1 /* 2 * Copyright (C) 2011 Fujitsu. All rights reserved. 3 * Written by Miao Xie <miaox@cn.fujitsu.com> 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public 7 * License v2 as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 12 * General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public 15 * License along with this program; if not, write to the 16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 17 * Boston, MA 021110-1307, USA. 18 */ 19 20 #include <linux/slab.h> 21 #include "delayed-inode.h" 22 #include "disk-io.h" 23 #include "transaction.h" 24 #include "ctree.h" 25 26 #define BTRFS_DELAYED_WRITEBACK 512 27 #define BTRFS_DELAYED_BACKGROUND 128 28 #define BTRFS_DELAYED_BATCH 16 29 30 static struct kmem_cache *delayed_node_cache; 31 32 int __init btrfs_delayed_inode_init(void) 33 { 34 delayed_node_cache = kmem_cache_create("btrfs_delayed_node", 35 sizeof(struct btrfs_delayed_node), 36 0, 37 SLAB_MEM_SPREAD, 38 NULL); 39 if (!delayed_node_cache) 40 return -ENOMEM; 41 return 0; 42 } 43 44 void btrfs_delayed_inode_exit(void) 45 { 46 kmem_cache_destroy(delayed_node_cache); 47 } 48 49 static inline void btrfs_init_delayed_node( 50 struct btrfs_delayed_node *delayed_node, 51 struct btrfs_root *root, u64 inode_id) 52 { 53 delayed_node->root = root; 54 delayed_node->inode_id = inode_id; 55 atomic_set(&delayed_node->refs, 0); 56 delayed_node->ins_root = RB_ROOT; 57 delayed_node->del_root = RB_ROOT; 58 mutex_init(&delayed_node->mutex); 59 INIT_LIST_HEAD(&delayed_node->n_list); 60 INIT_LIST_HEAD(&delayed_node->p_list); 61 } 62 63 static inline int btrfs_is_continuous_delayed_item( 64 struct btrfs_delayed_item *item1, 65 struct btrfs_delayed_item *item2) 66 { 67 if (item1->key.type == BTRFS_DIR_INDEX_KEY && 68 item1->key.objectid == item2->key.objectid && 69 item1->key.type == item2->key.type && 70 item1->key.offset + 1 == item2->key.offset) 71 return 1; 72 return 0; 73 } 74 75 static struct btrfs_delayed_node *btrfs_get_delayed_node( 76 struct btrfs_inode *btrfs_inode) 77 { 78 struct btrfs_root *root = btrfs_inode->root; 79 u64 ino = btrfs_ino(btrfs_inode); 80 struct btrfs_delayed_node *node; 81 82 node = READ_ONCE(btrfs_inode->delayed_node); 83 if (node) { 84 atomic_inc(&node->refs); 85 return node; 86 } 87 88 spin_lock(&root->inode_lock); 89 node = radix_tree_lookup(&root->delayed_nodes_tree, ino); 90 if (node) { 91 if (btrfs_inode->delayed_node) { 92 atomic_inc(&node->refs); /* can be accessed */ 93 BUG_ON(btrfs_inode->delayed_node != node); 94 spin_unlock(&root->inode_lock); 95 return node; 96 } 97 btrfs_inode->delayed_node = node; 98 /* can be accessed and cached in the inode */ 99 atomic_add(2, &node->refs); 100 spin_unlock(&root->inode_lock); 101 return node; 102 } 103 spin_unlock(&root->inode_lock); 104 105 return NULL; 106 } 107 108 /* Will return either the node or PTR_ERR(-ENOMEM) */ 109 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node( 110 struct btrfs_inode *btrfs_inode) 111 { 112 struct btrfs_delayed_node *node; 113 struct btrfs_root *root = btrfs_inode->root; 114 u64 ino = btrfs_ino(btrfs_inode); 115 int ret; 116 117 again: 118 node = btrfs_get_delayed_node(btrfs_inode); 119 if (node) 120 return node; 121 122 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS); 123 if (!node) 124 return ERR_PTR(-ENOMEM); 125 btrfs_init_delayed_node(node, root, ino); 126 127 /* cached in the btrfs inode and can be accessed */ 128 atomic_add(2, &node->refs); 129 130 ret = radix_tree_preload(GFP_NOFS); 131 if (ret) { 132 kmem_cache_free(delayed_node_cache, node); 133 return ERR_PTR(ret); 134 } 135 136 spin_lock(&root->inode_lock); 137 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node); 138 if (ret == -EEXIST) { 139 spin_unlock(&root->inode_lock); 140 kmem_cache_free(delayed_node_cache, node); 141 radix_tree_preload_end(); 142 goto again; 143 } 144 btrfs_inode->delayed_node = node; 145 spin_unlock(&root->inode_lock); 146 radix_tree_preload_end(); 147 148 return node; 149 } 150 151 /* 152 * Call it when holding delayed_node->mutex 153 * 154 * If mod = 1, add this node into the prepared list. 155 */ 156 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root, 157 struct btrfs_delayed_node *node, 158 int mod) 159 { 160 spin_lock(&root->lock); 161 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 162 if (!list_empty(&node->p_list)) 163 list_move_tail(&node->p_list, &root->prepare_list); 164 else if (mod) 165 list_add_tail(&node->p_list, &root->prepare_list); 166 } else { 167 list_add_tail(&node->n_list, &root->node_list); 168 list_add_tail(&node->p_list, &root->prepare_list); 169 atomic_inc(&node->refs); /* inserted into list */ 170 root->nodes++; 171 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); 172 } 173 spin_unlock(&root->lock); 174 } 175 176 /* Call it when holding delayed_node->mutex */ 177 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root, 178 struct btrfs_delayed_node *node) 179 { 180 spin_lock(&root->lock); 181 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 182 root->nodes--; 183 atomic_dec(&node->refs); /* not in the list */ 184 list_del_init(&node->n_list); 185 if (!list_empty(&node->p_list)) 186 list_del_init(&node->p_list); 187 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); 188 } 189 spin_unlock(&root->lock); 190 } 191 192 static struct btrfs_delayed_node *btrfs_first_delayed_node( 193 struct btrfs_delayed_root *delayed_root) 194 { 195 struct list_head *p; 196 struct btrfs_delayed_node *node = NULL; 197 198 spin_lock(&delayed_root->lock); 199 if (list_empty(&delayed_root->node_list)) 200 goto out; 201 202 p = delayed_root->node_list.next; 203 node = list_entry(p, struct btrfs_delayed_node, n_list); 204 atomic_inc(&node->refs); 205 out: 206 spin_unlock(&delayed_root->lock); 207 208 return node; 209 } 210 211 static struct btrfs_delayed_node *btrfs_next_delayed_node( 212 struct btrfs_delayed_node *node) 213 { 214 struct btrfs_delayed_root *delayed_root; 215 struct list_head *p; 216 struct btrfs_delayed_node *next = NULL; 217 218 delayed_root = node->root->fs_info->delayed_root; 219 spin_lock(&delayed_root->lock); 220 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 221 /* not in the list */ 222 if (list_empty(&delayed_root->node_list)) 223 goto out; 224 p = delayed_root->node_list.next; 225 } else if (list_is_last(&node->n_list, &delayed_root->node_list)) 226 goto out; 227 else 228 p = node->n_list.next; 229 230 next = list_entry(p, struct btrfs_delayed_node, n_list); 231 atomic_inc(&next->refs); 232 out: 233 spin_unlock(&delayed_root->lock); 234 235 return next; 236 } 237 238 static void __btrfs_release_delayed_node( 239 struct btrfs_delayed_node *delayed_node, 240 int mod) 241 { 242 struct btrfs_delayed_root *delayed_root; 243 244 if (!delayed_node) 245 return; 246 247 delayed_root = delayed_node->root->fs_info->delayed_root; 248 249 mutex_lock(&delayed_node->mutex); 250 if (delayed_node->count) 251 btrfs_queue_delayed_node(delayed_root, delayed_node, mod); 252 else 253 btrfs_dequeue_delayed_node(delayed_root, delayed_node); 254 mutex_unlock(&delayed_node->mutex); 255 256 if (atomic_dec_and_test(&delayed_node->refs)) { 257 bool free = false; 258 struct btrfs_root *root = delayed_node->root; 259 spin_lock(&root->inode_lock); 260 if (atomic_read(&delayed_node->refs) == 0) { 261 radix_tree_delete(&root->delayed_nodes_tree, 262 delayed_node->inode_id); 263 free = true; 264 } 265 spin_unlock(&root->inode_lock); 266 if (free) 267 kmem_cache_free(delayed_node_cache, delayed_node); 268 } 269 } 270 271 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node) 272 { 273 __btrfs_release_delayed_node(node, 0); 274 } 275 276 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node( 277 struct btrfs_delayed_root *delayed_root) 278 { 279 struct list_head *p; 280 struct btrfs_delayed_node *node = NULL; 281 282 spin_lock(&delayed_root->lock); 283 if (list_empty(&delayed_root->prepare_list)) 284 goto out; 285 286 p = delayed_root->prepare_list.next; 287 list_del_init(p); 288 node = list_entry(p, struct btrfs_delayed_node, p_list); 289 atomic_inc(&node->refs); 290 out: 291 spin_unlock(&delayed_root->lock); 292 293 return node; 294 } 295 296 static inline void btrfs_release_prepared_delayed_node( 297 struct btrfs_delayed_node *node) 298 { 299 __btrfs_release_delayed_node(node, 1); 300 } 301 302 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len) 303 { 304 struct btrfs_delayed_item *item; 305 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS); 306 if (item) { 307 item->data_len = data_len; 308 item->ins_or_del = 0; 309 item->bytes_reserved = 0; 310 item->delayed_node = NULL; 311 atomic_set(&item->refs, 1); 312 } 313 return item; 314 } 315 316 /* 317 * __btrfs_lookup_delayed_item - look up the delayed item by key 318 * @delayed_node: pointer to the delayed node 319 * @key: the key to look up 320 * @prev: used to store the prev item if the right item isn't found 321 * @next: used to store the next item if the right item isn't found 322 * 323 * Note: if we don't find the right item, we will return the prev item and 324 * the next item. 325 */ 326 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item( 327 struct rb_root *root, 328 struct btrfs_key *key, 329 struct btrfs_delayed_item **prev, 330 struct btrfs_delayed_item **next) 331 { 332 struct rb_node *node, *prev_node = NULL; 333 struct btrfs_delayed_item *delayed_item = NULL; 334 int ret = 0; 335 336 node = root->rb_node; 337 338 while (node) { 339 delayed_item = rb_entry(node, struct btrfs_delayed_item, 340 rb_node); 341 prev_node = node; 342 ret = btrfs_comp_cpu_keys(&delayed_item->key, key); 343 if (ret < 0) 344 node = node->rb_right; 345 else if (ret > 0) 346 node = node->rb_left; 347 else 348 return delayed_item; 349 } 350 351 if (prev) { 352 if (!prev_node) 353 *prev = NULL; 354 else if (ret < 0) 355 *prev = delayed_item; 356 else if ((node = rb_prev(prev_node)) != NULL) { 357 *prev = rb_entry(node, struct btrfs_delayed_item, 358 rb_node); 359 } else 360 *prev = NULL; 361 } 362 363 if (next) { 364 if (!prev_node) 365 *next = NULL; 366 else if (ret > 0) 367 *next = delayed_item; 368 else if ((node = rb_next(prev_node)) != NULL) { 369 *next = rb_entry(node, struct btrfs_delayed_item, 370 rb_node); 371 } else 372 *next = NULL; 373 } 374 return NULL; 375 } 376 377 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item( 378 struct btrfs_delayed_node *delayed_node, 379 struct btrfs_key *key) 380 { 381 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, 382 NULL, NULL); 383 } 384 385 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node, 386 struct btrfs_delayed_item *ins, 387 int action) 388 { 389 struct rb_node **p, *node; 390 struct rb_node *parent_node = NULL; 391 struct rb_root *root; 392 struct btrfs_delayed_item *item; 393 int cmp; 394 395 if (action == BTRFS_DELAYED_INSERTION_ITEM) 396 root = &delayed_node->ins_root; 397 else if (action == BTRFS_DELAYED_DELETION_ITEM) 398 root = &delayed_node->del_root; 399 else 400 BUG(); 401 p = &root->rb_node; 402 node = &ins->rb_node; 403 404 while (*p) { 405 parent_node = *p; 406 item = rb_entry(parent_node, struct btrfs_delayed_item, 407 rb_node); 408 409 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key); 410 if (cmp < 0) 411 p = &(*p)->rb_right; 412 else if (cmp > 0) 413 p = &(*p)->rb_left; 414 else 415 return -EEXIST; 416 } 417 418 rb_link_node(node, parent_node, p); 419 rb_insert_color(node, root); 420 ins->delayed_node = delayed_node; 421 ins->ins_or_del = action; 422 423 if (ins->key.type == BTRFS_DIR_INDEX_KEY && 424 action == BTRFS_DELAYED_INSERTION_ITEM && 425 ins->key.offset >= delayed_node->index_cnt) 426 delayed_node->index_cnt = ins->key.offset + 1; 427 428 delayed_node->count++; 429 atomic_inc(&delayed_node->root->fs_info->delayed_root->items); 430 return 0; 431 } 432 433 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node, 434 struct btrfs_delayed_item *item) 435 { 436 return __btrfs_add_delayed_item(node, item, 437 BTRFS_DELAYED_INSERTION_ITEM); 438 } 439 440 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node, 441 struct btrfs_delayed_item *item) 442 { 443 return __btrfs_add_delayed_item(node, item, 444 BTRFS_DELAYED_DELETION_ITEM); 445 } 446 447 static void finish_one_item(struct btrfs_delayed_root *delayed_root) 448 { 449 int seq = atomic_inc_return(&delayed_root->items_seq); 450 451 /* 452 * atomic_dec_return implies a barrier for waitqueue_active 453 */ 454 if ((atomic_dec_return(&delayed_root->items) < 455 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) && 456 waitqueue_active(&delayed_root->wait)) 457 wake_up(&delayed_root->wait); 458 } 459 460 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item) 461 { 462 struct rb_root *root; 463 struct btrfs_delayed_root *delayed_root; 464 465 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root; 466 467 BUG_ON(!delayed_root); 468 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM && 469 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM); 470 471 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM) 472 root = &delayed_item->delayed_node->ins_root; 473 else 474 root = &delayed_item->delayed_node->del_root; 475 476 rb_erase(&delayed_item->rb_node, root); 477 delayed_item->delayed_node->count--; 478 479 finish_one_item(delayed_root); 480 } 481 482 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item) 483 { 484 if (item) { 485 __btrfs_remove_delayed_item(item); 486 if (atomic_dec_and_test(&item->refs)) 487 kfree(item); 488 } 489 } 490 491 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item( 492 struct btrfs_delayed_node *delayed_node) 493 { 494 struct rb_node *p; 495 struct btrfs_delayed_item *item = NULL; 496 497 p = rb_first(&delayed_node->ins_root); 498 if (p) 499 item = rb_entry(p, struct btrfs_delayed_item, rb_node); 500 501 return item; 502 } 503 504 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item( 505 struct btrfs_delayed_node *delayed_node) 506 { 507 struct rb_node *p; 508 struct btrfs_delayed_item *item = NULL; 509 510 p = rb_first(&delayed_node->del_root); 511 if (p) 512 item = rb_entry(p, struct btrfs_delayed_item, rb_node); 513 514 return item; 515 } 516 517 static struct btrfs_delayed_item *__btrfs_next_delayed_item( 518 struct btrfs_delayed_item *item) 519 { 520 struct rb_node *p; 521 struct btrfs_delayed_item *next = NULL; 522 523 p = rb_next(&item->rb_node); 524 if (p) 525 next = rb_entry(p, struct btrfs_delayed_item, rb_node); 526 527 return next; 528 } 529 530 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans, 531 struct btrfs_fs_info *fs_info, 532 struct btrfs_delayed_item *item) 533 { 534 struct btrfs_block_rsv *src_rsv; 535 struct btrfs_block_rsv *dst_rsv; 536 u64 num_bytes; 537 int ret; 538 539 if (!trans->bytes_reserved) 540 return 0; 541 542 src_rsv = trans->block_rsv; 543 dst_rsv = &fs_info->delayed_block_rsv; 544 545 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); 546 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); 547 if (!ret) { 548 trace_btrfs_space_reservation(fs_info, "delayed_item", 549 item->key.objectid, 550 num_bytes, 1); 551 item->bytes_reserved = num_bytes; 552 } 553 554 return ret; 555 } 556 557 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info, 558 struct btrfs_delayed_item *item) 559 { 560 struct btrfs_block_rsv *rsv; 561 562 if (!item->bytes_reserved) 563 return; 564 565 rsv = &fs_info->delayed_block_rsv; 566 trace_btrfs_space_reservation(fs_info, "delayed_item", 567 item->key.objectid, item->bytes_reserved, 568 0); 569 btrfs_block_rsv_release(fs_info, rsv, 570 item->bytes_reserved); 571 } 572 573 static int btrfs_delayed_inode_reserve_metadata( 574 struct btrfs_trans_handle *trans, 575 struct btrfs_root *root, 576 struct btrfs_inode *inode, 577 struct btrfs_delayed_node *node) 578 { 579 struct btrfs_fs_info *fs_info = root->fs_info; 580 struct btrfs_block_rsv *src_rsv; 581 struct btrfs_block_rsv *dst_rsv; 582 u64 num_bytes; 583 int ret; 584 bool release = false; 585 586 src_rsv = trans->block_rsv; 587 dst_rsv = &fs_info->delayed_block_rsv; 588 589 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); 590 591 /* 592 * If our block_rsv is the delalloc block reserve then check and see if 593 * we have our extra reservation for updating the inode. If not fall 594 * through and try to reserve space quickly. 595 * 596 * We used to try and steal from the delalloc block rsv or the global 597 * reserve, but we'd steal a full reservation, which isn't kind. We are 598 * here through delalloc which means we've likely just cowed down close 599 * to the leaf that contains the inode, so we would steal less just 600 * doing the fallback inode update, so if we do end up having to steal 601 * from the global block rsv we hopefully only steal one or two blocks 602 * worth which is less likely to hurt us. 603 */ 604 if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) { 605 spin_lock(&inode->lock); 606 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, 607 &inode->runtime_flags)) 608 release = true; 609 else 610 src_rsv = NULL; 611 spin_unlock(&inode->lock); 612 } 613 614 /* 615 * btrfs_dirty_inode will update the inode under btrfs_join_transaction 616 * which doesn't reserve space for speed. This is a problem since we 617 * still need to reserve space for this update, so try to reserve the 618 * space. 619 * 620 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since 621 * we're accounted for. 622 */ 623 if (!src_rsv || (!trans->bytes_reserved && 624 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) { 625 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, 626 BTRFS_RESERVE_NO_FLUSH); 627 /* 628 * Since we're under a transaction reserve_metadata_bytes could 629 * try to commit the transaction which will make it return 630 * EAGAIN to make us stop the transaction we have, so return 631 * ENOSPC instead so that btrfs_dirty_inode knows what to do. 632 */ 633 if (ret == -EAGAIN) 634 ret = -ENOSPC; 635 if (!ret) { 636 node->bytes_reserved = num_bytes; 637 trace_btrfs_space_reservation(fs_info, 638 "delayed_inode", 639 btrfs_ino(inode), 640 num_bytes, 1); 641 } 642 return ret; 643 } 644 645 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); 646 647 /* 648 * Migrate only takes a reservation, it doesn't touch the size of the 649 * block_rsv. This is to simplify people who don't normally have things 650 * migrated from their block rsv. If they go to release their 651 * reservation, that will decrease the size as well, so if migrate 652 * reduced size we'd end up with a negative size. But for the 653 * delalloc_meta_reserved stuff we will only know to drop 1 reservation, 654 * but we could in fact do this reserve/migrate dance several times 655 * between the time we did the original reservation and we'd clean it 656 * up. So to take care of this, release the space for the meta 657 * reservation here. I think it may be time for a documentation page on 658 * how block rsvs. work. 659 */ 660 if (!ret) { 661 trace_btrfs_space_reservation(fs_info, "delayed_inode", 662 btrfs_ino(inode), num_bytes, 1); 663 node->bytes_reserved = num_bytes; 664 } 665 666 if (release) { 667 trace_btrfs_space_reservation(fs_info, "delalloc", 668 btrfs_ino(inode), num_bytes, 0); 669 btrfs_block_rsv_release(fs_info, src_rsv, num_bytes); 670 } 671 672 return ret; 673 } 674 675 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info, 676 struct btrfs_delayed_node *node) 677 { 678 struct btrfs_block_rsv *rsv; 679 680 if (!node->bytes_reserved) 681 return; 682 683 rsv = &fs_info->delayed_block_rsv; 684 trace_btrfs_space_reservation(fs_info, "delayed_inode", 685 node->inode_id, node->bytes_reserved, 0); 686 btrfs_block_rsv_release(fs_info, rsv, 687 node->bytes_reserved); 688 node->bytes_reserved = 0; 689 } 690 691 /* 692 * This helper will insert some continuous items into the same leaf according 693 * to the free space of the leaf. 694 */ 695 static int btrfs_batch_insert_items(struct btrfs_root *root, 696 struct btrfs_path *path, 697 struct btrfs_delayed_item *item) 698 { 699 struct btrfs_fs_info *fs_info = root->fs_info; 700 struct btrfs_delayed_item *curr, *next; 701 int free_space; 702 int total_data_size = 0, total_size = 0; 703 struct extent_buffer *leaf; 704 char *data_ptr; 705 struct btrfs_key *keys; 706 u32 *data_size; 707 struct list_head head; 708 int slot; 709 int nitems; 710 int i; 711 int ret = 0; 712 713 BUG_ON(!path->nodes[0]); 714 715 leaf = path->nodes[0]; 716 free_space = btrfs_leaf_free_space(fs_info, leaf); 717 INIT_LIST_HEAD(&head); 718 719 next = item; 720 nitems = 0; 721 722 /* 723 * count the number of the continuous items that we can insert in batch 724 */ 725 while (total_size + next->data_len + sizeof(struct btrfs_item) <= 726 free_space) { 727 total_data_size += next->data_len; 728 total_size += next->data_len + sizeof(struct btrfs_item); 729 list_add_tail(&next->tree_list, &head); 730 nitems++; 731 732 curr = next; 733 next = __btrfs_next_delayed_item(curr); 734 if (!next) 735 break; 736 737 if (!btrfs_is_continuous_delayed_item(curr, next)) 738 break; 739 } 740 741 if (!nitems) { 742 ret = 0; 743 goto out; 744 } 745 746 /* 747 * we need allocate some memory space, but it might cause the task 748 * to sleep, so we set all locked nodes in the path to blocking locks 749 * first. 750 */ 751 btrfs_set_path_blocking(path); 752 753 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS); 754 if (!keys) { 755 ret = -ENOMEM; 756 goto out; 757 } 758 759 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS); 760 if (!data_size) { 761 ret = -ENOMEM; 762 goto error; 763 } 764 765 /* get keys of all the delayed items */ 766 i = 0; 767 list_for_each_entry(next, &head, tree_list) { 768 keys[i] = next->key; 769 data_size[i] = next->data_len; 770 i++; 771 } 772 773 /* reset all the locked nodes in the patch to spinning locks. */ 774 btrfs_clear_path_blocking(path, NULL, 0); 775 776 /* insert the keys of the items */ 777 setup_items_for_insert(root, path, keys, data_size, 778 total_data_size, total_size, nitems); 779 780 /* insert the dir index items */ 781 slot = path->slots[0]; 782 list_for_each_entry_safe(curr, next, &head, tree_list) { 783 data_ptr = btrfs_item_ptr(leaf, slot, char); 784 write_extent_buffer(leaf, &curr->data, 785 (unsigned long)data_ptr, 786 curr->data_len); 787 slot++; 788 789 btrfs_delayed_item_release_metadata(fs_info, curr); 790 791 list_del(&curr->tree_list); 792 btrfs_release_delayed_item(curr); 793 } 794 795 error: 796 kfree(data_size); 797 kfree(keys); 798 out: 799 return ret; 800 } 801 802 /* 803 * This helper can just do simple insertion that needn't extend item for new 804 * data, such as directory name index insertion, inode insertion. 805 */ 806 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans, 807 struct btrfs_root *root, 808 struct btrfs_path *path, 809 struct btrfs_delayed_item *delayed_item) 810 { 811 struct btrfs_fs_info *fs_info = root->fs_info; 812 struct extent_buffer *leaf; 813 char *ptr; 814 int ret; 815 816 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key, 817 delayed_item->data_len); 818 if (ret < 0 && ret != -EEXIST) 819 return ret; 820 821 leaf = path->nodes[0]; 822 823 ptr = btrfs_item_ptr(leaf, path->slots[0], char); 824 825 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr, 826 delayed_item->data_len); 827 btrfs_mark_buffer_dirty(leaf); 828 829 btrfs_delayed_item_release_metadata(fs_info, delayed_item); 830 return 0; 831 } 832 833 /* 834 * we insert an item first, then if there are some continuous items, we try 835 * to insert those items into the same leaf. 836 */ 837 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans, 838 struct btrfs_path *path, 839 struct btrfs_root *root, 840 struct btrfs_delayed_node *node) 841 { 842 struct btrfs_delayed_item *curr, *prev; 843 int ret = 0; 844 845 do_again: 846 mutex_lock(&node->mutex); 847 curr = __btrfs_first_delayed_insertion_item(node); 848 if (!curr) 849 goto insert_end; 850 851 ret = btrfs_insert_delayed_item(trans, root, path, curr); 852 if (ret < 0) { 853 btrfs_release_path(path); 854 goto insert_end; 855 } 856 857 prev = curr; 858 curr = __btrfs_next_delayed_item(prev); 859 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) { 860 /* insert the continuous items into the same leaf */ 861 path->slots[0]++; 862 btrfs_batch_insert_items(root, path, curr); 863 } 864 btrfs_release_delayed_item(prev); 865 btrfs_mark_buffer_dirty(path->nodes[0]); 866 867 btrfs_release_path(path); 868 mutex_unlock(&node->mutex); 869 goto do_again; 870 871 insert_end: 872 mutex_unlock(&node->mutex); 873 return ret; 874 } 875 876 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans, 877 struct btrfs_root *root, 878 struct btrfs_path *path, 879 struct btrfs_delayed_item *item) 880 { 881 struct btrfs_fs_info *fs_info = root->fs_info; 882 struct btrfs_delayed_item *curr, *next; 883 struct extent_buffer *leaf; 884 struct btrfs_key key; 885 struct list_head head; 886 int nitems, i, last_item; 887 int ret = 0; 888 889 BUG_ON(!path->nodes[0]); 890 891 leaf = path->nodes[0]; 892 893 i = path->slots[0]; 894 last_item = btrfs_header_nritems(leaf) - 1; 895 if (i > last_item) 896 return -ENOENT; /* FIXME: Is errno suitable? */ 897 898 next = item; 899 INIT_LIST_HEAD(&head); 900 btrfs_item_key_to_cpu(leaf, &key, i); 901 nitems = 0; 902 /* 903 * count the number of the dir index items that we can delete in batch 904 */ 905 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) { 906 list_add_tail(&next->tree_list, &head); 907 nitems++; 908 909 curr = next; 910 next = __btrfs_next_delayed_item(curr); 911 if (!next) 912 break; 913 914 if (!btrfs_is_continuous_delayed_item(curr, next)) 915 break; 916 917 i++; 918 if (i > last_item) 919 break; 920 btrfs_item_key_to_cpu(leaf, &key, i); 921 } 922 923 if (!nitems) 924 return 0; 925 926 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems); 927 if (ret) 928 goto out; 929 930 list_for_each_entry_safe(curr, next, &head, tree_list) { 931 btrfs_delayed_item_release_metadata(fs_info, curr); 932 list_del(&curr->tree_list); 933 btrfs_release_delayed_item(curr); 934 } 935 936 out: 937 return ret; 938 } 939 940 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans, 941 struct btrfs_path *path, 942 struct btrfs_root *root, 943 struct btrfs_delayed_node *node) 944 { 945 struct btrfs_delayed_item *curr, *prev; 946 int ret = 0; 947 948 do_again: 949 mutex_lock(&node->mutex); 950 curr = __btrfs_first_delayed_deletion_item(node); 951 if (!curr) 952 goto delete_fail; 953 954 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1); 955 if (ret < 0) 956 goto delete_fail; 957 else if (ret > 0) { 958 /* 959 * can't find the item which the node points to, so this node 960 * is invalid, just drop it. 961 */ 962 prev = curr; 963 curr = __btrfs_next_delayed_item(prev); 964 btrfs_release_delayed_item(prev); 965 ret = 0; 966 btrfs_release_path(path); 967 if (curr) { 968 mutex_unlock(&node->mutex); 969 goto do_again; 970 } else 971 goto delete_fail; 972 } 973 974 btrfs_batch_delete_items(trans, root, path, curr); 975 btrfs_release_path(path); 976 mutex_unlock(&node->mutex); 977 goto do_again; 978 979 delete_fail: 980 btrfs_release_path(path); 981 mutex_unlock(&node->mutex); 982 return ret; 983 } 984 985 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node) 986 { 987 struct btrfs_delayed_root *delayed_root; 988 989 if (delayed_node && 990 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 991 BUG_ON(!delayed_node->root); 992 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); 993 delayed_node->count--; 994 995 delayed_root = delayed_node->root->fs_info->delayed_root; 996 finish_one_item(delayed_root); 997 } 998 } 999 1000 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node) 1001 { 1002 struct btrfs_delayed_root *delayed_root; 1003 1004 ASSERT(delayed_node->root); 1005 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); 1006 delayed_node->count--; 1007 1008 delayed_root = delayed_node->root->fs_info->delayed_root; 1009 finish_one_item(delayed_root); 1010 } 1011 1012 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, 1013 struct btrfs_root *root, 1014 struct btrfs_path *path, 1015 struct btrfs_delayed_node *node) 1016 { 1017 struct btrfs_fs_info *fs_info = root->fs_info; 1018 struct btrfs_key key; 1019 struct btrfs_inode_item *inode_item; 1020 struct extent_buffer *leaf; 1021 int mod; 1022 int ret; 1023 1024 key.objectid = node->inode_id; 1025 key.type = BTRFS_INODE_ITEM_KEY; 1026 key.offset = 0; 1027 1028 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) 1029 mod = -1; 1030 else 1031 mod = 1; 1032 1033 ret = btrfs_lookup_inode(trans, root, path, &key, mod); 1034 if (ret > 0) { 1035 btrfs_release_path(path); 1036 return -ENOENT; 1037 } else if (ret < 0) { 1038 return ret; 1039 } 1040 1041 leaf = path->nodes[0]; 1042 inode_item = btrfs_item_ptr(leaf, path->slots[0], 1043 struct btrfs_inode_item); 1044 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item, 1045 sizeof(struct btrfs_inode_item)); 1046 btrfs_mark_buffer_dirty(leaf); 1047 1048 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) 1049 goto no_iref; 1050 1051 path->slots[0]++; 1052 if (path->slots[0] >= btrfs_header_nritems(leaf)) 1053 goto search; 1054 again: 1055 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 1056 if (key.objectid != node->inode_id) 1057 goto out; 1058 1059 if (key.type != BTRFS_INODE_REF_KEY && 1060 key.type != BTRFS_INODE_EXTREF_KEY) 1061 goto out; 1062 1063 /* 1064 * Delayed iref deletion is for the inode who has only one link, 1065 * so there is only one iref. The case that several irefs are 1066 * in the same item doesn't exist. 1067 */ 1068 btrfs_del_item(trans, root, path); 1069 out: 1070 btrfs_release_delayed_iref(node); 1071 no_iref: 1072 btrfs_release_path(path); 1073 err_out: 1074 btrfs_delayed_inode_release_metadata(fs_info, node); 1075 btrfs_release_delayed_inode(node); 1076 1077 return ret; 1078 1079 search: 1080 btrfs_release_path(path); 1081 1082 key.type = BTRFS_INODE_EXTREF_KEY; 1083 key.offset = -1; 1084 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1085 if (ret < 0) 1086 goto err_out; 1087 ASSERT(ret); 1088 1089 ret = 0; 1090 leaf = path->nodes[0]; 1091 path->slots[0]--; 1092 goto again; 1093 } 1094 1095 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, 1096 struct btrfs_root *root, 1097 struct btrfs_path *path, 1098 struct btrfs_delayed_node *node) 1099 { 1100 int ret; 1101 1102 mutex_lock(&node->mutex); 1103 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) { 1104 mutex_unlock(&node->mutex); 1105 return 0; 1106 } 1107 1108 ret = __btrfs_update_delayed_inode(trans, root, path, node); 1109 mutex_unlock(&node->mutex); 1110 return ret; 1111 } 1112 1113 static inline int 1114 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, 1115 struct btrfs_path *path, 1116 struct btrfs_delayed_node *node) 1117 { 1118 int ret; 1119 1120 ret = btrfs_insert_delayed_items(trans, path, node->root, node); 1121 if (ret) 1122 return ret; 1123 1124 ret = btrfs_delete_delayed_items(trans, path, node->root, node); 1125 if (ret) 1126 return ret; 1127 1128 ret = btrfs_update_delayed_inode(trans, node->root, path, node); 1129 return ret; 1130 } 1131 1132 /* 1133 * Called when committing the transaction. 1134 * Returns 0 on success. 1135 * Returns < 0 on error and returns with an aborted transaction with any 1136 * outstanding delayed items cleaned up. 1137 */ 1138 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, 1139 struct btrfs_fs_info *fs_info, int nr) 1140 { 1141 struct btrfs_delayed_root *delayed_root; 1142 struct btrfs_delayed_node *curr_node, *prev_node; 1143 struct btrfs_path *path; 1144 struct btrfs_block_rsv *block_rsv; 1145 int ret = 0; 1146 bool count = (nr > 0); 1147 1148 if (trans->aborted) 1149 return -EIO; 1150 1151 path = btrfs_alloc_path(); 1152 if (!path) 1153 return -ENOMEM; 1154 path->leave_spinning = 1; 1155 1156 block_rsv = trans->block_rsv; 1157 trans->block_rsv = &fs_info->delayed_block_rsv; 1158 1159 delayed_root = fs_info->delayed_root; 1160 1161 curr_node = btrfs_first_delayed_node(delayed_root); 1162 while (curr_node && (!count || (count && nr--))) { 1163 ret = __btrfs_commit_inode_delayed_items(trans, path, 1164 curr_node); 1165 if (ret) { 1166 btrfs_release_delayed_node(curr_node); 1167 curr_node = NULL; 1168 btrfs_abort_transaction(trans, ret); 1169 break; 1170 } 1171 1172 prev_node = curr_node; 1173 curr_node = btrfs_next_delayed_node(curr_node); 1174 btrfs_release_delayed_node(prev_node); 1175 } 1176 1177 if (curr_node) 1178 btrfs_release_delayed_node(curr_node); 1179 btrfs_free_path(path); 1180 trans->block_rsv = block_rsv; 1181 1182 return ret; 1183 } 1184 1185 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans, 1186 struct btrfs_fs_info *fs_info) 1187 { 1188 return __btrfs_run_delayed_items(trans, fs_info, -1); 1189 } 1190 1191 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, 1192 struct btrfs_fs_info *fs_info, int nr) 1193 { 1194 return __btrfs_run_delayed_items(trans, fs_info, nr); 1195 } 1196 1197 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, 1198 struct btrfs_inode *inode) 1199 { 1200 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1201 struct btrfs_path *path; 1202 struct btrfs_block_rsv *block_rsv; 1203 int ret; 1204 1205 if (!delayed_node) 1206 return 0; 1207 1208 mutex_lock(&delayed_node->mutex); 1209 if (!delayed_node->count) { 1210 mutex_unlock(&delayed_node->mutex); 1211 btrfs_release_delayed_node(delayed_node); 1212 return 0; 1213 } 1214 mutex_unlock(&delayed_node->mutex); 1215 1216 path = btrfs_alloc_path(); 1217 if (!path) { 1218 btrfs_release_delayed_node(delayed_node); 1219 return -ENOMEM; 1220 } 1221 path->leave_spinning = 1; 1222 1223 block_rsv = trans->block_rsv; 1224 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; 1225 1226 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node); 1227 1228 btrfs_release_delayed_node(delayed_node); 1229 btrfs_free_path(path); 1230 trans->block_rsv = block_rsv; 1231 1232 return ret; 1233 } 1234 1235 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode) 1236 { 1237 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); 1238 struct btrfs_trans_handle *trans; 1239 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1240 struct btrfs_path *path; 1241 struct btrfs_block_rsv *block_rsv; 1242 int ret; 1243 1244 if (!delayed_node) 1245 return 0; 1246 1247 mutex_lock(&delayed_node->mutex); 1248 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1249 mutex_unlock(&delayed_node->mutex); 1250 btrfs_release_delayed_node(delayed_node); 1251 return 0; 1252 } 1253 mutex_unlock(&delayed_node->mutex); 1254 1255 trans = btrfs_join_transaction(delayed_node->root); 1256 if (IS_ERR(trans)) { 1257 ret = PTR_ERR(trans); 1258 goto out; 1259 } 1260 1261 path = btrfs_alloc_path(); 1262 if (!path) { 1263 ret = -ENOMEM; 1264 goto trans_out; 1265 } 1266 path->leave_spinning = 1; 1267 1268 block_rsv = trans->block_rsv; 1269 trans->block_rsv = &fs_info->delayed_block_rsv; 1270 1271 mutex_lock(&delayed_node->mutex); 1272 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) 1273 ret = __btrfs_update_delayed_inode(trans, delayed_node->root, 1274 path, delayed_node); 1275 else 1276 ret = 0; 1277 mutex_unlock(&delayed_node->mutex); 1278 1279 btrfs_free_path(path); 1280 trans->block_rsv = block_rsv; 1281 trans_out: 1282 btrfs_end_transaction(trans); 1283 btrfs_btree_balance_dirty(fs_info); 1284 out: 1285 btrfs_release_delayed_node(delayed_node); 1286 1287 return ret; 1288 } 1289 1290 void btrfs_remove_delayed_node(struct btrfs_inode *inode) 1291 { 1292 struct btrfs_delayed_node *delayed_node; 1293 1294 delayed_node = READ_ONCE(inode->delayed_node); 1295 if (!delayed_node) 1296 return; 1297 1298 inode->delayed_node = NULL; 1299 btrfs_release_delayed_node(delayed_node); 1300 } 1301 1302 struct btrfs_async_delayed_work { 1303 struct btrfs_delayed_root *delayed_root; 1304 int nr; 1305 struct btrfs_work work; 1306 }; 1307 1308 static void btrfs_async_run_delayed_root(struct btrfs_work *work) 1309 { 1310 struct btrfs_async_delayed_work *async_work; 1311 struct btrfs_delayed_root *delayed_root; 1312 struct btrfs_trans_handle *trans; 1313 struct btrfs_path *path; 1314 struct btrfs_delayed_node *delayed_node = NULL; 1315 struct btrfs_root *root; 1316 struct btrfs_block_rsv *block_rsv; 1317 int total_done = 0; 1318 1319 async_work = container_of(work, struct btrfs_async_delayed_work, work); 1320 delayed_root = async_work->delayed_root; 1321 1322 path = btrfs_alloc_path(); 1323 if (!path) 1324 goto out; 1325 1326 again: 1327 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2) 1328 goto free_path; 1329 1330 delayed_node = btrfs_first_prepared_delayed_node(delayed_root); 1331 if (!delayed_node) 1332 goto free_path; 1333 1334 path->leave_spinning = 1; 1335 root = delayed_node->root; 1336 1337 trans = btrfs_join_transaction(root); 1338 if (IS_ERR(trans)) 1339 goto release_path; 1340 1341 block_rsv = trans->block_rsv; 1342 trans->block_rsv = &root->fs_info->delayed_block_rsv; 1343 1344 __btrfs_commit_inode_delayed_items(trans, path, delayed_node); 1345 1346 trans->block_rsv = block_rsv; 1347 btrfs_end_transaction(trans); 1348 btrfs_btree_balance_dirty_nodelay(root->fs_info); 1349 1350 release_path: 1351 btrfs_release_path(path); 1352 total_done++; 1353 1354 btrfs_release_prepared_delayed_node(delayed_node); 1355 if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) || 1356 total_done < async_work->nr) 1357 goto again; 1358 1359 free_path: 1360 btrfs_free_path(path); 1361 out: 1362 wake_up(&delayed_root->wait); 1363 kfree(async_work); 1364 } 1365 1366 1367 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root, 1368 struct btrfs_fs_info *fs_info, int nr) 1369 { 1370 struct btrfs_async_delayed_work *async_work; 1371 1372 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND || 1373 btrfs_workqueue_normal_congested(fs_info->delayed_workers)) 1374 return 0; 1375 1376 async_work = kmalloc(sizeof(*async_work), GFP_NOFS); 1377 if (!async_work) 1378 return -ENOMEM; 1379 1380 async_work->delayed_root = delayed_root; 1381 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper, 1382 btrfs_async_run_delayed_root, NULL, NULL); 1383 async_work->nr = nr; 1384 1385 btrfs_queue_work(fs_info->delayed_workers, &async_work->work); 1386 return 0; 1387 } 1388 1389 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info) 1390 { 1391 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root)); 1392 } 1393 1394 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq) 1395 { 1396 int val = atomic_read(&delayed_root->items_seq); 1397 1398 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH) 1399 return 1; 1400 1401 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) 1402 return 1; 1403 1404 return 0; 1405 } 1406 1407 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info) 1408 { 1409 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root; 1410 1411 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) 1412 return; 1413 1414 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) { 1415 int seq; 1416 int ret; 1417 1418 seq = atomic_read(&delayed_root->items_seq); 1419 1420 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0); 1421 if (ret) 1422 return; 1423 1424 wait_event_interruptible(delayed_root->wait, 1425 could_end_wait(delayed_root, seq)); 1426 return; 1427 } 1428 1429 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH); 1430 } 1431 1432 /* Will return 0 or -ENOMEM */ 1433 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans, 1434 struct btrfs_fs_info *fs_info, 1435 const char *name, int name_len, 1436 struct btrfs_inode *dir, 1437 struct btrfs_disk_key *disk_key, u8 type, 1438 u64 index) 1439 { 1440 struct btrfs_delayed_node *delayed_node; 1441 struct btrfs_delayed_item *delayed_item; 1442 struct btrfs_dir_item *dir_item; 1443 int ret; 1444 1445 delayed_node = btrfs_get_or_create_delayed_node(dir); 1446 if (IS_ERR(delayed_node)) 1447 return PTR_ERR(delayed_node); 1448 1449 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len); 1450 if (!delayed_item) { 1451 ret = -ENOMEM; 1452 goto release_node; 1453 } 1454 1455 delayed_item->key.objectid = btrfs_ino(dir); 1456 delayed_item->key.type = BTRFS_DIR_INDEX_KEY; 1457 delayed_item->key.offset = index; 1458 1459 dir_item = (struct btrfs_dir_item *)delayed_item->data; 1460 dir_item->location = *disk_key; 1461 btrfs_set_stack_dir_transid(dir_item, trans->transid); 1462 btrfs_set_stack_dir_data_len(dir_item, 0); 1463 btrfs_set_stack_dir_name_len(dir_item, name_len); 1464 btrfs_set_stack_dir_type(dir_item, type); 1465 memcpy((char *)(dir_item + 1), name, name_len); 1466 1467 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item); 1468 /* 1469 * we have reserved enough space when we start a new transaction, 1470 * so reserving metadata failure is impossible 1471 */ 1472 BUG_ON(ret); 1473 1474 1475 mutex_lock(&delayed_node->mutex); 1476 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item); 1477 if (unlikely(ret)) { 1478 btrfs_err(fs_info, 1479 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)", 1480 name_len, name, delayed_node->root->objectid, 1481 delayed_node->inode_id, ret); 1482 BUG(); 1483 } 1484 mutex_unlock(&delayed_node->mutex); 1485 1486 release_node: 1487 btrfs_release_delayed_node(delayed_node); 1488 return ret; 1489 } 1490 1491 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info, 1492 struct btrfs_delayed_node *node, 1493 struct btrfs_key *key) 1494 { 1495 struct btrfs_delayed_item *item; 1496 1497 mutex_lock(&node->mutex); 1498 item = __btrfs_lookup_delayed_insertion_item(node, key); 1499 if (!item) { 1500 mutex_unlock(&node->mutex); 1501 return 1; 1502 } 1503 1504 btrfs_delayed_item_release_metadata(fs_info, item); 1505 btrfs_release_delayed_item(item); 1506 mutex_unlock(&node->mutex); 1507 return 0; 1508 } 1509 1510 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans, 1511 struct btrfs_fs_info *fs_info, 1512 struct btrfs_inode *dir, u64 index) 1513 { 1514 struct btrfs_delayed_node *node; 1515 struct btrfs_delayed_item *item; 1516 struct btrfs_key item_key; 1517 int ret; 1518 1519 node = btrfs_get_or_create_delayed_node(dir); 1520 if (IS_ERR(node)) 1521 return PTR_ERR(node); 1522 1523 item_key.objectid = btrfs_ino(dir); 1524 item_key.type = BTRFS_DIR_INDEX_KEY; 1525 item_key.offset = index; 1526 1527 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key); 1528 if (!ret) 1529 goto end; 1530 1531 item = btrfs_alloc_delayed_item(0); 1532 if (!item) { 1533 ret = -ENOMEM; 1534 goto end; 1535 } 1536 1537 item->key = item_key; 1538 1539 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item); 1540 /* 1541 * we have reserved enough space when we start a new transaction, 1542 * so reserving metadata failure is impossible. 1543 */ 1544 BUG_ON(ret); 1545 1546 mutex_lock(&node->mutex); 1547 ret = __btrfs_add_delayed_deletion_item(node, item); 1548 if (unlikely(ret)) { 1549 btrfs_err(fs_info, 1550 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)", 1551 index, node->root->objectid, node->inode_id, ret); 1552 BUG(); 1553 } 1554 mutex_unlock(&node->mutex); 1555 end: 1556 btrfs_release_delayed_node(node); 1557 return ret; 1558 } 1559 1560 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode) 1561 { 1562 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1563 1564 if (!delayed_node) 1565 return -ENOENT; 1566 1567 /* 1568 * Since we have held i_mutex of this directory, it is impossible that 1569 * a new directory index is added into the delayed node and index_cnt 1570 * is updated now. So we needn't lock the delayed node. 1571 */ 1572 if (!delayed_node->index_cnt) { 1573 btrfs_release_delayed_node(delayed_node); 1574 return -EINVAL; 1575 } 1576 1577 inode->index_cnt = delayed_node->index_cnt; 1578 btrfs_release_delayed_node(delayed_node); 1579 return 0; 1580 } 1581 1582 bool btrfs_readdir_get_delayed_items(struct inode *inode, 1583 struct list_head *ins_list, 1584 struct list_head *del_list) 1585 { 1586 struct btrfs_delayed_node *delayed_node; 1587 struct btrfs_delayed_item *item; 1588 1589 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode)); 1590 if (!delayed_node) 1591 return false; 1592 1593 /* 1594 * We can only do one readdir with delayed items at a time because of 1595 * item->readdir_list. 1596 */ 1597 inode_unlock_shared(inode); 1598 inode_lock(inode); 1599 1600 mutex_lock(&delayed_node->mutex); 1601 item = __btrfs_first_delayed_insertion_item(delayed_node); 1602 while (item) { 1603 atomic_inc(&item->refs); 1604 list_add_tail(&item->readdir_list, ins_list); 1605 item = __btrfs_next_delayed_item(item); 1606 } 1607 1608 item = __btrfs_first_delayed_deletion_item(delayed_node); 1609 while (item) { 1610 atomic_inc(&item->refs); 1611 list_add_tail(&item->readdir_list, del_list); 1612 item = __btrfs_next_delayed_item(item); 1613 } 1614 mutex_unlock(&delayed_node->mutex); 1615 /* 1616 * This delayed node is still cached in the btrfs inode, so refs 1617 * must be > 1 now, and we needn't check it is going to be freed 1618 * or not. 1619 * 1620 * Besides that, this function is used to read dir, we do not 1621 * insert/delete delayed items in this period. So we also needn't 1622 * requeue or dequeue this delayed node. 1623 */ 1624 atomic_dec(&delayed_node->refs); 1625 1626 return true; 1627 } 1628 1629 void btrfs_readdir_put_delayed_items(struct inode *inode, 1630 struct list_head *ins_list, 1631 struct list_head *del_list) 1632 { 1633 struct btrfs_delayed_item *curr, *next; 1634 1635 list_for_each_entry_safe(curr, next, ins_list, readdir_list) { 1636 list_del(&curr->readdir_list); 1637 if (atomic_dec_and_test(&curr->refs)) 1638 kfree(curr); 1639 } 1640 1641 list_for_each_entry_safe(curr, next, del_list, readdir_list) { 1642 list_del(&curr->readdir_list); 1643 if (atomic_dec_and_test(&curr->refs)) 1644 kfree(curr); 1645 } 1646 1647 /* 1648 * The VFS is going to do up_read(), so we need to downgrade back to a 1649 * read lock. 1650 */ 1651 downgrade_write(&inode->i_rwsem); 1652 } 1653 1654 int btrfs_should_delete_dir_index(struct list_head *del_list, 1655 u64 index) 1656 { 1657 struct btrfs_delayed_item *curr, *next; 1658 int ret; 1659 1660 if (list_empty(del_list)) 1661 return 0; 1662 1663 list_for_each_entry_safe(curr, next, del_list, readdir_list) { 1664 if (curr->key.offset > index) 1665 break; 1666 1667 list_del(&curr->readdir_list); 1668 ret = (curr->key.offset == index); 1669 1670 if (atomic_dec_and_test(&curr->refs)) 1671 kfree(curr); 1672 1673 if (ret) 1674 return 1; 1675 else 1676 continue; 1677 } 1678 return 0; 1679 } 1680 1681 /* 1682 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree 1683 * 1684 */ 1685 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx, 1686 struct list_head *ins_list) 1687 { 1688 struct btrfs_dir_item *di; 1689 struct btrfs_delayed_item *curr, *next; 1690 struct btrfs_key location; 1691 char *name; 1692 int name_len; 1693 int over = 0; 1694 unsigned char d_type; 1695 1696 if (list_empty(ins_list)) 1697 return 0; 1698 1699 /* 1700 * Changing the data of the delayed item is impossible. So 1701 * we needn't lock them. And we have held i_mutex of the 1702 * directory, nobody can delete any directory indexes now. 1703 */ 1704 list_for_each_entry_safe(curr, next, ins_list, readdir_list) { 1705 list_del(&curr->readdir_list); 1706 1707 if (curr->key.offset < ctx->pos) { 1708 if (atomic_dec_and_test(&curr->refs)) 1709 kfree(curr); 1710 continue; 1711 } 1712 1713 ctx->pos = curr->key.offset; 1714 1715 di = (struct btrfs_dir_item *)curr->data; 1716 name = (char *)(di + 1); 1717 name_len = btrfs_stack_dir_name_len(di); 1718 1719 d_type = btrfs_filetype_table[di->type]; 1720 btrfs_disk_key_to_cpu(&location, &di->location); 1721 1722 over = !dir_emit(ctx, name, name_len, 1723 location.objectid, d_type); 1724 1725 if (atomic_dec_and_test(&curr->refs)) 1726 kfree(curr); 1727 1728 if (over) 1729 return 1; 1730 } 1731 return 0; 1732 } 1733 1734 static void fill_stack_inode_item(struct btrfs_trans_handle *trans, 1735 struct btrfs_inode_item *inode_item, 1736 struct inode *inode) 1737 { 1738 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode)); 1739 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode)); 1740 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size); 1741 btrfs_set_stack_inode_mode(inode_item, inode->i_mode); 1742 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink); 1743 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode)); 1744 btrfs_set_stack_inode_generation(inode_item, 1745 BTRFS_I(inode)->generation); 1746 btrfs_set_stack_inode_sequence(inode_item, inode->i_version); 1747 btrfs_set_stack_inode_transid(inode_item, trans->transid); 1748 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev); 1749 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags); 1750 btrfs_set_stack_inode_block_group(inode_item, 0); 1751 1752 btrfs_set_stack_timespec_sec(&inode_item->atime, 1753 inode->i_atime.tv_sec); 1754 btrfs_set_stack_timespec_nsec(&inode_item->atime, 1755 inode->i_atime.tv_nsec); 1756 1757 btrfs_set_stack_timespec_sec(&inode_item->mtime, 1758 inode->i_mtime.tv_sec); 1759 btrfs_set_stack_timespec_nsec(&inode_item->mtime, 1760 inode->i_mtime.tv_nsec); 1761 1762 btrfs_set_stack_timespec_sec(&inode_item->ctime, 1763 inode->i_ctime.tv_sec); 1764 btrfs_set_stack_timespec_nsec(&inode_item->ctime, 1765 inode->i_ctime.tv_nsec); 1766 1767 btrfs_set_stack_timespec_sec(&inode_item->otime, 1768 BTRFS_I(inode)->i_otime.tv_sec); 1769 btrfs_set_stack_timespec_nsec(&inode_item->otime, 1770 BTRFS_I(inode)->i_otime.tv_nsec); 1771 } 1772 1773 int btrfs_fill_inode(struct inode *inode, u32 *rdev) 1774 { 1775 struct btrfs_delayed_node *delayed_node; 1776 struct btrfs_inode_item *inode_item; 1777 1778 delayed_node = btrfs_get_delayed_node(BTRFS_I(inode)); 1779 if (!delayed_node) 1780 return -ENOENT; 1781 1782 mutex_lock(&delayed_node->mutex); 1783 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1784 mutex_unlock(&delayed_node->mutex); 1785 btrfs_release_delayed_node(delayed_node); 1786 return -ENOENT; 1787 } 1788 1789 inode_item = &delayed_node->inode_item; 1790 1791 i_uid_write(inode, btrfs_stack_inode_uid(inode_item)); 1792 i_gid_write(inode, btrfs_stack_inode_gid(inode_item)); 1793 btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item)); 1794 inode->i_mode = btrfs_stack_inode_mode(inode_item); 1795 set_nlink(inode, btrfs_stack_inode_nlink(inode_item)); 1796 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item)); 1797 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item); 1798 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item); 1799 1800 inode->i_version = btrfs_stack_inode_sequence(inode_item); 1801 inode->i_rdev = 0; 1802 *rdev = btrfs_stack_inode_rdev(inode_item); 1803 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item); 1804 1805 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime); 1806 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime); 1807 1808 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime); 1809 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime); 1810 1811 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime); 1812 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime); 1813 1814 BTRFS_I(inode)->i_otime.tv_sec = 1815 btrfs_stack_timespec_sec(&inode_item->otime); 1816 BTRFS_I(inode)->i_otime.tv_nsec = 1817 btrfs_stack_timespec_nsec(&inode_item->otime); 1818 1819 inode->i_generation = BTRFS_I(inode)->generation; 1820 BTRFS_I(inode)->index_cnt = (u64)-1; 1821 1822 mutex_unlock(&delayed_node->mutex); 1823 btrfs_release_delayed_node(delayed_node); 1824 return 0; 1825 } 1826 1827 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans, 1828 struct btrfs_root *root, struct inode *inode) 1829 { 1830 struct btrfs_delayed_node *delayed_node; 1831 int ret = 0; 1832 1833 delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode)); 1834 if (IS_ERR(delayed_node)) 1835 return PTR_ERR(delayed_node); 1836 1837 mutex_lock(&delayed_node->mutex); 1838 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1839 fill_stack_inode_item(trans, &delayed_node->inode_item, inode); 1840 goto release_node; 1841 } 1842 1843 ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode), 1844 delayed_node); 1845 if (ret) 1846 goto release_node; 1847 1848 fill_stack_inode_item(trans, &delayed_node->inode_item, inode); 1849 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); 1850 delayed_node->count++; 1851 atomic_inc(&root->fs_info->delayed_root->items); 1852 release_node: 1853 mutex_unlock(&delayed_node->mutex); 1854 btrfs_release_delayed_node(delayed_node); 1855 return ret; 1856 } 1857 1858 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode) 1859 { 1860 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); 1861 struct btrfs_delayed_node *delayed_node; 1862 1863 /* 1864 * we don't do delayed inode updates during log recovery because it 1865 * leads to enospc problems. This means we also can't do 1866 * delayed inode refs 1867 */ 1868 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 1869 return -EAGAIN; 1870 1871 delayed_node = btrfs_get_or_create_delayed_node(inode); 1872 if (IS_ERR(delayed_node)) 1873 return PTR_ERR(delayed_node); 1874 1875 /* 1876 * We don't reserve space for inode ref deletion is because: 1877 * - We ONLY do async inode ref deletion for the inode who has only 1878 * one link(i_nlink == 1), it means there is only one inode ref. 1879 * And in most case, the inode ref and the inode item are in the 1880 * same leaf, and we will deal with them at the same time. 1881 * Since we are sure we will reserve the space for the inode item, 1882 * it is unnecessary to reserve space for inode ref deletion. 1883 * - If the inode ref and the inode item are not in the same leaf, 1884 * We also needn't worry about enospc problem, because we reserve 1885 * much more space for the inode update than it needs. 1886 * - At the worst, we can steal some space from the global reservation. 1887 * It is very rare. 1888 */ 1889 mutex_lock(&delayed_node->mutex); 1890 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) 1891 goto release_node; 1892 1893 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); 1894 delayed_node->count++; 1895 atomic_inc(&fs_info->delayed_root->items); 1896 release_node: 1897 mutex_unlock(&delayed_node->mutex); 1898 btrfs_release_delayed_node(delayed_node); 1899 return 0; 1900 } 1901 1902 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node) 1903 { 1904 struct btrfs_root *root = delayed_node->root; 1905 struct btrfs_fs_info *fs_info = root->fs_info; 1906 struct btrfs_delayed_item *curr_item, *prev_item; 1907 1908 mutex_lock(&delayed_node->mutex); 1909 curr_item = __btrfs_first_delayed_insertion_item(delayed_node); 1910 while (curr_item) { 1911 btrfs_delayed_item_release_metadata(fs_info, curr_item); 1912 prev_item = curr_item; 1913 curr_item = __btrfs_next_delayed_item(prev_item); 1914 btrfs_release_delayed_item(prev_item); 1915 } 1916 1917 curr_item = __btrfs_first_delayed_deletion_item(delayed_node); 1918 while (curr_item) { 1919 btrfs_delayed_item_release_metadata(fs_info, curr_item); 1920 prev_item = curr_item; 1921 curr_item = __btrfs_next_delayed_item(prev_item); 1922 btrfs_release_delayed_item(prev_item); 1923 } 1924 1925 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) 1926 btrfs_release_delayed_iref(delayed_node); 1927 1928 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1929 btrfs_delayed_inode_release_metadata(fs_info, delayed_node); 1930 btrfs_release_delayed_inode(delayed_node); 1931 } 1932 mutex_unlock(&delayed_node->mutex); 1933 } 1934 1935 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode) 1936 { 1937 struct btrfs_delayed_node *delayed_node; 1938 1939 delayed_node = btrfs_get_delayed_node(inode); 1940 if (!delayed_node) 1941 return; 1942 1943 __btrfs_kill_delayed_node(delayed_node); 1944 btrfs_release_delayed_node(delayed_node); 1945 } 1946 1947 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root) 1948 { 1949 u64 inode_id = 0; 1950 struct btrfs_delayed_node *delayed_nodes[8]; 1951 int i, n; 1952 1953 while (1) { 1954 spin_lock(&root->inode_lock); 1955 n = radix_tree_gang_lookup(&root->delayed_nodes_tree, 1956 (void **)delayed_nodes, inode_id, 1957 ARRAY_SIZE(delayed_nodes)); 1958 if (!n) { 1959 spin_unlock(&root->inode_lock); 1960 break; 1961 } 1962 1963 inode_id = delayed_nodes[n - 1]->inode_id + 1; 1964 1965 for (i = 0; i < n; i++) 1966 atomic_inc(&delayed_nodes[i]->refs); 1967 spin_unlock(&root->inode_lock); 1968 1969 for (i = 0; i < n; i++) { 1970 __btrfs_kill_delayed_node(delayed_nodes[i]); 1971 btrfs_release_delayed_node(delayed_nodes[i]); 1972 } 1973 } 1974 } 1975 1976 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info) 1977 { 1978 struct btrfs_delayed_node *curr_node, *prev_node; 1979 1980 curr_node = btrfs_first_delayed_node(fs_info->delayed_root); 1981 while (curr_node) { 1982 __btrfs_kill_delayed_node(curr_node); 1983 1984 prev_node = curr_node; 1985 curr_node = btrfs_next_delayed_node(curr_node); 1986 btrfs_release_delayed_node(prev_node); 1987 } 1988 } 1989 1990