Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 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 | |
| 25 | #define BTRFS_DELAYED_WRITEBACK 400 |
| 26 | #define BTRFS_DELAYED_BACKGROUND 100 |
| 27 | |
| 28 | static struct kmem_cache *delayed_node_cache; |
| 29 | |
| 30 | int __init btrfs_delayed_inode_init(void) |
| 31 | { |
| 32 | delayed_node_cache = kmem_cache_create("delayed_node", |
| 33 | sizeof(struct btrfs_delayed_node), |
| 34 | 0, |
| 35 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, |
| 36 | NULL); |
| 37 | if (!delayed_node_cache) |
| 38 | return -ENOMEM; |
| 39 | return 0; |
| 40 | } |
| 41 | |
| 42 | void btrfs_delayed_inode_exit(void) |
| 43 | { |
| 44 | if (delayed_node_cache) |
| 45 | kmem_cache_destroy(delayed_node_cache); |
| 46 | } |
| 47 | |
| 48 | static inline void btrfs_init_delayed_node( |
| 49 | struct btrfs_delayed_node *delayed_node, |
| 50 | struct btrfs_root *root, u64 inode_id) |
| 51 | { |
| 52 | delayed_node->root = root; |
| 53 | delayed_node->inode_id = inode_id; |
| 54 | atomic_set(&delayed_node->refs, 0); |
| 55 | delayed_node->count = 0; |
| 56 | delayed_node->in_list = 0; |
| 57 | delayed_node->inode_dirty = 0; |
| 58 | delayed_node->ins_root = RB_ROOT; |
| 59 | delayed_node->del_root = RB_ROOT; |
| 60 | mutex_init(&delayed_node->mutex); |
| 61 | delayed_node->index_cnt = 0; |
| 62 | INIT_LIST_HEAD(&delayed_node->n_list); |
| 63 | INIT_LIST_HEAD(&delayed_node->p_list); |
| 64 | delayed_node->bytes_reserved = 0; |
| 65 | } |
| 66 | |
| 67 | static inline int btrfs_is_continuous_delayed_item( |
| 68 | struct btrfs_delayed_item *item1, |
| 69 | struct btrfs_delayed_item *item2) |
| 70 | { |
| 71 | if (item1->key.type == BTRFS_DIR_INDEX_KEY && |
| 72 | item1->key.objectid == item2->key.objectid && |
| 73 | item1->key.type == item2->key.type && |
| 74 | item1->key.offset + 1 == item2->key.offset) |
| 75 | return 1; |
| 76 | return 0; |
| 77 | } |
| 78 | |
| 79 | static inline struct btrfs_delayed_root *btrfs_get_delayed_root( |
| 80 | struct btrfs_root *root) |
| 81 | { |
| 82 | return root->fs_info->delayed_root; |
| 83 | } |
| 84 | |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 85 | static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode) |
| 86 | { |
| 87 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); |
| 88 | struct btrfs_root *root = btrfs_inode->root; |
| 89 | u64 ino = btrfs_ino(inode); |
| 90 | struct btrfs_delayed_node *node; |
| 91 | |
| 92 | node = ACCESS_ONCE(btrfs_inode->delayed_node); |
| 93 | if (node) { |
| 94 | atomic_inc(&node->refs); |
| 95 | return node; |
| 96 | } |
| 97 | |
| 98 | spin_lock(&root->inode_lock); |
| 99 | node = radix_tree_lookup(&root->delayed_nodes_tree, ino); |
| 100 | if (node) { |
| 101 | if (btrfs_inode->delayed_node) { |
| 102 | atomic_inc(&node->refs); /* can be accessed */ |
| 103 | BUG_ON(btrfs_inode->delayed_node != node); |
| 104 | spin_unlock(&root->inode_lock); |
| 105 | return node; |
| 106 | } |
| 107 | btrfs_inode->delayed_node = node; |
| 108 | atomic_inc(&node->refs); /* can be accessed */ |
| 109 | atomic_inc(&node->refs); /* cached in the inode */ |
| 110 | spin_unlock(&root->inode_lock); |
| 111 | return node; |
| 112 | } |
| 113 | spin_unlock(&root->inode_lock); |
| 114 | |
| 115 | return NULL; |
| 116 | } |
| 117 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 118 | static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node( |
| 119 | struct inode *inode) |
| 120 | { |
| 121 | struct btrfs_delayed_node *node; |
| 122 | struct btrfs_inode *btrfs_inode = BTRFS_I(inode); |
| 123 | struct btrfs_root *root = btrfs_inode->root; |
Chris Mason | 0d0ca30 | 2011-05-22 07:11:22 -0400 | [diff] [blame] | 124 | u64 ino = btrfs_ino(inode); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 125 | int ret; |
| 126 | |
| 127 | again: |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 128 | node = btrfs_get_delayed_node(inode); |
| 129 | if (node) |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 130 | return node; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 131 | |
| 132 | node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS); |
| 133 | if (!node) |
| 134 | return ERR_PTR(-ENOMEM); |
Chris Mason | 0d0ca30 | 2011-05-22 07:11:22 -0400 | [diff] [blame] | 135 | btrfs_init_delayed_node(node, root, ino); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 136 | |
| 137 | atomic_inc(&node->refs); /* cached in the btrfs inode */ |
| 138 | atomic_inc(&node->refs); /* can be accessed */ |
| 139 | |
| 140 | ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
| 141 | if (ret) { |
| 142 | kmem_cache_free(delayed_node_cache, node); |
| 143 | return ERR_PTR(ret); |
| 144 | } |
| 145 | |
| 146 | spin_lock(&root->inode_lock); |
Chris Mason | 0d0ca30 | 2011-05-22 07:11:22 -0400 | [diff] [blame] | 147 | ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 148 | if (ret == -EEXIST) { |
| 149 | kmem_cache_free(delayed_node_cache, node); |
| 150 | spin_unlock(&root->inode_lock); |
| 151 | radix_tree_preload_end(); |
| 152 | goto again; |
| 153 | } |
| 154 | btrfs_inode->delayed_node = node; |
| 155 | spin_unlock(&root->inode_lock); |
| 156 | radix_tree_preload_end(); |
| 157 | |
| 158 | return node; |
| 159 | } |
| 160 | |
| 161 | /* |
| 162 | * Call it when holding delayed_node->mutex |
| 163 | * |
| 164 | * If mod = 1, add this node into the prepared list. |
| 165 | */ |
| 166 | static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root, |
| 167 | struct btrfs_delayed_node *node, |
| 168 | int mod) |
| 169 | { |
| 170 | spin_lock(&root->lock); |
| 171 | if (node->in_list) { |
| 172 | if (!list_empty(&node->p_list)) |
| 173 | list_move_tail(&node->p_list, &root->prepare_list); |
| 174 | else if (mod) |
| 175 | list_add_tail(&node->p_list, &root->prepare_list); |
| 176 | } else { |
| 177 | list_add_tail(&node->n_list, &root->node_list); |
| 178 | list_add_tail(&node->p_list, &root->prepare_list); |
| 179 | atomic_inc(&node->refs); /* inserted into list */ |
| 180 | root->nodes++; |
| 181 | node->in_list = 1; |
| 182 | } |
| 183 | spin_unlock(&root->lock); |
| 184 | } |
| 185 | |
| 186 | /* Call it when holding delayed_node->mutex */ |
| 187 | static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root, |
| 188 | struct btrfs_delayed_node *node) |
| 189 | { |
| 190 | spin_lock(&root->lock); |
| 191 | if (node->in_list) { |
| 192 | root->nodes--; |
| 193 | atomic_dec(&node->refs); /* not in the list */ |
| 194 | list_del_init(&node->n_list); |
| 195 | if (!list_empty(&node->p_list)) |
| 196 | list_del_init(&node->p_list); |
| 197 | node->in_list = 0; |
| 198 | } |
| 199 | spin_unlock(&root->lock); |
| 200 | } |
| 201 | |
| 202 | struct btrfs_delayed_node *btrfs_first_delayed_node( |
| 203 | struct btrfs_delayed_root *delayed_root) |
| 204 | { |
| 205 | struct list_head *p; |
| 206 | struct btrfs_delayed_node *node = NULL; |
| 207 | |
| 208 | spin_lock(&delayed_root->lock); |
| 209 | if (list_empty(&delayed_root->node_list)) |
| 210 | goto out; |
| 211 | |
| 212 | p = delayed_root->node_list.next; |
| 213 | node = list_entry(p, struct btrfs_delayed_node, n_list); |
| 214 | atomic_inc(&node->refs); |
| 215 | out: |
| 216 | spin_unlock(&delayed_root->lock); |
| 217 | |
| 218 | return node; |
| 219 | } |
| 220 | |
| 221 | struct btrfs_delayed_node *btrfs_next_delayed_node( |
| 222 | struct btrfs_delayed_node *node) |
| 223 | { |
| 224 | struct btrfs_delayed_root *delayed_root; |
| 225 | struct list_head *p; |
| 226 | struct btrfs_delayed_node *next = NULL; |
| 227 | |
| 228 | delayed_root = node->root->fs_info->delayed_root; |
| 229 | spin_lock(&delayed_root->lock); |
| 230 | if (!node->in_list) { /* not in the list */ |
| 231 | if (list_empty(&delayed_root->node_list)) |
| 232 | goto out; |
| 233 | p = delayed_root->node_list.next; |
| 234 | } else if (list_is_last(&node->n_list, &delayed_root->node_list)) |
| 235 | goto out; |
| 236 | else |
| 237 | p = node->n_list.next; |
| 238 | |
| 239 | next = list_entry(p, struct btrfs_delayed_node, n_list); |
| 240 | atomic_inc(&next->refs); |
| 241 | out: |
| 242 | spin_unlock(&delayed_root->lock); |
| 243 | |
| 244 | return next; |
| 245 | } |
| 246 | |
| 247 | static void __btrfs_release_delayed_node( |
| 248 | struct btrfs_delayed_node *delayed_node, |
| 249 | int mod) |
| 250 | { |
| 251 | struct btrfs_delayed_root *delayed_root; |
| 252 | |
| 253 | if (!delayed_node) |
| 254 | return; |
| 255 | |
| 256 | delayed_root = delayed_node->root->fs_info->delayed_root; |
| 257 | |
| 258 | mutex_lock(&delayed_node->mutex); |
| 259 | if (delayed_node->count) |
| 260 | btrfs_queue_delayed_node(delayed_root, delayed_node, mod); |
| 261 | else |
| 262 | btrfs_dequeue_delayed_node(delayed_root, delayed_node); |
| 263 | mutex_unlock(&delayed_node->mutex); |
| 264 | |
| 265 | if (atomic_dec_and_test(&delayed_node->refs)) { |
| 266 | struct btrfs_root *root = delayed_node->root; |
| 267 | spin_lock(&root->inode_lock); |
| 268 | if (atomic_read(&delayed_node->refs) == 0) { |
| 269 | radix_tree_delete(&root->delayed_nodes_tree, |
| 270 | delayed_node->inode_id); |
| 271 | kmem_cache_free(delayed_node_cache, delayed_node); |
| 272 | } |
| 273 | spin_unlock(&root->inode_lock); |
| 274 | } |
| 275 | } |
| 276 | |
| 277 | static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node) |
| 278 | { |
| 279 | __btrfs_release_delayed_node(node, 0); |
| 280 | } |
| 281 | |
| 282 | struct btrfs_delayed_node *btrfs_first_prepared_delayed_node( |
| 283 | struct btrfs_delayed_root *delayed_root) |
| 284 | { |
| 285 | struct list_head *p; |
| 286 | struct btrfs_delayed_node *node = NULL; |
| 287 | |
| 288 | spin_lock(&delayed_root->lock); |
| 289 | if (list_empty(&delayed_root->prepare_list)) |
| 290 | goto out; |
| 291 | |
| 292 | p = delayed_root->prepare_list.next; |
| 293 | list_del_init(p); |
| 294 | node = list_entry(p, struct btrfs_delayed_node, p_list); |
| 295 | atomic_inc(&node->refs); |
| 296 | out: |
| 297 | spin_unlock(&delayed_root->lock); |
| 298 | |
| 299 | return node; |
| 300 | } |
| 301 | |
| 302 | static inline void btrfs_release_prepared_delayed_node( |
| 303 | struct btrfs_delayed_node *node) |
| 304 | { |
| 305 | __btrfs_release_delayed_node(node, 1); |
| 306 | } |
| 307 | |
| 308 | struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len) |
| 309 | { |
| 310 | struct btrfs_delayed_item *item; |
| 311 | item = kmalloc(sizeof(*item) + data_len, GFP_NOFS); |
| 312 | if (item) { |
| 313 | item->data_len = data_len; |
| 314 | item->ins_or_del = 0; |
| 315 | item->bytes_reserved = 0; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 316 | item->delayed_node = NULL; |
| 317 | atomic_set(&item->refs, 1); |
| 318 | } |
| 319 | return item; |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * __btrfs_lookup_delayed_item - look up the delayed item by key |
| 324 | * @delayed_node: pointer to the delayed node |
| 325 | * @key: the key to look up |
| 326 | * @prev: used to store the prev item if the right item isn't found |
| 327 | * @next: used to store the next item if the right item isn't found |
| 328 | * |
| 329 | * Note: if we don't find the right item, we will return the prev item and |
| 330 | * the next item. |
| 331 | */ |
| 332 | static struct btrfs_delayed_item *__btrfs_lookup_delayed_item( |
| 333 | struct rb_root *root, |
| 334 | struct btrfs_key *key, |
| 335 | struct btrfs_delayed_item **prev, |
| 336 | struct btrfs_delayed_item **next) |
| 337 | { |
| 338 | struct rb_node *node, *prev_node = NULL; |
| 339 | struct btrfs_delayed_item *delayed_item = NULL; |
| 340 | int ret = 0; |
| 341 | |
| 342 | node = root->rb_node; |
| 343 | |
| 344 | while (node) { |
| 345 | delayed_item = rb_entry(node, struct btrfs_delayed_item, |
| 346 | rb_node); |
| 347 | prev_node = node; |
| 348 | ret = btrfs_comp_cpu_keys(&delayed_item->key, key); |
| 349 | if (ret < 0) |
| 350 | node = node->rb_right; |
| 351 | else if (ret > 0) |
| 352 | node = node->rb_left; |
| 353 | else |
| 354 | return delayed_item; |
| 355 | } |
| 356 | |
| 357 | if (prev) { |
| 358 | if (!prev_node) |
| 359 | *prev = NULL; |
| 360 | else if (ret < 0) |
| 361 | *prev = delayed_item; |
| 362 | else if ((node = rb_prev(prev_node)) != NULL) { |
| 363 | *prev = rb_entry(node, struct btrfs_delayed_item, |
| 364 | rb_node); |
| 365 | } else |
| 366 | *prev = NULL; |
| 367 | } |
| 368 | |
| 369 | if (next) { |
| 370 | if (!prev_node) |
| 371 | *next = NULL; |
| 372 | else if (ret > 0) |
| 373 | *next = delayed_item; |
| 374 | else if ((node = rb_next(prev_node)) != NULL) { |
| 375 | *next = rb_entry(node, struct btrfs_delayed_item, |
| 376 | rb_node); |
| 377 | } else |
| 378 | *next = NULL; |
| 379 | } |
| 380 | return NULL; |
| 381 | } |
| 382 | |
| 383 | struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item( |
| 384 | struct btrfs_delayed_node *delayed_node, |
| 385 | struct btrfs_key *key) |
| 386 | { |
| 387 | struct btrfs_delayed_item *item; |
| 388 | |
| 389 | item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, |
| 390 | NULL, NULL); |
| 391 | return item; |
| 392 | } |
| 393 | |
| 394 | struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item( |
| 395 | struct btrfs_delayed_node *delayed_node, |
| 396 | struct btrfs_key *key) |
| 397 | { |
| 398 | struct btrfs_delayed_item *item; |
| 399 | |
| 400 | item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key, |
| 401 | NULL, NULL); |
| 402 | return item; |
| 403 | } |
| 404 | |
| 405 | struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item( |
| 406 | struct btrfs_delayed_node *delayed_node, |
| 407 | struct btrfs_key *key) |
| 408 | { |
| 409 | struct btrfs_delayed_item *item, *next; |
| 410 | |
| 411 | item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key, |
| 412 | NULL, &next); |
| 413 | if (!item) |
| 414 | item = next; |
| 415 | |
| 416 | return item; |
| 417 | } |
| 418 | |
| 419 | struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item( |
| 420 | struct btrfs_delayed_node *delayed_node, |
| 421 | struct btrfs_key *key) |
| 422 | { |
| 423 | struct btrfs_delayed_item *item, *next; |
| 424 | |
| 425 | item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key, |
| 426 | NULL, &next); |
| 427 | if (!item) |
| 428 | item = next; |
| 429 | |
| 430 | return item; |
| 431 | } |
| 432 | |
| 433 | static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node, |
| 434 | struct btrfs_delayed_item *ins, |
| 435 | int action) |
| 436 | { |
| 437 | struct rb_node **p, *node; |
| 438 | struct rb_node *parent_node = NULL; |
| 439 | struct rb_root *root; |
| 440 | struct btrfs_delayed_item *item; |
| 441 | int cmp; |
| 442 | |
| 443 | if (action == BTRFS_DELAYED_INSERTION_ITEM) |
| 444 | root = &delayed_node->ins_root; |
| 445 | else if (action == BTRFS_DELAYED_DELETION_ITEM) |
| 446 | root = &delayed_node->del_root; |
| 447 | else |
| 448 | BUG(); |
| 449 | p = &root->rb_node; |
| 450 | node = &ins->rb_node; |
| 451 | |
| 452 | while (*p) { |
| 453 | parent_node = *p; |
| 454 | item = rb_entry(parent_node, struct btrfs_delayed_item, |
| 455 | rb_node); |
| 456 | |
| 457 | cmp = btrfs_comp_cpu_keys(&item->key, &ins->key); |
| 458 | if (cmp < 0) |
| 459 | p = &(*p)->rb_right; |
| 460 | else if (cmp > 0) |
| 461 | p = &(*p)->rb_left; |
| 462 | else |
| 463 | return -EEXIST; |
| 464 | } |
| 465 | |
| 466 | rb_link_node(node, parent_node, p); |
| 467 | rb_insert_color(node, root); |
| 468 | ins->delayed_node = delayed_node; |
| 469 | ins->ins_or_del = action; |
| 470 | |
| 471 | if (ins->key.type == BTRFS_DIR_INDEX_KEY && |
| 472 | action == BTRFS_DELAYED_INSERTION_ITEM && |
| 473 | ins->key.offset >= delayed_node->index_cnt) |
| 474 | delayed_node->index_cnt = ins->key.offset + 1; |
| 475 | |
| 476 | delayed_node->count++; |
| 477 | atomic_inc(&delayed_node->root->fs_info->delayed_root->items); |
| 478 | return 0; |
| 479 | } |
| 480 | |
| 481 | static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node, |
| 482 | struct btrfs_delayed_item *item) |
| 483 | { |
| 484 | return __btrfs_add_delayed_item(node, item, |
| 485 | BTRFS_DELAYED_INSERTION_ITEM); |
| 486 | } |
| 487 | |
| 488 | static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node, |
| 489 | struct btrfs_delayed_item *item) |
| 490 | { |
| 491 | return __btrfs_add_delayed_item(node, item, |
| 492 | BTRFS_DELAYED_DELETION_ITEM); |
| 493 | } |
| 494 | |
| 495 | static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item) |
| 496 | { |
| 497 | struct rb_root *root; |
| 498 | struct btrfs_delayed_root *delayed_root; |
| 499 | |
| 500 | delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root; |
| 501 | |
| 502 | BUG_ON(!delayed_root); |
| 503 | BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM && |
| 504 | delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM); |
| 505 | |
| 506 | if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM) |
| 507 | root = &delayed_item->delayed_node->ins_root; |
| 508 | else |
| 509 | root = &delayed_item->delayed_node->del_root; |
| 510 | |
| 511 | rb_erase(&delayed_item->rb_node, root); |
| 512 | delayed_item->delayed_node->count--; |
| 513 | atomic_dec(&delayed_root->items); |
| 514 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND && |
| 515 | waitqueue_active(&delayed_root->wait)) |
| 516 | wake_up(&delayed_root->wait); |
| 517 | } |
| 518 | |
| 519 | static void btrfs_release_delayed_item(struct btrfs_delayed_item *item) |
| 520 | { |
| 521 | if (item) { |
| 522 | __btrfs_remove_delayed_item(item); |
| 523 | if (atomic_dec_and_test(&item->refs)) |
| 524 | kfree(item); |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item( |
| 529 | struct btrfs_delayed_node *delayed_node) |
| 530 | { |
| 531 | struct rb_node *p; |
| 532 | struct btrfs_delayed_item *item = NULL; |
| 533 | |
| 534 | p = rb_first(&delayed_node->ins_root); |
| 535 | if (p) |
| 536 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); |
| 537 | |
| 538 | return item; |
| 539 | } |
| 540 | |
| 541 | struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item( |
| 542 | struct btrfs_delayed_node *delayed_node) |
| 543 | { |
| 544 | struct rb_node *p; |
| 545 | struct btrfs_delayed_item *item = NULL; |
| 546 | |
| 547 | p = rb_first(&delayed_node->del_root); |
| 548 | if (p) |
| 549 | item = rb_entry(p, struct btrfs_delayed_item, rb_node); |
| 550 | |
| 551 | return item; |
| 552 | } |
| 553 | |
| 554 | struct btrfs_delayed_item *__btrfs_next_delayed_item( |
| 555 | struct btrfs_delayed_item *item) |
| 556 | { |
| 557 | struct rb_node *p; |
| 558 | struct btrfs_delayed_item *next = NULL; |
| 559 | |
| 560 | p = rb_next(&item->rb_node); |
| 561 | if (p) |
| 562 | next = rb_entry(p, struct btrfs_delayed_item, rb_node); |
| 563 | |
| 564 | return next; |
| 565 | } |
| 566 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 567 | static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root, |
| 568 | u64 root_id) |
| 569 | { |
| 570 | struct btrfs_key root_key; |
| 571 | |
| 572 | if (root->objectid == root_id) |
| 573 | return root; |
| 574 | |
| 575 | root_key.objectid = root_id; |
| 576 | root_key.type = BTRFS_ROOT_ITEM_KEY; |
| 577 | root_key.offset = (u64)-1; |
| 578 | return btrfs_read_fs_root_no_name(root->fs_info, &root_key); |
| 579 | } |
| 580 | |
| 581 | static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans, |
| 582 | struct btrfs_root *root, |
| 583 | struct btrfs_delayed_item *item) |
| 584 | { |
| 585 | struct btrfs_block_rsv *src_rsv; |
| 586 | struct btrfs_block_rsv *dst_rsv; |
| 587 | u64 num_bytes; |
| 588 | int ret; |
| 589 | |
| 590 | if (!trans->bytes_reserved) |
| 591 | return 0; |
| 592 | |
| 593 | src_rsv = trans->block_rsv; |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 594 | dst_rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 595 | |
| 596 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
| 597 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 598 | if (!ret) |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 599 | item->bytes_reserved = num_bytes; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 600 | |
| 601 | return ret; |
| 602 | } |
| 603 | |
| 604 | static void btrfs_delayed_item_release_metadata(struct btrfs_root *root, |
| 605 | struct btrfs_delayed_item *item) |
| 606 | { |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 607 | struct btrfs_block_rsv *rsv; |
| 608 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 609 | if (!item->bytes_reserved) |
| 610 | return; |
| 611 | |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 612 | rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 613 | btrfs_block_rsv_release(root, rsv, |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 614 | item->bytes_reserved); |
| 615 | } |
| 616 | |
| 617 | static int btrfs_delayed_inode_reserve_metadata( |
| 618 | struct btrfs_trans_handle *trans, |
| 619 | struct btrfs_root *root, |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 620 | struct inode *inode, |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 621 | struct btrfs_delayed_node *node) |
| 622 | { |
| 623 | struct btrfs_block_rsv *src_rsv; |
| 624 | struct btrfs_block_rsv *dst_rsv; |
| 625 | u64 num_bytes; |
| 626 | int ret; |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 627 | int release = false; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 628 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 629 | src_rsv = trans->block_rsv; |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 630 | dst_rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 631 | |
| 632 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
Josef Bacik | c06a0e1 | 2011-11-04 19:56:02 -0400 | [diff] [blame] | 633 | |
| 634 | /* |
| 635 | * btrfs_dirty_inode will update the inode under btrfs_join_transaction |
| 636 | * which doesn't reserve space for speed. This is a problem since we |
| 637 | * still need to reserve space for this update, so try to reserve the |
| 638 | * space. |
| 639 | * |
| 640 | * Now if src_rsv == delalloc_block_rsv we'll let it just steal since |
| 641 | * we're accounted for. |
| 642 | */ |
| 643 | if (!trans->bytes_reserved && |
| 644 | src_rsv != &root->fs_info->delalloc_block_rsv) { |
| 645 | ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes); |
| 646 | /* |
| 647 | * Since we're under a transaction reserve_metadata_bytes could |
| 648 | * try to commit the transaction which will make it return |
| 649 | * EAGAIN to make us stop the transaction we have, so return |
| 650 | * ENOSPC instead so that btrfs_dirty_inode knows what to do. |
| 651 | */ |
| 652 | if (ret == -EAGAIN) |
| 653 | ret = -ENOSPC; |
| 654 | if (!ret) |
| 655 | node->bytes_reserved = num_bytes; |
| 656 | return ret; |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 657 | } else if (src_rsv == &root->fs_info->delalloc_block_rsv) { |
| 658 | spin_lock(&BTRFS_I(inode)->lock); |
| 659 | if (BTRFS_I(inode)->delalloc_meta_reserved) { |
| 660 | BTRFS_I(inode)->delalloc_meta_reserved = 0; |
| 661 | spin_unlock(&BTRFS_I(inode)->lock); |
| 662 | release = true; |
| 663 | goto migrate; |
| 664 | } |
| 665 | spin_unlock(&BTRFS_I(inode)->lock); |
| 666 | |
| 667 | /* Ok we didn't have space pre-reserved. This shouldn't happen |
| 668 | * too often but it can happen if we do delalloc to an existing |
| 669 | * inode which gets dirtied because of the time update, and then |
| 670 | * isn't touched again until after the transaction commits and |
| 671 | * then we try to write out the data. First try to be nice and |
| 672 | * reserve something strictly for us. If not be a pain and try |
| 673 | * to steal from the delalloc block rsv. |
| 674 | */ |
| 675 | ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes); |
| 676 | if (!ret) |
| 677 | goto out; |
| 678 | |
| 679 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); |
| 680 | if (!ret) |
| 681 | goto out; |
| 682 | |
| 683 | /* |
| 684 | * Ok this is a problem, let's just steal from the global rsv |
| 685 | * since this really shouldn't happen that often. |
| 686 | */ |
| 687 | WARN_ON(1); |
| 688 | ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv, |
| 689 | dst_rsv, num_bytes); |
| 690 | goto out; |
Josef Bacik | c06a0e1 | 2011-11-04 19:56:02 -0400 | [diff] [blame] | 691 | } |
| 692 | |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 693 | migrate: |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 694 | ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes); |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 695 | if (unlikely(ret)) { |
| 696 | /* This shouldn't happen */ |
| 697 | BUG_ON(release); |
| 698 | return ret; |
| 699 | } |
| 700 | |
| 701 | out: |
| 702 | /* |
| 703 | * Migrate only takes a reservation, it doesn't touch the size of the |
| 704 | * block_rsv. This is to simplify people who don't normally have things |
| 705 | * migrated from their block rsv. If they go to release their |
| 706 | * reservation, that will decrease the size as well, so if migrate |
| 707 | * reduced size we'd end up with a negative size. But for the |
| 708 | * delalloc_meta_reserved stuff we will only know to drop 1 reservation, |
| 709 | * but we could in fact do this reserve/migrate dance several times |
| 710 | * between the time we did the original reservation and we'd clean it |
| 711 | * up. So to take care of this, release the space for the meta |
| 712 | * reservation here. I think it may be time for a documentation page on |
| 713 | * how block rsvs. work. |
| 714 | */ |
| 715 | if (release) |
| 716 | btrfs_block_rsv_release(root, src_rsv, num_bytes); |
| 717 | node->bytes_reserved = num_bytes; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 718 | |
| 719 | return ret; |
| 720 | } |
| 721 | |
| 722 | static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root, |
| 723 | struct btrfs_delayed_node *node) |
| 724 | { |
| 725 | struct btrfs_block_rsv *rsv; |
| 726 | |
| 727 | if (!node->bytes_reserved) |
| 728 | return; |
| 729 | |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 730 | rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 731 | btrfs_block_rsv_release(root, rsv, |
| 732 | node->bytes_reserved); |
| 733 | node->bytes_reserved = 0; |
| 734 | } |
| 735 | |
| 736 | /* |
| 737 | * This helper will insert some continuous items into the same leaf according |
| 738 | * to the free space of the leaf. |
| 739 | */ |
| 740 | static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans, |
| 741 | struct btrfs_root *root, |
| 742 | struct btrfs_path *path, |
| 743 | struct btrfs_delayed_item *item) |
| 744 | { |
| 745 | struct btrfs_delayed_item *curr, *next; |
| 746 | int free_space; |
| 747 | int total_data_size = 0, total_size = 0; |
| 748 | struct extent_buffer *leaf; |
| 749 | char *data_ptr; |
| 750 | struct btrfs_key *keys; |
| 751 | u32 *data_size; |
| 752 | struct list_head head; |
| 753 | int slot; |
| 754 | int nitems; |
| 755 | int i; |
| 756 | int ret = 0; |
| 757 | |
| 758 | BUG_ON(!path->nodes[0]); |
| 759 | |
| 760 | leaf = path->nodes[0]; |
| 761 | free_space = btrfs_leaf_free_space(root, leaf); |
| 762 | INIT_LIST_HEAD(&head); |
| 763 | |
| 764 | next = item; |
Chris Mason | 17aca1c | 2011-06-03 01:13:45 -0400 | [diff] [blame] | 765 | nitems = 0; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 766 | |
| 767 | /* |
| 768 | * count the number of the continuous items that we can insert in batch |
| 769 | */ |
| 770 | while (total_size + next->data_len + sizeof(struct btrfs_item) <= |
| 771 | free_space) { |
| 772 | total_data_size += next->data_len; |
| 773 | total_size += next->data_len + sizeof(struct btrfs_item); |
| 774 | list_add_tail(&next->tree_list, &head); |
| 775 | nitems++; |
| 776 | |
| 777 | curr = next; |
| 778 | next = __btrfs_next_delayed_item(curr); |
| 779 | if (!next) |
| 780 | break; |
| 781 | |
| 782 | if (!btrfs_is_continuous_delayed_item(curr, next)) |
| 783 | break; |
| 784 | } |
| 785 | |
| 786 | if (!nitems) { |
| 787 | ret = 0; |
| 788 | goto out; |
| 789 | } |
| 790 | |
| 791 | /* |
| 792 | * we need allocate some memory space, but it might cause the task |
| 793 | * to sleep, so we set all locked nodes in the path to blocking locks |
| 794 | * first. |
| 795 | */ |
| 796 | btrfs_set_path_blocking(path); |
| 797 | |
| 798 | keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS); |
| 799 | if (!keys) { |
| 800 | ret = -ENOMEM; |
| 801 | goto out; |
| 802 | } |
| 803 | |
| 804 | data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS); |
| 805 | if (!data_size) { |
| 806 | ret = -ENOMEM; |
| 807 | goto error; |
| 808 | } |
| 809 | |
| 810 | /* get keys of all the delayed items */ |
| 811 | i = 0; |
| 812 | list_for_each_entry(next, &head, tree_list) { |
| 813 | keys[i] = next->key; |
| 814 | data_size[i] = next->data_len; |
| 815 | i++; |
| 816 | } |
| 817 | |
| 818 | /* reset all the locked nodes in the patch to spinning locks. */ |
Chris Mason | bd68151 | 2011-07-16 15:23:14 -0400 | [diff] [blame] | 819 | btrfs_clear_path_blocking(path, NULL, 0); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 820 | |
| 821 | /* insert the keys of the items */ |
| 822 | ret = setup_items_for_insert(trans, root, path, keys, data_size, |
| 823 | total_data_size, total_size, nitems); |
| 824 | if (ret) |
| 825 | goto error; |
| 826 | |
| 827 | /* insert the dir index items */ |
| 828 | slot = path->slots[0]; |
| 829 | list_for_each_entry_safe(curr, next, &head, tree_list) { |
| 830 | data_ptr = btrfs_item_ptr(leaf, slot, char); |
| 831 | write_extent_buffer(leaf, &curr->data, |
| 832 | (unsigned long)data_ptr, |
| 833 | curr->data_len); |
| 834 | slot++; |
| 835 | |
| 836 | btrfs_delayed_item_release_metadata(root, curr); |
| 837 | |
| 838 | list_del(&curr->tree_list); |
| 839 | btrfs_release_delayed_item(curr); |
| 840 | } |
| 841 | |
| 842 | error: |
| 843 | kfree(data_size); |
| 844 | kfree(keys); |
| 845 | out: |
| 846 | return ret; |
| 847 | } |
| 848 | |
| 849 | /* |
| 850 | * This helper can just do simple insertion that needn't extend item for new |
| 851 | * data, such as directory name index insertion, inode insertion. |
| 852 | */ |
| 853 | static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans, |
| 854 | struct btrfs_root *root, |
| 855 | struct btrfs_path *path, |
| 856 | struct btrfs_delayed_item *delayed_item) |
| 857 | { |
| 858 | struct extent_buffer *leaf; |
| 859 | struct btrfs_item *item; |
| 860 | char *ptr; |
| 861 | int ret; |
| 862 | |
| 863 | ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key, |
| 864 | delayed_item->data_len); |
| 865 | if (ret < 0 && ret != -EEXIST) |
| 866 | return ret; |
| 867 | |
| 868 | leaf = path->nodes[0]; |
| 869 | |
| 870 | item = btrfs_item_nr(leaf, path->slots[0]); |
| 871 | ptr = btrfs_item_ptr(leaf, path->slots[0], char); |
| 872 | |
| 873 | write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr, |
| 874 | delayed_item->data_len); |
| 875 | btrfs_mark_buffer_dirty(leaf); |
| 876 | |
| 877 | btrfs_delayed_item_release_metadata(root, delayed_item); |
| 878 | return 0; |
| 879 | } |
| 880 | |
| 881 | /* |
| 882 | * we insert an item first, then if there are some continuous items, we try |
| 883 | * to insert those items into the same leaf. |
| 884 | */ |
| 885 | static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans, |
| 886 | struct btrfs_path *path, |
| 887 | struct btrfs_root *root, |
| 888 | struct btrfs_delayed_node *node) |
| 889 | { |
| 890 | struct btrfs_delayed_item *curr, *prev; |
| 891 | int ret = 0; |
| 892 | |
| 893 | do_again: |
| 894 | mutex_lock(&node->mutex); |
| 895 | curr = __btrfs_first_delayed_insertion_item(node); |
| 896 | if (!curr) |
| 897 | goto insert_end; |
| 898 | |
| 899 | ret = btrfs_insert_delayed_item(trans, root, path, curr); |
| 900 | if (ret < 0) { |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 901 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 902 | goto insert_end; |
| 903 | } |
| 904 | |
| 905 | prev = curr; |
| 906 | curr = __btrfs_next_delayed_item(prev); |
| 907 | if (curr && btrfs_is_continuous_delayed_item(prev, curr)) { |
| 908 | /* insert the continuous items into the same leaf */ |
| 909 | path->slots[0]++; |
| 910 | btrfs_batch_insert_items(trans, root, path, curr); |
| 911 | } |
| 912 | btrfs_release_delayed_item(prev); |
| 913 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 914 | |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 915 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 916 | mutex_unlock(&node->mutex); |
| 917 | goto do_again; |
| 918 | |
| 919 | insert_end: |
| 920 | mutex_unlock(&node->mutex); |
| 921 | return ret; |
| 922 | } |
| 923 | |
| 924 | static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans, |
| 925 | struct btrfs_root *root, |
| 926 | struct btrfs_path *path, |
| 927 | struct btrfs_delayed_item *item) |
| 928 | { |
| 929 | struct btrfs_delayed_item *curr, *next; |
| 930 | struct extent_buffer *leaf; |
| 931 | struct btrfs_key key; |
| 932 | struct list_head head; |
| 933 | int nitems, i, last_item; |
| 934 | int ret = 0; |
| 935 | |
| 936 | BUG_ON(!path->nodes[0]); |
| 937 | |
| 938 | leaf = path->nodes[0]; |
| 939 | |
| 940 | i = path->slots[0]; |
| 941 | last_item = btrfs_header_nritems(leaf) - 1; |
| 942 | if (i > last_item) |
| 943 | return -ENOENT; /* FIXME: Is errno suitable? */ |
| 944 | |
| 945 | next = item; |
| 946 | INIT_LIST_HEAD(&head); |
| 947 | btrfs_item_key_to_cpu(leaf, &key, i); |
| 948 | nitems = 0; |
| 949 | /* |
| 950 | * count the number of the dir index items that we can delete in batch |
| 951 | */ |
| 952 | while (btrfs_comp_cpu_keys(&next->key, &key) == 0) { |
| 953 | list_add_tail(&next->tree_list, &head); |
| 954 | nitems++; |
| 955 | |
| 956 | curr = next; |
| 957 | next = __btrfs_next_delayed_item(curr); |
| 958 | if (!next) |
| 959 | break; |
| 960 | |
| 961 | if (!btrfs_is_continuous_delayed_item(curr, next)) |
| 962 | break; |
| 963 | |
| 964 | i++; |
| 965 | if (i > last_item) |
| 966 | break; |
| 967 | btrfs_item_key_to_cpu(leaf, &key, i); |
| 968 | } |
| 969 | |
| 970 | if (!nitems) |
| 971 | return 0; |
| 972 | |
| 973 | ret = btrfs_del_items(trans, root, path, path->slots[0], nitems); |
| 974 | if (ret) |
| 975 | goto out; |
| 976 | |
| 977 | list_for_each_entry_safe(curr, next, &head, tree_list) { |
| 978 | btrfs_delayed_item_release_metadata(root, curr); |
| 979 | list_del(&curr->tree_list); |
| 980 | btrfs_release_delayed_item(curr); |
| 981 | } |
| 982 | |
| 983 | out: |
| 984 | return ret; |
| 985 | } |
| 986 | |
| 987 | static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans, |
| 988 | struct btrfs_path *path, |
| 989 | struct btrfs_root *root, |
| 990 | struct btrfs_delayed_node *node) |
| 991 | { |
| 992 | struct btrfs_delayed_item *curr, *prev; |
| 993 | int ret = 0; |
| 994 | |
| 995 | do_again: |
| 996 | mutex_lock(&node->mutex); |
| 997 | curr = __btrfs_first_delayed_deletion_item(node); |
| 998 | if (!curr) |
| 999 | goto delete_fail; |
| 1000 | |
| 1001 | ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1); |
| 1002 | if (ret < 0) |
| 1003 | goto delete_fail; |
| 1004 | else if (ret > 0) { |
| 1005 | /* |
| 1006 | * can't find the item which the node points to, so this node |
| 1007 | * is invalid, just drop it. |
| 1008 | */ |
| 1009 | prev = curr; |
| 1010 | curr = __btrfs_next_delayed_item(prev); |
| 1011 | btrfs_release_delayed_item(prev); |
| 1012 | ret = 0; |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 1013 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1014 | if (curr) |
| 1015 | goto do_again; |
| 1016 | else |
| 1017 | goto delete_fail; |
| 1018 | } |
| 1019 | |
| 1020 | btrfs_batch_delete_items(trans, root, path, curr); |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 1021 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1022 | mutex_unlock(&node->mutex); |
| 1023 | goto do_again; |
| 1024 | |
| 1025 | delete_fail: |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 1026 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1027 | mutex_unlock(&node->mutex); |
| 1028 | return ret; |
| 1029 | } |
| 1030 | |
| 1031 | static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node) |
| 1032 | { |
| 1033 | struct btrfs_delayed_root *delayed_root; |
| 1034 | |
| 1035 | if (delayed_node && delayed_node->inode_dirty) { |
| 1036 | BUG_ON(!delayed_node->root); |
| 1037 | delayed_node->inode_dirty = 0; |
| 1038 | delayed_node->count--; |
| 1039 | |
| 1040 | delayed_root = delayed_node->root->fs_info->delayed_root; |
| 1041 | atomic_dec(&delayed_root->items); |
| 1042 | if (atomic_read(&delayed_root->items) < |
| 1043 | BTRFS_DELAYED_BACKGROUND && |
| 1044 | waitqueue_active(&delayed_root->wait)) |
| 1045 | wake_up(&delayed_root->wait); |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans, |
| 1050 | struct btrfs_root *root, |
| 1051 | struct btrfs_path *path, |
| 1052 | struct btrfs_delayed_node *node) |
| 1053 | { |
| 1054 | struct btrfs_key key; |
| 1055 | struct btrfs_inode_item *inode_item; |
| 1056 | struct extent_buffer *leaf; |
| 1057 | int ret; |
| 1058 | |
| 1059 | mutex_lock(&node->mutex); |
| 1060 | if (!node->inode_dirty) { |
| 1061 | mutex_unlock(&node->mutex); |
| 1062 | return 0; |
| 1063 | } |
| 1064 | |
| 1065 | key.objectid = node->inode_id; |
| 1066 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
| 1067 | key.offset = 0; |
| 1068 | ret = btrfs_lookup_inode(trans, root, path, &key, 1); |
| 1069 | if (ret > 0) { |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 1070 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1071 | mutex_unlock(&node->mutex); |
| 1072 | return -ENOENT; |
| 1073 | } else if (ret < 0) { |
| 1074 | mutex_unlock(&node->mutex); |
| 1075 | return ret; |
| 1076 | } |
| 1077 | |
| 1078 | btrfs_unlock_up_safe(path, 1); |
| 1079 | leaf = path->nodes[0]; |
| 1080 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
| 1081 | struct btrfs_inode_item); |
| 1082 | write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item, |
| 1083 | sizeof(struct btrfs_inode_item)); |
| 1084 | btrfs_mark_buffer_dirty(leaf); |
Chris Mason | 945d896 | 2011-05-22 12:33:42 -0400 | [diff] [blame] | 1085 | btrfs_release_path(path); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1086 | |
| 1087 | btrfs_delayed_inode_release_metadata(root, node); |
| 1088 | btrfs_release_delayed_inode(node); |
| 1089 | mutex_unlock(&node->mutex); |
| 1090 | |
| 1091 | return 0; |
| 1092 | } |
| 1093 | |
| 1094 | /* Called when committing the transaction. */ |
| 1095 | int btrfs_run_delayed_items(struct btrfs_trans_handle *trans, |
| 1096 | struct btrfs_root *root) |
| 1097 | { |
| 1098 | struct btrfs_delayed_root *delayed_root; |
| 1099 | struct btrfs_delayed_node *curr_node, *prev_node; |
| 1100 | struct btrfs_path *path; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1101 | struct btrfs_block_rsv *block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1102 | int ret = 0; |
| 1103 | |
| 1104 | path = btrfs_alloc_path(); |
| 1105 | if (!path) |
| 1106 | return -ENOMEM; |
| 1107 | path->leave_spinning = 1; |
| 1108 | |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1109 | block_rsv = trans->block_rsv; |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 1110 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1111 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1112 | delayed_root = btrfs_get_delayed_root(root); |
| 1113 | |
| 1114 | curr_node = btrfs_first_delayed_node(delayed_root); |
| 1115 | while (curr_node) { |
| 1116 | root = curr_node->root; |
| 1117 | ret = btrfs_insert_delayed_items(trans, path, root, |
| 1118 | curr_node); |
| 1119 | if (!ret) |
| 1120 | ret = btrfs_delete_delayed_items(trans, path, root, |
| 1121 | curr_node); |
| 1122 | if (!ret) |
| 1123 | ret = btrfs_update_delayed_inode(trans, root, path, |
| 1124 | curr_node); |
| 1125 | if (ret) { |
| 1126 | btrfs_release_delayed_node(curr_node); |
| 1127 | break; |
| 1128 | } |
| 1129 | |
| 1130 | prev_node = curr_node; |
| 1131 | curr_node = btrfs_next_delayed_node(curr_node); |
| 1132 | btrfs_release_delayed_node(prev_node); |
| 1133 | } |
| 1134 | |
| 1135 | btrfs_free_path(path); |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1136 | trans->block_rsv = block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1137 | return ret; |
| 1138 | } |
| 1139 | |
| 1140 | static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, |
| 1141 | struct btrfs_delayed_node *node) |
| 1142 | { |
| 1143 | struct btrfs_path *path; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1144 | struct btrfs_block_rsv *block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1145 | int ret; |
| 1146 | |
| 1147 | path = btrfs_alloc_path(); |
| 1148 | if (!path) |
| 1149 | return -ENOMEM; |
| 1150 | path->leave_spinning = 1; |
| 1151 | |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1152 | block_rsv = trans->block_rsv; |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 1153 | trans->block_rsv = &node->root->fs_info->delayed_block_rsv; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1154 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1155 | ret = btrfs_insert_delayed_items(trans, path, node->root, node); |
| 1156 | if (!ret) |
| 1157 | ret = btrfs_delete_delayed_items(trans, path, node->root, node); |
| 1158 | if (!ret) |
| 1159 | ret = btrfs_update_delayed_inode(trans, node->root, path, node); |
| 1160 | btrfs_free_path(path); |
| 1161 | |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1162 | trans->block_rsv = block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1163 | return ret; |
| 1164 | } |
| 1165 | |
| 1166 | int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans, |
| 1167 | struct inode *inode) |
| 1168 | { |
| 1169 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); |
| 1170 | int ret; |
| 1171 | |
| 1172 | if (!delayed_node) |
| 1173 | return 0; |
| 1174 | |
| 1175 | mutex_lock(&delayed_node->mutex); |
| 1176 | if (!delayed_node->count) { |
| 1177 | mutex_unlock(&delayed_node->mutex); |
| 1178 | btrfs_release_delayed_node(delayed_node); |
| 1179 | return 0; |
| 1180 | } |
| 1181 | mutex_unlock(&delayed_node->mutex); |
| 1182 | |
| 1183 | ret = __btrfs_commit_inode_delayed_items(trans, delayed_node); |
| 1184 | btrfs_release_delayed_node(delayed_node); |
| 1185 | return ret; |
| 1186 | } |
| 1187 | |
| 1188 | void btrfs_remove_delayed_node(struct inode *inode) |
| 1189 | { |
| 1190 | struct btrfs_delayed_node *delayed_node; |
| 1191 | |
| 1192 | delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node); |
| 1193 | if (!delayed_node) |
| 1194 | return; |
| 1195 | |
| 1196 | BTRFS_I(inode)->delayed_node = NULL; |
| 1197 | btrfs_release_delayed_node(delayed_node); |
| 1198 | } |
| 1199 | |
| 1200 | struct btrfs_async_delayed_node { |
| 1201 | struct btrfs_root *root; |
| 1202 | struct btrfs_delayed_node *delayed_node; |
| 1203 | struct btrfs_work work; |
| 1204 | }; |
| 1205 | |
| 1206 | static void btrfs_async_run_delayed_node_done(struct btrfs_work *work) |
| 1207 | { |
| 1208 | struct btrfs_async_delayed_node *async_node; |
| 1209 | struct btrfs_trans_handle *trans; |
| 1210 | struct btrfs_path *path; |
| 1211 | struct btrfs_delayed_node *delayed_node = NULL; |
| 1212 | struct btrfs_root *root; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1213 | struct btrfs_block_rsv *block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1214 | unsigned long nr = 0; |
| 1215 | int need_requeue = 0; |
| 1216 | int ret; |
| 1217 | |
| 1218 | async_node = container_of(work, struct btrfs_async_delayed_node, work); |
| 1219 | |
| 1220 | path = btrfs_alloc_path(); |
| 1221 | if (!path) |
| 1222 | goto out; |
| 1223 | path->leave_spinning = 1; |
| 1224 | |
| 1225 | delayed_node = async_node->delayed_node; |
| 1226 | root = delayed_node->root; |
| 1227 | |
Chris Mason | ff5714c | 2011-05-28 07:00:39 -0400 | [diff] [blame] | 1228 | trans = btrfs_join_transaction(root); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1229 | if (IS_ERR(trans)) |
| 1230 | goto free_path; |
| 1231 | |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1232 | block_rsv = trans->block_rsv; |
Josef Bacik | 6d668dd | 2011-11-03 22:54:25 -0400 | [diff] [blame] | 1233 | trans->block_rsv = &root->fs_info->delayed_block_rsv; |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1234 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1235 | ret = btrfs_insert_delayed_items(trans, path, root, delayed_node); |
| 1236 | if (!ret) |
| 1237 | ret = btrfs_delete_delayed_items(trans, path, root, |
| 1238 | delayed_node); |
| 1239 | |
| 1240 | if (!ret) |
| 1241 | btrfs_update_delayed_inode(trans, root, path, delayed_node); |
| 1242 | |
| 1243 | /* |
| 1244 | * Maybe new delayed items have been inserted, so we need requeue |
| 1245 | * the work. Besides that, we must dequeue the empty delayed nodes |
| 1246 | * to avoid the race between delayed items balance and the worker. |
| 1247 | * The race like this: |
| 1248 | * Task1 Worker thread |
| 1249 | * count == 0, needn't requeue |
| 1250 | * also needn't insert the |
| 1251 | * delayed node into prepare |
| 1252 | * list again. |
| 1253 | * add lots of delayed items |
| 1254 | * queue the delayed node |
| 1255 | * already in the list, |
| 1256 | * and not in the prepare |
| 1257 | * list, it means the delayed |
| 1258 | * node is being dealt with |
| 1259 | * by the worker. |
| 1260 | * do delayed items balance |
| 1261 | * the delayed node is being |
| 1262 | * dealt with by the worker |
| 1263 | * now, just wait. |
| 1264 | * the worker goto idle. |
| 1265 | * Task1 will sleep until the transaction is commited. |
| 1266 | */ |
| 1267 | mutex_lock(&delayed_node->mutex); |
| 1268 | if (delayed_node->count) |
| 1269 | need_requeue = 1; |
| 1270 | else |
| 1271 | btrfs_dequeue_delayed_node(root->fs_info->delayed_root, |
| 1272 | delayed_node); |
| 1273 | mutex_unlock(&delayed_node->mutex); |
| 1274 | |
| 1275 | nr = trans->blocks_used; |
| 1276 | |
Miao Xie | 19fd294 | 2011-06-15 10:47:30 +0000 | [diff] [blame] | 1277 | trans->block_rsv = block_rsv; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1278 | btrfs_end_transaction_dmeta(trans, root); |
| 1279 | __btrfs_btree_balance_dirty(root, nr); |
| 1280 | free_path: |
| 1281 | btrfs_free_path(path); |
| 1282 | out: |
| 1283 | if (need_requeue) |
| 1284 | btrfs_requeue_work(&async_node->work); |
| 1285 | else { |
| 1286 | btrfs_release_prepared_delayed_node(delayed_node); |
| 1287 | kfree(async_node); |
| 1288 | } |
| 1289 | } |
| 1290 | |
| 1291 | static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root, |
| 1292 | struct btrfs_root *root, int all) |
| 1293 | { |
| 1294 | struct btrfs_async_delayed_node *async_node; |
| 1295 | struct btrfs_delayed_node *curr; |
| 1296 | int count = 0; |
| 1297 | |
| 1298 | again: |
| 1299 | curr = btrfs_first_prepared_delayed_node(delayed_root); |
| 1300 | if (!curr) |
| 1301 | return 0; |
| 1302 | |
| 1303 | async_node = kmalloc(sizeof(*async_node), GFP_NOFS); |
| 1304 | if (!async_node) { |
| 1305 | btrfs_release_prepared_delayed_node(curr); |
| 1306 | return -ENOMEM; |
| 1307 | } |
| 1308 | |
| 1309 | async_node->root = root; |
| 1310 | async_node->delayed_node = curr; |
| 1311 | |
| 1312 | async_node->work.func = btrfs_async_run_delayed_node_done; |
| 1313 | async_node->work.flags = 0; |
| 1314 | |
| 1315 | btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work); |
| 1316 | count++; |
| 1317 | |
| 1318 | if (all || count < 4) |
| 1319 | goto again; |
| 1320 | |
| 1321 | return 0; |
| 1322 | } |
| 1323 | |
Chris Mason | e999376 | 2011-06-17 16:14:09 -0400 | [diff] [blame] | 1324 | void btrfs_assert_delayed_root_empty(struct btrfs_root *root) |
| 1325 | { |
| 1326 | struct btrfs_delayed_root *delayed_root; |
| 1327 | delayed_root = btrfs_get_delayed_root(root); |
| 1328 | WARN_ON(btrfs_first_delayed_node(delayed_root)); |
| 1329 | } |
| 1330 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1331 | void btrfs_balance_delayed_items(struct btrfs_root *root) |
| 1332 | { |
| 1333 | struct btrfs_delayed_root *delayed_root; |
| 1334 | |
| 1335 | delayed_root = btrfs_get_delayed_root(root); |
| 1336 | |
| 1337 | if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) |
| 1338 | return; |
| 1339 | |
| 1340 | if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) { |
| 1341 | int ret; |
| 1342 | ret = btrfs_wq_run_delayed_node(delayed_root, root, 1); |
| 1343 | if (ret) |
| 1344 | return; |
| 1345 | |
| 1346 | wait_event_interruptible_timeout( |
| 1347 | delayed_root->wait, |
| 1348 | (atomic_read(&delayed_root->items) < |
| 1349 | BTRFS_DELAYED_BACKGROUND), |
| 1350 | HZ); |
| 1351 | return; |
| 1352 | } |
| 1353 | |
| 1354 | btrfs_wq_run_delayed_node(delayed_root, root, 0); |
| 1355 | } |
| 1356 | |
| 1357 | int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans, |
| 1358 | struct btrfs_root *root, const char *name, |
| 1359 | int name_len, struct inode *dir, |
| 1360 | struct btrfs_disk_key *disk_key, u8 type, |
| 1361 | u64 index) |
| 1362 | { |
| 1363 | struct btrfs_delayed_node *delayed_node; |
| 1364 | struct btrfs_delayed_item *delayed_item; |
| 1365 | struct btrfs_dir_item *dir_item; |
| 1366 | int ret; |
| 1367 | |
| 1368 | delayed_node = btrfs_get_or_create_delayed_node(dir); |
| 1369 | if (IS_ERR(delayed_node)) |
| 1370 | return PTR_ERR(delayed_node); |
| 1371 | |
| 1372 | delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len); |
| 1373 | if (!delayed_item) { |
| 1374 | ret = -ENOMEM; |
| 1375 | goto release_node; |
| 1376 | } |
| 1377 | |
| 1378 | ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item); |
| 1379 | /* |
| 1380 | * we have reserved enough space when we start a new transaction, |
| 1381 | * so reserving metadata failure is impossible |
| 1382 | */ |
| 1383 | BUG_ON(ret); |
| 1384 | |
Chris Mason | 0d0ca30 | 2011-05-22 07:11:22 -0400 | [diff] [blame] | 1385 | delayed_item->key.objectid = btrfs_ino(dir); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1386 | btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY); |
| 1387 | delayed_item->key.offset = index; |
| 1388 | |
| 1389 | dir_item = (struct btrfs_dir_item *)delayed_item->data; |
| 1390 | dir_item->location = *disk_key; |
| 1391 | dir_item->transid = cpu_to_le64(trans->transid); |
| 1392 | dir_item->data_len = 0; |
| 1393 | dir_item->name_len = cpu_to_le16(name_len); |
| 1394 | dir_item->type = type; |
| 1395 | memcpy((char *)(dir_item + 1), name, name_len); |
| 1396 | |
| 1397 | mutex_lock(&delayed_node->mutex); |
| 1398 | ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item); |
| 1399 | if (unlikely(ret)) { |
| 1400 | printk(KERN_ERR "err add delayed dir index item(name: %s) into " |
| 1401 | "the insertion tree of the delayed node" |
| 1402 | "(root id: %llu, inode id: %llu, errno: %d)\n", |
| 1403 | name, |
| 1404 | (unsigned long long)delayed_node->root->objectid, |
| 1405 | (unsigned long long)delayed_node->inode_id, |
| 1406 | ret); |
| 1407 | BUG(); |
| 1408 | } |
| 1409 | mutex_unlock(&delayed_node->mutex); |
| 1410 | |
| 1411 | release_node: |
| 1412 | btrfs_release_delayed_node(delayed_node); |
| 1413 | return ret; |
| 1414 | } |
| 1415 | |
| 1416 | static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root, |
| 1417 | struct btrfs_delayed_node *node, |
| 1418 | struct btrfs_key *key) |
| 1419 | { |
| 1420 | struct btrfs_delayed_item *item; |
| 1421 | |
| 1422 | mutex_lock(&node->mutex); |
| 1423 | item = __btrfs_lookup_delayed_insertion_item(node, key); |
| 1424 | if (!item) { |
| 1425 | mutex_unlock(&node->mutex); |
| 1426 | return 1; |
| 1427 | } |
| 1428 | |
| 1429 | btrfs_delayed_item_release_metadata(root, item); |
| 1430 | btrfs_release_delayed_item(item); |
| 1431 | mutex_unlock(&node->mutex); |
| 1432 | return 0; |
| 1433 | } |
| 1434 | |
| 1435 | int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans, |
| 1436 | struct btrfs_root *root, struct inode *dir, |
| 1437 | u64 index) |
| 1438 | { |
| 1439 | struct btrfs_delayed_node *node; |
| 1440 | struct btrfs_delayed_item *item; |
| 1441 | struct btrfs_key item_key; |
| 1442 | int ret; |
| 1443 | |
| 1444 | node = btrfs_get_or_create_delayed_node(dir); |
| 1445 | if (IS_ERR(node)) |
| 1446 | return PTR_ERR(node); |
| 1447 | |
Chris Mason | 0d0ca30 | 2011-05-22 07:11:22 -0400 | [diff] [blame] | 1448 | item_key.objectid = btrfs_ino(dir); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1449 | btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY); |
| 1450 | item_key.offset = index; |
| 1451 | |
| 1452 | ret = btrfs_delete_delayed_insertion_item(root, node, &item_key); |
| 1453 | if (!ret) |
| 1454 | goto end; |
| 1455 | |
| 1456 | item = btrfs_alloc_delayed_item(0); |
| 1457 | if (!item) { |
| 1458 | ret = -ENOMEM; |
| 1459 | goto end; |
| 1460 | } |
| 1461 | |
| 1462 | item->key = item_key; |
| 1463 | |
| 1464 | ret = btrfs_delayed_item_reserve_metadata(trans, root, item); |
| 1465 | /* |
| 1466 | * we have reserved enough space when we start a new transaction, |
| 1467 | * so reserving metadata failure is impossible. |
| 1468 | */ |
| 1469 | BUG_ON(ret); |
| 1470 | |
| 1471 | mutex_lock(&node->mutex); |
| 1472 | ret = __btrfs_add_delayed_deletion_item(node, item); |
| 1473 | if (unlikely(ret)) { |
| 1474 | printk(KERN_ERR "err add delayed dir index item(index: %llu) " |
| 1475 | "into the deletion tree of the delayed node" |
| 1476 | "(root id: %llu, inode id: %llu, errno: %d)\n", |
| 1477 | (unsigned long long)index, |
| 1478 | (unsigned long long)node->root->objectid, |
| 1479 | (unsigned long long)node->inode_id, |
| 1480 | ret); |
| 1481 | BUG(); |
| 1482 | } |
| 1483 | mutex_unlock(&node->mutex); |
| 1484 | end: |
| 1485 | btrfs_release_delayed_node(node); |
| 1486 | return ret; |
| 1487 | } |
| 1488 | |
| 1489 | int btrfs_inode_delayed_dir_index_count(struct inode *inode) |
| 1490 | { |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 1491 | struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1492 | |
| 1493 | if (!delayed_node) |
| 1494 | return -ENOENT; |
| 1495 | |
| 1496 | /* |
| 1497 | * Since we have held i_mutex of this directory, it is impossible that |
| 1498 | * a new directory index is added into the delayed node and index_cnt |
| 1499 | * is updated now. So we needn't lock the delayed node. |
| 1500 | */ |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 1501 | if (!delayed_node->index_cnt) { |
| 1502 | btrfs_release_delayed_node(delayed_node); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1503 | return -EINVAL; |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 1504 | } |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1505 | |
| 1506 | BTRFS_I(inode)->index_cnt = delayed_node->index_cnt; |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 1507 | btrfs_release_delayed_node(delayed_node); |
| 1508 | return 0; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1509 | } |
| 1510 | |
| 1511 | void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list, |
| 1512 | struct list_head *del_list) |
| 1513 | { |
| 1514 | struct btrfs_delayed_node *delayed_node; |
| 1515 | struct btrfs_delayed_item *item; |
| 1516 | |
| 1517 | delayed_node = btrfs_get_delayed_node(inode); |
| 1518 | if (!delayed_node) |
| 1519 | return; |
| 1520 | |
| 1521 | mutex_lock(&delayed_node->mutex); |
| 1522 | item = __btrfs_first_delayed_insertion_item(delayed_node); |
| 1523 | while (item) { |
| 1524 | atomic_inc(&item->refs); |
| 1525 | list_add_tail(&item->readdir_list, ins_list); |
| 1526 | item = __btrfs_next_delayed_item(item); |
| 1527 | } |
| 1528 | |
| 1529 | item = __btrfs_first_delayed_deletion_item(delayed_node); |
| 1530 | while (item) { |
| 1531 | atomic_inc(&item->refs); |
| 1532 | list_add_tail(&item->readdir_list, del_list); |
| 1533 | item = __btrfs_next_delayed_item(item); |
| 1534 | } |
| 1535 | mutex_unlock(&delayed_node->mutex); |
| 1536 | /* |
| 1537 | * This delayed node is still cached in the btrfs inode, so refs |
| 1538 | * must be > 1 now, and we needn't check it is going to be freed |
| 1539 | * or not. |
| 1540 | * |
| 1541 | * Besides that, this function is used to read dir, we do not |
| 1542 | * insert/delete delayed items in this period. So we also needn't |
| 1543 | * requeue or dequeue this delayed node. |
| 1544 | */ |
| 1545 | atomic_dec(&delayed_node->refs); |
| 1546 | } |
| 1547 | |
| 1548 | void btrfs_put_delayed_items(struct list_head *ins_list, |
| 1549 | struct list_head *del_list) |
| 1550 | { |
| 1551 | struct btrfs_delayed_item *curr, *next; |
| 1552 | |
| 1553 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { |
| 1554 | list_del(&curr->readdir_list); |
| 1555 | if (atomic_dec_and_test(&curr->refs)) |
| 1556 | kfree(curr); |
| 1557 | } |
| 1558 | |
| 1559 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { |
| 1560 | list_del(&curr->readdir_list); |
| 1561 | if (atomic_dec_and_test(&curr->refs)) |
| 1562 | kfree(curr); |
| 1563 | } |
| 1564 | } |
| 1565 | |
| 1566 | int btrfs_should_delete_dir_index(struct list_head *del_list, |
| 1567 | u64 index) |
| 1568 | { |
| 1569 | struct btrfs_delayed_item *curr, *next; |
| 1570 | int ret; |
| 1571 | |
| 1572 | if (list_empty(del_list)) |
| 1573 | return 0; |
| 1574 | |
| 1575 | list_for_each_entry_safe(curr, next, del_list, readdir_list) { |
| 1576 | if (curr->key.offset > index) |
| 1577 | break; |
| 1578 | |
| 1579 | list_del(&curr->readdir_list); |
| 1580 | ret = (curr->key.offset == index); |
| 1581 | |
| 1582 | if (atomic_dec_and_test(&curr->refs)) |
| 1583 | kfree(curr); |
| 1584 | |
| 1585 | if (ret) |
| 1586 | return 1; |
| 1587 | else |
| 1588 | continue; |
| 1589 | } |
| 1590 | return 0; |
| 1591 | } |
| 1592 | |
| 1593 | /* |
| 1594 | * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree |
| 1595 | * |
| 1596 | */ |
| 1597 | int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent, |
| 1598 | filldir_t filldir, |
| 1599 | struct list_head *ins_list) |
| 1600 | { |
| 1601 | struct btrfs_dir_item *di; |
| 1602 | struct btrfs_delayed_item *curr, *next; |
| 1603 | struct btrfs_key location; |
| 1604 | char *name; |
| 1605 | int name_len; |
| 1606 | int over = 0; |
| 1607 | unsigned char d_type; |
| 1608 | |
| 1609 | if (list_empty(ins_list)) |
| 1610 | return 0; |
| 1611 | |
| 1612 | /* |
| 1613 | * Changing the data of the delayed item is impossible. So |
| 1614 | * we needn't lock them. And we have held i_mutex of the |
| 1615 | * directory, nobody can delete any directory indexes now. |
| 1616 | */ |
| 1617 | list_for_each_entry_safe(curr, next, ins_list, readdir_list) { |
| 1618 | list_del(&curr->readdir_list); |
| 1619 | |
| 1620 | if (curr->key.offset < filp->f_pos) { |
| 1621 | if (atomic_dec_and_test(&curr->refs)) |
| 1622 | kfree(curr); |
| 1623 | continue; |
| 1624 | } |
| 1625 | |
| 1626 | filp->f_pos = curr->key.offset; |
| 1627 | |
| 1628 | di = (struct btrfs_dir_item *)curr->data; |
| 1629 | name = (char *)(di + 1); |
| 1630 | name_len = le16_to_cpu(di->name_len); |
| 1631 | |
| 1632 | d_type = btrfs_filetype_table[di->type]; |
| 1633 | btrfs_disk_key_to_cpu(&location, &di->location); |
| 1634 | |
| 1635 | over = filldir(dirent, name, name_len, curr->key.offset, |
| 1636 | location.objectid, d_type); |
| 1637 | |
| 1638 | if (atomic_dec_and_test(&curr->refs)) |
| 1639 | kfree(curr); |
| 1640 | |
| 1641 | if (over) |
| 1642 | return 1; |
| 1643 | } |
| 1644 | return 0; |
| 1645 | } |
| 1646 | |
| 1647 | BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item, |
| 1648 | generation, 64); |
| 1649 | BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item, |
| 1650 | sequence, 64); |
| 1651 | BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item, |
| 1652 | transid, 64); |
| 1653 | BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64); |
| 1654 | BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item, |
| 1655 | nbytes, 64); |
| 1656 | BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item, |
| 1657 | block_group, 64); |
| 1658 | BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32); |
| 1659 | BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32); |
| 1660 | BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32); |
| 1661 | BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32); |
| 1662 | BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64); |
| 1663 | BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64); |
| 1664 | |
| 1665 | BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64); |
| 1666 | BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32); |
| 1667 | |
| 1668 | static void fill_stack_inode_item(struct btrfs_trans_handle *trans, |
| 1669 | struct btrfs_inode_item *inode_item, |
| 1670 | struct inode *inode) |
| 1671 | { |
| 1672 | btrfs_set_stack_inode_uid(inode_item, inode->i_uid); |
| 1673 | btrfs_set_stack_inode_gid(inode_item, inode->i_gid); |
| 1674 | btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size); |
| 1675 | btrfs_set_stack_inode_mode(inode_item, inode->i_mode); |
| 1676 | btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink); |
| 1677 | btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode)); |
| 1678 | btrfs_set_stack_inode_generation(inode_item, |
| 1679 | BTRFS_I(inode)->generation); |
| 1680 | btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence); |
| 1681 | btrfs_set_stack_inode_transid(inode_item, trans->transid); |
| 1682 | btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev); |
| 1683 | btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags); |
Chris Mason | ff5714c | 2011-05-28 07:00:39 -0400 | [diff] [blame] | 1684 | btrfs_set_stack_inode_block_group(inode_item, 0); |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1685 | |
| 1686 | btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item), |
| 1687 | inode->i_atime.tv_sec); |
| 1688 | btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item), |
| 1689 | inode->i_atime.tv_nsec); |
| 1690 | |
| 1691 | btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item), |
| 1692 | inode->i_mtime.tv_sec); |
| 1693 | btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item), |
| 1694 | inode->i_mtime.tv_nsec); |
| 1695 | |
| 1696 | btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item), |
| 1697 | inode->i_ctime.tv_sec); |
| 1698 | btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item), |
| 1699 | inode->i_ctime.tv_nsec); |
| 1700 | } |
| 1701 | |
Miao Xie | 2f7e33d | 2011-06-23 07:27:13 +0000 | [diff] [blame] | 1702 | int btrfs_fill_inode(struct inode *inode, u32 *rdev) |
| 1703 | { |
| 1704 | struct btrfs_delayed_node *delayed_node; |
| 1705 | struct btrfs_inode_item *inode_item; |
| 1706 | struct btrfs_timespec *tspec; |
| 1707 | |
| 1708 | delayed_node = btrfs_get_delayed_node(inode); |
| 1709 | if (!delayed_node) |
| 1710 | return -ENOENT; |
| 1711 | |
| 1712 | mutex_lock(&delayed_node->mutex); |
| 1713 | if (!delayed_node->inode_dirty) { |
| 1714 | mutex_unlock(&delayed_node->mutex); |
| 1715 | btrfs_release_delayed_node(delayed_node); |
| 1716 | return -ENOENT; |
| 1717 | } |
| 1718 | |
| 1719 | inode_item = &delayed_node->inode_item; |
| 1720 | |
| 1721 | inode->i_uid = btrfs_stack_inode_uid(inode_item); |
| 1722 | inode->i_gid = btrfs_stack_inode_gid(inode_item); |
| 1723 | btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item)); |
| 1724 | inode->i_mode = btrfs_stack_inode_mode(inode_item); |
| 1725 | inode->i_nlink = btrfs_stack_inode_nlink(inode_item); |
| 1726 | inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item)); |
| 1727 | BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item); |
| 1728 | BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item); |
| 1729 | inode->i_rdev = 0; |
| 1730 | *rdev = btrfs_stack_inode_rdev(inode_item); |
| 1731 | BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item); |
| 1732 | |
| 1733 | tspec = btrfs_inode_atime(inode_item); |
| 1734 | inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec); |
| 1735 | inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec); |
| 1736 | |
| 1737 | tspec = btrfs_inode_mtime(inode_item); |
| 1738 | inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec); |
| 1739 | inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec); |
| 1740 | |
| 1741 | tspec = btrfs_inode_ctime(inode_item); |
| 1742 | inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec); |
| 1743 | inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec); |
| 1744 | |
| 1745 | inode->i_generation = BTRFS_I(inode)->generation; |
| 1746 | BTRFS_I(inode)->index_cnt = (u64)-1; |
| 1747 | |
| 1748 | mutex_unlock(&delayed_node->mutex); |
| 1749 | btrfs_release_delayed_node(delayed_node); |
| 1750 | return 0; |
| 1751 | } |
| 1752 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1753 | int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans, |
| 1754 | struct btrfs_root *root, struct inode *inode) |
| 1755 | { |
| 1756 | struct btrfs_delayed_node *delayed_node; |
David Sterba | aa0467d | 2011-06-03 16:29:08 +0200 | [diff] [blame] | 1757 | int ret = 0; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1758 | |
| 1759 | delayed_node = btrfs_get_or_create_delayed_node(inode); |
| 1760 | if (IS_ERR(delayed_node)) |
| 1761 | return PTR_ERR(delayed_node); |
| 1762 | |
| 1763 | mutex_lock(&delayed_node->mutex); |
| 1764 | if (delayed_node->inode_dirty) { |
| 1765 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); |
| 1766 | goto release_node; |
| 1767 | } |
| 1768 | |
Josef Bacik | 7fd2ae2 | 2011-11-08 15:47:34 -0500 | [diff] [blame^] | 1769 | ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode, |
| 1770 | delayed_node); |
Josef Bacik | c06a0e1 | 2011-11-04 19:56:02 -0400 | [diff] [blame] | 1771 | if (ret) |
| 1772 | goto release_node; |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 1773 | |
| 1774 | fill_stack_inode_item(trans, &delayed_node->inode_item, inode); |
| 1775 | delayed_node->inode_dirty = 1; |
| 1776 | delayed_node->count++; |
| 1777 | atomic_inc(&root->fs_info->delayed_root->items); |
| 1778 | release_node: |
| 1779 | mutex_unlock(&delayed_node->mutex); |
| 1780 | btrfs_release_delayed_node(delayed_node); |
| 1781 | return ret; |
| 1782 | } |
| 1783 | |
| 1784 | static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node) |
| 1785 | { |
| 1786 | struct btrfs_root *root = delayed_node->root; |
| 1787 | struct btrfs_delayed_item *curr_item, *prev_item; |
| 1788 | |
| 1789 | mutex_lock(&delayed_node->mutex); |
| 1790 | curr_item = __btrfs_first_delayed_insertion_item(delayed_node); |
| 1791 | while (curr_item) { |
| 1792 | btrfs_delayed_item_release_metadata(root, curr_item); |
| 1793 | prev_item = curr_item; |
| 1794 | curr_item = __btrfs_next_delayed_item(prev_item); |
| 1795 | btrfs_release_delayed_item(prev_item); |
| 1796 | } |
| 1797 | |
| 1798 | curr_item = __btrfs_first_delayed_deletion_item(delayed_node); |
| 1799 | while (curr_item) { |
| 1800 | btrfs_delayed_item_release_metadata(root, curr_item); |
| 1801 | prev_item = curr_item; |
| 1802 | curr_item = __btrfs_next_delayed_item(prev_item); |
| 1803 | btrfs_release_delayed_item(prev_item); |
| 1804 | } |
| 1805 | |
| 1806 | if (delayed_node->inode_dirty) { |
| 1807 | btrfs_delayed_inode_release_metadata(root, delayed_node); |
| 1808 | btrfs_release_delayed_inode(delayed_node); |
| 1809 | } |
| 1810 | mutex_unlock(&delayed_node->mutex); |
| 1811 | } |
| 1812 | |
| 1813 | void btrfs_kill_delayed_inode_items(struct inode *inode) |
| 1814 | { |
| 1815 | struct btrfs_delayed_node *delayed_node; |
| 1816 | |
| 1817 | delayed_node = btrfs_get_delayed_node(inode); |
| 1818 | if (!delayed_node) |
| 1819 | return; |
| 1820 | |
| 1821 | __btrfs_kill_delayed_node(delayed_node); |
| 1822 | btrfs_release_delayed_node(delayed_node); |
| 1823 | } |
| 1824 | |
| 1825 | void btrfs_kill_all_delayed_nodes(struct btrfs_root *root) |
| 1826 | { |
| 1827 | u64 inode_id = 0; |
| 1828 | struct btrfs_delayed_node *delayed_nodes[8]; |
| 1829 | int i, n; |
| 1830 | |
| 1831 | while (1) { |
| 1832 | spin_lock(&root->inode_lock); |
| 1833 | n = radix_tree_gang_lookup(&root->delayed_nodes_tree, |
| 1834 | (void **)delayed_nodes, inode_id, |
| 1835 | ARRAY_SIZE(delayed_nodes)); |
| 1836 | if (!n) { |
| 1837 | spin_unlock(&root->inode_lock); |
| 1838 | break; |
| 1839 | } |
| 1840 | |
| 1841 | inode_id = delayed_nodes[n - 1]->inode_id + 1; |
| 1842 | |
| 1843 | for (i = 0; i < n; i++) |
| 1844 | atomic_inc(&delayed_nodes[i]->refs); |
| 1845 | spin_unlock(&root->inode_lock); |
| 1846 | |
| 1847 | for (i = 0; i < n; i++) { |
| 1848 | __btrfs_kill_delayed_node(delayed_nodes[i]); |
| 1849 | btrfs_release_delayed_node(delayed_nodes[i]); |
| 1850 | } |
| 1851 | } |
| 1852 | } |