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 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  90 struct btrfs_ioctl_send_args_32 {
  91         __s64 send_fd;                  /* in */
  92         __u64 clone_sources_count;      /* in */
  93         compat_uptr_t clone_sources;    /* in */
  94         __u64 parent_root;              /* in */
  95         __u64 flags;                    /* in */
  96         __u64 reserved[4];              /* in */
  97 } __attribute__ ((__packed__));
  98
  99 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
 100                                struct btrfs_ioctl_send_args_32)
 101 #endif
 102
 103 static int btrfs_clone(struct inode *src, struct inode *inode,
 104                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
 105                        int no_time_update);
 106
 107 /* Mask out flags that are inappropriate for the given type of inode. */
 108 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
 109 {
 110         if (S_ISDIR(mode))
 111                 return flags;
 112         else if (S_ISREG(mode))
 113                 return flags & ~FS_DIRSYNC_FL;
 114         else
 115                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 116 }
 117
 118 /*
 119  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
 120  */
 121 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
 122 {
 123         unsigned int iflags = 0;
 124
 125         if (flags & BTRFS_INODE_SYNC)
 126                 iflags |= FS_SYNC_FL;
 127         if (flags & BTRFS_INODE_IMMUTABLE)
 128                 iflags |= FS_IMMUTABLE_FL;
 129         if (flags & BTRFS_INODE_APPEND)
 130                 iflags |= FS_APPEND_FL;
 131         if (flags & BTRFS_INODE_NODUMP)
 132                 iflags |= FS_NODUMP_FL;
 133         if (flags & BTRFS_INODE_NOATIME)
 134                 iflags |= FS_NOATIME_FL;
 135         if (flags & BTRFS_INODE_DIRSYNC)
 136                 iflags |= FS_DIRSYNC_FL;
 137         if (flags & BTRFS_INODE_NODATACOW)
 138                 iflags |= FS_NOCOW_FL;
 139
 140         if (flags & BTRFS_INODE_NOCOMPRESS)
 141                 iflags |= FS_NOCOMP_FL;
 142         else if (flags & BTRFS_INODE_COMPRESS)
 143                 iflags |= FS_COMPR_FL;
 144
 145         return iflags;
 146 }
 147
 148 /*
 149  * Update inode->i_flags based on the btrfs internal flags.
 150  */
 151 void btrfs_update_iflags(struct inode *inode)
 152 {
 153         struct btrfs_inode *ip = BTRFS_I(inode);
 154         unsigned int new_fl = 0;
 155
 156         if (ip->flags & BTRFS_INODE_SYNC)
 157                 new_fl |= S_SYNC;
 158         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 159                 new_fl |= S_IMMUTABLE;
 160         if (ip->flags & BTRFS_INODE_APPEND)
 161                 new_fl |= S_APPEND;
 162         if (ip->flags & BTRFS_INODE_NOATIME)
 163                 new_fl |= S_NOATIME;
 164         if (ip->flags & BTRFS_INODE_DIRSYNC)
 165                 new_fl |= S_DIRSYNC;
 166
 167         set_mask_bits(&inode->i_flags,
 168                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 169                       new_fl);
 170 }
 171
 172 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 173 {
 174         struct btrfs_inode *ip = BTRFS_I(file_inode(file));
 175         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 176
 177         if (copy_to_user(arg, &flags, sizeof(flags)))
 178                 return -EFAULT;
 179         return 0;
 180 }
 181
 182 static int check_flags(unsigned int flags)
 183 {
 184         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 185                       FS_NOATIME_FL | FS_NODUMP_FL | \
 186                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 187                       FS_NOCOMP_FL | FS_COMPR_FL |
 188                       FS_NOCOW_FL))
 189                 return -EOPNOTSUPP;
 190
 191         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 192                 return -EINVAL;
 193
 194         return 0;
 195 }
 196
 197 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 198 {
 199         struct inode *inode = file_inode(file);
 200         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 201         struct btrfs_inode *ip = BTRFS_I(inode);
 202         struct btrfs_root *root = ip->root;
 203         struct btrfs_trans_handle *trans;
 204         unsigned int flags, oldflags;
 205         int ret;
 206         u64 ip_oldflags;
 207         unsigned int i_oldflags;
 208         umode_t mode;
 209
 210         if (!inode_owner_or_capable(inode))
 211                 return -EPERM;
 212
 213         if (btrfs_root_readonly(root))
 214                 return -EROFS;
 215
 216         if (copy_from_user(&flags, arg, sizeof(flags)))
 217                 return -EFAULT;
 218
 219         ret = check_flags(flags);
 220         if (ret)
 221                 return ret;
 222
 223         ret = mnt_want_write_file(file);
 224         if (ret)
 225                 return ret;
 226
 227         inode_lock(inode);
 228
 229         ip_oldflags = ip->flags;
 230         i_oldflags = inode->i_flags;
 231         mode = inode->i_mode;
 232
 233         flags = btrfs_mask_flags(inode->i_mode, flags);
 234         oldflags = btrfs_flags_to_ioctl(ip->flags);
 235         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 236                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 237                         ret = -EPERM;
 238                         goto out_unlock;
 239                 }
 240         }
 241
 242         if (flags & FS_SYNC_FL)
 243                 ip->flags |= BTRFS_INODE_SYNC;
 244         else
 245                 ip->flags &= ~BTRFS_INODE_SYNC;
 246         if (flags & FS_IMMUTABLE_FL)
 247                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 248         else
 249                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 250         if (flags & FS_APPEND_FL)
 251                 ip->flags |= BTRFS_INODE_APPEND;
 252         else
 253                 ip->flags &= ~BTRFS_INODE_APPEND;
 254         if (flags & FS_NODUMP_FL)
 255                 ip->flags |= BTRFS_INODE_NODUMP;
 256         else
 257                 ip->flags &= ~BTRFS_INODE_NODUMP;
 258         if (flags & FS_NOATIME_FL)
 259                 ip->flags |= BTRFS_INODE_NOATIME;
 260         else
 261                 ip->flags &= ~BTRFS_INODE_NOATIME;
 262         if (flags & FS_DIRSYNC_FL)
 263                 ip->flags |= BTRFS_INODE_DIRSYNC;
 264         else
 265                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 266         if (flags & FS_NOCOW_FL) {
 267                 if (S_ISREG(mode)) {
 268                         /*
 269                          * It's safe to turn csums off here, no extents exist.
 270                          * Otherwise we want the flag to reflect the real COW
 271                          * status of the file and will not set it.
 272                          */
 273                         if (inode->i_size == 0)
 274                                 ip->flags |= BTRFS_INODE_NODATACOW
 275                                            | BTRFS_INODE_NODATASUM;
 276                 } else {
 277                         ip->flags |= BTRFS_INODE_NODATACOW;
 278                 }
 279         } else {
 280                 /*
 281                  * Revert back under same assumptions as above
 282                  */
 283                 if (S_ISREG(mode)) {
 284                         if (inode->i_size == 0)
 285                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
 286                                              | BTRFS_INODE_NODATASUM);
 287                 } else {
 288                         ip->flags &= ~BTRFS_INODE_NODATACOW;
 289                 }
 290         }
 291
 292         /*
 293          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 294          * flag may be changed automatically if compression code won't make
 295          * things smaller.
 296          */
 297         if (flags & FS_NOCOMP_FL) {
 298                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 299                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 300
 301                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 302                 if (ret && ret != -ENODATA)
 303                         goto out_drop;
 304         } else if (flags & FS_COMPR_FL) {
 305                 const char *comp;
 306
 307                 ip->flags |= BTRFS_INODE_COMPRESS;
 308                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 309
 310                 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
 311                         comp = "lzo";
 312                 else if (fs_info->compress_type == BTRFS_COMPRESS_ZLIB)
 313                         comp = "zlib";
 314                 else
 315                         comp = "zstd";
 316                 ret = btrfs_set_prop(inode, "btrfs.compression",
 317                                      comp, strlen(comp), 0);
 318                 if (ret)
 319                         goto out_drop;
 320
 321         } else {
 322                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 323                 if (ret && ret != -ENODATA)
 324                         goto out_drop;
 325                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 326         }
 327
 328         trans = btrfs_start_transaction(root, 1);
 329         if (IS_ERR(trans)) {
 330                 ret = PTR_ERR(trans);
 331                 goto out_drop;
 332         }
 333
 334         btrfs_update_iflags(inode);
 335         inode_inc_iversion(inode);
 336         inode->i_ctime = current_time(inode);
 337         ret = btrfs_update_inode(trans, root, inode);
 338
 339         btrfs_end_transaction(trans);
 340  out_drop:
 341         if (ret) {
 342                 ip->flags = ip_oldflags;
 343                 inode->i_flags = i_oldflags;
 344         }
 345
 346  out_unlock:
 347         inode_unlock(inode);
 348         mnt_drop_write_file(file);
 349         return ret;
 350 }
 351
 352 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 353 {
 354         struct inode *inode = file_inode(file);
 355
 356         return put_user(inode->i_generation, arg);
 357 }
 358
 359 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 360 {
 361         struct inode *inode = file_inode(file);
 362         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 363         struct btrfs_device *device;
 364         struct request_queue *q;
 365         struct fstrim_range range;
 366         u64 minlen = ULLONG_MAX;
 367         u64 num_devices = 0;
 368         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 369         int ret;
 370
 371         if (!capable(CAP_SYS_ADMIN))
 372                 return -EPERM;
 373
 374         rcu_read_lock();
 375         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 376                                 dev_list) {
 377                 if (!device->bdev)
 378                         continue;
 379                 q = bdev_get_queue(device->bdev);
 380                 if (blk_queue_discard(q)) {
 381                         num_devices++;
 382                         minlen = min_t(u64, q->limits.discard_granularity,
 383                                      minlen);
 384                 }
 385         }
 386         rcu_read_unlock();
 387
 388         if (!num_devices)
 389                 return -EOPNOTSUPP;
 390         if (copy_from_user(&range, arg, sizeof(range)))
 391                 return -EFAULT;
 392         if (range.start > total_bytes ||
 393             range.len < fs_info->sb->s_blocksize)
 394                 return -EINVAL;
 395
 396         range.len = min(range.len, total_bytes - range.start);
 397         range.minlen = max(range.minlen, minlen);
 398         ret = btrfs_trim_fs(fs_info, &range);
 399         if (ret < 0)
 400                 return ret;
 401
 402         if (copy_to_user(arg, &range, sizeof(range)))
 403                 return -EFAULT;
 404
 405         return 0;
 406 }
 407
 408 int btrfs_is_empty_uuid(u8 *uuid)
 409 {
 410         int i;
 411
 412         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 413                 if (uuid[i])
 414                         return 0;
 415         }
 416         return 1;
 417 }
 418
 419 static noinline int create_subvol(struct inode *dir,
 420                                   struct dentry *dentry,
 421                                   const char *name, int namelen,
 422                                   u64 *async_transid,
 423                                   struct btrfs_qgroup_inherit *inherit)
 424 {
 425         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 426         struct btrfs_trans_handle *trans;
 427         struct btrfs_key key;
 428         struct btrfs_root_item *root_item;
 429         struct btrfs_inode_item *inode_item;
 430         struct extent_buffer *leaf;
 431         struct btrfs_root *root = BTRFS_I(dir)->root;
 432         struct btrfs_root *new_root;
 433         struct btrfs_block_rsv block_rsv;
 434         struct timespec cur_time = current_time(dir);
 435         struct inode *inode;
 436         int ret;
 437         int err;
 438         u64 objectid;
 439         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 440         u64 index = 0;
 441         u64 qgroup_reserved;
 442         uuid_le new_uuid;
 443
 444         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 445         if (!root_item)
 446                 return -ENOMEM;
 447
 448         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 449         if (ret)
 450                 goto fail_free;
 451
 452         /*
 453          * Don't create subvolume whose level is not zero. Or qgroup will be
 454          * screwed up since it assumes subvolume qgroup's level to be 0.
 455          */
 456         if (btrfs_qgroup_level(objectid)) {
 457                 ret = -ENOSPC;
 458                 goto fail_free;
 459         }
 460
 461         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 462         /*
 463          * The same as the snapshot creation, please see the comment
 464          * of create_snapshot().
 465          */
 466         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 467                                                8, &qgroup_reserved, false);
 468         if (ret)
 469                 goto fail_free;
 470
 471         trans = btrfs_start_transaction(root, 0);
 472         if (IS_ERR(trans)) {
 473                 ret = PTR_ERR(trans);
 474                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 475                 goto fail_free;
 476         }
 477         trans->block_rsv = &block_rsv;
 478         trans->bytes_reserved = block_rsv.size;
 479
 480         ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit);
 481         if (ret)
 482                 goto fail;
 483
 484         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 485         if (IS_ERR(leaf)) {
 486                 ret = PTR_ERR(leaf);
 487                 goto fail;
 488         }
 489
 490         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
 491         btrfs_set_header_bytenr(leaf, leaf->start);
 492         btrfs_set_header_generation(leaf, trans->transid);
 493         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 494         btrfs_set_header_owner(leaf, objectid);
 495
 496         write_extent_buffer_fsid(leaf, fs_info->fsid);
 497         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
 498         btrfs_mark_buffer_dirty(leaf);
 499
 500         inode_item = &root_item->inode;
 501         btrfs_set_stack_inode_generation(inode_item, 1);
 502         btrfs_set_stack_inode_size(inode_item, 3);
 503         btrfs_set_stack_inode_nlink(inode_item, 1);
 504         btrfs_set_stack_inode_nbytes(inode_item,
 505                                      fs_info->nodesize);
 506         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 507
 508         btrfs_set_root_flags(root_item, 0);
 509         btrfs_set_root_limit(root_item, 0);
 510         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 511
 512         btrfs_set_root_bytenr(root_item, leaf->start);
 513         btrfs_set_root_generation(root_item, trans->transid);
 514         btrfs_set_root_level(root_item, 0);
 515         btrfs_set_root_refs(root_item, 1);
 516         btrfs_set_root_used(root_item, leaf->len);
 517         btrfs_set_root_last_snapshot(root_item, 0);
 518
 519         btrfs_set_root_generation_v2(root_item,
 520                         btrfs_root_generation(root_item));
 521         uuid_le_gen(&new_uuid);
 522         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 523         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 524         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 525         root_item->ctime = root_item->otime;
 526         btrfs_set_root_ctransid(root_item, trans->transid);
 527         btrfs_set_root_otransid(root_item, trans->transid);
 528
 529         btrfs_tree_unlock(leaf);
 530         free_extent_buffer(leaf);
 531         leaf = NULL;
 532
 533         btrfs_set_root_dirid(root_item, new_dirid);
 534
 535         key.objectid = objectid;
 536         key.offset = 0;
 537         key.type = BTRFS_ROOT_ITEM_KEY;
 538         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 539                                 root_item);
 540         if (ret)
 541                 goto fail;
 542
 543         key.offset = (u64)-1;
 544         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 545         if (IS_ERR(new_root)) {
 546                 ret = PTR_ERR(new_root);
 547                 btrfs_abort_transaction(trans, ret);
 548                 goto fail;
 549         }
 550
 551         btrfs_record_root_in_trans(trans, new_root);
 552
 553         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 554         if (ret) {
 555                 /* We potentially lose an unused inode item here */
 556                 btrfs_abort_transaction(trans, ret);
 557                 goto fail;
 558         }
 559
 560         mutex_lock(&new_root->objectid_mutex);
 561         new_root->highest_objectid = new_dirid;
 562         mutex_unlock(&new_root->objectid_mutex);
 563
 564         /*
 565          * insert the directory item
 566          */
 567         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 568         if (ret) {
 569                 btrfs_abort_transaction(trans, ret);
 570                 goto fail;
 571         }
 572
 573         ret = btrfs_insert_dir_item(trans, root,
 574                                     name, namelen, BTRFS_I(dir), &key,
 575                                     BTRFS_FT_DIR, index);
 576         if (ret) {
 577                 btrfs_abort_transaction(trans, ret);
 578                 goto fail;
 579         }
 580
 581         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 582         ret = btrfs_update_inode(trans, root, dir);
 583         BUG_ON(ret);
 584
 585         ret = btrfs_add_root_ref(trans, fs_info,
 586                                  objectid, root->root_key.objectid,
 587                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 588         BUG_ON(ret);
 589
 590         ret = btrfs_uuid_tree_add(trans, fs_info, root_item->uuid,
 591                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 592         if (ret)
 593                 btrfs_abort_transaction(trans, ret);
 594
 595 fail:
 596         kfree(root_item);
 597         trans->block_rsv = NULL;
 598         trans->bytes_reserved = 0;
 599         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 600
 601         if (async_transid) {
 602                 *async_transid = trans->transid;
 603                 err = btrfs_commit_transaction_async(trans, 1);
 604                 if (err)
 605                         err = btrfs_commit_transaction(trans);
 606         } else {
 607                 err = btrfs_commit_transaction(trans);
 608         }
 609         if (err && !ret)
 610                 ret = err;
 611
 612         if (!ret) {
 613                 inode = btrfs_lookup_dentry(dir, dentry);
 614                 if (IS_ERR(inode))
 615                         return PTR_ERR(inode);
 616                 d_instantiate(dentry, inode);
 617         }
 618         return ret;
 619
 620 fail_free:
 621         kfree(root_item);
 622         return ret;
 623 }
 624
 625 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 626                            struct dentry *dentry,
 627                            u64 *async_transid, bool readonly,
 628                            struct btrfs_qgroup_inherit *inherit)
 629 {
 630         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 631         struct inode *inode;
 632         struct btrfs_pending_snapshot *pending_snapshot;
 633         struct btrfs_trans_handle *trans;
 634         int ret;
 635
 636         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 637                 return -EINVAL;
 638
 639         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 640         if (!pending_snapshot)
 641                 return -ENOMEM;
 642
 643         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 644                         GFP_KERNEL);
 645         pending_snapshot->path = btrfs_alloc_path();
 646         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 647                 ret = -ENOMEM;
 648                 goto free_pending;
 649         }
 650
 651         atomic_inc(&root->will_be_snapshotted);
 652         smp_mb__after_atomic();
 653         /* wait for no snapshot writes */
 654         wait_event(root->subv_writers->wait,
 655                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 656
 657         ret = btrfs_start_delalloc_inodes(root, 0);
 658         if (ret)
 659                 goto dec_and_free;
 660
 661         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 662
 663         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 664                              BTRFS_BLOCK_RSV_TEMP);
 665         /*
 666          * 1 - parent dir inode
 667          * 2 - dir entries
 668          * 1 - root item
 669          * 2 - root ref/backref
 670          * 1 - root of snapshot
 671          * 1 - UUID item
 672          */
 673         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 674                                         &pending_snapshot->block_rsv, 8,
 675                                         &pending_snapshot->qgroup_reserved,
 676                                         false);
 677         if (ret)
 678                 goto dec_and_free;
 679
 680         pending_snapshot->dentry = dentry;
 681         pending_snapshot->root = root;
 682         pending_snapshot->readonly = readonly;
 683         pending_snapshot->dir = dir;
 684         pending_snapshot->inherit = inherit;
 685
 686         trans = btrfs_start_transaction(root, 0);
 687         if (IS_ERR(trans)) {
 688                 ret = PTR_ERR(trans);
 689                 goto fail;
 690         }
 691
 692         spin_lock(&fs_info->trans_lock);
 693         list_add(&pending_snapshot->list,
 694                  &trans->transaction->pending_snapshots);
 695         spin_unlock(&fs_info->trans_lock);
 696         if (async_transid) {
 697                 *async_transid = trans->transid;
 698                 ret = btrfs_commit_transaction_async(trans, 1);
 699                 if (ret)
 700                         ret = btrfs_commit_transaction(trans);
 701         } else {
 702                 ret = btrfs_commit_transaction(trans);
 703         }
 704         if (ret)
 705                 goto fail;
 706
 707         ret = pending_snapshot->error;
 708         if (ret)
 709                 goto fail;
 710
 711         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 712         if (ret)
 713                 goto fail;
 714
 715         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 716         if (IS_ERR(inode)) {
 717                 ret = PTR_ERR(inode);
 718                 goto fail;
 719         }
 720
 721         d_instantiate(dentry, inode);
 722         ret = 0;
 723 fail:
 724         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 725 dec_and_free:
 726         if (atomic_dec_and_test(&root->will_be_snapshotted))
 727                 wake_up_atomic_t(&root->will_be_snapshotted);
 728 free_pending:
 729         kfree(pending_snapshot->root_item);
 730         btrfs_free_path(pending_snapshot->path);
 731         kfree(pending_snapshot);
 732
 733         return ret;
 734 }
 735
 736 /*  copy of may_delete in fs/namei.c()
 737  *      Check whether we can remove a link victim from directory dir, check
 738  *  whether the type of victim is right.
 739  *  1. We can't do it if dir is read-only (done in permission())
 740  *  2. We should have write and exec permissions on dir
 741  *  3. We can't remove anything from append-only dir
 742  *  4. We can't do anything with immutable dir (done in permission())
 743  *  5. If the sticky bit on dir is set we should either
 744  *      a. be owner of dir, or
 745  *      b. be owner of victim, or
 746  *      c. have CAP_FOWNER capability
 747  *  6. If the victim is append-only or immutable we can't do anything with
 748  *     links pointing to it.
 749  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 750  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 751  *  9. We can't remove a root or mountpoint.
 752  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 753  *     nfs_async_unlink().
 754  */
 755
 756 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 757 {
 758         int error;
 759
 760         if (d_really_is_negative(victim))
 761                 return -ENOENT;
 762
 763         BUG_ON(d_inode(victim->d_parent) != dir);
 764         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 765
 766         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 767         if (error)
 768                 return error;
 769         if (IS_APPEND(dir))
 770                 return -EPERM;
 771         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 772             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 773                 return -EPERM;
 774         if (isdir) {
 775                 if (!d_is_dir(victim))
 776                         return -ENOTDIR;
 777                 if (IS_ROOT(victim))
 778                         return -EBUSY;
 779         } else if (d_is_dir(victim))
 780                 return -EISDIR;
 781         if (IS_DEADDIR(dir))
 782                 return -ENOENT;
 783         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 784                 return -EBUSY;
 785         return 0;
 786 }
 787
 788 /* copy of may_create in fs/namei.c() */
 789 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 790 {
 791         if (d_really_is_positive(child))
 792                 return -EEXIST;
 793         if (IS_DEADDIR(dir))
 794                 return -ENOENT;
 795         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 796 }
 797
 798 /*
 799  * Create a new subvolume below @parent.  This is largely modeled after
 800  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 801  * inside this filesystem so it's quite a bit simpler.
 802  */
 803 static noinline int btrfs_mksubvol(const struct path *parent,
 804                                    const char *name, int namelen,
 805                                    struct btrfs_root *snap_src,
 806                                    u64 *async_transid, bool readonly,
 807                                    struct btrfs_qgroup_inherit *inherit)
 808 {
 809         struct inode *dir = d_inode(parent->dentry);
 810         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 811         struct dentry *dentry;
 812         int error;
 813
 814         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 815         if (error == -EINTR)
 816                 return error;
 817
 818         dentry = lookup_one_len(name, parent->dentry, namelen);
 819         error = PTR_ERR(dentry);
 820         if (IS_ERR(dentry))
 821                 goto out_unlock;
 822
 823         error = btrfs_may_create(dir, dentry);
 824         if (error)
 825                 goto out_dput;
 826
 827         /*
 828          * even if this name doesn't exist, we may get hash collisions.
 829          * check for them now when we can safely fail
 830          */
 831         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 832                                                dir->i_ino, name,
 833                                                namelen);
 834         if (error)
 835                 goto out_dput;
 836
 837         down_read(&fs_info->subvol_sem);
 838
 839         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 840                 goto out_up_read;
 841
 842         if (snap_src) {
 843                 error = create_snapshot(snap_src, dir, dentry,
 844                                         async_transid, readonly, inherit);
 845         } else {
 846                 error = create_subvol(dir, dentry, name, namelen,
 847                                       async_transid, inherit);
 848         }
 849         if (!error)
 850                 fsnotify_mkdir(dir, dentry);
 851 out_up_read:
 852         up_read(&fs_info->subvol_sem);
 853 out_dput:
 854         dput(dentry);
 855 out_unlock:
 856         inode_unlock(dir);
 857         return error;
 858 }
 859
 860 /*
 861  * When we're defragging a range, we don't want to kick it off again
 862  * if it is really just waiting for delalloc to send it down.
 863  * If we find a nice big extent or delalloc range for the bytes in the
 864  * file you want to defrag, we return 0 to let you know to skip this
 865  * part of the file
 866  */
 867 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 868 {
 869         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 870         struct extent_map *em = NULL;
 871         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 872         u64 end;
 873
 874         read_lock(&em_tree->lock);
 875         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 876         read_unlock(&em_tree->lock);
 877
 878         if (em) {
 879                 end = extent_map_end(em);
 880                 free_extent_map(em);
 881                 if (end - offset > thresh)
 882                         return 0;
 883         }
 884         /* if we already have a nice delalloc here, just stop */
 885         thresh /= 2;
 886         end = count_range_bits(io_tree, &offset, offset + thresh,
 887                                thresh, EXTENT_DELALLOC, 1);
 888         if (end >= thresh)
 889                 return 0;
 890         return 1;
 891 }
 892
 893 /*
 894  * helper function to walk through a file and find extents
 895  * newer than a specific transid, and smaller than thresh.
 896  *
 897  * This is used by the defragging code to find new and small
 898  * extents
 899  */
 900 static int find_new_extents(struct btrfs_root *root,
 901                             struct inode *inode, u64 newer_than,
 902                             u64 *off, u32 thresh)
 903 {
 904         struct btrfs_path *path;
 905         struct btrfs_key min_key;
 906         struct extent_buffer *leaf;
 907         struct btrfs_file_extent_item *extent;
 908         int type;
 909         int ret;
 910         u64 ino = btrfs_ino(BTRFS_I(inode));
 911
 912         path = btrfs_alloc_path();
 913         if (!path)
 914                 return -ENOMEM;
 915
 916         min_key.objectid = ino;
 917         min_key.type = BTRFS_EXTENT_DATA_KEY;
 918         min_key.offset = *off;
 919
 920         while (1) {
 921                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
 922                 if (ret != 0)
 923                         goto none;
 924 process_slot:
 925                 if (min_key.objectid != ino)
 926                         goto none;
 927                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 928                         goto none;
 929
 930                 leaf = path->nodes[0];
 931                 extent = btrfs_item_ptr(leaf, path->slots[0],
 932                                         struct btrfs_file_extent_item);
 933
 934                 type = btrfs_file_extent_type(leaf, extent);
 935                 if (type == BTRFS_FILE_EXTENT_REG &&
 936                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 937                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 938                         *off = min_key.offset;
 939                         btrfs_free_path(path);
 940                         return 0;
 941                 }
 942
 943                 path->slots[0]++;
 944                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
 945                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
 946                         goto process_slot;
 947                 }
 948
 949                 if (min_key.offset == (u64)-1)
 950                         goto none;
 951
 952                 min_key.offset++;
 953                 btrfs_release_path(path);
 954         }
 955 none:
 956         btrfs_free_path(path);
 957         return -ENOENT;
 958 }
 959
 960 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 961 {
 962         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 963         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 964         struct extent_map *em;
 965         u64 len = PAGE_SIZE;
 966
 967         /*
 968          * hopefully we have this extent in the tree already, try without
 969          * the full extent lock
 970          */
 971         read_lock(&em_tree->lock);
 972         em = lookup_extent_mapping(em_tree, start, len);
 973         read_unlock(&em_tree->lock);
 974
 975         if (!em) {
 976                 struct extent_state *cached = NULL;
 977                 u64 end = start + len - 1;
 978
 979                 /* get the big lock and read metadata off disk */
 980                 lock_extent_bits(io_tree, start, end, &cached);
 981                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
 982                 unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS);
 983
 984                 if (IS_ERR(em))
 985                         return NULL;
 986         }
 987
 988         return em;
 989 }
 990
 991 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 992 {
 993         struct extent_map *next;
 994         bool ret = true;
 995
 996         /* this is the last extent */
 997         if (em->start + em->len >= i_size_read(inode))
 998                 return false;
 999
1000         next = defrag_lookup_extent(inode, em->start + em->len);
1001         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1002                 ret = false;
1003         else if ((em->block_start + em->block_len == next->block_start) &&
1004                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1005                 ret = false;
1006
1007         free_extent_map(next);
1008         return ret;
1009 }
1010
1011 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1012                                u64 *last_len, u64 *skip, u64 *defrag_end,
1013                                int compress)
1014 {
1015         struct extent_map *em;
1016         int ret = 1;
1017         bool next_mergeable = true;
1018         bool prev_mergeable = true;
1019
1020         /*
1021          * make sure that once we start defragging an extent, we keep on
1022          * defragging it
1023          */
1024         if (start < *defrag_end)
1025                 return 1;
1026
1027         *skip = 0;
1028
1029         em = defrag_lookup_extent(inode, start);
1030         if (!em)
1031                 return 0;
1032
1033         /* this will cover holes, and inline extents */
1034         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1035                 ret = 0;
1036                 goto out;
1037         }
1038
1039         if (!*defrag_end)
1040                 prev_mergeable = false;
1041
1042         next_mergeable = defrag_check_next_extent(inode, em);
1043         /*
1044          * we hit a real extent, if it is big or the next extent is not a
1045          * real extent, don't bother defragging it
1046          */
1047         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1048             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1049                 ret = 0;
1050 out:
1051         /*
1052          * last_len ends up being a counter of how many bytes we've defragged.
1053          * every time we choose not to defrag an extent, we reset *last_len
1054          * so that the next tiny extent will force a defrag.
1055          *
1056          * The end result of this is that tiny extents before a single big
1057          * extent will force at least part of that big extent to be defragged.
1058          */
1059         if (ret) {
1060                 *defrag_end = extent_map_end(em);
1061         } else {
1062                 *last_len = 0;
1063                 *skip = extent_map_end(em);
1064                 *defrag_end = 0;
1065         }
1066
1067         free_extent_map(em);
1068         return ret;
1069 }
1070
1071 /*
1072  * it doesn't do much good to defrag one or two pages
1073  * at a time.  This pulls in a nice chunk of pages
1074  * to COW and defrag.
1075  *
1076  * It also makes sure the delalloc code has enough
1077  * dirty data to avoid making new small extents as part
1078  * of the defrag
1079  *
1080  * It's a good idea to start RA on this range
1081  * before calling this.
1082  */
1083 static int cluster_pages_for_defrag(struct inode *inode,
1084                                     struct page **pages,
1085                                     unsigned long start_index,
1086                                     unsigned long num_pages)
1087 {
1088         unsigned long file_end;
1089         u64 isize = i_size_read(inode);
1090         u64 page_start;
1091         u64 page_end;
1092         u64 page_cnt;
1093         int ret;
1094         int i;
1095         int i_done;
1096         struct btrfs_ordered_extent *ordered;
1097         struct extent_state *cached_state = NULL;
1098         struct extent_io_tree *tree;
1099         struct extent_changeset *data_reserved = NULL;
1100         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1101
1102         file_end = (isize - 1) >> PAGE_SHIFT;
1103         if (!isize || start_index > file_end)
1104                 return 0;
1105
1106         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1107
1108         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1109                         start_index << PAGE_SHIFT,
1110                         page_cnt << PAGE_SHIFT);
1111         if (ret)
1112                 return ret;
1113         i_done = 0;
1114         tree = &BTRFS_I(inode)->io_tree;
1115
1116         /* step one, lock all the pages */
1117         for (i = 0; i < page_cnt; i++) {
1118                 struct page *page;
1119 again:
1120                 page = find_or_create_page(inode->i_mapping,
1121                                            start_index + i, mask);
1122                 if (!page)
1123                         break;
1124
1125                 page_start = page_offset(page);
1126                 page_end = page_start + PAGE_SIZE - 1;
1127                 while (1) {
1128                         lock_extent_bits(tree, page_start, page_end,
1129                                          &cached_state);
1130                         ordered = btrfs_lookup_ordered_extent(inode,
1131                                                               page_start);
1132                         unlock_extent_cached(tree, page_start, page_end,
1133                                              &cached_state, GFP_NOFS);
1134                         if (!ordered)
1135                                 break;
1136
1137                         unlock_page(page);
1138                         btrfs_start_ordered_extent(inode, ordered, 1);
1139                         btrfs_put_ordered_extent(ordered);
1140                         lock_page(page);
1141                         /*
1142                          * we unlocked the page above, so we need check if
1143                          * it was released or not.
1144                          */
1145                         if (page->mapping != inode->i_mapping) {
1146                                 unlock_page(page);
1147                                 put_page(page);
1148                                 goto again;
1149                         }
1150                 }
1151
1152                 if (!PageUptodate(page)) {
1153                         btrfs_readpage(NULL, page);
1154                         lock_page(page);
1155                         if (!PageUptodate(page)) {
1156                                 unlock_page(page);
1157                                 put_page(page);
1158                                 ret = -EIO;
1159                                 break;
1160                         }
1161                 }
1162
1163                 if (page->mapping != inode->i_mapping) {
1164                         unlock_page(page);
1165                         put_page(page);
1166                         goto again;
1167                 }
1168
1169                 pages[i] = page;
1170                 i_done++;
1171         }
1172         if (!i_done || ret)
1173                 goto out;
1174
1175         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1176                 goto out;
1177
1178         /*
1179          * so now we have a nice long stream of locked
1180          * and up to date pages, lets wait on them
1181          */
1182         for (i = 0; i < i_done; i++)
1183                 wait_on_page_writeback(pages[i]);
1184
1185         page_start = page_offset(pages[0]);
1186         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1187
1188         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1189                          page_start, page_end - 1, &cached_state);
1190         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1191                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1192                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1193                           &cached_state);
1194
1195         if (i_done != page_cnt) {
1196                 spin_lock(&BTRFS_I(inode)->lock);
1197                 BTRFS_I(inode)->outstanding_extents++;
1198                 spin_unlock(&BTRFS_I(inode)->lock);
1199                 btrfs_delalloc_release_space(inode, data_reserved,
1200                                 start_index << PAGE_SHIFT,
1201                                 (page_cnt - i_done) << PAGE_SHIFT);
1202         }
1203
1204
1205         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1206                           &cached_state);
1207
1208         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1209                              page_start, page_end - 1, &cached_state,
1210                              GFP_NOFS);
1211
1212         for (i = 0; i < i_done; i++) {
1213                 clear_page_dirty_for_io(pages[i]);
1214                 ClearPageChecked(pages[i]);
1215                 set_page_extent_mapped(pages[i]);
1216                 set_page_dirty(pages[i]);
1217                 unlock_page(pages[i]);
1218                 put_page(pages[i]);
1219         }
1220         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1221         extent_changeset_free(data_reserved);
1222         return i_done;
1223 out:
1224         for (i = 0; i < i_done; i++) {
1225                 unlock_page(pages[i]);
1226                 put_page(pages[i]);
1227         }
1228         btrfs_delalloc_release_space(inode, data_reserved,
1229                         start_index << PAGE_SHIFT,
1230                         page_cnt << PAGE_SHIFT);
1231         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1232         extent_changeset_free(data_reserved);
1233         return ret;
1234
1235 }
1236
1237 int btrfs_defrag_file(struct inode *inode, struct file *file,
1238                       struct btrfs_ioctl_defrag_range_args *range,
1239                       u64 newer_than, unsigned long max_to_defrag)
1240 {
1241         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1242         struct btrfs_root *root = BTRFS_I(inode)->root;
1243         struct file_ra_state *ra = NULL;
1244         unsigned long last_index;
1245         u64 isize = i_size_read(inode);
1246         u64 last_len = 0;
1247         u64 skip = 0;
1248         u64 defrag_end = 0;
1249         u64 newer_off = range->start;
1250         unsigned long i;
1251         unsigned long ra_index = 0;
1252         int ret;
1253         int defrag_count = 0;
1254         int compress_type = BTRFS_COMPRESS_ZLIB;
1255         u32 extent_thresh = range->extent_thresh;
1256         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1257         unsigned long cluster = max_cluster;
1258         u64 new_align = ~((u64)SZ_128K - 1);
1259         struct page **pages = NULL;
1260         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1261
1262         if (isize == 0)
1263                 return 0;
1264
1265         if (range->start >= isize)
1266                 return -EINVAL;
1267
1268         if (do_compress) {
1269                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1270                         return -EINVAL;
1271                 if (range->compress_type)
1272                         compress_type = range->compress_type;
1273         }
1274
1275         if (extent_thresh == 0)
1276                 extent_thresh = SZ_256K;
1277
1278         /*
1279          * If we were not given a file, allocate a readahead context. As
1280          * readahead is just an optimization, defrag will work without it so
1281          * we don't error out.
1282          */
1283         if (!file) {
1284                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1285                 if (ra)
1286                         file_ra_state_init(ra, inode->i_mapping);
1287         } else {
1288                 ra = &file->f_ra;
1289         }
1290
1291         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1292         if (!pages) {
1293                 ret = -ENOMEM;
1294                 goto out_ra;
1295         }
1296
1297         /* find the last page to defrag */
1298         if (range->start + range->len > range->start) {
1299                 last_index = min_t(u64, isize - 1,
1300                          range->start + range->len - 1) >> PAGE_SHIFT;
1301         } else {
1302                 last_index = (isize - 1) >> PAGE_SHIFT;
1303         }
1304
1305         if (newer_than) {
1306                 ret = find_new_extents(root, inode, newer_than,
1307                                        &newer_off, SZ_64K);
1308                 if (!ret) {
1309                         range->start = newer_off;
1310                         /*
1311                          * we always align our defrag to help keep
1312                          * the extents in the file evenly spaced
1313                          */
1314                         i = (newer_off & new_align) >> PAGE_SHIFT;
1315                 } else
1316                         goto out_ra;
1317         } else {
1318                 i = range->start >> PAGE_SHIFT;
1319         }
1320         if (!max_to_defrag)
1321                 max_to_defrag = last_index - i + 1;
1322
1323         /*
1324          * make writeback starts from i, so the defrag range can be
1325          * written sequentially.
1326          */
1327         if (i < inode->i_mapping->writeback_index)
1328                 inode->i_mapping->writeback_index = i;
1329
1330         while (i <= last_index && defrag_count < max_to_defrag &&
1331                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1332                 /*
1333                  * make sure we stop running if someone unmounts
1334                  * the FS
1335                  */
1336                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1337                         break;
1338
1339                 if (btrfs_defrag_cancelled(fs_info)) {
1340                         btrfs_debug(fs_info, "defrag_file cancelled");
1341                         ret = -EAGAIN;
1342                         break;
1343                 }
1344
1345                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1346                                          extent_thresh, &last_len, &skip,
1347                                          &defrag_end, do_compress)){
1348                         unsigned long next;
1349                         /*
1350                          * the should_defrag function tells us how much to skip
1351                          * bump our counter by the suggested amount
1352                          */
1353                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1354                         i = max(i + 1, next);
1355                         continue;
1356                 }
1357
1358                 if (!newer_than) {
1359                         cluster = (PAGE_ALIGN(defrag_end) >>
1360                                    PAGE_SHIFT) - i;
1361                         cluster = min(cluster, max_cluster);
1362                 } else {
1363                         cluster = max_cluster;
1364                 }
1365
1366                 if (i + cluster > ra_index) {
1367                         ra_index = max(i, ra_index);
1368                         if (ra)
1369                                 page_cache_sync_readahead(inode->i_mapping, ra,
1370                                                 file, ra_index, cluster);
1371                         ra_index += cluster;
1372                 }
1373
1374                 inode_lock(inode);
1375                 if (do_compress)
1376                         BTRFS_I(inode)->defrag_compress = compress_type;
1377                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1378                 if (ret < 0) {
1379                         inode_unlock(inode);
1380                         goto out_ra;
1381                 }
1382
1383                 defrag_count += ret;
1384                 balance_dirty_pages_ratelimited(inode->i_mapping);
1385                 inode_unlock(inode);
1386
1387                 if (newer_than) {
1388                         if (newer_off == (u64)-1)
1389                                 break;
1390
1391                         if (ret > 0)
1392                                 i += ret;
1393
1394                         newer_off = max(newer_off + 1,
1395                                         (u64)i << PAGE_SHIFT);
1396
1397                         ret = find_new_extents(root, inode, newer_than,
1398                                                &newer_off, SZ_64K);
1399                         if (!ret) {
1400                                 range->start = newer_off;
1401                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1402                         } else {
1403                                 break;
1404                         }
1405                 } else {
1406                         if (ret > 0) {
1407                                 i += ret;
1408                                 last_len += ret << PAGE_SHIFT;
1409                         } else {
1410                                 i++;
1411                                 last_len = 0;
1412                         }
1413                 }
1414         }
1415
1416         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1417                 filemap_flush(inode->i_mapping);
1418                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1419                              &BTRFS_I(inode)->runtime_flags))
1420                         filemap_flush(inode->i_mapping);
1421         }
1422
1423         if (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 (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
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 - 1];
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         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2679                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
2680         } else {
2681                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2682                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2683         }
2684         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2685
2686         if (!ret) {
2687                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2688                         btrfs_info(fs_info, "device deleted: id %llu",
2689                                         vol_args->devid);
2690                 else
2691                         btrfs_info(fs_info, "device deleted: %s",
2692                                         vol_args->name);
2693         }
2694 out:
2695         kfree(vol_args);
2696 err_drop:
2697         mnt_drop_write_file(file);
2698         return ret;
2699 }
2700
2701 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2702 {
2703         struct inode *inode = file_inode(file);
2704         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2705         struct btrfs_ioctl_vol_args *vol_args;
2706         int ret;
2707
2708         if (!capable(CAP_SYS_ADMIN))
2709                 return -EPERM;
2710
2711         ret = mnt_want_write_file(file);
2712         if (ret)
2713                 return ret;
2714
2715         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
2716                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2717                 goto out_drop_write;
2718         }
2719
2720         vol_args = memdup_user(arg, sizeof(*vol_args));
2721         if (IS_ERR(vol_args)) {
2722                 ret = PTR_ERR(vol_args);
2723                 goto out;
2724         }
2725
2726         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2727         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
2728
2729         if (!ret)
2730                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2731         kfree(vol_args);
2732 out:
2733         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
2734 out_drop_write:
2735         mnt_drop_write_file(file);
2736
2737         return ret;
2738 }
2739
2740 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2741                                 void __user *arg)
2742 {
2743         struct btrfs_ioctl_fs_info_args *fi_args;
2744         struct btrfs_device *device;
2745         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2746         int ret = 0;
2747
2748         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2749         if (!fi_args)
2750                 return -ENOMEM;
2751
2752         rcu_read_lock();
2753         fi_args->num_devices = fs_devices->num_devices;
2754
2755         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2756                 if (device->devid > fi_args->max_id)
2757                         fi_args->max_id = device->devid;
2758         }
2759         rcu_read_unlock();
2760
2761         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
2762         fi_args->nodesize = fs_info->nodesize;
2763         fi_args->sectorsize = fs_info->sectorsize;
2764         fi_args->clone_alignment = fs_info->sectorsize;
2765
2766         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2767                 ret = -EFAULT;
2768
2769         kfree(fi_args);
2770         return ret;
2771 }
2772
2773 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2774                                  void __user *arg)
2775 {
2776         struct btrfs_ioctl_dev_info_args *di_args;
2777         struct btrfs_device *dev;
2778         int ret = 0;
2779         char *s_uuid = NULL;
2780
2781         di_args = memdup_user(arg, sizeof(*di_args));
2782         if (IS_ERR(di_args))
2783                 return PTR_ERR(di_args);
2784
2785         if (!btrfs_is_empty_uuid(di_args->uuid))
2786                 s_uuid = di_args->uuid;
2787
2788         rcu_read_lock();
2789         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
2790
2791         if (!dev) {
2792                 ret = -ENODEV;
2793                 goto out;
2794         }
2795
2796         di_args->devid = dev->devid;
2797         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2798         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2799         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2800         if (dev->name) {
2801                 struct rcu_string *name;
2802
2803                 name = rcu_dereference(dev->name);
2804                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2805                 di_args->path[sizeof(di_args->path) - 1] = 0;
2806         } else {
2807                 di_args->path[0] = '\0';
2808         }
2809
2810 out:
2811         rcu_read_unlock();
2812         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2813                 ret = -EFAULT;
2814
2815         kfree(di_args);
2816         return ret;
2817 }
2818
2819 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
2820 {
2821         struct page *page;
2822
2823         page = grab_cache_page(inode->i_mapping, index);
2824         if (!page)
2825                 return ERR_PTR(-ENOMEM);
2826
2827         if (!PageUptodate(page)) {
2828                 int ret;
2829
2830                 ret = btrfs_readpage(NULL, page);
2831                 if (ret)
2832                         return ERR_PTR(ret);
2833                 lock_page(page);
2834                 if (!PageUptodate(page)) {
2835                         unlock_page(page);
2836                         put_page(page);
2837                         return ERR_PTR(-EIO);
2838                 }
2839                 if (page->mapping != inode->i_mapping) {
2840                         unlock_page(page);
2841                         put_page(page);
2842                         return ERR_PTR(-EAGAIN);
2843                 }
2844         }
2845
2846         return page;
2847 }
2848
2849 static int gather_extent_pages(struct inode *inode, struct page **pages,
2850                                int num_pages, u64 off)
2851 {
2852         int i;
2853         pgoff_t index = off >> PAGE_SHIFT;
2854
2855         for (i = 0; i < num_pages; i++) {
2856 again:
2857                 pages[i] = extent_same_get_page(inode, index + i);
2858                 if (IS_ERR(pages[i])) {
2859                         int err = PTR_ERR(pages[i]);
2860
2861                         if (err == -EAGAIN)
2862                                 goto again;
2863                         pages[i] = NULL;
2864                         return err;
2865                 }
2866         }
2867         return 0;
2868 }
2869
2870 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
2871                              bool retry_range_locking)
2872 {
2873         /*
2874          * Do any pending delalloc/csum calculations on inode, one way or
2875          * another, and lock file content.
2876          * The locking order is:
2877          *
2878          *   1) pages
2879          *   2) range in the inode's io tree
2880          */
2881         while (1) {
2882                 struct btrfs_ordered_extent *ordered;
2883                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2884                 ordered = btrfs_lookup_first_ordered_extent(inode,
2885                                                             off + len - 1);
2886                 if ((!ordered ||
2887                      ordered->file_offset + ordered->len <= off ||
2888                      ordered->file_offset >= off + len) &&
2889                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
2890                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
2891                         if (ordered)
2892                                 btrfs_put_ordered_extent(ordered);
2893                         break;
2894                 }
2895                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
2896                 if (ordered)
2897                         btrfs_put_ordered_extent(ordered);
2898                 if (!retry_range_locking)
2899                         return -EAGAIN;
2900                 btrfs_wait_ordered_range(inode, off, len);
2901         }
2902         return 0;
2903 }
2904
2905 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
2906 {
2907         inode_unlock(inode1);
2908         inode_unlock(inode2);
2909 }
2910
2911 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
2912 {
2913         if (inode1 < inode2)
2914                 swap(inode1, inode2);
2915
2916         inode_lock_nested(inode1, I_MUTEX_PARENT);
2917         inode_lock_nested(inode2, I_MUTEX_CHILD);
2918 }
2919
2920 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
2921                                       struct inode *inode2, u64 loff2, u64 len)
2922 {
2923         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
2924         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
2925 }
2926
2927 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
2928                                     struct inode *inode2, u64 loff2, u64 len,
2929                                     bool retry_range_locking)
2930 {
2931         int ret;
2932
2933         if (inode1 < inode2) {
2934                 swap(inode1, inode2);
2935                 swap(loff1, loff2);
2936         }
2937         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
2938         if (ret)
2939                 return ret;
2940         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
2941         if (ret)
2942                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
2943                               loff1 + len - 1);
2944         return ret;
2945 }
2946
2947 struct cmp_pages {
2948         int             num_pages;
2949         struct page     **src_pages;
2950         struct page     **dst_pages;
2951 };
2952
2953 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
2954 {
2955         int i;
2956         struct page *pg;
2957
2958         for (i = 0; i < cmp->num_pages; i++) {
2959                 pg = cmp->src_pages[i];
2960                 if (pg) {
2961                         unlock_page(pg);
2962                         put_page(pg);
2963                 }
2964                 pg = cmp->dst_pages[i];
2965                 if (pg) {
2966                         unlock_page(pg);
2967                         put_page(pg);
2968                 }
2969         }
2970         kfree(cmp->src_pages);
2971         kfree(cmp->dst_pages);
2972 }
2973
2974 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
2975                                   struct inode *dst, u64 dst_loff,
2976                                   u64 len, struct cmp_pages *cmp)
2977 {
2978         int ret;
2979         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
2980         struct page **src_pgarr, **dst_pgarr;
2981
2982         /*
2983          * We must gather up all the pages before we initiate our
2984          * extent locking. We use an array for the page pointers. Size
2985          * of the array is bounded by len, which is in turn bounded by
2986          * BTRFS_MAX_DEDUPE_LEN.
2987          */
2988         src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2989         dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL);
2990         if (!src_pgarr || !dst_pgarr) {
2991                 kfree(src_pgarr);
2992                 kfree(dst_pgarr);
2993                 return -ENOMEM;
2994         }
2995         cmp->num_pages = num_pages;
2996         cmp->src_pages = src_pgarr;
2997         cmp->dst_pages = dst_pgarr;
2998
2999         /*
3000          * If deduping ranges in the same inode, locking rules make it mandatory
3001          * to always lock pages in ascending order to avoid deadlocks with
3002          * concurrent tasks (such as starting writeback/delalloc).
3003          */
3004         if (src == dst && dst_loff < loff) {
3005                 swap(src_pgarr, dst_pgarr);
3006                 swap(loff, dst_loff);
3007         }
3008
3009         ret = gather_extent_pages(src, src_pgarr, cmp->num_pages, loff);
3010         if (ret)
3011                 goto out;
3012
3013         ret = gather_extent_pages(dst, dst_pgarr, cmp->num_pages, dst_loff);
3014
3015 out:
3016         if (ret)
3017                 btrfs_cmp_data_free(cmp);
3018         return ret;
3019 }
3020
3021 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3022 {
3023         int ret = 0;
3024         int i;
3025         struct page *src_page, *dst_page;
3026         unsigned int cmp_len = PAGE_SIZE;
3027         void *addr, *dst_addr;
3028
3029         i = 0;
3030         while (len) {
3031                 if (len < PAGE_SIZE)
3032                         cmp_len = len;
3033
3034                 BUG_ON(i >= cmp->num_pages);
3035
3036                 src_page = cmp->src_pages[i];
3037                 dst_page = cmp->dst_pages[i];
3038                 ASSERT(PageLocked(src_page));
3039                 ASSERT(PageLocked(dst_page));
3040
3041                 addr = kmap_atomic(src_page);
3042                 dst_addr = kmap_atomic(dst_page);
3043
3044                 flush_dcache_page(src_page);
3045                 flush_dcache_page(dst_page);
3046
3047                 if (memcmp(addr, dst_addr, cmp_len))
3048                         ret = -EBADE;
3049
3050                 kunmap_atomic(addr);
3051                 kunmap_atomic(dst_addr);
3052
3053                 if (ret)
3054                         break;
3055
3056                 len -= cmp_len;
3057                 i++;
3058         }
3059
3060         return ret;
3061 }
3062
3063 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3064                                      u64 olen)
3065 {
3066         u64 len = *plen;
3067         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3068
3069         if (off + olen > inode->i_size || off + olen < off)
3070                 return -EINVAL;
3071
3072         /* if we extend to eof, continue to block boundary */
3073         if (off + len == inode->i_size)
3074                 *plen = len = ALIGN(inode->i_size, bs) - off;
3075
3076         /* Check that we are block aligned - btrfs_clone() requires this */
3077         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3078                 return -EINVAL;
3079
3080         return 0;
3081 }
3082
3083 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3084                              struct inode *dst, u64 dst_loff)
3085 {
3086         int ret;
3087         u64 len = olen;
3088         struct cmp_pages cmp;
3089         bool same_inode = (src == dst);
3090         u64 same_lock_start = 0;
3091         u64 same_lock_len = 0;
3092
3093         if (len == 0)
3094                 return 0;
3095
3096         if (same_inode)
3097                 inode_lock(src);
3098         else
3099                 btrfs_double_inode_lock(src, dst);
3100
3101         ret = extent_same_check_offsets(src, loff, &len, olen);
3102         if (ret)
3103                 goto out_unlock;
3104
3105         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3106         if (ret)
3107                 goto out_unlock;
3108
3109         if (same_inode) {
3110                 /*
3111                  * Single inode case wants the same checks, except we
3112                  * don't want our length pushed out past i_size as
3113                  * comparing that data range makes no sense.
3114                  *
3115                  * extent_same_check_offsets() will do this for an
3116                  * unaligned length at i_size, so catch it here and
3117                  * reject the request.
3118                  *
3119                  * This effectively means we require aligned extents
3120                  * for the single-inode case, whereas the other cases
3121                  * allow an unaligned length so long as it ends at
3122                  * i_size.
3123                  */
3124                 if (len != olen) {
3125                         ret = -EINVAL;
3126                         goto out_unlock;
3127                 }
3128
3129                 /* Check for overlapping ranges */
3130                 if (dst_loff + len > loff && dst_loff < loff + len) {
3131                         ret = -EINVAL;
3132                         goto out_unlock;
3133                 }
3134
3135                 same_lock_start = min_t(u64, loff, dst_loff);
3136                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3137         }
3138
3139         /* don't make the dst file partly checksummed */
3140         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3141             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3142                 ret = -EINVAL;
3143                 goto out_unlock;
3144         }
3145
3146 again:
3147         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp);
3148         if (ret)
3149                 goto out_unlock;
3150
3151         if (same_inode)
3152                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3153                                         false);
3154         else
3155                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3156                                                false);
3157         /*
3158          * If one of the inodes has dirty pages in the respective range or
3159          * ordered extents, we need to flush dellaloc and wait for all ordered
3160          * extents in the range. We must unlock the pages and the ranges in the
3161          * io trees to avoid deadlocks when flushing delalloc (requires locking
3162          * pages) and when waiting for ordered extents to complete (they require
3163          * range locking).
3164          */
3165         if (ret == -EAGAIN) {
3166                 /*
3167                  * Ranges in the io trees already unlocked. Now unlock all
3168                  * pages before waiting for all IO to complete.
3169                  */
3170                 btrfs_cmp_data_free(&cmp);
3171                 if (same_inode) {
3172                         btrfs_wait_ordered_range(src, same_lock_start,
3173                                                  same_lock_len);
3174                 } else {
3175                         btrfs_wait_ordered_range(src, loff, len);
3176                         btrfs_wait_ordered_range(dst, dst_loff, len);
3177                 }
3178                 goto again;
3179         }
3180         ASSERT(ret == 0);
3181         if (WARN_ON(ret)) {
3182                 /* ranges in the io trees already unlocked */
3183                 btrfs_cmp_data_free(&cmp);
3184                 return ret;
3185         }
3186
3187         /* pass original length for comparison so we stay within i_size */
3188         ret = btrfs_cmp_data(olen, &cmp);
3189         if (ret == 0)
3190                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3191
3192         if (same_inode)
3193                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3194                               same_lock_start + same_lock_len - 1);
3195         else
3196                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3197
3198         btrfs_cmp_data_free(&cmp);
3199 out_unlock:
3200         if (same_inode)
3201                 inode_unlock(src);
3202         else
3203                 btrfs_double_inode_unlock(src, dst);
3204
3205         return ret;
3206 }
3207
3208 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3209
3210 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen,
3211                                 struct file *dst_file, u64 dst_loff)
3212 {
3213         struct inode *src = file_inode(src_file);
3214         struct inode *dst = file_inode(dst_file);
3215         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3216         ssize_t res;
3217
3218         if (olen > BTRFS_MAX_DEDUPE_LEN)
3219                 olen = BTRFS_MAX_DEDUPE_LEN;
3220
3221         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3222                 /*
3223                  * Btrfs does not support blocksize < page_size. As a
3224                  * result, btrfs_cmp_data() won't correctly handle
3225                  * this situation without an update.
3226                  */
3227                 return -EINVAL;
3228         }
3229
3230         res = btrfs_extent_same(src, loff, olen, dst, dst_loff);
3231         if (res)
3232                 return res;
3233         return olen;
3234 }
3235
3236 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3237                                      struct inode *inode,
3238                                      u64 endoff,
3239                                      const u64 destoff,
3240                                      const u64 olen,
3241                                      int no_time_update)
3242 {
3243         struct btrfs_root *root = BTRFS_I(inode)->root;
3244         int ret;
3245
3246         inode_inc_iversion(inode);
3247         if (!no_time_update)
3248                 inode->i_mtime = inode->i_ctime = current_time(inode);
3249         /*
3250          * We round up to the block size at eof when determining which
3251          * extents to clone above, but shouldn't round up the file size.
3252          */
3253         if (endoff > destoff + olen)
3254                 endoff = destoff + olen;
3255         if (endoff > inode->i_size)
3256                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3257
3258         ret = btrfs_update_inode(trans, root, inode);
3259         if (ret) {
3260                 btrfs_abort_transaction(trans, ret);
3261                 btrfs_end_transaction(trans);
3262                 goto out;
3263         }
3264         ret = btrfs_end_transaction(trans);
3265 out:
3266         return ret;
3267 }
3268
3269 static void clone_update_extent_map(struct btrfs_inode *inode,
3270                                     const struct btrfs_trans_handle *trans,
3271                                     const struct btrfs_path *path,
3272                                     const u64 hole_offset,
3273                                     const u64 hole_len)
3274 {
3275         struct extent_map_tree *em_tree = &inode->extent_tree;
3276         struct extent_map *em;
3277         int ret;
3278
3279         em = alloc_extent_map();
3280         if (!em) {
3281                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3282                 return;
3283         }
3284
3285         if (path) {
3286                 struct btrfs_file_extent_item *fi;
3287
3288                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3289                                     struct btrfs_file_extent_item);
3290                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3291                 em->generation = -1;
3292                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3293                     BTRFS_FILE_EXTENT_INLINE)
3294                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3295                                         &inode->runtime_flags);
3296         } else {
3297                 em->start = hole_offset;
3298                 em->len = hole_len;
3299                 em->ram_bytes = em->len;
3300                 em->orig_start = hole_offset;
3301                 em->block_start = EXTENT_MAP_HOLE;
3302                 em->block_len = 0;
3303                 em->orig_block_len = 0;
3304                 em->compress_type = BTRFS_COMPRESS_NONE;
3305                 em->generation = trans->transid;
3306         }
3307
3308         while (1) {
3309                 write_lock(&em_tree->lock);
3310                 ret = add_extent_mapping(em_tree, em, 1);
3311                 write_unlock(&em_tree->lock);
3312                 if (ret != -EEXIST) {
3313                         free_extent_map(em);
3314                         break;
3315                 }
3316                 btrfs_drop_extent_cache(inode, em->start,
3317                                         em->start + em->len - 1, 0);
3318         }
3319
3320         if (ret)
3321                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3322 }
3323
3324 /*
3325  * Make sure we do not end up inserting an inline extent into a file that has
3326  * already other (non-inline) extents. If a file has an inline extent it can
3327  * not have any other extents and the (single) inline extent must start at the
3328  * file offset 0. Failing to respect these rules will lead to file corruption,
3329  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3330  *
3331  * We can have extents that have been already written to disk or we can have
3332  * dirty ranges still in delalloc, in which case the extent maps and items are
3333  * created only when we run delalloc, and the delalloc ranges might fall outside
3334  * the range we are currently locking in the inode's io tree. So we check the
3335  * inode's i_size because of that (i_size updates are done while holding the
3336  * i_mutex, which we are holding here).
3337  * We also check to see if the inode has a size not greater than "datal" but has
3338  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3339  * protected against such concurrent fallocate calls by the i_mutex).
3340  *
3341  * If the file has no extents but a size greater than datal, do not allow the
3342  * copy because we would need turn the inline extent into a non-inline one (even
3343  * with NO_HOLES enabled). If we find our destination inode only has one inline
3344  * extent, just overwrite it with the source inline extent if its size is less
3345  * than the source extent's size, or we could copy the source inline extent's
3346  * data into the destination inode's inline extent if the later is greater then
3347  * the former.
3348  */
3349 static int clone_copy_inline_extent(struct inode *dst,
3350                                     struct btrfs_trans_handle *trans,
3351                                     struct btrfs_path *path,
3352                                     struct btrfs_key *new_key,
3353                                     const u64 drop_start,
3354                                     const u64 datal,
3355                                     const u64 skip,
3356                                     const u64 size,
3357                                     char *inline_data)
3358 {
3359         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3360         struct btrfs_root *root = BTRFS_I(dst)->root;
3361         const u64 aligned_end = ALIGN(new_key->offset + datal,
3362                                       fs_info->sectorsize);
3363         int ret;
3364         struct btrfs_key key;
3365
3366         if (new_key->offset > 0)
3367                 return -EOPNOTSUPP;
3368
3369         key.objectid = btrfs_ino(BTRFS_I(dst));
3370         key.type = BTRFS_EXTENT_DATA_KEY;
3371         key.offset = 0;
3372         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3373         if (ret < 0) {
3374                 return ret;
3375         } else if (ret > 0) {
3376                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3377                         ret = btrfs_next_leaf(root, path);
3378                         if (ret < 0)
3379                                 return ret;
3380                         else if (ret > 0)
3381                                 goto copy_inline_extent;
3382                 }
3383                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3384                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3385                     key.type == BTRFS_EXTENT_DATA_KEY) {
3386                         ASSERT(key.offset > 0);
3387                         return -EOPNOTSUPP;
3388                 }
3389         } else if (i_size_read(dst) <= datal) {
3390                 struct btrfs_file_extent_item *ei;
3391                 u64 ext_len;
3392
3393                 /*
3394                  * If the file size is <= datal, make sure there are no other
3395                  * extents following (can happen do to an fallocate call with
3396                  * the flag FALLOC_FL_KEEP_SIZE).
3397                  */
3398                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3399                                     struct btrfs_file_extent_item);
3400                 /*
3401                  * If it's an inline extent, it can not have other extents
3402                  * following it.
3403                  */
3404                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3405                     BTRFS_FILE_EXTENT_INLINE)
3406                         goto copy_inline_extent;
3407
3408                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3409                 if (ext_len > aligned_end)
3410                         return -EOPNOTSUPP;
3411
3412                 ret = btrfs_next_item(root, path);
3413                 if (ret < 0) {
3414                         return ret;
3415                 } else if (ret == 0) {
3416                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3417                                               path->slots[0]);
3418                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3419                             key.type == BTRFS_EXTENT_DATA_KEY)
3420                                 return -EOPNOTSUPP;
3421                 }
3422         }
3423
3424 copy_inline_extent:
3425         /*
3426          * We have no extent items, or we have an extent at offset 0 which may
3427          * or may not be inlined. All these cases are dealt the same way.
3428          */
3429         if (i_size_read(dst) > datal) {
3430                 /*
3431                  * If the destination inode has an inline extent...
3432                  * This would require copying the data from the source inline
3433                  * extent into the beginning of the destination's inline extent.
3434                  * But this is really complex, both extents can be compressed
3435                  * or just one of them, which would require decompressing and
3436                  * re-compressing data (which could increase the new compressed
3437                  * size, not allowing the compressed data to fit anymore in an
3438                  * inline extent).
3439                  * So just don't support this case for now (it should be rare,
3440                  * we are not really saving space when cloning inline extents).
3441                  */
3442                 return -EOPNOTSUPP;
3443         }
3444
3445         btrfs_release_path(path);
3446         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3447         if (ret)
3448                 return ret;
3449         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3450         if (ret)
3451                 return ret;
3452
3453         if (skip) {
3454                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3455
3456                 memmove(inline_data + start, inline_data + start + skip, datal);
3457         }
3458
3459         write_extent_buffer(path->nodes[0], inline_data,
3460                             btrfs_item_ptr_offset(path->nodes[0],
3461                                                   path->slots[0]),
3462                             size);
3463         inode_add_bytes(dst, datal);
3464
3465         return 0;
3466 }
3467
3468 /**
3469  * btrfs_clone() - clone a range from inode file to another
3470  *
3471  * @src: Inode to clone from
3472  * @inode: Inode to clone to
3473  * @off: Offset within source to start clone from
3474  * @olen: Original length, passed by user, of range to clone
3475  * @olen_aligned: Block-aligned value of olen
3476  * @destoff: Offset within @inode to start clone
3477  * @no_time_update: Whether to update mtime/ctime on the target inode
3478  */
3479 static int btrfs_clone(struct inode *src, struct inode *inode,
3480                        const u64 off, const u64 olen, const u64 olen_aligned,
3481                        const u64 destoff, int no_time_update)
3482 {
3483         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3484         struct btrfs_root *root = BTRFS_I(inode)->root;
3485         struct btrfs_path *path = NULL;
3486         struct extent_buffer *leaf;
3487         struct btrfs_trans_handle *trans;
3488         char *buf = NULL;
3489         struct btrfs_key key;
3490         u32 nritems;
3491         int slot;
3492         int ret;
3493         const u64 len = olen_aligned;
3494         u64 last_dest_end = destoff;
3495
3496         ret = -ENOMEM;
3497         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3498         if (!buf)
3499                 return ret;
3500
3501         path = btrfs_alloc_path();
3502         if (!path) {
3503                 kvfree(buf);
3504                 return ret;
3505         }
3506
3507         path->reada = READA_FORWARD;
3508         /* clone data */
3509         key.objectid = btrfs_ino(BTRFS_I(src));
3510         key.type = BTRFS_EXTENT_DATA_KEY;
3511         key.offset = off;
3512
3513         while (1) {
3514                 u64 next_key_min_offset = key.offset + 1;
3515
3516                 /*
3517                  * note the key will change type as we walk through the
3518                  * tree.
3519                  */
3520                 path->leave_spinning = 1;
3521                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3522                                 0, 0);
3523                 if (ret < 0)
3524                         goto out;
3525                 /*
3526                  * First search, if no extent item that starts at offset off was
3527                  * found but the previous item is an extent item, it's possible
3528                  * it might overlap our target range, therefore process it.
3529                  */
3530                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3531                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3532                                               path->slots[0] - 1);
3533                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3534                                 path->slots[0]--;
3535                 }
3536
3537                 nritems = btrfs_header_nritems(path->nodes[0]);
3538 process_slot:
3539                 if (path->slots[0] >= nritems) {
3540                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3541                         if (ret < 0)
3542                                 goto out;
3543                         if (ret > 0)
3544                                 break;
3545                         nritems = btrfs_header_nritems(path->nodes[0]);
3546                 }
3547                 leaf = path->nodes[0];
3548                 slot = path->slots[0];
3549
3550                 btrfs_item_key_to_cpu(leaf, &key, slot);
3551                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3552                     key.objectid != btrfs_ino(BTRFS_I(src)))
3553                         break;
3554
3555                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3556                         struct btrfs_file_extent_item *extent;
3557                         int type;
3558                         u32 size;
3559                         struct btrfs_key new_key;
3560                         u64 disko = 0, diskl = 0;
3561                         u64 datao = 0, datal = 0;
3562                         u8 comp;
3563                         u64 drop_start;
3564
3565                         extent = btrfs_item_ptr(leaf, slot,
3566                                                 struct btrfs_file_extent_item);
3567                         comp = btrfs_file_extent_compression(leaf, extent);
3568                         type = btrfs_file_extent_type(leaf, extent);
3569                         if (type == BTRFS_FILE_EXTENT_REG ||
3570                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3571                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3572                                                                       extent);
3573                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3574                                                                  extent);
3575                                 datao = btrfs_file_extent_offset(leaf, extent);
3576                                 datal = btrfs_file_extent_num_bytes(leaf,
3577                                                                     extent);
3578                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3579                                 /* take upper bound, may be compressed */
3580                                 datal = btrfs_file_extent_ram_bytes(leaf,
3581                                                                     extent);
3582                         }
3583
3584                         /*
3585                          * The first search might have left us at an extent
3586                          * item that ends before our target range's start, can
3587                          * happen if we have holes and NO_HOLES feature enabled.
3588                          */
3589                         if (key.offset + datal <= off) {
3590                                 path->slots[0]++;
3591                                 goto process_slot;
3592                         } else if (key.offset >= off + len) {
3593                                 break;
3594                         }
3595                         next_key_min_offset = key.offset + datal;
3596                         size = btrfs_item_size_nr(leaf, slot);
3597                         read_extent_buffer(leaf, buf,
3598                                            btrfs_item_ptr_offset(leaf, slot),
3599                                            size);
3600
3601                         btrfs_release_path(path);
3602                         path->leave_spinning = 0;
3603
3604                         memcpy(&new_key, &key, sizeof(new_key));
3605                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3606                         if (off <= key.offset)
3607                                 new_key.offset = key.offset + destoff - off;
3608                         else
3609                                 new_key.offset = destoff;
3610
3611                         /*
3612                          * Deal with a hole that doesn't have an extent item
3613                          * that represents it (NO_HOLES feature enabled).
3614                          * This hole is either in the middle of the cloning
3615                          * range or at the beginning (fully overlaps it or
3616                          * partially overlaps it).
3617                          */
3618                         if (new_key.offset != last_dest_end)
3619                                 drop_start = last_dest_end;
3620                         else
3621                                 drop_start = new_key.offset;
3622
3623                         /*
3624                          * 1 - adjusting old extent (we may have to split it)
3625                          * 1 - add new extent
3626                          * 1 - inode update
3627                          */
3628                         trans = btrfs_start_transaction(root, 3);
3629                         if (IS_ERR(trans)) {
3630                                 ret = PTR_ERR(trans);
3631                                 goto out;
3632                         }
3633
3634                         if (type == BTRFS_FILE_EXTENT_REG ||
3635                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3636                                 /*
3637                                  *    a  | --- range to clone ---|  b
3638                                  * | ------------- extent ------------- |
3639                                  */
3640
3641                                 /* subtract range b */
3642                                 if (key.offset + datal > off + len)
3643                                         datal = off + len - key.offset;
3644
3645                                 /* subtract range a */
3646                                 if (off > key.offset) {
3647                                         datao += off - key.offset;
3648                                         datal -= off - key.offset;
3649                                 }
3650
3651                                 ret = btrfs_drop_extents(trans, root, inode,
3652                                                          drop_start,
3653                                                          new_key.offset + datal,
3654                                                          1);
3655                                 if (ret) {
3656                                         if (ret != -EOPNOTSUPP)
3657                                                 btrfs_abort_transaction(trans,
3658                                                                         ret);
3659                                         btrfs_end_transaction(trans);
3660                                         goto out;
3661                                 }
3662
3663                                 ret = btrfs_insert_empty_item(trans, root, path,
3664                                                               &new_key, size);
3665                                 if (ret) {
3666                                         btrfs_abort_transaction(trans, ret);
3667                                         btrfs_end_transaction(trans);
3668                                         goto out;
3669                                 }
3670
3671                                 leaf = path->nodes[0];
3672                                 slot = path->slots[0];
3673                                 write_extent_buffer(leaf, buf,
3674                                             btrfs_item_ptr_offset(leaf, slot),
3675                                             size);
3676
3677                                 extent = btrfs_item_ptr(leaf, slot,
3678                                                 struct btrfs_file_extent_item);
3679
3680                                 /* disko == 0 means it's a hole */
3681                                 if (!disko)
3682                                         datao = 0;
3683
3684                                 btrfs_set_file_extent_offset(leaf, extent,
3685                                                              datao);
3686                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3687                                                                 datal);
3688
3689                                 if (disko) {
3690                                         inode_add_bytes(inode, datal);
3691                                         ret = btrfs_inc_extent_ref(trans,
3692                                                         root,
3693                                                         disko, diskl, 0,
3694                                                         root->root_key.objectid,
3695                                                         btrfs_ino(BTRFS_I(inode)),
3696                                                         new_key.offset - datao);
3697                                         if (ret) {
3698                                                 btrfs_abort_transaction(trans,
3699                                                                         ret);
3700                                                 btrfs_end_transaction(trans);
3701                                                 goto out;
3702
3703                                         }
3704                                 }
3705                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3706                                 u64 skip = 0;
3707                                 u64 trim = 0;
3708
3709                                 if (off > key.offset) {
3710                                         skip = off - key.offset;
3711                                         new_key.offset += skip;
3712                                 }
3713
3714                                 if (key.offset + datal > off + len)
3715                                         trim = key.offset + datal - (off + len);
3716
3717                                 if (comp && (skip || trim)) {
3718                                         ret = -EINVAL;
3719                                         btrfs_end_transaction(trans);
3720                                         goto out;
3721                                 }
3722                                 size -= skip + trim;
3723                                 datal -= skip + trim;
3724
3725                                 ret = clone_copy_inline_extent(inode,
3726                                                                trans, path,
3727                                                                &new_key,
3728                                                                drop_start,
3729                                                                datal,
3730                                                                skip, size, buf);
3731                                 if (ret) {
3732                                         if (ret != -EOPNOTSUPP)
3733                                                 btrfs_abort_transaction(trans,
3734                                                                         ret);
3735                                         btrfs_end_transaction(trans);
3736                                         goto out;
3737                                 }
3738                                 leaf = path->nodes[0];
3739                                 slot = path->slots[0];
3740                         }
3741
3742                         /* If we have an implicit hole (NO_HOLES feature). */
3743                         if (drop_start < new_key.offset)
3744                                 clone_update_extent_map(BTRFS_I(inode), trans,
3745                                                 NULL, drop_start,
3746                                                 new_key.offset - drop_start);
3747
3748                         clone_update_extent_map(BTRFS_I(inode), trans,
3749                                         path, 0, 0);
3750
3751                         btrfs_mark_buffer_dirty(leaf);
3752                         btrfs_release_path(path);
3753
3754                         last_dest_end = ALIGN(new_key.offset + datal,
3755                                               fs_info->sectorsize);
3756                         ret = clone_finish_inode_update(trans, inode,
3757                                                         last_dest_end,
3758                                                         destoff, olen,
3759                                                         no_time_update);
3760                         if (ret)
3761                                 goto out;
3762                         if (new_key.offset + datal >= destoff + len)
3763                                 break;
3764                 }
3765                 btrfs_release_path(path);
3766                 key.offset = next_key_min_offset;
3767
3768                 if (fatal_signal_pending(current)) {
3769                         ret = -EINTR;
3770                         goto out;
3771                 }
3772         }
3773         ret = 0;
3774
3775         if (last_dest_end < destoff + len) {
3776                 /*
3777                  * We have an implicit hole (NO_HOLES feature is enabled) that
3778                  * fully or partially overlaps our cloning range at its end.
3779                  */
3780                 btrfs_release_path(path);
3781
3782                 /*
3783                  * 1 - remove extent(s)
3784                  * 1 - inode update
3785                  */
3786                 trans = btrfs_start_transaction(root, 2);
3787                 if (IS_ERR(trans)) {
3788                         ret = PTR_ERR(trans);
3789                         goto out;
3790                 }
3791                 ret = btrfs_drop_extents(trans, root, inode,
3792                                          last_dest_end, destoff + len, 1);
3793                 if (ret) {
3794                         if (ret != -EOPNOTSUPP)
3795                                 btrfs_abort_transaction(trans, ret);
3796                         btrfs_end_transaction(trans);
3797                         goto out;
3798                 }
3799                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3800                                 last_dest_end,
3801                                 destoff + len - last_dest_end);
3802                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3803                                                 destoff, olen, no_time_update);
3804         }
3805
3806 out:
3807         btrfs_free_path(path);
3808         kvfree(buf);
3809         return ret;
3810 }
3811
3812 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3813                                         u64 off, u64 olen, u64 destoff)
3814 {
3815         struct inode *inode = file_inode(file);
3816         struct inode *src = file_inode(file_src);
3817         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3818         struct btrfs_root *root = BTRFS_I(inode)->root;
3819         int ret;
3820         u64 len = olen;
3821         u64 bs = fs_info->sb->s_blocksize;
3822         int same_inode = src == inode;
3823
3824         /*
3825          * TODO:
3826          * - split compressed inline extents.  annoying: we need to
3827          *   decompress into destination's address_space (the file offset
3828          *   may change, so source mapping won't do), then recompress (or
3829          *   otherwise reinsert) a subrange.
3830          *
3831          * - split destination inode's inline extents.  The inline extents can
3832          *   be either compressed or non-compressed.
3833          */
3834
3835         if (btrfs_root_readonly(root))
3836                 return -EROFS;
3837
3838         if (file_src->f_path.mnt != file->f_path.mnt ||
3839             src->i_sb != inode->i_sb)
3840                 return -EXDEV;
3841
3842         /* don't make the dst file partly checksummed */
3843         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3844             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
3845                 return -EINVAL;
3846
3847         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3848                 return -EISDIR;
3849
3850         if (!same_inode) {
3851                 btrfs_double_inode_lock(src, inode);
3852         } else {
3853                 inode_lock(src);
3854         }
3855
3856         /* determine range to clone */
3857         ret = -EINVAL;
3858         if (off + len > src->i_size || off + len < off)
3859                 goto out_unlock;
3860         if (len == 0)
3861                 olen = len = src->i_size - off;
3862         /* if we extend to eof, continue to block boundary */
3863         if (off + len == src->i_size)
3864                 len = ALIGN(src->i_size, bs) - off;
3865
3866         if (len == 0) {
3867                 ret = 0;
3868                 goto out_unlock;
3869         }
3870
3871         /* verify the end result is block aligned */
3872         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
3873             !IS_ALIGNED(destoff, bs))
3874                 goto out_unlock;
3875
3876         /* verify if ranges are overlapped within the same file */
3877         if (same_inode) {
3878                 if (destoff + len > off && destoff < off + len)
3879                         goto out_unlock;
3880         }
3881
3882         if (destoff > inode->i_size) {
3883                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3884                 if (ret)
3885                         goto out_unlock;
3886         }
3887
3888         /*
3889          * Lock the target range too. Right after we replace the file extent
3890          * items in the fs tree (which now point to the cloned data), we might
3891          * have a worker replace them with extent items relative to a write
3892          * operation that was issued before this clone operation (i.e. confront
3893          * with inode.c:btrfs_finish_ordered_io).
3894          */
3895         if (same_inode) {
3896                 u64 lock_start = min_t(u64, off, destoff);
3897                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
3898
3899                 ret = lock_extent_range(src, lock_start, lock_len, true);
3900         } else {
3901                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
3902                                                true);
3903         }
3904         ASSERT(ret == 0);
3905         if (WARN_ON(ret)) {
3906                 /* ranges in the io trees already unlocked */
3907                 goto out_unlock;
3908         }
3909
3910         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3911
3912         if (same_inode) {
3913                 u64 lock_start = min_t(u64, off, destoff);
3914                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
3915
3916                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
3917         } else {
3918                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
3919         }
3920         /*
3921          * Truncate page cache pages so that future reads will see the cloned
3922          * data immediately and not the previous data.
3923          */
3924         truncate_inode_pages_range(&inode->i_data,
3925                                 round_down(destoff, PAGE_SIZE),
3926                                 round_up(destoff + len, PAGE_SIZE) - 1);
3927 out_unlock:
3928         if (!same_inode)
3929                 btrfs_double_inode_unlock(src, inode);
3930         else
3931                 inode_unlock(src);
3932         return ret;
3933 }
3934
3935 int btrfs_clone_file_range(struct file *src_file, loff_t off,
3936                 struct file *dst_file, loff_t destoff, u64 len)
3937 {
3938         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
3939 }
3940
3941 /*
3942  * there are many ways the trans_start and trans_end ioctls can lead
3943  * to deadlocks.  They should only be used by applications that
3944  * basically own the machine, and have a very in depth understanding
3945  * of all the possible deadlocks and enospc problems.
3946  */
3947 static long btrfs_ioctl_trans_start(struct file *file)
3948 {
3949         struct inode *inode = file_inode(file);
3950         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3951         struct btrfs_root *root = BTRFS_I(inode)->root;
3952         struct btrfs_trans_handle *trans;
3953         struct btrfs_file_private *private;
3954         int ret;
3955         static bool warned = false;
3956
3957         ret = -EPERM;
3958         if (!capable(CAP_SYS_ADMIN))
3959                 goto out;
3960
3961         if (!warned) {
3962                 btrfs_warn(fs_info,
3963                         "Userspace transaction mechanism is considered "
3964                         "deprecated and slated to be removed in 4.17. "
3965                         "If you have a valid use case please "
3966                         "speak up on the mailing list");
3967                 WARN_ON(1);
3968                 warned = true;
3969         }
3970
3971         ret = -EINPROGRESS;
3972         private = file->private_data;
3973         if (private && private->trans)
3974                 goto out;
3975         if (!private) {
3976                 private = kzalloc(sizeof(struct btrfs_file_private),
3977                                   GFP_KERNEL);
3978                 if (!private)
3979                         return -ENOMEM;
3980                 file->private_data = private;
3981         }
3982
3983         ret = -EROFS;
3984         if (btrfs_root_readonly(root))
3985                 goto out;
3986
3987         ret = mnt_want_write_file(file);
3988         if (ret)
3989                 goto out;
3990
3991         atomic_inc(&fs_info->open_ioctl_trans);
3992
3993         ret = -ENOMEM;
3994         trans = btrfs_start_ioctl_transaction(root);
3995         if (IS_ERR(trans))
3996                 goto out_drop;
3997
3998         private->trans = trans;
3999         return 0;
4000
4001 out_drop:
4002         atomic_dec(&fs_info->open_ioctl_trans);
4003         mnt_drop_write_file(file);
4004 out:
4005         return ret;
4006 }
4007
4008 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4009 {
4010         struct inode *inode = file_inode(file);
4011         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4012         struct btrfs_root *root = BTRFS_I(inode)->root;
4013         struct btrfs_root *new_root;
4014         struct btrfs_dir_item *di;
4015         struct btrfs_trans_handle *trans;
4016         struct btrfs_path *path;
4017         struct btrfs_key location;
4018         struct btrfs_disk_key disk_key;
4019         u64 objectid = 0;
4020         u64 dir_id;
4021         int ret;
4022
4023         if (!capable(CAP_SYS_ADMIN))
4024                 return -EPERM;
4025
4026         ret = mnt_want_write_file(file);
4027         if (ret)
4028                 return ret;
4029
4030         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4031                 ret = -EFAULT;
4032                 goto out;
4033         }
4034
4035         if (!objectid)
4036                 objectid = BTRFS_FS_TREE_OBJECTID;
4037
4038         location.objectid = objectid;
4039         location.type = BTRFS_ROOT_ITEM_KEY;
4040         location.offset = (u64)-1;
4041
4042         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4043         if (IS_ERR(new_root)) {
4044                 ret = PTR_ERR(new_root);
4045                 goto out;
4046         }
4047         if (!is_fstree(new_root->objectid)) {
4048                 ret = -ENOENT;
4049                 goto out;
4050         }
4051
4052         path = btrfs_alloc_path();
4053         if (!path) {
4054                 ret = -ENOMEM;
4055                 goto out;
4056         }
4057         path->leave_spinning = 1;
4058
4059         trans = btrfs_start_transaction(root, 1);
4060         if (IS_ERR(trans)) {
4061                 btrfs_free_path(path);
4062                 ret = PTR_ERR(trans);
4063                 goto out;
4064         }
4065
4066         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4067         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4068                                    dir_id, "default", 7, 1);
4069         if (IS_ERR_OR_NULL(di)) {
4070                 btrfs_free_path(path);
4071                 btrfs_end_transaction(trans);
4072                 btrfs_err(fs_info,
4073                           "Umm, you don't have the default diritem, this isn't going to work");
4074                 ret = -ENOENT;
4075                 goto out;
4076         }
4077
4078         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4079         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4080         btrfs_mark_buffer_dirty(path->nodes[0]);
4081         btrfs_free_path(path);
4082
4083         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4084         btrfs_end_transaction(trans);
4085 out:
4086         mnt_drop_write_file(file);
4087         return ret;
4088 }
4089
4090 void btrfs_get_block_group_info(struct list_head *groups_list,
4091                                 struct btrfs_ioctl_space_info *space)
4092 {
4093         struct btrfs_block_group_cache *block_group;
4094
4095         space->total_bytes = 0;
4096         space->used_bytes = 0;
4097         space->flags = 0;
4098         list_for_each_entry(block_group, groups_list, list) {
4099                 space->flags = block_group->flags;
4100                 space->total_bytes += block_group->key.offset;
4101                 space->used_bytes +=
4102                         btrfs_block_group_used(&block_group->item);
4103         }
4104 }
4105
4106 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4107                                    void __user *arg)
4108 {
4109         struct btrfs_ioctl_space_args space_args;
4110         struct btrfs_ioctl_space_info space;
4111         struct btrfs_ioctl_space_info *dest;
4112         struct btrfs_ioctl_space_info *dest_orig;
4113         struct btrfs_ioctl_space_info __user *user_dest;
4114         struct btrfs_space_info *info;
4115         static const u64 types[] = {
4116                 BTRFS_BLOCK_GROUP_DATA,
4117                 BTRFS_BLOCK_GROUP_SYSTEM,
4118                 BTRFS_BLOCK_GROUP_METADATA,
4119                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4120         };
4121         int num_types = 4;
4122         int alloc_size;
4123         int ret = 0;
4124         u64 slot_count = 0;
4125         int i, c;
4126
4127         if (copy_from_user(&space_args,
4128                            (struct btrfs_ioctl_space_args __user *)arg,
4129                            sizeof(space_args)))
4130                 return -EFAULT;
4131
4132         for (i = 0; i < num_types; i++) {
4133                 struct btrfs_space_info *tmp;
4134
4135                 info = NULL;
4136                 rcu_read_lock();
4137                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4138                                         list) {
4139                         if (tmp->flags == types[i]) {
4140                                 info = tmp;
4141                                 break;
4142                         }
4143                 }
4144                 rcu_read_unlock();
4145
4146                 if (!info)
4147                         continue;
4148
4149                 down_read(&info->groups_sem);
4150                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4151                         if (!list_empty(&info->block_groups[c]))
4152                                 slot_count++;
4153                 }
4154                 up_read(&info->groups_sem);
4155         }
4156
4157         /*
4158          * Global block reserve, exported as a space_info
4159          */
4160         slot_count++;
4161
4162         /* space_slots == 0 means they are asking for a count */
4163         if (space_args.space_slots == 0) {
4164                 space_args.total_spaces = slot_count;
4165                 goto out;
4166         }
4167
4168         slot_count = min_t(u64, space_args.space_slots, slot_count);
4169
4170         alloc_size = sizeof(*dest) * slot_count;
4171
4172         /* we generally have at most 6 or so space infos, one for each raid
4173          * level.  So, a whole page should be more than enough for everyone
4174          */
4175         if (alloc_size > PAGE_SIZE)
4176                 return -ENOMEM;
4177
4178         space_args.total_spaces = 0;
4179         dest = kmalloc(alloc_size, GFP_KERNEL);
4180         if (!dest)
4181                 return -ENOMEM;
4182         dest_orig = dest;
4183
4184         /* now we have a buffer to copy into */
4185         for (i = 0; i < num_types; i++) {
4186                 struct btrfs_space_info *tmp;
4187
4188                 if (!slot_count)
4189                         break;
4190
4191                 info = NULL;
4192                 rcu_read_lock();
4193                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4194                                         list) {
4195                         if (tmp->flags == types[i]) {
4196                                 info = tmp;
4197                                 break;
4198                         }
4199                 }
4200                 rcu_read_unlock();
4201
4202                 if (!info)
4203                         continue;
4204                 down_read(&info->groups_sem);
4205                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4206                         if (!list_empty(&info->block_groups[c])) {
4207                                 btrfs_get_block_group_info(
4208                                         &info->block_groups[c], &space);
4209                                 memcpy(dest, &space, sizeof(space));
4210                                 dest++;
4211                                 space_args.total_spaces++;
4212                                 slot_count--;
4213                         }
4214                         if (!slot_count)
4215                                 break;
4216                 }
4217                 up_read(&info->groups_sem);
4218         }
4219
4220         /*
4221          * Add global block reserve
4222          */
4223         if (slot_count) {
4224                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4225
4226                 spin_lock(&block_rsv->lock);
4227                 space.total_bytes = block_rsv->size;
4228                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4229                 spin_unlock(&block_rsv->lock);
4230                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4231                 memcpy(dest, &space, sizeof(space));
4232                 space_args.total_spaces++;
4233         }
4234
4235         user_dest = (struct btrfs_ioctl_space_info __user *)
4236                 (arg + sizeof(struct btrfs_ioctl_space_args));
4237
4238         if (copy_to_user(user_dest, dest_orig, alloc_size))
4239                 ret = -EFAULT;
4240
4241         kfree(dest_orig);
4242 out:
4243         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4244                 ret = -EFAULT;
4245
4246         return ret;
4247 }
4248
4249 /*
4250  * there are many ways the trans_start and trans_end ioctls can lead
4251  * to deadlocks.  They should only be used by applications that
4252  * basically own the machine, and have a very in depth understanding
4253  * of all the possible deadlocks and enospc problems.
4254  */
4255 long btrfs_ioctl_trans_end(struct file *file)
4256 {
4257         struct inode *inode = file_inode(file);
4258         struct btrfs_root *root = BTRFS_I(inode)->root;
4259         struct btrfs_file_private *private = file->private_data;
4260
4261         if (!private || !private->trans)
4262                 return -EINVAL;
4263
4264         btrfs_end_transaction(private->trans);
4265         private->trans = NULL;
4266
4267         atomic_dec(&root->fs_info->open_ioctl_trans);
4268
4269         mnt_drop_write_file(file);
4270         return 0;
4271 }
4272
4273 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4274                                             void __user *argp)
4275 {
4276         struct btrfs_trans_handle *trans;
4277         u64 transid;
4278         int ret;
4279
4280         trans = btrfs_attach_transaction_barrier(root);
4281         if (IS_ERR(trans)) {
4282                 if (PTR_ERR(trans) != -ENOENT)
4283                         return PTR_ERR(trans);
4284
4285                 /* No running transaction, don't bother */
4286                 transid = root->fs_info->last_trans_committed;
4287                 goto out;
4288         }
4289         transid = trans->transid;
4290         ret = btrfs_commit_transaction_async(trans, 0);
4291         if (ret) {
4292                 btrfs_end_transaction(trans);
4293                 return ret;
4294         }
4295 out:
4296         if (argp)
4297                 if (copy_to_user(argp, &transid, sizeof(transid)))
4298                         return -EFAULT;
4299         return 0;
4300 }
4301
4302 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4303                                            void __user *argp)
4304 {
4305         u64 transid;
4306
4307         if (argp) {
4308                 if (copy_from_user(&transid, argp, sizeof(transid)))
4309                         return -EFAULT;
4310         } else {
4311                 transid = 0;  /* current trans */
4312         }
4313         return btrfs_wait_for_commit(fs_info, transid);
4314 }
4315
4316 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4317 {
4318         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4319         struct btrfs_ioctl_scrub_args *sa;
4320         int ret;
4321
4322         if (!capable(CAP_SYS_ADMIN))
4323                 return -EPERM;
4324
4325         sa = memdup_user(arg, sizeof(*sa));
4326         if (IS_ERR(sa))
4327                 return PTR_ERR(sa);
4328
4329         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4330                 ret = mnt_want_write_file(file);
4331                 if (ret)
4332                         goto out;
4333         }
4334
4335         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4336                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4337                               0);
4338
4339         if (copy_to_user(arg, sa, sizeof(*sa)))
4340                 ret = -EFAULT;
4341
4342         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4343                 mnt_drop_write_file(file);
4344 out:
4345         kfree(sa);
4346         return ret;
4347 }
4348
4349 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4350 {
4351         if (!capable(CAP_SYS_ADMIN))
4352                 return -EPERM;
4353
4354         return btrfs_scrub_cancel(fs_info);
4355 }
4356
4357 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4358                                        void __user *arg)
4359 {
4360         struct btrfs_ioctl_scrub_args *sa;
4361         int ret;
4362
4363         if (!capable(CAP_SYS_ADMIN))
4364                 return -EPERM;
4365
4366         sa = memdup_user(arg, sizeof(*sa));
4367         if (IS_ERR(sa))
4368                 return PTR_ERR(sa);
4369
4370         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4371
4372         if (copy_to_user(arg, sa, sizeof(*sa)))
4373                 ret = -EFAULT;
4374
4375         kfree(sa);
4376         return ret;
4377 }
4378
4379 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4380                                       void __user *arg)
4381 {
4382         struct btrfs_ioctl_get_dev_stats *sa;
4383         int ret;
4384
4385         sa = memdup_user(arg, sizeof(*sa));
4386         if (IS_ERR(sa))
4387                 return PTR_ERR(sa);
4388
4389         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4390                 kfree(sa);
4391                 return -EPERM;
4392         }
4393
4394         ret = btrfs_get_dev_stats(fs_info, sa);
4395
4396         if (copy_to_user(arg, sa, sizeof(*sa)))
4397                 ret = -EFAULT;
4398
4399         kfree(sa);
4400         return ret;
4401 }
4402
4403 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4404                                     void __user *arg)
4405 {
4406         struct btrfs_ioctl_dev_replace_args *p;
4407         int ret;
4408
4409         if (!capable(CAP_SYS_ADMIN))
4410                 return -EPERM;
4411
4412         p = memdup_user(arg, sizeof(*p));
4413         if (IS_ERR(p))
4414                 return PTR_ERR(p);
4415
4416         switch (p->cmd) {
4417         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4418                 if (sb_rdonly(fs_info->sb)) {
4419                         ret = -EROFS;
4420                         goto out;
4421                 }
4422                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4423                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4424                 } else {
4425                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4426                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4427                 }
4428                 break;
4429         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4430                 btrfs_dev_replace_status(fs_info, p);
4431                 ret = 0;
4432                 break;
4433         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4434                 ret = btrfs_dev_replace_cancel(fs_info, p);
4435                 break;
4436         default:
4437                 ret = -EINVAL;
4438                 break;
4439         }
4440
4441         if (copy_to_user(arg, p, sizeof(*p)))
4442                 ret = -EFAULT;
4443 out:
4444         kfree(p);
4445         return ret;
4446 }
4447
4448 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4449 {
4450         int ret = 0;
4451         int i;
4452         u64 rel_ptr;
4453         int size;
4454         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4455         struct inode_fs_paths *ipath = NULL;
4456         struct btrfs_path *path;
4457
4458         if (!capable(CAP_DAC_READ_SEARCH))
4459                 return -EPERM;
4460
4461         path = btrfs_alloc_path();
4462         if (!path) {
4463                 ret = -ENOMEM;
4464                 goto out;
4465         }
4466
4467         ipa = memdup_user(arg, sizeof(*ipa));
4468         if (IS_ERR(ipa)) {
4469                 ret = PTR_ERR(ipa);
4470                 ipa = NULL;
4471                 goto out;
4472         }
4473
4474         size = min_t(u32, ipa->size, 4096);
4475         ipath = init_ipath(size, root, path);
4476         if (IS_ERR(ipath)) {
4477                 ret = PTR_ERR(ipath);
4478                 ipath = NULL;
4479                 goto out;
4480         }
4481
4482         ret = paths_from_inode(ipa->inum, ipath);
4483         if (ret < 0)
4484                 goto out;
4485
4486         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4487                 rel_ptr = ipath->fspath->val[i] -
4488                           (u64)(unsigned long)ipath->fspath->val;
4489                 ipath->fspath->val[i] = rel_ptr;
4490         }
4491
4492         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4493                            ipath->fspath, size);
4494         if (ret) {
4495                 ret = -EFAULT;
4496                 goto out;
4497         }
4498
4499 out:
4500         btrfs_free_path(path);
4501         free_ipath(ipath);
4502         kfree(ipa);
4503
4504         return ret;
4505 }
4506
4507 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4508 {
4509         struct btrfs_data_container *inodes = ctx;
4510         const size_t c = 3 * sizeof(u64);
4511
4512         if (inodes->bytes_left >= c) {
4513                 inodes->bytes_left -= c;
4514                 inodes->val[inodes->elem_cnt] = inum;
4515                 inodes->val[inodes->elem_cnt + 1] = offset;
4516                 inodes->val[inodes->elem_cnt + 2] = root;
4517                 inodes->elem_cnt += 3;
4518         } else {
4519                 inodes->bytes_missing += c - inodes->bytes_left;
4520                 inodes->bytes_left = 0;
4521                 inodes->elem_missed += 3;
4522         }
4523
4524         return 0;
4525 }
4526
4527 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4528                                         void __user *arg, int version)
4529 {
4530         int ret = 0;
4531         int size;
4532         struct btrfs_ioctl_logical_ino_args *loi;
4533         struct btrfs_data_container *inodes = NULL;
4534         struct btrfs_path *path = NULL;
4535         bool ignore_offset;
4536
4537         if (!capable(CAP_SYS_ADMIN))
4538                 return -EPERM;
4539
4540         loi = memdup_user(arg, sizeof(*loi));
4541         if (IS_ERR(loi))
4542                 return PTR_ERR(loi);
4543
4544         if (version == 1) {
4545                 ignore_offset = false;
4546                 size = min_t(u32, loi->size, SZ_64K);
4547         } else {
4548                 /* All reserved bits must be 0 for now */
4549                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4550                         ret = -EINVAL;
4551                         goto out_loi;
4552                 }
4553                 /* Only accept flags we have defined so far */
4554                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4555                         ret = -EINVAL;
4556                         goto out_loi;
4557                 }
4558                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4559                 size = min_t(u32, loi->size, SZ_16M);
4560         }
4561
4562         path = btrfs_alloc_path();
4563         if (!path) {
4564                 ret = -ENOMEM;
4565                 goto out;
4566         }
4567
4568         inodes = init_data_container(size);
4569         if (IS_ERR(inodes)) {
4570                 ret = PTR_ERR(inodes);
4571                 inodes = NULL;
4572                 goto out;
4573         }
4574
4575         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4576                                           build_ino_list, inodes, ignore_offset);
4577         if (ret == -EINVAL)
4578                 ret = -ENOENT;
4579         if (ret < 0)
4580                 goto out;
4581
4582         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4583                            size);
4584         if (ret)
4585                 ret = -EFAULT;
4586
4587 out:
4588         btrfs_free_path(path);
4589         kvfree(inodes);
4590 out_loi:
4591         kfree(loi);
4592
4593         return ret;
4594 }
4595
4596 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
4597                                struct btrfs_ioctl_balance_args *bargs)
4598 {
4599         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4600
4601         bargs->flags = bctl->flags;
4602
4603         if (atomic_read(&fs_info->balance_running))
4604                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4605         if (atomic_read(&fs_info->balance_pause_req))
4606                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4607         if (atomic_read(&fs_info->balance_cancel_req))
4608                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4609
4610         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4611         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4612         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4613
4614         if (lock) {
4615                 spin_lock(&fs_info->balance_lock);
4616                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4617                 spin_unlock(&fs_info->balance_lock);
4618         } else {
4619                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4620         }
4621 }
4622
4623 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4624 {
4625         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4626         struct btrfs_fs_info *fs_info = root->fs_info;
4627         struct btrfs_ioctl_balance_args *bargs;
4628         struct btrfs_balance_control *bctl;
4629         bool need_unlock; /* for mut. excl. ops lock */
4630         int ret;
4631
4632         if (!capable(CAP_SYS_ADMIN))
4633                 return -EPERM;
4634
4635         ret = mnt_want_write_file(file);
4636         if (ret)
4637                 return ret;
4638
4639 again:
4640         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4641                 mutex_lock(&fs_info->volume_mutex);
4642                 mutex_lock(&fs_info->balance_mutex);
4643                 need_unlock = true;
4644                 goto locked;
4645         }
4646
4647         /*
4648          * mut. excl. ops lock is locked.  Three possibilities:
4649          *   (1) some other op is running
4650          *   (2) balance is running
4651          *   (3) balance is paused -- special case (think resume)
4652          */
4653         mutex_lock(&fs_info->balance_mutex);
4654         if (fs_info->balance_ctl) {
4655                 /* this is either (2) or (3) */
4656                 if (!atomic_read(&fs_info->balance_running)) {
4657                         mutex_unlock(&fs_info->balance_mutex);
4658                         if (!mutex_trylock(&fs_info->volume_mutex))
4659                                 goto again;
4660                         mutex_lock(&fs_info->balance_mutex);
4661
4662                         if (fs_info->balance_ctl &&
4663                             !atomic_read(&fs_info->balance_running)) {
4664                                 /* this is (3) */
4665                                 need_unlock = false;
4666                                 goto locked;
4667                         }
4668
4669                         mutex_unlock(&fs_info->balance_mutex);
4670                         mutex_unlock(&fs_info->volume_mutex);
4671                         goto again;
4672                 } else {
4673                         /* this is (2) */
4674                         mutex_unlock(&fs_info->balance_mutex);
4675                         ret = -EINPROGRESS;
4676                         goto out;
4677                 }
4678         } else {
4679                 /* this is (1) */
4680                 mutex_unlock(&fs_info->balance_mutex);
4681                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4682                 goto out;
4683         }
4684
4685 locked:
4686         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4687
4688         if (arg) {
4689                 bargs = memdup_user(arg, sizeof(*bargs));
4690                 if (IS_ERR(bargs)) {
4691                         ret = PTR_ERR(bargs);
4692                         goto out_unlock;
4693                 }
4694
4695                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4696                         if (!fs_info->balance_ctl) {
4697                                 ret = -ENOTCONN;
4698                                 goto out_bargs;
4699                         }
4700
4701                         bctl = fs_info->balance_ctl;
4702                         spin_lock(&fs_info->balance_lock);
4703                         bctl->flags |= BTRFS_BALANCE_RESUME;
4704                         spin_unlock(&fs_info->balance_lock);
4705
4706                         goto do_balance;
4707                 }
4708         } else {
4709                 bargs = NULL;
4710         }
4711
4712         if (fs_info->balance_ctl) {
4713                 ret = -EINPROGRESS;
4714                 goto out_bargs;
4715         }
4716
4717         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4718         if (!bctl) {
4719                 ret = -ENOMEM;
4720                 goto out_bargs;
4721         }
4722
4723         bctl->fs_info = fs_info;
4724         if (arg) {
4725                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4726                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4727                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4728
4729                 bctl->flags = bargs->flags;
4730         } else {
4731                 /* balance everything - no filters */
4732                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4733         }
4734
4735         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4736                 ret = -EINVAL;
4737                 goto out_bctl;
4738         }
4739
4740 do_balance:
4741         /*
4742          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP
4743          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
4744          * or, if restriper was paused all the way until unmount, in
4745          * free_fs_info.  The flag is cleared in __cancel_balance.
4746          */
4747         need_unlock = false;
4748
4749         ret = btrfs_balance(bctl, bargs);
4750         bctl = NULL;
4751
4752         if (arg) {
4753                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4754                         ret = -EFAULT;
4755         }
4756
4757 out_bctl:
4758         kfree(bctl);
4759 out_bargs:
4760         kfree(bargs);
4761 out_unlock:
4762         mutex_unlock(&fs_info->balance_mutex);
4763         mutex_unlock(&fs_info->volume_mutex);
4764         if (need_unlock)
4765                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4766 out:
4767         mnt_drop_write_file(file);
4768         return ret;
4769 }
4770
4771 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4772 {
4773         if (!capable(CAP_SYS_ADMIN))
4774                 return -EPERM;
4775
4776         switch (cmd) {
4777         case BTRFS_BALANCE_CTL_PAUSE:
4778                 return btrfs_pause_balance(fs_info);
4779         case BTRFS_BALANCE_CTL_CANCEL:
4780                 return btrfs_cancel_balance(fs_info);
4781         }
4782
4783         return -EINVAL;
4784 }
4785
4786 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4787                                          void __user *arg)
4788 {
4789         struct btrfs_ioctl_balance_args *bargs;
4790         int ret = 0;
4791
4792         if (!capable(CAP_SYS_ADMIN))
4793                 return -EPERM;
4794
4795         mutex_lock(&fs_info->balance_mutex);
4796         if (!fs_info->balance_ctl) {
4797                 ret = -ENOTCONN;
4798                 goto out;
4799         }
4800
4801         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4802         if (!bargs) {
4803                 ret = -ENOMEM;
4804                 goto out;
4805         }
4806
4807         update_ioctl_balance_args(fs_info, 1, bargs);
4808
4809         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4810                 ret = -EFAULT;
4811
4812         kfree(bargs);
4813 out:
4814         mutex_unlock(&fs_info->balance_mutex);
4815         return ret;
4816 }
4817
4818 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4819 {
4820         struct inode *inode = file_inode(file);
4821         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4822         struct btrfs_ioctl_quota_ctl_args *sa;
4823         struct btrfs_trans_handle *trans = NULL;
4824         int ret;
4825         int err;
4826
4827         if (!capable(CAP_SYS_ADMIN))
4828                 return -EPERM;
4829
4830         ret = mnt_want_write_file(file);
4831         if (ret)
4832                 return ret;
4833
4834         sa = memdup_user(arg, sizeof(*sa));
4835         if (IS_ERR(sa)) {
4836                 ret = PTR_ERR(sa);
4837                 goto drop_write;
4838         }
4839
4840         down_write(&fs_info->subvol_sem);
4841         trans = btrfs_start_transaction(fs_info->tree_root, 2);
4842         if (IS_ERR(trans)) {
4843                 ret = PTR_ERR(trans);
4844                 goto out;
4845         }
4846
4847         switch (sa->cmd) {
4848         case BTRFS_QUOTA_CTL_ENABLE:
4849                 ret = btrfs_quota_enable(trans, fs_info);
4850                 break;
4851         case BTRFS_QUOTA_CTL_DISABLE:
4852                 ret = btrfs_quota_disable(trans, fs_info);
4853                 break;
4854         default:
4855                 ret = -EINVAL;
4856                 break;
4857         }
4858
4859         err = btrfs_commit_transaction(trans);
4860         if (err && !ret)
4861                 ret = err;
4862 out:
4863         kfree(sa);
4864         up_write(&fs_info->subvol_sem);
4865 drop_write:
4866         mnt_drop_write_file(file);
4867         return ret;
4868 }
4869
4870 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4871 {
4872         struct inode *inode = file_inode(file);
4873         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4874         struct btrfs_root *root = BTRFS_I(inode)->root;
4875         struct btrfs_ioctl_qgroup_assign_args *sa;
4876         struct btrfs_trans_handle *trans;
4877         int ret;
4878         int err;
4879
4880         if (!capable(CAP_SYS_ADMIN))
4881                 return -EPERM;
4882
4883         ret = mnt_want_write_file(file);
4884         if (ret)
4885                 return ret;
4886
4887         sa = memdup_user(arg, sizeof(*sa));
4888         if (IS_ERR(sa)) {
4889                 ret = PTR_ERR(sa);
4890                 goto drop_write;
4891         }
4892
4893         trans = btrfs_join_transaction(root);
4894         if (IS_ERR(trans)) {
4895                 ret = PTR_ERR(trans);
4896                 goto out;
4897         }
4898
4899         if (sa->assign) {
4900                 ret = btrfs_add_qgroup_relation(trans, fs_info,
4901                                                 sa->src, sa->dst);
4902         } else {
4903                 ret = btrfs_del_qgroup_relation(trans, fs_info,
4904                                                 sa->src, sa->dst);
4905         }
4906
4907         /* update qgroup status and info */
4908         err = btrfs_run_qgroups(trans, fs_info);
4909         if (err < 0)
4910                 btrfs_handle_fs_error(fs_info, err,
4911                                       "failed to update qgroup status and info");
4912         err = btrfs_end_transaction(trans);
4913         if (err && !ret)
4914                 ret = err;
4915
4916 out:
4917         kfree(sa);
4918 drop_write:
4919         mnt_drop_write_file(file);
4920         return ret;
4921 }
4922
4923 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4924 {
4925         struct inode *inode = file_inode(file);
4926         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4927         struct btrfs_root *root = BTRFS_I(inode)->root;
4928         struct btrfs_ioctl_qgroup_create_args *sa;
4929         struct btrfs_trans_handle *trans;
4930         int ret;
4931         int err;
4932
4933         if (!capable(CAP_SYS_ADMIN))
4934                 return -EPERM;
4935
4936         ret = mnt_want_write_file(file);
4937         if (ret)
4938                 return ret;
4939
4940         sa = memdup_user(arg, sizeof(*sa));
4941         if (IS_ERR(sa)) {
4942                 ret = PTR_ERR(sa);
4943                 goto drop_write;
4944         }
4945
4946         if (!sa->qgroupid) {
4947                 ret = -EINVAL;
4948                 goto out;
4949         }
4950
4951         trans = btrfs_join_transaction(root);
4952         if (IS_ERR(trans)) {
4953                 ret = PTR_ERR(trans);
4954                 goto out;
4955         }
4956
4957         if (sa->create) {
4958                 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid);
4959         } else {
4960                 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid);
4961         }
4962
4963         err = btrfs_end_transaction(trans);
4964         if (err && !ret)
4965                 ret = err;
4966
4967 out:
4968         kfree(sa);
4969 drop_write:
4970         mnt_drop_write_file(file);
4971         return ret;
4972 }
4973
4974 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4975 {
4976         struct inode *inode = file_inode(file);
4977         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4978         struct btrfs_root *root = BTRFS_I(inode)->root;
4979         struct btrfs_ioctl_qgroup_limit_args *sa;
4980         struct btrfs_trans_handle *trans;
4981         int ret;
4982         int err;
4983         u64 qgroupid;
4984
4985         if (!capable(CAP_SYS_ADMIN))
4986                 return -EPERM;
4987
4988         ret = mnt_want_write_file(file);
4989         if (ret)
4990                 return ret;
4991
4992         sa = memdup_user(arg, sizeof(*sa));
4993         if (IS_ERR(sa)) {
4994                 ret = PTR_ERR(sa);
4995                 goto drop_write;
4996         }
4997
4998         trans = btrfs_join_transaction(root);
4999         if (IS_ERR(trans)) {
5000                 ret = PTR_ERR(trans);
5001                 goto out;
5002         }
5003
5004         qgroupid = sa->qgroupid;
5005         if (!qgroupid) {
5006                 /* take the current subvol as qgroup */
5007                 qgroupid = root->root_key.objectid;
5008         }
5009
5010         ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim);
5011
5012         err = btrfs_end_transaction(trans);
5013         if (err && !ret)
5014                 ret = err;
5015
5016 out:
5017         kfree(sa);
5018 drop_write:
5019         mnt_drop_write_file(file);
5020         return ret;
5021 }
5022
5023 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5024 {
5025         struct inode *inode = file_inode(file);
5026         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5027         struct btrfs_ioctl_quota_rescan_args *qsa;
5028         int ret;
5029
5030         if (!capable(CAP_SYS_ADMIN))
5031                 return -EPERM;
5032
5033         ret = mnt_want_write_file(file);
5034         if (ret)
5035                 return ret;
5036
5037         qsa = memdup_user(arg, sizeof(*qsa));
5038         if (IS_ERR(qsa)) {
5039                 ret = PTR_ERR(qsa);
5040                 goto drop_write;
5041         }
5042
5043         if (qsa->flags) {
5044                 ret = -EINVAL;
5045                 goto out;
5046         }
5047
5048         ret = btrfs_qgroup_rescan(fs_info);
5049
5050 out:
5051         kfree(qsa);
5052 drop_write:
5053         mnt_drop_write_file(file);
5054         return ret;
5055 }
5056
5057 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5058 {
5059         struct inode *inode = file_inode(file);
5060         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5061         struct btrfs_ioctl_quota_rescan_args *qsa;
5062         int ret = 0;
5063
5064         if (!capable(CAP_SYS_ADMIN))
5065                 return -EPERM;
5066
5067         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5068         if (!qsa)
5069                 return -ENOMEM;
5070
5071         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5072                 qsa->flags = 1;
5073                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5074         }
5075
5076         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5077                 ret = -EFAULT;
5078
5079         kfree(qsa);
5080         return ret;
5081 }
5082
5083 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5084 {
5085         struct inode *inode = file_inode(file);
5086         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5087
5088         if (!capable(CAP_SYS_ADMIN))
5089                 return -EPERM;
5090
5091         return btrfs_qgroup_wait_for_completion(fs_info, true);
5092 }
5093
5094 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5095                                             struct btrfs_ioctl_received_subvol_args *sa)
5096 {
5097         struct inode *inode = file_inode(file);
5098         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5099         struct btrfs_root *root = BTRFS_I(inode)->root;
5100         struct btrfs_root_item *root_item = &root->root_item;
5101         struct btrfs_trans_handle *trans;
5102         struct timespec ct = current_time(inode);
5103         int ret = 0;
5104         int received_uuid_changed;
5105
5106         if (!inode_owner_or_capable(inode))
5107                 return -EPERM;
5108
5109         ret = mnt_want_write_file(file);
5110         if (ret < 0)
5111                 return ret;
5112
5113         down_write(&fs_info->subvol_sem);
5114
5115         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5116                 ret = -EINVAL;
5117                 goto out;
5118         }
5119
5120         if (btrfs_root_readonly(root)) {
5121                 ret = -EROFS;
5122                 goto out;
5123         }
5124
5125         /*
5126          * 1 - root item
5127          * 2 - uuid items (received uuid + subvol uuid)
5128          */
5129         trans = btrfs_start_transaction(root, 3);
5130         if (IS_ERR(trans)) {
5131                 ret = PTR_ERR(trans);
5132                 trans = NULL;
5133                 goto out;
5134         }
5135
5136         sa->rtransid = trans->transid;
5137         sa->rtime.sec = ct.tv_sec;
5138         sa->rtime.nsec = ct.tv_nsec;
5139
5140         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5141                                        BTRFS_UUID_SIZE);
5142         if (received_uuid_changed &&
5143             !btrfs_is_empty_uuid(root_item->received_uuid))
5144                 btrfs_uuid_tree_rem(trans, fs_info, root_item->received_uuid,
5145                                     BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5146                                     root->root_key.objectid);
5147         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5148         btrfs_set_root_stransid(root_item, sa->stransid);
5149         btrfs_set_root_rtransid(root_item, sa->rtransid);
5150         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5151         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5152         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5153         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5154
5155         ret = btrfs_update_root(trans, fs_info->tree_root,
5156                                 &root->root_key, &root->root_item);
5157         if (ret < 0) {
5158                 btrfs_end_transaction(trans);
5159                 goto out;
5160         }
5161         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5162                 ret = btrfs_uuid_tree_add(trans, fs_info, sa->uuid,
5163                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5164                                           root->root_key.objectid);
5165                 if (ret < 0 && ret != -EEXIST) {
5166                         btrfs_abort_transaction(trans, ret);
5167                         btrfs_end_transaction(trans);
5168                         goto out;
5169                 }
5170         }
5171         ret = btrfs_commit_transaction(trans);
5172 out:
5173         up_write(&fs_info->subvol_sem);
5174         mnt_drop_write_file(file);
5175         return ret;
5176 }
5177
5178 #ifdef CONFIG_64BIT
5179 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5180                                                 void __user *arg)
5181 {
5182         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5183         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5184         int ret = 0;
5185
5186         args32 = memdup_user(arg, sizeof(*args32));
5187         if (IS_ERR(args32))
5188                 return PTR_ERR(args32);
5189
5190         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5191         if (!args64) {
5192                 ret = -ENOMEM;
5193                 goto out;
5194         }
5195
5196         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5197         args64->stransid = args32->stransid;
5198         args64->rtransid = args32->rtransid;
5199         args64->stime.sec = args32->stime.sec;
5200         args64->stime.nsec = args32->stime.nsec;
5201         args64->rtime.sec = args32->rtime.sec;
5202         args64->rtime.nsec = args32->rtime.nsec;
5203         args64->flags = args32->flags;
5204
5205         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5206         if (ret)
5207                 goto out;
5208
5209         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5210         args32->stransid = args64->stransid;
5211         args32->rtransid = args64->rtransid;
5212         args32->stime.sec = args64->stime.sec;
5213         args32->stime.nsec = args64->stime.nsec;
5214         args32->rtime.sec = args64->rtime.sec;
5215         args32->rtime.nsec = args64->rtime.nsec;
5216         args32->flags = args64->flags;
5217
5218         ret = copy_to_user(arg, args32, sizeof(*args32));
5219         if (ret)
5220                 ret = -EFAULT;
5221
5222 out:
5223         kfree(args32);
5224         kfree(args64);
5225         return ret;
5226 }
5227 #endif
5228
5229 static long btrfs_ioctl_set_received_subvol(struct file *file,
5230                                             void __user *arg)
5231 {
5232         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5233         int ret = 0;
5234
5235         sa = memdup_user(arg, sizeof(*sa));
5236         if (IS_ERR(sa))
5237                 return PTR_ERR(sa);
5238
5239         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5240
5241         if (ret)
5242                 goto out;
5243
5244         ret = copy_to_user(arg, sa, sizeof(*sa));
5245         if (ret)
5246                 ret = -EFAULT;
5247
5248 out:
5249         kfree(sa);
5250         return ret;
5251 }
5252
5253 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5254 {
5255         struct inode *inode = file_inode(file);
5256         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5257         size_t len;
5258         int ret;
5259         char label[BTRFS_LABEL_SIZE];
5260
5261         spin_lock(&fs_info->super_lock);
5262         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5263         spin_unlock(&fs_info->super_lock);
5264
5265         len = strnlen(label, BTRFS_LABEL_SIZE);
5266
5267         if (len == BTRFS_LABEL_SIZE) {
5268                 btrfs_warn(fs_info,
5269                            "label is too long, return the first %zu bytes",
5270                            --len);
5271         }
5272
5273         ret = copy_to_user(arg, label, len);
5274
5275         return ret ? -EFAULT : 0;
5276 }
5277
5278 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5279 {
5280         struct inode *inode = file_inode(file);
5281         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5282         struct btrfs_root *root = BTRFS_I(inode)->root;
5283         struct btrfs_super_block *super_block = fs_info->super_copy;
5284         struct btrfs_trans_handle *trans;
5285         char label[BTRFS_LABEL_SIZE];
5286         int ret;
5287
5288         if (!capable(CAP_SYS_ADMIN))
5289                 return -EPERM;
5290
5291         if (copy_from_user(label, arg, sizeof(label)))
5292                 return -EFAULT;
5293
5294         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5295                 btrfs_err(fs_info,
5296                           "unable to set label with more than %d bytes",
5297                           BTRFS_LABEL_SIZE - 1);
5298                 return -EINVAL;
5299         }
5300
5301         ret = mnt_want_write_file(file);
5302         if (ret)
5303                 return ret;
5304
5305         trans = btrfs_start_transaction(root, 0);
5306         if (IS_ERR(trans)) {
5307                 ret = PTR_ERR(trans);
5308                 goto out_unlock;
5309         }
5310
5311         spin_lock(&fs_info->super_lock);
5312         strcpy(super_block->label, label);
5313         spin_unlock(&fs_info->super_lock);
5314         ret = btrfs_commit_transaction(trans);
5315
5316 out_unlock:
5317         mnt_drop_write_file(file);
5318         return ret;
5319 }
5320
5321 #define INIT_FEATURE_FLAGS(suffix) \
5322         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5323           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5324           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5325
5326 int btrfs_ioctl_get_supported_features(void __user *arg)
5327 {
5328         static const struct btrfs_ioctl_feature_flags features[3] = {
5329                 INIT_FEATURE_FLAGS(SUPP),
5330                 INIT_FEATURE_FLAGS(SAFE_SET),
5331                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5332         };
5333
5334         if (copy_to_user(arg, &features, sizeof(features)))
5335                 return -EFAULT;
5336
5337         return 0;
5338 }
5339
5340 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5341 {
5342         struct inode *inode = file_inode(file);
5343         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5344         struct btrfs_super_block *super_block = fs_info->super_copy;
5345         struct btrfs_ioctl_feature_flags features;
5346
5347         features.compat_flags = btrfs_super_compat_flags(super_block);
5348         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5349         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5350
5351         if (copy_to_user(arg, &features, sizeof(features)))
5352                 return -EFAULT;
5353
5354         return 0;
5355 }
5356
5357 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5358                               enum btrfs_feature_set set,
5359                               u64 change_mask, u64 flags, u64 supported_flags,
5360                               u64 safe_set, u64 safe_clear)
5361 {
5362         const char *type = btrfs_feature_set_names[set];
5363         char *names;
5364         u64 disallowed, unsupported;
5365         u64 set_mask = flags & change_mask;
5366         u64 clear_mask = ~flags & change_mask;
5367
5368         unsupported = set_mask & ~supported_flags;
5369         if (unsupported) {
5370                 names = btrfs_printable_features(set, unsupported);
5371                 if (names) {
5372                         btrfs_warn(fs_info,
5373                                    "this kernel does not support the %s feature bit%s",
5374                                    names, strchr(names, ',') ? "s" : "");
5375                         kfree(names);
5376                 } else
5377                         btrfs_warn(fs_info,
5378                                    "this kernel does not support %s bits 0x%llx",
5379                                    type, unsupported);
5380                 return -EOPNOTSUPP;
5381         }
5382
5383         disallowed = set_mask & ~safe_set;
5384         if (disallowed) {
5385                 names = btrfs_printable_features(set, disallowed);
5386                 if (names) {
5387                         btrfs_warn(fs_info,
5388                                    "can't set the %s feature bit%s while mounted",
5389                                    names, strchr(names, ',') ? "s" : "");
5390                         kfree(names);
5391                 } else
5392                         btrfs_warn(fs_info,
5393                                    "can't set %s bits 0x%llx while mounted",
5394                                    type, disallowed);
5395                 return -EPERM;
5396         }
5397
5398         disallowed = clear_mask & ~safe_clear;
5399         if (disallowed) {
5400                 names = btrfs_printable_features(set, disallowed);
5401                 if (names) {
5402                         btrfs_warn(fs_info,
5403                                    "can't clear the %s feature bit%s while mounted",
5404                                    names, strchr(names, ',') ? "s" : "");
5405                         kfree(names);
5406                 } else
5407                         btrfs_warn(fs_info,
5408                                    "can't clear %s bits 0x%llx while mounted",
5409                                    type, disallowed);
5410                 return -EPERM;
5411         }
5412
5413         return 0;
5414 }
5415
5416 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5417 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5418                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5419                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5420                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5421
5422 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5423 {
5424         struct inode *inode = file_inode(file);
5425         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5426         struct btrfs_root *root = BTRFS_I(inode)->root;
5427         struct btrfs_super_block *super_block = fs_info->super_copy;
5428         struct btrfs_ioctl_feature_flags flags[2];
5429         struct btrfs_trans_handle *trans;
5430         u64 newflags;
5431         int ret;
5432
5433         if (!capable(CAP_SYS_ADMIN))
5434                 return -EPERM;
5435
5436         if (copy_from_user(flags, arg, sizeof(flags)))
5437                 return -EFAULT;
5438
5439         /* Nothing to do */
5440         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5441             !flags[0].incompat_flags)
5442                 return 0;
5443
5444         ret = check_feature(fs_info, flags[0].compat_flags,
5445                             flags[1].compat_flags, COMPAT);
5446         if (ret)
5447                 return ret;
5448
5449         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5450                             flags[1].compat_ro_flags, COMPAT_RO);
5451         if (ret)
5452                 return ret;
5453
5454         ret = check_feature(fs_info, flags[0].incompat_flags,
5455                             flags[1].incompat_flags, INCOMPAT);
5456         if (ret)
5457                 return ret;
5458
5459         ret = mnt_want_write_file(file);
5460         if (ret)
5461                 return ret;
5462
5463         trans = btrfs_start_transaction(root, 0);
5464         if (IS_ERR(trans)) {
5465                 ret = PTR_ERR(trans);
5466                 goto out_drop_write;
5467         }
5468
5469         spin_lock(&fs_info->super_lock);
5470         newflags = btrfs_super_compat_flags(super_block);
5471         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5472         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5473         btrfs_set_super_compat_flags(super_block, newflags);
5474
5475         newflags = btrfs_super_compat_ro_flags(super_block);
5476         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5477         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5478         btrfs_set_super_compat_ro_flags(super_block, newflags);
5479
5480         newflags = btrfs_super_incompat_flags(super_block);
5481         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5482         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5483         btrfs_set_super_incompat_flags(super_block, newflags);
5484         spin_unlock(&fs_info->super_lock);
5485
5486         ret = btrfs_commit_transaction(trans);
5487 out_drop_write:
5488         mnt_drop_write_file(file);
5489
5490         return ret;
5491 }
5492
5493 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5494 {
5495         struct btrfs_ioctl_send_args *arg;
5496         int ret;
5497
5498         if (compat) {
5499 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5500                 struct btrfs_ioctl_send_args_32 args32;
5501
5502                 ret = copy_from_user(&args32, argp, sizeof(args32));
5503                 if (ret)
5504                         return -EFAULT;
5505                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5506                 if (!arg)
5507                         return -ENOMEM;
5508                 arg->send_fd = args32.send_fd;
5509                 arg->clone_sources_count = args32.clone_sources_count;
5510                 arg->clone_sources = compat_ptr(args32.clone_sources);
5511                 arg->parent_root = args32.parent_root;
5512                 arg->flags = args32.flags;
5513                 memcpy(arg->reserved, args32.reserved,
5514                        sizeof(args32.reserved));
5515 #else
5516                 return -ENOTTY;
5517 #endif
5518         } else {
5519                 arg = memdup_user(argp, sizeof(*arg));
5520                 if (IS_ERR(arg))
5521                         return PTR_ERR(arg);
5522         }
5523         ret = btrfs_ioctl_send(file, arg);
5524         kfree(arg);
5525         return ret;
5526 }
5527
5528 long btrfs_ioctl(struct file *file, unsigned int
5529                 cmd, unsigned long arg)
5530 {
5531         struct inode *inode = file_inode(file);
5532         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5533         struct btrfs_root *root = BTRFS_I(inode)->root;
5534         void __user *argp = (void __user *)arg;
5535
5536         switch (cmd) {
5537         case FS_IOC_GETFLAGS:
5538                 return btrfs_ioctl_getflags(file, argp);
5539         case FS_IOC_SETFLAGS:
5540                 return btrfs_ioctl_setflags(file, argp);
5541         case FS_IOC_GETVERSION:
5542                 return btrfs_ioctl_getversion(file, argp);
5543         case FITRIM:
5544                 return btrfs_ioctl_fitrim(file, argp);
5545         case BTRFS_IOC_SNAP_CREATE:
5546                 return btrfs_ioctl_snap_create(file, argp, 0);
5547         case BTRFS_IOC_SNAP_CREATE_V2:
5548                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5549         case BTRFS_IOC_SUBVOL_CREATE:
5550                 return btrfs_ioctl_snap_create(file, argp, 1);
5551         case BTRFS_IOC_SUBVOL_CREATE_V2:
5552                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5553         case BTRFS_IOC_SNAP_DESTROY:
5554                 return btrfs_ioctl_snap_destroy(file, argp);
5555         case BTRFS_IOC_SUBVOL_GETFLAGS:
5556                 return btrfs_ioctl_subvol_getflags(file, argp);
5557         case BTRFS_IOC_SUBVOL_SETFLAGS:
5558                 return btrfs_ioctl_subvol_setflags(file, argp);
5559         case BTRFS_IOC_DEFAULT_SUBVOL:
5560                 return btrfs_ioctl_default_subvol(file, argp);
5561         case BTRFS_IOC_DEFRAG:
5562                 return btrfs_ioctl_defrag(file, NULL);
5563         case BTRFS_IOC_DEFRAG_RANGE:
5564                 return btrfs_ioctl_defrag(file, argp);
5565         case BTRFS_IOC_RESIZE:
5566                 return btrfs_ioctl_resize(file, argp);
5567         case BTRFS_IOC_ADD_DEV:
5568                 return btrfs_ioctl_add_dev(fs_info, argp);
5569         case BTRFS_IOC_RM_DEV:
5570                 return btrfs_ioctl_rm_dev(file, argp);
5571         case BTRFS_IOC_RM_DEV_V2:
5572                 return btrfs_ioctl_rm_dev_v2(file, argp);
5573         case BTRFS_IOC_FS_INFO:
5574                 return btrfs_ioctl_fs_info(fs_info, argp);
5575         case BTRFS_IOC_DEV_INFO:
5576                 return btrfs_ioctl_dev_info(fs_info, argp);
5577         case BTRFS_IOC_BALANCE:
5578                 return btrfs_ioctl_balance(file, NULL);
5579         case BTRFS_IOC_TRANS_START:
5580                 return btrfs_ioctl_trans_start(file);
5581         case BTRFS_IOC_TRANS_END:
5582                 return btrfs_ioctl_trans_end(file);
5583         case BTRFS_IOC_TREE_SEARCH:
5584                 return btrfs_ioctl_tree_search(file, argp);
5585         case BTRFS_IOC_TREE_SEARCH_V2:
5586                 return btrfs_ioctl_tree_search_v2(file, argp);
5587         case BTRFS_IOC_INO_LOOKUP:
5588                 return btrfs_ioctl_ino_lookup(file, argp);
5589         case BTRFS_IOC_INO_PATHS:
5590                 return btrfs_ioctl_ino_to_path(root, argp);
5591         case BTRFS_IOC_LOGICAL_INO:
5592                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5593         case BTRFS_IOC_LOGICAL_INO_V2:
5594                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5595         case BTRFS_IOC_SPACE_INFO:
5596                 return btrfs_ioctl_space_info(fs_info, argp);
5597         case BTRFS_IOC_SYNC: {
5598                 int ret;
5599
5600                 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
5601                 if (ret)
5602                         return ret;
5603                 ret = btrfs_sync_fs(inode->i_sb, 1);
5604                 /*
5605                  * The transaction thread may want to do more work,
5606                  * namely it pokes the cleaner kthread that will start
5607                  * processing uncleaned subvols.
5608                  */
5609                 wake_up_process(fs_info->transaction_kthread);
5610                 return ret;
5611         }
5612         case BTRFS_IOC_START_SYNC:
5613                 return btrfs_ioctl_start_sync(root, argp);
5614         case BTRFS_IOC_WAIT_SYNC:
5615                 return btrfs_ioctl_wait_sync(fs_info, argp);
5616         case BTRFS_IOC_SCRUB:
5617                 return btrfs_ioctl_scrub(file, argp);
5618         case BTRFS_IOC_SCRUB_CANCEL:
5619                 return btrfs_ioctl_scrub_cancel(fs_info);
5620         case BTRFS_IOC_SCRUB_PROGRESS:
5621                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5622         case BTRFS_IOC_BALANCE_V2:
5623                 return btrfs_ioctl_balance(file, argp);
5624         case BTRFS_IOC_BALANCE_CTL:
5625                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5626         case BTRFS_IOC_BALANCE_PROGRESS:
5627                 return btrfs_ioctl_balance_progress(fs_info, argp);
5628         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5629                 return btrfs_ioctl_set_received_subvol(file, argp);
5630 #ifdef CONFIG_64BIT
5631         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5632                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5633 #endif
5634         case BTRFS_IOC_SEND:
5635                 return _btrfs_ioctl_send(file, argp, false);
5636 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5637         case BTRFS_IOC_SEND_32:
5638                 return _btrfs_ioctl_send(file, argp, true);
5639 #endif
5640         case BTRFS_IOC_GET_DEV_STATS:
5641                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5642         case BTRFS_IOC_QUOTA_CTL:
5643                 return btrfs_ioctl_quota_ctl(file, argp);
5644         case BTRFS_IOC_QGROUP_ASSIGN:
5645                 return btrfs_ioctl_qgroup_assign(file, argp);
5646         case BTRFS_IOC_QGROUP_CREATE:
5647                 return btrfs_ioctl_qgroup_create(file, argp);
5648         case BTRFS_IOC_QGROUP_LIMIT:
5649                 return btrfs_ioctl_qgroup_limit(file, argp);
5650         case BTRFS_IOC_QUOTA_RESCAN:
5651                 return btrfs_ioctl_quota_rescan(file, argp);
5652         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5653                 return btrfs_ioctl_quota_rescan_status(file, argp);
5654         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5655                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5656         case BTRFS_IOC_DEV_REPLACE:
5657                 return btrfs_ioctl_dev_replace(fs_info, argp);
5658         case BTRFS_IOC_GET_FSLABEL:
5659                 return btrfs_ioctl_get_fslabel(file, argp);
5660         case BTRFS_IOC_SET_FSLABEL:
5661                 return btrfs_ioctl_set_fslabel(file, argp);
5662         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5663                 return btrfs_ioctl_get_supported_features(argp);
5664         case BTRFS_IOC_GET_FEATURES:
5665                 return btrfs_ioctl_get_features(file, argp);
5666         case BTRFS_IOC_SET_FEATURES:
5667                 return btrfs_ioctl_set_features(file, argp);
5668         }
5669
5670         return -ENOTTY;
5671 }
5672
5673 #ifdef CONFIG_COMPAT
5674 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5675 {
5676         /*
5677          * These all access 32-bit values anyway so no further
5678          * handling is necessary.
5679          */
5680         switch (cmd) {
5681         case FS_IOC32_GETFLAGS:
5682                 cmd = FS_IOC_GETFLAGS;
5683                 break;
5684         case FS_IOC32_SETFLAGS:
5685                 cmd = FS_IOC_SETFLAGS;
5686                 break;
5687         case FS_IOC32_GETVERSION:
5688                 cmd = FS_IOC_GETVERSION;
5689                 break;
5690         }
5691
5692         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5693 }
5694 #endif