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 void btrfs_wait_for_no_snapshotting_writes(struct btrfs_root *root)
 613 {
 614         s64 writers;
 615         DEFINE_WAIT(wait);
 616
 617         do {
 618                 prepare_to_wait(&root->subv_writers->wait, &wait,
 619                                 TASK_UNINTERRUPTIBLE);
 620
 621                 writers = percpu_counter_sum(&root->subv_writers->counter);
 622                 if (writers)
 623                         schedule();
 624
 625                 finish_wait(&root->subv_writers->wait, &wait);
 626         } while (writers);
 627 }
 628
 629 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 630                            struct dentry *dentry,
 631                            u64 *async_transid, bool readonly,
 632                            struct btrfs_qgroup_inherit *inherit)
 633 {
 634         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 635         struct inode *inode;
 636         struct btrfs_pending_snapshot *pending_snapshot;
 637         struct btrfs_trans_handle *trans;
 638         int ret;
 639
 640         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 641                 return -EINVAL;
 642
 643         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 644         if (!pending_snapshot)
 645                 return -ENOMEM;
 646
 647         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 648                         GFP_KERNEL);
 649         pending_snapshot->path = btrfs_alloc_path();
 650         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 651                 ret = -ENOMEM;
 652                 goto free_pending;
 653         }
 654
 655         atomic_inc(&root->will_be_snapshotted);
 656         smp_mb__after_atomic();
 657         btrfs_wait_for_no_snapshotting_writes(root);
 658
 659         ret = btrfs_start_delalloc_inodes(root, 0);
 660         if (ret)
 661                 goto dec_and_free;
 662
 663         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 664
 665         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 666                              BTRFS_BLOCK_RSV_TEMP);
 667         /*
 668          * 1 - parent dir inode
 669          * 2 - dir entries
 670          * 1 - root item
 671          * 2 - root ref/backref
 672          * 1 - root of snapshot
 673          * 1 - UUID item
 674          */
 675         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 676                                         &pending_snapshot->block_rsv, 8,
 677                                         &pending_snapshot->qgroup_reserved,
 678                                         false);
 679         if (ret)
 680                 goto dec_and_free;
 681
 682         pending_snapshot->dentry = dentry;
 683         pending_snapshot->root = root;
 684         pending_snapshot->readonly = readonly;
 685         pending_snapshot->dir = dir;
 686         pending_snapshot->inherit = inherit;
 687
 688         trans = btrfs_start_transaction(root, 0);
 689         if (IS_ERR(trans)) {
 690                 ret = PTR_ERR(trans);
 691                 goto fail;
 692         }
 693
 694         spin_lock(&fs_info->trans_lock);
 695         list_add(&pending_snapshot->list,
 696                  &trans->transaction->pending_snapshots);
 697         spin_unlock(&fs_info->trans_lock);
 698         if (async_transid) {
 699                 *async_transid = trans->transid;
 700                 ret = btrfs_commit_transaction_async(trans, 1);
 701                 if (ret)
 702                         ret = btrfs_commit_transaction(trans);
 703         } else {
 704                 ret = btrfs_commit_transaction(trans);
 705         }
 706         if (ret)
 707                 goto fail;
 708
 709         ret = pending_snapshot->error;
 710         if (ret)
 711                 goto fail;
 712
 713         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 714         if (ret)
 715                 goto fail;
 716
 717         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 718         if (IS_ERR(inode)) {
 719                 ret = PTR_ERR(inode);
 720                 goto fail;
 721         }
 722
 723         d_instantiate(dentry, inode);
 724         ret = 0;
 725 fail:
 726         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 727 dec_and_free:
 728         if (atomic_dec_and_test(&root->will_be_snapshotted))
 729                 wake_up_atomic_t(&root->will_be_snapshotted);
 730 free_pending:
 731         kfree(pending_snapshot->root_item);
 732         btrfs_free_path(pending_snapshot->path);
 733         kfree(pending_snapshot);
 734
 735         return ret;
 736 }
 737
 738 /*  copy of may_delete in fs/namei.c()
 739  *      Check whether we can remove a link victim from directory dir, check
 740  *  whether the type of victim is right.
 741  *  1. We can't do it if dir is read-only (done in permission())
 742  *  2. We should have write and exec permissions on dir
 743  *  3. We can't remove anything from append-only dir
 744  *  4. We can't do anything with immutable dir (done in permission())
 745  *  5. If the sticky bit on dir is set we should either
 746  *      a. be owner of dir, or
 747  *      b. be owner of victim, or
 748  *      c. have CAP_FOWNER capability
 749  *  6. If the victim is append-only or immutable we can't do anything with
 750  *     links pointing to it.
 751  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 752  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 753  *  9. We can't remove a root or mountpoint.
 754  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 755  *     nfs_async_unlink().
 756  */
 757
 758 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 759 {
 760         int error;
 761
 762         if (d_really_is_negative(victim))
 763                 return -ENOENT;
 764
 765         BUG_ON(d_inode(victim->d_parent) != dir);
 766         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 767
 768         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 769         if (error)
 770                 return error;
 771         if (IS_APPEND(dir))
 772                 return -EPERM;
 773         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 774             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 775                 return -EPERM;
 776         if (isdir) {
 777                 if (!d_is_dir(victim))
 778                         return -ENOTDIR;
 779                 if (IS_ROOT(victim))
 780                         return -EBUSY;
 781         } else if (d_is_dir(victim))
 782                 return -EISDIR;
 783         if (IS_DEADDIR(dir))
 784                 return -ENOENT;
 785         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 786                 return -EBUSY;
 787         return 0;
 788 }
 789
 790 /* copy of may_create in fs/namei.c() */
 791 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 792 {
 793         if (d_really_is_positive(child))
 794                 return -EEXIST;
 795         if (IS_DEADDIR(dir))
 796                 return -ENOENT;
 797         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 798 }
 799
 800 /*
 801  * Create a new subvolume below @parent.  This is largely modeled after
 802  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 803  * inside this filesystem so it's quite a bit simpler.
 804  */
 805 static noinline int btrfs_mksubvol(const struct path *parent,
 806                                    const char *name, int namelen,
 807                                    struct btrfs_root *snap_src,
 808                                    u64 *async_transid, bool readonly,
 809                                    struct btrfs_qgroup_inherit *inherit)
 810 {
 811         struct inode *dir = d_inode(parent->dentry);
 812         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 813         struct dentry *dentry;
 814         int error;
 815
 816         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 817         if (error == -EINTR)
 818                 return error;
 819
 820         dentry = lookup_one_len(name, parent->dentry, namelen);
 821         error = PTR_ERR(dentry);
 822         if (IS_ERR(dentry))
 823                 goto out_unlock;
 824
 825         error = btrfs_may_create(dir, dentry);
 826         if (error)
 827                 goto out_dput;
 828
 829         /*
 830          * even if this name doesn't exist, we may get hash collisions.
 831          * check for them now when we can safely fail
 832          */
 833         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 834                                                dir->i_ino, name,
 835                                                namelen);
 836         if (error)
 837                 goto out_dput;
 838
 839         down_read(&fs_info->subvol_sem);
 840
 841         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 842                 goto out_up_read;
 843
 844         if (snap_src) {
 845                 error = create_snapshot(snap_src, dir, dentry,
 846                                         async_transid, readonly, inherit);
 847         } else {
 848                 error = create_subvol(dir, dentry, name, namelen,
 849                                       async_transid, inherit);
 850         }
 851         if (!error)
 852                 fsnotify_mkdir(dir, dentry);
 853 out_up_read:
 854         up_read(&fs_info->subvol_sem);
 855 out_dput:
 856         dput(dentry);
 857 out_unlock:
 858         inode_unlock(dir);
 859         return error;
 860 }
 861
 862 /*
 863  * When we're defragging a range, we don't want to kick it off again
 864  * if it is really just waiting for delalloc to send it down.
 865  * If we find a nice big extent or delalloc range for the bytes in the
 866  * file you want to defrag, we return 0 to let you know to skip this
 867  * part of the file
 868  */
 869 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 870 {
 871         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 872         struct extent_map *em = NULL;
 873         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 874         u64 end;
 875
 876         read_lock(&em_tree->lock);
 877         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 878         read_unlock(&em_tree->lock);
 879
 880         if (em) {
 881                 end = extent_map_end(em);
 882                 free_extent_map(em);
 883                 if (end - offset > thresh)
 884                         return 0;
 885         }
 886         /* if we already have a nice delalloc here, just stop */
 887         thresh /= 2;
 888         end = count_range_bits(io_tree, &offset, offset + thresh,
 889                                thresh, EXTENT_DELALLOC, 1);
 890         if (end >= thresh)
 891                 return 0;
 892         return 1;
 893 }
 894
 895 /*
 896  * helper function to walk through a file and find extents
 897  * newer than a specific transid, and smaller than thresh.
 898  *
 899  * This is used by the defragging code to find new and small
 900  * extents
 901  */
 902 static int find_new_extents(struct btrfs_root *root,
 903                             struct inode *inode, u64 newer_than,
 904                             u64 *off, u32 thresh)
 905 {
 906         struct btrfs_path *path;
 907         struct btrfs_key min_key;
 908         struct extent_buffer *leaf;
 909         struct btrfs_file_extent_item *extent;
 910         int type;
 911         int ret;
 912         u64 ino = btrfs_ino(BTRFS_I(inode));
 913
 914         path = btrfs_alloc_path();
 915         if (!path)
 916                 return -ENOMEM;
 917
 918         min_key.objectid = ino;
 919         min_key.type = BTRFS_EXTENT_DATA_KEY;
 920         min_key.offset = *off;
 921
 922         while (1) {
 923                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 924                 if (ret != 0)
 925                         goto none;
 926 process_slot:
 927                 if (min_key.objectid != ino)
 928                         goto none;
 929                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 930                         goto none;
 931
 932                 leaf = path->nodes[0];
 933                 extent = btrfs_item_ptr(leaf, path->slots[0],
 934                                         struct btrfs_file_extent_item);
 935
 936                 type = btrfs_file_extent_type(leaf, extent);
 937                 if (type == BTRFS_FILE_EXTENT_REG &&
 938                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 939                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 940                         *off = min_key.offset;
 941                         btrfs_free_path(path);
 942                         return 0;
 943                 }
 944
 945                 path->slots[0]++;
 946                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 947                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 948                         goto process_slot;
 949                 }
 950
 951                 if (min_key.offset == (u64)-1)
 952                         goto none;
 953
 954                 min_key.offset++;
 955                 btrfs_release_path(path);
 956         }
 957 none:
 958         btrfs_free_path(path);
 959         return -ENOENT;
 960 }
 961
 962 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 963 {
 964         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 965         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 966         struct extent_map *em;
 967         u64 len = PAGE_SIZE;
 968
 969         /*
 970          * hopefully we have this extent in the tree already, try without
 971          * the full extent lock
 972          */
 973         read_lock(&em_tree->lock);
 974         em = lookup_extent_mapping(em_tree, start, len);
 975         read_unlock(&em_tree->lock);
 976
 977         if (!em) {
 978                 struct extent_state *cached = NULL;
 979                 u64 end = start + len - 1;
 980
 981                 /* get the big lock and read metadata off disk */
 982                 lock_extent_bits(io_tree, start, end, &cached);
 983                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 984                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 985
 986                 if (IS_ERR(em))
 987                         return NULL;
 988         }
 989
 990         return em;
 991 }
 992
 993 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 994 {
 995         struct extent_map *next;
 996         bool ret = true;
 997
 998         /* this is the last extent */
 999         if (em->start + em->len >= i_size_read(inode))
1000                 return false;
1001
1002         next = defrag_lookup_extent(inode, em->start + em->len);
1003         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1004                 ret = false;
1005         else if ((em->block_start + em->block_len == next->block_start) &&
1006                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1007                 ret = false;
1008
1009         free_extent_map(next);
1010         return ret;
1011 }
1012
1013 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1014                                u64 *last_len, u64 *skip, u64 *defrag_end,
1015                                int compress)
1016 {
1017         struct extent_map *em;
1018         int ret = 1;
1019         bool next_mergeable = true;
1020         bool prev_mergeable = true;
1021
1022         /*
1023          * make sure that once we start defragging an extent, we keep on
1024          * defragging it
1025          */
1026         if (start < *defrag_end)
1027                 return 1;
1028
1029         *skip = 0;
1030
1031         em = defrag_lookup_extent(inode, start);
1032         if (!em)
1033                 return 0;
1034
1035         /* this will cover holes, and inline extents */
1036         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1037                 ret = 0;
1038                 goto out;
1039         }
1040
1041         if (!*defrag_end)
1042                 prev_mergeable = false;
1043
1044         next_mergeable = defrag_check_next_extent(inode, em);
1045         /*
1046          * we hit a real extent, if it is big or the next extent is not a
1047          * real extent, don't bother defragging it
1048          */
1049         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1050             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1051                 ret = 0;
1052 out:
1053         /*
1054          * last_len ends up being a counter of how many bytes we've defragged.
1055          * every time we choose not to defrag an extent, we reset *last_len
1056          * so that the next tiny extent will force a defrag.
1057          *
1058          * The end result of this is that tiny extents before a single big
1059          * extent will force at least part of that big extent to be defragged.
1060          */
1061         if (ret) {
1062                 *defrag_end = extent_map_end(em);
1063         } else {
1064                 *last_len = 0;
1065                 *skip = extent_map_end(em);
1066                 *defrag_end = 0;
1067         }
1068
1069         free_extent_map(em);
1070         return ret;
1071 }
1072
1073 /*
1074  * it doesn't do much good to defrag one or two pages
1075  * at a time.  This pulls in a nice chunk of pages
1076  * to COW and defrag.
1077  *
1078  * It also makes sure the delalloc code has enough
1079  * dirty data to avoid making new small extents as part
1080  * of the defrag
1081  *
1082  * It's a good idea to start RA on this range
1083  * before calling this.
1084  */
1085 static int cluster_pages_for_defrag(struct inode *inode,
1086                                     struct page **pages,
1087                                     unsigned long start_index,
1088                                     unsigned long num_pages)
1089 {
1090         unsigned long file_end;
1091         u64 isize = i_size_read(inode);
1092         u64 page_start;
1093         u64 page_end;
1094         u64 page_cnt;
1095         int ret;
1096         int i;
1097         int i_done;
1098         struct btrfs_ordered_extent *ordered;
1099         struct extent_state *cached_state = NULL;
1100         struct extent_io_tree *tree;
1101         struct extent_changeset *data_reserved = NULL;
1102         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1103
1104         file_end = (isize - 1) >> PAGE_SHIFT;
1105         if (!isize || start_index > file_end)
1106                 return 0;
1107
1108         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1109
1110         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1111                         start_index << PAGE_SHIFT,
1112                         page_cnt << PAGE_SHIFT);
1113         if (ret)
1114                 return ret;
1115         i_done = 0;
1116         tree = &BTRFS_I(inode)->io_tree;
1117
1118         /* step one, lock all the pages */
1119         for (i = 0; i < page_cnt; i++) {
1120                 struct page *page;
1121 again:
1122                 page = find_or_create_page(inode->i_mapping,
1123                                            start_index + i, mask);
1124                 if (!page)
1125                         break;
1126
1127                 page_start = page_offset(page);
1128                 page_end = page_start + PAGE_SIZE - 1;
1129                 while (1) {
1130                         lock_extent_bits(tree, page_start, page_end,
1131                                          &cached_state);
1132                         ordered = btrfs_lookup_ordered_extent(inode,
1133                                                               page_start);
1134                         unlock_extent_cached(tree, page_start, page_end,
1135                                              &cached_state, GFP_NOFS);
1136                         if (!ordered)
1137                                 break;
1138
1139                         unlock_page(page);
1140                         btrfs_start_ordered_extent(inode, ordered, 1);
1141                         btrfs_put_ordered_extent(ordered);
1142                         lock_page(page);
1143                         /*
1144                          * we unlocked the page above, so we need check if
1145                          * it was released or not.
1146                          */
1147                         if (page->mapping != inode->i_mapping) {
1148                                 unlock_page(page);
1149                                 put_page(page);
1150                                 goto again;
1151                         }
1152                 }
1153
1154                 if (!PageUptodate(page)) {
1155                         btrfs_readpage(NULL, page);
1156                         lock_page(page);
1157                         if (!PageUptodate(page)) {
1158                                 unlock_page(page);
1159                                 put_page(page);
1160                                 ret = -EIO;
1161                                 break;
1162                         }
1163                 }
1164
1165                 if (page->mapping != inode->i_mapping) {
1166                         unlock_page(page);
1167                         put_page(page);
1168                         goto again;
1169                 }
1170
1171                 pages[i] = page;
1172                 i_done++;
1173         }
1174         if (!i_done || ret)
1175                 goto out;
1176
1177         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1178                 goto out;
1179
1180         /*
1181          * so now we have a nice long stream of locked
1182          * and up to date pages, lets wait on them
1183          */
1184         for (i = 0; i < i_done; i++)
1185                 wait_on_page_writeback(pages[i]);
1186
1187         page_start = page_offset(pages[0]);
1188         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1189
1190         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1191                          page_start, page_end - 1, &cached_state);
1192         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1193                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1194                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1195                           &cached_state, GFP_NOFS);
1196
1197         if (i_done != page_cnt) {
1198                 spin_lock(&BTRFS_I(inode)->lock);
1199                 BTRFS_I(inode)->outstanding_extents++;
1200                 spin_unlock(&BTRFS_I(inode)->lock);
1201                 btrfs_delalloc_release_space(inode, data_reserved,
1202                                 start_index << PAGE_SHIFT,
1203                                 (page_cnt - i_done) << PAGE_SHIFT);
1204         }
1205
1206
1207         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1208                           &cached_state);
1209
1210         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1211                              page_start, page_end - 1, &cached_state,
1212                              GFP_NOFS);
1213
1214         for (i = 0; i < i_done; i++) {
1215                 clear_page_dirty_for_io(pages[i]);
1216                 ClearPageChecked(pages[i]);
1217                 set_page_extent_mapped(pages[i]);
1218                 set_page_dirty(pages[i]);
1219                 unlock_page(pages[i]);
1220                 put_page(pages[i]);
1221         }
1222         extent_changeset_free(data_reserved);
1223         return i_done;
1224 out:
1225         for (i = 0; i < i_done; i++) {
1226                 unlock_page(pages[i]);
1227                 put_page(pages[i]);
1228         }
1229         btrfs_delalloc_release_space(inode, data_reserved,
1230                         start_index << PAGE_SHIFT,
1231                         page_cnt << PAGE_SHIFT);
1232         extent_changeset_free(data_reserved);
1233         return ret;
1234
1235 }
1236
1237 int btrfs_defrag_file(struct inode *inode, struct file *file,
1238                       struct btrfs_ioctl_defrag_range_args *range,
1239                       u64 newer_than, unsigned long max_to_defrag)
1240 {
1241         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1242         struct btrfs_root *root = BTRFS_I(inode)->root;
1243         struct file_ra_state *ra = NULL;
1244         unsigned long last_index;
1245         u64 isize = i_size_read(inode);
1246         u64 last_len = 0;
1247         u64 skip = 0;
1248         u64 defrag_end = 0;
1249         u64 newer_off = range->start;
1250         unsigned long i;
1251         unsigned long ra_index = 0;
1252         int ret;
1253         int defrag_count = 0;
1254         int compress_type = BTRFS_COMPRESS_ZLIB;
1255         u32 extent_thresh = range->extent_thresh;
1256         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1257         unsigned long cluster = max_cluster;
1258         u64 new_align = ~((u64)SZ_128K - 1);
1259         struct page **pages = NULL;
1260         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1261
1262         if (isize == 0)
1263                 return 0;
1264
1265         if (range->start >= isize)
1266                 return -EINVAL;
1267
1268         if (do_compress) {
1269                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1270                         return -EINVAL;
1271                 if (range->compress_type)
1272                         compress_type = range->compress_type;
1273         }
1274
1275         if (extent_thresh == 0)
1276                 extent_thresh = SZ_256K;
1277
1278         /*
1279          * If we were not given a file, allocate a readahead context. As
1280          * readahead is just an optimization, defrag will work without it so
1281          * we don't error out.
1282          */
1283         if (!file) {
1284                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1285                 if (ra)
1286                         file_ra_state_init(ra, inode->i_mapping);
1287         } else {
1288                 ra = &file->f_ra;
1289         }
1290
1291         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1292         if (!pages) {
1293                 ret = -ENOMEM;
1294                 goto out_ra;
1295         }
1296
1297         /* find the last page to defrag */
1298         if (range->start + range->len > range->start) {
1299                 last_index = min_t(u64, isize - 1,
1300                          range->start + range->len - 1) >> PAGE_SHIFT;
1301         } else {
1302                 last_index = (isize - 1) >> PAGE_SHIFT;
1303         }
1304
1305         if (newer_than) {
1306                 ret = find_new_extents(root, inode, newer_than,
1307                                        &newer_off, SZ_64K);
1308                 if (!ret) {
1309                         range->start = newer_off;
1310                         /*
1311                          * we always align our defrag to help keep
1312                          * the extents in the file evenly spaced
1313                          */
1314                         i = (newer_off & new_align) >> PAGE_SHIFT;
1315                 } else
1316                         goto out_ra;
1317         } else {
1318                 i = range->start >> PAGE_SHIFT;
1319         }
1320         if (!max_to_defrag)
1321                 max_to_defrag = last_index - i + 1;
1322
1323         /*
1324          * make writeback starts from i, so the defrag range can be
1325          * written sequentially.
1326          */
1327         if (i < inode->i_mapping->writeback_index)
1328                 inode->i_mapping->writeback_index = i;
1329
1330         while (i <= last_index && defrag_count < max_to_defrag &&
1331                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1332                 /*
1333                  * make sure we stop running if someone unmounts
1334                  * the FS
1335                  */
1336                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1337                         break;
1338
1339                 if (btrfs_defrag_cancelled(fs_info)) {
1340                         btrfs_debug(fs_info, "defrag_file cancelled");
1341                         ret = -EAGAIN;
1342                         break;
1343                 }
1344
1345                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1346                                          extent_thresh, &last_len, &skip,
1347                                          &defrag_end, do_compress)){
1348                         unsigned long next;
1349                         /*
1350                          * the should_defrag function tells us how much to skip
1351                          * bump our counter by the suggested amount
1352                          */
1353                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1354                         i = max(i + 1, next);
1355                         continue;
1356                 }
1357
1358                 if (!newer_than) {
1359                         cluster = (PAGE_ALIGN(defrag_end) >>
1360                                    PAGE_SHIFT) - i;
1361                         cluster = min(cluster, max_cluster);
1362                 } else {
1363                         cluster = max_cluster;
1364                 }
1365
1366                 if (i + cluster > ra_index) {
1367                         ra_index = max(i, ra_index);
1368                         if (ra)
1369                                 page_cache_sync_readahead(inode->i_mapping, ra,
1370                                                 file, ra_index, cluster);
1371                         ra_index += cluster;
1372                 }
1373
1374                 inode_lock(inode);
1375                 if (do_compress)
1376                         BTRFS_I(inode)->defrag_compress = compress_type;
1377                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378                 if (ret < 0) {
1379                         inode_unlock(inode);
1380                         goto out_ra;
1381                 }
1382
1383                 defrag_count += ret;
1384                 balance_dirty_pages_ratelimited(inode->i_mapping);
1385                 inode_unlock(inode);
1386
1387                 if (newer_than) {
1388                         if (newer_off == (u64)-1)
1389                                 break;
1390
1391                         if (ret > 0)
1392                                 i += ret;
1393
1394                         newer_off = max(newer_off + 1,
1395                                         (u64)i << PAGE_SHIFT);
1396
1397                         ret = find_new_extents(root, inode, newer_than,
1398                                                &newer_off, SZ_64K);
1399                         if (!ret) {
1400                                 range->start = newer_off;
1401                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1402                         } else {
1403                                 break;
1404                         }
1405                 } else {
1406                         if (ret > 0) {
1407                                 i += ret;
1408                                 last_len += ret << PAGE_SHIFT;
1409                         } else {
1410                                 i++;
1411                                 last_len = 0;
1412                         }
1413                 }
1414         }
1415
1416         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1417                 filemap_flush(inode->i_mapping);
1418                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1419                              &BTRFS_I(inode)->runtime_flags))
1420                         filemap_flush(inode->i_mapping);
1421         }
1422
1423         if (do_compress) {
1424                 /* the filemap_flush will queue IO into the worker threads, but
1425                  * we have to make sure the IO is actually started and that
1426                  * ordered extents get created before we return
1427                  */
1428                 atomic_inc(&fs_info->async_submit_draining);
1429                 while (atomic_read(&fs_info->nr_async_submits) ||
1430                        atomic_read(&fs_info->async_delalloc_pages)) {
1431                         wait_event(fs_info->async_submit_wait,
1432                                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
1433                                     atomic_read(&fs_info->async_delalloc_pages) == 0));
1434                 }
1435                 atomic_dec(&fs_info->async_submit_draining);
1436         }
1437
1438         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1439                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1440         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1441                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1442         }
1443
1444         ret = defrag_count;
1445
1446 out_ra:
1447         if (do_compress) {
1448                 inode_lock(inode);
1449                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1450                 inode_unlock(inode);
1451         }
1452         if (!file)
1453                 kfree(ra);
1454         kfree(pages);
1455         return ret;
1456 }
1457
1458 static noinline int btrfs_ioctl_resize(struct file *file,
1459                                         void __user *arg)
1460 {
1461         struct inode *inode = file_inode(file);
1462         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1463         u64 new_size;
1464         u64 old_size;
1465         u64 devid = 1;
1466         struct btrfs_root *root = BTRFS_I(inode)->root;
1467         struct btrfs_ioctl_vol_args *vol_args;
1468         struct btrfs_trans_handle *trans;
1469         struct btrfs_device *device = NULL;
1470         char *sizestr;
1471         char *retptr;
1472         char *devstr = NULL;
1473         int ret = 0;
1474         int mod = 0;
1475
1476         if (!capable(CAP_SYS_ADMIN))
1477                 return -EPERM;
1478
1479         ret = mnt_want_write_file(file);
1480         if (ret)
1481                 return ret;
1482
1483         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1484                 mnt_drop_write_file(file);
1485                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1486         }
1487
1488         mutex_lock(&fs_info->volume_mutex);
1489         vol_args = memdup_user(arg, sizeof(*vol_args));
1490         if (IS_ERR(vol_args)) {
1491                 ret = PTR_ERR(vol_args);
1492                 goto out;
1493         }
1494
1495         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1496
1497         sizestr = vol_args->name;
1498         devstr = strchr(sizestr, ':');
1499         if (devstr) {
1500                 sizestr = devstr + 1;
1501                 *devstr = '\0';
1502                 devstr = vol_args->name;
1503                 ret = kstrtoull(devstr, 10, &devid);
1504                 if (ret)
1505                         goto out_free;
1506                 if (!devid) {
1507                         ret = -EINVAL;
1508                         goto out_free;
1509                 }
1510                 btrfs_info(fs_info, "resizing devid %llu", devid);
1511         }
1512
1513         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1514         if (!device) {
1515                 btrfs_info(fs_info, "resizer unable to find device %llu",
1516                            devid);
1517                 ret = -ENODEV;
1518                 goto out_free;
1519         }
1520
1521         if (!device->writeable) {
1522                 btrfs_info(fs_info,
1523                            "resizer unable to apply on readonly device %llu",
1524                        devid);
1525                 ret = -EPERM;
1526                 goto out_free;
1527         }
1528
1529         if (!strcmp(sizestr, "max"))
1530                 new_size = device->bdev->bd_inode->i_size;
1531         else {
1532                 if (sizestr[0] == '-') {
1533                         mod = -1;
1534                         sizestr++;
1535                 } else if (sizestr[0] == '+') {
1536                         mod = 1;
1537                         sizestr++;
1538                 }
1539                 new_size = memparse(sizestr, &retptr);
1540                 if (*retptr != '\0' || new_size == 0) {
1541                         ret = -EINVAL;
1542                         goto out_free;
1543                 }
1544         }
1545
1546         if (device->is_tgtdev_for_dev_replace) {
1547                 ret = -EPERM;
1548                 goto out_free;
1549         }
1550
1551         old_size = btrfs_device_get_total_bytes(device);
1552
1553         if (mod < 0) {
1554                 if (new_size > old_size) {
1555                         ret = -EINVAL;
1556                         goto out_free;
1557                 }
1558                 new_size = old_size - new_size;
1559         } else if (mod > 0) {
1560                 if (new_size > ULLONG_MAX - old_size) {
1561                         ret = -ERANGE;
1562                         goto out_free;
1563                 }
1564                 new_size = old_size + new_size;
1565         }
1566
1567         if (new_size < SZ_256M) {
1568                 ret = -EINVAL;
1569                 goto out_free;
1570         }
1571         if (new_size > device->bdev->bd_inode->i_size) {
1572                 ret = -EFBIG;
1573                 goto out_free;
1574         }
1575
1576         new_size = round_down(new_size, fs_info->sectorsize);
1577
1578         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1579                           rcu_str_deref(device->name), new_size);
1580
1581         if (new_size > old_size) {
1582                 trans = btrfs_start_transaction(root, 0);
1583                 if (IS_ERR(trans)) {
1584                         ret = PTR_ERR(trans);
1585                         goto out_free;
1586                 }
1587                 ret = btrfs_grow_device(trans, device, new_size);
1588                 btrfs_commit_transaction(trans);
1589         } else if (new_size < old_size) {
1590                 ret = btrfs_shrink_device(device, new_size);
1591         } /* equal, nothing need to do */
1592
1593 out_free:
1594         kfree(vol_args);
1595 out:
1596         mutex_unlock(&fs_info->volume_mutex);
1597         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1598         mnt_drop_write_file(file);
1599         return ret;
1600 }
1601
1602 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1603                                 const char *name, unsigned long fd, int subvol,
1604                                 u64 *transid, bool readonly,
1605                                 struct btrfs_qgroup_inherit *inherit)
1606 {
1607         int namelen;
1608         int ret = 0;
1609
1610         if (!S_ISDIR(file_inode(file)->i_mode))
1611                 return -ENOTDIR;
1612
1613         ret = mnt_want_write_file(file);
1614         if (ret)
1615                 goto out;
1616
1617         namelen = strlen(name);
1618         if (strchr(name, '/')) {
1619                 ret = -EINVAL;
1620                 goto out_drop_write;
1621         }
1622
1623         if (name[0] == '.' &&
1624            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1625                 ret = -EEXIST;
1626                 goto out_drop_write;
1627         }
1628
1629         if (subvol) {
1630                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1631                                      NULL, transid, readonly, inherit);
1632         } else {
1633                 struct fd src = fdget(fd);
1634                 struct inode *src_inode;
1635                 if (!src.file) {
1636                         ret = -EINVAL;
1637                         goto out_drop_write;
1638                 }
1639
1640                 src_inode = file_inode(src.file);
1641                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1642                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1643                                    "Snapshot src from another FS");
1644                         ret = -EXDEV;
1645                 } else if (!inode_owner_or_capable(src_inode)) {
1646                         /*
1647                          * Subvolume creation is not restricted, but snapshots
1648                          * are limited to own subvolumes only
1649                          */
1650                         ret = -EPERM;
1651                 } else {
1652                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1653                                              BTRFS_I(src_inode)->root,
1654                                              transid, readonly, inherit);
1655                 }
1656                 fdput(src);
1657         }
1658 out_drop_write:
1659         mnt_drop_write_file(file);
1660 out:
1661         return ret;
1662 }
1663
1664 static noinline int btrfs_ioctl_snap_create(struct file *file,
1665                                             void __user *arg, int subvol)
1666 {
1667         struct btrfs_ioctl_vol_args *vol_args;
1668         int ret;
1669
1670         if (!S_ISDIR(file_inode(file)->i_mode))
1671                 return -ENOTDIR;
1672
1673         vol_args = memdup_user(arg, sizeof(*vol_args));
1674         if (IS_ERR(vol_args))
1675                 return PTR_ERR(vol_args);
1676         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1677
1678         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1679                                               vol_args->fd, subvol,
1680                                               NULL, false, NULL);
1681
1682         kfree(vol_args);
1683         return ret;
1684 }
1685
1686 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1687                                                void __user *arg, int subvol)
1688 {
1689         struct btrfs_ioctl_vol_args_v2 *vol_args;
1690         int ret;
1691         u64 transid = 0;
1692         u64 *ptr = NULL;
1693         bool readonly = false;
1694         struct btrfs_qgroup_inherit *inherit = NULL;
1695
1696         if (!S_ISDIR(file_inode(file)->i_mode))
1697                 return -ENOTDIR;
1698
1699         vol_args = memdup_user(arg, sizeof(*vol_args));
1700         if (IS_ERR(vol_args))
1701                 return PTR_ERR(vol_args);
1702         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1703
1704         if (vol_args->flags &
1705             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1706               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1707                 ret = -EOPNOTSUPP;
1708                 goto free_args;
1709         }
1710
1711         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1712                 ptr = &transid;
1713         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1714                 readonly = true;
1715         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1716                 if (vol_args->size > PAGE_SIZE) {
1717                         ret = -EINVAL;
1718                         goto free_args;
1719                 }
1720                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1721                 if (IS_ERR(inherit)) {
1722                         ret = PTR_ERR(inherit);
1723                         goto free_args;
1724                 }
1725         }
1726
1727         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1728                                               vol_args->fd, subvol, ptr,
1729                                               readonly, inherit);
1730         if (ret)
1731                 goto free_inherit;
1732
1733         if (ptr && copy_to_user(arg +
1734                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1735                                         transid),
1736                                 ptr, sizeof(*ptr)))
1737                 ret = -EFAULT;
1738
1739 free_inherit:
1740         kfree(inherit);
1741 free_args:
1742         kfree(vol_args);
1743         return ret;
1744 }
1745
1746 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1747                                                 void __user *arg)
1748 {
1749         struct inode *inode = file_inode(file);
1750         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1751         struct btrfs_root *root = BTRFS_I(inode)->root;
1752         int ret = 0;
1753         u64 flags = 0;
1754
1755         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1756                 return -EINVAL;
1757
1758         down_read(&fs_info->subvol_sem);
1759         if (btrfs_root_readonly(root))
1760                 flags |= BTRFS_SUBVOL_RDONLY;
1761         up_read(&fs_info->subvol_sem);
1762
1763         if (copy_to_user(arg, &flags, sizeof(flags)))
1764                 ret = -EFAULT;
1765
1766         return ret;
1767 }
1768
1769 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1770                                               void __user *arg)
1771 {
1772         struct inode *inode = file_inode(file);
1773         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1774         struct btrfs_root *root = BTRFS_I(inode)->root;
1775         struct btrfs_trans_handle *trans;
1776         u64 root_flags;
1777         u64 flags;
1778         int ret = 0;
1779
1780         if (!inode_owner_or_capable(inode))
1781                 return -EPERM;
1782
1783         ret = mnt_want_write_file(file);
1784         if (ret)
1785                 goto out;
1786
1787         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1788                 ret = -EINVAL;
1789                 goto out_drop_write;
1790         }
1791
1792         if (copy_from_user(&flags, arg, sizeof(flags))) {
1793                 ret = -EFAULT;
1794                 goto out_drop_write;
1795         }
1796
1797         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1798                 ret = -EINVAL;
1799                 goto out_drop_write;
1800         }
1801
1802         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1803                 ret = -EOPNOTSUPP;
1804                 goto out_drop_write;
1805         }
1806
1807         down_write(&fs_info->subvol_sem);
1808
1809         /* nothing to do */
1810         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1811                 goto out_drop_sem;
1812
1813         root_flags = btrfs_root_flags(&root->root_item);
1814         if (flags & BTRFS_SUBVOL_RDONLY) {
1815                 btrfs_set_root_flags(&root->root_item,
1816                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1817         } else {
1818                 /*
1819                  * Block RO -> RW transition if this subvolume is involved in
1820                  * send
1821                  */
1822                 spin_lock(&root->root_item_lock);
1823                 if (root->send_in_progress == 0) {
1824                         btrfs_set_root_flags(&root->root_item,
1825                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1826                         spin_unlock(&root->root_item_lock);
1827                 } else {
1828                         spin_unlock(&root->root_item_lock);
1829                         btrfs_warn(fs_info,
1830                                    "Attempt to set subvolume %llu read-write during send",
1831                                    root->root_key.objectid);
1832                         ret = -EPERM;
1833                         goto out_drop_sem;
1834                 }
1835         }
1836
1837         trans = btrfs_start_transaction(root, 1);
1838         if (IS_ERR(trans)) {
1839                 ret = PTR_ERR(trans);
1840                 goto out_reset;
1841         }
1842
1843         ret = btrfs_update_root(trans, fs_info->tree_root,
1844                                 &root->root_key, &root->root_item);
1845         if (ret < 0) {
1846                 btrfs_end_transaction(trans);
1847                 goto out_reset;
1848         }
1849
1850         ret = btrfs_commit_transaction(trans);
1851
1852 out_reset:
1853         if (ret)
1854                 btrfs_set_root_flags(&root->root_item, root_flags);
1855 out_drop_sem:
1856         up_write(&fs_info->subvol_sem);
1857 out_drop_write:
1858         mnt_drop_write_file(file);
1859 out:
1860         return ret;
1861 }
1862
1863 /*
1864  * helper to check if the subvolume references other subvolumes
1865  */
1866 static noinline int may_destroy_subvol(struct btrfs_root *root)
1867 {
1868         struct btrfs_fs_info *fs_info = root->fs_info;
1869         struct btrfs_path *path;
1870         struct btrfs_dir_item *di;
1871         struct btrfs_key key;
1872         u64 dir_id;
1873         int ret;
1874
1875         path = btrfs_alloc_path();
1876         if (!path)
1877                 return -ENOMEM;
1878
1879         /* Make sure this root isn't set as the default subvol */
1880         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1881         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1882                                    dir_id, "default", 7, 0);
1883         if (di && !IS_ERR(di)) {
1884                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1885                 if (key.objectid == root->root_key.objectid) {
1886                         ret = -EPERM;
1887                         btrfs_err(fs_info,
1888                                   "deleting default subvolume %llu is not allowed",
1889                                   key.objectid);
1890                         goto out;
1891                 }
1892                 btrfs_release_path(path);
1893         }
1894
1895         key.objectid = root->root_key.objectid;
1896         key.type = BTRFS_ROOT_REF_KEY;
1897         key.offset = (u64)-1;
1898
1899         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1900         if (ret < 0)
1901                 goto out;
1902         BUG_ON(ret == 0);
1903
1904         ret = 0;
1905         if (path->slots[0] > 0) {
1906                 path->slots[0]--;
1907                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1908                 if (key.objectid == root->root_key.objectid &&
1909                     key.type == BTRFS_ROOT_REF_KEY)
1910                         ret = -ENOTEMPTY;
1911         }
1912 out:
1913         btrfs_free_path(path);
1914         return ret;
1915 }
1916
1917 static noinline int key_in_sk(struct btrfs_key *key,
1918                               struct btrfs_ioctl_search_key *sk)
1919 {
1920         struct btrfs_key test;
1921         int ret;
1922
1923         test.objectid = sk->min_objectid;
1924         test.type = sk->min_type;
1925         test.offset = sk->min_offset;
1926
1927         ret = btrfs_comp_cpu_keys(key, &test);
1928         if (ret < 0)
1929                 return 0;
1930
1931         test.objectid = sk->max_objectid;
1932         test.type = sk->max_type;
1933         test.offset = sk->max_offset;
1934
1935         ret = btrfs_comp_cpu_keys(key, &test);
1936         if (ret > 0)
1937                 return 0;
1938         return 1;
1939 }
1940
1941 static noinline int copy_to_sk(struct btrfs_path *path,
1942                                struct btrfs_key *key,
1943                                struct btrfs_ioctl_search_key *sk,
1944                                size_t *buf_size,
1945                                char __user *ubuf,
1946                                unsigned long *sk_offset,
1947                                int *num_found)
1948 {
1949         u64 found_transid;
1950         struct extent_buffer *leaf;
1951         struct btrfs_ioctl_search_header sh;
1952         struct btrfs_key test;
1953         unsigned long item_off;
1954         unsigned long item_len;
1955         int nritems;
1956         int i;
1957         int slot;
1958         int ret = 0;
1959
1960         leaf = path->nodes[0];
1961         slot = path->slots[0];
1962         nritems = btrfs_header_nritems(leaf);
1963
1964         if (btrfs_header_generation(leaf) > sk->max_transid) {
1965                 i = nritems;
1966                 goto advance_key;
1967         }
1968         found_transid = btrfs_header_generation(leaf);
1969
1970         for (i = slot; i < nritems; i++) {
1971                 item_off = btrfs_item_ptr_offset(leaf, i);
1972                 item_len = btrfs_item_size_nr(leaf, i);
1973
1974                 btrfs_item_key_to_cpu(leaf, key, i);
1975                 if (!key_in_sk(key, sk))
1976                         continue;
1977
1978                 if (sizeof(sh) + item_len > *buf_size) {
1979                         if (*num_found) {
1980                                 ret = 1;
1981                                 goto out;
1982                         }
1983
1984                         /*
1985                          * return one empty item back for v1, which does not
1986                          * handle -EOVERFLOW
1987                          */
1988
1989                         *buf_size = sizeof(sh) + item_len;
1990                         item_len = 0;
1991                         ret = -EOVERFLOW;
1992                 }
1993
1994                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1995                         ret = 1;
1996                         goto out;
1997                 }
1998
1999                 sh.objectid = key->objectid;
2000                 sh.offset = key->offset;
2001                 sh.type = key->type;
2002                 sh.len = item_len;
2003                 sh.transid = found_transid;
2004
2005                 /* copy search result header */
2006                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2007                         ret = -EFAULT;
2008                         goto out;
2009                 }
2010
2011                 *sk_offset += sizeof(sh);
2012
2013                 if (item_len) {
2014                         char __user *up = ubuf + *sk_offset;
2015                         /* copy the item */
2016                         if (read_extent_buffer_to_user(leaf, up,
2017                                                        item_off, item_len)) {
2018                                 ret = -EFAULT;
2019                                 goto out;
2020                         }
2021
2022                         *sk_offset += item_len;
2023                 }
2024                 (*num_found)++;
2025
2026                 if (ret) /* -EOVERFLOW from above */
2027                         goto out;
2028
2029                 if (*num_found >= sk->nr_items) {
2030                         ret = 1;
2031                         goto out;
2032                 }
2033         }
2034 advance_key:
2035         ret = 0;
2036         test.objectid = sk->max_objectid;
2037         test.type = sk->max_type;
2038         test.offset = sk->max_offset;
2039         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2040                 ret = 1;
2041         else if (key->offset < (u64)-1)
2042                 key->offset++;
2043         else if (key->type < (u8)-1) {
2044                 key->offset = 0;
2045                 key->type++;
2046         } else if (key->objectid < (u64)-1) {
2047                 key->offset = 0;
2048                 key->type = 0;
2049                 key->objectid++;
2050         } else
2051                 ret = 1;
2052 out:
2053         /*
2054          *  0: all items from this leaf copied, continue with next
2055          *  1: * more items can be copied, but unused buffer is too small
2056          *     * all items were found
2057          *     Either way, it will stops the loop which iterates to the next
2058          *     leaf
2059          *  -EOVERFLOW: item was to large for buffer
2060          *  -EFAULT: could not copy extent buffer back to userspace
2061          */
2062         return ret;
2063 }
2064
2065 static noinline int search_ioctl(struct inode *inode,
2066                                  struct btrfs_ioctl_search_key *sk,
2067                                  size_t *buf_size,
2068                                  char __user *ubuf)
2069 {
2070         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2071         struct btrfs_root *root;
2072         struct btrfs_key key;
2073         struct btrfs_path *path;
2074         int ret;
2075         int num_found = 0;
2076         unsigned long sk_offset = 0;
2077
2078         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2079                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2080                 return -EOVERFLOW;
2081         }
2082
2083         path = btrfs_alloc_path();
2084         if (!path)
2085                 return -ENOMEM;
2086
2087         if (sk->tree_id == 0) {
2088                 /* search the root of the inode that was passed */
2089                 root = BTRFS_I(inode)->root;
2090         } else {
2091                 key.objectid = sk->tree_id;
2092                 key.type = BTRFS_ROOT_ITEM_KEY;
2093                 key.offset = (u64)-1;
2094                 root = btrfs_read_fs_root_no_name(info, &key);
2095                 if (IS_ERR(root)) {
2096                         btrfs_free_path(path);
2097                         return -ENOENT;
2098                 }
2099         }
2100
2101         key.objectid = sk->min_objectid;
2102         key.type = sk->min_type;
2103         key.offset = sk->min_offset;
2104
2105         while (1) {
2106                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2107                 if (ret != 0) {
2108                         if (ret > 0)
2109                                 ret = 0;
2110                         goto err;
2111                 }
2112                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2113                                  &sk_offset, &num_found);
2114                 btrfs_release_path(path);
2115                 if (ret)
2116                         break;
2117
2118         }
2119         if (ret > 0)
2120                 ret = 0;
2121 err:
2122         sk->nr_items = num_found;
2123         btrfs_free_path(path);
2124         return ret;
2125 }
2126
2127 static noinline int btrfs_ioctl_tree_search(struct file *file,
2128                                            void __user *argp)
2129 {
2130         struct btrfs_ioctl_search_args __user *uargs;
2131         struct btrfs_ioctl_search_key sk;
2132         struct inode *inode;
2133         int ret;
2134         size_t buf_size;
2135
2136         if (!capable(CAP_SYS_ADMIN))
2137                 return -EPERM;
2138
2139         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2140
2141         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2142                 return -EFAULT;
2143
2144         buf_size = sizeof(uargs->buf);
2145
2146         inode = file_inode(file);
2147         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2148
2149         /*
2150          * In the origin implementation an overflow is handled by returning a
2151          * search header with a len of zero, so reset ret.
2152          */
2153         if (ret == -EOVERFLOW)
2154                 ret = 0;
2155
2156         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2157                 ret = -EFAULT;
2158         return ret;
2159 }
2160
2161 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2162                                                void __user *argp)
2163 {
2164         struct btrfs_ioctl_search_args_v2 __user *uarg;
2165         struct btrfs_ioctl_search_args_v2 args;
2166         struct inode *inode;
2167         int ret;
2168         size_t buf_size;
2169         const size_t buf_limit = SZ_16M;
2170
2171         if (!capable(CAP_SYS_ADMIN))
2172                 return -EPERM;
2173
2174         /* copy search header and buffer size */
2175         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2176         if (copy_from_user(&args, uarg, sizeof(args)))
2177                 return -EFAULT;
2178
2179         buf_size = args.buf_size;
2180
2181         /* limit result size to 16MB */
2182         if (buf_size > buf_limit)
2183                 buf_size = buf_limit;
2184
2185         inode = file_inode(file);
2186         ret = search_ioctl(inode, &args.key, &buf_size,
2187                            (char *)(&uarg->buf[0]));
2188         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2189                 ret = -EFAULT;
2190         else if (ret == -EOVERFLOW &&
2191                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2192                 ret = -EFAULT;
2193
2194         return ret;
2195 }
2196
2197 /*
2198  * Search INODE_REFs to identify path name of 'dirid' directory
2199  * in a 'tree_id' tree. and sets path name to 'name'.
2200  */
2201 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2202                                 u64 tree_id, u64 dirid, char *name)
2203 {
2204         struct btrfs_root *root;
2205         struct btrfs_key key;
2206         char *ptr;
2207         int ret = -1;
2208         int slot;
2209         int len;
2210         int total_len = 0;
2211         struct btrfs_inode_ref *iref;
2212         struct extent_buffer *l;
2213         struct btrfs_path *path;
2214
2215         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2216                 name[0]='\0';
2217                 return 0;
2218         }
2219
2220         path = btrfs_alloc_path();
2221         if (!path)
2222                 return -ENOMEM;
2223
2224         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2225
2226         key.objectid = tree_id;
2227         key.type = BTRFS_ROOT_ITEM_KEY;
2228         key.offset = (u64)-1;
2229         root = btrfs_read_fs_root_no_name(info, &key);
2230         if (IS_ERR(root)) {
2231                 btrfs_err(info, "could not find root %llu", tree_id);
2232                 ret = -ENOENT;
2233                 goto out;
2234         }
2235
2236         key.objectid = dirid;
2237         key.type = BTRFS_INODE_REF_KEY;
2238         key.offset = (u64)-1;
2239
2240         while (1) {
2241                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2242                 if (ret < 0)
2243                         goto out;
2244                 else if (ret > 0) {
2245                         ret = btrfs_previous_item(root, path, dirid,
2246                                                   BTRFS_INODE_REF_KEY);
2247                         if (ret < 0)
2248                                 goto out;
2249                         else if (ret > 0) {
2250                                 ret = -ENOENT;
2251                                 goto out;
2252                         }
2253                 }
2254
2255                 l = path->nodes[0];
2256                 slot = path->slots[0];
2257                 btrfs_item_key_to_cpu(l, &key, slot);
2258
2259                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2260                 len = btrfs_inode_ref_name_len(l, iref);
2261                 ptr -= len + 1;
2262                 total_len += len + 1;
2263                 if (ptr < name) {
2264                         ret = -ENAMETOOLONG;
2265                         goto out;
2266                 }
2267
2268                 *(ptr + len) = '/';
2269                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2270
2271                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2272                         break;
2273
2274                 btrfs_release_path(path);
2275                 key.objectid = key.offset;
2276                 key.offset = (u64)-1;
2277                 dirid = key.objectid;
2278         }
2279         memmove(name, ptr, total_len);
2280         name[total_len] = '\0';
2281         ret = 0;
2282 out:
2283         btrfs_free_path(path);
2284         return ret;
2285 }
2286
2287 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2288                                            void __user *argp)
2289 {
2290          struct btrfs_ioctl_ino_lookup_args *args;
2291          struct inode *inode;
2292         int ret = 0;
2293
2294         args = memdup_user(argp, sizeof(*args));
2295         if (IS_ERR(args))
2296                 return PTR_ERR(args);
2297
2298         inode = file_inode(file);
2299
2300         /*
2301          * Unprivileged query to obtain the containing subvolume root id. The
2302          * path is reset so it's consistent with btrfs_search_path_in_tree.
2303          */
2304         if (args->treeid == 0)
2305                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2306
2307         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2308                 args->name[0] = 0;
2309                 goto out;
2310         }
2311
2312         if (!capable(CAP_SYS_ADMIN)) {
2313                 ret = -EPERM;
2314                 goto out;
2315         }
2316
2317         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2318                                         args->treeid, args->objectid,
2319                                         args->name);
2320
2321 out:
2322         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2323                 ret = -EFAULT;
2324
2325         kfree(args);
2326         return ret;
2327 }
2328
2329 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2330                                              void __user *arg)
2331 {
2332         struct dentry *parent = file->f_path.dentry;
2333         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2334         struct dentry *dentry;
2335         struct inode *dir = d_inode(parent);
2336         struct inode *inode;
2337         struct btrfs_root *root = BTRFS_I(dir)->root;
2338         struct btrfs_root *dest = NULL;
2339         struct btrfs_ioctl_vol_args *vol_args;
2340         struct btrfs_trans_handle *trans;
2341         struct btrfs_block_rsv block_rsv;
2342         u64 root_flags;
2343         u64 qgroup_reserved;
2344         int namelen;
2345         int ret;
2346         int err = 0;
2347
2348         if (!S_ISDIR(dir->i_mode))
2349                 return -ENOTDIR;
2350
2351         vol_args = memdup_user(arg, sizeof(*vol_args));
2352         if (IS_ERR(vol_args))
2353                 return PTR_ERR(vol_args);
2354
2355         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2356         namelen = strlen(vol_args->name);
2357         if (strchr(vol_args->name, '/') ||
2358             strncmp(vol_args->name, "..", namelen) == 0) {
2359                 err = -EINVAL;
2360                 goto out;
2361         }
2362
2363         err = mnt_want_write_file(file);
2364         if (err)
2365                 goto out;
2366
2367
2368         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2369         if (err == -EINTR)
2370                 goto out_drop_write;
2371         dentry = lookup_one_len(vol_args->name, parent, namelen);
2372         if (IS_ERR(dentry)) {
2373                 err = PTR_ERR(dentry);
2374                 goto out_unlock_dir;
2375         }
2376
2377         if (d_really_is_negative(dentry)) {
2378                 err = -ENOENT;
2379                 goto out_dput;
2380         }
2381
2382         inode = d_inode(dentry);
2383         dest = BTRFS_I(inode)->root;
2384         if (!capable(CAP_SYS_ADMIN)) {
2385                 /*
2386                  * Regular user.  Only allow this with a special mount
2387                  * option, when the user has write+exec access to the
2388                  * subvol root, and when rmdir(2) would have been
2389                  * allowed.
2390                  *
2391                  * Note that this is _not_ check that the subvol is
2392                  * empty or doesn't contain data that we wouldn't
2393                  * otherwise be able to delete.
2394                  *
2395                  * Users who want to delete empty subvols should try
2396                  * rmdir(2).
2397                  */
2398                 err = -EPERM;
2399                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2400                         goto out_dput;
2401
2402                 /*
2403                  * Do not allow deletion if the parent dir is the same
2404                  * as the dir to be deleted.  That means the ioctl
2405                  * must be called on the dentry referencing the root
2406                  * of the subvol, not a random directory contained
2407                  * within it.
2408                  */
2409                 err = -EINVAL;
2410                 if (root == dest)
2411                         goto out_dput;
2412
2413                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2414                 if (err)
2415                         goto out_dput;
2416         }
2417
2418         /* check if subvolume may be deleted by a user */
2419         err = btrfs_may_delete(dir, dentry, 1);
2420         if (err)
2421                 goto out_dput;
2422
2423         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2424                 err = -EINVAL;
2425                 goto out_dput;
2426         }
2427
2428         inode_lock(inode);
2429
2430         /*
2431          * Don't allow to delete a subvolume with send in progress. This is
2432          * inside the i_mutex so the error handling that has to drop the bit
2433          * again is not run concurrently.
2434          */
2435         spin_lock(&dest->root_item_lock);
2436         root_flags = btrfs_root_flags(&dest->root_item);
2437         if (dest->send_in_progress == 0) {
2438                 btrfs_set_root_flags(&dest->root_item,
2439                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2440                 spin_unlock(&dest->root_item_lock);
2441         } else {
2442                 spin_unlock(&dest->root_item_lock);
2443                 btrfs_warn(fs_info,
2444                            "Attempt to delete subvolume %llu during send",
2445                            dest->root_key.objectid);
2446                 err = -EPERM;
2447                 goto out_unlock_inode;
2448         }
2449
2450         down_write(&fs_info->subvol_sem);
2451
2452         err = may_destroy_subvol(dest);
2453         if (err)
2454                 goto out_up_write;
2455
2456         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2457         /*
2458          * One for dir inode, two for dir entries, two for root
2459          * ref/backref.
2460          */
2461         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2462                                                5, &qgroup_reserved, true);
2463         if (err)
2464                 goto out_up_write;
2465
2466         trans = btrfs_start_transaction(root, 0);
2467         if (IS_ERR(trans)) {
2468                 err = PTR_ERR(trans);
2469                 goto out_release;
2470         }
2471         trans->block_rsv = &block_rsv;
2472         trans->bytes_reserved = block_rsv.size;
2473
2474         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2475
2476         ret = btrfs_unlink_subvol(trans, root, dir,
2477                                 dest->root_key.objectid,
2478                                 dentry->d_name.name,
2479                                 dentry->d_name.len);
2480         if (ret) {
2481                 err = ret;
2482                 btrfs_abort_transaction(trans, ret);
2483                 goto out_end_trans;
2484         }
2485
2486         btrfs_record_root_in_trans(trans, dest);
2487
2488         memset(&dest->root_item.drop_progress, 0,
2489                 sizeof(dest->root_item.drop_progress));
2490         dest->root_item.drop_level = 0;
2491         btrfs_set_root_refs(&dest->root_item, 0);
2492
2493         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2494                 ret = btrfs_insert_orphan_item(trans,
2495                                         fs_info->tree_root,
2496                                         dest->root_key.objectid);
2497                 if (ret) {
2498                         btrfs_abort_transaction(trans, ret);
2499                         err = ret;
2500                         goto out_end_trans;
2501                 }
2502         }
2503
2504         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2505                                   BTRFS_UUID_KEY_SUBVOL,
2506                                   dest->root_key.objectid);
2507         if (ret && ret != -ENOENT) {
2508                 btrfs_abort_transaction(trans, ret);
2509                 err = ret;
2510                 goto out_end_trans;
2511         }
2512         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2513                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2514                                           dest->root_item.received_uuid,
2515                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2516                                           dest->root_key.objectid);
2517                 if (ret && ret != -ENOENT) {
2518                         btrfs_abort_transaction(trans, ret);
2519                         err = ret;
2520                         goto out_end_trans;
2521                 }
2522         }
2523
2524 out_end_trans:
2525         trans->block_rsv = NULL;
2526         trans->bytes_reserved = 0;
2527         ret = btrfs_end_transaction(trans);
2528         if (ret && !err)
2529                 err = ret;
2530         inode->i_flags |= S_DEAD;
2531 out_release:
2532         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2533 out_up_write:
2534         up_write(&fs_info->subvol_sem);
2535         if (err) {
2536                 spin_lock(&dest->root_item_lock);
2537                 root_flags = btrfs_root_flags(&dest->root_item);
2538                 btrfs_set_root_flags(&dest->root_item,
2539                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2540                 spin_unlock(&dest->root_item_lock);
2541         }
2542 out_unlock_inode:
2543         inode_unlock(inode);
2544         if (!err) {
2545                 d_invalidate(dentry);
2546                 btrfs_invalidate_inodes(dest);
2547                 d_delete(dentry);
2548                 ASSERT(dest->send_in_progress == 0);
2549
2550                 /* the last ref */
2551                 if (dest->ino_cache_inode) {
2552                         iput(dest->ino_cache_inode);
2553                         dest->ino_cache_inode = NULL;
2554                 }
2555         }
2556 out_dput:
2557         dput(dentry);
2558 out_unlock_dir:
2559         inode_unlock(dir);
2560 out_drop_write:
2561         mnt_drop_write_file(file);
2562 out:
2563         kfree(vol_args);
2564         return err;
2565 }
2566
2567 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2568 {
2569         struct inode *inode = file_inode(file);
2570         struct btrfs_root *root = BTRFS_I(inode)->root;
2571         struct btrfs_ioctl_defrag_range_args *range;
2572         int ret;
2573
2574         ret = mnt_want_write_file(file);
2575         if (ret)
2576                 return ret;
2577
2578         if (btrfs_root_readonly(root)) {
2579                 ret = -EROFS;
2580                 goto out;
2581         }
2582
2583         switch (inode->i_mode & S_IFMT) {
2584         case S_IFDIR:
2585                 if (!capable(CAP_SYS_ADMIN)) {
2586                         ret = -EPERM;
2587                         goto out;
2588                 }
2589                 ret = btrfs_defrag_root(root);
2590                 break;
2591         case S_IFREG:
2592                 if (!(file->f_mode & FMODE_WRITE)) {
2593                         ret = -EINVAL;
2594                         goto out;
2595                 }
2596
2597                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2598                 if (!range) {
2599                         ret = -ENOMEM;
2600                         goto out;
2601                 }
2602
2603                 if (argp) {
2604                         if (copy_from_user(range, argp,
2605                                            sizeof(*range))) {
2606                                 ret = -EFAULT;
2607                                 kfree(range);
2608                                 goto out;
2609                         }
2610                         /* compression requires us to start the IO */
2611                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2612                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2613                                 range->extent_thresh = (u32)-1;
2614                         }
2615                 } else {
2616                         /* the rest are all set to zero by kzalloc */
2617                         range->len = (u64)-1;
2618                 }
2619                 ret = btrfs_defrag_file(file_inode(file), file,
2620                                         range, 0, 0);
2621                 if (ret > 0)
2622                         ret = 0;
2623                 kfree(range);
2624                 break;
2625         default:
2626                 ret = -EINVAL;
2627         }
2628 out:
2629         mnt_drop_write_file(file);
2630         return ret;
2631 }
2632
2633 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2634 {
2635         struct btrfs_ioctl_vol_args *vol_args;
2636         int ret;
2637
2638         if (!capable(CAP_SYS_ADMIN))
2639                 return -EPERM;
2640
2641         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2642                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2643
2644         mutex_lock(&fs_info->volume_mutex);
2645         vol_args = memdup_user(arg, sizeof(*vol_args));
2646         if (IS_ERR(vol_args)) {
2647                 ret = PTR_ERR(vol_args);
2648                 goto out;
2649         }
2650
2651         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2652         ret = btrfs_init_new_device(fs_info, vol_args->name);
2653
2654         if (!ret)
2655                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2656
2657         kfree(vol_args);
2658 out:
2659         mutex_unlock(&fs_info->volume_mutex);
2660         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2661         return ret;
2662 }
2663
2664 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2665 {
2666         struct inode *inode = file_inode(file);
2667         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2668         struct btrfs_ioctl_vol_args_v2 *vol_args;
2669         int ret;
2670
2671         if (!capable(CAP_SYS_ADMIN))
2672                 return -EPERM;
2673
2674         ret = mnt_want_write_file(file);
2675         if (ret)
2676                 return ret;
2677
2678         vol_args = memdup_user(arg, sizeof(*vol_args));
2679         if (IS_ERR(vol_args)) {
2680                 ret = PTR_ERR(vol_args);
2681                 goto err_drop;
2682         }
2683
2684         /* Check for compatibility reject unknown flags */
2685         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2686                 return -EOPNOTSUPP;
2687
2688         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2689                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2690                 goto out;
2691         }
2692
2693         mutex_lock(&fs_info->volume_mutex);
2694         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2695                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2696         } else {
2697                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2698                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2699         }
2700         mutex_unlock(&fs_info->volume_mutex);
2701         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2702
2703         if (!ret) {
2704                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2705                         btrfs_info(fs_info, "device deleted: id %llu",
2706                                         vol_args->devid);
2707                 else
2708                         btrfs_info(fs_info, "device deleted: %s",
2709                                         vol_args->name);
2710         }
2711 out:
2712         kfree(vol_args);
2713 err_drop:
2714         mnt_drop_write_file(file);
2715         return ret;
2716 }
2717
2718 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2719 {
2720         struct inode *inode = file_inode(file);
2721         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2722         struct btrfs_ioctl_vol_args *vol_args;
2723         int ret;
2724
2725         if (!capable(CAP_SYS_ADMIN))
2726                 return -EPERM;
2727
2728         ret = mnt_want_write_file(file);
2729         if (ret)
2730                 return ret;
2731
2732         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2733                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2734                 goto out_drop_write;
2735         }
2736
2737         vol_args = memdup_user(arg, sizeof(*vol_args));
2738         if (IS_ERR(vol_args)) {
2739                 ret = PTR_ERR(vol_args);
2740                 goto out;
2741         }
2742
2743         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2744         mutex_lock(&fs_info->volume_mutex);
2745         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2746         mutex_unlock(&fs_info->volume_mutex);
2747
2748         if (!ret)
2749                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2750         kfree(vol_args);
2751 out:
2752         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2753 out_drop_write:
2754         mnt_drop_write_file(file);
2755
2756         return ret;
2757 }
2758
2759 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2760                                 void __user *arg)
2761 {
2762         struct btrfs_ioctl_fs_info_args *fi_args;
2763         struct btrfs_device *device;
2764         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2765         int ret = 0;
2766
2767         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2768         if (!fi_args)
2769                 return -ENOMEM;
2770
2771         mutex_lock(&fs_devices->device_list_mutex);
2772         fi_args->num_devices = fs_devices->num_devices;
2773         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2774
2775         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2776                 if (device->devid > fi_args->max_id)
2777                         fi_args->max_id = device->devid;
2778         }
2779         mutex_unlock(&fs_devices->device_list_mutex);
2780
2781         fi_args->nodesize = fs_info->nodesize;
2782         fi_args->sectorsize = fs_info->sectorsize;
2783         fi_args->clone_alignment = fs_info->sectorsize;
2784
2785         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2786                 ret = -EFAULT;
2787
2788         kfree(fi_args);
2789         return ret;
2790 }
2791
2792 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2793                                  void __user *arg)
2794 {
2795         struct btrfs_ioctl_dev_info_args *di_args;
2796         struct btrfs_device *dev;
2797         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2798         int ret = 0;
2799         char *s_uuid = NULL;
2800
2801         di_args = memdup_user(arg, sizeof(*di_args));
2802         if (IS_ERR(di_args))
2803                 return PTR_ERR(di_args);
2804
2805         if (!btrfs_is_empty_uuid(di_args->uuid))
2806                 s_uuid = di_args->uuid;
2807
2808         mutex_lock(&fs_devices->device_list_mutex);
2809         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2810
2811         if (!dev) {
2812                 ret = -ENODEV;
2813                 goto out;
2814         }
2815
2816         di_args->devid = dev->devid;
2817         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2818         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2819         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2820         if (dev->name) {
2821                 struct rcu_string *name;
2822
2823                 rcu_read_lock();
2824                 name = rcu_dereference(dev->name);
2825                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2826                 rcu_read_unlock();
2827                 di_args->path[sizeof(di_args->path) - 1] = 0;
2828         } else {
2829                 di_args->path[0] = '\0';
2830         }
2831
2832 out:
2833         mutex_unlock(&fs_devices->device_list_mutex);
2834         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2835                 ret = -EFAULT;
2836
2837         kfree(di_args);
2838         return ret;
2839 }
2840
2841 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2842 {
2843         struct page *page;
2844
2845         page = grab_cache_page(inode->i_mapping, index);
2846         if (!page)
2847                 return ERR_PTR(-ENOMEM);
2848
2849         if (!PageUptodate(page)) {
2850                 int ret;
2851
2852                 ret = btrfs_readpage(NULL, page);
2853                 if (ret)
2854                         return ERR_PTR(ret);
2855                 lock_page(page);
2856                 if (!PageUptodate(page)) {
2857                         unlock_page(page);
2858                         put_page(page);
2859                         return ERR_PTR(-EIO);
2860                 }
2861                 if (page->mapping != inode->i_mapping) {
2862                         unlock_page(page);
2863                         put_page(page);
2864                         return ERR_PTR(-EAGAIN);
2865                 }
2866         }
2867
2868         return page;
2869 }
2870
2871 static int gather_extent_pages(struct inode *inode, struct page **pages,
2872                                int num_pages, u64 off)
2873 {
2874         int i;
2875         pgoff_t index = off >> PAGE_SHIFT;
2876
2877         for (i = 0; i < num_pages; i++) {
2878 again:
2879                 pages[i] = extent_same_get_page(inode, index + i);
2880                 if (IS_ERR(pages[i])) {
2881                         int err = PTR_ERR(pages[i]);
2882
2883                         if (err == -EAGAIN)
2884                                 goto again;
2885                         pages[i] = NULL;
2886                         return err;
2887                 }
2888         }
2889         return 0;
2890 }
2891
2892 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2893                              bool retry_range_locking)
2894 {
2895         /*
2896          * Do any pending delalloc/csum calculations on inode, one way or
2897          * another, and lock file content.
2898          * The locking order is:
2899          *
2900          *   1) pages
2901          *   2) range in the inode's io tree
2902          */
2903         while (1) {
2904                 struct btrfs_ordered_extent *ordered;
2905                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2906                 ordered = btrfs_lookup_first_ordered_extent(inode,
2907                                                             off + len - 1);
2908                 if ((!ordered ||
2909                      ordered->file_offset + ordered->len <= off ||
2910                      ordered->file_offset >= off + len) &&
2911                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2912                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2913                         if (ordered)
2914                                 btrfs_put_ordered_extent(ordered);
2915                         break;
2916                 }
2917                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2918                 if (ordered)
2919                         btrfs_put_ordered_extent(ordered);
2920                 if (!retry_range_locking)
2921                         return -EAGAIN;
2922                 btrfs_wait_ordered_range(inode, off, len);
2923         }
2924         return 0;
2925 }
2926
2927 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2928 {
2929         inode_unlock(inode1);
2930         inode_unlock(inode2);
2931 }
2932
2933 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2934 {
2935         if (inode1 < inode2)
2936                 swap(inode1, inode2);
2937
2938         inode_lock_nested(inode1, I_MUTEX_PARENT);
2939         inode_lock_nested(inode2, I_MUTEX_CHILD);
2940 }
2941
2942 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2943                                       struct inode *inode2, u64 loff2, u64 len)
2944 {
2945         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2946         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2947 }
2948
2949 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2950                                     struct inode *inode2, u64 loff2, u64 len,
2951                                     bool retry_range_locking)
2952 {
2953         int ret;
2954
2955         if (inode1 < inode2) {
2956                 swap(inode1, inode2);
2957                 swap(loff1, loff2);
2958         }
2959         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2960         if (ret)
2961                 return ret;
2962         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2963         if (ret)
2964                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2965                               loff1 + len - 1);
2966         return ret;
2967 }
2968
2969 struct cmp_pages {
2970         int             num_pages;
2971         struct page     **src_pages;
2972         struct page     **dst_pages;
2973 };
2974
2975 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2976 {
2977         int i;
2978         struct page *pg;
2979
2980         for (i = 0; i < cmp->num_pages; i++) {
2981                 pg = cmp->src_pages[i];
2982                 if (pg) {
2983                         unlock_page(pg);
2984                         put_page(pg);
2985                 }
2986                 pg = cmp->dst_pages[i];
2987                 if (pg) {
2988                         unlock_page(pg);
2989                         put_page(pg);
2990                 }
2991         }
2992         kfree(cmp->src_pages);
2993         kfree(cmp->dst_pages);
2994 }
2995
2996 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2997                                   struct inode *dst, u64 dst_loff,
2998                                   u64 len, struct cmp_pages *cmp)
2999 {
3000         int ret;
3001         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3002         struct page **src_pgarr, **dst_pgarr;
3003
3004         /*
3005          * We must gather up all the pages before we initiate our
3006          * extent locking. We use an array for the page pointers. Size
3007          * of the array is bounded by len, which is in turn bounded by
3008          * BTRFS_MAX_DEDUPE_LEN.
3009          */
3010         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3011         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
3012         if (!src_pgarr || !dst_pgarr) {
3013                 kfree(src_pgarr);
3014                 kfree(dst_pgarr);
3015                 return -ENOMEM;
3016         }
3017         cmp->num_pages = num_pages;
3018         cmp->src_pages = src_pgarr;
3019         cmp->dst_pages = dst_pgarr;
3020
3021         /*
3022          * If deduping ranges in the same inode, locking rules make it mandatory
3023          * to always lock pages in ascending order to avoid deadlocks with
3024          * concurrent tasks (such as starting writeback/delalloc).
3025          */
3026         if (src == dst && dst_loff < loff) {
3027                 swap(src_pgarr, dst_pgarr);
3028                 swap(loff, dst_loff);
3029         }
3030
3031         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3032         if (ret)
3033                 goto out;
3034
3035         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3036
3037 out:
3038         if (ret)
3039                 btrfs_cmp_data_free(cmp);
3040         return ret;
3041 }
3042
3043 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3044 {
3045         int ret = 0;
3046         int i;
3047         struct page *src_page, *dst_page;
3048         unsigned int cmp_len = PAGE_SIZE;
3049         void *addr, *dst_addr;
3050
3051         i = 0;
3052         while (len) {
3053                 if (len < PAGE_SIZE)
3054                         cmp_len = len;
3055
3056                 BUG_ON(i >= cmp->num_pages);
3057
3058                 src_page = cmp->src_pages[i];
3059                 dst_page = cmp->dst_pages[i];
3060                 ASSERT(PageLocked(src_page));
3061                 ASSERT(PageLocked(dst_page));
3062
3063                 addr = kmap_atomic(src_page);
3064                 dst_addr = kmap_atomic(dst_page);
3065
3066                 flush_dcache_page(src_page);
3067                 flush_dcache_page(dst_page);
3068
3069                 if (memcmp(addr, dst_addr, cmp_len))
3070                         ret = -EBADE;
3071
3072                 kunmap_atomic(addr);
3073                 kunmap_atomic(dst_addr);
3074
3075                 if (ret)
3076                         break;
3077
3078                 len -= cmp_len;
3079                 i++;
3080         }
3081
3082         return ret;
3083 }
3084
3085 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3086                                      u64 olen)
3087 {
3088         u64 len = *plen;
3089         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3090
3091         if (off + olen > inode->i_size || off + olen < off)
3092                 return -EINVAL;
3093
3094         /* if we extend to eof, continue to block boundary */
3095         if (off + len == inode->i_size)
3096                 *plen = len = ALIGN(inode->i_size, bs) - off;
3097
3098         /* Check that we are block aligned - btrfs_clone() requires this */
3099         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3100                 return -EINVAL;
3101
3102         return 0;
3103 }
3104
3105 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3106                              struct inode *dst, u64 dst_loff)
3107 {
3108         int ret;
3109         u64 len = olen;
3110         struct cmp_pages cmp;
3111         bool same_inode = (src == dst);
3112         u64 same_lock_start = 0;
3113         u64 same_lock_len = 0;
3114
3115         if (len == 0)
3116                 return 0;
3117
3118         if (same_inode)
3119                 inode_lock(src);
3120         else
3121                 btrfs_double_inode_lock(src, dst);
3122
3123         ret = extent_same_check_offsets(src, loff, &len, olen);
3124         if (ret)
3125                 goto out_unlock;
3126
3127         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3128         if (ret)
3129                 goto out_unlock;
3130
3131         if (same_inode) {
3132                 /*
3133                  * Single inode case wants the same checks, except we
3134                  * don't want our length pushed out past i_size as
3135                  * comparing that data range makes no sense.
3136                  *
3137                  * extent_same_check_offsets() will do this for an
3138                  * unaligned length at i_size, so catch it here and
3139                  * reject the request.
3140                  *
3141                  * This effectively means we require aligned extents
3142                  * for the single-inode case, whereas the other cases
3143                  * allow an unaligned length so long as it ends at
3144                  * i_size.
3145                  */
3146                 if (len != olen) {
3147                         ret = -EINVAL;
3148                         goto out_unlock;
3149                 }
3150
3151                 /* Check for overlapping ranges */
3152                 if (dst_loff + len > loff && dst_loff < loff + len) {
3153                         ret = -EINVAL;
3154                         goto out_unlock;
3155                 }
3156
3157                 same_lock_start = min_t(u64, loff, dst_loff);
3158                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3159         }
3160
3161         /* don't make the dst file partly checksummed */
3162         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3163             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3164                 ret = -EINVAL;
3165                 goto out_unlock;
3166         }
3167
3168 again:
3169         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3170         if (ret)
3171                 goto out_unlock;
3172
3173         if (same_inode)
3174                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3175                                         false);
3176         else
3177                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3178                                                false);
3179         /*
3180          * If one of the inodes has dirty pages in the respective range or
3181          * ordered extents, we need to flush dellaloc and wait for all ordered
3182          * extents in the range. We must unlock the pages and the ranges in the
3183          * io trees to avoid deadlocks when flushing delalloc (requires locking
3184          * pages) and when waiting for ordered extents to complete (they require
3185          * range locking).
3186          */
3187         if (ret == -EAGAIN) {
3188                 /*
3189                  * Ranges in the io trees already unlocked. Now unlock all
3190                  * pages before waiting for all IO to complete.
3191                  */
3192                 btrfs_cmp_data_free(&cmp);
3193                 if (same_inode) {
3194                         btrfs_wait_ordered_range(src, same_lock_start,
3195                                                  same_lock_len);
3196                 } else {
3197                         btrfs_wait_ordered_range(src, loff, len);
3198                         btrfs_wait_ordered_range(dst, dst_loff, len);
3199                 }
3200                 goto again;
3201         }
3202         ASSERT(ret == 0);
3203         if (WARN_ON(ret)) {
3204                 /* ranges in the io trees already unlocked */
3205                 btrfs_cmp_data_free(&cmp);
3206                 return ret;
3207         }
3208
3209         /* pass original length for comparison so we stay within i_size */
3210         ret = btrfs_cmp_data(olen, &cmp);
3211         if (ret == 0)
3212                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3213
3214         if (same_inode)
3215                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3216                               same_lock_start + same_lock_len - 1);
3217         else
3218                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3219
3220         btrfs_cmp_data_free(&cmp);
3221 out_unlock:
3222         if (same_inode)
3223                 inode_unlock(src);
3224         else
3225                 btrfs_double_inode_unlock(src, dst);
3226
3227         return ret;
3228 }
3229
3230 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3231
3232 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3233                                 struct file *dst_file, u64 dst_loff)
3234 {
3235         struct inode *src = file_inode(src_file);
3236         struct inode *dst = file_inode(dst_file);
3237         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3238         ssize_t res;
3239
3240         if (olen > BTRFS_MAX_DEDUPE_LEN)
3241                 olen = BTRFS_MAX_DEDUPE_LEN;
3242
3243         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3244                 /*
3245                  * Btrfs does not support blocksize < page_size. As a
3246                  * result, btrfs_cmp_data() won't correctly handle
3247                  * this situation without an update.
3248                  */
3249                 return -EINVAL;
3250         }
3251
3252         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3253         if (res)
3254                 return res;
3255         return olen;
3256 }
3257
3258 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3259                                      struct inode *inode,
3260                                      u64 endoff,
3261                                      const u64 destoff,
3262                                      const u64 olen,
3263                                      int no_time_update)
3264 {
3265         struct btrfs_root *root = BTRFS_I(inode)->root;
3266         int ret;
3267
3268         inode_inc_iversion(inode);
3269         if (!no_time_update)
3270                 inode->i_mtime = inode->i_ctime = current_time(inode);
3271         /*
3272          * We round up to the block size at eof when determining which
3273          * extents to clone above, but shouldn't round up the file size.
3274          */
3275         if (endoff > destoff + olen)
3276                 endoff = destoff + olen;
3277         if (endoff > inode->i_size)
3278                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3279
3280         ret = btrfs_update_inode(trans, root, inode);
3281         if (ret) {
3282                 btrfs_abort_transaction(trans, ret);
3283                 btrfs_end_transaction(trans);
3284                 goto out;
3285         }
3286         ret = btrfs_end_transaction(trans);
3287 out:
3288         return ret;
3289 }
3290
3291 static void clone_update_extent_map(struct btrfs_inode *inode,
3292                                     const struct btrfs_trans_handle *trans,
3293                                     const struct btrfs_path *path,
3294                                     const u64 hole_offset,
3295                                     const u64 hole_len)
3296 {
3297         struct extent_map_tree *em_tree = &inode->extent_tree;
3298         struct extent_map *em;
3299         int ret;
3300
3301         em = alloc_extent_map();
3302         if (!em) {
3303                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3304                 return;
3305         }
3306
3307         if (path) {
3308                 struct btrfs_file_extent_item *fi;
3309
3310                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3311                                     struct btrfs_file_extent_item);
3312                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3313                 em->generation = -1;
3314                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3315                     BTRFS_FILE_EXTENT_INLINE)
3316                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3317                                         &inode->runtime_flags);
3318         } else {
3319                 em->start = hole_offset;
3320                 em->len = hole_len;
3321                 em->ram_bytes = em->len;
3322                 em->orig_start = hole_offset;
3323                 em->block_start = EXTENT_MAP_HOLE;
3324                 em->block_len = 0;
3325                 em->orig_block_len = 0;
3326                 em->compress_type = BTRFS_COMPRESS_NONE;
3327                 em->generation = trans->transid;
3328         }
3329
3330         while (1) {
3331                 write_lock(&em_tree->lock);
3332                 ret = add_extent_mapping(em_tree, em, 1);
3333                 write_unlock(&em_tree->lock);
3334                 if (ret != -EEXIST) {
3335                         free_extent_map(em);
3336                         break;
3337                 }
3338                 btrfs_drop_extent_cache(inode, em->start,
3339                                         em->start + em->len - 1, 0);
3340         }
3341
3342         if (ret)
3343                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3344 }
3345
3346 /*
3347  * Make sure we do not end up inserting an inline extent into a file that has
3348  * already other (non-inline) extents. If a file has an inline extent it can
3349  * not have any other extents and the (single) inline extent must start at the
3350  * file offset 0. Failing to respect these rules will lead to file corruption,
3351  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3352  *
3353  * We can have extents that have been already written to disk or we can have
3354  * dirty ranges still in delalloc, in which case the extent maps and items are
3355  * created only when we run delalloc, and the delalloc ranges might fall outside
3356  * the range we are currently locking in the inode's io tree. So we check the
3357  * inode's i_size because of that (i_size updates are done while holding the
3358  * i_mutex, which we are holding here).
3359  * We also check to see if the inode has a size not greater than "datal" but has
3360  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3361  * protected against such concurrent fallocate calls by the i_mutex).
3362  *
3363  * If the file has no extents but a size greater than datal, do not allow the
3364  * copy because we would need turn the inline extent into a non-inline one (even
3365  * with NO_HOLES enabled). If we find our destination inode only has one inline
3366  * extent, just overwrite it with the source inline extent if its size is less
3367  * than the source extent's size, or we could copy the source inline extent's
3368  * data into the destination inode's inline extent if the later is greater then
3369  * the former.
3370  */
3371 static int clone_copy_inline_extent(struct inode *dst,
3372                                     struct btrfs_trans_handle *trans,
3373                                     struct btrfs_path *path,
3374                                     struct btrfs_key *new_key,
3375                                     const u64 drop_start,
3376                                     const u64 datal,
3377                                     const u64 skip,
3378                                     const u64 size,
3379                                     char *inline_data)
3380 {
3381         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3382         struct btrfs_root *root = BTRFS_I(dst)->root;
3383         const u64 aligned_end = ALIGN(new_key->offset + datal,
3384                                       fs_info->sectorsize);
3385         int ret;
3386         struct btrfs_key key;
3387
3388         if (new_key->offset > 0)
3389                 return -EOPNOTSUPP;
3390
3391         key.objectid = btrfs_ino(BTRFS_I(dst));
3392         key.type = BTRFS_EXTENT_DATA_KEY;
3393         key.offset = 0;
3394         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3395         if (ret < 0) {
3396                 return ret;
3397         } else if (ret > 0) {
3398                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3399                         ret = btrfs_next_leaf(root, path);
3400                         if (ret < 0)
3401                                 return ret;
3402                         else if (ret > 0)
3403                                 goto copy_inline_extent;
3404                 }
3405                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3406                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3407                     key.type == BTRFS_EXTENT_DATA_KEY) {
3408                         ASSERT(key.offset > 0);
3409                         return -EOPNOTSUPP;
3410                 }
3411         } else if (i_size_read(dst) <= datal) {
3412                 struct btrfs_file_extent_item *ei;
3413                 u64 ext_len;
3414
3415                 /*
3416                  * If the file size is <= datal, make sure there are no other
3417                  * extents following (can happen do to an fallocate call with
3418                  * the flag FALLOC_FL_KEEP_SIZE).
3419                  */
3420                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3421                                     struct btrfs_file_extent_item);
3422                 /*
3423                  * If it's an inline extent, it can not have other extents
3424                  * following it.
3425                  */
3426                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3427                     BTRFS_FILE_EXTENT_INLINE)
3428                         goto copy_inline_extent;
3429
3430                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3431                 if (ext_len > aligned_end)
3432                         return -EOPNOTSUPP;
3433
3434                 ret = btrfs_next_item(root, path);
3435                 if (ret < 0) {
3436                         return ret;
3437                 } else if (ret == 0) {
3438                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3439                                               path->slots[0]);
3440                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3441                             key.type == BTRFS_EXTENT_DATA_KEY)
3442                                 return -EOPNOTSUPP;
3443                 }
3444         }
3445
3446 copy_inline_extent:
3447         /*
3448          * We have no extent items, or we have an extent at offset 0 which may
3449          * or may not be inlined. All these cases are dealt the same way.
3450          */
3451         if (i_size_read(dst) > datal) {
3452                 /*
3453                  * If the destination inode has an inline extent...
3454                  * This would require copying the data from the source inline
3455                  * extent into the beginning of the destination's inline extent.
3456                  * But this is really complex, both extents can be compressed
3457                  * or just one of them, which would require decompressing and
3458                  * re-compressing data (which could increase the new compressed
3459                  * size, not allowing the compressed data to fit anymore in an
3460                  * inline extent).
3461                  * So just don't support this case for now (it should be rare,
3462                  * we are not really saving space when cloning inline extents).
3463                  */
3464                 return -EOPNOTSUPP;
3465         }
3466
3467         btrfs_release_path(path);
3468         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3469         if (ret)
3470                 return ret;
3471         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3472         if (ret)
3473                 return ret;
3474
3475         if (skip) {
3476                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3477
3478                 memmove(inline_data + start, inline_data + start + skip, datal);
3479         }
3480
3481         write_extent_buffer(path->nodes[0], inline_data,
3482                             btrfs_item_ptr_offset(path->nodes[0],
3483                                                   path->slots[0]),
3484                             size);
3485         inode_add_bytes(dst, datal);
3486
3487         return 0;
3488 }
3489
3490 /**
3491  * btrfs_clone() - clone a range from inode file to another
3492  *
3493  * @src: Inode to clone from
3494  * @inode: Inode to clone to
3495  * @off: Offset within source to start clone from
3496  * @olen: Original length, passed by user, of range to clone
3497  * @olen_aligned: Block-aligned value of olen
3498  * @destoff: Offset within @inode to start clone
3499  * @no_time_update: Whether to update mtime/ctime on the target inode
3500  */
3501 static int btrfs_clone(struct inode *src, struct inode *inode,
3502                        const u64 off, const u64 olen, const u64 olen_aligned,
3503                        const u64 destoff, int no_time_update)
3504 {
3505         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3506         struct btrfs_root *root = BTRFS_I(inode)->root;
3507         struct btrfs_path *path = NULL;
3508         struct extent_buffer *leaf;
3509         struct btrfs_trans_handle *trans;
3510         char *buf = NULL;
3511         struct btrfs_key key;
3512         u32 nritems;
3513         int slot;
3514         int ret;
3515         const u64 len = olen_aligned;
3516         u64 last_dest_end = destoff;
3517
3518         ret = -ENOMEM;
3519         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3520         if (!buf)
3521                 return ret;
3522
3523         path = btrfs_alloc_path();
3524         if (!path) {
3525                 kvfree(buf);
3526                 return ret;
3527         }
3528
3529         path->reada = READA_FORWARD;
3530         /* clone data */
3531         key.objectid = btrfs_ino(BTRFS_I(src));
3532         key.type = BTRFS_EXTENT_DATA_KEY;
3533         key.offset = off;
3534
3535         while (1) {
3536                 u64 next_key_min_offset = key.offset + 1;
3537
3538                 /*
3539                  * note the key will change type as we walk through the
3540                  * tree.
3541                  */
3542                 path->leave_spinning = 1;
3543                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3544                                 0, 0);
3545                 if (ret < 0)
3546                         goto out;
3547                 /*
3548                  * First search, if no extent item that starts at offset off was
3549                  * found but the previous item is an extent item, it's possible
3550                  * it might overlap our target range, therefore process it.
3551                  */
3552                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3553                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3554                                               path->slots[0] - 1);
3555                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3556                                 path->slots[0]--;
3557                 }
3558
3559                 nritems = btrfs_header_nritems(path->nodes[0]);
3560 process_slot:
3561                 if (path->slots[0] >= nritems) {
3562                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3563                         if (ret < 0)
3564                                 goto out;
3565                         if (ret > 0)
3566                                 break;
3567                         nritems = btrfs_header_nritems(path->nodes[0]);
3568                 }
3569                 leaf = path->nodes[0];
3570                 slot = path->slots[0];
3571
3572                 btrfs_item_key_to_cpu(leaf, &key, slot);
3573                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3574                     key.objectid != btrfs_ino(BTRFS_I(src)))
3575                         break;
3576
3577                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3578                         struct btrfs_file_extent_item *extent;
3579                         int type;
3580                         u32 size;
3581                         struct btrfs_key new_key;
3582                         u64 disko = 0, diskl = 0;
3583                         u64 datao = 0, datal = 0;
3584                         u8 comp;
3585                         u64 drop_start;
3586
3587                         extent = btrfs_item_ptr(leaf, slot,
3588                                                 struct btrfs_file_extent_item);
3589                         comp = btrfs_file_extent_compression(leaf, extent);
3590                         type = btrfs_file_extent_type(leaf, extent);
3591                         if (type == BTRFS_FILE_EXTENT_REG ||
3592                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3593                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3594                                                                       extent);
3595                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3596                                                                  extent);
3597                                 datao = btrfs_file_extent_offset(leaf, extent);
3598                                 datal = btrfs_file_extent_num_bytes(leaf,
3599                                                                     extent);
3600                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3601                                 /* take upper bound, may be compressed */
3602                                 datal = btrfs_file_extent_ram_bytes(leaf,
3603                                                                     extent);
3604                         }
3605
3606                         /*
3607                          * The first search might have left us at an extent
3608                          * item that ends before our target range's start, can
3609                          * happen if we have holes and NO_HOLES feature enabled.
3610                          */
3611                         if (key.offset + datal <= off) {
3612                                 path->slots[0]++;
3613                                 goto process_slot;
3614                         } else if (key.offset >= off + len) {
3615                                 break;
3616                         }
3617                         next_key_min_offset = key.offset + datal;
3618                         size = btrfs_item_size_nr(leaf, slot);
3619                         read_extent_buffer(leaf, buf,
3620                                            btrfs_item_ptr_offset(leaf, slot),
3621                                            size);
3622
3623                         btrfs_release_path(path);
3624                         path->leave_spinning = 0;
3625
3626                         memcpy(&new_key, &key, sizeof(new_key));
3627                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3628                         if (off <= key.offset)
3629                                 new_key.offset = key.offset + destoff - off;
3630                         else
3631                                 new_key.offset = destoff;
3632
3633                         /*
3634                          * Deal with a hole that doesn't have an extent item
3635                          * that represents it (NO_HOLES feature enabled).
3636                          * This hole is either in the middle of the cloning
3637                          * range or at the beginning (fully overlaps it or
3638                          * partially overlaps it).
3639                          */
3640                         if (new_key.offset != last_dest_end)
3641                                 drop_start = last_dest_end;
3642                         else
3643                                 drop_start = new_key.offset;
3644
3645                         /*
3646                          * 1 - adjusting old extent (we may have to split it)
3647                          * 1 - add new extent
3648                          * 1 - inode update
3649                          */
3650                         trans = btrfs_start_transaction(root, 3);
3651                         if (IS_ERR(trans)) {
3652                                 ret = PTR_ERR(trans);
3653                                 goto out;
3654                         }
3655
3656                         if (type == BTRFS_FILE_EXTENT_REG ||
3657                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3658                                 /*
3659                                  *    a  | --- range to clone ---|  b
3660                                  * | ------------- extent ------------- |
3661                                  */
3662
3663                                 /* subtract range b */
3664                                 if (key.offset + datal > off + len)
3665                                         datal = off + len - key.offset;
3666
3667                                 /* subtract range a */
3668                                 if (off > key.offset) {
3669                                         datao += off - key.offset;
3670                                         datal -= off - key.offset;
3671                                 }
3672
3673                                 ret = btrfs_drop_extents(trans, root, inode,
3674                                                          drop_start,
3675                                                          new_key.offset + datal,
3676                                                          1);
3677                                 if (ret) {
3678                                         if (ret != -EOPNOTSUPP)
3679                                                 btrfs_abort_transaction(trans,
3680                                                                         ret);
3681                                         btrfs_end_transaction(trans);
3682                                         goto out;
3683                                 }
3684
3685                                 ret = btrfs_insert_empty_item(trans, root, path,
3686                                                               &new_key, size);
3687                                 if (ret) {
3688                                         btrfs_abort_transaction(trans, ret);
3689                                         btrfs_end_transaction(trans);
3690                                         goto out;
3691                                 }
3692
3693                                 leaf = path->nodes[0];
3694                                 slot = path->slots[0];
3695                                 write_extent_buffer(leaf, buf,
3696                                             btrfs_item_ptr_offset(leaf, slot),
3697                                             size);
3698
3699                                 extent = btrfs_item_ptr(leaf, slot,
3700                                                 struct btrfs_file_extent_item);
3701
3702                                 /* disko == 0 means it's a hole */
3703                                 if (!disko)
3704                                         datao = 0;
3705
3706                                 btrfs_set_file_extent_offset(leaf, extent,
3707                                                              datao);
3708                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3709                                                                 datal);
3710
3711                                 if (disko) {
3712                                         inode_add_bytes(inode, datal);
3713                                         ret = btrfs_inc_extent_ref(trans,
3714                                                         fs_info,
3715                                                         disko, diskl, 0,
3716                                                         root->root_key.objectid,
3717                                                         btrfs_ino(BTRFS_I(inode)),
3718                                                         new_key.offset - datao);
3719                                         if (ret) {
3720                                                 btrfs_abort_transaction(trans,
3721                                                                         ret);
3722                                                 btrfs_end_transaction(trans);
3723                                                 goto out;
3724
3725                                         }
3726                                 }
3727                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3728                                 u64 skip = 0;
3729                                 u64 trim = 0;
3730
3731                                 if (off > key.offset) {
3732                                         skip = off - key.offset;
3733                                         new_key.offset += skip;
3734                                 }
3735
3736                                 if (key.offset + datal > off + len)
3737                                         trim = key.offset + datal - (off + len);
3738
3739                                 if (comp && (skip || trim)) {
3740                                         ret = -EINVAL;
3741                                         btrfs_end_transaction(trans);
3742                                         goto out;
3743                                 }
3744                                 size -= skip + trim;
3745                                 datal -= skip + trim;
3746
3747                                 ret = clone_copy_inline_extent(inode,
3748                                                                trans, path,
3749                                                                &new_key,
3750                                                                drop_start,
3751                                                                datal,
3752                                                                skip, size, buf);
3753                                 if (ret) {
3754                                         if (ret != -EOPNOTSUPP)
3755                                                 btrfs_abort_transaction(trans,
3756                                                                         ret);
3757                                         btrfs_end_transaction(trans);
3758                                         goto out;
3759                                 }
3760                                 leaf = path->nodes[0];
3761                                 slot = path->slots[0];
3762                         }
3763
3764                         /* If we have an implicit hole (NO_HOLES feature). */
3765                         if (drop_start < new_key.offset)
3766                                 clone_update_extent_map(BTRFS_I(inode), trans,
3767                                                 NULL, drop_start,
3768                                                 new_key.offset - drop_start);
3769
3770                         clone_update_extent_map(BTRFS_I(inode), trans,
3771                                         path, 0, 0);
3772
3773                         btrfs_mark_buffer_dirty(leaf);
3774                         btrfs_release_path(path);
3775
3776                         last_dest_end = ALIGN(new_key.offset + datal,
3777                                               fs_info->sectorsize);
3778                         ret = clone_finish_inode_update(trans, inode,
3779                                                         last_dest_end,
3780                                                         destoff, olen,
3781                                                         no_time_update);
3782                         if (ret)
3783                                 goto out;
3784                         if (new_key.offset + datal >= destoff + len)
3785                                 break;
3786                 }
3787                 btrfs_release_path(path);
3788                 key.offset = next_key_min_offset;
3789
3790                 if (fatal_signal_pending(current)) {
3791                         ret = -EINTR;
3792                         goto out;
3793                 }
3794         }
3795         ret = 0;
3796
3797         if (last_dest_end < destoff + len) {
3798                 /*
3799                  * We have an implicit hole (NO_HOLES feature is enabled) that
3800                  * fully or partially overlaps our cloning range at its end.
3801                  */
3802                 btrfs_release_path(path);
3803
3804                 /*
3805                  * 1 - remove extent(s)
3806                  * 1 - inode update
3807                  */
3808                 trans = btrfs_start_transaction(root, 2);
3809                 if (IS_ERR(trans)) {
3810                         ret = PTR_ERR(trans);
3811                         goto out;
3812                 }
3813                 ret = btrfs_drop_extents(trans, root, inode,
3814                                          last_dest_end, destoff + len, 1);
3815                 if (ret) {
3816                         if (ret != -EOPNOTSUPP)
3817                                 btrfs_abort_transaction(trans, ret);
3818                         btrfs_end_transaction(trans);
3819                         goto out;
3820                 }
3821                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3822                                 last_dest_end,
3823                                 destoff + len - last_dest_end);
3824                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3825                                                 destoff, olen, no_time_update);
3826         }
3827
3828 out:
3829         btrfs_free_path(path);
3830         kvfree(buf);
3831         return ret;
3832 }
3833
3834 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3835                                         u64 off, u64 olen, u64 destoff)
3836 {
3837         struct inode *inode = file_inode(file);
3838         struct inode *src = file_inode(file_src);
3839         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3840         struct btrfs_root *root = BTRFS_I(inode)->root;
3841         int ret;
3842         u64 len = olen;
3843         u64 bs = fs_info->sb->s_blocksize;
3844         int same_inode = src == inode;
3845
3846         /*
3847          * TODO:
3848          * - split compressed inline extents.  annoying: we need to
3849          *   decompress into destination's address_space (the file offset
3850          *   may change, so source mapping won't do), then recompress (or
3851          *   otherwise reinsert) a subrange.
3852          *
3853          * - split destination inode's inline extents.  The inline extents can
3854          *   be either compressed or non-compressed.
3855          */
3856
3857         if (btrfs_root_readonly(root))
3858                 return -EROFS;
3859
3860         if (file_src->f_path.mnt != file->f_path.mnt ||
3861             src->i_sb != inode->i_sb)
3862                 return -EXDEV;
3863
3864         /* don't make the dst file partly checksummed */
3865         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3866             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3867                 return -EINVAL;
3868
3869         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3870                 return -EISDIR;
3871
3872         if (!same_inode) {
3873                 btrfs_double_inode_lock(src, inode);
3874         } else {
3875                 inode_lock(src);
3876         }
3877
3878         /* determine range to clone */
3879         ret = -EINVAL;
3880         if (off + len > src->i_size || off + len < off)
3881                 goto out_unlock;
3882         if (len == 0)
3883                 olen = len = src->i_size - off;
3884         /* if we extend to eof, continue to block boundary */
3885         if (off + len == src->i_size)
3886                 len = ALIGN(src->i_size, bs) - off;
3887
3888         if (len == 0) {
3889                 ret = 0;
3890                 goto out_unlock;
3891         }
3892
3893         /* verify the end result is block aligned */
3894         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3895             !IS_ALIGNED(destoff, bs))
3896                 goto out_unlock;
3897
3898         /* verify if ranges are overlapped within the same file */
3899         if (same_inode) {
3900                 if (destoff + len > off && destoff < off + len)
3901                         goto out_unlock;
3902         }
3903
3904         if (destoff > inode->i_size) {
3905                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3906                 if (ret)
3907                         goto out_unlock;
3908         }
3909
3910         /*
3911          * Lock the target range too. Right after we replace the file extent
3912          * items in the fs tree (which now point to the cloned data), we might
3913          * have a worker replace them with extent items relative to a write
3914          * operation that was issued before this clone operation (i.e. confront
3915          * with inode.c:btrfs_finish_ordered_io).
3916          */
3917         if (same_inode) {
3918                 u64 lock_start = min_t(u64, off, destoff);
3919                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3920
3921                 ret = lock_extent_range(src, lock_start, lock_len, true);
3922         } else {
3923                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3924                                                true);
3925         }
3926         ASSERT(ret == 0);
3927         if (WARN_ON(ret)) {
3928                 /* ranges in the io trees already unlocked */
3929                 goto out_unlock;
3930         }
3931
3932         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3933
3934         if (same_inode) {
3935                 u64 lock_start = min_t(u64, off, destoff);
3936                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3937
3938                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3939         } else {
3940                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3941         }
3942         /*
3943          * Truncate page cache pages so that future reads will see the cloned
3944          * data immediately and not the previous data.
3945          */
3946         truncate_inode_pages_range(&inode->i_data,
3947                                 round_down(destoff, PAGE_SIZE),
3948                                 round_up(destoff + len, PAGE_SIZE) - 1);
3949 out_unlock:
3950         if (!same_inode)
3951                 btrfs_double_inode_unlock(src, inode);
3952         else
3953                 inode_unlock(src);
3954         return ret;
3955 }
3956
3957 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3958                 struct file *dst_file, loff_t destoff, u64 len)
3959 {
3960         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3961 }
3962
3963 /*
3964  * there are many ways the trans_start and trans_end ioctls can lead
3965  * to deadlocks.  They should only be used by applications that
3966  * basically own the machine, and have a very in depth understanding
3967  * of all the possible deadlocks and enospc problems.
3968  */
3969 static long btrfs_ioctl_trans_start(struct file *file)
3970 {
3971         struct inode *inode = file_inode(file);
3972         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3973         struct btrfs_root *root = BTRFS_I(inode)->root;
3974         struct btrfs_trans_handle *trans;
3975         struct btrfs_file_private *private;
3976         int ret;
3977         static bool warned = false;
3978
3979         ret = -EPERM;
3980         if (!capable(CAP_SYS_ADMIN))
3981                 goto out;
3982
3983         if (!warned) {
3984                 btrfs_warn(fs_info,
3985                         "Userspace transaction mechanism is considered "
3986                         "deprecated and slated to be removed in 4.17. "
3987                         "If you have a valid use case please "
3988                         "speak up on the mailing list");
3989                 WARN_ON(1);
3990                 warned = true;
3991         }
3992
3993         ret = -EINPROGRESS;
3994         private = file->private_data;
3995         if (private && private->trans)
3996                 goto out;
3997         if (!private) {
3998                 private = kzalloc(sizeof(struct btrfs_file_private),
3999                                   GFP_KERNEL);
4000                 if (!private)
4001                         return -ENOMEM;
4002                 file->private_data = private;
4003         }
4004
4005         ret = -EROFS;
4006         if (btrfs_root_readonly(root))
4007                 goto out;
4008
4009         ret = mnt_want_write_file(file);
4010         if (ret)
4011                 goto out;
4012
4013         atomic_inc(&fs_info->open_ioctl_trans);
4014
4015         ret = -ENOMEM;
4016         trans = btrfs_start_ioctl_transaction(root);
4017         if (IS_ERR(trans))
4018                 goto out_drop;
4019
4020         private->trans = trans;
4021         return 0;
4022
4023 out_drop:
4024         atomic_dec(&fs_info->open_ioctl_trans);
4025         mnt_drop_write_file(file);
4026 out:
4027         return ret;
4028 }
4029
4030 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4031 {
4032         struct inode *inode = file_inode(file);
4033         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4034         struct btrfs_root *root = BTRFS_I(inode)->root;
4035         struct btrfs_root *new_root;
4036         struct btrfs_dir_item *di;
4037         struct btrfs_trans_handle *trans;
4038         struct btrfs_path *path;
4039         struct btrfs_key location;
4040         struct btrfs_disk_key disk_key;
4041         u64 objectid = 0;
4042         u64 dir_id;
4043         int ret;
4044
4045         if (!capable(CAP_SYS_ADMIN))
4046                 return -EPERM;
4047
4048         ret = mnt_want_write_file(file);
4049         if (ret)
4050                 return ret;
4051
4052         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4053                 ret = -EFAULT;
4054                 goto out;
4055         }
4056
4057         if (!objectid)
4058                 objectid = BTRFS_FS_TREE_OBJECTID;
4059
4060         location.objectid = objectid;
4061         location.type = BTRFS_ROOT_ITEM_KEY;
4062         location.offset = (u64)-1;
4063
4064         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4065         if (IS_ERR(new_root)) {
4066                 ret = PTR_ERR(new_root);
4067                 goto out;
4068         }
4069         if (!is_fstree(new_root->objectid)) {
4070                 ret = -ENOENT;
4071                 goto out;
4072         }
4073
4074         path = btrfs_alloc_path();
4075         if (!path) {
4076                 ret = -ENOMEM;
4077                 goto out;
4078         }
4079         path->leave_spinning = 1;
4080
4081         trans = btrfs_start_transaction(root, 1);
4082         if (IS_ERR(trans)) {
4083                 btrfs_free_path(path);
4084                 ret = PTR_ERR(trans);
4085                 goto out;
4086         }
4087
4088         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4089         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4090                                    dir_id, "default", 7, 1);
4091         if (IS_ERR_OR_NULL(di)) {
4092                 btrfs_free_path(path);
4093                 btrfs_end_transaction(trans);
4094                 btrfs_err(fs_info,
4095                           "Umm, you don't have the default diritem, this isn't going to work");
4096                 ret = -ENOENT;
4097                 goto out;
4098         }
4099
4100         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4101         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4102         btrfs_mark_buffer_dirty(path->nodes[0]);
4103         btrfs_free_path(path);
4104
4105         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4106         btrfs_end_transaction(trans);
4107 out:
4108         mnt_drop_write_file(file);
4109         return ret;
4110 }
4111
4112 void btrfs_get_block_group_info(struct list_head *groups_list,
4113                                 struct btrfs_ioctl_space_info *space)
4114 {
4115         struct btrfs_block_group_cache *block_group;
4116
4117         space->total_bytes = 0;
4118         space->used_bytes = 0;
4119         space->flags = 0;
4120         list_for_each_entry(block_group, groups_list, list) {
4121                 space->flags = block_group->flags;
4122                 space->total_bytes += block_group->key.offset;
4123                 space->used_bytes +=
4124                         btrfs_block_group_used(&block_group->item);
4125         }
4126 }
4127
4128 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4129                                    void __user *arg)
4130 {
4131         struct btrfs_ioctl_space_args space_args;
4132         struct btrfs_ioctl_space_info space;
4133         struct btrfs_ioctl_space_info *dest;
4134         struct btrfs_ioctl_space_info *dest_orig;
4135         struct btrfs_ioctl_space_info __user *user_dest;
4136         struct btrfs_space_info *info;
4137         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
4138                        BTRFS_BLOCK_GROUP_SYSTEM,
4139                        BTRFS_BLOCK_GROUP_METADATA,
4140                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
4141         int num_types = 4;
4142         int alloc_size;
4143         int ret = 0;
4144         u64 slot_count = 0;
4145         int i, c;
4146
4147         if (copy_from_user(&space_args,
4148                            (struct btrfs_ioctl_space_args __user *)arg,
4149                            sizeof(space_args)))
4150                 return -EFAULT;
4151
4152         for (i = 0; i < num_types; i++) {
4153                 struct btrfs_space_info *tmp;
4154
4155                 info = NULL;
4156                 rcu_read_lock();
4157                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4158                                         list) {
4159                         if (tmp->flags == types[i]) {
4160                                 info = tmp;
4161                                 break;
4162                         }
4163                 }
4164                 rcu_read_unlock();
4165
4166                 if (!info)
4167                         continue;
4168
4169                 down_read(&info->groups_sem);
4170                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4171                         if (!list_empty(&info->block_groups[c]))
4172                                 slot_count++;
4173                 }
4174                 up_read(&info->groups_sem);
4175         }
4176
4177         /*
4178          * Global block reserve, exported as a space_info
4179          */
4180         slot_count++;
4181
4182         /* space_slots == 0 means they are asking for a count */
4183         if (space_args.space_slots == 0) {
4184                 space_args.total_spaces = slot_count;
4185                 goto out;
4186         }
4187
4188         slot_count = min_t(u64, space_args.space_slots, slot_count);
4189
4190         alloc_size = sizeof(*dest) * slot_count;
4191
4192         /* we generally have at most 6 or so space infos, one for each raid
4193          * level.  So, a whole page should be more than enough for everyone
4194          */
4195         if (alloc_size > PAGE_SIZE)
4196                 return -ENOMEM;
4197
4198         space_args.total_spaces = 0;
4199         dest = kmalloc(alloc_size, GFP_KERNEL);
4200         if (!dest)
4201                 return -ENOMEM;
4202         dest_orig = dest;
4203
4204         /* now we have a buffer to copy into */
4205         for (i = 0; i < num_types; i++) {
4206                 struct btrfs_space_info *tmp;
4207
4208                 if (!slot_count)
4209                         break;
4210
4211                 info = NULL;
4212                 rcu_read_lock();
4213                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4214                                         list) {
4215                         if (tmp->flags == types[i]) {
4216                                 info = tmp;
4217                                 break;
4218                         }
4219                 }
4220                 rcu_read_unlock();
4221
4222                 if (!info)
4223                         continue;
4224                 down_read(&info->groups_sem);
4225                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4226                         if (!list_empty(&info->block_groups[c])) {
4227                                 btrfs_get_block_group_info(
4228                                         &info->block_groups[c], &space);
4229                                 memcpy(dest, &space, sizeof(space));
4230                                 dest++;
4231                                 space_args.total_spaces++;
4232                                 slot_count--;
4233                         }
4234                         if (!slot_count)
4235                                 break;
4236                 }
4237                 up_read(&info->groups_sem);
4238         }
4239
4240         /*
4241          * Add global block reserve
4242          */
4243         if (slot_count) {
4244                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4245
4246                 spin_lock(&block_rsv->lock);
4247                 space.total_bytes = block_rsv->size;
4248                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4249                 spin_unlock(&block_rsv->lock);
4250                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4251                 memcpy(dest, &space, sizeof(space));
4252                 space_args.total_spaces++;
4253         }
4254
4255         user_dest = (struct btrfs_ioctl_space_info __user *)
4256                 (arg + sizeof(struct btrfs_ioctl_space_args));
4257
4258         if (copy_to_user(user_dest, dest_orig, alloc_size))
4259                 ret = -EFAULT;
4260
4261         kfree(dest_orig);
4262 out:
4263         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4264                 ret = -EFAULT;
4265
4266         return ret;
4267 }
4268
4269 /*
4270  * there are many ways the trans_start and trans_end ioctls can lead
4271  * to deadlocks.  They should only be used by applications that
4272  * basically own the machine, and have a very in depth understanding
4273  * of all the possible deadlocks and enospc problems.
4274  */
4275 long btrfs_ioctl_trans_end(struct file *file)
4276 {
4277         struct inode *inode = file_inode(file);
4278         struct btrfs_root *root = BTRFS_I(inode)->root;
4279         struct btrfs_file_private *private = file->private_data;
4280
4281         if (!private || !private->trans)
4282                 return -EINVAL;
4283
4284         btrfs_end_transaction(private->trans);
4285         private->trans = NULL;
4286
4287         atomic_dec(&root->fs_info->open_ioctl_trans);
4288
4289         mnt_drop_write_file(file);
4290         return 0;
4291 }
4292
4293 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4294                                             void __user *argp)
4295 {
4296         struct btrfs_trans_handle *trans;
4297         u64 transid;
4298         int ret;
4299
4300         trans = btrfs_attach_transaction_barrier(root);
4301         if (IS_ERR(trans)) {
4302                 if (PTR_ERR(trans) != -ENOENT)
4303                         return PTR_ERR(trans);
4304
4305                 /* No running transaction, don't bother */
4306                 transid = root->fs_info->last_trans_committed;
4307                 goto out;
4308         }
4309         transid = trans->transid;
4310         ret = btrfs_commit_transaction_async(trans, 0);
4311         if (ret) {
4312                 btrfs_end_transaction(trans);
4313                 return ret;
4314         }
4315 out:
4316         if (argp)
4317                 if (copy_to_user(argp, &transid, sizeof(transid)))
4318                         return -EFAULT;
4319         return 0;
4320 }
4321
4322 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4323                                            void __user *argp)
4324 {
4325         u64 transid;
4326
4327         if (argp) {
4328                 if (copy_from_user(&transid, argp, sizeof(transid)))
4329                         return -EFAULT;
4330         } else {
4331                 transid = 0;  /* current trans */
4332         }
4333         return btrfs_wait_for_commit(fs_info, transid);
4334 }
4335
4336 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4337 {
4338         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4339         struct btrfs_ioctl_scrub_args *sa;
4340         int ret;
4341
4342         if (!capable(CAP_SYS_ADMIN))
4343                 return -EPERM;
4344
4345         sa = memdup_user(arg, sizeof(*sa));
4346         if (IS_ERR(sa))
4347                 return PTR_ERR(sa);
4348
4349         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4350                 ret = mnt_want_write_file(file);
4351                 if (ret)
4352                         goto out;
4353         }
4354
4355         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4356                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4357                               0);
4358
4359         if (copy_to_user(arg, sa, sizeof(*sa)))
4360                 ret = -EFAULT;
4361
4362         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4363                 mnt_drop_write_file(file);
4364 out:
4365         kfree(sa);
4366         return ret;
4367 }
4368
4369 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4370 {
4371         if (!capable(CAP_SYS_ADMIN))
4372                 return -EPERM;
4373
4374         return btrfs_scrub_cancel(fs_info);
4375 }
4376
4377 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4378                                        void __user *arg)
4379 {
4380         struct btrfs_ioctl_scrub_args *sa;
4381         int ret;
4382
4383         if (!capable(CAP_SYS_ADMIN))
4384                 return -EPERM;
4385
4386         sa = memdup_user(arg, sizeof(*sa));
4387         if (IS_ERR(sa))
4388                 return PTR_ERR(sa);
4389
4390         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4391
4392         if (copy_to_user(arg, sa, sizeof(*sa)))
4393                 ret = -EFAULT;
4394
4395         kfree(sa);
4396         return ret;
4397 }
4398
4399 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4400                                       void __user *arg)
4401 {
4402         struct btrfs_ioctl_get_dev_stats *sa;
4403         int ret;
4404
4405         sa = memdup_user(arg, sizeof(*sa));
4406         if (IS_ERR(sa))
4407                 return PTR_ERR(sa);
4408
4409         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4410                 kfree(sa);
4411                 return -EPERM;
4412         }
4413
4414         ret = btrfs_get_dev_stats(fs_info, sa);
4415
4416         if (copy_to_user(arg, sa, sizeof(*sa)))
4417                 ret = -EFAULT;
4418
4419         kfree(sa);
4420         return ret;
4421 }
4422
4423 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4424                                     void __user *arg)
4425 {
4426         struct btrfs_ioctl_dev_replace_args *p;
4427         int ret;
4428
4429         if (!capable(CAP_SYS_ADMIN))
4430                 return -EPERM;
4431
4432         p = memdup_user(arg, sizeof(*p));
4433         if (IS_ERR(p))
4434                 return PTR_ERR(p);
4435
4436         switch (p->cmd) {
4437         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4438                 if (sb_rdonly(fs_info->sb)) {
4439                         ret = -EROFS;
4440                         goto out;
4441                 }
4442                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4443                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4444                 } else {
4445                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4446                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4447                 }
4448                 break;
4449         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4450                 btrfs_dev_replace_status(fs_info, p);
4451                 ret = 0;
4452                 break;
4453         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4454                 ret = btrfs_dev_replace_cancel(fs_info, p);
4455                 break;
4456         default:
4457                 ret = -EINVAL;
4458                 break;
4459         }
4460
4461         if (copy_to_user(arg, p, sizeof(*p)))
4462                 ret = -EFAULT;
4463 out:
4464         kfree(p);
4465         return ret;
4466 }
4467
4468 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4469 {
4470         int ret = 0;
4471         int i;
4472         u64 rel_ptr;
4473         int size;
4474         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4475         struct inode_fs_paths *ipath = NULL;
4476         struct btrfs_path *path;
4477
4478         if (!capable(CAP_DAC_READ_SEARCH))
4479                 return -EPERM;
4480
4481         path = btrfs_alloc_path();
4482         if (!path) {
4483                 ret = -ENOMEM;
4484                 goto out;
4485         }
4486
4487         ipa = memdup_user(arg, sizeof(*ipa));
4488         if (IS_ERR(ipa)) {
4489                 ret = PTR_ERR(ipa);
4490                 ipa = NULL;
4491                 goto out;
4492         }
4493
4494         size = min_t(u32, ipa->size, 4096);
4495         ipath = init_ipath(size, root, path);
4496         if (IS_ERR(ipath)) {
4497                 ret = PTR_ERR(ipath);
4498                 ipath = NULL;
4499                 goto out;
4500         }
4501
4502         ret = paths_from_inode(ipa->inum, ipath);
4503         if (ret < 0)
4504                 goto out;
4505
4506         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4507                 rel_ptr = ipath->fspath->val[i] -
4508                           (u64)(unsigned long)ipath->fspath->val;
4509                 ipath->fspath->val[i] = rel_ptr;
4510         }
4511
4512         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
4513                            (void *)(unsigned long)ipath->fspath, size);
4514         if (ret) {
4515                 ret = -EFAULT;
4516                 goto out;
4517         }
4518
4519 out:
4520         btrfs_free_path(path);
4521         free_ipath(ipath);
4522         kfree(ipa);
4523
4524         return ret;
4525 }
4526
4527 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4528 {
4529         struct btrfs_data_container *inodes = ctx;
4530         const size_t c = 3 * sizeof(u64);
4531
4532         if (inodes->bytes_left >= c) {
4533                 inodes->bytes_left -= c;
4534                 inodes->val[inodes->elem_cnt] = inum;
4535                 inodes->val[inodes->elem_cnt + 1] = offset;
4536                 inodes->val[inodes->elem_cnt + 2] = root;
4537                 inodes->elem_cnt += 3;
4538         } else {
4539                 inodes->bytes_missing += c - inodes->bytes_left;
4540                 inodes->bytes_left = 0;
4541                 inodes->elem_missed += 3;
4542         }
4543
4544         return 0;
4545 }
4546
4547 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4548                                         void __user *arg)
4549 {
4550         int ret = 0;
4551         int size;
4552         struct btrfs_ioctl_logical_ino_args *loi;
4553         struct btrfs_data_container *inodes = NULL;
4554         struct btrfs_path *path = NULL;
4555
4556         if (!capable(CAP_SYS_ADMIN))
4557                 return -EPERM;
4558
4559         loi = memdup_user(arg, sizeof(*loi));
4560         if (IS_ERR(loi))
4561                 return PTR_ERR(loi);
4562
4563         path = btrfs_alloc_path();
4564         if (!path) {
4565                 ret = -ENOMEM;
4566                 goto out;
4567         }
4568
4569         size = min_t(u32, loi->size, SZ_64K);
4570         inodes = init_data_container(size);
4571         if (IS_ERR(inodes)) {
4572                 ret = PTR_ERR(inodes);
4573                 inodes = NULL;
4574                 goto out;
4575         }
4576
4577         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4578                                           build_ino_list, inodes);
4579         if (ret == -EINVAL)
4580                 ret = -ENOENT;
4581         if (ret < 0)
4582                 goto out;
4583
4584         ret = copy_to_user((void *)(unsigned long)loi->inodes,
4585                            (void *)(unsigned long)inodes, size);
4586         if (ret)
4587                 ret = -EFAULT;
4588
4589 out:
4590         btrfs_free_path(path);
4591         kvfree(inodes);
4592         kfree(loi);
4593
4594         return ret;
4595 }
4596
4597 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4598                                struct btrfs_ioctl_balance_args *bargs)
4599 {
4600         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4601
4602         bargs->flags = bctl->flags;
4603
4604         if (atomic_read(&fs_info->balance_running))
4605                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4606         if (atomic_read(&fs_info->balance_pause_req))
4607                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4608         if (atomic_read(&fs_info->balance_cancel_req))
4609                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4610
4611         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4612         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4613         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4614
4615         if (lock) {
4616                 spin_lock(&fs_info->balance_lock);
4617                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4618                 spin_unlock(&fs_info->balance_lock);
4619         } else {
4620                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4621         }
4622 }
4623
4624 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4625 {
4626         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4627         struct btrfs_fs_info *fs_info = root->fs_info;
4628         struct btrfs_ioctl_balance_args *bargs;
4629         struct btrfs_balance_control *bctl;
4630         bool need_unlock; /* for mut. excl. ops lock */
4631         int ret;
4632
4633         if (!capable(CAP_SYS_ADMIN))
4634                 return -EPERM;
4635
4636         ret = mnt_want_write_file(file);
4637         if (ret)
4638                 return ret;
4639
4640 again:
4641         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4642                 mutex_lock(&fs_info->volume_mutex);
4643                 mutex_lock(&fs_info->balance_mutex);
4644                 need_unlock = true;
4645                 goto locked;
4646         }
4647
4648         /*
4649          * mut. excl. ops lock is locked.  Three possibilities:
4650          *   (1) some other op is running
4651          *   (2) balance is running
4652          *   (3) balance is paused -- special case (think resume)
4653          */
4654         mutex_lock(&fs_info->balance_mutex);
4655         if (fs_info->balance_ctl) {
4656                 /* this is either (2) or (3) */
4657                 if (!atomic_read(&fs_info->balance_running)) {
4658                         mutex_unlock(&fs_info->balance_mutex);
4659                         if (!mutex_trylock(&fs_info->volume_mutex))
4660                                 goto again;
4661                         mutex_lock(&fs_info->balance_mutex);
4662
4663                         if (fs_info->balance_ctl &&
4664                             !atomic_read(&fs_info->balance_running)) {
4665                                 /* this is (3) */
4666                                 need_unlock = false;
4667                                 goto locked;
4668                         }
4669
4670                         mutex_unlock(&fs_info->balance_mutex);
4671                         mutex_unlock(&fs_info->volume_mutex);
4672                         goto again;
4673                 } else {
4674                         /* this is (2) */
4675                         mutex_unlock(&fs_info->balance_mutex);
4676                         ret = -EINPROGRESS;
4677                         goto out;
4678                 }
4679         } else {
4680                 /* this is (1) */
4681                 mutex_unlock(&fs_info->balance_mutex);
4682                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4683                 goto out;
4684         }
4685
4686 locked:
4687         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4688
4689         if (arg) {
4690                 bargs = memdup_user(arg, sizeof(*bargs));
4691                 if (IS_ERR(bargs)) {
4692                         ret = PTR_ERR(bargs);
4693                         goto out_unlock;
4694                 }
4695
4696                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4697                         if (!fs_info->balance_ctl) {
4698                                 ret = -ENOTCONN;
4699                                 goto out_bargs;
4700                         }
4701
4702                         bctl = fs_info->balance_ctl;
4703                         spin_lock(&fs_info->balance_lock);
4704                         bctl->flags |= BTRFS_BALANCE_RESUME;
4705                         spin_unlock(&fs_info->balance_lock);
4706
4707                         goto do_balance;
4708                 }
4709         } else {
4710                 bargs = NULL;
4711         }
4712
4713         if (fs_info->balance_ctl) {
4714                 ret = -EINPROGRESS;
4715                 goto out_bargs;
4716         }
4717
4718         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4719         if (!bctl) {
4720                 ret = -ENOMEM;
4721                 goto out_bargs;
4722         }
4723
4724         bctl->fs_info = fs_info;
4725         if (arg) {
4726                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4727                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4728                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4729
4730                 bctl->flags = bargs->flags;
4731         } else {
4732                 /* balance everything - no filters */
4733                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4734         }
4735
4736         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4737                 ret = -EINVAL;
4738                 goto out_bctl;
4739         }
4740
4741 do_balance:
4742         /*
4743          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4744          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4745          * or, if restriper was paused all the way until unmount, in
4746          * free_fs_info.  The flag is cleared in __cancel_balance.
4747          */
4748         need_unlock = false;
4749
4750         ret = btrfs_balance(bctl, bargs);
4751         bctl = NULL;
4752
4753         if (arg) {
4754                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4755                         ret = -EFAULT;
4756         }
4757
4758 out_bctl:
4759         kfree(bctl);
4760 out_bargs:
4761         kfree(bargs);
4762 out_unlock:
4763         mutex_unlock(&fs_info->balance_mutex);
4764         mutex_unlock(&fs_info->volume_mutex);
4765         if (need_unlock)
4766                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4767 out:
4768         mnt_drop_write_file(file);
4769         return ret;
4770 }
4771
4772 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4773 {
4774         if (!capable(CAP_SYS_ADMIN))
4775                 return -EPERM;
4776
4777         switch (cmd) {
4778         case BTRFS_BALANCE_CTL_PAUSE:
4779                 return btrfs_pause_balance(fs_info);
4780         case BTRFS_BALANCE_CTL_CANCEL:
4781                 return btrfs_cancel_balance(fs_info);
4782         }
4783
4784         return -EINVAL;
4785 }
4786
4787 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4788                                          void __user *arg)
4789 {
4790         struct btrfs_ioctl_balance_args *bargs;
4791         int ret = 0;
4792
4793         if (!capable(CAP_SYS_ADMIN))
4794                 return -EPERM;
4795
4796         mutex_lock(&fs_info->balance_mutex);
4797         if (!fs_info->balance_ctl) {
4798                 ret = -ENOTCONN;
4799                 goto out;
4800         }
4801
4802         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4803         if (!bargs) {
4804                 ret = -ENOMEM;
4805                 goto out;
4806         }
4807
4808         update_ioctl_balance_args(fs_info, 1, bargs);
4809
4810         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4811                 ret = -EFAULT;
4812
4813         kfree(bargs);
4814 out:
4815         mutex_unlock(&fs_info->balance_mutex);
4816         return ret;
4817 }
4818
4819 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4820 {
4821         struct inode *inode = file_inode(file);
4822         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4823         struct btrfs_ioctl_quota_ctl_args *sa;
4824         struct btrfs_trans_handle *trans = NULL;
4825         int ret;
4826         int err;
4827
4828         if (!capable(CAP_SYS_ADMIN))
4829                 return -EPERM;
4830
4831         ret = mnt_want_write_file(file);
4832         if (ret)
4833                 return ret;
4834
4835         sa = memdup_user(arg, sizeof(*sa));
4836         if (IS_ERR(sa)) {
4837                 ret = PTR_ERR(sa);
4838                 goto drop_write;
4839         }
4840
4841         down_write(&fs_info->subvol_sem);
4842         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4843         if (IS_ERR(trans)) {
4844                 ret = PTR_ERR(trans);
4845                 goto out;
4846         }
4847
4848         switch (sa->cmd) {
4849         case BTRFS_QUOTA_CTL_ENABLE:
4850                 ret = btrfs_quota_enable(trans, fs_info);
4851                 break;
4852         case BTRFS_QUOTA_CTL_DISABLE:
4853                 ret = btrfs_quota_disable(trans, fs_info);
4854                 break;
4855         default:
4856                 ret = -EINVAL;
4857                 break;
4858         }
4859
4860         err = btrfs_commit_transaction(trans);
4861         if (err && !ret)
4862                 ret = err;
4863 out:
4864         kfree(sa);
4865         up_write(&fs_info->subvol_sem);
4866 drop_write:
4867         mnt_drop_write_file(file);
4868         return ret;
4869 }
4870
4871 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4872 {
4873         struct inode *inode = file_inode(file);
4874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4875         struct btrfs_root *root = BTRFS_I(inode)->root;
4876         struct btrfs_ioctl_qgroup_assign_args *sa;
4877         struct btrfs_trans_handle *trans;
4878         int ret;
4879         int err;
4880
4881         if (!capable(CAP_SYS_ADMIN))
4882                 return -EPERM;
4883
4884         ret = mnt_want_write_file(file);
4885         if (ret)
4886                 return ret;
4887
4888         sa = memdup_user(arg, sizeof(*sa));
4889         if (IS_ERR(sa)) {
4890                 ret = PTR_ERR(sa);
4891                 goto drop_write;
4892         }
4893
4894         trans = btrfs_join_transaction(root);
4895         if (IS_ERR(trans)) {
4896                 ret = PTR_ERR(trans);
4897                 goto out;
4898         }
4899
4900         if (sa->assign) {
4901                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4902                                                 sa->src, sa->dst);
4903         } else {
4904                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4905                                                 sa->src, sa->dst);
4906         }
4907
4908         /* update qgroup status and info */
4909         err = btrfs_run_qgroups(trans, fs_info);
4910         if (err < 0)
4911                 btrfs_handle_fs_error(fs_info, err,
4912                                       "failed to update qgroup status and info");
4913         err = btrfs_end_transaction(trans);
4914         if (err && !ret)
4915                 ret = err;
4916
4917 out:
4918         kfree(sa);
4919 drop_write:
4920         mnt_drop_write_file(file);
4921         return ret;
4922 }
4923
4924 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4925 {
4926         struct inode *inode = file_inode(file);
4927         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4928         struct btrfs_root *root = BTRFS_I(inode)->root;
4929         struct btrfs_ioctl_qgroup_create_args *sa;
4930         struct btrfs_trans_handle *trans;
4931         int ret;
4932         int err;
4933
4934         if (!capable(CAP_SYS_ADMIN))
4935                 return -EPERM;
4936
4937         ret = mnt_want_write_file(file);
4938         if (ret)
4939                 return ret;
4940
4941         sa = memdup_user(arg, sizeof(*sa));
4942         if (IS_ERR(sa)) {
4943                 ret = PTR_ERR(sa);
4944                 goto drop_write;
4945         }
4946
4947         if (!sa->qgroupid) {
4948                 ret = -EINVAL;
4949                 goto out;
4950         }
4951
4952         trans = btrfs_join_transaction(root);
4953         if (IS_ERR(trans)) {
4954                 ret = PTR_ERR(trans);
4955                 goto out;
4956         }
4957
4958         if (sa->create) {
4959                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4960         } else {
4961                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4962         }
4963
4964         err = btrfs_end_transaction(trans);
4965         if (err && !ret)
4966                 ret = err;
4967
4968 out:
4969         kfree(sa);
4970 drop_write:
4971         mnt_drop_write_file(file);
4972         return ret;
4973 }
4974
4975 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4976 {
4977         struct inode *inode = file_inode(file);
4978         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4979         struct btrfs_root *root = BTRFS_I(inode)->root;
4980         struct btrfs_ioctl_qgroup_limit_args *sa;
4981         struct btrfs_trans_handle *trans;
4982         int ret;
4983         int err;
4984         u64 qgroupid;
4985
4986         if (!capable(CAP_SYS_ADMIN))
4987                 return -EPERM;
4988
4989         ret = mnt_want_write_file(file);
4990         if (ret)
4991                 return ret;
4992
4993         sa = memdup_user(arg, sizeof(*sa));
4994         if (IS_ERR(sa)) {
4995                 ret = PTR_ERR(sa);
4996                 goto drop_write;
4997         }
4998
4999         trans = btrfs_join_transaction(root);
5000         if (IS_ERR(trans)) {
5001                 ret = PTR_ERR(trans);
5002                 goto out;
5003         }
5004
5005         qgroupid = sa->qgroupid;
5006         if (!qgroupid) {
5007                 /* take the current subvol as qgroup */
5008                 qgroupid = root->root_key.objectid;
5009         }
5010
5011         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5012
5013         err = btrfs_end_transaction(trans);
5014         if (err && !ret)
5015                 ret = err;
5016
5017 out:
5018         kfree(sa);
5019 drop_write:
5020         mnt_drop_write_file(file);
5021         return ret;
5022 }
5023
5024 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5025 {
5026         struct inode *inode = file_inode(file);
5027         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5028         struct btrfs_ioctl_quota_rescan_args *qsa;
5029         int ret;
5030
5031         if (!capable(CAP_SYS_ADMIN))
5032                 return -EPERM;
5033
5034         ret = mnt_want_write_file(file);
5035         if (ret)
5036                 return ret;
5037
5038         qsa = memdup_user(arg, sizeof(*qsa));
5039         if (IS_ERR(qsa)) {
5040                 ret = PTR_ERR(qsa);
5041                 goto drop_write;
5042         }
5043
5044         if (qsa->flags) {
5045                 ret = -EINVAL;
5046                 goto out;
5047         }
5048
5049         ret = btrfs_qgroup_rescan(fs_info);
5050
5051 out:
5052         kfree(qsa);
5053 drop_write:
5054         mnt_drop_write_file(file);
5055         return ret;
5056 }
5057
5058 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5059 {
5060         struct inode *inode = file_inode(file);
5061         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5062         struct btrfs_ioctl_quota_rescan_args *qsa;
5063         int ret = 0;
5064
5065         if (!capable(CAP_SYS_ADMIN))
5066                 return -EPERM;
5067
5068         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5069         if (!qsa)
5070                 return -ENOMEM;
5071
5072         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5073                 qsa->flags = 1;
5074                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5075         }
5076
5077         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5078                 ret = -EFAULT;
5079
5080         kfree(qsa);
5081         return ret;
5082 }
5083
5084 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5085 {
5086         struct inode *inode = file_inode(file);
5087         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5088
5089         if (!capable(CAP_SYS_ADMIN))
5090                 return -EPERM;
5091
5092         return btrfs_qgroup_wait_for_completion(fs_info, true);
5093 }
5094
5095 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5096                                             struct btrfs_ioctl_received_subvol_args *sa)
5097 {
5098         struct inode *inode = file_inode(file);
5099         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5100         struct btrfs_root *root = BTRFS_I(inode)->root;
5101         struct btrfs_root_item *root_item = &root->root_item;
5102         struct btrfs_trans_handle *trans;
5103         struct timespec ct = current_time(inode);
5104         int ret = 0;
5105         int received_uuid_changed;
5106
5107         if (!inode_owner_or_capable(inode))
5108                 return -EPERM;
5109
5110         ret = mnt_want_write_file(file);
5111         if (ret < 0)
5112                 return ret;
5113
5114         down_write(&fs_info->subvol_sem);
5115
5116         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5117                 ret = -EINVAL;
5118                 goto out;
5119         }
5120
5121         if (btrfs_root_readonly(root)) {
5122                 ret = -EROFS;
5123                 goto out;
5124         }
5125
5126         /*
5127          * 1 - root item
5128          * 2 - uuid items (received uuid + subvol uuid)
5129          */
5130         trans = btrfs_start_transaction(root, 3);
5131         if (IS_ERR(trans)) {
5132                 ret = PTR_ERR(trans);
5133                 trans = NULL;
5134                 goto out;
5135         }
5136
5137         sa->rtransid = trans->transid;
5138         sa->rtime.sec = ct.tv_sec;
5139         sa->rtime.nsec = ct.tv_nsec;
5140
5141         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5142                                        BTRFS_UUID_SIZE);
5143         if (received_uuid_changed &&
5144             !btrfs_is_empty_uuid(root_item->received_uuid))
5145                 btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5146                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5147                                     root->root_key.objectid);
5148         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5149         btrfs_set_root_stransid(root_item, sa->stransid);
5150         btrfs_set_root_rtransid(root_item, sa->rtransid);
5151         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5152         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5153         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5154         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5155
5156         ret = btrfs_update_root(trans, fs_info->tree_root,
5157                                 &root->root_key, &root->root_item);
5158         if (ret < 0) {
5159                 btrfs_end_transaction(trans);
5160                 goto out;
5161         }
5162         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5163                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5164                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5165                                           root->root_key.objectid);
5166                 if (ret < 0 && ret != -EEXIST) {
5167                         btrfs_abort_transaction(trans, ret);
5168                         goto out;
5169                 }
5170         }
5171         ret = btrfs_commit_transaction(trans);
5172         if (ret < 0) {
5173                 btrfs_abort_transaction(trans, ret);
5174                 goto out;
5175         }
5176
5177 out:
5178         up_write(&fs_info->subvol_sem);
5179         mnt_drop_write_file(file);
5180         return ret;
5181 }
5182
5183 #ifdef CONFIG_64BIT
5184 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5185                                                 void __user *arg)
5186 {
5187         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5188         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5189         int ret = 0;
5190
5191         args32 = memdup_user(arg, sizeof(*args32));
5192         if (IS_ERR(args32))
5193                 return PTR_ERR(args32);
5194
5195         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5196         if (!args64) {
5197                 ret = -ENOMEM;
5198                 goto out;
5199         }
5200
5201         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5202         args64->stransid = args32->stransid;
5203         args64->rtransid = args32->rtransid;
5204         args64->stime.sec = args32->stime.sec;
5205         args64->stime.nsec = args32->stime.nsec;
5206         args64->rtime.sec = args32->rtime.sec;
5207         args64->rtime.nsec = args32->rtime.nsec;
5208         args64->flags = args32->flags;
5209
5210         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5211         if (ret)
5212                 goto out;
5213
5214         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5215         args32->stransid = args64->stransid;
5216         args32->rtransid = args64->rtransid;
5217         args32->stime.sec = args64->stime.sec;
5218         args32->stime.nsec = args64->stime.nsec;
5219         args32->rtime.sec = args64->rtime.sec;
5220         args32->rtime.nsec = args64->rtime.nsec;
5221         args32->flags = args64->flags;
5222
5223         ret = copy_to_user(arg, args32, sizeof(*args32));
5224         if (ret)
5225                 ret = -EFAULT;
5226
5227 out:
5228         kfree(args32);
5229         kfree(args64);
5230         return ret;
5231 }
5232 #endif
5233
5234 static long btrfs_ioctl_set_received_subvol(struct file *file,
5235                                             void __user *arg)
5236 {
5237         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5238         int ret = 0;
5239
5240         sa = memdup_user(arg, sizeof(*sa));
5241         if (IS_ERR(sa))
5242                 return PTR_ERR(sa);
5243
5244         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5245
5246         if (ret)
5247                 goto out;
5248
5249         ret = copy_to_user(arg, sa, sizeof(*sa));
5250         if (ret)
5251                 ret = -EFAULT;
5252
5253 out:
5254         kfree(sa);
5255         return ret;
5256 }
5257
5258 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5259 {
5260         struct inode *inode = file_inode(file);
5261         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5262         size_t len;
5263         int ret;
5264         char label[BTRFS_LABEL_SIZE];
5265
5266         spin_lock(&fs_info->super_lock);
5267         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5268         spin_unlock(&fs_info->super_lock);
5269
5270         len = strnlen(label, BTRFS_LABEL_SIZE);
5271
5272         if (len == BTRFS_LABEL_SIZE) {
5273                 btrfs_warn(fs_info,
5274                            "label is too long, return the first %zu bytes",
5275                            --len);
5276         }
5277
5278         ret = copy_to_user(arg, label, len);
5279
5280         return ret ? -EFAULT : 0;
5281 }
5282
5283 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5284 {
5285         struct inode *inode = file_inode(file);
5286         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5287         struct btrfs_root *root = BTRFS_I(inode)->root;
5288         struct btrfs_super_block *super_block = fs_info->super_copy;
5289         struct btrfs_trans_handle *trans;
5290         char label[BTRFS_LABEL_SIZE];
5291         int ret;
5292
5293         if (!capable(CAP_SYS_ADMIN))
5294                 return -EPERM;
5295
5296         if (copy_from_user(label, arg, sizeof(label)))
5297                 return -EFAULT;
5298
5299         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5300                 btrfs_err(fs_info,
5301                           "unable to set label with more than %d bytes",
5302                           BTRFS_LABEL_SIZE - 1);
5303                 return -EINVAL;
5304         }
5305
5306         ret = mnt_want_write_file(file);
5307         if (ret)
5308                 return ret;
5309
5310         trans = btrfs_start_transaction(root, 0);
5311         if (IS_ERR(trans)) {
5312                 ret = PTR_ERR(trans);
5313                 goto out_unlock;
5314         }
5315
5316         spin_lock(&fs_info->super_lock);
5317         strcpy(super_block->label, label);
5318         spin_unlock(&fs_info->super_lock);
5319         ret = btrfs_commit_transaction(trans);
5320
5321 out_unlock:
5322         mnt_drop_write_file(file);
5323         return ret;
5324 }
5325
5326 #define INIT_FEATURE_FLAGS(suffix) \
5327         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5328           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5329           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5330
5331 int btrfs_ioctl_get_supported_features(void __user *arg)
5332 {
5333         static const struct btrfs_ioctl_feature_flags features[3] = {
5334                 INIT_FEATURE_FLAGS(SUPP),
5335                 INIT_FEATURE_FLAGS(SAFE_SET),
5336                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5337         };
5338
5339         if (copy_to_user(arg, &features, sizeof(features)))
5340                 return -EFAULT;
5341
5342         return 0;
5343 }
5344
5345 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5346 {
5347         struct inode *inode = file_inode(file);
5348         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5349         struct btrfs_super_block *super_block = fs_info->super_copy;
5350         struct btrfs_ioctl_feature_flags features;
5351
5352         features.compat_flags = btrfs_super_compat_flags(super_block);
5353         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5354         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5355
5356         if (copy_to_user(arg, &features, sizeof(features)))
5357                 return -EFAULT;
5358
5359         return 0;
5360 }
5361
5362 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5363                               enum btrfs_feature_set set,
5364                               u64 change_mask, u64 flags, u64 supported_flags,
5365                               u64 safe_set, u64 safe_clear)
5366 {
5367         const char *type = btrfs_feature_set_names[set];
5368         char *names;
5369         u64 disallowed, unsupported;
5370         u64 set_mask = flags & change_mask;
5371         u64 clear_mask = ~flags & change_mask;
5372
5373         unsupported = set_mask & ~supported_flags;
5374         if (unsupported) {
5375                 names = btrfs_printable_features(set, unsupported);
5376                 if (names) {
5377                         btrfs_warn(fs_info,
5378                                    "this kernel does not support the %s feature bit%s",
5379                                    names, strchr(names, ',') ? "s" : "");
5380                         kfree(names);
5381                 } else
5382                         btrfs_warn(fs_info,
5383                                    "this kernel does not support %s bits 0x%llx",
5384                                    type, unsupported);
5385                 return -EOPNOTSUPP;
5386         }
5387
5388         disallowed = set_mask & ~safe_set;
5389         if (disallowed) {
5390                 names = btrfs_printable_features(set, disallowed);
5391                 if (names) {
5392                         btrfs_warn(fs_info,
5393                                    "can't set the %s feature bit%s while mounted",
5394                                    names, strchr(names, ',') ? "s" : "");
5395                         kfree(names);
5396                 } else
5397                         btrfs_warn(fs_info,
5398                                    "can't set %s bits 0x%llx while mounted",
5399                                    type, disallowed);
5400                 return -EPERM;
5401         }
5402
5403         disallowed = clear_mask & ~safe_clear;
5404         if (disallowed) {
5405                 names = btrfs_printable_features(set, disallowed);
5406                 if (names) {
5407                         btrfs_warn(fs_info,
5408                                    "can't clear the %s feature bit%s while mounted",
5409                                    names, strchr(names, ',') ? "s" : "");
5410                         kfree(names);
5411                 } else
5412                         btrfs_warn(fs_info,
5413                                    "can't clear %s bits 0x%llx while mounted",
5414                                    type, disallowed);
5415                 return -EPERM;
5416         }
5417
5418         return 0;
5419 }
5420
5421 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5422 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5423                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5424                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5425                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5426
5427 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5428 {
5429         struct inode *inode = file_inode(file);
5430         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5431         struct btrfs_root *root = BTRFS_I(inode)->root;
5432         struct btrfs_super_block *super_block = fs_info->super_copy;
5433         struct btrfs_ioctl_feature_flags flags[2];
5434         struct btrfs_trans_handle *trans;
5435         u64 newflags;
5436         int ret;
5437
5438         if (!capable(CAP_SYS_ADMIN))
5439                 return -EPERM;
5440
5441         if (copy_from_user(flags, arg, sizeof(flags)))
5442                 return -EFAULT;
5443
5444         /* Nothing to do */
5445         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5446             !flags[0].incompat_flags)
5447                 return 0;
5448
5449         ret = check_feature(fs_info, flags[0].compat_flags,
5450                             flags[1].compat_flags, COMPAT);
5451         if (ret)
5452                 return ret;
5453
5454         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5455                             flags[1].compat_ro_flags, COMPAT_RO);
5456         if (ret)
5457                 return ret;
5458
5459         ret = check_feature(fs_info, flags[0].incompat_flags,
5460                             flags[1].incompat_flags, INCOMPAT);
5461         if (ret)
5462                 return ret;
5463
5464         ret = mnt_want_write_file(file);
5465         if (ret)
5466                 return ret;
5467
5468         trans = btrfs_start_transaction(root, 0);
5469         if (IS_ERR(trans)) {
5470                 ret = PTR_ERR(trans);
5471                 goto out_drop_write;
5472         }
5473
5474         spin_lock(&fs_info->super_lock);
5475         newflags = btrfs_super_compat_flags(super_block);
5476         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5477         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5478         btrfs_set_super_compat_flags(super_block, newflags);
5479
5480         newflags = btrfs_super_compat_ro_flags(super_block);
5481         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5482         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5483         btrfs_set_super_compat_ro_flags(super_block, newflags);
5484
5485         newflags = btrfs_super_incompat_flags(super_block);
5486         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5487         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5488         btrfs_set_super_incompat_flags(super_block, newflags);
5489         spin_unlock(&fs_info->super_lock);
5490
5491         ret = btrfs_commit_transaction(trans);
5492 out_drop_write:
5493         mnt_drop_write_file(file);
5494
5495         return ret;
5496 }
5497
5498 long btrfs_ioctl(struct file *file, unsigned int
5499                 cmd, unsigned long arg)
5500 {
5501         struct inode *inode = file_inode(file);
5502         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5503         struct btrfs_root *root = BTRFS_I(inode)->root;
5504         void __user *argp = (void __user *)arg;
5505
5506         switch (cmd) {
5507         case FS_IOC_GETFLAGS:
5508                 return btrfs_ioctl_getflags(file, argp);
5509         case FS_IOC_SETFLAGS:
5510                 return btrfs_ioctl_setflags(file, argp);
5511         case FS_IOC_GETVERSION:
5512                 return btrfs_ioctl_getversion(file, argp);
5513         case FITRIM:
5514                 return btrfs_ioctl_fitrim(file, argp);
5515         case BTRFS_IOC_SNAP_CREATE:
5516                 return btrfs_ioctl_snap_create(file, argp, 0);
5517         case BTRFS_IOC_SNAP_CREATE_V2:
5518                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5519         case BTRFS_IOC_SUBVOL_CREATE:
5520                 return btrfs_ioctl_snap_create(file, argp, 1);
5521         case BTRFS_IOC_SUBVOL_CREATE_V2:
5522                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5523         case BTRFS_IOC_SNAP_DESTROY:
5524                 return btrfs_ioctl_snap_destroy(file, argp);
5525         case BTRFS_IOC_SUBVOL_GETFLAGS:
5526                 return btrfs_ioctl_subvol_getflags(file, argp);
5527         case BTRFS_IOC_SUBVOL_SETFLAGS:
5528                 return btrfs_ioctl_subvol_setflags(file, argp);
5529         case BTRFS_IOC_DEFAULT_SUBVOL:
5530                 return btrfs_ioctl_default_subvol(file, argp);
5531         case BTRFS_IOC_DEFRAG:
5532                 return btrfs_ioctl_defrag(file, NULL);
5533         case BTRFS_IOC_DEFRAG_RANGE:
5534                 return btrfs_ioctl_defrag(file, argp);
5535         case BTRFS_IOC_RESIZE:
5536                 return btrfs_ioctl_resize(file, argp);
5537         case BTRFS_IOC_ADD_DEV:
5538                 return btrfs_ioctl_add_dev(fs_info, argp);
5539         case BTRFS_IOC_RM_DEV:
5540                 return btrfs_ioctl_rm_dev(file, argp);
5541         case BTRFS_IOC_RM_DEV_V2:
5542                 return btrfs_ioctl_rm_dev_v2(file, argp);
5543         case BTRFS_IOC_FS_INFO:
5544                 return btrfs_ioctl_fs_info(fs_info, argp);
5545         case BTRFS_IOC_DEV_INFO:
5546                 return btrfs_ioctl_dev_info(fs_info, argp);
5547         case BTRFS_IOC_BALANCE:
5548                 return btrfs_ioctl_balance(file, NULL);
5549         case BTRFS_IOC_TRANS_START:
5550                 return btrfs_ioctl_trans_start(file);
5551         case BTRFS_IOC_TRANS_END:
5552                 return btrfs_ioctl_trans_end(file);
5553         case BTRFS_IOC_TREE_SEARCH:
5554                 return btrfs_ioctl_tree_search(file, argp);
5555         case BTRFS_IOC_TREE_SEARCH_V2:
5556                 return btrfs_ioctl_tree_search_v2(file, argp);
5557         case BTRFS_IOC_INO_LOOKUP:
5558                 return btrfs_ioctl_ino_lookup(file, argp);
5559         case BTRFS_IOC_INO_PATHS:
5560                 return btrfs_ioctl_ino_to_path(root, argp);
5561         case BTRFS_IOC_LOGICAL_INO:
5562                 return btrfs_ioctl_logical_to_ino(fs_info, argp);
5563         case BTRFS_IOC_SPACE_INFO:
5564                 return btrfs_ioctl_space_info(fs_info, argp);
5565         case BTRFS_IOC_SYNC: {
5566                 int ret;
5567
5568                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5569                 if (ret)
5570                         return ret;
5571                 ret = btrfs_sync_fs(inode->i_sb, 1);
5572                 /*
5573                  * The transaction thread may want to do more work,
5574                  * namely it pokes the cleaner kthread that will start
5575                  * processing uncleaned subvols.
5576                  */
5577                 wake_up_process(fs_info->transaction_kthread);
5578                 return ret;
5579         }
5580         case BTRFS_IOC_START_SYNC:
5581                 return btrfs_ioctl_start_sync(root, argp);
5582         case BTRFS_IOC_WAIT_SYNC:
5583                 return btrfs_ioctl_wait_sync(fs_info, argp);
5584         case BTRFS_IOC_SCRUB:
5585                 return btrfs_ioctl_scrub(file, argp);
5586         case BTRFS_IOC_SCRUB_CANCEL:
5587                 return btrfs_ioctl_scrub_cancel(fs_info);
5588         case BTRFS_IOC_SCRUB_PROGRESS:
5589                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5590         case BTRFS_IOC_BALANCE_V2:
5591                 return btrfs_ioctl_balance(file, argp);
5592         case BTRFS_IOC_BALANCE_CTL:
5593                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5594         case BTRFS_IOC_BALANCE_PROGRESS:
5595                 return btrfs_ioctl_balance_progress(fs_info, argp);
5596         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5597                 return btrfs_ioctl_set_received_subvol(file, argp);
5598 #ifdef CONFIG_64BIT
5599         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5600                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5601 #endif
5602         case BTRFS_IOC_SEND:
5603                 return btrfs_ioctl_send(file, argp);
5604         case BTRFS_IOC_GET_DEV_STATS:
5605                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5606         case BTRFS_IOC_QUOTA_CTL:
5607                 return btrfs_ioctl_quota_ctl(file, argp);
5608         case BTRFS_IOC_QGROUP_ASSIGN:
5609                 return btrfs_ioctl_qgroup_assign(file, argp);
5610         case BTRFS_IOC_QGROUP_CREATE:
5611                 return btrfs_ioctl_qgroup_create(file, argp);
5612         case BTRFS_IOC_QGROUP_LIMIT:
5613                 return btrfs_ioctl_qgroup_limit(file, argp);
5614         case BTRFS_IOC_QUOTA_RESCAN:
5615                 return btrfs_ioctl_quota_rescan(file, argp);
5616         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5617                 return btrfs_ioctl_quota_rescan_status(file, argp);
5618         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5619                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5620         case BTRFS_IOC_DEV_REPLACE:
5621                 return btrfs_ioctl_dev_replace(fs_info, argp);
5622         case BTRFS_IOC_GET_FSLABEL:
5623                 return btrfs_ioctl_get_fslabel(file, argp);
5624         case BTRFS_IOC_SET_FSLABEL:
5625                 return btrfs_ioctl_set_fslabel(file, argp);
5626         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5627                 return btrfs_ioctl_get_supported_features(argp);
5628         case BTRFS_IOC_GET_FEATURES:
5629                 return btrfs_ioctl_get_features(file, argp);
5630         case BTRFS_IOC_SET_FEATURES:
5631                 return btrfs_ioctl_set_features(file, argp);
5632         }
5633
5634         return -ENOTTY;
5635 }
5636
5637 #ifdef CONFIG_COMPAT
5638 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5639 {
5640         /*
5641          * These all access 32-bit values anyway so no further
5642          * handling is necessary.
5643          */
5644         switch (cmd) {
5645         case FS_IOC32_GETFLAGS:
5646                 cmd = FS_IOC_GETFLAGS;
5647                 break;
5648         case FS_IOC32_SETFLAGS:
5649                 cmd = FS_IOC_SETFLAGS;
5650                 break;
5651         case FS_IOC32_GETVERSION:
5652                 cmd = FS_IOC_GETVERSION;
5653                 break;
5654         }
5655
5656         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5657 }
5658 #endif