fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/compat.h>
  37 #include <linux/bit_spinlock.h>
  38 #include <linux/security.h>
  39 #include <linux/xattr.h>
  40 #include <linux/mm.h>
  41 #include <linux/slab.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/uuid.h>
  44 #include <linux/btrfs.h>
  45 #include <linux/uaccess.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55 #include "rcu-string.h"
  56 #include "send.h"
  57 #include "dev-replace.h"
  58 #include "props.h"
  59 #include "sysfs.h"
  60 #include "qgroup.h"
  61 #include "tree-log.h"
  62 #include "compression.h"
  63
  64 #ifdef CONFIG_64BIT
  65 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  66  * structures are incorrect, as the timespec structure from userspace
  67  * is 4 bytes too small. We define these alternatives here to teach
  68  * the kernel about the 32-bit struct packing.
  69  */
  70 struct btrfs_ioctl_timespec_32 {
  71         __u64 sec;
  72         __u32 nsec;
  73 } __attribute__ ((__packed__));
  74
  75 struct btrfs_ioctl_received_subvol_args_32 {
  76         char    uuid[BTRFS_UUID_SIZE];  /* in */
  77         __u64   stransid;               /* in */
  78         __u64   rtransid;               /* out */
  79         struct btrfs_ioctl_timespec_32 stime; /* in */
  80         struct btrfs_ioctl_timespec_32 rtime; /* out */
  81         __u64   flags;                  /* in */
  82         __u64   reserved[16];           /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  86                                 struct btrfs_ioctl_received_subvol_args_32)
  87 #endif
  88
  89
  90 static int btrfs_clone(struct inode *src, struct inode *inode,
  91                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  92                        int no_time_update);
  93
  94 /* Mask out flags that are inappropriate for the given type of inode. */
  95 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  96 {
  97         if (S_ISDIR(mode))
  98                 return flags;
  99         else if (S_ISREG(mode))
 100                 return flags & ~FS_DIRSYNC_FL;
 101         else
 102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103 }
 104
 105 /*
 106  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 107  */
 108 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 109 {
 110         unsigned int iflags = 0;
 111
 112         if (flags & BTRFS_INODE_SYNC)
 113                 iflags |= FS_SYNC_FL;
 114         if (flags & BTRFS_INODE_IMMUTABLE)
 115                 iflags |= FS_IMMUTABLE_FL;
 116         if (flags & BTRFS_INODE_APPEND)
 117                 iflags |= FS_APPEND_FL;
 118         if (flags & BTRFS_INODE_NODUMP)
 119                 iflags |= FS_NODUMP_FL;
 120         if (flags & BTRFS_INODE_NOATIME)
 121                 iflags |= FS_NOATIME_FL;
 122         if (flags & BTRFS_INODE_DIRSYNC)
 123                 iflags |= FS_DIRSYNC_FL;
 124         if (flags & BTRFS_INODE_NODATACOW)
 125                 iflags |= FS_NOCOW_FL;
 126
 127         if (flags & BTRFS_INODE_NOCOMPRESS)
 128                 iflags |= FS_NOCOMP_FL;
 129         else if (flags & BTRFS_INODE_COMPRESS)
 130                 iflags |= FS_COMPR_FL;
 131
 132         return iflags;
 133 }
 134
 135 /*
 136  * Update inode->i_flags based on the btrfs internal flags.
 137  */
 138 void btrfs_update_iflags(struct inode *inode)
 139 {
 140         struct btrfs_inode *ip = BTRFS_I(inode);
 141         unsigned int new_fl = 0;
 142
 143         if (ip->flags & BTRFS_INODE_SYNC)
 144                 new_fl |= S_SYNC;
 145         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 146                 new_fl |= S_IMMUTABLE;
 147         if (ip->flags & BTRFS_INODE_APPEND)
 148                 new_fl |= S_APPEND;
 149         if (ip->flags & BTRFS_INODE_NOATIME)
 150                 new_fl |= S_NOATIME;
 151         if (ip->flags & BTRFS_INODE_DIRSYNC)
 152                 new_fl |= S_DIRSYNC;
 153
 154         set_mask_bits(&inode->i_flags,
 155                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 156                       new_fl);
 157 }
 158
 159 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 160 {
 161         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 162         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 163
 164         if (copy_to_user(arg, &flags, sizeof(flags)))
 165                 return -EFAULT;
 166         return 0;
 167 }
 168
 169 static int check_flags(unsigned int flags)
 170 {
 171         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 172                       FS_NOATIME_FL | FS_NODUMP_FL | \
 173                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 174                       FS_NOCOMP_FL | FS_COMPR_FL |
 175                       FS_NOCOW_FL))
 176                 return -EOPNOTSUPP;
 177
 178         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 179                 return -EINVAL;
 180
 181         return 0;
 182 }
 183
 184 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 185 {
 186         struct inode *inode = file_inode(file);
 187         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 188         struct btrfs_inode *ip = BTRFS_I(inode);
 189         struct btrfs_root *root = ip->root;
 190         struct btrfs_trans_handle *trans;
 191         unsigned int flags, oldflags;
 192         int ret;
 193         u64 ip_oldflags;
 194         unsigned int i_oldflags;
 195         umode_t mode;
 196
 197         if (!inode_owner_or_capable(inode))
 198                 return -EPERM;
 199
 200         if (btrfs_root_readonly(root))
 201                 return -EROFS;
 202
 203         if (copy_from_user(&flags, arg, sizeof(flags)))
 204                 return -EFAULT;
 205
 206         ret = check_flags(flags);
 207         if (ret)
 208                 return ret;
 209
 210         ret = mnt_want_write_file(file);
 211         if (ret)
 212                 return ret;
 213
 214         inode_lock(inode);
 215
 216         ip_oldflags = ip->flags;
 217         i_oldflags = inode->i_flags;
 218         mode = inode->i_mode;
 219
 220         flags = btrfs_mask_flags(inode->i_mode, flags);
 221         oldflags = btrfs_flags_to_ioctl(ip->flags);
 222         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 223                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 224                         ret = -EPERM;
 225                         goto out_unlock;
 226                 }
 227         }
 228
 229         if (flags & FS_SYNC_FL)
 230                 ip->flags |= BTRFS_INODE_SYNC;
 231         else
 232                 ip->flags &= ~BTRFS_INODE_SYNC;
 233         if (flags & FS_IMMUTABLE_FL)
 234                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 235         else
 236                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 237         if (flags & FS_APPEND_FL)
 238                 ip->flags |= BTRFS_INODE_APPEND;
 239         else
 240                 ip->flags &= ~BTRFS_INODE_APPEND;
 241         if (flags & FS_NODUMP_FL)
 242                 ip->flags |= BTRFS_INODE_NODUMP;
 243         else
 244                 ip->flags &= ~BTRFS_INODE_NODUMP;
 245         if (flags & FS_NOATIME_FL)
 246                 ip->flags |= BTRFS_INODE_NOATIME;
 247         else
 248                 ip->flags &= ~BTRFS_INODE_NOATIME;
 249         if (flags & FS_DIRSYNC_FL)
 250                 ip->flags |= BTRFS_INODE_DIRSYNC;
 251         else
 252                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 253         if (flags & FS_NOCOW_FL) {
 254                 if (S_ISREG(mode)) {
 255                         /*
 256                          * It's safe to turn csums off here, no extents exist.
 257                          * Otherwise we want the flag to reflect the real COW
 258                          * status of the file and will not set it.
 259                          */
 260                         if (inode->i_size == 0)
 261                                 ip->flags |= BTRFS_INODE_NODATACOW
 262                                            | BTRFS_INODE_NODATASUM;
 263                 } else {
 264                         ip->flags |= BTRFS_INODE_NODATACOW;
 265                 }
 266         } else {
 267                 /*
 268                  * Revert back under same assumptions as above
 269                  */
 270                 if (S_ISREG(mode)) {
 271                         if (inode->i_size == 0)
 272                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 273                                              | BTRFS_INODE_NODATASUM);
 274                 } else {
 275                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 276                 }
 277         }
 278
 279         /*
 280          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 281          * flag may be changed automatically if compression code won't make
 282          * things smaller.
 283          */
 284         if (flags & FS_NOCOMP_FL) {
 285                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 286                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 287
 288                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 289                 if (ret && ret != -ENODATA)
 290                         goto out_drop;
 291         } else if (flags & FS_COMPR_FL) {
 292                 const char *comp;
 293
 294                 ip->flags |= BTRFS_INODE_COMPRESS;
 295                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 296
 297                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
 298                         comp = "lzo";
 299                 else if (fs_info->compress_type == BTRFS_COMPRESS_ZLIB)
 300                         comp = "zlib";
 301                 else
 302                         comp = "zstd";
 303                 ret = btrfs_set_prop(inode, "btrfs.compression",
 304                                      comp, strlen(comp), 0);
 305                 if (ret)
 306                         goto out_drop;
 307
 308         } else {
 309                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 310                 if (ret && ret != -ENODATA)
 311                         goto out_drop;
 312                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 313         }
 314
 315         trans = btrfs_start_transaction(root, 1);
 316         if (IS_ERR(trans)) {
 317                 ret = PTR_ERR(trans);
 318                 goto out_drop;
 319         }
 320
 321         btrfs_update_iflags(inode);
 322         inode_inc_iversion(inode);
 323         inode->i_ctime = current_time(inode);
 324         ret = btrfs_update_inode(trans, root, inode);
 325
 326         btrfs_end_transaction(trans);
 327  out_drop:
 328         if (ret) {
 329                 ip->flags = ip_oldflags;
 330                 inode->i_flags = i_oldflags;
 331         }
 332
 333  out_unlock:
 334         inode_unlock(inode);
 335         mnt_drop_write_file(file);
 336         return ret;
 337 }
 338
 339 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 340 {
 341         struct inode *inode = file_inode(file);
 342
 343         return put_user(inode->i_generation, arg);
 344 }
 345
 346 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 347 {
 348         struct inode *inode = file_inode(file);
 349         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 350         struct btrfs_device *device;
 351         struct request_queue *q;
 352         struct fstrim_range range;
 353         u64 minlen = ULLONG_MAX;
 354         u64 num_devices = 0;
 355         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 356         int ret;
 357
 358         if (!capable(CAP_SYS_ADMIN))
 359                 return -EPERM;
 360
 361         rcu_read_lock();
 362         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 363                                 dev_list) {
 364                 if (!device->bdev)
 365                         continue;
 366                 q = bdev_get_queue(device->bdev);
 367                 if (blk_queue_discard(q)) {
 368                         num_devices++;
 369                         minlen = min_t(u64, q->limits.discard_granularity,
 370                                      minlen);
 371                 }
 372         }
 373         rcu_read_unlock();
 374
 375         if (!num_devices)
 376                 return -EOPNOTSUPP;
 377         if (copy_from_user(&range, arg, sizeof(range)))
 378                 return -EFAULT;
 379         if (range.start > total_bytes ||
 380             range.len < fs_info->sb->s_blocksize)
 381                 return -EINVAL;
 382
 383         range.len = min(range.len, total_bytes - range.start);
 384         range.minlen = max(range.minlen, minlen);
 385         ret = btrfs_trim_fs(fs_info, &range);
 386         if (ret < 0)
 387                 return ret;
 388
 389         if (copy_to_user(arg, &range, sizeof(range)))
 390                 return -EFAULT;
 391
 392         return 0;
 393 }
 394
 395 int btrfs_is_empty_uuid(u8 *uuid)
 396 {
 397         int i;
 398
 399         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 400                 if (uuid[i])
 401                         return 0;
 402         }
 403         return 1;
 404 }
 405
 406 static noinline int create_subvol(struct inode *dir,
 407                                   struct dentry *dentry,
 408                                   const char *name, int namelen,
 409                                   u64 *async_transid,
 410                                   struct btrfs_qgroup_inherit *inherit)
 411 {
 412         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 413         struct btrfs_trans_handle *trans;
 414         struct btrfs_key key;
 415         struct btrfs_root_item *root_item;
 416         struct btrfs_inode_item *inode_item;
 417         struct extent_buffer *leaf;
 418         struct btrfs_root *root = BTRFS_I(dir)->root;
 419         struct btrfs_root *new_root;
 420         struct btrfs_block_rsv block_rsv;
 421         struct timespec cur_time = current_time(dir);
 422         struct inode *inode;
 423         int ret;
 424         int err;
 425         u64 objectid;
 426         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 427         u64 index = 0;
 428         u64 qgroup_reserved;
 429         uuid_le new_uuid;
 430
 431         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 432         if (!root_item)
 433                 return -ENOMEM;
 434
 435         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 436         if (ret)
 437                 goto fail_free;
 438
 439         /*
 440          * Don't create subvolume whose level is not zero. Or qgroup will be
 441          * screwed up since it assumes subvolume qgroup's level to be 0.
 442          */
 443         if (btrfs_qgroup_level(objectid)) {
 444                 ret = -ENOSPC;
 445                 goto fail_free;
 446         }
 447
 448         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 449         /*
 450          * The same as the snapshot creation, please see the comment
 451          * of create_snapshot().
 452          */
 453         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 454                                                8, &qgroup_reserved, false);
 455         if (ret)
 456                 goto fail_free;
 457
 458         trans = btrfs_start_transaction(root, 0);
 459         if (IS_ERR(trans)) {
 460                 ret = PTR_ERR(trans);
 461                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 462                 goto fail_free;
 463         }
 464         trans->block_rsv = &block_rsv;
 465         trans->bytes_reserved = block_rsv.size;
 466
 467         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 468         if (ret)
 469                 goto fail;
 470
 471         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 472         if (IS_ERR(leaf)) {
 473                 ret = PTR_ERR(leaf);
 474                 goto fail;
 475         }
 476
 477         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 478         btrfs_set_header_bytenr(leaf, leaf->start);
 479         btrfs_set_header_generation(leaf, trans->transid);
 480         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 481         btrfs_set_header_owner(leaf, objectid);
 482
 483         write_extent_buffer_fsid(leaf, fs_info->fsid);
 484         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 485         btrfs_mark_buffer_dirty(leaf);
 486
 487         inode_item = &root_item->inode;
 488         btrfs_set_stack_inode_generation(inode_item, 1);
 489         btrfs_set_stack_inode_size(inode_item, 3);
 490         btrfs_set_stack_inode_nlink(inode_item, 1);
 491         btrfs_set_stack_inode_nbytes(inode_item,
 492                                      fs_info->nodesize);
 493         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 494
 495         btrfs_set_root_flags(root_item, 0);
 496         btrfs_set_root_limit(root_item, 0);
 497         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 498
 499         btrfs_set_root_bytenr(root_item, leaf->start);
 500         btrfs_set_root_generation(root_item, trans->transid);
 501         btrfs_set_root_level(root_item, 0);
 502         btrfs_set_root_refs(root_item, 1);
 503         btrfs_set_root_used(root_item, leaf->len);
 504         btrfs_set_root_last_snapshot(root_item, 0);
 505
 506         btrfs_set_root_generation_v2(root_item,
 507                         btrfs_root_generation(root_item));
 508         uuid_le_gen(&new_uuid);
 509         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 510         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 511         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 512         root_item->ctime = root_item->otime;
 513         btrfs_set_root_ctransid(root_item, trans->transid);
 514         btrfs_set_root_otransid(root_item, trans->transid);
 515
 516         btrfs_tree_unlock(leaf);
 517         free_extent_buffer(leaf);
 518         leaf = NULL;
 519
 520         btrfs_set_root_dirid(root_item, new_dirid);
 521
 522         key.objectid = objectid;
 523         key.offset = 0;
 524         key.type = BTRFS_ROOT_ITEM_KEY;
 525         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 526                                 root_item);
 527         if (ret)
 528                 goto fail;
 529
 530         key.offset = (u64)-1;
 531         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 532         if (IS_ERR(new_root)) {
 533                 ret = PTR_ERR(new_root);
 534                 btrfs_abort_transaction(trans, ret);
 535                 goto fail;
 536         }
 537
 538         btrfs_record_root_in_trans(trans, new_root);
 539
 540         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 541         if (ret) {
 542                 /* We potentially lose an unused inode item here */
 543                 btrfs_abort_transaction(trans, ret);
 544                 goto fail;
 545         }
 546
 547         mutex_lock(&new_root->objectid_mutex);
 548         new_root->highest_objectid = new_dirid;
 549         mutex_unlock(&new_root->objectid_mutex);
 550
 551         /*
 552          * insert the directory item
 553          */
 554         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 555         if (ret) {
 556                 btrfs_abort_transaction(trans, ret);
 557                 goto fail;
 558         }
 559
 560         ret = btrfs_insert_dir_item(trans, root,
 561                                     name, namelen, BTRFS_I(dir), &key,
 562                                     BTRFS_FT_DIR, index);
 563         if (ret) {
 564                 btrfs_abort_transaction(trans, ret);
 565                 goto fail;
 566         }
 567
 568         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 569         ret = btrfs_update_inode(trans, root, dir);
 570         BUG_ON(ret);
 571
 572         ret = btrfs_add_root_ref(trans, fs_info,
 573                                  objectid, root->root_key.objectid,
 574                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 575         BUG_ON(ret);
 576
 577         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 578                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 579         if (ret)
 580                 btrfs_abort_transaction(trans, ret);
 581
 582 fail:
 583         kfree(root_item);
 584         trans->block_rsv = NULL;
 585         trans->bytes_reserved = 0;
 586         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 587
 588         if (async_transid) {
 589                 *async_transid = trans->transid;
 590                 err = btrfs_commit_transaction_async(trans, 1);
 591                 if (err)
 592                         err = btrfs_commit_transaction(trans);
 593         } else {
 594                 err = btrfs_commit_transaction(trans);
 595         }
 596         if (err && !ret)
 597                 ret = err;
 598
 599         if (!ret) {
 600                 inode = btrfs_lookup_dentry(dir, dentry);
 601                 if (IS_ERR(inode))
 602                         return PTR_ERR(inode);
 603                 d_instantiate(dentry, inode);
 604         }
 605         return ret;
 606
 607 fail_free:
 608         kfree(root_item);
 609         return ret;
 610 }
 611
 612 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 613                            struct dentry *dentry,
 614                            u64 *async_transid, bool readonly,
 615                            struct btrfs_qgroup_inherit *inherit)
 616 {
 617         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 618         struct inode *inode;
 619         struct btrfs_pending_snapshot *pending_snapshot;
 620         struct btrfs_trans_handle *trans;
 621         int ret;
 622
 623         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 624                 return -EINVAL;
 625
 626         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 627         if (!pending_snapshot)
 628                 return -ENOMEM;
 629
 630         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 631                         GFP_KERNEL);
 632         pending_snapshot->path = btrfs_alloc_path();
 633         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 634                 ret = -ENOMEM;
 635                 goto free_pending;
 636         }
 637
 638         atomic_inc(&root->will_be_snapshotted);
 639         smp_mb__after_atomic();
 640         /* wait for no snapshot writes */
 641         wait_event(root->subv_writers->wait,
 642                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 643
 644         ret = btrfs_start_delalloc_inodes(root, 0);
 645         if (ret)
 646                 goto dec_and_free;
 647
 648         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 649
 650         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 651                              BTRFS_BLOCK_RSV_TEMP);
 652         /*
 653          * 1 - parent dir inode
 654          * 2 - dir entries
 655          * 1 - root item
 656          * 2 - root ref/backref
 657          * 1 - root of snapshot
 658          * 1 - UUID item
 659          */
 660         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 661                                         &pending_snapshot->block_rsv, 8,
 662                                         &pending_snapshot->qgroup_reserved,
 663                                         false);
 664         if (ret)
 665                 goto dec_and_free;
 666
 667         pending_snapshot->dentry = dentry;
 668         pending_snapshot->root = root;
 669         pending_snapshot->readonly = readonly;
 670         pending_snapshot->dir = dir;
 671         pending_snapshot->inherit = inherit;
 672
 673         trans = btrfs_start_transaction(root, 0);
 674         if (IS_ERR(trans)) {
 675                 ret = PTR_ERR(trans);
 676                 goto fail;
 677         }
 678
 679         spin_lock(&fs_info->trans_lock);
 680         list_add(&pending_snapshot->list,
 681                  &trans->transaction->pending_snapshots);
 682         spin_unlock(&fs_info->trans_lock);
 683         if (async_transid) {
 684                 *async_transid = trans->transid;
 685                 ret = btrfs_commit_transaction_async(trans, 1);
 686                 if (ret)
 687                         ret = btrfs_commit_transaction(trans);
 688         } else {
 689                 ret = btrfs_commit_transaction(trans);
 690         }
 691         if (ret)
 692                 goto fail;
 693
 694         ret = pending_snapshot->error;
 695         if (ret)
 696                 goto fail;
 697
 698         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 699         if (ret)
 700                 goto fail;
 701
 702         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 703         if (IS_ERR(inode)) {
 704                 ret = PTR_ERR(inode);
 705                 goto fail;
 706         }
 707
 708         d_instantiate(dentry, inode);
 709         ret = 0;
 710 fail:
 711         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 712 dec_and_free:
 713         if (atomic_dec_and_test(&root->will_be_snapshotted))
 714                 wake_up_atomic_t(&root->will_be_snapshotted);
 715 free_pending:
 716         kfree(pending_snapshot->root_item);
 717         btrfs_free_path(pending_snapshot->path);
 718         kfree(pending_snapshot);
 719
 720         return ret;
 721 }
 722
 723 /*  copy of may_delete in fs/namei.c()
 724  *      Check whether we can remove a link victim from directory dir, check
 725  *  whether the type of victim is right.
 726  *  1. We can't do it if dir is read-only (done in permission())
 727  *  2. We should have write and exec permissions on dir
 728  *  3. We can't remove anything from append-only dir
 729  *  4. We can't do anything with immutable dir (done in permission())
 730  *  5. If the sticky bit on dir is set we should either
 731  *      a. be owner of dir, or
 732  *      b. be owner of victim, or
 733  *      c. have CAP_FOWNER capability
 734  *  6. If the victim is append-only or immutable we can't do anything with
 735  *     links pointing to it.
 736  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 737  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 738  *  9. We can't remove a root or mountpoint.
 739  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 740  *     nfs_async_unlink().
 741  */
 742
 743 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 744 {
 745         int error;
 746
 747         if (d_really_is_negative(victim))
 748                 return -ENOENT;
 749
 750         BUG_ON(d_inode(victim->d_parent) != dir);
 751         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 752
 753         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 754         if (error)
 755                 return error;
 756         if (IS_APPEND(dir))
 757                 return -EPERM;
 758         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 759             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 760                 return -EPERM;
 761         if (isdir) {
 762                 if (!d_is_dir(victim))
 763                         return -ENOTDIR;
 764                 if (IS_ROOT(victim))
 765                         return -EBUSY;
 766         } else if (d_is_dir(victim))
 767                 return -EISDIR;
 768         if (IS_DEADDIR(dir))
 769                 return -ENOENT;
 770         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 771                 return -EBUSY;
 772         return 0;
 773 }
 774
 775 /* copy of may_create in fs/namei.c() */
 776 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 777 {
 778         if (d_really_is_positive(child))
 779                 return -EEXIST;
 780         if (IS_DEADDIR(dir))
 781                 return -ENOENT;
 782         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 783 }
 784
 785 /*
 786  * Create a new subvolume below @parent.  This is largely modeled after
 787  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 788  * inside this filesystem so it's quite a bit simpler.
 789  */
 790 static noinline int btrfs_mksubvol(const struct path *parent,
 791                                    const char *name, int namelen,
 792                                    struct btrfs_root *snap_src,
 793                                    u64 *async_transid, bool readonly,
 794                                    struct btrfs_qgroup_inherit *inherit)
 795 {
 796         struct inode *dir = d_inode(parent->dentry);
 797         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 798         struct dentry *dentry;
 799         int error;
 800
 801         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 802         if (error == -EINTR)
 803                 return error;
 804
 805         dentry = lookup_one_len(name, parent->dentry, namelen);
 806         error = PTR_ERR(dentry);
 807         if (IS_ERR(dentry))
 808                 goto out_unlock;
 809
 810         error = btrfs_may_create(dir, dentry);
 811         if (error)
 812                 goto out_dput;
 813
 814         /*
 815          * even if this name doesn't exist, we may get hash collisions.
 816          * check for them now when we can safely fail
 817          */
 818         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 819                                                dir->i_ino, name,
 820                                                namelen);
 821         if (error)
 822                 goto out_dput;
 823
 824         down_read(&fs_info->subvol_sem);
 825
 826         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 827                 goto out_up_read;
 828
 829         if (snap_src) {
 830                 error = create_snapshot(snap_src, dir, dentry,
 831                                         async_transid, readonly, inherit);
 832         } else {
 833                 error = create_subvol(dir, dentry, name, namelen,
 834                                       async_transid, inherit);
 835         }
 836         if (!error)
 837                 fsnotify_mkdir(dir, dentry);
 838 out_up_read:
 839         up_read(&fs_info->subvol_sem);
 840 out_dput:
 841         dput(dentry);
 842 out_unlock:
 843         inode_unlock(dir);
 844         return error;
 845 }
 846
 847 /*
 848  * When we're defragging a range, we don't want to kick it off again
 849  * if it is really just waiting for delalloc to send it down.
 850  * If we find a nice big extent or delalloc range for the bytes in the
 851  * file you want to defrag, we return 0 to let you know to skip this
 852  * part of the file
 853  */
 854 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 855 {
 856         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 857         struct extent_map *em = NULL;
 858         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 859         u64 end;
 860
 861         read_lock(&em_tree->lock);
 862         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 863         read_unlock(&em_tree->lock);
 864
 865         if (em) {
 866                 end = extent_map_end(em);
 867                 free_extent_map(em);
 868                 if (end - offset > thresh)
 869                         return 0;
 870         }
 871         /* if we already have a nice delalloc here, just stop */
 872         thresh /= 2;
 873         end = count_range_bits(io_tree, &offset, offset + thresh,
 874                                thresh, EXTENT_DELALLOC, 1);
 875         if (end >= thresh)
 876                 return 0;
 877         return 1;
 878 }
 879
 880 /*
 881  * helper function to walk through a file and find extents
 882  * newer than a specific transid, and smaller than thresh.
 883  *
 884  * This is used by the defragging code to find new and small
 885  * extents
 886  */
 887 static int find_new_extents(struct btrfs_root *root,
 888                             struct inode *inode, u64 newer_than,
 889                             u64 *off, u32 thresh)
 890 {
 891         struct btrfs_path *path;
 892         struct btrfs_key min_key;
 893         struct extent_buffer *leaf;
 894         struct btrfs_file_extent_item *extent;
 895         int type;
 896         int ret;
 897         u64 ino = btrfs_ino(BTRFS_I(inode));
 898
 899         path = btrfs_alloc_path();
 900         if (!path)
 901                 return -ENOMEM;
 902
 903         min_key.objectid = ino;
 904         min_key.type = BTRFS_EXTENT_DATA_KEY;
 905         min_key.offset = *off;
 906
 907         while (1) {
 908                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 909                 if (ret != 0)
 910                         goto none;
 911 process_slot:
 912                 if (min_key.objectid != ino)
 913                         goto none;
 914                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 915                         goto none;
 916
 917                 leaf = path->nodes[0];
 918                 extent = btrfs_item_ptr(leaf, path->slots[0],
 919                                         struct btrfs_file_extent_item);
 920
 921                 type = btrfs_file_extent_type(leaf, extent);
 922                 if (type == BTRFS_FILE_EXTENT_REG &&
 923                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 924                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 925                         *off = min_key.offset;
 926                         btrfs_free_path(path);
 927                         return 0;
 928                 }
 929
 930                 path->slots[0]++;
 931                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 932                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 933                         goto process_slot;
 934                 }
 935
 936                 if (min_key.offset == (u64)-1)
 937                         goto none;
 938
 939                 min_key.offset++;
 940                 btrfs_release_path(path);
 941         }
 942 none:
 943         btrfs_free_path(path);
 944         return -ENOENT;
 945 }
 946
 947 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 948 {
 949         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 950         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 951         struct extent_map *em;
 952         u64 len = PAGE_SIZE;
 953
 954         /*
 955          * hopefully we have this extent in the tree already, try without
 956          * the full extent lock
 957          */
 958         read_lock(&em_tree->lock);
 959         em = lookup_extent_mapping(em_tree, start, len);
 960         read_unlock(&em_tree->lock);
 961
 962         if (!em) {
 963                 struct extent_state *cached = NULL;
 964                 u64 end = start + len - 1;
 965
 966                 /* get the big lock and read metadata off disk */
 967                 lock_extent_bits(io_tree, start, end, &cached);
 968                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 969                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 970
 971                 if (IS_ERR(em))
 972                         return NULL;
 973         }
 974
 975         return em;
 976 }
 977
 978 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 979 {
 980         struct extent_map *next;
 981         bool ret = true;
 982
 983         /* this is the last extent */
 984         if (em->start + em->len >= i_size_read(inode))
 985                 return false;
 986
 987         next = defrag_lookup_extent(inode, em->start + em->len);
 988         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 989                 ret = false;
 990         else if ((em->block_start + em->block_len == next->block_start) &&
 991                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
 992                 ret = false;
 993
 994         free_extent_map(next);
 995         return ret;
 996 }
 997
 998 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
 999                                u64 *last_len, u64 *skip, u64 *defrag_end,
1000                                int compress)
1001 {
1002         struct extent_map *em;
1003         int ret = 1;
1004         bool next_mergeable = true;
1005         bool prev_mergeable = true;
1006
1007         /*
1008          * make sure that once we start defragging an extent, we keep on
1009          * defragging it
1010          */
1011         if (start < *defrag_end)
1012                 return 1;
1013
1014         *skip = 0;
1015
1016         em = defrag_lookup_extent(inode, start);
1017         if (!em)
1018                 return 0;
1019
1020         /* this will cover holes, and inline extents */
1021         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1022                 ret = 0;
1023                 goto out;
1024         }
1025
1026         if (!*defrag_end)
1027                 prev_mergeable = false;
1028
1029         next_mergeable = defrag_check_next_extent(inode, em);
1030         /*
1031          * we hit a real extent, if it is big or the next extent is not a
1032          * real extent, don't bother defragging it
1033          */
1034         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1035             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1036                 ret = 0;
1037 out:
1038         /*
1039          * last_len ends up being a counter of how many bytes we've defragged.
1040          * every time we choose not to defrag an extent, we reset *last_len
1041          * so that the next tiny extent will force a defrag.
1042          *
1043          * The end result of this is that tiny extents before a single big
1044          * extent will force at least part of that big extent to be defragged.
1045          */
1046         if (ret) {
1047                 *defrag_end = extent_map_end(em);
1048         } else {
1049                 *last_len = 0;
1050                 *skip = extent_map_end(em);
1051                 *defrag_end = 0;
1052         }
1053
1054         free_extent_map(em);
1055         return ret;
1056 }
1057
1058 /*
1059  * it doesn't do much good to defrag one or two pages
1060  * at a time.  This pulls in a nice chunk of pages
1061  * to COW and defrag.
1062  *
1063  * It also makes sure the delalloc code has enough
1064  * dirty data to avoid making new small extents as part
1065  * of the defrag
1066  *
1067  * It's a good idea to start RA on this range
1068  * before calling this.
1069  */
1070 static int cluster_pages_for_defrag(struct inode *inode,
1071                                     struct page **pages,
1072                                     unsigned long start_index,
1073                                     unsigned long num_pages)
1074 {
1075         unsigned long file_end;
1076         u64 isize = i_size_read(inode);
1077         u64 page_start;
1078         u64 page_end;
1079         u64 page_cnt;
1080         int ret;
1081         int i;
1082         int i_done;
1083         struct btrfs_ordered_extent *ordered;
1084         struct extent_state *cached_state = NULL;
1085         struct extent_io_tree *tree;
1086         struct extent_changeset *data_reserved = NULL;
1087         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1088
1089         file_end = (isize - 1) >> PAGE_SHIFT;
1090         if (!isize || start_index > file_end)
1091                 return 0;
1092
1093         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1094
1095         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1096                         start_index << PAGE_SHIFT,
1097                         page_cnt << PAGE_SHIFT);
1098         if (ret)
1099                 return ret;
1100         i_done = 0;
1101         tree = &BTRFS_I(inode)->io_tree;
1102
1103         /* step one, lock all the pages */
1104         for (i = 0; i < page_cnt; i++) {
1105                 struct page *page;
1106 again:
1107                 page = find_or_create_page(inode->i_mapping,
1108                                            start_index + i, mask);
1109                 if (!page)
1110                         break;
1111
1112                 page_start = page_offset(page);
1113                 page_end = page_start + PAGE_SIZE - 1;
1114                 while (1) {
1115                         lock_extent_bits(tree, page_start, page_end,
1116                                          &cached_state);
1117                         ordered = btrfs_lookup_ordered_extent(inode,
1118                                                               page_start);
1119                         unlock_extent_cached(tree, page_start, page_end,
1120                                              &cached_state, GFP_NOFS);
1121                         if (!ordered)
1122                                 break;
1123
1124                         unlock_page(page);
1125                         btrfs_start_ordered_extent(inode, ordered, 1);
1126                         btrfs_put_ordered_extent(ordered);
1127                         lock_page(page);
1128                         /*
1129                          * we unlocked the page above, so we need check if
1130                          * it was released or not.
1131                          */
1132                         if (page->mapping != inode->i_mapping) {
1133                                 unlock_page(page);
1134                                 put_page(page);
1135                                 goto again;
1136                         }
1137                 }
1138
1139                 if (!PageUptodate(page)) {
1140                         btrfs_readpage(NULL, page);
1141                         lock_page(page);
1142                         if (!PageUptodate(page)) {
1143                                 unlock_page(page);
1144                                 put_page(page);
1145                                 ret = -EIO;
1146                                 break;
1147                         }
1148                 }
1149
1150                 if (page->mapping != inode->i_mapping) {
1151                         unlock_page(page);
1152                         put_page(page);
1153                         goto again;
1154                 }
1155
1156                 pages[i] = page;
1157                 i_done++;
1158         }
1159         if (!i_done || ret)
1160                 goto out;
1161
1162         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1163                 goto out;
1164
1165         /*
1166          * so now we have a nice long stream of locked
1167          * and up to date pages, lets wait on them
1168          */
1169         for (i = 0; i < i_done; i++)
1170                 wait_on_page_writeback(pages[i]);
1171
1172         page_start = page_offset(pages[0]);
1173         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1174
1175         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1176                          page_start, page_end - 1, &cached_state);
1177         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1178                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1179                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1180                           &cached_state, GFP_NOFS);
1181
1182         if (i_done != page_cnt) {
1183                 spin_lock(&BTRFS_I(inode)->lock);
1184                 BTRFS_I(inode)->outstanding_extents++;
1185                 spin_unlock(&BTRFS_I(inode)->lock);
1186                 btrfs_delalloc_release_space(inode, data_reserved,
1187                                 start_index << PAGE_SHIFT,
1188                                 (page_cnt - i_done) << PAGE_SHIFT);
1189         }
1190
1191
1192         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1193                           &cached_state);
1194
1195         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1196                              page_start, page_end - 1, &cached_state,
1197                              GFP_NOFS);
1198
1199         for (i = 0; i < i_done; i++) {
1200                 clear_page_dirty_for_io(pages[i]);
1201                 ClearPageChecked(pages[i]);
1202                 set_page_extent_mapped(pages[i]);
1203                 set_page_dirty(pages[i]);
1204                 unlock_page(pages[i]);
1205                 put_page(pages[i]);
1206         }
1207         extent_changeset_free(data_reserved);
1208         return i_done;
1209 out:
1210         for (i = 0; i < i_done; i++) {
1211                 unlock_page(pages[i]);
1212                 put_page(pages[i]);
1213         }
1214         btrfs_delalloc_release_space(inode, data_reserved,
1215                         start_index << PAGE_SHIFT,
1216                         page_cnt << PAGE_SHIFT);
1217         extent_changeset_free(data_reserved);
1218         return ret;
1219
1220 }
1221
1222 int btrfs_defrag_file(struct inode *inode, struct file *file,
1223                       struct btrfs_ioctl_defrag_range_args *range,
1224                       u64 newer_than, unsigned long max_to_defrag)
1225 {
1226         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1227         struct btrfs_root *root = BTRFS_I(inode)->root;
1228         struct file_ra_state *ra = NULL;
1229         unsigned long last_index;
1230         u64 isize = i_size_read(inode);
1231         u64 last_len = 0;
1232         u64 skip = 0;
1233         u64 defrag_end = 0;
1234         u64 newer_off = range->start;
1235         unsigned long i;
1236         unsigned long ra_index = 0;
1237         int ret;
1238         int defrag_count = 0;
1239         int compress_type = BTRFS_COMPRESS_ZLIB;
1240         u32 extent_thresh = range->extent_thresh;
1241         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1242         unsigned long cluster = max_cluster;
1243         u64 new_align = ~((u64)SZ_128K - 1);
1244         struct page **pages = NULL;
1245         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1246
1247         if (isize == 0)
1248                 return 0;
1249
1250         if (range->start >= isize)
1251                 return -EINVAL;
1252
1253         if (do_compress) {
1254                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1255                         return -EINVAL;
1256                 if (range->compress_type)
1257                         compress_type = range->compress_type;
1258         }
1259
1260         if (extent_thresh == 0)
1261                 extent_thresh = SZ_256K;
1262
1263         /*
1264          * If we were not given a file, allocate a readahead context. As
1265          * readahead is just an optimization, defrag will work without it so
1266          * we don't error out.
1267          */
1268         if (!file) {
1269                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1270                 if (ra)
1271                         file_ra_state_init(ra, inode->i_mapping);
1272         } else {
1273                 ra = &file->f_ra;
1274         }
1275
1276         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1277         if (!pages) {
1278                 ret = -ENOMEM;
1279                 goto out_ra;
1280         }
1281
1282         /* find the last page to defrag */
1283         if (range->start + range->len > range->start) {
1284                 last_index = min_t(u64, isize - 1,
1285                          range->start + range->len - 1) >> PAGE_SHIFT;
1286         } else {
1287                 last_index = (isize - 1) >> PAGE_SHIFT;
1288         }
1289
1290         if (newer_than) {
1291                 ret = find_new_extents(root, inode, newer_than,
1292                                        &newer_off, SZ_64K);
1293                 if (!ret) {
1294                         range->start = newer_off;
1295                         /*
1296                          * we always align our defrag to help keep
1297                          * the extents in the file evenly spaced
1298                          */
1299                         i = (newer_off & new_align) >> PAGE_SHIFT;
1300                 } else
1301                         goto out_ra;
1302         } else {
1303                 i = range->start >> PAGE_SHIFT;
1304         }
1305         if (!max_to_defrag)
1306                 max_to_defrag = last_index - i + 1;
1307
1308         /*
1309          * make writeback starts from i, so the defrag range can be
1310          * written sequentially.
1311          */
1312         if (i < inode->i_mapping->writeback_index)
1313                 inode->i_mapping->writeback_index = i;
1314
1315         while (i <= last_index && defrag_count < max_to_defrag &&
1316                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1317                 /*
1318                  * make sure we stop running if someone unmounts
1319                  * the FS
1320                  */
1321                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1322                         break;
1323
1324                 if (btrfs_defrag_cancelled(fs_info)) {
1325                         btrfs_debug(fs_info, "defrag_file cancelled");
1326                         ret = -EAGAIN;
1327                         break;
1328                 }
1329
1330                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1331                                          extent_thresh, &last_len, &skip,
1332                                          &defrag_end, do_compress)){
1333                         unsigned long next;
1334                         /*
1335                          * the should_defrag function tells us how much to skip
1336                          * bump our counter by the suggested amount
1337                          */
1338                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1339                         i = max(i + 1, next);
1340                         continue;
1341                 }
1342
1343                 if (!newer_than) {
1344                         cluster = (PAGE_ALIGN(defrag_end) >>
1345                                    PAGE_SHIFT) - i;
1346                         cluster = min(cluster, max_cluster);
1347                 } else {
1348                         cluster = max_cluster;
1349                 }
1350
1351                 if (i + cluster > ra_index) {
1352                         ra_index = max(i, ra_index);
1353                         if (ra)
1354                                 page_cache_sync_readahead(inode->i_mapping, ra,
1355                                                 file, ra_index, cluster);
1356                         ra_index += cluster;
1357                 }
1358
1359                 inode_lock(inode);
1360                 if (do_compress)
1361                         BTRFS_I(inode)->defrag_compress = compress_type;
1362                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1363                 if (ret < 0) {
1364                         inode_unlock(inode);
1365                         goto out_ra;
1366                 }
1367
1368                 defrag_count += ret;
1369                 balance_dirty_pages_ratelimited(inode->i_mapping);
1370                 inode_unlock(inode);
1371
1372                 if (newer_than) {
1373                         if (newer_off == (u64)-1)
1374                                 break;
1375
1376                         if (ret > 0)
1377                                 i += ret;
1378
1379                         newer_off = max(newer_off + 1,
1380                                         (u64)i << PAGE_SHIFT);
1381
1382                         ret = find_new_extents(root, inode, newer_than,
1383                                                &newer_off, SZ_64K);
1384                         if (!ret) {
1385                                 range->start = newer_off;
1386                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1387                         } else {
1388                                 break;
1389                         }
1390                 } else {
1391                         if (ret > 0) {
1392                                 i += ret;
1393                                 last_len += ret << PAGE_SHIFT;
1394                         } else {
1395                                 i++;
1396                                 last_len = 0;
1397                         }
1398                 }
1399         }
1400
1401         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1402                 filemap_flush(inode->i_mapping);
1403                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1404                              &BTRFS_I(inode)->runtime_flags))
1405                         filemap_flush(inode->i_mapping);
1406         }
1407
1408         if (do_compress) {
1409                 /* the filemap_flush will queue IO into the worker threads, but
1410                  * we have to make sure the IO is actually started and that
1411                  * ordered extents get created before we return
1412                  */
1413                 atomic_inc(&fs_info->async_submit_draining);
1414                 while (atomic_read(&fs_info->nr_async_submits) ||
1415                        atomic_read(&fs_info->async_delalloc_pages)) {
1416                         wait_event(fs_info->async_submit_wait,
1417                                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
1418                                     atomic_read(&fs_info->async_delalloc_pages) == 0));
1419                 }
1420                 atomic_dec(&fs_info->async_submit_draining);
1421         }
1422
1423         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1424                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1425         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1426                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1427         }
1428
1429         ret = defrag_count;
1430
1431 out_ra:
1432         if (do_compress) {
1433                 inode_lock(inode);
1434                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1435                 inode_unlock(inode);
1436         }
1437         if (!file)
1438                 kfree(ra);
1439         kfree(pages);
1440         return ret;
1441 }
1442
1443 static noinline int btrfs_ioctl_resize(struct file *file,
1444                                         void __user *arg)
1445 {
1446         struct inode *inode = file_inode(file);
1447         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1448         u64 new_size;
1449         u64 old_size;
1450         u64 devid = 1;
1451         struct btrfs_root *root = BTRFS_I(inode)->root;
1452         struct btrfs_ioctl_vol_args *vol_args;
1453         struct btrfs_trans_handle *trans;
1454         struct btrfs_device *device = NULL;
1455         char *sizestr;
1456         char *retptr;
1457         char *devstr = NULL;
1458         int ret = 0;
1459         int mod = 0;
1460
1461         if (!capable(CAP_SYS_ADMIN))
1462                 return -EPERM;
1463
1464         ret = mnt_want_write_file(file);
1465         if (ret)
1466                 return ret;
1467
1468         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1469                 mnt_drop_write_file(file);
1470                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1471         }
1472
1473         mutex_lock(&fs_info->volume_mutex);
1474         vol_args = memdup_user(arg, sizeof(*vol_args));
1475         if (IS_ERR(vol_args)) {
1476                 ret = PTR_ERR(vol_args);
1477                 goto out;
1478         }
1479
1480         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1481
1482         sizestr = vol_args->name;
1483         devstr = strchr(sizestr, ':');
1484         if (devstr) {
1485                 sizestr = devstr + 1;
1486                 *devstr = '\0';
1487                 devstr = vol_args->name;
1488                 ret = kstrtoull(devstr, 10, &devid);
1489                 if (ret)
1490                         goto out_free;
1491                 if (!devid) {
1492                         ret = -EINVAL;
1493                         goto out_free;
1494                 }
1495                 btrfs_info(fs_info, "resizing devid %llu", devid);
1496         }
1497
1498         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1499         if (!device) {
1500                 btrfs_info(fs_info, "resizer unable to find device %llu",
1501                            devid);
1502                 ret = -ENODEV;
1503                 goto out_free;
1504         }
1505
1506         if (!device->writeable) {
1507                 btrfs_info(fs_info,
1508                            "resizer unable to apply on readonly device %llu",
1509                        devid);
1510                 ret = -EPERM;
1511                 goto out_free;
1512         }
1513
1514         if (!strcmp(sizestr, "max"))
1515                 new_size = device->bdev->bd_inode->i_size;
1516         else {
1517                 if (sizestr[0] == '-') {
1518                         mod = -1;
1519                         sizestr++;
1520                 } else if (sizestr[0] == '+') {
1521                         mod = 1;
1522                         sizestr++;
1523                 }
1524                 new_size = memparse(sizestr, &retptr);
1525                 if (*retptr != '\0' || new_size == 0) {
1526                         ret = -EINVAL;
1527                         goto out_free;
1528                 }
1529         }
1530
1531         if (device->is_tgtdev_for_dev_replace) {
1532                 ret = -EPERM;
1533                 goto out_free;
1534         }
1535
1536         old_size = btrfs_device_get_total_bytes(device);
1537
1538         if (mod < 0) {
1539                 if (new_size > old_size) {
1540                         ret = -EINVAL;
1541                         goto out_free;
1542                 }
1543                 new_size = old_size - new_size;
1544         } else if (mod > 0) {
1545                 if (new_size > ULLONG_MAX - old_size) {
1546                         ret = -ERANGE;
1547                         goto out_free;
1548                 }
1549                 new_size = old_size + new_size;
1550         }
1551
1552         if (new_size < SZ_256M) {
1553                 ret = -EINVAL;
1554                 goto out_free;
1555         }
1556         if (new_size > device->bdev->bd_inode->i_size) {
1557                 ret = -EFBIG;
1558                 goto out_free;
1559         }
1560
1561         new_size = round_down(new_size, fs_info->sectorsize);
1562
1563         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1564                           rcu_str_deref(device->name), new_size);
1565
1566         if (new_size > old_size) {
1567                 trans = btrfs_start_transaction(root, 0);
1568                 if (IS_ERR(trans)) {
1569                         ret = PTR_ERR(trans);
1570                         goto out_free;
1571                 }
1572                 ret = btrfs_grow_device(trans, device, new_size);
1573                 btrfs_commit_transaction(trans);
1574         } else if (new_size < old_size) {
1575                 ret = btrfs_shrink_device(device, new_size);
1576         } /* equal, nothing need to do */
1577
1578 out_free:
1579         kfree(vol_args);
1580 out:
1581         mutex_unlock(&fs_info->volume_mutex);
1582         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1583         mnt_drop_write_file(file);
1584         return ret;
1585 }
1586
1587 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1588                                 const char *name, unsigned long fd, int subvol,
1589                                 u64 *transid, bool readonly,
1590                                 struct btrfs_qgroup_inherit *inherit)
1591 {
1592         int namelen;
1593         int ret = 0;
1594
1595         if (!S_ISDIR(file_inode(file)->i_mode))
1596                 return -ENOTDIR;
1597
1598         ret = mnt_want_write_file(file);
1599         if (ret)
1600                 goto out;
1601
1602         namelen = strlen(name);
1603         if (strchr(name, '/')) {
1604                 ret = -EINVAL;
1605                 goto out_drop_write;
1606         }
1607
1608         if (name[0] == '.' &&
1609            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1610                 ret = -EEXIST;
1611                 goto out_drop_write;
1612         }
1613
1614         if (subvol) {
1615                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1616                                      NULL, transid, readonly, inherit);
1617         } else {
1618                 struct fd src = fdget(fd);
1619                 struct inode *src_inode;
1620                 if (!src.file) {
1621                         ret = -EINVAL;
1622                         goto out_drop_write;
1623                 }
1624
1625                 src_inode = file_inode(src.file);
1626                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1627                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1628                                    "Snapshot src from another FS");
1629                         ret = -EXDEV;
1630                 } else if (!inode_owner_or_capable(src_inode)) {
1631                         /*
1632                          * Subvolume creation is not restricted, but snapshots
1633                          * are limited to own subvolumes only
1634                          */
1635                         ret = -EPERM;
1636                 } else {
1637                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1638                                              BTRFS_I(src_inode)->root,
1639                                              transid, readonly, inherit);
1640                 }
1641                 fdput(src);
1642         }
1643 out_drop_write:
1644         mnt_drop_write_file(file);
1645 out:
1646         return ret;
1647 }
1648
1649 static noinline int btrfs_ioctl_snap_create(struct file *file,
1650                                             void __user *arg, int subvol)
1651 {
1652         struct btrfs_ioctl_vol_args *vol_args;
1653         int ret;
1654
1655         if (!S_ISDIR(file_inode(file)->i_mode))
1656                 return -ENOTDIR;
1657
1658         vol_args = memdup_user(arg, sizeof(*vol_args));
1659         if (IS_ERR(vol_args))
1660                 return PTR_ERR(vol_args);
1661         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1662
1663         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1664                                               vol_args->fd, subvol,
1665                                               NULL, false, NULL);
1666
1667         kfree(vol_args);
1668         return ret;
1669 }
1670
1671 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1672                                                void __user *arg, int subvol)
1673 {
1674         struct btrfs_ioctl_vol_args_v2 *vol_args;
1675         int ret;
1676         u64 transid = 0;
1677         u64 *ptr = NULL;
1678         bool readonly = false;
1679         struct btrfs_qgroup_inherit *inherit = NULL;
1680
1681         if (!S_ISDIR(file_inode(file)->i_mode))
1682                 return -ENOTDIR;
1683
1684         vol_args = memdup_user(arg, sizeof(*vol_args));
1685         if (IS_ERR(vol_args))
1686                 return PTR_ERR(vol_args);
1687         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1688
1689         if (vol_args->flags &
1690             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1691               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1692                 ret = -EOPNOTSUPP;
1693                 goto free_args;
1694         }
1695
1696         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1697                 ptr = &transid;
1698         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1699                 readonly = true;
1700         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1701                 if (vol_args->size > PAGE_SIZE) {
1702                         ret = -EINVAL;
1703                         goto free_args;
1704                 }
1705                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1706                 if (IS_ERR(inherit)) {
1707                         ret = PTR_ERR(inherit);
1708                         goto free_args;
1709                 }
1710         }
1711
1712         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1713                                               vol_args->fd, subvol, ptr,
1714                                               readonly, inherit);
1715         if (ret)
1716                 goto free_inherit;
1717
1718         if (ptr && copy_to_user(arg +
1719                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1720                                         transid),
1721                                 ptr, sizeof(*ptr)))
1722                 ret = -EFAULT;
1723
1724 free_inherit:
1725         kfree(inherit);
1726 free_args:
1727         kfree(vol_args);
1728         return ret;
1729 }
1730
1731 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1732                                                 void __user *arg)
1733 {
1734         struct inode *inode = file_inode(file);
1735         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1736         struct btrfs_root *root = BTRFS_I(inode)->root;
1737         int ret = 0;
1738         u64 flags = 0;
1739
1740         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1741                 return -EINVAL;
1742
1743         down_read(&fs_info->subvol_sem);
1744         if (btrfs_root_readonly(root))
1745                 flags |= BTRFS_SUBVOL_RDONLY;
1746         up_read(&fs_info->subvol_sem);
1747
1748         if (copy_to_user(arg, &flags, sizeof(flags)))
1749                 ret = -EFAULT;
1750
1751         return ret;
1752 }
1753
1754 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1755                                               void __user *arg)
1756 {
1757         struct inode *inode = file_inode(file);
1758         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1759         struct btrfs_root *root = BTRFS_I(inode)->root;
1760         struct btrfs_trans_handle *trans;
1761         u64 root_flags;
1762         u64 flags;
1763         int ret = 0;
1764
1765         if (!inode_owner_or_capable(inode))
1766                 return -EPERM;
1767
1768         ret = mnt_want_write_file(file);
1769         if (ret)
1770                 goto out;
1771
1772         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1773                 ret = -EINVAL;
1774                 goto out_drop_write;
1775         }
1776
1777         if (copy_from_user(&flags, arg, sizeof(flags))) {
1778                 ret = -EFAULT;
1779                 goto out_drop_write;
1780         }
1781
1782         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1783                 ret = -EINVAL;
1784                 goto out_drop_write;
1785         }
1786
1787         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1788                 ret = -EOPNOTSUPP;
1789                 goto out_drop_write;
1790         }
1791
1792         down_write(&fs_info->subvol_sem);
1793
1794         /* nothing to do */
1795         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1796                 goto out_drop_sem;
1797
1798         root_flags = btrfs_root_flags(&root->root_item);
1799         if (flags & BTRFS_SUBVOL_RDONLY) {
1800                 btrfs_set_root_flags(&root->root_item,
1801                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1802         } else {
1803                 /*
1804                  * Block RO -> RW transition if this subvolume is involved in
1805                  * send
1806                  */
1807                 spin_lock(&root->root_item_lock);
1808                 if (root->send_in_progress == 0) {
1809                         btrfs_set_root_flags(&root->root_item,
1810                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1811                         spin_unlock(&root->root_item_lock);
1812                 } else {
1813                         spin_unlock(&root->root_item_lock);
1814                         btrfs_warn(fs_info,
1815                                    "Attempt to set subvolume %llu read-write during send",
1816                                    root->root_key.objectid);
1817                         ret = -EPERM;
1818                         goto out_drop_sem;
1819                 }
1820         }
1821
1822         trans = btrfs_start_transaction(root, 1);
1823         if (IS_ERR(trans)) {
1824                 ret = PTR_ERR(trans);
1825                 goto out_reset;
1826         }
1827
1828         ret = btrfs_update_root(trans, fs_info->tree_root,
1829                                 &root->root_key, &root->root_item);
1830         if (ret < 0) {
1831                 btrfs_end_transaction(trans);
1832                 goto out_reset;
1833         }
1834
1835         ret = btrfs_commit_transaction(trans);
1836
1837 out_reset:
1838         if (ret)
1839                 btrfs_set_root_flags(&root->root_item, root_flags);
1840 out_drop_sem:
1841         up_write(&fs_info->subvol_sem);
1842 out_drop_write:
1843         mnt_drop_write_file(file);
1844 out:
1845         return ret;
1846 }
1847
1848 /*
1849  * helper to check if the subvolume references other subvolumes
1850  */
1851 static noinline int may_destroy_subvol(struct btrfs_root *root)
1852 {
1853         struct btrfs_fs_info *fs_info = root->fs_info;
1854         struct btrfs_path *path;
1855         struct btrfs_dir_item *di;
1856         struct btrfs_key key;
1857         u64 dir_id;
1858         int ret;
1859
1860         path = btrfs_alloc_path();
1861         if (!path)
1862                 return -ENOMEM;
1863
1864         /* Make sure this root isn't set as the default subvol */
1865         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1866         di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
1867                                    dir_id, "default", 7, 0);
1868         if (di && !IS_ERR(di)) {
1869                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1870                 if (key.objectid == root->root_key.objectid) {
1871                         ret = -EPERM;
1872                         btrfs_err(fs_info,
1873                                   "deleting default subvolume %llu is not allowed",
1874                                   key.objectid);
1875                         goto out;
1876                 }
1877                 btrfs_release_path(path);
1878         }
1879
1880         key.objectid = root->root_key.objectid;
1881         key.type = BTRFS_ROOT_REF_KEY;
1882         key.offset = (u64)-1;
1883
1884         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1885         if (ret < 0)
1886                 goto out;
1887         BUG_ON(ret == 0);
1888
1889         ret = 0;
1890         if (path->slots[0] > 0) {
1891                 path->slots[0]--;
1892                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1893                 if (key.objectid == root->root_key.objectid &&
1894                     key.type == BTRFS_ROOT_REF_KEY)
1895                         ret = -ENOTEMPTY;
1896         }
1897 out:
1898         btrfs_free_path(path);
1899         return ret;
1900 }
1901
1902 static noinline int key_in_sk(struct btrfs_key *key,
1903                               struct btrfs_ioctl_search_key *sk)
1904 {
1905         struct btrfs_key test;
1906         int ret;
1907
1908         test.objectid = sk->min_objectid;
1909         test.type = sk->min_type;
1910         test.offset = sk->min_offset;
1911
1912         ret = btrfs_comp_cpu_keys(key, &test);
1913         if (ret < 0)
1914                 return 0;
1915
1916         test.objectid = sk->max_objectid;
1917         test.type = sk->max_type;
1918         test.offset = sk->max_offset;
1919
1920         ret = btrfs_comp_cpu_keys(key, &test);
1921         if (ret > 0)
1922                 return 0;
1923         return 1;
1924 }
1925
1926 static noinline int copy_to_sk(struct btrfs_path *path,
1927                                struct btrfs_key *key,
1928                                struct btrfs_ioctl_search_key *sk,
1929                                size_t *buf_size,
1930                                char __user *ubuf,
1931                                unsigned long *sk_offset,
1932                                int *num_found)
1933 {
1934         u64 found_transid;
1935         struct extent_buffer *leaf;
1936         struct btrfs_ioctl_search_header sh;
1937         struct btrfs_key test;
1938         unsigned long item_off;
1939         unsigned long item_len;
1940         int nritems;
1941         int i;
1942         int slot;
1943         int ret = 0;
1944
1945         leaf = path->nodes[0];
1946         slot = path->slots[0];
1947         nritems = btrfs_header_nritems(leaf);
1948
1949         if (btrfs_header_generation(leaf) > sk->max_transid) {
1950                 i = nritems;
1951                 goto advance_key;
1952         }
1953         found_transid = btrfs_header_generation(leaf);
1954
1955         for (i = slot; i < nritems; i++) {
1956                 item_off = btrfs_item_ptr_offset(leaf, i);
1957                 item_len = btrfs_item_size_nr(leaf, i);
1958
1959                 btrfs_item_key_to_cpu(leaf, key, i);
1960                 if (!key_in_sk(key, sk))
1961                         continue;
1962
1963                 if (sizeof(sh) + item_len > *buf_size) {
1964                         if (*num_found) {
1965                                 ret = 1;
1966                                 goto out;
1967                         }
1968
1969                         /*
1970                          * return one empty item back for v1, which does not
1971                          * handle -EOVERFLOW
1972                          */
1973
1974                         *buf_size = sizeof(sh) + item_len;
1975                         item_len = 0;
1976                         ret = -EOVERFLOW;
1977                 }
1978
1979                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1980                         ret = 1;
1981                         goto out;
1982                 }
1983
1984                 sh.objectid = key->objectid;
1985                 sh.offset = key->offset;
1986                 sh.type = key->type;
1987                 sh.len = item_len;
1988                 sh.transid = found_transid;
1989
1990                 /* copy search result header */
1991                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
1992                         ret = -EFAULT;
1993                         goto out;
1994                 }
1995
1996                 *sk_offset += sizeof(sh);
1997
1998                 if (item_len) {
1999                         char __user *up = ubuf + *sk_offset;
2000                         /* copy the item */
2001                         if (read_extent_buffer_to_user(leaf, up,
2002                                                        item_off, item_len)) {
2003                                 ret = -EFAULT;
2004                                 goto out;
2005                         }
2006
2007                         *sk_offset += item_len;
2008                 }
2009                 (*num_found)++;
2010
2011                 if (ret) /* -EOVERFLOW from above */
2012                         goto out;
2013
2014                 if (*num_found >= sk->nr_items) {
2015                         ret = 1;
2016                         goto out;
2017                 }
2018         }
2019 advance_key:
2020         ret = 0;
2021         test.objectid = sk->max_objectid;
2022         test.type = sk->max_type;
2023         test.offset = sk->max_offset;
2024         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2025                 ret = 1;
2026         else if (key->offset < (u64)-1)
2027                 key->offset++;
2028         else if (key->type < (u8)-1) {
2029                 key->offset = 0;
2030                 key->type++;
2031         } else if (key->objectid < (u64)-1) {
2032                 key->offset = 0;
2033                 key->type = 0;
2034                 key->objectid++;
2035         } else
2036                 ret = 1;
2037 out:
2038         /*
2039          *  0: all items from this leaf copied, continue with next
2040          *  1: * more items can be copied, but unused buffer is too small
2041          *     * all items were found
2042          *     Either way, it will stops the loop which iterates to the next
2043          *     leaf
2044          *  -EOVERFLOW: item was to large for buffer
2045          *  -EFAULT: could not copy extent buffer back to userspace
2046          */
2047         return ret;
2048 }
2049
2050 static noinline int search_ioctl(struct inode *inode,
2051                                  struct btrfs_ioctl_search_key *sk,
2052                                  size_t *buf_size,
2053                                  char __user *ubuf)
2054 {
2055         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2056         struct btrfs_root *root;
2057         struct btrfs_key key;
2058         struct btrfs_path *path;
2059         int ret;
2060         int num_found = 0;
2061         unsigned long sk_offset = 0;
2062
2063         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2064                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2065                 return -EOVERFLOW;
2066         }
2067
2068         path = btrfs_alloc_path();
2069         if (!path)
2070                 return -ENOMEM;
2071
2072         if (sk->tree_id == 0) {
2073                 /* search the root of the inode that was passed */
2074                 root = BTRFS_I(inode)->root;
2075         } else {
2076                 key.objectid = sk->tree_id;
2077                 key.type = BTRFS_ROOT_ITEM_KEY;
2078                 key.offset = (u64)-1;
2079                 root = btrfs_read_fs_root_no_name(info, &key);
2080                 if (IS_ERR(root)) {
2081                         btrfs_free_path(path);
2082                         return -ENOENT;
2083                 }
2084         }
2085
2086         key.objectid = sk->min_objectid;
2087         key.type = sk->min_type;
2088         key.offset = sk->min_offset;
2089
2090         while (1) {
2091                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2092                 if (ret != 0) {
2093                         if (ret > 0)
2094                                 ret = 0;
2095                         goto err;
2096                 }
2097                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2098                                  &sk_offset, &num_found);
2099                 btrfs_release_path(path);
2100                 if (ret)
2101                         break;
2102
2103         }
2104         if (ret > 0)
2105                 ret = 0;
2106 err:
2107         sk->nr_items = num_found;
2108         btrfs_free_path(path);
2109         return ret;
2110 }
2111
2112 static noinline int btrfs_ioctl_tree_search(struct file *file,
2113                                            void __user *argp)
2114 {
2115         struct btrfs_ioctl_search_args __user *uargs;
2116         struct btrfs_ioctl_search_key sk;
2117         struct inode *inode;
2118         int ret;
2119         size_t buf_size;
2120
2121         if (!capable(CAP_SYS_ADMIN))
2122                 return -EPERM;
2123
2124         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2125
2126         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2127                 return -EFAULT;
2128
2129         buf_size = sizeof(uargs->buf);
2130
2131         inode = file_inode(file);
2132         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2133
2134         /*
2135          * In the origin implementation an overflow is handled by returning a
2136          * search header with a len of zero, so reset ret.
2137          */
2138         if (ret == -EOVERFLOW)
2139                 ret = 0;
2140
2141         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2142                 ret = -EFAULT;
2143         return ret;
2144 }
2145
2146 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2147                                                void __user *argp)
2148 {
2149         struct btrfs_ioctl_search_args_v2 __user *uarg;
2150         struct btrfs_ioctl_search_args_v2 args;
2151         struct inode *inode;
2152         int ret;
2153         size_t buf_size;
2154         const size_t buf_limit = SZ_16M;
2155
2156         if (!capable(CAP_SYS_ADMIN))
2157                 return -EPERM;
2158
2159         /* copy search header and buffer size */
2160         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2161         if (copy_from_user(&args, uarg, sizeof(args)))
2162                 return -EFAULT;
2163
2164         buf_size = args.buf_size;
2165
2166         /* limit result size to 16MB */
2167         if (buf_size > buf_limit)
2168                 buf_size = buf_limit;
2169
2170         inode = file_inode(file);
2171         ret = search_ioctl(inode, &args.key, &buf_size,
2172                            (char __user *)(&uarg->buf[0]));
2173         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2174                 ret = -EFAULT;
2175         else if (ret == -EOVERFLOW &&
2176                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2177                 ret = -EFAULT;
2178
2179         return ret;
2180 }
2181
2182 /*
2183  * Search INODE_REFs to identify path name of 'dirid' directory
2184  * in a 'tree_id' tree. and sets path name to 'name'.
2185  */
2186 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2187                                 u64 tree_id, u64 dirid, char *name)
2188 {
2189         struct btrfs_root *root;
2190         struct btrfs_key key;
2191         char *ptr;
2192         int ret = -1;
2193         int slot;
2194         int len;
2195         int total_len = 0;
2196         struct btrfs_inode_ref *iref;
2197         struct extent_buffer *l;
2198         struct btrfs_path *path;
2199
2200         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2201                 name[0]='\0';
2202                 return 0;
2203         }
2204
2205         path = btrfs_alloc_path();
2206         if (!path)
2207                 return -ENOMEM;
2208
2209         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
2210
2211         key.objectid = tree_id;
2212         key.type = BTRFS_ROOT_ITEM_KEY;
2213         key.offset = (u64)-1;
2214         root = btrfs_read_fs_root_no_name(info, &key);
2215         if (IS_ERR(root)) {
2216                 btrfs_err(info, "could not find root %llu", tree_id);
2217                 ret = -ENOENT;
2218                 goto out;
2219         }
2220
2221         key.objectid = dirid;
2222         key.type = BTRFS_INODE_REF_KEY;
2223         key.offset = (u64)-1;
2224
2225         while (1) {
2226                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2227                 if (ret < 0)
2228                         goto out;
2229                 else if (ret > 0) {
2230                         ret = btrfs_previous_item(root, path, dirid,
2231                                                   BTRFS_INODE_REF_KEY);
2232                         if (ret < 0)
2233                                 goto out;
2234                         else if (ret > 0) {
2235                                 ret = -ENOENT;
2236                                 goto out;
2237                         }
2238                 }
2239
2240                 l = path->nodes[0];
2241                 slot = path->slots[0];
2242                 btrfs_item_key_to_cpu(l, &key, slot);
2243
2244                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2245                 len = btrfs_inode_ref_name_len(l, iref);
2246                 ptr -= len + 1;
2247                 total_len += len + 1;
2248                 if (ptr < name) {
2249                         ret = -ENAMETOOLONG;
2250                         goto out;
2251                 }
2252
2253                 *(ptr + len) = '/';
2254                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2255
2256                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2257                         break;
2258
2259                 btrfs_release_path(path);
2260                 key.objectid = key.offset;
2261                 key.offset = (u64)-1;
2262                 dirid = key.objectid;
2263         }
2264         memmove(name, ptr, total_len);
2265         name[total_len] = '\0';
2266         ret = 0;
2267 out:
2268         btrfs_free_path(path);
2269         return ret;
2270 }
2271
2272 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2273                                            void __user *argp)
2274 {
2275          struct btrfs_ioctl_ino_lookup_args *args;
2276          struct inode *inode;
2277         int ret = 0;
2278
2279         args = memdup_user(argp, sizeof(*args));
2280         if (IS_ERR(args))
2281                 return PTR_ERR(args);
2282
2283         inode = file_inode(file);
2284
2285         /*
2286          * Unprivileged query to obtain the containing subvolume root id. The
2287          * path is reset so it's consistent with btrfs_search_path_in_tree.
2288          */
2289         if (args->treeid == 0)
2290                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2291
2292         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2293                 args->name[0] = 0;
2294                 goto out;
2295         }
2296
2297         if (!capable(CAP_SYS_ADMIN)) {
2298                 ret = -EPERM;
2299                 goto out;
2300         }
2301
2302         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2303                                         args->treeid, args->objectid,
2304                                         args->name);
2305
2306 out:
2307         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2308                 ret = -EFAULT;
2309
2310         kfree(args);
2311         return ret;
2312 }
2313
2314 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2315                                              void __user *arg)
2316 {
2317         struct dentry *parent = file->f_path.dentry;
2318         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2319         struct dentry *dentry;
2320         struct inode *dir = d_inode(parent);
2321         struct inode *inode;
2322         struct btrfs_root *root = BTRFS_I(dir)->root;
2323         struct btrfs_root *dest = NULL;
2324         struct btrfs_ioctl_vol_args *vol_args;
2325         struct btrfs_trans_handle *trans;
2326         struct btrfs_block_rsv block_rsv;
2327         u64 root_flags;
2328         u64 qgroup_reserved;
2329         int namelen;
2330         int ret;
2331         int err = 0;
2332
2333         if (!S_ISDIR(dir->i_mode))
2334                 return -ENOTDIR;
2335
2336         vol_args = memdup_user(arg, sizeof(*vol_args));
2337         if (IS_ERR(vol_args))
2338                 return PTR_ERR(vol_args);
2339
2340         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2341         namelen = strlen(vol_args->name);
2342         if (strchr(vol_args->name, '/') ||
2343             strncmp(vol_args->name, "..", namelen) == 0) {
2344                 err = -EINVAL;
2345                 goto out;
2346         }
2347
2348         err = mnt_want_write_file(file);
2349         if (err)
2350                 goto out;
2351
2352
2353         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2354         if (err == -EINTR)
2355                 goto out_drop_write;
2356         dentry = lookup_one_len(vol_args->name, parent, namelen);
2357         if (IS_ERR(dentry)) {
2358                 err = PTR_ERR(dentry);
2359                 goto out_unlock_dir;
2360         }
2361
2362         if (d_really_is_negative(dentry)) {
2363                 err = -ENOENT;
2364                 goto out_dput;
2365         }
2366
2367         inode = d_inode(dentry);
2368         dest = BTRFS_I(inode)->root;
2369         if (!capable(CAP_SYS_ADMIN)) {
2370                 /*
2371                  * Regular user.  Only allow this with a special mount
2372                  * option, when the user has write+exec access to the
2373                  * subvol root, and when rmdir(2) would have been
2374                  * allowed.
2375                  *
2376                  * Note that this is _not_ check that the subvol is
2377                  * empty or doesn't contain data that we wouldn't
2378                  * otherwise be able to delete.
2379                  *
2380                  * Users who want to delete empty subvols should try
2381                  * rmdir(2).
2382                  */
2383                 err = -EPERM;
2384                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2385                         goto out_dput;
2386
2387                 /*
2388                  * Do not allow deletion if the parent dir is the same
2389                  * as the dir to be deleted.  That means the ioctl
2390                  * must be called on the dentry referencing the root
2391                  * of the subvol, not a random directory contained
2392                  * within it.
2393                  */
2394                 err = -EINVAL;
2395                 if (root == dest)
2396                         goto out_dput;
2397
2398                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2399                 if (err)
2400                         goto out_dput;
2401         }
2402
2403         /* check if subvolume may be deleted by a user */
2404         err = btrfs_may_delete(dir, dentry, 1);
2405         if (err)
2406                 goto out_dput;
2407
2408         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2409                 err = -EINVAL;
2410                 goto out_dput;
2411         }
2412
2413         inode_lock(inode);
2414
2415         /*
2416          * Don't allow to delete a subvolume with send in progress. This is
2417          * inside the i_mutex so the error handling that has to drop the bit
2418          * again is not run concurrently.
2419          */
2420         spin_lock(&dest->root_item_lock);
2421         root_flags = btrfs_root_flags(&dest->root_item);
2422         if (dest->send_in_progress == 0) {
2423                 btrfs_set_root_flags(&dest->root_item,
2424                                 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
2425                 spin_unlock(&dest->root_item_lock);
2426         } else {
2427                 spin_unlock(&dest->root_item_lock);
2428                 btrfs_warn(fs_info,
2429                            "Attempt to delete subvolume %llu during send",
2430                            dest->root_key.objectid);
2431                 err = -EPERM;
2432                 goto out_unlock_inode;
2433         }
2434
2435         down_write(&fs_info->subvol_sem);
2436
2437         err = may_destroy_subvol(dest);
2438         if (err)
2439                 goto out_up_write;
2440
2441         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
2442         /*
2443          * One for dir inode, two for dir entries, two for root
2444          * ref/backref.
2445          */
2446         err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
2447                                                5, &qgroup_reserved, true);
2448         if (err)
2449                 goto out_up_write;
2450
2451         trans = btrfs_start_transaction(root, 0);
2452         if (IS_ERR(trans)) {
2453                 err = PTR_ERR(trans);
2454                 goto out_release;
2455         }
2456         trans->block_rsv = &block_rsv;
2457         trans->bytes_reserved = block_rsv.size;
2458
2459         btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
2460
2461         ret = btrfs_unlink_subvol(trans, root, dir,
2462                                 dest->root_key.objectid,
2463                                 dentry->d_name.name,
2464                                 dentry->d_name.len);
2465         if (ret) {
2466                 err = ret;
2467                 btrfs_abort_transaction(trans, ret);
2468                 goto out_end_trans;
2469         }
2470
2471         btrfs_record_root_in_trans(trans, dest);
2472
2473         memset(&dest->root_item.drop_progress, 0,
2474                 sizeof(dest->root_item.drop_progress));
2475         dest->root_item.drop_level = 0;
2476         btrfs_set_root_refs(&dest->root_item, 0);
2477
2478         if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
2479                 ret = btrfs_insert_orphan_item(trans,
2480                                         fs_info->tree_root,
2481                                         dest->root_key.objectid);
2482                 if (ret) {
2483                         btrfs_abort_transaction(trans, ret);
2484                         err = ret;
2485                         goto out_end_trans;
2486                 }
2487         }
2488
2489         ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
2490                                   BTRFS_UUID_KEY_SUBVOL,
2491                                   dest->root_key.objectid);
2492         if (ret && ret != -ENOENT) {
2493                 btrfs_abort_transaction(trans, ret);
2494                 err = ret;
2495                 goto out_end_trans;
2496         }
2497         if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
2498                 ret = btrfs_uuid_tree_rem(trans, fs_info,
2499                                           dest->root_item.received_uuid,
2500                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
2501                                           dest->root_key.objectid);
2502                 if (ret && ret != -ENOENT) {
2503                         btrfs_abort_transaction(trans, ret);
2504                         err = ret;
2505                         goto out_end_trans;
2506                 }
2507         }
2508
2509 out_end_trans:
2510         trans->block_rsv = NULL;
2511         trans->bytes_reserved = 0;
2512         ret = btrfs_end_transaction(trans);
2513         if (ret && !err)
2514                 err = ret;
2515         inode->i_flags |= S_DEAD;
2516 out_release:
2517         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
2518 out_up_write:
2519         up_write(&fs_info->subvol_sem);
2520         if (err) {
2521                 spin_lock(&dest->root_item_lock);
2522                 root_flags = btrfs_root_flags(&dest->root_item);
2523                 btrfs_set_root_flags(&dest->root_item,
2524                                 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
2525                 spin_unlock(&dest->root_item_lock);
2526         }
2527 out_unlock_inode:
2528         inode_unlock(inode);
2529         if (!err) {
2530                 d_invalidate(dentry);
2531                 btrfs_invalidate_inodes(dest);
2532                 d_delete(dentry);
2533                 ASSERT(dest->send_in_progress == 0);
2534
2535                 /* the last ref */
2536                 if (dest->ino_cache_inode) {
2537                         iput(dest->ino_cache_inode);
2538                         dest->ino_cache_inode = NULL;
2539                 }
2540         }
2541 out_dput:
2542         dput(dentry);
2543 out_unlock_dir:
2544         inode_unlock(dir);
2545 out_drop_write:
2546         mnt_drop_write_file(file);
2547 out:
2548         kfree(vol_args);
2549         return err;
2550 }
2551
2552 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2553 {
2554         struct inode *inode = file_inode(file);
2555         struct btrfs_root *root = BTRFS_I(inode)->root;
2556         struct btrfs_ioctl_defrag_range_args *range;
2557         int ret;
2558
2559         ret = mnt_want_write_file(file);
2560         if (ret)
2561                 return ret;
2562
2563         if (btrfs_root_readonly(root)) {
2564                 ret = -EROFS;
2565                 goto out;
2566         }
2567
2568         switch (inode->i_mode & S_IFMT) {
2569         case S_IFDIR:
2570                 if (!capable(CAP_SYS_ADMIN)) {
2571                         ret = -EPERM;
2572                         goto out;
2573                 }
2574                 ret = btrfs_defrag_root(root);
2575                 break;
2576         case S_IFREG:
2577                 if (!(file->f_mode & FMODE_WRITE)) {
2578                         ret = -EINVAL;
2579                         goto out;
2580                 }
2581
2582                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2583                 if (!range) {
2584                         ret = -ENOMEM;
2585                         goto out;
2586                 }
2587
2588                 if (argp) {
2589                         if (copy_from_user(range, argp,
2590                                            sizeof(*range))) {
2591                                 ret = -EFAULT;
2592                                 kfree(range);
2593                                 goto out;
2594                         }
2595                         /* compression requires us to start the IO */
2596                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2597                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2598                                 range->extent_thresh = (u32)-1;
2599                         }
2600                 } else {
2601                         /* the rest are all set to zero by kzalloc */
2602                         range->len = (u64)-1;
2603                 }
2604                 ret = btrfs_defrag_file(file_inode(file), file,
2605                                         range, 0, 0);
2606                 if (ret > 0)
2607                         ret = 0;
2608                 kfree(range);
2609                 break;
2610         default:
2611                 ret = -EINVAL;
2612         }
2613 out:
2614         mnt_drop_write_file(file);
2615         return ret;
2616 }
2617
2618 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2619 {
2620         struct btrfs_ioctl_vol_args *vol_args;
2621         int ret;
2622
2623         if (!capable(CAP_SYS_ADMIN))
2624                 return -EPERM;
2625
2626         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
2627                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2628
2629         mutex_lock(&fs_info->volume_mutex);
2630         vol_args = memdup_user(arg, sizeof(*vol_args));
2631         if (IS_ERR(vol_args)) {
2632                 ret = PTR_ERR(vol_args);
2633                 goto out;
2634         }
2635
2636         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2637         ret = btrfs_init_new_device(fs_info, vol_args->name);
2638
2639         if (!ret)
2640                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2641
2642         kfree(vol_args);
2643 out:
2644         mutex_unlock(&fs_info->volume_mutex);
2645         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2646         return ret;
2647 }
2648
2649 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2650 {
2651         struct inode *inode = file_inode(file);
2652         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2653         struct btrfs_ioctl_vol_args_v2 *vol_args;
2654         int ret;
2655
2656         if (!capable(CAP_SYS_ADMIN))
2657                 return -EPERM;
2658
2659         ret = mnt_want_write_file(file);
2660         if (ret)
2661                 return ret;
2662
2663         vol_args = memdup_user(arg, sizeof(*vol_args));
2664         if (IS_ERR(vol_args)) {
2665                 ret = PTR_ERR(vol_args);
2666                 goto err_drop;
2667         }
2668
2669         /* Check for compatibility reject unknown flags */
2670         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED)
2671                 return -EOPNOTSUPP;
2672
2673         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2674                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2675                 goto out;
2676         }
2677
2678         mutex_lock(&fs_info->volume_mutex);
2679         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2680                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2681         } else {
2682                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2683                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2684         }
2685         mutex_unlock(&fs_info->volume_mutex);
2686         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2687
2688         if (!ret) {
2689                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2690                         btrfs_info(fs_info, "device deleted: id %llu",
2691                                         vol_args->devid);
2692                 else
2693                         btrfs_info(fs_info, "device deleted: %s",
2694                                         vol_args->name);
2695         }
2696 out:
2697         kfree(vol_args);
2698 err_drop:
2699         mnt_drop_write_file(file);
2700         return ret;
2701 }
2702
2703 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2704 {
2705         struct inode *inode = file_inode(file);
2706         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2707         struct btrfs_ioctl_vol_args *vol_args;
2708         int ret;
2709
2710         if (!capable(CAP_SYS_ADMIN))
2711                 return -EPERM;
2712
2713         ret = mnt_want_write_file(file);
2714         if (ret)
2715                 return ret;
2716
2717         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2718                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2719                 goto out_drop_write;
2720         }
2721
2722         vol_args = memdup_user(arg, sizeof(*vol_args));
2723         if (IS_ERR(vol_args)) {
2724                 ret = PTR_ERR(vol_args);
2725                 goto out;
2726         }
2727
2728         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2729         mutex_lock(&fs_info->volume_mutex);
2730         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2731         mutex_unlock(&fs_info->volume_mutex);
2732
2733         if (!ret)
2734                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2735         kfree(vol_args);
2736 out:
2737         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2738 out_drop_write:
2739         mnt_drop_write_file(file);
2740
2741         return ret;
2742 }
2743
2744 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2745                                 void __user *arg)
2746 {
2747         struct btrfs_ioctl_fs_info_args *fi_args;
2748         struct btrfs_device *device;
2749         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2750         int ret = 0;
2751
2752         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2753         if (!fi_args)
2754                 return -ENOMEM;
2755
2756         mutex_lock(&fs_devices->device_list_mutex);
2757         fi_args->num_devices = fs_devices->num_devices;
2758         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2759
2760         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2761                 if (device->devid > fi_args->max_id)
2762                         fi_args->max_id = device->devid;
2763         }
2764         mutex_unlock(&fs_devices->device_list_mutex);
2765
2766         fi_args->nodesize = fs_info->nodesize;
2767         fi_args->sectorsize = fs_info->sectorsize;
2768         fi_args->clone_alignment = fs_info->sectorsize;
2769
2770         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2771                 ret = -EFAULT;
2772
2773         kfree(fi_args);
2774         return ret;
2775 }
2776
2777 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2778                                  void __user *arg)
2779 {
2780         struct btrfs_ioctl_dev_info_args *di_args;
2781         struct btrfs_device *dev;
2782         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2783         int ret = 0;
2784         char *s_uuid = NULL;
2785
2786         di_args = memdup_user(arg, sizeof(*di_args));
2787         if (IS_ERR(di_args))
2788                 return PTR_ERR(di_args);
2789
2790         if (!btrfs_is_empty_uuid(di_args->uuid))
2791                 s_uuid = di_args->uuid;
2792
2793         mutex_lock(&fs_devices->device_list_mutex);
2794         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2795
2796         if (!dev) {
2797                 ret = -ENODEV;
2798                 goto out;
2799         }
2800
2801         di_args->devid = dev->devid;
2802         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2803         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2804         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2805         if (dev->name) {
2806                 struct rcu_string *name;
2807
2808                 rcu_read_lock();
2809                 name = rcu_dereference(dev->name);
2810                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2811                 rcu_read_unlock();
2812                 di_args->path[sizeof(di_args->path) - 1] = 0;
2813         } else {
2814                 di_args->path[0] = '\0';
2815         }
2816
2817 out:
2818         mutex_unlock(&fs_devices->device_list_mutex);
2819         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2820                 ret = -EFAULT;
2821
2822         kfree(di_args);
2823         return ret;
2824 }
2825
2826 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2827 {
2828         struct page *page;
2829
2830         page = grab_cache_page(inode->i_mapping, index);
2831         if (!page)
2832                 return ERR_PTR(-ENOMEM);
2833
2834         if (!PageUptodate(page)) {
2835                 int ret;
2836
2837                 ret = btrfs_readpage(NULL, page);
2838                 if (ret)
2839                         return ERR_PTR(ret);
2840                 lock_page(page);
2841                 if (!PageUptodate(page)) {
2842                         unlock_page(page);
2843                         put_page(page);
2844                         return ERR_PTR(-EIO);
2845                 }
2846                 if (page->mapping != inode->i_mapping) {
2847                         unlock_page(page);
2848                         put_page(page);
2849                         return ERR_PTR(-EAGAIN);
2850                 }
2851         }
2852
2853         return page;
2854 }
2855
2856 static int gather_extent_pages(struct inode *inode, struct page **pages,
2857                                int num_pages, u64 off)
2858 {
2859         int i;
2860         pgoff_t index = off >> PAGE_SHIFT;
2861
2862         for (i = 0; i < num_pages; i++) {
2863 again:
2864                 pages[i] = extent_same_get_page(inode, index + i);
2865                 if (IS_ERR(pages[i])) {
2866                         int err = PTR_ERR(pages[i]);
2867
2868                         if (err == -EAGAIN)
2869                                 goto again;
2870                         pages[i] = NULL;
2871                         return err;
2872                 }
2873         }
2874         return 0;
2875 }
2876
2877 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2878                              bool retry_range_locking)
2879 {
2880         /*
2881          * Do any pending delalloc/csum calculations on inode, one way or
2882          * another, and lock file content.
2883          * The locking order is:
2884          *
2885          *   1) pages
2886          *   2) range in the inode's io tree
2887          */
2888         while (1) {
2889                 struct btrfs_ordered_extent *ordered;
2890                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2891                 ordered = btrfs_lookup_first_ordered_extent(inode,
2892                                                             off + len - 1);
2893                 if ((!ordered ||
2894                      ordered->file_offset + ordered->len <= off ||
2895                      ordered->file_offset >= off + len) &&
2896                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2897                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2898                         if (ordered)
2899                                 btrfs_put_ordered_extent(ordered);
2900                         break;
2901                 }
2902                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2903                 if (ordered)
2904                         btrfs_put_ordered_extent(ordered);
2905                 if (!retry_range_locking)
2906                         return -EAGAIN;
2907                 btrfs_wait_ordered_range(inode, off, len);
2908         }
2909         return 0;
2910 }
2911
2912 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2913 {
2914         inode_unlock(inode1);
2915         inode_unlock(inode2);
2916 }
2917
2918 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2919 {
2920         if (inode1 < inode2)
2921                 swap(inode1, inode2);
2922
2923         inode_lock_nested(inode1, I_MUTEX_PARENT);
2924         inode_lock_nested(inode2, I_MUTEX_CHILD);
2925 }
2926
2927 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2928                                       struct inode *inode2, u64 loff2, u64 len)
2929 {
2930         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2931         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2932 }
2933
2934 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2935                                     struct inode *inode2, u64 loff2, u64 len,
2936                                     bool retry_range_locking)
2937 {
2938         int ret;
2939
2940         if (inode1 < inode2) {
2941                 swap(inode1, inode2);
2942                 swap(loff1, loff2);
2943         }
2944         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2945         if (ret)
2946                 return ret;
2947         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2948         if (ret)
2949                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2950                               loff1 + len - 1);
2951         return ret;
2952 }
2953
2954 struct cmp_pages {
2955         int             num_pages;
2956         struct page     **src_pages;
2957         struct page     **dst_pages;
2958 };
2959
2960 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2961 {
2962         int i;
2963         struct page *pg;
2964
2965         for (i = 0; i < cmp->num_pages; i++) {
2966                 pg = cmp->src_pages[i];
2967                 if (pg) {
2968                         unlock_page(pg);
2969                         put_page(pg);
2970                 }
2971                 pg = cmp->dst_pages[i];
2972                 if (pg) {
2973                         unlock_page(pg);
2974                         put_page(pg);
2975                 }
2976         }
2977         kfree(cmp->src_pages);
2978         kfree(cmp->dst_pages);
2979 }
2980
2981 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2982                                   struct inode *dst, u64 dst_loff,
2983                                   u64 len, struct cmp_pages *cmp)
2984 {
2985         int ret;
2986         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2987         struct page **src_pgarr, **dst_pgarr;
2988
2989         /*
2990          * We must gather up all the pages before we initiate our
2991          * extent locking. We use an array for the page pointers. Size
2992          * of the array is bounded by len, which is in turn bounded by
2993          * BTRFS_MAX_DEDUPE_LEN.
2994          */
2995         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2996         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2997         if (!src_pgarr || !dst_pgarr) {
2998                 kfree(src_pgarr);
2999                 kfree(dst_pgarr);
3000                 return -ENOMEM;
3001         }
3002         cmp->num_pages = num_pages;
3003         cmp->src_pages = src_pgarr;
3004         cmp->dst_pages = dst_pgarr;
3005
3006         /*
3007          * If deduping ranges in the same inode, locking rules make it mandatory
3008          * to always lock pages in ascending order to avoid deadlocks with
3009          * concurrent tasks (such as starting writeback/delalloc).
3010          */
3011         if (src == dst && dst_loff < loff) {
3012                 swap(src_pgarr, dst_pgarr);
3013                 swap(loff, dst_loff);
3014         }
3015
3016         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3017         if (ret)
3018                 goto out;
3019
3020         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3021
3022 out:
3023         if (ret)
3024                 btrfs_cmp_data_free(cmp);
3025         return ret;
3026 }
3027
3028 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3029 {
3030         int ret = 0;
3031         int i;
3032         struct page *src_page, *dst_page;
3033         unsigned int cmp_len = PAGE_SIZE;
3034         void *addr, *dst_addr;
3035
3036         i = 0;
3037         while (len) {
3038                 if (len < PAGE_SIZE)
3039                         cmp_len = len;
3040
3041                 BUG_ON(i >= cmp->num_pages);
3042
3043                 src_page = cmp->src_pages[i];
3044                 dst_page = cmp->dst_pages[i];
3045                 ASSERT(PageLocked(src_page));
3046                 ASSERT(PageLocked(dst_page));
3047
3048                 addr = kmap_atomic(src_page);
3049                 dst_addr = kmap_atomic(dst_page);
3050
3051                 flush_dcache_page(src_page);
3052                 flush_dcache_page(dst_page);
3053
3054                 if (memcmp(addr, dst_addr, cmp_len))
3055                         ret = -EBADE;
3056
3057                 kunmap_atomic(addr);
3058                 kunmap_atomic(dst_addr);
3059
3060                 if (ret)
3061                         break;
3062
3063                 len -= cmp_len;
3064                 i++;
3065         }
3066
3067         return ret;
3068 }
3069
3070 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3071                                      u64 olen)
3072 {
3073         u64 len = *plen;
3074         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3075
3076         if (off + olen > inode->i_size || off + olen < off)
3077                 return -EINVAL;
3078
3079         /* if we extend to eof, continue to block boundary */
3080         if (off + len == inode->i_size)
3081                 *plen = len = ALIGN(inode->i_size, bs) - off;
3082
3083         /* Check that we are block aligned - btrfs_clone() requires this */
3084         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3085                 return -EINVAL;
3086
3087         return 0;
3088 }
3089
3090 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3091                              struct inode *dst, u64 dst_loff)
3092 {
3093         int ret;
3094         u64 len = olen;
3095         struct cmp_pages cmp;
3096         bool same_inode = (src == dst);
3097         u64 same_lock_start = 0;
3098         u64 same_lock_len = 0;
3099
3100         if (len == 0)
3101                 return 0;
3102
3103         if (same_inode)
3104                 inode_lock(src);
3105         else
3106                 btrfs_double_inode_lock(src, dst);
3107
3108         ret = extent_same_check_offsets(src, loff, &len, olen);
3109         if (ret)
3110                 goto out_unlock;
3111
3112         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3113         if (ret)
3114                 goto out_unlock;
3115
3116         if (same_inode) {
3117                 /*
3118                  * Single inode case wants the same checks, except we
3119                  * don't want our length pushed out past i_size as
3120                  * comparing that data range makes no sense.
3121                  *
3122                  * extent_same_check_offsets() will do this for an
3123                  * unaligned length at i_size, so catch it here and
3124                  * reject the request.
3125                  *
3126                  * This effectively means we require aligned extents
3127                  * for the single-inode case, whereas the other cases
3128                  * allow an unaligned length so long as it ends at
3129                  * i_size.
3130                  */
3131                 if (len != olen) {
3132                         ret = -EINVAL;
3133                         goto out_unlock;
3134                 }
3135
3136                 /* Check for overlapping ranges */
3137                 if (dst_loff + len > loff && dst_loff < loff + len) {
3138                         ret = -EINVAL;
3139                         goto out_unlock;
3140                 }
3141
3142                 same_lock_start = min_t(u64, loff, dst_loff);
3143                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3144         }
3145
3146         /* don't make the dst file partly checksummed */
3147         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3148             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3149                 ret = -EINVAL;
3150                 goto out_unlock;
3151         }
3152
3153 again:
3154         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3155         if (ret)
3156                 goto out_unlock;
3157
3158         if (same_inode)
3159                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3160                                         false);
3161         else
3162                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3163                                                false);
3164         /*
3165          * If one of the inodes has dirty pages in the respective range or
3166          * ordered extents, we need to flush dellaloc and wait for all ordered
3167          * extents in the range. We must unlock the pages and the ranges in the
3168          * io trees to avoid deadlocks when flushing delalloc (requires locking
3169          * pages) and when waiting for ordered extents to complete (they require
3170          * range locking).
3171          */
3172         if (ret == -EAGAIN) {
3173                 /*
3174                  * Ranges in the io trees already unlocked. Now unlock all
3175                  * pages before waiting for all IO to complete.
3176                  */
3177                 btrfs_cmp_data_free(&cmp);
3178                 if (same_inode) {
3179                         btrfs_wait_ordered_range(src, same_lock_start,
3180                                                  same_lock_len);
3181                 } else {
3182                         btrfs_wait_ordered_range(src, loff, len);
3183                         btrfs_wait_ordered_range(dst, dst_loff, len);
3184                 }
3185                 goto again;
3186         }
3187         ASSERT(ret == 0);
3188         if (WARN_ON(ret)) {
3189                 /* ranges in the io trees already unlocked */
3190                 btrfs_cmp_data_free(&cmp);
3191                 return ret;
3192         }
3193
3194         /* pass original length for comparison so we stay within i_size */
3195         ret = btrfs_cmp_data(olen, &cmp);
3196         if (ret == 0)
3197                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3198
3199         if (same_inode)
3200                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3201                               same_lock_start + same_lock_len - 1);
3202         else
3203                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3204
3205         btrfs_cmp_data_free(&cmp);
3206 out_unlock:
3207         if (same_inode)
3208                 inode_unlock(src);
3209         else
3210                 btrfs_double_inode_unlock(src, dst);
3211
3212         return ret;
3213 }
3214
3215 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3216
3217 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3218                                 struct file *dst_file, u64 dst_loff)
3219 {
3220         struct inode *src = file_inode(src_file);
3221         struct inode *dst = file_inode(dst_file);
3222         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3223         ssize_t res;
3224
3225         if (olen > BTRFS_MAX_DEDUPE_LEN)
3226                 olen = BTRFS_MAX_DEDUPE_LEN;
3227
3228         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3229                 /*
3230                  * Btrfs does not support blocksize < page_size. As a
3231                  * result, btrfs_cmp_data() won't correctly handle
3232                  * this situation without an update.
3233                  */
3234                 return -EINVAL;
3235         }
3236
3237         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3238         if (res)
3239                 return res;
3240         return olen;
3241 }
3242
3243 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3244                                      struct inode *inode,
3245                                      u64 endoff,
3246                                      const u64 destoff,
3247                                      const u64 olen,
3248                                      int no_time_update)
3249 {
3250         struct btrfs_root *root = BTRFS_I(inode)->root;
3251         int ret;
3252
3253         inode_inc_iversion(inode);
3254         if (!no_time_update)
3255                 inode->i_mtime = inode->i_ctime = current_time(inode);
3256         /*
3257          * We round up to the block size at eof when determining which
3258          * extents to clone above, but shouldn't round up the file size.
3259          */