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/slab.h> 20 #include <linux/blkdev.h> 21 #include <linux/writeback.h> 22 #include <linux/pagevec.h> 23 #include "ctree.h" 24 #include "transaction.h" 25 #include "btrfs_inode.h" 26 #include "extent_io.h" 27 #include "disk-io.h" 28 29 static struct kmem_cache *btrfs_ordered_extent_cache; 30 31 static u64 entry_end(struct btrfs_ordered_extent *entry) 32 { 33 if (entry->file_offset + entry->len < entry->file_offset) 34 return (u64)-1; 35 return entry->file_offset + entry->len; 36 } 37 38 /* returns NULL if the insertion worked, or it returns the node it did find 39 * in the tree 40 */ 41 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, 42 struct rb_node *node) 43 { 44 struct rb_node **p = &root->rb_node; 45 struct rb_node *parent = NULL; 46 struct btrfs_ordered_extent *entry; 47 48 while (*p) { 49 parent = *p; 50 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); 51 52 if (file_offset < entry->file_offset) 53 p = &(*p)->rb_left; 54 else if (file_offset >= entry_end(entry)) 55 p = &(*p)->rb_right; 56 else 57 return parent; 58 } 59 60 rb_link_node(node, parent, p); 61 rb_insert_color(node, root); 62 return NULL; 63 } 64 65 static void ordered_data_tree_panic(struct inode *inode, int errno, 66 u64 offset) 67 { 68 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 69 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset " 70 "%llu", offset); 71 } 72 73 /* 74 * look for a given offset in the tree, and if it can't be found return the 75 * first lesser offset 76 */ 77 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, 78 struct rb_node **prev_ret) 79 { 80 struct rb_node *n = root->rb_node; 81 struct rb_node *prev = NULL; 82 struct rb_node *test; 83 struct btrfs_ordered_extent *entry; 84 struct btrfs_ordered_extent *prev_entry = NULL; 85 86 while (n) { 87 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); 88 prev = n; 89 prev_entry = entry; 90 91 if (file_offset < entry->file_offset) 92 n = n->rb_left; 93 else if (file_offset >= entry_end(entry)) 94 n = n->rb_right; 95 else 96 return n; 97 } 98 if (!prev_ret) 99 return NULL; 100 101 while (prev && file_offset >= entry_end(prev_entry)) { 102 test = rb_next(prev); 103 if (!test) 104 break; 105 prev_entry = rb_entry(test, struct btrfs_ordered_extent, 106 rb_node); 107 if (file_offset < entry_end(prev_entry)) 108 break; 109 110 prev = test; 111 } 112 if (prev) 113 prev_entry = rb_entry(prev, struct btrfs_ordered_extent, 114 rb_node); 115 while (prev && file_offset < entry_end(prev_entry)) { 116 test = rb_prev(prev); 117 if (!test) 118 break; 119 prev_entry = rb_entry(test, struct btrfs_ordered_extent, 120 rb_node); 121 prev = test; 122 } 123 *prev_ret = prev; 124 return NULL; 125 } 126 127 /* 128 * helper to check if a given offset is inside a given entry 129 */ 130 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) 131 { 132 if (file_offset < entry->file_offset || 133 entry->file_offset + entry->len <= file_offset) 134 return 0; 135 return 1; 136 } 137 138 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, 139 u64 len) 140 { 141 if (file_offset + len <= entry->file_offset || 142 entry->file_offset + entry->len <= file_offset) 143 return 0; 144 return 1; 145 } 146 147 /* 148 * look find the first ordered struct that has this offset, otherwise 149 * the first one less than this offset 150 */ 151 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, 152 u64 file_offset) 153 { 154 struct rb_root *root = &tree->tree; 155 struct rb_node *prev = NULL; 156 struct rb_node *ret; 157 struct btrfs_ordered_extent *entry; 158 159 if (tree->last) { 160 entry = rb_entry(tree->last, struct btrfs_ordered_extent, 161 rb_node); 162 if (offset_in_entry(entry, file_offset)) 163 return tree->last; 164 } 165 ret = __tree_search(root, file_offset, &prev); 166 if (!ret) 167 ret = prev; 168 if (ret) 169 tree->last = ret; 170 return ret; 171 } 172 173 /* allocate and add a new ordered_extent into the per-inode tree. 174 * file_offset is the logical offset in the file 175 * 176 * start is the disk block number of an extent already reserved in the 177 * extent allocation tree 178 * 179 * len is the length of the extent 180 * 181 * The tree is given a single reference on the ordered extent that was 182 * inserted. 183 */ 184 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, 185 u64 start, u64 len, u64 disk_len, 186 int type, int dio, int compress_type) 187 { 188 struct btrfs_root *root = BTRFS_I(inode)->root; 189 struct btrfs_ordered_inode_tree *tree; 190 struct rb_node *node; 191 struct btrfs_ordered_extent *entry; 192 193 tree = &BTRFS_I(inode)->ordered_tree; 194 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); 195 if (!entry) 196 return -ENOMEM; 197 198 entry->file_offset = file_offset; 199 entry->start = start; 200 entry->len = len; 201 entry->disk_len = disk_len; 202 entry->bytes_left = len; 203 entry->inode = igrab(inode); 204 entry->compress_type = compress_type; 205 entry->truncated_len = (u64)-1; 206 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) 207 set_bit(type, &entry->flags); 208 209 if (dio) 210 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); 211 212 /* one ref for the tree */ 213 atomic_set(&entry->refs, 1); 214 init_waitqueue_head(&entry->wait); 215 INIT_LIST_HEAD(&entry->list); 216 INIT_LIST_HEAD(&entry->root_extent_list); 217 INIT_LIST_HEAD(&entry->work_list); 218 init_completion(&entry->completion); 219 INIT_LIST_HEAD(&entry->log_list); 220 INIT_LIST_HEAD(&entry->trans_list); 221 222 trace_btrfs_ordered_extent_add(inode, entry); 223 224 spin_lock_irq(&tree->lock); 225 node = tree_insert(&tree->tree, file_offset, 226 &entry->rb_node); 227 if (node) 228 ordered_data_tree_panic(inode, -EEXIST, file_offset); 229 spin_unlock_irq(&tree->lock); 230 231 spin_lock(&root->ordered_extent_lock); 232 list_add_tail(&entry->root_extent_list, 233 &root->ordered_extents); 234 root->nr_ordered_extents++; 235 if (root->nr_ordered_extents == 1) { 236 spin_lock(&root->fs_info->ordered_root_lock); 237 BUG_ON(!list_empty(&root->ordered_root)); 238 list_add_tail(&root->ordered_root, 239 &root->fs_info->ordered_roots); 240 spin_unlock(&root->fs_info->ordered_root_lock); 241 } 242 spin_unlock(&root->ordered_extent_lock); 243 244 return 0; 245 } 246 247 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, 248 u64 start, u64 len, u64 disk_len, int type) 249 { 250 return __btrfs_add_ordered_extent(inode, file_offset, start, len, 251 disk_len, type, 0, 252 BTRFS_COMPRESS_NONE); 253 } 254 255 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, 256 u64 start, u64 len, u64 disk_len, int type) 257 { 258 return __btrfs_add_ordered_extent(inode, file_offset, start, len, 259 disk_len, type, 1, 260 BTRFS_COMPRESS_NONE); 261 } 262 263 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, 264 u64 start, u64 len, u64 disk_len, 265 int type, int compress_type) 266 { 267 return __btrfs_add_ordered_extent(inode, file_offset, start, len, 268 disk_len, type, 0, 269 compress_type); 270 } 271 272 /* 273 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted 274 * when an ordered extent is finished. If the list covers more than one 275 * ordered extent, it is split across multiples. 276 */ 277 void btrfs_add_ordered_sum(struct inode *inode, 278 struct btrfs_ordered_extent *entry, 279 struct btrfs_ordered_sum *sum) 280 { 281 struct btrfs_ordered_inode_tree *tree; 282 283 tree = &BTRFS_I(inode)->ordered_tree; 284 spin_lock_irq(&tree->lock); 285 list_add_tail(&sum->list, &entry->list); 286 spin_unlock_irq(&tree->lock); 287 } 288 289 /* 290 * this is used to account for finished IO across a given range 291 * of the file. The IO may span ordered extents. If 292 * a given ordered_extent is completely done, 1 is returned, otherwise 293 * 0. 294 * 295 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used 296 * to make sure this function only returns 1 once for a given ordered extent. 297 * 298 * file_offset is updated to one byte past the range that is recorded as 299 * complete. This allows you to walk forward in the file. 300 */ 301 int btrfs_dec_test_first_ordered_pending(struct inode *inode, 302 struct btrfs_ordered_extent **cached, 303 u64 *file_offset, u64 io_size, int uptodate) 304 { 305 struct btrfs_ordered_inode_tree *tree; 306 struct rb_node *node; 307 struct btrfs_ordered_extent *entry = NULL; 308 int ret; 309 unsigned long flags; 310 u64 dec_end; 311 u64 dec_start; 312 u64 to_dec; 313 314 tree = &BTRFS_I(inode)->ordered_tree; 315 spin_lock_irqsave(&tree->lock, flags); 316 node = tree_search(tree, *file_offset); 317 if (!node) { 318 ret = 1; 319 goto out; 320 } 321 322 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); 323 if (!offset_in_entry(entry, *file_offset)) { 324 ret = 1; 325 goto out; 326 } 327 328 dec_start = max(*file_offset, entry->file_offset); 329 dec_end = min(*file_offset + io_size, entry->file_offset + 330 entry->len); 331 *file_offset = dec_end; 332 if (dec_start > dec_end) { 333 btrfs_crit(BTRFS_I(inode)->root->fs_info, 334 "bad ordering dec_start %llu end %llu", dec_start, dec_end); 335 } 336 to_dec = dec_end - dec_start; 337 if (to_dec > entry->bytes_left) { 338 btrfs_crit(BTRFS_I(inode)->root->fs_info, 339 "bad ordered accounting left %llu size %llu", 340 entry->bytes_left, to_dec); 341 } 342 entry->bytes_left -= to_dec; 343 if (!uptodate) 344 set_bit(BTRFS_ORDERED_IOERR, &entry->flags); 345 346 if (entry->bytes_left == 0) { 347 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); 348 if (waitqueue_active(&entry->wait)) 349 wake_up(&entry->wait); 350 } else { 351 ret = 1; 352 } 353 out: 354 if (!ret && cached && entry) { 355 *cached = entry; 356 atomic_inc(&entry->refs); 357 } 358 spin_unlock_irqrestore(&tree->lock, flags); 359 return ret == 0; 360 } 361 362 /* 363 * this is used to account for finished IO across a given range 364 * of the file. The IO should not span ordered extents. If 365 * a given ordered_extent is completely done, 1 is returned, otherwise 366 * 0. 367 * 368 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used 369 * to make sure this function only returns 1 once for a given ordered extent. 370 */ 371 int btrfs_dec_test_ordered_pending(struct inode *inode, 372 struct btrfs_ordered_extent **cached, 373 u64 file_offset, u64 io_size, int uptodate) 374 { 375 struct btrfs_ordered_inode_tree *tree; 376 struct rb_node *node; 377 struct btrfs_ordered_extent *entry = NULL; 378 unsigned long flags; 379 int ret; 380 381 tree = &BTRFS_I(inode)->ordered_tree; 382 spin_lock_irqsave(&tree->lock, flags); 383 if (cached && *cached) { 384 entry = *cached; 385 goto have_entry; 386 } 387 388 node = tree_search(tree, file_offset); 389 if (!node) { 390 ret = 1; 391 goto out; 392 } 393 394 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); 395 have_entry: 396 if (!offset_in_entry(entry, file_offset)) { 397 ret = 1; 398 goto out; 399 } 400 401 if (io_size > entry->bytes_left) { 402 btrfs_crit(BTRFS_I(inode)->root->fs_info, 403 "bad ordered accounting left %llu size %llu", 404 entry->bytes_left, io_size); 405 } 406 entry->bytes_left -= io_size; 407 if (!uptodate) 408 set_bit(BTRFS_ORDERED_IOERR, &entry->flags); 409 410 if (entry->bytes_left == 0) { 411 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); 412 if (waitqueue_active(&entry->wait)) 413 wake_up(&entry->wait); 414 } else { 415 ret = 1; 416 } 417 out: 418 if (!ret && cached && entry) { 419 *cached = entry; 420 atomic_inc(&entry->refs); 421 } 422 spin_unlock_irqrestore(&tree->lock, flags); 423 return ret == 0; 424 } 425 426 /* Needs to either be called under a log transaction or the log_mutex */ 427 void btrfs_get_logged_extents(struct inode *inode, 428 struct list_head *logged_list, 429 const loff_t start, 430 const loff_t end) 431 { 432 struct btrfs_ordered_inode_tree *tree; 433 struct btrfs_ordered_extent *ordered; 434 struct rb_node *n; 435 struct rb_node *prev; 436 437 tree = &BTRFS_I(inode)->ordered_tree; 438 spin_lock_irq(&tree->lock); 439 n = __tree_search(&tree->tree, end, &prev); 440 if (!n) 441 n = prev; 442 for (; n; n = rb_prev(n)) { 443 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); 444 if (ordered->file_offset > end) 445 continue; 446 if (entry_end(ordered) <= start) 447 break; 448 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags)) 449 continue; 450 list_add(&ordered->log_list, logged_list); 451 atomic_inc(&ordered->refs); 452 } 453 spin_unlock_irq(&tree->lock); 454 } 455 456 void btrfs_put_logged_extents(struct list_head *logged_list) 457 { 458 struct btrfs_ordered_extent *ordered; 459 460 while (!list_empty(logged_list)) { 461 ordered = list_first_entry(logged_list, 462 struct btrfs_ordered_extent, 463 log_list); 464 list_del_init(&ordered->log_list); 465 btrfs_put_ordered_extent(ordered); 466 } 467 } 468 469 void btrfs_submit_logged_extents(struct list_head *logged_list, 470 struct btrfs_root *log) 471 { 472 int index = log->log_transid % 2; 473 474 spin_lock_irq(&log->log_extents_lock[index]); 475 list_splice_tail(logged_list, &log->logged_list[index]); 476 spin_unlock_irq(&log->log_extents_lock[index]); 477 } 478 479 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans, 480 struct btrfs_root *log, u64 transid) 481 { 482 struct btrfs_ordered_extent *ordered; 483 int index = transid % 2; 484 485 spin_lock_irq(&log->log_extents_lock[index]); 486 while (!list_empty(&log->logged_list[index])) { 487 ordered = list_first_entry(&log->logged_list[index], 488 struct btrfs_ordered_extent, 489 log_list); 490 list_del_init(&ordered->log_list); 491 spin_unlock_irq(&log->log_extents_lock[index]); 492 493 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && 494 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { 495 struct inode *inode = ordered->inode; 496 u64 start = ordered->file_offset; 497 u64 end = ordered->file_offset + ordered->len - 1; 498 499 WARN_ON(!inode); 500 filemap_fdatawrite_range(inode->i_mapping, start, end); 501 } 502 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, 503 &ordered->flags)); 504 505 /* 506 * If our ordered extent completed it means it updated the 507 * fs/subvol and csum trees already, so no need to make the 508 * current transaction's commit wait for it, as we end up 509 * holding memory unnecessarily and delaying the inode's iput 510 * until the transaction commit (we schedule an iput for the 511 * inode when the ordered extent's refcount drops to 0), which 512 * prevents it from being evictable until the transaction 513 * commits. 514 */ 515 if (test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) 516 btrfs_put_ordered_extent(ordered); 517 else 518 list_add_tail(&ordered->trans_list, &trans->ordered); 519 520 spin_lock_irq(&log->log_extents_lock[index]); 521 } 522 spin_unlock_irq(&log->log_extents_lock[index]); 523 } 524 525 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) 526 { 527 struct btrfs_ordered_extent *ordered; 528 int index = transid % 2; 529 530 spin_lock_irq(&log->log_extents_lock[index]); 531 while (!list_empty(&log->logged_list[index])) { 532 ordered = list_first_entry(&log->logged_list[index], 533 struct btrfs_ordered_extent, 534 log_list); 535 list_del_init(&ordered->log_list); 536 spin_unlock_irq(&log->log_extents_lock[index]); 537 btrfs_put_ordered_extent(ordered); 538 spin_lock_irq(&log->log_extents_lock[index]); 539 } 540 spin_unlock_irq(&log->log_extents_lock[index]); 541 } 542 543 /* 544 * used to drop a reference on an ordered extent. This will free 545 * the extent if the last reference is dropped 546 */ 547 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) 548 { 549 struct list_head *cur; 550 struct btrfs_ordered_sum *sum; 551 552 trace_btrfs_ordered_extent_put(entry->inode, entry); 553 554 if (atomic_dec_and_test(&entry->refs)) { 555 if (entry->inode) 556 btrfs_add_delayed_iput(entry->inode); 557 while (!list_empty(&entry->list)) { 558 cur = entry->list.next; 559 sum = list_entry(cur, struct btrfs_ordered_sum, list); 560 list_del(&sum->list); 561 kfree(sum); 562 } 563 kmem_cache_free(btrfs_ordered_extent_cache, entry); 564 } 565 } 566 567 /* 568 * remove an ordered extent from the tree. No references are dropped 569 * and waiters are woken up. 570 */ 571 void btrfs_remove_ordered_extent(struct inode *inode, 572 struct btrfs_ordered_extent *entry) 573 { 574 struct btrfs_ordered_inode_tree *tree; 575 struct btrfs_root *root = BTRFS_I(inode)->root; 576 struct rb_node *node; 577 578 tree = &BTRFS_I(inode)->ordered_tree; 579 spin_lock_irq(&tree->lock); 580 node = &entry->rb_node; 581 rb_erase(node, &tree->tree); 582 if (tree->last == node) 583 tree->last = NULL; 584 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); 585 spin_unlock_irq(&tree->lock); 586 587 spin_lock(&root->ordered_extent_lock); 588 list_del_init(&entry->root_extent_list); 589 root->nr_ordered_extents--; 590 591 trace_btrfs_ordered_extent_remove(inode, entry); 592 593 if (!root->nr_ordered_extents) { 594 spin_lock(&root->fs_info->ordered_root_lock); 595 BUG_ON(list_empty(&root->ordered_root)); 596 list_del_init(&root->ordered_root); 597 spin_unlock(&root->fs_info->ordered_root_lock); 598 } 599 spin_unlock(&root->ordered_extent_lock); 600 wake_up(&entry->wait); 601 } 602 603 static void btrfs_run_ordered_extent_work(struct btrfs_work *work) 604 { 605 struct btrfs_ordered_extent *ordered; 606 607 ordered = container_of(work, struct btrfs_ordered_extent, flush_work); 608 btrfs_start_ordered_extent(ordered->inode, ordered, 1); 609 complete(&ordered->completion); 610 } 611 612 /* 613 * wait for all the ordered extents in a root. This is done when balancing 614 * space between drives. 615 */ 616 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr) 617 { 618 struct list_head splice, works; 619 struct btrfs_ordered_extent *ordered, *next; 620 int count = 0; 621 622 INIT_LIST_HEAD(&splice); 623 INIT_LIST_HEAD(&works); 624 625 mutex_lock(&root->ordered_extent_mutex); 626 spin_lock(&root->ordered_extent_lock); 627 list_splice_init(&root->ordered_extents, &splice); 628 while (!list_empty(&splice) && nr) { 629 ordered = list_first_entry(&splice, struct btrfs_ordered_extent, 630 root_extent_list); 631 list_move_tail(&ordered->root_extent_list, 632 &root->ordered_extents); 633 atomic_inc(&ordered->refs); 634 spin_unlock(&root->ordered_extent_lock); 635 636 btrfs_init_work(&ordered->flush_work, 637 btrfs_flush_delalloc_helper, 638 btrfs_run_ordered_extent_work, NULL, NULL); 639 list_add_tail(&ordered->work_list, &works); 640 btrfs_queue_work(root->fs_info->flush_workers, 641 &ordered->flush_work); 642 643 cond_resched(); 644 spin_lock(&root->ordered_extent_lock); 645 if (nr != -1) 646 nr--; 647 count++; 648 } 649 list_splice_tail(&splice, &root->ordered_extents); 650 spin_unlock(&root->ordered_extent_lock); 651 652 list_for_each_entry_safe(ordered, next, &works, work_list) { 653 list_del_init(&ordered->work_list); 654 wait_for_completion(&ordered->completion); 655 btrfs_put_ordered_extent(ordered); 656 cond_resched(); 657 } 658 mutex_unlock(&root->ordered_extent_mutex); 659 660 return count; 661 } 662 663 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr) 664 { 665 struct btrfs_root *root; 666 struct list_head splice; 667 int done; 668 669 INIT_LIST_HEAD(&splice); 670 671 mutex_lock(&fs_info->ordered_operations_mutex); 672 spin_lock(&fs_info->ordered_root_lock); 673 list_splice_init(&fs_info->ordered_roots, &splice); 674 while (!list_empty(&splice) && nr) { 675 root = list_first_entry(&splice, struct btrfs_root, 676 ordered_root); 677 root = btrfs_grab_fs_root(root); 678 BUG_ON(!root); 679 list_move_tail(&root->ordered_root, 680 &fs_info->ordered_roots); 681 spin_unlock(&fs_info->ordered_root_lock); 682 683 done = btrfs_wait_ordered_extents(root, nr); 684 btrfs_put_fs_root(root); 685 686 spin_lock(&fs_info->ordered_root_lock); 687 if (nr != -1) { 688 nr -= done; 689 WARN_ON(nr < 0); 690 } 691 } 692 list_splice_tail(&splice, &fs_info->ordered_roots); 693 spin_unlock(&fs_info->ordered_root_lock); 694 mutex_unlock(&fs_info->ordered_operations_mutex); 695 } 696 697 /* 698 * Used to start IO or wait for a given ordered extent to finish. 699 * 700 * If wait is one, this effectively waits on page writeback for all the pages 701 * in the extent, and it waits on the io completion code to insert 702 * metadata into the btree corresponding to the extent 703 */ 704 void btrfs_start_ordered_extent(struct inode *inode, 705 struct btrfs_ordered_extent *entry, 706 int wait) 707 { 708 u64 start = entry->file_offset; 709 u64 end = start + entry->len - 1; 710 711 trace_btrfs_ordered_extent_start(inode, entry); 712 713 /* 714 * pages in the range can be dirty, clean or writeback. We 715 * start IO on any dirty ones so the wait doesn't stall waiting 716 * for the flusher thread to find them 717 */ 718 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) 719 filemap_fdatawrite_range(inode->i_mapping, start, end); 720 if (wait) { 721 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, 722 &entry->flags)); 723 } 724 } 725 726 /* 727 * Used to wait on ordered extents across a large range of bytes. 728 */ 729 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) 730 { 731 int ret = 0; 732 int ret_wb = 0; 733 u64 end; 734 u64 orig_end; 735 struct btrfs_ordered_extent *ordered; 736 737 if (start + len < start) { 738 orig_end = INT_LIMIT(loff_t); 739 } else { 740 orig_end = start + len - 1; 741 if (orig_end > INT_LIMIT(loff_t)) 742 orig_end = INT_LIMIT(loff_t); 743 } 744 745 /* start IO across the range first to instantiate any delalloc 746 * extents 747 */ 748 ret = btrfs_fdatawrite_range(inode, start, orig_end); 749 if (ret) 750 return ret; 751 752 /* 753 * If we have a writeback error don't return immediately. Wait first 754 * for any ordered extents that haven't completed yet. This is to make 755 * sure no one can dirty the same page ranges and call writepages() 756 * before the ordered extents complete - to avoid failures (-EEXIST) 757 * when adding the new ordered extents to the ordered tree. 758 */ 759 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end); 760 761 end = orig_end; 762 while (1) { 763 ordered = btrfs_lookup_first_ordered_extent(inode, end); 764 if (!ordered) 765 break; 766 if (ordered->file_offset > orig_end) { 767 btrfs_put_ordered_extent(ordered); 768 break; 769 } 770 if (ordered->file_offset + ordered->len <= start) { 771 btrfs_put_ordered_extent(ordered); 772 break; 773 } 774 btrfs_start_ordered_extent(inode, ordered, 1); 775 end = ordered->file_offset; 776 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) 777 ret = -EIO; 778 btrfs_put_ordered_extent(ordered); 779 if (ret || end == 0 || end == start) 780 break; 781 end--; 782 } 783 return ret_wb ? ret_wb : ret; 784 } 785 786 /* 787 * find an ordered extent corresponding to file_offset. return NULL if 788 * nothing is found, otherwise take a reference on the extent and return it 789 */ 790 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, 791 u64 file_offset) 792 { 793 struct btrfs_ordered_inode_tree *tree; 794 struct rb_node *node; 795 struct btrfs_ordered_extent *entry = NULL; 796 797 tree = &BTRFS_I(inode)->ordered_tree; 798 spin_lock_irq(&tree->lock); 799 node = tree_search(tree, file_offset); 800 if (!node) 801 goto out; 802 803 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); 804 if (!offset_in_entry(entry, file_offset)) 805 entry = NULL; 806 if (entry) 807 atomic_inc(&entry->refs); 808 out: 809 spin_unlock_irq(&tree->lock); 810 return entry; 811 } 812 813 /* Since the DIO code tries to lock a wide area we need to look for any ordered 814 * extents that exist in the range, rather than just the start of the range. 815 */ 816 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, 817 u64 file_offset, 818 u64 len) 819 { 820 struct btrfs_ordered_inode_tree *tree; 821 struct rb_node *node; 822 struct btrfs_ordered_extent *entry = NULL; 823 824 tree = &BTRFS_I(inode)->ordered_tree; 825 spin_lock_irq(&tree->lock); 826 node = tree_search(tree, file_offset); 827 if (!node) { 828 node = tree_search(tree, file_offset + len); 829 if (!node) 830 goto out; 831 } 832 833 while (1) { 834 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); 835 if (range_overlaps(entry, file_offset, len)) 836 break; 837 838 if (entry->file_offset >= file_offset + len) { 839 entry = NULL; 840 break; 841 } 842 entry = NULL; 843 node = rb_next(node); 844 if (!node) 845 break; 846 } 847 out: 848 if (entry) 849 atomic_inc(&entry->refs); 850 spin_unlock_irq(&tree->lock); 851 return entry; 852 } 853 854 bool btrfs_have_ordered_extents_in_range(struct inode *inode, 855 u64 file_offset, 856 u64 len) 857 { 858 struct btrfs_ordered_extent *oe; 859 860 oe = btrfs_lookup_ordered_range(inode, file_offset, len); 861 if (oe) { 862 btrfs_put_ordered_extent(oe); 863 return true; 864 } 865 return false; 866 } 867 868 /* 869 * lookup and return any extent before 'file_offset'. NULL is returned 870 * if none is found 871 */ 872 struct btrfs_ordered_extent * 873 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) 874 { 875 struct btrfs_ordered_inode_tree *tree; 876 struct rb_node *node; 877 struct btrfs_ordered_extent *entry = NULL; 878 879 tree = &BTRFS_I(inode)->ordered_tree; 880 spin_lock_irq(&tree->lock); 881 node = tree_search(tree, file_offset); 882 if (!node) 883 goto out; 884 885 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); 886 atomic_inc(&entry->refs); 887 out: 888 spin_unlock_irq(&tree->lock); 889 return entry; 890 } 891 892 /* 893 * After an extent is done, call this to conditionally update the on disk 894 * i_size. i_size is updated to cover any fully written part of the file. 895 */ 896 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, 897 struct btrfs_ordered_extent *ordered) 898 { 899 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; 900 u64 disk_i_size; 901 u64 new_i_size; 902 u64 i_size = i_size_read(inode); 903 struct rb_node *node; 904 struct rb_node *prev = NULL; 905 struct btrfs_ordered_extent *test; 906 int ret = 1; 907 908 spin_lock_irq(&tree->lock); 909 if (ordered) { 910 offset = entry_end(ordered); 911 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) 912 offset = min(offset, 913 ordered->file_offset + 914 ordered->truncated_len); 915 } else { 916 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); 917 } 918 disk_i_size = BTRFS_I(inode)->disk_i_size; 919 920 /* truncate file */ 921 if (disk_i_size > i_size) { 922 BTRFS_I(inode)->disk_i_size = i_size; 923 ret = 0; 924 goto out; 925 } 926 927 /* 928 * if the disk i_size is already at the inode->i_size, or 929 * this ordered extent is inside the disk i_size, we're done 930 */ 931 if (disk_i_size == i_size) 932 goto out; 933 934 /* 935 * We still need to update disk_i_size if outstanding_isize is greater 936 * than disk_i_size. 937 */ 938 if (offset <= disk_i_size && 939 (!ordered || ordered->outstanding_isize <= disk_i_size)) 940 goto out; 941 942 /* 943 * walk backward from this ordered extent to disk_i_size. 944 * if we find an ordered extent then we can't update disk i_size 945 * yet 946 */ 947 if (ordered) { 948 node = rb_prev(&ordered->rb_node); 949 } else { 950 prev = tree_search(tree, offset); 951 /* 952 * we insert file extents without involving ordered struct, 953 * so there should be no ordered struct cover this offset 954 */ 955 if (prev) { 956 test = rb_entry(prev, struct btrfs_ordered_extent, 957 rb_node); 958 BUG_ON(offset_in_entry(test, offset)); 959 } 960 node = prev; 961 } 962 for (; node; node = rb_prev(node)) { 963 test = rb_entry(node, struct btrfs_ordered_extent, rb_node); 964 965 /* We treat this entry as if it doesnt exist */ 966 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) 967 continue; 968 if (test->file_offset + test->len <= disk_i_size) 969 break; 970 if (test->file_offset >= i_size) 971 break; 972 if (entry_end(test) > disk_i_size) { 973 /* 974 * we don't update disk_i_size now, so record this 975 * undealt i_size. Or we will not know the real 976 * i_size. 977 */ 978 if (test->outstanding_isize < offset) 979 test->outstanding_isize = offset; 980 if (ordered && 981 ordered->outstanding_isize > 982 test->outstanding_isize) 983 test->outstanding_isize = 984 ordered->outstanding_isize; 985 goto out; 986 } 987 } 988 new_i_size = min_t(u64, offset, i_size); 989 990 /* 991 * Some ordered extents may completed before the current one, and 992 * we hold the real i_size in ->outstanding_isize. 993 */ 994 if (ordered && ordered->outstanding_isize > new_i_size) 995 new_i_size = min_t(u64, ordered->outstanding_isize, i_size); 996 BTRFS_I(inode)->disk_i_size = new_i_size; 997 ret = 0; 998 out: 999 /* 1000 * We need to do this because we can't remove ordered extents until 1001 * after the i_disk_size has been updated and then the inode has been 1002 * updated to reflect the change, so we need to tell anybody who finds 1003 * this ordered extent that we've already done all the real work, we 1004 * just haven't completed all the other work. 1005 */ 1006 if (ordered) 1007 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); 1008 spin_unlock_irq(&tree->lock); 1009 return ret; 1010 } 1011 1012 /* 1013 * search the ordered extents for one corresponding to 'offset' and 1014 * try to find a checksum. This is used because we allow pages to 1015 * be reclaimed before their checksum is actually put into the btree 1016 */ 1017 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, 1018 u32 *sum, int len) 1019 { 1020 struct btrfs_ordered_sum *ordered_sum; 1021 struct btrfs_ordered_extent *ordered; 1022 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; 1023 unsigned long num_sectors; 1024 unsigned long i; 1025 u32 sectorsize = BTRFS_I(inode)->root->sectorsize; 1026 int index = 0; 1027 1028 ordered = btrfs_lookup_ordered_extent(inode, offset); 1029 if (!ordered) 1030 return 0; 1031 1032 spin_lock_irq(&tree->lock); 1033 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { 1034 if (disk_bytenr >= ordered_sum->bytenr && 1035 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) { 1036 i = (disk_bytenr - ordered_sum->bytenr) >> 1037 inode->i_sb->s_blocksize_bits; 1038 num_sectors = ordered_sum->len >> 1039 inode->i_sb->s_blocksize_bits; 1040 num_sectors = min_t(int, len - index, num_sectors - i); 1041 memcpy(sum + index, ordered_sum->sums + i, 1042 num_sectors); 1043 1044 index += (int)num_sectors; 1045 if (index == len) 1046 goto out; 1047 disk_bytenr += num_sectors * sectorsize; 1048 } 1049 } 1050 out: 1051 spin_unlock_irq(&tree->lock); 1052 btrfs_put_ordered_extent(ordered); 1053 return index; 1054 } 1055 1056 int __init ordered_data_init(void) 1057 { 1058 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", 1059 sizeof(struct btrfs_ordered_extent), 0, 1060 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, 1061 NULL); 1062 if (!btrfs_ordered_extent_cache) 1063 return -ENOMEM; 1064 1065 return 0; 1066 } 1067 1068 void ordered_data_exit(void) 1069 { 1070 if (btrfs_ordered_extent_cache) 1071 kmem_cache_destroy(btrfs_ordered_extent_cache); 1072 } 1073