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