt_want_write_file(file);
2334         if (err)
2335                 goto out;
2336
2337
2338         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2339         if (err == -EINTR)
2340                 goto out_drop_write;
2341         dentry = lookup_one_len(vol_args->name, parent, namelen);
2342         if (IS_ERR(dentry)) {
2343                 err = PTR_ERR(dentry);
2344                 goto out_unlock_dir;
2345         }
2346
2347         if (d_really_is_negative(dentry)) {
2348                 err = -ENOENT;
2349                 goto out_dput;
2350         }
2351
2352         inode = d_inode(dentry);
2353         dest = BTRFS_I(inode)->root;
2354         if (!capable(CAP_SYS_ADMIN)) {
2355                 /*
2356                  * Regular user.  Only allow this with a special mount
2357                  * option, when the user has write+exec access to the
2358                  * subvol root, and when rmdir(2) would have been
2359                  * allowed.
2360                  *
2361                  * Note that this is _not_ check that the subvol is
2362                  * empty or doesn't contain data that we wouldn't
2363                  * otherwise be able to delete.
2364                  *
2365                  * Users who want to delete empty subvols should try
2366                  * rmdir(2).
2367                  */
2368                 err = -EPERM;
2369                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2370                         goto out_dput;
2371
2372                 /*
2373                  * Do not allow deletion if the parent dir is the same
2374                  * as the dir to be deleted.  That means the ioctl
2375                  * must be called on the dentry referencing the root
2376                  * of the subvol, not a random directory contained
2377                  * within it.
2378                  */
2379                 err = -EINVAL;
2380                 if (root == dest)
2381                         goto out_dput;
2382
2383                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2384                 if (err)
2385                         goto out_dput;
2386         }
2387
2388         /* check if subvolume may be deleted by a user */
2389         err = btrfs_may_delete(dir, dentry, 1);
2390         if (err)
2391                 goto out_dput;
2392
2393         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2394                 err = -EINVAL;
2395                 goto out_dput;
2396         }
2397
2398         inode_lock(inode);
2399
2400         /*
2401          * Don't allow to delete a subvolume with send in progress. This is
2402          * inside the i_mutex so the error handling that has to drop the bit
2403          * again is not run concurrently.
2404          */
2405         spin_lock(&dest->root_item_lock);
2406         root_flags = btrfs_root_flags(&dest->root_item);
2407         if (dest->send_in_progress == 0) {
2408                 btrfs_set_root_flags(&dest->root_item,
2409                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2410                 spin_unlock(&dest->root_item_lock);
2411         } else {
2412                 spin_unlock(&dest->root_item_lock);
2413                 btrfs_warn(fs_info,
2414                            "Attempt to delete subvolume %llu during send",
2415                            dest->root_key.objectid);
2416                 err = -EPERM;
2417                 goto out_unlock_inode;
2418         }
2419
2420         down_write(&fs_info->subvol_sem);
2421
2422         err = may_destroy_subvol(dest);
2423         if (err)
2424                 goto out_up_write;
2425
2426         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2427         /*
2428          * One for dir inode, two for dir entries, two for root
2429          * ref/backref.
2430          */
2431         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2432                                                5, &qgroup_reserved, true);
2433         if (err)
2434                 goto out_up_write;
2435
2436         trans = btrfs_start_transaction(root, 0);
2437         if (IS_ERR(trans)) {
2438                 err = PTR_ERR(trans);
2439                 goto out_release;
2440         }
2441         trans->block_rsv = &block_rsv;
2442         trans->bytes_reserved = block_rsv.size;
2443
2444         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2445
2446         ret = btrfs_unlink_subvol(trans, root, dir,
2447                                 dest->root_key.objectid,
2448                                 dentry->d_name.name,
2449                                 dentry->d_name.len);
2450         if (ret) {
2451                 err = ret;
2452                 btrfs_abort_transaction(trans, ret);
2453                 goto out_end_trans;
2454         }
2455
2456         btrfs_record_root_in_trans(trans, dest);
2457
2458         memset(&dest->root_item.drop_progress, 0,
2459                 sizeof(dest->root_item.drop_progress));
2460         dest->root_item.drop_level = 0;
2461         btrfs_set_root_refs(&dest->root_item, 0);
2462
2463         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2464                 ret = btrfs_insert_orphan_item(trans,
2465                                         fs_info->tree_root,
2466                                         dest->root_key.objectid);
2467                 if (ret) {
2468                         btrfs_abort_transaction(trans, ret);
2469                         err = ret;
2470                         goto out_end_trans;
2471                 }
2472         }
2473
2474         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2475                                   BTRFS_UUID_KEY_SUBVOL,
2476                                   dest->root_key.objectid);
2477         if (ret && ret != -ENOENT) {
2478                 btrfs_abort_transaction(trans, ret);
2479                 err = ret;
2480                 goto out_end_trans;
2481         }
2482         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2483                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2484                                           dest->root_item.received_uuid,
2485                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2486                                           dest->root_key.objectid);
2487                 if (ret && ret != -ENOENT) {
2488                         btrfs_abort_transaction(trans, ret);
2489                         err = ret;
2490                         goto out_end_trans;
2491                 }
2492         }
2493
2494 out_end_trans:
2495         trans->block_rsv = NULL;
2496         trans->bytes_reserved = 0;
2497         ret = btrfs_end_transaction(trans);
2498         if (ret && !err)
2499                 err = ret;
2500         inode->i_flags |= S_DEAD;
2501 out_release:
2502         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2503 out_up_write:
2504         up_write(&fs_info->subvol_sem);
2505         if (err) {
2506                 spin_lock(&dest->root_item_lock);
2507                 root_flags = btrfs_root_flags(&dest->root_item);
2508                 btrfs_set_root_flags(&dest->root_item,
2509                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2510                 spin_unlock(&dest->root_item_lock);
2511         }
2512 out_unlock_inode:
2513         inode_unlock(inode);
2514         if (!err) {
2515                 d_invalidate(dentry);
2516                 btrfs_invalidate_inodes(dest);
2517                 d_delete(dentry);
2518                 ASSERT(dest->send_in_progress == 0);
2519
2520                 /* the last ref */
2521                 if (dest->ino_cache_inode) {
2522                         iput(dest->ino_cache_inode);
2523                         dest->ino_cache_inode = NULL;
2524                 }
2525         }
2526 out_dput:
2527         dput(dentry);
2528 out_unlock_dir:
2529         inode_unlock(dir);
2530 out_drop_write:
2531         mnt_drop_write_file(file);
2532 out:
2533         kfree(vol_args);
2534         return err;
2535 }
2536
2537 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2538 {
2539         struct inode *inode = file_inode(file);
2540         struct btrfs_root *root = BTRFS_I(inode)->root;
2541         struct btrfs_ioctl_defrag_range_args *range;
2542         int ret;
2543
2544         ret = mnt_want_write_file(file);
2545         if (ret)
2546                 return ret;
2547
2548         if (btrfs_root_readonly(root)) {
2549                 ret = -EROFS;
2550                 goto out;
2551         }
2552
2553         switch (inode->i_mode & S_IFMT) {
2554         case S_IFDIR:
2555                 if (!capable(CAP_SYS_ADMIN)) {
2556                         ret = -EPERM;
2557                         goto out;
2558                 }
2559                 ret = btrfs_defrag_root(root);
2560                 break;
2561         case S_IFREG:
2562                 if (!(file->f_mode & FMODE_WRITE)) {
2563                         ret = -EINVAL;
2564                         goto out;
2565                 }
2566
2567                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2568                 if (!range) {
2569                         ret = -ENOMEM;
2570                         goto out;
2571                 }
2572
2573                 if (argp) {
2574                         if (copy_from_user(range, argp,
2575                                            sizeof(*range))) {
2576                                 ret = -EFAULT;
2577                                 kfree(range);
2578                                 goto out;
2579                         }
2580                         /* compression requires us to start the IO */
2581                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2582                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2583                                 range->extent_thresh = (u32)-1;
2584                         }
2585                 } else {
2586                         /* the rest are all set to zero by kzalloc */
2587                         range->len = (u64)-1;
2588                 }
2589                 ret = btrfs_defrag_file(file_inode(file), file,
2590                                         range, 0, 0);
2591                 if (ret > 0)
2592                         ret = 0;
2593                 kfree(range);
2594                 break;
2595         default:
2596                 ret = -EINVAL;
2597         }
2598 out:
2599         mnt_drop_write_file(file);
2600         return ret;
2601 }
2602
2603 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2604 {
2605         struct btrfs_ioctl_vol_args *vol_args;
2606         int ret;
2607
2608         if (!capable(CAP_SYS_ADMIN))
2609                 return -EPERM;
2610
2611         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2612                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2613
2614         mutex_lock(&fs_info->volume_mutex);
2615         vol_args = memdup_user(arg, sizeof(*vol_args));
2616         if (IS_ERR(vol_args)) {
2617                 ret = PTR_ERR(vol_args);
2618                 goto out;
2619         }
2620
2621         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2622         ret = btrfs_init_new_device(fs_info, vol_args->name);
2623
2624         if (!ret)
2625                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2626
2627         kfree(vol_args);
2628 out:
2629         mutex_unlock(&fs_info->volume_mutex);
2630         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2631         return ret;
2632 }
2633
2634 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2635 {
2636         struct inode *inode = file_inode(file);
2637         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2638         struct btrfs_ioctl_vol_args_v2 *vol_args;
2639         int ret;
2640
2641         if (!capable(CAP_SYS_ADMIN))
2642                 return -EPERM;
2643
2644         ret = mnt_want_write_file(file);
2645         if (ret)
2646                 return ret;
2647
2648         vol_args = memdup_user(arg, sizeof(*vol_args));
2649         if (IS_ERR(vol_args)) {
2650                 ret = PTR_ERR(vol_args);
2651                 goto err_drop;
2652         }
2653
2654         /* Check for compatibility reject unknown flags */
2655         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2656                 return -EOPNOTSUPP;
2657
2658         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2659                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2660                 goto out;
2661         }
2662
2663         mutex_lock(&fs_info->volume_mutex);
2664         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2665                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2666         } else {
2667                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2668                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2669         }
2670         mutex_unlock(&fs_info->volume_mutex);
2671         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2672
2673         if (!ret) {
2674                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2675                         btrfs_info(fs_info, "device deleted: id %llu",
2676                                         vol_args->devid);
2677                 else
2678                         btrfs_info(fs_info, "device deleted: %s",
2679                                         vol_args->name);
2680         }
2681 out:
2682         kfree(vol_args);
2683 err_drop:
2684         mnt_drop_write_file(file);
2685         return ret;
2686 }
2687
2688 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2689 {
2690         struct inode *inode = file_inode(file);
2691         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2692         struct btrfs_ioctl_vol_args *vol_args;
2693         int ret;
2694
2695         if (!capable(CAP_SYS_ADMIN))
2696                 return -EPERM;
2697
2698         ret = mnt_want_write_file(file);
2699         if (ret)
2700                 return ret;
2701
2702         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2703                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2704                 goto out_drop_write;
2705         }
2706
2707         vol_args = memdup_user(arg, sizeof(*vol_args));
2708         if (IS_ERR(vol_args)) {
2709                 ret = PTR_ERR(vol_args);
2710                 goto out;
2711         }
2712
2713         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2714         mutex_lock(&fs_info->volume_mutex);
2715         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2716         mutex_unlock(&fs_info->volume_mutex);
2717
2718         if (!ret)
2719                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2720         kfree(vol_args);
2721 out:
2722         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2723 out_drop_write:
2724         mnt_drop_write_file(file);
2725
2726         return ret;
2727 }
2728
2729 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2730                                 void __user *arg)
2731 {
2732         struct btrfs_ioctl_fs_info_args *fi_args;
2733         struct btrfs_device *device;
2734         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2735         int ret = 0;
2736
2737         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2738         if (!fi_args)
2739                 return -ENOMEM;
2740
2741         mutex_lock(&fs_devices->device_list_mutex);
2742         fi_args->num_devices = fs_devices->num_devices;
2743         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2744
2745         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2746                 if (device->devid > fi_args->max_id)
2747                         fi_args->max_id = device->devid;
2748         }
2749         mutex_unlock(&fs_devices->device_list_mutex);
2750
2751         fi_args->nodesize = fs_info->nodesize;
2752         fi_args->sectorsize = fs_info->sectorsize;
2753         fi_args->clone_alignment = fs_info->sectorsize;
2754
2755         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2756                 ret = -EFAULT;
2757
2758         kfree(fi_args);
2759         return ret;
2760 }
2761
2762 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2763                                  void __user *arg)
2764 {
2765         struct btrfs_ioctl_dev_info_args *di_args;
2766         struct btrfs_device *dev;
2767         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2768         int ret = 0;
2769         char *s_uuid = NULL;
2770
2771         di_args = memdup_user(arg, sizeof(*di_args));
2772         if (IS_ERR(di_args))
2773                 return PTR_ERR(di_args);
2774
2775         if (!btrfs_is_empty_uuid(di_args->uuid))
2776                 s_uuid = di_args->uuid;
2777
2778         mutex_lock(&fs_devices->device_list_mutex);
2779         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2780
2781         if (!dev) {
2782                 ret = -ENODEV;
2783                 goto out;
2784         }
2785
2786         di_args->devid = dev->devid;
2787         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2788         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2789         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2790         if (dev->name) {
2791                 struct rcu_string *name;
2792
2793                 rcu_read_lock();
2794                 name = rcu_dereference(dev->name);
2795                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2796                 rcu_read_unlock();
2797                 di_args->path[sizeof(di_args->path) - 1] = 0;
2798         } else {
2799                 di_args->path[0] = '\0';
2800         }
2801
2802 out:
2803         mutex_unlock(&fs_devices->device_list_mutex);
2804         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2805                 ret = -EFAULT;
2806
2807         kfree(di_args);
2808         return ret;
2809 }
2810
2811 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2812 {
2813         struct page *page;
2814
2815         page = grab_cache_page(inode->i_mapping, index);
2816         if (!page)
2817                 return ERR_PTR(-ENOMEM);
2818
2819         if (!PageUptodate(page)) {
2820                 int ret;
2821
2822                 ret = btrfs_readpage(NULL, page);
2823                 if (ret)
2824                         return ERR_PTR(ret);
2825                 lock_page(page);
2826                 if (!PageUptodate(page)) {
2827                         unlock_page(page);
2828                         put_page(page);
2829                         return ERR_PTR(-EIO);
2830                 }
2831                 if (page->mapping != inode->i_mapping) {
2832                         unlock_page(page);
2833                         put_page(page);
2834                         return ERR_PTR(-EAGAIN);
2835                 }
2836         }
2837
2838         return page;
2839 }
2840
2841 static int gather_extent_pages(struct inode *inode, struct page **pages,
2842                                int num_pages, u64 off)
2843 {
2844         int i;
2845         pgoff_t index = off >> PAGE_SHIFT;
2846
2847         for (i = 0; i < num_pages; i++) {
2848 again:
2849                 pages[i] = extent_same_get_page(inode, index + i);
2850                 if (IS_ERR(pages[i])) {
2851                         int err = PTR_ERR(pages[i]);
2852
2853                         if (err == -EAGAIN)
2854                                 goto again;
2855                         pages[i] = NULL;
2856                         return err;
2857                 }
2858         }
2859         return 0;
2860 }
2861
2862 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2863                              bool retry_range_locking)
2864 {
2865         /*
2866          * Do any pending delalloc/csum calculations on inode, one way or
2867          * another, and lock file content.
2868          * The locking order is:
2869          *
2870          *   1) pages
2871          *   2) range in the inode's io tree
2872          */
2873         while (1) {
2874                 struct btrfs_ordered_extent *ordered;
2875                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2876                 ordered = btrfs_lookup_first_ordered_extent(inode,
2877                                                             off + len - 1);
2878                 if ((!ordered ||
2879                      ordered->file_offset + ordered->len <= off ||
2880                      ordered->file_offset >= off + len) &&
2881                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2882                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2883                         if (ordered)
2884                                 btrfs_put_ordered_extent(ordered);
2885                         break;
2886                 }
2887                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2888                 if (ordered)
2889                         btrfs_put_ordered_extent(ordered);
2890                 if (!retry_range_locking)
2891                         return -EAGAIN;
2892                 btrfs_wait_ordered_range(inode, off, len);
2893         }
2894         return 0;
2895 }
2896
2897 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2898 {
2899         inode_unlock(inode1);
2900         inode_unlock(inode2);
2901 }
2902
2903 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2904 {
2905         if (inode1 < inode2)
2906                 swap(inode1, inode2);
2907
2908         inode_lock_nested(inode1, I_MUTEX_PARENT);
2909         inode_lock_nested(inode2, I_MUTEX_CHILD);
2910 }
2911
2912 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2913                                       struct inode *inode2, u64 loff2, u64 len)
2914 {
2915         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2916         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2917 }
2918
2919 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2920                                     struct inode *inode2, u64 loff2, u64 len,
2921                                     bool retry_range_locking)
2922 {
2923         int ret;
2924
2925         if (inode1 < inode2) {
2926                 swap(inode1, inode2);
2927                 swap(loff1, loff2);
2928         }
2929         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2930         if (ret)
2931                 return ret;
2932         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2933         if (ret)
2934                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2935                               loff1 + len - 1);
2936         return ret;
2937 }
2938
2939 struct cmp_pages {
2940         int             num_pages;
2941         struct page     **src_pages;
2942         struct page     **dst_pages;
2943 };
2944
2945 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2946 {
2947         int i;
2948         struct page *pg;
2949
2950         for (i = 0; i < cmp->num_pages; i++) {
2951                 pg = cmp->src_pages[i];
2952                 if (pg) {
2953                         unlock_page(pg);
2954                         put_page(pg);
2955                 }
2956                 pg = cmp->dst_pages[i];
2957                 if (pg) {
2958                         unlock_page(pg);
2959                         put_page(pg);
2960                 }
2961         }
2962         kfree(cmp->src_pages);
2963         kfree(cmp->dst_pages);
2964 }
2965
2966 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2967                                   struct inode *dst, u64 dst_loff,
2968                                   u64 len, struct cmp_pages *cmp)
2969 {
2970         int ret;
2971         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2972         struct page **src_pgarr, **dst_pgarr;
2973
2974         /*
2975          * We must gather up all the pages before we initiate our
2976          * extent locking. We use an array for the page pointers. Size
2977          * of the array is bounded by len, which is in turn bounded by
2978          * BTRFS_MAX_DEDUPE_LEN.
2979          */
2980         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2981         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2982         if (!src_pgarr || !dst_pgarr) {
2983                 kfree(src_pgarr);
2984                 kfree(dst_pgarr);
2985                 return -ENOMEM;
2986         }
2987         cmp->num_pages = num_pages;
2988         cmp->src_pages = src_pgarr;
2989         cmp->dst_pages = dst_pgarr;
2990
2991         /*
2992          * If deduping ranges in the same inode, locking rules make it mandatory
2993          * to always lock pages in ascending order to avoid deadlocks with
2994          * concurrent tasks (such as starting writeback/delalloc).
2995          */
2996         if (src == dst && dst_loff < loff) {
2997                 swap(src_pgarr, dst_pgarr);
2998                 swap(loff, dst_loff);
2999         }
3000
3001         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3002         if (ret)
3003                 goto out;
3004
3005         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3006
3007 out:
3008         if (ret)
3009                 btrfs_cmp_data_free(cmp);
3010         return ret;
3011 }
3012
3013 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3014 {
3015         int ret = 0;
3016         int i;
3017         struct page *src_page, *dst_page;
3018         unsigned int cmp_len = PAGE_SIZE;
3019         void *addr, *dst_addr;
3020
3021         i = 0;
3022         while (len) {
3023                 if (len < PAGE_SIZE)
3024                         cmp_len = len;
3025
3026                 BUG_ON(i >= cmp->num_pages);
3027
3028                 src_page = cmp->src_pages[i];
3029                 dst_page = cmp->dst_pages[i];
3030                 ASSERT(PageLocked(src_page));
3031                 ASSERT(PageLocked(dst_page));
3032
3033                 addr = kmap_atomic(src_page);
3034                 dst_addr = kmap_atomic(dst_page);
3035
3036                 flush_dcache_page(src_page);
3037                 flush_dcache_page(dst_page);
3038
3039                 if (memcmp(addr, dst_addr, cmp_len))
3040                         ret = -EBADE;
3041
3042                 kunmap_atomic(addr);
3043                 kunmap_atomic(dst_addr);
3044
3045                 if (ret)
3046                         break;
3047
3048                 len -= cmp_len;
3049                 i++;
3050         }
3051
3052         return ret;
3053 }
3054
3055 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3056                                      u64 olen)
3057 {
3058         u64 len = *plen;
3059         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3060
3061         if (off + olen > inode->i_size || off + olen < off)
3062                 return -EINVAL;
3063
3064         /* if we extend to eof, continue to block boundary */
3065         if (off + len == inode->i_size)
3066                 *plen = len = ALIGN(inode->i_size, bs) - off;
3067
3068         /* Check that we are block aligned - btrfs_clone() requires this */
3069         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3070                 return -EINVAL;
3071
3072         return 0;
3073 }
3074
3075 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3076                              struct inode *dst, u64 dst_loff)
3077 {
3078         int ret;
3079         u64 len = olen;
3080         struct cmp_pages cmp;
3081         bool same_inode = (src == dst);
3082         u64 same_lock_start = 0;
3083         u64 same_lock_len = 0;
3084
3085         if (len == 0)
3086                 return 0;
3087
3088         if (same_inode)
3089                 inode_lock(src);
3090         else
3091                 btrfs_double_inode_lock(src, dst);
3092
3093         ret = extent_same_check_offsets(src, loff, &len, olen);
3094         if (ret)
3095                 goto out_unlock;
3096
3097         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3098         if (ret)
3099                 goto out_unlock;
3100
3101         if (same_inode) {
3102                 /*
3103                  * Single inode case wants the same checks, except we
3104                  * don't want our length pushed out past i_size as
3105                  * comparing that data range makes no sense.
3106                  *
3107                  * extent_same_check_offsets() will do this for an
3108                  * unaligned length at i_size, so catch it here and
3109                  * reject the request.
3110                  *
3111                  * This effectively means we require aligned extents
3112                  * for the single-inode case, whereas the other cases
3113                  * allow an unaligned length so long as it ends at
3114                  * i_size.
3115                  */
3116                 if (len != olen) {
3117                         ret = -EINVAL;
3118                         goto out_unlock;
3119                 }
3120
3121                 /* Check for overlapping ranges */
3122                 if (dst_loff + len > loff && dst_loff < loff + len) {
3123                         ret = -EINVAL;
3124                         goto out_unlock;
3125                 }
3126
3127                 same_lock_start = min_t(u64, loff, dst_loff);
3128                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3129         }
3130
3131         /* don't make the dst file partly checksummed */
3132         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3133             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3134                 ret = -EINVAL;
3135                 goto out_unlock;
3136         }
3137
3138 again:
3139         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3140         if (ret)
3141                 goto out_unlock;
3142
3143         if (same_inode)
3144                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3145                                         false);
3146         else
3147                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3148                                                false);
3149         /*
3150          * If one of the inodes has dirty pages in the respective range or
3151          * ordered extents, we need to flush dellaloc and wait for all ordered
3152          * extents in the range. We must unlock the pages and the ranges in the
3153          * io trees to avoid deadlocks when flushing delalloc (requires locking
3154          * pages) and when waiting for ordered extents to complete (they require
3155          * range locking).
3156          */
3157         if (ret == -EAGAIN) {
3158                 /*
3159                  * Ranges in the io trees already unlocked. Now unlock all
3160                  * pages before waiting for all IO to complete.
3161                  */
3162                 btrfs_cmp_data_free(&cmp);
3163                 if (same_inode) {
3164                         btrfs_wait_ordered_range(src, same_lock_start,
3165                                                  same_lock_len);
3166                 } else {
3167                         btrfs_wait_ordered_range(src, loff, len);
3168                         btrfs_wait_ordered_range(dst, dst_loff, len);
3169                 }
3170                 goto again;
3171         }
3172         ASSERT(ret == 0);
3173         if (WARN_ON(ret)) {
3174                 /* ranges in the io trees already unlocked */
3175                 btrfs_cmp_data_free(&cmp);
3176                 return ret;
3177         }
3178
3179         /* pass original length for comparison so we stay within i_size */
3180         ret = btrfs_cmp_data(olen, &cmp);
3181         if (ret == 0)
3182                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3183
3184         if (same_inode)
3185                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3186                               same_lock_start + same_lock_len - 1);
3187         else
3188                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3189
3190         btrfs_cmp_data_free(&cmp);
3191 out_unlock:
3192         if (same_inode)
3193                 inode_unlock(src);
3194         else
3195                 btrfs_double_inode_unlock(src, dst);
3196
3197         return ret;
3198 }
3199
3200 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3201
3202 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3203                                 struct file *dst_file, u64 dst_loff)
3204 {
3205         struct inode *src = file_inode(src_file);
3206         struct inode *dst = file_inode(dst_file);
3207         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3208         ssize_t res;
3209
3210         if (olen > BTRFS_MAX_DEDUPE_LEN)
3211                 olen = BTRFS_MAX_DEDUPE_LEN;
3212
3213         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3214                 /*
3215                  * Btrfs does not support blocksize < page_size. As a
3216                  * result, btrfs_cmp_data() won't correctly handle
3217                  * this situation without an update.
3218                  */
3219                 return -EINVAL;
3220         }
3221
3222         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3223         if (res)
3224                 return res;
3225         return olen;
3226 }
3227
3228 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3229                                      struct inode *inode,
3230                                      u64 endoff,
3231                                      const u64 destoff,
3232                                      const u64 olen,
3233                                      int no_time_update)
3234 {
3235         struct btrfs_root *root = BTRFS_I(inode)->root;
3236         int ret;
3237
3238         inode_inc_iversion(inode);
3239         if (!no_time_update)
3240                 inode->i_mtime = inode->i_ctime = current_time(inode);
3241         /*
3242          * We round up to the block size at eof when determining which
3243          * extents to clone above, but shouldn't round up the file size.
3244          */
3245         if (endoff > destoff + olen)
3246                 endoff = destoff + olen;
3247         if (endoff > inode->i_size)
3248                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3249
3250         ret = btrfs_update_inode(trans, root, inode);
3251         if (ret) {
3252                 btrfs_abort_transaction(trans, ret);
3253                 btrfs_end_transaction(trans);
3254                 goto out;
3255         }
3256         ret = btrfs_end_transaction(trans);
3257 out:
3258         return ret;
3259 }
3260
3261 static void clone_update_extent_map(struct btrfs_inode *inode,
3262                                     const struct btrfs_trans_handle *trans,
3263                                     const struct btrfs_path *path,
3264                                     const u64 hole_offset,
3265                                     const u64 hole_len)
3266 {
3267         struct extent_map_tree *em_tree = &inode->extent_tree;
3268         struct extent_map *em;
3269         int ret;
3270
3271         em = alloc_extent_map();
3272         if (!em) {
3273                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3274                 return;
3275         }
3276
3277         if (path) {
3278                 struct btrfs_file_extent_item *fi;
3279
3280                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3281                                     struct btrfs_file_extent_item);
3282                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3283                 em->generation = -1;
3284                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3285                     BTRFS_FILE_EXTENT_INLINE)
3286                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3287                                         &inode->runtime_flags);
3288         } else {
3289                 em->start = hole_offset;
3290                 em->len = hole_len;
3291                 em->ram_bytes = em->len;
3292                 em->orig_start = hole_offset;
3293                 em->block_start = EXTENT_MAP_HOLE;
3294                 em->block_len = 0;
3295                 em->orig_block_len = 0;
3296                 em->compress_type = BTRFS_COMPRESS_NONE;
3297                 em->generation = trans->transid;
3298         }
3299
3300         while (1) {
3301                 write_lock(&em_tree->lock);
3302                 ret = add_extent_mapping(em_tree, em, 1);
3303                 write_unlock(&em_tree->lock);
3304                 if (ret != -EEXIST) {
3305                         free_extent_map(em);
3306                         break;
3307                 }
3308                 btrfs_drop_extent_cache(inode, em->start,
3309                                         em->start + em->len - 1, 0);
3310         }
3311
3312         if (ret)
3313                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3314 }
3315
3316 /*
3317  * Make sure we do not end up inserting an inline extent into a file that has
3318  * already other (non-inline) extents. If a file has an inline extent it can
3319  * not have any other extents and the (single) inline extent must start at the
3320  * file offset 0. Failing to respect these rules will lead to file corruption,
3321  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3322  *
3323  * We can have extents that have been already written to disk or we can have
3324  * dirty ranges still in delalloc, in which case the extent maps and items are
3325  * created only when we run delalloc, and the delalloc ranges might fall outside
3326  * the range we are currently locking in the inode's io tree. So we check the
3327  * inode's i_size because of that (i_size updates are done while holding the
3328  * i_mutex, which we are holding here).
3329  * We also check to see if the inode has a size not greater than "datal" but has
3330  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3331  * protected against such concurrent fallocate calls by the i_mutex).
3332  *
3333  * If the file has no extents but a size greater than datal, do not allow the
3334  * copy because we would need turn the inline extent into a non-inline one (even
3335  * with NO_HOLES enabled). If we find our destination inode only has one inline
3336  * extent, just overwrite it with the source inline extent if its size is less
3337  * than the source extent's size, or we could copy the source inline extent's
3338  * data into the destination inode's inline extent if the later is greater then
3339  * the former.
3340  */
3341 static int clone_copy_inline_extent(struct inode *dst,
3342                                     struct btrfs_trans_handle *trans,
3343                                     struct btrfs_path *path,
3344                                     struct btrfs_key *new_key,
3345                                     const u64 drop_start,
3346                                     const u64 datal,
3347                                     const u64 skip,
3348                                     const u64 size,
3349                                     char *inline_data)
3350 {
3351         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3352         struct btrfs_root *root = BTRFS_I(dst)->root;
3353         const u64 aligned_end = ALIGN(new_key->offset + datal,
3354                                       fs_info->sectorsize);
3355         int ret;
3356         struct btrfs_key key;
3357
3358         if (new_key->offset > 0)
3359                 return -EOPNOTSUPP;
3360
3361         key.objectid = btrfs_ino(BTRFS_I(dst));
3362         key.type = BTRFS_EXTENT_DATA_KEY;
3363         key.offset = 0;
3364         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3365         if (ret < 0) {
3366                 return ret;
3367         } else if (ret > 0) {
3368                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3369                         ret = btrfs_next_leaf(root, path);
3370                         if (ret < 0)
3371                                 return ret;
3372                         else if (ret > 0)
3373                                 goto copy_inline_extent;
3374                 }
3375                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3376                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3377                     key.type == BTRFS_EXTENT_DATA_KEY) {
3378                         ASSERT(key.offset > 0);
3379                         return -EOPNOTSUPP;
3380                 }
3381         } else if (i_size_read(dst) <= datal) {
3382                 struct btrfs_file_extent_item *ei;
3383                 u64 ext_len;
3384
3385                 /*
3386                  * If the file size is <= datal, make sure there are no other
3387                  * extents following (can happen do to an fallocate call with
3388                  * the flag FALLOC_FL_KEEP_SIZE).
3389                  */
3390                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3391                                     struct btrfs_file_extent_item);
3392                 /*
3393                  * If it's an inline extent, it can not have other extents
3394                  * following it.
3395                  */
3396                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3397                     BTRFS_FILE_EXTENT_INLINE)
3398                         goto copy_inline_extent;
3399
3400                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3401                 if (ext_len > aligned_end)
3402                         return -EOPNOTSUPP;
3403
3404                 ret = btrfs_next_item(root, path);
3405                 if (ret < 0) {
3406                         return ret;
3407                 } else if (ret == 0) {
3408                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3409                                               path->slots[0]);
3410                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3411                             key.type == BTRFS_EXTENT_DATA_KEY)
3412                                 return -EOPNOTSUPP;
3413                 }
3414         }
3415
3416 copy_inline_extent:
3417         /*
3418          * We have no extent items, or we have an extent at offset 0 which may
3419          * or may not be inlined. All these cases are dealt the same way.
3420          */
3421         if (i_size_read(dst) > datal) {
3422                 /*
3423                  * If the destination inode has an inline extent...
3424                  * This would require copying the data from the source inline
3425                  * extent into the beginning of the destination's inline extent.
3426                  * But this is really complex, both extents can be compressed
3427                  * or just one of them, which would require decompressing and
3428                  * re-compressing data (which could increase the new compressed
3429                  * size, not allowing the compressed data to fit anymore in an
3430                  * inline extent).
3431                  * So just don't support this case for now (it should be rare,
3432                  * we are not really saving space when cloning inline extents).
3433                  */
3434                 return -EOPNOTSUPP;
3435         }
3436
3437         btrfs_release_path(path);
3438         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3439         if (ret)
3440                 return ret;
3441         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3442         if (ret)
3443                 return ret;
3444
3445         if (skip) {
3446                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3447
3448                 memmove(inline_data + start, inline_data + start + skip, datal);
3449         }
3450
3451         write_extent_buffer(path->nodes[0], inline_data,
3452                             btrfs_item_ptr_offset(path->nodes[0],
3453                                                   path->slots[0]),
3454                             size);
3455         inode_add_bytes(dst, datal);
3456
3457         return 0;
3458 }
3459
3460 /**
3461  * btrfs_clone() - clone a range from inode file to another
3462  *
3463  * @src: Inode to clone from
3464  * @inode: Inode to clone to
3465  * @off: Offset within source to start clone from
3466  * @olen: Original length, passed by user, of range to clone
3467  * @olen_aligned: Block-aligned value of olen
3468  * @destoff: Offset within @inode to start clone
3469  * @no_time_update: Whether to update mtime/ctime on the target inode
3470  */
3471 static int btrfs_clone(struct inode *src, struct inode *inode,
3472                        const u64 off, const u64 olen, const u64 olen_aligned,
3473                        const u64 destoff, int no_time_update)
3474 {
3475         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3476         struct btrfs_root *root = BTRFS_I(inode)->root;
3477         struct btrfs_path *path = NULL;
3478         struct extent_buffer *leaf;
3479         struct btrfs_trans_handle *trans;
3480         char *buf = NULL;
3481         struct btrfs_key key;
3482         u32 nritems;
3483         int slot;
3484         int ret;
3485         const u64 len = olen_aligned;
3486         u64 last_dest_end = destoff;
3487
3488         ret = -ENOMEM;
3489         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3490         if (!buf)
3491                 return ret;
3492
3493         path = btrfs_alloc_path();
3494         if (!path) {
3495                 kvfree(buf);
3496                 return ret;
3497         }
3498
3499         path->reada = READA_FORWARD;
3500         /* clone data */
3501         key.objectid = btrfs_ino(BTRFS_I(src));
3502         key.type = BTRFS_EXTENT_DATA_KEY;
3503         key.offset = off;
3504
3505         while (1) {
3506                 u64 next_key_min_offset = key.offset + 1;
3507
3508                 /*
3509                  * note the key will change type as we walk through the
3510                  * tree.
3511                  */
3512                 path->leave_spinning = 1;
3513                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3514                                 0, 0);
3515                 if (ret < 0)
3516                         goto out;
3517                 /*
3518                  * First search, if no extent item that starts at offset off was
3519                  * found but the previous item is an extent item, it's possible
3520                  * it might overlap our target range, therefore process it.
3521                  */
3522                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3523                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3524                                               path->slots[0] - 1);
3525                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3526                                 path->slots[0]--;
3527                 }
3528
3529                 nritems = btrfs_header_nritems(path->nodes[0]);
3530 process_slot:
3531                 if (path->slots[0] >= nritems) {
3532                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3533                         if (ret < 0)
3534                                 goto out;
3535                         if (ret > 0)
3536                                 break;
3537                         nritems = btrfs_header_nritems(path->nodes[0]);
3538                 }
3539                 leaf = path->nodes[0];
3540                 slot = path->slots[0];
3541
3542                 btrfs_item_key_to_cpu(leaf, &key, slot);
3543                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3544                     key.objectid != btrfs_ino(BTRFS_I(src)))
3545                         break;
3546
3547                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3548                         struct btrfs_file_extent_item *extent;
3549                         int type;
3550                         u32 size;
3551                         struct btrfs_key new_key;
3552                         u64 disko = 0, diskl = 0;
3553                         u64 datao = 0, datal = 0;
3554                         u8 comp;
3555                         u64 drop_start;
3556
3557                         extent = btrfs_item_ptr(leaf, slot,
3558                                                 struct btrfs_file_extent_item);
3559                         comp = btrfs_file_extent_compression(leaf, extent);
3560                         type = btrfs_file_extent_type(leaf, extent);
3561                         if (type == BTRFS_FILE_EXTENT_REG ||
3562                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3563                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3564                                                                       extent);
3565                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3566                                                                  extent);
3567                                 datao = btrfs_file_extent_offset(leaf, extent);
3568                                 datal = btrfs_file_extent_num_bytes(leaf,
3569                                                                     extent);
3570                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3571                                 /* take upper bound, may be compressed */
3572                                 datal = btrfs_file_extent_ram_bytes(leaf,
3573                                                                     extent);
3574                         }
3575
3576                         /*
3577                          * The first search might have left us at an extent
3578                          * item that ends before our target range's start, can
3579                          * happen if we have holes and NO_HOLES feature enabled.
3580                          */
3581                         if (key.offset + datal <= off) {
3582                                 path->slots[0]++;
3583                                 goto process_slot;
3584                         } else if (key.offset >= off + len) {
3585                                 break;
3586                         }
3587                         next_key_min_offset = key.offset + datal;
3588                         size = btrfs_item_size_nr(leaf, slot);
3589                         read_extent_buffer(leaf, buf,
3590                                            btrfs_item_ptr_offset(leaf, slot),
3591                                            size);
3592
3593                         btrfs_release_path(path);
3594                         path->leave_spinning = 0;
3595
3596                         memcpy(&new_key, &key, sizeof(new_key));
3597                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3598                         if (off <= key.offset)
3599                                 new_key.offset = key.offset + destoff - off;
3600                         else
3601                                 new_key.offset = destoff;
3602
3603                         /*
3604                          * Deal with a hole that doesn't have an extent item
3605                          * that represents it (NO_HOLES feature enabled).
3606                          * This hole is either in the middle of the cloning
3607                          * range or at the beginning (fully overlaps it or
3608                          * partially overlaps it).
3609                          */
3610                         if (new_key.offset != last_dest_end)
3611                                 drop_start = last_dest_end;
3612                         else
3613                                 drop_start = new_key.offset;
3614
3615                         /*
3616                          * 1 - adjusting old extent (we may have to split it)
3617                          * 1 - add new extent
3618                          * 1 - inode update
3619                          */
3620                         trans = btrfs_start_transaction(root, 3);
3621                         if (IS_ERR(trans)) {
3622                                 ret = PTR_ERR(trans);
3623                                 goto out;
3624                         }
3625
3626                         if (type == BTRFS_FILE_EXTENT_REG ||
3627                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3628                                 /*
3629                                  *    a  | --- range to clone ---|  b
3630                                  * | ------------- extent ------------- |
3631                                  */
3632
3633                                 /* subtract range b */
3634                                 if (key.offset + datal > off + len)
3635                                         datal = off + len - key.offset;
3636
3637                                 /* subtract range a */
3638                                 if (off > key.offset) {
3639                                         datao += off - key.offset;
3640                                         datal -= off - key.offset;
3641                                 }
3642
3643                                 ret = btrfs_drop_extents(trans, root, inode,
3644                                                          drop_start,
3645                                                          new_key.offset + datal,
3646                                                          1);
3647                                 if (ret) {
3648                                         if (ret != -EOPNOTSUPP)
3649                                                 btrfs_abort_transaction(trans,
3650                                                                         ret);
3651                                         btrfs_end_transaction(trans);
3652                                         goto out;
3653                                 }
3654
3655                                 ret = btrfs_insert_empty_item(trans, root, path,
3656                                                               &new_key, size);
3657                                 if (ret) {
3658                                         btrfs_abort_transaction(trans, ret);
3659                                         btrfs_end_transaction(trans);
3660                                         goto out;
3661                                 }
3662
3663                                 leaf = path->nodes[0];
3664                                 slot = path->slots[0];
3665                                 write_extent_buffer(leaf, buf,
3666                                             btrfs_item_ptr_offset(leaf, slot),
3667                                             size);
3668
3669                                 extent = btrfs_item_ptr(leaf, slot,
3670                                                 struct btrfs_file_extent_item);
3671
3672                                 /* disko == 0 means it's a hole */
3673                                 if (!disko)
3674                                         datao = 0;
3675
3676                                 btrfs_set_file_extent_offset(leaf, extent,
3677                                                              datao);
3678                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3679                                                                 datal);
3680
3681                                 if (disko) {
3682                                         inode_add_bytes(inode, datal);
3683                                         ret = btrfs_inc_extent_ref(trans,
3684                                                         fs_info,
3685                                                         disko, diskl, 0,
3686                                                         root->root_key.objectid,
3687                                                         btrfs_ino(BTRFS_I(inode)),
3688                                                         new_key.offset - datao);
3689                                         if (ret) {
3690                                                 btrfs_abort_transaction(trans,
3691                                                                         ret);
3692                                                 btrfs_end_transaction(trans);
3693                                                 goto out;
3694
3695                                         }
3696                                 }
3697                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3698                                 u64 skip = 0;
3699                                 u64 trim = 0;
3700
3701                                 if (off > key.offset) {
3702                                         skip = off - key.offset;
3703                                         new_key.offset += skip;
3704                                 }
3705
3706                                 if (key.offset + datal > off + len)
3707                                         trim = key.offset + datal - (off + len);
3708
3709                                 if (comp && (skip || trim)) {
3710                                         ret = -EINVAL;
3711                                         btrfs_end_transaction(trans);
3712                                         goto out;
3713                                 }
3714                                 size -= skip + trim;
3715                                 datal -= skip + trim;
3716
3717                                 ret = clone_copy_inline_extent(inode,
3718                                                                trans, path,
3719                                                                &new_key,
3720                                                                drop_start,
3721                                                                datal,
3722                                                                skip, size, buf);
3723                                 if (ret) {
3724                                         if (ret != -EOPNOTSUPP)
3725                                                 btrfs_abort_transaction(trans,
3726                                                                         ret);
3727                                         btrfs_end_transaction(trans);
3728                                         goto out;
3729                                 }
3730                                 leaf = path->nodes[0];
3731                                 slot = path->slots[0];
3732                         }
3733
3734                         /* If we have an implicit hole (NO_HOLES feature). */
3735                         if (drop_start < new_key.offset)
3736                                 clone_update_extent_map(BTRFS_I(inode), trans,
3737                                                 NULL, drop_start,
3738                                                 new_key.offset - drop_start);
3739
3740                         clone_update_extent_map(BTRFS_I(inode), trans,
3741                                         path, 0, 0);
3742
3743                         btrfs_mark_buffer_dirty(leaf);
3744                         btrfs_release_path(path);
3745
3746                         last_dest_end = ALIGN(new_key.offset + datal,
3747                                               fs_info->sectorsize);
3748                         ret = clone_finish_inode_update(trans, inode,
3749                                                         last_dest_end,
3750                                                         destoff, olen,
3751                                                         no_time_update);
3752                         if (ret)
3753                                 goto out;
3754                         if (new_key.offset + datal >= destoff + len)
3755                                 break;
3756                 }
3757                 btrfs_release_path(path);
3758                 key.offset = next_key_min_offset;
3759
3760                 if (fatal_signal_pending(current)) {
3761                         ret = -EINTR;
3762                         goto out;
3763                 }
3764         }
3765         ret = 0;
3766
3767         if (last_dest_end < destoff + len) {
3768                 /*
3769                  * We have an implicit hole (NO_HOLES feature is enabled) that
3770                  * fully or partially overlaps our cloning range at its end.
3771                  */
3772                 btrfs_release_path(path);
3773
3774                 /*
3775                  * 1 - remove extent(s)
3776                  * 1 - inode update
3777                  */
3778                 trans = btrfs_start_transaction(root, 2);
3779                 if (IS_ERR(trans)) {
3780                         ret = PTR_ERR(trans);
3781                         goto out;
3782                 }
3783                 ret = btrfs_drop_extents(trans, root, inode,
3784                                          last_dest_end, destoff + len, 1);
3785                 if (ret) {
3786                         if (ret != -EOPNOTSUPP)
3787                                 btrfs_abort_transaction(trans, ret);
3788                         btrfs_end_transaction(trans);
3789                         goto out;
3790                 }
3791                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3792                                 last_dest_end,
3793                                 destoff + len - last_dest_end);
3794                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3795                                                 destoff, olen, no_time_update);
3796         }
3797
3798 out:
3799         btrfs_free_path(path);
3800         kvfree(buf);
3801         return ret;
3802 }
3803
3804 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3805                                         u64 off, u64 olen, u64 destoff)
3806 {
3807         struct inode *inode = file_inode(file);
3808         struct inode *src = file_inode(file_src);
3809         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3810         struct btrfs_root *root = BTRFS_I(inode)->root;
3811         int ret;
3812         u64 len = olen;
3813         u64 bs = fs_info->sb->s_blocksize;
3814         int same_inode = src == inode;
3815
3816         /*
3817          * TODO:
3818          * - split compressed inline extents.  annoying: we need to
3819          *   decompress into destination's address_space (the file offset
3820          *   may change, so source mapping won't do), then recompress (or
3821          *   otherwise reinsert) a subrange.
3822          *
3823          * - split destination inode's inline extents.  The inline extents can
3824          *   be either compressed or non-compressed.
3825          */
3826
3827         if (btrfs_root_readonly(root))
3828                 return -EROFS;
3829
3830         if (file_src->f_path.mnt != file->f_path.mnt ||
3831             src->i_sb != inode->i_sb)
3832                 return -EXDEV;
3833
3834         /* don't make the dst file partly checksummed */
3835         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3836             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3837                 return -EINVAL;
3838
3839         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3840                 return -EISDIR;
3841
3842         if (!same_inode) {
3843                 btrfs_double_inode_lock(src, inode);
3844         } else {
3845                 inode_lock(src);
3846         }
3847
3848         /* determine range to clone */
3849         ret = -EINVAL;
3850         if (off + len > src->i_size || off + len < off)
3851                 goto out_unlock;
3852         if (len == 0)
3853                 olen = len = src->i_size - off;
3854         /* if we extend to eof, continue to block boundary */
3855         if (off + len == src->i_size)
3856                 len = ALIGN(src->i_size, bs) - off;
3857
3858         if (len == 0) {
3859                 ret = 0;
3860                 goto out_unlock;
3861         }
3862
3863         /* verify the end result is block aligned */
3864         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3865             !IS_ALIGNED(destoff, bs))
3866                 goto out_unlock;
3867
3868         /* verify if ranges are overlapped within the same file */
3869         if (same_inode) {
3870                 if (destoff + len > off && destoff < off + len)
3871                         goto out_unlock;
3872         }
3873
3874         if (destoff > inode->i_size) {
3875                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3876                 if (ret)
3877                         goto out_unlock;
3878         }
3879
3880         /*
3881          * Lock the target range too. Right after we replace the file extent
3882          * items in the fs tree (which now point to the cloned data), we might
3883          * have a worker replace them with extent items relative to a write
3884          * operation that was issued before this clone operation (i.e. confront
3885          * with inode.c:btrfs_finish_ordered_io).
3886          */
3887         if (same_inode) {
3888                 u64 lock_start = min_t(u64, off, destoff);
3889                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3890
3891                 ret = lock_extent_range(src, lock_start, lock_len, true);
3892         } else {
3893                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3894                                                true);
3895         }
3896         ASSERT(ret == 0);
3897         if (WARN_ON(ret)) {
3898                 /* ranges in the io trees already unlocked */
3899                 goto out_unlock;
3900         }
3901
3902         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3903
3904         if (same_inode) {
3905                 u64 lock_start = min_t(u64, off, destoff);
3906                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3907
3908                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3909         } else {
3910                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3911         }
3912         /*
3913          * Truncate page cache pages so that future reads will see the cloned
3914          * data immediately and not the previous data.
3915          */
3916         truncate_inode_pages_range(&inode->i_data,
3917                                 round_down(destoff, PAGE_SIZE),
3918                                 round_up(destoff + len, PAGE_SIZE) - 1);
3919 out_unlock:
3920         if (!same_inode)
3921                 btrfs_double_inode_unlock(src, inode);
3922         else
3923                 inode_unlock(src);
3924         return ret;
3925 }
3926
3927 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3928                 struct file *dst_file, loff_t destoff, u64 len)
3929 {
3930         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3931 }
3932
3933 /*
3934  * there are many ways the trans_start and trans_end ioctls can lead
3935  * to deadlocks.  They should only be used by applications that
3936  * basically own the machine, and have a very in depth understanding
3937  * of all the possible deadlocks and enospc problems.
3938  */
3939 static long btrfs_ioctl_trans_start(struct file *file)
3940 {
3941         struct inode *inode = file_inode(file);
3942         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3943         struct btrfs_root *root = BTRFS_I(inode)->root;
3944         struct btrfs_trans_handle *trans;
3945         struct btrfs_file_private *private;
3946         int ret;
3947         static bool warned = false;
3948
3949         ret = -EPERM;
3950         if (!capable(CAP_SYS_ADMIN))
3951                 goto out;
3952
3953         if (!warned) {
3954                 btrfs_warn(fs_info,
3955                         "Userspace transaction mechanism is considered "
3956                         "deprecated and slated to be removed in 4.17. "
3957                         "If you have a valid use case please "
3958                         "speak up on the mailing list");
3959                 WARN_ON(1);
3960                 warned = true;
3961         }
3962
3963         ret = -EINPROGRESS;
3964         private = file->private_data;
3965         if (private && private->trans)
3966                 goto out;
3967         if (!private) {
3968                 private = kzalloc(sizeof(struct btrfs_file_private),
3969                                   GFP_KERNEL);
3970                 if (!private)
3971                         return -ENOMEM;
3972                 file->private_data = private;
3973         }
3974
3975         ret = -EROFS;
3976         if (btrfs_root_readonly(root))
3977                 goto out;
3978
3979         ret = mnt_want_write_file(file);
3980         if (ret)
3981                 goto out;
3982
3983         atomic_inc(&fs_info->open_ioctl_trans);
3984
3985         ret = -ENOMEM;
3986         trans = btrfs_start_ioctl_transaction(root);
3987         if (IS_ERR(trans))
3988                 goto out_drop;
3989
3990         private->trans = trans;
3991         return 0;
3992
3993 out_drop:
3994         atomic_dec(&fs_info->open_ioctl_trans);
3995         mnt_drop_write_file(file);
3996 out:
3997         return ret;
3998 }
3999
4000 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4001 {
4002         struct inode *inode = file_inode(file);
4003         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4004         struct btrfs_root *root = BTRFS_I(inode)->root;
4005         struct btrfs_root *new_root;
4006         struct btrfs_dir_item *di;
4007         struct btrfs_trans_handle *trans;
4008         struct btrfs_path *path;
4009         struct btrfs_key location;
4010         struct btrfs_disk_key disk_key;
4011         u64 objectid = 0;
4012         u64 dir_id;
4013         int ret;
4014
4015         if (!capable(CAP_SYS_ADMIN))
4016                 return -EPERM;
4017
4018         ret = mnt_want_write_file(file);
4019         if (ret)
4020                 return ret;
4021
4022         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4023                 ret = -EFAULT;
4024                 goto out;
4025         }
4026
4027         if (!objectid)
4028                 objectid = BTRFS_FS_TREE_OBJECTID;
4029
4030         location.objectid = objectid;
4031         location.type = BTRFS_ROOT_ITEM_KEY;
4032         location.offset = (u64)-1;
4033
4034         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4035         if (IS_ERR(new_root)) {
4036                 ret = PTR_ERR(new_root);
4037                 goto out;
4038         }
4039         if (!is_fstree(new_root->objectid)) {
4040                 ret = -ENOENT;
4041                 goto out;
4042         }
4043
4044         path = btrfs_alloc_path();
4045         if (!path) {
4046                 ret = -ENOMEM;
4047                 goto out;
4048         }
4049         path->leave_spinning = 1;
4050
4051         trans = btrfs_start_transaction(root, 1);
4052         if (IS_ERR(trans)) {
4053                 btrfs_free_path(path);
4054                 ret = PTR_ERR(trans);
4055                 goto out;
4056         }
4057
4058         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4059         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4060                                    dir_id, "default", 7, 1);
4061         if (IS_ERR_OR_NULL(di)) {
4062                 btrfs_free_path(path);
4063                 btrfs_end_transaction(trans);
4064                 btrfs_err(fs_info,
4065                           "Umm, you don't have the default diritem, this isn't going to work");
4066                 ret = -ENOENT;
4067                 goto out;
4068         }
4069
4070         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4071         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4072         btrfs_mark_buffer_dirty(path->nodes[0]);
4073         btrfs_free_path(path);
4074
4075         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4076         btrfs_end_transaction(trans);
4077 out:
4078         mnt_drop_write_file(file);
4079         return ret;
4080 }
4081
4082 void btrfs_get_block_group_info(struct list_head *groups_list,
4083                                 struct btrfs_ioctl_space_info *space)
4084 {
4085         struct btrfs_block_group_cache *block_group;
4086
4087         space->total_bytes = 0;
4088         space->used_bytes = 0;
4089         space->flags = 0;
4090         list_for_each_entry(block_group, groups_list, list) {
4091                 space->flags = block_group->flags;
4092                 space->total_bytes += block_group->key.offset;
4093                 space->used_bytes +=
4094                         btrfs_block_group_used(&block_group->item);
4095         }
4096 }
4097
4098 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4099                                    void __user *arg)
4100 {
4101         struct btrfs_ioctl_space_args space_args;
4102         struct btrfs_ioctl_space_info space;
4103         struct btrfs_ioctl_space_info *dest;
4104         struct btrfs_ioctl_space_info *dest_orig;
4105         struct btrfs_ioctl_space_info __user *user_dest;
4106         struct btrfs_space_info *info;
4107         static const u64 types[] = {
4108                 BTRFS_BLOCK_GROUP_DATA,
4109                 BTRFS_BLOCK_GROUP_SYSTEM,
4110                 BTRFS_BLOCK_GROUP_METADATA,
4111                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4112         };
4113         int num_types = 4;
4114         int alloc_size;
4115         int ret = 0;
4116         u64 slot_count = 0;
4117         int i, c;
4118
4119         if (copy_from_user(&space_args,
4120                            (struct btrfs_ioctl_space_args __user *)arg,
4121                            sizeof(space_args)))
4122                 return -EFAULT;
4123
4124         for (i = 0; i < num_types; i++) {
4125                 struct btrfs_space_info *tmp;
4126
4127                 info = NULL;
4128                 rcu_read_lock();
4129                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4130                                         list) {
4131                         if (tmp->flags == types[i]) {
4132                                 info = tmp;
4133                                 break;
4134                         }
4135                 }
4136                 rcu_read_unlock();
4137
4138                 if (!info)
4139                         continue;
4140
4141                 down_read(&info->groups_sem);
4142                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4143                         if (!list_empty(&info->block_groups[c]))
4144                                 slot_count++;
4145                 }
4146                 up_read(&info->groups_sem);
4147         }
4148
4149         /*
4150          * Global block reserve, exported as a space_info
4151          */
4152         slot_count++;
4153
4154         /* space_slots == 0 means they are asking for a count */
4155         if (space_args.space_slots == 0) {
4156                 space_args.total_spaces = slot_count;
4157                 goto out;
4158         }
4159
4160         slot_count = min_t(u64, space_args.space_slots, slot_count);
4161
4162         alloc_size = sizeof(*dest) * slot_count;
4163
4164         /* we generally have at most 6 or so space infos, one for each raid
4165          * level.  So, a whole page should be more than enough for everyone
4166          */
4167         if (alloc_size > PAGE_SIZE)
4168                 return -ENOMEM;
4169
4170         space_args.total_spaces = 0;
4171         dest = kmalloc(alloc_size, GFP_KERNEL);
4172         if (!dest)
4173                 return -ENOMEM;
4174         dest_orig = dest;
4175
4176         /* now we have a buffer to copy into */
4177         for (i = 0; i < num_types; i++) {
4178                 struct btrfs_space_info *tmp;
4179
4180                 if (!slot_count)
4181                         break;
4182
4183                 info = NULL;
4184                 rcu_read_lock();
4185                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4186                                         list) {
4187                         if (tmp->flags == types[i]) {
4188                                 info = tmp;
4189                                 break;
4190                         }
4191                 }
4192                 rcu_read_unlock();
4193
4194                 if (!info)
4195                         continue;
4196                 down_read(&info->groups_sem);
4197                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4198                         if (!list_empty(&info->block_groups[c])) {
4199                                 btrfs_get_block_group_info(
4200                                         &info->block_groups[c], &space);
4201                                 memcpy(dest, &space, sizeof(space));
4202                                 dest++;
4203                                 space_args.total_spaces++;
4204                                 slot_count--;
4205                         }
4206                         if (!slot_count)
4207                                 break;
4208                 }
4209                 up_read(&info->groups_sem);
4210         }
4211
4212         /*
4213          * Add global block reserve
4214          */
4215         if (slot_count) {
4216                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4217
4218                 spin_lock(&block_rsv->lock);
4219                 space.total_bytes = block_rsv->size;
4220                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4221                 spin_unlock(&block_rsv->lock);
4222                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4223                 memcpy(dest, &space, sizeof(space));
4224                 space_args.total_spaces++;
4225         }
4226
4227         user_dest = (struct btrfs_ioctl_space_info __user *)
4228                 (arg + sizeof(struct btrfs_ioctl_space_args));
4229
4230         if (copy_to_user(user_dest, dest_orig, alloc_size))
4231                 ret = -EFAULT;
4232
4233         kfree(dest_orig);
4234 out:
4235         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4236                 ret = -EFAULT;
4237
4238         return ret;
4239 }
4240
4241 /*
4242  * there are many ways the trans_start and trans_end ioctls can lead
4243  * to deadlocks.  They should only be used by applications that
4244  * basically own the machine, and have a very in depth understanding
4245  * of all the possible deadlocks and enospc problems.
4246  */
4247 long btrfs_ioctl_trans_end(struct file *file)
4248 {
4249         struct inode *inode = file_inode(file);
4250         struct btrfs_root *root = BTRFS_I(inode)->root;
4251         struct btrfs_file_private *private = file->private_data;
4252
4253         if (!private || !private->trans)
4254                 return -EINVAL;
4255
4256         btrfs_end_transaction(private->trans);
4257         private->trans = NULL;
4258
4259         atomic_dec(&root->fs_info->open_ioctl_trans);
4260
4261         mnt_drop_write_file(file);
4262         return 0;
4263 }
4264
4265 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4266                                             void __user *argp)
4267 {
4268         struct btrfs_trans_handle *trans;
4269         u64 transid;
4270         int ret;
4271
4272         trans = btrfs_attach_transaction_barrier(root);
4273         if (IS_ERR(trans)) {
4274                 if (PTR_ERR(trans) != -ENOENT)
4275                         return PTR_ERR(trans);
4276
4277                 /* No running transaction, don't bother */
4278                 transid = root->fs_info->last_trans_committed;
4279                 goto out;
4280         }
4281         transid = trans->transid;
4282         ret = btrfs_commit_transaction_async(trans, 0);
4283         if (ret) {
4284                 btrfs_end_transaction(trans);
4285                 return ret;
4286         }
4287 out:
4288         if (argp)
4289                 if (copy_to_user(argp, &transid, sizeof(transid)))
4290                         return -EFAULT;
4291         return 0;
4292 }
4293
4294 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4295                                            void __user *argp)
4296 {
4297         u64 transid;
4298
4299         if (argp) {
4300                 if (copy_from_user(&transid, argp, sizeof(transid)))
4301                         return -EFAULT;
4302         } else {
4303                 transid = 0;  /* current trans */
4304         }
4305         return btrfs_wait_for_commit(fs_info, transid);
4306 }
4307
4308 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4309 {
4310         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4311         struct btrfs_ioctl_scrub_args *sa;
4312         int ret;
4313
4314         if (!capable(CAP_SYS_ADMIN))
4315                 return -EPERM;
4316
4317         sa = memdup_user(arg, sizeof(*sa));
4318         if (IS_ERR(sa))
4319                 return PTR_ERR(sa);
4320
4321         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4322                 ret = mnt_want_write_file(file);
4323                 if (ret)
4324                         goto out;
4325         }
4326
4327         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4328                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4329                               0);
4330
4331         if (copy_to_user(arg, sa, sizeof(*sa)))
4332                 ret = -EFAULT;
4333
4334         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4335                 mnt_drop_write_file(file);
4336 out:
4337         kfree(sa);
4338         return ret;
4339 }
4340
4341 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4342 {
4343         if (!capable(CAP_SYS_ADMIN))
4344                 return -EPERM;
4345
4346         return btrfs_scrub_cancel(fs_info);
4347 }
4348
4349 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4350                                        void __user *arg)
4351 {
4352         struct btrfs_ioctl_scrub_args *sa;
4353         int ret;
4354
4355         if (!capable(CAP_SYS_ADMIN))
4356                 return -EPERM;
4357
4358         sa = memdup_user(arg, sizeof(*sa));
4359         if (IS_ERR(sa))
4360                 return PTR_ERR(sa);
4361
4362         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4363
4364         if (copy_to_user(arg, sa, sizeof(*sa)))
4365                 ret = -EFAULT;
4366
4367         kfree(sa);
4368         return ret;
4369 }
4370
4371 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4372                                       void __user *arg)
4373 {
4374         struct btrfs_ioctl_get_dev_stats *sa;
4375         int ret;
4376
4377         sa = memdup_user(arg, sizeof(*sa));
4378         if (IS_ERR(sa))
4379                 return PTR_ERR(sa);
4380
4381         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4382                 kfree(sa);
4383                 return -EPERM;
4384         }
4385
4386         ret = btrfs_get_dev_stats(fs_info, sa);
4387
4388         if (copy_to_user(arg, sa, sizeof(*sa)))
4389                 ret = -EFAULT;
4390
4391         kfree(sa);
4392         return ret;
4393 }
4394
4395 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4396                                     void __user *arg)
4397 {
4398         struct btrfs_ioctl_dev_replace_args *p;
4399         int ret;
4400
4401         if (!capable(CAP_SYS_ADMIN))
4402                 return -EPERM;
4403
4404         p = memdup_user(arg, sizeof(*p));
4405         if (IS_ERR(p))
4406                 return PTR_ERR(p);
4407
4408         switch (p->cmd) {
4409         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4410                 if (sb_rdonly(fs_info->sb)) {
4411                         ret = -EROFS;
4412                         goto out;
4413                 }
4414                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4415                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4416                 } else {
4417                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4418                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4419                 }
4420                 break;
4421         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4422                 btrfs_dev_replace_status(fs_info, p);
4423                 ret = 0;
4424                 break;
4425         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4426                 ret = btrfs_dev_replace_cancel(fs_info, p);
4427                 break;
4428         default:
4429                 ret = -EINVAL;
4430                 break;
4431         }
4432
4433         if (copy_to_user(arg, p, sizeof(*p)))
4434                 ret = -EFAULT;
4435 out:
4436         kfree(p);
4437         return ret;
4438 }
4439
4440 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4441 {
4442         int ret = 0;
4443         int i;
4444         u64 rel_ptr;
4445         int size;
4446         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4447         struct inode_fs_paths *ipath = NULL;
4448         struct btrfs_path *path;
4449
4450         if (!capable(CAP_DAC_READ_SEARCH))
4451                 return -EPERM;
4452
4453         path = btrfs_alloc_path();
4454         if (!path) {
4455                 ret = -ENOMEM;
4456                 goto out;
4457         }
4458
4459         ipa = memdup_user(arg, sizeof(*ipa));
4460         if (IS_ERR(ipa)) {
4461                 ret = PTR_ERR(ipa);
4462                 ipa = NULL;
4463                 goto out;
4464         }
4465
4466         size = min_t(u32, ipa->size, 4096);
4467         ipath = init_ipath(size, root, path);
4468         if (IS_ERR(ipath)) {
4469                 ret = PTR_ERR(ipath);
4470                 ipath = NULL;
4471                 goto out;
4472         }
4473
4474         ret = paths_from_inode(ipa->inum, ipath);
4475         if (ret < 0)
4476                 goto out;
4477
4478         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4479                 rel_ptr = ipath->fspath->val[i] -
4480                           (u64)(unsigned long)ipath->fspath->val;
4481                 ipath->fspath->val[i] = rel_ptr;
4482         }
4483
4484         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4485                            ipath->fspath, size);
4486         if (ret) {
4487                 ret = -EFAULT;
4488                 goto out;
4489         }
4490
4491 out:
4492         btrfs_free_path(path);
4493         free_ipath(ipath);
4494         kfree(ipa);
4495
4496         return ret;
4497 }
4498
4499 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4500 {
4501         struct btrfs_data_container *inodes = ctx;
4502         const size_t c = 3 * sizeof(u64);
4503
4504         if (inodes->bytes_left >= c) {
4505                 inodes->bytes_left -= c;
4506                 inodes->val[inodes->elem_cnt] = inum;
4507                 inodes->val[inodes->elem_cnt + 1] = offset;
4508                 inodes->val[inodes->elem_cnt + 2] = root;
4509                 inodes->elem_cnt += 3;
4510         } else {
4511                 inodes->bytes_missing += c - inodes->bytes_left;
4512                 inodes->bytes_left = 0;
4513                 inodes->elem_missed += 3;
4514         }
4515
4516         return 0;
4517 }
4518
4519 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4520                                         void __user *arg)
4521 {
4522         int ret = 0;
4523         int size;
4524         struct btrfs_ioctl_logical_ino_args *loi;
4525         struct btrfs_data_container *inodes = NULL;
4526         struct btrfs_path *path = NULL;
4527
4528         if (!capable(CAP_SYS_ADMIN))
4529                 return -EPERM;
4530
4531         loi = memdup_user(arg, sizeof(*loi));
4532         if (IS_ERR(loi))
4533                 return PTR_ERR(loi);
4534
4535         path = btrfs_alloc_path();
4536         if (!path) {
4537                 ret = -ENOMEM;
4538                 goto out;
4539         }
4540
4541         size = min_t(u32, loi->size, SZ_64K);
4542         inodes = init_data_container(size);
4543         if (IS_ERR(inodes)) {
4544                 ret = PTR_ERR(inodes);
4545                 inodes = NULL;
4546                 goto out;
4547         }
4548
4549         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4550                                           build_ino_list, inodes);
4551         if (ret == -EINVAL)
4552                 ret = -ENOENT;
4553         if (ret < 0)
4554                 goto out;
4555
4556         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4557                            size);
4558         if (ret)
4559                 ret = -EFAULT;
4560
4561 out:
4562         btrfs_free_path(path);
4563         kvfree(inodes);
4564         kfree(loi);
4565
4566         return ret;
4567 }
4568
4569 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4570                                struct btrfs_ioctl_balance_args *bargs)
4571 {
4572         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4573
4574         bargs->flags = bctl->flags;
4575
4576         if (atomic_read(&fs_info->balance_running))
4577                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4578         if (atomic_read(&fs_info->balance_pause_req))
4579                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4580         if (atomic_read(&fs_info->balance_cancel_req))
4581                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4582
4583         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4584         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4585         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4586
4587         if (lock) {
4588                 spin_lock(&fs_info->balance_lock);
4589                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4590                 spin_unlock(&fs_info->balance_lock);
4591         } else {
4592                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4593         }
4594 }
4595
4596 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4597 {
4598         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4599         struct btrfs_fs_info *fs_info = root->fs_info;
4600         struct btrfs_ioctl_balance_args *bargs;
4601         struct btrfs_balance_control *bctl;
4602         bool need_unlock; /* for mut. excl. ops lock */
4603         int ret;
4604
4605         if (!capable(CAP_SYS_ADMIN))
4606                 return -EPERM;
4607
4608         ret = mnt_want_write_file(file);
4609         if (ret)
4610                 return ret;
4611
4612 again:
4613         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4614                 mutex_lock(&fs_info->volume_mutex);
4615                 mutex_lock(&fs_info->balance_mutex);
4616                 need_unlock = true;
4617                 goto locked;
4618         }
4619
4620         /*
4621          * mut. excl. ops lock is locked.  Three possibilities:
4622          *   (1) some other op is running
4623          *   (2) balance is running
4624          *   (3) balance is paused -- special case (think resume)
4625          */
4626         mutex_lock(&fs_info->balance_mutex);
4627         if (fs_info->balance_ctl) {
4628                 /* this is either (2) or (3) */
4629                 if (!atomic_read(&fs_info->balance_running)) {
4630                         mutex_unlock(&fs_info->balance_mutex);
4631                         if (!mutex_trylock(&fs_info->volume_mutex))
4632                                 goto again;
4633                         mutex_lock(&fs_info->balance_mutex);
4634
4635                         if (fs_info->balance_ctl &&
4636                             !atomic_read(&fs_info->balance_running)) {
4637                                 /* this is (3) */
4638                                 need_unlock = false;
4639                                 goto locked;
4640                         }
4641
4642                         mutex_unlock(&fs_info->balance_mutex);
4643                         mutex_unlock(&fs_info->volume_mutex);
4644                         goto again;
4645                 } else {
4646                         /* this is (2) */
4647                         mutex_unlock(&fs_info->balance_mutex);
4648                         ret = -EINPROGRESS;
4649                         goto out;
4650                 }
4651         } else {
4652                 /* this is (1) */
4653                 mutex_unlock(&fs_info->balance_mutex);
4654                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4655                 goto out;
4656         }
4657
4658 locked:
4659         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4660
4661         if (arg) {
4662                 bargs = memdup_user(arg, sizeof(*bargs));
4663                 if (IS_ERR(bargs)) {
4664                         ret = PTR_ERR(bargs);
4665                         goto out_unlock;
4666                 }
4667
4668                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4669                         if (!fs_info->balance_ctl) {
4670                                 ret = -ENOTCONN;
4671                                 goto out_bargs;
4672                         }
4673
4674                         bctl = fs_info->balance_ctl;
4675                         spin_lock(&fs_info->balance_lock);
4676                         bctl->flags |= BTRFS_BALANCE_RESUME;
4677                         spin_unlock(&fs_info->balance_lock);
4678
4679                         goto do_balance;
4680                 }
4681         } else {
4682                 bargs = NULL;
4683         }
4684
4685         if (fs_info->balance_ctl) {
4686                 ret = -EINPROGRESS;
4687                 goto out_bargs;
4688         }
4689
4690         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4691         if (!bctl) {
4692                 ret = -ENOMEM;
4693                 goto out_bargs;
4694         }
4695
4696         bctl->fs_info = fs_info;
4697         if (arg) {
4698                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4699                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4700                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4701
4702                 bctl->flags = bargs->flags;
4703         } else {
4704                 /* balance everything - no filters */
4705                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4706         }
4707
4708         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4709                 ret = -EINVAL;
4710                 goto out_bctl;
4711         }
4712
4713 do_balance:
4714         /*
4715          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4716          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4717          * or, if restriper was paused all the way until unmount, in
4718          * free_fs_info.  The flag is cleared in __cancel_balance.
4719          */
4720         need_unlock = false;
4721
4722         ret = btrfs_balance(bctl, bargs);
4723         bctl = NULL;
4724
4725         if (arg) {
4726                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4727                         ret = -EFAULT;
4728         }
4729
4730 out_bctl:
4731         kfree(bctl);
4732 out_bargs:
4733         kfree(bargs);
4734 out_unlock:
4735         mutex_unlock(&fs_info->balance_mutex);
4736         mutex_unlock(&fs_info->volume_mutex);
4737         if (need_unlock)
4738                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4739 out:
4740         mnt_drop_write_file(file);
4741         return ret;
4742 }
4743
4744 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4745 {
4746         if (!capable(CAP_SYS_ADMIN))
4747                 return -EPERM;
4748
4749         switch (cmd) {
4750         case BTRFS_BALANCE_CTL_PAUSE:
4751                 return btrfs_pause_balance(fs_info);
4752         case BTRFS_BALANCE_CTL_CANCEL:
4753                 return btrfs_cancel_balance(fs_info);
4754         }
4755
4756         return -EINVAL;
4757 }
4758
4759 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4760                                          void __user *arg)
4761 {
4762         struct btrfs_ioctl_balance_args *bargs;
4763         int ret = 0;
4764
4765         if (!capable(CAP_SYS_ADMIN))
4766                 return -EPERM;
4767
4768         mutex_lock(&fs_info->balance_mutex);
4769         if (!fs_info->balance_ctl) {
4770                 ret = -ENOTCONN;
4771                 goto out;
4772         }
4773
4774         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4775         if (!bargs) {
4776                 ret = -ENOMEM;
4777                 goto out;
4778         }
4779
4780         update_ioctl_balance_args(fs_info, 1, bargs);
4781
4782         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4783                 ret = -EFAULT;
4784
4785         kfree(bargs);
4786 out:
4787         mutex_unlock(&fs_info->balance_mutex);
4788         return ret;
4789 }
4790
4791 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4792 {
4793         struct inode *inode = file_inode(file);
4794         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4795         struct btrfs_ioctl_quota_ctl_args *sa;
4796         struct btrfs_trans_handle *trans = NULL;
4797         int ret;
4798         int err;
4799
4800         if (!capable(CAP_SYS_ADMIN))
4801                 return -EPERM;
4802
4803         ret = mnt_want_write_file(file);
4804         if (ret)
4805                 return ret;
4806
4807         sa = memdup_user(arg, sizeof(*sa));
4808         if (IS_ERR(sa)) {
4809                 ret = PTR_ERR(sa);
4810                 goto drop_write;
4811         }
4812
4813         down_write(&fs_info->subvol_sem);
4814         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4815         if (IS_ERR(trans)) {
4816                 ret = PTR_ERR(trans);
4817                 goto out;
4818         }
4819
4820         switch (sa->cmd) {
4821         case BTRFS_QUOTA_CTL_ENABLE:
4822                 ret = btrfs_quota_enable(trans, fs_info);
4823                 break;
4824         case BTRFS_QUOTA_CTL_DISABLE:
4825                 ret = btrfs_quota_disable(trans, fs_info);
4826                 break;
4827         default:
4828                 ret = -EINVAL;
4829                 break;
4830         }
4831
4832         err = btrfs_commit_transaction(trans);
4833         if (err && !ret)
4834                 ret = err;
4835 out:
4836         kfree(sa);
4837         up_write(&fs_info->subvol_sem);
4838 drop_write:
4839         mnt_drop_write_file(file);
4840         return ret;
4841 }
4842
4843 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4844 {
4845         struct inode *inode = file_inode(file);
4846         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4847         struct btrfs_root *root = BTRFS_I(inode)->root;
4848         struct btrfs_ioctl_qgroup_assign_args *sa;
4849         struct btrfs_trans_handle *trans;
4850         int ret;
4851         int err;
4852
4853         if (!capable(CAP_SYS_ADMIN))
4854                 return -EPERM;
4855
4856         ret = mnt_want_write_file(file);
4857         if (ret)
4858                 return ret;
4859
4860         sa = memdup_user(arg, sizeof(*sa));
4861         if (IS_ERR(sa)) {
4862                 ret = PTR_ERR(sa);
4863                 goto drop_write;
4864         }
4865
4866         trans = btrfs_join_transaction(root);
4867         if (IS_ERR(trans)) {
4868                 ret = PTR_ERR(trans);
4869                 goto out;
4870         }
4871
4872         if (sa->assign) {
4873                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4874                                                 sa->src, sa->dst);
4875         } else {
4876                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4877                                                 sa->src, sa->dst);
4878         }
4879
4880         /* update qgroup status and info */
4881         err = btrfs_run_qgroups(trans, fs_info);
4882         if (err < 0)
4883                 btrfs_handle_fs_error(fs_info, err,
4884                                       "failed to update qgroup status and info");
4885         err = btrfs_end_transaction(trans);
4886         if (err && !ret)
4887                 ret = err;
4888
4889 out:
4890         kfree(sa);
4891 drop_write:
4892         mnt_drop_write_file(file);
4893         return ret;
4894 }
4895
4896 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4897 {
4898         struct inode *inode = file_inode(file);
4899         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4900         struct btrfs_root *root = BTRFS_I(inode)->root;
4901         struct btrfs_ioctl_qgroup_create_args *sa;
4902         struct btrfs_trans_handle *trans;
4903         int ret;
4904         int err;
4905
4906         if (!capable(CAP_SYS_ADMIN))
4907                 return -EPERM;
4908
4909         ret = mnt_want_write_file(file);
4910         if (ret)
4911                 return ret;
4912
4913         sa = memdup_user(arg, sizeof(*sa));
4914         if (IS_ERR(sa)) {
4915                 ret = PTR_ERR(sa);
4916                 goto drop_write;
4917         }
4918
4919         if (!sa->qgroupid) {
4920                 ret = -EINVAL;
4921                 goto out;
4922         }
4923
4924         trans = btrfs_join_transaction(root);
4925         if (IS_ERR(trans)) {
4926                 ret = PTR_ERR(trans);
4927                 goto out;
4928         }
4929
4930         if (sa->create) {
4931                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4932         } else {
4933                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4934         }
4935
4936         err = btrfs_end_transaction(trans);
4937         if (err && !ret)
4938                 ret = err;
4939
4940 out:
4941         kfree(sa);
4942 drop_write:
4943         mnt_drop_write_file(file);
4944         return ret;
4945 }
4946
4947 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4948 {
4949         struct inode *inode = file_inode(file);
4950         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4951         struct btrfs_root *root = BTRFS_I(inode)->root;
4952         struct btrfs_ioctl_qgroup_limit_args *sa;
4953         struct btrfs_trans_handle *trans;
4954         int ret;
4955         int err;
4956         u64 qgroupid;
4957
4958         if (!capable(CAP_SYS_ADMIN))
4959                 return -EPERM;
4960
4961         ret = mnt_want_write_file(file);
4962         if (ret)
4963                 return ret;
4964
4965         sa = memdup_user(arg, sizeof(*sa));
4966         if (IS_ERR(sa)) {
4967                 ret = PTR_ERR(sa);
4968                 goto drop_write;
4969         }
4970
4971         trans = btrfs_join_transaction(root);
4972         if (IS_ERR(trans)) {
4973                 ret = PTR_ERR(trans);
4974                 goto out;
4975         }
4976
4977         qgroupid = sa->qgroupid;
4978         if (!qgroupid) {
4979                 /* take the current subvol as qgroup */
4980                 qgroupid = root->root_key.objectid;
4981         }
4982
4983         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
4984
4985         err = btrfs_end_transaction(trans);
4986         if (err && !ret)
4987                 ret = err;
4988
4989 out:
4990         kfree(sa);
4991 drop_write:
4992         mnt_drop_write_file(file);
4993         return ret;
4994 }
4995
4996 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4997 {
4998         struct inode *inode = file_inode(file);
4999         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5000         struct btrfs_ioctl_quota_rescan_args *qsa;
5001         int ret;
5002
5003         if (!capable(CAP_SYS_ADMIN))
5004                 return -EPERM;
5005
5006         ret = mnt_want_write_file(file);
5007         if (ret)
5008                 return ret;
5009
5010         qsa = memdup_user(arg, sizeof(*qsa));
5011         if (IS_ERR(qsa)) {
5012                 ret = PTR_ERR(qsa);
5013                 goto drop_write;
5014         }
5015
5016         if (qsa->flags) {
5017                 ret = -EINVAL;
5018                 goto out;
5019         }
5020
5021         ret = btrfs_qgroup_rescan(fs_info);
5022
5023 out:
5024         kfree(qsa);
5025 drop_write:
5026         mnt_drop_write_file(file);
5027         return ret;
5028 }
5029
5030 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5031 {
5032         struct inode *inode = file_inode(file);
5033         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5034         struct btrfs_ioctl_quota_rescan_args *qsa;
5035         int ret = 0;
5036
5037         if (!capable(CAP_SYS_ADMIN))
5038                 return -EPERM;
5039
5040         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5041         if (!qsa)
5042                 return -ENOMEM;
5043
5044         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5045                 qsa->flags = 1;
5046                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5047         }
5048
5049         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5050                 ret = -EFAULT;
5051
5052         kfree(qsa);
5053         return ret;
5054 }
5055
5056 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5057 {
5058         struct inode *inode = file_inode(file);
5059         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5060
5061         if (!capable(CAP_SYS_ADMIN))
5062                 return -EPERM;
5063
5064         return btrfs_qgroup_wait_for_completion(fs_info, true);
5065 }
5066
5067 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5068                                             struct btrfs_ioctl_received_subvol_args *sa)
5069 {
5070         struct inode *inode = file_inode(file);
5071         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5072         struct btrfs_root *root = BTRFS_I(inode)->root;
5073         struct btrfs_root_item *root_item = &root->root_item;
5074         struct btrfs_trans_handle *trans;
5075         struct timespec ct = current_time(inode);
5076         int ret = 0;
5077         int received_uuid_changed;
5078
5079         if (!inode_owner_or_capable(inode))
5080                 return -EPERM;
5081
5082         ret = mnt_want_write_file(file);
5083         if (ret < 0)
5084                 return ret;
5085
5086         down_write(&fs_info->subvol_sem);
5087
5088         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5089                 ret = -EINVAL;
5090                 goto out;
5091         }
5092
5093         if (btrfs_root_readonly(root)) {
5094                 ret = -EROFS;
5095                 goto out;
5096         }
5097
5098         /*
5099          * 1 - root item
5100          * 2 - uuid items (received uuid + subvol uuid)
5101          */
5102         trans = btrfs_start_transaction(root, 3);
5103         if (IS_ERR(trans)) {
5104                 ret = PTR_ERR(trans);
5105                 trans = NULL;
5106                 goto out;
5107         }
5108
5109         sa->rtransid = trans->transid;
5110         sa->rtime.sec = ct.tv_sec;
5111         sa->rtime.nsec = ct.tv_nsec;
5112
5113         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5114                                        BTRFS_UUID_SIZE);
5115         if (received_uuid_changed &&
5116             !btrfs_is_empty_uuid(root_item->received_uuid))
5117                 btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5118                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5119                                     root->root_key.objectid);
5120         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5121         btrfs_set_root_stransid(root_item, sa->stransid);
5122         btrfs_set_root_rtransid(root_item, sa->rtransid);
5123         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5124         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5125         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5126         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5127
5128         ret = btrfs_update_root(trans, fs_info->tree_root,
5129                                 &root->root_key, &root->root_item);
5130         if (ret < 0) {
5131                 btrfs_end_transaction(trans);
5132                 goto out;
5133         }
5134         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5135                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5136                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5137                                           root->root_key.objectid);
5138                 if (ret < 0 && ret != -EEXIST) {
5139                         btrfs_abort_transaction(trans, ret);
5140                         btrfs_end_transaction(trans);
5141                         goto out;
5142                 }
5143         }
5144         ret = btrfs_commit_transaction(trans);
5145 out:
5146         up_write(&fs_info->subvol_sem);
5147         mnt_drop_write_file(file);
5148         return ret;
5149 }
5150
5151 #ifdef CONFIG_64BIT
5152 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5153                                                 void __user *arg)
5154 {
5155         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5156         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5157         int ret = 0;
5158
5159         args32 = memdup_user(arg, sizeof(*args32));
5160         if (IS_ERR(args32))
5161                 return PTR_ERR(args32);
5162
5163         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5164         if (!args64) {
5165                 ret = -ENOMEM;
5166                 goto out;
5167         }
5168
5169         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5170         args64->stransid = args32->stransid;
5171         args64->rtransid = args32->rtransid;
5172         args64->stime.sec = args32->stime.sec;
5173         args64->stime.nsec = args32->stime.nsec;
5174         args64->rtime.sec = args32->rtime.sec;
5175         args64->rtime.nsec = args32->rtime.nsec;
5176         args64->flags = args32->flags;
5177
5178         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5179         if (ret)
5180                 goto out;
5181
5182         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5183         args32->stransid = args64->stransid;
5184         args32->rtransid = args64->rtransid;
5185         args32->stime.sec = args64->stime.sec;
5186         args32->stime.nsec = args64->stime.nsec;
5187         args32->rtime.sec = args64->rtime.sec;
5188         args32->rtime.nsec = args64->rtime.nsec;
5189         args32->flags = args64->flags;
5190
5191         ret = copy_to_user(arg, args32, sizeof(*args32));
5192         if (ret)
5193                 ret = -EFAULT;
5194
5195 out:
5196         kfree(args32);
5197         kfree(args64);
5198         return ret;
5199 }
5200 #endif
5201
5202 static long btrfs_ioctl_set_received_subvol(struct file *file,
5203                                             void __user *arg)
5204 {
5205         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5206         int ret = 0;
5207
5208         sa = memdup_user(arg, sizeof(*sa));
5209         if (IS_ERR(sa))
5210                 return PTR_ERR(sa);
5211
5212         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5213
5214         if (ret)
5215                 goto out;
5216
5217         ret = copy_to_user(arg, sa, sizeof(*sa));
5218         if (ret)
5219                 ret = -EFAULT;
5220
5221 out:
5222         kfree(sa);
5223         return ret;
5224 }
5225
5226 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5227 {
5228         struct inode *inode = file_inode(file);
5229         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5230         size_t len;
5231         int ret;
5232         char label[BTRFS_LABEL_SIZE];
5233
5234         spin_lock(&fs_info->super_lock);
5235         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5236         spin_unlock(&fs_info->super_lock);
5237
5238         len = strnlen(label, BTRFS_LABEL_SIZE);
5239
5240         if (len == BTRFS_LABEL_SIZE) {
5241                 btrfs_warn(fs_info,
5242                            "label is too long, return the first %zu bytes",
5243                            --len);
5244         }
5245
5246         ret = copy_to_user(arg, label, len);
5247
5248         return ret ? -EFAULT : 0;
5249 }
5250
5251 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5252 {
5253         struct inode *inode = file_inode(file);
5254         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5255         struct btrfs_root *root = BTRFS_I(inode)->root;
5256         struct btrfs_super_block *super_block = fs_info->super_copy;
5257         struct btrfs_trans_handle *trans;
5258         char label[BTRFS_LABEL_SIZE];
5259         int ret;
5260
5261         if (!capable(CAP_SYS_ADMIN))
5262                 return -EPERM;
5263
5264         if (copy_from_user(label, arg, sizeof(label)))
5265                 return -EFAULT;
5266
5267         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5268                 btrfs_err(fs_info,
5269                           "unable to set label with more than %d bytes",
5270                           BTRFS_LABEL_SIZE - 1);
5271                 return -EINVAL;
5272         }
5273
5274         ret = mnt_want_write_file(file);
5275         if (ret)
5276                 return ret;
5277
5278         trans = btrfs_start_transaction(root, 0);
5279         if (IS_ERR(trans)) {
5280                 ret = PTR_ERR(trans);
5281                 goto out_unlock;
5282         }
5283
5284         spin_lock(&fs_info->super_lock);
5285         strcpy(super_block->label, label);
5286         spin_unlock(&fs_info->super_lock);
5287         ret = btrfs_commit_transaction(trans);
5288
5289 out_unlock:
5290         mnt_drop_write_file(file);
5291         return ret;
5292 }
5293
5294 #define INIT_FEATURE_FLAGS(suffix) \
5295         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5296           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5297           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5298
5299 int btrfs_ioctl_get_supported_features(void __user *arg)
5300 {
5301         static const struct btrfs_ioctl_feature_flags features[3] = {
5302                 INIT_FEATURE_FLAGS(SUPP),
5303                 INIT_FEATURE_FLAGS(SAFE_SET),
5304                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5305         };
5306
5307         if (copy_to_user(arg, &features, sizeof(features)))
5308                 return -EFAULT;
5309
5310         return 0;
5311 }
5312
5313 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5314 {
5315         struct inode *inode = file_inode(file);
5316         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5317         struct btrfs_super_block *super_block = fs_info->super_copy;
5318         struct btrfs_ioctl_feature_flags features;
5319
5320         features.compat_flags = btrfs_super_compat_flags(super_block);
5321         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5322         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5323
5324         if (copy_to_user(arg, &features, sizeof(features)))
5325                 return -EFAULT;
5326
5327         return 0;
5328 }
5329
5330 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5331                               enum btrfs_feature_set set,
5332                               u64 change_mask, u64 flags, u64 supported_flags,
5333                               u64 safe_set, u64 safe_clear)
5334 {
5335         const char *type = btrfs_feature_set_names[set];
5336         char *names;
5337         u64 disallowed, unsupported;
5338         u64 set_mask = flags & change_mask;
5339         u64 clear_mask = ~flags & change_mask;
5340
5341         unsupported = set_mask & ~supported_flags;
5342         if (unsupported) {
5343                 names = btrfs_printable_features(set, unsupported);
5344                 if (names) {
5345                         btrfs_warn(fs_info,
5346                                    "this kernel does not support the %s feature bit%s",
5347                                    names, strchr(names, ',') ? "s" : "");
5348                         kfree(names);
5349                 } else
5350                         btrfs_warn(fs_info,
5351                                    "this kernel does not support %s bits 0x%llx",
5352                                    type, unsupported);
5353                 return -EOPNOTSUPP;
5354         }
5355
5356         disallowed = set_mask & ~safe_set;
5357         if (disallowed) {
5358                 names = btrfs_printable_features(set, disallowed);
5359                 if (names) {
5360                         btrfs_warn(fs_info,
5361                                    "can't set the %s feature bit%s while mounted",
5362                                    names, strchr(names, ',') ? "s" : "");
5363                         kfree(names);
5364                 } else
5365                         btrfs_warn(fs_info,
5366                                    "can't set %s bits 0x%llx while mounted",
5367                                    type, disallowed);
5368                 return -EPERM;
5369         }
5370
5371         disallowed = clear_mask & ~safe_clear;
5372         if (disallowed) {
5373                 names = btrfs_printable_features(set, disallowed);
5374                 if (names) {
5375                         btrfs_warn(fs_info,
5376                                    "can't clear the %s feature bit%s while mounted",
5377                                    names, strchr(names, ',') ? "s" : "");
5378                         kfree(names);
5379                 } else
5380                         btrfs_warn(fs_info,
5381                                    "can't clear %s bits 0x%llx while mounted",
5382                                    type, disallowed);
5383                 return -EPERM;
5384         }
5385
5386         return 0;
5387 }
5388
5389 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5390 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5391                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5392                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5393                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5394
5395 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5396 {
5397         struct inode *inode = file_inode(file);
5398         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5399         struct btrfs_root *root = BTRFS_I(inode)->root;
5400         struct btrfs_super_block *super_block = fs_info->super_copy;
5401         struct btrfs_ioctl_feature_flags flags[2];
5402         struct btrfs_trans_handle *trans;
5403         u64 newflags;
5404         int ret;
5405
5406         if (!capable(CAP_SYS_ADMIN))
5407                 return -EPERM;
5408
5409         if (copy_from_user(flags, arg, sizeof(flags)))
5410                 return -EFAULT;
5411
5412         /* Nothing to do */
5413         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5414             !flags[0].incompat_flags)
5415                 return 0;
5416
5417         ret = check_feature(fs_info, flags[0].compat_flags,
5418                             flags[1].compat_flags, COMPAT);
5419         if (ret)
5420                 return ret;
5421
5422         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5423                             flags[1].compat_ro_flags, COMPAT_RO);
5424         if (ret)
5425                 return ret;
5426
5427         ret = check_feature(fs_info, flags[0].incompat_flags,
5428                             flags[1].incompat_flags, INCOMPAT);
5429         if (ret)
5430                 return ret;
5431
5432         ret = mnt_want_write_file(file);
5433         if (ret)
5434                 return ret;
5435
5436         trans = btrfs_start_transaction(root, 0);
5437         if (IS_ERR(trans)) {
5438                 ret = PTR_ERR(trans);
5439                 goto out_drop_write;
5440         }
5441
5442         spin_lock(&fs_info->super_lock);
5443         newflags = btrfs_super_compat_flags(super_block);
5444         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5445         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5446         btrfs_set_super_compat_flags(super_block, newflags);
5447
5448         newflags = btrfs_super_compat_ro_flags(super_block);
5449         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5450         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5451         btrfs_set_super_compat_ro_flags(super_block, newflags);
5452
5453         newflags = btrfs_super_incompat_flags(super_block);
5454         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5455         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5456         btrfs_set_super_incompat_flags(super_block, newflags);
5457         spin_unlock(&fs_info->super_lock);
5458
5459         ret = btrfs_commit_transaction(trans);
5460 out_drop_write:
5461         mnt_drop_write_file(file);
5462
5463         return ret;
5464 }
5465
5466 long btrfs_ioctl(struct file *file, unsigned int
5467                 cmd, unsigned long arg)
5468 {
5469         struct inode *inode = file_inode(file);
5470         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5471         struct btrfs_root *root = BTRFS_I(inode)->root;
5472         void __user *argp = (void __user *)arg;
5473
5474         switch (cmd) {
5475         case FS_IOC_GETFLAGS:
5476                 return btrfs_ioctl_getflags(file, argp);
5477         case FS_IOC_SETFLAGS:
5478                 return btrfs_ioctl_setflags(file, argp);
5479         case FS_IOC_GETVERSION:
5480                 return btrfs_ioctl_getversion(file, argp);
5481         case FITRIM:
5482                 return btrfs_ioctl_fitrim(file, argp);
5483         case BTRFS_IOC_SNAP_CREATE:
5484                 return btrfs_ioctl_snap_create(file, argp, 0);
5485         case BTRFS_IOC_SNAP_CREATE_V2:
5486                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5487         case BTRFS_IOC_SUBVOL_CREATE:
5488                 return btrfs_ioctl_snap_create(file, argp, 1);
5489         case BTRFS_IOC_SUBVOL_CREATE_V2:
5490                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5491         case BTRFS_IOC_SNAP_DESTROY:
5492                 return btrfs_ioctl_snap_destroy(file, argp);
5493         case BTRFS_IOC_SUBVOL_GETFLAGS:
5494                 return btrfs_ioctl_subvol_getflags(file, argp);
5495         case BTRFS_IOC_SUBVOL_SETFLAGS:
5496                 return btrfs_ioctl_subvol_setflags(file, argp);
5497         case BTRFS_IOC_DEFAULT_SUBVOL:
5498                 return btrfs_ioctl_default_subvol(file, argp);
5499         case BTRFS_IOC_DEFRAG:
5500                 return btrfs_ioctl_defrag(file, NULL);
5501         case BTRFS_IOC_DEFRAG_RANGE:
5502                 return btrfs_ioctl_defrag(file, argp);
5503         case BTRFS_IOC_RESIZE:
5504                 return btrfs_ioctl_resize(file, argp);
5505         case BTRFS_IOC_ADD_DEV:
5506                 return btrfs_ioctl_add_dev(fs_info, argp);
5507         case BTRFS_IOC_RM_DEV:
5508                 return btrfs_ioctl_rm_dev(file, argp);
5509         case BTRFS_IOC_RM_DEV_V2:
5510                 return btrfs_ioctl_rm_dev_v2(file, argp);
5511         case BTRFS_IOC_FS_INFO:
5512                 return btrfs_ioctl_fs_info(fs_info, argp);
5513         case BTRFS_IOC_DEV_INFO:
5514                 return btrfs_ioctl_dev_info(fs_info, argp);
5515         case BTRFS_IOC_BALANCE:
5516                 return btrfs_ioctl_balance(file, NULL);
5517         case BTRFS_IOC_TRANS_START:
5518                 return btrfs_ioctl_trans_start(file);
5519         case BTRFS_IOC_TRANS_END:
5520                 return btrfs_ioctl_trans_end(file);
5521         case BTRFS_IOC_TREE_SEARCH:
5522                 return btrfs_ioctl_tree_search(file, argp);
5523         case BTRFS_IOC_TREE_SEARCH_V2:
5524                 return btrfs_ioctl_tree_search_v2(file, argp);
5525         case BTRFS_IOC_INO_LOOKUP:
5526                 return btrfs_ioctl_ino_lookup(file, argp);
5527         case BTRFS_IOC_INO_PATHS:
5528                 return btrfs_ioctl_ino_to_path(root, argp);
5529         case BTRFS_IOC_LOGICAL_INO:
5530                 return btrfs_ioctl_logical_to_ino(fs_info, argp);
5531         case BTRFS_IOC_SPACE_INFO:
5532                 return btrfs_ioctl_space_info(fs_info, argp);
5533         case BTRFS_IOC_SYNC: {
5534                 int ret;
5535
5536                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5537                 if (ret)
5538                         return ret;
5539                 ret = btrfs_sync_fs(inode->i_sb, 1);
5540                 /*
5541                  * The transaction thread may want to do more work,
5542                  * namely it pokes the cleaner kthread that will start
5543                  * processing uncleaned subvols.
5544                  */
5545                 wake_up_process(fs_info->transaction_kthread);
5546                 return ret;
5547         }
5548         case BTRFS_IOC_START_SYNC:
5549                 return btrfs_ioctl_start_sync(root, argp);
5550         case BTRFS_IOC_WAIT_SYNC:
5551                 return btrfs_ioctl_wait_sync(fs_info, argp);
5552         case BTRFS_IOC_SCRUB:
5553                 return btrfs_ioctl_scrub(file, argp);
5554         case BTRFS_IOC_SCRUB_CANCEL:
5555                 return btrfs_ioctl_scrub_cancel(fs_info);
5556         case BTRFS_IOC_SCRUB_PROGRESS:
5557                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5558         case BTRFS_IOC_BALANCE_V2:
5559                 return btrfs_ioctl_balance(file, argp);
5560         case BTRFS_IOC_BALANCE_CTL:
5561                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5562         case BTRFS_IOC_BALANCE_PROGRESS:
5563                 return btrfs_ioctl_balance_progress(fs_info, argp);
5564         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5565                 return btrfs_ioctl_set_received_subvol(file, argp);
5566 #ifdef CONFIG_64BIT
5567         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5568                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5569 #endif
5570         case BTRFS_IOC_SEND:
5571                 return btrfs_ioctl_send(file, argp);
5572         case BTRFS_IOC_GET_DEV_STATS:
5573                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5574         case BTRFS_IOC_QUOTA_CTL:
5575                 return btrfs_ioctl_quota_ctl(file, argp);
5576         case BTRFS_IOC_QGROUP_ASSIGN:
5577                 return btrfs_ioctl_qgroup_assign(file, argp);
5578         case BTRFS_IOC_QGROUP_CREATE:
5579                 return btrfs_ioctl_qgroup_create(file, argp);
5580         case BTRFS_IOC_QGROUP_LIMIT:
5581                 return btrfs_ioctl_qgroup_limit(file, argp);
5582         case BTRFS_IOC_QUOTA_RESCAN:
5583                 return btrfs_ioctl_quota_rescan(file, argp);
5584         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5585                 return btrfs_ioctl_quota_rescan_status(file, argp);
5586         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5587                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5588         case BTRFS_IOC_DEV_REPLACE:
5589                 return btrfs_ioctl_dev_replace(fs_info, argp);
5590         case BTRFS_IOC_GET_FSLABEL:
5591                 return btrfs_ioctl_get_fslabel(file, argp);
5592         case BTRFS_IOC_SET_FSLABEL:
5593                 return btrfs_ioctl_set_fslabel(file, argp);
5594         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5595                 return btrfs_ioctl_get_supported_features(argp);
5596         case BTRFS_IOC_GET_FEATURES:
5597                 return btrfs_ioctl_get_features(file, argp);
5598         case BTRFS_IOC_SET_FEATURES:
5599                 return btrfs_ioctl_set_features(file, argp);
5600         }
5601
5602         return -ENOTTY;
5603 }
5604
5605 #ifdef CONFIG_COMPAT
5606 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5607 {
5608         /*
5609          * These all access 32-bit values anyway so no further
5610          * handling is necessary.
5611          */
5612         switch (cmd) {
5613         case FS_IOC32_GETFLAGS:
5614                 cmd = FS_IOC_GETFLAGS;
5615                 break;
5616         case FS_IOC32_SETFLAGS:
5617                 cmd = FS_IOC_SETFLAGS;
5618                 break;
5619         case FS_IOC32_GETVERSION:
5620                 cmd = FS_IOC_GETVERSION;
5621                 break;
5622         }
5623
5624         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5625 }
5626 #endif