1 /* 2 * Copyright (C) 2008 Red Hat. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/pagemap.h> 20 #include <linux/sched.h> 21 #include <linux/slab.h> 22 #include <linux/math64.h> 23 #include <linux/ratelimit.h> 24 #include "ctree.h" 25 #include "free-space-cache.h" 26 #include "transaction.h" 27 #include "disk-io.h" 28 #include "extent_io.h" 29 #include "inode-map.h" 30 31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) 32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) 33 34 static int link_free_space(struct btrfs_free_space_ctl *ctl, 35 struct btrfs_free_space *info); 36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl, 37 struct btrfs_free_space *info); 38 39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root, 40 struct btrfs_path *path, 41 u64 offset) 42 { 43 struct btrfs_key key; 44 struct btrfs_key location; 45 struct btrfs_disk_key disk_key; 46 struct btrfs_free_space_header *header; 47 struct extent_buffer *leaf; 48 struct inode *inode = NULL; 49 int ret; 50 51 key.objectid = BTRFS_FREE_SPACE_OBJECTID; 52 key.offset = offset; 53 key.type = 0; 54 55 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 56 if (ret < 0) 57 return ERR_PTR(ret); 58 if (ret > 0) { 59 btrfs_release_path(path); 60 return ERR_PTR(-ENOENT); 61 } 62 63 leaf = path->nodes[0]; 64 header = btrfs_item_ptr(leaf, path->slots[0], 65 struct btrfs_free_space_header); 66 btrfs_free_space_key(leaf, header, &disk_key); 67 btrfs_disk_key_to_cpu(&location, &disk_key); 68 btrfs_release_path(path); 69 70 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); 71 if (!inode) 72 return ERR_PTR(-ENOENT); 73 if (IS_ERR(inode)) 74 return inode; 75 if (is_bad_inode(inode)) { 76 iput(inode); 77 return ERR_PTR(-ENOENT); 78 } 79 80 mapping_set_gfp_mask(inode->i_mapping, 81 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); 82 83 return inode; 84 } 85 86 struct inode *lookup_free_space_inode(struct btrfs_root *root, 87 struct btrfs_block_group_cache 88 *block_group, struct btrfs_path *path) 89 { 90 struct inode *inode = NULL; 91 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; 92 93 spin_lock(&block_group->lock); 94 if (block_group->inode) 95 inode = igrab(block_group->inode); 96 spin_unlock(&block_group->lock); 97 if (inode) 98 return inode; 99 100 inode = __lookup_free_space_inode(root, path, 101 block_group->key.objectid); 102 if (IS_ERR(inode)) 103 return inode; 104 105 spin_lock(&block_group->lock); 106 if (!((BTRFS_I(inode)->flags & flags) == flags)) { 107 printk(KERN_INFO "Old style space inode found, converting.\n"); 108 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM | 109 BTRFS_INODE_NODATACOW; 110 block_group->disk_cache_state = BTRFS_DC_CLEAR; 111 } 112 113 if (!block_group->iref) { 114 block_group->inode = igrab(inode); 115 block_group->iref = 1; 116 } 117 spin_unlock(&block_group->lock); 118 119 return inode; 120 } 121 122 int __create_free_space_inode(struct btrfs_root *root, 123 struct btrfs_trans_handle *trans, 124 struct btrfs_path *path, u64 ino, u64 offset) 125 { 126 struct btrfs_key key; 127 struct btrfs_disk_key disk_key; 128 struct btrfs_free_space_header *header; 129 struct btrfs_inode_item *inode_item; 130 struct extent_buffer *leaf; 131 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC; 132 int ret; 133 134 ret = btrfs_insert_empty_inode(trans, root, path, ino); 135 if (ret) 136 return ret; 137 138 /* We inline crc's for the free disk space cache */ 139 if (ino != BTRFS_FREE_INO_OBJECTID) 140 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW; 141 142 leaf = path->nodes[0]; 143 inode_item = btrfs_item_ptr(leaf, path->slots[0], 144 struct btrfs_inode_item); 145 btrfs_item_key(leaf, &disk_key, path->slots[0]); 146 memset_extent_buffer(leaf, 0, (unsigned long)inode_item, 147 sizeof(*inode_item)); 148 btrfs_set_inode_generation(leaf, inode_item, trans->transid); 149 btrfs_set_inode_size(leaf, inode_item, 0); 150 btrfs_set_inode_nbytes(leaf, inode_item, 0); 151 btrfs_set_inode_uid(leaf, inode_item, 0); 152 btrfs_set_inode_gid(leaf, inode_item, 0); 153 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); 154 btrfs_set_inode_flags(leaf, inode_item, flags); 155 btrfs_set_inode_nlink(leaf, inode_item, 1); 156 btrfs_set_inode_transid(leaf, inode_item, trans->transid); 157 btrfs_set_inode_block_group(leaf, inode_item, offset); 158 btrfs_mark_buffer_dirty(leaf); 159 btrfs_release_path(path); 160 161 key.objectid = BTRFS_FREE_SPACE_OBJECTID; 162 key.offset = offset; 163 key.type = 0; 164 165 ret = btrfs_insert_empty_item(trans, root, path, &key, 166 sizeof(struct btrfs_free_space_header)); 167 if (ret < 0) { 168 btrfs_release_path(path); 169 return ret; 170 } 171 leaf = path->nodes[0]; 172 header = btrfs_item_ptr(leaf, path->slots[0], 173 struct btrfs_free_space_header); 174 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); 175 btrfs_set_free_space_key(leaf, header, &disk_key); 176 btrfs_mark_buffer_dirty(leaf); 177 btrfs_release_path(path); 178 179 return 0; 180 } 181 182 int create_free_space_inode(struct btrfs_root *root, 183 struct btrfs_trans_handle *trans, 184 struct btrfs_block_group_cache *block_group, 185 struct btrfs_path *path) 186 { 187 int ret; 188 u64 ino; 189 190 ret = btrfs_find_free_objectid(root, &ino); 191 if (ret < 0) 192 return ret; 193 194 return __create_free_space_inode(root, trans, path, ino, 195 block_group->key.objectid); 196 } 197 198 int btrfs_truncate_free_space_cache(struct btrfs_root *root, 199 struct btrfs_trans_handle *trans, 200 struct btrfs_path *path, 201 struct inode *inode) 202 { 203 struct btrfs_block_rsv *rsv; 204 u64 needed_bytes; 205 loff_t oldsize; 206 int ret = 0; 207 208 rsv = trans->block_rsv; 209 trans->block_rsv = &root->fs_info->global_block_rsv; 210 211 /* 1 for slack space, 1 for updating the inode */ 212 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) + 213 btrfs_calc_trans_metadata_size(root, 1); 214 215 spin_lock(&trans->block_rsv->lock); 216 if (trans->block_rsv->reserved < needed_bytes) { 217 spin_unlock(&trans->block_rsv->lock); 218 trans->block_rsv = rsv; 219 return -ENOSPC; 220 } 221 spin_unlock(&trans->block_rsv->lock); 222 223 oldsize = i_size_read(inode); 224 btrfs_i_size_write(inode, 0); 225 truncate_pagecache(inode, oldsize, 0); 226 227 /* 228 * We don't need an orphan item because truncating the free space cache 229 * will never be split across transactions. 230 */ 231 ret = btrfs_truncate_inode_items(trans, root, inode, 232 0, BTRFS_EXTENT_DATA_KEY); 233 234 if (ret) { 235 trans->block_rsv = rsv; 236 btrfs_abort_transaction(trans, root, ret); 237 return ret; 238 } 239 240 ret = btrfs_update_inode(trans, root, inode); 241 if (ret) 242 btrfs_abort_transaction(trans, root, ret); 243 trans->block_rsv = rsv; 244 245 return ret; 246 } 247 248 static int readahead_cache(struct inode *inode) 249 { 250 struct file_ra_state *ra; 251 unsigned long last_index; 252 253 ra = kzalloc(sizeof(*ra), GFP_NOFS); 254 if (!ra) 255 return -ENOMEM; 256 257 file_ra_state_init(ra, inode->i_mapping); 258 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; 259 260 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); 261 262 kfree(ra); 263 264 return 0; 265 } 266 267 struct io_ctl { 268 void *cur, *orig; 269 struct page *page; 270 struct page **pages; 271 struct btrfs_root *root; 272 unsigned long size; 273 int index; 274 int num_pages; 275 unsigned check_crcs:1; 276 }; 277 278 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode, 279 struct btrfs_root *root) 280 { 281 memset(io_ctl, 0, sizeof(struct io_ctl)); 282 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> 283 PAGE_CACHE_SHIFT; 284 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages, 285 GFP_NOFS); 286 if (!io_ctl->pages) 287 return -ENOMEM; 288 io_ctl->root = root; 289 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID) 290 io_ctl->check_crcs = 1; 291 return 0; 292 } 293 294 static void io_ctl_free(struct io_ctl *io_ctl) 295 { 296 kfree(io_ctl->pages); 297 } 298 299 static void io_ctl_unmap_page(struct io_ctl *io_ctl) 300 { 301 if (io_ctl->cur) { 302 kunmap(io_ctl->page); 303 io_ctl->cur = NULL; 304 io_ctl->orig = NULL; 305 } 306 } 307 308 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear) 309 { 310 BUG_ON(io_ctl->index >= io_ctl->num_pages); 311 io_ctl->page = io_ctl->pages[io_ctl->index++]; 312 io_ctl->cur = kmap(io_ctl->page); 313 io_ctl->orig = io_ctl->cur; 314 io_ctl->size = PAGE_CACHE_SIZE; 315 if (clear) 316 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE); 317 } 318 319 static void io_ctl_drop_pages(struct io_ctl *io_ctl) 320 { 321 int i; 322 323 io_ctl_unmap_page(io_ctl); 324 325 for (i = 0; i < io_ctl->num_pages; i++) { 326 if (io_ctl->pages[i]) { 327 ClearPageChecked(io_ctl->pages[i]); 328 unlock_page(io_ctl->pages[i]); 329 page_cache_release(io_ctl->pages[i]); 330 } 331 } 332 } 333 334 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode, 335 int uptodate) 336 { 337 struct page *page; 338 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 339 int i; 340 341 for (i = 0; i < io_ctl->num_pages; i++) { 342 page = find_or_create_page(inode->i_mapping, i, mask); 343 if (!page) { 344 io_ctl_drop_pages(io_ctl); 345 return -ENOMEM; 346 } 347 io_ctl->pages[i] = page; 348 if (uptodate && !PageUptodate(page)) { 349 btrfs_readpage(NULL, page); 350 lock_page(page); 351 if (!PageUptodate(page)) { 352 printk(KERN_ERR "btrfs: error reading free " 353 "space cache\n"); 354 io_ctl_drop_pages(io_ctl); 355 return -EIO; 356 } 357 } 358 } 359 360 for (i = 0; i < io_ctl->num_pages; i++) { 361 clear_page_dirty_for_io(io_ctl->pages[i]); 362 set_page_extent_mapped(io_ctl->pages[i]); 363 } 364 365 return 0; 366 } 367 368 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation) 369 { 370 __le64 *val; 371 372 io_ctl_map_page(io_ctl, 1); 373 374 /* 375 * Skip the csum areas. If we don't check crcs then we just have a 376 * 64bit chunk at the front of the first page. 377 */ 378 if (io_ctl->check_crcs) { 379 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages); 380 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages); 381 } else { 382 io_ctl->cur += sizeof(u64); 383 io_ctl->size -= sizeof(u64) * 2; 384 } 385 386 val = io_ctl->cur; 387 *val = cpu_to_le64(generation); 388 io_ctl->cur += sizeof(u64); 389 } 390 391 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation) 392 { 393 __le64 *gen; 394 395 /* 396 * Skip the crc area. If we don't check crcs then we just have a 64bit 397 * chunk at the front of the first page. 398 */ 399 if (io_ctl->check_crcs) { 400 io_ctl->cur += sizeof(u32) * io_ctl->num_pages; 401 io_ctl->size -= sizeof(u64) + 402 (sizeof(u32) * io_ctl->num_pages); 403 } else { 404 io_ctl->cur += sizeof(u64); 405 io_ctl->size -= sizeof(u64) * 2; 406 } 407 408 gen = io_ctl->cur; 409 if (le64_to_cpu(*gen) != generation) { 410 printk_ratelimited(KERN_ERR "btrfs: space cache generation " 411 "(%Lu) does not match inode (%Lu)\n", *gen, 412 generation); 413 io_ctl_unmap_page(io_ctl); 414 return -EIO; 415 } 416 io_ctl->cur += sizeof(u64); 417 return 0; 418 } 419 420 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index) 421 { 422 u32 *tmp; 423 u32 crc = ~(u32)0; 424 unsigned offset = 0; 425 426 if (!io_ctl->check_crcs) { 427 io_ctl_unmap_page(io_ctl); 428 return; 429 } 430 431 if (index == 0) 432 offset = sizeof(u32) * io_ctl->num_pages; 433 434 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, 435 PAGE_CACHE_SIZE - offset); 436 btrfs_csum_final(crc, (char *)&crc); 437 io_ctl_unmap_page(io_ctl); 438 tmp = kmap(io_ctl->pages[0]); 439 tmp += index; 440 *tmp = crc; 441 kunmap(io_ctl->pages[0]); 442 } 443 444 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index) 445 { 446 u32 *tmp, val; 447 u32 crc = ~(u32)0; 448 unsigned offset = 0; 449 450 if (!io_ctl->check_crcs) { 451 io_ctl_map_page(io_ctl, 0); 452 return 0; 453 } 454 455 if (index == 0) 456 offset = sizeof(u32) * io_ctl->num_pages; 457 458 tmp = kmap(io_ctl->pages[0]); 459 tmp += index; 460 val = *tmp; 461 kunmap(io_ctl->pages[0]); 462 463 io_ctl_map_page(io_ctl, 0); 464 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc, 465 PAGE_CACHE_SIZE - offset); 466 btrfs_csum_final(crc, (char *)&crc); 467 if (val != crc) { 468 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free " 469 "space cache\n"); 470 io_ctl_unmap_page(io_ctl); 471 return -EIO; 472 } 473 474 return 0; 475 } 476 477 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes, 478 void *bitmap) 479 { 480 struct btrfs_free_space_entry *entry; 481 482 if (!io_ctl->cur) 483 return -ENOSPC; 484 485 entry = io_ctl->cur; 486 entry->offset = cpu_to_le64(offset); 487 entry->bytes = cpu_to_le64(bytes); 488 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP : 489 BTRFS_FREE_SPACE_EXTENT; 490 io_ctl->cur += sizeof(struct btrfs_free_space_entry); 491 io_ctl->size -= sizeof(struct btrfs_free_space_entry); 492 493 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) 494 return 0; 495 496 io_ctl_set_crc(io_ctl, io_ctl->index - 1); 497 498 /* No more pages to map */ 499 if (io_ctl->index >= io_ctl->num_pages) 500 return 0; 501 502 /* map the next page */ 503 io_ctl_map_page(io_ctl, 1); 504 return 0; 505 } 506 507 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap) 508 { 509 if (!io_ctl->cur) 510 return -ENOSPC; 511 512 /* 513 * If we aren't at the start of the current page, unmap this one and 514 * map the next one if there is any left. 515 */ 516 if (io_ctl->cur != io_ctl->orig) { 517 io_ctl_set_crc(io_ctl, io_ctl->index - 1); 518 if (io_ctl->index >= io_ctl->num_pages) 519 return -ENOSPC; 520 io_ctl_map_page(io_ctl, 0); 521 } 522 523 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE); 524 io_ctl_set_crc(io_ctl, io_ctl->index - 1); 525 if (io_ctl->index < io_ctl->num_pages) 526 io_ctl_map_page(io_ctl, 0); 527 return 0; 528 } 529 530 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl) 531 { 532 /* 533 * If we're not on the boundary we know we've modified the page and we 534 * need to crc the page. 535 */ 536 if (io_ctl->cur != io_ctl->orig) 537 io_ctl_set_crc(io_ctl, io_ctl->index - 1); 538 else 539 io_ctl_unmap_page(io_ctl); 540 541 while (io_ctl->index < io_ctl->num_pages) { 542 io_ctl_map_page(io_ctl, 1); 543 io_ctl_set_crc(io_ctl, io_ctl->index - 1); 544 } 545 } 546 547 static int io_ctl_read_entry(struct io_ctl *io_ctl, 548 struct btrfs_free_space *entry, u8 *type) 549 { 550 struct btrfs_free_space_entry *e; 551 int ret; 552 553 if (!io_ctl->cur) { 554 ret = io_ctl_check_crc(io_ctl, io_ctl->index); 555 if (ret) 556 return ret; 557 } 558 559 e = io_ctl->cur; 560 entry->offset = le64_to_cpu(e->offset); 561 entry->bytes = le64_to_cpu(e->bytes); 562 *type = e->type; 563 io_ctl->cur += sizeof(struct btrfs_free_space_entry); 564 io_ctl->size -= sizeof(struct btrfs_free_space_entry); 565 566 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry)) 567 return 0; 568 569 io_ctl_unmap_page(io_ctl); 570 571 return 0; 572 } 573 574 static int io_ctl_read_bitmap(struct io_ctl *io_ctl, 575 struct btrfs_free_space *entry) 576 { 577 int ret; 578 579 ret = io_ctl_check_crc(io_ctl, io_ctl->index); 580 if (ret) 581 return ret; 582 583 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE); 584 io_ctl_unmap_page(io_ctl); 585 586 return 0; 587 } 588 589 /* 590 * Since we attach pinned extents after the fact we can have contiguous sections 591 * of free space that are split up in entries. This poses a problem with the 592 * tree logging stuff since it could have allocated across what appears to be 2 593 * entries since we would have merged the entries when adding the pinned extents 594 * back to the free space cache. So run through the space cache that we just 595 * loaded and merge contiguous entries. This will make the log replay stuff not 596 * blow up and it will make for nicer allocator behavior. 597 */ 598 static void merge_space_tree(struct btrfs_free_space_ctl *ctl) 599 { 600 struct btrfs_free_space *e, *prev = NULL; 601 struct rb_node *n; 602 603 again: 604 spin_lock(&ctl->tree_lock); 605 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { 606 e = rb_entry(n, struct btrfs_free_space, offset_index); 607 if (!prev) 608 goto next; 609 if (e->bitmap || prev->bitmap) 610 goto next; 611 if (prev->offset + prev->bytes == e->offset) { 612 unlink_free_space(ctl, prev); 613 unlink_free_space(ctl, e); 614 prev->bytes += e->bytes; 615 kmem_cache_free(btrfs_free_space_cachep, e); 616 link_free_space(ctl, prev); 617 prev = NULL; 618 spin_unlock(&ctl->tree_lock); 619 goto again; 620 } 621 next: 622 prev = e; 623 } 624 spin_unlock(&ctl->tree_lock); 625 } 626 627 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode, 628 struct btrfs_free_space_ctl *ctl, 629 struct btrfs_path *path, u64 offset) 630 { 631 struct btrfs_free_space_header *header; 632 struct extent_buffer *leaf; 633 struct io_ctl io_ctl; 634 struct btrfs_key key; 635 struct btrfs_free_space *e, *n; 636 struct list_head bitmaps; 637 u64 num_entries; 638 u64 num_bitmaps; 639 u64 generation; 640 u8 type; 641 int ret = 0; 642 643 INIT_LIST_HEAD(&bitmaps); 644 645 /* Nothing in the space cache, goodbye */ 646 if (!i_size_read(inode)) 647 return 0; 648 649 key.objectid = BTRFS_FREE_SPACE_OBJECTID; 650 key.offset = offset; 651 key.type = 0; 652 653 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 654 if (ret < 0) 655 return 0; 656 else if (ret > 0) { 657 btrfs_release_path(path); 658 return 0; 659 } 660 661 ret = -1; 662 663 leaf = path->nodes[0]; 664 header = btrfs_item_ptr(leaf, path->slots[0], 665 struct btrfs_free_space_header); 666 num_entries = btrfs_free_space_entries(leaf, header); 667 num_bitmaps = btrfs_free_space_bitmaps(leaf, header); 668 generation = btrfs_free_space_generation(leaf, header); 669 btrfs_release_path(path); 670 671 if (BTRFS_I(inode)->generation != generation) { 672 printk(KERN_ERR "btrfs: free space inode generation (%llu) did" 673 " not match free space cache generation (%llu)\n", 674 (unsigned long long)BTRFS_I(inode)->generation, 675 (unsigned long long)generation); 676 return 0; 677 } 678 679 if (!num_entries) 680 return 0; 681 682 ret = io_ctl_init(&io_ctl, inode, root); 683 if (ret) 684 return ret; 685 686 ret = readahead_cache(inode); 687 if (ret) 688 goto out; 689 690 ret = io_ctl_prepare_pages(&io_ctl, inode, 1); 691 if (ret) 692 goto out; 693 694 ret = io_ctl_check_crc(&io_ctl, 0); 695 if (ret) 696 goto free_cache; 697 698 ret = io_ctl_check_generation(&io_ctl, generation); 699 if (ret) 700 goto free_cache; 701 702 while (num_entries) { 703 e = kmem_cache_zalloc(btrfs_free_space_cachep, 704 GFP_NOFS); 705 if (!e) 706 goto free_cache; 707 708 ret = io_ctl_read_entry(&io_ctl, e, &type); 709 if (ret) { 710 kmem_cache_free(btrfs_free_space_cachep, e); 711 goto free_cache; 712 } 713 714 if (!e->bytes) { 715 kmem_cache_free(btrfs_free_space_cachep, e); 716 goto free_cache; 717 } 718 719 if (type == BTRFS_FREE_SPACE_EXTENT) { 720 spin_lock(&ctl->tree_lock); 721 ret = link_free_space(ctl, e); 722 spin_unlock(&ctl->tree_lock); 723 if (ret) { 724 printk(KERN_ERR "Duplicate entries in " 725 "free space cache, dumping\n"); 726 kmem_cache_free(btrfs_free_space_cachep, e); 727 goto free_cache; 728 } 729 } else { 730 BUG_ON(!num_bitmaps); 731 num_bitmaps--; 732 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); 733 if (!e->bitmap) { 734 kmem_cache_free( 735 btrfs_free_space_cachep, e); 736 goto free_cache; 737 } 738 spin_lock(&ctl->tree_lock); 739 ret = link_free_space(ctl, e); 740 ctl->total_bitmaps++; 741 ctl->op->recalc_thresholds(ctl); 742 spin_unlock(&ctl->tree_lock); 743 if (ret) { 744 printk(KERN_ERR "Duplicate entries in " 745 "free space cache, dumping\n"); 746 kmem_cache_free(btrfs_free_space_cachep, e); 747 goto free_cache; 748 } 749 list_add_tail(&e->list, &bitmaps); 750 } 751 752 num_entries--; 753 } 754 755 io_ctl_unmap_page(&io_ctl); 756 757 /* 758 * We add the bitmaps at the end of the entries in order that 759 * the bitmap entries are added to the cache. 760 */ 761 list_for_each_entry_safe(e, n, &bitmaps, list) { 762 list_del_init(&e->list); 763 ret = io_ctl_read_bitmap(&io_ctl, e); 764 if (ret) 765 goto free_cache; 766 } 767 768 io_ctl_drop_pages(&io_ctl); 769 merge_space_tree(ctl); 770 ret = 1; 771 out: 772 io_ctl_free(&io_ctl); 773 return ret; 774 free_cache: 775 io_ctl_drop_pages(&io_ctl); 776 __btrfs_remove_free_space_cache(ctl); 777 goto out; 778 } 779 780 int load_free_space_cache(struct btrfs_fs_info *fs_info, 781 struct btrfs_block_group_cache *block_group) 782 { 783 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 784 struct btrfs_root *root = fs_info->tree_root; 785 struct inode *inode; 786 struct btrfs_path *path; 787 int ret = 0; 788 bool matched; 789 u64 used = btrfs_block_group_used(&block_group->item); 790 791 /* 792 * If this block group has been marked to be cleared for one reason or 793 * another then we can't trust the on disk cache, so just return. 794 */ 795 spin_lock(&block_group->lock); 796 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { 797 spin_unlock(&block_group->lock); 798 return 0; 799 } 800 spin_unlock(&block_group->lock); 801 802 path = btrfs_alloc_path(); 803 if (!path) 804 return 0; 805 path->search_commit_root = 1; 806 path->skip_locking = 1; 807 808 inode = lookup_free_space_inode(root, block_group, path); 809 if (IS_ERR(inode)) { 810 btrfs_free_path(path); 811 return 0; 812 } 813 814 /* We may have converted the inode and made the cache invalid. */ 815 spin_lock(&block_group->lock); 816 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { 817 spin_unlock(&block_group->lock); 818 btrfs_free_path(path); 819 goto out; 820 } 821 spin_unlock(&block_group->lock); 822 823 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl, 824 path, block_group->key.objectid); 825 btrfs_free_path(path); 826 if (ret <= 0) 827 goto out; 828 829 spin_lock(&ctl->tree_lock); 830 matched = (ctl->free_space == (block_group->key.offset - used - 831 block_group->bytes_super)); 832 spin_unlock(&ctl->tree_lock); 833 834 if (!matched) { 835 __btrfs_remove_free_space_cache(ctl); 836 printk(KERN_ERR "block group %llu has an wrong amount of free " 837 "space\n", block_group->key.objectid); 838 ret = -1; 839 } 840 out: 841 if (ret < 0) { 842 /* This cache is bogus, make sure it gets cleared */ 843 spin_lock(&block_group->lock); 844 block_group->disk_cache_state = BTRFS_DC_CLEAR; 845 spin_unlock(&block_group->lock); 846 ret = 0; 847 848 printk(KERN_ERR "btrfs: failed to load free space cache " 849 "for block group %llu\n", block_group->key.objectid); 850 } 851 852 iput(inode); 853 return ret; 854 } 855 856 /** 857 * __btrfs_write_out_cache - write out cached info to an inode 858 * @root - the root the inode belongs to 859 * @ctl - the free space cache we are going to write out 860 * @block_group - the block_group for this cache if it belongs to a block_group 861 * @trans - the trans handle 862 * @path - the path to use 863 * @offset - the offset for the key we'll insert 864 * 865 * This function writes out a free space cache struct to disk for quick recovery 866 * on mount. This will return 0 if it was successfull in writing the cache out, 867 * and -1 if it was not. 868 */ 869 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, 870 struct btrfs_free_space_ctl *ctl, 871 struct btrfs_block_group_cache *block_group, 872 struct btrfs_trans_handle *trans, 873 struct btrfs_path *path, u64 offset) 874 { 875 struct btrfs_free_space_header *header; 876 struct extent_buffer *leaf; 877 struct rb_node *node; 878 struct list_head *pos, *n; 879 struct extent_state *cached_state = NULL; 880 struct btrfs_free_cluster *cluster = NULL; 881 struct extent_io_tree *unpin = NULL; 882 struct io_ctl io_ctl; 883 struct list_head bitmap_list; 884 struct btrfs_key key; 885 u64 start, extent_start, extent_end, len; 886 int entries = 0; 887 int bitmaps = 0; 888 int ret; 889 int err = -1; 890 891 INIT_LIST_HEAD(&bitmap_list); 892 893 if (!i_size_read(inode)) 894 return -1; 895 896 ret = io_ctl_init(&io_ctl, inode, root); 897 if (ret) 898 return -1; 899 900 /* Get the cluster for this block_group if it exists */ 901 if (block_group && !list_empty(&block_group->cluster_list)) 902 cluster = list_entry(block_group->cluster_list.next, 903 struct btrfs_free_cluster, 904 block_group_list); 905 906 /* Lock all pages first so we can lock the extent safely. */ 907 io_ctl_prepare_pages(&io_ctl, inode, 0); 908 909 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, 910 0, &cached_state); 911 912 node = rb_first(&ctl->free_space_offset); 913 if (!node && cluster) { 914 node = rb_first(&cluster->root); 915 cluster = NULL; 916 } 917 918 /* Make sure we can fit our crcs into the first page */ 919 if (io_ctl.check_crcs && 920 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) { 921 WARN_ON(1); 922 goto out_nospc; 923 } 924 925 io_ctl_set_generation(&io_ctl, trans->transid); 926 927 /* Write out the extent entries */ 928 while (node) { 929 struct btrfs_free_space *e; 930 931 e = rb_entry(node, struct btrfs_free_space, offset_index); 932 entries++; 933 934 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes, 935 e->bitmap); 936 if (ret) 937 goto out_nospc; 938 939 if (e->bitmap) { 940 list_add_tail(&e->list, &bitmap_list); 941 bitmaps++; 942 } 943 node = rb_next(node); 944 if (!node && cluster) { 945 node = rb_first(&cluster->root); 946 cluster = NULL; 947 } 948 } 949 950 /* 951 * We want to add any pinned extents to our free space cache 952 * so we don't leak the space 953 */ 954 955 /* 956 * We shouldn't have switched the pinned extents yet so this is the 957 * right one 958 */ 959 unpin = root->fs_info->pinned_extents; 960 961 if (block_group) 962 start = block_group->key.objectid; 963 964 while (block_group && (start < block_group->key.objectid + 965 block_group->key.offset)) { 966 ret = find_first_extent_bit(unpin, start, 967 &extent_start, &extent_end, 968 EXTENT_DIRTY, NULL); 969 if (ret) { 970 ret = 0; 971 break; 972 } 973 974 /* This pinned extent is out of our range */ 975 if (extent_start >= block_group->key.objectid + 976 block_group->key.offset) 977 break; 978 979 extent_start = max(extent_start, start); 980 extent_end = min(block_group->key.objectid + 981 block_group->key.offset, extent_end + 1); 982 len = extent_end - extent_start; 983 984 entries++; 985 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL); 986 if (ret) 987 goto out_nospc; 988 989 start = extent_end; 990 } 991 992 /* Write out the bitmaps */ 993 list_for_each_safe(pos, n, &bitmap_list) { 994 struct btrfs_free_space *entry = 995 list_entry(pos, struct btrfs_free_space, list); 996 997 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap); 998 if (ret) 999 goto out_nospc; 1000 list_del_init(&entry->list); 1001 } 1002 1003 /* Zero out the rest of the pages just to make sure */ 1004 io_ctl_zero_remaining_pages(&io_ctl); 1005 1006 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages, 1007 0, i_size_read(inode), &cached_state); 1008 io_ctl_drop_pages(&io_ctl); 1009 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, 1010 i_size_read(inode) - 1, &cached_state, GFP_NOFS); 1011 1012 if (ret) 1013 goto out; 1014 1015 1016 btrfs_wait_ordered_range(inode, 0, (u64)-1); 1017 1018 key.objectid = BTRFS_FREE_SPACE_OBJECTID; 1019 key.offset = offset; 1020 key.type = 0; 1021 1022 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 1023 if (ret < 0) { 1024 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, 1025 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, 1026 GFP_NOFS); 1027 goto out; 1028 } 1029 leaf = path->nodes[0]; 1030 if (ret > 0) { 1031 struct btrfs_key found_key; 1032 BUG_ON(!path->slots[0]); 1033 path->slots[0]--; 1034 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1035 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || 1036 found_key.offset != offset) { 1037 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, 1038 inode->i_size - 1, 1039 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, 1040 NULL, GFP_NOFS); 1041 btrfs_release_path(path); 1042 goto out; 1043 } 1044 } 1045 1046 BTRFS_I(inode)->generation = trans->transid; 1047 header = btrfs_item_ptr(leaf, path->slots[0], 1048 struct btrfs_free_space_header); 1049 btrfs_set_free_space_entries(leaf, header, entries); 1050 btrfs_set_free_space_bitmaps(leaf, header, bitmaps); 1051 btrfs_set_free_space_generation(leaf, header, trans->transid); 1052 btrfs_mark_buffer_dirty(leaf); 1053 btrfs_release_path(path); 1054 1055 err = 0; 1056 out: 1057 io_ctl_free(&io_ctl); 1058 if (err) { 1059 invalidate_inode_pages2(inode->i_mapping); 1060 BTRFS_I(inode)->generation = 0; 1061 } 1062 btrfs_update_inode(trans, root, inode); 1063 return err; 1064 1065 out_nospc: 1066 list_for_each_safe(pos, n, &bitmap_list) { 1067 struct btrfs_free_space *entry = 1068 list_entry(pos, struct btrfs_free_space, list); 1069 list_del_init(&entry->list); 1070 } 1071 io_ctl_drop_pages(&io_ctl); 1072 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, 1073 i_size_read(inode) - 1, &cached_state, GFP_NOFS); 1074 goto out; 1075 } 1076 1077 int btrfs_write_out_cache(struct btrfs_root *root, 1078 struct btrfs_trans_handle *trans, 1079 struct btrfs_block_group_cache *block_group, 1080 struct btrfs_path *path) 1081 { 1082 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 1083 struct inode *inode; 1084 int ret = 0; 1085 1086 root = root->fs_info->tree_root; 1087 1088 spin_lock(&block_group->lock); 1089 if (block_group->disk_cache_state < BTRFS_DC_SETUP) { 1090 spin_unlock(&block_group->lock); 1091 return 0; 1092 } 1093 spin_unlock(&block_group->lock); 1094 1095 inode = lookup_free_space_inode(root, block_group, path); 1096 if (IS_ERR(inode)) 1097 return 0; 1098 1099 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans, 1100 path, block_group->key.objectid); 1101 if (ret) { 1102 spin_lock(&block_group->lock); 1103 block_group->disk_cache_state = BTRFS_DC_ERROR; 1104 spin_unlock(&block_group->lock); 1105 ret = 0; 1106 #ifdef DEBUG 1107 printk(KERN_ERR "btrfs: failed to write free space cache " 1108 "for block group %llu\n", block_group->key.objectid); 1109 #endif 1110 } 1111 1112 iput(inode); 1113 return ret; 1114 } 1115 1116 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit, 1117 u64 offset) 1118 { 1119 BUG_ON(offset < bitmap_start); 1120 offset -= bitmap_start; 1121 return (unsigned long)(div_u64(offset, unit)); 1122 } 1123 1124 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit) 1125 { 1126 return (unsigned long)(div_u64(bytes, unit)); 1127 } 1128 1129 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl, 1130 u64 offset) 1131 { 1132 u64 bitmap_start; 1133 u64 bytes_per_bitmap; 1134 1135 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit; 1136 bitmap_start = offset - ctl->start; 1137 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); 1138 bitmap_start *= bytes_per_bitmap; 1139 bitmap_start += ctl->start; 1140 1141 return bitmap_start; 1142 } 1143 1144 static int tree_insert_offset(struct rb_root *root, u64 offset, 1145 struct rb_node *node, int bitmap) 1146 { 1147 struct rb_node **p = &root->rb_node; 1148 struct rb_node *parent = NULL; 1149 struct btrfs_free_space *info; 1150 1151 while (*p) { 1152 parent = *p; 1153 info = rb_entry(parent, struct btrfs_free_space, offset_index); 1154 1155 if (offset < info->offset) { 1156 p = &(*p)->rb_left; 1157 } else if (offset > info->offset) { 1158 p = &(*p)->rb_right; 1159 } else { 1160 /* 1161 * we could have a bitmap entry and an extent entry 1162 * share the same offset. If this is the case, we want 1163 * the extent entry to always be found first if we do a 1164 * linear search through the tree, since we want to have 1165 * the quickest allocation time, and allocating from an 1166 * extent is faster than allocating from a bitmap. So 1167 * if we're inserting a bitmap and we find an entry at 1168 * this offset, we want to go right, or after this entry 1169 * logically. If we are inserting an extent and we've 1170 * found a bitmap, we want to go left, or before 1171 * logically. 1172 */ 1173 if (bitmap) { 1174 if (info->bitmap) { 1175 WARN_ON_ONCE(1); 1176 return -EEXIST; 1177 } 1178 p = &(*p)->rb_right; 1179 } else { 1180 if (!info->bitmap) { 1181 WARN_ON_ONCE(1); 1182 return -EEXIST; 1183 } 1184 p = &(*p)->rb_left; 1185 } 1186 } 1187 } 1188 1189 rb_link_node(node, parent, p); 1190 rb_insert_color(node, root); 1191 1192 return 0; 1193 } 1194 1195 /* 1196 * searches the tree for the given offset. 1197 * 1198 * fuzzy - If this is set, then we are trying to make an allocation, and we just 1199 * want a section that has at least bytes size and comes at or after the given 1200 * offset. 1201 */ 1202 static struct btrfs_free_space * 1203 tree_search_offset(struct btrfs_free_space_ctl *ctl, 1204 u64 offset, int bitmap_only, int fuzzy) 1205 { 1206 struct rb_node *n = ctl->free_space_offset.rb_node; 1207 struct btrfs_free_space *entry, *prev = NULL; 1208 1209 /* find entry that is closest to the 'offset' */ 1210 while (1) { 1211 if (!n) { 1212 entry = NULL; 1213 break; 1214 } 1215 1216 entry = rb_entry(n, struct btrfs_free_space, offset_index); 1217 prev = entry; 1218 1219 if (offset < entry->offset) 1220 n = n->rb_left; 1221 else if (offset > entry->offset) 1222 n = n->rb_right; 1223 else 1224 break; 1225 } 1226 1227 if (bitmap_only) { 1228 if (!entry) 1229 return NULL; 1230 if (entry->bitmap) 1231 return entry; 1232 1233 /* 1234 * bitmap entry and extent entry may share same offset, 1235 * in that case, bitmap entry comes after extent entry. 1236 */ 1237 n = rb_next(n); 1238 if (!n) 1239 return NULL; 1240 entry = rb_entry(n, struct btrfs_free_space, offset_index); 1241 if (entry->offset != offset) 1242 return NULL; 1243 1244 WARN_ON(!entry->bitmap); 1245 return entry; 1246 } else if (entry) { 1247 if (entry->bitmap) { 1248 /* 1249 * if previous extent entry covers the offset, 1250 * we should return it instead of the bitmap entry 1251 */ 1252 n = rb_prev(&entry->offset_index); 1253 if (n) { 1254 prev = rb_entry(n, struct btrfs_free_space, 1255 offset_index); 1256 if (!prev->bitmap && 1257 prev->offset + prev->bytes > offset) 1258 entry = prev; 1259 } 1260 } 1261 return entry; 1262 } 1263 1264 if (!prev) 1265 return NULL; 1266 1267 /* find last entry before the 'offset' */ 1268 entry = prev; 1269 if (entry->offset > offset) { 1270 n = rb_prev(&entry->offset_index); 1271 if (n) { 1272 entry = rb_entry(n, struct btrfs_free_space, 1273 offset_index); 1274 BUG_ON(entry->offset > offset); 1275 } else { 1276 if (fuzzy) 1277 return entry; 1278 else 1279 return NULL; 1280 } 1281 } 1282 1283 if (entry->bitmap) { 1284 n = rb_prev(&entry->offset_index); 1285 if (n) { 1286 prev = rb_entry(n, struct btrfs_free_space, 1287 offset_index); 1288 if (!prev->bitmap && 1289 prev->offset + prev->bytes > offset) 1290 return prev; 1291 } 1292 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset) 1293 return entry; 1294 } else if (entry->offset + entry->bytes > offset) 1295 return entry; 1296 1297 if (!fuzzy) 1298 return NULL; 1299 1300 while (1) { 1301 if (entry->bitmap) { 1302 if (entry->offset + BITS_PER_BITMAP * 1303 ctl->unit > offset) 1304 break; 1305 } else { 1306 if (entry->offset + entry->bytes > offset) 1307 break; 1308 } 1309 1310 n = rb_next(&entry->offset_index); 1311 if (!n) 1312 return NULL; 1313 entry = rb_entry(n, struct btrfs_free_space, offset_index); 1314 } 1315 return entry; 1316 } 1317 1318 static inline void 1319 __unlink_free_space(struct btrfs_free_space_ctl *ctl, 1320 struct btrfs_free_space *info) 1321 { 1322 rb_erase(&info->offset_index, &ctl->free_space_offset); 1323 ctl->free_extents--; 1324 } 1325 1326 static void unlink_free_space(struct btrfs_free_space_ctl *ctl, 1327 struct btrfs_free_space *info) 1328 { 1329 __unlink_free_space(ctl, info); 1330 ctl->free_space -= info->bytes; 1331 } 1332 1333 static int link_free_space(struct btrfs_free_space_ctl *ctl, 1334 struct btrfs_free_space *info) 1335 { 1336 int ret = 0; 1337 1338 BUG_ON(!info->bitmap && !info->bytes); 1339 ret = tree_insert_offset(&ctl->free_space_offset, info->offset, 1340 &info->offset_index, (info->bitmap != NULL)); 1341 if (ret) 1342 return ret; 1343 1344 ctl->free_space += info->bytes; 1345 ctl->free_extents++; 1346 return ret; 1347 } 1348 1349 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl) 1350 { 1351 struct btrfs_block_group_cache *block_group = ctl->private; 1352 u64 max_bytes; 1353 u64 bitmap_bytes; 1354 u64 extent_bytes; 1355 u64 size = block_group->key.offset; 1356 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit; 1357 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg); 1358 1359 BUG_ON(ctl->total_bitmaps > max_bitmaps); 1360 1361 /* 1362 * The goal is to keep the total amount of memory used per 1gb of space 1363 * at or below 32k, so we need to adjust how much memory we allow to be 1364 * used by extent based free space tracking 1365 */ 1366 if (size < 1024 * 1024 * 1024) 1367 max_bytes = MAX_CACHE_BYTES_PER_GIG; 1368 else 1369 max_bytes = MAX_CACHE_BYTES_PER_GIG * 1370 div64_u64(size, 1024 * 1024 * 1024); 1371 1372 /* 1373 * we want to account for 1 more bitmap than what we have so we can make 1374 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as 1375 * we add more bitmaps. 1376 */ 1377 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE; 1378 1379 if (bitmap_bytes >= max_bytes) { 1380 ctl->extents_thresh = 0; 1381 return; 1382 } 1383 1384 /* 1385 * we want the extent entry threshold to always be at most 1/2 the maxw 1386 * bytes we can have, or whatever is less than that. 1387 */ 1388 extent_bytes = max_bytes - bitmap_bytes; 1389 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); 1390 1391 ctl->extents_thresh = 1392 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); 1393 } 1394 1395 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, 1396 struct btrfs_free_space *info, 1397 u64 offset, u64 bytes) 1398 { 1399 unsigned long start, count; 1400 1401 start = offset_to_bit(info->offset, ctl->unit, offset); 1402 count = bytes_to_bits(bytes, ctl->unit); 1403 BUG_ON(start + count > BITS_PER_BITMAP); 1404 1405 bitmap_clear(info->bitmap, start, count); 1406 1407 info->bytes -= bytes; 1408 } 1409 1410 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl, 1411 struct btrfs_free_space *info, u64 offset, 1412 u64 bytes) 1413 { 1414 __bitmap_clear_bits(ctl, info, offset, bytes); 1415 ctl->free_space -= bytes; 1416 } 1417 1418 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl, 1419 struct btrfs_free_space *info, u64 offset, 1420 u64 bytes) 1421 { 1422 unsigned long start, count; 1423 1424 start = offset_to_bit(info->offset, ctl->unit, offset); 1425 count = bytes_to_bits(bytes, ctl->unit); 1426 BUG_ON(start + count > BITS_PER_BITMAP); 1427 1428 bitmap_set(info->bitmap, start, count); 1429 1430 info->bytes += bytes; 1431 ctl->free_space += bytes; 1432 } 1433 1434 static int search_bitmap(struct btrfs_free_space_ctl *ctl, 1435 struct btrfs_free_space *bitmap_info, u64 *offset, 1436 u64 *bytes) 1437 { 1438 unsigned long found_bits = 0; 1439 unsigned long bits, i; 1440 unsigned long next_zero; 1441 1442 i = offset_to_bit(bitmap_info->offset, ctl->unit, 1443 max_t(u64, *offset, bitmap_info->offset)); 1444 bits = bytes_to_bits(*bytes, ctl->unit); 1445 1446 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) { 1447 next_zero = find_next_zero_bit(bitmap_info->bitmap, 1448 BITS_PER_BITMAP, i); 1449 if ((next_zero - i) >= bits) { 1450 found_bits = next_zero - i; 1451 break; 1452 } 1453 i = next_zero; 1454 } 1455 1456 if (found_bits) { 1457 *offset = (u64)(i * ctl->unit) + bitmap_info->offset; 1458 *bytes = (u64)(found_bits) * ctl->unit; 1459 return 0; 1460 } 1461 1462 return -1; 1463 } 1464 1465 static struct btrfs_free_space * 1466 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes) 1467 { 1468 struct btrfs_free_space *entry; 1469 struct rb_node *node; 1470 int ret; 1471 1472 if (!ctl->free_space_offset.rb_node) 1473 return NULL; 1474 1475 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1); 1476 if (!entry) 1477 return NULL; 1478 1479 for (node = &entry->offset_index; node; node = rb_next(node)) { 1480 entry = rb_entry(node, struct btrfs_free_space, offset_index); 1481 if (entry->bytes < *bytes) 1482 continue; 1483 1484 if (entry->bitmap) { 1485 ret = search_bitmap(ctl, entry, offset, bytes); 1486 if (!ret) 1487 return entry; 1488 continue; 1489 } 1490 1491 *offset = entry->offset; 1492 *bytes = entry->bytes; 1493 return entry; 1494 } 1495 1496 return NULL; 1497 } 1498 1499 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl, 1500 struct btrfs_free_space *info, u64 offset) 1501 { 1502 info->offset = offset_to_bitmap(ctl, offset); 1503 info->bytes = 0; 1504 INIT_LIST_HEAD(&info->list); 1505 link_free_space(ctl, info); 1506 ctl->total_bitmaps++; 1507 1508 ctl->op->recalc_thresholds(ctl); 1509 } 1510 1511 static void free_bitmap(struct btrfs_free_space_ctl *ctl, 1512 struct btrfs_free_space *bitmap_info) 1513 { 1514 unlink_free_space(ctl, bitmap_info); 1515 kfree(bitmap_info->bitmap); 1516 kmem_cache_free(btrfs_free_space_cachep, bitmap_info); 1517 ctl->total_bitmaps--; 1518 ctl->op->recalc_thresholds(ctl); 1519 } 1520 1521 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl, 1522 struct btrfs_free_space *bitmap_info, 1523 u64 *offset, u64 *bytes) 1524 { 1525 u64 end; 1526 u64 search_start, search_bytes; 1527 int ret; 1528 1529 again: 1530 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1; 1531 1532 /* 1533 * We need to search for bits in this bitmap. We could only cover some 1534 * of the extent in this bitmap thanks to how we add space, so we need 1535 * to search for as much as it as we can and clear that amount, and then 1536 * go searching for the next bit. 1537 */ 1538 search_start = *offset; 1539 search_bytes = ctl->unit; 1540 search_bytes = min(search_bytes, end - search_start + 1); 1541 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes); 1542 BUG_ON(ret < 0 || search_start != *offset); 1543 1544 /* We may have found more bits than what we need */ 1545 search_bytes = min(search_bytes, *bytes); 1546 1547 /* Cannot clear past the end of the bitmap */ 1548 search_bytes = min(search_bytes, end - search_start + 1); 1549 1550 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes); 1551 *offset += search_bytes; 1552 *bytes -= search_bytes; 1553 1554 if (*bytes) { 1555 struct rb_node *next = rb_next(&bitmap_info->offset_index); 1556 if (!bitmap_info->bytes) 1557 free_bitmap(ctl, bitmap_info); 1558 1559 /* 1560 * no entry after this bitmap, but we still have bytes to 1561 * remove, so something has gone wrong. 1562 */ 1563 if (!next) 1564 return -EINVAL; 1565 1566 bitmap_info = rb_entry(next, struct btrfs_free_space, 1567 offset_index); 1568 1569 /* 1570 * if the next entry isn't a bitmap we need to return to let the 1571 * extent stuff do its work. 1572 */ 1573 if (!bitmap_info->bitmap) 1574 return -EAGAIN; 1575 1576 /* 1577 * Ok the next item is a bitmap, but it may not actually hold 1578 * the information for the rest of this free space stuff, so 1579 * look for it, and if we don't find it return so we can try 1580 * everything over again. 1581 */ 1582 search_start = *offset; 1583 search_bytes = ctl->unit; 1584 ret = search_bitmap(ctl, bitmap_info, &search_start, 1585 &search_bytes); 1586 if (ret < 0 || search_start != *offset) 1587 return -EAGAIN; 1588 1589 goto again; 1590 } else if (!bitmap_info->bytes) 1591 free_bitmap(ctl, bitmap_info); 1592 1593 return 0; 1594 } 1595 1596 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl, 1597 struct btrfs_free_space *info, u64 offset, 1598 u64 bytes) 1599 { 1600 u64 bytes_to_set = 0; 1601 u64 end; 1602 1603 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit); 1604 1605 bytes_to_set = min(end - offset, bytes); 1606 1607 bitmap_set_bits(ctl, info, offset, bytes_to_set); 1608 1609 return bytes_to_set; 1610 1611 } 1612 1613 static bool use_bitmap(struct btrfs_free_space_ctl *ctl, 1614 struct btrfs_free_space *info) 1615 { 1616 struct btrfs_block_group_cache *block_group = ctl->private; 1617 1618 /* 1619 * If we are below the extents threshold then we can add this as an 1620 * extent, and don't have to deal with the bitmap 1621 */ 1622 if (ctl->free_extents < ctl->extents_thresh) { 1623 /* 1624 * If this block group has some small extents we don't want to 1625 * use up all of our free slots in the cache with them, we want 1626 * to reserve them to larger extents, however if we have plent 1627 * of cache left then go ahead an dadd them, no sense in adding 1628 * the overhead of a bitmap if we don't have to. 1629 */ 1630 if (info->bytes <= block_group->sectorsize * 4) { 1631 if (ctl->free_extents * 2 <= ctl->extents_thresh) 1632 return false; 1633 } else { 1634 return false; 1635 } 1636 } 1637 1638 /* 1639 * some block groups are so tiny they can't be enveloped by a bitmap, so 1640 * don't even bother to create a bitmap for this 1641 */ 1642 if (BITS_PER_BITMAP * ctl->unit > block_group->key.offset) 1643 return false; 1644 1645 return true; 1646 } 1647 1648 static struct btrfs_free_space_op free_space_op = { 1649 .recalc_thresholds = recalculate_thresholds, 1650 .use_bitmap = use_bitmap, 1651 }; 1652 1653 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl, 1654 struct btrfs_free_space *info) 1655 { 1656 struct btrfs_free_space *bitmap_info; 1657 struct btrfs_block_group_cache *block_group = NULL; 1658 int added = 0; 1659 u64 bytes, offset, bytes_added; 1660 int ret; 1661 1662 bytes = info->bytes; 1663 offset = info->offset; 1664 1665 if (!ctl->op->use_bitmap(ctl, info)) 1666 return 0; 1667 1668 if (ctl->op == &free_space_op) 1669 block_group = ctl->private; 1670 again: 1671 /* 1672 * Since we link bitmaps right into the cluster we need to see if we 1673 * have a cluster here, and if so and it has our bitmap we need to add 1674 * the free space to that bitmap. 1675 */ 1676 if (block_group && !list_empty(&block_group->cluster_list)) { 1677 struct btrfs_free_cluster *cluster; 1678 struct rb_node *node; 1679 struct btrfs_free_space *entry; 1680 1681 cluster = list_entry(block_group->cluster_list.next, 1682 struct btrfs_free_cluster, 1683 block_group_list); 1684 spin_lock(&cluster->lock); 1685 node = rb_first(&cluster->root); 1686 if (!node) { 1687 spin_unlock(&cluster->lock); 1688 goto no_cluster_bitmap; 1689 } 1690 1691 entry = rb_entry(node, struct btrfs_free_space, offset_index); 1692 if (!entry->bitmap) { 1693 spin_unlock(&cluster->lock); 1694 goto no_cluster_bitmap; 1695 } 1696 1697 if (entry->offset == offset_to_bitmap(ctl, offset)) { 1698 bytes_added = add_bytes_to_bitmap(ctl, entry, 1699 offset, bytes); 1700 bytes -= bytes_added; 1701 offset += bytes_added; 1702 } 1703 spin_unlock(&cluster->lock); 1704 if (!bytes) { 1705 ret = 1; 1706 goto out; 1707 } 1708 } 1709 1710 no_cluster_bitmap: 1711 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 1712 1, 0); 1713 if (!bitmap_info) { 1714 BUG_ON(added); 1715 goto new_bitmap; 1716 } 1717 1718 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes); 1719 bytes -= bytes_added; 1720 offset += bytes_added; 1721 added = 0; 1722 1723 if (!bytes) { 1724 ret = 1; 1725 goto out; 1726 } else 1727 goto again; 1728 1729 new_bitmap: 1730 if (info && info->bitmap) { 1731 add_new_bitmap(ctl, info, offset); 1732 added = 1; 1733 info = NULL; 1734 goto again; 1735 } else { 1736 spin_unlock(&ctl->tree_lock); 1737 1738 /* no pre-allocated info, allocate a new one */ 1739 if (!info) { 1740 info = kmem_cache_zalloc(btrfs_free_space_cachep, 1741 GFP_NOFS); 1742 if (!info) { 1743 spin_lock(&ctl->tree_lock); 1744 ret = -ENOMEM; 1745 goto out; 1746 } 1747 } 1748 1749 /* allocate the bitmap */ 1750 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); 1751 spin_lock(&ctl->tree_lock); 1752 if (!info->bitmap) { 1753 ret = -ENOMEM; 1754 goto out; 1755 } 1756 goto again; 1757 } 1758 1759 out: 1760 if (info) { 1761 if (info->bitmap) 1762 kfree(info->bitmap); 1763 kmem_cache_free(btrfs_free_space_cachep, info); 1764 } 1765 1766 return ret; 1767 } 1768 1769 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl, 1770 struct btrfs_free_space *info, bool update_stat) 1771 { 1772 struct btrfs_free_space *left_info; 1773 struct btrfs_free_space *right_info; 1774 bool merged = false; 1775 u64 offset = info->offset; 1776 u64 bytes = info->bytes; 1777 1778 /* 1779 * first we want to see if there is free space adjacent to the range we 1780 * are adding, if there is remove that struct and add a new one to 1781 * cover the entire range 1782 */ 1783 right_info = tree_search_offset(ctl, offset + bytes, 0, 0); 1784 if (right_info && rb_prev(&right_info->offset_index)) 1785 left_info = rb_entry(rb_prev(&right_info->offset_index), 1786 struct btrfs_free_space, offset_index); 1787 else 1788 left_info = tree_search_offset(ctl, offset - 1, 0, 0); 1789 1790 if (right_info && !right_info->bitmap) { 1791 if (update_stat) 1792 unlink_free_space(ctl, right_info); 1793 else 1794 __unlink_free_space(ctl, right_info); 1795 info->bytes += right_info->bytes; 1796 kmem_cache_free(btrfs_free_space_cachep, right_info); 1797 merged = true; 1798 } 1799 1800 if (left_info && !left_info->bitmap && 1801 left_info->offset + left_info->bytes == offset) { 1802 if (update_stat) 1803 unlink_free_space(ctl, left_info); 1804 else 1805 __unlink_free_space(ctl, left_info); 1806 info->offset = left_info->offset; 1807 info->bytes += left_info->bytes; 1808 kmem_cache_free(btrfs_free_space_cachep, left_info); 1809 merged = true; 1810 } 1811 1812 return merged; 1813 } 1814 1815 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl, 1816 u64 offset, u64 bytes) 1817 { 1818 struct btrfs_free_space *info; 1819 int ret = 0; 1820 1821 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS); 1822 if (!info) 1823 return -ENOMEM; 1824 1825 info->offset = offset; 1826 info->bytes = bytes; 1827 1828 spin_lock(&ctl->tree_lock); 1829 1830 if (try_merge_free_space(ctl, info, true)) 1831 goto link; 1832 1833 /* 1834 * There was no extent directly to the left or right of this new 1835 * extent then we know we're going to have to allocate a new extent, so 1836 * before we do that see if we need to drop this into a bitmap 1837 */ 1838 ret = insert_into_bitmap(ctl, info); 1839 if (ret < 0) { 1840 goto out; 1841 } else if (ret) { 1842 ret = 0; 1843 goto out; 1844 } 1845 link: 1846 ret = link_free_space(ctl, info); 1847 if (ret) 1848 kmem_cache_free(btrfs_free_space_cachep, info); 1849 out: 1850 spin_unlock(&ctl->tree_lock); 1851 1852 if (ret) { 1853 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); 1854 BUG_ON(ret == -EEXIST); 1855 } 1856 1857 return ret; 1858 } 1859 1860 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, 1861 u64 offset, u64 bytes) 1862 { 1863 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 1864 struct btrfs_free_space *info; 1865 int ret = 0; 1866 1867 spin_lock(&ctl->tree_lock); 1868 1869 again: 1870 if (!bytes) 1871 goto out_lock; 1872 1873 info = tree_search_offset(ctl, offset, 0, 0); 1874 if (!info) { 1875 /* 1876 * oops didn't find an extent that matched the space we wanted 1877 * to remove, look for a bitmap instead 1878 */ 1879 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset), 1880 1, 0); 1881 if (!info) { 1882 /* the tree logging code might be calling us before we 1883 * have fully loaded the free space rbtree for this 1884 * block group. So it is possible the entry won't 1885 * be in the rbtree yet at all. The caching code 1886 * will make sure not to put it in the rbtree if 1887 * the logging code has pinned it. 1888 */ 1889 goto out_lock; 1890 } 1891 } 1892 1893 if (!info->bitmap) { 1894 unlink_free_space(ctl, info); 1895 if (offset == info->offset) { 1896 u64 to_free = min(bytes, info->bytes); 1897 1898 info->bytes -= to_free; 1899 info->offset += to_free; 1900 if (info->bytes) { 1901 ret = link_free_space(ctl, info); 1902 WARN_ON(ret); 1903 } else { 1904 kmem_cache_free(btrfs_free_space_cachep, info); 1905 } 1906 1907 offset += to_free; 1908 bytes -= to_free; 1909 goto again; 1910 } else { 1911 u64 old_end = info->bytes + info->offset; 1912 1913 info->bytes = offset - info->offset; 1914 ret = link_free_space(ctl, info); 1915 WARN_ON(ret); 1916 if (ret) 1917 goto out_lock; 1918 1919 /* Not enough bytes in this entry to satisfy us */ 1920 if (old_end < offset + bytes) { 1921 bytes -= old_end - offset; 1922 offset = old_end; 1923 goto again; 1924 } else if (old_end == offset + bytes) { 1925 /* all done */ 1926 goto out_lock; 1927 } 1928 spin_unlock(&ctl->tree_lock); 1929 1930 ret = btrfs_add_free_space(block_group, offset + bytes, 1931 old_end - (offset + bytes)); 1932 WARN_ON(ret); 1933 goto out; 1934 } 1935 } 1936 1937 ret = remove_from_bitmap(ctl, info, &offset, &bytes); 1938 if (ret == -EAGAIN) 1939 goto again; 1940 BUG_ON(ret); /* logic error */ 1941 out_lock: 1942 spin_unlock(&ctl->tree_lock); 1943 out: 1944 return ret; 1945 } 1946 1947 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, 1948 u64 bytes) 1949 { 1950 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 1951 struct btrfs_free_space *info; 1952 struct rb_node *n; 1953 int count = 0; 1954 1955 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) { 1956 info = rb_entry(n, struct btrfs_free_space, offset_index); 1957 if (info->bytes >= bytes && !block_group->ro) 1958 count++; 1959 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", 1960 (unsigned long long)info->offset, 1961 (unsigned long long)info->bytes, 1962 (info->bitmap) ? "yes" : "no"); 1963 } 1964 printk(KERN_INFO "block group has cluster?: %s\n", 1965 list_empty(&block_group->cluster_list) ? "no" : "yes"); 1966 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" 1967 "\n", count); 1968 } 1969 1970 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group) 1971 { 1972 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 1973 1974 spin_lock_init(&ctl->tree_lock); 1975 ctl->unit = block_group->sectorsize; 1976 ctl->start = block_group->key.objectid; 1977 ctl->private = block_group; 1978 ctl->op = &free_space_op; 1979 1980 /* 1981 * we only want to have 32k of ram per block group for keeping 1982 * track of free space, and if we pass 1/2 of that we want to 1983 * start converting things over to using bitmaps 1984 */ 1985 ctl->extents_thresh = ((1024 * 32) / 2) / 1986 sizeof(struct btrfs_free_space); 1987 } 1988 1989 /* 1990 * for a given cluster, put all of its extents back into the free 1991 * space cache. If the block group passed doesn't match the block group 1992 * pointed to by the cluster, someone else raced in and freed the 1993 * cluster already. In that case, we just return without changing anything 1994 */ 1995 static int 1996 __btrfs_return_cluster_to_free_space( 1997 struct btrfs_block_group_cache *block_group, 1998 struct btrfs_free_cluster *cluster) 1999 { 2000 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2001 struct btrfs_free_space *entry; 2002 struct rb_node *node; 2003 2004 spin_lock(&cluster->lock); 2005 if (cluster->block_group != block_group) 2006 goto out; 2007 2008 cluster->block_group = NULL; 2009 cluster->window_start = 0; 2010 list_del_init(&cluster->block_group_list); 2011 2012 node = rb_first(&cluster->root); 2013 while (node) { 2014 bool bitmap; 2015 2016 entry = rb_entry(node, struct btrfs_free_space, offset_index); 2017 node = rb_next(&entry->offset_index); 2018 rb_erase(&entry->offset_index, &cluster->root); 2019 2020 bitmap = (entry->bitmap != NULL); 2021 if (!bitmap) 2022 try_merge_free_space(ctl, entry, false); 2023 tree_insert_offset(&ctl->free_space_offset, 2024 entry->offset, &entry->offset_index, bitmap); 2025 } 2026 cluster->root = RB_ROOT; 2027 2028 out: 2029 spin_unlock(&cluster->lock); 2030 btrfs_put_block_group(block_group); 2031 return 0; 2032 } 2033 2034 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl) 2035 { 2036 struct btrfs_free_space *info; 2037 struct rb_node *node; 2038 2039 while ((node = rb_last(&ctl->free_space_offset)) != NULL) { 2040 info = rb_entry(node, struct btrfs_free_space, offset_index); 2041 if (!info->bitmap) { 2042 unlink_free_space(ctl, info); 2043 kmem_cache_free(btrfs_free_space_cachep, info); 2044 } else { 2045 free_bitmap(ctl, info); 2046 } 2047 if (need_resched()) { 2048 spin_unlock(&ctl->tree_lock); 2049 cond_resched(); 2050 spin_lock(&ctl->tree_lock); 2051 } 2052 } 2053 } 2054 2055 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl) 2056 { 2057 spin_lock(&ctl->tree_lock); 2058 __btrfs_remove_free_space_cache_locked(ctl); 2059 spin_unlock(&ctl->tree_lock); 2060 } 2061 2062 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) 2063 { 2064 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2065 struct btrfs_free_cluster *cluster; 2066 struct list_head *head; 2067 2068 spin_lock(&ctl->tree_lock); 2069 while ((head = block_group->cluster_list.next) != 2070 &block_group->cluster_list) { 2071 cluster = list_entry(head, struct btrfs_free_cluster, 2072 block_group_list); 2073 2074 WARN_ON(cluster->block_group != block_group); 2075 __btrfs_return_cluster_to_free_space(block_group, cluster); 2076 if (need_resched()) { 2077 spin_unlock(&ctl->tree_lock); 2078 cond_resched(); 2079 spin_lock(&ctl->tree_lock); 2080 } 2081 } 2082 __btrfs_remove_free_space_cache_locked(ctl); 2083 spin_unlock(&ctl->tree_lock); 2084 2085 } 2086 2087 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, 2088 u64 offset, u64 bytes, u64 empty_size) 2089 { 2090 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2091 struct btrfs_free_space *entry = NULL; 2092 u64 bytes_search = bytes + empty_size; 2093 u64 ret = 0; 2094 2095 spin_lock(&ctl->tree_lock); 2096 entry = find_free_space(ctl, &offset, &bytes_search); 2097 if (!entry) 2098 goto out; 2099 2100 ret = offset; 2101 if (entry->bitmap) { 2102 bitmap_clear_bits(ctl, entry, offset, bytes); 2103 if (!entry->bytes) 2104 free_bitmap(ctl, entry); 2105 } else { 2106 unlink_free_space(ctl, entry); 2107 entry->offset += bytes; 2108 entry->bytes -= bytes; 2109 if (!entry->bytes) 2110 kmem_cache_free(btrfs_free_space_cachep, entry); 2111 else 2112 link_free_space(ctl, entry); 2113 } 2114 2115 out: 2116 spin_unlock(&ctl->tree_lock); 2117 2118 return ret; 2119 } 2120 2121 /* 2122 * given a cluster, put all of its extents back into the free space 2123 * cache. If a block group is passed, this function will only free 2124 * a cluster that belongs to the passed block group. 2125 * 2126 * Otherwise, it'll get a reference on the block group pointed to by the 2127 * cluster and remove the cluster from it. 2128 */ 2129 int btrfs_return_cluster_to_free_space( 2130 struct btrfs_block_group_cache *block_group, 2131 struct btrfs_free_cluster *cluster) 2132 { 2133 struct btrfs_free_space_ctl *ctl; 2134 int ret; 2135 2136 /* first, get a safe pointer to the block group */ 2137 spin_lock(&cluster->lock); 2138 if (!block_group) { 2139 block_group = cluster->block_group; 2140 if (!block_group) { 2141 spin_unlock(&cluster->lock); 2142 return 0; 2143 } 2144 } else if (cluster->block_group != block_group) { 2145 /* someone else has already freed it don't redo their work */ 2146 spin_unlock(&cluster->lock); 2147 return 0; 2148 } 2149 atomic_inc(&block_group->count); 2150 spin_unlock(&cluster->lock); 2151 2152 ctl = block_group->free_space_ctl; 2153 2154 /* now return any extents the cluster had on it */ 2155 spin_lock(&ctl->tree_lock); 2156 ret = __btrfs_return_cluster_to_free_space(block_group, cluster); 2157 spin_unlock(&ctl->tree_lock); 2158 2159 /* finally drop our ref */ 2160 btrfs_put_block_group(block_group); 2161 return ret; 2162 } 2163 2164 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, 2165 struct btrfs_free_cluster *cluster, 2166 struct btrfs_free_space *entry, 2167 u64 bytes, u64 min_start) 2168 { 2169 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2170 int err; 2171 u64 search_start = cluster->window_start; 2172 u64 search_bytes = bytes; 2173 u64 ret = 0; 2174 2175 search_start = min_start; 2176 search_bytes = bytes; 2177 2178 err = search_bitmap(ctl, entry, &search_start, &search_bytes); 2179 if (err) 2180 return 0; 2181 2182 ret = search_start; 2183 __bitmap_clear_bits(ctl, entry, ret, bytes); 2184 2185 return ret; 2186 } 2187 2188 /* 2189 * given a cluster, try to allocate 'bytes' from it, returns 0 2190 * if it couldn't find anything suitably large, or a logical disk offset 2191 * if things worked out 2192 */ 2193 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, 2194 struct btrfs_free_cluster *cluster, u64 bytes, 2195 u64 min_start) 2196 { 2197 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2198 struct btrfs_free_space *entry = NULL; 2199 struct rb_node *node; 2200 u64 ret = 0; 2201 2202 spin_lock(&cluster->lock); 2203 if (bytes > cluster->max_size) 2204 goto out; 2205 2206 if (cluster->block_group != block_group) 2207 goto out; 2208 2209 node = rb_first(&cluster->root); 2210 if (!node) 2211 goto out; 2212 2213 entry = rb_entry(node, struct btrfs_free_space, offset_index); 2214 while(1) { 2215 if (entry->bytes < bytes || 2216 (!entry->bitmap && entry->offset < min_start)) { 2217 node = rb_next(&entry->offset_index); 2218 if (!node) 2219 break; 2220 entry = rb_entry(node, struct btrfs_free_space, 2221 offset_index); 2222 continue; 2223 } 2224 2225 if (entry->bitmap) { 2226 ret = btrfs_alloc_from_bitmap(block_group, 2227 cluster, entry, bytes, 2228 cluster->window_start); 2229 if (ret == 0) { 2230 node = rb_next(&entry->offset_index); 2231 if (!node) 2232 break; 2233 entry = rb_entry(node, struct btrfs_free_space, 2234 offset_index); 2235 continue; 2236 } 2237 cluster->window_start += bytes; 2238 } else { 2239 ret = entry->offset; 2240 2241 entry->offset += bytes; 2242 entry->bytes -= bytes; 2243 } 2244 2245 if (entry->bytes == 0) 2246 rb_erase(&entry->offset_index, &cluster->root); 2247 break; 2248 } 2249 out: 2250 spin_unlock(&cluster->lock); 2251 2252 if (!ret) 2253 return 0; 2254 2255 spin_lock(&ctl->tree_lock); 2256 2257 ctl->free_space -= bytes; 2258 if (entry->bytes == 0) { 2259 ctl->free_extents--; 2260 if (entry->bitmap) { 2261 kfree(entry->bitmap); 2262 ctl->total_bitmaps--; 2263 ctl->op->recalc_thresholds(ctl); 2264 } 2265 kmem_cache_free(btrfs_free_space_cachep, entry); 2266 } 2267 2268 spin_unlock(&ctl->tree_lock); 2269 2270 return ret; 2271 } 2272 2273 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, 2274 struct btrfs_free_space *entry, 2275 struct btrfs_free_cluster *cluster, 2276 u64 offset, u64 bytes, 2277 u64 cont1_bytes, u64 min_bytes) 2278 { 2279 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2280 unsigned long next_zero; 2281 unsigned long i; 2282 unsigned long want_bits; 2283 unsigned long min_bits; 2284 unsigned long found_bits; 2285 unsigned long start = 0; 2286 unsigned long total_found = 0; 2287 int ret; 2288 2289 i = offset_to_bit(entry->offset, ctl->unit, 2290 max_t(u64, offset, entry->offset)); 2291 want_bits = bytes_to_bits(bytes, ctl->unit); 2292 min_bits = bytes_to_bits(min_bytes, ctl->unit); 2293 2294 again: 2295 found_bits = 0; 2296 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) { 2297 next_zero = find_next_zero_bit(entry->bitmap, 2298 BITS_PER_BITMAP, i); 2299 if (next_zero - i >= min_bits) { 2300 found_bits = next_zero - i; 2301 break; 2302 } 2303 i = next_zero; 2304 } 2305 2306 if (!found_bits) 2307 return -ENOSPC; 2308 2309 if (!total_found) { 2310 start = i; 2311 cluster->max_size = 0; 2312 } 2313 2314 total_found += found_bits; 2315 2316 if (cluster->max_size < found_bits * ctl->unit) 2317 cluster->max_size = found_bits * ctl->unit; 2318 2319 if (total_found < want_bits || cluster->max_size < cont1_bytes) { 2320 i = next_zero + 1; 2321 goto again; 2322 } 2323 2324 cluster->window_start = start * ctl->unit + entry->offset; 2325 rb_erase(&entry->offset_index, &ctl->free_space_offset); 2326 ret = tree_insert_offset(&cluster->root, entry->offset, 2327 &entry->offset_index, 1); 2328 BUG_ON(ret); /* -EEXIST; Logic error */ 2329 2330 trace_btrfs_setup_cluster(block_group, cluster, 2331 total_found * ctl->unit, 1); 2332 return 0; 2333 } 2334 2335 /* 2336 * This searches the block group for just extents to fill the cluster with. 2337 * Try to find a cluster with at least bytes total bytes, at least one 2338 * extent of cont1_bytes, and other clusters of at least min_bytes. 2339 */ 2340 static noinline int 2341 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group, 2342 struct btrfs_free_cluster *cluster, 2343 struct list_head *bitmaps, u64 offset, u64 bytes, 2344 u64 cont1_bytes, u64 min_bytes) 2345 { 2346 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2347 struct btrfs_free_space *first = NULL; 2348 struct btrfs_free_space *entry = NULL; 2349 struct btrfs_free_space *last; 2350 struct rb_node *node; 2351 u64 window_start; 2352 u64 window_free; 2353 u64 max_extent; 2354 u64 total_size = 0; 2355 2356 entry = tree_search_offset(ctl, offset, 0, 1); 2357 if (!entry) 2358 return -ENOSPC; 2359 2360 /* 2361 * We don't want bitmaps, so just move along until we find a normal 2362 * extent entry. 2363 */ 2364 while (entry->bitmap || entry->bytes < min_bytes) { 2365 if (entry->bitmap && list_empty(&entry->list)) 2366 list_add_tail(&entry->list, bitmaps); 2367 node = rb_next(&entry->offset_index); 2368 if (!node) 2369 return -ENOSPC; 2370 entry = rb_entry(node, struct btrfs_free_space, offset_index); 2371 } 2372 2373 window_start = entry->offset; 2374 window_free = entry->bytes; 2375 max_extent = entry->bytes; 2376 first = entry; 2377 last = entry; 2378 2379 for (node = rb_next(&entry->offset_index); node; 2380 node = rb_next(&entry->offset_index)) { 2381 entry = rb_entry(node, struct btrfs_free_space, offset_index); 2382 2383 if (entry->bitmap) { 2384 if (list_empty(&entry->list)) 2385 list_add_tail(&entry->list, bitmaps); 2386 continue; 2387 } 2388 2389 if (entry->bytes < min_bytes) 2390 continue; 2391 2392 last = entry; 2393 window_free += entry->bytes; 2394 if (entry->bytes > max_extent) 2395 max_extent = entry->bytes; 2396 } 2397 2398 if (window_free < bytes || max_extent < cont1_bytes) 2399 return -ENOSPC; 2400 2401 cluster->window_start = first->offset; 2402 2403 node = &first->offset_index; 2404 2405 /* 2406 * now we've found our entries, pull them out of the free space 2407 * cache and put them into the cluster rbtree 2408 */ 2409 do { 2410 int ret; 2411 2412 entry = rb_entry(node, struct btrfs_free_space, offset_index); 2413 node = rb_next(&entry->offset_index); 2414 if (entry->bitmap || entry->bytes < min_bytes) 2415 continue; 2416 2417 rb_erase(&entry->offset_index, &ctl->free_space_offset); 2418 ret = tree_insert_offset(&cluster->root, entry->offset, 2419 &entry->offset_index, 0); 2420 total_size += entry->bytes; 2421 BUG_ON(ret); /* -EEXIST; Logic error */ 2422 } while (node && entry != last); 2423 2424 cluster->max_size = max_extent; 2425 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0); 2426 return 0; 2427 } 2428 2429 /* 2430 * This specifically looks for bitmaps that may work in the cluster, we assume 2431 * that we have already failed to find extents that will work. 2432 */ 2433 static noinline int 2434 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group, 2435 struct btrfs_free_cluster *cluster, 2436 struct list_head *bitmaps, u64 offset, u64 bytes, 2437 u64 cont1_bytes, u64 min_bytes) 2438 { 2439 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2440 struct btrfs_free_space *entry; 2441 int ret = -ENOSPC; 2442 u64 bitmap_offset = offset_to_bitmap(ctl, offset); 2443 2444 if (ctl->total_bitmaps == 0) 2445 return -ENOSPC; 2446 2447 /* 2448 * The bitmap that covers offset won't be in the list unless offset 2449 * is just its start offset. 2450 */ 2451 entry = list_first_entry(bitmaps, struct btrfs_free_space, list); 2452 if (entry->offset != bitmap_offset) { 2453 entry = tree_search_offset(ctl, bitmap_offset, 1, 0); 2454 if (entry && list_empty(&entry->list)) 2455 list_add(&entry->list, bitmaps); 2456 } 2457 2458 list_for_each_entry(entry, bitmaps, list) { 2459 if (entry->bytes < bytes) 2460 continue; 2461 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset, 2462 bytes, cont1_bytes, min_bytes); 2463 if (!ret) 2464 return 0; 2465 } 2466 2467 /* 2468 * The bitmaps list has all the bitmaps that record free space 2469 * starting after offset, so no more search is required. 2470 */ 2471 return -ENOSPC; 2472 } 2473 2474 /* 2475 * here we try to find a cluster of blocks in a block group. The goal 2476 * is to find at least bytes+empty_size. 2477 * We might not find them all in one contiguous area. 2478 * 2479 * returns zero and sets up cluster if things worked out, otherwise 2480 * it returns -enospc 2481 */ 2482 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, 2483 struct btrfs_root *root, 2484 struct btrfs_block_group_cache *block_group, 2485 struct btrfs_free_cluster *cluster, 2486 u64 offset, u64 bytes, u64 empty_size) 2487 { 2488 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2489 struct btrfs_free_space *entry, *tmp; 2490 LIST_HEAD(bitmaps); 2491 u64 min_bytes; 2492 u64 cont1_bytes; 2493 int ret; 2494 2495 /* 2496 * Choose the minimum extent size we'll require for this 2497 * cluster. For SSD_SPREAD, don't allow any fragmentation. 2498 * For metadata, allow allocates with smaller extents. For 2499 * data, keep it dense. 2500 */ 2501 if (btrfs_test_opt(root, SSD_SPREAD)) { 2502 cont1_bytes = min_bytes = bytes + empty_size; 2503 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { 2504 cont1_bytes = bytes; 2505 min_bytes = block_group->sectorsize; 2506 } else { 2507 cont1_bytes = max(bytes, (bytes + empty_size) >> 2); 2508 min_bytes = block_group->sectorsize; 2509 } 2510 2511 spin_lock(&ctl->tree_lock); 2512 2513 /* 2514 * If we know we don't have enough space to make a cluster don't even 2515 * bother doing all the work to try and find one. 2516 */ 2517 if (ctl->free_space < bytes) { 2518 spin_unlock(&ctl->tree_lock); 2519 return -ENOSPC; 2520 } 2521 2522 spin_lock(&cluster->lock); 2523 2524 /* someone already found a cluster, hooray */ 2525 if (cluster->block_group) { 2526 ret = 0; 2527 goto out; 2528 } 2529 2530 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size, 2531 min_bytes); 2532 2533 INIT_LIST_HEAD(&bitmaps); 2534 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset, 2535 bytes + empty_size, 2536 cont1_bytes, min_bytes); 2537 if (ret) 2538 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps, 2539 offset, bytes + empty_size, 2540 cont1_bytes, min_bytes); 2541 2542 /* Clear our temporary list */ 2543 list_for_each_entry_safe(entry, tmp, &bitmaps, list) 2544 list_del_init(&entry->list); 2545 2546 if (!ret) { 2547 atomic_inc(&block_group->count); 2548 list_add_tail(&cluster->block_group_list, 2549 &block_group->cluster_list); 2550 cluster->block_group = block_group; 2551 } else { 2552 trace_btrfs_failed_cluster_setup(block_group); 2553 } 2554 out: 2555 spin_unlock(&cluster->lock); 2556 spin_unlock(&ctl->tree_lock); 2557 2558 return ret; 2559 } 2560 2561 /* 2562 * simple code to zero out a cluster 2563 */ 2564 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) 2565 { 2566 spin_lock_init(&cluster->lock); 2567 spin_lock_init(&cluster->refill_lock); 2568 cluster->root = RB_ROOT; 2569 cluster->max_size = 0; 2570 INIT_LIST_HEAD(&cluster->block_group_list); 2571 cluster->block_group = NULL; 2572 } 2573 2574 static int do_trimming(struct btrfs_block_group_cache *block_group, 2575 u64 *total_trimmed, u64 start, u64 bytes, 2576 u64 reserved_start, u64 reserved_bytes) 2577 { 2578 struct btrfs_space_info *space_info = block_group->space_info; 2579 struct btrfs_fs_info *fs_info = block_group->fs_info; 2580 int ret; 2581 int update = 0; 2582 u64 trimmed = 0; 2583 2584 spin_lock(&space_info->lock); 2585 spin_lock(&block_group->lock); 2586 if (!block_group->ro) { 2587 block_group->reserved += reserved_bytes; 2588 space_info->bytes_reserved += reserved_bytes; 2589 update = 1; 2590 } 2591 spin_unlock(&block_group->lock); 2592 spin_unlock(&space_info->lock); 2593 2594 ret = btrfs_error_discard_extent(fs_info->extent_root, 2595 start, bytes, &trimmed); 2596 if (!ret) 2597 *total_trimmed += trimmed; 2598 2599 btrfs_add_free_space(block_group, reserved_start, reserved_bytes); 2600 2601 if (update) { 2602 spin_lock(&space_info->lock); 2603 spin_lock(&block_group->lock); 2604 if (block_group->ro) 2605 space_info->bytes_readonly += reserved_bytes; 2606 block_group->reserved -= reserved_bytes; 2607 space_info->bytes_reserved -= reserved_bytes; 2608 spin_unlock(&space_info->lock); 2609 spin_unlock(&block_group->lock); 2610 } 2611 2612 return ret; 2613 } 2614 2615 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group, 2616 u64 *total_trimmed, u64 start, u64 end, u64 minlen) 2617 { 2618 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2619 struct btrfs_free_space *entry; 2620 struct rb_node *node; 2621 int ret = 0; 2622 u64 extent_start; 2623 u64 extent_bytes; 2624 u64 bytes; 2625 2626 while (start < end) { 2627 spin_lock(&ctl->tree_lock); 2628 2629 if (ctl->free_space < minlen) { 2630 spin_unlock(&ctl->tree_lock); 2631 break; 2632 } 2633 2634 entry = tree_search_offset(ctl, start, 0, 1); 2635 if (!entry) { 2636 spin_unlock(&ctl->tree_lock); 2637 break; 2638 } 2639 2640 /* skip bitmaps */ 2641 while (entry->bitmap) { 2642 node = rb_next(&entry->offset_index); 2643 if (!node) { 2644 spin_unlock(&ctl->tree_lock); 2645 goto out; 2646 } 2647 entry = rb_entry(node, struct btrfs_free_space, 2648 offset_index); 2649 } 2650 2651 if (entry->offset >= end) { 2652 spin_unlock(&ctl->tree_lock); 2653 break; 2654 } 2655 2656 extent_start = entry->offset; 2657 extent_bytes = entry->bytes; 2658 start = max(start, extent_start); 2659 bytes = min(extent_start + extent_bytes, end) - start; 2660 if (bytes < minlen) { 2661 spin_unlock(&ctl->tree_lock); 2662 goto next; 2663 } 2664 2665 unlink_free_space(ctl, entry); 2666 kmem_cache_free(btrfs_free_space_cachep, entry); 2667 2668 spin_unlock(&ctl->tree_lock); 2669 2670 ret = do_trimming(block_group, total_trimmed, start, bytes, 2671 extent_start, extent_bytes); 2672 if (ret) 2673 break; 2674 next: 2675 start += bytes; 2676 2677 if (fatal_signal_pending(current)) { 2678 ret = -ERESTARTSYS; 2679 break; 2680 } 2681 2682 cond_resched(); 2683 } 2684 out: 2685 return ret; 2686 } 2687 2688 static int trim_bitmaps(struct btrfs_block_group_cache *block_group, 2689 u64 *total_trimmed, u64 start, u64 end, u64 minlen) 2690 { 2691 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; 2692 struct btrfs_free_space *entry; 2693 int ret = 0; 2694 int ret2; 2695 u64 bytes; 2696 u64 offset = offset_to_bitmap(ctl, start); 2697 2698 while (offset < end) { 2699 bool next_bitmap = false; 2700 2701 spin_lock(&ctl->tree_lock); 2702 2703 if (ctl->free_space < minlen) { 2704 spin_unlock(&ctl->tree_lock); 2705 break; 2706 } 2707 2708 entry = tree_search_offset(ctl, offset, 1, 0); 2709 if (!entry) { 2710 spin_unlock(&ctl->tree_lock); 2711 next_bitmap = true; 2712 goto next; 2713 } 2714 2715 bytes = minlen; 2716 ret2 = search_bitmap(ctl, entry, &start, &bytes); 2717 if (ret2 || start >= end) { 2718 spin_unlock(&ctl->tree_lock); 2719 next_bitmap = true; 2720 goto next; 2721 } 2722 2723 bytes = min(bytes, end - start); 2724 if (bytes < minlen) { 2725 spin_unlock(&ctl->tree_lock); 2726 goto next; 2727 } 2728 2729 bitmap_clear_bits(ctl, entry, start, bytes); 2730 if (entry->bytes == 0) 2731 free_bitmap(ctl, entry); 2732 2733 spin_unlock(&ctl->tree_lock); 2734 2735 ret = do_trimming(block_group, total_trimmed, start, bytes, 2736 start, bytes); 2737 if (ret) 2738 break; 2739 next: 2740 if (next_bitmap) { 2741 offset += BITS_PER_BITMAP * ctl->unit; 2742 } else { 2743 start += bytes; 2744 if (start >= offset + BITS_PER_BITMAP * ctl->unit) 2745 offset += BITS_PER_BITMAP * ctl->unit; 2746 } 2747 2748 if (fatal_signal_pending(current)) { 2749 ret = -ERESTARTSYS; 2750 break; 2751 } 2752 2753 cond_resched(); 2754 } 2755 2756 return ret; 2757 } 2758 2759 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group, 2760 u64 *trimmed, u64 start, u64 end, u64 minlen) 2761 { 2762 int ret; 2763 2764 *trimmed = 0; 2765 2766 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen); 2767 if (ret) 2768 return ret; 2769 2770 ret = trim_bitmaps(block_group, trimmed, start, end, minlen); 2771 2772 return ret; 2773 } 2774 2775 /* 2776 * Find the left-most item in the cache tree, and then return the 2777 * smallest inode number in the item. 2778 * 2779 * Note: the returned inode number may not be the smallest one in 2780 * the tree, if the left-most item is a bitmap. 2781 */ 2782 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root) 2783 { 2784 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl; 2785 struct btrfs_free_space *entry = NULL; 2786 u64 ino = 0; 2787 2788 spin_lock(&ctl->tree_lock); 2789 2790 if (RB_EMPTY_ROOT(&ctl->free_space_offset)) 2791 goto out; 2792 2793 entry = rb_entry(rb_first(&ctl->free_space_offset), 2794 struct btrfs_free_space, offset_index); 2795 2796 if (!entry->bitmap) { 2797 ino = entry->offset; 2798 2799 unlink_free_space(ctl, entry); 2800 entry->offset++; 2801 entry->bytes--; 2802 if (!entry->bytes) 2803 kmem_cache_free(btrfs_free_space_cachep, entry); 2804 else 2805 link_free_space(ctl, entry); 2806 } else { 2807 u64 offset = 0; 2808 u64 count = 1; 2809 int ret; 2810 2811 ret = search_bitmap(ctl, entry, &offset, &count); 2812 /* Logic error; Should be empty if it can't find anything */ 2813 BUG_ON(ret); 2814 2815 ino = offset; 2816 bitmap_clear_bits(ctl, entry, offset, 1); 2817 if (entry->bytes == 0) 2818 free_bitmap(ctl, entry); 2819 } 2820 out: 2821 spin_unlock(&ctl->tree_lock); 2822 2823 return ino; 2824 } 2825 2826 struct inode *lookup_free_ino_inode(struct btrfs_root *root, 2827 struct btrfs_path *path) 2828 { 2829 struct inode *inode = NULL; 2830 2831 spin_lock(&root->cache_lock); 2832 if (root->cache_inode) 2833 inode = igrab(root->cache_inode); 2834 spin_unlock(&root->cache_lock); 2835 if (inode) 2836 return inode; 2837 2838 inode = __lookup_free_space_inode(root, path, 0); 2839 if (IS_ERR(inode)) 2840 return inode; 2841 2842 spin_lock(&root->cache_lock); 2843 if (!btrfs_fs_closing(root->fs_info)) 2844 root->cache_inode = igrab(inode); 2845 spin_unlock(&root->cache_lock); 2846 2847 return inode; 2848 } 2849 2850 int create_free_ino_inode(struct btrfs_root *root, 2851 struct btrfs_trans_handle *trans, 2852 struct btrfs_path *path) 2853 { 2854 return __create_free_space_inode(root, trans, path, 2855 BTRFS_FREE_INO_OBJECTID, 0); 2856 } 2857 2858 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 2859 { 2860 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; 2861 struct btrfs_path *path; 2862 struct inode *inode; 2863 int ret = 0; 2864 u64 root_gen = btrfs_root_generation(&root->root_item); 2865 2866 if (!btrfs_test_opt(root, INODE_MAP_CACHE)) 2867 return 0; 2868 2869 /* 2870 * If we're unmounting then just return, since this does a search on the 2871 * normal root and not the commit root and we could deadlock. 2872 */ 2873 if (btrfs_fs_closing(fs_info)) 2874 return 0; 2875 2876 path = btrfs_alloc_path(); 2877 if (!path) 2878 return 0; 2879 2880 inode = lookup_free_ino_inode(root, path); 2881 if (IS_ERR(inode)) 2882 goto out; 2883 2884 if (root_gen != BTRFS_I(inode)->generation) 2885 goto out_put; 2886 2887 ret = __load_free_space_cache(root, inode, ctl, path, 0); 2888 2889 if (ret < 0) 2890 printk(KERN_ERR "btrfs: failed to load free ino cache for " 2891 "root %llu\n", root->root_key.objectid); 2892 out_put: 2893 iput(inode); 2894 out: 2895 btrfs_free_path(path); 2896 return ret; 2897 } 2898 2899 int btrfs_write_out_ino_cache(struct btrfs_root *root, 2900 struct btrfs_trans_handle *trans, 2901 struct btrfs_path *path) 2902 { 2903 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl; 2904 struct inode *inode; 2905 int ret; 2906 2907 if (!btrfs_test_opt(root, INODE_MAP_CACHE)) 2908 return 0; 2909 2910 inode = lookup_free_ino_inode(root, path); 2911 if (IS_ERR(inode)) 2912 return 0; 2913 2914 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0); 2915 if (ret) { 2916 btrfs_delalloc_release_metadata(inode, inode->i_size); 2917 #ifdef DEBUG 2918 printk(KERN_ERR "btrfs: failed to write free ino cache " 2919 "for root %llu\n", root->root_key.objectid); 2920 #endif 2921 } 2922 2923 iput(inode); 2924 return ret; 2925 } 2926