1 #include <linux/bitops.h> 2 #include <linux/slab.h> 3 #include <linux/bio.h> 4 #include <linux/mm.h> 5 #include <linux/gfp.h> 6 #include <linux/pagemap.h> 7 #include <linux/page-flags.h> 8 #include <linux/module.h> 9 #include <linux/spinlock.h> 10 #include <linux/blkdev.h> 11 #include <linux/swap.h> 12 #include <linux/version.h> 13 #include <linux/writeback.h> 14 #include <linux/pagevec.h> 15 #include "extent_map.h" 16 17 /* temporary define until extent_map moves out of btrfs */ 18 struct kmem_cache *btrfs_cache_create(const char *name, size_t size, 19 unsigned long extra_flags, 20 void (*ctor)(void *, struct kmem_cache *, 21 unsigned long)); 22 23 static struct kmem_cache *extent_map_cache; 24 static struct kmem_cache *extent_state_cache; 25 static struct kmem_cache *extent_buffer_cache; 26 27 static LIST_HEAD(buffers); 28 static LIST_HEAD(states); 29 30 static spinlock_t state_lock = SPIN_LOCK_UNLOCKED; 31 #define BUFFER_LRU_MAX 64 32 33 struct tree_entry { 34 u64 start; 35 u64 end; 36 int in_tree; 37 struct rb_node rb_node; 38 }; 39 40 struct extent_page_data { 41 struct bio *bio; 42 struct extent_map_tree *tree; 43 get_extent_t *get_extent; 44 }; 45 46 int __init extent_map_init(void) 47 { 48 extent_map_cache = btrfs_cache_create("extent_map", 49 sizeof(struct extent_map), 0, 50 NULL); 51 if (!extent_map_cache) 52 return -ENOMEM; 53 extent_state_cache = btrfs_cache_create("extent_state", 54 sizeof(struct extent_state), 0, 55 NULL); 56 if (!extent_state_cache) 57 goto free_map_cache; 58 extent_buffer_cache = btrfs_cache_create("extent_buffers", 59 sizeof(struct extent_buffer), 0, 60 NULL); 61 if (!extent_buffer_cache) 62 goto free_state_cache; 63 return 0; 64 65 free_state_cache: 66 kmem_cache_destroy(extent_state_cache); 67 free_map_cache: 68 kmem_cache_destroy(extent_map_cache); 69 return -ENOMEM; 70 } 71 72 void extent_map_exit(void) 73 { 74 struct extent_state *state; 75 76 while (!list_empty(&states)) { 77 state = list_entry(states.next, struct extent_state, list); 78 printk("state leak: start %Lu end %Lu state %lu in tree %d refs %d\n", state->start, state->end, state->state, state->in_tree, atomic_read(&state->refs)); 79 list_del(&state->list); 80 kmem_cache_free(extent_state_cache, state); 81 82 } 83 84 if (extent_map_cache) 85 kmem_cache_destroy(extent_map_cache); 86 if (extent_state_cache) 87 kmem_cache_destroy(extent_state_cache); 88 if (extent_buffer_cache) 89 kmem_cache_destroy(extent_buffer_cache); 90 } 91 92 void extent_map_tree_init(struct extent_map_tree *tree, 93 struct address_space *mapping, gfp_t mask) 94 { 95 tree->map.rb_node = NULL; 96 tree->state.rb_node = NULL; 97 tree->ops = NULL; 98 tree->dirty_bytes = 0; 99 rwlock_init(&tree->lock); 100 spin_lock_init(&tree->lru_lock); 101 tree->mapping = mapping; 102 INIT_LIST_HEAD(&tree->buffer_lru); 103 tree->lru_size = 0; 104 } 105 EXPORT_SYMBOL(extent_map_tree_init); 106 107 void extent_map_tree_empty_lru(struct extent_map_tree *tree) 108 { 109 struct extent_buffer *eb; 110 while(!list_empty(&tree->buffer_lru)) { 111 eb = list_entry(tree->buffer_lru.next, struct extent_buffer, 112 lru); 113 list_del_init(&eb->lru); 114 free_extent_buffer(eb); 115 } 116 } 117 EXPORT_SYMBOL(extent_map_tree_empty_lru); 118 119 struct extent_map *alloc_extent_map(gfp_t mask) 120 { 121 struct extent_map *em; 122 em = kmem_cache_alloc(extent_map_cache, mask); 123 if (!em || IS_ERR(em)) 124 return em; 125 em->in_tree = 0; 126 atomic_set(&em->refs, 1); 127 return em; 128 } 129 EXPORT_SYMBOL(alloc_extent_map); 130 131 void free_extent_map(struct extent_map *em) 132 { 133 if (!em) 134 return; 135 if (atomic_dec_and_test(&em->refs)) { 136 WARN_ON(em->in_tree); 137 kmem_cache_free(extent_map_cache, em); 138 } 139 } 140 EXPORT_SYMBOL(free_extent_map); 141 142 143 struct extent_state *alloc_extent_state(gfp_t mask) 144 { 145 struct extent_state *state; 146 unsigned long flags; 147 148 state = kmem_cache_alloc(extent_state_cache, mask); 149 if (!state || IS_ERR(state)) 150 return state; 151 state->state = 0; 152 state->in_tree = 0; 153 state->private = 0; 154 155 spin_lock_irqsave(&state_lock, flags); 156 list_add(&state->list, &states); 157 spin_unlock_irqrestore(&state_lock, flags); 158 159 atomic_set(&state->refs, 1); 160 init_waitqueue_head(&state->wq); 161 return state; 162 } 163 EXPORT_SYMBOL(alloc_extent_state); 164 165 void free_extent_state(struct extent_state *state) 166 { 167 unsigned long flags; 168 if (!state) 169 return; 170 if (atomic_dec_and_test(&state->refs)) { 171 WARN_ON(state->in_tree); 172 spin_lock_irqsave(&state_lock, flags); 173 list_del(&state->list); 174 spin_unlock_irqrestore(&state_lock, flags); 175 kmem_cache_free(extent_state_cache, state); 176 } 177 } 178 EXPORT_SYMBOL(free_extent_state); 179 180 static struct rb_node *tree_insert(struct rb_root *root, u64 offset, 181 struct rb_node *node) 182 { 183 struct rb_node ** p = &root->rb_node; 184 struct rb_node * parent = NULL; 185 struct tree_entry *entry; 186 187 while(*p) { 188 parent = *p; 189 entry = rb_entry(parent, struct tree_entry, rb_node); 190 191 if (offset < entry->start) 192 p = &(*p)->rb_left; 193 else if (offset > entry->end) 194 p = &(*p)->rb_right; 195 else 196 return parent; 197 } 198 199 entry = rb_entry(node, struct tree_entry, rb_node); 200 entry->in_tree = 1; 201 rb_link_node(node, parent, p); 202 rb_insert_color(node, root); 203 return NULL; 204 } 205 206 static struct rb_node *__tree_search(struct rb_root *root, u64 offset, 207 struct rb_node **prev_ret) 208 { 209 struct rb_node * n = root->rb_node; 210 struct rb_node *prev = NULL; 211 struct tree_entry *entry; 212 struct tree_entry *prev_entry = NULL; 213 214 while(n) { 215 entry = rb_entry(n, struct tree_entry, rb_node); 216 prev = n; 217 prev_entry = entry; 218 219 if (offset < entry->start) 220 n = n->rb_left; 221 else if (offset > entry->end) 222 n = n->rb_right; 223 else 224 return n; 225 } 226 if (!prev_ret) 227 return NULL; 228 while(prev && offset > prev_entry->end) { 229 prev = rb_next(prev); 230 prev_entry = rb_entry(prev, struct tree_entry, rb_node); 231 } 232 *prev_ret = prev; 233 return NULL; 234 } 235 236 static inline struct rb_node *tree_search(struct rb_root *root, u64 offset) 237 { 238 struct rb_node *prev; 239 struct rb_node *ret; 240 ret = __tree_search(root, offset, &prev); 241 if (!ret) 242 return prev; 243 return ret; 244 } 245 246 static int tree_delete(struct rb_root *root, u64 offset) 247 { 248 struct rb_node *node; 249 struct tree_entry *entry; 250 251 node = __tree_search(root, offset, NULL); 252 if (!node) 253 return -ENOENT; 254 entry = rb_entry(node, struct tree_entry, rb_node); 255 entry->in_tree = 0; 256 rb_erase(node, root); 257 return 0; 258 } 259 260 /* 261 * add_extent_mapping tries a simple backward merge with existing 262 * mappings. The extent_map struct passed in will be inserted into 263 * the tree directly (no copies made, just a reference taken). 264 */ 265 int add_extent_mapping(struct extent_map_tree *tree, 266 struct extent_map *em) 267 { 268 int ret = 0; 269 struct extent_map *prev = NULL; 270 struct rb_node *rb; 271 272 write_lock_irq(&tree->lock); 273 rb = tree_insert(&tree->map, em->end, &em->rb_node); 274 if (rb) { 275 prev = rb_entry(rb, struct extent_map, rb_node); 276 ret = -EEXIST; 277 goto out; 278 } 279 atomic_inc(&em->refs); 280 if (em->start != 0) { 281 rb = rb_prev(&em->rb_node); 282 if (rb) 283 prev = rb_entry(rb, struct extent_map, rb_node); 284 if (prev && prev->end + 1 == em->start && 285 ((em->block_start == EXTENT_MAP_HOLE && 286 prev->block_start == EXTENT_MAP_HOLE) || 287 (em->block_start == EXTENT_MAP_INLINE && 288 prev->block_start == EXTENT_MAP_INLINE) || 289 (em->block_start == EXTENT_MAP_DELALLOC && 290 prev->block_start == EXTENT_MAP_DELALLOC) || 291 (em->block_start < EXTENT_MAP_DELALLOC - 1 && 292 em->block_start == prev->block_end + 1))) { 293 em->start = prev->start; 294 em->block_start = prev->block_start; 295 rb_erase(&prev->rb_node, &tree->map); 296 prev->in_tree = 0; 297 free_extent_map(prev); 298 } 299 } 300 out: 301 write_unlock_irq(&tree->lock); 302 return ret; 303 } 304 EXPORT_SYMBOL(add_extent_mapping); 305 306 /* 307 * lookup_extent_mapping returns the first extent_map struct in the 308 * tree that intersects the [start, end] (inclusive) range. There may 309 * be additional objects in the tree that intersect, so check the object 310 * returned carefully to make sure you don't need additional lookups. 311 */ 312 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree, 313 u64 start, u64 end) 314 { 315 struct extent_map *em; 316 struct rb_node *rb_node; 317 318 read_lock_irq(&tree->lock); 319 rb_node = tree_search(&tree->map, start); 320 if (!rb_node) { 321 em = NULL; 322 goto out; 323 } 324 if (IS_ERR(rb_node)) { 325 em = ERR_PTR(PTR_ERR(rb_node)); 326 goto out; 327 } 328 em = rb_entry(rb_node, struct extent_map, rb_node); 329 if (em->end < start || em->start > end) { 330 em = NULL; 331 goto out; 332 } 333 atomic_inc(&em->refs); 334 out: 335 read_unlock_irq(&tree->lock); 336 return em; 337 } 338 EXPORT_SYMBOL(lookup_extent_mapping); 339 340 /* 341 * removes an extent_map struct from the tree. No reference counts are 342 * dropped, and no checks are done to see if the range is in use 343 */ 344 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em) 345 { 346 int ret; 347 348 write_lock_irq(&tree->lock); 349 ret = tree_delete(&tree->map, em->end); 350 write_unlock_irq(&tree->lock); 351 return ret; 352 } 353 EXPORT_SYMBOL(remove_extent_mapping); 354 355 /* 356 * utility function to look for merge candidates inside a given range. 357 * Any extents with matching state are merged together into a single 358 * extent in the tree. Extents with EXTENT_IO in their state field 359 * are not merged because the end_io handlers need to be able to do 360 * operations on them without sleeping (or doing allocations/splits). 361 * 362 * This should be called with the tree lock held. 363 */ 364 static int merge_state(struct extent_map_tree *tree, 365 struct extent_state *state) 366 { 367 struct extent_state *other; 368 struct rb_node *other_node; 369 370 if (state->state & EXTENT_IOBITS) 371 return 0; 372 373 other_node = rb_prev(&state->rb_node); 374 if (other_node) { 375 other = rb_entry(other_node, struct extent_state, rb_node); 376 if (other->end == state->start - 1 && 377 other->state == state->state) { 378 state->start = other->start; 379 other->in_tree = 0; 380 rb_erase(&other->rb_node, &tree->state); 381 free_extent_state(other); 382 } 383 } 384 other_node = rb_next(&state->rb_node); 385 if (other_node) { 386 other = rb_entry(other_node, struct extent_state, rb_node); 387 if (other->start == state->end + 1 && 388 other->state == state->state) { 389 other->start = state->start; 390 state->in_tree = 0; 391 rb_erase(&state->rb_node, &tree->state); 392 free_extent_state(state); 393 } 394 } 395 return 0; 396 } 397 398 /* 399 * insert an extent_state struct into the tree. 'bits' are set on the 400 * struct before it is inserted. 401 * 402 * This may return -EEXIST if the extent is already there, in which case the 403 * state struct is freed. 404 * 405 * The tree lock is not taken internally. This is a utility function and 406 * probably isn't what you want to call (see set/clear_extent_bit). 407 */ 408 static int insert_state(struct extent_map_tree *tree, 409 struct extent_state *state, u64 start, u64 end, 410 int bits) 411 { 412 struct rb_node *node; 413 414 if (end < start) { 415 printk("end < start %Lu %Lu\n", end, start); 416 WARN_ON(1); 417 } 418 if (bits & EXTENT_DIRTY) 419 tree->dirty_bytes += end - start + 1; 420 state->state |= bits; 421 state->start = start; 422 state->end = end; 423 node = tree_insert(&tree->state, end, &state->rb_node); 424 if (node) { 425 struct extent_state *found; 426 found = rb_entry(node, struct extent_state, rb_node); 427 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end); 428 free_extent_state(state); 429 return -EEXIST; 430 } 431 merge_state(tree, state); 432 return 0; 433 } 434 435 /* 436 * split a given extent state struct in two, inserting the preallocated 437 * struct 'prealloc' as the newly created second half. 'split' indicates an 438 * offset inside 'orig' where it should be split. 439 * 440 * Before calling, 441 * the tree has 'orig' at [orig->start, orig->end]. After calling, there 442 * are two extent state structs in the tree: 443 * prealloc: [orig->start, split - 1] 444 * orig: [ split, orig->end ] 445 * 446 * The tree locks are not taken by this function. They need to be held 447 * by the caller. 448 */ 449 static int split_state(struct extent_map_tree *tree, struct extent_state *orig, 450 struct extent_state *prealloc, u64 split) 451 { 452 struct rb_node *node; 453 prealloc->start = orig->start; 454 prealloc->end = split - 1; 455 prealloc->state = orig->state; 456 orig->start = split; 457 458 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node); 459 if (node) { 460 struct extent_state *found; 461 found = rb_entry(node, struct extent_state, rb_node); 462 printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end); 463 free_extent_state(prealloc); 464 return -EEXIST; 465 } 466 return 0; 467 } 468 469 /* 470 * utility function to clear some bits in an extent state struct. 471 * it will optionally wake up any one waiting on this state (wake == 1), or 472 * forcibly remove the state from the tree (delete == 1). 473 * 474 * If no bits are set on the state struct after clearing things, the 475 * struct is freed and removed from the tree 476 */ 477 static int clear_state_bit(struct extent_map_tree *tree, 478 struct extent_state *state, int bits, int wake, 479 int delete) 480 { 481 int ret = state->state & bits; 482 483 if ((bits & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { 484 u64 range = state->end - state->start + 1; 485 WARN_ON(range > tree->dirty_bytes); 486 tree->dirty_bytes -= range; 487 } 488 state->state &= ~bits; 489 if (wake) 490 wake_up(&state->wq); 491 if (delete || state->state == 0) { 492 if (state->in_tree) { 493 rb_erase(&state->rb_node, &tree->state); 494 state->in_tree = 0; 495 free_extent_state(state); 496 } else { 497 WARN_ON(1); 498 } 499 } else { 500 merge_state(tree, state); 501 } 502 return ret; 503 } 504 505 /* 506 * clear some bits on a range in the tree. This may require splitting 507 * or inserting elements in the tree, so the gfp mask is used to 508 * indicate which allocations or sleeping are allowed. 509 * 510 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove 511 * the given range from the tree regardless of state (ie for truncate). 512 * 513 * the range [start, end] is inclusive. 514 * 515 * This takes the tree lock, and returns < 0 on error, > 0 if any of the 516 * bits were already set, or zero if none of the bits were already set. 517 */ 518 int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, 519 int bits, int wake, int delete, gfp_t mask) 520 { 521 struct extent_state *state; 522 struct extent_state *prealloc = NULL; 523 struct rb_node *node; 524 unsigned long flags; 525 int err; 526 int set = 0; 527 528 again: 529 if (!prealloc && (mask & __GFP_WAIT)) { 530 prealloc = alloc_extent_state(mask); 531 if (!prealloc) 532 return -ENOMEM; 533 } 534 535 write_lock_irqsave(&tree->lock, flags); 536 /* 537 * this search will find the extents that end after 538 * our range starts 539 */ 540 node = tree_search(&tree->state, start); 541 if (!node) 542 goto out; 543 state = rb_entry(node, struct extent_state, rb_node); 544 if (state->start > end) 545 goto out; 546 WARN_ON(state->end < start); 547 548 /* 549 * | ---- desired range ---- | 550 * | state | or 551 * | ------------- state -------------- | 552 * 553 * We need to split the extent we found, and may flip 554 * bits on second half. 555 * 556 * If the extent we found extends past our range, we 557 * just split and search again. It'll get split again 558 * the next time though. 559 * 560 * If the extent we found is inside our range, we clear 561 * the desired bit on it. 562 */ 563 564 if (state->start < start) { 565 err = split_state(tree, state, prealloc, start); 566 BUG_ON(err == -EEXIST); 567 prealloc = NULL; 568 if (err) 569 goto out; 570 if (state->end <= end) { 571 start = state->end + 1; 572 set |= clear_state_bit(tree, state, bits, 573 wake, delete); 574 } else { 575 start = state->start; 576 } 577 goto search_again; 578 } 579 /* 580 * | ---- desired range ---- | 581 * | state | 582 * We need to split the extent, and clear the bit 583 * on the first half 584 */ 585 if (state->start <= end && state->end > end) { 586 err = split_state(tree, state, prealloc, end + 1); 587 BUG_ON(err == -EEXIST); 588 589 if (wake) 590 wake_up(&state->wq); 591 set |= clear_state_bit(tree, prealloc, bits, 592 wake, delete); 593 prealloc = NULL; 594 goto out; 595 } 596 597 start = state->end + 1; 598 set |= clear_state_bit(tree, state, bits, wake, delete); 599 goto search_again; 600 601 out: 602 write_unlock_irqrestore(&tree->lock, flags); 603 if (prealloc) 604 free_extent_state(prealloc); 605 606 return set; 607 608 search_again: 609 if (start > end) 610 goto out; 611 write_unlock_irqrestore(&tree->lock, flags); 612 if (mask & __GFP_WAIT) 613 cond_resched(); 614 goto again; 615 } 616 EXPORT_SYMBOL(clear_extent_bit); 617 618 static int wait_on_state(struct extent_map_tree *tree, 619 struct extent_state *state) 620 { 621 DEFINE_WAIT(wait); 622 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); 623 read_unlock_irq(&tree->lock); 624 schedule(); 625 read_lock_irq(&tree->lock); 626 finish_wait(&state->wq, &wait); 627 return 0; 628 } 629 630 /* 631 * waits for one or more bits to clear on a range in the state tree. 632 * The range [start, end] is inclusive. 633 * The tree lock is taken by this function 634 */ 635 int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits) 636 { 637 struct extent_state *state; 638 struct rb_node *node; 639 640 read_lock_irq(&tree->lock); 641 again: 642 while (1) { 643 /* 644 * this search will find all the extents that end after 645 * our range starts 646 */ 647 node = tree_search(&tree->state, start); 648 if (!node) 649 break; 650 651 state = rb_entry(node, struct extent_state, rb_node); 652 653 if (state->start > end) 654 goto out; 655 656 if (state->state & bits) { 657 start = state->start; 658 atomic_inc(&state->refs); 659 wait_on_state(tree, state); 660 free_extent_state(state); 661 goto again; 662 } 663 start = state->end + 1; 664 665 if (start > end) 666 break; 667 668 if (need_resched()) { 669 read_unlock_irq(&tree->lock); 670 cond_resched(); 671 read_lock_irq(&tree->lock); 672 } 673 } 674 out: 675 read_unlock_irq(&tree->lock); 676 return 0; 677 } 678 EXPORT_SYMBOL(wait_extent_bit); 679 680 static void set_state_bits(struct extent_map_tree *tree, 681 struct extent_state *state, 682 int bits) 683 { 684 if ((bits & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { 685 u64 range = state->end - state->start + 1; 686 tree->dirty_bytes += range; 687 } 688 state->state |= bits; 689 } 690 691 /* 692 * set some bits on a range in the tree. This may require allocations 693 * or sleeping, so the gfp mask is used to indicate what is allowed. 694 * 695 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the 696 * range already has the desired bits set. The start of the existing 697 * range is returned in failed_start in this case. 698 * 699 * [start, end] is inclusive 700 * This takes the tree lock. 701 */ 702 int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits, 703 int exclusive, u64 *failed_start, gfp_t mask) 704 { 705 struct extent_state *state; 706 struct extent_state *prealloc = NULL; 707 struct rb_node *node; 708 unsigned long flags; 709 int err = 0; 710 int set; 711 u64 last_start; 712 u64 last_end; 713 again: 714 if (!prealloc && (mask & __GFP_WAIT)) { 715 prealloc = alloc_extent_state(mask); 716 if (!prealloc) 717 return -ENOMEM; 718 } 719 720 write_lock_irqsave(&tree->lock, flags); 721 /* 722 * this search will find all the extents that end after 723 * our range starts. 724 */ 725 node = tree_search(&tree->state, start); 726 if (!node) { 727 err = insert_state(tree, prealloc, start, end, bits); 728 prealloc = NULL; 729 BUG_ON(err == -EEXIST); 730 goto out; 731 } 732 733 state = rb_entry(node, struct extent_state, rb_node); 734 last_start = state->start; 735 last_end = state->end; 736 737 /* 738 * | ---- desired range ---- | 739 * | state | 740 * 741 * Just lock what we found and keep going 742 */ 743 if (state->start == start && state->end <= end) { 744 set = state->state & bits; 745 if (set && exclusive) { 746 *failed_start = state->start; 747 err = -EEXIST; 748 goto out; 749 } 750 set_state_bits(tree, state, bits); 751 start = state->end + 1; 752 merge_state(tree, state); 753 goto search_again; 754 } 755 756 /* 757 * | ---- desired range ---- | 758 * | state | 759 * or 760 * | ------------- state -------------- | 761 * 762 * We need to split the extent we found, and may flip bits on 763 * second half. 764 * 765 * If the extent we found extends past our 766 * range, we just split and search again. It'll get split 767 * again the next time though. 768 * 769 * If the extent we found is inside our range, we set the 770 * desired bit on it. 771 */ 772 if (state->start < start) { 773 set = state->state & bits; 774 if (exclusive && set) { 775 *failed_start = start; 776 err = -EEXIST; 777 goto out; 778 } 779 err = split_state(tree, state, prealloc, start); 780 BUG_ON(err == -EEXIST); 781 prealloc = NULL; 782 if (err) 783 goto out; 784 if (state->end <= end) { 785 set_state_bits(tree, state, bits); 786 start = state->end + 1; 787 merge_state(tree, state); 788 } else { 789 start = state->start; 790 } 791 goto search_again; 792 } 793 /* 794 * | ---- desired range ---- | 795 * | state | or | state | 796 * 797 * There's a hole, we need to insert something in it and 798 * ignore the extent we found. 799 */ 800 if (state->start > start) { 801 u64 this_end; 802 if (end < last_start) 803 this_end = end; 804 else 805 this_end = last_start -1; 806 err = insert_state(tree, prealloc, start, this_end, 807 bits); 808 prealloc = NULL; 809 BUG_ON(err == -EEXIST); 810 if (err) 811 goto out; 812 start = this_end + 1; 813 goto search_again; 814 } 815 /* 816 * | ---- desired range ---- | 817 * | state | 818 * We need to split the extent, and set the bit 819 * on the first half 820 */ 821 if (state->start <= end && state->end > end) { 822 set = state->state & bits; 823 if (exclusive && set) { 824 *failed_start = start; 825 err = -EEXIST; 826 goto out; 827 } 828 err = split_state(tree, state, prealloc, end + 1); 829 BUG_ON(err == -EEXIST); 830 831 set_state_bits(tree, prealloc, bits); 832 merge_state(tree, prealloc); 833 prealloc = NULL; 834 goto out; 835 } 836 837 goto search_again; 838 839 out: 840 write_unlock_irqrestore(&tree->lock, flags); 841 if (prealloc) 842 free_extent_state(prealloc); 843 844 return err; 845 846 search_again: 847 if (start > end) 848 goto out; 849 write_unlock_irqrestore(&tree->lock, flags); 850 if (mask & __GFP_WAIT) 851 cond_resched(); 852 goto again; 853 } 854 EXPORT_SYMBOL(set_extent_bit); 855 856 /* wrappers around set/clear extent bit */ 857 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end, 858 gfp_t mask) 859 { 860 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL, 861 mask); 862 } 863 EXPORT_SYMBOL(set_extent_dirty); 864 865 int set_extent_bits(struct extent_map_tree *tree, u64 start, u64 end, 866 int bits, gfp_t mask) 867 { 868 return set_extent_bit(tree, start, end, bits, 0, NULL, 869 mask); 870 } 871 EXPORT_SYMBOL(set_extent_bits); 872 873 int clear_extent_bits(struct extent_map_tree *tree, u64 start, u64 end, 874 int bits, gfp_t mask) 875 { 876 return clear_extent_bit(tree, start, end, bits, 0, 0, mask); 877 } 878 EXPORT_SYMBOL(clear_extent_bits); 879 880 int set_extent_delalloc(struct extent_map_tree *tree, u64 start, u64 end, 881 gfp_t mask) 882 { 883 return set_extent_bit(tree, start, end, 884 EXTENT_DELALLOC | EXTENT_DIRTY, 0, NULL, 885 mask); 886 } 887 EXPORT_SYMBOL(set_extent_delalloc); 888 889 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end, 890 gfp_t mask) 891 { 892 return clear_extent_bit(tree, start, end, 893 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, mask); 894 } 895 EXPORT_SYMBOL(clear_extent_dirty); 896 897 int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end, 898 gfp_t mask) 899 { 900 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL, 901 mask); 902 } 903 EXPORT_SYMBOL(set_extent_new); 904 905 int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end, 906 gfp_t mask) 907 { 908 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask); 909 } 910 EXPORT_SYMBOL(clear_extent_new); 911 912 int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end, 913 gfp_t mask) 914 { 915 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL, 916 mask); 917 } 918 EXPORT_SYMBOL(set_extent_uptodate); 919 920 int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end, 921 gfp_t mask) 922 { 923 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask); 924 } 925 EXPORT_SYMBOL(clear_extent_uptodate); 926 927 int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end, 928 gfp_t mask) 929 { 930 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK, 931 0, NULL, mask); 932 } 933 EXPORT_SYMBOL(set_extent_writeback); 934 935 int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end, 936 gfp_t mask) 937 { 938 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask); 939 } 940 EXPORT_SYMBOL(clear_extent_writeback); 941 942 int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end) 943 { 944 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK); 945 } 946 EXPORT_SYMBOL(wait_on_extent_writeback); 947 948 /* 949 * locks a range in ascending order, waiting for any locked regions 950 * it hits on the way. [start,end] are inclusive, and this will sleep. 951 */ 952 int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask) 953 { 954 int err; 955 u64 failed_start; 956 while (1) { 957 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 958 &failed_start, mask); 959 if (err == -EEXIST && (mask & __GFP_WAIT)) { 960 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); 961 start = failed_start; 962 } else { 963 break; 964 } 965 WARN_ON(start > end); 966 } 967 return err; 968 } 969 EXPORT_SYMBOL(lock_extent); 970 971 int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end, 972 gfp_t mask) 973 { 974 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask); 975 } 976 EXPORT_SYMBOL(unlock_extent); 977 978 /* 979 * helper function to set pages and extents in the tree dirty 980 */ 981 int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end) 982 { 983 unsigned long index = start >> PAGE_CACHE_SHIFT; 984 unsigned long end_index = end >> PAGE_CACHE_SHIFT; 985 struct page *page; 986 987 while (index <= end_index) { 988 page = find_get_page(tree->mapping, index); 989 BUG_ON(!page); 990 __set_page_dirty_nobuffers(page); 991 page_cache_release(page); 992 index++; 993 } 994 set_extent_dirty(tree, start, end, GFP_NOFS); 995 return 0; 996 } 997 EXPORT_SYMBOL(set_range_dirty); 998 999 /* 1000 * helper function to set both pages and extents in the tree writeback 1001 */ 1002 int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end) 1003 { 1004 unsigned long index = start >> PAGE_CACHE_SHIFT; 1005 unsigned long end_index = end >> PAGE_CACHE_SHIFT; 1006 struct page *page; 1007 1008 while (index <= end_index) { 1009 page = find_get_page(tree->mapping, index); 1010 BUG_ON(!page); 1011 set_page_writeback(page); 1012 page_cache_release(page); 1013 index++; 1014 } 1015 set_extent_writeback(tree, start, end, GFP_NOFS); 1016 return 0; 1017 } 1018 EXPORT_SYMBOL(set_range_writeback); 1019 1020 int find_first_extent_bit(struct extent_map_tree *tree, u64 start, 1021 u64 *start_ret, u64 *end_ret, int bits) 1022 { 1023 struct rb_node *node; 1024 struct extent_state *state; 1025 int ret = 1; 1026 1027 read_lock_irq(&tree->lock); 1028 /* 1029 * this search will find all the extents that end after 1030 * our range starts. 1031 */ 1032 node = tree_search(&tree->state, start); 1033 if (!node || IS_ERR(node)) { 1034 goto out; 1035 } 1036 1037 while(1) { 1038 state = rb_entry(node, struct extent_state, rb_node); 1039 if (state->end >= start && (state->state & bits)) { 1040 *start_ret = state->start; 1041 *end_ret = state->end; 1042 ret = 0; 1043 break; 1044 } 1045 node = rb_next(node); 1046 if (!node) 1047 break; 1048 } 1049 out: 1050 read_unlock_irq(&tree->lock); 1051 return ret; 1052 } 1053 EXPORT_SYMBOL(find_first_extent_bit); 1054 1055 u64 find_lock_delalloc_range(struct extent_map_tree *tree, 1056 u64 *start, u64 *end, u64 max_bytes) 1057 { 1058 struct rb_node *node; 1059 struct extent_state *state; 1060 u64 cur_start = *start; 1061 u64 found = 0; 1062 u64 total_bytes = 0; 1063 1064 write_lock_irq(&tree->lock); 1065 /* 1066 * this search will find all the extents that end after 1067 * our range starts. 1068 */ 1069 search_again: 1070 node = tree_search(&tree->state, cur_start); 1071 if (!node || IS_ERR(node)) { 1072 *end = (u64)-1; 1073 goto out; 1074 } 1075 1076 while(1) { 1077 state = rb_entry(node, struct extent_state, rb_node); 1078 if (found && state->start != cur_start) { 1079 goto out; 1080 } 1081 if (!(state->state & EXTENT_DELALLOC)) { 1082 if (!found) 1083 *end = state->end; 1084 goto out; 1085 } 1086 if (!found) { 1087 struct extent_state *prev_state; 1088 struct rb_node *prev_node = node; 1089 while(1) { 1090 prev_node = rb_prev(prev_node); 1091 if (!prev_node) 1092 break; 1093 prev_state = rb_entry(prev_node, 1094 struct extent_state, 1095 rb_node); 1096 if (!(prev_state->state & EXTENT_DELALLOC)) 1097 break; 1098 state = prev_state; 1099 node = prev_node; 1100 } 1101 } 1102 if (state->state & EXTENT_LOCKED) { 1103 DEFINE_WAIT(wait); 1104 atomic_inc(&state->refs); 1105 prepare_to_wait(&state->wq, &wait, 1106 TASK_UNINTERRUPTIBLE); 1107 write_unlock_irq(&tree->lock); 1108 schedule(); 1109 write_lock_irq(&tree->lock); 1110 finish_wait(&state->wq, &wait); 1111 free_extent_state(state); 1112 goto search_again; 1113 } 1114 state->state |= EXTENT_LOCKED; 1115 if (!found) 1116 *start = state->start; 1117 found++; 1118 *end = state->end; 1119 cur_start = state->end + 1; 1120 node = rb_next(node); 1121 if (!node) 1122 break; 1123 total_bytes += state->end - state->start + 1; 1124 if (total_bytes >= max_bytes) 1125 break; 1126 } 1127 out: 1128 write_unlock_irq(&tree->lock); 1129 return found; 1130 } 1131 1132 u64 count_range_bits(struct extent_map_tree *tree, 1133 u64 *start, u64 search_end, u64 max_bytes, 1134 unsigned long bits) 1135 { 1136 struct rb_node *node; 1137 struct extent_state *state; 1138 u64 cur_start = *start; 1139 u64 total_bytes = 0; 1140 int found = 0; 1141 1142 if (search_end <= cur_start) { 1143 printk("search_end %Lu start %Lu\n", search_end, cur_start); 1144 WARN_ON(1); 1145 return 0; 1146 } 1147 1148 write_lock_irq(&tree->lock); 1149 if (cur_start == 0 && bits == EXTENT_DIRTY) { 1150 total_bytes = tree->dirty_bytes; 1151 goto out; 1152 } 1153 /* 1154 * this search will find all the extents that end after 1155 * our range starts. 1156 */ 1157 node = tree_search(&tree->state, cur_start); 1158 if (!node || IS_ERR(node)) { 1159 goto out; 1160 } 1161 1162 while(1) { 1163 state = rb_entry(node, struct extent_state, rb_node); 1164 if (state->start > search_end) 1165 break; 1166 if (state->end >= cur_start && (state->state & bits)) { 1167 total_bytes += min(search_end, state->end) + 1 - 1168 max(cur_start, state->start); 1169 if (total_bytes >= max_bytes) 1170 break; 1171 if (!found) { 1172 *start = state->start; 1173 found = 1; 1174 } 1175 } 1176 node = rb_next(node); 1177 if (!node) 1178 break; 1179 } 1180 out: 1181 write_unlock_irq(&tree->lock); 1182 return total_bytes; 1183 } 1184 /* 1185 * helper function to lock both pages and extents in the tree. 1186 * pages must be locked first. 1187 */ 1188 int lock_range(struct extent_map_tree *tree, u64 start, u64 end) 1189 { 1190 unsigned long index = start >> PAGE_CACHE_SHIFT; 1191 unsigned long end_index = end >> PAGE_CACHE_SHIFT; 1192 struct page *page; 1193 int err; 1194 1195 while (index <= end_index) { 1196 page = grab_cache_page(tree->mapping, index); 1197 if (!page) { 1198 err = -ENOMEM; 1199 goto failed; 1200 } 1201 if (IS_ERR(page)) { 1202 err = PTR_ERR(page); 1203 goto failed; 1204 } 1205 index++; 1206 } 1207 lock_extent(tree, start, end, GFP_NOFS); 1208 return 0; 1209 1210 failed: 1211 /* 1212 * we failed above in getting the page at 'index', so we undo here 1213 * up to but not including the page at 'index' 1214 */ 1215 end_index = index; 1216 index = start >> PAGE_CACHE_SHIFT; 1217 while (index < end_index) { 1218 page = find_get_page(tree->mapping, index); 1219 unlock_page(page); 1220 page_cache_release(page); 1221 index++; 1222 } 1223 return err; 1224 } 1225 EXPORT_SYMBOL(lock_range); 1226 1227 /* 1228 * helper function to unlock both pages and extents in the tree. 1229 */ 1230 int unlock_range(struct extent_map_tree *tree, u64 start, u64 end) 1231 { 1232 unsigned long index = start >> PAGE_CACHE_SHIFT; 1233 unsigned long end_index = end >> PAGE_CACHE_SHIFT; 1234 struct page *page; 1235 1236 while (index <= end_index) { 1237 page = find_get_page(tree->mapping, index); 1238 unlock_page(page); 1239 page_cache_release(page); 1240 index++; 1241 } 1242 unlock_extent(tree, start, end, GFP_NOFS); 1243 return 0; 1244 } 1245 EXPORT_SYMBOL(unlock_range); 1246 1247 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private) 1248 { 1249 struct rb_node *node; 1250 struct extent_state *state; 1251 int ret = 0; 1252 1253 write_lock_irq(&tree->lock); 1254 /* 1255 * this search will find all the extents that end after 1256 * our range starts. 1257 */ 1258 node = tree_search(&tree->state, start); 1259 if (!node || IS_ERR(node)) { 1260 ret = -ENOENT; 1261 goto out; 1262 } 1263 state = rb_entry(node, struct extent_state, rb_node); 1264 if (state->start != start) { 1265 ret = -ENOENT; 1266 goto out; 1267 } 1268 state->private = private; 1269 out: 1270 write_unlock_irq(&tree->lock); 1271 return ret; 1272 } 1273 1274 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private) 1275 { 1276 struct rb_node *node; 1277 struct extent_state *state; 1278 int ret = 0; 1279 1280 read_lock_irq(&tree->lock); 1281 /* 1282 * this search will find all the extents that end after 1283 * our range starts. 1284 */ 1285 node = tree_search(&tree->state, start); 1286 if (!node || IS_ERR(node)) { 1287 ret = -ENOENT; 1288 goto out; 1289 } 1290 state = rb_entry(node, struct extent_state, rb_node); 1291 if (state->start != start) { 1292 ret = -ENOENT; 1293 goto out; 1294 } 1295 *private = state->private; 1296 out: 1297 read_unlock_irq(&tree->lock); 1298 return ret; 1299 } 1300 1301 /* 1302 * searches a range in the state tree for a given mask. 1303 * If 'filled' == 1, this returns 1 only if ever extent in the tree 1304 * has the bits set. Otherwise, 1 is returned if any bit in the 1305 * range is found set. 1306 */ 1307 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end, 1308 int bits, int filled) 1309 { 1310 struct extent_state *state = NULL; 1311 struct rb_node *node; 1312 int bitset = 0; 1313 1314 read_lock_irq(&tree->lock); 1315 node = tree_search(&tree->state, start); 1316 while (node && start <= end) { 1317 state = rb_entry(node, struct extent_state, rb_node); 1318 1319 if (filled && state->start > start) { 1320 bitset = 0; 1321 break; 1322 } 1323 1324 if (state->start > end) 1325 break; 1326 1327 if (state->state & bits) { 1328 bitset = 1; 1329 if (!filled) 1330 break; 1331 } else if (filled) { 1332 bitset = 0; 1333 break; 1334 } 1335 start = state->end + 1; 1336 if (start > end) 1337 break; 1338 node = rb_next(node); 1339 } 1340 read_unlock_irq(&tree->lock); 1341 return bitset; 1342 } 1343 EXPORT_SYMBOL(test_range_bit); 1344 1345 /* 1346 * helper function to set a given page up to date if all the 1347 * extents in the tree for that page are up to date 1348 */ 1349 static int check_page_uptodate(struct extent_map_tree *tree, 1350 struct page *page) 1351 { 1352 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1353 u64 end = start + PAGE_CACHE_SIZE - 1; 1354 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1)) 1355 SetPageUptodate(page); 1356 return 0; 1357 } 1358 1359 /* 1360 * helper function to unlock a page if all the extents in the tree 1361 * for that page are unlocked 1362 */ 1363 static int check_page_locked(struct extent_map_tree *tree, 1364 struct page *page) 1365 { 1366 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1367 u64 end = start + PAGE_CACHE_SIZE - 1; 1368 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0)) 1369 unlock_page(page); 1370 return 0; 1371 } 1372 1373 /* 1374 * helper function to end page writeback if all the extents 1375 * in the tree for that page are done with writeback 1376 */ 1377 static int check_page_writeback(struct extent_map_tree *tree, 1378 struct page *page) 1379 { 1380 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1381 u64 end = start + PAGE_CACHE_SIZE - 1; 1382 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0)) 1383 end_page_writeback(page); 1384 return 0; 1385 } 1386 1387 /* lots and lots of room for performance fixes in the end_bio funcs */ 1388 1389 /* 1390 * after a writepage IO is done, we need to: 1391 * clear the uptodate bits on error 1392 * clear the writeback bits in the extent tree for this IO 1393 * end_page_writeback if the page has no more pending IO 1394 * 1395 * Scheduling is not allowed, so the extent state tree is expected 1396 * to have one and only one object corresponding to this IO. 1397 */ 1398 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) 1399 static void end_bio_extent_writepage(struct bio *bio, int err) 1400 #else 1401 static int end_bio_extent_writepage(struct bio *bio, 1402 unsigned int bytes_done, int err) 1403 #endif 1404 { 1405 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 1406 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; 1407 struct extent_map_tree *tree = bio->bi_private; 1408 u64 start; 1409 u64 end; 1410 int whole_page; 1411 1412 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1413 if (bio->bi_size) 1414 return 1; 1415 #endif 1416 1417 do { 1418 struct page *page = bvec->bv_page; 1419 start = ((u64)page->index << PAGE_CACHE_SHIFT) + 1420 bvec->bv_offset; 1421 end = start + bvec->bv_len - 1; 1422 1423 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) 1424 whole_page = 1; 1425 else 1426 whole_page = 0; 1427 1428 if (--bvec >= bio->bi_io_vec) 1429 prefetchw(&bvec->bv_page->flags); 1430 1431 if (!uptodate) { 1432 clear_extent_uptodate(tree, start, end, GFP_ATOMIC); 1433 ClearPageUptodate(page); 1434 SetPageError(page); 1435 } 1436 clear_extent_writeback(tree, start, end, GFP_ATOMIC); 1437 1438 if (whole_page) 1439 end_page_writeback(page); 1440 else 1441 check_page_writeback(tree, page); 1442 if (tree->ops && tree->ops->writepage_end_io_hook) 1443 tree->ops->writepage_end_io_hook(page, start, end); 1444 } while (bvec >= bio->bi_io_vec); 1445 1446 bio_put(bio); 1447 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1448 return 0; 1449 #endif 1450 } 1451 1452 /* 1453 * after a readpage IO is done, we need to: 1454 * clear the uptodate bits on error 1455 * set the uptodate bits if things worked 1456 * set the page up to date if all extents in the tree are uptodate 1457 * clear the lock bit in the extent tree 1458 * unlock the page if there are no other extents locked for it 1459 * 1460 * Scheduling is not allowed, so the extent state tree is expected 1461 * to have one and only one object corresponding to this IO. 1462 */ 1463 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) 1464 static void end_bio_extent_readpage(struct bio *bio, int err) 1465 #else 1466 static int end_bio_extent_readpage(struct bio *bio, 1467 unsigned int bytes_done, int err) 1468 #endif 1469 { 1470 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 1471 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; 1472 struct extent_map_tree *tree = bio->bi_private; 1473 u64 start; 1474 u64 end; 1475 int whole_page; 1476 int ret; 1477 1478 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1479 if (bio->bi_size) 1480 return 1; 1481 #endif 1482 1483 do { 1484 struct page *page = bvec->bv_page; 1485 start = ((u64)page->index << PAGE_CACHE_SHIFT) + 1486 bvec->bv_offset; 1487 end = start + bvec->bv_len - 1; 1488 1489 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE) 1490 whole_page = 1; 1491 else 1492 whole_page = 0; 1493 1494 if (--bvec >= bio->bi_io_vec) 1495 prefetchw(&bvec->bv_page->flags); 1496 1497 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) { 1498 ret = tree->ops->readpage_end_io_hook(page, start, end); 1499 if (ret) 1500 uptodate = 0; 1501 } 1502 if (uptodate) { 1503 set_extent_uptodate(tree, start, end, GFP_ATOMIC); 1504 if (whole_page) 1505 SetPageUptodate(page); 1506 else 1507 check_page_uptodate(tree, page); 1508 } else { 1509 ClearPageUptodate(page); 1510 SetPageError(page); 1511 } 1512 1513 unlock_extent(tree, start, end, GFP_ATOMIC); 1514 1515 if (whole_page) 1516 unlock_page(page); 1517 else 1518 check_page_locked(tree, page); 1519 } while (bvec >= bio->bi_io_vec); 1520 1521 bio_put(bio); 1522 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1523 return 0; 1524 #endif 1525 } 1526 1527 /* 1528 * IO done from prepare_write is pretty simple, we just unlock 1529 * the structs in the extent tree when done, and set the uptodate bits 1530 * as appropriate. 1531 */ 1532 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) 1533 static void end_bio_extent_preparewrite(struct bio *bio, int err) 1534 #else 1535 static int end_bio_extent_preparewrite(struct bio *bio, 1536 unsigned int bytes_done, int err) 1537 #endif 1538 { 1539 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 1540 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; 1541 struct extent_map_tree *tree = bio->bi_private; 1542 u64 start; 1543 u64 end; 1544 1545 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1546 if (bio->bi_size) 1547 return 1; 1548 #endif 1549 1550 do { 1551 struct page *page = bvec->bv_page; 1552 start = ((u64)page->index << PAGE_CACHE_SHIFT) + 1553 bvec->bv_offset; 1554 end = start + bvec->bv_len - 1; 1555 1556 if (--bvec >= bio->bi_io_vec) 1557 prefetchw(&bvec->bv_page->flags); 1558 1559 if (uptodate) { 1560 set_extent_uptodate(tree, start, end, GFP_ATOMIC); 1561 } else { 1562 ClearPageUptodate(page); 1563 SetPageError(page); 1564 } 1565 1566 unlock_extent(tree, start, end, GFP_ATOMIC); 1567 1568 } while (bvec >= bio->bi_io_vec); 1569 1570 bio_put(bio); 1571 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) 1572 return 0; 1573 #endif 1574 } 1575 1576 static struct bio * 1577 extent_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs, 1578 gfp_t gfp_flags) 1579 { 1580 struct bio *bio; 1581 1582 bio = bio_alloc(gfp_flags, nr_vecs); 1583 1584 if (bio == NULL && (current->flags & PF_MEMALLOC)) { 1585 while (!bio && (nr_vecs /= 2)) 1586 bio = bio_alloc(gfp_flags, nr_vecs); 1587 } 1588 1589 if (bio) { 1590 bio->bi_bdev = bdev; 1591 bio->bi_sector = first_sector; 1592 } 1593 return bio; 1594 } 1595 1596 static int submit_one_bio(int rw, struct bio *bio) 1597 { 1598 u64 maxsector; 1599 int ret = 0; 1600 1601 bio_get(bio); 1602 1603 maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 1604 if (maxsector < bio->bi_sector) { 1605 printk("sector too large max %Lu got %llu\n", maxsector, 1606 (unsigned long long)bio->bi_sector); 1607 WARN_ON(1); 1608 } 1609 1610 submit_bio(rw, bio); 1611 if (bio_flagged(bio, BIO_EOPNOTSUPP)) 1612 ret = -EOPNOTSUPP; 1613 bio_put(bio); 1614 return ret; 1615 } 1616 1617 static int submit_extent_page(int rw, struct extent_map_tree *tree, 1618 struct page *page, sector_t sector, 1619 size_t size, unsigned long offset, 1620 struct block_device *bdev, 1621 struct bio **bio_ret, 1622 unsigned long max_pages, 1623 bio_end_io_t end_io_func) 1624 { 1625 int ret = 0; 1626 struct bio *bio; 1627 int nr; 1628 1629 if (bio_ret && *bio_ret) { 1630 bio = *bio_ret; 1631 if (bio->bi_sector + (bio->bi_size >> 9) != sector || 1632 bio_add_page(bio, page, size, offset) < size) { 1633 ret = submit_one_bio(rw, bio); 1634 bio = NULL; 1635 } else { 1636 return 0; 1637 } 1638 } 1639 nr = min_t(int, max_pages, bio_get_nr_vecs(bdev)); 1640 bio = extent_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH); 1641 if (!bio) { 1642 printk("failed to allocate bio nr %d\n", nr); 1643 } 1644 bio_add_page(bio, page, size, offset); 1645 bio->bi_end_io = end_io_func; 1646 bio->bi_private = tree; 1647 if (bio_ret) { 1648 *bio_ret = bio; 1649 } else { 1650 ret = submit_one_bio(rw, bio); 1651 } 1652 1653 return ret; 1654 } 1655 1656 void set_page_extent_mapped(struct page *page) 1657 { 1658 if (!PagePrivate(page)) { 1659 SetPagePrivate(page); 1660 WARN_ON(!page->mapping->a_ops->invalidatepage); 1661 set_page_private(page, EXTENT_PAGE_PRIVATE); 1662 page_cache_get(page); 1663 } 1664 } 1665 1666 void set_page_extent_head(struct page *page, unsigned long len) 1667 { 1668 WARN_ON(page->private && page->private == EXTENT_PAGE_PRIVATE && 1669 PageDirty(page)); 1670 set_page_private(page, EXTENT_PAGE_PRIVATE_FIRST_PAGE | len << 2); 1671 } 1672 1673 /* 1674 * basic readpage implementation. Locked extent state structs are inserted 1675 * into the tree that are removed when the IO is done (by the end_io 1676 * handlers) 1677 */ 1678 static int __extent_read_full_page(struct extent_map_tree *tree, 1679 struct page *page, 1680 get_extent_t *get_extent, 1681 struct bio **bio) 1682 { 1683 struct inode *inode = page->mapping->host; 1684 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1685 u64 page_end = start + PAGE_CACHE_SIZE - 1; 1686 u64 end; 1687 u64 cur = start; 1688 u64 extent_offset; 1689 u64 last_byte = i_size_read(inode); 1690 u64 block_start; 1691 u64 cur_end; 1692 sector_t sector; 1693 struct extent_map *em; 1694 struct block_device *bdev; 1695 int ret; 1696 int nr = 0; 1697 size_t page_offset = 0; 1698 size_t iosize; 1699 size_t blocksize = inode->i_sb->s_blocksize; 1700 1701 set_page_extent_mapped(page); 1702 1703 end = page_end; 1704 lock_extent(tree, start, end, GFP_NOFS); 1705 1706 while (cur <= end) { 1707 if (cur >= last_byte) { 1708 char *userpage; 1709 iosize = PAGE_CACHE_SIZE - page_offset; 1710 userpage = kmap_atomic(page, KM_USER0); 1711 memset(userpage + page_offset, 0, iosize); 1712 flush_dcache_page(page); 1713 kunmap_atomic(userpage, KM_USER0); 1714 set_extent_uptodate(tree, cur, cur + iosize - 1, 1715 GFP_NOFS); 1716 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); 1717 break; 1718 } 1719 em = get_extent(inode, page, page_offset, cur, end, 0); 1720 if (IS_ERR(em) || !em) { 1721 SetPageError(page); 1722 unlock_extent(tree, cur, end, GFP_NOFS); 1723 break; 1724 } 1725 1726 extent_offset = cur - em->start; 1727 BUG_ON(em->end < cur); 1728 BUG_ON(end < cur); 1729 1730 iosize = min(em->end - cur, end - cur) + 1; 1731 cur_end = min(em->end, end); 1732 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); 1733 sector = (em->block_start + extent_offset) >> 9; 1734 bdev = em->bdev; 1735 block_start = em->block_start; 1736 free_extent_map(em); 1737 em = NULL; 1738 1739 /* we've found a hole, just zero and go on */ 1740 if (block_start == EXTENT_MAP_HOLE) { 1741 char *userpage; 1742 userpage = kmap_atomic(page, KM_USER0); 1743 memset(userpage + page_offset, 0, iosize); 1744 flush_dcache_page(page); 1745 kunmap_atomic(userpage, KM_USER0); 1746 1747 set_extent_uptodate(tree, cur, cur + iosize - 1, 1748 GFP_NOFS); 1749 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); 1750 cur = cur + iosize; 1751 page_offset += iosize; 1752 continue; 1753 } 1754 /* the get_extent function already copied into the page */ 1755 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) { 1756 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS); 1757 cur = cur + iosize; 1758 page_offset += iosize; 1759 continue; 1760 } 1761 1762 ret = 0; 1763 if (tree->ops && tree->ops->readpage_io_hook) { 1764 ret = tree->ops->readpage_io_hook(page, cur, 1765 cur + iosize - 1); 1766 } 1767 if (!ret) { 1768 unsigned long nr = (last_byte >> PAGE_CACHE_SHIFT) + 1; 1769 nr -= page->index; 1770 ret = submit_extent_page(READ, tree, page, 1771 sector, iosize, page_offset, 1772 bdev, bio, nr, 1773 end_bio_extent_readpage); 1774 } 1775 if (ret) 1776 SetPageError(page); 1777 cur = cur + iosize; 1778 page_offset += iosize; 1779 nr++; 1780 } 1781 if (!nr) { 1782 if (!PageError(page)) 1783 SetPageUptodate(page); 1784 unlock_page(page); 1785 } 1786 return 0; 1787 } 1788 1789 int extent_read_full_page(struct extent_map_tree *tree, struct page *page, 1790 get_extent_t *get_extent) 1791 { 1792 struct bio *bio = NULL; 1793 int ret; 1794 1795 ret = __extent_read_full_page(tree, page, get_extent, &bio); 1796 if (bio) 1797 submit_one_bio(READ, bio); 1798 return ret; 1799 } 1800 EXPORT_SYMBOL(extent_read_full_page); 1801 1802 /* 1803 * the writepage semantics are similar to regular writepage. extent 1804 * records are inserted to lock ranges in the tree, and as dirty areas 1805 * are found, they are marked writeback. Then the lock bits are removed 1806 * and the end_io handler clears the writeback ranges 1807 */ 1808 static int __extent_writepage(struct page *page, struct writeback_control *wbc, 1809 void *data) 1810 { 1811 struct inode *inode = page->mapping->host; 1812 struct extent_page_data *epd = data; 1813 struct extent_map_tree *tree = epd->tree; 1814 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 1815 u64 delalloc_start; 1816 u64 page_end = start + PAGE_CACHE_SIZE - 1; 1817 u64 end; 1818 u64 cur = start; 1819 u64 extent_offset; 1820 u64 last_byte = i_size_read(inode); 1821 u64 block_start; 1822 u64 iosize; 1823 sector_t sector; 1824 struct extent_map *em; 1825 struct block_device *bdev; 1826 int ret; 1827 int nr = 0; 1828 size_t page_offset = 0; 1829 size_t blocksize; 1830 loff_t i_size = i_size_read(inode); 1831 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT; 1832 u64 nr_delalloc; 1833 u64 delalloc_end; 1834 1835 WARN_ON(!PageLocked(page)); 1836 if (page->index > end_index) { 1837 clear_extent_dirty(tree, start, page_end, GFP_NOFS); 1838 unlock_page(page); 1839 return 0; 1840 } 1841 1842 if (page->index == end_index) { 1843 char *userpage; 1844 1845 size_t offset = i_size & (PAGE_CACHE_SIZE - 1); 1846 1847 userpage = kmap_atomic(page, KM_USER0); 1848 memset(userpage + offset, 0, PAGE_CACHE_SIZE - offset); 1849 flush_dcache_page(page); 1850 kunmap_atomic(userpage, KM_USER0); 1851 } 1852 1853 set_page_extent_mapped(page); 1854 1855 delalloc_start = start; 1856 delalloc_end = 0; 1857 while(delalloc_end < page_end) { 1858 nr_delalloc = find_lock_delalloc_range(tree, &delalloc_start, 1859 &delalloc_end, 1860 128 * 1024 * 1024); 1861 if (nr_delalloc == 0) { 1862 delalloc_start = delalloc_end + 1; 1863 continue; 1864 } 1865 tree->ops->fill_delalloc(inode, delalloc_start, 1866 delalloc_end); 1867 clear_extent_bit(tree, delalloc_start, 1868 delalloc_end, 1869 EXTENT_LOCKED | EXTENT_DELALLOC, 1870 1, 0, GFP_NOFS); 1871 delalloc_start = delalloc_end + 1; 1872 } 1873 lock_extent(tree, start, page_end, GFP_NOFS); 1874 1875 end = page_end; 1876 if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) { 1877 printk("found delalloc bits after lock_extent\n"); 1878 } 1879 1880 if (last_byte <= start) { 1881 clear_extent_dirty(tree, start, page_end, GFP_NOFS); 1882 goto done; 1883 } 1884 1885 set_extent_uptodate(tree, start, page_end, GFP_NOFS); 1886 blocksize = inode->i_sb->s_blocksize; 1887 1888 while (cur <= end) { 1889 if (cur >= last_byte) { 1890 clear_extent_dirty(tree, cur, page_end, GFP_NOFS); 1891 break; 1892 } 1893 em = epd->get_extent(inode, page, page_offset, cur, end, 1); 1894 if (IS_ERR(em) || !em) { 1895 SetPageError(page); 1896 break; 1897 } 1898 1899 extent_offset = cur - em->start; 1900 BUG_ON(em->end < cur); 1901 BUG_ON(end < cur); 1902 iosize = min(em->end - cur, end - cur) + 1; 1903 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1); 1904 sector = (em->block_start + extent_offset) >> 9; 1905 bdev = em->bdev; 1906 block_start = em->block_start; 1907 free_extent_map(em); 1908 em = NULL; 1909 1910 if (block_start == EXTENT_MAP_HOLE || 1911 block_start == EXTENT_MAP_INLINE) { 1912 clear_extent_dirty(tree, cur, 1913 cur + iosize - 1, GFP_NOFS); 1914 cur = cur + iosize; 1915 page_offset += iosize; 1916 continue; 1917 } 1918 1919 /* leave this out until we have a page_mkwrite call */ 1920 if (0 && !test_range_bit(tree, cur, cur + iosize - 1, 1921 EXTENT_DIRTY, 0)) { 1922 cur = cur + iosize; 1923 page_offset += iosize; 1924 continue; 1925 } 1926 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS); 1927 if (tree->ops && tree->ops->writepage_io_hook) { 1928 ret = tree->ops->writepage_io_hook(page, cur, 1929 cur + iosize - 1); 1930 } else { 1931 ret = 0; 1932 } 1933 if (ret) 1934 SetPageError(page); 1935 else { 1936 unsigned long max_nr = end_index + 1; 1937 set_range_writeback(tree, cur, cur + iosize - 1); 1938 if (!PageWriteback(page)) { 1939 printk("warning page %lu not writeback, " 1940 "cur %llu end %llu\n", page->index, 1941 (unsigned long long)cur, 1942 (unsigned long long)end); 1943 } 1944 1945 ret = submit_extent_page(WRITE, tree, page, sector, 1946 iosize, page_offset, bdev, 1947 &epd->bio, max_nr, 1948 end_bio_extent_writepage); 1949 if (ret) 1950 SetPageError(page); 1951 } 1952 cur = cur + iosize; 1953 page_offset += iosize; 1954 nr++; 1955 } 1956 done: 1957 if (nr == 0) { 1958 /* make sure the mapping tag for page dirty gets cleared */ 1959 set_page_writeback(page); 1960 end_page_writeback(page); 1961 } 1962 unlock_extent(tree, start, page_end, GFP_NOFS); 1963 unlock_page(page); 1964 return 0; 1965 } 1966 1967 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) 1968 1969 /* Taken directly from 2.6.23 for 2.6.18 back port */ 1970 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc, 1971 void *data); 1972 1973 /** 1974 * write_cache_pages - walk the list of dirty pages of the given address space 1975 * and write all of them. 1976 * @mapping: address space structure to write 1977 * @wbc: subtract the number of written pages from *@wbc->nr_to_write 1978 * @writepage: function called for each page 1979 * @data: data passed to writepage function 1980 * 1981 * If a page is already under I/O, write_cache_pages() skips it, even 1982 * if it's dirty. This is desirable behaviour for memory-cleaning writeback, 1983 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() 1984 * and msync() need to guarantee that all the data which was dirty at the time 1985 * the call was made get new I/O started against them. If wbc->sync_mode is 1986 * WB_SYNC_ALL then we were called for data integrity and we must wait for 1987 * existing IO to complete. 1988 */ 1989 static int write_cache_pages(struct address_space *mapping, 1990 struct writeback_control *wbc, writepage_t writepage, 1991 void *data) 1992 { 1993 struct backing_dev_info *bdi = mapping->backing_dev_info; 1994 int ret = 0; 1995 int done = 0; 1996 struct pagevec pvec; 1997 int nr_pages; 1998 pgoff_t index; 1999 pgoff_t end; /* Inclusive */ 2000 int scanned = 0; 2001 int range_whole = 0; 2002 2003 if (wbc->nonblocking && bdi_write_congested(bdi)) { 2004 wbc->encountered_congestion = 1; 2005 return 0; 2006 } 2007 2008 pagevec_init(&pvec, 0); 2009 if (wbc->range_cyclic) { 2010 index = mapping->writeback_index; /* Start from prev offset */ 2011 end = -1; 2012 } else { 2013 index = wbc->range_start >> PAGE_CACHE_SHIFT; 2014 end = wbc->range_end >> PAGE_CACHE_SHIFT; 2015 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 2016 range_whole = 1; 2017 scanned = 1; 2018 } 2019 retry: 2020 while (!done && (index <= end) && 2021 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 2022 PAGECACHE_TAG_DIRTY, 2023 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { 2024 unsigned i; 2025 2026 scanned = 1; 2027 for (i = 0; i < nr_pages; i++) { 2028 struct page *page = pvec.pages[i]; 2029 2030 /* 2031 * At this point we hold neither mapping->tree_lock nor 2032 * lock on the page itself: the page may be truncated or 2033 * invalidated (changing page->mapping to NULL), or even 2034 * swizzled back from swapper_space to tmpfs file 2035 * mapping 2036 */ 2037 lock_page(page); 2038 2039 if (unlikely(page->mapping != mapping)) { 2040 unlock_page(page); 2041 continue; 2042 } 2043 2044 if (!wbc->range_cyclic && page->index > end) { 2045 done = 1; 2046 unlock_page(page); 2047 continue; 2048 } 2049 2050 if (wbc->sync_mode != WB_SYNC_NONE) 2051 wait_on_page_writeback(page); 2052 2053 if (PageWriteback(page) || 2054 !clear_page_dirty_for_io(page)) { 2055 unlock_page(page); 2056 continue; 2057 } 2058 2059 ret = (*writepage)(page, wbc, data); 2060 2061 if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) { 2062 unlock_page(page); 2063 ret = 0; 2064 } 2065 if (ret || (--(wbc->nr_to_write) <= 0)) 2066 done = 1; 2067 if (wbc->nonblocking && bdi_write_congested(bdi)) { 2068 wbc->encountered_congestion = 1; 2069 done = 1; 2070 } 2071 } 2072 pagevec_release(&pvec); 2073 cond_resched(); 2074 } 2075 if (!scanned && !done) { 2076 /* 2077 * We hit the last page and there is more work to be done: wrap 2078 * back to the start of the file 2079 */ 2080 scanned = 1; 2081 index = 0; 2082 goto retry; 2083 } 2084 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 2085 mapping->writeback_index = index; 2086 return ret; 2087 } 2088 #endif 2089 2090 int extent_write_full_page(struct extent_map_tree *tree, struct page *page, 2091 get_extent_t *get_extent, 2092 struct writeback_control *wbc) 2093 { 2094 int ret; 2095 struct address_space *mapping = page->mapping; 2096 struct extent_page_data epd = { 2097 .bio = NULL, 2098 .tree = tree, 2099 .get_extent = get_extent, 2100 }; 2101 struct writeback_control wbc_writepages = { 2102 .bdi = wbc->bdi, 2103 .sync_mode = WB_SYNC_NONE, 2104 .older_than_this = NULL, 2105 .nr_to_write = 64, 2106 .range_start = page_offset(page) + PAGE_CACHE_SIZE, 2107 .range_end = (loff_t)-1, 2108 }; 2109 2110 2111 ret = __extent_writepage(page, wbc, &epd); 2112 2113 write_cache_pages(mapping, &wbc_writepages, __extent_writepage, &epd); 2114 if (epd.bio) { 2115 submit_one_bio(WRITE, epd.bio); 2116 } 2117 return ret; 2118 } 2119 EXPORT_SYMBOL(extent_write_full_page); 2120 2121 2122 int extent_writepages(struct extent_map_tree *tree, 2123 struct address_space *mapping, 2124 get_extent_t *get_extent, 2125 struct writeback_control *wbc) 2126 { 2127 int ret = 0; 2128 struct extent_page_data epd = { 2129 .bio = NULL, 2130 .tree = tree, 2131 .get_extent = get_extent, 2132 }; 2133 2134 ret = write_cache_pages(mapping, wbc, __extent_writepage, &epd); 2135 if (epd.bio) { 2136 submit_one_bio(WRITE, epd.bio); 2137 } 2138 return ret; 2139 } 2140 EXPORT_SYMBOL(extent_writepages); 2141 2142 int extent_readpages(struct extent_map_tree *tree, 2143 struct address_space *mapping, 2144 struct list_head *pages, unsigned nr_pages, 2145 get_extent_t get_extent) 2146 { 2147 struct bio *bio = NULL; 2148 unsigned page_idx; 2149 struct pagevec pvec; 2150 2151 pagevec_init(&pvec, 0); 2152 for (page_idx = 0; page_idx < nr_pages; page_idx++) { 2153 struct page *page = list_entry(pages->prev, struct page, lru); 2154 2155 prefetchw(&page->flags); 2156 list_del(&page->lru); 2157 /* 2158 * what we want to do here is call add_to_page_cache_lru, 2159 * but that isn't exported, so we reproduce it here 2160 */ 2161 if (!add_to_page_cache(page, mapping, 2162 page->index, GFP_KERNEL)) { 2163 2164 /* open coding of lru_cache_add, also not exported */ 2165 page_cache_get(page); 2166 if (!pagevec_add(&pvec, page)) 2167 __pagevec_lru_add(&pvec); 2168 __extent_read_full_page(tree, page, get_extent, &bio); 2169 } 2170 page_cache_release(page); 2171 } 2172 if (pagevec_count(&pvec)) 2173 __pagevec_lru_add(&pvec); 2174 BUG_ON(!list_empty(pages)); 2175 if (bio) 2176 submit_one_bio(READ, bio); 2177 return 0; 2178 } 2179 EXPORT_SYMBOL(extent_readpages); 2180 2181 /* 2182 * basic invalidatepage code, this waits on any locked or writeback 2183 * ranges corresponding to the page, and then deletes any extent state 2184 * records from the tree 2185 */ 2186 int extent_invalidatepage(struct extent_map_tree *tree, 2187 struct page *page, unsigned long offset) 2188 { 2189 u64 start = ((u64)page->index << PAGE_CACHE_SHIFT); 2190 u64 end = start + PAGE_CACHE_SIZE - 1; 2191 size_t blocksize = page->mapping->host->i_sb->s_blocksize; 2192 2193 start += (offset + blocksize -1) & ~(blocksize - 1); 2194 if (start > end) 2195 return 0; 2196 2197 lock_extent(tree, start, end, GFP_NOFS); 2198 wait_on_extent_writeback(tree, start, end); 2199 clear_extent_bit(tree, start, end, 2200 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, 2201 1, 1, GFP_NOFS); 2202 return 0; 2203 } 2204 EXPORT_SYMBOL(extent_invalidatepage); 2205 2206 /* 2207 * simple commit_write call, set_range_dirty is used to mark both 2208 * the pages and the extent records as dirty 2209 */ 2210 int extent_commit_write(struct extent_map_tree *tree, 2211 struct inode *inode, struct page *page, 2212 unsigned from, unsigned to) 2213 { 2214 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; 2215 2216 set_page_extent_mapped(page); 2217 set_page_dirty(page); 2218 2219 if (pos > inode->i_size) { 2220 i_size_write(inode, pos); 2221 mark_inode_dirty(inode); 2222 } 2223 return 0; 2224 } 2225 EXPORT_SYMBOL(extent_commit_write); 2226 2227 int extent_prepare_write(struct extent_map_tree *tree, 2228 struct inode *inode, struct page *page, 2229 unsigned from, unsigned to, get_extent_t *get_extent) 2230 { 2231 u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT; 2232 u64 page_end = page_start + PAGE_CACHE_SIZE - 1; 2233 u64 block_start; 2234 u64 orig_block_start; 2235 u64 block_end; 2236 u64 cur_end; 2237 struct extent_map *em; 2238 unsigned blocksize = 1 << inode->i_blkbits; 2239 size_t page_offset = 0; 2240 size_t block_off_start; 2241 size_t block_off_end; 2242 int err = 0; 2243 int iocount = 0; 2244 int ret = 0; 2245 int isnew; 2246 2247 set_page_extent_mapped(page); 2248 2249 block_start = (page_start + from) & ~((u64)blocksize - 1); 2250 block_end = (page_start + to - 1) | (blocksize - 1); 2251 orig_block_start = block_start; 2252 2253 lock_extent(tree, page_start, page_end, GFP_NOFS); 2254 while(block_start <= block_end) { 2255 em = get_extent(inode, page, page_offset, block_start, 2256 block_end, 1); 2257 if (IS_ERR(em) || !em) { 2258 goto err; 2259 } 2260 cur_end = min(block_end, em->end); 2261 block_off_start = block_start & (PAGE_CACHE_SIZE - 1); 2262 block_off_end = block_off_start + blocksize; 2263 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS); 2264 2265 if (!PageUptodate(page) && isnew && 2266 (block_off_end > to || block_off_start < from)) { 2267 void *kaddr; 2268 2269 kaddr = kmap_atomic(page, KM_USER0); 2270 if (block_off_end > to) 2271 memset(kaddr + to, 0, block_off_end - to); 2272 if (block_off_start < from) 2273 memset(kaddr + block_off_start, 0, 2274 from - block_off_start); 2275 flush_dcache_page(page); 2276 kunmap_atomic(kaddr, KM_USER0); 2277 } 2278 if ((em->block_start != EXTENT_MAP_HOLE && 2279 em->block_start != EXTENT_MAP_INLINE) && 2280 !isnew && !PageUptodate(page) && 2281 (block_off_end > to || block_off_start < from) && 2282 !test_range_bit(tree, block_start, cur_end, 2283 EXTENT_UPTODATE, 1)) { 2284 u64 sector; 2285 u64 extent_offset = block_start - em->start; 2286 size_t iosize; 2287 sector = (em->block_start + extent_offset) >> 9; 2288 iosize = (cur_end - block_start + blocksize) & 2289 ~((u64)blocksize - 1); 2290 /* 2291 * we've already got the extent locked, but we 2292 * need to split the state such that our end_bio 2293 * handler can clear the lock. 2294 */ 2295 set_extent_bit(tree, block_start, 2296 block_start + iosize - 1, 2297 EXTENT_LOCKED, 0, NULL, GFP_NOFS); 2298 ret = submit_extent_page(READ, tree, page, 2299 sector, iosize, page_offset, em->bdev, 2300 NULL, 1, 2301 end_bio_extent_preparewrite); 2302 iocount++; 2303 block_start = block_start + iosize; 2304 } else { 2305 set_extent_uptodate(tree, block_start, cur_end, 2306 GFP_NOFS); 2307 unlock_extent(tree, block_start, cur_end, GFP_NOFS); 2308 block_start = cur_end + 1; 2309 } 2310 page_offset = block_start & (PAGE_CACHE_SIZE - 1); 2311 free_extent_map(em); 2312 } 2313 if (iocount) { 2314 wait_extent_bit(tree, orig_block_start, 2315 block_end, EXTENT_LOCKED); 2316 } 2317 check_page_uptodate(tree, page); 2318 err: 2319 /* FIXME, zero out newly allocated blocks on error */ 2320 return err; 2321 } 2322 EXPORT_SYMBOL(extent_prepare_write); 2323 2324 /* 2325 * a helper for releasepage. As long as there are no locked extents 2326 * in the range corresponding to the page, both state records and extent 2327 * map records are removed 2328 */ 2329 int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page) 2330 { 2331 struct extent_map *em; 2332 u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 2333 u64 end = start + PAGE_CACHE_SIZE - 1; 2334 u64 orig_start = start; 2335 int ret = 1; 2336 2337 while (start <= end) { 2338 em = lookup_extent_mapping(tree, start, end); 2339 if (!em || IS_ERR(em)) 2340 break; 2341 if (!test_range_bit(tree, em->start, em->end, 2342 EXTENT_LOCKED, 0)) { 2343 remove_extent_mapping(tree, em); 2344 /* once for the rb tree */ 2345 free_extent_map(em); 2346 } 2347 start = em->end + 1; 2348 /* once for us */ 2349 free_extent_map(em); 2350 } 2351 if (test_range_bit(tree, orig_start, end, EXTENT_LOCKED, 0)) 2352 ret = 0; 2353 else 2354 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE, 2355 1, 1, GFP_NOFS); 2356 return ret; 2357 } 2358 EXPORT_SYMBOL(try_release_extent_mapping); 2359 2360 sector_t extent_bmap(struct address_space *mapping, sector_t iblock, 2361 get_extent_t *get_extent) 2362 { 2363 struct inode *inode = mapping->host; 2364 u64 start = iblock << inode->i_blkbits; 2365 u64 end = start + (1 << inode->i_blkbits) - 1; 2366 sector_t sector = 0; 2367 struct extent_map *em; 2368 2369 em = get_extent(inode, NULL, 0, start, end, 0); 2370 if (!em || IS_ERR(em)) 2371 return 0; 2372 2373 if (em->block_start == EXTENT_MAP_INLINE || 2374 em->block_start == EXTENT_MAP_HOLE) 2375 goto out; 2376 2377 sector = (em->block_start + start - em->start) >> inode->i_blkbits; 2378 out: 2379 free_extent_map(em); 2380 return sector; 2381 } 2382 2383 static int add_lru(struct extent_map_tree *tree, struct extent_buffer *eb) 2384 { 2385 if (list_empty(&eb->lru)) { 2386 extent_buffer_get(eb); 2387 list_add(&eb->lru, &tree->buffer_lru); 2388 tree->lru_size++; 2389 if (tree->lru_size >= BUFFER_LRU_MAX) { 2390 struct extent_buffer *rm; 2391 rm = list_entry(tree->buffer_lru.prev, 2392 struct extent_buffer, lru); 2393 tree->lru_size--; 2394 list_del_init(&rm->lru); 2395 free_extent_buffer(rm); 2396 } 2397 } else 2398 list_move(&eb->lru, &tree->buffer_lru); 2399 return 0; 2400 } 2401 static struct extent_buffer *find_lru(struct extent_map_tree *tree, 2402 u64 start, unsigned long len) 2403 { 2404 struct list_head *lru = &tree->buffer_lru; 2405 struct list_head *cur = lru->next; 2406 struct extent_buffer *eb; 2407 2408 if (list_empty(lru)) 2409 return NULL; 2410 2411 do { 2412 eb = list_entry(cur, struct extent_buffer, lru); 2413 if (eb->start == start && eb->len == len) { 2414 extent_buffer_get(eb); 2415 return eb; 2416 } 2417 cur = cur->next; 2418 } while (cur != lru); 2419 return NULL; 2420 } 2421 2422 static inline unsigned long num_extent_pages(u64 start, u64 len) 2423 { 2424 return ((start + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - 2425 (start >> PAGE_CACHE_SHIFT); 2426 } 2427 2428 static inline struct page *extent_buffer_page(struct extent_buffer *eb, 2429 unsigned long i) 2430 { 2431 struct page *p; 2432 struct address_space *mapping; 2433 2434 if (i == 0) 2435 return eb->first_page; 2436 i += eb->start >> PAGE_CACHE_SHIFT; 2437 mapping = eb->first_page->mapping; 2438 read_lock_irq(&mapping->tree_lock); 2439 p = radix_tree_lookup(&mapping->page_tree, i); 2440 read_unlock_irq(&mapping->tree_lock); 2441 return p; 2442 } 2443 2444 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree, 2445 u64 start, 2446 unsigned long len, 2447 gfp_t mask) 2448 { 2449 struct extent_buffer *eb = NULL; 2450 2451 spin_lock(&tree->lru_lock); 2452 eb = find_lru(tree, start, len); 2453 spin_unlock(&tree->lru_lock); 2454 if (eb) { 2455 return eb; 2456 } 2457 2458 eb = kmem_cache_zalloc(extent_buffer_cache, mask); 2459 INIT_LIST_HEAD(&eb->lru); 2460 eb->start = start; 2461 eb->len = len; 2462 atomic_set(&eb->refs, 1); 2463 2464 return eb; 2465 } 2466 2467 static void __free_extent_buffer(struct extent_buffer *eb) 2468 { 2469 kmem_cache_free(extent_buffer_cache, eb); 2470 } 2471 2472 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree, 2473 u64 start, unsigned long len, 2474 struct page *page0, 2475 gfp_t mask) 2476 { 2477 unsigned long num_pages = num_extent_pages(start, len); 2478 unsigned long i; 2479 unsigned long index = start >> PAGE_CACHE_SHIFT; 2480 struct extent_buffer *eb; 2481 struct page *p; 2482 struct address_space *mapping = tree->mapping; 2483 int uptodate = 1; 2484 2485 eb = __alloc_extent_buffer(tree, start, len, mask); 2486 if (!eb || IS_ERR(eb)) 2487 return NULL; 2488 2489 if (eb->flags & EXTENT_BUFFER_FILLED) 2490 goto lru_add; 2491 2492 if (page0) { 2493 eb->first_page = page0; 2494 i = 1; 2495 index++; 2496 page_cache_get(page0); 2497 mark_page_accessed(page0); 2498 set_page_extent_mapped(page0); 2499 WARN_ON(!PageUptodate(page0)); 2500 set_page_extent_head(page0, len); 2501 } else { 2502 i = 0; 2503 } 2504 for (; i < num_pages; i++, index++) { 2505 p = find_or_create_page(mapping, index, mask | __GFP_HIGHMEM); 2506 if (!p) { 2507 WARN_ON(1); 2508 goto fail; 2509 } 2510 set_page_extent_mapped(p); 2511 mark_page_accessed(p); 2512 if (i == 0) { 2513 eb->first_page = p; 2514 set_page_extent_head(p, len); 2515 } else { 2516 set_page_private(p, EXTENT_PAGE_PRIVATE); 2517 } 2518 if (!PageUptodate(p)) 2519 uptodate = 0; 2520 unlock_page(p); 2521 } 2522 if (uptodate) 2523 eb->flags |= EXTENT_UPTODATE; 2524 eb->flags |= EXTENT_BUFFER_FILLED; 2525 2526 lru_add: 2527 spin_lock(&tree->lru_lock); 2528 add_lru(tree, eb); 2529 spin_unlock(&tree->lru_lock); 2530 return eb; 2531 2532 fail: 2533 spin_lock(&tree->lru_lock); 2534 list_del_init(&eb->lru); 2535 spin_unlock(&tree->lru_lock); 2536 if (!atomic_dec_and_test(&eb->refs)) 2537 return NULL; 2538 for (index = 1; index < i; index++) { 2539 page_cache_release(extent_buffer_page(eb, index)); 2540 } 2541 if (i > 0) 2542 page_cache_release(extent_buffer_page(eb, 0)); 2543 __free_extent_buffer(eb); 2544 return NULL; 2545 } 2546 EXPORT_SYMBOL(alloc_extent_buffer); 2547 2548 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree, 2549 u64 start, unsigned long len, 2550 gfp_t mask) 2551 { 2552 unsigned long num_pages = num_extent_pages(start, len); 2553 unsigned long i; 2554 unsigned long index = start >> PAGE_CACHE_SHIFT; 2555 struct extent_buffer *eb; 2556 struct page *p; 2557 struct address_space *mapping = tree->mapping; 2558 int uptodate = 1; 2559 2560 eb = __alloc_extent_buffer(tree, start, len, mask); 2561 if (!eb || IS_ERR(eb)) 2562 return NULL; 2563 2564 if (eb->flags & EXTENT_BUFFER_FILLED) 2565 goto lru_add; 2566 2567 for (i = 0; i < num_pages; i++, index++) { 2568 p = find_lock_page(mapping, index); 2569 if (!p) { 2570 goto fail; 2571 } 2572 set_page_extent_mapped(p); 2573 mark_page_accessed(p); 2574 2575 if (i == 0) { 2576 eb->first_page = p; 2577 set_page_extent_head(p, len); 2578 } else { 2579 set_page_private(p, EXTENT_PAGE_PRIVATE); 2580 } 2581 2582 if (!PageUptodate(p)) 2583 uptodate = 0; 2584 unlock_page(p); 2585 } 2586 if (uptodate) 2587 eb->flags |= EXTENT_UPTODATE; 2588 eb->flags |= EXTENT_BUFFER_FILLED; 2589 2590 lru_add: 2591 spin_lock(&tree->lru_lock); 2592 add_lru(tree, eb); 2593 spin_unlock(&tree->lru_lock); 2594 return eb; 2595 fail: 2596 spin_lock(&tree->lru_lock); 2597 list_del_init(&eb->lru); 2598 spin_unlock(&tree->lru_lock); 2599 if (!atomic_dec_and_test(&eb->refs)) 2600 return NULL; 2601 for (index = 1; index < i; index++) { 2602 page_cache_release(extent_buffer_page(eb, index)); 2603 } 2604 if (i > 0) 2605 page_cache_release(extent_buffer_page(eb, 0)); 2606 __free_extent_buffer(eb); 2607 return NULL; 2608 } 2609 EXPORT_SYMBOL(find_extent_buffer); 2610 2611 void free_extent_buffer(struct extent_buffer *eb) 2612 { 2613 unsigned long i; 2614 unsigned long num_pages; 2615 2616 if (!eb) 2617 return; 2618 2619 if (!atomic_dec_and_test(&eb->refs)) 2620 return; 2621 2622 WARN_ON(!list_empty(&eb->lru)); 2623 num_pages = num_extent_pages(eb->start, eb->len); 2624 2625 for (i = 1; i < num_pages; i++) { 2626 page_cache_release(extent_buffer_page(eb, i)); 2627 } 2628 page_cache_release(extent_buffer_page(eb, 0)); 2629 __free_extent_buffer(eb); 2630 } 2631 EXPORT_SYMBOL(free_extent_buffer); 2632 2633 int clear_extent_buffer_dirty(struct extent_map_tree *tree, 2634 struct extent_buffer *eb) 2635 { 2636 int set; 2637 unsigned long i; 2638 unsigned long num_pages; 2639 struct page *page; 2640 2641 u64 start = eb->start; 2642 u64 end = start + eb->len - 1; 2643 2644 set = clear_extent_dirty(tree, start, end, GFP_NOFS); 2645 num_pages = num_extent_pages(eb->start, eb->len); 2646 2647 for (i = 0; i < num_pages; i++) { 2648 page = extent_buffer_page(eb, i); 2649 lock_page(page); 2650 if (i == 0) 2651 set_page_extent_head(page, eb->len); 2652 else 2653 set_page_private(page, EXTENT_PAGE_PRIVATE); 2654 2655 /* 2656 * if we're on the last page or the first page and the 2657 * block isn't aligned on a page boundary, do extra checks 2658 * to make sure we don't clean page that is partially dirty 2659 */ 2660 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || 2661 ((i == num_pages - 1) && 2662 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { 2663 start = (u64)page->index << PAGE_CACHE_SHIFT; 2664 end = start + PAGE_CACHE_SIZE - 1; 2665 if (test_range_bit(tree, start, end, 2666 EXTENT_DIRTY, 0)) { 2667 unlock_page(page); 2668 continue; 2669 } 2670 } 2671 clear_page_dirty_for_io(page); 2672 write_lock_irq(&page->mapping->tree_lock); 2673 if (!PageDirty(page)) { 2674 radix_tree_tag_clear(&page->mapping->page_tree, 2675 page_index(page), 2676 PAGECACHE_TAG_DIRTY); 2677 } 2678 write_unlock_irq(&page->mapping->tree_lock); 2679 unlock_page(page); 2680 } 2681 return 0; 2682 } 2683 EXPORT_SYMBOL(clear_extent_buffer_dirty); 2684 2685 int wait_on_extent_buffer_writeback(struct extent_map_tree *tree, 2686 struct extent_buffer *eb) 2687 { 2688 return wait_on_extent_writeback(tree, eb->start, 2689 eb->start + eb->len - 1); 2690 } 2691 EXPORT_SYMBOL(wait_on_extent_buffer_writeback); 2692 2693 int set_extent_buffer_dirty(struct extent_map_tree *tree, 2694 struct extent_buffer *eb) 2695 { 2696 unsigned long i; 2697 unsigned long num_pages; 2698 2699 num_pages = num_extent_pages(eb->start, eb->len); 2700 for (i = 0; i < num_pages; i++) { 2701 struct page *page = extent_buffer_page(eb, i); 2702 /* writepage may need to do something special for the 2703 * first page, we have to make sure page->private is 2704 * properly set. releasepage may drop page->private 2705 * on us if the page isn't already dirty. 2706 */ 2707 if (i == 0) { 2708 lock_page(page); 2709 set_page_extent_head(page, eb->len); 2710 } else if (PagePrivate(page) && 2711 page->private != EXTENT_PAGE_PRIVATE) { 2712 lock_page(page); 2713 set_page_extent_mapped(page); 2714 unlock_page(page); 2715 } 2716 __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); 2717 if (i == 0) 2718 unlock_page(page); 2719 } 2720 return set_extent_dirty(tree, eb->start, 2721 eb->start + eb->len - 1, GFP_NOFS); 2722 } 2723 EXPORT_SYMBOL(set_extent_buffer_dirty); 2724 2725 int set_extent_buffer_uptodate(struct extent_map_tree *tree, 2726 struct extent_buffer *eb) 2727 { 2728 unsigned long i; 2729 struct page *page; 2730 unsigned long num_pages; 2731 2732 num_pages = num_extent_pages(eb->start, eb->len); 2733 2734 set_extent_uptodate(tree, eb->start, eb->start + eb->len - 1, 2735 GFP_NOFS); 2736 for (i = 0; i < num_pages; i++) { 2737 page = extent_buffer_page(eb, i); 2738 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) || 2739 ((i == num_pages - 1) && 2740 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) { 2741 check_page_uptodate(tree, page); 2742 continue; 2743 } 2744 SetPageUptodate(page); 2745 } 2746 return 0; 2747 } 2748 EXPORT_SYMBOL(set_extent_buffer_uptodate); 2749 2750 int extent_buffer_uptodate(struct extent_map_tree *tree, 2751 struct extent_buffer *eb) 2752 { 2753 if (eb->flags & EXTENT_UPTODATE) 2754 return 1; 2755 return test_range_bit(tree, eb->start, eb->start + eb->len - 1, 2756 EXTENT_UPTODATE, 1); 2757 } 2758 EXPORT_SYMBOL(extent_buffer_uptodate); 2759 2760 int read_extent_buffer_pages(struct extent_map_tree *tree, 2761 struct extent_buffer *eb, 2762 u64 start, 2763 int wait) 2764 { 2765 unsigned long i; 2766 unsigned long start_i; 2767 struct page *page; 2768 int err; 2769 int ret = 0; 2770 unsigned long num_pages; 2771 2772 if (eb->flags & EXTENT_UPTODATE) 2773 return 0; 2774 2775 if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1, 2776 EXTENT_UPTODATE, 1)) { 2777 return 0; 2778 } 2779 2780 if (start) { 2781 WARN_ON(start < eb->start); 2782 start_i = (start >> PAGE_CACHE_SHIFT) - 2783 (eb->start >> PAGE_CACHE_SHIFT); 2784 } else { 2785 start_i = 0; 2786 } 2787 2788 num_pages = num_extent_pages(eb->start, eb->len); 2789 for (i = start_i; i < num_pages; i++) { 2790 page = extent_buffer_page(eb, i); 2791 if (PageUptodate(page)) { 2792 continue; 2793 } 2794 if (!wait) { 2795 if (TestSetPageLocked(page)) { 2796 continue; 2797 } 2798 } else { 2799 lock_page(page); 2800 } 2801 if (!PageUptodate(page)) { 2802 err = page->mapping->a_ops->readpage(NULL, page); 2803 if (err) { 2804 ret = err; 2805 } 2806 } else { 2807 unlock_page(page); 2808 } 2809 } 2810 2811 if (ret || !wait) { 2812 return ret; 2813 } 2814 2815 for (i = start_i; i < num_pages; i++) { 2816 page = extent_buffer_page(eb, i); 2817 wait_on_page_locked(page); 2818 if (!PageUptodate(page)) { 2819 ret = -EIO; 2820 } 2821 } 2822 if (!ret) 2823 eb->flags |= EXTENT_UPTODATE; 2824 return ret; 2825 } 2826 EXPORT_SYMBOL(read_extent_buffer_pages); 2827 2828 void read_extent_buffer(struct extent_buffer *eb, void *dstv, 2829 unsigned long start, 2830 unsigned long len) 2831 { 2832 size_t cur; 2833 size_t offset; 2834 struct page *page; 2835 char *kaddr; 2836 char *dst = (char *)dstv; 2837 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 2838 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 2839 unsigned long num_pages = num_extent_pages(eb->start, eb->len); 2840 2841 WARN_ON(start > eb->len); 2842 WARN_ON(start + len > eb->start + eb->len); 2843 2844 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); 2845 2846 while(len > 0) { 2847 page = extent_buffer_page(eb, i); 2848 if (!PageUptodate(page)) { 2849 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len); 2850 WARN_ON(1); 2851 } 2852 WARN_ON(!PageUptodate(page)); 2853 2854 cur = min(len, (PAGE_CACHE_SIZE - offset)); 2855 kaddr = kmap_atomic(page, KM_USER1); 2856 memcpy(dst, kaddr + offset, cur); 2857 kunmap_atomic(kaddr, KM_USER1); 2858 2859 dst += cur; 2860 len -= cur; 2861 offset = 0; 2862 i++; 2863 } 2864 } 2865 EXPORT_SYMBOL(read_extent_buffer); 2866 2867 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start, 2868 unsigned long min_len, char **token, char **map, 2869 unsigned long *map_start, 2870 unsigned long *map_len, int km) 2871 { 2872 size_t offset = start & (PAGE_CACHE_SIZE - 1); 2873 char *kaddr; 2874 struct page *p; 2875 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 2876 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 2877 unsigned long end_i = (start_offset + start + min_len - 1) >> 2878 PAGE_CACHE_SHIFT; 2879 2880 if (i != end_i) 2881 return -EINVAL; 2882 2883 if (i == 0) { 2884 offset = start_offset; 2885 *map_start = 0; 2886 } else { 2887 offset = 0; 2888 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset; 2889 } 2890 if (start + min_len > eb->len) { 2891 printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len, start, min_len); 2892 WARN_ON(1); 2893 } 2894 2895 p = extent_buffer_page(eb, i); 2896 WARN_ON(!PageUptodate(p)); 2897 kaddr = kmap_atomic(p, km); 2898 *token = kaddr; 2899 *map = kaddr + offset; 2900 *map_len = PAGE_CACHE_SIZE - offset; 2901 return 0; 2902 } 2903 EXPORT_SYMBOL(map_private_extent_buffer); 2904 2905 int map_extent_buffer(struct extent_buffer *eb, unsigned long start, 2906 unsigned long min_len, 2907 char **token, char **map, 2908 unsigned long *map_start, 2909 unsigned long *map_len, int km) 2910 { 2911 int err; 2912 int save = 0; 2913 if (eb->map_token) { 2914 unmap_extent_buffer(eb, eb->map_token, km); 2915 eb->map_token = NULL; 2916 save = 1; 2917 } 2918 err = map_private_extent_buffer(eb, start, min_len, token, map, 2919 map_start, map_len, km); 2920 if (!err && save) { 2921 eb->map_token = *token; 2922 eb->kaddr = *map; 2923 eb->map_start = *map_start; 2924 eb->map_len = *map_len; 2925 } 2926 return err; 2927 } 2928 EXPORT_SYMBOL(map_extent_buffer); 2929 2930 void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km) 2931 { 2932 kunmap_atomic(token, km); 2933 } 2934 EXPORT_SYMBOL(unmap_extent_buffer); 2935 2936 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv, 2937 unsigned long start, 2938 unsigned long len) 2939 { 2940 size_t cur; 2941 size_t offset; 2942 struct page *page; 2943 char *kaddr; 2944 char *ptr = (char *)ptrv; 2945 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 2946 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 2947 int ret = 0; 2948 2949 WARN_ON(start > eb->len); 2950 WARN_ON(start + len > eb->start + eb->len); 2951 2952 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); 2953 2954 while(len > 0) { 2955 page = extent_buffer_page(eb, i); 2956 WARN_ON(!PageUptodate(page)); 2957 2958 cur = min(len, (PAGE_CACHE_SIZE - offset)); 2959 2960 kaddr = kmap_atomic(page, KM_USER0); 2961 ret = memcmp(ptr, kaddr + offset, cur); 2962 kunmap_atomic(kaddr, KM_USER0); 2963 if (ret) 2964 break; 2965 2966 ptr += cur; 2967 len -= cur; 2968 offset = 0; 2969 i++; 2970 } 2971 return ret; 2972 } 2973 EXPORT_SYMBOL(memcmp_extent_buffer); 2974 2975 void write_extent_buffer(struct extent_buffer *eb, const void *srcv, 2976 unsigned long start, unsigned long len) 2977 { 2978 size_t cur; 2979 size_t offset; 2980 struct page *page; 2981 char *kaddr; 2982 char *src = (char *)srcv; 2983 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 2984 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 2985 2986 WARN_ON(start > eb->len); 2987 WARN_ON(start + len > eb->start + eb->len); 2988 2989 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); 2990 2991 while(len > 0) { 2992 page = extent_buffer_page(eb, i); 2993 WARN_ON(!PageUptodate(page)); 2994 2995 cur = min(len, PAGE_CACHE_SIZE - offset); 2996 kaddr = kmap_atomic(page, KM_USER1); 2997 memcpy(kaddr + offset, src, cur); 2998 kunmap_atomic(kaddr, KM_USER1); 2999 3000 src += cur; 3001 len -= cur; 3002 offset = 0; 3003 i++; 3004 } 3005 } 3006 EXPORT_SYMBOL(write_extent_buffer); 3007 3008 void memset_extent_buffer(struct extent_buffer *eb, char c, 3009 unsigned long start, unsigned long len) 3010 { 3011 size_t cur; 3012 size_t offset; 3013 struct page *page; 3014 char *kaddr; 3015 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 3016 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 3017 3018 WARN_ON(start > eb->len); 3019 WARN_ON(start + len > eb->start + eb->len); 3020 3021 offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1); 3022 3023 while(len > 0) { 3024 page = extent_buffer_page(eb, i); 3025 WARN_ON(!PageUptodate(page)); 3026 3027 cur = min(len, PAGE_CACHE_SIZE - offset); 3028 kaddr = kmap_atomic(page, KM_USER0); 3029 memset(kaddr + offset, c, cur); 3030 kunmap_atomic(kaddr, KM_USER0); 3031 3032 len -= cur; 3033 offset = 0; 3034 i++; 3035 } 3036 } 3037 EXPORT_SYMBOL(memset_extent_buffer); 3038 3039 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src, 3040 unsigned long dst_offset, unsigned long src_offset, 3041 unsigned long len) 3042 { 3043 u64 dst_len = dst->len; 3044 size_t cur; 3045 size_t offset; 3046 struct page *page; 3047 char *kaddr; 3048 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); 3049 unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; 3050 3051 WARN_ON(src->len != dst_len); 3052 3053 offset = (start_offset + dst_offset) & 3054 ((unsigned long)PAGE_CACHE_SIZE - 1); 3055 3056 while(len > 0) { 3057 page = extent_buffer_page(dst, i); 3058 WARN_ON(!PageUptodate(page)); 3059 3060 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset)); 3061 3062 kaddr = kmap_atomic(page, KM_USER0); 3063 read_extent_buffer(src, kaddr + offset, src_offset, cur); 3064 kunmap_atomic(kaddr, KM_USER0); 3065 3066 src_offset += cur; 3067 len -= cur; 3068 offset = 0; 3069 i++; 3070 } 3071 } 3072 EXPORT_SYMBOL(copy_extent_buffer); 3073 3074 static void move_pages(struct page *dst_page, struct page *src_page, 3075 unsigned long dst_off, unsigned long src_off, 3076 unsigned long len) 3077 { 3078 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); 3079 if (dst_page == src_page) { 3080 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len); 3081 } else { 3082 char *src_kaddr = kmap_atomic(src_page, KM_USER1); 3083 char *p = dst_kaddr + dst_off + len; 3084 char *s = src_kaddr + src_off + len; 3085 3086 while (len--) 3087 *--p = *--s; 3088 3089 kunmap_atomic(src_kaddr, KM_USER1); 3090 } 3091 kunmap_atomic(dst_kaddr, KM_USER0); 3092 } 3093 3094 static void copy_pages(struct page *dst_page, struct page *src_page, 3095 unsigned long dst_off, unsigned long src_off, 3096 unsigned long len) 3097 { 3098 char *dst_kaddr = kmap_atomic(dst_page, KM_USER0); 3099 char *src_kaddr; 3100 3101 if (dst_page != src_page) 3102 src_kaddr = kmap_atomic(src_page, KM_USER1); 3103 else 3104 src_kaddr = dst_kaddr; 3105 3106 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); 3107 kunmap_atomic(dst_kaddr, KM_USER0); 3108 if (dst_page != src_page) 3109 kunmap_atomic(src_kaddr, KM_USER1); 3110 } 3111 3112 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 3113 unsigned long src_offset, unsigned long len) 3114 { 3115 size_t cur; 3116 size_t dst_off_in_page; 3117 size_t src_off_in_page; 3118 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); 3119 unsigned long dst_i; 3120 unsigned long src_i; 3121 3122 if (src_offset + len > dst->len) { 3123 printk("memmove bogus src_offset %lu move len %lu len %lu\n", 3124 src_offset, len, dst->len); 3125 BUG_ON(1); 3126 } 3127 if (dst_offset + len > dst->len) { 3128 printk("memmove bogus dst_offset %lu move len %lu len %lu\n", 3129 dst_offset, len, dst->len); 3130 BUG_ON(1); 3131 } 3132 3133 while(len > 0) { 3134 dst_off_in_page = (start_offset + dst_offset) & 3135 ((unsigned long)PAGE_CACHE_SIZE - 1); 3136 src_off_in_page = (start_offset + src_offset) & 3137 ((unsigned long)PAGE_CACHE_SIZE - 1); 3138 3139 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT; 3140 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT; 3141 3142 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - 3143 src_off_in_page)); 3144 cur = min_t(unsigned long, cur, 3145 (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page)); 3146 3147 copy_pages(extent_buffer_page(dst, dst_i), 3148 extent_buffer_page(dst, src_i), 3149 dst_off_in_page, src_off_in_page, cur); 3150 3151 src_offset += cur; 3152 dst_offset += cur; 3153 len -= cur; 3154 } 3155 } 3156 EXPORT_SYMBOL(memcpy_extent_buffer); 3157 3158 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset, 3159 unsigned long src_offset, unsigned long len) 3160 { 3161 size_t cur; 3162 size_t dst_off_in_page; 3163 size_t src_off_in_page; 3164 unsigned long dst_end = dst_offset + len - 1; 3165 unsigned long src_end = src_offset + len - 1; 3166 size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1); 3167 unsigned long dst_i; 3168 unsigned long src_i; 3169 3170 if (src_offset + len > dst->len) { 3171 printk("memmove bogus src_offset %lu move len %lu len %lu\n", 3172 src_offset, len, dst->len); 3173 BUG_ON(1); 3174 } 3175 if (dst_offset + len > dst->len) { 3176 printk("memmove bogus dst_offset %lu move len %lu len %lu\n", 3177 dst_offset, len, dst->len); 3178 BUG_ON(1); 3179 } 3180 if (dst_offset < src_offset) { 3181 memcpy_extent_buffer(dst, dst_offset, src_offset, len); 3182 return; 3183 } 3184 while(len > 0) { 3185 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT; 3186 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT; 3187 3188 dst_off_in_page = (start_offset + dst_end) & 3189 ((unsigned long)PAGE_CACHE_SIZE - 1); 3190 src_off_in_page = (start_offset + src_end) & 3191 ((unsigned long)PAGE_CACHE_SIZE - 1); 3192 3193 cur = min_t(unsigned long, len, src_off_in_page + 1); 3194 cur = min(cur, dst_off_in_page + 1); 3195 move_pages(extent_buffer_page(dst, dst_i), 3196 extent_buffer_page(dst, src_i), 3197 dst_off_in_page - cur + 1, 3198 src_off_in_page - cur + 1, cur); 3199 3200 dst_end -= cur; 3201 src_end -= cur; 3202 len -= cur; 3203 } 3204 } 3205 EXPORT_SYMBOL(memmove_extent_buffer); 3206