1 /* 2 * Copyright (C) 2007 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/kernel.h> 20 #include <linux/bio.h> 21 #include <linux/buffer_head.h> 22 #include <linux/file.h> 23 #include <linux/fs.h> 24 #include <linux/pagemap.h> 25 #include <linux/highmem.h> 26 #include <linux/time.h> 27 #include <linux/init.h> 28 #include <linux/string.h> 29 #include <linux/backing-dev.h> 30 #include <linux/mpage.h> 31 #include <linux/swap.h> 32 #include <linux/writeback.h> 33 #include <linux/statfs.h> 34 #include <linux/compat.h> 35 #include <linux/bit_spinlock.h> 36 #include <linux/xattr.h> 37 #include <linux/posix_acl.h> 38 #include <linux/falloc.h> 39 #include "compat.h" 40 #include "ctree.h" 41 #include "disk-io.h" 42 #include "transaction.h" 43 #include "btrfs_inode.h" 44 #include "ioctl.h" 45 #include "print-tree.h" 46 #include "volumes.h" 47 #include "ordered-data.h" 48 #include "xattr.h" 49 #include "tree-log.h" 50 #include "compression.h" 51 #include "locking.h" 52 53 struct btrfs_iget_args { 54 u64 ino; 55 struct btrfs_root *root; 56 }; 57 58 static const struct inode_operations btrfs_dir_inode_operations; 59 static const struct inode_operations btrfs_symlink_inode_operations; 60 static const struct inode_operations btrfs_dir_ro_inode_operations; 61 static const struct inode_operations btrfs_special_inode_operations; 62 static const struct inode_operations btrfs_file_inode_operations; 63 static const struct address_space_operations btrfs_aops; 64 static const struct address_space_operations btrfs_symlink_aops; 65 static const struct file_operations btrfs_dir_file_operations; 66 static struct extent_io_ops btrfs_extent_io_ops; 67 68 static struct kmem_cache *btrfs_inode_cachep; 69 struct kmem_cache *btrfs_trans_handle_cachep; 70 struct kmem_cache *btrfs_transaction_cachep; 71 struct kmem_cache *btrfs_path_cachep; 72 73 #define S_SHIFT 12 74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { 75 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, 76 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, 77 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, 78 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, 79 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, 80 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, 81 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, 82 }; 83 84 static void btrfs_truncate(struct inode *inode); 85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); 86 static noinline int cow_file_range(struct inode *inode, 87 struct page *locked_page, 88 u64 start, u64 end, int *page_started, 89 unsigned long *nr_written, int unlock); 90 91 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, 92 struct inode *inode, struct inode *dir) 93 { 94 int err; 95 96 err = btrfs_init_acl(trans, inode, dir); 97 if (!err) 98 err = btrfs_xattr_security_init(trans, inode, dir); 99 return err; 100 } 101 102 /* 103 * this does all the hard work for inserting an inline extent into 104 * the btree. The caller should have done a btrfs_drop_extents so that 105 * no overlapping inline items exist in the btree 106 */ 107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, 108 struct btrfs_root *root, struct inode *inode, 109 u64 start, size_t size, size_t compressed_size, 110 struct page **compressed_pages) 111 { 112 struct btrfs_key key; 113 struct btrfs_path *path; 114 struct extent_buffer *leaf; 115 struct page *page = NULL; 116 char *kaddr; 117 unsigned long ptr; 118 struct btrfs_file_extent_item *ei; 119 int err = 0; 120 int ret; 121 size_t cur_size = size; 122 size_t datasize; 123 unsigned long offset; 124 int use_compress = 0; 125 126 if (compressed_size && compressed_pages) { 127 use_compress = 1; 128 cur_size = compressed_size; 129 } 130 131 path = btrfs_alloc_path(); 132 if (!path) 133 return -ENOMEM; 134 135 path->leave_spinning = 1; 136 btrfs_set_trans_block_group(trans, inode); 137 138 key.objectid = inode->i_ino; 139 key.offset = start; 140 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); 141 datasize = btrfs_file_extent_calc_inline_size(cur_size); 142 143 inode_add_bytes(inode, size); 144 ret = btrfs_insert_empty_item(trans, root, path, &key, 145 datasize); 146 BUG_ON(ret); 147 if (ret) { 148 err = ret; 149 goto fail; 150 } 151 leaf = path->nodes[0]; 152 ei = btrfs_item_ptr(leaf, path->slots[0], 153 struct btrfs_file_extent_item); 154 btrfs_set_file_extent_generation(leaf, ei, trans->transid); 155 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); 156 btrfs_set_file_extent_encryption(leaf, ei, 0); 157 btrfs_set_file_extent_other_encoding(leaf, ei, 0); 158 btrfs_set_file_extent_ram_bytes(leaf, ei, size); 159 ptr = btrfs_file_extent_inline_start(ei); 160 161 if (use_compress) { 162 struct page *cpage; 163 int i = 0; 164 while (compressed_size > 0) { 165 cpage = compressed_pages[i]; 166 cur_size = min_t(unsigned long, compressed_size, 167 PAGE_CACHE_SIZE); 168 169 kaddr = kmap_atomic(cpage, KM_USER0); 170 write_extent_buffer(leaf, kaddr, ptr, cur_size); 171 kunmap_atomic(kaddr, KM_USER0); 172 173 i++; 174 ptr += cur_size; 175 compressed_size -= cur_size; 176 } 177 btrfs_set_file_extent_compression(leaf, ei, 178 BTRFS_COMPRESS_ZLIB); 179 } else { 180 page = find_get_page(inode->i_mapping, 181 start >> PAGE_CACHE_SHIFT); 182 btrfs_set_file_extent_compression(leaf, ei, 0); 183 kaddr = kmap_atomic(page, KM_USER0); 184 offset = start & (PAGE_CACHE_SIZE - 1); 185 write_extent_buffer(leaf, kaddr + offset, ptr, size); 186 kunmap_atomic(kaddr, KM_USER0); 187 page_cache_release(page); 188 } 189 btrfs_mark_buffer_dirty(leaf); 190 btrfs_free_path(path); 191 192 /* 193 * we're an inline extent, so nobody can 194 * extend the file past i_size without locking 195 * a page we already have locked. 196 * 197 * We must do any isize and inode updates 198 * before we unlock the pages. Otherwise we 199 * could end up racing with unlink. 200 */ 201 BTRFS_I(inode)->disk_i_size = inode->i_size; 202 btrfs_update_inode(trans, root, inode); 203 204 return 0; 205 fail: 206 btrfs_free_path(path); 207 return err; 208 } 209 210 211 /* 212 * conditionally insert an inline extent into the file. This 213 * does the checks required to make sure the data is small enough 214 * to fit as an inline extent. 215 */ 216 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, 217 struct btrfs_root *root, 218 struct inode *inode, u64 start, u64 end, 219 size_t compressed_size, 220 struct page **compressed_pages) 221 { 222 u64 isize = i_size_read(inode); 223 u64 actual_end = min(end + 1, isize); 224 u64 inline_len = actual_end - start; 225 u64 aligned_end = (end + root->sectorsize - 1) & 226 ~((u64)root->sectorsize - 1); 227 u64 hint_byte; 228 u64 data_len = inline_len; 229 int ret; 230 231 if (compressed_size) 232 data_len = compressed_size; 233 234 if (start > 0 || 235 actual_end >= PAGE_CACHE_SIZE || 236 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || 237 (!compressed_size && 238 (actual_end & (root->sectorsize - 1)) == 0) || 239 end + 1 < isize || 240 data_len > root->fs_info->max_inline) { 241 return 1; 242 } 243 244 ret = btrfs_drop_extents(trans, inode, start, aligned_end, 245 &hint_byte, 1); 246 BUG_ON(ret); 247 248 if (isize > actual_end) 249 inline_len = min_t(u64, isize, actual_end); 250 ret = insert_inline_extent(trans, root, inode, start, 251 inline_len, compressed_size, 252 compressed_pages); 253 BUG_ON(ret); 254 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); 255 return 0; 256 } 257 258 struct async_extent { 259 u64 start; 260 u64 ram_size; 261 u64 compressed_size; 262 struct page **pages; 263 unsigned long nr_pages; 264 struct list_head list; 265 }; 266 267 struct async_cow { 268 struct inode *inode; 269 struct btrfs_root *root; 270 struct page *locked_page; 271 u64 start; 272 u64 end; 273 struct list_head extents; 274 struct btrfs_work work; 275 }; 276 277 static noinline int add_async_extent(struct async_cow *cow, 278 u64 start, u64 ram_size, 279 u64 compressed_size, 280 struct page **pages, 281 unsigned long nr_pages) 282 { 283 struct async_extent *async_extent; 284 285 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); 286 async_extent->start = start; 287 async_extent->ram_size = ram_size; 288 async_extent->compressed_size = compressed_size; 289 async_extent->pages = pages; 290 async_extent->nr_pages = nr_pages; 291 list_add_tail(&async_extent->list, &cow->extents); 292 return 0; 293 } 294 295 /* 296 * we create compressed extents in two phases. The first 297 * phase compresses a range of pages that have already been 298 * locked (both pages and state bits are locked). 299 * 300 * This is done inside an ordered work queue, and the compression 301 * is spread across many cpus. The actual IO submission is step 302 * two, and the ordered work queue takes care of making sure that 303 * happens in the same order things were put onto the queue by 304 * writepages and friends. 305 * 306 * If this code finds it can't get good compression, it puts an 307 * entry onto the work queue to write the uncompressed bytes. This 308 * makes sure that both compressed inodes and uncompressed inodes 309 * are written in the same order that pdflush sent them down. 310 */ 311 static noinline int compress_file_range(struct inode *inode, 312 struct page *locked_page, 313 u64 start, u64 end, 314 struct async_cow *async_cow, 315 int *num_added) 316 { 317 struct btrfs_root *root = BTRFS_I(inode)->root; 318 struct btrfs_trans_handle *trans; 319 u64 num_bytes; 320 u64 orig_start; 321 u64 disk_num_bytes; 322 u64 blocksize = root->sectorsize; 323 u64 actual_end; 324 u64 isize = i_size_read(inode); 325 int ret = 0; 326 struct page **pages = NULL; 327 unsigned long nr_pages; 328 unsigned long nr_pages_ret = 0; 329 unsigned long total_compressed = 0; 330 unsigned long total_in = 0; 331 unsigned long max_compressed = 128 * 1024; 332 unsigned long max_uncompressed = 128 * 1024; 333 int i; 334 int will_compress; 335 336 orig_start = start; 337 338 actual_end = min_t(u64, isize, end + 1); 339 again: 340 will_compress = 0; 341 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; 342 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); 343 344 /* 345 * we don't want to send crud past the end of i_size through 346 * compression, that's just a waste of CPU time. So, if the 347 * end of the file is before the start of our current 348 * requested range of bytes, we bail out to the uncompressed 349 * cleanup code that can deal with all of this. 350 * 351 * It isn't really the fastest way to fix things, but this is a 352 * very uncommon corner. 353 */ 354 if (actual_end <= start) 355 goto cleanup_and_bail_uncompressed; 356 357 total_compressed = actual_end - start; 358 359 /* we want to make sure that amount of ram required to uncompress 360 * an extent is reasonable, so we limit the total size in ram 361 * of a compressed extent to 128k. This is a crucial number 362 * because it also controls how easily we can spread reads across 363 * cpus for decompression. 364 * 365 * We also want to make sure the amount of IO required to do 366 * a random read is reasonably small, so we limit the size of 367 * a compressed extent to 128k. 368 */ 369 total_compressed = min(total_compressed, max_uncompressed); 370 num_bytes = (end - start + blocksize) & ~(blocksize - 1); 371 num_bytes = max(blocksize, num_bytes); 372 disk_num_bytes = num_bytes; 373 total_in = 0; 374 ret = 0; 375 376 /* 377 * we do compression for mount -o compress and when the 378 * inode has not been flagged as nocompress. This flag can 379 * change at any time if we discover bad compression ratios. 380 */ 381 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && 382 btrfs_test_opt(root, COMPRESS)) { 383 WARN_ON(pages); 384 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); 385 386 ret = btrfs_zlib_compress_pages(inode->i_mapping, start, 387 total_compressed, pages, 388 nr_pages, &nr_pages_ret, 389 &total_in, 390 &total_compressed, 391 max_compressed); 392 393 if (!ret) { 394 unsigned long offset = total_compressed & 395 (PAGE_CACHE_SIZE - 1); 396 struct page *page = pages[nr_pages_ret - 1]; 397 char *kaddr; 398 399 /* zero the tail end of the last page, we might be 400 * sending it down to disk 401 */ 402 if (offset) { 403 kaddr = kmap_atomic(page, KM_USER0); 404 memset(kaddr + offset, 0, 405 PAGE_CACHE_SIZE - offset); 406 kunmap_atomic(kaddr, KM_USER0); 407 } 408 will_compress = 1; 409 } 410 } 411 if (start == 0) { 412 trans = btrfs_join_transaction(root, 1); 413 BUG_ON(!trans); 414 btrfs_set_trans_block_group(trans, inode); 415 416 /* lets try to make an inline extent */ 417 if (ret || total_in < (actual_end - start)) { 418 /* we didn't compress the entire range, try 419 * to make an uncompressed inline extent. 420 */ 421 ret = cow_file_range_inline(trans, root, inode, 422 start, end, 0, NULL); 423 } else { 424 /* try making a compressed inline extent */ 425 ret = cow_file_range_inline(trans, root, inode, 426 start, end, 427 total_compressed, pages); 428 } 429 if (ret == 0) { 430 /* 431 * inline extent creation worked, we don't need 432 * to create any more async work items. Unlock 433 * and free up our temp pages. 434 */ 435 extent_clear_unlock_delalloc(inode, 436 &BTRFS_I(inode)->io_tree, 437 start, end, NULL, 438 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | 439 EXTENT_CLEAR_DELALLOC | 440 EXTENT_CLEAR_ACCOUNTING | 441 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); 442 443 btrfs_end_transaction(trans, root); 444 goto free_pages_out; 445 } 446 btrfs_end_transaction(trans, root); 447 } 448 449 if (will_compress) { 450 /* 451 * we aren't doing an inline extent round the compressed size 452 * up to a block size boundary so the allocator does sane 453 * things 454 */ 455 total_compressed = (total_compressed + blocksize - 1) & 456 ~(blocksize - 1); 457 458 /* 459 * one last check to make sure the compression is really a 460 * win, compare the page count read with the blocks on disk 461 */ 462 total_in = (total_in + PAGE_CACHE_SIZE - 1) & 463 ~(PAGE_CACHE_SIZE - 1); 464 if (total_compressed >= total_in) { 465 will_compress = 0; 466 } else { 467 disk_num_bytes = total_compressed; 468 num_bytes = total_in; 469 } 470 } 471 if (!will_compress && pages) { 472 /* 473 * the compression code ran but failed to make things smaller, 474 * free any pages it allocated and our page pointer array 475 */ 476 for (i = 0; i < nr_pages_ret; i++) { 477 WARN_ON(pages[i]->mapping); 478 page_cache_release(pages[i]); 479 } 480 kfree(pages); 481 pages = NULL; 482 total_compressed = 0; 483 nr_pages_ret = 0; 484 485 /* flag the file so we don't compress in the future */ 486 if (!btrfs_test_opt(root, FORCE_COMPRESS)) 487 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; 488 } 489 if (will_compress) { 490 *num_added += 1; 491 492 /* the async work queues will take care of doing actual 493 * allocation on disk for these compressed pages, 494 * and will submit them to the elevator. 495 */ 496 add_async_extent(async_cow, start, num_bytes, 497 total_compressed, pages, nr_pages_ret); 498 499 if (start + num_bytes < end && start + num_bytes < actual_end) { 500 start += num_bytes; 501 pages = NULL; 502 cond_resched(); 503 goto again; 504 } 505 } else { 506 cleanup_and_bail_uncompressed: 507 /* 508 * No compression, but we still need to write the pages in 509 * the file we've been given so far. redirty the locked 510 * page if it corresponds to our extent and set things up 511 * for the async work queue to run cow_file_range to do 512 * the normal delalloc dance 513 */ 514 if (page_offset(locked_page) >= start && 515 page_offset(locked_page) <= end) { 516 __set_page_dirty_nobuffers(locked_page); 517 /* unlocked later on in the async handlers */ 518 } 519 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); 520 *num_added += 1; 521 } 522 523 out: 524 return 0; 525 526 free_pages_out: 527 for (i = 0; i < nr_pages_ret; i++) { 528 WARN_ON(pages[i]->mapping); 529 page_cache_release(pages[i]); 530 } 531 kfree(pages); 532 533 goto out; 534 } 535 536 /* 537 * phase two of compressed writeback. This is the ordered portion 538 * of the code, which only gets called in the order the work was 539 * queued. We walk all the async extents created by compress_file_range 540 * and send them down to the disk. 541 */ 542 static noinline int submit_compressed_extents(struct inode *inode, 543 struct async_cow *async_cow) 544 { 545 struct async_extent *async_extent; 546 u64 alloc_hint = 0; 547 struct btrfs_trans_handle *trans; 548 struct btrfs_key ins; 549 struct extent_map *em; 550 struct btrfs_root *root = BTRFS_I(inode)->root; 551 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 552 struct extent_io_tree *io_tree; 553 int ret = 0; 554 555 if (list_empty(&async_cow->extents)) 556 return 0; 557 558 559 while (!list_empty(&async_cow->extents)) { 560 async_extent = list_entry(async_cow->extents.next, 561 struct async_extent, list); 562 list_del(&async_extent->list); 563 564 io_tree = &BTRFS_I(inode)->io_tree; 565 566 retry: 567 /* did the compression code fall back to uncompressed IO? */ 568 if (!async_extent->pages) { 569 int page_started = 0; 570 unsigned long nr_written = 0; 571 572 lock_extent(io_tree, async_extent->start, 573 async_extent->start + 574 async_extent->ram_size - 1, GFP_NOFS); 575 576 /* allocate blocks */ 577 ret = cow_file_range(inode, async_cow->locked_page, 578 async_extent->start, 579 async_extent->start + 580 async_extent->ram_size - 1, 581 &page_started, &nr_written, 0); 582 583 /* 584 * if page_started, cow_file_range inserted an 585 * inline extent and took care of all the unlocking 586 * and IO for us. Otherwise, we need to submit 587 * all those pages down to the drive. 588 */ 589 if (!page_started && !ret) 590 extent_write_locked_range(io_tree, 591 inode, async_extent->start, 592 async_extent->start + 593 async_extent->ram_size - 1, 594 btrfs_get_extent, 595 WB_SYNC_ALL); 596 kfree(async_extent); 597 cond_resched(); 598 continue; 599 } 600 601 lock_extent(io_tree, async_extent->start, 602 async_extent->start + async_extent->ram_size - 1, 603 GFP_NOFS); 604 605 trans = btrfs_join_transaction(root, 1); 606 ret = btrfs_reserve_extent(trans, root, 607 async_extent->compressed_size, 608 async_extent->compressed_size, 609 0, alloc_hint, 610 (u64)-1, &ins, 1); 611 btrfs_end_transaction(trans, root); 612 613 if (ret) { 614 int i; 615 for (i = 0; i < async_extent->nr_pages; i++) { 616 WARN_ON(async_extent->pages[i]->mapping); 617 page_cache_release(async_extent->pages[i]); 618 } 619 kfree(async_extent->pages); 620 async_extent->nr_pages = 0; 621 async_extent->pages = NULL; 622 unlock_extent(io_tree, async_extent->start, 623 async_extent->start + 624 async_extent->ram_size - 1, GFP_NOFS); 625 goto retry; 626 } 627 628 /* 629 * here we're doing allocation and writeback of the 630 * compressed pages 631 */ 632 btrfs_drop_extent_cache(inode, async_extent->start, 633 async_extent->start + 634 async_extent->ram_size - 1, 0); 635 636 em = alloc_extent_map(GFP_NOFS); 637 em->start = async_extent->start; 638 em->len = async_extent->ram_size; 639 em->orig_start = em->start; 640 641 em->block_start = ins.objectid; 642 em->block_len = ins.offset; 643 em->bdev = root->fs_info->fs_devices->latest_bdev; 644 set_bit(EXTENT_FLAG_PINNED, &em->flags); 645 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 646 647 while (1) { 648 write_lock(&em_tree->lock); 649 ret = add_extent_mapping(em_tree, em); 650 write_unlock(&em_tree->lock); 651 if (ret != -EEXIST) { 652 free_extent_map(em); 653 break; 654 } 655 btrfs_drop_extent_cache(inode, async_extent->start, 656 async_extent->start + 657 async_extent->ram_size - 1, 0); 658 } 659 660 ret = btrfs_add_ordered_extent(inode, async_extent->start, 661 ins.objectid, 662 async_extent->ram_size, 663 ins.offset, 664 BTRFS_ORDERED_COMPRESSED); 665 BUG_ON(ret); 666 667 /* 668 * clear dirty, set writeback and unlock the pages. 669 */ 670 extent_clear_unlock_delalloc(inode, 671 &BTRFS_I(inode)->io_tree, 672 async_extent->start, 673 async_extent->start + 674 async_extent->ram_size - 1, 675 NULL, EXTENT_CLEAR_UNLOCK_PAGE | 676 EXTENT_CLEAR_UNLOCK | 677 EXTENT_CLEAR_DELALLOC | 678 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); 679 680 ret = btrfs_submit_compressed_write(inode, 681 async_extent->start, 682 async_extent->ram_size, 683 ins.objectid, 684 ins.offset, async_extent->pages, 685 async_extent->nr_pages); 686 687 BUG_ON(ret); 688 alloc_hint = ins.objectid + ins.offset; 689 kfree(async_extent); 690 cond_resched(); 691 } 692 693 return 0; 694 } 695 696 /* 697 * when extent_io.c finds a delayed allocation range in the file, 698 * the call backs end up in this code. The basic idea is to 699 * allocate extents on disk for the range, and create ordered data structs 700 * in ram to track those extents. 701 * 702 * locked_page is the page that writepage had locked already. We use 703 * it to make sure we don't do extra locks or unlocks. 704 * 705 * *page_started is set to one if we unlock locked_page and do everything 706 * required to start IO on it. It may be clean and already done with 707 * IO when we return. 708 */ 709 static noinline int cow_file_range(struct inode *inode, 710 struct page *locked_page, 711 u64 start, u64 end, int *page_started, 712 unsigned long *nr_written, 713 int unlock) 714 { 715 struct btrfs_root *root = BTRFS_I(inode)->root; 716 struct btrfs_trans_handle *trans; 717 u64 alloc_hint = 0; 718 u64 num_bytes; 719 unsigned long ram_size; 720 u64 disk_num_bytes; 721 u64 cur_alloc_size; 722 u64 blocksize = root->sectorsize; 723 u64 actual_end; 724 u64 isize = i_size_read(inode); 725 struct btrfs_key ins; 726 struct extent_map *em; 727 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 728 int ret = 0; 729 730 trans = btrfs_join_transaction(root, 1); 731 BUG_ON(!trans); 732 btrfs_set_trans_block_group(trans, inode); 733 734 actual_end = min_t(u64, isize, end + 1); 735 736 num_bytes = (end - start + blocksize) & ~(blocksize - 1); 737 num_bytes = max(blocksize, num_bytes); 738 disk_num_bytes = num_bytes; 739 ret = 0; 740 741 if (start == 0) { 742 /* lets try to make an inline extent */ 743 ret = cow_file_range_inline(trans, root, inode, 744 start, end, 0, NULL); 745 if (ret == 0) { 746 extent_clear_unlock_delalloc(inode, 747 &BTRFS_I(inode)->io_tree, 748 start, end, NULL, 749 EXTENT_CLEAR_UNLOCK_PAGE | 750 EXTENT_CLEAR_UNLOCK | 751 EXTENT_CLEAR_DELALLOC | 752 EXTENT_CLEAR_ACCOUNTING | 753 EXTENT_CLEAR_DIRTY | 754 EXTENT_SET_WRITEBACK | 755 EXTENT_END_WRITEBACK); 756 757 *nr_written = *nr_written + 758 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; 759 *page_started = 1; 760 ret = 0; 761 goto out; 762 } 763 } 764 765 BUG_ON(disk_num_bytes > 766 btrfs_super_total_bytes(&root->fs_info->super_copy)); 767 768 769 read_lock(&BTRFS_I(inode)->extent_tree.lock); 770 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, 771 start, num_bytes); 772 if (em) { 773 /* 774 * if block start isn't an actual block number then find the 775 * first block in this inode and use that as a hint. If that 776 * block is also bogus then just don't worry about it. 777 */ 778 if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 779 free_extent_map(em); 780 em = search_extent_mapping(em_tree, 0, 0); 781 if (em && em->block_start < EXTENT_MAP_LAST_BYTE) 782 alloc_hint = em->block_start; 783 if (em) 784 free_extent_map(em); 785 } else { 786 alloc_hint = em->block_start; 787 free_extent_map(em); 788 } 789 } 790 read_unlock(&BTRFS_I(inode)->extent_tree.lock); 791 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); 792 793 while (disk_num_bytes > 0) { 794 unsigned long op; 795 796 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); 797 ret = btrfs_reserve_extent(trans, root, cur_alloc_size, 798 root->sectorsize, 0, alloc_hint, 799 (u64)-1, &ins, 1); 800 BUG_ON(ret); 801 802 em = alloc_extent_map(GFP_NOFS); 803 em->start = start; 804 em->orig_start = em->start; 805 ram_size = ins.offset; 806 em->len = ins.offset; 807 808 em->block_start = ins.objectid; 809 em->block_len = ins.offset; 810 em->bdev = root->fs_info->fs_devices->latest_bdev; 811 set_bit(EXTENT_FLAG_PINNED, &em->flags); 812 813 while (1) { 814 write_lock(&em_tree->lock); 815 ret = add_extent_mapping(em_tree, em); 816 write_unlock(&em_tree->lock); 817 if (ret != -EEXIST) { 818 free_extent_map(em); 819 break; 820 } 821 btrfs_drop_extent_cache(inode, start, 822 start + ram_size - 1, 0); 823 } 824 825 cur_alloc_size = ins.offset; 826 ret = btrfs_add_ordered_extent(inode, start, ins.objectid, 827 ram_size, cur_alloc_size, 0); 828 BUG_ON(ret); 829 830 if (root->root_key.objectid == 831 BTRFS_DATA_RELOC_TREE_OBJECTID) { 832 ret = btrfs_reloc_clone_csums(inode, start, 833 cur_alloc_size); 834 BUG_ON(ret); 835 } 836 837 if (disk_num_bytes < cur_alloc_size) 838 break; 839 840 /* we're not doing compressed IO, don't unlock the first 841 * page (which the caller expects to stay locked), don't 842 * clear any dirty bits and don't set any writeback bits 843 * 844 * Do set the Private2 bit so we know this page was properly 845 * setup for writepage 846 */ 847 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; 848 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | 849 EXTENT_SET_PRIVATE2; 850 851 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 852 start, start + ram_size - 1, 853 locked_page, op); 854 disk_num_bytes -= cur_alloc_size; 855 num_bytes -= cur_alloc_size; 856 alloc_hint = ins.objectid + ins.offset; 857 start += cur_alloc_size; 858 } 859 out: 860 ret = 0; 861 btrfs_end_transaction(trans, root); 862 863 return ret; 864 } 865 866 /* 867 * work queue call back to started compression on a file and pages 868 */ 869 static noinline void async_cow_start(struct btrfs_work *work) 870 { 871 struct async_cow *async_cow; 872 int num_added = 0; 873 async_cow = container_of(work, struct async_cow, work); 874 875 compress_file_range(async_cow->inode, async_cow->locked_page, 876 async_cow->start, async_cow->end, async_cow, 877 &num_added); 878 if (num_added == 0) 879 async_cow->inode = NULL; 880 } 881 882 /* 883 * work queue call back to submit previously compressed pages 884 */ 885 static noinline void async_cow_submit(struct btrfs_work *work) 886 { 887 struct async_cow *async_cow; 888 struct btrfs_root *root; 889 unsigned long nr_pages; 890 891 async_cow = container_of(work, struct async_cow, work); 892 893 root = async_cow->root; 894 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> 895 PAGE_CACHE_SHIFT; 896 897 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); 898 899 if (atomic_read(&root->fs_info->async_delalloc_pages) < 900 5 * 1042 * 1024 && 901 waitqueue_active(&root->fs_info->async_submit_wait)) 902 wake_up(&root->fs_info->async_submit_wait); 903 904 if (async_cow->inode) 905 submit_compressed_extents(async_cow->inode, async_cow); 906 } 907 908 static noinline void async_cow_free(struct btrfs_work *work) 909 { 910 struct async_cow *async_cow; 911 async_cow = container_of(work, struct async_cow, work); 912 kfree(async_cow); 913 } 914 915 static int cow_file_range_async(struct inode *inode, struct page *locked_page, 916 u64 start, u64 end, int *page_started, 917 unsigned long *nr_written) 918 { 919 struct async_cow *async_cow; 920 struct btrfs_root *root = BTRFS_I(inode)->root; 921 unsigned long nr_pages; 922 u64 cur_end; 923 int limit = 10 * 1024 * 1042; 924 925 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, 926 1, 0, NULL, GFP_NOFS); 927 while (start < end) { 928 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); 929 async_cow->inode = inode; 930 async_cow->root = root; 931 async_cow->locked_page = locked_page; 932 async_cow->start = start; 933 934 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) 935 cur_end = end; 936 else 937 cur_end = min(end, start + 512 * 1024 - 1); 938 939 async_cow->end = cur_end; 940 INIT_LIST_HEAD(&async_cow->extents); 941 942 async_cow->work.func = async_cow_start; 943 async_cow->work.ordered_func = async_cow_submit; 944 async_cow->work.ordered_free = async_cow_free; 945 async_cow->work.flags = 0; 946 947 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> 948 PAGE_CACHE_SHIFT; 949 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); 950 951 btrfs_queue_worker(&root->fs_info->delalloc_workers, 952 &async_cow->work); 953 954 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { 955 wait_event(root->fs_info->async_submit_wait, 956 (atomic_read(&root->fs_info->async_delalloc_pages) < 957 limit)); 958 } 959 960 while (atomic_read(&root->fs_info->async_submit_draining) && 961 atomic_read(&root->fs_info->async_delalloc_pages)) { 962 wait_event(root->fs_info->async_submit_wait, 963 (atomic_read(&root->fs_info->async_delalloc_pages) == 964 0)); 965 } 966 967 *nr_written += nr_pages; 968 start = cur_end + 1; 969 } 970 *page_started = 1; 971 return 0; 972 } 973 974 static noinline int csum_exist_in_range(struct btrfs_root *root, 975 u64 bytenr, u64 num_bytes) 976 { 977 int ret; 978 struct btrfs_ordered_sum *sums; 979 LIST_HEAD(list); 980 981 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, 982 bytenr + num_bytes - 1, &list); 983 if (ret == 0 && list_empty(&list)) 984 return 0; 985 986 while (!list_empty(&list)) { 987 sums = list_entry(list.next, struct btrfs_ordered_sum, list); 988 list_del(&sums->list); 989 kfree(sums); 990 } 991 return 1; 992 } 993 994 /* 995 * when nowcow writeback call back. This checks for snapshots or COW copies 996 * of the extents that exist in the file, and COWs the file as required. 997 * 998 * If no cow copies or snapshots exist, we write directly to the existing 999 * blocks on disk 1000 */ 1001 static noinline int run_delalloc_nocow(struct inode *inode, 1002 struct page *locked_page, 1003 u64 start, u64 end, int *page_started, int force, 1004 unsigned long *nr_written) 1005 { 1006 struct btrfs_root *root = BTRFS_I(inode)->root; 1007 struct btrfs_trans_handle *trans; 1008 struct extent_buffer *leaf; 1009 struct btrfs_path *path; 1010 struct btrfs_file_extent_item *fi; 1011 struct btrfs_key found_key; 1012 u64 cow_start; 1013 u64 cur_offset; 1014 u64 extent_end; 1015 u64 extent_offset; 1016 u64 disk_bytenr; 1017 u64 num_bytes; 1018 int extent_type; 1019 int ret; 1020 int type; 1021 int nocow; 1022 int check_prev = 1; 1023 1024 path = btrfs_alloc_path(); 1025 BUG_ON(!path); 1026 trans = btrfs_join_transaction(root, 1); 1027 BUG_ON(!trans); 1028 1029 cow_start = (u64)-1; 1030 cur_offset = start; 1031 while (1) { 1032 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, 1033 cur_offset, 0); 1034 BUG_ON(ret < 0); 1035 if (ret > 0 && path->slots[0] > 0 && check_prev) { 1036 leaf = path->nodes[0]; 1037 btrfs_item_key_to_cpu(leaf, &found_key, 1038 path->slots[0] - 1); 1039 if (found_key.objectid == inode->i_ino && 1040 found_key.type == BTRFS_EXTENT_DATA_KEY) 1041 path->slots[0]--; 1042 } 1043 check_prev = 0; 1044 next_slot: 1045 leaf = path->nodes[0]; 1046 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 1047 ret = btrfs_next_leaf(root, path); 1048 if (ret < 0) 1049 BUG_ON(1); 1050 if (ret > 0) 1051 break; 1052 leaf = path->nodes[0]; 1053 } 1054 1055 nocow = 0; 1056 disk_bytenr = 0; 1057 num_bytes = 0; 1058 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1059 1060 if (found_key.objectid > inode->i_ino || 1061 found_key.type > BTRFS_EXTENT_DATA_KEY || 1062 found_key.offset > end) 1063 break; 1064 1065 if (found_key.offset > cur_offset) { 1066 extent_end = found_key.offset; 1067 extent_type = 0; 1068 goto out_check; 1069 } 1070 1071 fi = btrfs_item_ptr(leaf, path->slots[0], 1072 struct btrfs_file_extent_item); 1073 extent_type = btrfs_file_extent_type(leaf, fi); 1074 1075 if (extent_type == BTRFS_FILE_EXTENT_REG || 1076 extent_type == BTRFS_FILE_EXTENT_PREALLOC) { 1077 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 1078 extent_offset = btrfs_file_extent_offset(leaf, fi); 1079 extent_end = found_key.offset + 1080 btrfs_file_extent_num_bytes(leaf, fi); 1081 if (extent_end <= start) { 1082 path->slots[0]++; 1083 goto next_slot; 1084 } 1085 if (disk_bytenr == 0) 1086 goto out_check; 1087 if (btrfs_file_extent_compression(leaf, fi) || 1088 btrfs_file_extent_encryption(leaf, fi) || 1089 btrfs_file_extent_other_encoding(leaf, fi)) 1090 goto out_check; 1091 if (extent_type == BTRFS_FILE_EXTENT_REG && !force) 1092 goto out_check; 1093 if (btrfs_extent_readonly(root, disk_bytenr)) 1094 goto out_check; 1095 if (btrfs_cross_ref_exist(trans, root, inode->i_ino, 1096 found_key.offset - 1097 extent_offset, disk_bytenr)) 1098 goto out_check; 1099 disk_bytenr += extent_offset; 1100 disk_bytenr += cur_offset - found_key.offset; 1101 num_bytes = min(end + 1, extent_end) - cur_offset; 1102 /* 1103 * force cow if csum exists in the range. 1104 * this ensure that csum for a given extent are 1105 * either valid or do not exist. 1106 */ 1107 if (csum_exist_in_range(root, disk_bytenr, num_bytes)) 1108 goto out_check; 1109 nocow = 1; 1110 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { 1111 extent_end = found_key.offset + 1112 btrfs_file_extent_inline_len(leaf, fi); 1113 extent_end = ALIGN(extent_end, root->sectorsize); 1114 } else { 1115 BUG_ON(1); 1116 } 1117 out_check: 1118 if (extent_end <= start) { 1119 path->slots[0]++; 1120 goto next_slot; 1121 } 1122 if (!nocow) { 1123 if (cow_start == (u64)-1) 1124 cow_start = cur_offset; 1125 cur_offset = extent_end; 1126 if (cur_offset > end) 1127 break; 1128 path->slots[0]++; 1129 goto next_slot; 1130 } 1131 1132 btrfs_release_path(root, path); 1133 if (cow_start != (u64)-1) { 1134 ret = cow_file_range(inode, locked_page, cow_start, 1135 found_key.offset - 1, page_started, 1136 nr_written, 1); 1137 BUG_ON(ret); 1138 cow_start = (u64)-1; 1139 } 1140 1141 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { 1142 struct extent_map *em; 1143 struct extent_map_tree *em_tree; 1144 em_tree = &BTRFS_I(inode)->extent_tree; 1145 em = alloc_extent_map(GFP_NOFS); 1146 em->start = cur_offset; 1147 em->orig_start = em->start; 1148 em->len = num_bytes; 1149 em->block_len = num_bytes; 1150 em->block_start = disk_bytenr; 1151 em->bdev = root->fs_info->fs_devices->latest_bdev; 1152 set_bit(EXTENT_FLAG_PINNED, &em->flags); 1153 while (1) { 1154 write_lock(&em_tree->lock); 1155 ret = add_extent_mapping(em_tree, em); 1156 write_unlock(&em_tree->lock); 1157 if (ret != -EEXIST) { 1158 free_extent_map(em); 1159 break; 1160 } 1161 btrfs_drop_extent_cache(inode, em->start, 1162 em->start + em->len - 1, 0); 1163 } 1164 type = BTRFS_ORDERED_PREALLOC; 1165 } else { 1166 type = BTRFS_ORDERED_NOCOW; 1167 } 1168 1169 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, 1170 num_bytes, num_bytes, type); 1171 BUG_ON(ret); 1172 1173 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, 1174 cur_offset, cur_offset + num_bytes - 1, 1175 locked_page, EXTENT_CLEAR_UNLOCK_PAGE | 1176 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | 1177 EXTENT_SET_PRIVATE2); 1178 cur_offset = extent_end; 1179 if (cur_offset > end) 1180 break; 1181 } 1182 btrfs_release_path(root, path); 1183 1184 if (cur_offset <= end && cow_start == (u64)-1) 1185 cow_start = cur_offset; 1186 if (cow_start != (u64)-1) { 1187 ret = cow_file_range(inode, locked_page, cow_start, end, 1188 page_started, nr_written, 1); 1189 BUG_ON(ret); 1190 } 1191 1192 ret = btrfs_end_transaction(trans, root); 1193 BUG_ON(ret); 1194 btrfs_free_path(path); 1195 return 0; 1196 } 1197 1198 /* 1199 * extent_io.c call back to do delayed allocation processing 1200 */ 1201 static int run_delalloc_range(struct inode *inode, struct page *locked_page, 1202 u64 start, u64 end, int *page_started, 1203 unsigned long *nr_written) 1204 { 1205 int ret; 1206 struct btrfs_root *root = BTRFS_I(inode)->root; 1207 1208 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) 1209 ret = run_delalloc_nocow(inode, locked_page, start, end, 1210 page_started, 1, nr_written); 1211 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) 1212 ret = run_delalloc_nocow(inode, locked_page, start, end, 1213 page_started, 0, nr_written); 1214 else if (!btrfs_test_opt(root, COMPRESS)) 1215 ret = cow_file_range(inode, locked_page, start, end, 1216 page_started, nr_written, 1); 1217 else 1218 ret = cow_file_range_async(inode, locked_page, start, end, 1219 page_started, nr_written); 1220 return ret; 1221 } 1222 1223 static int btrfs_split_extent_hook(struct inode *inode, 1224 struct extent_state *orig, u64 split) 1225 { 1226 struct btrfs_root *root = BTRFS_I(inode)->root; 1227 u64 size; 1228 1229 if (!(orig->state & EXTENT_DELALLOC)) 1230 return 0; 1231 1232 size = orig->end - orig->start + 1; 1233 if (size > root->fs_info->max_extent) { 1234 u64 num_extents; 1235 u64 new_size; 1236 1237 new_size = orig->end - split + 1; 1238 num_extents = div64_u64(size + root->fs_info->max_extent - 1, 1239 root->fs_info->max_extent); 1240 1241 /* 1242 * if we break a large extent up then leave oustanding_extents 1243 * be, since we've already accounted for the large extent. 1244 */ 1245 if (div64_u64(new_size + root->fs_info->max_extent - 1, 1246 root->fs_info->max_extent) < num_extents) 1247 return 0; 1248 } 1249 1250 spin_lock(&BTRFS_I(inode)->accounting_lock); 1251 BTRFS_I(inode)->outstanding_extents++; 1252 spin_unlock(&BTRFS_I(inode)->accounting_lock); 1253 1254 return 0; 1255 } 1256 1257 /* 1258 * extent_io.c merge_extent_hook, used to track merged delayed allocation 1259 * extents so we can keep track of new extents that are just merged onto old 1260 * extents, such as when we are doing sequential writes, so we can properly 1261 * account for the metadata space we'll need. 1262 */ 1263 static int btrfs_merge_extent_hook(struct inode *inode, 1264 struct extent_state *new, 1265 struct extent_state *other) 1266 { 1267 struct btrfs_root *root = BTRFS_I(inode)->root; 1268 u64 new_size, old_size; 1269 u64 num_extents; 1270 1271 /* not delalloc, ignore it */ 1272 if (!(other->state & EXTENT_DELALLOC)) 1273 return 0; 1274 1275 old_size = other->end - other->start + 1; 1276 if (new->start < other->start) 1277 new_size = other->end - new->start + 1; 1278 else 1279 new_size = new->end - other->start + 1; 1280 1281 /* we're not bigger than the max, unreserve the space and go */ 1282 if (new_size <= root->fs_info->max_extent) { 1283 spin_lock(&BTRFS_I(inode)->accounting_lock); 1284 BTRFS_I(inode)->outstanding_extents--; 1285 spin_unlock(&BTRFS_I(inode)->accounting_lock); 1286 return 0; 1287 } 1288 1289 /* 1290 * If we grew by another max_extent, just return, we want to keep that 1291 * reserved amount. 1292 */ 1293 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, 1294 root->fs_info->max_extent); 1295 if (div64_u64(new_size + root->fs_info->max_extent - 1, 1296 root->fs_info->max_extent) > num_extents) 1297 return 0; 1298 1299 spin_lock(&BTRFS_I(inode)->accounting_lock); 1300 BTRFS_I(inode)->outstanding_extents--; 1301 spin_unlock(&BTRFS_I(inode)->accounting_lock); 1302 1303 return 0; 1304 } 1305 1306 /* 1307 * extent_io.c set_bit_hook, used to track delayed allocation 1308 * bytes in this file, and to maintain the list of inodes that 1309 * have pending delalloc work to be done. 1310 */ 1311 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, 1312 unsigned long old, unsigned long bits) 1313 { 1314 1315 /* 1316 * set_bit and clear bit hooks normally require _irqsave/restore 1317 * but in this case, we are only testeing for the DELALLOC 1318 * bit, which is only set or cleared with irqs on 1319 */ 1320 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1321 struct btrfs_root *root = BTRFS_I(inode)->root; 1322 1323 spin_lock(&BTRFS_I(inode)->accounting_lock); 1324 BTRFS_I(inode)->outstanding_extents++; 1325 spin_unlock(&BTRFS_I(inode)->accounting_lock); 1326 btrfs_delalloc_reserve_space(root, inode, end - start + 1); 1327 spin_lock(&root->fs_info->delalloc_lock); 1328 BTRFS_I(inode)->delalloc_bytes += end - start + 1; 1329 root->fs_info->delalloc_bytes += end - start + 1; 1330 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { 1331 list_add_tail(&BTRFS_I(inode)->delalloc_inodes, 1332 &root->fs_info->delalloc_inodes); 1333 } 1334 spin_unlock(&root->fs_info->delalloc_lock); 1335 } 1336 return 0; 1337 } 1338 1339 /* 1340 * extent_io.c clear_bit_hook, see set_bit_hook for why 1341 */ 1342 static int btrfs_clear_bit_hook(struct inode *inode, 1343 struct extent_state *state, unsigned long bits) 1344 { 1345 /* 1346 * set_bit and clear bit hooks normally require _irqsave/restore 1347 * but in this case, we are only testeing for the DELALLOC 1348 * bit, which is only set or cleared with irqs on 1349 */ 1350 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { 1351 struct btrfs_root *root = BTRFS_I(inode)->root; 1352 1353 if (bits & EXTENT_DO_ACCOUNTING) { 1354 spin_lock(&BTRFS_I(inode)->accounting_lock); 1355 BTRFS_I(inode)->outstanding_extents--; 1356 spin_unlock(&BTRFS_I(inode)->accounting_lock); 1357 btrfs_unreserve_metadata_for_delalloc(root, inode, 1); 1358 } 1359 1360 spin_lock(&root->fs_info->delalloc_lock); 1361 if (state->end - state->start + 1 > 1362 root->fs_info->delalloc_bytes) { 1363 printk(KERN_INFO "btrfs warning: delalloc account " 1364 "%llu %llu\n", 1365 (unsigned long long) 1366 state->end - state->start + 1, 1367 (unsigned long long) 1368 root->fs_info->delalloc_bytes); 1369 btrfs_delalloc_free_space(root, inode, (u64)-1); 1370 root->fs_info->delalloc_bytes = 0; 1371 BTRFS_I(inode)->delalloc_bytes = 0; 1372 } else { 1373 btrfs_delalloc_free_space(root, inode, 1374 state->end - 1375 state->start + 1); 1376 root->fs_info->delalloc_bytes -= state->end - 1377 state->start + 1; 1378 BTRFS_I(inode)->delalloc_bytes -= state->end - 1379 state->start + 1; 1380 } 1381 if (BTRFS_I(inode)->delalloc_bytes == 0 && 1382 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { 1383 list_del_init(&BTRFS_I(inode)->delalloc_inodes); 1384 } 1385 spin_unlock(&root->fs_info->delalloc_lock); 1386 } 1387 return 0; 1388 } 1389 1390 /* 1391 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure 1392 * we don't create bios that span stripes or chunks 1393 */ 1394 int btrfs_merge_bio_hook(struct page *page, unsigned long offset, 1395 size_t size, struct bio *bio, 1396 unsigned long bio_flags) 1397 { 1398 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 1399 struct btrfs_mapping_tree *map_tree; 1400 u64 logical = (u64)bio->bi_sector << 9; 1401 u64 length = 0; 1402 u64 map_length; 1403 int ret; 1404 1405 if (bio_flags & EXTENT_BIO_COMPRESSED) 1406 return 0; 1407 1408 length = bio->bi_size; 1409 map_tree = &root->fs_info->mapping_tree; 1410 map_length = length; 1411 ret = btrfs_map_block(map_tree, READ, logical, 1412 &map_length, NULL, 0); 1413 1414 if (map_length < length + size) 1415 return 1; 1416 return 0; 1417 } 1418 1419 /* 1420 * in order to insert checksums into the metadata in large chunks, 1421 * we wait until bio submission time. All the pages in the bio are 1422 * checksummed and sums are attached onto the ordered extent record. 1423 * 1424 * At IO completion time the cums attached on the ordered extent record 1425 * are inserted into the btree 1426 */ 1427 static int __btrfs_submit_bio_start(struct inode *inode, int rw, 1428 struct bio *bio, int mirror_num, 1429 unsigned long bio_flags) 1430 { 1431 struct btrfs_root *root = BTRFS_I(inode)->root; 1432 int ret = 0; 1433 1434 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); 1435 BUG_ON(ret); 1436 return 0; 1437 } 1438 1439 /* 1440 * in order to insert checksums into the metadata in large chunks, 1441 * we wait until bio submission time. All the pages in the bio are 1442 * checksummed and sums are attached onto the ordered extent record. 1443 * 1444 * At IO completion time the cums attached on the ordered extent record 1445 * are inserted into the btree 1446 */ 1447 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 1448 int mirror_num, unsigned long bio_flags) 1449 { 1450 struct btrfs_root *root = BTRFS_I(inode)->root; 1451 return btrfs_map_bio(root, rw, bio, mirror_num, 1); 1452 } 1453 1454 /* 1455 * extent_io.c submission hook. This does the right thing for csum calculation 1456 * on write, or reading the csums from the tree before a read 1457 */ 1458 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 1459 int mirror_num, unsigned long bio_flags) 1460 { 1461 struct btrfs_root *root = BTRFS_I(inode)->root; 1462 int ret = 0; 1463 int skip_sum; 1464 1465 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; 1466 1467 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); 1468 BUG_ON(ret); 1469 1470 if (!(rw & (1 << BIO_RW))) { 1471 if (bio_flags & EXTENT_BIO_COMPRESSED) { 1472 return btrfs_submit_compressed_read(inode, bio, 1473 mirror_num, bio_flags); 1474 } else if (!skip_sum) 1475 btrfs_lookup_bio_sums(root, inode, bio, NULL); 1476 goto mapit; 1477 } else if (!skip_sum) { 1478 /* csum items have already been cloned */ 1479 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) 1480 goto mapit; 1481 /* we're doing a write, do the async checksumming */ 1482 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 1483 inode, rw, bio, mirror_num, 1484 bio_flags, __btrfs_submit_bio_start, 1485 __btrfs_submit_bio_done); 1486 } 1487 1488 mapit: 1489 return btrfs_map_bio(root, rw, bio, mirror_num, 0); 1490 } 1491 1492 /* 1493 * given a list of ordered sums record them in the inode. This happens 1494 * at IO completion time based on sums calculated at bio submission time. 1495 */ 1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans, 1497 struct inode *inode, u64 file_offset, 1498 struct list_head *list) 1499 { 1500 struct btrfs_ordered_sum *sum; 1501 1502 btrfs_set_trans_block_group(trans, inode); 1503 1504 list_for_each_entry(sum, list, list) { 1505 btrfs_csum_file_blocks(trans, 1506 BTRFS_I(inode)->root->fs_info->csum_root, sum); 1507 } 1508 return 0; 1509 } 1510 1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) 1512 { 1513 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) 1514 WARN_ON(1); 1515 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, 1516 GFP_NOFS); 1517 } 1518 1519 /* see btrfs_writepage_start_hook for details on why this is required */ 1520 struct btrfs_writepage_fixup { 1521 struct page *page; 1522 struct btrfs_work work; 1523 }; 1524 1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work) 1526 { 1527 struct btrfs_writepage_fixup *fixup; 1528 struct btrfs_ordered_extent *ordered; 1529 struct page *page; 1530 struct inode *inode; 1531 u64 page_start; 1532 u64 page_end; 1533 1534 fixup = container_of(work, struct btrfs_writepage_fixup, work); 1535 page = fixup->page; 1536 again: 1537 lock_page(page); 1538 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { 1539 ClearPageChecked(page); 1540 goto out_page; 1541 } 1542 1543 inode = page->mapping->host; 1544 page_start = page_offset(page); 1545 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; 1546 1547 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); 1548 1549 /* already ordered? We're done */ 1550 if (PagePrivate2(page)) 1551 goto out; 1552 1553 ordered = btrfs_lookup_ordered_extent(inode, page_start); 1554 if (ordered) { 1555 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, 1556 page_end, GFP_NOFS); 1557 unlock_page(page); 1558 btrfs_start_ordered_extent(inode, ordered, 1); 1559 goto again; 1560 } 1561 1562 btrfs_set_extent_delalloc(inode, page_start, page_end); 1563 ClearPageChecked(page); 1564 out: 1565 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); 1566 out_page: 1567 unlock_page(page); 1568 page_cache_release(page); 1569 } 1570 1571 /* 1572 * There are a few paths in the higher layers of the kernel that directly 1573 * set the page dirty bit without asking the filesystem if it is a 1574 * good idea. This causes problems because we want to make sure COW 1575 * properly happens and the data=ordered rules are followed. 1576 * 1577 * In our case any range that doesn't have the ORDERED bit set 1578 * hasn't been properly setup for IO. We kick off an async process 1579 * to fix it up. The async helper will wait for ordered extents, set 1580 * the delalloc bit and make it safe to write the page. 1581 */ 1582 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) 1583 { 1584 struct inode *inode = page->mapping->host; 1585 struct btrfs_writepage_fixup *fixup; 1586 struct btrfs_root *root = BTRFS_I(inode)->root; 1587 1588 /* this page is properly in the ordered list */ 1589 if (TestClearPagePrivate2(page)) 1590 return 0; 1591 1592 if (PageChecked(page)) 1593 return -EAGAIN; 1594 1595 fixup = kzalloc(sizeof(*fixup), GFP_NOFS); 1596 if (!fixup) 1597 return -EAGAIN; 1598 1599 SetPageChecked(page); 1600 page_cache_get(page); 1601 fixup->work.func = btrfs_writepage_fixup_worker; 1602 fixup->page = page; 1603 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); 1604 return -EAGAIN; 1605 } 1606 1607 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, 1608 struct inode *inode, u64 file_pos, 1609 u64 disk_bytenr, u64 disk_num_bytes, 1610 u64 num_bytes, u64 ram_bytes, 1611 u8 compression, u8 encryption, 1612 u16 other_encoding, int extent_type) 1613 { 1614 struct btrfs_root *root = BTRFS_I(inode)->root; 1615 struct btrfs_file_extent_item *fi; 1616 struct btrfs_path *path; 1617 struct extent_buffer *leaf; 1618 struct btrfs_key ins; 1619 u64 hint; 1620 int ret; 1621 1622 path = btrfs_alloc_path(); 1623 BUG_ON(!path); 1624 1625 path->leave_spinning = 1; 1626 1627 /* 1628 * we may be replacing one extent in the tree with another. 1629 * The new extent is pinned in the extent map, and we don't want 1630 * to drop it from the cache until it is completely in the btree. 1631 * 1632 * So, tell btrfs_drop_extents to leave this extent in the cache. 1633 * the caller is expected to unpin it and allow it to be merged 1634 * with the others. 1635 */ 1636 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, 1637 &hint, 0); 1638 BUG_ON(ret); 1639 1640 ins.objectid = inode->i_ino; 1641 ins.offset = file_pos; 1642 ins.type = BTRFS_EXTENT_DATA_KEY; 1643 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); 1644 BUG_ON(ret); 1645 leaf = path->nodes[0]; 1646 fi = btrfs_item_ptr(leaf, path->slots[0], 1647 struct btrfs_file_extent_item); 1648 btrfs_set_file_extent_generation(leaf, fi, trans->transid); 1649 btrfs_set_file_extent_type(leaf, fi, extent_type); 1650 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); 1651 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); 1652 btrfs_set_file_extent_offset(leaf, fi, 0); 1653 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); 1654 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); 1655 btrfs_set_file_extent_compression(leaf, fi, compression); 1656 btrfs_set_file_extent_encryption(leaf, fi, encryption); 1657 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); 1658 1659 btrfs_unlock_up_safe(path, 1); 1660 btrfs_set_lock_blocking(leaf); 1661 1662 btrfs_mark_buffer_dirty(leaf); 1663 1664 inode_add_bytes(inode, num_bytes); 1665 1666 ins.objectid = disk_bytenr; 1667 ins.offset = disk_num_bytes; 1668 ins.type = BTRFS_EXTENT_ITEM_KEY; 1669 ret = btrfs_alloc_reserved_file_extent(trans, root, 1670 root->root_key.objectid, 1671 inode->i_ino, file_pos, &ins); 1672 BUG_ON(ret); 1673 btrfs_free_path(path); 1674 1675 return 0; 1676 } 1677 1678 /* 1679 * helper function for btrfs_finish_ordered_io, this 1680 * just reads in some of the csum leaves to prime them into ram 1681 * before we start the transaction. It limits the amount of btree 1682 * reads required while inside the transaction. 1683 */ 1684 /* as ordered data IO finishes, this gets called so we can finish 1685 * an ordered extent if the range of bytes in the file it covers are 1686 * fully written. 1687 */ 1688 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) 1689 { 1690 struct btrfs_root *root = BTRFS_I(inode)->root; 1691 struct btrfs_trans_handle *trans; 1692 struct btrfs_ordered_extent *ordered_extent = NULL; 1693 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 1694 int compressed = 0; 1695 int ret; 1696 1697 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); 1698 if (!ret) 1699 return 0; 1700 1701 ordered_extent = btrfs_lookup_ordered_extent(inode, start); 1702 BUG_ON(!ordered_extent); 1703 1704 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { 1705 BUG_ON(!list_empty(&ordered_extent->list)); 1706 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); 1707 if (!ret) { 1708 trans = btrfs_join_transaction(root, 1); 1709 ret = btrfs_update_inode(trans, root, inode); 1710 BUG_ON(ret); 1711 btrfs_end_transaction(trans, root); 1712 } 1713 goto out; 1714 } 1715 1716 lock_extent(io_tree, ordered_extent->file_offset, 1717 ordered_extent->file_offset + ordered_extent->len - 1, 1718 GFP_NOFS); 1719 1720 trans = btrfs_join_transaction(root, 1); 1721 1722 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) 1723 compressed = 1; 1724 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { 1725 BUG_ON(compressed); 1726 ret = btrfs_mark_extent_written(trans, inode, 1727 ordered_extent->file_offset, 1728 ordered_extent->file_offset + 1729 ordered_extent->len); 1730 BUG_ON(ret); 1731 } else { 1732 ret = insert_reserved_file_extent(trans, inode, 1733 ordered_extent->file_offset, 1734 ordered_extent->start, 1735 ordered_extent->disk_len, 1736 ordered_extent->len, 1737 ordered_extent->len, 1738 compressed, 0, 0, 1739 BTRFS_FILE_EXTENT_REG); 1740 unpin_extent_cache(&BTRFS_I(inode)->extent_tree, 1741 ordered_extent->file_offset, 1742 ordered_extent->len); 1743 BUG_ON(ret); 1744 } 1745 unlock_extent(io_tree, ordered_extent->file_offset, 1746 ordered_extent->file_offset + ordered_extent->len - 1, 1747 GFP_NOFS); 1748 add_pending_csums(trans, inode, ordered_extent->file_offset, 1749 &ordered_extent->list); 1750 1751 /* this also removes the ordered extent from the tree */ 1752 btrfs_ordered_update_i_size(inode, 0, ordered_extent); 1753 ret = btrfs_update_inode(trans, root, inode); 1754 BUG_ON(ret); 1755 btrfs_end_transaction(trans, root); 1756 out: 1757 /* once for us */ 1758 btrfs_put_ordered_extent(ordered_extent); 1759 /* once for the tree */ 1760 btrfs_put_ordered_extent(ordered_extent); 1761 1762 return 0; 1763 } 1764 1765 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, 1766 struct extent_state *state, int uptodate) 1767 { 1768 ClearPagePrivate2(page); 1769 return btrfs_finish_ordered_io(page->mapping->host, start, end); 1770 } 1771 1772 /* 1773 * When IO fails, either with EIO or csum verification fails, we 1774 * try other mirrors that might have a good copy of the data. This 1775 * io_failure_record is used to record state as we go through all the 1776 * mirrors. If another mirror has good data, the page is set up to date 1777 * and things continue. If a good mirror can't be found, the original 1778 * bio end_io callback is called to indicate things have failed. 1779 */ 1780 struct io_failure_record { 1781 struct page *page; 1782 u64 start; 1783 u64 len; 1784 u64 logical; 1785 unsigned long bio_flags; 1786 int last_mirror; 1787 }; 1788 1789 static int btrfs_io_failed_hook(struct bio *failed_bio, 1790 struct page *page, u64 start, u64 end, 1791 struct extent_state *state) 1792 { 1793 struct io_failure_record *failrec = NULL; 1794 u64 private; 1795 struct extent_map *em; 1796 struct inode *inode = page->mapping->host; 1797 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; 1798 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1799 struct bio *bio; 1800 int num_copies; 1801 int ret; 1802 int rw; 1803 u64 logical; 1804 1805 ret = get_state_private(failure_tree, start, &private); 1806 if (ret) { 1807 failrec = kmalloc(sizeof(*failrec), GFP_NOFS); 1808 if (!failrec) 1809 return -ENOMEM; 1810 failrec->start = start; 1811 failrec->len = end - start + 1; 1812 failrec->last_mirror = 0; 1813 failrec->bio_flags = 0; 1814 1815 read_lock(&em_tree->lock); 1816 em = lookup_extent_mapping(em_tree, start, failrec->len); 1817 if (em->start > start || em->start + em->len < start) { 1818 free_extent_map(em); 1819 em = NULL; 1820 } 1821 read_unlock(&em_tree->lock); 1822 1823 if (!em || IS_ERR(em)) { 1824 kfree(failrec); 1825 return -EIO; 1826 } 1827 logical = start - em->start; 1828 logical = em->block_start + logical; 1829 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { 1830 logical = em->block_start; 1831 failrec->bio_flags = EXTENT_BIO_COMPRESSED; 1832 } 1833 failrec->logical = logical; 1834 free_extent_map(em); 1835 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | 1836 EXTENT_DIRTY, GFP_NOFS); 1837 set_state_private(failure_tree, start, 1838 (u64)(unsigned long)failrec); 1839 } else { 1840 failrec = (struct io_failure_record *)(unsigned long)private; 1841 } 1842 num_copies = btrfs_num_copies( 1843 &BTRFS_I(inode)->root->fs_info->mapping_tree, 1844 failrec->logical, failrec->len); 1845 failrec->last_mirror++; 1846 if (!state) { 1847 spin_lock(&BTRFS_I(inode)->io_tree.lock); 1848 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, 1849 failrec->start, 1850 EXTENT_LOCKED); 1851 if (state && state->start != failrec->start) 1852 state = NULL; 1853 spin_unlock(&BTRFS_I(inode)->io_tree.lock); 1854 } 1855 if (!state || failrec->last_mirror > num_copies) { 1856 set_state_private(failure_tree, failrec->start, 0); 1857 clear_extent_bits(failure_tree, failrec->start, 1858 failrec->start + failrec->len - 1, 1859 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); 1860 kfree(failrec); 1861 return -EIO; 1862 } 1863 bio = bio_alloc(GFP_NOFS, 1); 1864 bio->bi_private = state; 1865 bio->bi_end_io = failed_bio->bi_end_io; 1866 bio->bi_sector = failrec->logical >> 9; 1867 bio->bi_bdev = failed_bio->bi_bdev; 1868 bio->bi_size = 0; 1869 1870 bio_add_page(bio, page, failrec->len, start - page_offset(page)); 1871 if (failed_bio->bi_rw & (1 << BIO_RW)) 1872 rw = WRITE; 1873 else 1874 rw = READ; 1875 1876 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, 1877 failrec->last_mirror, 1878 failrec->bio_flags); 1879 return 0; 1880 } 1881 1882 /* 1883 * each time an IO finishes, we do a fast check in the IO failure tree 1884 * to see if we need to process or clean up an io_failure_record 1885 */ 1886 static int btrfs_clean_io_failures(struct inode *inode, u64 start) 1887 { 1888 u64 private; 1889 u64 private_failure; 1890 struct io_failure_record *failure; 1891 int ret; 1892 1893 private = 0; 1894 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, 1895 (u64)-1, 1, EXTENT_DIRTY)) { 1896 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, 1897 start, &private_failure); 1898 if (ret == 0) { 1899 failure = (struct io_failure_record *)(unsigned long) 1900 private_failure; 1901 set_state_private(&BTRFS_I(inode)->io_failure_tree, 1902 failure->start, 0); 1903 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, 1904 failure->start, 1905 failure->start + failure->len - 1, 1906 EXTENT_DIRTY | EXTENT_LOCKED, 1907 GFP_NOFS); 1908 kfree(failure); 1909 } 1910 } 1911 return 0; 1912 } 1913 1914 /* 1915 * when reads are done, we need to check csums to verify the data is correct 1916 * if there's a match, we allow the bio to finish. If not, we go through 1917 * the io_failure_record routines to find good copies 1918 */ 1919 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, 1920 struct extent_state *state) 1921 { 1922 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); 1923 struct inode *inode = page->mapping->host; 1924 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 1925 char *kaddr; 1926 u64 private = ~(u32)0; 1927 int ret; 1928 struct btrfs_root *root = BTRFS_I(inode)->root; 1929 u32 csum = ~(u32)0; 1930 1931 if (PageChecked(page)) { 1932 ClearPageChecked(page); 1933 goto good; 1934 } 1935 1936 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) 1937 return 0; 1938 1939 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && 1940 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { 1941 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, 1942 GFP_NOFS); 1943 return 0; 1944 } 1945 1946 if (state && state->start == start) { 1947 private = state->private; 1948 ret = 0; 1949 } else { 1950 ret = get_state_private(io_tree, start, &private); 1951 } 1952 kaddr = kmap_atomic(page, KM_USER0); 1953 if (ret) 1954 goto zeroit; 1955 1956 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); 1957 btrfs_csum_final(csum, (char *)&csum); 1958 if (csum != private) 1959 goto zeroit; 1960 1961 kunmap_atomic(kaddr, KM_USER0); 1962 good: 1963 /* if the io failure tree for this inode is non-empty, 1964 * check to see if we've recovered from a failed IO 1965 */ 1966 btrfs_clean_io_failures(inode, start); 1967 return 0; 1968 1969 zeroit: 1970 if (printk_ratelimit()) { 1971 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " 1972 "private %llu\n", page->mapping->host->i_ino, 1973 (unsigned long long)start, csum, 1974 (unsigned long long)private); 1975 } 1976 memset(kaddr + offset, 1, end - start + 1); 1977 flush_dcache_page(page); 1978 kunmap_atomic(kaddr, KM_USER0); 1979 if (private == 0) 1980 return 0; 1981 return -EIO; 1982 } 1983 1984 struct delayed_iput { 1985 struct list_head list; 1986 struct inode *inode; 1987 }; 1988 1989 void btrfs_add_delayed_iput(struct inode *inode) 1990 { 1991 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 1992 struct delayed_iput *delayed; 1993 1994 if (atomic_add_unless(&inode->i_count, -1, 1)) 1995 return; 1996 1997 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); 1998 delayed->inode = inode; 1999 2000 spin_lock(&fs_info->delayed_iput_lock); 2001 list_add_tail(&delayed->list, &fs_info->delayed_iputs); 2002 spin_unlock(&fs_info->delayed_iput_lock); 2003 } 2004 2005 void btrfs_run_delayed_iputs(struct btrfs_root *root) 2006 { 2007 LIST_HEAD(list); 2008 struct btrfs_fs_info *fs_info = root->fs_info; 2009 struct delayed_iput *delayed; 2010 int empty; 2011 2012 spin_lock(&fs_info->delayed_iput_lock); 2013 empty = list_empty(&fs_info->delayed_iputs); 2014 spin_unlock(&fs_info->delayed_iput_lock); 2015 if (empty) 2016 return; 2017 2018 down_read(&root->fs_info->cleanup_work_sem); 2019 spin_lock(&fs_info->delayed_iput_lock); 2020 list_splice_init(&fs_info->delayed_iputs, &list); 2021 spin_unlock(&fs_info->delayed_iput_lock); 2022 2023 while (!list_empty(&list)) { 2024 delayed = list_entry(list.next, struct delayed_iput, list); 2025 list_del(&delayed->list); 2026 iput(delayed->inode); 2027 kfree(delayed); 2028 } 2029 up_read(&root->fs_info->cleanup_work_sem); 2030 } 2031 2032 /* 2033 * This creates an orphan entry for the given inode in case something goes 2034 * wrong in the middle of an unlink/truncate. 2035 */ 2036 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) 2037 { 2038 struct btrfs_root *root = BTRFS_I(inode)->root; 2039 int ret = 0; 2040 2041 spin_lock(&root->list_lock); 2042 2043 /* already on the orphan list, we're good */ 2044 if (!list_empty(&BTRFS_I(inode)->i_orphan)) { 2045 spin_unlock(&root->list_lock); 2046 return 0; 2047 } 2048 2049 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); 2050 2051 spin_unlock(&root->list_lock); 2052 2053 /* 2054 * insert an orphan item to track this unlinked/truncated file 2055 */ 2056 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); 2057 2058 return ret; 2059 } 2060 2061 /* 2062 * We have done the truncate/delete so we can go ahead and remove the orphan 2063 * item for this particular inode. 2064 */ 2065 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) 2066 { 2067 struct btrfs_root *root = BTRFS_I(inode)->root; 2068 int ret = 0; 2069 2070 spin_lock(&root->list_lock); 2071 2072 if (list_empty(&BTRFS_I(inode)->i_orphan)) { 2073 spin_unlock(&root->list_lock); 2074 return 0; 2075 } 2076 2077 list_del_init(&BTRFS_I(inode)->i_orphan); 2078 if (!trans) { 2079 spin_unlock(&root->list_lock); 2080 return 0; 2081 } 2082 2083 spin_unlock(&root->list_lock); 2084 2085 ret = btrfs_del_orphan_item(trans, root, inode->i_ino); 2086 2087 return ret; 2088 } 2089 2090 /* 2091 * this cleans up any orphans that may be left on the list from the last use 2092 * of this root. 2093 */ 2094 void btrfs_orphan_cleanup(struct btrfs_root *root) 2095 { 2096 struct btrfs_path *path; 2097 struct extent_buffer *leaf; 2098 struct btrfs_item *item; 2099 struct btrfs_key key, found_key; 2100 struct btrfs_trans_handle *trans; 2101 struct inode *inode; 2102 int ret = 0, nr_unlink = 0, nr_truncate = 0; 2103 2104 if (!xchg(&root->clean_orphans, 0)) 2105 return; 2106 2107 path = btrfs_alloc_path(); 2108 BUG_ON(!path); 2109 path->reada = -1; 2110 2111 key.objectid = BTRFS_ORPHAN_OBJECTID; 2112 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); 2113 key.offset = (u64)-1; 2114 2115 while (1) { 2116 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2117 if (ret < 0) { 2118 printk(KERN_ERR "Error searching slot for orphan: %d" 2119 "\n", ret); 2120 break; 2121 } 2122 2123 /* 2124 * if ret == 0 means we found what we were searching for, which 2125 * is weird, but possible, so only screw with path if we didnt 2126 * find the key and see if we have stuff that matches 2127 */ 2128 if (ret > 0) { 2129 if (path->slots[0] == 0) 2130 break; 2131 path->slots[0]--; 2132 } 2133 2134 /* pull out the item */ 2135 leaf = path->nodes[0]; 2136 item = btrfs_item_nr(leaf, path->slots[0]); 2137 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 2138 2139 /* make sure the item matches what we want */ 2140 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) 2141 break; 2142 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) 2143 break; 2144 2145 /* release the path since we're done with it */ 2146 btrfs_release_path(root, path); 2147 2148 /* 2149 * this is where we are basically btrfs_lookup, without the 2150 * crossing root thing. we store the inode number in the 2151 * offset of the orphan item. 2152 */ 2153 found_key.objectid = found_key.offset; 2154 found_key.type = BTRFS_INODE_ITEM_KEY; 2155 found_key.offset = 0; 2156 inode = btrfs_iget(root->fs_info->sb, &found_key, root); 2157 if (IS_ERR(inode)) 2158 break; 2159 2160 /* 2161 * add this inode to the orphan list so btrfs_orphan_del does 2162 * the proper thing when we hit it 2163 */ 2164 spin_lock(&root->list_lock); 2165 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); 2166 spin_unlock(&root->list_lock); 2167 2168 /* 2169 * if this is a bad inode, means we actually succeeded in 2170 * removing the inode, but not the orphan record, which means 2171 * we need to manually delete the orphan since iput will just 2172 * do a destroy_inode 2173 */ 2174 if (is_bad_inode(inode)) { 2175 trans = btrfs_start_transaction(root, 1); 2176 btrfs_orphan_del(trans, inode); 2177 btrfs_end_transaction(trans, root); 2178 iput(inode); 2179 continue; 2180 } 2181 2182 /* if we have links, this was a truncate, lets do that */ 2183 if (inode->i_nlink) { 2184 nr_truncate++; 2185 btrfs_truncate(inode); 2186 } else { 2187 nr_unlink++; 2188 } 2189 2190 /* this will do delete_inode and everything for us */ 2191 iput(inode); 2192 } 2193 2194 if (nr_unlink) 2195 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); 2196 if (nr_truncate) 2197 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); 2198 2199 btrfs_free_path(path); 2200 } 2201 2202 /* 2203 * very simple check to peek ahead in the leaf looking for xattrs. If we 2204 * don't find any xattrs, we know there can't be any acls. 2205 * 2206 * slot is the slot the inode is in, objectid is the objectid of the inode 2207 */ 2208 static noinline int acls_after_inode_item(struct extent_buffer *leaf, 2209 int slot, u64 objectid) 2210 { 2211 u32 nritems = btrfs_header_nritems(leaf); 2212 struct btrfs_key found_key; 2213 int scanned = 0; 2214 2215 slot++; 2216 while (slot < nritems) { 2217 btrfs_item_key_to_cpu(leaf, &found_key, slot); 2218 2219 /* we found a different objectid, there must not be acls */ 2220 if (found_key.objectid != objectid) 2221 return 0; 2222 2223 /* we found an xattr, assume we've got an acl */ 2224 if (found_key.type == BTRFS_XATTR_ITEM_KEY) 2225 return 1; 2226 2227 /* 2228 * we found a key greater than an xattr key, there can't 2229 * be any acls later on 2230 */ 2231 if (found_key.type > BTRFS_XATTR_ITEM_KEY) 2232 return 0; 2233 2234 slot++; 2235 scanned++; 2236 2237 /* 2238 * it goes inode, inode backrefs, xattrs, extents, 2239 * so if there are a ton of hard links to an inode there can 2240 * be a lot of backrefs. Don't waste time searching too hard, 2241 * this is just an optimization 2242 */ 2243 if (scanned >= 8) 2244 break; 2245 } 2246 /* we hit the end of the leaf before we found an xattr or 2247 * something larger than an xattr. We have to assume the inode 2248 * has acls 2249 */ 2250 return 1; 2251 } 2252 2253 /* 2254 * read an inode from the btree into the in-memory inode 2255 */ 2256 static void btrfs_read_locked_inode(struct inode *inode) 2257 { 2258 struct btrfs_path *path; 2259 struct extent_buffer *leaf; 2260 struct btrfs_inode_item *inode_item; 2261 struct btrfs_timespec *tspec; 2262 struct btrfs_root *root = BTRFS_I(inode)->root; 2263 struct btrfs_key location; 2264 int maybe_acls; 2265 u64 alloc_group_block; 2266 u32 rdev; 2267 int ret; 2268 2269 path = btrfs_alloc_path(); 2270 BUG_ON(!path); 2271 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); 2272 2273 ret = btrfs_lookup_inode(NULL, root, path, &location, 0); 2274 if (ret) 2275 goto make_bad; 2276 2277 leaf = path->nodes[0]; 2278 inode_item = btrfs_item_ptr(leaf, path->slots[0], 2279 struct btrfs_inode_item); 2280 2281 inode->i_mode = btrfs_inode_mode(leaf, inode_item); 2282 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); 2283 inode->i_uid = btrfs_inode_uid(leaf, inode_item); 2284 inode->i_gid = btrfs_inode_gid(leaf, inode_item); 2285 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); 2286 2287 tspec = btrfs_inode_atime(inode_item); 2288 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); 2289 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); 2290 2291 tspec = btrfs_inode_mtime(inode_item); 2292 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); 2293 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); 2294 2295 tspec = btrfs_inode_ctime(inode_item); 2296 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); 2297 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); 2298 2299 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); 2300 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); 2301 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); 2302 inode->i_generation = BTRFS_I(inode)->generation; 2303 inode->i_rdev = 0; 2304 rdev = btrfs_inode_rdev(leaf, inode_item); 2305 2306 BTRFS_I(inode)->index_cnt = (u64)-1; 2307 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); 2308 2309 alloc_group_block = btrfs_inode_block_group(leaf, inode_item); 2310 2311 /* 2312 * try to precache a NULL acl entry for files that don't have 2313 * any xattrs or acls 2314 */ 2315 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); 2316 if (!maybe_acls) 2317 cache_no_acl(inode); 2318 2319 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, 2320 alloc_group_block, 0); 2321 btrfs_free_path(path); 2322 inode_item = NULL; 2323 2324 switch (inode->i_mode & S_IFMT) { 2325 case S_IFREG: 2326 inode->i_mapping->a_ops = &btrfs_aops; 2327 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 2328 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; 2329 inode->i_fop = &btrfs_file_operations; 2330 inode->i_op = &btrfs_file_inode_operations; 2331 break; 2332 case S_IFDIR: 2333 inode->i_fop = &btrfs_dir_file_operations; 2334 if (root == root->fs_info->tree_root) 2335 inode->i_op = &btrfs_dir_ro_inode_operations; 2336 else 2337 inode->i_op = &btrfs_dir_inode_operations; 2338 break; 2339 case S_IFLNK: 2340 inode->i_op = &btrfs_symlink_inode_operations; 2341 inode->i_mapping->a_ops = &btrfs_symlink_aops; 2342 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 2343 break; 2344 default: 2345 inode->i_op = &btrfs_special_inode_operations; 2346 init_special_inode(inode, inode->i_mode, rdev); 2347 break; 2348 } 2349 2350 btrfs_update_iflags(inode); 2351 return; 2352 2353 make_bad: 2354 btrfs_free_path(path); 2355 make_bad_inode(inode); 2356 } 2357 2358 /* 2359 * given a leaf and an inode, copy the inode fields into the leaf 2360 */ 2361 static void fill_inode_item(struct btrfs_trans_handle *trans, 2362 struct extent_buffer *leaf, 2363 struct btrfs_inode_item *item, 2364 struct inode *inode) 2365 { 2366 btrfs_set_inode_uid(leaf, item, inode->i_uid); 2367 btrfs_set_inode_gid(leaf, item, inode->i_gid); 2368 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); 2369 btrfs_set_inode_mode(leaf, item, inode->i_mode); 2370 btrfs_set_inode_nlink(leaf, item, inode->i_nlink); 2371 2372 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), 2373 inode->i_atime.tv_sec); 2374 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), 2375 inode->i_atime.tv_nsec); 2376 2377 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), 2378 inode->i_mtime.tv_sec); 2379 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), 2380 inode->i_mtime.tv_nsec); 2381 2382 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), 2383 inode->i_ctime.tv_sec); 2384 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), 2385 inode->i_ctime.tv_nsec); 2386 2387 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); 2388 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); 2389 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); 2390 btrfs_set_inode_transid(leaf, item, trans->transid); 2391 btrfs_set_inode_rdev(leaf, item, inode->i_rdev); 2392 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); 2393 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); 2394 } 2395 2396 /* 2397 * copy everything in the in-memory inode into the btree. 2398 */ 2399 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, 2400 struct btrfs_root *root, struct inode *inode) 2401 { 2402 struct btrfs_inode_item *inode_item; 2403 struct btrfs_path *path; 2404 struct extent_buffer *leaf; 2405 int ret; 2406 2407 path = btrfs_alloc_path(); 2408 BUG_ON(!path); 2409 path->leave_spinning = 1; 2410 ret = btrfs_lookup_inode(trans, root, path, 2411 &BTRFS_I(inode)->location, 1); 2412 if (ret) { 2413 if (ret > 0) 2414 ret = -ENOENT; 2415 goto failed; 2416 } 2417 2418 btrfs_unlock_up_safe(path, 1); 2419 leaf = path->nodes[0]; 2420 inode_item = btrfs_item_ptr(leaf, path->slots[0], 2421 struct btrfs_inode_item); 2422 2423 fill_inode_item(trans, leaf, inode_item, inode); 2424 btrfs_mark_buffer_dirty(leaf); 2425 btrfs_set_inode_last_trans(trans, inode); 2426 ret = 0; 2427 failed: 2428 btrfs_free_path(path); 2429 return ret; 2430 } 2431 2432 2433 /* 2434 * unlink helper that gets used here in inode.c and in the tree logging 2435 * recovery code. It remove a link in a directory with a given name, and 2436 * also drops the back refs in the inode to the directory 2437 */ 2438 int btrfs_unlink_inode(struct btrfs_trans_handle *trans, 2439 struct btrfs_root *root, 2440 struct inode *dir, struct inode *inode, 2441 const char *name, int name_len) 2442 { 2443 struct btrfs_path *path; 2444 int ret = 0; 2445 struct extent_buffer *leaf; 2446 struct btrfs_dir_item *di; 2447 struct btrfs_key key; 2448 u64 index; 2449 2450 path = btrfs_alloc_path(); 2451 if (!path) { 2452 ret = -ENOMEM; 2453 goto err; 2454 } 2455 2456 path->leave_spinning = 1; 2457 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, 2458 name, name_len, -1); 2459 if (IS_ERR(di)) { 2460 ret = PTR_ERR(di); 2461 goto err; 2462 } 2463 if (!di) { 2464 ret = -ENOENT; 2465 goto err; 2466 } 2467 leaf = path->nodes[0]; 2468 btrfs_dir_item_key_to_cpu(leaf, di, &key); 2469 ret = btrfs_delete_one_dir_name(trans, root, path, di); 2470 if (ret) 2471 goto err; 2472 btrfs_release_path(root, path); 2473 2474 ret = btrfs_del_inode_ref(trans, root, name, name_len, 2475 inode->i_ino, 2476 dir->i_ino, &index); 2477 if (ret) { 2478 printk(KERN_INFO "btrfs failed to delete reference to %.*s, " 2479 "inode %lu parent %lu\n", name_len, name, 2480 inode->i_ino, dir->i_ino); 2481 goto err; 2482 } 2483 2484 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, 2485 index, name, name_len, -1); 2486 if (IS_ERR(di)) { 2487 ret = PTR_ERR(di); 2488 goto err; 2489 } 2490 if (!di) { 2491 ret = -ENOENT; 2492 goto err; 2493 } 2494 ret = btrfs_delete_one_dir_name(trans, root, path, di); 2495 btrfs_release_path(root, path); 2496 2497 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, 2498 inode, dir->i_ino); 2499 BUG_ON(ret != 0 && ret != -ENOENT); 2500 2501 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, 2502 dir, index); 2503 BUG_ON(ret); 2504 err: 2505 btrfs_free_path(path); 2506 if (ret) 2507 goto out; 2508 2509 btrfs_i_size_write(dir, dir->i_size - name_len * 2); 2510 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; 2511 btrfs_update_inode(trans, root, dir); 2512 btrfs_drop_nlink(inode); 2513 ret = btrfs_update_inode(trans, root, inode); 2514 out: 2515 return ret; 2516 } 2517 2518 static int btrfs_unlink(struct inode *dir, struct dentry *dentry) 2519 { 2520 struct btrfs_root *root; 2521 struct btrfs_trans_handle *trans; 2522 struct inode *inode = dentry->d_inode; 2523 int ret; 2524 unsigned long nr = 0; 2525 2526 root = BTRFS_I(dir)->root; 2527 2528 /* 2529 * 5 items for unlink inode 2530 * 1 for orphan 2531 */ 2532 ret = btrfs_reserve_metadata_space(root, 6); 2533 if (ret) 2534 return ret; 2535 2536 trans = btrfs_start_transaction(root, 1); 2537 if (IS_ERR(trans)) { 2538 btrfs_unreserve_metadata_space(root, 6); 2539 return PTR_ERR(trans); 2540 } 2541 2542 btrfs_set_trans_block_group(trans, dir); 2543 2544 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); 2545 2546 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, 2547 dentry->d_name.name, dentry->d_name.len); 2548 2549 if (inode->i_nlink == 0) 2550 ret = btrfs_orphan_add(trans, inode); 2551 2552 nr = trans->blocks_used; 2553 2554 btrfs_end_transaction_throttle(trans, root); 2555 btrfs_unreserve_metadata_space(root, 6); 2556 btrfs_btree_balance_dirty(root, nr); 2557 return ret; 2558 } 2559 2560 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, 2561 struct btrfs_root *root, 2562 struct inode *dir, u64 objectid, 2563 const char *name, int name_len) 2564 { 2565 struct btrfs_path *path; 2566 struct extent_buffer *leaf; 2567 struct btrfs_dir_item *di; 2568 struct btrfs_key key; 2569 u64 index; 2570 int ret; 2571 2572 path = btrfs_alloc_path(); 2573 if (!path) 2574 return -ENOMEM; 2575 2576 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, 2577 name, name_len, -1); 2578 BUG_ON(!di || IS_ERR(di)); 2579 2580 leaf = path->nodes[0]; 2581 btrfs_dir_item_key_to_cpu(leaf, di, &key); 2582 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); 2583 ret = btrfs_delete_one_dir_name(trans, root, path, di); 2584 BUG_ON(ret); 2585 btrfs_release_path(root, path); 2586 2587 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, 2588 objectid, root->root_key.objectid, 2589 dir->i_ino, &index, name, name_len); 2590 if (ret < 0) { 2591 BUG_ON(ret != -ENOENT); 2592 di = btrfs_search_dir_index_item(root, path, dir->i_ino, 2593 name, name_len); 2594 BUG_ON(!di || IS_ERR(di)); 2595 2596 leaf = path->nodes[0]; 2597 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2598 btrfs_release_path(root, path); 2599 index = key.offset; 2600 } 2601 2602 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, 2603 index, name, name_len, -1); 2604 BUG_ON(!di || IS_ERR(di)); 2605 2606 leaf = path->nodes[0]; 2607 btrfs_dir_item_key_to_cpu(leaf, di, &key); 2608 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); 2609 ret = btrfs_delete_one_dir_name(trans, root, path, di); 2610 BUG_ON(ret); 2611 btrfs_release_path(root, path); 2612 2613 btrfs_i_size_write(dir, dir->i_size - name_len * 2); 2614 dir->i_mtime = dir->i_ctime = CURRENT_TIME; 2615 ret = btrfs_update_inode(trans, root, dir); 2616 BUG_ON(ret); 2617 dir->i_sb->s_dirt = 1; 2618 2619 btrfs_free_path(path); 2620 return 0; 2621 } 2622 2623 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) 2624 { 2625 struct inode *inode = dentry->d_inode; 2626 int err = 0; 2627 int ret; 2628 struct btrfs_root *root = BTRFS_I(dir)->root; 2629 struct btrfs_trans_handle *trans; 2630 unsigned long nr = 0; 2631 2632 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || 2633 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 2634 return -ENOTEMPTY; 2635 2636 ret = btrfs_reserve_metadata_space(root, 5); 2637 if (ret) 2638 return ret; 2639 2640 trans = btrfs_start_transaction(root, 1); 2641 if (IS_ERR(trans)) { 2642 btrfs_unreserve_metadata_space(root, 5); 2643 return PTR_ERR(trans); 2644 } 2645 2646 btrfs_set_trans_block_group(trans, dir); 2647 2648 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { 2649 err = btrfs_unlink_subvol(trans, root, dir, 2650 BTRFS_I(inode)->location.objectid, 2651 dentry->d_name.name, 2652 dentry->d_name.len); 2653 goto out; 2654 } 2655 2656 err = btrfs_orphan_add(trans, inode); 2657 if (err) 2658 goto out; 2659 2660 /* now the directory is empty */ 2661 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, 2662 dentry->d_name.name, dentry->d_name.len); 2663 if (!err) 2664 btrfs_i_size_write(inode, 0); 2665 out: 2666 nr = trans->blocks_used; 2667 ret = btrfs_end_transaction_throttle(trans, root); 2668 btrfs_unreserve_metadata_space(root, 5); 2669 btrfs_btree_balance_dirty(root, nr); 2670 2671 if (ret && !err) 2672 err = ret; 2673 return err; 2674 } 2675 2676 #if 0 2677 /* 2678 * when truncating bytes in a file, it is possible to avoid reading 2679 * the leaves that contain only checksum items. This can be the 2680 * majority of the IO required to delete a large file, but it must 2681 * be done carefully. 2682 * 2683 * The keys in the level just above the leaves are checked to make sure 2684 * the lowest key in a given leaf is a csum key, and starts at an offset 2685 * after the new size. 2686 * 2687 * Then the key for the next leaf is checked to make sure it also has 2688 * a checksum item for the same file. If it does, we know our target leaf 2689 * contains only checksum items, and it can be safely freed without reading 2690 * it. 2691 * 2692 * This is just an optimization targeted at large files. It may do 2693 * nothing. It will return 0 unless things went badly. 2694 */ 2695 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, 2696 struct btrfs_root *root, 2697 struct btrfs_path *path, 2698 struct inode *inode, u64 new_size) 2699 { 2700 struct btrfs_key key; 2701 int ret; 2702 int nritems; 2703 struct btrfs_key found_key; 2704 struct btrfs_key other_key; 2705 struct btrfs_leaf_ref *ref; 2706 u64 leaf_gen; 2707 u64 leaf_start; 2708 2709 path->lowest_level = 1; 2710 key.objectid = inode->i_ino; 2711 key.type = BTRFS_CSUM_ITEM_KEY; 2712 key.offset = new_size; 2713 again: 2714 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2715 if (ret < 0) 2716 goto out; 2717 2718 if (path->nodes[1] == NULL) { 2719 ret = 0; 2720 goto out; 2721 } 2722 ret = 0; 2723 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); 2724 nritems = btrfs_header_nritems(path->nodes[1]); 2725 2726 if (!nritems) 2727 goto out; 2728 2729 if (path->slots[1] >= nritems) 2730 goto next_node; 2731 2732 /* did we find a key greater than anything we want to delete? */ 2733 if (found_key.objectid > inode->i_ino || 2734 (found_key.objectid == inode->i_ino && found_key.type > key.type)) 2735 goto out; 2736 2737 /* we check the next key in the node to make sure the leave contains 2738 * only checksum items. This comparison doesn't work if our 2739 * leaf is the last one in the node 2740 */ 2741 if (path->slots[1] + 1 >= nritems) { 2742 next_node: 2743 /* search forward from the last key in the node, this 2744 * will bring us into the next node in the tree 2745 */ 2746 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); 2747 2748 /* unlikely, but we inc below, so check to be safe */ 2749 if (found_key.offset == (u64)-1) 2750 goto out; 2751 2752 /* search_forward needs a path with locks held, do the 2753 * search again for the original key. It is possible 2754 * this will race with a balance and return a path that 2755 * we could modify, but this drop is just an optimization 2756 * and is allowed to miss some leaves. 2757 */ 2758 btrfs_release_path(root, path); 2759 found_key.offset++; 2760 2761 /* setup a max key for search_forward */ 2762 other_key.offset = (u64)-1; 2763 other_key.type = key.type; 2764 other_key.objectid = key.objectid; 2765 2766 path->keep_locks = 1; 2767 ret = btrfs_search_forward(root, &found_key, &other_key, 2768 path, 0, 0); 2769 path->keep_locks = 0; 2770 if (ret || found_key.objectid != key.objectid || 2771 found_key.type != key.type) { 2772 ret = 0; 2773 goto out; 2774 } 2775 2776 key.offset = found_key.offset; 2777 btrfs_release_path(root, path); 2778 cond_resched(); 2779 goto again; 2780 } 2781 2782 /* we know there's one more slot after us in the tree, 2783 * read that key so we can verify it is also a checksum item 2784 */ 2785 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); 2786 2787 if (found_key.objectid < inode->i_ino) 2788 goto next_key; 2789 2790 if (found_key.type != key.type || found_key.offset < new_size) 2791 goto next_key; 2792 2793 /* 2794 * if the key for the next leaf isn't a csum key from this objectid, 2795 * we can't be sure there aren't good items inside this leaf. 2796 * Bail out 2797 */ 2798 if (other_key.objectid != inode->i_ino || other_key.type != key.type) 2799 goto out; 2800 2801 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); 2802 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); 2803 /* 2804 * it is safe to delete this leaf, it contains only 2805 * csum items from this inode at an offset >= new_size 2806 */ 2807 ret = btrfs_del_leaf(trans, root, path, leaf_start); 2808 BUG_ON(ret); 2809 2810 if (root->ref_cows && leaf_gen < trans->transid) { 2811 ref = btrfs_alloc_leaf_ref(root, 0); 2812 if (ref) { 2813 ref->root_gen = root->root_key.offset; 2814 ref->bytenr = leaf_start; 2815 ref->owner = 0; 2816 ref->generation = leaf_gen; 2817 ref->nritems = 0; 2818 2819 btrfs_sort_leaf_ref(ref); 2820 2821 ret = btrfs_add_leaf_ref(root, ref, 0); 2822 WARN_ON(ret); 2823 btrfs_free_leaf_ref(root, ref); 2824 } else { 2825 WARN_ON(1); 2826 } 2827 } 2828 next_key: 2829 btrfs_release_path(root, path); 2830 2831 if (other_key.objectid == inode->i_ino && 2832 other_key.type == key.type && other_key.offset > key.offset) { 2833 key.offset = other_key.offset; 2834 cond_resched(); 2835 goto again; 2836 } 2837 ret = 0; 2838 out: 2839 /* fixup any changes we've made to the path */ 2840 path->lowest_level = 0; 2841 path->keep_locks = 0; 2842 btrfs_release_path(root, path); 2843 return ret; 2844 } 2845 2846 #endif 2847 2848 /* 2849 * this can truncate away extent items, csum items and directory items. 2850 * It starts at a high offset and removes keys until it can't find 2851 * any higher than new_size 2852 * 2853 * csum items that cross the new i_size are truncated to the new size 2854 * as well. 2855 * 2856 * min_type is the minimum key type to truncate down to. If set to 0, this 2857 * will kill all the items on this inode, including the INODE_ITEM_KEY. 2858 */ 2859 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, 2860 struct btrfs_root *root, 2861 struct inode *inode, 2862 u64 new_size, u32 min_type) 2863 { 2864 struct btrfs_path *path; 2865 struct extent_buffer *leaf; 2866 struct btrfs_file_extent_item *fi; 2867 struct btrfs_key key; 2868 struct btrfs_key found_key; 2869 u64 extent_start = 0; 2870 u64 extent_num_bytes = 0; 2871 u64 extent_offset = 0; 2872 u64 item_end = 0; 2873 u64 mask = root->sectorsize - 1; 2874 u32 found_type = (u8)-1; 2875 int found_extent; 2876 int del_item; 2877 int pending_del_nr = 0; 2878 int pending_del_slot = 0; 2879 int extent_type = -1; 2880 int encoding; 2881 int ret; 2882 int err = 0; 2883 2884 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); 2885 2886 if (root->ref_cows) 2887 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); 2888 2889 path = btrfs_alloc_path(); 2890 BUG_ON(!path); 2891 path->reada = -1; 2892 2893 key.objectid = inode->i_ino; 2894 key.offset = (u64)-1; 2895 key.type = (u8)-1; 2896 2897 search_again: 2898 path->leave_spinning = 1; 2899 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2900 if (ret < 0) { 2901 err = ret; 2902 goto out; 2903 } 2904 2905 if (ret > 0) { 2906 /* there are no items in the tree for us to truncate, we're 2907 * done 2908 */ 2909 if (path->slots[0] == 0) 2910 goto out; 2911 path->slots[0]--; 2912 } 2913 2914 while (1) { 2915 fi = NULL; 2916 leaf = path->nodes[0]; 2917 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 2918 found_type = btrfs_key_type(&found_key); 2919 encoding = 0; 2920 2921 if (found_key.objectid != inode->i_ino) 2922 break; 2923 2924 if (found_type < min_type) 2925 break; 2926 2927 item_end = found_key.offset; 2928 if (found_type == BTRFS_EXTENT_DATA_KEY) { 2929 fi = btrfs_item_ptr(leaf, path->slots[0], 2930 struct btrfs_file_extent_item); 2931 extent_type = btrfs_file_extent_type(leaf, fi); 2932 encoding = btrfs_file_extent_compression(leaf, fi); 2933 encoding |= btrfs_file_extent_encryption(leaf, fi); 2934 encoding |= btrfs_file_extent_other_encoding(leaf, fi); 2935 2936 if (extent_type != BTRFS_FILE_EXTENT_INLINE) { 2937 item_end += 2938 btrfs_file_extent_num_bytes(leaf, fi); 2939 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { 2940 item_end += btrfs_file_extent_inline_len(leaf, 2941 fi); 2942 } 2943 item_end--; 2944 } 2945 if (found_type > min_type) { 2946 del_item = 1; 2947 } else { 2948 if (item_end < new_size) 2949 break; 2950 if (found_key.offset >= new_size) 2951 del_item = 1; 2952 else 2953 del_item = 0; 2954 } 2955 found_extent = 0; 2956 /* FIXME, shrink the extent if the ref count is only 1 */ 2957 if (found_type != BTRFS_EXTENT_DATA_KEY) 2958 goto delete; 2959 2960 if (extent_type != BTRFS_FILE_EXTENT_INLINE) { 2961 u64 num_dec; 2962 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); 2963 if (!del_item && !encoding) { 2964 u64 orig_num_bytes = 2965 btrfs_file_extent_num_bytes(leaf, fi); 2966 extent_num_bytes = new_size - 2967 found_key.offset + root->sectorsize - 1; 2968 extent_num_bytes = extent_num_bytes & 2969 ~((u64)root->sectorsize - 1); 2970 btrfs_set_file_extent_num_bytes(leaf, fi, 2971 extent_num_bytes); 2972 num_dec = (orig_num_bytes - 2973 extent_num_bytes); 2974 if (root->ref_cows && extent_start != 0) 2975 inode_sub_bytes(inode, num_dec); 2976 btrfs_mark_buffer_dirty(leaf); 2977 } else { 2978 extent_num_bytes = 2979 btrfs_file_extent_disk_num_bytes(leaf, 2980 fi); 2981 extent_offset = found_key.offset - 2982 btrfs_file_extent_offset(leaf, fi); 2983 2984 /* FIXME blocksize != 4096 */ 2985 num_dec = btrfs_file_extent_num_bytes(leaf, fi); 2986 if (extent_start != 0) { 2987 found_extent = 1; 2988 if (root->ref_cows) 2989 inode_sub_bytes(inode, num_dec); 2990 } 2991 } 2992 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { 2993 /* 2994 * we can't truncate inline items that have had 2995 * special encodings 2996 */ 2997 if (!del_item && 2998 btrfs_file_extent_compression(leaf, fi) == 0 && 2999 btrfs_file_extent_encryption(leaf, fi) == 0 && 3000 btrfs_file_extent_other_encoding(leaf, fi) == 0) { 3001 u32 size = new_size - found_key.offset; 3002 3003 if (root->ref_cows) { 3004 inode_sub_bytes(inode, item_end + 1 - 3005 new_size); 3006 } 3007 size = 3008 btrfs_file_extent_calc_inline_size(size); 3009 ret = btrfs_truncate_item(trans, root, path, 3010 size, 1); 3011 BUG_ON(ret); 3012 } else if (root->ref_cows) { 3013 inode_sub_bytes(inode, item_end + 1 - 3014 found_key.offset); 3015 } 3016 } 3017 delete: 3018 if (del_item) { 3019 if (!pending_del_nr) { 3020 /* no pending yet, add ourselves */ 3021 pending_del_slot = path->slots[0]; 3022 pending_del_nr = 1; 3023 } else if (pending_del_nr && 3024 path->slots[0] + 1 == pending_del_slot) { 3025 /* hop on the pending chunk */ 3026 pending_del_nr++; 3027 pending_del_slot = path->slots[0]; 3028 } else { 3029 BUG(); 3030 } 3031 } else { 3032 break; 3033 } 3034 if (found_extent && root->ref_cows) { 3035 btrfs_set_path_blocking(path); 3036 ret = btrfs_free_extent(trans, root, extent_start, 3037 extent_num_bytes, 0, 3038 btrfs_header_owner(leaf), 3039 inode->i_ino, extent_offset); 3040 BUG_ON(ret); 3041 } 3042 3043 if (found_type == BTRFS_INODE_ITEM_KEY) 3044 break; 3045 3046 if (path->slots[0] == 0 || 3047 path->slots[0] != pending_del_slot) { 3048 if (root->ref_cows) { 3049 err = -EAGAIN; 3050 goto out; 3051 } 3052 if (pending_del_nr) { 3053 ret = btrfs_del_items(trans, root, path, 3054 pending_del_slot, 3055 pending_del_nr); 3056 BUG_ON(ret); 3057 pending_del_nr = 0; 3058 } 3059 btrfs_release_path(root, path); 3060 goto search_again; 3061 } else { 3062 path->slots[0]--; 3063 } 3064 } 3065 out: 3066 if (pending_del_nr) { 3067 ret = btrfs_del_items(trans, root, path, pending_del_slot, 3068 pending_del_nr); 3069 } 3070 btrfs_free_path(path); 3071 return err; 3072 } 3073 3074 /* 3075 * taken from block_truncate_page, but does cow as it zeros out 3076 * any bytes left in the last page in the file. 3077 */ 3078 static int btrfs_truncate_page(struct address_space *mapping, loff_t from) 3079 { 3080 struct inode *inode = mapping->host; 3081 struct btrfs_root *root = BTRFS_I(inode)->root; 3082 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 3083 struct btrfs_ordered_extent *ordered; 3084 char *kaddr; 3085 u32 blocksize = root->sectorsize; 3086 pgoff_t index = from >> PAGE_CACHE_SHIFT; 3087 unsigned offset = from & (PAGE_CACHE_SIZE-1); 3088 struct page *page; 3089 int ret = 0; 3090 u64 page_start; 3091 u64 page_end; 3092 3093 if ((offset & (blocksize - 1)) == 0) 3094 goto out; 3095 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); 3096 if (ret) 3097 goto out; 3098 3099 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); 3100 if (ret) 3101 goto out; 3102 3103 ret = -ENOMEM; 3104 again: 3105 page = grab_cache_page(mapping, index); 3106 if (!page) { 3107 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); 3108 btrfs_unreserve_metadata_for_delalloc(root, inode, 1); 3109 goto out; 3110 } 3111 3112 page_start = page_offset(page); 3113 page_end = page_start + PAGE_CACHE_SIZE - 1; 3114 3115 if (!PageUptodate(page)) { 3116 ret = btrfs_readpage(NULL, page); 3117 lock_page(page); 3118 if (page->mapping != mapping) { 3119 unlock_page(page); 3120 page_cache_release(page); 3121 goto again; 3122 } 3123 if (!PageUptodate(page)) { 3124 ret = -EIO; 3125 goto out_unlock; 3126 } 3127 } 3128 wait_on_page_writeback(page); 3129 3130 lock_extent(io_tree, page_start, page_end, GFP_NOFS); 3131 set_page_extent_mapped(page); 3132 3133 ordered = btrfs_lookup_ordered_extent(inode, page_start); 3134 if (ordered) { 3135 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 3136 unlock_page(page); 3137 page_cache_release(page); 3138 btrfs_start_ordered_extent(inode, ordered, 1); 3139 btrfs_put_ordered_extent(ordered); 3140 goto again; 3141 } 3142 3143 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 3144 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, 3145 GFP_NOFS); 3146 3147 ret = btrfs_set_extent_delalloc(inode, page_start, page_end); 3148 if (ret) { 3149 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 3150 goto out_unlock; 3151 } 3152 3153 ret = 0; 3154 if (offset != PAGE_CACHE_SIZE) { 3155 kaddr = kmap(page); 3156 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); 3157 flush_dcache_page(page); 3158 kunmap(page); 3159 } 3160 ClearPageChecked(page); 3161 set_page_dirty(page); 3162 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 3163 3164 out_unlock: 3165 if (ret) 3166 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); 3167 btrfs_unreserve_metadata_for_delalloc(root, inode, 1); 3168 unlock_page(page); 3169 page_cache_release(page); 3170 out: 3171 return ret; 3172 } 3173 3174 int btrfs_cont_expand(struct inode *inode, loff_t size) 3175 { 3176 struct btrfs_trans_handle *trans; 3177 struct btrfs_root *root = BTRFS_I(inode)->root; 3178 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 3179 struct extent_map *em; 3180 u64 mask = root->sectorsize - 1; 3181 u64 hole_start = (inode->i_size + mask) & ~mask; 3182 u64 block_end = (size + mask) & ~mask; 3183 u64 last_byte; 3184 u64 cur_offset; 3185 u64 hole_size; 3186 int err = 0; 3187 3188 if (size <= hole_start) 3189 return 0; 3190 3191 while (1) { 3192 struct btrfs_ordered_extent *ordered; 3193 btrfs_wait_ordered_range(inode, hole_start, 3194 block_end - hole_start); 3195 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); 3196 ordered = btrfs_lookup_ordered_extent(inode, hole_start); 3197 if (!ordered) 3198 break; 3199 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); 3200 btrfs_put_ordered_extent(ordered); 3201 } 3202 3203 cur_offset = hole_start; 3204 while (1) { 3205 em = btrfs_get_extent(inode, NULL, 0, cur_offset, 3206 block_end - cur_offset, 0); 3207 BUG_ON(IS_ERR(em) || !em); 3208 last_byte = min(extent_map_end(em), block_end); 3209 last_byte = (last_byte + mask) & ~mask; 3210 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { 3211 u64 hint_byte = 0; 3212 hole_size = last_byte - cur_offset; 3213 3214 err = btrfs_reserve_metadata_space(root, 2); 3215 if (err) 3216 break; 3217 3218 trans = btrfs_start_transaction(root, 1); 3219 btrfs_set_trans_block_group(trans, inode); 3220 3221 err = btrfs_drop_extents(trans, inode, cur_offset, 3222 cur_offset + hole_size, 3223 &hint_byte, 1); 3224 BUG_ON(err); 3225 3226 err = btrfs_insert_file_extent(trans, root, 3227 inode->i_ino, cur_offset, 0, 3228 0, hole_size, 0, hole_size, 3229 0, 0, 0); 3230 BUG_ON(err); 3231 3232 btrfs_drop_extent_cache(inode, hole_start, 3233 last_byte - 1, 0); 3234 3235 btrfs_end_transaction(trans, root); 3236 btrfs_unreserve_metadata_space(root, 2); 3237 } 3238 free_extent_map(em); 3239 cur_offset = last_byte; 3240 if (cur_offset >= block_end) 3241 break; 3242 } 3243 3244 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); 3245 return err; 3246 } 3247 3248 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr) 3249 { 3250 struct btrfs_root *root = BTRFS_I(inode)->root; 3251 struct btrfs_trans_handle *trans; 3252 unsigned long nr; 3253 int ret; 3254 3255 if (attr->ia_size == inode->i_size) 3256 return 0; 3257 3258 if (attr->ia_size > inode->i_size) { 3259 unsigned long limit; 3260 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 3261 if (attr->ia_size > inode->i_sb->s_maxbytes) 3262 return -EFBIG; 3263 if (limit != RLIM_INFINITY && attr->ia_size > limit) { 3264 send_sig(SIGXFSZ, current, 0); 3265 return -EFBIG; 3266 } 3267 } 3268 3269 ret = btrfs_reserve_metadata_space(root, 1); 3270 if (ret) 3271 return ret; 3272 3273 trans = btrfs_start_transaction(root, 1); 3274 btrfs_set_trans_block_group(trans, inode); 3275 3276 ret = btrfs_orphan_add(trans, inode); 3277 BUG_ON(ret); 3278 3279 nr = trans->blocks_used; 3280 btrfs_end_transaction(trans, root); 3281 btrfs_unreserve_metadata_space(root, 1); 3282 btrfs_btree_balance_dirty(root, nr); 3283 3284 if (attr->ia_size > inode->i_size) { 3285 ret = btrfs_cont_expand(inode, attr->ia_size); 3286 if (ret) { 3287 btrfs_truncate(inode); 3288 return ret; 3289 } 3290 3291 i_size_write(inode, attr->ia_size); 3292 btrfs_ordered_update_i_size(inode, inode->i_size, NULL); 3293 3294 trans = btrfs_start_transaction(root, 1); 3295 btrfs_set_trans_block_group(trans, inode); 3296 3297 ret = btrfs_update_inode(trans, root, inode); 3298 BUG_ON(ret); 3299 if (inode->i_nlink > 0) { 3300 ret = btrfs_orphan_del(trans, inode); 3301 BUG_ON(ret); 3302 } 3303 nr = trans->blocks_used; 3304 btrfs_end_transaction(trans, root); 3305 btrfs_btree_balance_dirty(root, nr); 3306 return 0; 3307 } 3308 3309 /* 3310 * We're truncating a file that used to have good data down to 3311 * zero. Make sure it gets into the ordered flush list so that 3312 * any new writes get down to disk quickly. 3313 */ 3314 if (attr->ia_size == 0) 3315 BTRFS_I(inode)->ordered_data_close = 1; 3316 3317 /* we don't support swapfiles, so vmtruncate shouldn't fail */ 3318 ret = vmtruncate(inode, attr->ia_size); 3319 BUG_ON(ret); 3320 3321 return 0; 3322 } 3323 3324 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) 3325 { 3326 struct inode *inode = dentry->d_inode; 3327 int err; 3328 3329 err = inode_change_ok(inode, attr); 3330 if (err) 3331 return err; 3332 3333 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 3334 err = btrfs_setattr_size(inode, attr); 3335 if (err) 3336 return err; 3337 } 3338 attr->ia_valid &= ~ATTR_SIZE; 3339 3340 if (attr->ia_valid) 3341 err = inode_setattr(inode, attr); 3342 3343 if (!err && ((attr->ia_valid & ATTR_MODE))) 3344 err = btrfs_acl_chmod(inode); 3345 return err; 3346 } 3347 3348 void btrfs_delete_inode(struct inode *inode) 3349 { 3350 struct btrfs_trans_handle *trans; 3351 struct btrfs_root *root = BTRFS_I(inode)->root; 3352 unsigned long nr; 3353 int ret; 3354 3355 truncate_inode_pages(&inode->i_data, 0); 3356 if (is_bad_inode(inode)) { 3357 btrfs_orphan_del(NULL, inode); 3358 goto no_delete; 3359 } 3360 btrfs_wait_ordered_range(inode, 0, (u64)-1); 3361 3362 if (root->fs_info->log_root_recovering) { 3363 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); 3364 goto no_delete; 3365 } 3366 3367 if (inode->i_nlink > 0) { 3368 BUG_ON(btrfs_root_refs(&root->root_item) != 0); 3369 goto no_delete; 3370 } 3371 3372 btrfs_i_size_write(inode, 0); 3373 3374 while (1) { 3375 trans = btrfs_start_transaction(root, 1); 3376 btrfs_set_trans_block_group(trans, inode); 3377 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); 3378 3379 if (ret != -EAGAIN) 3380 break; 3381 3382 nr = trans->blocks_used; 3383 btrfs_end_transaction(trans, root); 3384 trans = NULL; 3385 btrfs_btree_balance_dirty(root, nr); 3386 } 3387 3388 if (ret == 0) { 3389 ret = btrfs_orphan_del(trans, inode); 3390 BUG_ON(ret); 3391 } 3392 3393 nr = trans->blocks_used; 3394 btrfs_end_transaction(trans, root); 3395 btrfs_btree_balance_dirty(root, nr); 3396 no_delete: 3397 clear_inode(inode); 3398 return; 3399 } 3400 3401 /* 3402 * this returns the key found in the dir entry in the location pointer. 3403 * If no dir entries were found, location->objectid is 0. 3404 */ 3405 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, 3406 struct btrfs_key *location) 3407 { 3408 const char *name = dentry->d_name.name; 3409 int namelen = dentry->d_name.len; 3410 struct btrfs_dir_item *di; 3411 struct btrfs_path *path; 3412 struct btrfs_root *root = BTRFS_I(dir)->root; 3413 int ret = 0; 3414 3415 path = btrfs_alloc_path(); 3416 BUG_ON(!path); 3417 3418 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, 3419 namelen, 0); 3420 if (IS_ERR(di)) 3421 ret = PTR_ERR(di); 3422 3423 if (!di || IS_ERR(di)) 3424 goto out_err; 3425 3426 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); 3427 out: 3428 btrfs_free_path(path); 3429 return ret; 3430 out_err: 3431 location->objectid = 0; 3432 goto out; 3433 } 3434 3435 /* 3436 * when we hit a tree root in a directory, the btrfs part of the inode 3437 * needs to be changed to reflect the root directory of the tree root. This 3438 * is kind of like crossing a mount point. 3439 */ 3440 static int fixup_tree_root_location(struct btrfs_root *root, 3441 struct inode *dir, 3442 struct dentry *dentry, 3443 struct btrfs_key *location, 3444 struct btrfs_root **sub_root) 3445 { 3446 struct btrfs_path *path; 3447 struct btrfs_root *new_root; 3448 struct btrfs_root_ref *ref; 3449 struct extent_buffer *leaf; 3450 int ret; 3451 int err = 0; 3452 3453 path = btrfs_alloc_path(); 3454 if (!path) { 3455 err = -ENOMEM; 3456 goto out; 3457 } 3458 3459 err = -ENOENT; 3460 ret = btrfs_find_root_ref(root->fs_info->tree_root, path, 3461 BTRFS_I(dir)->root->root_key.objectid, 3462 location->objectid); 3463 if (ret) { 3464 if (ret < 0) 3465 err = ret; 3466 goto out; 3467 } 3468 3469 leaf = path->nodes[0]; 3470 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); 3471 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || 3472 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) 3473 goto out; 3474 3475 ret = memcmp_extent_buffer(leaf, dentry->d_name.name, 3476 (unsigned long)(ref + 1), 3477 dentry->d_name.len); 3478 if (ret) 3479 goto out; 3480 3481 btrfs_release_path(root->fs_info->tree_root, path); 3482 3483 new_root = btrfs_read_fs_root_no_name(root->fs_info, location); 3484 if (IS_ERR(new_root)) { 3485 err = PTR_ERR(new_root); 3486 goto out; 3487 } 3488 3489 if (btrfs_root_refs(&new_root->root_item) == 0) { 3490 err = -ENOENT; 3491 goto out; 3492 } 3493 3494 *sub_root = new_root; 3495 location->objectid = btrfs_root_dirid(&new_root->root_item); 3496 location->type = BTRFS_INODE_ITEM_KEY; 3497 location->offset = 0; 3498 err = 0; 3499 out: 3500 btrfs_free_path(path); 3501 return err; 3502 } 3503 3504 static void inode_tree_add(struct inode *inode) 3505 { 3506 struct btrfs_root *root = BTRFS_I(inode)->root; 3507 struct btrfs_inode *entry; 3508 struct rb_node **p; 3509 struct rb_node *parent; 3510 again: 3511 p = &root->inode_tree.rb_node; 3512 parent = NULL; 3513 3514 if (hlist_unhashed(&inode->i_hash)) 3515 return; 3516 3517 spin_lock(&root->inode_lock); 3518 while (*p) { 3519 parent = *p; 3520 entry = rb_entry(parent, struct btrfs_inode, rb_node); 3521 3522 if (inode->i_ino < entry->vfs_inode.i_ino) 3523 p = &parent->rb_left; 3524 else if (inode->i_ino > entry->vfs_inode.i_ino) 3525 p = &parent->rb_right; 3526 else { 3527 WARN_ON(!(entry->vfs_inode.i_state & 3528 (I_WILL_FREE | I_FREEING | I_CLEAR))); 3529 rb_erase(parent, &root->inode_tree); 3530 RB_CLEAR_NODE(parent); 3531 spin_unlock(&root->inode_lock); 3532 goto again; 3533 } 3534 } 3535 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); 3536 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); 3537 spin_unlock(&root->inode_lock); 3538 } 3539 3540 static void inode_tree_del(struct inode *inode) 3541 { 3542 struct btrfs_root *root = BTRFS_I(inode)->root; 3543 int empty = 0; 3544 3545 spin_lock(&root->inode_lock); 3546 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { 3547 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); 3548 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 3549 empty = RB_EMPTY_ROOT(&root->inode_tree); 3550 } 3551 spin_unlock(&root->inode_lock); 3552 3553 if (empty && btrfs_root_refs(&root->root_item) == 0) { 3554 synchronize_srcu(&root->fs_info->subvol_srcu); 3555 spin_lock(&root->inode_lock); 3556 empty = RB_EMPTY_ROOT(&root->inode_tree); 3557 spin_unlock(&root->inode_lock); 3558 if (empty) 3559 btrfs_add_dead_root(root); 3560 } 3561 } 3562 3563 int btrfs_invalidate_inodes(struct btrfs_root *root) 3564 { 3565 struct rb_node *node; 3566 struct rb_node *prev; 3567 struct btrfs_inode *entry; 3568 struct inode *inode; 3569 u64 objectid = 0; 3570 3571 WARN_ON(btrfs_root_refs(&root->root_item) != 0); 3572 3573 spin_lock(&root->inode_lock); 3574 again: 3575 node = root->inode_tree.rb_node; 3576 prev = NULL; 3577 while (node) { 3578 prev = node; 3579 entry = rb_entry(node, struct btrfs_inode, rb_node); 3580 3581 if (objectid < entry->vfs_inode.i_ino) 3582 node = node->rb_left; 3583 else if (objectid > entry->vfs_inode.i_ino) 3584 node = node->rb_right; 3585 else 3586 break; 3587 } 3588 if (!node) { 3589 while (prev) { 3590 entry = rb_entry(prev, struct btrfs_inode, rb_node); 3591 if (objectid <= entry->vfs_inode.i_ino) { 3592 node = prev; 3593 break; 3594 } 3595 prev = rb_next(prev); 3596 } 3597 } 3598 while (node) { 3599 entry = rb_entry(node, struct btrfs_inode, rb_node); 3600 objectid = entry->vfs_inode.i_ino + 1; 3601 inode = igrab(&entry->vfs_inode); 3602 if (inode) { 3603 spin_unlock(&root->inode_lock); 3604 if (atomic_read(&inode->i_count) > 1) 3605 d_prune_aliases(inode); 3606 /* 3607 * btrfs_drop_inode will remove it from 3608 * the inode cache when its usage count 3609 * hits zero. 3610 */ 3611 iput(inode); 3612 cond_resched(); 3613 spin_lock(&root->inode_lock); 3614 goto again; 3615 } 3616 3617 if (cond_resched_lock(&root->inode_lock)) 3618 goto again; 3619 3620 node = rb_next(node); 3621 } 3622 spin_unlock(&root->inode_lock); 3623 return 0; 3624 } 3625 3626 static noinline void init_btrfs_i(struct inode *inode) 3627 { 3628 struct btrfs_inode *bi = BTRFS_I(inode); 3629 3630 bi->generation = 0; 3631 bi->sequence = 0; 3632 bi->last_trans = 0; 3633 bi->last_sub_trans = 0; 3634 bi->logged_trans = 0; 3635 bi->delalloc_bytes = 0; 3636 bi->reserved_bytes = 0; 3637 bi->disk_i_size = 0; 3638 bi->flags = 0; 3639 bi->index_cnt = (u64)-1; 3640 bi->last_unlink_trans = 0; 3641 bi->ordered_data_close = 0; 3642 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); 3643 extent_io_tree_init(&BTRFS_I(inode)->io_tree, 3644 inode->i_mapping, GFP_NOFS); 3645 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, 3646 inode->i_mapping, GFP_NOFS); 3647 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); 3648 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); 3649 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 3650 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); 3651 mutex_init(&BTRFS_I(inode)->log_mutex); 3652 } 3653 3654 static int btrfs_init_locked_inode(struct inode *inode, void *p) 3655 { 3656 struct btrfs_iget_args *args = p; 3657 inode->i_ino = args->ino; 3658 init_btrfs_i(inode); 3659 BTRFS_I(inode)->root = args->root; 3660 btrfs_set_inode_space_info(args->root, inode); 3661 return 0; 3662 } 3663 3664 static int btrfs_find_actor(struct inode *inode, void *opaque) 3665 { 3666 struct btrfs_iget_args *args = opaque; 3667 return args->ino == inode->i_ino && 3668 args->root == BTRFS_I(inode)->root; 3669 } 3670 3671 static struct inode *btrfs_iget_locked(struct super_block *s, 3672 u64 objectid, 3673 struct btrfs_root *root) 3674 { 3675 struct inode *inode; 3676 struct btrfs_iget_args args; 3677 args.ino = objectid; 3678 args.root = root; 3679 3680 inode = iget5_locked(s, objectid, btrfs_find_actor, 3681 btrfs_init_locked_inode, 3682 (void *)&args); 3683 return inode; 3684 } 3685 3686 /* Get an inode object given its location and corresponding root. 3687 * Returns in *is_new if the inode was read from disk 3688 */ 3689 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, 3690 struct btrfs_root *root) 3691 { 3692 struct inode *inode; 3693 3694 inode = btrfs_iget_locked(s, location->objectid, root); 3695 if (!inode) 3696 return ERR_PTR(-ENOMEM); 3697 3698 if (inode->i_state & I_NEW) { 3699 BTRFS_I(inode)->root = root; 3700 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); 3701 btrfs_read_locked_inode(inode); 3702 3703 inode_tree_add(inode); 3704 unlock_new_inode(inode); 3705 } 3706 3707 return inode; 3708 } 3709 3710 static struct inode *new_simple_dir(struct super_block *s, 3711 struct btrfs_key *key, 3712 struct btrfs_root *root) 3713 { 3714 struct inode *inode = new_inode(s); 3715 3716 if (!inode) 3717 return ERR_PTR(-ENOMEM); 3718 3719 init_btrfs_i(inode); 3720 3721 BTRFS_I(inode)->root = root; 3722 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); 3723 BTRFS_I(inode)->dummy_inode = 1; 3724 3725 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; 3726 inode->i_op = &simple_dir_inode_operations; 3727 inode->i_fop = &simple_dir_operations; 3728 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; 3729 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 3730 3731 return inode; 3732 } 3733 3734 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) 3735 { 3736 struct inode *inode; 3737 struct btrfs_root *root = BTRFS_I(dir)->root; 3738 struct btrfs_root *sub_root = root; 3739 struct btrfs_key location; 3740 int index; 3741 int ret; 3742 3743 dentry->d_op = &btrfs_dentry_operations; 3744 3745 if (dentry->d_name.len > BTRFS_NAME_LEN) 3746 return ERR_PTR(-ENAMETOOLONG); 3747 3748 ret = btrfs_inode_by_name(dir, dentry, &location); 3749 3750 if (ret < 0) 3751 return ERR_PTR(ret); 3752 3753 if (location.objectid == 0) 3754 return NULL; 3755 3756 if (location.type == BTRFS_INODE_ITEM_KEY) { 3757 inode = btrfs_iget(dir->i_sb, &location, root); 3758 return inode; 3759 } 3760 3761 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); 3762 3763 index = srcu_read_lock(&root->fs_info->subvol_srcu); 3764 ret = fixup_tree_root_location(root, dir, dentry, 3765 &location, &sub_root); 3766 if (ret < 0) { 3767 if (ret != -ENOENT) 3768 inode = ERR_PTR(ret); 3769 else 3770 inode = new_simple_dir(dir->i_sb, &location, sub_root); 3771 } else { 3772 inode = btrfs_iget(dir->i_sb, &location, sub_root); 3773 } 3774 srcu_read_unlock(&root->fs_info->subvol_srcu, index); 3775 3776 if (root != sub_root) { 3777 down_read(&root->fs_info->cleanup_work_sem); 3778 if (!(inode->i_sb->s_flags & MS_RDONLY)) 3779 btrfs_orphan_cleanup(sub_root); 3780 up_read(&root->fs_info->cleanup_work_sem); 3781 } 3782 3783 return inode; 3784 } 3785 3786 static int btrfs_dentry_delete(struct dentry *dentry) 3787 { 3788 struct btrfs_root *root; 3789 3790 if (!dentry->d_inode && !IS_ROOT(dentry)) 3791 dentry = dentry->d_parent; 3792 3793 if (dentry->d_inode) { 3794 root = BTRFS_I(dentry->d_inode)->root; 3795 if (btrfs_root_refs(&root->root_item) == 0) 3796 return 1; 3797 } 3798 return 0; 3799 } 3800 3801 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, 3802 struct nameidata *nd) 3803 { 3804 struct inode *inode; 3805 3806 inode = btrfs_lookup_dentry(dir, dentry); 3807 if (IS_ERR(inode)) 3808 return ERR_CAST(inode); 3809 3810 return d_splice_alias(inode, dentry); 3811 } 3812 3813 static unsigned char btrfs_filetype_table[] = { 3814 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK 3815 }; 3816 3817 static int btrfs_real_readdir(struct file *filp, void *dirent, 3818 filldir_t filldir) 3819 { 3820 struct inode *inode = filp->f_dentry->d_inode; 3821 struct btrfs_root *root = BTRFS_I(inode)->root; 3822 struct btrfs_item *item; 3823 struct btrfs_dir_item *di; 3824 struct btrfs_key key; 3825 struct btrfs_key found_key; 3826 struct btrfs_path *path; 3827 int ret; 3828 u32 nritems; 3829 struct extent_buffer *leaf; 3830 int slot; 3831 int advance; 3832 unsigned char d_type; 3833 int over = 0; 3834 u32 di_cur; 3835 u32 di_total; 3836 u32 di_len; 3837 int key_type = BTRFS_DIR_INDEX_KEY; 3838 char tmp_name[32]; 3839 char *name_ptr; 3840 int name_len; 3841 3842 /* FIXME, use a real flag for deciding about the key type */ 3843 if (root->fs_info->tree_root == root) 3844 key_type = BTRFS_DIR_ITEM_KEY; 3845 3846 /* special case for "." */ 3847 if (filp->f_pos == 0) { 3848 over = filldir(dirent, ".", 1, 3849 1, inode->i_ino, 3850 DT_DIR); 3851 if (over) 3852 return 0; 3853 filp->f_pos = 1; 3854 } 3855 /* special case for .., just use the back ref */ 3856 if (filp->f_pos == 1) { 3857 u64 pino = parent_ino(filp->f_path.dentry); 3858 over = filldir(dirent, "..", 2, 3859 2, pino, DT_DIR); 3860 if (over) 3861 return 0; 3862 filp->f_pos = 2; 3863 } 3864 path = btrfs_alloc_path(); 3865 path->reada = 2; 3866 3867 btrfs_set_key_type(&key, key_type); 3868 key.offset = filp->f_pos; 3869 key.objectid = inode->i_ino; 3870 3871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 3872 if (ret < 0) 3873 goto err; 3874 advance = 0; 3875 3876 while (1) { 3877 leaf = path->nodes[0]; 3878 nritems = btrfs_header_nritems(leaf); 3879 slot = path->slots[0]; 3880 if (advance || slot >= nritems) { 3881 if (slot >= nritems - 1) { 3882 ret = btrfs_next_leaf(root, path); 3883 if (ret) 3884 break; 3885 leaf = path->nodes[0]; 3886 nritems = btrfs_header_nritems(leaf); 3887 slot = path->slots[0]; 3888 } else { 3889 slot++; 3890 path->slots[0]++; 3891 } 3892 } 3893 3894 advance = 1; 3895 item = btrfs_item_nr(leaf, slot); 3896 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3897 3898 if (found_key.objectid != key.objectid) 3899 break; 3900 if (btrfs_key_type(&found_key) != key_type) 3901 break; 3902 if (found_key.offset < filp->f_pos) 3903 continue; 3904 3905 filp->f_pos = found_key.offset; 3906 3907 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); 3908 di_cur = 0; 3909 di_total = btrfs_item_size(leaf, item); 3910 3911 while (di_cur < di_total) { 3912 struct btrfs_key location; 3913 3914 name_len = btrfs_dir_name_len(leaf, di); 3915 if (name_len <= sizeof(tmp_name)) { 3916 name_ptr = tmp_name; 3917 } else { 3918 name_ptr = kmalloc(name_len, GFP_NOFS); 3919 if (!name_ptr) { 3920 ret = -ENOMEM; 3921 goto err; 3922 } 3923 } 3924 read_extent_buffer(leaf, name_ptr, 3925 (unsigned long)(di + 1), name_len); 3926 3927 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; 3928 btrfs_dir_item_key_to_cpu(leaf, di, &location); 3929 3930 /* is this a reference to our own snapshot? If so 3931 * skip it 3932 */ 3933 if (location.type == BTRFS_ROOT_ITEM_KEY && 3934 location.objectid == root->root_key.objectid) { 3935 over = 0; 3936 goto skip; 3937 } 3938 over = filldir(dirent, name_ptr, name_len, 3939 found_key.offset, location.objectid, 3940 d_type); 3941 3942 skip: 3943 if (name_ptr != tmp_name) 3944 kfree(name_ptr); 3945 3946 if (over) 3947 goto nopos; 3948 di_len = btrfs_dir_name_len(leaf, di) + 3949 btrfs_dir_data_len(leaf, di) + sizeof(*di); 3950 di_cur += di_len; 3951 di = (struct btrfs_dir_item *)((char *)di + di_len); 3952 } 3953 } 3954 3955 /* Reached end of directory/root. Bump pos past the last item. */ 3956 if (key_type == BTRFS_DIR_INDEX_KEY) 3957 /* 3958 * 32-bit glibc will use getdents64, but then strtol - 3959 * so the last number we can serve is this. 3960 */ 3961 filp->f_pos = 0x7fffffff; 3962 else 3963 filp->f_pos++; 3964 nopos: 3965 ret = 0; 3966 err: 3967 btrfs_free_path(path); 3968 return ret; 3969 } 3970 3971 int btrfs_write_inode(struct inode *inode, int wait) 3972 { 3973 struct btrfs_root *root = BTRFS_I(inode)->root; 3974 struct btrfs_trans_handle *trans; 3975 int ret = 0; 3976 3977 if (root->fs_info->btree_inode == inode) 3978 return 0; 3979 3980 if (wait) { 3981 trans = btrfs_join_transaction(root, 1); 3982 btrfs_set_trans_block_group(trans, inode); 3983 ret = btrfs_commit_transaction(trans, root); 3984 } 3985 return ret; 3986 } 3987 3988 /* 3989 * This is somewhat expensive, updating the tree every time the 3990 * inode changes. But, it is most likely to find the inode in cache. 3991 * FIXME, needs more benchmarking...there are no reasons other than performance 3992 * to keep or drop this code. 3993 */ 3994 void btrfs_dirty_inode(struct inode *inode) 3995 { 3996 struct btrfs_root *root = BTRFS_I(inode)->root; 3997 struct btrfs_trans_handle *trans; 3998 3999 trans = btrfs_join_transaction(root, 1); 4000 btrfs_set_trans_block_group(trans, inode); 4001 btrfs_update_inode(trans, root, inode); 4002 btrfs_end_transaction(trans, root); 4003 } 4004 4005 /* 4006 * find the highest existing sequence number in a directory 4007 * and then set the in-memory index_cnt variable to reflect 4008 * free sequence numbers 4009 */ 4010 static int btrfs_set_inode_index_count(struct inode *inode) 4011 { 4012 struct btrfs_root *root = BTRFS_I(inode)->root; 4013 struct btrfs_key key, found_key; 4014 struct btrfs_path *path; 4015 struct extent_buffer *leaf; 4016 int ret; 4017 4018 key.objectid = inode->i_ino; 4019 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); 4020 key.offset = (u64)-1; 4021 4022 path = btrfs_alloc_path(); 4023 if (!path) 4024 return -ENOMEM; 4025 4026 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4027 if (ret < 0) 4028 goto out; 4029 /* FIXME: we should be able to handle this */ 4030 if (ret == 0) 4031 goto out; 4032 ret = 0; 4033 4034 /* 4035 * MAGIC NUMBER EXPLANATION: 4036 * since we search a directory based on f_pos we have to start at 2 4037 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody 4038 * else has to start at 2 4039 */ 4040 if (path->slots[0] == 0) { 4041 BTRFS_I(inode)->index_cnt = 2; 4042 goto out; 4043 } 4044 4045 path->slots[0]--; 4046 4047 leaf = path->nodes[0]; 4048 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4049 4050 if (found_key.objectid != inode->i_ino || 4051 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { 4052 BTRFS_I(inode)->index_cnt = 2; 4053 goto out; 4054 } 4055 4056 BTRFS_I(inode)->index_cnt = found_key.offset + 1; 4057 out: 4058 btrfs_free_path(path); 4059 return ret; 4060 } 4061 4062 /* 4063 * helper to find a free sequence number in a given directory. This current 4064 * code is very simple, later versions will do smarter things in the btree 4065 */ 4066 int btrfs_set_inode_index(struct inode *dir, u64 *index) 4067 { 4068 int ret = 0; 4069 4070 if (BTRFS_I(dir)->index_cnt == (u64)-1) { 4071 ret = btrfs_set_inode_index_count(dir); 4072 if (ret) 4073 return ret; 4074 } 4075 4076 *index = BTRFS_I(dir)->index_cnt; 4077 BTRFS_I(dir)->index_cnt++; 4078 4079 return ret; 4080 } 4081 4082 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, 4083 struct btrfs_root *root, 4084 struct inode *dir, 4085 const char *name, int name_len, 4086 u64 ref_objectid, u64 objectid, 4087 u64 alloc_hint, int mode, u64 *index) 4088 { 4089 struct inode *inode; 4090 struct btrfs_inode_item *inode_item; 4091 struct btrfs_key *location; 4092 struct btrfs_path *path; 4093 struct btrfs_inode_ref *ref; 4094 struct btrfs_key key[2]; 4095 u32 sizes[2]; 4096 unsigned long ptr; 4097 int ret; 4098 int owner; 4099 4100 path = btrfs_alloc_path(); 4101 BUG_ON(!path); 4102 4103 inode = new_inode(root->fs_info->sb); 4104 if (!inode) 4105 return ERR_PTR(-ENOMEM); 4106 4107 if (dir) { 4108 ret = btrfs_set_inode_index(dir, index); 4109 if (ret) { 4110 iput(inode); 4111 return ERR_PTR(ret); 4112 } 4113 } 4114 /* 4115 * index_cnt is ignored for everything but a dir, 4116 * btrfs_get_inode_index_count has an explanation for the magic 4117 * number 4118 */ 4119 init_btrfs_i(inode); 4120 BTRFS_I(inode)->index_cnt = 2; 4121 BTRFS_I(inode)->root = root; 4122 BTRFS_I(inode)->generation = trans->transid; 4123 btrfs_set_inode_space_info(root, inode); 4124 4125 if (mode & S_IFDIR) 4126 owner = 0; 4127 else 4128 owner = 1; 4129 BTRFS_I(inode)->block_group = 4130 btrfs_find_block_group(root, 0, alloc_hint, owner); 4131 4132 key[0].objectid = objectid; 4133 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); 4134 key[0].offset = 0; 4135 4136 key[1].objectid = objectid; 4137 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); 4138 key[1].offset = ref_objectid; 4139 4140 sizes[0] = sizeof(struct btrfs_inode_item); 4141 sizes[1] = name_len + sizeof(*ref); 4142 4143 path->leave_spinning = 1; 4144 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); 4145 if (ret != 0) 4146 goto fail; 4147 4148 inode->i_uid = current_fsuid(); 4149 4150 if (dir && (dir->i_mode & S_ISGID)) { 4151 inode->i_gid = dir->i_gid; 4152 if (S_ISDIR(mode)) 4153 mode |= S_ISGID; 4154 } else 4155 inode->i_gid = current_fsgid(); 4156 4157 inode->i_mode = mode; 4158 inode->i_ino = objectid; 4159 inode_set_bytes(inode, 0); 4160 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 4161 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], 4162 struct btrfs_inode_item); 4163 fill_inode_item(trans, path->nodes[0], inode_item, inode); 4164 4165 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, 4166 struct btrfs_inode_ref); 4167 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); 4168 btrfs_set_inode_ref_index(path->nodes[0], ref, *index); 4169 ptr = (unsigned long)(ref + 1); 4170 write_extent_buffer(path->nodes[0], name, ptr, name_len); 4171 4172 btrfs_mark_buffer_dirty(path->nodes[0]); 4173 btrfs_free_path(path); 4174 4175 location = &BTRFS_I(inode)->location; 4176 location->objectid = objectid; 4177 location->offset = 0; 4178 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); 4179 4180 btrfs_inherit_iflags(inode, dir); 4181 4182 if ((mode & S_IFREG)) { 4183 if (btrfs_test_opt(root, NODATASUM)) 4184 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; 4185 if (btrfs_test_opt(root, NODATACOW)) 4186 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; 4187 } 4188 4189 insert_inode_hash(inode); 4190 inode_tree_add(inode); 4191 return inode; 4192 fail: 4193 if (dir) 4194 BTRFS_I(dir)->index_cnt--; 4195 btrfs_free_path(path); 4196 iput(inode); 4197 return ERR_PTR(ret); 4198 } 4199 4200 static inline u8 btrfs_inode_type(struct inode *inode) 4201 { 4202 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; 4203 } 4204 4205 /* 4206 * utility function to add 'inode' into 'parent_inode' with 4207 * a give name and a given sequence number. 4208 * if 'add_backref' is true, also insert a backref from the 4209 * inode to the parent directory. 4210 */ 4211 int btrfs_add_link(struct btrfs_trans_handle *trans, 4212 struct inode *parent_inode, struct inode *inode, 4213 const char *name, int name_len, int add_backref, u64 index) 4214 { 4215 int ret = 0; 4216 struct btrfs_key key; 4217 struct btrfs_root *root = BTRFS_I(parent_inode)->root; 4218 4219 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { 4220 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); 4221 } else { 4222 key.objectid = inode->i_ino; 4223 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); 4224 key.offset = 0; 4225 } 4226 4227 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { 4228 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 4229 key.objectid, root->root_key.objectid, 4230 parent_inode->i_ino, 4231 index, name, name_len); 4232 } else if (add_backref) { 4233 ret = btrfs_insert_inode_ref(trans, root, 4234 name, name_len, inode->i_ino, 4235 parent_inode->i_ino, index); 4236 } 4237 4238 if (ret == 0) { 4239 ret = btrfs_insert_dir_item(trans, root, name, name_len, 4240 parent_inode->i_ino, &key, 4241 btrfs_inode_type(inode), index); 4242 BUG_ON(ret); 4243 4244 btrfs_i_size_write(parent_inode, parent_inode->i_size + 4245 name_len * 2); 4246 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; 4247 ret = btrfs_update_inode(trans, root, parent_inode); 4248 } 4249 return ret; 4250 } 4251 4252 static int btrfs_add_nondir(struct btrfs_trans_handle *trans, 4253 struct dentry *dentry, struct inode *inode, 4254 int backref, u64 index) 4255 { 4256 int err = btrfs_add_link(trans, dentry->d_parent->d_inode, 4257 inode, dentry->d_name.name, 4258 dentry->d_name.len, backref, index); 4259 if (!err) { 4260 d_instantiate(dentry, inode); 4261 return 0; 4262 } 4263 if (err > 0) 4264 err = -EEXIST; 4265 return err; 4266 } 4267 4268 static int btrfs_mknod(struct inode *dir, struct dentry *dentry, 4269 int mode, dev_t rdev) 4270 { 4271 struct btrfs_trans_handle *trans; 4272 struct btrfs_root *root = BTRFS_I(dir)->root; 4273 struct inode *inode = NULL; 4274 int err; 4275 int drop_inode = 0; 4276 u64 objectid; 4277 unsigned long nr = 0; 4278 u64 index = 0; 4279 4280 if (!new_valid_dev(rdev)) 4281 return -EINVAL; 4282 4283 /* 4284 * 2 for inode item and ref 4285 * 2 for dir items 4286 * 1 for xattr if selinux is on 4287 */ 4288 err = btrfs_reserve_metadata_space(root, 5); 4289 if (err) 4290 return err; 4291 4292 trans = btrfs_start_transaction(root, 1); 4293 if (!trans) 4294 goto fail; 4295 btrfs_set_trans_block_group(trans, dir); 4296 4297 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4298 if (err) { 4299 err = -ENOSPC; 4300 goto out_unlock; 4301 } 4302 4303 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, 4304 dentry->d_name.len, 4305 dentry->d_parent->d_inode->i_ino, objectid, 4306 BTRFS_I(dir)->block_group, mode, &index); 4307 err = PTR_ERR(inode); 4308 if (IS_ERR(inode)) 4309 goto out_unlock; 4310 4311 err = btrfs_init_inode_security(trans, inode, dir); 4312 if (err) { 4313 drop_inode = 1; 4314 goto out_unlock; 4315 } 4316 4317 btrfs_set_trans_block_group(trans, inode); 4318 err = btrfs_add_nondir(trans, dentry, inode, 0, index); 4319 if (err) 4320 drop_inode = 1; 4321 else { 4322 inode->i_op = &btrfs_special_inode_operations; 4323 init_special_inode(inode, inode->i_mode, rdev); 4324 btrfs_update_inode(trans, root, inode); 4325 } 4326 btrfs_update_inode_block_group(trans, inode); 4327 btrfs_update_inode_block_group(trans, dir); 4328 out_unlock: 4329 nr = trans->blocks_used; 4330 btrfs_end_transaction_throttle(trans, root); 4331 fail: 4332 btrfs_unreserve_metadata_space(root, 5); 4333 if (drop_inode) { 4334 inode_dec_link_count(inode); 4335 iput(inode); 4336 } 4337 btrfs_btree_balance_dirty(root, nr); 4338 return err; 4339 } 4340 4341 static int btrfs_create(struct inode *dir, struct dentry *dentry, 4342 int mode, struct nameidata *nd) 4343 { 4344 struct btrfs_trans_handle *trans; 4345 struct btrfs_root *root = BTRFS_I(dir)->root; 4346 struct inode *inode = NULL; 4347 int err; 4348 int drop_inode = 0; 4349 unsigned long nr = 0; 4350 u64 objectid; 4351 u64 index = 0; 4352 4353 /* 4354 * 2 for inode item and ref 4355 * 2 for dir items 4356 * 1 for xattr if selinux is on 4357 */ 4358 err = btrfs_reserve_metadata_space(root, 5); 4359 if (err) 4360 return err; 4361 4362 trans = btrfs_start_transaction(root, 1); 4363 if (!trans) 4364 goto fail; 4365 btrfs_set_trans_block_group(trans, dir); 4366 4367 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4368 if (err) { 4369 err = -ENOSPC; 4370 goto out_unlock; 4371 } 4372 4373 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, 4374 dentry->d_name.len, 4375 dentry->d_parent->d_inode->i_ino, 4376 objectid, BTRFS_I(dir)->block_group, mode, 4377 &index); 4378 err = PTR_ERR(inode); 4379 if (IS_ERR(inode)) 4380 goto out_unlock; 4381 4382 err = btrfs_init_inode_security(trans, inode, dir); 4383 if (err) { 4384 drop_inode = 1; 4385 goto out_unlock; 4386 } 4387 4388 btrfs_set_trans_block_group(trans, inode); 4389 err = btrfs_add_nondir(trans, dentry, inode, 0, index); 4390 if (err) 4391 drop_inode = 1; 4392 else { 4393 inode->i_mapping->a_ops = &btrfs_aops; 4394 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 4395 inode->i_fop = &btrfs_file_operations; 4396 inode->i_op = &btrfs_file_inode_operations; 4397 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; 4398 } 4399 btrfs_update_inode_block_group(trans, inode); 4400 btrfs_update_inode_block_group(trans, dir); 4401 out_unlock: 4402 nr = trans->blocks_used; 4403 btrfs_end_transaction_throttle(trans, root); 4404 fail: 4405 btrfs_unreserve_metadata_space(root, 5); 4406 if (drop_inode) { 4407 inode_dec_link_count(inode); 4408 iput(inode); 4409 } 4410 btrfs_btree_balance_dirty(root, nr); 4411 return err; 4412 } 4413 4414 static int btrfs_link(struct dentry *old_dentry, struct inode *dir, 4415 struct dentry *dentry) 4416 { 4417 struct btrfs_trans_handle *trans; 4418 struct btrfs_root *root = BTRFS_I(dir)->root; 4419 struct inode *inode = old_dentry->d_inode; 4420 u64 index; 4421 unsigned long nr = 0; 4422 int err; 4423 int drop_inode = 0; 4424 4425 if (inode->i_nlink == 0) 4426 return -ENOENT; 4427 4428 /* do not allow sys_link's with other subvols of the same device */ 4429 if (root->objectid != BTRFS_I(inode)->root->objectid) 4430 return -EPERM; 4431 4432 /* 4433 * 1 item for inode ref 4434 * 2 items for dir items 4435 */ 4436 err = btrfs_reserve_metadata_space(root, 3); 4437 if (err) 4438 return err; 4439 4440 btrfs_inc_nlink(inode); 4441 4442 err = btrfs_set_inode_index(dir, &index); 4443 if (err) 4444 goto fail; 4445 4446 trans = btrfs_start_transaction(root, 1); 4447 4448 btrfs_set_trans_block_group(trans, dir); 4449 atomic_inc(&inode->i_count); 4450 4451 err = btrfs_add_nondir(trans, dentry, inode, 1, index); 4452 4453 if (err) { 4454 drop_inode = 1; 4455 } else { 4456 btrfs_update_inode_block_group(trans, dir); 4457 err = btrfs_update_inode(trans, root, inode); 4458 BUG_ON(err); 4459 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); 4460 } 4461 4462 nr = trans->blocks_used; 4463 btrfs_end_transaction_throttle(trans, root); 4464 fail: 4465 btrfs_unreserve_metadata_space(root, 3); 4466 if (drop_inode) { 4467 inode_dec_link_count(inode); 4468 iput(inode); 4469 } 4470 btrfs_btree_balance_dirty(root, nr); 4471 return err; 4472 } 4473 4474 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 4475 { 4476 struct inode *inode = NULL; 4477 struct btrfs_trans_handle *trans; 4478 struct btrfs_root *root = BTRFS_I(dir)->root; 4479 int err = 0; 4480 int drop_on_err = 0; 4481 u64 objectid = 0; 4482 u64 index = 0; 4483 unsigned long nr = 1; 4484 4485 /* 4486 * 2 items for inode and ref 4487 * 2 items for dir items 4488 * 1 for xattr if selinux is on 4489 */ 4490 err = btrfs_reserve_metadata_space(root, 5); 4491 if (err) 4492 return err; 4493 4494 trans = btrfs_start_transaction(root, 1); 4495 if (!trans) { 4496 err = -ENOMEM; 4497 goto out_unlock; 4498 } 4499 btrfs_set_trans_block_group(trans, dir); 4500 4501 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 4502 if (err) { 4503 err = -ENOSPC; 4504 goto out_unlock; 4505 } 4506 4507 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, 4508 dentry->d_name.len, 4509 dentry->d_parent->d_inode->i_ino, objectid, 4510 BTRFS_I(dir)->block_group, S_IFDIR | mode, 4511 &index); 4512 if (IS_ERR(inode)) { 4513 err = PTR_ERR(inode); 4514 goto out_fail; 4515 } 4516 4517 drop_on_err = 1; 4518 4519 err = btrfs_init_inode_security(trans, inode, dir); 4520 if (err) 4521 goto out_fail; 4522 4523 inode->i_op = &btrfs_dir_inode_operations; 4524 inode->i_fop = &btrfs_dir_file_operations; 4525 btrfs_set_trans_block_group(trans, inode); 4526 4527 btrfs_i_size_write(inode, 0); 4528 err = btrfs_update_inode(trans, root, inode); 4529 if (err) 4530 goto out_fail; 4531 4532 err = btrfs_add_link(trans, dentry->d_parent->d_inode, 4533 inode, dentry->d_name.name, 4534 dentry->d_name.len, 0, index); 4535 if (err) 4536 goto out_fail; 4537 4538 d_instantiate(dentry, inode); 4539 drop_on_err = 0; 4540 btrfs_update_inode_block_group(trans, inode); 4541 btrfs_update_inode_block_group(trans, dir); 4542 4543 out_fail: 4544 nr = trans->blocks_used; 4545 btrfs_end_transaction_throttle(trans, root); 4546 4547 out_unlock: 4548 btrfs_unreserve_metadata_space(root, 5); 4549 if (drop_on_err) 4550 iput(inode); 4551 btrfs_btree_balance_dirty(root, nr); 4552 return err; 4553 } 4554 4555 /* helper for btfs_get_extent. Given an existing extent in the tree, 4556 * and an extent that you want to insert, deal with overlap and insert 4557 * the new extent into the tree. 4558 */ 4559 static int merge_extent_mapping(struct extent_map_tree *em_tree, 4560 struct extent_map *existing, 4561 struct extent_map *em, 4562 u64 map_start, u64 map_len) 4563 { 4564 u64 start_diff; 4565 4566 BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); 4567 start_diff = map_start - em->start; 4568 em->start = map_start; 4569 em->len = map_len; 4570 if (em->block_start < EXTENT_MAP_LAST_BYTE && 4571 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { 4572 em->block_start += start_diff; 4573 em->block_len -= start_diff; 4574 } 4575 return add_extent_mapping(em_tree, em); 4576 } 4577 4578 static noinline int uncompress_inline(struct btrfs_path *path, 4579 struct inode *inode, struct page *page, 4580 size_t pg_offset, u64 extent_offset, 4581 struct btrfs_file_extent_item *item) 4582 { 4583 int ret; 4584 struct extent_buffer *leaf = path->nodes[0]; 4585 char *tmp; 4586 size_t max_size; 4587 unsigned long inline_size; 4588 unsigned long ptr; 4589 4590 WARN_ON(pg_offset != 0); 4591 max_size = btrfs_file_extent_ram_bytes(leaf, item); 4592 inline_size = btrfs_file_extent_inline_item_len(leaf, 4593 btrfs_item_nr(leaf, path->slots[0])); 4594 tmp = kmalloc(inline_size, GFP_NOFS); 4595 ptr = btrfs_file_extent_inline_start(item); 4596 4597 read_extent_buffer(leaf, tmp, ptr, inline_size); 4598 4599 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); 4600 ret = btrfs_zlib_decompress(tmp, page, extent_offset, 4601 inline_size, max_size); 4602 if (ret) { 4603 char *kaddr = kmap_atomic(page, KM_USER0); 4604 unsigned long copy_size = min_t(u64, 4605 PAGE_CACHE_SIZE - pg_offset, 4606 max_size - extent_offset); 4607 memset(kaddr + pg_offset, 0, copy_size); 4608 kunmap_atomic(kaddr, KM_USER0); 4609 } 4610 kfree(tmp); 4611 return 0; 4612 } 4613 4614 /* 4615 * a bit scary, this does extent mapping from logical file offset to the disk. 4616 * the ugly parts come from merging extents from the disk with the in-ram 4617 * representation. This gets more complex because of the data=ordered code, 4618 * where the in-ram extents might be locked pending data=ordered completion. 4619 * 4620 * This also copies inline extents directly into the page. 4621 */ 4622 4623 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, 4624 size_t pg_offset, u64 start, u64 len, 4625 int create) 4626 { 4627 int ret; 4628 int err = 0; 4629 u64 bytenr; 4630 u64 extent_start = 0; 4631 u64 extent_end = 0; 4632 u64 objectid = inode->i_ino; 4633 u32 found_type; 4634 struct btrfs_path *path = NULL; 4635 struct btrfs_root *root = BTRFS_I(inode)->root; 4636 struct btrfs_file_extent_item *item; 4637 struct extent_buffer *leaf; 4638 struct btrfs_key found_key; 4639 struct extent_map *em = NULL; 4640 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 4641 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 4642 struct btrfs_trans_handle *trans = NULL; 4643 int compressed; 4644 4645 again: 4646 read_lock(&em_tree->lock); 4647 em = lookup_extent_mapping(em_tree, start, len); 4648 if (em) 4649 em->bdev = root->fs_info->fs_devices->latest_bdev; 4650 read_unlock(&em_tree->lock); 4651 4652 if (em) { 4653 if (em->start > start || em->start + em->len <= start) 4654 free_extent_map(em); 4655 else if (em->block_start == EXTENT_MAP_INLINE && page) 4656 free_extent_map(em); 4657 else 4658 goto out; 4659 } 4660 em = alloc_extent_map(GFP_NOFS); 4661 if (!em) { 4662 err = -ENOMEM; 4663 goto out; 4664 } 4665 em->bdev = root->fs_info->fs_devices->latest_bdev; 4666 em->start = EXTENT_MAP_HOLE; 4667 em->orig_start = EXTENT_MAP_HOLE; 4668 em->len = (u64)-1; 4669 em->block_len = (u64)-1; 4670 4671 if (!path) { 4672 path = btrfs_alloc_path(); 4673 BUG_ON(!path); 4674 } 4675 4676 ret = btrfs_lookup_file_extent(trans, root, path, 4677 objectid, start, trans != NULL); 4678 if (ret < 0) { 4679 err = ret; 4680 goto out; 4681 } 4682 4683 if (ret != 0) { 4684 if (path->slots[0] == 0) 4685 goto not_found; 4686 path->slots[0]--; 4687 } 4688 4689 leaf = path->nodes[0]; 4690 item = btrfs_item_ptr(leaf, path->slots[0], 4691 struct btrfs_file_extent_item); 4692 /* are we inside the extent that was found? */ 4693 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4694 found_type = btrfs_key_type(&found_key); 4695 if (found_key.objectid != objectid || 4696 found_type != BTRFS_EXTENT_DATA_KEY) { 4697 goto not_found; 4698 } 4699 4700 found_type = btrfs_file_extent_type(leaf, item); 4701 extent_start = found_key.offset; 4702 compressed = btrfs_file_extent_compression(leaf, item); 4703 if (found_type == BTRFS_FILE_EXTENT_REG || 4704 found_type == BTRFS_FILE_EXTENT_PREALLOC) { 4705 extent_end = extent_start + 4706 btrfs_file_extent_num_bytes(leaf, item); 4707 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { 4708 size_t size; 4709 size = btrfs_file_extent_inline_len(leaf, item); 4710 extent_end = (extent_start + size + root->sectorsize - 1) & 4711 ~((u64)root->sectorsize - 1); 4712 } 4713 4714 if (start >= extent_end) { 4715 path->slots[0]++; 4716 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 4717 ret = btrfs_next_leaf(root, path); 4718 if (ret < 0) { 4719 err = ret; 4720 goto out; 4721 } 4722 if (ret > 0) 4723 goto not_found; 4724 leaf = path->nodes[0]; 4725 } 4726 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 4727 if (found_key.objectid != objectid || 4728 found_key.type != BTRFS_EXTENT_DATA_KEY) 4729 goto not_found; 4730 if (start + len <= found_key.offset) 4731 goto not_found; 4732 em->start = start; 4733 em->len = found_key.offset - start; 4734 goto not_found_em; 4735 } 4736 4737 if (found_type == BTRFS_FILE_EXTENT_REG || 4738 found_type == BTRFS_FILE_EXTENT_PREALLOC) { 4739 em->start = extent_start; 4740 em->len = extent_end - extent_start; 4741 em->orig_start = extent_start - 4742 btrfs_file_extent_offset(leaf, item); 4743 bytenr = btrfs_file_extent_disk_bytenr(leaf, item); 4744 if (bytenr == 0) { 4745 em->block_start = EXTENT_MAP_HOLE; 4746 goto insert; 4747 } 4748 if (compressed) { 4749 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 4750 em->block_start = bytenr; 4751 em->block_len = btrfs_file_extent_disk_num_bytes(leaf, 4752 item); 4753 } else { 4754 bytenr += btrfs_file_extent_offset(leaf, item); 4755 em->block_start = bytenr; 4756 em->block_len = em->len; 4757 if (found_type == BTRFS_FILE_EXTENT_PREALLOC) 4758 set_bit(EXTENT_FLAG_PREALLOC, &em->flags); 4759 } 4760 goto insert; 4761 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { 4762 unsigned long ptr; 4763 char *map; 4764 size_t size; 4765 size_t extent_offset; 4766 size_t copy_size; 4767 4768 em->block_start = EXTENT_MAP_INLINE; 4769 if (!page || create) { 4770 em->start = extent_start; 4771 em->len = extent_end - extent_start; 4772 goto out; 4773 } 4774 4775 size = btrfs_file_extent_inline_len(leaf, item); 4776 extent_offset = page_offset(page) + pg_offset - extent_start; 4777 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, 4778 size - extent_offset); 4779 em->start = extent_start + extent_offset; 4780 em->len = (copy_size + root->sectorsize - 1) & 4781 ~((u64)root->sectorsize - 1); 4782 em->orig_start = EXTENT_MAP_INLINE; 4783 if (compressed) 4784 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 4785 ptr = btrfs_file_extent_inline_start(item) + extent_offset; 4786 if (create == 0 && !PageUptodate(page)) { 4787 if (btrfs_file_extent_compression(leaf, item) == 4788 BTRFS_COMPRESS_ZLIB) { 4789 ret = uncompress_inline(path, inode, page, 4790 pg_offset, 4791 extent_offset, item); 4792 BUG_ON(ret); 4793 } else { 4794 map = kmap(page); 4795 read_extent_buffer(leaf, map + pg_offset, ptr, 4796 copy_size); 4797 if (pg_offset + copy_size < PAGE_CACHE_SIZE) { 4798 memset(map + pg_offset + copy_size, 0, 4799 PAGE_CACHE_SIZE - pg_offset - 4800 copy_size); 4801 } 4802 kunmap(page); 4803 } 4804 flush_dcache_page(page); 4805 } else if (create && PageUptodate(page)) { 4806 if (!trans) { 4807 kunmap(page); 4808 free_extent_map(em); 4809 em = NULL; 4810 btrfs_release_path(root, path); 4811 trans = btrfs_join_transaction(root, 1); 4812 goto again; 4813 } 4814 map = kmap(page); 4815 write_extent_buffer(leaf, map + pg_offset, ptr, 4816 copy_size); 4817 kunmap(page); 4818 btrfs_mark_buffer_dirty(leaf); 4819 } 4820 set_extent_uptodate(io_tree, em->start, 4821 extent_map_end(em) - 1, GFP_NOFS); 4822 goto insert; 4823 } else { 4824 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); 4825 WARN_ON(1); 4826 } 4827 not_found: 4828 em->start = start; 4829 em->len = len; 4830 not_found_em: 4831 em->block_start = EXTENT_MAP_HOLE; 4832 set_bit(EXTENT_FLAG_VACANCY, &em->flags); 4833 insert: 4834 btrfs_release_path(root, path); 4835 if (em->start > start || extent_map_end(em) <= start) { 4836 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " 4837 "[%llu %llu]\n", (unsigned long long)em->start, 4838 (unsigned long long)em->len, 4839 (unsigned long long)start, 4840 (unsigned long long)len); 4841 err = -EIO; 4842 goto out; 4843 } 4844 4845 err = 0; 4846 write_lock(&em_tree->lock); 4847 ret = add_extent_mapping(em_tree, em); 4848 /* it is possible that someone inserted the extent into the tree 4849 * while we had the lock dropped. It is also possible that 4850 * an overlapping map exists in the tree 4851 */ 4852 if (ret == -EEXIST) { 4853 struct extent_map *existing; 4854 4855 ret = 0; 4856 4857 existing = lookup_extent_mapping(em_tree, start, len); 4858 if (existing && (existing->start > start || 4859 existing->start + existing->len <= start)) { 4860 free_extent_map(existing); 4861 existing = NULL; 4862 } 4863 if (!existing) { 4864 existing = lookup_extent_mapping(em_tree, em->start, 4865 em->len); 4866 if (existing) { 4867 err = merge_extent_mapping(em_tree, existing, 4868 em, start, 4869 root->sectorsize); 4870 free_extent_map(existing); 4871 if (err) { 4872 free_extent_map(em); 4873 em = NULL; 4874 } 4875 } else { 4876 err = -EIO; 4877 free_extent_map(em); 4878 em = NULL; 4879 } 4880 } else { 4881 free_extent_map(em); 4882 em = existing; 4883 err = 0; 4884 } 4885 } 4886 write_unlock(&em_tree->lock); 4887 out: 4888 if (path) 4889 btrfs_free_path(path); 4890 if (trans) { 4891 ret = btrfs_end_transaction(trans, root); 4892 if (!err) 4893 err = ret; 4894 } 4895 if (err) { 4896 free_extent_map(em); 4897 return ERR_PTR(err); 4898 } 4899 return em; 4900 } 4901 4902 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, 4903 const struct iovec *iov, loff_t offset, 4904 unsigned long nr_segs) 4905 { 4906 return -EINVAL; 4907 } 4908 4909 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 4910 __u64 start, __u64 len) 4911 { 4912 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); 4913 } 4914 4915 int btrfs_readpage(struct file *file, struct page *page) 4916 { 4917 struct extent_io_tree *tree; 4918 tree = &BTRFS_I(page->mapping->host)->io_tree; 4919 return extent_read_full_page(tree, page, btrfs_get_extent); 4920 } 4921 4922 static int btrfs_writepage(struct page *page, struct writeback_control *wbc) 4923 { 4924 struct extent_io_tree *tree; 4925 4926 4927 if (current->flags & PF_MEMALLOC) { 4928 redirty_page_for_writepage(wbc, page); 4929 unlock_page(page); 4930 return 0; 4931 } 4932 tree = &BTRFS_I(page->mapping->host)->io_tree; 4933 return extent_write_full_page(tree, page, btrfs_get_extent, wbc); 4934 } 4935 4936 int btrfs_writepages(struct address_space *mapping, 4937 struct writeback_control *wbc) 4938 { 4939 struct extent_io_tree *tree; 4940 4941 tree = &BTRFS_I(mapping->host)->io_tree; 4942 return extent_writepages(tree, mapping, btrfs_get_extent, wbc); 4943 } 4944 4945 static int 4946 btrfs_readpages(struct file *file, struct address_space *mapping, 4947 struct list_head *pages, unsigned nr_pages) 4948 { 4949 struct extent_io_tree *tree; 4950 tree = &BTRFS_I(mapping->host)->io_tree; 4951 return extent_readpages(tree, mapping, pages, nr_pages, 4952 btrfs_get_extent); 4953 } 4954 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) 4955 { 4956 struct extent_io_tree *tree; 4957 struct extent_map_tree *map; 4958 int ret; 4959 4960 tree = &BTRFS_I(page->mapping->host)->io_tree; 4961 map = &BTRFS_I(page->mapping->host)->extent_tree; 4962 ret = try_release_extent_mapping(map, tree, page, gfp_flags); 4963 if (ret == 1) { 4964 ClearPagePrivate(page); 4965 set_page_private(page, 0); 4966 page_cache_release(page); 4967 } 4968 return ret; 4969 } 4970 4971 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) 4972 { 4973 if (PageWriteback(page) || PageDirty(page)) 4974 return 0; 4975 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); 4976 } 4977 4978 static void btrfs_invalidatepage(struct page *page, unsigned long offset) 4979 { 4980 struct extent_io_tree *tree; 4981 struct btrfs_ordered_extent *ordered; 4982 u64 page_start = page_offset(page); 4983 u64 page_end = page_start + PAGE_CACHE_SIZE - 1; 4984 4985 4986 /* 4987 * we have the page locked, so new writeback can't start, 4988 * and the dirty bit won't be cleared while we are here. 4989 * 4990 * Wait for IO on this page so that we can safely clear 4991 * the PagePrivate2 bit and do ordered accounting 4992 */ 4993 wait_on_page_writeback(page); 4994 4995 tree = &BTRFS_I(page->mapping->host)->io_tree; 4996 if (offset) { 4997 btrfs_releasepage(page, GFP_NOFS); 4998 return; 4999 } 5000 lock_extent(tree, page_start, page_end, GFP_NOFS); 5001 ordered = btrfs_lookup_ordered_extent(page->mapping->host, 5002 page_offset(page)); 5003 if (ordered) { 5004 /* 5005 * IO on this page will never be started, so we need 5006 * to account for any ordered extents now 5007 */ 5008 clear_extent_bit(tree, page_start, page_end, 5009 EXTENT_DIRTY | EXTENT_DELALLOC | 5010 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, 5011 NULL, GFP_NOFS); 5012 /* 5013 * whoever cleared the private bit is responsible 5014 * for the finish_ordered_io 5015 */ 5016 if (TestClearPagePrivate2(page)) { 5017 btrfs_finish_ordered_io(page->mapping->host, 5018 page_start, page_end); 5019 } 5020 btrfs_put_ordered_extent(ordered); 5021 lock_extent(tree, page_start, page_end, GFP_NOFS); 5022 } 5023 clear_extent_bit(tree, page_start, page_end, 5024 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | 5025 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); 5026 __btrfs_releasepage(page, GFP_NOFS); 5027 5028 ClearPageChecked(page); 5029 if (PagePrivate(page)) { 5030 ClearPagePrivate(page); 5031 set_page_private(page, 0); 5032 page_cache_release(page); 5033 } 5034 } 5035 5036 /* 5037 * btrfs_page_mkwrite() is not allowed to change the file size as it gets 5038 * called from a page fault handler when a page is first dirtied. Hence we must 5039 * be careful to check for EOF conditions here. We set the page up correctly 5040 * for a written page which means we get ENOSPC checking when writing into 5041 * holes and correct delalloc and unwritten extent mapping on filesystems that 5042 * support these features. 5043 * 5044 * We are not allowed to take the i_mutex here so we have to play games to 5045 * protect against truncate races as the page could now be beyond EOF. Because 5046 * vmtruncate() writes the inode size before removing pages, once we have the 5047 * page lock we can determine safely if the page is beyond EOF. If it is not 5048 * beyond EOF, then the page is guaranteed safe against truncation until we 5049 * unlock the page. 5050 */ 5051 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 5052 { 5053 struct page *page = vmf->page; 5054 struct inode *inode = fdentry(vma->vm_file)->d_inode; 5055 struct btrfs_root *root = BTRFS_I(inode)->root; 5056 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 5057 struct btrfs_ordered_extent *ordered; 5058 char *kaddr; 5059 unsigned long zero_start; 5060 loff_t size; 5061 int ret; 5062 u64 page_start; 5063 u64 page_end; 5064 5065 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); 5066 if (ret) { 5067 if (ret == -ENOMEM) 5068 ret = VM_FAULT_OOM; 5069 else /* -ENOSPC, -EIO, etc */ 5070 ret = VM_FAULT_SIGBUS; 5071 goto out; 5072 } 5073 5074 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); 5075 if (ret) { 5076 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); 5077 ret = VM_FAULT_SIGBUS; 5078 goto out; 5079 } 5080 5081 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ 5082 again: 5083 lock_page(page); 5084 size = i_size_read(inode); 5085 page_start = page_offset(page); 5086 page_end = page_start + PAGE_CACHE_SIZE - 1; 5087 5088 if ((page->mapping != inode->i_mapping) || 5089 (page_start >= size)) { 5090 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); 5091 /* page got truncated out from underneath us */ 5092 goto out_unlock; 5093 } 5094 wait_on_page_writeback(page); 5095 5096 lock_extent(io_tree, page_start, page_end, GFP_NOFS); 5097 set_page_extent_mapped(page); 5098 5099 /* 5100 * we can't set the delalloc bits if there are pending ordered 5101 * extents. Drop our locks and wait for them to finish 5102 */ 5103 ordered = btrfs_lookup_ordered_extent(inode, page_start); 5104 if (ordered) { 5105 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 5106 unlock_page(page); 5107 btrfs_start_ordered_extent(inode, ordered, 1); 5108 btrfs_put_ordered_extent(ordered); 5109 goto again; 5110 } 5111 5112 /* 5113 * XXX - page_mkwrite gets called every time the page is dirtied, even 5114 * if it was already dirty, so for space accounting reasons we need to 5115 * clear any delalloc bits for the range we are fixing to save. There 5116 * is probably a better way to do this, but for now keep consistent with 5117 * prepare_pages in the normal write path. 5118 */ 5119 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 5120 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, 5121 GFP_NOFS); 5122 5123 ret = btrfs_set_extent_delalloc(inode, page_start, page_end); 5124 if (ret) { 5125 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 5126 ret = VM_FAULT_SIGBUS; 5127 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); 5128 goto out_unlock; 5129 } 5130 ret = 0; 5131 5132 /* page is wholly or partially inside EOF */ 5133 if (page_start + PAGE_CACHE_SIZE > size) 5134 zero_start = size & ~PAGE_CACHE_MASK; 5135 else 5136 zero_start = PAGE_CACHE_SIZE; 5137 5138 if (zero_start != PAGE_CACHE_SIZE) { 5139 kaddr = kmap(page); 5140 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); 5141 flush_dcache_page(page); 5142 kunmap(page); 5143 } 5144 ClearPageChecked(page); 5145 set_page_dirty(page); 5146 SetPageUptodate(page); 5147 5148 BTRFS_I(inode)->last_trans = root->fs_info->generation; 5149 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; 5150 5151 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 5152 5153 out_unlock: 5154 btrfs_unreserve_metadata_for_delalloc(root, inode, 1); 5155 if (!ret) 5156 return VM_FAULT_LOCKED; 5157 unlock_page(page); 5158 out: 5159 return ret; 5160 } 5161 5162 static void btrfs_truncate(struct inode *inode) 5163 { 5164 struct btrfs_root *root = BTRFS_I(inode)->root; 5165 int ret; 5166 struct btrfs_trans_handle *trans; 5167 unsigned long nr; 5168 u64 mask = root->sectorsize - 1; 5169 5170 if (!S_ISREG(inode->i_mode)) { 5171 WARN_ON(1); 5172 return; 5173 } 5174 5175 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); 5176 if (ret) 5177 return; 5178 5179 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); 5180 btrfs_ordered_update_i_size(inode, inode->i_size, NULL); 5181 5182 trans = btrfs_start_transaction(root, 1); 5183 btrfs_set_trans_block_group(trans, inode); 5184 5185 /* 5186 * setattr is responsible for setting the ordered_data_close flag, 5187 * but that is only tested during the last file release. That 5188 * could happen well after the next commit, leaving a great big 5189 * window where new writes may get lost if someone chooses to write 5190 * to this file after truncating to zero 5191 * 5192 * The inode doesn't have any dirty data here, and so if we commit 5193 * this is a noop. If someone immediately starts writing to the inode 5194 * it is very likely we'll catch some of their writes in this 5195 * transaction, and the commit will find this file on the ordered 5196 * data list with good things to send down. 5197 * 5198 * This is a best effort solution, there is still a window where 5199 * using truncate to replace the contents of the file will 5200 * end up with a zero length file after a crash. 5201 */ 5202 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) 5203 btrfs_add_ordered_operation(trans, root, inode); 5204 5205 while (1) { 5206 ret = btrfs_truncate_inode_items(trans, root, inode, 5207 inode->i_size, 5208 BTRFS_EXTENT_DATA_KEY); 5209 if (ret != -EAGAIN) 5210 break; 5211 5212 ret = btrfs_update_inode(trans, root, inode); 5213 BUG_ON(ret); 5214 5215 nr = trans->blocks_used; 5216 btrfs_end_transaction(trans, root); 5217 btrfs_btree_balance_dirty(root, nr); 5218 5219 trans = btrfs_start_transaction(root, 1); 5220 btrfs_set_trans_block_group(trans, inode); 5221 } 5222 5223 if (ret == 0 && inode->i_nlink > 0) { 5224 ret = btrfs_orphan_del(trans, inode); 5225 BUG_ON(ret); 5226 } 5227 5228 ret = btrfs_update_inode(trans, root, inode); 5229 BUG_ON(ret); 5230 5231 nr = trans->blocks_used; 5232 ret = btrfs_end_transaction_throttle(trans, root); 5233 BUG_ON(ret); 5234 btrfs_btree_balance_dirty(root, nr); 5235 } 5236 5237 /* 5238 * create a new subvolume directory/inode (helper for the ioctl). 5239 */ 5240 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, 5241 struct btrfs_root *new_root, 5242 u64 new_dirid, u64 alloc_hint) 5243 { 5244 struct inode *inode; 5245 int err; 5246 u64 index = 0; 5247 5248 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, 5249 new_dirid, alloc_hint, S_IFDIR | 0700, &index); 5250 if (IS_ERR(inode)) 5251 return PTR_ERR(inode); 5252 inode->i_op = &btrfs_dir_inode_operations; 5253 inode->i_fop = &btrfs_dir_file_operations; 5254 5255 inode->i_nlink = 1; 5256 btrfs_i_size_write(inode, 0); 5257 5258 err = btrfs_update_inode(trans, new_root, inode); 5259 BUG_ON(err); 5260 5261 iput(inode); 5262 return 0; 5263 } 5264 5265 /* helper function for file defrag and space balancing. This 5266 * forces readahead on a given range of bytes in an inode 5267 */ 5268 unsigned long btrfs_force_ra(struct address_space *mapping, 5269 struct file_ra_state *ra, struct file *file, 5270 pgoff_t offset, pgoff_t last_index) 5271 { 5272 pgoff_t req_size = last_index - offset + 1; 5273 5274 page_cache_sync_readahead(mapping, ra, file, offset, req_size); 5275 return offset + req_size; 5276 } 5277 5278 struct inode *btrfs_alloc_inode(struct super_block *sb) 5279 { 5280 struct btrfs_inode *ei; 5281 5282 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); 5283 if (!ei) 5284 return NULL; 5285 ei->last_trans = 0; 5286 ei->last_sub_trans = 0; 5287 ei->logged_trans = 0; 5288 ei->outstanding_extents = 0; 5289 ei->reserved_extents = 0; 5290 ei->root = NULL; 5291 spin_lock_init(&ei->accounting_lock); 5292 btrfs_ordered_inode_tree_init(&ei->ordered_tree); 5293 INIT_LIST_HEAD(&ei->i_orphan); 5294 INIT_LIST_HEAD(&ei->ordered_operations); 5295 return &ei->vfs_inode; 5296 } 5297 5298 void btrfs_destroy_inode(struct inode *inode) 5299 { 5300 struct btrfs_ordered_extent *ordered; 5301 struct btrfs_root *root = BTRFS_I(inode)->root; 5302 5303 WARN_ON(!list_empty(&inode->i_dentry)); 5304 WARN_ON(inode->i_data.nrpages); 5305 5306 /* 5307 * This can happen where we create an inode, but somebody else also 5308 * created the same inode and we need to destroy the one we already 5309 * created. 5310 */ 5311 if (!root) 5312 goto free; 5313 5314 /* 5315 * Make sure we're properly removed from the ordered operation 5316 * lists. 5317 */ 5318 smp_mb(); 5319 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { 5320 spin_lock(&root->fs_info->ordered_extent_lock); 5321 list_del_init(&BTRFS_I(inode)->ordered_operations); 5322 spin_unlock(&root->fs_info->ordered_extent_lock); 5323 } 5324 5325 spin_lock(&root->list_lock); 5326 if (!list_empty(&BTRFS_I(inode)->i_orphan)) { 5327 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", 5328 inode->i_ino); 5329 list_del_init(&BTRFS_I(inode)->i_orphan); 5330 } 5331 spin_unlock(&root->list_lock); 5332 5333 while (1) { 5334 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); 5335 if (!ordered) 5336 break; 5337 else { 5338 printk(KERN_ERR "btrfs found ordered " 5339 "extent %llu %llu on inode cleanup\n", 5340 (unsigned long long)ordered->file_offset, 5341 (unsigned long long)ordered->len); 5342 btrfs_remove_ordered_extent(inode, ordered); 5343 btrfs_put_ordered_extent(ordered); 5344 btrfs_put_ordered_extent(ordered); 5345 } 5346 } 5347 inode_tree_del(inode); 5348 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); 5349 free: 5350 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); 5351 } 5352 5353 void btrfs_drop_inode(struct inode *inode) 5354 { 5355 struct btrfs_root *root = BTRFS_I(inode)->root; 5356 5357 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) 5358 generic_delete_inode(inode); 5359 else 5360 generic_drop_inode(inode); 5361 } 5362 5363 static void init_once(void *foo) 5364 { 5365 struct btrfs_inode *ei = (struct btrfs_inode *) foo; 5366 5367 inode_init_once(&ei->vfs_inode); 5368 } 5369 5370 void btrfs_destroy_cachep(void) 5371 { 5372 if (btrfs_inode_cachep) 5373 kmem_cache_destroy(btrfs_inode_cachep); 5374 if (btrfs_trans_handle_cachep) 5375 kmem_cache_destroy(btrfs_trans_handle_cachep); 5376 if (btrfs_transaction_cachep) 5377 kmem_cache_destroy(btrfs_transaction_cachep); 5378 if (btrfs_path_cachep) 5379 kmem_cache_destroy(btrfs_path_cachep); 5380 } 5381 5382 int btrfs_init_cachep(void) 5383 { 5384 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", 5385 sizeof(struct btrfs_inode), 0, 5386 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); 5387 if (!btrfs_inode_cachep) 5388 goto fail; 5389 5390 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", 5391 sizeof(struct btrfs_trans_handle), 0, 5392 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); 5393 if (!btrfs_trans_handle_cachep) 5394 goto fail; 5395 5396 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", 5397 sizeof(struct btrfs_transaction), 0, 5398 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); 5399 if (!btrfs_transaction_cachep) 5400 goto fail; 5401 5402 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", 5403 sizeof(struct btrfs_path), 0, 5404 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); 5405 if (!btrfs_path_cachep) 5406 goto fail; 5407 5408 return 0; 5409 fail: 5410 btrfs_destroy_cachep(); 5411 return -ENOMEM; 5412 } 5413 5414 static int btrfs_getattr(struct vfsmount *mnt, 5415 struct dentry *dentry, struct kstat *stat) 5416 { 5417 struct inode *inode = dentry->d_inode; 5418 generic_fillattr(inode, stat); 5419 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; 5420 stat->blksize = PAGE_CACHE_SIZE; 5421 stat->blocks = (inode_get_bytes(inode) + 5422 BTRFS_I(inode)->delalloc_bytes) >> 9; 5423 return 0; 5424 } 5425 5426 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, 5427 struct inode *new_dir, struct dentry *new_dentry) 5428 { 5429 struct btrfs_trans_handle *trans; 5430 struct btrfs_root *root = BTRFS_I(old_dir)->root; 5431 struct btrfs_root *dest = BTRFS_I(new_dir)->root; 5432 struct inode *new_inode = new_dentry->d_inode; 5433 struct inode *old_inode = old_dentry->d_inode; 5434 struct timespec ctime = CURRENT_TIME; 5435 u64 index = 0; 5436 u64 root_objectid; 5437 int ret; 5438 5439 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) 5440 return -EPERM; 5441 5442 /* we only allow rename subvolume link between subvolumes */ 5443 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) 5444 return -EXDEV; 5445 5446 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || 5447 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) 5448 return -ENOTEMPTY; 5449 5450 if (S_ISDIR(old_inode->i_mode) && new_inode && 5451 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) 5452 return -ENOTEMPTY; 5453 5454 /* 5455 * We want to reserve the absolute worst case amount of items. So if 5456 * both inodes are subvols and we need to unlink them then that would 5457 * require 4 item modifications, but if they are both normal inodes it 5458 * would require 5 item modifications, so we'll assume their normal 5459 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items 5460 * should cover the worst case number of items we'll modify. 5461 */ 5462 ret = btrfs_reserve_metadata_space(root, 11); 5463 if (ret) 5464 return ret; 5465 5466 /* 5467 * we're using rename to replace one file with another. 5468 * and the replacement file is large. Start IO on it now so 5469 * we don't add too much work to the end of the transaction 5470 */ 5471 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && 5472 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) 5473 filemap_flush(old_inode->i_mapping); 5474 5475 /* close the racy window with snapshot create/destroy ioctl */ 5476 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 5477 down_read(&root->fs_info->subvol_sem); 5478 5479 trans = btrfs_start_transaction(root, 1); 5480 btrfs_set_trans_block_group(trans, new_dir); 5481 5482 if (dest != root) 5483 btrfs_record_root_in_trans(trans, dest); 5484 5485 ret = btrfs_set_inode_index(new_dir, &index); 5486 if (ret) 5487 goto out_fail; 5488 5489 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { 5490 /* force full log commit if subvolume involved. */ 5491 root->fs_info->last_trans_log_full_commit = trans->transid; 5492 } else { 5493 ret = btrfs_insert_inode_ref(trans, dest, 5494 new_dentry->d_name.name, 5495 new_dentry->d_name.len, 5496 old_inode->i_ino, 5497 new_dir->i_ino, index); 5498 if (ret) 5499 goto out_fail; 5500 /* 5501 * this is an ugly little race, but the rename is required 5502 * to make sure that if we crash, the inode is either at the 5503 * old name or the new one. pinning the log transaction lets 5504 * us make sure we don't allow a log commit to come in after 5505 * we unlink the name but before we add the new name back in. 5506 */ 5507 btrfs_pin_log_trans(root); 5508 } 5509 /* 5510 * make sure the inode gets flushed if it is replacing 5511 * something. 5512 */ 5513 if (new_inode && new_inode->i_size && 5514 old_inode && S_ISREG(old_inode->i_mode)) { 5515 btrfs_add_ordered_operation(trans, root, old_inode); 5516 } 5517 5518 old_dir->i_ctime = old_dir->i_mtime = ctime; 5519 new_dir->i_ctime = new_dir->i_mtime = ctime; 5520 old_inode->i_ctime = ctime; 5521 5522 if (old_dentry->d_parent != new_dentry->d_parent) 5523 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); 5524 5525 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { 5526 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; 5527 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, 5528 old_dentry->d_name.name, 5529 old_dentry->d_name.len); 5530 } else { 5531 btrfs_inc_nlink(old_dentry->d_inode); 5532 ret = btrfs_unlink_inode(trans, root, old_dir, 5533 old_dentry->d_inode, 5534 old_dentry->d_name.name, 5535 old_dentry->d_name.len); 5536 } 5537 BUG_ON(ret); 5538 5539 if (new_inode) { 5540 new_inode->i_ctime = CURRENT_TIME; 5541 if (unlikely(new_inode->i_ino == 5542 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { 5543 root_objectid = BTRFS_I(new_inode)->location.objectid; 5544 ret = btrfs_unlink_subvol(trans, dest, new_dir, 5545 root_objectid, 5546 new_dentry->d_name.name, 5547 new_dentry->d_name.len); 5548 BUG_ON(new_inode->i_nlink == 0); 5549 } else { 5550 ret = btrfs_unlink_inode(trans, dest, new_dir, 5551 new_dentry->d_inode, 5552 new_dentry->d_name.name, 5553 new_dentry->d_name.len); 5554 } 5555 BUG_ON(ret); 5556 if (new_inode->i_nlink == 0) { 5557 ret = btrfs_orphan_add(trans, new_dentry->d_inode); 5558 BUG_ON(ret); 5559 } 5560 } 5561 5562 ret = btrfs_add_link(trans, new_dir, old_inode, 5563 new_dentry->d_name.name, 5564 new_dentry->d_name.len, 0, index); 5565 BUG_ON(ret); 5566 5567 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { 5568 btrfs_log_new_name(trans, old_inode, old_dir, 5569 new_dentry->d_parent); 5570 btrfs_end_log_trans(root); 5571 } 5572 out_fail: 5573 btrfs_end_transaction_throttle(trans, root); 5574 5575 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) 5576 up_read(&root->fs_info->subvol_sem); 5577 5578 btrfs_unreserve_metadata_space(root, 11); 5579 return ret; 5580 } 5581 5582 /* 5583 * some fairly slow code that needs optimization. This walks the list 5584 * of all the inodes with pending delalloc and forces them to disk. 5585 */ 5586 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) 5587 { 5588 struct list_head *head = &root->fs_info->delalloc_inodes; 5589 struct btrfs_inode *binode; 5590 struct inode *inode; 5591 5592 if (root->fs_info->sb->s_flags & MS_RDONLY) 5593 return -EROFS; 5594 5595 spin_lock(&root->fs_info->delalloc_lock); 5596 while (!list_empty(head)) { 5597 binode = list_entry(head->next, struct btrfs_inode, 5598 delalloc_inodes); 5599 inode = igrab(&binode->vfs_inode); 5600 if (!inode) 5601 list_del_init(&binode->delalloc_inodes); 5602 spin_unlock(&root->fs_info->delalloc_lock); 5603 if (inode) { 5604 filemap_flush(inode->i_mapping); 5605 if (delay_iput) 5606 btrfs_add_delayed_iput(inode); 5607 else 5608 iput(inode); 5609 } 5610 cond_resched(); 5611 spin_lock(&root->fs_info->delalloc_lock); 5612 } 5613 spin_unlock(&root->fs_info->delalloc_lock); 5614 5615 /* the filemap_flush will queue IO into the worker threads, but 5616 * we have to make sure the IO is actually started and that 5617 * ordered extents get created before we return 5618 */ 5619 atomic_inc(&root->fs_info->async_submit_draining); 5620 while (atomic_read(&root->fs_info->nr_async_submits) || 5621 atomic_read(&root->fs_info->async_delalloc_pages)) { 5622 wait_event(root->fs_info->async_submit_wait, 5623 (atomic_read(&root->fs_info->nr_async_submits) == 0 && 5624 atomic_read(&root->fs_info->async_delalloc_pages) == 0)); 5625 } 5626 atomic_dec(&root->fs_info->async_submit_draining); 5627 return 0; 5628 } 5629 5630 static int btrfs_symlink(struct inode *dir, struct dentry *dentry, 5631 const char *symname) 5632 { 5633 struct btrfs_trans_handle *trans; 5634 struct btrfs_root *root = BTRFS_I(dir)->root; 5635 struct btrfs_path *path; 5636 struct btrfs_key key; 5637 struct inode *inode = NULL; 5638 int err; 5639 int drop_inode = 0; 5640 u64 objectid; 5641 u64 index = 0 ; 5642 int name_len; 5643 int datasize; 5644 unsigned long ptr; 5645 struct btrfs_file_extent_item *ei; 5646 struct extent_buffer *leaf; 5647 unsigned long nr = 0; 5648 5649 name_len = strlen(symname) + 1; 5650 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) 5651 return -ENAMETOOLONG; 5652 5653 /* 5654 * 2 items for inode item and ref 5655 * 2 items for dir items 5656 * 1 item for xattr if selinux is on 5657 */ 5658 err = btrfs_reserve_metadata_space(root, 5); 5659 if (err) 5660 return err; 5661 5662 trans = btrfs_start_transaction(root, 1); 5663 if (!trans) 5664 goto out_fail; 5665 btrfs_set_trans_block_group(trans, dir); 5666 5667 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); 5668 if (err) { 5669 err = -ENOSPC; 5670 goto out_unlock; 5671 } 5672 5673 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, 5674 dentry->d_name.len, 5675 dentry->d_parent->d_inode->i_ino, objectid, 5676 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, 5677 &index); 5678 err = PTR_ERR(inode); 5679 if (IS_ERR(inode)) 5680 goto out_unlock; 5681 5682 err = btrfs_init_inode_security(trans, inode, dir); 5683 if (err) { 5684 drop_inode = 1; 5685 goto out_unlock; 5686 } 5687 5688 btrfs_set_trans_block_group(trans, inode); 5689 err = btrfs_add_nondir(trans, dentry, inode, 0, index); 5690 if (err) 5691 drop_inode = 1; 5692 else { 5693 inode->i_mapping->a_ops = &btrfs_aops; 5694 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 5695 inode->i_fop = &btrfs_file_operations; 5696 inode->i_op = &btrfs_file_inode_operations; 5697 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; 5698 } 5699 btrfs_update_inode_block_group(trans, inode); 5700 btrfs_update_inode_block_group(trans, dir); 5701 if (drop_inode) 5702 goto out_unlock; 5703 5704 path = btrfs_alloc_path(); 5705 BUG_ON(!path); 5706 key.objectid = inode->i_ino; 5707 key.offset = 0; 5708 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); 5709 datasize = btrfs_file_extent_calc_inline_size(name_len); 5710 err = btrfs_insert_empty_item(trans, root, path, &key, 5711 datasize); 5712 if (err) { 5713 drop_inode = 1; 5714 goto out_unlock; 5715 } 5716 leaf = path->nodes[0]; 5717 ei = btrfs_item_ptr(leaf, path->slots[0], 5718 struct btrfs_file_extent_item); 5719 btrfs_set_file_extent_generation(leaf, ei, trans->transid); 5720 btrfs_set_file_extent_type(leaf, ei, 5721 BTRFS_FILE_EXTENT_INLINE); 5722 btrfs_set_file_extent_encryption(leaf, ei, 0); 5723 btrfs_set_file_extent_compression(leaf, ei, 0); 5724 btrfs_set_file_extent_other_encoding(leaf, ei, 0); 5725 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); 5726 5727 ptr = btrfs_file_extent_inline_start(ei); 5728 write_extent_buffer(leaf, symname, ptr, name_len); 5729 btrfs_mark_buffer_dirty(leaf); 5730 btrfs_free_path(path); 5731 5732 inode->i_op = &btrfs_symlink_inode_operations; 5733 inode->i_mapping->a_ops = &btrfs_symlink_aops; 5734 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 5735 inode_set_bytes(inode, name_len); 5736 btrfs_i_size_write(inode, name_len - 1); 5737 err = btrfs_update_inode(trans, root, inode); 5738 if (err) 5739 drop_inode = 1; 5740 5741 out_unlock: 5742 nr = trans->blocks_used; 5743 btrfs_end_transaction_throttle(trans, root); 5744 out_fail: 5745 btrfs_unreserve_metadata_space(root, 5); 5746 if (drop_inode) { 5747 inode_dec_link_count(inode); 5748 iput(inode); 5749 } 5750 btrfs_btree_balance_dirty(root, nr); 5751 return err; 5752 } 5753 5754 static int prealloc_file_range(struct inode *inode, u64 start, u64 end, 5755 u64 alloc_hint, int mode, loff_t actual_len) 5756 { 5757 struct btrfs_trans_handle *trans; 5758 struct btrfs_root *root = BTRFS_I(inode)->root; 5759 struct btrfs_key ins; 5760 u64 alloc_size; 5761 u64 cur_offset = start; 5762 u64 num_bytes = end - start; 5763 int ret = 0; 5764 u64 i_size; 5765 5766 while (num_bytes > 0) { 5767 alloc_size = min(num_bytes, root->fs_info->max_extent); 5768 5769 trans = btrfs_start_transaction(root, 1); 5770 5771 ret = btrfs_reserve_extent(trans, root, alloc_size, 5772 root->sectorsize, 0, alloc_hint, 5773 (u64)-1, &ins, 1); 5774 if (ret) { 5775 WARN_ON(1); 5776 goto stop_trans; 5777 } 5778 5779 ret = btrfs_reserve_metadata_space(root, 3); 5780 if (ret) { 5781 btrfs_free_reserved_extent(root, ins.objectid, 5782 ins.offset); 5783 goto stop_trans; 5784 } 5785 5786 ret = insert_reserved_file_extent(trans, inode, 5787 cur_offset, ins.objectid, 5788 ins.offset, ins.offset, 5789 ins.offset, 0, 0, 0, 5790 BTRFS_FILE_EXTENT_PREALLOC); 5791 BUG_ON(ret); 5792 btrfs_drop_extent_cache(inode, cur_offset, 5793 cur_offset + ins.offset -1, 0); 5794 5795 num_bytes -= ins.offset; 5796 cur_offset += ins.offset; 5797 alloc_hint = ins.objectid + ins.offset; 5798 5799 inode->i_ctime = CURRENT_TIME; 5800 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; 5801 if (!(mode & FALLOC_FL_KEEP_SIZE) && 5802 (actual_len > inode->i_size) && 5803 (cur_offset > inode->i_size)) { 5804 5805 if (cur_offset > actual_len) 5806 i_size = actual_len; 5807 else 5808 i_size = cur_offset; 5809 i_size_write(inode, i_size); 5810 btrfs_ordered_update_i_size(inode, i_size, NULL); 5811 } 5812 5813 ret = btrfs_update_inode(trans, root, inode); 5814 BUG_ON(ret); 5815 5816 btrfs_end_transaction(trans, root); 5817 btrfs_unreserve_metadata_space(root, 3); 5818 } 5819 return ret; 5820 5821 stop_trans: 5822 btrfs_end_transaction(trans, root); 5823 return ret; 5824 5825 } 5826 5827 static long btrfs_fallocate(struct inode *inode, int mode, 5828 loff_t offset, loff_t len) 5829 { 5830 u64 cur_offset; 5831 u64 last_byte; 5832 u64 alloc_start; 5833 u64 alloc_end; 5834 u64 alloc_hint = 0; 5835 u64 locked_end; 5836 u64 mask = BTRFS_I(inode)->root->sectorsize - 1; 5837 struct extent_map *em; 5838 int ret; 5839 5840 alloc_start = offset & ~mask; 5841 alloc_end = (offset + len + mask) & ~mask; 5842 5843 /* 5844 * wait for ordered IO before we have any locks. We'll loop again 5845 * below with the locks held. 5846 */ 5847 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); 5848 5849 mutex_lock(&inode->i_mutex); 5850 if (alloc_start > inode->i_size) { 5851 ret = btrfs_cont_expand(inode, alloc_start); 5852 if (ret) 5853 goto out; 5854 } 5855 5856 ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode, 5857 alloc_end - alloc_start); 5858 if (ret) 5859 goto out; 5860 5861 locked_end = alloc_end - 1; 5862 while (1) { 5863 struct btrfs_ordered_extent *ordered; 5864 5865 /* the extent lock is ordered inside the running 5866 * transaction 5867 */ 5868 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, 5869 GFP_NOFS); 5870 ordered = btrfs_lookup_first_ordered_extent(inode, 5871 alloc_end - 1); 5872 if (ordered && 5873 ordered->file_offset + ordered->len > alloc_start && 5874 ordered->file_offset < alloc_end) { 5875 btrfs_put_ordered_extent(ordered); 5876 unlock_extent(&BTRFS_I(inode)->io_tree, 5877 alloc_start, locked_end, GFP_NOFS); 5878 /* 5879 * we can't wait on the range with the transaction 5880 * running or with the extent lock held 5881 */ 5882 btrfs_wait_ordered_range(inode, alloc_start, 5883 alloc_end - alloc_start); 5884 } else { 5885 if (ordered) 5886 btrfs_put_ordered_extent(ordered); 5887 break; 5888 } 5889 } 5890 5891 cur_offset = alloc_start; 5892 while (1) { 5893 em = btrfs_get_extent(inode, NULL, 0, cur_offset, 5894 alloc_end - cur_offset, 0); 5895 BUG_ON(IS_ERR(em) || !em); 5896 last_byte = min(extent_map_end(em), alloc_end); 5897 last_byte = (last_byte + mask) & ~mask; 5898 if (em->block_start == EXTENT_MAP_HOLE || 5899 (cur_offset >= inode->i_size && 5900 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { 5901 ret = prealloc_file_range(inode, 5902 cur_offset, last_byte, 5903 alloc_hint, mode, offset+len); 5904 if (ret < 0) { 5905 free_extent_map(em); 5906 break; 5907 } 5908 } 5909 if (em->block_start <= EXTENT_MAP_LAST_BYTE) 5910 alloc_hint = em->block_start; 5911 free_extent_map(em); 5912 5913 cur_offset = last_byte; 5914 if (cur_offset >= alloc_end) { 5915 ret = 0; 5916 break; 5917 } 5918 } 5919 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, 5920 GFP_NOFS); 5921 5922 btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode, 5923 alloc_end - alloc_start); 5924 out: 5925 mutex_unlock(&inode->i_mutex); 5926 return ret; 5927 } 5928 5929 static int btrfs_set_page_dirty(struct page *page) 5930 { 5931 return __set_page_dirty_nobuffers(page); 5932 } 5933 5934 static int btrfs_permission(struct inode *inode, int mask) 5935 { 5936 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) 5937 return -EACCES; 5938 return generic_permission(inode, mask, btrfs_check_acl); 5939 } 5940 5941 static const struct inode_operations btrfs_dir_inode_operations = { 5942 .getattr = btrfs_getattr, 5943 .lookup = btrfs_lookup, 5944 .create = btrfs_create, 5945 .unlink = btrfs_unlink, 5946 .link = btrfs_link, 5947 .mkdir = btrfs_mkdir, 5948 .rmdir = btrfs_rmdir, 5949 .rename = btrfs_rename, 5950 .symlink = btrfs_symlink, 5951 .setattr = btrfs_setattr, 5952 .mknod = btrfs_mknod, 5953 .setxattr = btrfs_setxattr, 5954 .getxattr = btrfs_getxattr, 5955 .listxattr = btrfs_listxattr, 5956 .removexattr = btrfs_removexattr, 5957 .permission = btrfs_permission, 5958 }; 5959 static const struct inode_operations btrfs_dir_ro_inode_operations = { 5960 .lookup = btrfs_lookup, 5961 .permission = btrfs_permission, 5962 }; 5963 5964 static const struct file_operations btrfs_dir_file_operations = { 5965 .llseek = generic_file_llseek, 5966 .read = generic_read_dir, 5967 .readdir = btrfs_real_readdir, 5968 .unlocked_ioctl = btrfs_ioctl, 5969 #ifdef CONFIG_COMPAT 5970 .compat_ioctl = btrfs_ioctl, 5971 #endif 5972 .release = btrfs_release_file, 5973 .fsync = btrfs_sync_file, 5974 }; 5975 5976 static struct extent_io_ops btrfs_extent_io_ops = { 5977 .fill_delalloc = run_delalloc_range, 5978 .submit_bio_hook = btrfs_submit_bio_hook, 5979 .merge_bio_hook = btrfs_merge_bio_hook, 5980 .readpage_end_io_hook = btrfs_readpage_end_io_hook, 5981 .writepage_end_io_hook = btrfs_writepage_end_io_hook, 5982 .writepage_start_hook = btrfs_writepage_start_hook, 5983 .readpage_io_failed_hook = btrfs_io_failed_hook, 5984 .set_bit_hook = btrfs_set_bit_hook, 5985 .clear_bit_hook = btrfs_clear_bit_hook, 5986 .merge_extent_hook = btrfs_merge_extent_hook, 5987 .split_extent_hook = btrfs_split_extent_hook, 5988 }; 5989 5990 /* 5991 * btrfs doesn't support the bmap operation because swapfiles 5992 * use bmap to make a mapping of extents in the file. They assume 5993 * these extents won't change over the life of the file and they 5994 * use the bmap result to do IO directly to the drive. 5995 * 5996 * the btrfs bmap call would return logical addresses that aren't 5997 * suitable for IO and they also will change frequently as COW 5998 * operations happen. So, swapfile + btrfs == corruption. 5999 * 6000 * For now we're avoiding this by dropping bmap. 6001 */ 6002 static const struct address_space_operations btrfs_aops = { 6003 .readpage = btrfs_readpage, 6004 .writepage = btrfs_writepage, 6005 .writepages = btrfs_writepages, 6006 .readpages = btrfs_readpages, 6007 .sync_page = block_sync_page, 6008 .direct_IO = btrfs_direct_IO, 6009 .invalidatepage = btrfs_invalidatepage, 6010 .releasepage = btrfs_releasepage, 6011 .set_page_dirty = btrfs_set_page_dirty, 6012 .error_remove_page = generic_error_remove_page, 6013 }; 6014 6015 static const struct address_space_operations btrfs_symlink_aops = { 6016 .readpage = btrfs_readpage, 6017 .writepage = btrfs_writepage, 6018 .invalidatepage = btrfs_invalidatepage, 6019 .releasepage = btrfs_releasepage, 6020 }; 6021 6022 static const struct inode_operations btrfs_file_inode_operations = { 6023 .truncate = btrfs_truncate, 6024 .getattr = btrfs_getattr, 6025 .setattr = btrfs_setattr, 6026 .setxattr = btrfs_setxattr, 6027 .getxattr = btrfs_getxattr, 6028 .listxattr = btrfs_listxattr, 6029 .removexattr = btrfs_removexattr, 6030 .permission = btrfs_permission, 6031 .fallocate = btrfs_fallocate, 6032 .fiemap = btrfs_fiemap, 6033 }; 6034 static const struct inode_operations btrfs_special_inode_operations = { 6035 .getattr = btrfs_getattr, 6036 .setattr = btrfs_setattr, 6037 .permission = btrfs_permission, 6038 .setxattr = btrfs_setxattr, 6039 .getxattr = btrfs_getxattr, 6040 .listxattr = btrfs_listxattr, 6041 .removexattr = btrfs_removexattr, 6042 }; 6043 static const struct inode_operations btrfs_symlink_inode_operations = { 6044 .readlink = generic_readlink, 6045 .follow_link = page_follow_link_light, 6046 .put_link = page_put_link, 6047 .permission = btrfs_permission, 6048 .setxattr = btrfs_setxattr, 6049 .getxattr = btrfs_getxattr, 6050 .listxattr = btrfs_listxattr, 6051 .removexattr = btrfs_removexattr, 6052 }; 6053 6054 const struct dentry_operations btrfs_dentry_operations = { 6055 .d_delete = btrfs_dentry_delete, 6056 }; 6057