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/fs.h> 20 #include <linux/blkdev.h> 21 #include <linux/scatterlist.h> 22 #include <linux/swap.h> 23 #include <linux/radix-tree.h> 24 #include <linux/writeback.h> 25 #include <linux/buffer_head.h> 26 #include <linux/workqueue.h> 27 #include <linux/kthread.h> 28 #include <linux/freezer.h> 29 #include <linux/crc32c.h> 30 #include "compat.h" 31 #include "ctree.h" 32 #include "disk-io.h" 33 #include "transaction.h" 34 #include "btrfs_inode.h" 35 #include "volumes.h" 36 #include "print-tree.h" 37 #include "async-thread.h" 38 #include "locking.h" 39 #include "tree-log.h" 40 #include "free-space-cache.h" 41 42 static struct extent_io_ops btree_extent_io_ops; 43 static void end_workqueue_fn(struct btrfs_work *work); 44 static void free_fs_root(struct btrfs_root *root); 45 46 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0); 47 48 /* 49 * end_io_wq structs are used to do processing in task context when an IO is 50 * complete. This is used during reads to verify checksums, and it is used 51 * by writes to insert metadata for new file extents after IO is complete. 52 */ 53 struct end_io_wq { 54 struct bio *bio; 55 bio_end_io_t *end_io; 56 void *private; 57 struct btrfs_fs_info *info; 58 int error; 59 int metadata; 60 struct list_head list; 61 struct btrfs_work work; 62 }; 63 64 /* 65 * async submit bios are used to offload expensive checksumming 66 * onto the worker threads. They checksum file and metadata bios 67 * just before they are sent down the IO stack. 68 */ 69 struct async_submit_bio { 70 struct inode *inode; 71 struct bio *bio; 72 struct list_head list; 73 extent_submit_bio_hook_t *submit_bio_start; 74 extent_submit_bio_hook_t *submit_bio_done; 75 int rw; 76 int mirror_num; 77 unsigned long bio_flags; 78 struct btrfs_work work; 79 }; 80 81 /* These are used to set the lockdep class on the extent buffer locks. 82 * The class is set by the readpage_end_io_hook after the buffer has 83 * passed csum validation but before the pages are unlocked. 84 * 85 * The lockdep class is also set by btrfs_init_new_buffer on freshly 86 * allocated blocks. 87 * 88 * The class is based on the level in the tree block, which allows lockdep 89 * to know that lower nodes nest inside the locks of higher nodes. 90 * 91 * We also add a check to make sure the highest level of the tree is 92 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this 93 * code needs update as well. 94 */ 95 #ifdef CONFIG_DEBUG_LOCK_ALLOC 96 # if BTRFS_MAX_LEVEL != 8 97 # error 98 # endif 99 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1]; 100 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = { 101 /* leaf */ 102 "btrfs-extent-00", 103 "btrfs-extent-01", 104 "btrfs-extent-02", 105 "btrfs-extent-03", 106 "btrfs-extent-04", 107 "btrfs-extent-05", 108 "btrfs-extent-06", 109 "btrfs-extent-07", 110 /* highest possible level */ 111 "btrfs-extent-08", 112 }; 113 #endif 114 115 /* 116 * extents on the btree inode are pretty simple, there's one extent 117 * that covers the entire device 118 */ 119 static struct extent_map *btree_get_extent(struct inode *inode, 120 struct page *page, size_t page_offset, u64 start, u64 len, 121 int create) 122 { 123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 124 struct extent_map *em; 125 int ret; 126 127 read_lock(&em_tree->lock); 128 em = lookup_extent_mapping(em_tree, start, len); 129 if (em) { 130 em->bdev = 131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 132 read_unlock(&em_tree->lock); 133 goto out; 134 } 135 read_unlock(&em_tree->lock); 136 137 em = alloc_extent_map(GFP_NOFS); 138 if (!em) { 139 em = ERR_PTR(-ENOMEM); 140 goto out; 141 } 142 em->start = 0; 143 em->len = (u64)-1; 144 em->block_len = (u64)-1; 145 em->block_start = 0; 146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 147 148 write_lock(&em_tree->lock); 149 ret = add_extent_mapping(em_tree, em); 150 if (ret == -EEXIST) { 151 u64 failed_start = em->start; 152 u64 failed_len = em->len; 153 154 free_extent_map(em); 155 em = lookup_extent_mapping(em_tree, start, len); 156 if (em) { 157 ret = 0; 158 } else { 159 em = lookup_extent_mapping(em_tree, failed_start, 160 failed_len); 161 ret = -EIO; 162 } 163 } else if (ret) { 164 free_extent_map(em); 165 em = NULL; 166 } 167 write_unlock(&em_tree->lock); 168 169 if (ret) 170 em = ERR_PTR(ret); 171 out: 172 return em; 173 } 174 175 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 176 { 177 return crc32c(seed, data, len); 178 } 179 180 void btrfs_csum_final(u32 crc, char *result) 181 { 182 *(__le32 *)result = ~cpu_to_le32(crc); 183 } 184 185 /* 186 * compute the csum for a btree block, and either verify it or write it 187 * into the csum field of the block. 188 */ 189 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 190 int verify) 191 { 192 u16 csum_size = 193 btrfs_super_csum_size(&root->fs_info->super_copy); 194 char *result = NULL; 195 unsigned long len; 196 unsigned long cur_len; 197 unsigned long offset = BTRFS_CSUM_SIZE; 198 char *map_token = NULL; 199 char *kaddr; 200 unsigned long map_start; 201 unsigned long map_len; 202 int err; 203 u32 crc = ~(u32)0; 204 unsigned long inline_result; 205 206 len = buf->len - offset; 207 while (len > 0) { 208 err = map_private_extent_buffer(buf, offset, 32, 209 &map_token, &kaddr, 210 &map_start, &map_len, KM_USER0); 211 if (err) 212 return 1; 213 cur_len = min(len, map_len - (offset - map_start)); 214 crc = btrfs_csum_data(root, kaddr + offset - map_start, 215 crc, cur_len); 216 len -= cur_len; 217 offset += cur_len; 218 unmap_extent_buffer(buf, map_token, KM_USER0); 219 } 220 if (csum_size > sizeof(inline_result)) { 221 result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 222 if (!result) 223 return 1; 224 } else { 225 result = (char *)&inline_result; 226 } 227 228 btrfs_csum_final(crc, result); 229 230 if (verify) { 231 if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 232 u32 val; 233 u32 found = 0; 234 memcpy(&found, result, csum_size); 235 236 read_extent_buffer(buf, &val, 0, csum_size); 237 if (printk_ratelimit()) { 238 printk(KERN_INFO "btrfs: %s checksum verify " 239 "failed on %llu wanted %X found %X " 240 "level %d\n", 241 root->fs_info->sb->s_id, 242 (unsigned long long)buf->start, val, found, 243 btrfs_header_level(buf)); 244 } 245 if (result != (char *)&inline_result) 246 kfree(result); 247 return 1; 248 } 249 } else { 250 write_extent_buffer(buf, result, 0, csum_size); 251 } 252 if (result != (char *)&inline_result) 253 kfree(result); 254 return 0; 255 } 256 257 /* 258 * we can't consider a given block up to date unless the transid of the 259 * block matches the transid in the parent node's pointer. This is how we 260 * detect blocks that either didn't get written at all or got written 261 * in the wrong place. 262 */ 263 static int verify_parent_transid(struct extent_io_tree *io_tree, 264 struct extent_buffer *eb, u64 parent_transid) 265 { 266 int ret; 267 268 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 269 return 0; 270 271 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS); 272 if (extent_buffer_uptodate(io_tree, eb) && 273 btrfs_header_generation(eb) == parent_transid) { 274 ret = 0; 275 goto out; 276 } 277 if (printk_ratelimit()) { 278 printk("parent transid verify failed on %llu wanted %llu " 279 "found %llu\n", 280 (unsigned long long)eb->start, 281 (unsigned long long)parent_transid, 282 (unsigned long long)btrfs_header_generation(eb)); 283 } 284 ret = 1; 285 clear_extent_buffer_uptodate(io_tree, eb); 286 out: 287 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1, 288 GFP_NOFS); 289 return ret; 290 } 291 292 /* 293 * helper to read a given tree block, doing retries as required when 294 * the checksums don't match and we have alternate mirrors to try. 295 */ 296 static int btree_read_extent_buffer_pages(struct btrfs_root *root, 297 struct extent_buffer *eb, 298 u64 start, u64 parent_transid) 299 { 300 struct extent_io_tree *io_tree; 301 int ret; 302 int num_copies = 0; 303 int mirror_num = 0; 304 305 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 306 while (1) { 307 ret = read_extent_buffer_pages(io_tree, eb, start, 1, 308 btree_get_extent, mirror_num); 309 if (!ret && 310 !verify_parent_transid(io_tree, eb, parent_transid)) 311 return ret; 312 313 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, 314 eb->start, eb->len); 315 if (num_copies == 1) 316 return ret; 317 318 mirror_num++; 319 if (mirror_num > num_copies) 320 return ret; 321 } 322 return -EIO; 323 } 324 325 /* 326 * checksum a dirty tree block before IO. This has extra checks to make sure 327 * we only fill in the checksum field in the first page of a multi-page block 328 */ 329 330 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 331 { 332 struct extent_io_tree *tree; 333 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 334 u64 found_start; 335 int found_level; 336 unsigned long len; 337 struct extent_buffer *eb; 338 int ret; 339 340 tree = &BTRFS_I(page->mapping->host)->io_tree; 341 342 if (page->private == EXTENT_PAGE_PRIVATE) 343 goto out; 344 if (!page->private) 345 goto out; 346 len = page->private >> 2; 347 WARN_ON(len == 0); 348 349 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 350 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE, 351 btrfs_header_generation(eb)); 352 BUG_ON(ret); 353 found_start = btrfs_header_bytenr(eb); 354 if (found_start != start) { 355 WARN_ON(1); 356 goto err; 357 } 358 if (eb->first_page != page) { 359 WARN_ON(1); 360 goto err; 361 } 362 if (!PageUptodate(page)) { 363 WARN_ON(1); 364 goto err; 365 } 366 found_level = btrfs_header_level(eb); 367 368 csum_tree_block(root, eb, 0); 369 err: 370 free_extent_buffer(eb); 371 out: 372 return 0; 373 } 374 375 static int check_tree_block_fsid(struct btrfs_root *root, 376 struct extent_buffer *eb) 377 { 378 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 379 u8 fsid[BTRFS_UUID_SIZE]; 380 int ret = 1; 381 382 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 383 BTRFS_FSID_SIZE); 384 while (fs_devices) { 385 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 386 ret = 0; 387 break; 388 } 389 fs_devices = fs_devices->seed; 390 } 391 return ret; 392 } 393 394 #ifdef CONFIG_DEBUG_LOCK_ALLOC 395 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level) 396 { 397 lockdep_set_class_and_name(&eb->lock, 398 &btrfs_eb_class[level], 399 btrfs_eb_name[level]); 400 } 401 #endif 402 403 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 404 struct extent_state *state) 405 { 406 struct extent_io_tree *tree; 407 u64 found_start; 408 int found_level; 409 unsigned long len; 410 struct extent_buffer *eb; 411 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 412 int ret = 0; 413 414 tree = &BTRFS_I(page->mapping->host)->io_tree; 415 if (page->private == EXTENT_PAGE_PRIVATE) 416 goto out; 417 if (!page->private) 418 goto out; 419 420 len = page->private >> 2; 421 WARN_ON(len == 0); 422 423 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 424 425 found_start = btrfs_header_bytenr(eb); 426 if (found_start != start) { 427 if (printk_ratelimit()) { 428 printk(KERN_INFO "btrfs bad tree block start " 429 "%llu %llu\n", 430 (unsigned long long)found_start, 431 (unsigned long long)eb->start); 432 } 433 ret = -EIO; 434 goto err; 435 } 436 if (eb->first_page != page) { 437 printk(KERN_INFO "btrfs bad first page %lu %lu\n", 438 eb->first_page->index, page->index); 439 WARN_ON(1); 440 ret = -EIO; 441 goto err; 442 } 443 if (check_tree_block_fsid(root, eb)) { 444 if (printk_ratelimit()) { 445 printk(KERN_INFO "btrfs bad fsid on block %llu\n", 446 (unsigned long long)eb->start); 447 } 448 ret = -EIO; 449 goto err; 450 } 451 found_level = btrfs_header_level(eb); 452 453 btrfs_set_buffer_lockdep_class(eb, found_level); 454 455 ret = csum_tree_block(root, eb, 1); 456 if (ret) 457 ret = -EIO; 458 459 end = min_t(u64, eb->len, PAGE_CACHE_SIZE); 460 end = eb->start + end - 1; 461 err: 462 free_extent_buffer(eb); 463 out: 464 return ret; 465 } 466 467 static void end_workqueue_bio(struct bio *bio, int err) 468 { 469 struct end_io_wq *end_io_wq = bio->bi_private; 470 struct btrfs_fs_info *fs_info; 471 472 fs_info = end_io_wq->info; 473 end_io_wq->error = err; 474 end_io_wq->work.func = end_workqueue_fn; 475 end_io_wq->work.flags = 0; 476 477 if (bio->bi_rw & (1 << BIO_RW)) { 478 if (end_io_wq->metadata) 479 btrfs_queue_worker(&fs_info->endio_meta_write_workers, 480 &end_io_wq->work); 481 else 482 btrfs_queue_worker(&fs_info->endio_write_workers, 483 &end_io_wq->work); 484 } else { 485 if (end_io_wq->metadata) 486 btrfs_queue_worker(&fs_info->endio_meta_workers, 487 &end_io_wq->work); 488 else 489 btrfs_queue_worker(&fs_info->endio_workers, 490 &end_io_wq->work); 491 } 492 } 493 494 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 495 int metadata) 496 { 497 struct end_io_wq *end_io_wq; 498 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 499 if (!end_io_wq) 500 return -ENOMEM; 501 502 end_io_wq->private = bio->bi_private; 503 end_io_wq->end_io = bio->bi_end_io; 504 end_io_wq->info = info; 505 end_io_wq->error = 0; 506 end_io_wq->bio = bio; 507 end_io_wq->metadata = metadata; 508 509 bio->bi_private = end_io_wq; 510 bio->bi_end_io = end_workqueue_bio; 511 return 0; 512 } 513 514 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 515 { 516 unsigned long limit = min_t(unsigned long, 517 info->workers.max_workers, 518 info->fs_devices->open_devices); 519 return 256 * limit; 520 } 521 522 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone) 523 { 524 return atomic_read(&info->nr_async_bios) > 525 btrfs_async_submit_limit(info); 526 } 527 528 static void run_one_async_start(struct btrfs_work *work) 529 { 530 struct btrfs_fs_info *fs_info; 531 struct async_submit_bio *async; 532 533 async = container_of(work, struct async_submit_bio, work); 534 fs_info = BTRFS_I(async->inode)->root->fs_info; 535 async->submit_bio_start(async->inode, async->rw, async->bio, 536 async->mirror_num, async->bio_flags); 537 } 538 539 static void run_one_async_done(struct btrfs_work *work) 540 { 541 struct btrfs_fs_info *fs_info; 542 struct async_submit_bio *async; 543 int limit; 544 545 async = container_of(work, struct async_submit_bio, work); 546 fs_info = BTRFS_I(async->inode)->root->fs_info; 547 548 limit = btrfs_async_submit_limit(fs_info); 549 limit = limit * 2 / 3; 550 551 atomic_dec(&fs_info->nr_async_submits); 552 553 if (atomic_read(&fs_info->nr_async_submits) < limit && 554 waitqueue_active(&fs_info->async_submit_wait)) 555 wake_up(&fs_info->async_submit_wait); 556 557 async->submit_bio_done(async->inode, async->rw, async->bio, 558 async->mirror_num, async->bio_flags); 559 } 560 561 static void run_one_async_free(struct btrfs_work *work) 562 { 563 struct async_submit_bio *async; 564 565 async = container_of(work, struct async_submit_bio, work); 566 kfree(async); 567 } 568 569 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 570 int rw, struct bio *bio, int mirror_num, 571 unsigned long bio_flags, 572 extent_submit_bio_hook_t *submit_bio_start, 573 extent_submit_bio_hook_t *submit_bio_done) 574 { 575 struct async_submit_bio *async; 576 577 async = kmalloc(sizeof(*async), GFP_NOFS); 578 if (!async) 579 return -ENOMEM; 580 581 async->inode = inode; 582 async->rw = rw; 583 async->bio = bio; 584 async->mirror_num = mirror_num; 585 async->submit_bio_start = submit_bio_start; 586 async->submit_bio_done = submit_bio_done; 587 588 async->work.func = run_one_async_start; 589 async->work.ordered_func = run_one_async_done; 590 async->work.ordered_free = run_one_async_free; 591 592 async->work.flags = 0; 593 async->bio_flags = bio_flags; 594 595 atomic_inc(&fs_info->nr_async_submits); 596 597 if (rw & (1 << BIO_RW_SYNCIO)) 598 btrfs_set_work_high_prio(&async->work); 599 600 btrfs_queue_worker(&fs_info->workers, &async->work); 601 602 while (atomic_read(&fs_info->async_submit_draining) && 603 atomic_read(&fs_info->nr_async_submits)) { 604 wait_event(fs_info->async_submit_wait, 605 (atomic_read(&fs_info->nr_async_submits) == 0)); 606 } 607 608 return 0; 609 } 610 611 static int btree_csum_one_bio(struct bio *bio) 612 { 613 struct bio_vec *bvec = bio->bi_io_vec; 614 int bio_index = 0; 615 struct btrfs_root *root; 616 617 WARN_ON(bio->bi_vcnt <= 0); 618 while (bio_index < bio->bi_vcnt) { 619 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 620 csum_dirty_buffer(root, bvec->bv_page); 621 bio_index++; 622 bvec++; 623 } 624 return 0; 625 } 626 627 static int __btree_submit_bio_start(struct inode *inode, int rw, 628 struct bio *bio, int mirror_num, 629 unsigned long bio_flags) 630 { 631 /* 632 * when we're called for a write, we're already in the async 633 * submission context. Just jump into btrfs_map_bio 634 */ 635 btree_csum_one_bio(bio); 636 return 0; 637 } 638 639 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 640 int mirror_num, unsigned long bio_flags) 641 { 642 /* 643 * when we're called for a write, we're already in the async 644 * submission context. Just jump into btrfs_map_bio 645 */ 646 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 647 } 648 649 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 650 int mirror_num, unsigned long bio_flags) 651 { 652 int ret; 653 654 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 655 bio, 1); 656 BUG_ON(ret); 657 658 if (!(rw & (1 << BIO_RW))) { 659 /* 660 * called for a read, do the setup so that checksum validation 661 * can happen in the async kernel threads 662 */ 663 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 664 mirror_num, 0); 665 } 666 667 /* 668 * kthread helpers are used to submit writes so that checksumming 669 * can happen in parallel across all CPUs 670 */ 671 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 672 inode, rw, bio, mirror_num, 0, 673 __btree_submit_bio_start, 674 __btree_submit_bio_done); 675 } 676 677 static int btree_writepage(struct page *page, struct writeback_control *wbc) 678 { 679 struct extent_io_tree *tree; 680 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 681 struct extent_buffer *eb; 682 int was_dirty; 683 684 tree = &BTRFS_I(page->mapping->host)->io_tree; 685 if (!(current->flags & PF_MEMALLOC)) { 686 return extent_write_full_page(tree, page, 687 btree_get_extent, wbc); 688 } 689 690 redirty_page_for_writepage(wbc, page); 691 eb = btrfs_find_tree_block(root, page_offset(page), 692 PAGE_CACHE_SIZE); 693 WARN_ON(!eb); 694 695 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); 696 if (!was_dirty) { 697 spin_lock(&root->fs_info->delalloc_lock); 698 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE; 699 spin_unlock(&root->fs_info->delalloc_lock); 700 } 701 free_extent_buffer(eb); 702 703 unlock_page(page); 704 return 0; 705 } 706 707 static int btree_writepages(struct address_space *mapping, 708 struct writeback_control *wbc) 709 { 710 struct extent_io_tree *tree; 711 tree = &BTRFS_I(mapping->host)->io_tree; 712 if (wbc->sync_mode == WB_SYNC_NONE) { 713 struct btrfs_root *root = BTRFS_I(mapping->host)->root; 714 u64 num_dirty; 715 unsigned long thresh = 32 * 1024 * 1024; 716 717 if (wbc->for_kupdate) 718 return 0; 719 720 /* this is a bit racy, but that's ok */ 721 num_dirty = root->fs_info->dirty_metadata_bytes; 722 if (num_dirty < thresh) 723 return 0; 724 } 725 return extent_writepages(tree, mapping, btree_get_extent, wbc); 726 } 727 728 static int btree_readpage(struct file *file, struct page *page) 729 { 730 struct extent_io_tree *tree; 731 tree = &BTRFS_I(page->mapping->host)->io_tree; 732 return extent_read_full_page(tree, page, btree_get_extent); 733 } 734 735 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 736 { 737 struct extent_io_tree *tree; 738 struct extent_map_tree *map; 739 int ret; 740 741 if (PageWriteback(page) || PageDirty(page)) 742 return 0; 743 744 tree = &BTRFS_I(page->mapping->host)->io_tree; 745 map = &BTRFS_I(page->mapping->host)->extent_tree; 746 747 ret = try_release_extent_state(map, tree, page, gfp_flags); 748 if (!ret) 749 return 0; 750 751 ret = try_release_extent_buffer(tree, page); 752 if (ret == 1) { 753 ClearPagePrivate(page); 754 set_page_private(page, 0); 755 page_cache_release(page); 756 } 757 758 return ret; 759 } 760 761 static void btree_invalidatepage(struct page *page, unsigned long offset) 762 { 763 struct extent_io_tree *tree; 764 tree = &BTRFS_I(page->mapping->host)->io_tree; 765 extent_invalidatepage(tree, page, offset); 766 btree_releasepage(page, GFP_NOFS); 767 if (PagePrivate(page)) { 768 printk(KERN_WARNING "btrfs warning page private not zero " 769 "on page %llu\n", (unsigned long long)page_offset(page)); 770 ClearPagePrivate(page); 771 set_page_private(page, 0); 772 page_cache_release(page); 773 } 774 } 775 776 static const struct address_space_operations btree_aops = { 777 .readpage = btree_readpage, 778 .writepage = btree_writepage, 779 .writepages = btree_writepages, 780 .releasepage = btree_releasepage, 781 .invalidatepage = btree_invalidatepage, 782 .sync_page = block_sync_page, 783 }; 784 785 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 786 u64 parent_transid) 787 { 788 struct extent_buffer *buf = NULL; 789 struct inode *btree_inode = root->fs_info->btree_inode; 790 int ret = 0; 791 792 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 793 if (!buf) 794 return 0; 795 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 796 buf, 0, 0, btree_get_extent, 0); 797 free_extent_buffer(buf); 798 return ret; 799 } 800 801 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 802 u64 bytenr, u32 blocksize) 803 { 804 struct inode *btree_inode = root->fs_info->btree_inode; 805 struct extent_buffer *eb; 806 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 807 bytenr, blocksize, GFP_NOFS); 808 return eb; 809 } 810 811 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 812 u64 bytenr, u32 blocksize) 813 { 814 struct inode *btree_inode = root->fs_info->btree_inode; 815 struct extent_buffer *eb; 816 817 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 818 bytenr, blocksize, NULL, GFP_NOFS); 819 return eb; 820 } 821 822 823 int btrfs_write_tree_block(struct extent_buffer *buf) 824 { 825 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start, 826 buf->start + buf->len - 1); 827 } 828 829 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 830 { 831 return filemap_fdatawait_range(buf->first_page->mapping, 832 buf->start, buf->start + buf->len - 1); 833 } 834 835 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 836 u32 blocksize, u64 parent_transid) 837 { 838 struct extent_buffer *buf = NULL; 839 struct inode *btree_inode = root->fs_info->btree_inode; 840 struct extent_io_tree *io_tree; 841 int ret; 842 843 io_tree = &BTRFS_I(btree_inode)->io_tree; 844 845 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 846 if (!buf) 847 return NULL; 848 849 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 850 851 if (ret == 0) 852 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 853 return buf; 854 855 } 856 857 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 858 struct extent_buffer *buf) 859 { 860 struct inode *btree_inode = root->fs_info->btree_inode; 861 if (btrfs_header_generation(buf) == 862 root->fs_info->running_transaction->transid) { 863 btrfs_assert_tree_locked(buf); 864 865 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 866 spin_lock(&root->fs_info->delalloc_lock); 867 if (root->fs_info->dirty_metadata_bytes >= buf->len) 868 root->fs_info->dirty_metadata_bytes -= buf->len; 869 else 870 WARN_ON(1); 871 spin_unlock(&root->fs_info->delalloc_lock); 872 } 873 874 /* ugh, clear_extent_buffer_dirty needs to lock the page */ 875 btrfs_set_lock_blocking(buf); 876 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 877 buf); 878 } 879 return 0; 880 } 881 882 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 883 u32 stripesize, struct btrfs_root *root, 884 struct btrfs_fs_info *fs_info, 885 u64 objectid) 886 { 887 root->node = NULL; 888 root->commit_root = NULL; 889 root->sectorsize = sectorsize; 890 root->nodesize = nodesize; 891 root->leafsize = leafsize; 892 root->stripesize = stripesize; 893 root->ref_cows = 0; 894 root->track_dirty = 0; 895 896 root->fs_info = fs_info; 897 root->objectid = objectid; 898 root->last_trans = 0; 899 root->highest_objectid = 0; 900 root->name = NULL; 901 root->in_sysfs = 0; 902 root->inode_tree.rb_node = NULL; 903 904 INIT_LIST_HEAD(&root->dirty_list); 905 INIT_LIST_HEAD(&root->orphan_list); 906 INIT_LIST_HEAD(&root->root_list); 907 spin_lock_init(&root->node_lock); 908 spin_lock_init(&root->list_lock); 909 spin_lock_init(&root->inode_lock); 910 mutex_init(&root->objectid_mutex); 911 mutex_init(&root->log_mutex); 912 init_waitqueue_head(&root->log_writer_wait); 913 init_waitqueue_head(&root->log_commit_wait[0]); 914 init_waitqueue_head(&root->log_commit_wait[1]); 915 atomic_set(&root->log_commit[0], 0); 916 atomic_set(&root->log_commit[1], 0); 917 atomic_set(&root->log_writers, 0); 918 root->log_batch = 0; 919 root->log_transid = 0; 920 root->last_log_commit = 0; 921 extent_io_tree_init(&root->dirty_log_pages, 922 fs_info->btree_inode->i_mapping, GFP_NOFS); 923 924 memset(&root->root_key, 0, sizeof(root->root_key)); 925 memset(&root->root_item, 0, sizeof(root->root_item)); 926 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 927 memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 928 root->defrag_trans_start = fs_info->generation; 929 init_completion(&root->kobj_unregister); 930 root->defrag_running = 0; 931 root->defrag_level = 0; 932 root->root_key.objectid = objectid; 933 root->anon_super.s_root = NULL; 934 root->anon_super.s_dev = 0; 935 INIT_LIST_HEAD(&root->anon_super.s_list); 936 INIT_LIST_HEAD(&root->anon_super.s_instances); 937 init_rwsem(&root->anon_super.s_umount); 938 939 return 0; 940 } 941 942 static int find_and_setup_root(struct btrfs_root *tree_root, 943 struct btrfs_fs_info *fs_info, 944 u64 objectid, 945 struct btrfs_root *root) 946 { 947 int ret; 948 u32 blocksize; 949 u64 generation; 950 951 __setup_root(tree_root->nodesize, tree_root->leafsize, 952 tree_root->sectorsize, tree_root->stripesize, 953 root, fs_info, objectid); 954 ret = btrfs_find_last_root(tree_root, objectid, 955 &root->root_item, &root->root_key); 956 if (ret > 0) 957 return -ENOENT; 958 BUG_ON(ret); 959 960 generation = btrfs_root_generation(&root->root_item); 961 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 962 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 963 blocksize, generation); 964 BUG_ON(!root->node); 965 root->commit_root = btrfs_root_node(root); 966 return 0; 967 } 968 969 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, 970 struct btrfs_fs_info *fs_info) 971 { 972 struct extent_buffer *eb; 973 struct btrfs_root *log_root_tree = fs_info->log_root_tree; 974 u64 start = 0; 975 u64 end = 0; 976 int ret; 977 978 if (!log_root_tree) 979 return 0; 980 981 while (1) { 982 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages, 983 0, &start, &end, EXTENT_DIRTY); 984 if (ret) 985 break; 986 987 clear_extent_dirty(&log_root_tree->dirty_log_pages, 988 start, end, GFP_NOFS); 989 } 990 eb = fs_info->log_root_tree->node; 991 992 WARN_ON(btrfs_header_level(eb) != 0); 993 WARN_ON(btrfs_header_nritems(eb) != 0); 994 995 ret = btrfs_free_reserved_extent(fs_info->tree_root, 996 eb->start, eb->len); 997 BUG_ON(ret); 998 999 free_extent_buffer(eb); 1000 kfree(fs_info->log_root_tree); 1001 fs_info->log_root_tree = NULL; 1002 return 0; 1003 } 1004 1005 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1006 struct btrfs_fs_info *fs_info) 1007 { 1008 struct btrfs_root *root; 1009 struct btrfs_root *tree_root = fs_info->tree_root; 1010 struct extent_buffer *leaf; 1011 1012 root = kzalloc(sizeof(*root), GFP_NOFS); 1013 if (!root) 1014 return ERR_PTR(-ENOMEM); 1015 1016 __setup_root(tree_root->nodesize, tree_root->leafsize, 1017 tree_root->sectorsize, tree_root->stripesize, 1018 root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1019 1020 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1021 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1022 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1023 /* 1024 * log trees do not get reference counted because they go away 1025 * before a real commit is actually done. They do store pointers 1026 * to file data extents, and those reference counts still get 1027 * updated (along with back refs to the log tree). 1028 */ 1029 root->ref_cows = 0; 1030 1031 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 1032 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0); 1033 if (IS_ERR(leaf)) { 1034 kfree(root); 1035 return ERR_CAST(leaf); 1036 } 1037 1038 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 1039 btrfs_set_header_bytenr(leaf, leaf->start); 1040 btrfs_set_header_generation(leaf, trans->transid); 1041 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1042 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 1043 root->node = leaf; 1044 1045 write_extent_buffer(root->node, root->fs_info->fsid, 1046 (unsigned long)btrfs_header_fsid(root->node), 1047 BTRFS_FSID_SIZE); 1048 btrfs_mark_buffer_dirty(root->node); 1049 btrfs_tree_unlock(root->node); 1050 return root; 1051 } 1052 1053 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1054 struct btrfs_fs_info *fs_info) 1055 { 1056 struct btrfs_root *log_root; 1057 1058 log_root = alloc_log_tree(trans, fs_info); 1059 if (IS_ERR(log_root)) 1060 return PTR_ERR(log_root); 1061 WARN_ON(fs_info->log_root_tree); 1062 fs_info->log_root_tree = log_root; 1063 return 0; 1064 } 1065 1066 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1067 struct btrfs_root *root) 1068 { 1069 struct btrfs_root *log_root; 1070 struct btrfs_inode_item *inode_item; 1071 1072 log_root = alloc_log_tree(trans, root->fs_info); 1073 if (IS_ERR(log_root)) 1074 return PTR_ERR(log_root); 1075 1076 log_root->last_trans = trans->transid; 1077 log_root->root_key.offset = root->root_key.objectid; 1078 1079 inode_item = &log_root->root_item.inode; 1080 inode_item->generation = cpu_to_le64(1); 1081 inode_item->size = cpu_to_le64(3); 1082 inode_item->nlink = cpu_to_le32(1); 1083 inode_item->nbytes = cpu_to_le64(root->leafsize); 1084 inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 1085 1086 btrfs_set_root_node(&log_root->root_item, log_root->node); 1087 1088 WARN_ON(root->log_root); 1089 root->log_root = log_root; 1090 root->log_transid = 0; 1091 root->last_log_commit = 0; 1092 return 0; 1093 } 1094 1095 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1096 struct btrfs_key *location) 1097 { 1098 struct btrfs_root *root; 1099 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1100 struct btrfs_path *path; 1101 struct extent_buffer *l; 1102 u64 generation; 1103 u32 blocksize; 1104 int ret = 0; 1105 1106 root = kzalloc(sizeof(*root), GFP_NOFS); 1107 if (!root) 1108 return ERR_PTR(-ENOMEM); 1109 if (location->offset == (u64)-1) { 1110 ret = find_and_setup_root(tree_root, fs_info, 1111 location->objectid, root); 1112 if (ret) { 1113 kfree(root); 1114 return ERR_PTR(ret); 1115 } 1116 goto out; 1117 } 1118 1119 __setup_root(tree_root->nodesize, tree_root->leafsize, 1120 tree_root->sectorsize, tree_root->stripesize, 1121 root, fs_info, location->objectid); 1122 1123 path = btrfs_alloc_path(); 1124 BUG_ON(!path); 1125 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 1126 if (ret == 0) { 1127 l = path->nodes[0]; 1128 read_extent_buffer(l, &root->root_item, 1129 btrfs_item_ptr_offset(l, path->slots[0]), 1130 sizeof(root->root_item)); 1131 memcpy(&root->root_key, location, sizeof(*location)); 1132 } 1133 btrfs_free_path(path); 1134 if (ret) { 1135 if (ret > 0) 1136 ret = -ENOENT; 1137 return ERR_PTR(ret); 1138 } 1139 1140 generation = btrfs_root_generation(&root->root_item); 1141 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1142 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1143 blocksize, generation); 1144 root->commit_root = btrfs_root_node(root); 1145 BUG_ON(!root->node); 1146 out: 1147 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) 1148 root->ref_cows = 1; 1149 1150 return root; 1151 } 1152 1153 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1154 u64 root_objectid) 1155 { 1156 struct btrfs_root *root; 1157 1158 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID) 1159 return fs_info->tree_root; 1160 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) 1161 return fs_info->extent_root; 1162 1163 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1164 (unsigned long)root_objectid); 1165 return root; 1166 } 1167 1168 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1169 struct btrfs_key *location) 1170 { 1171 struct btrfs_root *root; 1172 int ret; 1173 1174 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1175 return fs_info->tree_root; 1176 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1177 return fs_info->extent_root; 1178 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 1179 return fs_info->chunk_root; 1180 if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 1181 return fs_info->dev_root; 1182 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 1183 return fs_info->csum_root; 1184 again: 1185 spin_lock(&fs_info->fs_roots_radix_lock); 1186 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1187 (unsigned long)location->objectid); 1188 spin_unlock(&fs_info->fs_roots_radix_lock); 1189 if (root) 1190 return root; 1191 1192 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1193 if (ret == 0) 1194 ret = -ENOENT; 1195 if (ret < 0) 1196 return ERR_PTR(ret); 1197 1198 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 1199 if (IS_ERR(root)) 1200 return root; 1201 1202 WARN_ON(btrfs_root_refs(&root->root_item) == 0); 1203 set_anon_super(&root->anon_super, NULL); 1204 1205 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1206 if (ret) 1207 goto fail; 1208 1209 spin_lock(&fs_info->fs_roots_radix_lock); 1210 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1211 (unsigned long)root->root_key.objectid, 1212 root); 1213 if (ret == 0) 1214 root->in_radix = 1; 1215 spin_unlock(&fs_info->fs_roots_radix_lock); 1216 radix_tree_preload_end(); 1217 if (ret) { 1218 if (ret == -EEXIST) { 1219 free_fs_root(root); 1220 goto again; 1221 } 1222 goto fail; 1223 } 1224 1225 ret = btrfs_find_dead_roots(fs_info->tree_root, 1226 root->root_key.objectid); 1227 WARN_ON(ret); 1228 1229 if (!(fs_info->sb->s_flags & MS_RDONLY)) 1230 btrfs_orphan_cleanup(root); 1231 1232 return root; 1233 fail: 1234 free_fs_root(root); 1235 return ERR_PTR(ret); 1236 } 1237 1238 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info, 1239 struct btrfs_key *location, 1240 const char *name, int namelen) 1241 { 1242 return btrfs_read_fs_root_no_name(fs_info, location); 1243 #if 0 1244 struct btrfs_root *root; 1245 int ret; 1246 1247 root = btrfs_read_fs_root_no_name(fs_info, location); 1248 if (!root) 1249 return NULL; 1250 1251 if (root->in_sysfs) 1252 return root; 1253 1254 ret = btrfs_set_root_name(root, name, namelen); 1255 if (ret) { 1256 free_extent_buffer(root->node); 1257 kfree(root); 1258 return ERR_PTR(ret); 1259 } 1260 1261 ret = btrfs_sysfs_add_root(root); 1262 if (ret) { 1263 free_extent_buffer(root->node); 1264 kfree(root->name); 1265 kfree(root); 1266 return ERR_PTR(ret); 1267 } 1268 root->in_sysfs = 1; 1269 return root; 1270 #endif 1271 } 1272 1273 static int btrfs_congested_fn(void *congested_data, int bdi_bits) 1274 { 1275 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 1276 int ret = 0; 1277 struct btrfs_device *device; 1278 struct backing_dev_info *bdi; 1279 1280 list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1281 if (!device->bdev) 1282 continue; 1283 bdi = blk_get_backing_dev_info(device->bdev); 1284 if (bdi && bdi_congested(bdi, bdi_bits)) { 1285 ret = 1; 1286 break; 1287 } 1288 } 1289 return ret; 1290 } 1291 1292 /* 1293 * this unplugs every device on the box, and it is only used when page 1294 * is null 1295 */ 1296 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 1297 { 1298 struct btrfs_device *device; 1299 struct btrfs_fs_info *info; 1300 1301 info = (struct btrfs_fs_info *)bdi->unplug_io_data; 1302 list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1303 if (!device->bdev) 1304 continue; 1305 1306 bdi = blk_get_backing_dev_info(device->bdev); 1307 if (bdi->unplug_io_fn) 1308 bdi->unplug_io_fn(bdi, page); 1309 } 1310 } 1311 1312 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 1313 { 1314 struct inode *inode; 1315 struct extent_map_tree *em_tree; 1316 struct extent_map *em; 1317 struct address_space *mapping; 1318 u64 offset; 1319 1320 /* the generic O_DIRECT read code does this */ 1321 if (1 || !page) { 1322 __unplug_io_fn(bdi, page); 1323 return; 1324 } 1325 1326 /* 1327 * page->mapping may change at any time. Get a consistent copy 1328 * and use that for everything below 1329 */ 1330 smp_mb(); 1331 mapping = page->mapping; 1332 if (!mapping) 1333 return; 1334 1335 inode = mapping->host; 1336 1337 /* 1338 * don't do the expensive searching for a small number of 1339 * devices 1340 */ 1341 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) { 1342 __unplug_io_fn(bdi, page); 1343 return; 1344 } 1345 1346 offset = page_offset(page); 1347 1348 em_tree = &BTRFS_I(inode)->extent_tree; 1349 read_lock(&em_tree->lock); 1350 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 1351 read_unlock(&em_tree->lock); 1352 if (!em) { 1353 __unplug_io_fn(bdi, page); 1354 return; 1355 } 1356 1357 if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 1358 free_extent_map(em); 1359 __unplug_io_fn(bdi, page); 1360 return; 1361 } 1362 offset = offset - em->start; 1363 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree, 1364 em->block_start + offset, page); 1365 free_extent_map(em); 1366 } 1367 1368 /* 1369 * If this fails, caller must call bdi_destroy() to get rid of the 1370 * bdi again. 1371 */ 1372 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 1373 { 1374 int err; 1375 1376 bdi->name = "btrfs"; 1377 bdi->capabilities = BDI_CAP_MAP_COPY; 1378 err = bdi_init(bdi); 1379 if (err) 1380 return err; 1381 1382 err = bdi_register(bdi, NULL, "btrfs-%d", 1383 atomic_inc_return(&btrfs_bdi_num)); 1384 if (err) { 1385 bdi_destroy(bdi); 1386 return err; 1387 } 1388 1389 bdi->ra_pages = default_backing_dev_info.ra_pages; 1390 bdi->unplug_io_fn = btrfs_unplug_io_fn; 1391 bdi->unplug_io_data = info; 1392 bdi->congested_fn = btrfs_congested_fn; 1393 bdi->congested_data = info; 1394 return 0; 1395 } 1396 1397 static int bio_ready_for_csum(struct bio *bio) 1398 { 1399 u64 length = 0; 1400 u64 buf_len = 0; 1401 u64 start = 0; 1402 struct page *page; 1403 struct extent_io_tree *io_tree = NULL; 1404 struct btrfs_fs_info *info = NULL; 1405 struct bio_vec *bvec; 1406 int i; 1407 int ret; 1408 1409 bio_for_each_segment(bvec, bio, i) { 1410 page = bvec->bv_page; 1411 if (page->private == EXTENT_PAGE_PRIVATE) { 1412 length += bvec->bv_len; 1413 continue; 1414 } 1415 if (!page->private) { 1416 length += bvec->bv_len; 1417 continue; 1418 } 1419 length = bvec->bv_len; 1420 buf_len = page->private >> 2; 1421 start = page_offset(page) + bvec->bv_offset; 1422 io_tree = &BTRFS_I(page->mapping->host)->io_tree; 1423 info = BTRFS_I(page->mapping->host)->root->fs_info; 1424 } 1425 /* are we fully contained in this bio? */ 1426 if (buf_len <= length) 1427 return 1; 1428 1429 ret = extent_range_uptodate(io_tree, start + length, 1430 start + buf_len - 1); 1431 return ret; 1432 } 1433 1434 /* 1435 * called by the kthread helper functions to finally call the bio end_io 1436 * functions. This is where read checksum verification actually happens 1437 */ 1438 static void end_workqueue_fn(struct btrfs_work *work) 1439 { 1440 struct bio *bio; 1441 struct end_io_wq *end_io_wq; 1442 struct btrfs_fs_info *fs_info; 1443 int error; 1444 1445 end_io_wq = container_of(work, struct end_io_wq, work); 1446 bio = end_io_wq->bio; 1447 fs_info = end_io_wq->info; 1448 1449 /* metadata bio reads are special because the whole tree block must 1450 * be checksummed at once. This makes sure the entire block is in 1451 * ram and up to date before trying to verify things. For 1452 * blocksize <= pagesize, it is basically a noop 1453 */ 1454 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata && 1455 !bio_ready_for_csum(bio)) { 1456 btrfs_queue_worker(&fs_info->endio_meta_workers, 1457 &end_io_wq->work); 1458 return; 1459 } 1460 error = end_io_wq->error; 1461 bio->bi_private = end_io_wq->private; 1462 bio->bi_end_io = end_io_wq->end_io; 1463 kfree(end_io_wq); 1464 bio_endio(bio, error); 1465 } 1466 1467 static int cleaner_kthread(void *arg) 1468 { 1469 struct btrfs_root *root = arg; 1470 1471 do { 1472 smp_mb(); 1473 if (root->fs_info->closing) 1474 break; 1475 1476 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1477 1478 if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 1479 mutex_trylock(&root->fs_info->cleaner_mutex)) { 1480 btrfs_clean_old_snapshots(root); 1481 mutex_unlock(&root->fs_info->cleaner_mutex); 1482 } 1483 1484 if (freezing(current)) { 1485 refrigerator(); 1486 } else { 1487 smp_mb(); 1488 if (root->fs_info->closing) 1489 break; 1490 set_current_state(TASK_INTERRUPTIBLE); 1491 schedule(); 1492 __set_current_state(TASK_RUNNING); 1493 } 1494 } while (!kthread_should_stop()); 1495 return 0; 1496 } 1497 1498 static int transaction_kthread(void *arg) 1499 { 1500 struct btrfs_root *root = arg; 1501 struct btrfs_trans_handle *trans; 1502 struct btrfs_transaction *cur; 1503 unsigned long now; 1504 unsigned long delay; 1505 int ret; 1506 1507 do { 1508 smp_mb(); 1509 if (root->fs_info->closing) 1510 break; 1511 1512 delay = HZ * 30; 1513 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1514 mutex_lock(&root->fs_info->transaction_kthread_mutex); 1515 1516 mutex_lock(&root->fs_info->trans_mutex); 1517 cur = root->fs_info->running_transaction; 1518 if (!cur) { 1519 mutex_unlock(&root->fs_info->trans_mutex); 1520 goto sleep; 1521 } 1522 1523 now = get_seconds(); 1524 if (now < cur->start_time || now - cur->start_time < 30) { 1525 mutex_unlock(&root->fs_info->trans_mutex); 1526 delay = HZ * 5; 1527 goto sleep; 1528 } 1529 mutex_unlock(&root->fs_info->trans_mutex); 1530 trans = btrfs_start_transaction(root, 1); 1531 ret = btrfs_commit_transaction(trans, root); 1532 1533 sleep: 1534 wake_up_process(root->fs_info->cleaner_kthread); 1535 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1536 1537 if (freezing(current)) { 1538 refrigerator(); 1539 } else { 1540 if (root->fs_info->closing) 1541 break; 1542 set_current_state(TASK_INTERRUPTIBLE); 1543 schedule_timeout(delay); 1544 __set_current_state(TASK_RUNNING); 1545 } 1546 } while (!kthread_should_stop()); 1547 return 0; 1548 } 1549 1550 struct btrfs_root *open_ctree(struct super_block *sb, 1551 struct btrfs_fs_devices *fs_devices, 1552 char *options) 1553 { 1554 u32 sectorsize; 1555 u32 nodesize; 1556 u32 leafsize; 1557 u32 blocksize; 1558 u32 stripesize; 1559 u64 generation; 1560 u64 features; 1561 struct btrfs_key location; 1562 struct buffer_head *bh; 1563 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root), 1564 GFP_NOFS); 1565 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root), 1566 GFP_NOFS); 1567 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root), 1568 GFP_NOFS); 1569 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info), 1570 GFP_NOFS); 1571 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root), 1572 GFP_NOFS); 1573 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root), 1574 GFP_NOFS); 1575 struct btrfs_root *log_tree_root; 1576 1577 int ret; 1578 int err = -EINVAL; 1579 1580 struct btrfs_super_block *disk_super; 1581 1582 if (!extent_root || !tree_root || !fs_info || 1583 !chunk_root || !dev_root || !csum_root) { 1584 err = -ENOMEM; 1585 goto fail; 1586 } 1587 1588 ret = init_srcu_struct(&fs_info->subvol_srcu); 1589 if (ret) { 1590 err = ret; 1591 goto fail; 1592 } 1593 1594 ret = setup_bdi(fs_info, &fs_info->bdi); 1595 if (ret) { 1596 err = ret; 1597 goto fail_srcu; 1598 } 1599 1600 fs_info->btree_inode = new_inode(sb); 1601 if (!fs_info->btree_inode) { 1602 err = -ENOMEM; 1603 goto fail_bdi; 1604 } 1605 1606 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 1607 INIT_LIST_HEAD(&fs_info->trans_list); 1608 INIT_LIST_HEAD(&fs_info->dead_roots); 1609 INIT_LIST_HEAD(&fs_info->hashers); 1610 INIT_LIST_HEAD(&fs_info->delalloc_inodes); 1611 INIT_LIST_HEAD(&fs_info->ordered_operations); 1612 INIT_LIST_HEAD(&fs_info->caching_block_groups); 1613 spin_lock_init(&fs_info->delalloc_lock); 1614 spin_lock_init(&fs_info->new_trans_lock); 1615 spin_lock_init(&fs_info->ref_cache_lock); 1616 spin_lock_init(&fs_info->fs_roots_radix_lock); 1617 1618 init_completion(&fs_info->kobj_unregister); 1619 fs_info->tree_root = tree_root; 1620 fs_info->extent_root = extent_root; 1621 fs_info->csum_root = csum_root; 1622 fs_info->chunk_root = chunk_root; 1623 fs_info->dev_root = dev_root; 1624 fs_info->fs_devices = fs_devices; 1625 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 1626 INIT_LIST_HEAD(&fs_info->space_info); 1627 btrfs_mapping_init(&fs_info->mapping_tree); 1628 atomic_set(&fs_info->nr_async_submits, 0); 1629 atomic_set(&fs_info->async_delalloc_pages, 0); 1630 atomic_set(&fs_info->async_submit_draining, 0); 1631 atomic_set(&fs_info->nr_async_bios, 0); 1632 fs_info->sb = sb; 1633 fs_info->max_extent = (u64)-1; 1634 fs_info->max_inline = 8192 * 1024; 1635 fs_info->metadata_ratio = 0; 1636 1637 fs_info->thread_pool_size = min_t(unsigned long, 1638 num_online_cpus() + 2, 8); 1639 1640 INIT_LIST_HEAD(&fs_info->ordered_extents); 1641 spin_lock_init(&fs_info->ordered_extent_lock); 1642 1643 sb->s_blocksize = 4096; 1644 sb->s_blocksize_bits = blksize_bits(4096); 1645 sb->s_bdi = &fs_info->bdi; 1646 1647 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 1648 fs_info->btree_inode->i_nlink = 1; 1649 /* 1650 * we set the i_size on the btree inode to the max possible int. 1651 * the real end of the address space is determined by all of 1652 * the devices in the system 1653 */ 1654 fs_info->btree_inode->i_size = OFFSET_MAX; 1655 fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 1656 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 1657 1658 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 1659 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 1660 fs_info->btree_inode->i_mapping, 1661 GFP_NOFS); 1662 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree, 1663 GFP_NOFS); 1664 1665 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 1666 1667 BTRFS_I(fs_info->btree_inode)->root = tree_root; 1668 memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 1669 sizeof(struct btrfs_key)); 1670 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1; 1671 insert_inode_hash(fs_info->btree_inode); 1672 1673 spin_lock_init(&fs_info->block_group_cache_lock); 1674 fs_info->block_group_cache_tree.rb_node = NULL; 1675 1676 extent_io_tree_init(&fs_info->freed_extents[0], 1677 fs_info->btree_inode->i_mapping, GFP_NOFS); 1678 extent_io_tree_init(&fs_info->freed_extents[1], 1679 fs_info->btree_inode->i_mapping, GFP_NOFS); 1680 fs_info->pinned_extents = &fs_info->freed_extents[0]; 1681 fs_info->do_barriers = 1; 1682 1683 1684 mutex_init(&fs_info->trans_mutex); 1685 mutex_init(&fs_info->ordered_operations_mutex); 1686 mutex_init(&fs_info->tree_log_mutex); 1687 mutex_init(&fs_info->chunk_mutex); 1688 mutex_init(&fs_info->transaction_kthread_mutex); 1689 mutex_init(&fs_info->cleaner_mutex); 1690 mutex_init(&fs_info->volume_mutex); 1691 init_rwsem(&fs_info->extent_commit_sem); 1692 init_rwsem(&fs_info->subvol_sem); 1693 1694 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1695 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1696 1697 init_waitqueue_head(&fs_info->transaction_throttle); 1698 init_waitqueue_head(&fs_info->transaction_wait); 1699 init_waitqueue_head(&fs_info->async_submit_wait); 1700 1701 __setup_root(4096, 4096, 4096, 4096, tree_root, 1702 fs_info, BTRFS_ROOT_TREE_OBJECTID); 1703 1704 1705 bh = btrfs_read_dev_super(fs_devices->latest_bdev); 1706 if (!bh) 1707 goto fail_iput; 1708 1709 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy)); 1710 memcpy(&fs_info->super_for_commit, &fs_info->super_copy, 1711 sizeof(fs_info->super_for_commit)); 1712 brelse(bh); 1713 1714 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE); 1715 1716 disk_super = &fs_info->super_copy; 1717 if (!btrfs_super_root(disk_super)) 1718 goto fail_iput; 1719 1720 ret = btrfs_parse_options(tree_root, options); 1721 if (ret) { 1722 err = ret; 1723 goto fail_iput; 1724 } 1725 1726 features = btrfs_super_incompat_flags(disk_super) & 1727 ~BTRFS_FEATURE_INCOMPAT_SUPP; 1728 if (features) { 1729 printk(KERN_ERR "BTRFS: couldn't mount because of " 1730 "unsupported optional features (%Lx).\n", 1731 (unsigned long long)features); 1732 err = -EINVAL; 1733 goto fail_iput; 1734 } 1735 1736 features = btrfs_super_incompat_flags(disk_super); 1737 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) { 1738 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 1739 btrfs_set_super_incompat_flags(disk_super, features); 1740 } 1741 1742 features = btrfs_super_compat_ro_flags(disk_super) & 1743 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 1744 if (!(sb->s_flags & MS_RDONLY) && features) { 1745 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 1746 "unsupported option features (%Lx).\n", 1747 (unsigned long long)features); 1748 err = -EINVAL; 1749 goto fail_iput; 1750 } 1751 1752 btrfs_init_workers(&fs_info->generic_worker, 1753 "genwork", 1, NULL); 1754 1755 btrfs_init_workers(&fs_info->workers, "worker", 1756 fs_info->thread_pool_size, 1757 &fs_info->generic_worker); 1758 1759 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 1760 fs_info->thread_pool_size, 1761 &fs_info->generic_worker); 1762 1763 btrfs_init_workers(&fs_info->submit_workers, "submit", 1764 min_t(u64, fs_devices->num_devices, 1765 fs_info->thread_pool_size), 1766 &fs_info->generic_worker); 1767 btrfs_init_workers(&fs_info->enospc_workers, "enospc", 1768 fs_info->thread_pool_size, 1769 &fs_info->generic_worker); 1770 1771 /* a higher idle thresh on the submit workers makes it much more 1772 * likely that bios will be send down in a sane order to the 1773 * devices 1774 */ 1775 fs_info->submit_workers.idle_thresh = 64; 1776 1777 fs_info->workers.idle_thresh = 16; 1778 fs_info->workers.ordered = 1; 1779 1780 fs_info->delalloc_workers.idle_thresh = 2; 1781 fs_info->delalloc_workers.ordered = 1; 1782 1783 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 1784 &fs_info->generic_worker); 1785 btrfs_init_workers(&fs_info->endio_workers, "endio", 1786 fs_info->thread_pool_size, 1787 &fs_info->generic_worker); 1788 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 1789 fs_info->thread_pool_size, 1790 &fs_info->generic_worker); 1791 btrfs_init_workers(&fs_info->endio_meta_write_workers, 1792 "endio-meta-write", fs_info->thread_pool_size, 1793 &fs_info->generic_worker); 1794 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 1795 fs_info->thread_pool_size, 1796 &fs_info->generic_worker); 1797 1798 /* 1799 * endios are largely parallel and should have a very 1800 * low idle thresh 1801 */ 1802 fs_info->endio_workers.idle_thresh = 4; 1803 fs_info->endio_meta_workers.idle_thresh = 4; 1804 1805 fs_info->endio_write_workers.idle_thresh = 2; 1806 fs_info->endio_meta_write_workers.idle_thresh = 2; 1807 1808 btrfs_start_workers(&fs_info->workers, 1); 1809 btrfs_start_workers(&fs_info->generic_worker, 1); 1810 btrfs_start_workers(&fs_info->submit_workers, 1); 1811 btrfs_start_workers(&fs_info->delalloc_workers, 1); 1812 btrfs_start_workers(&fs_info->fixup_workers, 1); 1813 btrfs_start_workers(&fs_info->endio_workers, 1); 1814 btrfs_start_workers(&fs_info->endio_meta_workers, 1); 1815 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); 1816 btrfs_start_workers(&fs_info->endio_write_workers, 1); 1817 btrfs_start_workers(&fs_info->enospc_workers, 1); 1818 1819 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1820 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1821 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 1822 1823 nodesize = btrfs_super_nodesize(disk_super); 1824 leafsize = btrfs_super_leafsize(disk_super); 1825 sectorsize = btrfs_super_sectorsize(disk_super); 1826 stripesize = btrfs_super_stripesize(disk_super); 1827 tree_root->nodesize = nodesize; 1828 tree_root->leafsize = leafsize; 1829 tree_root->sectorsize = sectorsize; 1830 tree_root->stripesize = stripesize; 1831 1832 sb->s_blocksize = sectorsize; 1833 sb->s_blocksize_bits = blksize_bits(sectorsize); 1834 1835 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 1836 sizeof(disk_super->magic))) { 1837 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 1838 goto fail_sb_buffer; 1839 } 1840 1841 mutex_lock(&fs_info->chunk_mutex); 1842 ret = btrfs_read_sys_array(tree_root); 1843 mutex_unlock(&fs_info->chunk_mutex); 1844 if (ret) { 1845 printk(KERN_WARNING "btrfs: failed to read the system " 1846 "array on %s\n", sb->s_id); 1847 goto fail_sb_buffer; 1848 } 1849 1850 blocksize = btrfs_level_size(tree_root, 1851 btrfs_super_chunk_root_level(disk_super)); 1852 generation = btrfs_super_chunk_root_generation(disk_super); 1853 1854 __setup_root(nodesize, leafsize, sectorsize, stripesize, 1855 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 1856 1857 chunk_root->node = read_tree_block(chunk_root, 1858 btrfs_super_chunk_root(disk_super), 1859 blocksize, generation); 1860 BUG_ON(!chunk_root->node); 1861 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 1862 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 1863 sb->s_id); 1864 goto fail_chunk_root; 1865 } 1866 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 1867 chunk_root->commit_root = btrfs_root_node(chunk_root); 1868 1869 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 1870 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 1871 BTRFS_UUID_SIZE); 1872 1873 mutex_lock(&fs_info->chunk_mutex); 1874 ret = btrfs_read_chunk_tree(chunk_root); 1875 mutex_unlock(&fs_info->chunk_mutex); 1876 if (ret) { 1877 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 1878 sb->s_id); 1879 goto fail_chunk_root; 1880 } 1881 1882 btrfs_close_extra_devices(fs_devices); 1883 1884 blocksize = btrfs_level_size(tree_root, 1885 btrfs_super_root_level(disk_super)); 1886 generation = btrfs_super_generation(disk_super); 1887 1888 tree_root->node = read_tree_block(tree_root, 1889 btrfs_super_root(disk_super), 1890 blocksize, generation); 1891 if (!tree_root->node) 1892 goto fail_chunk_root; 1893 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 1894 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 1895 sb->s_id); 1896 goto fail_tree_root; 1897 } 1898 btrfs_set_root_node(&tree_root->root_item, tree_root->node); 1899 tree_root->commit_root = btrfs_root_node(tree_root); 1900 1901 ret = find_and_setup_root(tree_root, fs_info, 1902 BTRFS_EXTENT_TREE_OBJECTID, extent_root); 1903 if (ret) 1904 goto fail_tree_root; 1905 extent_root->track_dirty = 1; 1906 1907 ret = find_and_setup_root(tree_root, fs_info, 1908 BTRFS_DEV_TREE_OBJECTID, dev_root); 1909 if (ret) 1910 goto fail_extent_root; 1911 dev_root->track_dirty = 1; 1912 1913 ret = find_and_setup_root(tree_root, fs_info, 1914 BTRFS_CSUM_TREE_OBJECTID, csum_root); 1915 if (ret) 1916 goto fail_dev_root; 1917 1918 csum_root->track_dirty = 1; 1919 1920 btrfs_read_block_groups(extent_root); 1921 1922 fs_info->generation = generation; 1923 fs_info->last_trans_committed = generation; 1924 fs_info->data_alloc_profile = (u64)-1; 1925 fs_info->metadata_alloc_profile = (u64)-1; 1926 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile; 1927 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 1928 "btrfs-cleaner"); 1929 if (IS_ERR(fs_info->cleaner_kthread)) 1930 goto fail_csum_root; 1931 1932 fs_info->transaction_kthread = kthread_run(transaction_kthread, 1933 tree_root, 1934 "btrfs-transaction"); 1935 if (IS_ERR(fs_info->transaction_kthread)) 1936 goto fail_cleaner; 1937 1938 if (!btrfs_test_opt(tree_root, SSD) && 1939 !btrfs_test_opt(tree_root, NOSSD) && 1940 !fs_info->fs_devices->rotating) { 1941 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 1942 "mode\n"); 1943 btrfs_set_opt(fs_info->mount_opt, SSD); 1944 } 1945 1946 if (btrfs_super_log_root(disk_super) != 0) { 1947 u64 bytenr = btrfs_super_log_root(disk_super); 1948 1949 if (fs_devices->rw_devices == 0) { 1950 printk(KERN_WARNING "Btrfs log replay required " 1951 "on RO media\n"); 1952 err = -EIO; 1953 goto fail_trans_kthread; 1954 } 1955 blocksize = 1956 btrfs_level_size(tree_root, 1957 btrfs_super_log_root_level(disk_super)); 1958 1959 log_tree_root = kzalloc(sizeof(struct btrfs_root), 1960 GFP_NOFS); 1961 1962 __setup_root(nodesize, leafsize, sectorsize, stripesize, 1963 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1964 1965 log_tree_root->node = read_tree_block(tree_root, bytenr, 1966 blocksize, 1967 generation + 1); 1968 ret = btrfs_recover_log_trees(log_tree_root); 1969 BUG_ON(ret); 1970 1971 if (sb->s_flags & MS_RDONLY) { 1972 ret = btrfs_commit_super(tree_root); 1973 BUG_ON(ret); 1974 } 1975 } 1976 1977 ret = btrfs_find_orphan_roots(tree_root); 1978 BUG_ON(ret); 1979 1980 if (!(sb->s_flags & MS_RDONLY)) { 1981 ret = btrfs_recover_relocation(tree_root); 1982 BUG_ON(ret); 1983 } 1984 1985 location.objectid = BTRFS_FS_TREE_OBJECTID; 1986 location.type = BTRFS_ROOT_ITEM_KEY; 1987 location.offset = (u64)-1; 1988 1989 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 1990 if (!fs_info->fs_root) 1991 goto fail_trans_kthread; 1992 1993 return tree_root; 1994 1995 fail_trans_kthread: 1996 kthread_stop(fs_info->transaction_kthread); 1997 fail_cleaner: 1998 kthread_stop(fs_info->cleaner_kthread); 1999 2000 /* 2001 * make sure we're done with the btree inode before we stop our 2002 * kthreads 2003 */ 2004 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 2005 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 2006 2007 fail_csum_root: 2008 free_extent_buffer(csum_root->node); 2009 free_extent_buffer(csum_root->commit_root); 2010 fail_dev_root: 2011 free_extent_buffer(dev_root->node); 2012 free_extent_buffer(dev_root->commit_root); 2013 fail_extent_root: 2014 free_extent_buffer(extent_root->node); 2015 free_extent_buffer(extent_root->commit_root); 2016 fail_tree_root: 2017 free_extent_buffer(tree_root->node); 2018 free_extent_buffer(tree_root->commit_root); 2019 fail_chunk_root: 2020 free_extent_buffer(chunk_root->node); 2021 free_extent_buffer(chunk_root->commit_root); 2022 fail_sb_buffer: 2023 btrfs_stop_workers(&fs_info->generic_worker); 2024 btrfs_stop_workers(&fs_info->fixup_workers); 2025 btrfs_stop_workers(&fs_info->delalloc_workers); 2026 btrfs_stop_workers(&fs_info->workers); 2027 btrfs_stop_workers(&fs_info->endio_workers); 2028 btrfs_stop_workers(&fs_info->endio_meta_workers); 2029 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2030 btrfs_stop_workers(&fs_info->endio_write_workers); 2031 btrfs_stop_workers(&fs_info->submit_workers); 2032 btrfs_stop_workers(&fs_info->enospc_workers); 2033 fail_iput: 2034 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 2035 iput(fs_info->btree_inode); 2036 2037 btrfs_close_devices(fs_info->fs_devices); 2038 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2039 fail_bdi: 2040 bdi_destroy(&fs_info->bdi); 2041 fail_srcu: 2042 cleanup_srcu_struct(&fs_info->subvol_srcu); 2043 fail: 2044 kfree(extent_root); 2045 kfree(tree_root); 2046 kfree(fs_info); 2047 kfree(chunk_root); 2048 kfree(dev_root); 2049 kfree(csum_root); 2050 return ERR_PTR(err); 2051 } 2052 2053 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2054 { 2055 char b[BDEVNAME_SIZE]; 2056 2057 if (uptodate) { 2058 set_buffer_uptodate(bh); 2059 } else { 2060 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) { 2061 printk(KERN_WARNING "lost page write due to " 2062 "I/O error on %s\n", 2063 bdevname(bh->b_bdev, b)); 2064 } 2065 /* note, we dont' set_buffer_write_io_error because we have 2066 * our own ways of dealing with the IO errors 2067 */ 2068 clear_buffer_uptodate(bh); 2069 } 2070 unlock_buffer(bh); 2071 put_bh(bh); 2072 } 2073 2074 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2075 { 2076 struct buffer_head *bh; 2077 struct buffer_head *latest = NULL; 2078 struct btrfs_super_block *super; 2079 int i; 2080 u64 transid = 0; 2081 u64 bytenr; 2082 2083 /* we would like to check all the supers, but that would make 2084 * a btrfs mount succeed after a mkfs from a different FS. 2085 * So, we need to add a special mount option to scan for 2086 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2087 */ 2088 for (i = 0; i < 1; i++) { 2089 bytenr = btrfs_sb_offset(i); 2090 if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2091 break; 2092 bh = __bread(bdev, bytenr / 4096, 4096); 2093 if (!bh) 2094 continue; 2095 2096 super = (struct btrfs_super_block *)bh->b_data; 2097 if (btrfs_super_bytenr(super) != bytenr || 2098 strncmp((char *)(&super->magic), BTRFS_MAGIC, 2099 sizeof(super->magic))) { 2100 brelse(bh); 2101 continue; 2102 } 2103 2104 if (!latest || btrfs_super_generation(super) > transid) { 2105 brelse(latest); 2106 latest = bh; 2107 transid = btrfs_super_generation(super); 2108 } else { 2109 brelse(bh); 2110 } 2111 } 2112 return latest; 2113 } 2114 2115 /* 2116 * this should be called twice, once with wait == 0 and 2117 * once with wait == 1. When wait == 0 is done, all the buffer heads 2118 * we write are pinned. 2119 * 2120 * They are released when wait == 1 is done. 2121 * max_mirrors must be the same for both runs, and it indicates how 2122 * many supers on this one device should be written. 2123 * 2124 * max_mirrors == 0 means to write them all. 2125 */ 2126 static int write_dev_supers(struct btrfs_device *device, 2127 struct btrfs_super_block *sb, 2128 int do_barriers, int wait, int max_mirrors) 2129 { 2130 struct buffer_head *bh; 2131 int i; 2132 int ret; 2133 int errors = 0; 2134 u32 crc; 2135 u64 bytenr; 2136 int last_barrier = 0; 2137 2138 if (max_mirrors == 0) 2139 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2140 2141 /* make sure only the last submit_bh does a barrier */ 2142 if (do_barriers) { 2143 for (i = 0; i < max_mirrors; i++) { 2144 bytenr = btrfs_sb_offset(i); 2145 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 2146 device->total_bytes) 2147 break; 2148 last_barrier = i; 2149 } 2150 } 2151 2152 for (i = 0; i < max_mirrors; i++) { 2153 bytenr = btrfs_sb_offset(i); 2154 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2155 break; 2156 2157 if (wait) { 2158 bh = __find_get_block(device->bdev, bytenr / 4096, 2159 BTRFS_SUPER_INFO_SIZE); 2160 BUG_ON(!bh); 2161 wait_on_buffer(bh); 2162 if (!buffer_uptodate(bh)) 2163 errors++; 2164 2165 /* drop our reference */ 2166 brelse(bh); 2167 2168 /* drop the reference from the wait == 0 run */ 2169 brelse(bh); 2170 continue; 2171 } else { 2172 btrfs_set_super_bytenr(sb, bytenr); 2173 2174 crc = ~(u32)0; 2175 crc = btrfs_csum_data(NULL, (char *)sb + 2176 BTRFS_CSUM_SIZE, crc, 2177 BTRFS_SUPER_INFO_SIZE - 2178 BTRFS_CSUM_SIZE); 2179 btrfs_csum_final(crc, sb->csum); 2180 2181 /* 2182 * one reference for us, and we leave it for the 2183 * caller 2184 */ 2185 bh = __getblk(device->bdev, bytenr / 4096, 2186 BTRFS_SUPER_INFO_SIZE); 2187 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2188 2189 /* one reference for submit_bh */ 2190 get_bh(bh); 2191 2192 set_buffer_uptodate(bh); 2193 lock_buffer(bh); 2194 bh->b_end_io = btrfs_end_buffer_write_sync; 2195 } 2196 2197 if (i == last_barrier && do_barriers && device->barriers) { 2198 ret = submit_bh(WRITE_BARRIER, bh); 2199 if (ret == -EOPNOTSUPP) { 2200 printk("btrfs: disabling barriers on dev %s\n", 2201 device->name); 2202 set_buffer_uptodate(bh); 2203 device->barriers = 0; 2204 /* one reference for submit_bh */ 2205 get_bh(bh); 2206 lock_buffer(bh); 2207 ret = submit_bh(WRITE_SYNC, bh); 2208 } 2209 } else { 2210 ret = submit_bh(WRITE_SYNC, bh); 2211 } 2212 2213 if (ret) 2214 errors++; 2215 } 2216 return errors < i ? 0 : -1; 2217 } 2218 2219 int write_all_supers(struct btrfs_root *root, int max_mirrors) 2220 { 2221 struct list_head *head; 2222 struct btrfs_device *dev; 2223 struct btrfs_super_block *sb; 2224 struct btrfs_dev_item *dev_item; 2225 int ret; 2226 int do_barriers; 2227 int max_errors; 2228 int total_errors = 0; 2229 u64 flags; 2230 2231 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; 2232 do_barriers = !btrfs_test_opt(root, NOBARRIER); 2233 2234 sb = &root->fs_info->super_for_commit; 2235 dev_item = &sb->dev_item; 2236 2237 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 2238 head = &root->fs_info->fs_devices->devices; 2239 list_for_each_entry(dev, head, dev_list) { 2240 if (!dev->bdev) { 2241 total_errors++; 2242 continue; 2243 } 2244 if (!dev->in_fs_metadata || !dev->writeable) 2245 continue; 2246 2247 btrfs_set_stack_device_generation(dev_item, 0); 2248 btrfs_set_stack_device_type(dev_item, dev->type); 2249 btrfs_set_stack_device_id(dev_item, dev->devid); 2250 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 2251 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 2252 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 2253 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 2254 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 2255 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 2256 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 2257 2258 flags = btrfs_super_flags(sb); 2259 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 2260 2261 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 2262 if (ret) 2263 total_errors++; 2264 } 2265 if (total_errors > max_errors) { 2266 printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2267 total_errors); 2268 BUG(); 2269 } 2270 2271 total_errors = 0; 2272 list_for_each_entry(dev, head, dev_list) { 2273 if (!dev->bdev) 2274 continue; 2275 if (!dev->in_fs_metadata || !dev->writeable) 2276 continue; 2277 2278 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 2279 if (ret) 2280 total_errors++; 2281 } 2282 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 2283 if (total_errors > max_errors) { 2284 printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2285 total_errors); 2286 BUG(); 2287 } 2288 return 0; 2289 } 2290 2291 int write_ctree_super(struct btrfs_trans_handle *trans, 2292 struct btrfs_root *root, int max_mirrors) 2293 { 2294 int ret; 2295 2296 ret = write_all_supers(root, max_mirrors); 2297 return ret; 2298 } 2299 2300 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 2301 { 2302 spin_lock(&fs_info->fs_roots_radix_lock); 2303 radix_tree_delete(&fs_info->fs_roots_radix, 2304 (unsigned long)root->root_key.objectid); 2305 spin_unlock(&fs_info->fs_roots_radix_lock); 2306 2307 if (btrfs_root_refs(&root->root_item) == 0) 2308 synchronize_srcu(&fs_info->subvol_srcu); 2309 2310 free_fs_root(root); 2311 return 0; 2312 } 2313 2314 static void free_fs_root(struct btrfs_root *root) 2315 { 2316 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 2317 if (root->anon_super.s_dev) { 2318 down_write(&root->anon_super.s_umount); 2319 kill_anon_super(&root->anon_super); 2320 } 2321 free_extent_buffer(root->node); 2322 free_extent_buffer(root->commit_root); 2323 kfree(root->name); 2324 kfree(root); 2325 } 2326 2327 static int del_fs_roots(struct btrfs_fs_info *fs_info) 2328 { 2329 int ret; 2330 struct btrfs_root *gang[8]; 2331 int i; 2332 2333 while (!list_empty(&fs_info->dead_roots)) { 2334 gang[0] = list_entry(fs_info->dead_roots.next, 2335 struct btrfs_root, root_list); 2336 list_del(&gang[0]->root_list); 2337 2338 if (gang[0]->in_radix) { 2339 btrfs_free_fs_root(fs_info, gang[0]); 2340 } else { 2341 free_extent_buffer(gang[0]->node); 2342 free_extent_buffer(gang[0]->commit_root); 2343 kfree(gang[0]); 2344 } 2345 } 2346 2347 while (1) { 2348 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2349 (void **)gang, 0, 2350 ARRAY_SIZE(gang)); 2351 if (!ret) 2352 break; 2353 for (i = 0; i < ret; i++) 2354 btrfs_free_fs_root(fs_info, gang[i]); 2355 } 2356 return 0; 2357 } 2358 2359 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 2360 { 2361 u64 root_objectid = 0; 2362 struct btrfs_root *gang[8]; 2363 int i; 2364 int ret; 2365 2366 while (1) { 2367 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2368 (void **)gang, root_objectid, 2369 ARRAY_SIZE(gang)); 2370 if (!ret) 2371 break; 2372 2373 root_objectid = gang[ret - 1]->root_key.objectid + 1; 2374 for (i = 0; i < ret; i++) { 2375 root_objectid = gang[i]->root_key.objectid; 2376 btrfs_orphan_cleanup(gang[i]); 2377 } 2378 root_objectid++; 2379 } 2380 return 0; 2381 } 2382 2383 int btrfs_commit_super(struct btrfs_root *root) 2384 { 2385 struct btrfs_trans_handle *trans; 2386 int ret; 2387 2388 mutex_lock(&root->fs_info->cleaner_mutex); 2389 btrfs_clean_old_snapshots(root); 2390 mutex_unlock(&root->fs_info->cleaner_mutex); 2391 trans = btrfs_start_transaction(root, 1); 2392 ret = btrfs_commit_transaction(trans, root); 2393 BUG_ON(ret); 2394 /* run commit again to drop the original snapshot */ 2395 trans = btrfs_start_transaction(root, 1); 2396 btrfs_commit_transaction(trans, root); 2397 ret = btrfs_write_and_wait_transaction(NULL, root); 2398 BUG_ON(ret); 2399 2400 ret = write_ctree_super(NULL, root, 0); 2401 return ret; 2402 } 2403 2404 int close_ctree(struct btrfs_root *root) 2405 { 2406 struct btrfs_fs_info *fs_info = root->fs_info; 2407 int ret; 2408 2409 fs_info->closing = 1; 2410 smp_mb(); 2411 2412 kthread_stop(root->fs_info->transaction_kthread); 2413 kthread_stop(root->fs_info->cleaner_kthread); 2414 2415 if (!(fs_info->sb->s_flags & MS_RDONLY)) { 2416 ret = btrfs_commit_super(root); 2417 if (ret) 2418 printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2419 } 2420 2421 fs_info->closing = 2; 2422 smp_mb(); 2423 2424 if (fs_info->delalloc_bytes) { 2425 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 2426 (unsigned long long)fs_info->delalloc_bytes); 2427 } 2428 if (fs_info->total_ref_cache_size) { 2429 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n", 2430 (unsigned long long)fs_info->total_ref_cache_size); 2431 } 2432 2433 free_extent_buffer(fs_info->extent_root->node); 2434 free_extent_buffer(fs_info->extent_root->commit_root); 2435 free_extent_buffer(fs_info->tree_root->node); 2436 free_extent_buffer(fs_info->tree_root->commit_root); 2437 free_extent_buffer(root->fs_info->chunk_root->node); 2438 free_extent_buffer(root->fs_info->chunk_root->commit_root); 2439 free_extent_buffer(root->fs_info->dev_root->node); 2440 free_extent_buffer(root->fs_info->dev_root->commit_root); 2441 free_extent_buffer(root->fs_info->csum_root->node); 2442 free_extent_buffer(root->fs_info->csum_root->commit_root); 2443 2444 btrfs_free_block_groups(root->fs_info); 2445 2446 del_fs_roots(fs_info); 2447 2448 iput(fs_info->btree_inode); 2449 2450 btrfs_stop_workers(&fs_info->generic_worker); 2451 btrfs_stop_workers(&fs_info->fixup_workers); 2452 btrfs_stop_workers(&fs_info->delalloc_workers); 2453 btrfs_stop_workers(&fs_info->workers); 2454 btrfs_stop_workers(&fs_info->endio_workers); 2455 btrfs_stop_workers(&fs_info->endio_meta_workers); 2456 btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2457 btrfs_stop_workers(&fs_info->endio_write_workers); 2458 btrfs_stop_workers(&fs_info->submit_workers); 2459 btrfs_stop_workers(&fs_info->enospc_workers); 2460 2461 btrfs_close_devices(fs_info->fs_devices); 2462 btrfs_mapping_tree_free(&fs_info->mapping_tree); 2463 2464 bdi_destroy(&fs_info->bdi); 2465 cleanup_srcu_struct(&fs_info->subvol_srcu); 2466 2467 kfree(fs_info->extent_root); 2468 kfree(fs_info->tree_root); 2469 kfree(fs_info->chunk_root); 2470 kfree(fs_info->dev_root); 2471 kfree(fs_info->csum_root); 2472 return 0; 2473 } 2474 2475 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid) 2476 { 2477 int ret; 2478 struct inode *btree_inode = buf->first_page->mapping->host; 2479 2480 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf); 2481 if (!ret) 2482 return ret; 2483 2484 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 2485 parent_transid); 2486 return !ret; 2487 } 2488 2489 int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 2490 { 2491 struct inode *btree_inode = buf->first_page->mapping->host; 2492 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, 2493 buf); 2494 } 2495 2496 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 2497 { 2498 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2499 u64 transid = btrfs_header_generation(buf); 2500 struct inode *btree_inode = root->fs_info->btree_inode; 2501 int was_dirty; 2502 2503 btrfs_assert_tree_locked(buf); 2504 if (transid != root->fs_info->generation) { 2505 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " 2506 "found %llu running %llu\n", 2507 (unsigned long long)buf->start, 2508 (unsigned long long)transid, 2509 (unsigned long long)root->fs_info->generation); 2510 WARN_ON(1); 2511 } 2512 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 2513 buf); 2514 if (!was_dirty) { 2515 spin_lock(&root->fs_info->delalloc_lock); 2516 root->fs_info->dirty_metadata_bytes += buf->len; 2517 spin_unlock(&root->fs_info->delalloc_lock); 2518 } 2519 } 2520 2521 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 2522 { 2523 /* 2524 * looks as though older kernels can get into trouble with 2525 * this code, they end up stuck in balance_dirty_pages forever 2526 */ 2527 u64 num_dirty; 2528 unsigned long thresh = 32 * 1024 * 1024; 2529 2530 if (current->flags & PF_MEMALLOC) 2531 return; 2532 2533 num_dirty = root->fs_info->dirty_metadata_bytes; 2534 2535 if (num_dirty > thresh) { 2536 balance_dirty_pages_ratelimited_nr( 2537 root->fs_info->btree_inode->i_mapping, 1); 2538 } 2539 return; 2540 } 2541 2542 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 2543 { 2544 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2545 int ret; 2546 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 2547 if (ret == 0) 2548 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 2549 return ret; 2550 } 2551 2552 int btree_lock_page_hook(struct page *page) 2553 { 2554 struct inode *inode = page->mapping->host; 2555 struct btrfs_root *root = BTRFS_I(inode)->root; 2556 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 2557 struct extent_buffer *eb; 2558 unsigned long len; 2559 u64 bytenr = page_offset(page); 2560 2561 if (page->private == EXTENT_PAGE_PRIVATE) 2562 goto out; 2563 2564 len = page->private >> 2; 2565 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS); 2566 if (!eb) 2567 goto out; 2568 2569 btrfs_tree_lock(eb); 2570 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 2571 2572 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { 2573 spin_lock(&root->fs_info->delalloc_lock); 2574 if (root->fs_info->dirty_metadata_bytes >= eb->len) 2575 root->fs_info->dirty_metadata_bytes -= eb->len; 2576 else 2577 WARN_ON(1); 2578 spin_unlock(&root->fs_info->delalloc_lock); 2579 } 2580 2581 btrfs_tree_unlock(eb); 2582 free_extent_buffer(eb); 2583 out: 2584 lock_page(page); 2585 return 0; 2586 } 2587 2588 static struct extent_io_ops btree_extent_io_ops = { 2589 .write_cache_pages_lock_hook = btree_lock_page_hook, 2590 .readpage_end_io_hook = btree_readpage_end_io_hook, 2591 .submit_bio_hook = btree_submit_bio_hook, 2592 /* note we're sharing with inode.c for the merge bio hook */ 2593 .merge_bio_hook = btrfs_merge_bio_hook, 2594 }; 2595