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(struct inode *inode) 76 { 77 struct btrfs_inode *btrfs_inode = BTRFS_I(inode); 78 struct btrfs_root *root = btrfs_inode->root; 79 u64 ino = btrfs_ino(inode); 80 struct btrfs_delayed_node *node; 81 82 node = ACCESS_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 inode *inode) 111 { 112 struct btrfs_delayed_node *node; 113 struct btrfs_inode *btrfs_inode = BTRFS_I(inode); 114 struct btrfs_root *root = btrfs_inode->root; 115 u64 ino = btrfs_ino(inode); 116 int ret; 117 118 again: 119 node = btrfs_get_delayed_node(inode); 120 if (node) 121 return node; 122 123 node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS); 124 if (!node) 125 return ERR_PTR(-ENOMEM); 126 btrfs_init_delayed_node(node, root, ino); 127 128 /* cached in the btrfs inode and can be accessed */ 129 atomic_add(2, &node->refs); 130 131 ret = radix_tree_preload(GFP_NOFS); 132 if (ret) { 133 kmem_cache_free(delayed_node_cache, node); 134 return ERR_PTR(ret); 135 } 136 137 spin_lock(&root->inode_lock); 138 ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node); 139 if (ret == -EEXIST) { 140 spin_unlock(&root->inode_lock); 141 kmem_cache_free(delayed_node_cache, node); 142 radix_tree_preload_end(); 143 goto again; 144 } 145 btrfs_inode->delayed_node = node; 146 spin_unlock(&root->inode_lock); 147 radix_tree_preload_end(); 148 149 return node; 150 } 151 152 /* 153 * Call it when holding delayed_node->mutex 154 * 155 * If mod = 1, add this node into the prepared list. 156 */ 157 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root, 158 struct btrfs_delayed_node *node, 159 int mod) 160 { 161 spin_lock(&root->lock); 162 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 163 if (!list_empty(&node->p_list)) 164 list_move_tail(&node->p_list, &root->prepare_list); 165 else if (mod) 166 list_add_tail(&node->p_list, &root->prepare_list); 167 } else { 168 list_add_tail(&node->n_list, &root->node_list); 169 list_add_tail(&node->p_list, &root->prepare_list); 170 atomic_inc(&node->refs); /* inserted into list */ 171 root->nodes++; 172 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); 173 } 174 spin_unlock(&root->lock); 175 } 176 177 /* Call it when holding delayed_node->mutex */ 178 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root, 179 struct btrfs_delayed_node *node) 180 { 181 spin_lock(&root->lock); 182 if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 183 root->nodes--; 184 atomic_dec(&node->refs); /* not in the list */ 185 list_del_init(&node->n_list); 186 if (!list_empty(&node->p_list)) 187 list_del_init(&node->p_list); 188 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags); 189 } 190 spin_unlock(&root->lock); 191 } 192 193 static struct btrfs_delayed_node *btrfs_first_delayed_node( 194 struct btrfs_delayed_root *delayed_root) 195 { 196 struct list_head *p; 197 struct btrfs_delayed_node *node = NULL; 198 199 spin_lock(&delayed_root->lock); 200 if (list_empty(&delayed_root->node_list)) 201 goto out; 202 203 p = delayed_root->node_list.next; 204 node = list_entry(p, struct btrfs_delayed_node, n_list); 205 atomic_inc(&node->refs); 206 out: 207 spin_unlock(&delayed_root->lock); 208 209 return node; 210 } 211 212 static struct btrfs_delayed_node *btrfs_next_delayed_node( 213 struct btrfs_delayed_node *node) 214 { 215 struct btrfs_delayed_root *delayed_root; 216 struct list_head *p; 217 struct btrfs_delayed_node *next = NULL; 218 219 delayed_root = node->root->fs_info->delayed_root; 220 spin_lock(&delayed_root->lock); 221 if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) { 222 /* not in the list */ 223 if (list_empty(&delayed_root->node_list)) 224 goto out; 225 p = delayed_root->node_list.next; 226 } else if (list_is_last(&node->n_list, &delayed_root->node_list)) 227 goto out; 228 else 229 p = node->n_list.next; 230 231 next = list_entry(p, struct btrfs_delayed_node, n_list); 232 atomic_inc(&next->refs); 233 out: 234 spin_unlock(&delayed_root->lock); 235 236 return next; 237 } 238 239 static void __btrfs_release_delayed_node( 240 struct btrfs_delayed_node *delayed_node, 241 int mod) 242 { 243 struct btrfs_delayed_root *delayed_root; 244 245 if (!delayed_node) 246 return; 247 248 delayed_root = delayed_node->root->fs_info->delayed_root; 249 250 mutex_lock(&delayed_node->mutex); 251 if (delayed_node->count) 252 btrfs_queue_delayed_node(delayed_root, delayed_node, mod); 253 else 254 btrfs_dequeue_delayed_node(delayed_root, delayed_node); 255 mutex_unlock(&delayed_node->mutex); 256 257 if (atomic_dec_and_test(&delayed_node->refs)) { 258 bool free = false; 259 struct btrfs_root *root = delayed_node->root; 260 spin_lock(&root->inode_lock); 261 if (atomic_read(&delayed_node->refs) == 0) { 262 radix_tree_delete(&root->delayed_nodes_tree, 263 delayed_node->inode_id); 264 free = true; 265 } 266 spin_unlock(&root->inode_lock); 267 if (free) 268 kmem_cache_free(delayed_node_cache, delayed_node); 269 } 270 } 271 272 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node) 273 { 274 __btrfs_release_delayed_node(node, 0); 275 } 276 277 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node( 278 struct btrfs_delayed_root *delayed_root) 279 { 280 struct list_head *p; 281 struct btrfs_delayed_node *node = NULL; 282 283 spin_lock(&delayed_root->lock); 284 if (list_empty(&delayed_root->prepare_list)) 285 goto out; 286 287 p = delayed_root->prepare_list.next; 288 list_del_init(p); 289 node = list_entry(p, struct btrfs_delayed_node, p_list); 290 atomic_inc(&node->refs); 291 out: 292 spin_unlock(&delayed_root->lock); 293 294 return node; 295 } 296 297 static inline void btrfs_release_prepared_delayed_node( 298 struct btrfs_delayed_node *node) 299 { 300 __btrfs_release_delayed_node(node, 1); 301 } 302 303 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len) 304 { 305 struct btrfs_delayed_item *item; 306 item = kmalloc(sizeof(*item) + data_len, GFP_NOFS); 307 if (item) { 308 item->data_len = data_len; 309 item->ins_or_del = 0; 310 item->bytes_reserved = 0; 311 item->delayed_node = NULL; 312 atomic_set(&item->refs, 1); 313 } 314 return item; 315 } 316 317 /* 318 * __btrfs_lookup_delayed_item - look up the delayed item by key 319 * @delayed_node: pointer to the delayed node 320 * @key: the key to look up 321 * @prev: used to store the prev item if the right item isn't found 322 * @next: used to store the next item if the right item isn't found 323 * 324 * Note: if we don't find the right item, we will return the prev item and 325 * the next item. 326 */ 327 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item( 328 struct rb_root *root, 329 struct btrfs_key *key, 330 struct btrfs_delayed_item **prev, 331 struct btrfs_delayed_item **next) 332 { 333 struct rb_node *node, *prev_node = NULL; 334 struct btrfs_delayed_item *delayed_item = NULL; 335 int ret = 0; 336 337 node = root->rb_node; 338 339 while (node) { 340 delayed_item = rb_entry(node, struct btrfs_delayed_item, 341 rb_node); 342 prev_node = node; 343 ret = btrfs_comp_cpu_keys(&delayed_item->key, key); 344 if (ret < 0) 345 node = node->rb_right; 346 else if (ret > 0) 347 node = node->rb_left; 348 else 349 return delayed_item; 350 } 351 352 if (prev) { 353 if (!prev_node) 354 *prev = NULL; 355 else if (ret < 0) 356 *prev = delayed_item; 357 else if ((node = rb_prev(prev_node)) != NULL) { 358 *prev = rb_entry(node, struct btrfs_delayed_item, 359 rb_node); 360 } else 361 *prev = NULL; 362 } 363 364 if (next) { 365 if (!prev_node) 366 *next = NULL; 367 else if (ret > 0) 368 *next = delayed_item; 369 else if ((node = rb_next(prev_node)) != NULL) { 370 *next = rb_entry(node, struct btrfs_delayed_item, 371 rb_node); 372 } else 373 *next = NULL; 374 } 375 return NULL; 376 } 377 378 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item( 379 struct btrfs_delayed_node *delayed_node, 380 struct btrfs_key *key) 381 { 382 return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, 383 NULL, NULL); 384 } 385 386 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node, 387 struct btrfs_delayed_item *ins, 388 int action) 389 { 390 struct rb_node **p, *node; 391 struct rb_node *parent_node = NULL; 392 struct rb_root *root; 393 struct btrfs_delayed_item *item; 394 int cmp; 395 396 if (action == BTRFS_DELAYED_INSERTION_ITEM) 397 root = &delayed_node->ins_root; 398 else if (action == BTRFS_DELAYED_DELETION_ITEM) 399 root = &delayed_node->del_root; 400 else 401 BUG(); 402 p = &root->rb_node; 403 node = &ins->rb_node; 404 405 while (*p) { 406 parent_node = *p; 407 item = rb_entry(parent_node, struct btrfs_delayed_item, 408 rb_node); 409 410 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key); 411 if (cmp < 0) 412 p = &(*p)->rb_right; 413 else if (cmp > 0) 414 p = &(*p)->rb_left; 415 else 416 return -EEXIST; 417 } 418 419 rb_link_node(node, parent_node, p); 420 rb_insert_color(node, root); 421 ins->delayed_node = delayed_node; 422 ins->ins_or_del = action; 423 424 if (ins->key.type == BTRFS_DIR_INDEX_KEY && 425 action == BTRFS_DELAYED_INSERTION_ITEM && 426 ins->key.offset >= delayed_node->index_cnt) 427 delayed_node->index_cnt = ins->key.offset + 1; 428 429 delayed_node->count++; 430 atomic_inc(&delayed_node->root->fs_info->delayed_root->items); 431 return 0; 432 } 433 434 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node, 435 struct btrfs_delayed_item *item) 436 { 437 return __btrfs_add_delayed_item(node, item, 438 BTRFS_DELAYED_INSERTION_ITEM); 439 } 440 441 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node, 442 struct btrfs_delayed_item *item) 443 { 444 return __btrfs_add_delayed_item(node, item, 445 BTRFS_DELAYED_DELETION_ITEM); 446 } 447 448 static void finish_one_item(struct btrfs_delayed_root *delayed_root) 449 { 450 int seq = atomic_inc_return(&delayed_root->items_seq); 451 452 /* 453 * atomic_dec_return implies a barrier for waitqueue_active 454 */ 455 if ((atomic_dec_return(&delayed_root->items) < 456 BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) && 457 waitqueue_active(&delayed_root->wait)) 458 wake_up(&delayed_root->wait); 459 } 460 461 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item) 462 { 463 struct rb_root *root; 464 struct btrfs_delayed_root *delayed_root; 465 466 delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root; 467 468 BUG_ON(!delayed_root); 469 BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM && 470 delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM); 471 472 if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM) 473 root = &delayed_item->delayed_node->ins_root; 474 else 475 root = &delayed_item->delayed_node->del_root; 476 477 rb_erase(&delayed_item->rb_node, root); 478 delayed_item->delayed_node->count--; 479 480 finish_one_item(delayed_root); 481 } 482 483 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item) 484 { 485 if (item) { 486 __btrfs_remove_delayed_item(item); 487 if (atomic_dec_and_test(&item->refs)) 488 kfree(item); 489 } 490 } 491 492 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item( 493 struct btrfs_delayed_node *delayed_node) 494 { 495 struct rb_node *p; 496 struct btrfs_delayed_item *item = NULL; 497 498 p = rb_first(&delayed_node->ins_root); 499 if (p) 500 item = rb_entry(p, struct btrfs_delayed_item, rb_node); 501 502 return item; 503 } 504 505 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item( 506 struct btrfs_delayed_node *delayed_node) 507 { 508 struct rb_node *p; 509 struct btrfs_delayed_item *item = NULL; 510 511 p = rb_first(&delayed_node->del_root); 512 if (p) 513 item = rb_entry(p, struct btrfs_delayed_item, rb_node); 514 515 return item; 516 } 517 518 static struct btrfs_delayed_item *__btrfs_next_delayed_item( 519 struct btrfs_delayed_item *item) 520 { 521 struct rb_node *p; 522 struct btrfs_delayed_item *next = NULL; 523 524 p = rb_next(&item->rb_node); 525 if (p) 526 next = rb_entry(p, struct btrfs_delayed_item, rb_node); 527 528 return next; 529 } 530 531 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans, 532 struct btrfs_fs_info *fs_info, 533 struct btrfs_delayed_item *item) 534 { 535 struct btrfs_block_rsv *src_rsv; 536 struct btrfs_block_rsv *dst_rsv; 537 u64 num_bytes; 538 int ret; 539 540 if (!trans->bytes_reserved) 541 return 0; 542 543 src_rsv = trans->block_rsv; 544 dst_rsv = &fs_info->delayed_block_rsv; 545 546 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); 547 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); 548 if (!ret) { 549 trace_btrfs_space_reservation(fs_info, "delayed_item", 550 item->key.objectid, 551 num_bytes, 1); 552 item->bytes_reserved = num_bytes; 553 } 554 555 return ret; 556 } 557 558 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info, 559 struct btrfs_delayed_item *item) 560 { 561 struct btrfs_block_rsv *rsv; 562 563 if (!item->bytes_reserved) 564 return; 565 566 rsv = &fs_info->delayed_block_rsv; 567 trace_btrfs_space_reservation(fs_info, "delayed_item", 568 item->key.objectid, item->bytes_reserved, 569 0); 570 btrfs_block_rsv_release(fs_info, rsv, 571 item->bytes_reserved); 572 } 573 574 static int btrfs_delayed_inode_reserve_metadata( 575 struct btrfs_trans_handle *trans, 576 struct btrfs_root *root, 577 struct inode *inode, 578 struct btrfs_delayed_node *node) 579 { 580 struct btrfs_fs_info *fs_info = root->fs_info; 581 struct btrfs_block_rsv *src_rsv; 582 struct btrfs_block_rsv *dst_rsv; 583 u64 num_bytes; 584 int ret; 585 bool release = false; 586 587 src_rsv = trans->block_rsv; 588 dst_rsv = &fs_info->delayed_block_rsv; 589 590 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); 591 592 /* 593 * If our block_rsv is the delalloc block reserve then check and see if 594 * we have our extra reservation for updating the inode. If not fall 595 * through and try to reserve space quickly. 596 * 597 * We used to try and steal from the delalloc block rsv or the global 598 * reserve, but we'd steal a full reservation, which isn't kind. We are 599 * here through delalloc which means we've likely just cowed down close 600 * to the leaf that contains the inode, so we would steal less just 601 * doing the fallback inode update, so if we do end up having to steal 602 * from the global block rsv we hopefully only steal one or two blocks 603 * worth which is less likely to hurt us. 604 */ 605 if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) { 606 spin_lock(&BTRFS_I(inode)->lock); 607 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, 608 &BTRFS_I(inode)->runtime_flags)) 609 release = true; 610 else 611 src_rsv = NULL; 612 spin_unlock(&BTRFS_I(inode)->lock); 613 } 614 615 /* 616 * btrfs_dirty_inode will update the inode under btrfs_join_transaction 617 * which doesn't reserve space for speed. This is a problem since we 618 * still need to reserve space for this update, so try to reserve the 619 * space. 620 * 621 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since 622 * we're accounted for. 623 */ 624 if (!src_rsv || (!trans->bytes_reserved && 625 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) { 626 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes, 627 BTRFS_RESERVE_NO_FLUSH); 628 /* 629 * Since we're under a transaction reserve_metadata_bytes could 630 * try to commit the transaction which will make it return 631 * EAGAIN to make us stop the transaction we have, so return 632 * ENOSPC instead so that btrfs_dirty_inode knows what to do. 633 */ 634 if (ret == -EAGAIN) 635 ret = -ENOSPC; 636 if (!ret) { 637 node->bytes_reserved = num_bytes; 638 trace_btrfs_space_reservation(fs_info, 639 "delayed_inode", 640 btrfs_ino(inode), 641 num_bytes, 1); 642 } 643 return ret; 644 } 645 646 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); 647 648 /* 649 * Migrate only takes a reservation, it doesn't touch the size of the 650 * block_rsv. This is to simplify people who don't normally have things 651 * migrated from their block rsv. If they go to release their 652 * reservation, that will decrease the size as well, so if migrate 653 * reduced size we'd end up with a negative size. But for the 654 * delalloc_meta_reserved stuff we will only know to drop 1 reservation, 655 * but we could in fact do this reserve/migrate dance several times 656 * between the time we did the original reservation and we'd clean it 657 * up. So to take care of this, release the space for the meta 658 * reservation here. I think it may be time for a documentation page on 659 * how block rsvs. work. 660 */ 661 if (!ret) { 662 trace_btrfs_space_reservation(fs_info, "delayed_inode", 663 btrfs_ino(inode), num_bytes, 1); 664 node->bytes_reserved = num_bytes; 665 } 666 667 if (release) { 668 trace_btrfs_space_reservation(fs_info, "delalloc", 669 btrfs_ino(inode), num_bytes, 0); 670 btrfs_block_rsv_release(fs_info, src_rsv, num_bytes); 671 } 672 673 return ret; 674 } 675 676 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info, 677 struct btrfs_delayed_node *node) 678 { 679 struct btrfs_block_rsv *rsv; 680 681 if (!node->bytes_reserved) 682 return; 683 684 rsv = &fs_info->delayed_block_rsv; 685 trace_btrfs_space_reservation(fs_info, "delayed_inode", 686 node->inode_id, node->bytes_reserved, 0); 687 btrfs_block_rsv_release(fs_info, rsv, 688 node->bytes_reserved); 689 node->bytes_reserved = 0; 690 } 691 692 /* 693 * This helper will insert some continuous items into the same leaf according 694 * to the free space of the leaf. 695 */ 696 static int btrfs_batch_insert_items(struct btrfs_root *root, 697 struct btrfs_path *path, 698 struct btrfs_delayed_item *item) 699 { 700 struct btrfs_fs_info *fs_info = root->fs_info; 701 struct btrfs_delayed_item *curr, *next; 702 int free_space; 703 int total_data_size = 0, total_size = 0; 704 struct extent_buffer *leaf; 705 char *data_ptr; 706 struct btrfs_key *keys; 707 u32 *data_size; 708 struct list_head head; 709 int slot; 710 int nitems; 711 int i; 712 int ret = 0; 713 714 BUG_ON(!path->nodes[0]); 715 716 leaf = path->nodes[0]; 717 free_space = btrfs_leaf_free_space(fs_info, leaf); 718 INIT_LIST_HEAD(&head); 719 720 next = item; 721 nitems = 0; 722 723 /* 724 * count the number of the continuous items that we can insert in batch 725 */ 726 while (total_size + next->data_len + sizeof(struct btrfs_item) <= 727 free_space) { 728 total_data_size += next->data_len; 729 total_size += next->data_len + sizeof(struct btrfs_item); 730 list_add_tail(&next->tree_list, &head); 731 nitems++; 732 733 curr = next; 734 next = __btrfs_next_delayed_item(curr); 735 if (!next) 736 break; 737 738 if (!btrfs_is_continuous_delayed_item(curr, next)) 739 break; 740 } 741 742 if (!nitems) { 743 ret = 0; 744 goto out; 745 } 746 747 /* 748 * we need allocate some memory space, but it might cause the task 749 * to sleep, so we set all locked nodes in the path to blocking locks 750 * first. 751 */ 752 btrfs_set_path_blocking(path); 753 754 keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS); 755 if (!keys) { 756 ret = -ENOMEM; 757 goto out; 758 } 759 760 data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS); 761 if (!data_size) { 762 ret = -ENOMEM; 763 goto error; 764 } 765 766 /* get keys of all the delayed items */ 767 i = 0; 768 list_for_each_entry(next, &head, tree_list) { 769 keys[i] = next->key; 770 data_size[i] = next->data_len; 771 i++; 772 } 773 774 /* reset all the locked nodes in the patch to spinning locks. */ 775 btrfs_clear_path_blocking(path, NULL, 0); 776 777 /* insert the keys of the items */ 778 setup_items_for_insert(root, path, keys, data_size, 779 total_data_size, total_size, nitems); 780 781 /* insert the dir index items */ 782 slot = path->slots[0]; 783 list_for_each_entry_safe(curr, next, &head, tree_list) { 784 data_ptr = btrfs_item_ptr(leaf, slot, char); 785 write_extent_buffer(leaf, &curr->data, 786 (unsigned long)data_ptr, 787 curr->data_len); 788 slot++; 789 790 btrfs_delayed_item_release_metadata(fs_info, curr); 791 792 list_del(&curr->tree_list); 793 btrfs_release_delayed_item(curr); 794 } 795 796 error: 797 kfree(data_size); 798 kfree(keys); 799 out: 800 return ret; 801 } 802 803 /* 804 * This helper can just do simple insertion that needn't extend item for new 805 * data, such as directory name index insertion, inode insertion. 806 */ 807 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans, 808 struct btrfs_root *root, 809 struct btrfs_path *path, 810 struct btrfs_delayed_item *delayed_item) 811 { 812 struct btrfs_fs_info *fs_info = root->fs_info; 813 struct extent_buffer *leaf; 814 char *ptr; 815 int ret; 816 817 ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key, 818 delayed_item->data_len); 819 if (ret < 0 && ret != -EEXIST) 820 return ret; 821 822 leaf = path->nodes[0]; 823 824 ptr = btrfs_item_ptr(leaf, path->slots[0], char); 825 826 write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr, 827 delayed_item->data_len); 828 btrfs_mark_buffer_dirty(leaf); 829 830 btrfs_delayed_item_release_metadata(fs_info, delayed_item); 831 return 0; 832 } 833 834 /* 835 * we insert an item first, then if there are some continuous items, we try 836 * to insert those items into the same leaf. 837 */ 838 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans, 839 struct btrfs_path *path, 840 struct btrfs_root *root, 841 struct btrfs_delayed_node *node) 842 { 843 struct btrfs_delayed_item *curr, *prev; 844 int ret = 0; 845 846 do_again: 847 mutex_lock(&node->mutex); 848 curr = __btrfs_first_delayed_insertion_item(node); 849 if (!curr) 850 goto insert_end; 851 852 ret = btrfs_insert_delayed_item(trans, root, path, curr); 853 if (ret < 0) { 854 btrfs_release_path(path); 855 goto insert_end; 856 } 857 858 prev = curr; 859 curr = __btrfs_next_delayed_item(prev); 860 if (curr && btrfs_is_continuous_delayed_item(prev, curr)) { 861 /* insert the continuous items into the same leaf */ 862 path->slots[0]++; 863 btrfs_batch_insert_items(root, path, curr); 864 } 865 btrfs_release_delayed_item(prev); 866 btrfs_mark_buffer_dirty(path->nodes[0]); 867 868 btrfs_release_path(path); 869 mutex_unlock(&node->mutex); 870 goto do_again; 871 872 insert_end: 873 mutex_unlock(&node->mutex); 874 return ret; 875 } 876 877 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans, 878 struct btrfs_root *root, 879 struct btrfs_path *path, 880 struct btrfs_delayed_item *item) 881 { 882 struct btrfs_fs_info *fs_info = root->fs_info; 883 struct btrfs_delayed_item *curr, *next; 884 struct extent_buffer *leaf; 885 struct btrfs_key key; 886 struct list_head head; 887 int nitems, i, last_item; 888 int ret = 0; 889 890 BUG_ON(!path->nodes[0]); 891 892 leaf = path->nodes[0]; 893 894 i = path->slots[0]; 895 last_item = btrfs_header_nritems(leaf) - 1; 896 if (i > last_item) 897 return -ENOENT; /* FIXME: Is errno suitable? */ 898 899 next = item; 900 INIT_LIST_HEAD(&head); 901 btrfs_item_key_to_cpu(leaf, &key, i); 902 nitems = 0; 903 /* 904 * count the number of the dir index items that we can delete in batch 905 */ 906 while (btrfs_comp_cpu_keys(&next->key, &key) == 0) { 907 list_add_tail(&next->tree_list, &head); 908 nitems++; 909 910 curr = next; 911 next = __btrfs_next_delayed_item(curr); 912 if (!next) 913 break; 914 915 if (!btrfs_is_continuous_delayed_item(curr, next)) 916 break; 917 918 i++; 919 if (i > last_item) 920 break; 921 btrfs_item_key_to_cpu(leaf, &key, i); 922 } 923 924 if (!nitems) 925 return 0; 926 927 ret = btrfs_del_items(trans, root, path, path->slots[0], nitems); 928 if (ret) 929 goto out; 930 931 list_for_each_entry_safe(curr, next, &head, tree_list) { 932 btrfs_delayed_item_release_metadata(fs_info, curr); 933 list_del(&curr->tree_list); 934 btrfs_release_delayed_item(curr); 935 } 936 937 out: 938 return ret; 939 } 940 941 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans, 942 struct btrfs_path *path, 943 struct btrfs_root *root, 944 struct btrfs_delayed_node *node) 945 { 946 struct btrfs_delayed_item *curr, *prev; 947 int ret = 0; 948 949 do_again: 950 mutex_lock(&node->mutex); 951 curr = __btrfs_first_delayed_deletion_item(node); 952 if (!curr) 953 goto delete_fail; 954 955 ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1); 956 if (ret < 0) 957 goto delete_fail; 958 else if (ret > 0) { 959 /* 960 * can't find the item which the node points to, so this node 961 * is invalid, just drop it. 962 */ 963 prev = curr; 964 curr = __btrfs_next_delayed_item(prev); 965 btrfs_release_delayed_item(prev); 966 ret = 0; 967 btrfs_release_path(path); 968 if (curr) { 969 mutex_unlock(&node->mutex); 970 goto do_again; 971 } else 972 goto delete_fail; 973 } 974 975 btrfs_batch_delete_items(trans, root, path, curr); 976 btrfs_release_path(path); 977 mutex_unlock(&node->mutex); 978 goto do_again; 979 980 delete_fail: 981 btrfs_release_path(path); 982 mutex_unlock(&node->mutex); 983 return ret; 984 } 985 986 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node) 987 { 988 struct btrfs_delayed_root *delayed_root; 989 990 if (delayed_node && 991 test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 992 BUG_ON(!delayed_node->root); 993 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); 994 delayed_node->count--; 995 996 delayed_root = delayed_node->root->fs_info->delayed_root; 997 finish_one_item(delayed_root); 998 } 999 } 1000 1001 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node) 1002 { 1003 struct btrfs_delayed_root *delayed_root; 1004 1005 ASSERT(delayed_node->root); 1006 clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); 1007 delayed_node->count--; 1008 1009 delayed_root = delayed_node->root->fs_info->delayed_root; 1010 finish_one_item(delayed_root); 1011 } 1012 1013 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, 1014 struct btrfs_root *root, 1015 struct btrfs_path *path, 1016 struct btrfs_delayed_node *node) 1017 { 1018 struct btrfs_fs_info *fs_info = root->fs_info; 1019 struct btrfs_key key; 1020 struct btrfs_inode_item *inode_item; 1021 struct extent_buffer *leaf; 1022 int mod; 1023 int ret; 1024 1025 key.objectid = node->inode_id; 1026 key.type = BTRFS_INODE_ITEM_KEY; 1027 key.offset = 0; 1028 1029 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) 1030 mod = -1; 1031 else 1032 mod = 1; 1033 1034 ret = btrfs_lookup_inode(trans, root, path, &key, mod); 1035 if (ret > 0) { 1036 btrfs_release_path(path); 1037 return -ENOENT; 1038 } else if (ret < 0) { 1039 return ret; 1040 } 1041 1042 leaf = path->nodes[0]; 1043 inode_item = btrfs_item_ptr(leaf, path->slots[0], 1044 struct btrfs_inode_item); 1045 write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item, 1046 sizeof(struct btrfs_inode_item)); 1047 btrfs_mark_buffer_dirty(leaf); 1048 1049 if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags)) 1050 goto no_iref; 1051 1052 path->slots[0]++; 1053 if (path->slots[0] >= btrfs_header_nritems(leaf)) 1054 goto search; 1055 again: 1056 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 1057 if (key.objectid != node->inode_id) 1058 goto out; 1059 1060 if (key.type != BTRFS_INODE_REF_KEY && 1061 key.type != BTRFS_INODE_EXTREF_KEY) 1062 goto out; 1063 1064 /* 1065 * Delayed iref deletion is for the inode who has only one link, 1066 * so there is only one iref. The case that several irefs are 1067 * in the same item doesn't exist. 1068 */ 1069 btrfs_del_item(trans, root, path); 1070 out: 1071 btrfs_release_delayed_iref(node); 1072 no_iref: 1073 btrfs_release_path(path); 1074 err_out: 1075 btrfs_delayed_inode_release_metadata(fs_info, node); 1076 btrfs_release_delayed_inode(node); 1077 1078 return ret; 1079 1080 search: 1081 btrfs_release_path(path); 1082 1083 key.type = BTRFS_INODE_EXTREF_KEY; 1084 key.offset = -1; 1085 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1086 if (ret < 0) 1087 goto err_out; 1088 ASSERT(ret); 1089 1090 ret = 0; 1091 leaf = path->nodes[0]; 1092 path->slots[0]--; 1093 goto again; 1094 } 1095 1096 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, 1097 struct btrfs_root *root, 1098 struct btrfs_path *path, 1099 struct btrfs_delayed_node *node) 1100 { 1101 int ret; 1102 1103 mutex_lock(&node->mutex); 1104 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) { 1105 mutex_unlock(&node->mutex); 1106 return 0; 1107 } 1108 1109 ret = __btrfs_update_delayed_inode(trans, root, path, node); 1110 mutex_unlock(&node->mutex); 1111 return ret; 1112 } 1113 1114 static inline int 1115 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, 1116 struct btrfs_path *path, 1117 struct btrfs_delayed_node *node) 1118 { 1119 int ret; 1120 1121 ret = btrfs_insert_delayed_items(trans, path, node->root, node); 1122 if (ret) 1123 return ret; 1124 1125 ret = btrfs_delete_delayed_items(trans, path, node->root, node); 1126 if (ret) 1127 return ret; 1128 1129 ret = btrfs_update_delayed_inode(trans, node->root, path, node); 1130 return ret; 1131 } 1132 1133 /* 1134 * Called when committing the transaction. 1135 * Returns 0 on success. 1136 * Returns < 0 on error and returns with an aborted transaction with any 1137 * outstanding delayed items cleaned up. 1138 */ 1139 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, 1140 struct btrfs_fs_info *fs_info, int nr) 1141 { 1142 struct btrfs_delayed_root *delayed_root; 1143 struct btrfs_delayed_node *curr_node, *prev_node; 1144 struct btrfs_path *path; 1145 struct btrfs_block_rsv *block_rsv; 1146 int ret = 0; 1147 bool count = (nr > 0); 1148 1149 if (trans->aborted) 1150 return -EIO; 1151 1152 path = btrfs_alloc_path(); 1153 if (!path) 1154 return -ENOMEM; 1155 path->leave_spinning = 1; 1156 1157 block_rsv = trans->block_rsv; 1158 trans->block_rsv = &fs_info->delayed_block_rsv; 1159 1160 delayed_root = fs_info->delayed_root; 1161 1162 curr_node = btrfs_first_delayed_node(delayed_root); 1163 while (curr_node && (!count || (count && nr--))) { 1164 ret = __btrfs_commit_inode_delayed_items(trans, path, 1165 curr_node); 1166 if (ret) { 1167 btrfs_release_delayed_node(curr_node); 1168 curr_node = NULL; 1169 btrfs_abort_transaction(trans, ret); 1170 break; 1171 } 1172 1173 prev_node = curr_node; 1174 curr_node = btrfs_next_delayed_node(curr_node); 1175 btrfs_release_delayed_node(prev_node); 1176 } 1177 1178 if (curr_node) 1179 btrfs_release_delayed_node(curr_node); 1180 btrfs_free_path(path); 1181 trans->block_rsv = block_rsv; 1182 1183 return ret; 1184 } 1185 1186 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans, 1187 struct btrfs_fs_info *fs_info) 1188 { 1189 return __btrfs_run_delayed_items(trans, fs_info, -1); 1190 } 1191 1192 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, 1193 struct btrfs_fs_info *fs_info, int nr) 1194 { 1195 return __btrfs_run_delayed_items(trans, fs_info, nr); 1196 } 1197 1198 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, 1199 struct inode *inode) 1200 { 1201 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1202 struct btrfs_path *path; 1203 struct btrfs_block_rsv *block_rsv; 1204 int ret; 1205 1206 if (!delayed_node) 1207 return 0; 1208 1209 mutex_lock(&delayed_node->mutex); 1210 if (!delayed_node->count) { 1211 mutex_unlock(&delayed_node->mutex); 1212 btrfs_release_delayed_node(delayed_node); 1213 return 0; 1214 } 1215 mutex_unlock(&delayed_node->mutex); 1216 1217 path = btrfs_alloc_path(); 1218 if (!path) { 1219 btrfs_release_delayed_node(delayed_node); 1220 return -ENOMEM; 1221 } 1222 path->leave_spinning = 1; 1223 1224 block_rsv = trans->block_rsv; 1225 trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv; 1226 1227 ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node); 1228 1229 btrfs_release_delayed_node(delayed_node); 1230 btrfs_free_path(path); 1231 trans->block_rsv = block_rsv; 1232 1233 return ret; 1234 } 1235 1236 int btrfs_commit_inode_delayed_inode(struct inode *inode) 1237 { 1238 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1239 struct btrfs_trans_handle *trans; 1240 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1241 struct btrfs_path *path; 1242 struct btrfs_block_rsv *block_rsv; 1243 int ret; 1244 1245 if (!delayed_node) 1246 return 0; 1247 1248 mutex_lock(&delayed_node->mutex); 1249 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1250 mutex_unlock(&delayed_node->mutex); 1251 btrfs_release_delayed_node(delayed_node); 1252 return 0; 1253 } 1254 mutex_unlock(&delayed_node->mutex); 1255 1256 trans = btrfs_join_transaction(delayed_node->root); 1257 if (IS_ERR(trans)) { 1258 ret = PTR_ERR(trans); 1259 goto out; 1260 } 1261 1262 path = btrfs_alloc_path(); 1263 if (!path) { 1264 ret = -ENOMEM; 1265 goto trans_out; 1266 } 1267 path->leave_spinning = 1; 1268 1269 block_rsv = trans->block_rsv; 1270 trans->block_rsv = &fs_info->delayed_block_rsv; 1271 1272 mutex_lock(&delayed_node->mutex); 1273 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) 1274 ret = __btrfs_update_delayed_inode(trans, delayed_node->root, 1275 path, delayed_node); 1276 else 1277 ret = 0; 1278 mutex_unlock(&delayed_node->mutex); 1279 1280 btrfs_free_path(path); 1281 trans->block_rsv = block_rsv; 1282 trans_out: 1283 btrfs_end_transaction(trans); 1284 btrfs_btree_balance_dirty(fs_info); 1285 out: 1286 btrfs_release_delayed_node(delayed_node); 1287 1288 return ret; 1289 } 1290 1291 void btrfs_remove_delayed_node(struct inode *inode) 1292 { 1293 struct btrfs_delayed_node *delayed_node; 1294 1295 delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node); 1296 if (!delayed_node) 1297 return; 1298 1299 BTRFS_I(inode)->delayed_node = NULL; 1300 btrfs_release_delayed_node(delayed_node); 1301 } 1302 1303 struct btrfs_async_delayed_work { 1304 struct btrfs_delayed_root *delayed_root; 1305 int nr; 1306 struct btrfs_work work; 1307 }; 1308 1309 static void btrfs_async_run_delayed_root(struct btrfs_work *work) 1310 { 1311 struct btrfs_async_delayed_work *async_work; 1312 struct btrfs_delayed_root *delayed_root; 1313 struct btrfs_trans_handle *trans; 1314 struct btrfs_path *path; 1315 struct btrfs_delayed_node *delayed_node = NULL; 1316 struct btrfs_root *root; 1317 struct btrfs_block_rsv *block_rsv; 1318 int total_done = 0; 1319 1320 async_work = container_of(work, struct btrfs_async_delayed_work, work); 1321 delayed_root = async_work->delayed_root; 1322 1323 path = btrfs_alloc_path(); 1324 if (!path) 1325 goto out; 1326 1327 again: 1328 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2) 1329 goto free_path; 1330 1331 delayed_node = btrfs_first_prepared_delayed_node(delayed_root); 1332 if (!delayed_node) 1333 goto free_path; 1334 1335 path->leave_spinning = 1; 1336 root = delayed_node->root; 1337 1338 trans = btrfs_join_transaction(root); 1339 if (IS_ERR(trans)) 1340 goto release_path; 1341 1342 block_rsv = trans->block_rsv; 1343 trans->block_rsv = &root->fs_info->delayed_block_rsv; 1344 1345 __btrfs_commit_inode_delayed_items(trans, path, delayed_node); 1346 1347 trans->block_rsv = block_rsv; 1348 btrfs_end_transaction(trans); 1349 btrfs_btree_balance_dirty_nodelay(root->fs_info); 1350 1351 release_path: 1352 btrfs_release_path(path); 1353 total_done++; 1354 1355 btrfs_release_prepared_delayed_node(delayed_node); 1356 if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) || 1357 total_done < async_work->nr) 1358 goto again; 1359 1360 free_path: 1361 btrfs_free_path(path); 1362 out: 1363 wake_up(&delayed_root->wait); 1364 kfree(async_work); 1365 } 1366 1367 1368 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root, 1369 struct btrfs_fs_info *fs_info, int nr) 1370 { 1371 struct btrfs_async_delayed_work *async_work; 1372 1373 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND || 1374 btrfs_workqueue_normal_congested(fs_info->delayed_workers)) 1375 return 0; 1376 1377 async_work = kmalloc(sizeof(*async_work), GFP_NOFS); 1378 if (!async_work) 1379 return -ENOMEM; 1380 1381 async_work->delayed_root = delayed_root; 1382 btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper, 1383 btrfs_async_run_delayed_root, NULL, NULL); 1384 async_work->nr = nr; 1385 1386 btrfs_queue_work(fs_info->delayed_workers, &async_work->work); 1387 return 0; 1388 } 1389 1390 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info) 1391 { 1392 WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root)); 1393 } 1394 1395 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq) 1396 { 1397 int val = atomic_read(&delayed_root->items_seq); 1398 1399 if (val < seq || val >= seq + BTRFS_DELAYED_BATCH) 1400 return 1; 1401 1402 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) 1403 return 1; 1404 1405 return 0; 1406 } 1407 1408 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info) 1409 { 1410 struct btrfs_delayed_root *delayed_root = fs_info->delayed_root; 1411 1412 if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) 1413 return; 1414 1415 if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) { 1416 int seq; 1417 int ret; 1418 1419 seq = atomic_read(&delayed_root->items_seq); 1420 1421 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0); 1422 if (ret) 1423 return; 1424 1425 wait_event_interruptible(delayed_root->wait, 1426 could_end_wait(delayed_root, seq)); 1427 return; 1428 } 1429 1430 btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH); 1431 } 1432 1433 /* Will return 0 or -ENOMEM */ 1434 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans, 1435 struct btrfs_fs_info *fs_info, 1436 const char *name, int name_len, 1437 struct inode *dir, 1438 struct btrfs_disk_key *disk_key, u8 type, 1439 u64 index) 1440 { 1441 struct btrfs_delayed_node *delayed_node; 1442 struct btrfs_delayed_item *delayed_item; 1443 struct btrfs_dir_item *dir_item; 1444 int ret; 1445 1446 delayed_node = btrfs_get_or_create_delayed_node(dir); 1447 if (IS_ERR(delayed_node)) 1448 return PTR_ERR(delayed_node); 1449 1450 delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len); 1451 if (!delayed_item) { 1452 ret = -ENOMEM; 1453 goto release_node; 1454 } 1455 1456 delayed_item->key.objectid = btrfs_ino(dir); 1457 delayed_item->key.type = BTRFS_DIR_INDEX_KEY; 1458 delayed_item->key.offset = index; 1459 1460 dir_item = (struct btrfs_dir_item *)delayed_item->data; 1461 dir_item->location = *disk_key; 1462 btrfs_set_stack_dir_transid(dir_item, trans->transid); 1463 btrfs_set_stack_dir_data_len(dir_item, 0); 1464 btrfs_set_stack_dir_name_len(dir_item, name_len); 1465 btrfs_set_stack_dir_type(dir_item, type); 1466 memcpy((char *)(dir_item + 1), name, name_len); 1467 1468 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item); 1469 /* 1470 * we have reserved enough space when we start a new transaction, 1471 * so reserving metadata failure is impossible 1472 */ 1473 BUG_ON(ret); 1474 1475 1476 mutex_lock(&delayed_node->mutex); 1477 ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item); 1478 if (unlikely(ret)) { 1479 btrfs_err(fs_info, 1480 "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)", 1481 name_len, name, delayed_node->root->objectid, 1482 delayed_node->inode_id, ret); 1483 BUG(); 1484 } 1485 mutex_unlock(&delayed_node->mutex); 1486 1487 release_node: 1488 btrfs_release_delayed_node(delayed_node); 1489 return ret; 1490 } 1491 1492 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info, 1493 struct btrfs_delayed_node *node, 1494 struct btrfs_key *key) 1495 { 1496 struct btrfs_delayed_item *item; 1497 1498 mutex_lock(&node->mutex); 1499 item = __btrfs_lookup_delayed_insertion_item(node, key); 1500 if (!item) { 1501 mutex_unlock(&node->mutex); 1502 return 1; 1503 } 1504 1505 btrfs_delayed_item_release_metadata(fs_info, item); 1506 btrfs_release_delayed_item(item); 1507 mutex_unlock(&node->mutex); 1508 return 0; 1509 } 1510 1511 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans, 1512 struct btrfs_fs_info *fs_info, 1513 struct inode *dir, u64 index) 1514 { 1515 struct btrfs_delayed_node *node; 1516 struct btrfs_delayed_item *item; 1517 struct btrfs_key item_key; 1518 int ret; 1519 1520 node = btrfs_get_or_create_delayed_node(dir); 1521 if (IS_ERR(node)) 1522 return PTR_ERR(node); 1523 1524 item_key.objectid = btrfs_ino(dir); 1525 item_key.type = BTRFS_DIR_INDEX_KEY; 1526 item_key.offset = index; 1527 1528 ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key); 1529 if (!ret) 1530 goto end; 1531 1532 item = btrfs_alloc_delayed_item(0); 1533 if (!item) { 1534 ret = -ENOMEM; 1535 goto end; 1536 } 1537 1538 item->key = item_key; 1539 1540 ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item); 1541 /* 1542 * we have reserved enough space when we start a new transaction, 1543 * so reserving metadata failure is impossible. 1544 */ 1545 BUG_ON(ret); 1546 1547 mutex_lock(&node->mutex); 1548 ret = __btrfs_add_delayed_deletion_item(node, item); 1549 if (unlikely(ret)) { 1550 btrfs_err(fs_info, 1551 "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)", 1552 index, node->root->objectid, node->inode_id, ret); 1553 BUG(); 1554 } 1555 mutex_unlock(&node->mutex); 1556 end: 1557 btrfs_release_delayed_node(node); 1558 return ret; 1559 } 1560 1561 int btrfs_inode_delayed_dir_index_count(struct inode *inode) 1562 { 1563 struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); 1564 1565 if (!delayed_node) 1566 return -ENOENT; 1567 1568 /* 1569 * Since we have held i_mutex of this directory, it is impossible that 1570 * a new directory index is added into the delayed node and index_cnt 1571 * is updated now. So we needn't lock the delayed node. 1572 */ 1573 if (!delayed_node->index_cnt) { 1574 btrfs_release_delayed_node(delayed_node); 1575 return -EINVAL; 1576 } 1577 1578 BTRFS_I(inode)->index_cnt = delayed_node->index_cnt; 1579 btrfs_release_delayed_node(delayed_node); 1580 return 0; 1581 } 1582 1583 bool btrfs_readdir_get_delayed_items(struct inode *inode, 1584 struct list_head *ins_list, 1585 struct list_head *del_list) 1586 { 1587 struct btrfs_delayed_node *delayed_node; 1588 struct btrfs_delayed_item *item; 1589 1590 delayed_node = btrfs_get_delayed_node(inode); 1591 if (!delayed_node) 1592 return false; 1593 1594 /* 1595 * We can only do one readdir with delayed items at a time because of 1596 * item->readdir_list. 1597 */ 1598 inode_unlock_shared(inode); 1599 inode_lock(inode); 1600 1601 mutex_lock(&delayed_node->mutex); 1602 item = __btrfs_first_delayed_insertion_item(delayed_node); 1603 while (item) { 1604 atomic_inc(&item->refs); 1605 list_add_tail(&item->readdir_list, ins_list); 1606 item = __btrfs_next_delayed_item(item); 1607 } 1608 1609 item = __btrfs_first_delayed_deletion_item(delayed_node); 1610 while (item) { 1611 atomic_inc(&item->refs); 1612 list_add_tail(&item->readdir_list, del_list); 1613 item = __btrfs_next_delayed_item(item); 1614 } 1615 mutex_unlock(&delayed_node->mutex); 1616 /* 1617 * This delayed node is still cached in the btrfs inode, so refs 1618 * must be > 1 now, and we needn't check it is going to be freed 1619 * or not. 1620 * 1621 * Besides that, this function is used to read dir, we do not 1622 * insert/delete delayed items in this period. So we also needn't 1623 * requeue or dequeue this delayed node. 1624 */ 1625 atomic_dec(&delayed_node->refs); 1626 1627 return true; 1628 } 1629 1630 void btrfs_readdir_put_delayed_items(struct inode *inode, 1631 struct list_head *ins_list, 1632 struct list_head *del_list) 1633 { 1634 struct btrfs_delayed_item *curr, *next; 1635 1636 list_for_each_entry_safe(curr, next, ins_list, readdir_list) { 1637 list_del(&curr->readdir_list); 1638 if (atomic_dec_and_test(&curr->refs)) 1639 kfree(curr); 1640 } 1641 1642 list_for_each_entry_safe(curr, next, del_list, readdir_list) { 1643 list_del(&curr->readdir_list); 1644 if (atomic_dec_and_test(&curr->refs)) 1645 kfree(curr); 1646 } 1647 1648 /* 1649 * The VFS is going to do up_read(), so we need to downgrade back to a 1650 * read lock. 1651 */ 1652 downgrade_write(&inode->i_rwsem); 1653 } 1654 1655 int btrfs_should_delete_dir_index(struct list_head *del_list, 1656 u64 index) 1657 { 1658 struct btrfs_delayed_item *curr, *next; 1659 int ret; 1660 1661 if (list_empty(del_list)) 1662 return 0; 1663 1664 list_for_each_entry_safe(curr, next, del_list, readdir_list) { 1665 if (curr->key.offset > index) 1666 break; 1667 1668 list_del(&curr->readdir_list); 1669 ret = (curr->key.offset == index); 1670 1671 if (atomic_dec_and_test(&curr->refs)) 1672 kfree(curr); 1673 1674 if (ret) 1675 return 1; 1676 else 1677 continue; 1678 } 1679 return 0; 1680 } 1681 1682 /* 1683 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree 1684 * 1685 */ 1686 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx, 1687 struct list_head *ins_list) 1688 { 1689 struct btrfs_dir_item *di; 1690 struct btrfs_delayed_item *curr, *next; 1691 struct btrfs_key location; 1692 char *name; 1693 int name_len; 1694 int over = 0; 1695 unsigned char d_type; 1696 1697 if (list_empty(ins_list)) 1698 return 0; 1699 1700 /* 1701 * Changing the data of the delayed item is impossible. So 1702 * we needn't lock them. And we have held i_mutex of the 1703 * directory, nobody can delete any directory indexes now. 1704 */ 1705 list_for_each_entry_safe(curr, next, ins_list, readdir_list) { 1706 list_del(&curr->readdir_list); 1707 1708 if (curr->key.offset < ctx->pos) { 1709 if (atomic_dec_and_test(&curr->refs)) 1710 kfree(curr); 1711 continue; 1712 } 1713 1714 ctx->pos = curr->key.offset; 1715 1716 di = (struct btrfs_dir_item *)curr->data; 1717 name = (char *)(di + 1); 1718 name_len = btrfs_stack_dir_name_len(di); 1719 1720 d_type = btrfs_filetype_table[di->type]; 1721 btrfs_disk_key_to_cpu(&location, &di->location); 1722 1723 over = !dir_emit(ctx, name, name_len, 1724 location.objectid, d_type); 1725 1726 if (atomic_dec_and_test(&curr->refs)) 1727 kfree(curr); 1728 1729 if (over) 1730 return 1; 1731 } 1732 return 0; 1733 } 1734 1735 static void fill_stack_inode_item(struct btrfs_trans_handle *trans, 1736 struct btrfs_inode_item *inode_item, 1737 struct inode *inode) 1738 { 1739 btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode)); 1740 btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode)); 1741 btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size); 1742 btrfs_set_stack_inode_mode(inode_item, inode->i_mode); 1743 btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink); 1744 btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode)); 1745 btrfs_set_stack_inode_generation(inode_item, 1746 BTRFS_I(inode)->generation); 1747 btrfs_set_stack_inode_sequence(inode_item, inode->i_version); 1748 btrfs_set_stack_inode_transid(inode_item, trans->transid); 1749 btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev); 1750 btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags); 1751 btrfs_set_stack_inode_block_group(inode_item, 0); 1752 1753 btrfs_set_stack_timespec_sec(&inode_item->atime, 1754 inode->i_atime.tv_sec); 1755 btrfs_set_stack_timespec_nsec(&inode_item->atime, 1756 inode->i_atime.tv_nsec); 1757 1758 btrfs_set_stack_timespec_sec(&inode_item->mtime, 1759 inode->i_mtime.tv_sec); 1760 btrfs_set_stack_timespec_nsec(&inode_item->mtime, 1761 inode->i_mtime.tv_nsec); 1762 1763 btrfs_set_stack_timespec_sec(&inode_item->ctime, 1764 inode->i_ctime.tv_sec); 1765 btrfs_set_stack_timespec_nsec(&inode_item->ctime, 1766 inode->i_ctime.tv_nsec); 1767 1768 btrfs_set_stack_timespec_sec(&inode_item->otime, 1769 BTRFS_I(inode)->i_otime.tv_sec); 1770 btrfs_set_stack_timespec_nsec(&inode_item->otime, 1771 BTRFS_I(inode)->i_otime.tv_nsec); 1772 } 1773 1774 int btrfs_fill_inode(struct inode *inode, u32 *rdev) 1775 { 1776 struct btrfs_delayed_node *delayed_node; 1777 struct btrfs_inode_item *inode_item; 1778 1779 delayed_node = btrfs_get_delayed_node(inode); 1780 if (!delayed_node) 1781 return -ENOENT; 1782 1783 mutex_lock(&delayed_node->mutex); 1784 if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1785 mutex_unlock(&delayed_node->mutex); 1786 btrfs_release_delayed_node(delayed_node); 1787 return -ENOENT; 1788 } 1789 1790 inode_item = &delayed_node->inode_item; 1791 1792 i_uid_write(inode, btrfs_stack_inode_uid(inode_item)); 1793 i_gid_write(inode, btrfs_stack_inode_gid(inode_item)); 1794 btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item)); 1795 inode->i_mode = btrfs_stack_inode_mode(inode_item); 1796 set_nlink(inode, btrfs_stack_inode_nlink(inode_item)); 1797 inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item)); 1798 BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item); 1799 BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item); 1800 1801 inode->i_version = btrfs_stack_inode_sequence(inode_item); 1802 inode->i_rdev = 0; 1803 *rdev = btrfs_stack_inode_rdev(inode_item); 1804 BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item); 1805 1806 inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime); 1807 inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime); 1808 1809 inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime); 1810 inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime); 1811 1812 inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime); 1813 inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime); 1814 1815 BTRFS_I(inode)->i_otime.tv_sec = 1816 btrfs_stack_timespec_sec(&inode_item->otime); 1817 BTRFS_I(inode)->i_otime.tv_nsec = 1818 btrfs_stack_timespec_nsec(&inode_item->otime); 1819 1820 inode->i_generation = BTRFS_I(inode)->generation; 1821 BTRFS_I(inode)->index_cnt = (u64)-1; 1822 1823 mutex_unlock(&delayed_node->mutex); 1824 btrfs_release_delayed_node(delayed_node); 1825 return 0; 1826 } 1827 1828 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans, 1829 struct btrfs_root *root, struct inode *inode) 1830 { 1831 struct btrfs_delayed_node *delayed_node; 1832 int ret = 0; 1833 1834 delayed_node = btrfs_get_or_create_delayed_node(inode); 1835 if (IS_ERR(delayed_node)) 1836 return PTR_ERR(delayed_node); 1837 1838 mutex_lock(&delayed_node->mutex); 1839 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1840 fill_stack_inode_item(trans, &delayed_node->inode_item, inode); 1841 goto release_node; 1842 } 1843 1844 ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode, 1845 delayed_node); 1846 if (ret) 1847 goto release_node; 1848 1849 fill_stack_inode_item(trans, &delayed_node->inode_item, inode); 1850 set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags); 1851 delayed_node->count++; 1852 atomic_inc(&root->fs_info->delayed_root->items); 1853 release_node: 1854 mutex_unlock(&delayed_node->mutex); 1855 btrfs_release_delayed_node(delayed_node); 1856 return ret; 1857 } 1858 1859 int btrfs_delayed_delete_inode_ref(struct inode *inode) 1860 { 1861 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1862 struct btrfs_delayed_node *delayed_node; 1863 1864 /* 1865 * we don't do delayed inode updates during log recovery because it 1866 * leads to enospc problems. This means we also can't do 1867 * delayed inode refs 1868 */ 1869 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 1870 return -EAGAIN; 1871 1872 delayed_node = btrfs_get_or_create_delayed_node(inode); 1873 if (IS_ERR(delayed_node)) 1874 return PTR_ERR(delayed_node); 1875 1876 /* 1877 * We don't reserve space for inode ref deletion is because: 1878 * - We ONLY do async inode ref deletion for the inode who has only 1879 * one link(i_nlink == 1), it means there is only one inode ref. 1880 * And in most case, the inode ref and the inode item are in the 1881 * same leaf, and we will deal with them at the same time. 1882 * Since we are sure we will reserve the space for the inode item, 1883 * it is unnecessary to reserve space for inode ref deletion. 1884 * - If the inode ref and the inode item are not in the same leaf, 1885 * We also needn't worry about enospc problem, because we reserve 1886 * much more space for the inode update than it needs. 1887 * - At the worst, we can steal some space from the global reservation. 1888 * It is very rare. 1889 */ 1890 mutex_lock(&delayed_node->mutex); 1891 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) 1892 goto release_node; 1893 1894 set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags); 1895 delayed_node->count++; 1896 atomic_inc(&fs_info->delayed_root->items); 1897 release_node: 1898 mutex_unlock(&delayed_node->mutex); 1899 btrfs_release_delayed_node(delayed_node); 1900 return 0; 1901 } 1902 1903 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node) 1904 { 1905 struct btrfs_root *root = delayed_node->root; 1906 struct btrfs_fs_info *fs_info = root->fs_info; 1907 struct btrfs_delayed_item *curr_item, *prev_item; 1908 1909 mutex_lock(&delayed_node->mutex); 1910 curr_item = __btrfs_first_delayed_insertion_item(delayed_node); 1911 while (curr_item) { 1912 btrfs_delayed_item_release_metadata(fs_info, curr_item); 1913 prev_item = curr_item; 1914 curr_item = __btrfs_next_delayed_item(prev_item); 1915 btrfs_release_delayed_item(prev_item); 1916 } 1917 1918 curr_item = __btrfs_first_delayed_deletion_item(delayed_node); 1919 while (curr_item) { 1920 btrfs_delayed_item_release_metadata(fs_info, curr_item); 1921 prev_item = curr_item; 1922 curr_item = __btrfs_next_delayed_item(prev_item); 1923 btrfs_release_delayed_item(prev_item); 1924 } 1925 1926 if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) 1927 btrfs_release_delayed_iref(delayed_node); 1928 1929 if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) { 1930 btrfs_delayed_inode_release_metadata(fs_info, delayed_node); 1931 btrfs_release_delayed_inode(delayed_node); 1932 } 1933 mutex_unlock(&delayed_node->mutex); 1934 } 1935 1936 void btrfs_kill_delayed_inode_items(struct inode *inode) 1937 { 1938 struct btrfs_delayed_node *delayed_node; 1939 1940 delayed_node = btrfs_get_delayed_node(inode); 1941 if (!delayed_node) 1942 return; 1943 1944 __btrfs_kill_delayed_node(delayed_node); 1945 btrfs_release_delayed_node(delayed_node); 1946 } 1947 1948 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root) 1949 { 1950 u64 inode_id = 0; 1951 struct btrfs_delayed_node *delayed_nodes[8]; 1952 int i, n; 1953 1954 while (1) { 1955 spin_lock(&root->inode_lock); 1956 n = radix_tree_gang_lookup(&root->delayed_nodes_tree, 1957 (void **)delayed_nodes, inode_id, 1958 ARRAY_SIZE(delayed_nodes)); 1959 if (!n) { 1960 spin_unlock(&root->inode_lock); 1961 break; 1962 } 1963 1964 inode_id = delayed_nodes[n - 1]->inode_id + 1; 1965 1966 for (i = 0; i < n; i++) 1967 atomic_inc(&delayed_nodes[i]->refs); 1968 spin_unlock(&root->inode_lock); 1969 1970 for (i = 0; i < n; i++) { 1971 __btrfs_kill_delayed_node(delayed_nodes[i]); 1972 btrfs_release_delayed_node(delayed_nodes[i]); 1973 } 1974 } 1975 } 1976 1977 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info) 1978 { 1979 struct btrfs_delayed_node *curr_node, *prev_node; 1980 1981 curr_node = btrfs_first_delayed_node(fs_info->delayed_root); 1982 while (curr_node) { 1983 __btrfs_kill_delayed_node(curr_node); 1984 1985 prev_node = curr_node; 1986 curr_node = btrfs_next_delayed_node(curr_node); 1987 btrfs_release_delayed_node(prev_node); 1988 } 1989 } 1990 1991