1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #include <linux/kernel.h> 7 #include <linux/bio.h> 8 #include <linux/buffer_head.h> 9 #include <linux/file.h> 10 #include <linux/fs.h> 11 #include <linux/fsnotify.h> 12 #include <linux/pagemap.h> 13 #include <linux/highmem.h> 14 #include <linux/time.h> 15 #include <linux/init.h> 16 #include <linux/string.h> 17 #include <linux/backing-dev.h> 18 #include <linux/mount.h> 19 #include <linux/mpage.h> 20 #include <linux/namei.h> 21 #include <linux/swap.h> 22 #include <linux/writeback.h> 23 #include <linux/compat.h> 24 #include <linux/bit_spinlock.h> 25 #include <linux/security.h> 26 #include <linux/xattr.h> 27 #include <linux/mm.h> 28 #include <linux/slab.h> 29 #include <linux/blkdev.h> 30 #include <linux/uuid.h> 31 #include <linux/btrfs.h> 32 #include <linux/uaccess.h> 33 #include <linux/iversion.h> 34 #include "ctree.h" 35 #include "disk-io.h" 36 #include "transaction.h" 37 #include "btrfs_inode.h" 38 #include "print-tree.h" 39 #include "volumes.h" 40 #include "locking.h" 41 #include "inode-map.h" 42 #include "backref.h" 43 #include "rcu-string.h" 44 #include "send.h" 45 #include "dev-replace.h" 46 #include "props.h" 47 #include "sysfs.h" 48 #include "qgroup.h" 49 #include "tree-log.h" 50 #include "compression.h" 51 52 #ifdef CONFIG_64BIT 53 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI 54 * structures are incorrect, as the timespec structure from userspace 55 * is 4 bytes too small. We define these alternatives here to teach 56 * the kernel about the 32-bit struct packing. 57 */ 58 struct btrfs_ioctl_timespec_32 { 59 __u64 sec; 60 __u32 nsec; 61 } __attribute__ ((__packed__)); 62 63 struct btrfs_ioctl_received_subvol_args_32 { 64 char uuid[BTRFS_UUID_SIZE]; /* in */ 65 __u64 stransid; /* in */ 66 __u64 rtransid; /* out */ 67 struct btrfs_ioctl_timespec_32 stime; /* in */ 68 struct btrfs_ioctl_timespec_32 rtime; /* out */ 69 __u64 flags; /* in */ 70 __u64 reserved[16]; /* in */ 71 } __attribute__ ((__packed__)); 72 73 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \ 74 struct btrfs_ioctl_received_subvol_args_32) 75 #endif 76 77 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 78 struct btrfs_ioctl_send_args_32 { 79 __s64 send_fd; /* in */ 80 __u64 clone_sources_count; /* in */ 81 compat_uptr_t clone_sources; /* in */ 82 __u64 parent_root; /* in */ 83 __u64 flags; /* in */ 84 __u64 reserved[4]; /* in */ 85 } __attribute__ ((__packed__)); 86 87 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \ 88 struct btrfs_ioctl_send_args_32) 89 #endif 90 91 static int btrfs_clone(struct inode *src, struct inode *inode, 92 u64 off, u64 olen, u64 olen_aligned, u64 destoff, 93 int no_time_update); 94 95 /* Mask out flags that are inappropriate for the given type of inode. */ 96 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode, 97 unsigned int flags) 98 { 99 if (S_ISDIR(inode->i_mode)) 100 return flags; 101 else if (S_ISREG(inode->i_mode)) 102 return flags & ~FS_DIRSYNC_FL; 103 else 104 return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 105 } 106 107 /* 108 * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS 109 * ioctl. 110 */ 111 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags) 112 { 113 unsigned int iflags = 0; 114 115 if (flags & BTRFS_INODE_SYNC) 116 iflags |= FS_SYNC_FL; 117 if (flags & BTRFS_INODE_IMMUTABLE) 118 iflags |= FS_IMMUTABLE_FL; 119 if (flags & BTRFS_INODE_APPEND) 120 iflags |= FS_APPEND_FL; 121 if (flags & BTRFS_INODE_NODUMP) 122 iflags |= FS_NODUMP_FL; 123 if (flags & BTRFS_INODE_NOATIME) 124 iflags |= FS_NOATIME_FL; 125 if (flags & BTRFS_INODE_DIRSYNC) 126 iflags |= FS_DIRSYNC_FL; 127 if (flags & BTRFS_INODE_NODATACOW) 128 iflags |= FS_NOCOW_FL; 129 130 if (flags & BTRFS_INODE_NOCOMPRESS) 131 iflags |= FS_NOCOMP_FL; 132 else if (flags & BTRFS_INODE_COMPRESS) 133 iflags |= FS_COMPR_FL; 134 135 return iflags; 136 } 137 138 /* 139 * Update inode->i_flags based on the btrfs internal flags. 140 */ 141 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode) 142 { 143 struct btrfs_inode *binode = BTRFS_I(inode); 144 unsigned int new_fl = 0; 145 146 if (binode->flags & BTRFS_INODE_SYNC) 147 new_fl |= S_SYNC; 148 if (binode->flags & BTRFS_INODE_IMMUTABLE) 149 new_fl |= S_IMMUTABLE; 150 if (binode->flags & BTRFS_INODE_APPEND) 151 new_fl |= S_APPEND; 152 if (binode->flags & BTRFS_INODE_NOATIME) 153 new_fl |= S_NOATIME; 154 if (binode->flags & BTRFS_INODE_DIRSYNC) 155 new_fl |= S_DIRSYNC; 156 157 set_mask_bits(&inode->i_flags, 158 S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC, 159 new_fl); 160 } 161 162 static int btrfs_ioctl_getflags(struct file *file, void __user *arg) 163 { 164 struct btrfs_inode *binode = BTRFS_I(file_inode(file)); 165 unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags); 166 167 if (copy_to_user(arg, &flags, sizeof(flags))) 168 return -EFAULT; 169 return 0; 170 } 171 172 /* Check if @flags are a supported and valid set of FS_*_FL flags */ 173 static int check_fsflags(unsigned int flags) 174 { 175 if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 176 FS_NOATIME_FL | FS_NODUMP_FL | \ 177 FS_SYNC_FL | FS_DIRSYNC_FL | \ 178 FS_NOCOMP_FL | FS_COMPR_FL | 179 FS_NOCOW_FL)) 180 return -EOPNOTSUPP; 181 182 if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 183 return -EINVAL; 184 185 return 0; 186 } 187 188 static int btrfs_ioctl_setflags(struct file *file, void __user *arg) 189 { 190 struct inode *inode = file_inode(file); 191 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 192 struct btrfs_inode *binode = BTRFS_I(inode); 193 struct btrfs_root *root = binode->root; 194 struct btrfs_trans_handle *trans; 195 unsigned int fsflags, old_fsflags; 196 int ret; 197 u64 old_flags; 198 unsigned int old_i_flags; 199 umode_t mode; 200 201 if (!inode_owner_or_capable(inode)) 202 return -EPERM; 203 204 if (btrfs_root_readonly(root)) 205 return -EROFS; 206 207 if (copy_from_user(&fsflags, arg, sizeof(fsflags))) 208 return -EFAULT; 209 210 ret = check_fsflags(fsflags); 211 if (ret) 212 return ret; 213 214 ret = mnt_want_write_file(file); 215 if (ret) 216 return ret; 217 218 inode_lock(inode); 219 220 old_flags = binode->flags; 221 old_i_flags = inode->i_flags; 222 mode = inode->i_mode; 223 224 fsflags = btrfs_mask_fsflags_for_type(inode, fsflags); 225 old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags); 226 if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 227 if (!capable(CAP_LINUX_IMMUTABLE)) { 228 ret = -EPERM; 229 goto out_unlock; 230 } 231 } 232 233 if (fsflags & FS_SYNC_FL) 234 binode->flags |= BTRFS_INODE_SYNC; 235 else 236 binode->flags &= ~BTRFS_INODE_SYNC; 237 if (fsflags & FS_IMMUTABLE_FL) 238 binode->flags |= BTRFS_INODE_IMMUTABLE; 239 else 240 binode->flags &= ~BTRFS_INODE_IMMUTABLE; 241 if (fsflags & FS_APPEND_FL) 242 binode->flags |= BTRFS_INODE_APPEND; 243 else 244 binode->flags &= ~BTRFS_INODE_APPEND; 245 if (fsflags & FS_NODUMP_FL) 246 binode->flags |= BTRFS_INODE_NODUMP; 247 else 248 binode->flags &= ~BTRFS_INODE_NODUMP; 249 if (fsflags & FS_NOATIME_FL) 250 binode->flags |= BTRFS_INODE_NOATIME; 251 else 252 binode->flags &= ~BTRFS_INODE_NOATIME; 253 if (fsflags & FS_DIRSYNC_FL) 254 binode->flags |= BTRFS_INODE_DIRSYNC; 255 else 256 binode->flags &= ~BTRFS_INODE_DIRSYNC; 257 if (fsflags & FS_NOCOW_FL) { 258 if (S_ISREG(mode)) { 259 /* 260 * It's safe to turn csums off here, no extents exist. 261 * Otherwise we want the flag to reflect the real COW 262 * status of the file and will not set it. 263 */ 264 if (inode->i_size == 0) 265 binode->flags |= BTRFS_INODE_NODATACOW 266 | BTRFS_INODE_NODATASUM; 267 } else { 268 binode->flags |= BTRFS_INODE_NODATACOW; 269 } 270 } else { 271 /* 272 * Revert back under same assumptions as above 273 */ 274 if (S_ISREG(mode)) { 275 if (inode->i_size == 0) 276 binode->flags &= ~(BTRFS_INODE_NODATACOW 277 | BTRFS_INODE_NODATASUM); 278 } else { 279 binode->flags &= ~BTRFS_INODE_NODATACOW; 280 } 281 } 282 283 /* 284 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 285 * flag may be changed automatically if compression code won't make 286 * things smaller. 287 */ 288 if (fsflags & FS_NOCOMP_FL) { 289 binode->flags &= ~BTRFS_INODE_COMPRESS; 290 binode->flags |= BTRFS_INODE_NOCOMPRESS; 291 292 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0); 293 if (ret && ret != -ENODATA) 294 goto out_drop; 295 } else if (fsflags & FS_COMPR_FL) { 296 const char *comp; 297 298 binode->flags |= BTRFS_INODE_COMPRESS; 299 binode->flags &= ~BTRFS_INODE_NOCOMPRESS; 300 301 comp = btrfs_compress_type2str(fs_info->compress_type); 302 if (!comp || comp[0] == 0) 303 comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB); 304 305 ret = btrfs_set_prop(inode, "btrfs.compression", 306 comp, strlen(comp), 0); 307 if (ret) 308 goto out_drop; 309 310 } else { 311 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0); 312 if (ret && ret != -ENODATA) 313 goto out_drop; 314 binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 315 } 316 317 trans = btrfs_start_transaction(root, 1); 318 if (IS_ERR(trans)) { 319 ret = PTR_ERR(trans); 320 goto out_drop; 321 } 322 323 btrfs_sync_inode_flags_to_i_flags(inode); 324 inode_inc_iversion(inode); 325 inode->i_ctime = current_time(inode); 326 ret = btrfs_update_inode(trans, root, inode); 327 328 btrfs_end_transaction(trans); 329 out_drop: 330 if (ret) { 331 binode->flags = old_flags; 332 inode->i_flags = old_i_flags; 333 } 334 335 out_unlock: 336 inode_unlock(inode); 337 mnt_drop_write_file(file); 338 return ret; 339 } 340 341 /* 342 * Translate btrfs internal inode flags to xflags as expected by the 343 * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are 344 * silently dropped. 345 */ 346 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags) 347 { 348 unsigned int xflags = 0; 349 350 if (flags & BTRFS_INODE_APPEND) 351 xflags |= FS_XFLAG_APPEND; 352 if (flags & BTRFS_INODE_IMMUTABLE) 353 xflags |= FS_XFLAG_IMMUTABLE; 354 if (flags & BTRFS_INODE_NOATIME) 355 xflags |= FS_XFLAG_NOATIME; 356 if (flags & BTRFS_INODE_NODUMP) 357 xflags |= FS_XFLAG_NODUMP; 358 if (flags & BTRFS_INODE_SYNC) 359 xflags |= FS_XFLAG_SYNC; 360 361 return xflags; 362 } 363 364 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */ 365 static int check_xflags(unsigned int flags) 366 { 367 if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME | 368 FS_XFLAG_NODUMP | FS_XFLAG_SYNC)) 369 return -EOPNOTSUPP; 370 return 0; 371 } 372 373 /* 374 * Set the xflags from the internal inode flags. The remaining items of fsxattr 375 * are zeroed. 376 */ 377 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg) 378 { 379 struct btrfs_inode *binode = BTRFS_I(file_inode(file)); 380 struct fsxattr fa; 381 382 memset(&fa, 0, sizeof(fa)); 383 fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags); 384 385 if (copy_to_user(arg, &fa, sizeof(fa))) 386 return -EFAULT; 387 388 return 0; 389 } 390 391 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg) 392 { 393 struct inode *inode = file_inode(file); 394 struct btrfs_inode *binode = BTRFS_I(inode); 395 struct btrfs_root *root = binode->root; 396 struct btrfs_trans_handle *trans; 397 struct fsxattr fa; 398 unsigned old_flags; 399 unsigned old_i_flags; 400 int ret = 0; 401 402 if (!inode_owner_or_capable(inode)) 403 return -EPERM; 404 405 if (btrfs_root_readonly(root)) 406 return -EROFS; 407 408 memset(&fa, 0, sizeof(fa)); 409 if (copy_from_user(&fa, arg, sizeof(fa))) 410 return -EFAULT; 411 412 ret = check_xflags(fa.fsx_xflags); 413 if (ret) 414 return ret; 415 416 if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0) 417 return -EOPNOTSUPP; 418 419 ret = mnt_want_write_file(file); 420 if (ret) 421 return ret; 422 423 inode_lock(inode); 424 425 old_flags = binode->flags; 426 old_i_flags = inode->i_flags; 427 428 /* We need the capabilities to change append-only or immutable inode */ 429 if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) || 430 (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) && 431 !capable(CAP_LINUX_IMMUTABLE)) { 432 ret = -EPERM; 433 goto out_unlock; 434 } 435 436 if (fa.fsx_xflags & FS_XFLAG_SYNC) 437 binode->flags |= BTRFS_INODE_SYNC; 438 else 439 binode->flags &= ~BTRFS_INODE_SYNC; 440 if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE) 441 binode->flags |= BTRFS_INODE_IMMUTABLE; 442 else 443 binode->flags &= ~BTRFS_INODE_IMMUTABLE; 444 if (fa.fsx_xflags & FS_XFLAG_APPEND) 445 binode->flags |= BTRFS_INODE_APPEND; 446 else 447 binode->flags &= ~BTRFS_INODE_APPEND; 448 if (fa.fsx_xflags & FS_XFLAG_NODUMP) 449 binode->flags |= BTRFS_INODE_NODUMP; 450 else 451 binode->flags &= ~BTRFS_INODE_NODUMP; 452 if (fa.fsx_xflags & FS_XFLAG_NOATIME) 453 binode->flags |= BTRFS_INODE_NOATIME; 454 else 455 binode->flags &= ~BTRFS_INODE_NOATIME; 456 457 /* 1 item for the inode */ 458 trans = btrfs_start_transaction(root, 1); 459 if (IS_ERR(trans)) { 460 ret = PTR_ERR(trans); 461 goto out_unlock; 462 } 463 464 btrfs_sync_inode_flags_to_i_flags(inode); 465 inode_inc_iversion(inode); 466 inode->i_ctime = current_time(inode); 467 ret = btrfs_update_inode(trans, root, inode); 468 469 btrfs_end_transaction(trans); 470 471 out_unlock: 472 if (ret) { 473 binode->flags = old_flags; 474 inode->i_flags = old_i_flags; 475 } 476 477 inode_unlock(inode); 478 mnt_drop_write_file(file); 479 480 return ret; 481 } 482 483 static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 484 { 485 struct inode *inode = file_inode(file); 486 487 return put_user(inode->i_generation, arg); 488 } 489 490 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg) 491 { 492 struct inode *inode = file_inode(file); 493 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 494 struct btrfs_device *device; 495 struct request_queue *q; 496 struct fstrim_range range; 497 u64 minlen = ULLONG_MAX; 498 u64 num_devices = 0; 499 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 500 int ret; 501 502 if (!capable(CAP_SYS_ADMIN)) 503 return -EPERM; 504 505 rcu_read_lock(); 506 list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 507 dev_list) { 508 if (!device->bdev) 509 continue; 510 q = bdev_get_queue(device->bdev); 511 if (blk_queue_discard(q)) { 512 num_devices++; 513 minlen = min_t(u64, q->limits.discard_granularity, 514 minlen); 515 } 516 } 517 rcu_read_unlock(); 518 519 if (!num_devices) 520 return -EOPNOTSUPP; 521 if (copy_from_user(&range, arg, sizeof(range))) 522 return -EFAULT; 523 if (range.start > total_bytes || 524 range.len < fs_info->sb->s_blocksize) 525 return -EINVAL; 526 527 range.len = min(range.len, total_bytes - range.start); 528 range.minlen = max(range.minlen, minlen); 529 ret = btrfs_trim_fs(fs_info, &range); 530 if (ret < 0) 531 return ret; 532 533 if (copy_to_user(arg, &range, sizeof(range))) 534 return -EFAULT; 535 536 return 0; 537 } 538 539 int btrfs_is_empty_uuid(u8 *uuid) 540 { 541 int i; 542 543 for (i = 0; i < BTRFS_UUID_SIZE; i++) { 544 if (uuid[i]) 545 return 0; 546 } 547 return 1; 548 } 549 550 static noinline int create_subvol(struct inode *dir, 551 struct dentry *dentry, 552 const char *name, int namelen, 553 u64 *async_transid, 554 struct btrfs_qgroup_inherit *inherit) 555 { 556 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 557 struct btrfs_trans_handle *trans; 558 struct btrfs_key key; 559 struct btrfs_root_item *root_item; 560 struct btrfs_inode_item *inode_item; 561 struct extent_buffer *leaf; 562 struct btrfs_root *root = BTRFS_I(dir)->root; 563 struct btrfs_root *new_root; 564 struct btrfs_block_rsv block_rsv; 565 struct timespec cur_time = current_time(dir); 566 struct inode *inode; 567 int ret; 568 int err; 569 u64 objectid; 570 u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 571 u64 index = 0; 572 uuid_le new_uuid; 573 574 root_item = kzalloc(sizeof(*root_item), GFP_KERNEL); 575 if (!root_item) 576 return -ENOMEM; 577 578 ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid); 579 if (ret) 580 goto fail_free; 581 582 /* 583 * Don't create subvolume whose level is not zero. Or qgroup will be 584 * screwed up since it assumes subvolume qgroup's level to be 0. 585 */ 586 if (btrfs_qgroup_level(objectid)) { 587 ret = -ENOSPC; 588 goto fail_free; 589 } 590 591 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 592 /* 593 * The same as the snapshot creation, please see the comment 594 * of create_snapshot(). 595 */ 596 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false); 597 if (ret) 598 goto fail_free; 599 600 trans = btrfs_start_transaction(root, 0); 601 if (IS_ERR(trans)) { 602 ret = PTR_ERR(trans); 603 btrfs_subvolume_release_metadata(fs_info, &block_rsv); 604 goto fail_free; 605 } 606 trans->block_rsv = &block_rsv; 607 trans->bytes_reserved = block_rsv.size; 608 609 ret = btrfs_qgroup_inherit(trans, fs_info, 0, objectid, inherit); 610 if (ret) 611 goto fail; 612 613 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 614 if (IS_ERR(leaf)) { 615 ret = PTR_ERR(leaf); 616 goto fail; 617 } 618 619 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header)); 620 btrfs_set_header_bytenr(leaf, leaf->start); 621 btrfs_set_header_generation(leaf, trans->transid); 622 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 623 btrfs_set_header_owner(leaf, objectid); 624 625 write_extent_buffer_fsid(leaf, fs_info->fsid); 626 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); 627 btrfs_mark_buffer_dirty(leaf); 628 629 inode_item = &root_item->inode; 630 btrfs_set_stack_inode_generation(inode_item, 1); 631 btrfs_set_stack_inode_size(inode_item, 3); 632 btrfs_set_stack_inode_nlink(inode_item, 1); 633 btrfs_set_stack_inode_nbytes(inode_item, 634 fs_info->nodesize); 635 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 636 637 btrfs_set_root_flags(root_item, 0); 638 btrfs_set_root_limit(root_item, 0); 639 btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 640 641 btrfs_set_root_bytenr(root_item, leaf->start); 642 btrfs_set_root_generation(root_item, trans->transid); 643 btrfs_set_root_level(root_item, 0); 644 btrfs_set_root_refs(root_item, 1); 645 btrfs_set_root_used(root_item, leaf->len); 646 btrfs_set_root_last_snapshot(root_item, 0); 647 648 btrfs_set_root_generation_v2(root_item, 649 btrfs_root_generation(root_item)); 650 uuid_le_gen(&new_uuid); 651 memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); 652 btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec); 653 btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec); 654 root_item->ctime = root_item->otime; 655 btrfs_set_root_ctransid(root_item, trans->transid); 656 btrfs_set_root_otransid(root_item, trans->transid); 657 658 btrfs_tree_unlock(leaf); 659 free_extent_buffer(leaf); 660 leaf = NULL; 661 662 btrfs_set_root_dirid(root_item, new_dirid); 663 664 key.objectid = objectid; 665 key.offset = 0; 666 key.type = BTRFS_ROOT_ITEM_KEY; 667 ret = btrfs_insert_root(trans, fs_info->tree_root, &key, 668 root_item); 669 if (ret) 670 goto fail; 671 672 key.offset = (u64)-1; 673 new_root = btrfs_read_fs_root_no_name(fs_info, &key); 674 if (IS_ERR(new_root)) { 675 ret = PTR_ERR(new_root); 676 btrfs_abort_transaction(trans, ret); 677 goto fail; 678 } 679 680 btrfs_record_root_in_trans(trans, new_root); 681 682 ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid); 683 if (ret) { 684 /* We potentially lose an unused inode item here */ 685 btrfs_abort_transaction(trans, ret); 686 goto fail; 687 } 688 689 mutex_lock(&new_root->objectid_mutex); 690 new_root->highest_objectid = new_dirid; 691 mutex_unlock(&new_root->objectid_mutex); 692 693 /* 694 * insert the directory item 695 */ 696 ret = btrfs_set_inode_index(BTRFS_I(dir), &index); 697 if (ret) { 698 btrfs_abort_transaction(trans, ret); 699 goto fail; 700 } 701 702 ret = btrfs_insert_dir_item(trans, root, 703 name, namelen, BTRFS_I(dir), &key, 704 BTRFS_FT_DIR, index); 705 if (ret) { 706 btrfs_abort_transaction(trans, ret); 707 goto fail; 708 } 709 710 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2); 711 ret = btrfs_update_inode(trans, root, dir); 712 BUG_ON(ret); 713 714 ret = btrfs_add_root_ref(trans, fs_info, 715 objectid, root->root_key.objectid, 716 btrfs_ino(BTRFS_I(dir)), index, name, namelen); 717 BUG_ON(ret); 718 719 ret = btrfs_uuid_tree_add(trans, root_item->uuid, 720 BTRFS_UUID_KEY_SUBVOL, objectid); 721 if (ret) 722 btrfs_abort_transaction(trans, ret); 723 724 fail: 725 kfree(root_item); 726 trans->block_rsv = NULL; 727 trans->bytes_reserved = 0; 728 btrfs_subvolume_release_metadata(fs_info, &block_rsv); 729 730 if (async_transid) { 731 *async_transid = trans->transid; 732 err = btrfs_commit_transaction_async(trans, 1); 733 if (err) 734 err = btrfs_commit_transaction(trans); 735 } else { 736 err = btrfs_commit_transaction(trans); 737 } 738 if (err && !ret) 739 ret = err; 740 741 if (!ret) { 742 inode = btrfs_lookup_dentry(dir, dentry); 743 if (IS_ERR(inode)) 744 return PTR_ERR(inode); 745 d_instantiate(dentry, inode); 746 } 747 return ret; 748 749 fail_free: 750 kfree(root_item); 751 return ret; 752 } 753 754 static int create_snapshot(struct btrfs_root *root, struct inode *dir, 755 struct dentry *dentry, 756 u64 *async_transid, bool readonly, 757 struct btrfs_qgroup_inherit *inherit) 758 { 759 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 760 struct inode *inode; 761 struct btrfs_pending_snapshot *pending_snapshot; 762 struct btrfs_trans_handle *trans; 763 int ret; 764 765 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 766 return -EINVAL; 767 768 pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL); 769 if (!pending_snapshot) 770 return -ENOMEM; 771 772 pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item), 773 GFP_KERNEL); 774 pending_snapshot->path = btrfs_alloc_path(); 775 if (!pending_snapshot->root_item || !pending_snapshot->path) { 776 ret = -ENOMEM; 777 goto free_pending; 778 } 779 780 atomic_inc(&root->will_be_snapshotted); 781 smp_mb__after_atomic(); 782 /* wait for no snapshot writes */ 783 wait_event(root->subv_writers->wait, 784 percpu_counter_sum(&root->subv_writers->counter) == 0); 785 786 ret = btrfs_start_delalloc_inodes(root); 787 if (ret) 788 goto dec_and_free; 789 790 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1); 791 792 btrfs_init_block_rsv(&pending_snapshot->block_rsv, 793 BTRFS_BLOCK_RSV_TEMP); 794 /* 795 * 1 - parent dir inode 796 * 2 - dir entries 797 * 1 - root item 798 * 2 - root ref/backref 799 * 1 - root of snapshot 800 * 1 - UUID item 801 */ 802 ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 803 &pending_snapshot->block_rsv, 8, 804 false); 805 if (ret) 806 goto dec_and_free; 807 808 pending_snapshot->dentry = dentry; 809 pending_snapshot->root = root; 810 pending_snapshot->readonly = readonly; 811 pending_snapshot->dir = dir; 812 pending_snapshot->inherit = inherit; 813 814 trans = btrfs_start_transaction(root, 0); 815 if (IS_ERR(trans)) { 816 ret = PTR_ERR(trans); 817 goto fail; 818 } 819 820 spin_lock(&fs_info->trans_lock); 821 list_add(&pending_snapshot->list, 822 &trans->transaction->pending_snapshots); 823 spin_unlock(&fs_info->trans_lock); 824 if (async_transid) { 825 *async_transid = trans->transid; 826 ret = btrfs_commit_transaction_async(trans, 1); 827 if (ret) 828 ret = btrfs_commit_transaction(trans); 829 } else { 830 ret = btrfs_commit_transaction(trans); 831 } 832 if (ret) 833 goto fail; 834 835 ret = pending_snapshot->error; 836 if (ret) 837 goto fail; 838 839 ret = btrfs_orphan_cleanup(pending_snapshot->snap); 840 if (ret) 841 goto fail; 842 843 inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry); 844 if (IS_ERR(inode)) { 845 ret = PTR_ERR(inode); 846 goto fail; 847 } 848 849 d_instantiate(dentry, inode); 850 ret = 0; 851 fail: 852 btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv); 853 dec_and_free: 854 if (atomic_dec_and_test(&root->will_be_snapshotted)) 855 wake_up_var(&root->will_be_snapshotted); 856 free_pending: 857 kfree(pending_snapshot->root_item); 858 btrfs_free_path(pending_snapshot->path); 859 kfree(pending_snapshot); 860 861 return ret; 862 } 863 864 /* copy of may_delete in fs/namei.c() 865 * Check whether we can remove a link victim from directory dir, check 866 * whether the type of victim is right. 867 * 1. We can't do it if dir is read-only (done in permission()) 868 * 2. We should have write and exec permissions on dir 869 * 3. We can't remove anything from append-only dir 870 * 4. We can't do anything with immutable dir (done in permission()) 871 * 5. If the sticky bit on dir is set we should either 872 * a. be owner of dir, or 873 * b. be owner of victim, or 874 * c. have CAP_FOWNER capability 875 * 6. If the victim is append-only or immutable we can't do anything with 876 * links pointing to it. 877 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 878 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 879 * 9. We can't remove a root or mountpoint. 880 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 881 * nfs_async_unlink(). 882 */ 883 884 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir) 885 { 886 int error; 887 888 if (d_really_is_negative(victim)) 889 return -ENOENT; 890 891 BUG_ON(d_inode(victim->d_parent) != dir); 892 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 893 894 error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 895 if (error) 896 return error; 897 if (IS_APPEND(dir)) 898 return -EPERM; 899 if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) || 900 IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim))) 901 return -EPERM; 902 if (isdir) { 903 if (!d_is_dir(victim)) 904 return -ENOTDIR; 905 if (IS_ROOT(victim)) 906 return -EBUSY; 907 } else if (d_is_dir(victim)) 908 return -EISDIR; 909 if (IS_DEADDIR(dir)) 910 return -ENOENT; 911 if (victim->d_flags & DCACHE_NFSFS_RENAMED) 912 return -EBUSY; 913 return 0; 914 } 915 916 /* copy of may_create in fs/namei.c() */ 917 static inline int btrfs_may_create(struct inode *dir, struct dentry *child) 918 { 919 if (d_really_is_positive(child)) 920 return -EEXIST; 921 if (IS_DEADDIR(dir)) 922 return -ENOENT; 923 return inode_permission(dir, MAY_WRITE | MAY_EXEC); 924 } 925 926 /* 927 * Create a new subvolume below @parent. This is largely modeled after 928 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 929 * inside this filesystem so it's quite a bit simpler. 930 */ 931 static noinline int btrfs_mksubvol(const struct path *parent, 932 const char *name, int namelen, 933 struct btrfs_root *snap_src, 934 u64 *async_transid, bool readonly, 935 struct btrfs_qgroup_inherit *inherit) 936 { 937 struct inode *dir = d_inode(parent->dentry); 938 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 939 struct dentry *dentry; 940 int error; 941 942 error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 943 if (error == -EINTR) 944 return error; 945 946 dentry = lookup_one_len(name, parent->dentry, namelen); 947 error = PTR_ERR(dentry); 948 if (IS_ERR(dentry)) 949 goto out_unlock; 950 951 error = btrfs_may_create(dir, dentry); 952 if (error) 953 goto out_dput; 954 955 /* 956 * even if this name doesn't exist, we may get hash collisions. 957 * check for them now when we can safely fail 958 */ 959 error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 960 dir->i_ino, name, 961 namelen); 962 if (error) 963 goto out_dput; 964 965 down_read(&fs_info->subvol_sem); 966 967 if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 968 goto out_up_read; 969 970 if (snap_src) { 971 error = create_snapshot(snap_src, dir, dentry, 972 async_transid, readonly, inherit); 973 } else { 974 error = create_subvol(dir, dentry, name, namelen, 975 async_transid, inherit); 976 } 977 if (!error) 978 fsnotify_mkdir(dir, dentry); 979 out_up_read: 980 up_read(&fs_info->subvol_sem); 981 out_dput: 982 dput(dentry); 983 out_unlock: 984 inode_unlock(dir); 985 return error; 986 } 987 988 /* 989 * When we're defragging a range, we don't want to kick it off again 990 * if it is really just waiting for delalloc to send it down. 991 * If we find a nice big extent or delalloc range for the bytes in the 992 * file you want to defrag, we return 0 to let you know to skip this 993 * part of the file 994 */ 995 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh) 996 { 997 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 998 struct extent_map *em = NULL; 999 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1000 u64 end; 1001 1002 read_lock(&em_tree->lock); 1003 em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE); 1004 read_unlock(&em_tree->lock); 1005 1006 if (em) { 1007 end = extent_map_end(em); 1008 free_extent_map(em); 1009 if (end - offset > thresh) 1010 return 0; 1011 } 1012 /* if we already have a nice delalloc here, just stop */ 1013 thresh /= 2; 1014 end = count_range_bits(io_tree, &offset, offset + thresh, 1015 thresh, EXTENT_DELALLOC, 1); 1016 if (end >= thresh) 1017 return 0; 1018 return 1; 1019 } 1020 1021 /* 1022 * helper function to walk through a file and find extents 1023 * newer than a specific transid, and smaller than thresh. 1024 * 1025 * This is used by the defragging code to find new and small 1026 * extents 1027 */ 1028 static int find_new_extents(struct btrfs_root *root, 1029 struct inode *inode, u64 newer_than, 1030 u64 *off, u32 thresh) 1031 { 1032 struct btrfs_path *path; 1033 struct btrfs_key min_key; 1034 struct extent_buffer *leaf; 1035 struct btrfs_file_extent_item *extent; 1036 int type; 1037 int ret; 1038 u64 ino = btrfs_ino(BTRFS_I(inode)); 1039 1040 path = btrfs_alloc_path(); 1041 if (!path) 1042 return -ENOMEM; 1043 1044 min_key.objectid = ino; 1045 min_key.type = BTRFS_EXTENT_DATA_KEY; 1046 min_key.offset = *off; 1047 1048 while (1) { 1049 ret = btrfs_search_forward(root, &min_key, path, newer_than); 1050 if (ret != 0) 1051 goto none; 1052 process_slot: 1053 if (min_key.objectid != ino) 1054 goto none; 1055 if (min_key.type != BTRFS_EXTENT_DATA_KEY) 1056 goto none; 1057 1058 leaf = path->nodes[0]; 1059 extent = btrfs_item_ptr(leaf, path->slots[0], 1060 struct btrfs_file_extent_item); 1061 1062 type = btrfs_file_extent_type(leaf, extent); 1063 if (type == BTRFS_FILE_EXTENT_REG && 1064 btrfs_file_extent_num_bytes(leaf, extent) < thresh && 1065 check_defrag_in_cache(inode, min_key.offset, thresh)) { 1066 *off = min_key.offset; 1067 btrfs_free_path(path); 1068 return 0; 1069 } 1070 1071 path->slots[0]++; 1072 if (path->slots[0] < btrfs_header_nritems(leaf)) { 1073 btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]); 1074 goto process_slot; 1075 } 1076 1077 if (min_key.offset == (u64)-1) 1078 goto none; 1079 1080 min_key.offset++; 1081 btrfs_release_path(path); 1082 } 1083 none: 1084 btrfs_free_path(path); 1085 return -ENOENT; 1086 } 1087 1088 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start) 1089 { 1090 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1091 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 1092 struct extent_map *em; 1093 u64 len = PAGE_SIZE; 1094 1095 /* 1096 * hopefully we have this extent in the tree already, try without 1097 * the full extent lock 1098 */ 1099 read_lock(&em_tree->lock); 1100 em = lookup_extent_mapping(em_tree, start, len); 1101 read_unlock(&em_tree->lock); 1102 1103 if (!em) { 1104 struct extent_state *cached = NULL; 1105 u64 end = start + len - 1; 1106 1107 /* get the big lock and read metadata off disk */ 1108 lock_extent_bits(io_tree, start, end, &cached); 1109 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); 1110 unlock_extent_cached(io_tree, start, end, &cached); 1111 1112 if (IS_ERR(em)) 1113 return NULL; 1114 } 1115 1116 return em; 1117 } 1118 1119 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em) 1120 { 1121 struct extent_map *next; 1122 bool ret = true; 1123 1124 /* this is the last extent */ 1125 if (em->start + em->len >= i_size_read(inode)) 1126 return false; 1127 1128 next = defrag_lookup_extent(inode, em->start + em->len); 1129 if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 1130 ret = false; 1131 else if ((em->block_start + em->block_len == next->block_start) && 1132 (em->block_len > SZ_128K && next->block_len > SZ_128K)) 1133 ret = false; 1134 1135 free_extent_map(next); 1136 return ret; 1137 } 1138 1139 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh, 1140 u64 *last_len, u64 *skip, u64 *defrag_end, 1141 int compress) 1142 { 1143 struct extent_map *em; 1144 int ret = 1; 1145 bool next_mergeable = true; 1146 bool prev_mergeable = true; 1147 1148 /* 1149 * make sure that once we start defragging an extent, we keep on 1150 * defragging it 1151 */ 1152 if (start < *defrag_end) 1153 return 1; 1154 1155 *skip = 0; 1156 1157 em = defrag_lookup_extent(inode, start); 1158 if (!em) 1159 return 0; 1160 1161 /* this will cover holes, and inline extents */ 1162 if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 1163 ret = 0; 1164 goto out; 1165 } 1166 1167 if (!*defrag_end) 1168 prev_mergeable = false; 1169 1170 next_mergeable = defrag_check_next_extent(inode, em); 1171 /* 1172 * we hit a real extent, if it is big or the next extent is not a 1173 * real extent, don't bother defragging it 1174 */ 1175 if (!compress && (*last_len == 0 || *last_len >= thresh) && 1176 (em->len >= thresh || (!next_mergeable && !prev_mergeable))) 1177 ret = 0; 1178 out: 1179 /* 1180 * last_len ends up being a counter of how many bytes we've defragged. 1181 * every time we choose not to defrag an extent, we reset *last_len 1182 * so that the next tiny extent will force a defrag. 1183 * 1184 * The end result of this is that tiny extents before a single big 1185 * extent will force at least part of that big extent to be defragged. 1186 */ 1187 if (ret) { 1188 *defrag_end = extent_map_end(em); 1189 } else { 1190 *last_len = 0; 1191 *skip = extent_map_end(em); 1192 *defrag_end = 0; 1193 } 1194 1195 free_extent_map(em); 1196 return ret; 1197 } 1198 1199 /* 1200 * it doesn't do much good to defrag one or two pages 1201 * at a time. This pulls in a nice chunk of pages 1202 * to COW and defrag. 1203 * 1204 * It also makes sure the delalloc code has enough 1205 * dirty data to avoid making new small extents as part 1206 * of the defrag 1207 * 1208 * It's a good idea to start RA on this range 1209 * before calling this. 1210 */ 1211 static int cluster_pages_for_defrag(struct inode *inode, 1212 struct page **pages, 1213 unsigned long start_index, 1214 unsigned long num_pages) 1215 { 1216 unsigned long file_end; 1217 u64 isize = i_size_read(inode); 1218 u64 page_start; 1219 u64 page_end; 1220 u64 page_cnt; 1221 int ret; 1222 int i; 1223 int i_done; 1224 struct btrfs_ordered_extent *ordered; 1225 struct extent_state *cached_state = NULL; 1226 struct extent_io_tree *tree; 1227 struct extent_changeset *data_reserved = NULL; 1228 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 1229 1230 file_end = (isize - 1) >> PAGE_SHIFT; 1231 if (!isize || start_index > file_end) 1232 return 0; 1233 1234 page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1); 1235 1236 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, 1237 start_index << PAGE_SHIFT, 1238 page_cnt << PAGE_SHIFT); 1239 if (ret) 1240 return ret; 1241 i_done = 0; 1242 tree = &BTRFS_I(inode)->io_tree; 1243 1244 /* step one, lock all the pages */ 1245 for (i = 0; i < page_cnt; i++) { 1246 struct page *page; 1247 again: 1248 page = find_or_create_page(inode->i_mapping, 1249 start_index + i, mask); 1250 if (!page) 1251 break; 1252 1253 page_start = page_offset(page); 1254 page_end = page_start + PAGE_SIZE - 1; 1255 while (1) { 1256 lock_extent_bits(tree, page_start, page_end, 1257 &cached_state); 1258 ordered = btrfs_lookup_ordered_extent(inode, 1259 page_start); 1260 unlock_extent_cached(tree, page_start, page_end, 1261 &cached_state); 1262 if (!ordered) 1263 break; 1264 1265 unlock_page(page); 1266 btrfs_start_ordered_extent(inode, ordered, 1); 1267 btrfs_put_ordered_extent(ordered); 1268 lock_page(page); 1269 /* 1270 * we unlocked the page above, so we need check if 1271 * it was released or not. 1272 */ 1273 if (page->mapping != inode->i_mapping) { 1274 unlock_page(page); 1275 put_page(page); 1276 goto again; 1277 } 1278 } 1279 1280 if (!PageUptodate(page)) { 1281 btrfs_readpage(NULL, page); 1282 lock_page(page); 1283 if (!PageUptodate(page)) { 1284 unlock_page(page); 1285 put_page(page); 1286 ret = -EIO; 1287 break; 1288 } 1289 } 1290 1291 if (page->mapping != inode->i_mapping) { 1292 unlock_page(page); 1293 put_page(page); 1294 goto again; 1295 } 1296 1297 pages[i] = page; 1298 i_done++; 1299 } 1300 if (!i_done || ret) 1301 goto out; 1302 1303 if (!(inode->i_sb->s_flags & SB_ACTIVE)) 1304 goto out; 1305 1306 /* 1307 * so now we have a nice long stream of locked 1308 * and up to date pages, lets wait on them 1309 */ 1310 for (i = 0; i < i_done; i++) 1311 wait_on_page_writeback(pages[i]); 1312 1313 page_start = page_offset(pages[0]); 1314 page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE; 1315 1316 lock_extent_bits(&BTRFS_I(inode)->io_tree, 1317 page_start, page_end - 1, &cached_state); 1318 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, 1319 page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | 1320 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, 1321 &cached_state); 1322 1323 if (i_done != page_cnt) { 1324 spin_lock(&BTRFS_I(inode)->lock); 1325 BTRFS_I(inode)->outstanding_extents++; 1326 spin_unlock(&BTRFS_I(inode)->lock); 1327 btrfs_delalloc_release_space(inode, data_reserved, 1328 start_index << PAGE_SHIFT, 1329 (page_cnt - i_done) << PAGE_SHIFT, true); 1330 } 1331 1332 1333 set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, 1334 &cached_state); 1335 1336 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 1337 page_start, page_end - 1, &cached_state); 1338 1339 for (i = 0; i < i_done; i++) { 1340 clear_page_dirty_for_io(pages[i]); 1341 ClearPageChecked(pages[i]); 1342 set_page_extent_mapped(pages[i]); 1343 set_page_dirty(pages[i]); 1344 unlock_page(pages[i]); 1345 put_page(pages[i]); 1346 } 1347 btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT, 1348 false); 1349 extent_changeset_free(data_reserved); 1350 return i_done; 1351 out: 1352 for (i = 0; i < i_done; i++) { 1353 unlock_page(pages[i]); 1354 put_page(pages[i]); 1355 } 1356 btrfs_delalloc_release_space(inode, data_reserved, 1357 start_index << PAGE_SHIFT, 1358 page_cnt << PAGE_SHIFT, true); 1359 btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT, 1360 true); 1361 extent_changeset_free(data_reserved); 1362 return ret; 1363 1364 } 1365 1366 int btrfs_defrag_file(struct inode *inode, struct file *file, 1367 struct btrfs_ioctl_defrag_range_args *range, 1368 u64 newer_than, unsigned long max_to_defrag) 1369 { 1370 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1371 struct btrfs_root *root = BTRFS_I(inode)->root; 1372 struct file_ra_state *ra = NULL; 1373 unsigned long last_index; 1374 u64 isize = i_size_read(inode); 1375 u64 last_len = 0; 1376 u64 skip = 0; 1377 u64 defrag_end = 0; 1378 u64 newer_off = range->start; 1379 unsigned long i; 1380 unsigned long ra_index = 0; 1381 int ret; 1382 int defrag_count = 0; 1383 int compress_type = BTRFS_COMPRESS_ZLIB; 1384 u32 extent_thresh = range->extent_thresh; 1385 unsigned long max_cluster = SZ_256K >> PAGE_SHIFT; 1386 unsigned long cluster = max_cluster; 1387 u64 new_align = ~((u64)SZ_128K - 1); 1388 struct page **pages = NULL; 1389 bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS; 1390 1391 if (isize == 0) 1392 return 0; 1393 1394 if (range->start >= isize) 1395 return -EINVAL; 1396 1397 if (do_compress) { 1398 if (range->compress_type > BTRFS_COMPRESS_TYPES) 1399 return -EINVAL; 1400 if (range->compress_type) 1401 compress_type = range->compress_type; 1402 } 1403 1404 if (extent_thresh == 0) 1405 extent_thresh = SZ_256K; 1406 1407 /* 1408 * If we were not given a file, allocate a readahead context. As 1409 * readahead is just an optimization, defrag will work without it so 1410 * we don't error out. 1411 */ 1412 if (!file) { 1413 ra = kzalloc(sizeof(*ra), GFP_KERNEL); 1414 if (ra) 1415 file_ra_state_init(ra, inode->i_mapping); 1416 } else { 1417 ra = &file->f_ra; 1418 } 1419 1420 pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL); 1421 if (!pages) { 1422 ret = -ENOMEM; 1423 goto out_ra; 1424 } 1425 1426 /* find the last page to defrag */ 1427 if (range->start + range->len > range->start) { 1428 last_index = min_t(u64, isize - 1, 1429 range->start + range->len - 1) >> PAGE_SHIFT; 1430 } else { 1431 last_index = (isize - 1) >> PAGE_SHIFT; 1432 } 1433 1434 if (newer_than) { 1435 ret = find_new_extents(root, inode, newer_than, 1436 &newer_off, SZ_64K); 1437 if (!ret) { 1438 range->start = newer_off; 1439 /* 1440 * we always align our defrag to help keep 1441 * the extents in the file evenly spaced 1442 */ 1443 i = (newer_off & new_align) >> PAGE_SHIFT; 1444 } else 1445 goto out_ra; 1446 } else { 1447 i = range->start >> PAGE_SHIFT; 1448 } 1449 if (!max_to_defrag) 1450 max_to_defrag = last_index - i + 1; 1451 1452 /* 1453 * make writeback starts from i, so the defrag range can be 1454 * written sequentially. 1455 */ 1456 if (i < inode->i_mapping->writeback_index) 1457 inode->i_mapping->writeback_index = i; 1458 1459 while (i <= last_index && defrag_count < max_to_defrag && 1460 (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) { 1461 /* 1462 * make sure we stop running if someone unmounts 1463 * the FS 1464 */ 1465 if (!(inode->i_sb->s_flags & SB_ACTIVE)) 1466 break; 1467 1468 if (btrfs_defrag_cancelled(fs_info)) { 1469 btrfs_debug(fs_info, "defrag_file cancelled"); 1470 ret = -EAGAIN; 1471 break; 1472 } 1473 1474 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT, 1475 extent_thresh, &last_len, &skip, 1476 &defrag_end, do_compress)){ 1477 unsigned long next; 1478 /* 1479 * the should_defrag function tells us how much to skip 1480 * bump our counter by the suggested amount 1481 */ 1482 next = DIV_ROUND_UP(skip, PAGE_SIZE); 1483 i = max(i + 1, next); 1484 continue; 1485 } 1486 1487 if (!newer_than) { 1488 cluster = (PAGE_ALIGN(defrag_end) >> 1489 PAGE_SHIFT) - i; 1490 cluster = min(cluster, max_cluster); 1491 } else { 1492 cluster = max_cluster; 1493 } 1494 1495 if (i + cluster > ra_index) { 1496 ra_index = max(i, ra_index); 1497 if (ra) 1498 page_cache_sync_readahead(inode->i_mapping, ra, 1499 file, ra_index, cluster); 1500 ra_index += cluster; 1501 } 1502 1503 inode_lock(inode); 1504 if (do_compress) 1505 BTRFS_I(inode)->defrag_compress = compress_type; 1506 ret = cluster_pages_for_defrag(inode, pages, i, cluster); 1507 if (ret < 0) { 1508 inode_unlock(inode); 1509 goto out_ra; 1510 } 1511 1512 defrag_count += ret; 1513 balance_dirty_pages_ratelimited(inode->i_mapping); 1514 inode_unlock(inode); 1515 1516 if (newer_than) { 1517 if (newer_off == (u64)-1) 1518 break; 1519 1520 if (ret > 0) 1521 i += ret; 1522 1523 newer_off = max(newer_off + 1, 1524 (u64)i << PAGE_SHIFT); 1525 1526 ret = find_new_extents(root, inode, newer_than, 1527 &newer_off, SZ_64K); 1528 if (!ret) { 1529 range->start = newer_off; 1530 i = (newer_off & new_align) >> PAGE_SHIFT; 1531 } else { 1532 break; 1533 } 1534 } else { 1535 if (ret > 0) { 1536 i += ret; 1537 last_len += ret << PAGE_SHIFT; 1538 } else { 1539 i++; 1540 last_len = 0; 1541 } 1542 } 1543 } 1544 1545 if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) { 1546 filemap_flush(inode->i_mapping); 1547 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, 1548 &BTRFS_I(inode)->runtime_flags)) 1549 filemap_flush(inode->i_mapping); 1550 } 1551 1552 if (range->compress_type == BTRFS_COMPRESS_LZO) { 1553 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO); 1554 } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) { 1555 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD); 1556 } 1557 1558 ret = defrag_count; 1559 1560 out_ra: 1561 if (do_compress) { 1562 inode_lock(inode); 1563 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE; 1564 inode_unlock(inode); 1565 } 1566 if (!file) 1567 kfree(ra); 1568 kfree(pages); 1569 return ret; 1570 } 1571 1572 static noinline int btrfs_ioctl_resize(struct file *file, 1573 void __user *arg) 1574 { 1575 struct inode *inode = file_inode(file); 1576 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1577 u64 new_size; 1578 u64 old_size; 1579 u64 devid = 1; 1580 struct btrfs_root *root = BTRFS_I(inode)->root; 1581 struct btrfs_ioctl_vol_args *vol_args; 1582 struct btrfs_trans_handle *trans; 1583 struct btrfs_device *device = NULL; 1584 char *sizestr; 1585 char *retptr; 1586 char *devstr = NULL; 1587 int ret = 0; 1588 int mod = 0; 1589 1590 if (!capable(CAP_SYS_ADMIN)) 1591 return -EPERM; 1592 1593 ret = mnt_want_write_file(file); 1594 if (ret) 1595 return ret; 1596 1597 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 1598 mnt_drop_write_file(file); 1599 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 1600 } 1601 1602 vol_args = memdup_user(arg, sizeof(*vol_args)); 1603 if (IS_ERR(vol_args)) { 1604 ret = PTR_ERR(vol_args); 1605 goto out; 1606 } 1607 1608 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1609 1610 sizestr = vol_args->name; 1611 devstr = strchr(sizestr, ':'); 1612 if (devstr) { 1613 sizestr = devstr + 1; 1614 *devstr = '\0'; 1615 devstr = vol_args->name; 1616 ret = kstrtoull(devstr, 10, &devid); 1617 if (ret) 1618 goto out_free; 1619 if (!devid) { 1620 ret = -EINVAL; 1621 goto out_free; 1622 } 1623 btrfs_info(fs_info, "resizing devid %llu", devid); 1624 } 1625 1626 device = btrfs_find_device(fs_info, devid, NULL, NULL); 1627 if (!device) { 1628 btrfs_info(fs_info, "resizer unable to find device %llu", 1629 devid); 1630 ret = -ENODEV; 1631 goto out_free; 1632 } 1633 1634 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1635 btrfs_info(fs_info, 1636 "resizer unable to apply on readonly device %llu", 1637 devid); 1638 ret = -EPERM; 1639 goto out_free; 1640 } 1641 1642 if (!strcmp(sizestr, "max")) 1643 new_size = device->bdev->bd_inode->i_size; 1644 else { 1645 if (sizestr[0] == '-') { 1646 mod = -1; 1647 sizestr++; 1648 } else if (sizestr[0] == '+') { 1649 mod = 1; 1650 sizestr++; 1651 } 1652 new_size = memparse(sizestr, &retptr); 1653 if (*retptr != '\0' || new_size == 0) { 1654 ret = -EINVAL; 1655 goto out_free; 1656 } 1657 } 1658 1659 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1660 ret = -EPERM; 1661 goto out_free; 1662 } 1663 1664 old_size = btrfs_device_get_total_bytes(device); 1665 1666 if (mod < 0) { 1667 if (new_size > old_size) { 1668 ret = -EINVAL; 1669 goto out_free; 1670 } 1671 new_size = old_size - new_size; 1672 } else if (mod > 0) { 1673 if (new_size > ULLONG_MAX - old_size) { 1674 ret = -ERANGE; 1675 goto out_free; 1676 } 1677 new_size = old_size + new_size; 1678 } 1679 1680 if (new_size < SZ_256M) { 1681 ret = -EINVAL; 1682 goto out_free; 1683 } 1684 if (new_size > device->bdev->bd_inode->i_size) { 1685 ret = -EFBIG; 1686 goto out_free; 1687 } 1688 1689 new_size = round_down(new_size, fs_info->sectorsize); 1690 1691 btrfs_info_in_rcu(fs_info, "new size for %s is %llu", 1692 rcu_str_deref(device->name), new_size); 1693 1694 if (new_size > old_size) { 1695 trans = btrfs_start_transaction(root, 0); 1696 if (IS_ERR(trans)) { 1697 ret = PTR_ERR(trans); 1698 goto out_free; 1699 } 1700 ret = btrfs_grow_device(trans, device, new_size); 1701 btrfs_commit_transaction(trans); 1702 } else if (new_size < old_size) { 1703 ret = btrfs_shrink_device(device, new_size); 1704 } /* equal, nothing need to do */ 1705 1706 out_free: 1707 kfree(vol_args); 1708 out: 1709 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 1710 mnt_drop_write_file(file); 1711 return ret; 1712 } 1713 1714 static noinline int btrfs_ioctl_snap_create_transid(struct file *file, 1715 const char *name, unsigned long fd, int subvol, 1716 u64 *transid, bool readonly, 1717 struct btrfs_qgroup_inherit *inherit) 1718 { 1719 int namelen; 1720 int ret = 0; 1721 1722 if (!S_ISDIR(file_inode(file)->i_mode)) 1723 return -ENOTDIR; 1724 1725 ret = mnt_want_write_file(file); 1726 if (ret) 1727 goto out; 1728 1729 namelen = strlen(name); 1730 if (strchr(name, '/')) { 1731 ret = -EINVAL; 1732 goto out_drop_write; 1733 } 1734 1735 if (name[0] == '.' && 1736 (namelen == 1 || (name[1] == '.' && namelen == 2))) { 1737 ret = -EEXIST; 1738 goto out_drop_write; 1739 } 1740 1741 if (subvol) { 1742 ret = btrfs_mksubvol(&file->f_path, name, namelen, 1743 NULL, transid, readonly, inherit); 1744 } else { 1745 struct fd src = fdget(fd); 1746 struct inode *src_inode; 1747 if (!src.file) { 1748 ret = -EINVAL; 1749 goto out_drop_write; 1750 } 1751 1752 src_inode = file_inode(src.file); 1753 if (src_inode->i_sb != file_inode(file)->i_sb) { 1754 btrfs_info(BTRFS_I(file_inode(file))->root->fs_info, 1755 "Snapshot src from another FS"); 1756 ret = -EXDEV; 1757 } else if (!inode_owner_or_capable(src_inode)) { 1758 /* 1759 * Subvolume creation is not restricted, but snapshots 1760 * are limited to own subvolumes only 1761 */ 1762 ret = -EPERM; 1763 } else { 1764 ret = btrfs_mksubvol(&file->f_path, name, namelen, 1765 BTRFS_I(src_inode)->root, 1766 transid, readonly, inherit); 1767 } 1768 fdput(src); 1769 } 1770 out_drop_write: 1771 mnt_drop_write_file(file); 1772 out: 1773 return ret; 1774 } 1775 1776 static noinline int btrfs_ioctl_snap_create(struct file *file, 1777 void __user *arg, int subvol) 1778 { 1779 struct btrfs_ioctl_vol_args *vol_args; 1780 int ret; 1781 1782 if (!S_ISDIR(file_inode(file)->i_mode)) 1783 return -ENOTDIR; 1784 1785 vol_args = memdup_user(arg, sizeof(*vol_args)); 1786 if (IS_ERR(vol_args)) 1787 return PTR_ERR(vol_args); 1788 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1789 1790 ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1791 vol_args->fd, subvol, 1792 NULL, false, NULL); 1793 1794 kfree(vol_args); 1795 return ret; 1796 } 1797 1798 static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1799 void __user *arg, int subvol) 1800 { 1801 struct btrfs_ioctl_vol_args_v2 *vol_args; 1802 int ret; 1803 u64 transid = 0; 1804 u64 *ptr = NULL; 1805 bool readonly = false; 1806 struct btrfs_qgroup_inherit *inherit = NULL; 1807 1808 if (!S_ISDIR(file_inode(file)->i_mode)) 1809 return -ENOTDIR; 1810 1811 vol_args = memdup_user(arg, sizeof(*vol_args)); 1812 if (IS_ERR(vol_args)) 1813 return PTR_ERR(vol_args); 1814 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1815 1816 if (vol_args->flags & 1817 ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY | 1818 BTRFS_SUBVOL_QGROUP_INHERIT)) { 1819 ret = -EOPNOTSUPP; 1820 goto free_args; 1821 } 1822 1823 if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) 1824 ptr = &transid; 1825 if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1826 readonly = true; 1827 if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 1828 if (vol_args->size > PAGE_SIZE) { 1829 ret = -EINVAL; 1830 goto free_args; 1831 } 1832 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 1833 if (IS_ERR(inherit)) { 1834 ret = PTR_ERR(inherit); 1835 goto free_args; 1836 } 1837 } 1838 1839 ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1840 vol_args->fd, subvol, ptr, 1841 readonly, inherit); 1842 if (ret) 1843 goto free_inherit; 1844 1845 if (ptr && copy_to_user(arg + 1846 offsetof(struct btrfs_ioctl_vol_args_v2, 1847 transid), 1848 ptr, sizeof(*ptr))) 1849 ret = -EFAULT; 1850 1851 free_inherit: 1852 kfree(inherit); 1853 free_args: 1854 kfree(vol_args); 1855 return ret; 1856 } 1857 1858 static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 1859 void __user *arg) 1860 { 1861 struct inode *inode = file_inode(file); 1862 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1863 struct btrfs_root *root = BTRFS_I(inode)->root; 1864 int ret = 0; 1865 u64 flags = 0; 1866 1867 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) 1868 return -EINVAL; 1869 1870 down_read(&fs_info->subvol_sem); 1871 if (btrfs_root_readonly(root)) 1872 flags |= BTRFS_SUBVOL_RDONLY; 1873 up_read(&fs_info->subvol_sem); 1874 1875 if (copy_to_user(arg, &flags, sizeof(flags))) 1876 ret = -EFAULT; 1877 1878 return ret; 1879 } 1880 1881 static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 1882 void __user *arg) 1883 { 1884 struct inode *inode = file_inode(file); 1885 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1886 struct btrfs_root *root = BTRFS_I(inode)->root; 1887 struct btrfs_trans_handle *trans; 1888 u64 root_flags; 1889 u64 flags; 1890 int ret = 0; 1891 1892 if (!inode_owner_or_capable(inode)) 1893 return -EPERM; 1894 1895 ret = mnt_want_write_file(file); 1896 if (ret) 1897 goto out; 1898 1899 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 1900 ret = -EINVAL; 1901 goto out_drop_write; 1902 } 1903 1904 if (copy_from_user(&flags, arg, sizeof(flags))) { 1905 ret = -EFAULT; 1906 goto out_drop_write; 1907 } 1908 1909 if (flags & BTRFS_SUBVOL_CREATE_ASYNC) { 1910 ret = -EINVAL; 1911 goto out_drop_write; 1912 } 1913 1914 if (flags & ~BTRFS_SUBVOL_RDONLY) { 1915 ret = -EOPNOTSUPP; 1916 goto out_drop_write; 1917 } 1918 1919 down_write(&fs_info->subvol_sem); 1920 1921 /* nothing to do */ 1922 if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 1923 goto out_drop_sem; 1924 1925 root_flags = btrfs_root_flags(&root->root_item); 1926 if (flags & BTRFS_SUBVOL_RDONLY) { 1927 btrfs_set_root_flags(&root->root_item, 1928 root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 1929 } else { 1930 /* 1931 * Block RO -> RW transition if this subvolume is involved in 1932 * send 1933 */ 1934 spin_lock(&root->root_item_lock); 1935 if (root->send_in_progress == 0) { 1936 btrfs_set_root_flags(&root->root_item, 1937 root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 1938 spin_unlock(&root->root_item_lock); 1939 } else { 1940 spin_unlock(&root->root_item_lock); 1941 btrfs_warn(fs_info, 1942 "Attempt to set subvolume %llu read-write during send", 1943 root->root_key.objectid); 1944 ret = -EPERM; 1945 goto out_drop_sem; 1946 } 1947 } 1948 1949 trans = btrfs_start_transaction(root, 1); 1950 if (IS_ERR(trans)) { 1951 ret = PTR_ERR(trans); 1952 goto out_reset; 1953 } 1954 1955 ret = btrfs_update_root(trans, fs_info->tree_root, 1956 &root->root_key, &root->root_item); 1957 if (ret < 0) { 1958 btrfs_end_transaction(trans); 1959 goto out_reset; 1960 } 1961 1962 ret = btrfs_commit_transaction(trans); 1963 1964 out_reset: 1965 if (ret) 1966 btrfs_set_root_flags(&root->root_item, root_flags); 1967 out_drop_sem: 1968 up_write(&fs_info->subvol_sem); 1969 out_drop_write: 1970 mnt_drop_write_file(file); 1971 out: 1972 return ret; 1973 } 1974 1975 static noinline int key_in_sk(struct btrfs_key *key, 1976 struct btrfs_ioctl_search_key *sk) 1977 { 1978 struct btrfs_key test; 1979 int ret; 1980 1981 test.objectid = sk->min_objectid; 1982 test.type = sk->min_type; 1983 test.offset = sk->min_offset; 1984 1985 ret = btrfs_comp_cpu_keys(key, &test); 1986 if (ret < 0) 1987 return 0; 1988 1989 test.objectid = sk->max_objectid; 1990 test.type = sk->max_type; 1991 test.offset = sk->max_offset; 1992 1993 ret = btrfs_comp_cpu_keys(key, &test); 1994 if (ret > 0) 1995 return 0; 1996 return 1; 1997 } 1998 1999 static noinline int copy_to_sk(struct btrfs_path *path, 2000 struct btrfs_key *key, 2001 struct btrfs_ioctl_search_key *sk, 2002 size_t *buf_size, 2003 char __user *ubuf, 2004 unsigned long *sk_offset, 2005 int *num_found) 2006 { 2007 u64 found_transid; 2008 struct extent_buffer *leaf; 2009 struct btrfs_ioctl_search_header sh; 2010 struct btrfs_key test; 2011 unsigned long item_off; 2012 unsigned long item_len; 2013 int nritems; 2014 int i; 2015 int slot; 2016 int ret = 0; 2017 2018 leaf = path->nodes[0]; 2019 slot = path->slots[0]; 2020 nritems = btrfs_header_nritems(leaf); 2021 2022 if (btrfs_header_generation(leaf) > sk->max_transid) { 2023 i = nritems; 2024 goto advance_key; 2025 } 2026 found_transid = btrfs_header_generation(leaf); 2027 2028 for (i = slot; i < nritems; i++) { 2029 item_off = btrfs_item_ptr_offset(leaf, i); 2030 item_len = btrfs_item_size_nr(leaf, i); 2031 2032 btrfs_item_key_to_cpu(leaf, key, i); 2033 if (!key_in_sk(key, sk)) 2034 continue; 2035 2036 if (sizeof(sh) + item_len > *buf_size) { 2037 if (*num_found) { 2038 ret = 1; 2039 goto out; 2040 } 2041 2042 /* 2043 * return one empty item back for v1, which does not 2044 * handle -EOVERFLOW 2045 */ 2046 2047 *buf_size = sizeof(sh) + item_len; 2048 item_len = 0; 2049 ret = -EOVERFLOW; 2050 } 2051 2052 if (sizeof(sh) + item_len + *sk_offset > *buf_size) { 2053 ret = 1; 2054 goto out; 2055 } 2056 2057 sh.objectid = key->objectid; 2058 sh.offset = key->offset; 2059 sh.type = key->type; 2060 sh.len = item_len; 2061 sh.transid = found_transid; 2062 2063 /* copy search result header */ 2064 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) { 2065 ret = -EFAULT; 2066 goto out; 2067 } 2068 2069 *sk_offset += sizeof(sh); 2070 2071 if (item_len) { 2072 char __user *up = ubuf + *sk_offset; 2073 /* copy the item */ 2074 if (read_extent_buffer_to_user(leaf, up, 2075 item_off, item_len)) { 2076 ret = -EFAULT; 2077 goto out; 2078 } 2079 2080 *sk_offset += item_len; 2081 } 2082 (*num_found)++; 2083 2084 if (ret) /* -EOVERFLOW from above */ 2085 goto out; 2086 2087 if (*num_found >= sk->nr_items) { 2088 ret = 1; 2089 goto out; 2090 } 2091 } 2092 advance_key: 2093 ret = 0; 2094 test.objectid = sk->max_objectid; 2095 test.type = sk->max_type; 2096 test.offset = sk->max_offset; 2097 if (btrfs_comp_cpu_keys(key, &test) >= 0) 2098 ret = 1; 2099 else if (key->offset < (u64)-1) 2100 key->offset++; 2101 else if (key->type < (u8)-1) { 2102 key->offset = 0; 2103 key->type++; 2104 } else if (key->objectid < (u64)-1) { 2105 key->offset = 0; 2106 key->type = 0; 2107 key->objectid++; 2108 } else 2109 ret = 1; 2110 out: 2111 /* 2112 * 0: all items from this leaf copied, continue with next 2113 * 1: * more items can be copied, but unused buffer is too small 2114 * * all items were found 2115 * Either way, it will stops the loop which iterates to the next 2116 * leaf 2117 * -EOVERFLOW: item was to large for buffer 2118 * -EFAULT: could not copy extent buffer back to userspace 2119 */ 2120 return ret; 2121 } 2122 2123 static noinline int search_ioctl(struct inode *inode, 2124 struct btrfs_ioctl_search_key *sk, 2125 size_t *buf_size, 2126 char __user *ubuf) 2127 { 2128 struct btrfs_fs_info *info = btrfs_sb(inode->i_sb); 2129 struct btrfs_root *root; 2130 struct btrfs_key key; 2131 struct btrfs_path *path; 2132 int ret; 2133 int num_found = 0; 2134 unsigned long sk_offset = 0; 2135 2136 if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) { 2137 *buf_size = sizeof(struct btrfs_ioctl_search_header); 2138 return -EOVERFLOW; 2139 } 2140 2141 path = btrfs_alloc_path(); 2142 if (!path) 2143 return -ENOMEM; 2144 2145 if (sk->tree_id == 0) { 2146 /* search the root of the inode that was passed */ 2147 root = BTRFS_I(inode)->root; 2148 } else { 2149 key.objectid = sk->tree_id; 2150 key.type = BTRFS_ROOT_ITEM_KEY; 2151 key.offset = (u64)-1; 2152 root = btrfs_read_fs_root_no_name(info, &key); 2153 if (IS_ERR(root)) { 2154 btrfs_free_path(path); 2155 return PTR_ERR(root); 2156 } 2157 } 2158 2159 key.objectid = sk->min_objectid; 2160 key.type = sk->min_type; 2161 key.offset = sk->min_offset; 2162 2163 while (1) { 2164 ret = btrfs_search_forward(root, &key, path, sk->min_transid); 2165 if (ret != 0) { 2166 if (ret > 0) 2167 ret = 0; 2168 goto err; 2169 } 2170 ret = copy_to_sk(path, &key, sk, buf_size, ubuf, 2171 &sk_offset, &num_found); 2172 btrfs_release_path(path); 2173 if (ret) 2174 break; 2175 2176 } 2177 if (ret > 0) 2178 ret = 0; 2179 err: 2180 sk->nr_items = num_found; 2181 btrfs_free_path(path); 2182 return ret; 2183 } 2184 2185 static noinline int btrfs_ioctl_tree_search(struct file *file, 2186 void __user *argp) 2187 { 2188 struct btrfs_ioctl_search_args __user *uargs; 2189 struct btrfs_ioctl_search_key sk; 2190 struct inode *inode; 2191 int ret; 2192 size_t buf_size; 2193 2194 if (!capable(CAP_SYS_ADMIN)) 2195 return -EPERM; 2196 2197 uargs = (struct btrfs_ioctl_search_args __user *)argp; 2198 2199 if (copy_from_user(&sk, &uargs->key, sizeof(sk))) 2200 return -EFAULT; 2201 2202 buf_size = sizeof(uargs->buf); 2203 2204 inode = file_inode(file); 2205 ret = search_ioctl(inode, &sk, &buf_size, uargs->buf); 2206 2207 /* 2208 * In the origin implementation an overflow is handled by returning a 2209 * search header with a len of zero, so reset ret. 2210 */ 2211 if (ret == -EOVERFLOW) 2212 ret = 0; 2213 2214 if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk))) 2215 ret = -EFAULT; 2216 return ret; 2217 } 2218 2219 static noinline int btrfs_ioctl_tree_search_v2(struct file *file, 2220 void __user *argp) 2221 { 2222 struct btrfs_ioctl_search_args_v2 __user *uarg; 2223 struct btrfs_ioctl_search_args_v2 args; 2224 struct inode *inode; 2225 int ret; 2226 size_t buf_size; 2227 const size_t buf_limit = SZ_16M; 2228 2229 if (!capable(CAP_SYS_ADMIN)) 2230 return -EPERM; 2231 2232 /* copy search header and buffer size */ 2233 uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp; 2234 if (copy_from_user(&args, uarg, sizeof(args))) 2235 return -EFAULT; 2236 2237 buf_size = args.buf_size; 2238 2239 /* limit result size to 16MB */ 2240 if (buf_size > buf_limit) 2241 buf_size = buf_limit; 2242 2243 inode = file_inode(file); 2244 ret = search_ioctl(inode, &args.key, &buf_size, 2245 (char __user *)(&uarg->buf[0])); 2246 if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key))) 2247 ret = -EFAULT; 2248 else if (ret == -EOVERFLOW && 2249 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size))) 2250 ret = -EFAULT; 2251 2252 return ret; 2253 } 2254 2255 /* 2256 * Search INODE_REFs to identify path name of 'dirid' directory 2257 * in a 'tree_id' tree. and sets path name to 'name'. 2258 */ 2259 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 2260 u64 tree_id, u64 dirid, char *name) 2261 { 2262 struct btrfs_root *root; 2263 struct btrfs_key key; 2264 char *ptr; 2265 int ret = -1; 2266 int slot; 2267 int len; 2268 int total_len = 0; 2269 struct btrfs_inode_ref *iref; 2270 struct extent_buffer *l; 2271 struct btrfs_path *path; 2272 2273 if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 2274 name[0]='\0'; 2275 return 0; 2276 } 2277 2278 path = btrfs_alloc_path(); 2279 if (!path) 2280 return -ENOMEM; 2281 2282 ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1]; 2283 2284 key.objectid = tree_id; 2285 key.type = BTRFS_ROOT_ITEM_KEY; 2286 key.offset = (u64)-1; 2287 root = btrfs_read_fs_root_no_name(info, &key); 2288 if (IS_ERR(root)) { 2289 ret = PTR_ERR(root); 2290 goto out; 2291 } 2292 2293 key.objectid = dirid; 2294 key.type = BTRFS_INODE_REF_KEY; 2295 key.offset = (u64)-1; 2296 2297 while (1) { 2298 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2299 if (ret < 0) 2300 goto out; 2301 else if (ret > 0) { 2302 ret = btrfs_previous_item(root, path, dirid, 2303 BTRFS_INODE_REF_KEY); 2304 if (ret < 0) 2305 goto out; 2306 else if (ret > 0) { 2307 ret = -ENOENT; 2308 goto out; 2309 } 2310 } 2311 2312 l = path->nodes[0]; 2313 slot = path->slots[0]; 2314 btrfs_item_key_to_cpu(l, &key, slot); 2315 2316 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 2317 len = btrfs_inode_ref_name_len(l, iref); 2318 ptr -= len + 1; 2319 total_len += len + 1; 2320 if (ptr < name) { 2321 ret = -ENAMETOOLONG; 2322 goto out; 2323 } 2324 2325 *(ptr + len) = '/'; 2326 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 2327 2328 if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 2329 break; 2330 2331 btrfs_release_path(path); 2332 key.objectid = key.offset; 2333 key.offset = (u64)-1; 2334 dirid = key.objectid; 2335 } 2336 memmove(name, ptr, total_len); 2337 name[total_len] = '\0'; 2338 ret = 0; 2339 out: 2340 btrfs_free_path(path); 2341 return ret; 2342 } 2343 2344 static int btrfs_search_path_in_tree_user(struct inode *inode, 2345 struct btrfs_ioctl_ino_lookup_user_args *args) 2346 { 2347 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 2348 struct super_block *sb = inode->i_sb; 2349 struct btrfs_key upper_limit = BTRFS_I(inode)->location; 2350 u64 treeid = BTRFS_I(inode)->root->root_key.objectid; 2351 u64 dirid = args->dirid; 2352 unsigned long item_off; 2353 unsigned long item_len; 2354 struct btrfs_inode_ref *iref; 2355 struct btrfs_root_ref *rref; 2356 struct btrfs_root *root; 2357 struct btrfs_path *path; 2358 struct btrfs_key key, key2; 2359 struct extent_buffer *leaf; 2360 struct inode *temp_inode; 2361 char *ptr; 2362 int slot; 2363 int len; 2364 int total_len = 0; 2365 int ret; 2366 2367 path = btrfs_alloc_path(); 2368 if (!path) 2369 return -ENOMEM; 2370 2371 /* 2372 * If the bottom subvolume does not exist directly under upper_limit, 2373 * construct the path in from the bottom up. 2374 */ 2375 if (dirid != upper_limit.objectid) { 2376 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1]; 2377 2378 key.objectid = treeid; 2379 key.type = BTRFS_ROOT_ITEM_KEY; 2380 key.offset = (u64)-1; 2381 root = btrfs_read_fs_root_no_name(fs_info, &key); 2382 if (IS_ERR(root)) { 2383 ret = PTR_ERR(root); 2384 goto out; 2385 } 2386 2387 key.objectid = dirid; 2388 key.type = BTRFS_INODE_REF_KEY; 2389 key.offset = (u64)-1; 2390 while (1) { 2391 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2392 if (ret < 0) { 2393 goto out; 2394 } else if (ret > 0) { 2395 ret = btrfs_previous_item(root, path, dirid, 2396 BTRFS_INODE_REF_KEY); 2397 if (ret < 0) { 2398 goto out; 2399 } else if (ret > 0) { 2400 ret = -ENOENT; 2401 goto out; 2402 } 2403 } 2404 2405 leaf = path->nodes[0]; 2406 slot = path->slots[0]; 2407 btrfs_item_key_to_cpu(leaf, &key, slot); 2408 2409 iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref); 2410 len = btrfs_inode_ref_name_len(leaf, iref); 2411 ptr -= len + 1; 2412 total_len += len + 1; 2413 if (ptr < args->path) { 2414 ret = -ENAMETOOLONG; 2415 goto out; 2416 } 2417 2418 *(ptr + len) = '/'; 2419 read_extent_buffer(leaf, ptr, 2420 (unsigned long)(iref + 1), len); 2421 2422 /* Check the read+exec permission of this directory */ 2423 ret = btrfs_previous_item(root, path, dirid, 2424 BTRFS_INODE_ITEM_KEY); 2425 if (ret < 0) { 2426 goto out; 2427 } else if (ret > 0) { 2428 ret = -ENOENT; 2429 goto out; 2430 } 2431 2432 leaf = path->nodes[0]; 2433 slot = path->slots[0]; 2434 btrfs_item_key_to_cpu(leaf, &key2, slot); 2435 if (key2.objectid != dirid) { 2436 ret = -ENOENT; 2437 goto out; 2438 } 2439 2440 temp_inode = btrfs_iget(sb, &key2, root, NULL); 2441 ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC); 2442 iput(temp_inode); 2443 if (ret) { 2444 ret = -EACCES; 2445 goto out; 2446 } 2447 2448 if (key.offset == upper_limit.objectid) 2449 break; 2450 if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) { 2451 ret = -EACCES; 2452 goto out; 2453 } 2454 2455 btrfs_release_path(path); 2456 key.objectid = key.offset; 2457 key.offset = (u64)-1; 2458 dirid = key.objectid; 2459 } 2460 2461 memmove(args->path, ptr, total_len); 2462 args->path[total_len] = '\0'; 2463 btrfs_release_path(path); 2464 } 2465 2466 /* Get the bottom subvolume's name from ROOT_REF */ 2467 root = fs_info->tree_root; 2468 key.objectid = treeid; 2469 key.type = BTRFS_ROOT_REF_KEY; 2470 key.offset = args->treeid; 2471 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2472 if (ret < 0) { 2473 goto out; 2474 } else if (ret > 0) { 2475 ret = -ENOENT; 2476 goto out; 2477 } 2478 2479 leaf = path->nodes[0]; 2480 slot = path->slots[0]; 2481 btrfs_item_key_to_cpu(leaf, &key, slot); 2482 2483 item_off = btrfs_item_ptr_offset(leaf, slot); 2484 item_len = btrfs_item_size_nr(leaf, slot); 2485 /* Check if dirid in ROOT_REF corresponds to passed dirid */ 2486 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 2487 if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) { 2488 ret = -EINVAL; 2489 goto out; 2490 } 2491 2492 /* Copy subvolume's name */ 2493 item_off += sizeof(struct btrfs_root_ref); 2494 item_len -= sizeof(struct btrfs_root_ref); 2495 read_extent_buffer(leaf, args->name, item_off, item_len); 2496 args->name[item_len] = 0; 2497 2498 out: 2499 btrfs_free_path(path); 2500 return ret; 2501 } 2502 2503 static noinline int btrfs_ioctl_ino_lookup(struct file *file, 2504 void __user *argp) 2505 { 2506 struct btrfs_ioctl_ino_lookup_args *args; 2507 struct inode *inode; 2508 int ret = 0; 2509 2510 args = memdup_user(argp, sizeof(*args)); 2511 if (IS_ERR(args)) 2512 return PTR_ERR(args); 2513 2514 inode = file_inode(file); 2515 2516 /* 2517 * Unprivileged query to obtain the containing subvolume root id. The 2518 * path is reset so it's consistent with btrfs_search_path_in_tree. 2519 */ 2520 if (args->treeid == 0) 2521 args->treeid = BTRFS_I(inode)->root->root_key.objectid; 2522 2523 if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) { 2524 args->name[0] = 0; 2525 goto out; 2526 } 2527 2528 if (!capable(CAP_SYS_ADMIN)) { 2529 ret = -EPERM; 2530 goto out; 2531 } 2532 2533 ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 2534 args->treeid, args->objectid, 2535 args->name); 2536 2537 out: 2538 if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2539 ret = -EFAULT; 2540 2541 kfree(args); 2542 return ret; 2543 } 2544 2545 /* 2546 * Version of ino_lookup ioctl (unprivileged) 2547 * 2548 * The main differences from ino_lookup ioctl are: 2549 * 2550 * 1. Read + Exec permission will be checked using inode_permission() during 2551 * path construction. -EACCES will be returned in case of failure. 2552 * 2. Path construction will be stopped at the inode number which corresponds 2553 * to the fd with which this ioctl is called. If constructed path does not 2554 * exist under fd's inode, -EACCES will be returned. 2555 * 3. The name of bottom subvolume is also searched and filled. 2556 */ 2557 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp) 2558 { 2559 struct btrfs_ioctl_ino_lookup_user_args *args; 2560 struct inode *inode; 2561 int ret; 2562 2563 args = memdup_user(argp, sizeof(*args)); 2564 if (IS_ERR(args)) 2565 return PTR_ERR(args); 2566 2567 inode = file_inode(file); 2568 2569 if (args->dirid == BTRFS_FIRST_FREE_OBJECTID && 2570 BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) { 2571 /* 2572 * The subvolume does not exist under fd with which this is 2573 * called 2574 */ 2575 kfree(args); 2576 return -EACCES; 2577 } 2578 2579 ret = btrfs_search_path_in_tree_user(inode, args); 2580 2581 if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2582 ret = -EFAULT; 2583 2584 kfree(args); 2585 return ret; 2586 } 2587 2588 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */ 2589 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp) 2590 { 2591 struct btrfs_ioctl_get_subvol_info_args *subvol_info; 2592 struct btrfs_fs_info *fs_info; 2593 struct btrfs_root *root; 2594 struct btrfs_path *path; 2595 struct btrfs_key key; 2596 struct btrfs_root_item *root_item; 2597 struct btrfs_root_ref *rref; 2598 struct extent_buffer *leaf; 2599 unsigned long item_off; 2600 unsigned long item_len; 2601 struct inode *inode; 2602 int slot; 2603 int ret = 0; 2604 2605 path = btrfs_alloc_path(); 2606 if (!path) 2607 return -ENOMEM; 2608 2609 subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL); 2610 if (!subvol_info) { 2611 btrfs_free_path(path); 2612 return -ENOMEM; 2613 } 2614 2615 inode = file_inode(file); 2616 fs_info = BTRFS_I(inode)->root->fs_info; 2617 2618 /* Get root_item of inode's subvolume */ 2619 key.objectid = BTRFS_I(inode)->root->root_key.objectid; 2620 key.type = BTRFS_ROOT_ITEM_KEY; 2621 key.offset = (u64)-1; 2622 root = btrfs_read_fs_root_no_name(fs_info, &key); 2623 if (IS_ERR(root)) { 2624 ret = PTR_ERR(root); 2625 goto out; 2626 } 2627 root_item = &root->root_item; 2628 2629 subvol_info->treeid = key.objectid; 2630 2631 subvol_info->generation = btrfs_root_generation(root_item); 2632 subvol_info->flags = btrfs_root_flags(root_item); 2633 2634 memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE); 2635 memcpy(subvol_info->parent_uuid, root_item->parent_uuid, 2636 BTRFS_UUID_SIZE); 2637 memcpy(subvol_info->received_uuid, root_item->received_uuid, 2638 BTRFS_UUID_SIZE); 2639 2640 subvol_info->ctransid = btrfs_root_ctransid(root_item); 2641 subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime); 2642 subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime); 2643 2644 subvol_info->otransid = btrfs_root_otransid(root_item); 2645 subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime); 2646 subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime); 2647 2648 subvol_info->stransid = btrfs_root_stransid(root_item); 2649 subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime); 2650 subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime); 2651 2652 subvol_info->rtransid = btrfs_root_rtransid(root_item); 2653 subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime); 2654 subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime); 2655 2656 if (key.objectid != BTRFS_FS_TREE_OBJECTID) { 2657 /* Search root tree for ROOT_BACKREF of this subvolume */ 2658 root = fs_info->tree_root; 2659 2660 key.type = BTRFS_ROOT_BACKREF_KEY; 2661 key.offset = 0; 2662 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2663 if (ret < 0) { 2664 goto out; 2665 } else if (path->slots[0] >= 2666 btrfs_header_nritems(path->nodes[0])) { 2667 ret = btrfs_next_leaf(root, path); 2668 if (ret < 0) { 2669 goto out; 2670 } else if (ret > 0) { 2671 ret = -EUCLEAN; 2672 goto out; 2673 } 2674 } 2675 2676 leaf = path->nodes[0]; 2677 slot = path->slots[0]; 2678 btrfs_item_key_to_cpu(leaf, &key, slot); 2679 if (key.objectid == subvol_info->treeid && 2680 key.type == BTRFS_ROOT_BACKREF_KEY) { 2681 subvol_info->parent_id = key.offset; 2682 2683 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 2684 subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref); 2685 2686 item_off = btrfs_item_ptr_offset(leaf, slot) 2687 + sizeof(struct btrfs_root_ref); 2688 item_len = btrfs_item_size_nr(leaf, slot) 2689 - sizeof(struct btrfs_root_ref); 2690 read_extent_buffer(leaf, subvol_info->name, 2691 item_off, item_len); 2692 } else { 2693 ret = -ENOENT; 2694 goto out; 2695 } 2696 } 2697 2698 if (copy_to_user(argp, subvol_info, sizeof(*subvol_info))) 2699 ret = -EFAULT; 2700 2701 out: 2702 btrfs_free_path(path); 2703 kzfree(subvol_info); 2704 return ret; 2705 } 2706 2707 /* 2708 * Return ROOT_REF information of the subvolume containing this inode 2709 * except the subvolume name. 2710 */ 2711 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp) 2712 { 2713 struct btrfs_ioctl_get_subvol_rootref_args *rootrefs; 2714 struct btrfs_root_ref *rref; 2715 struct btrfs_root *root; 2716 struct btrfs_path *path; 2717 struct btrfs_key key; 2718 struct extent_buffer *leaf; 2719 struct inode *inode; 2720 u64 objectid; 2721 int slot; 2722 int ret; 2723 u8 found; 2724 2725 path = btrfs_alloc_path(); 2726 if (!path) 2727 return -ENOMEM; 2728 2729 rootrefs = memdup_user(argp, sizeof(*rootrefs)); 2730 if (IS_ERR(rootrefs)) { 2731 btrfs_free_path(path); 2732 return PTR_ERR(rootrefs); 2733 } 2734 2735 inode = file_inode(file); 2736 root = BTRFS_I(inode)->root->fs_info->tree_root; 2737 objectid = BTRFS_I(inode)->root->root_key.objectid; 2738 2739 key.objectid = objectid; 2740 key.type = BTRFS_ROOT_REF_KEY; 2741 key.offset = rootrefs->min_treeid; 2742 found = 0; 2743 2744 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2745 if (ret < 0) { 2746 goto out; 2747 } else if (path->slots[0] >= 2748 btrfs_header_nritems(path->nodes[0])) { 2749 ret = btrfs_next_leaf(root, path); 2750 if (ret < 0) { 2751 goto out; 2752 } else if (ret > 0) { 2753 ret = -EUCLEAN; 2754 goto out; 2755 } 2756 } 2757 while (1) { 2758 leaf = path->nodes[0]; 2759 slot = path->slots[0]; 2760 2761 btrfs_item_key_to_cpu(leaf, &key, slot); 2762 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) { 2763 ret = 0; 2764 goto out; 2765 } 2766 2767 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) { 2768 ret = -EOVERFLOW; 2769 goto out; 2770 } 2771 2772 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 2773 rootrefs->rootref[found].treeid = key.offset; 2774 rootrefs->rootref[found].dirid = 2775 btrfs_root_ref_dirid(leaf, rref); 2776 found++; 2777 2778 ret = btrfs_next_item(root, path); 2779 if (ret < 0) { 2780 goto out; 2781 } else if (ret > 0) { 2782 ret = -EUCLEAN; 2783 goto out; 2784 } 2785 } 2786 2787 out: 2788 if (!ret || ret == -EOVERFLOW) { 2789 rootrefs->num_items = found; 2790 /* update min_treeid for next search */ 2791 if (found) 2792 rootrefs->min_treeid = 2793 rootrefs->rootref[found - 1].treeid + 1; 2794 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs))) 2795 ret = -EFAULT; 2796 } 2797 2798 kfree(rootrefs); 2799 btrfs_free_path(path); 2800 2801 return ret; 2802 } 2803 2804 static noinline int btrfs_ioctl_snap_destroy(struct file *file, 2805 void __user *arg) 2806 { 2807 struct dentry *parent = file->f_path.dentry; 2808 struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb); 2809 struct dentry *dentry; 2810 struct inode *dir = d_inode(parent); 2811 struct inode *inode; 2812 struct btrfs_root *root = BTRFS_I(dir)->root; 2813 struct btrfs_root *dest = NULL; 2814 struct btrfs_ioctl_vol_args *vol_args; 2815 int namelen; 2816 int err = 0; 2817 2818 if (!S_ISDIR(dir->i_mode)) 2819 return -ENOTDIR; 2820 2821 vol_args = memdup_user(arg, sizeof(*vol_args)); 2822 if (IS_ERR(vol_args)) 2823 return PTR_ERR(vol_args); 2824 2825 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2826 namelen = strlen(vol_args->name); 2827 if (strchr(vol_args->name, '/') || 2828 strncmp(vol_args->name, "..", namelen) == 0) { 2829 err = -EINVAL; 2830 goto out; 2831 } 2832 2833 err = mnt_want_write_file(file); 2834 if (err) 2835 goto out; 2836 2837 2838 err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 2839 if (err == -EINTR) 2840 goto out_drop_write; 2841 dentry = lookup_one_len(vol_args->name, parent, namelen); 2842 if (IS_ERR(dentry)) { 2843 err = PTR_ERR(dentry); 2844 goto out_unlock_dir; 2845 } 2846 2847 if (d_really_is_negative(dentry)) { 2848 err = -ENOENT; 2849 goto out_dput; 2850 } 2851 2852 inode = d_inode(dentry); 2853 dest = BTRFS_I(inode)->root; 2854 if (!capable(CAP_SYS_ADMIN)) { 2855 /* 2856 * Regular user. Only allow this with a special mount 2857 * option, when the user has write+exec access to the 2858 * subvol root, and when rmdir(2) would have been 2859 * allowed. 2860 * 2861 * Note that this is _not_ check that the subvol is 2862 * empty or doesn't contain data that we wouldn't 2863 * otherwise be able to delete. 2864 * 2865 * Users who want to delete empty subvols should try 2866 * rmdir(2). 2867 */ 2868 err = -EPERM; 2869 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED)) 2870 goto out_dput; 2871 2872 /* 2873 * Do not allow deletion if the parent dir is the same 2874 * as the dir to be deleted. That means the ioctl 2875 * must be called on the dentry referencing the root 2876 * of the subvol, not a random directory contained 2877 * within it. 2878 */ 2879 err = -EINVAL; 2880 if (root == dest) 2881 goto out_dput; 2882 2883 err = inode_permission(inode, MAY_WRITE | MAY_EXEC); 2884 if (err) 2885 goto out_dput; 2886 } 2887 2888 /* check if subvolume may be deleted by a user */ 2889 err = btrfs_may_delete(dir, dentry, 1); 2890 if (err) 2891 goto out_dput; 2892 2893 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 2894 err = -EINVAL; 2895 goto out_dput; 2896 } 2897 2898 inode_lock(inode); 2899 err = btrfs_delete_subvolume(dir, dentry); 2900 inode_unlock(inode); 2901 if (!err) 2902 d_delete(dentry); 2903 2904 out_dput: 2905 dput(dentry); 2906 out_unlock_dir: 2907 inode_unlock(dir); 2908 out_drop_write: 2909 mnt_drop_write_file(file); 2910 out: 2911 kfree(vol_args); 2912 return err; 2913 } 2914 2915 static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 2916 { 2917 struct inode *inode = file_inode(file); 2918 struct btrfs_root *root = BTRFS_I(inode)->root; 2919 struct btrfs_ioctl_defrag_range_args *range; 2920 int ret; 2921 2922 ret = mnt_want_write_file(file); 2923 if (ret) 2924 return ret; 2925 2926 if (btrfs_root_readonly(root)) { 2927 ret = -EROFS; 2928 goto out; 2929 } 2930 2931 switch (inode->i_mode & S_IFMT) { 2932 case S_IFDIR: 2933 if (!capable(CAP_SYS_ADMIN)) { 2934 ret = -EPERM; 2935 goto out; 2936 } 2937 ret = btrfs_defrag_root(root); 2938 break; 2939 case S_IFREG: 2940 if (!(file->f_mode & FMODE_WRITE)) { 2941 ret = -EINVAL; 2942 goto out; 2943 } 2944 2945 range = kzalloc(sizeof(*range), GFP_KERNEL); 2946 if (!range) { 2947 ret = -ENOMEM; 2948 goto out; 2949 } 2950 2951 if (argp) { 2952 if (copy_from_user(range, argp, 2953 sizeof(*range))) { 2954 ret = -EFAULT; 2955 kfree(range); 2956 goto out; 2957 } 2958 /* compression requires us to start the IO */ 2959 if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 2960 range->flags |= BTRFS_DEFRAG_RANGE_START_IO; 2961 range->extent_thresh = (u32)-1; 2962 } 2963 } else { 2964 /* the rest are all set to zero by kzalloc */ 2965 range->len = (u64)-1; 2966 } 2967 ret = btrfs_defrag_file(file_inode(file), file, 2968 range, BTRFS_OLDEST_GENERATION, 0); 2969 if (ret > 0) 2970 ret = 0; 2971 kfree(range); 2972 break; 2973 default: 2974 ret = -EINVAL; 2975 } 2976 out: 2977 mnt_drop_write_file(file); 2978 return ret; 2979 } 2980 2981 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg) 2982 { 2983 struct btrfs_ioctl_vol_args *vol_args; 2984 int ret; 2985 2986 if (!capable(CAP_SYS_ADMIN)) 2987 return -EPERM; 2988 2989 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) 2990 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 2991 2992 vol_args = memdup_user(arg, sizeof(*vol_args)); 2993 if (IS_ERR(vol_args)) { 2994 ret = PTR_ERR(vol_args); 2995 goto out; 2996 } 2997 2998 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2999 ret = btrfs_init_new_device(fs_info, vol_args->name); 3000 3001 if (!ret) 3002 btrfs_info(fs_info, "disk added %s", vol_args->name); 3003 3004 kfree(vol_args); 3005 out: 3006 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 3007 return ret; 3008 } 3009 3010 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg) 3011 { 3012 struct inode *inode = file_inode(file); 3013 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3014 struct btrfs_ioctl_vol_args_v2 *vol_args; 3015 int ret; 3016 3017 if (!capable(CAP_SYS_ADMIN)) 3018 return -EPERM; 3019 3020 ret = mnt_want_write_file(file); 3021 if (ret) 3022 return ret; 3023 3024 vol_args = memdup_user(arg, sizeof(*vol_args)); 3025 if (IS_ERR(vol_args)) { 3026 ret = PTR_ERR(vol_args); 3027 goto err_drop; 3028 } 3029 3030 /* Check for compatibility reject unknown flags */ 3031 if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) { 3032 ret = -EOPNOTSUPP; 3033 goto out; 3034 } 3035 3036 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 3037 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3038 goto out; 3039 } 3040 3041 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) { 3042 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid); 3043 } else { 3044 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 3045 ret = btrfs_rm_device(fs_info, vol_args->name, 0); 3046 } 3047 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 3048 3049 if (!ret) { 3050 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) 3051 btrfs_info(fs_info, "device deleted: id %llu", 3052 vol_args->devid); 3053 else 3054 btrfs_info(fs_info, "device deleted: %s", 3055 vol_args->name); 3056 } 3057 out: 3058 kfree(vol_args); 3059 err_drop: 3060 mnt_drop_write_file(file); 3061 return ret; 3062 } 3063 3064 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 3065 { 3066 struct inode *inode = file_inode(file); 3067 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3068 struct btrfs_ioctl_vol_args *vol_args; 3069 int ret; 3070 3071 if (!capable(CAP_SYS_ADMIN)) 3072 return -EPERM; 3073 3074 ret = mnt_want_write_file(file); 3075 if (ret) 3076 return ret; 3077 3078 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 3079 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3080 goto out_drop_write; 3081 } 3082 3083 vol_args = memdup_user(arg, sizeof(*vol_args)); 3084 if (IS_ERR(vol_args)) { 3085 ret = PTR_ERR(vol_args); 3086 goto out; 3087 } 3088 3089 vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 3090 ret = btrfs_rm_device(fs_info, vol_args->name, 0); 3091 3092 if (!ret) 3093 btrfs_info(fs_info, "disk deleted %s", vol_args->name); 3094 kfree(vol_args); 3095 out: 3096 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 3097 out_drop_write: 3098 mnt_drop_write_file(file); 3099 3100 return ret; 3101 } 3102 3103 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info, 3104 void __user *arg) 3105 { 3106 struct btrfs_ioctl_fs_info_args *fi_args; 3107 struct btrfs_device *device; 3108 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3109 int ret = 0; 3110 3111 fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL); 3112 if (!fi_args) 3113 return -ENOMEM; 3114 3115 rcu_read_lock(); 3116 fi_args->num_devices = fs_devices->num_devices; 3117 3118 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 3119 if (device->devid > fi_args->max_id) 3120 fi_args->max_id = device->devid; 3121 } 3122 rcu_read_unlock(); 3123 3124 memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid)); 3125 fi_args->nodesize = fs_info->nodesize; 3126 fi_args->sectorsize = fs_info->sectorsize; 3127 fi_args->clone_alignment = fs_info->sectorsize; 3128 3129 if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 3130 ret = -EFAULT; 3131 3132 kfree(fi_args); 3133 return ret; 3134 } 3135 3136 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info, 3137 void __user *arg) 3138 { 3139 struct btrfs_ioctl_dev_info_args *di_args; 3140 struct btrfs_device *dev; 3141 int ret = 0; 3142 char *s_uuid = NULL; 3143 3144 di_args = memdup_user(arg, sizeof(*di_args)); 3145 if (IS_ERR(di_args)) 3146 return PTR_ERR(di_args); 3147 3148 if (!btrfs_is_empty_uuid(di_args->uuid)) 3149 s_uuid = di_args->uuid; 3150 3151 rcu_read_lock(); 3152 dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL); 3153 3154 if (!dev) { 3155 ret = -ENODEV; 3156 goto out; 3157 } 3158 3159 di_args->devid = dev->devid; 3160 di_args->bytes_used = btrfs_device_get_bytes_used(dev); 3161 di_args->total_bytes = btrfs_device_get_total_bytes(dev); 3162 memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 3163 if (dev->name) { 3164 struct rcu_string *name; 3165 3166 name = rcu_dereference(dev->name); 3167 strncpy(di_args->path, name->str, sizeof(di_args->path) - 1); 3168 di_args->path[sizeof(di_args->path) - 1] = 0; 3169 } else { 3170 di_args->path[0] = '\0'; 3171 } 3172 3173 out: 3174 rcu_read_unlock(); 3175 if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 3176 ret = -EFAULT; 3177 3178 kfree(di_args); 3179 return ret; 3180 } 3181 3182 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index) 3183 { 3184 struct page *page; 3185 3186 page = grab_cache_page(inode->i_mapping, index); 3187 if (!page) 3188 return ERR_PTR(-ENOMEM); 3189 3190 if (!PageUptodate(page)) { 3191 int ret; 3192 3193 ret = btrfs_readpage(NULL, page); 3194 if (ret) 3195 return ERR_PTR(ret); 3196 lock_page(page); 3197 if (!PageUptodate(page)) { 3198 unlock_page(page); 3199 put_page(page); 3200 return ERR_PTR(-EIO); 3201 } 3202 if (page->mapping != inode->i_mapping) { 3203 unlock_page(page); 3204 put_page(page); 3205 return ERR_PTR(-EAGAIN); 3206 } 3207 } 3208 3209 return page; 3210 } 3211 3212 static int gather_extent_pages(struct inode *inode, struct page **pages, 3213 int num_pages, u64 off) 3214 { 3215 int i; 3216 pgoff_t index = off >> PAGE_SHIFT; 3217 3218 for (i = 0; i < num_pages; i++) { 3219 again: 3220 pages[i] = extent_same_get_page(inode, index + i); 3221 if (IS_ERR(pages[i])) { 3222 int err = PTR_ERR(pages[i]); 3223 3224 if (err == -EAGAIN) 3225 goto again; 3226 pages[i] = NULL; 3227 return err; 3228 } 3229 } 3230 return 0; 3231 } 3232 3233 static int lock_extent_range(struct inode *inode, u64 off, u64 len, 3234 bool retry_range_locking) 3235 { 3236 /* 3237 * Do any pending delalloc/csum calculations on inode, one way or 3238 * another, and lock file content. 3239 * The locking order is: 3240 * 3241 * 1) pages 3242 * 2) range in the inode's io tree 3243 */ 3244 while (1) { 3245 struct btrfs_ordered_extent *ordered; 3246 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 3247 ordered = btrfs_lookup_first_ordered_extent(inode, 3248 off + len - 1); 3249 if ((!ordered || 3250 ordered->file_offset + ordered->len <= off || 3251 ordered->file_offset >= off + len) && 3252 !test_range_bit(&BTRFS_I(inode)->io_tree, off, 3253 off + len - 1, EXTENT_DELALLOC, 0, NULL)) { 3254 if (ordered) 3255 btrfs_put_ordered_extent(ordered); 3256 break; 3257 } 3258 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 3259 if (ordered) 3260 btrfs_put_ordered_extent(ordered); 3261 if (!retry_range_locking) 3262 return -EAGAIN; 3263 btrfs_wait_ordered_range(inode, off, len); 3264 } 3265 return 0; 3266 } 3267 3268 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2) 3269 { 3270 inode_unlock(inode1); 3271 inode_unlock(inode2); 3272 } 3273 3274 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2) 3275 { 3276 if (inode1 < inode2) 3277 swap(inode1, inode2); 3278 3279 inode_lock_nested(inode1, I_MUTEX_PARENT); 3280 inode_lock_nested(inode2, I_MUTEX_CHILD); 3281 } 3282 3283 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1, 3284 struct inode *inode2, u64 loff2, u64 len) 3285 { 3286 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 3287 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 3288 } 3289 3290 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1, 3291 struct inode *inode2, u64 loff2, u64 len, 3292 bool retry_range_locking) 3293 { 3294 int ret; 3295 3296 if (inode1 < inode2) { 3297 swap(inode1, inode2); 3298 swap(loff1, loff2); 3299 } 3300 ret = lock_extent_range(inode1, loff1, len, retry_range_locking); 3301 if (ret) 3302 return ret; 3303 ret = lock_extent_range(inode2, loff2, len, retry_range_locking); 3304 if (ret) 3305 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, 3306 loff1 + len - 1); 3307 return ret; 3308 } 3309 3310 struct cmp_pages { 3311 int num_pages; 3312 struct page **src_pages; 3313 struct page **dst_pages; 3314 }; 3315 3316 static void btrfs_cmp_data_free(struct cmp_pages *cmp) 3317 { 3318 int i; 3319 struct page *pg; 3320 3321 for (i = 0; i < cmp->num_pages; i++) { 3322 pg = cmp->src_pages[i]; 3323 if (pg) { 3324 unlock_page(pg); 3325 put_page(pg); 3326 } 3327 pg = cmp->dst_pages[i]; 3328 if (pg) { 3329 unlock_page(pg); 3330 put_page(pg); 3331 } 3332 } 3333 } 3334 3335 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff, 3336 struct inode *dst, u64 dst_loff, 3337 u64 len, struct cmp_pages *cmp) 3338 { 3339 int ret; 3340 int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT; 3341 3342 cmp->num_pages = num_pages; 3343 3344 ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff); 3345 if (ret) 3346 goto out; 3347 3348 ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff); 3349 3350 out: 3351 if (ret) 3352 btrfs_cmp_data_free(cmp); 3353 return ret; 3354 } 3355 3356 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp) 3357 { 3358 int ret = 0; 3359 int i; 3360 struct page *src_page, *dst_page; 3361 unsigned int cmp_len = PAGE_SIZE; 3362 void *addr, *dst_addr; 3363 3364 i = 0; 3365 while (len) { 3366 if (len < PAGE_SIZE) 3367 cmp_len = len; 3368 3369 BUG_ON(i >= cmp->num_pages); 3370 3371 src_page = cmp->src_pages[i]; 3372 dst_page = cmp->dst_pages[i]; 3373 ASSERT(PageLocked(src_page)); 3374 ASSERT(PageLocked(dst_page)); 3375 3376 addr = kmap_atomic(src_page); 3377 dst_addr = kmap_atomic(dst_page); 3378 3379 flush_dcache_page(src_page); 3380 flush_dcache_page(dst_page); 3381 3382 if (memcmp(addr, dst_addr, cmp_len)) 3383 ret = -EBADE; 3384 3385 kunmap_atomic(addr); 3386 kunmap_atomic(dst_addr); 3387 3388 if (ret) 3389 break; 3390 3391 len -= cmp_len; 3392 i++; 3393 } 3394 3395 return ret; 3396 } 3397 3398 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen, 3399 u64 olen) 3400 { 3401 u64 len = *plen; 3402 u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize; 3403 3404 if (off + olen > inode->i_size || off + olen < off) 3405 return -EINVAL; 3406 3407 /* if we extend to eof, continue to block boundary */ 3408 if (off + len == inode->i_size) 3409 *plen = len = ALIGN(inode->i_size, bs) - off; 3410 3411 /* Check that we are block aligned - btrfs_clone() requires this */ 3412 if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs)) 3413 return -EINVAL; 3414 3415 return 0; 3416 } 3417 3418 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen, 3419 struct inode *dst, u64 dst_loff, 3420 struct cmp_pages *cmp) 3421 { 3422 int ret; 3423 u64 len = olen; 3424 bool same_inode = (src == dst); 3425 u64 same_lock_start = 0; 3426 u64 same_lock_len = 0; 3427 3428 ret = extent_same_check_offsets(src, loff, &len, olen); 3429 if (ret) 3430 return ret; 3431 3432 ret = extent_same_check_offsets(dst, dst_loff, &len, olen); 3433 if (ret) 3434 return ret; 3435 3436 if (same_inode) { 3437 /* 3438 * Single inode case wants the same checks, except we 3439 * don't want our length pushed out past i_size as 3440 * comparing that data range makes no sense. 3441 * 3442 * extent_same_check_offsets() will do this for an 3443 * unaligned length at i_size, so catch it here and 3444 * reject the request. 3445 * 3446 * This effectively means we require aligned extents 3447 * for the single-inode case, whereas the other cases 3448 * allow an unaligned length so long as it ends at 3449 * i_size. 3450 */ 3451 if (len != olen) 3452 return -EINVAL; 3453 3454 /* Check for overlapping ranges */ 3455 if (dst_loff + len > loff && dst_loff < loff + len) 3456 return -EINVAL; 3457 3458 same_lock_start = min_t(u64, loff, dst_loff); 3459 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start; 3460 } 3461 3462 again: 3463 ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp); 3464 if (ret) 3465 return ret; 3466 3467 if (same_inode) 3468 ret = lock_extent_range(src, same_lock_start, same_lock_len, 3469 false); 3470 else 3471 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len, 3472 false); 3473 /* 3474 * If one of the inodes has dirty pages in the respective range or 3475 * ordered extents, we need to flush dellaloc and wait for all ordered 3476 * extents in the range. We must unlock the pages and the ranges in the 3477 * io trees to avoid deadlocks when flushing delalloc (requires locking 3478 * pages) and when waiting for ordered extents to complete (they require 3479 * range locking). 3480 */ 3481 if (ret == -EAGAIN) { 3482 /* 3483 * Ranges in the io trees already unlocked. Now unlock all 3484 * pages before waiting for all IO to complete. 3485 */ 3486 btrfs_cmp_data_free(cmp); 3487 if (same_inode) { 3488 btrfs_wait_ordered_range(src, same_lock_start, 3489 same_lock_len); 3490 } else { 3491 btrfs_wait_ordered_range(src, loff, len); 3492 btrfs_wait_ordered_range(dst, dst_loff, len); 3493 } 3494 goto again; 3495 } 3496 ASSERT(ret == 0); 3497 if (WARN_ON(ret)) { 3498 /* ranges in the io trees already unlocked */ 3499 btrfs_cmp_data_free(cmp); 3500 return ret; 3501 } 3502 3503 /* pass original length for comparison so we stay within i_size */ 3504 ret = btrfs_cmp_data(olen, cmp); 3505 if (ret == 0) 3506 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1); 3507 3508 if (same_inode) 3509 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start, 3510 same_lock_start + same_lock_len - 1); 3511 else 3512 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len); 3513 3514 btrfs_cmp_data_free(cmp); 3515 3516 return ret; 3517 } 3518 3519 #define BTRFS_MAX_DEDUPE_LEN SZ_16M 3520 3521 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen, 3522 struct inode *dst, u64 dst_loff) 3523 { 3524 int ret; 3525 struct cmp_pages cmp; 3526 int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT; 3527 bool same_inode = (src == dst); 3528 u64 i, tail_len, chunk_count; 3529 3530 if (olen == 0) 3531 return 0; 3532 3533 if (same_inode) 3534 inode_lock(src); 3535 else 3536 btrfs_double_inode_lock(src, dst); 3537 3538 /* don't make the dst file partly checksummed */ 3539 if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 3540 (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) { 3541 ret = -EINVAL; 3542 goto out_unlock; 3543 } 3544 3545 tail_len = olen % BTRFS_MAX_DEDUPE_LEN; 3546 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN); 3547 if (chunk_count == 0) 3548 num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT; 3549 3550 /* 3551 * If deduping ranges in the same inode, locking rules make it 3552 * mandatory to always lock pages in ascending order to avoid deadlocks 3553 * with concurrent tasks (such as starting writeback/delalloc). 3554 */ 3555 if (same_inode && dst_loff < loff) 3556 swap(loff, dst_loff); 3557 3558 /* 3559 * We must gather up all the pages before we initiate our extent 3560 * locking. We use an array for the page pointers. Size of the array is 3561 * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN. 3562 */ 3563 cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *), 3564 GFP_KERNEL | __GFP_ZERO); 3565 cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *), 3566 GFP_KERNEL | __GFP_ZERO); 3567 if (!cmp.src_pages || !cmp.dst_pages) { 3568 ret = -ENOMEM; 3569 goto out_free; 3570 } 3571 3572 for (i = 0; i < chunk_count; i++) { 3573 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN, 3574 dst, dst_loff, &cmp); 3575 if (ret) 3576 goto out_unlock; 3577 3578 loff += BTRFS_MAX_DEDUPE_LEN; 3579 dst_loff += BTRFS_MAX_DEDUPE_LEN; 3580 } 3581 3582 if (tail_len > 0) 3583 ret = btrfs_extent_same_range(src, loff, tail_len, dst, 3584 dst_loff, &cmp); 3585 3586 out_unlock: 3587 if (same_inode) 3588 inode_unlock(src); 3589 else 3590 btrfs_double_inode_unlock(src, dst); 3591 3592 out_free: 3593 kvfree(cmp.src_pages); 3594 kvfree(cmp.dst_pages); 3595 3596 return ret; 3597 } 3598 3599 ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen, 3600 struct file *dst_file, u64 dst_loff) 3601 { 3602 struct inode *src = file_inode(src_file); 3603 struct inode *dst = file_inode(dst_file); 3604 u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize; 3605 ssize_t res; 3606 3607 if (WARN_ON_ONCE(bs < PAGE_SIZE)) { 3608 /* 3609 * Btrfs does not support blocksize < page_size. As a 3610 * result, btrfs_cmp_data() won't correctly handle 3611 * this situation without an update. 3612 */ 3613 return -EINVAL; 3614 } 3615 3616 res = btrfs_extent_same(src, loff, olen, dst, dst_loff); 3617 if (res) 3618 return res; 3619 return olen; 3620 } 3621 3622 static int clone_finish_inode_update(struct btrfs_trans_handle *trans, 3623 struct inode *inode, 3624 u64 endoff, 3625 const u64 destoff, 3626 const u64 olen, 3627 int no_time_update) 3628 { 3629 struct btrfs_root *root = BTRFS_I(inode)->root; 3630 int ret; 3631 3632 inode_inc_iversion(inode); 3633 if (!no_time_update) 3634 inode->i_mtime = inode->i_ctime = current_time(inode); 3635 /* 3636 * We round up to the block size at eof when determining which 3637 * extents to clone above, but shouldn't round up the file size. 3638 */ 3639 if (endoff > destoff + olen) 3640 endoff = destoff + olen; 3641 if (endoff > inode->i_size) 3642 btrfs_i_size_write(BTRFS_I(inode), endoff); 3643 3644 ret = btrfs_update_inode(trans, root, inode); 3645 if (ret) { 3646 btrfs_abort_transaction(trans, ret); 3647 btrfs_end_transaction(trans); 3648 goto out; 3649 } 3650 ret = btrfs_end_transaction(trans); 3651 out: 3652 return ret; 3653 } 3654 3655 static void clone_update_extent_map(struct btrfs_inode *inode, 3656 const struct btrfs_trans_handle *trans, 3657 const struct btrfs_path *path, 3658 const u64 hole_offset, 3659 const u64 hole_len) 3660 { 3661 struct extent_map_tree *em_tree = &inode->extent_tree; 3662 struct extent_map *em; 3663 int ret; 3664 3665 em = alloc_extent_map(); 3666 if (!em) { 3667 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); 3668 return; 3669 } 3670 3671 if (path) { 3672 struct btrfs_file_extent_item *fi; 3673 3674 fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 3675 struct btrfs_file_extent_item); 3676 btrfs_extent_item_to_extent_map(inode, path, fi, false, em); 3677 em->generation = -1; 3678 if (btrfs_file_extent_type(path->nodes[0], fi) == 3679 BTRFS_FILE_EXTENT_INLINE) 3680 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, 3681 &inode->runtime_flags); 3682 } else { 3683 em->start = hole_offset; 3684 em->len = hole_len; 3685 em->ram_bytes = em->len; 3686 em->orig_start = hole_offset; 3687 em->block_start = EXTENT_MAP_HOLE; 3688 em->block_len = 0; 3689 em->orig_block_len = 0; 3690 em->compress_type = BTRFS_COMPRESS_NONE; 3691 em->generation = trans->transid; 3692 } 3693 3694 while (1) { 3695 write_lock(&em_tree->lock); 3696 ret = add_extent_mapping(em_tree, em, 1); 3697 write_unlock(&em_tree->lock); 3698 if (ret != -EEXIST) { 3699 free_extent_map(em); 3700 break; 3701 } 3702 btrfs_drop_extent_cache(inode, em->start, 3703 em->start + em->len - 1, 0); 3704 } 3705 3706 if (ret) 3707 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); 3708 } 3709 3710 /* 3711 * Make sure we do not end up inserting an inline extent into a file that has 3712 * already other (non-inline) extents. If a file has an inline extent it can 3713 * not have any other extents and the (single) inline extent must start at the 3714 * file offset 0. Failing to respect these rules will lead to file corruption, 3715 * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc 3716 * 3717 * We can have extents that have been already written to disk or we can have 3718 * dirty ranges still in delalloc, in which case the extent maps and items are 3719 * created only when we run delalloc, and the delalloc ranges might fall outside 3720 * the range we are currently locking in the inode's io tree. So we check the 3721 * inode's i_size because of that (i_size updates are done while holding the 3722 * i_mutex, which we are holding here). 3723 * We also check to see if the inode has a size not greater than "datal" but has 3724 * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are 3725 * protected against such concurrent fallocate calls by the i_mutex). 3726 * 3727 * If the file has no extents but a size greater than datal, do not allow the 3728 * copy because we would need turn the inline extent into a non-inline one (even 3729 * with NO_HOLES enabled). If we find our destination inode only has one inline 3730 * extent, just overwrite it with the source inline extent if its size is less 3731 * than the source extent's size, or we could copy the source inline extent's 3732 * data into the destination inode's inline extent if the later is greater then 3733 * the former. 3734 */ 3735 static int clone_copy_inline_extent(struct inode *dst, 3736 struct btrfs_trans_handle *trans, 3737 struct btrfs_path *path, 3738 struct btrfs_key *new_key, 3739 const u64 drop_start, 3740 const u64 datal, 3741 const u64 skip, 3742 const u64 size, 3743 char *inline_data) 3744 { 3745 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb); 3746 struct btrfs_root *root = BTRFS_I(dst)->root; 3747 const u64 aligned_end = ALIGN(new_key->offset + datal, 3748 fs_info->sectorsize); 3749 int ret; 3750 struct btrfs_key key; 3751 3752 if (new_key->offset > 0) 3753 return -EOPNOTSUPP; 3754 3755 key.objectid = btrfs_ino(BTRFS_I(dst)); 3756 key.type = BTRFS_EXTENT_DATA_KEY; 3757 key.offset = 0; 3758 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 3759 if (ret < 0) { 3760 return ret; 3761 } else if (ret > 0) { 3762 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 3763 ret = btrfs_next_leaf(root, path); 3764 if (ret < 0) 3765 return ret; 3766 else if (ret > 0) 3767 goto copy_inline_extent; 3768 } 3769 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 3770 if (key.objectid == btrfs_ino(BTRFS_I(dst)) && 3771 key.type == BTRFS_EXTENT_DATA_KEY) { 3772 ASSERT(key.offset > 0); 3773 return -EOPNOTSUPP; 3774 } 3775 } else if (i_size_read(dst) <= datal) { 3776 struct btrfs_file_extent_item *ei; 3777 u64 ext_len; 3778 3779 /* 3780 * If the file size is <= datal, make sure there are no other 3781 * extents following (can happen do to an fallocate call with 3782 * the flag FALLOC_FL_KEEP_SIZE). 3783 */ 3784 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 3785 struct btrfs_file_extent_item); 3786 /* 3787 * If it's an inline extent, it can not have other extents 3788 * following it. 3789 */ 3790 if (btrfs_file_extent_type(path->nodes[0], ei) == 3791 BTRFS_FILE_EXTENT_INLINE) 3792 goto copy_inline_extent; 3793 3794 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei); 3795 if (ext_len > aligned_end) 3796 return -EOPNOTSUPP; 3797 3798 ret = btrfs_next_item(root, path); 3799 if (ret < 0) { 3800 return ret; 3801 } else if (ret == 0) { 3802 btrfs_item_key_to_cpu(path->nodes[0], &key, 3803 path->slots[0]); 3804 if (key.objectid == btrfs_ino(BTRFS_I(dst)) && 3805 key.type == BTRFS_EXTENT_DATA_KEY) 3806 return -EOPNOTSUPP; 3807 } 3808 } 3809 3810 copy_inline_extent: 3811 /* 3812 * We have no extent items, or we have an extent at offset 0 which may 3813 * or may not be inlined. All these cases are dealt the same way. 3814 */ 3815 if (i_size_read(dst) > datal) { 3816 /* 3817 * If the destination inode has an inline extent... 3818 * This would require copying the data from the source inline 3819 * extent into the beginning of the destination's inline extent. 3820 * But this is really complex, both extents can be compressed 3821 * or just one of them, which would require decompressing and 3822 * re-compressing data (which could increase the new compressed 3823 * size, not allowing the compressed data to fit anymore in an 3824 * inline extent). 3825 * So just don't support this case for now (it should be rare, 3826 * we are not really saving space when cloning inline extents). 3827 */ 3828 return -EOPNOTSUPP; 3829 } 3830 3831 btrfs_release_path(path); 3832 ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1); 3833 if (ret) 3834 return ret; 3835 ret = btrfs_insert_empty_item(trans, root, path, new_key, size); 3836 if (ret) 3837 return ret; 3838 3839 if (skip) { 3840 const u32 start = btrfs_file_extent_calc_inline_size(0); 3841 3842 memmove(inline_data + start, inline_data + start + skip, datal); 3843 } 3844 3845 write_extent_buffer(path->nodes[0], inline_data, 3846 btrfs_item_ptr_offset(path->nodes[0], 3847 path->slots[0]), 3848 size); 3849 inode_add_bytes(dst, datal); 3850 3851 return 0; 3852 } 3853 3854 /** 3855 * btrfs_clone() - clone a range from inode file to another 3856 * 3857 * @src: Inode to clone from 3858 * @inode: Inode to clone to 3859 * @off: Offset within source to start clone from 3860 * @olen: Original length, passed by user, of range to clone 3861 * @olen_aligned: Block-aligned value of olen 3862 * @destoff: Offset within @inode to start clone 3863 * @no_time_update: Whether to update mtime/ctime on the target inode 3864 */ 3865 static int btrfs_clone(struct inode *src, struct inode *inode, 3866 const u64 off, const u64 olen, const u64 olen_aligned, 3867 const u64 destoff, int no_time_update) 3868 { 3869 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3870 struct btrfs_root *root = BTRFS_I(inode)->root; 3871 struct btrfs_path *path = NULL; 3872 struct extent_buffer *leaf; 3873 struct btrfs_trans_handle *trans; 3874 char *buf = NULL; 3875 struct btrfs_key key; 3876 u32 nritems; 3877 int slot; 3878 int ret; 3879 const u64 len = olen_aligned; 3880 u64 last_dest_end = destoff; 3881 3882 ret = -ENOMEM; 3883 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); 3884 if (!buf) 3885 return ret; 3886 3887 path = btrfs_alloc_path(); 3888 if (!path) { 3889 kvfree(buf); 3890 return ret; 3891 } 3892 3893 path->reada = READA_FORWARD; 3894 /* clone data */ 3895 key.objectid = btrfs_ino(BTRFS_I(src)); 3896 key.type = BTRFS_EXTENT_DATA_KEY; 3897 key.offset = off; 3898 3899 while (1) { 3900 u64 next_key_min_offset = key.offset + 1; 3901 3902 /* 3903 * note the key will change type as we walk through the 3904 * tree. 3905 */ 3906 path->leave_spinning = 1; 3907 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 3908 0, 0); 3909 if (ret < 0) 3910 goto out; 3911 /* 3912 * First search, if no extent item that starts at offset off was 3913 * found but the previous item is an extent item, it's possible 3914 * it might overlap our target range, therefore process it. 3915 */ 3916 if (key.offset == off && ret > 0 && path->slots[0] > 0) { 3917 btrfs_item_key_to_cpu(path->nodes[0], &key, 3918 path->slots[0] - 1); 3919 if (key.type == BTRFS_EXTENT_DATA_KEY) 3920 path->slots[0]--; 3921 } 3922 3923 nritems = btrfs_header_nritems(path->nodes[0]); 3924 process_slot: 3925 if (path->slots[0] >= nritems) { 3926 ret = btrfs_next_leaf(BTRFS_I(src)->root, path); 3927 if (ret < 0) 3928 goto out; 3929 if (ret > 0) 3930 break; 3931 nritems = btrfs_header_nritems(path->nodes[0]); 3932 } 3933 leaf = path->nodes[0]; 3934 slot = path->slots[0]; 3935 3936 btrfs_item_key_to_cpu(leaf, &key, slot); 3937 if (key.type > BTRFS_EXTENT_DATA_KEY || 3938 key.objectid != btrfs_ino(BTRFS_I(src))) 3939 break; 3940 3941 if (key.type == BTRFS_EXTENT_DATA_KEY) { 3942 struct btrfs_file_extent_item *extent; 3943 int type; 3944 u32 size; 3945 struct btrfs_key new_key; 3946 u64 disko = 0, diskl = 0; 3947 u64 datao = 0, datal = 0; 3948 u8 comp; 3949 u64 drop_start; 3950 3951 extent = btrfs_item_ptr(leaf, slot, 3952 struct btrfs_file_extent_item); 3953 comp = btrfs_file_extent_compression(leaf, extent); 3954 type = btrfs_file_extent_type(leaf, extent); 3955 if (type == BTRFS_FILE_EXTENT_REG || 3956 type == BTRFS_FILE_EXTENT_PREALLOC) { 3957 disko = btrfs_file_extent_disk_bytenr(leaf, 3958 extent); 3959 diskl = btrfs_file_extent_disk_num_bytes(leaf, 3960 extent); 3961 datao = btrfs_file_extent_offset(leaf, extent); 3962 datal = btrfs_file_extent_num_bytes(leaf, 3963 extent); 3964 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 3965 /* take upper bound, may be compressed */ 3966 datal = btrfs_file_extent_ram_bytes(leaf, 3967 extent); 3968 } 3969 3970 /* 3971 * The first search might have left us at an extent 3972 * item that ends before our target range's start, can 3973 * happen if we have holes and NO_HOLES feature enabled. 3974 */ 3975 if (key.offset + datal <= off) { 3976 path->slots[0]++; 3977 goto process_slot; 3978 } else if (key.offset >= off + len) { 3979 break; 3980 } 3981 next_key_min_offset = key.offset + datal; 3982 size = btrfs_item_size_nr(leaf, slot); 3983 read_extent_buffer(leaf, buf, 3984 btrfs_item_ptr_offset(leaf, slot), 3985 size); 3986 3987 btrfs_release_path(path); 3988 path->leave_spinning = 0; 3989 3990 memcpy(&new_key, &key, sizeof(new_key)); 3991 new_key.objectid = btrfs_ino(BTRFS_I(inode)); 3992 if (off <= key.offset) 3993 new_key.offset = key.offset + destoff - off; 3994 else 3995 new_key.offset = destoff; 3996 3997 /* 3998 * Deal with a hole that doesn't have an extent item 3999 * that represents it (NO_HOLES feature enabled). 4000 * This hole is either in the middle of the cloning 4001 * range or at the beginning (fully overlaps it or 4002 * partially overlaps it). 4003 */ 4004 if (new_key.offset != last_dest_end) 4005 drop_start = last_dest_end; 4006 else 4007 drop_start = new_key.offset; 4008 4009 /* 4010 * 1 - adjusting old extent (we may have to split it) 4011 * 1 - add new extent 4012 * 1 - inode update 4013 */ 4014 trans = btrfs_start_transaction(root, 3); 4015 if (IS_ERR(trans)) { 4016 ret = PTR_ERR(trans); 4017 goto out; 4018 } 4019 4020 if (type == BTRFS_FILE_EXTENT_REG || 4021 type == BTRFS_FILE_EXTENT_PREALLOC) { 4022 /* 4023 * a | --- range to clone ---| b 4024 * | ------------- extent ------------- | 4025 */ 4026 4027 /* subtract range b */ 4028 if (key.offset + datal > off + len) 4029 datal = off + len - key.offset; 4030 4031 /* subtract range a */ 4032 if (off > key.offset) { 4033 datao += off - key.offset; 4034 datal -= off - key.offset; 4035 } 4036 4037 ret = btrfs_drop_extents(trans, root, inode, 4038 drop_start, 4039 new_key.offset + datal, 4040 1); 4041 if (ret) { 4042 if (ret != -EOPNOTSUPP) 4043 btrfs_abort_transaction(trans, 4044 ret); 4045 btrfs_end_transaction(trans); 4046 goto out; 4047 } 4048 4049 ret = btrfs_insert_empty_item(trans, root, path, 4050 &new_key, size); 4051 if (ret) { 4052 btrfs_abort_transaction(trans, ret); 4053 btrfs_end_transaction(trans); 4054 goto out; 4055 } 4056 4057 leaf = path->nodes[0]; 4058 slot = path->slots[0]; 4059 write_extent_buffer(leaf, buf, 4060 btrfs_item_ptr_offset(leaf, slot), 4061 size); 4062 4063 extent = btrfs_item_ptr(leaf, slot, 4064 struct btrfs_file_extent_item); 4065 4066 /* disko == 0 means it's a hole */ 4067 if (!disko) 4068 datao = 0; 4069 4070 btrfs_set_file_extent_offset(leaf, extent, 4071 datao); 4072 btrfs_set_file_extent_num_bytes(leaf, extent, 4073 datal); 4074 4075 if (disko) { 4076 inode_add_bytes(inode, datal); 4077 ret = btrfs_inc_extent_ref(trans, 4078 root, 4079 disko, diskl, 0, 4080 root->root_key.objectid, 4081 btrfs_ino(BTRFS_I(inode)), 4082 new_key.offset - datao); 4083 if (ret) { 4084 btrfs_abort_transaction(trans, 4085 ret); 4086 btrfs_end_transaction(trans); 4087 goto out; 4088 4089 } 4090 } 4091 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 4092 u64 skip = 0; 4093 u64 trim = 0; 4094 4095 if (off > key.offset) { 4096 skip = off - key.offset; 4097 new_key.offset += skip; 4098 } 4099 4100 if (key.offset + datal > off + len) 4101 trim = key.offset + datal - (off + len); 4102 4103 if (comp && (skip || trim)) { 4104 ret = -EINVAL; 4105 btrfs_end_transaction(trans); 4106 goto out; 4107 } 4108 size -= skip + trim; 4109 datal -= skip + trim; 4110 4111 ret = clone_copy_inline_extent(inode, 4112 trans, path, 4113 &new_key, 4114 drop_start, 4115 datal, 4116 skip, size, buf); 4117 if (ret) { 4118 if (ret != -EOPNOTSUPP) 4119 btrfs_abort_transaction(trans, 4120 ret); 4121 btrfs_end_transaction(trans); 4122 goto out; 4123 } 4124 leaf = path->nodes[0]; 4125 slot = path->slots[0]; 4126 } 4127 4128 /* If we have an implicit hole (NO_HOLES feature). */ 4129 if (drop_start < new_key.offset) 4130 clone_update_extent_map(BTRFS_I(inode), trans, 4131 NULL, drop_start, 4132 new_key.offset - drop_start); 4133 4134 clone_update_extent_map(BTRFS_I(inode), trans, 4135 path, 0, 0); 4136 4137 btrfs_mark_buffer_dirty(leaf); 4138 btrfs_release_path(path); 4139 4140 last_dest_end = ALIGN(new_key.offset + datal, 4141 fs_info->sectorsize); 4142 ret = clone_finish_inode_update(trans, inode, 4143 last_dest_end, 4144 destoff, olen, 4145 no_time_update); 4146 if (ret) 4147 goto out; 4148 if (new_key.offset + datal >= destoff + len) 4149 break; 4150 } 4151 btrfs_release_path(path); 4152 key.offset = next_key_min_offset; 4153 4154 if (fatal_signal_pending(current)) { 4155 ret = -EINTR; 4156 goto out; 4157 } 4158 } 4159 ret = 0; 4160 4161 if (last_dest_end < destoff + len) { 4162 /* 4163 * We have an implicit hole (NO_HOLES feature is enabled) that 4164 * fully or partially overlaps our cloning range at its end. 4165 */ 4166 btrfs_release_path(path); 4167 4168 /* 4169 * 1 - remove extent(s) 4170 * 1 - inode update 4171 */ 4172 trans = btrfs_start_transaction(root, 2); 4173 if (IS_ERR(trans)) { 4174 ret = PTR_ERR(trans); 4175 goto out; 4176 } 4177 ret = btrfs_drop_extents(trans, root, inode, 4178 last_dest_end, destoff + len, 1); 4179 if (ret) { 4180 if (ret != -EOPNOTSUPP) 4181 btrfs_abort_transaction(trans, ret); 4182 btrfs_end_transaction(trans); 4183 goto out; 4184 } 4185 clone_update_extent_map(BTRFS_I(inode), trans, NULL, 4186 last_dest_end, 4187 destoff + len - last_dest_end); 4188 ret = clone_finish_inode_update(trans, inode, destoff + len, 4189 destoff, olen, no_time_update); 4190 } 4191 4192 out: 4193 btrfs_free_path(path); 4194 kvfree(buf); 4195 return ret; 4196 } 4197 4198 static noinline int btrfs_clone_files(struct file *file, struct file *file_src, 4199 u64 off, u64 olen, u64 destoff) 4200 { 4201 struct inode *inode = file_inode(file); 4202 struct inode *src = file_inode(file_src); 4203 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 4204 struct btrfs_root *root = BTRFS_I(inode)->root; 4205 int ret; 4206 u64 len = olen; 4207 u64 bs = fs_info->sb->s_blocksize; 4208 int same_inode = src == inode; 4209 4210 /* 4211 * TODO: 4212 * - split compressed inline extents. annoying: we need to 4213 * decompress into destination's address_space (the file offset 4214 * may change, so source mapping won't do), then recompress (or 4215 * otherwise reinsert) a subrange. 4216 * 4217 * - split destination inode's inline extents. The inline extents can 4218 * be either compressed or non-compressed. 4219 */ 4220 4221 if (btrfs_root_readonly(root)) 4222 return -EROFS; 4223 4224 if (file_src->f_path.mnt != file->f_path.mnt || 4225 src->i_sb != inode->i_sb) 4226 return -EXDEV; 4227 4228 if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode)) 4229 return -EISDIR; 4230 4231 if (!same_inode) { 4232 btrfs_double_inode_lock(src, inode); 4233 } else { 4234 inode_lock(src); 4235 } 4236 4237 /* don't make the dst file partly checksummed */ 4238 if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 4239 (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { 4240 ret = -EINVAL; 4241 goto out_unlock; 4242 } 4243 4244 /* determine range to clone */ 4245 ret = -EINVAL; 4246 if (off + len > src->i_size || off + len < off) 4247 goto out_unlock; 4248 if (len == 0) 4249 olen = len = src->i_size - off; 4250 /* if we extend to eof, continue to block boundary */ 4251 if (off + len == src->i_size) 4252 len = ALIGN(src->i_size, bs) - off; 4253 4254 if (len == 0) { 4255 ret = 0; 4256 goto out_unlock; 4257 } 4258 4259 /* verify the end result is block aligned */ 4260 if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) || 4261 !IS_ALIGNED(destoff, bs)) 4262 goto out_unlock; 4263 4264 /* verify if ranges are overlapped within the same file */ 4265 if (same_inode) { 4266 if (destoff + len > off && destoff < off + len) 4267 goto out_unlock; 4268 } 4269 4270 if (destoff > inode->i_size) { 4271 ret = btrfs_cont_expand(inode, inode->i_size, destoff); 4272 if (ret) 4273 goto out_unlock; 4274 } 4275 4276 /* 4277 * Lock the target range too. Right after we replace the file extent 4278 * items in the fs tree (which now point to the cloned data), we might 4279 * have a worker replace them with extent items relative to a write 4280 * operation that was issued before this clone operation (i.e. confront 4281 * with inode.c:btrfs_finish_ordered_io). 4282 */ 4283 if (same_inode) { 4284 u64 lock_start = min_t(u64, off, destoff); 4285 u64 lock_len = max_t(u64, off, destoff) + len - lock_start; 4286 4287 ret = lock_extent_range(src, lock_start, lock_len, true); 4288 } else { 4289 ret = btrfs_double_extent_lock(src, off, inode, destoff, len, 4290 true); 4291 } 4292 ASSERT(ret == 0); 4293 if (WARN_ON(ret)) { 4294 /* ranges in the io trees already unlocked */ 4295 goto out_unlock; 4296 } 4297 4298 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0); 4299 4300 if (same_inode) { 4301 u64 lock_start = min_t(u64, off, destoff); 4302 u64 lock_end = max_t(u64, off, destoff) + len - 1; 4303 4304 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end); 4305 } else { 4306 btrfs_double_extent_unlock(src, off, inode, destoff, len); 4307 } 4308 /* 4309 * Truncate page cache pages so that future reads will see the cloned 4310 * data immediately and not the previous data. 4311 */ 4312 truncate_inode_pages_range(&inode->i_data, 4313 round_down(destoff, PAGE_SIZE), 4314 round_up(destoff + len, PAGE_SIZE) - 1); 4315 out_unlock: 4316 if (!same_inode) 4317 btrfs_double_inode_unlock(src, inode); 4318 else 4319 inode_unlock(src); 4320 return ret; 4321 } 4322 4323 int btrfs_clone_file_range(struct file *src_file, loff_t off, 4324 struct file *dst_file, loff_t destoff, u64 len) 4325 { 4326 return btrfs_clone_files(dst_file, src_file, off, len, destoff); 4327 } 4328 4329 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 4330 { 4331 struct inode *inode = file_inode(file); 4332 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 4333 struct btrfs_root *root = BTRFS_I(inode)->root; 4334 struct btrfs_root *new_root; 4335 struct btrfs_dir_item *di; 4336 struct btrfs_trans_handle *trans; 4337 struct btrfs_path *path; 4338 struct btrfs_key location; 4339 struct btrfs_disk_key disk_key; 4340 u64 objectid = 0; 4341 u64 dir_id; 4342 int ret; 4343 4344 if (!capable(CAP_SYS_ADMIN)) 4345 return -EPERM; 4346 4347 ret = mnt_want_write_file(file); 4348 if (ret) 4349 return ret; 4350 4351 if (copy_from_user(&objectid, argp, sizeof(objectid))) { 4352 ret = -EFAULT; 4353 goto out; 4354 } 4355 4356 if (!objectid) 4357 objectid = BTRFS_FS_TREE_OBJECTID; 4358 4359 location.objectid = objectid; 4360 location.type = BTRFS_ROOT_ITEM_KEY; 4361 location.offset = (u64)-1; 4362 4363 new_root = btrfs_read_fs_root_no_name(fs_info, &location); 4364 if (IS_ERR(new_root)) { 4365 ret = PTR_ERR(new_root); 4366 goto out; 4367 } 4368 if (!is_fstree(new_root->objectid)) { 4369 ret = -ENOENT; 4370 goto out; 4371 } 4372 4373 path = btrfs_alloc_path(); 4374 if (!path) { 4375 ret = -ENOMEM; 4376 goto out; 4377 } 4378 path->leave_spinning = 1; 4379 4380 trans = btrfs_start_transaction(root, 1); 4381 if (IS_ERR(trans)) { 4382 btrfs_free_path(path); 4383 ret = PTR_ERR(trans); 4384 goto out; 4385 } 4386 4387 dir_id = btrfs_super_root_dir(fs_info->super_copy); 4388 di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path, 4389 dir_id, "default", 7, 1); 4390 if (IS_ERR_OR_NULL(di)) { 4391 btrfs_free_path(path); 4392 btrfs_end_transaction(trans); 4393 btrfs_err(fs_info, 4394 "Umm, you don't have the default diritem, this isn't going to work"); 4395 ret = -ENOENT; 4396 goto out; 4397 } 4398 4399 btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 4400 btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 4401 btrfs_mark_buffer_dirty(path->nodes[0]); 4402 btrfs_free_path(path); 4403 4404 btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL); 4405 btrfs_end_transaction(trans); 4406 out: 4407 mnt_drop_write_file(file); 4408 return ret; 4409 } 4410 4411 static void get_block_group_info(struct list_head *groups_list, 4412 struct btrfs_ioctl_space_info *space) 4413 { 4414 struct btrfs_block_group_cache *block_group; 4415 4416 space->total_bytes = 0; 4417 space->used_bytes = 0; 4418 space->flags = 0; 4419 list_for_each_entry(block_group, groups_list, list) { 4420 space->flags = block_group->flags; 4421 space->total_bytes += block_group->key.offset; 4422 space->used_bytes += 4423 btrfs_block_group_used(&block_group->item); 4424 } 4425 } 4426 4427 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info, 4428 void __user *arg) 4429 { 4430 struct btrfs_ioctl_space_args space_args; 4431 struct btrfs_ioctl_space_info space; 4432 struct btrfs_ioctl_space_info *dest; 4433 struct btrfs_ioctl_space_info *dest_orig; 4434 struct btrfs_ioctl_space_info __user *user_dest; 4435 struct btrfs_space_info *info; 4436 static const u64 types[] = { 4437 BTRFS_BLOCK_GROUP_DATA, 4438 BTRFS_BLOCK_GROUP_SYSTEM, 4439 BTRFS_BLOCK_GROUP_METADATA, 4440 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA 4441 }; 4442 int num_types = 4; 4443 int alloc_size; 4444 int ret = 0; 4445 u64 slot_count = 0; 4446 int i, c; 4447 4448 if (copy_from_user(&space_args, 4449 (struct btrfs_ioctl_space_args __user *)arg, 4450 sizeof(space_args))) 4451 return -EFAULT; 4452 4453 for (i = 0; i < num_types; i++) { 4454 struct btrfs_space_info *tmp; 4455 4456 info = NULL; 4457 rcu_read_lock(); 4458 list_for_each_entry_rcu(tmp, &fs_info->space_info, 4459 list) { 4460 if (tmp->flags == types[i]) { 4461 info = tmp; 4462 break; 4463 } 4464 } 4465 rcu_read_unlock(); 4466 4467 if (!info) 4468 continue; 4469 4470 down_read(&info->groups_sem); 4471 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 4472 if (!list_empty(&info->block_groups[c])) 4473 slot_count++; 4474 } 4475 up_read(&info->groups_sem); 4476 } 4477 4478 /* 4479 * Global block reserve, exported as a space_info 4480 */ 4481 slot_count++; 4482 4483 /* space_slots == 0 means they are asking for a count */ 4484 if (space_args.space_slots == 0) { 4485 space_args.total_spaces = slot_count; 4486 goto out; 4487 } 4488 4489 slot_count = min_t(u64, space_args.space_slots, slot_count); 4490 4491 alloc_size = sizeof(*dest) * slot_count; 4492 4493 /* we generally have at most 6 or so space infos, one for each raid 4494 * level. So, a whole page should be more than enough for everyone 4495 */ 4496 if (alloc_size > PAGE_SIZE) 4497 return -ENOMEM; 4498 4499 space_args.total_spaces = 0; 4500 dest = kmalloc(alloc_size, GFP_KERNEL); 4501 if (!dest) 4502 return -ENOMEM; 4503 dest_orig = dest; 4504 4505 /* now we have a buffer to copy into */ 4506 for (i = 0; i < num_types; i++) { 4507 struct btrfs_space_info *tmp; 4508 4509 if (!slot_count) 4510 break; 4511 4512 info = NULL; 4513 rcu_read_lock(); 4514 list_for_each_entry_rcu(tmp, &fs_info->space_info, 4515 list) { 4516 if (tmp->flags == types[i]) { 4517 info = tmp; 4518 break; 4519 } 4520 } 4521 rcu_read_unlock(); 4522 4523 if (!info) 4524 continue; 4525 down_read(&info->groups_sem); 4526 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 4527 if (!list_empty(&info->block_groups[c])) { 4528 get_block_group_info(&info->block_groups[c], 4529 &space); 4530 memcpy(dest, &space, sizeof(space)); 4531 dest++; 4532 space_args.total_spaces++; 4533 slot_count--; 4534 } 4535 if (!slot_count) 4536 break; 4537 } 4538 up_read(&info->groups_sem); 4539 } 4540 4541 /* 4542 * Add global block reserve 4543 */ 4544 if (slot_count) { 4545 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; 4546 4547 spin_lock(&block_rsv->lock); 4548 space.total_bytes = block_rsv->size; 4549 space.used_bytes = block_rsv->size - block_rsv->reserved; 4550 spin_unlock(&block_rsv->lock); 4551 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV; 4552 memcpy(dest, &space, sizeof(space)); 4553 space_args.total_spaces++; 4554 } 4555 4556 user_dest = (struct btrfs_ioctl_space_info __user *) 4557 (arg + sizeof(struct btrfs_ioctl_space_args)); 4558 4559 if (copy_to_user(user_dest, dest_orig, alloc_size)) 4560 ret = -EFAULT; 4561 4562 kfree(dest_orig); 4563 out: 4564 if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 4565 ret = -EFAULT; 4566 4567 return ret; 4568 } 4569 4570 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 4571 void __user *argp) 4572 { 4573 struct btrfs_trans_handle *trans; 4574 u64 transid; 4575 int ret; 4576 4577 trans = btrfs_attach_transaction_barrier(root); 4578 if (IS_ERR(trans)) { 4579 if (PTR_ERR(trans) != -ENOENT) 4580 return PTR_ERR(trans); 4581 4582 /* No running transaction, don't bother */ 4583 transid = root->fs_info->last_trans_committed; 4584 goto out; 4585 } 4586 transid = trans->transid; 4587 ret = btrfs_commit_transaction_async(trans, 0); 4588 if (ret) { 4589 btrfs_end_transaction(trans); 4590 return ret; 4591 } 4592 out: 4593 if (argp) 4594 if (copy_to_user(argp, &transid, sizeof(transid))) 4595 return -EFAULT; 4596 return 0; 4597 } 4598 4599 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info, 4600 void __user *argp) 4601 { 4602 u64 transid; 4603 4604 if (argp) { 4605 if (copy_from_user(&transid, argp, sizeof(transid))) 4606 return -EFAULT; 4607 } else { 4608 transid = 0; /* current trans */ 4609 } 4610 return btrfs_wait_for_commit(fs_info, transid); 4611 } 4612 4613 static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 4614 { 4615 struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 4616 struct btrfs_ioctl_scrub_args *sa; 4617 int ret; 4618 4619 if (!capable(CAP_SYS_ADMIN)) 4620 return -EPERM; 4621 4622 sa = memdup_user(arg, sizeof(*sa)); 4623 if (IS_ERR(sa)) 4624 return PTR_ERR(sa); 4625 4626 if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 4627 ret = mnt_want_write_file(file); 4628 if (ret) 4629 goto out; 4630 } 4631 4632 ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end, 4633 &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 4634 0); 4635 4636 if (copy_to_user(arg, sa, sizeof(*sa))) 4637 ret = -EFAULT; 4638 4639 if (!(sa->flags & BTRFS_SCRUB_READONLY)) 4640 mnt_drop_write_file(file); 4641 out: 4642 kfree(sa); 4643 return ret; 4644 } 4645 4646 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info) 4647 { 4648 if (!capable(CAP_SYS_ADMIN)) 4649 return -EPERM; 4650 4651 return btrfs_scrub_cancel(fs_info); 4652 } 4653 4654 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info, 4655 void __user *arg) 4656 { 4657 struct btrfs_ioctl_scrub_args *sa; 4658 int ret; 4659 4660 if (!capable(CAP_SYS_ADMIN)) 4661 return -EPERM; 4662 4663 sa = memdup_user(arg, sizeof(*sa)); 4664 if (IS_ERR(sa)) 4665 return PTR_ERR(sa); 4666 4667 ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress); 4668 4669 if (copy_to_user(arg, sa, sizeof(*sa))) 4670 ret = -EFAULT; 4671 4672 kfree(sa); 4673 return ret; 4674 } 4675 4676 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info, 4677 void __user *arg) 4678 { 4679 struct btrfs_ioctl_get_dev_stats *sa; 4680 int ret; 4681 4682 sa = memdup_user(arg, sizeof(*sa)); 4683 if (IS_ERR(sa)) 4684 return PTR_ERR(sa); 4685 4686 if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 4687 kfree(sa); 4688 return -EPERM; 4689 } 4690 4691 ret = btrfs_get_dev_stats(fs_info, sa); 4692 4693 if (copy_to_user(arg, sa, sizeof(*sa))) 4694 ret = -EFAULT; 4695 4696 kfree(sa); 4697 return ret; 4698 } 4699 4700 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info, 4701 void __user *arg) 4702 { 4703 struct btrfs_ioctl_dev_replace_args *p; 4704 int ret; 4705 4706 if (!capable(CAP_SYS_ADMIN)) 4707 return -EPERM; 4708 4709 p = memdup_user(arg, sizeof(*p)); 4710 if (IS_ERR(p)) 4711 return PTR_ERR(p); 4712 4713 switch (p->cmd) { 4714 case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 4715 if (sb_rdonly(fs_info->sb)) { 4716 ret = -EROFS; 4717 goto out; 4718 } 4719 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 4720 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4721 } else { 4722 ret = btrfs_dev_replace_by_ioctl(fs_info, p); 4723 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 4724 } 4725 break; 4726 case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 4727 btrfs_dev_replace_status(fs_info, p); 4728 ret = 0; 4729 break; 4730 case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 4731 p->result = btrfs_dev_replace_cancel(fs_info); 4732 ret = 0; 4733 break; 4734 default: 4735 ret = -EINVAL; 4736 break; 4737 } 4738 4739 if (copy_to_user(arg, p, sizeof(*p))) 4740 ret = -EFAULT; 4741 out: 4742 kfree(p); 4743 return ret; 4744 } 4745 4746 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 4747 { 4748 int ret = 0; 4749 int i; 4750 u64 rel_ptr; 4751 int size; 4752 struct btrfs_ioctl_ino_path_args *ipa = NULL; 4753 struct inode_fs_paths *ipath = NULL; 4754 struct btrfs_path *path; 4755 4756 if (!capable(CAP_DAC_READ_SEARCH)) 4757 return -EPERM; 4758 4759 path = btrfs_alloc_path(); 4760 if (!path) { 4761 ret = -ENOMEM; 4762 goto out; 4763 } 4764 4765 ipa = memdup_user(arg, sizeof(*ipa)); 4766 if (IS_ERR(ipa)) { 4767 ret = PTR_ERR(ipa); 4768 ipa = NULL; 4769 goto out; 4770 } 4771 4772 size = min_t(u32, ipa->size, 4096); 4773 ipath = init_ipath(size, root, path); 4774 if (IS_ERR(ipath)) { 4775 ret = PTR_ERR(ipath); 4776 ipath = NULL; 4777 goto out; 4778 } 4779 4780 ret = paths_from_inode(ipa->inum, ipath); 4781 if (ret < 0) 4782 goto out; 4783 4784 for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 4785 rel_ptr = ipath->fspath->val[i] - 4786 (u64)(unsigned long)ipath->fspath->val; 4787 ipath->fspath->val[i] = rel_ptr; 4788 } 4789 4790 ret = copy_to_user((void __user *)(unsigned long)ipa->fspath, 4791 ipath->fspath, size); 4792 if (ret) { 4793 ret = -EFAULT; 4794 goto out; 4795 } 4796 4797 out: 4798 btrfs_free_path(path); 4799 free_ipath(ipath); 4800 kfree(ipa); 4801 4802 return ret; 4803 } 4804 4805 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 4806 { 4807 struct btrfs_data_container *inodes = ctx; 4808 const size_t c = 3 * sizeof(u64); 4809 4810 if (inodes->bytes_left >= c) { 4811 inodes->bytes_left -= c; 4812 inodes->val[inodes->elem_cnt] = inum; 4813 inodes->val[inodes->elem_cnt + 1] = offset; 4814 inodes->val[inodes->elem_cnt + 2] = root; 4815 inodes->elem_cnt += 3; 4816 } else { 4817 inodes->bytes_missing += c - inodes->bytes_left; 4818 inodes->bytes_left = 0; 4819 inodes->elem_missed += 3; 4820 } 4821 4822 return 0; 4823 } 4824 4825 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info, 4826 void __user *arg, int version) 4827 { 4828 int ret = 0; 4829 int size; 4830 struct btrfs_ioctl_logical_ino_args *loi; 4831 struct btrfs_data_container *inodes = NULL; 4832 struct btrfs_path *path = NULL; 4833 bool ignore_offset; 4834 4835 if (!capable(CAP_SYS_ADMIN)) 4836 return -EPERM; 4837 4838 loi = memdup_user(arg, sizeof(*loi)); 4839 if (IS_ERR(loi)) 4840 return PTR_ERR(loi); 4841 4842 if (version == 1) { 4843 ignore_offset = false; 4844 size = min_t(u32, loi->size, SZ_64K); 4845 } else { 4846 /* All reserved bits must be 0 for now */ 4847 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) { 4848 ret = -EINVAL; 4849 goto out_loi; 4850 } 4851 /* Only accept flags we have defined so far */ 4852 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) { 4853 ret = -EINVAL; 4854 goto out_loi; 4855 } 4856 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET; 4857 size = min_t(u32, loi->size, SZ_16M); 4858 } 4859 4860 path = btrfs_alloc_path(); 4861 if (!path) { 4862 ret = -ENOMEM; 4863 goto out; 4864 } 4865 4866 inodes = init_data_container(size); 4867 if (IS_ERR(inodes)) { 4868 ret = PTR_ERR(inodes); 4869 inodes = NULL; 4870 goto out; 4871 } 4872 4873 ret = iterate_inodes_from_logical(loi->logical, fs_info, path, 4874 build_ino_list, inodes, ignore_offset); 4875 if (ret == -EINVAL) 4876 ret = -ENOENT; 4877 if (ret < 0) 4878 goto out; 4879 4880 ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes, 4881 size); 4882 if (ret) 4883 ret = -EFAULT; 4884 4885 out: 4886 btrfs_free_path(path); 4887 kvfree(inodes); 4888 out_loi: 4889 kfree(loi); 4890 4891 return ret; 4892 } 4893 4894 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info, 4895 struct btrfs_ioctl_balance_args *bargs) 4896 { 4897 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4898 4899 bargs->flags = bctl->flags; 4900 4901 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) 4902 bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 4903 if (atomic_read(&fs_info->balance_pause_req)) 4904 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 4905 if (atomic_read(&fs_info->balance_cancel_req)) 4906 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 4907 4908 memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 4909 memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 4910 memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 4911 4912 spin_lock(&fs_info->balance_lock); 4913 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 4914 spin_unlock(&fs_info->balance_lock); 4915 } 4916 4917 static long btrfs_ioctl_balance(struct file *file, void __user *arg) 4918 { 4919 struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4920 struct btrfs_fs_info *fs_info = root->fs_info; 4921 struct btrfs_ioctl_balance_args *bargs; 4922 struct btrfs_balance_control *bctl; 4923 bool need_unlock; /* for mut. excl. ops lock */ 4924 int ret; 4925 4926 if (!capable(CAP_SYS_ADMIN)) 4927 return -EPERM; 4928 4929 ret = mnt_want_write_file(file); 4930 if (ret) 4931 return ret; 4932 4933 again: 4934 if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { 4935 mutex_lock(&fs_info->balance_mutex); 4936 need_unlock = true; 4937 goto locked; 4938 } 4939 4940 /* 4941 * mut. excl. ops lock is locked. Three possibilities: 4942 * (1) some other op is running 4943 * (2) balance is running 4944 * (3) balance is paused -- special case (think resume) 4945 */ 4946 mutex_lock(&fs_info->balance_mutex); 4947 if (fs_info->balance_ctl) { 4948 /* this is either (2) or (3) */ 4949 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4950 mutex_unlock(&fs_info->balance_mutex); 4951 /* 4952 * Lock released to allow other waiters to continue, 4953 * we'll reexamine the status again. 4954 */ 4955 mutex_lock(&fs_info->balance_mutex); 4956 4957 if (fs_info->balance_ctl && 4958 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4959 /* this is (3) */ 4960 need_unlock = false; 4961 goto locked; 4962 } 4963 4964 mutex_unlock(&fs_info->balance_mutex); 4965 goto again; 4966 } else { 4967 /* this is (2) */ 4968 mutex_unlock(&fs_info->balance_mutex); 4969 ret = -EINPROGRESS; 4970 goto out; 4971 } 4972 } else { 4973 /* this is (1) */ 4974 mutex_unlock(&fs_info->balance_mutex); 4975 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4976 goto out; 4977 } 4978 4979 locked: 4980 BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)); 4981 4982 if (arg) { 4983 bargs = memdup_user(arg, sizeof(*bargs)); 4984 if (IS_ERR(bargs)) { 4985 ret = PTR_ERR(bargs); 4986 goto out_unlock; 4987 } 4988 4989 if (bargs->flags & BTRFS_BALANCE_RESUME) { 4990 if (!fs_info->balance_ctl) { 4991 ret = -ENOTCONN; 4992 goto out_bargs; 4993 } 4994 4995 bctl = fs_info->balance_ctl; 4996 spin_lock(&fs_info->balance_lock); 4997 bctl->flags |= BTRFS_BALANCE_RESUME; 4998 spin_unlock(&fs_info->balance_lock); 4999 5000 goto do_balance; 5001 } 5002 } else { 5003 bargs = NULL; 5004 } 5005 5006 if (fs_info->balance_ctl) { 5007 ret = -EINPROGRESS; 5008 goto out_bargs; 5009 } 5010 5011 bctl = kzalloc(sizeof(*bctl), GFP_KERNEL); 5012 if (!bctl) { 5013 ret = -ENOMEM; 5014 goto out_bargs; 5015 } 5016 5017 if (arg) { 5018 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 5019 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 5020 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 5021 5022 bctl->flags = bargs->flags; 5023 } else { 5024 /* balance everything - no filters */ 5025 bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 5026 } 5027 5028 if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) { 5029 ret = -EINVAL; 5030 goto out_bctl; 5031 } 5032 5033 do_balance: 5034 /* 5035 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to 5036 * btrfs_balance. bctl is freed in reset_balance_state, or, if 5037 * restriper was paused all the way until unmount, in free_fs_info. 5038 * The flag should be cleared after reset_balance_state. 5039 */ 5040 need_unlock = false; 5041 5042 ret = btrfs_balance(fs_info, bctl, bargs); 5043 bctl = NULL; 5044 5045 if (arg) { 5046 if (copy_to_user(arg, bargs, sizeof(*bargs))) 5047 ret = -EFAULT; 5048 } 5049 5050 out_bctl: 5051 kfree(bctl); 5052 out_bargs: 5053 kfree(bargs); 5054 out_unlock: 5055 mutex_unlock(&fs_info->balance_mutex); 5056 if (need_unlock) 5057 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); 5058 out: 5059 mnt_drop_write_file(file); 5060 return ret; 5061 } 5062 5063 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd) 5064 { 5065 if (!capable(CAP_SYS_ADMIN)) 5066 return -EPERM; 5067 5068 switch (cmd) { 5069 case BTRFS_BALANCE_CTL_PAUSE: 5070 return btrfs_pause_balance(fs_info); 5071 case BTRFS_BALANCE_CTL_CANCEL: 5072 return btrfs_cancel_balance(fs_info); 5073 } 5074 5075 return -EINVAL; 5076 } 5077 5078 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info, 5079 void __user *arg) 5080 { 5081 struct btrfs_ioctl_balance_args *bargs; 5082 int ret = 0; 5083 5084 if (!capable(CAP_SYS_ADMIN)) 5085 return -EPERM; 5086 5087 mutex_lock(&fs_info->balance_mutex); 5088 if (!fs_info->balance_ctl) { 5089 ret = -ENOTCONN; 5090 goto out; 5091 } 5092 5093 bargs = kzalloc(sizeof(*bargs), GFP_KERNEL); 5094 if (!bargs) { 5095 ret = -ENOMEM; 5096 goto out; 5097 } 5098 5099 btrfs_update_ioctl_balance_args(fs_info, bargs); 5100 5101 if (copy_to_user(arg, bargs, sizeof(*bargs))) 5102 ret = -EFAULT; 5103 5104 kfree(bargs); 5105 out: 5106 mutex_unlock(&fs_info->balance_mutex); 5107 return ret; 5108 } 5109 5110 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 5111 { 5112 struct inode *inode = file_inode(file); 5113 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5114 struct btrfs_ioctl_quota_ctl_args *sa; 5115 struct btrfs_trans_handle *trans = NULL; 5116 int ret; 5117 int err; 5118 5119 if (!capable(CAP_SYS_ADMIN)) 5120 return -EPERM; 5121 5122 ret = mnt_want_write_file(file); 5123 if (ret) 5124 return ret; 5125 5126 sa = memdup_user(arg, sizeof(*sa)); 5127 if (IS_ERR(sa)) { 5128 ret = PTR_ERR(sa); 5129 goto drop_write; 5130 } 5131 5132 down_write(&fs_info->subvol_sem); 5133 trans = btrfs_start_transaction(fs_info->tree_root, 2); 5134 if (IS_ERR(trans)) { 5135 ret = PTR_ERR(trans); 5136 goto out; 5137 } 5138 5139 switch (sa->cmd) { 5140 case BTRFS_QUOTA_CTL_ENABLE: 5141 ret = btrfs_quota_enable(trans, fs_info); 5142 break; 5143 case BTRFS_QUOTA_CTL_DISABLE: 5144 ret = btrfs_quota_disable(trans, fs_info); 5145 break; 5146 default: 5147 ret = -EINVAL; 5148 break; 5149 } 5150 5151 err = btrfs_commit_transaction(trans); 5152 if (err && !ret) 5153 ret = err; 5154 out: 5155 kfree(sa); 5156 up_write(&fs_info->subvol_sem); 5157 drop_write: 5158 mnt_drop_write_file(file); 5159 return ret; 5160 } 5161 5162 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 5163 { 5164 struct inode *inode = file_inode(file); 5165 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5166 struct btrfs_root *root = BTRFS_I(inode)->root; 5167 struct btrfs_ioctl_qgroup_assign_args *sa; 5168 struct btrfs_trans_handle *trans; 5169 int ret; 5170 int err; 5171 5172 if (!capable(CAP_SYS_ADMIN)) 5173 return -EPERM; 5174 5175 ret = mnt_want_write_file(file); 5176 if (ret) 5177 return ret; 5178 5179 sa = memdup_user(arg, sizeof(*sa)); 5180 if (IS_ERR(sa)) { 5181 ret = PTR_ERR(sa); 5182 goto drop_write; 5183 } 5184 5185 trans = btrfs_join_transaction(root); 5186 if (IS_ERR(trans)) { 5187 ret = PTR_ERR(trans); 5188 goto out; 5189 } 5190 5191 if (sa->assign) { 5192 ret = btrfs_add_qgroup_relation(trans, fs_info, 5193 sa->src, sa->dst); 5194 } else { 5195 ret = btrfs_del_qgroup_relation(trans, fs_info, 5196 sa->src, sa->dst); 5197 } 5198 5199 /* update qgroup status and info */ 5200 err = btrfs_run_qgroups(trans, fs_info); 5201 if (err < 0) 5202 btrfs_handle_fs_error(fs_info, err, 5203 "failed to update qgroup status and info"); 5204 err = btrfs_end_transaction(trans); 5205 if (err && !ret) 5206 ret = err; 5207 5208 out: 5209 kfree(sa); 5210 drop_write: 5211 mnt_drop_write_file(file); 5212 return ret; 5213 } 5214 5215 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 5216 { 5217 struct inode *inode = file_inode(file); 5218 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5219 struct btrfs_root *root = BTRFS_I(inode)->root; 5220 struct btrfs_ioctl_qgroup_create_args *sa; 5221 struct btrfs_trans_handle *trans; 5222 int ret; 5223 int err; 5224 5225 if (!capable(CAP_SYS_ADMIN)) 5226 return -EPERM; 5227 5228 ret = mnt_want_write_file(file); 5229 if (ret) 5230 return ret; 5231 5232 sa = memdup_user(arg, sizeof(*sa)); 5233 if (IS_ERR(sa)) { 5234 ret = PTR_ERR(sa); 5235 goto drop_write; 5236 } 5237 5238 if (!sa->qgroupid) { 5239 ret = -EINVAL; 5240 goto out; 5241 } 5242 5243 trans = btrfs_join_transaction(root); 5244 if (IS_ERR(trans)) { 5245 ret = PTR_ERR(trans); 5246 goto out; 5247 } 5248 5249 if (sa->create) { 5250 ret = btrfs_create_qgroup(trans, fs_info, sa->qgroupid); 5251 } else { 5252 ret = btrfs_remove_qgroup(trans, fs_info, sa->qgroupid); 5253 } 5254 5255 err = btrfs_end_transaction(trans); 5256 if (err && !ret) 5257 ret = err; 5258 5259 out: 5260 kfree(sa); 5261 drop_write: 5262 mnt_drop_write_file(file); 5263 return ret; 5264 } 5265 5266 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 5267 { 5268 struct inode *inode = file_inode(file); 5269 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5270 struct btrfs_root *root = BTRFS_I(inode)->root; 5271 struct btrfs_ioctl_qgroup_limit_args *sa; 5272 struct btrfs_trans_handle *trans; 5273 int ret; 5274 int err; 5275 u64 qgroupid; 5276 5277 if (!capable(CAP_SYS_ADMIN)) 5278 return -EPERM; 5279 5280 ret = mnt_want_write_file(file); 5281 if (ret) 5282 return ret; 5283 5284 sa = memdup_user(arg, sizeof(*sa)); 5285 if (IS_ERR(sa)) { 5286 ret = PTR_ERR(sa); 5287 goto drop_write; 5288 } 5289 5290 trans = btrfs_join_transaction(root); 5291 if (IS_ERR(trans)) { 5292 ret = PTR_ERR(trans); 5293 goto out; 5294 } 5295 5296 qgroupid = sa->qgroupid; 5297 if (!qgroupid) { 5298 /* take the current subvol as qgroup */ 5299 qgroupid = root->root_key.objectid; 5300 } 5301 5302 ret = btrfs_limit_qgroup(trans, fs_info, qgroupid, &sa->lim); 5303 5304 err = btrfs_end_transaction(trans); 5305 if (err && !ret) 5306 ret = err; 5307 5308 out: 5309 kfree(sa); 5310 drop_write: 5311 mnt_drop_write_file(file); 5312 return ret; 5313 } 5314 5315 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 5316 { 5317 struct inode *inode = file_inode(file); 5318 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5319 struct btrfs_ioctl_quota_rescan_args *qsa; 5320 int ret; 5321 5322 if (!capable(CAP_SYS_ADMIN)) 5323 return -EPERM; 5324 5325 ret = mnt_want_write_file(file); 5326 if (ret) 5327 return ret; 5328 5329 qsa = memdup_user(arg, sizeof(*qsa)); 5330 if (IS_ERR(qsa)) { 5331 ret = PTR_ERR(qsa); 5332 goto drop_write; 5333 } 5334 5335 if (qsa->flags) { 5336 ret = -EINVAL; 5337 goto out; 5338 } 5339 5340 ret = btrfs_qgroup_rescan(fs_info); 5341 5342 out: 5343 kfree(qsa); 5344 drop_write: 5345 mnt_drop_write_file(file); 5346 return ret; 5347 } 5348 5349 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg) 5350 { 5351 struct inode *inode = file_inode(file); 5352 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5353 struct btrfs_ioctl_quota_rescan_args *qsa; 5354 int ret = 0; 5355 5356 if (!capable(CAP_SYS_ADMIN)) 5357 return -EPERM; 5358 5359 qsa = kzalloc(sizeof(*qsa), GFP_KERNEL); 5360 if (!qsa) 5361 return -ENOMEM; 5362 5363 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 5364 qsa->flags = 1; 5365 qsa->progress = fs_info->qgroup_rescan_progress.objectid; 5366 } 5367 5368 if (copy_to_user(arg, qsa, sizeof(*qsa))) 5369 ret = -EFAULT; 5370 5371 kfree(qsa); 5372 return ret; 5373 } 5374 5375 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg) 5376 { 5377 struct inode *inode = file_inode(file); 5378 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5379 5380 if (!capable(CAP_SYS_ADMIN)) 5381 return -EPERM; 5382 5383 return btrfs_qgroup_wait_for_completion(fs_info, true); 5384 } 5385 5386 static long _btrfs_ioctl_set_received_subvol(struct file *file, 5387 struct btrfs_ioctl_received_subvol_args *sa) 5388 { 5389 struct inode *inode = file_inode(file); 5390 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5391 struct btrfs_root *root = BTRFS_I(inode)->root; 5392 struct btrfs_root_item *root_item = &root->root_item; 5393 struct btrfs_trans_handle *trans; 5394 struct timespec ct = current_time(inode); 5395 int ret = 0; 5396 int received_uuid_changed; 5397 5398 if (!inode_owner_or_capable(inode)) 5399 return -EPERM; 5400 5401 ret = mnt_want_write_file(file); 5402 if (ret < 0) 5403 return ret; 5404 5405 down_write(&fs_info->subvol_sem); 5406 5407 if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 5408 ret = -EINVAL; 5409 goto out; 5410 } 5411 5412 if (btrfs_root_readonly(root)) { 5413 ret = -EROFS; 5414 goto out; 5415 } 5416 5417 /* 5418 * 1 - root item 5419 * 2 - uuid items (received uuid + subvol uuid) 5420 */ 5421 trans = btrfs_start_transaction(root, 3); 5422 if (IS_ERR(trans)) { 5423 ret = PTR_ERR(trans); 5424 trans = NULL; 5425 goto out; 5426 } 5427 5428 sa->rtransid = trans->transid; 5429 sa->rtime.sec = ct.tv_sec; 5430 sa->rtime.nsec = ct.tv_nsec; 5431 5432 received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 5433 BTRFS_UUID_SIZE); 5434 if (received_uuid_changed && 5435 !btrfs_is_empty_uuid(root_item->received_uuid)) { 5436 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid, 5437 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 5438 root->root_key.objectid); 5439 if (ret && ret != -ENOENT) { 5440 btrfs_abort_transaction(trans, ret); 5441 btrfs_end_transaction(trans); 5442 goto out; 5443 } 5444 } 5445 memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 5446 btrfs_set_root_stransid(root_item, sa->stransid); 5447 btrfs_set_root_rtransid(root_item, sa->rtransid); 5448 btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 5449 btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 5450 btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 5451 btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 5452 5453 ret = btrfs_update_root(trans, fs_info->tree_root, 5454 &root->root_key, &root->root_item); 5455 if (ret < 0) { 5456 btrfs_end_transaction(trans); 5457 goto out; 5458 } 5459 if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 5460 ret = btrfs_uuid_tree_add(trans, sa->uuid, 5461 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 5462 root->root_key.objectid); 5463 if (ret < 0 && ret != -EEXIST) { 5464 btrfs_abort_transaction(trans, ret); 5465 btrfs_end_transaction(trans); 5466 goto out; 5467 } 5468 } 5469 ret = btrfs_commit_transaction(trans); 5470 out: 5471 up_write(&fs_info->subvol_sem); 5472 mnt_drop_write_file(file); 5473 return ret; 5474 } 5475 5476 #ifdef CONFIG_64BIT 5477 static long btrfs_ioctl_set_received_subvol_32(struct file *file, 5478 void __user *arg) 5479 { 5480 struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL; 5481 struct btrfs_ioctl_received_subvol_args *args64 = NULL; 5482 int ret = 0; 5483 5484 args32 = memdup_user(arg, sizeof(*args32)); 5485 if (IS_ERR(args32)) 5486 return PTR_ERR(args32); 5487 5488 args64 = kmalloc(sizeof(*args64), GFP_KERNEL); 5489 if (!args64) { 5490 ret = -ENOMEM; 5491 goto out; 5492 } 5493 5494 memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE); 5495 args64->stransid = args32->stransid; 5496 args64->rtransid = args32->rtransid; 5497 args64->stime.sec = args32->stime.sec; 5498 args64->stime.nsec = args32->stime.nsec; 5499 args64->rtime.sec = args32->rtime.sec; 5500 args64->rtime.nsec = args32->rtime.nsec; 5501 args64->flags = args32->flags; 5502 5503 ret = _btrfs_ioctl_set_received_subvol(file, args64); 5504 if (ret) 5505 goto out; 5506 5507 memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE); 5508 args32->stransid = args64->stransid; 5509 args32->rtransid = args64->rtransid; 5510 args32->stime.sec = args64->stime.sec; 5511 args32->stime.nsec = args64->stime.nsec; 5512 args32->rtime.sec = args64->rtime.sec; 5513 args32->rtime.nsec = args64->rtime.nsec; 5514 args32->flags = args64->flags; 5515 5516 ret = copy_to_user(arg, args32, sizeof(*args32)); 5517 if (ret) 5518 ret = -EFAULT; 5519 5520 out: 5521 kfree(args32); 5522 kfree(args64); 5523 return ret; 5524 } 5525 #endif 5526 5527 static long btrfs_ioctl_set_received_subvol(struct file *file, 5528 void __user *arg) 5529 { 5530 struct btrfs_ioctl_received_subvol_args *sa = NULL; 5531 int ret = 0; 5532 5533 sa = memdup_user(arg, sizeof(*sa)); 5534 if (IS_ERR(sa)) 5535 return PTR_ERR(sa); 5536 5537 ret = _btrfs_ioctl_set_received_subvol(file, sa); 5538 5539 if (ret) 5540 goto out; 5541 5542 ret = copy_to_user(arg, sa, sizeof(*sa)); 5543 if (ret) 5544 ret = -EFAULT; 5545 5546 out: 5547 kfree(sa); 5548 return ret; 5549 } 5550 5551 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg) 5552 { 5553 struct inode *inode = file_inode(file); 5554 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5555 size_t len; 5556 int ret; 5557 char label[BTRFS_LABEL_SIZE]; 5558 5559 spin_lock(&fs_info->super_lock); 5560 memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE); 5561 spin_unlock(&fs_info->super_lock); 5562 5563 len = strnlen(label, BTRFS_LABEL_SIZE); 5564 5565 if (len == BTRFS_LABEL_SIZE) { 5566 btrfs_warn(fs_info, 5567 "label is too long, return the first %zu bytes", 5568 --len); 5569 } 5570 5571 ret = copy_to_user(arg, label, len); 5572 5573 return ret ? -EFAULT : 0; 5574 } 5575 5576 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 5577 { 5578 struct inode *inode = file_inode(file); 5579 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5580 struct btrfs_root *root = BTRFS_I(inode)->root; 5581 struct btrfs_super_block *super_block = fs_info->super_copy; 5582 struct btrfs_trans_handle *trans; 5583 char label[BTRFS_LABEL_SIZE]; 5584 int ret; 5585 5586 if (!capable(CAP_SYS_ADMIN)) 5587 return -EPERM; 5588 5589 if (copy_from_user(label, arg, sizeof(label))) 5590 return -EFAULT; 5591 5592 if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 5593 btrfs_err(fs_info, 5594 "unable to set label with more than %d bytes", 5595 BTRFS_LABEL_SIZE - 1); 5596 return -EINVAL; 5597 } 5598 5599 ret = mnt_want_write_file(file); 5600 if (ret) 5601 return ret; 5602 5603 trans = btrfs_start_transaction(root, 0); 5604 if (IS_ERR(trans)) { 5605 ret = PTR_ERR(trans); 5606 goto out_unlock; 5607 } 5608 5609 spin_lock(&fs_info->super_lock); 5610 strcpy(super_block->label, label); 5611 spin_unlock(&fs_info->super_lock); 5612 ret = btrfs_commit_transaction(trans); 5613 5614 out_unlock: 5615 mnt_drop_write_file(file); 5616 return ret; 5617 } 5618 5619 #define INIT_FEATURE_FLAGS(suffix) \ 5620 { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 5621 .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 5622 .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 5623 5624 int btrfs_ioctl_get_supported_features(void __user *arg) 5625 { 5626 static const struct btrfs_ioctl_feature_flags features[3] = { 5627 INIT_FEATURE_FLAGS(SUPP), 5628 INIT_FEATURE_FLAGS(SAFE_SET), 5629 INIT_FEATURE_FLAGS(SAFE_CLEAR) 5630 }; 5631 5632 if (copy_to_user(arg, &features, sizeof(features))) 5633 return -EFAULT; 5634 5635 return 0; 5636 } 5637 5638 static int btrfs_ioctl_get_features(struct file *file, void __user *arg) 5639 { 5640 struct inode *inode = file_inode(file); 5641 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5642 struct btrfs_super_block *super_block = fs_info->super_copy; 5643 struct btrfs_ioctl_feature_flags features; 5644 5645 features.compat_flags = btrfs_super_compat_flags(super_block); 5646 features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 5647 features.incompat_flags = btrfs_super_incompat_flags(super_block); 5648 5649 if (copy_to_user(arg, &features, sizeof(features))) 5650 return -EFAULT; 5651 5652 return 0; 5653 } 5654 5655 static int check_feature_bits(struct btrfs_fs_info *fs_info, 5656 enum btrfs_feature_set set, 5657 u64 change_mask, u64 flags, u64 supported_flags, 5658 u64 safe_set, u64 safe_clear) 5659 { 5660 const char *type = btrfs_feature_set_names[set]; 5661 char *names; 5662 u64 disallowed, unsupported; 5663 u64 set_mask = flags & change_mask; 5664 u64 clear_mask = ~flags & change_mask; 5665 5666 unsupported = set_mask & ~supported_flags; 5667 if (unsupported) { 5668 names = btrfs_printable_features(set, unsupported); 5669 if (names) { 5670 btrfs_warn(fs_info, 5671 "this kernel does not support the %s feature bit%s", 5672 names, strchr(names, ',') ? "s" : ""); 5673 kfree(names); 5674 } else 5675 btrfs_warn(fs_info, 5676 "this kernel does not support %s bits 0x%llx", 5677 type, unsupported); 5678 return -EOPNOTSUPP; 5679 } 5680 5681 disallowed = set_mask & ~safe_set; 5682 if (disallowed) { 5683 names = btrfs_printable_features(set, disallowed); 5684 if (names) { 5685 btrfs_warn(fs_info, 5686 "can't set the %s feature bit%s while mounted", 5687 names, strchr(names, ',') ? "s" : ""); 5688 kfree(names); 5689 } else 5690 btrfs_warn(fs_info, 5691 "can't set %s bits 0x%llx while mounted", 5692 type, disallowed); 5693 return -EPERM; 5694 } 5695 5696 disallowed = clear_mask & ~safe_clear; 5697 if (disallowed) { 5698 names = btrfs_printable_features(set, disallowed); 5699 if (names) { 5700 btrfs_warn(fs_info, 5701 "can't clear the %s feature bit%s while mounted", 5702 names, strchr(names, ',') ? "s" : ""); 5703 kfree(names); 5704 } else 5705 btrfs_warn(fs_info, 5706 "can't clear %s bits 0x%llx while mounted", 5707 type, disallowed); 5708 return -EPERM; 5709 } 5710 5711 return 0; 5712 } 5713 5714 #define check_feature(fs_info, change_mask, flags, mask_base) \ 5715 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags, \ 5716 BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 5717 BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 5718 BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 5719 5720 static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 5721 { 5722 struct inode *inode = file_inode(file); 5723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5724 struct btrfs_root *root = BTRFS_I(inode)->root; 5725 struct btrfs_super_block *super_block = fs_info->super_copy; 5726 struct btrfs_ioctl_feature_flags flags[2]; 5727 struct btrfs_trans_handle *trans; 5728 u64 newflags; 5729 int ret; 5730 5731 if (!capable(CAP_SYS_ADMIN)) 5732 return -EPERM; 5733 5734 if (copy_from_user(flags, arg, sizeof(flags))) 5735 return -EFAULT; 5736 5737 /* Nothing to do */ 5738 if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 5739 !flags[0].incompat_flags) 5740 return 0; 5741 5742 ret = check_feature(fs_info, flags[0].compat_flags, 5743 flags[1].compat_flags, COMPAT); 5744 if (ret) 5745 return ret; 5746 5747 ret = check_feature(fs_info, flags[0].compat_ro_flags, 5748 flags[1].compat_ro_flags, COMPAT_RO); 5749 if (ret) 5750 return ret; 5751 5752 ret = check_feature(fs_info, flags[0].incompat_flags, 5753 flags[1].incompat_flags, INCOMPAT); 5754 if (ret) 5755 return ret; 5756 5757 ret = mnt_want_write_file(file); 5758 if (ret) 5759 return ret; 5760 5761 trans = btrfs_start_transaction(root, 0); 5762 if (IS_ERR(trans)) { 5763 ret = PTR_ERR(trans); 5764 goto out_drop_write; 5765 } 5766 5767 spin_lock(&fs_info->super_lock); 5768 newflags = btrfs_super_compat_flags(super_block); 5769 newflags |= flags[0].compat_flags & flags[1].compat_flags; 5770 newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 5771 btrfs_set_super_compat_flags(super_block, newflags); 5772 5773 newflags = btrfs_super_compat_ro_flags(super_block); 5774 newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 5775 newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 5776 btrfs_set_super_compat_ro_flags(super_block, newflags); 5777 5778 newflags = btrfs_super_incompat_flags(super_block); 5779 newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 5780 newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 5781 btrfs_set_super_incompat_flags(super_block, newflags); 5782 spin_unlock(&fs_info->super_lock); 5783 5784 ret = btrfs_commit_transaction(trans); 5785 out_drop_write: 5786 mnt_drop_write_file(file); 5787 5788 return ret; 5789 } 5790 5791 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat) 5792 { 5793 struct btrfs_ioctl_send_args *arg; 5794 int ret; 5795 5796 if (compat) { 5797 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 5798 struct btrfs_ioctl_send_args_32 args32; 5799 5800 ret = copy_from_user(&args32, argp, sizeof(args32)); 5801 if (ret) 5802 return -EFAULT; 5803 arg = kzalloc(sizeof(*arg), GFP_KERNEL); 5804 if (!arg) 5805 return -ENOMEM; 5806 arg->send_fd = args32.send_fd; 5807 arg->clone_sources_count = args32.clone_sources_count; 5808 arg->clone_sources = compat_ptr(args32.clone_sources); 5809 arg->parent_root = args32.parent_root; 5810 arg->flags = args32.flags; 5811 memcpy(arg->reserved, args32.reserved, 5812 sizeof(args32.reserved)); 5813 #else 5814 return -ENOTTY; 5815 #endif 5816 } else { 5817 arg = memdup_user(argp, sizeof(*arg)); 5818 if (IS_ERR(arg)) 5819 return PTR_ERR(arg); 5820 } 5821 ret = btrfs_ioctl_send(file, arg); 5822 kfree(arg); 5823 return ret; 5824 } 5825 5826 long btrfs_ioctl(struct file *file, unsigned int 5827 cmd, unsigned long arg) 5828 { 5829 struct inode *inode = file_inode(file); 5830 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 5831 struct btrfs_root *root = BTRFS_I(inode)->root; 5832 void __user *argp = (void __user *)arg; 5833 5834 switch (cmd) { 5835 case FS_IOC_GETFLAGS: 5836 return btrfs_ioctl_getflags(file, argp); 5837 case FS_IOC_SETFLAGS: 5838 return btrfs_ioctl_setflags(file, argp); 5839 case FS_IOC_GETVERSION: 5840 return btrfs_ioctl_getversion(file, argp); 5841 case FITRIM: 5842 return btrfs_ioctl_fitrim(file, argp); 5843 case BTRFS_IOC_SNAP_CREATE: 5844 return btrfs_ioctl_snap_create(file, argp, 0); 5845 case BTRFS_IOC_SNAP_CREATE_V2: 5846 return btrfs_ioctl_snap_create_v2(file, argp, 0); 5847 case BTRFS_IOC_SUBVOL_CREATE: 5848 return btrfs_ioctl_snap_create(file, argp, 1); 5849 case BTRFS_IOC_SUBVOL_CREATE_V2: 5850 return btrfs_ioctl_snap_create_v2(file, argp, 1); 5851 case BTRFS_IOC_SNAP_DESTROY: 5852 return btrfs_ioctl_snap_destroy(file, argp); 5853 case BTRFS_IOC_SUBVOL_GETFLAGS: 5854 return btrfs_ioctl_subvol_getflags(file, argp); 5855 case BTRFS_IOC_SUBVOL_SETFLAGS: 5856 return btrfs_ioctl_subvol_setflags(file, argp); 5857 case BTRFS_IOC_DEFAULT_SUBVOL: 5858 return btrfs_ioctl_default_subvol(file, argp); 5859 case BTRFS_IOC_DEFRAG: 5860 return btrfs_ioctl_defrag(file, NULL); 5861 case BTRFS_IOC_DEFRAG_RANGE: 5862 return btrfs_ioctl_defrag(file, argp); 5863 case BTRFS_IOC_RESIZE: 5864 return btrfs_ioctl_resize(file, argp); 5865 case BTRFS_IOC_ADD_DEV: 5866 return btrfs_ioctl_add_dev(fs_info, argp); 5867 case BTRFS_IOC_RM_DEV: 5868 return btrfs_ioctl_rm_dev(file, argp); 5869 case BTRFS_IOC_RM_DEV_V2: 5870 return btrfs_ioctl_rm_dev_v2(file, argp); 5871 case BTRFS_IOC_FS_INFO: 5872 return btrfs_ioctl_fs_info(fs_info, argp); 5873 case BTRFS_IOC_DEV_INFO: 5874 return btrfs_ioctl_dev_info(fs_info, argp); 5875 case BTRFS_IOC_BALANCE: 5876 return btrfs_ioctl_balance(file, NULL); 5877 case BTRFS_IOC_TREE_SEARCH: 5878 return btrfs_ioctl_tree_search(file, argp); 5879 case BTRFS_IOC_TREE_SEARCH_V2: 5880 return btrfs_ioctl_tree_search_v2(file, argp); 5881 case BTRFS_IOC_INO_LOOKUP: 5882 return btrfs_ioctl_ino_lookup(file, argp); 5883 case BTRFS_IOC_INO_PATHS: 5884 return btrfs_ioctl_ino_to_path(root, argp); 5885 case BTRFS_IOC_LOGICAL_INO: 5886 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1); 5887 case BTRFS_IOC_LOGICAL_INO_V2: 5888 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2); 5889 case BTRFS_IOC_SPACE_INFO: 5890 return btrfs_ioctl_space_info(fs_info, argp); 5891 case BTRFS_IOC_SYNC: { 5892 int ret; 5893 5894 ret = btrfs_start_delalloc_roots(fs_info, -1); 5895 if (ret) 5896 return ret; 5897 ret = btrfs_sync_fs(inode->i_sb, 1); 5898 /* 5899 * The transaction thread may want to do more work, 5900 * namely it pokes the cleaner kthread that will start 5901 * processing uncleaned subvols. 5902 */ 5903 wake_up_process(fs_info->transaction_kthread); 5904 return ret; 5905 } 5906 case BTRFS_IOC_START_SYNC: 5907 return btrfs_ioctl_start_sync(root, argp); 5908 case BTRFS_IOC_WAIT_SYNC: 5909 return btrfs_ioctl_wait_sync(fs_info, argp); 5910 case BTRFS_IOC_SCRUB: 5911 return btrfs_ioctl_scrub(file, argp); 5912 case BTRFS_IOC_SCRUB_CANCEL: 5913 return btrfs_ioctl_scrub_cancel(fs_info); 5914 case BTRFS_IOC_SCRUB_PROGRESS: 5915 return btrfs_ioctl_scrub_progress(fs_info, argp); 5916 case BTRFS_IOC_BALANCE_V2: 5917 return btrfs_ioctl_balance(file, argp); 5918 case BTRFS_IOC_BALANCE_CTL: 5919 return btrfs_ioctl_balance_ctl(fs_info, arg); 5920 case BTRFS_IOC_BALANCE_PROGRESS: 5921 return btrfs_ioctl_balance_progress(fs_info, argp); 5922 case BTRFS_IOC_SET_RECEIVED_SUBVOL: 5923 return btrfs_ioctl_set_received_subvol(file, argp); 5924 #ifdef CONFIG_64BIT 5925 case BTRFS_IOC_SET_RECEIVED_SUBVOL_32: 5926 return btrfs_ioctl_set_received_subvol_32(file, argp); 5927 #endif 5928 case BTRFS_IOC_SEND: 5929 return _btrfs_ioctl_send(file, argp, false); 5930 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 5931 case BTRFS_IOC_SEND_32: 5932 return _btrfs_ioctl_send(file, argp, true); 5933 #endif 5934 case BTRFS_IOC_GET_DEV_STATS: 5935 return btrfs_ioctl_get_dev_stats(fs_info, argp); 5936 case BTRFS_IOC_QUOTA_CTL: 5937 return btrfs_ioctl_quota_ctl(file, argp); 5938 case BTRFS_IOC_QGROUP_ASSIGN: 5939 return btrfs_ioctl_qgroup_assign(file, argp); 5940 case BTRFS_IOC_QGROUP_CREATE: 5941 return btrfs_ioctl_qgroup_create(file, argp); 5942 case BTRFS_IOC_QGROUP_LIMIT: 5943 return btrfs_ioctl_qgroup_limit(file, argp); 5944 case BTRFS_IOC_QUOTA_RESCAN: 5945 return btrfs_ioctl_quota_rescan(file, argp); 5946 case BTRFS_IOC_QUOTA_RESCAN_STATUS: 5947 return btrfs_ioctl_quota_rescan_status(file, argp); 5948 case BTRFS_IOC_QUOTA_RESCAN_WAIT: 5949 return btrfs_ioctl_quota_rescan_wait(file, argp); 5950 case BTRFS_IOC_DEV_REPLACE: 5951 return btrfs_ioctl_dev_replace(fs_info, argp); 5952 case BTRFS_IOC_GET_FSLABEL: 5953 return btrfs_ioctl_get_fslabel(file, argp); 5954 case BTRFS_IOC_SET_FSLABEL: 5955 return btrfs_ioctl_set_fslabel(file, argp); 5956 case BTRFS_IOC_GET_SUPPORTED_FEATURES: 5957 return btrfs_ioctl_get_supported_features(argp); 5958 case BTRFS_IOC_GET_FEATURES: 5959 return btrfs_ioctl_get_features(file, argp); 5960 case BTRFS_IOC_SET_FEATURES: 5961 return btrfs_ioctl_set_features(file, argp); 5962 case FS_IOC_FSGETXATTR: 5963 return btrfs_ioctl_fsgetxattr(file, argp); 5964 case FS_IOC_FSSETXATTR: 5965 return btrfs_ioctl_fssetxattr(file, argp); 5966 case BTRFS_IOC_GET_SUBVOL_INFO: 5967 return btrfs_ioctl_get_subvol_info(file, argp); 5968 case BTRFS_IOC_GET_SUBVOL_ROOTREF: 5969 return btrfs_ioctl_get_subvol_rootref(file, argp); 5970 case BTRFS_IOC_INO_LOOKUP_USER: 5971 return btrfs_ioctl_ino_lookup_user(file, argp); 5972 } 5973 5974 return -ENOTTY; 5975 } 5976 5977 #ifdef CONFIG_COMPAT 5978 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 5979 { 5980 /* 5981 * These all access 32-bit values anyway so no further 5982 * handling is necessary. 5983 */ 5984 switch (cmd) { 5985 case FS_IOC32_GETFLAGS: 5986 cmd = FS_IOC_GETFLAGS; 5987 break; 5988 case FS_IOC32_SETFLAGS: 5989 cmd = FS_IOC_SETFLAGS; 5990 break; 5991 case FS_IOC32_GETVERSION: 5992 cmd = FS_IOC_GETVERSION; 5993 break; 5994 } 5995 5996 return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 5997 } 5998 #endif 5999