Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 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 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 19 | #include <linux/pagemap.h> |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 20 | #include <linux/sched.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 21 | #include <linux/slab.h> |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 22 | #include <linux/math64.h> |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 23 | #include "ctree.h" |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 24 | #include "free-space-cache.h" |
| 25 | #include "transaction.h" |
Josef Bacik | 0af3d00 | 2010-06-21 14:48:16 -0400 | [diff] [blame] | 26 | #include "disk-io.h" |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 27 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 28 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
| 29 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) |
| 30 | |
Josef Bacik | 0cb59c9 | 2010-07-02 12:14:14 -0400 | [diff] [blame^] | 31 | static void recalculate_thresholds(struct btrfs_block_group_cache |
| 32 | *block_group); |
| 33 | static int link_free_space(struct btrfs_block_group_cache *block_group, |
| 34 | struct btrfs_free_space *info); |
| 35 | |
Josef Bacik | 0af3d00 | 2010-06-21 14:48:16 -0400 | [diff] [blame] | 36 | struct inode *lookup_free_space_inode(struct btrfs_root *root, |
| 37 | struct btrfs_block_group_cache |
| 38 | *block_group, struct btrfs_path *path) |
| 39 | { |
| 40 | struct btrfs_key key; |
| 41 | struct btrfs_key location; |
| 42 | struct btrfs_disk_key disk_key; |
| 43 | struct btrfs_free_space_header *header; |
| 44 | struct extent_buffer *leaf; |
| 45 | struct inode *inode = NULL; |
| 46 | int ret; |
| 47 | |
| 48 | spin_lock(&block_group->lock); |
| 49 | if (block_group->inode) |
| 50 | inode = igrab(block_group->inode); |
| 51 | spin_unlock(&block_group->lock); |
| 52 | if (inode) |
| 53 | return inode; |
| 54 | |
| 55 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
| 56 | key.offset = block_group->key.objectid; |
| 57 | key.type = 0; |
| 58 | |
| 59 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 60 | if (ret < 0) |
| 61 | return ERR_PTR(ret); |
| 62 | if (ret > 0) { |
| 63 | btrfs_release_path(root, path); |
| 64 | return ERR_PTR(-ENOENT); |
| 65 | } |
| 66 | |
| 67 | leaf = path->nodes[0]; |
| 68 | header = btrfs_item_ptr(leaf, path->slots[0], |
| 69 | struct btrfs_free_space_header); |
| 70 | btrfs_free_space_key(leaf, header, &disk_key); |
| 71 | btrfs_disk_key_to_cpu(&location, &disk_key); |
| 72 | btrfs_release_path(root, path); |
| 73 | |
| 74 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); |
| 75 | if (!inode) |
| 76 | return ERR_PTR(-ENOENT); |
| 77 | if (IS_ERR(inode)) |
| 78 | return inode; |
| 79 | if (is_bad_inode(inode)) { |
| 80 | iput(inode); |
| 81 | return ERR_PTR(-ENOENT); |
| 82 | } |
| 83 | |
| 84 | spin_lock(&block_group->lock); |
| 85 | if (!root->fs_info->closing) { |
| 86 | block_group->inode = igrab(inode); |
| 87 | block_group->iref = 1; |
| 88 | } |
| 89 | spin_unlock(&block_group->lock); |
| 90 | |
| 91 | return inode; |
| 92 | } |
| 93 | |
| 94 | int create_free_space_inode(struct btrfs_root *root, |
| 95 | struct btrfs_trans_handle *trans, |
| 96 | struct btrfs_block_group_cache *block_group, |
| 97 | struct btrfs_path *path) |
| 98 | { |
| 99 | struct btrfs_key key; |
| 100 | struct btrfs_disk_key disk_key; |
| 101 | struct btrfs_free_space_header *header; |
| 102 | struct btrfs_inode_item *inode_item; |
| 103 | struct extent_buffer *leaf; |
| 104 | u64 objectid; |
| 105 | int ret; |
| 106 | |
| 107 | ret = btrfs_find_free_objectid(trans, root, 0, &objectid); |
| 108 | if (ret < 0) |
| 109 | return ret; |
| 110 | |
| 111 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); |
| 112 | if (ret) |
| 113 | return ret; |
| 114 | |
| 115 | leaf = path->nodes[0]; |
| 116 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
| 117 | struct btrfs_inode_item); |
| 118 | btrfs_item_key(leaf, &disk_key, path->slots[0]); |
| 119 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, |
| 120 | sizeof(*inode_item)); |
| 121 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); |
| 122 | btrfs_set_inode_size(leaf, inode_item, 0); |
| 123 | btrfs_set_inode_nbytes(leaf, inode_item, 0); |
| 124 | btrfs_set_inode_uid(leaf, inode_item, 0); |
| 125 | btrfs_set_inode_gid(leaf, inode_item, 0); |
| 126 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); |
| 127 | btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS | |
| 128 | BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM); |
| 129 | btrfs_set_inode_nlink(leaf, inode_item, 1); |
| 130 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); |
| 131 | btrfs_set_inode_block_group(leaf, inode_item, |
| 132 | block_group->key.objectid); |
| 133 | btrfs_mark_buffer_dirty(leaf); |
| 134 | btrfs_release_path(root, path); |
| 135 | |
| 136 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
| 137 | key.offset = block_group->key.objectid; |
| 138 | key.type = 0; |
| 139 | |
| 140 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
| 141 | sizeof(struct btrfs_free_space_header)); |
| 142 | if (ret < 0) { |
| 143 | btrfs_release_path(root, path); |
| 144 | return ret; |
| 145 | } |
| 146 | leaf = path->nodes[0]; |
| 147 | header = btrfs_item_ptr(leaf, path->slots[0], |
| 148 | struct btrfs_free_space_header); |
| 149 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); |
| 150 | btrfs_set_free_space_key(leaf, header, &disk_key); |
| 151 | btrfs_mark_buffer_dirty(leaf); |
| 152 | btrfs_release_path(root, path); |
| 153 | |
| 154 | return 0; |
| 155 | } |
| 156 | |
| 157 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, |
| 158 | struct btrfs_trans_handle *trans, |
| 159 | struct btrfs_path *path, |
| 160 | struct inode *inode) |
| 161 | { |
| 162 | loff_t oldsize; |
| 163 | int ret = 0; |
| 164 | |
| 165 | trans->block_rsv = root->orphan_block_rsv; |
| 166 | ret = btrfs_block_rsv_check(trans, root, |
| 167 | root->orphan_block_rsv, |
| 168 | 0, 5); |
| 169 | if (ret) |
| 170 | return ret; |
| 171 | |
| 172 | oldsize = i_size_read(inode); |
| 173 | btrfs_i_size_write(inode, 0); |
| 174 | truncate_pagecache(inode, oldsize, 0); |
| 175 | |
| 176 | /* |
| 177 | * We don't need an orphan item because truncating the free space cache |
| 178 | * will never be split across transactions. |
| 179 | */ |
| 180 | ret = btrfs_truncate_inode_items(trans, root, inode, |
| 181 | 0, BTRFS_EXTENT_DATA_KEY); |
| 182 | if (ret) { |
| 183 | WARN_ON(1); |
| 184 | return ret; |
| 185 | } |
| 186 | |
| 187 | return btrfs_update_inode(trans, root, inode); |
| 188 | } |
| 189 | |
Josef Bacik | 0cb59c9 | 2010-07-02 12:14:14 -0400 | [diff] [blame^] | 190 | int btrfs_write_out_cache(struct btrfs_root *root, |
| 191 | struct btrfs_trans_handle *trans, |
| 192 | struct btrfs_block_group_cache *block_group, |
| 193 | struct btrfs_path *path) |
| 194 | { |
| 195 | struct btrfs_free_space_header *header; |
| 196 | struct extent_buffer *leaf; |
| 197 | struct inode *inode; |
| 198 | struct rb_node *node; |
| 199 | struct list_head *pos, *n; |
| 200 | struct page *page; |
| 201 | struct extent_state *cached_state = NULL; |
| 202 | struct list_head bitmap_list; |
| 203 | struct btrfs_key key; |
| 204 | u64 bytes = 0; |
| 205 | u32 *crc, *checksums; |
| 206 | pgoff_t index = 0, last_index = 0; |
| 207 | unsigned long first_page_offset; |
| 208 | int num_checksums; |
| 209 | int entries = 0; |
| 210 | int bitmaps = 0; |
| 211 | int ret = 0; |
| 212 | |
| 213 | root = root->fs_info->tree_root; |
| 214 | |
| 215 | INIT_LIST_HEAD(&bitmap_list); |
| 216 | |
| 217 | spin_lock(&block_group->lock); |
| 218 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { |
| 219 | spin_unlock(&block_group->lock); |
| 220 | return 0; |
| 221 | } |
| 222 | spin_unlock(&block_group->lock); |
| 223 | |
| 224 | inode = lookup_free_space_inode(root, block_group, path); |
| 225 | if (IS_ERR(inode)) |
| 226 | return 0; |
| 227 | |
| 228 | if (!i_size_read(inode)) { |
| 229 | iput(inode); |
| 230 | return 0; |
| 231 | } |
| 232 | |
| 233 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; |
| 234 | filemap_write_and_wait(inode->i_mapping); |
| 235 | btrfs_wait_ordered_range(inode, inode->i_size & |
| 236 | ~(root->sectorsize - 1), (u64)-1); |
| 237 | |
| 238 | /* We need a checksum per page. */ |
| 239 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; |
| 240 | crc = checksums = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); |
| 241 | if (!crc) { |
| 242 | iput(inode); |
| 243 | return 0; |
| 244 | } |
| 245 | |
| 246 | /* Since the first page has all of our checksums and our generation we |
| 247 | * need to calculate the offset into the page that we can start writing |
| 248 | * our entries. |
| 249 | */ |
| 250 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); |
| 251 | |
| 252 | node = rb_first(&block_group->free_space_offset); |
| 253 | if (!node) |
| 254 | goto out_free; |
| 255 | |
| 256 | /* |
| 257 | * Lock all pages first so we can lock the extent safely. |
| 258 | * |
| 259 | * NOTE: Because we hold the ref the entire time we're going to write to |
| 260 | * the page find_get_page should never fail, so we don't do a check |
| 261 | * after find_get_page at this point. Just putting this here so people |
| 262 | * know and don't freak out. |
| 263 | */ |
| 264 | while (index <= last_index) { |
| 265 | page = grab_cache_page(inode->i_mapping, index); |
| 266 | if (!page) { |
| 267 | pgoff_t i = 0; |
| 268 | |
| 269 | while (i < index) { |
| 270 | page = find_get_page(inode->i_mapping, i); |
| 271 | unlock_page(page); |
| 272 | page_cache_release(page); |
| 273 | page_cache_release(page); |
| 274 | i++; |
| 275 | } |
| 276 | goto out_free; |
| 277 | } |
| 278 | index++; |
| 279 | } |
| 280 | |
| 281 | index = 0; |
| 282 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, |
| 283 | 0, &cached_state, GFP_NOFS); |
| 284 | |
| 285 | /* Write out the extent entries */ |
| 286 | do { |
| 287 | struct btrfs_free_space_entry *entry; |
| 288 | void *addr; |
| 289 | unsigned long offset = 0; |
| 290 | unsigned long start_offset = 0; |
| 291 | |
| 292 | if (index == 0) { |
| 293 | start_offset = first_page_offset; |
| 294 | offset = start_offset; |
| 295 | } |
| 296 | |
| 297 | page = find_get_page(inode->i_mapping, index); |
| 298 | |
| 299 | addr = kmap(page); |
| 300 | entry = addr + start_offset; |
| 301 | |
| 302 | memset(addr, 0, PAGE_CACHE_SIZE); |
| 303 | while (1) { |
| 304 | struct btrfs_free_space *e; |
| 305 | |
| 306 | e = rb_entry(node, struct btrfs_free_space, offset_index); |
| 307 | entries++; |
| 308 | |
| 309 | entry->offset = cpu_to_le64(e->offset); |
| 310 | entry->bytes = cpu_to_le64(e->bytes); |
| 311 | if (e->bitmap) { |
| 312 | entry->type = BTRFS_FREE_SPACE_BITMAP; |
| 313 | list_add_tail(&e->list, &bitmap_list); |
| 314 | bitmaps++; |
| 315 | } else { |
| 316 | entry->type = BTRFS_FREE_SPACE_EXTENT; |
| 317 | } |
| 318 | node = rb_next(node); |
| 319 | if (!node) |
| 320 | break; |
| 321 | offset += sizeof(struct btrfs_free_space_entry); |
| 322 | if (offset + sizeof(struct btrfs_free_space_entry) >= |
| 323 | PAGE_CACHE_SIZE) |
| 324 | break; |
| 325 | entry++; |
| 326 | } |
| 327 | *crc = ~(u32)0; |
| 328 | *crc = btrfs_csum_data(root, addr + start_offset, *crc, |
| 329 | PAGE_CACHE_SIZE - start_offset); |
| 330 | kunmap(page); |
| 331 | |
| 332 | btrfs_csum_final(*crc, (char *)crc); |
| 333 | crc++; |
| 334 | |
| 335 | bytes += PAGE_CACHE_SIZE; |
| 336 | |
| 337 | ClearPageChecked(page); |
| 338 | set_page_extent_mapped(page); |
| 339 | SetPageUptodate(page); |
| 340 | set_page_dirty(page); |
| 341 | |
| 342 | /* |
| 343 | * We need to release our reference we got for grab_cache_page, |
| 344 | * except for the first page which will hold our checksums, we |
| 345 | * do that below. |
| 346 | */ |
| 347 | if (index != 0) { |
| 348 | unlock_page(page); |
| 349 | page_cache_release(page); |
| 350 | } |
| 351 | |
| 352 | page_cache_release(page); |
| 353 | |
| 354 | index++; |
| 355 | } while (node); |
| 356 | |
| 357 | /* Write out the bitmaps */ |
| 358 | list_for_each_safe(pos, n, &bitmap_list) { |
| 359 | void *addr; |
| 360 | struct btrfs_free_space *entry = |
| 361 | list_entry(pos, struct btrfs_free_space, list); |
| 362 | |
| 363 | page = find_get_page(inode->i_mapping, index); |
| 364 | |
| 365 | addr = kmap(page); |
| 366 | memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE); |
| 367 | *crc = ~(u32)0; |
| 368 | *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE); |
| 369 | kunmap(page); |
| 370 | btrfs_csum_final(*crc, (char *)crc); |
| 371 | crc++; |
| 372 | bytes += PAGE_CACHE_SIZE; |
| 373 | |
| 374 | ClearPageChecked(page); |
| 375 | set_page_extent_mapped(page); |
| 376 | SetPageUptodate(page); |
| 377 | set_page_dirty(page); |
| 378 | unlock_page(page); |
| 379 | page_cache_release(page); |
| 380 | page_cache_release(page); |
| 381 | list_del_init(&entry->list); |
| 382 | index++; |
| 383 | } |
| 384 | |
| 385 | /* Zero out the rest of the pages just to make sure */ |
| 386 | while (index <= last_index) { |
| 387 | void *addr; |
| 388 | |
| 389 | page = find_get_page(inode->i_mapping, index); |
| 390 | |
| 391 | addr = kmap(page); |
| 392 | memset(addr, 0, PAGE_CACHE_SIZE); |
| 393 | kunmap(page); |
| 394 | ClearPageChecked(page); |
| 395 | set_page_extent_mapped(page); |
| 396 | SetPageUptodate(page); |
| 397 | set_page_dirty(page); |
| 398 | unlock_page(page); |
| 399 | page_cache_release(page); |
| 400 | page_cache_release(page); |
| 401 | bytes += PAGE_CACHE_SIZE; |
| 402 | index++; |
| 403 | } |
| 404 | |
| 405 | btrfs_set_extent_delalloc(inode, 0, bytes - 1, &cached_state); |
| 406 | |
| 407 | /* Write the checksums and trans id to the first page */ |
| 408 | { |
| 409 | void *addr; |
| 410 | u64 *gen; |
| 411 | |
| 412 | page = find_get_page(inode->i_mapping, 0); |
| 413 | |
| 414 | addr = kmap(page); |
| 415 | memcpy(addr, checksums, sizeof(u32) * num_checksums); |
| 416 | gen = addr + (sizeof(u32) * num_checksums); |
| 417 | *gen = trans->transid; |
| 418 | kunmap(page); |
| 419 | ClearPageChecked(page); |
| 420 | set_page_extent_mapped(page); |
| 421 | SetPageUptodate(page); |
| 422 | set_page_dirty(page); |
| 423 | unlock_page(page); |
| 424 | page_cache_release(page); |
| 425 | page_cache_release(page); |
| 426 | } |
| 427 | BTRFS_I(inode)->generation = trans->transid; |
| 428 | |
| 429 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, |
| 430 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); |
| 431 | |
| 432 | filemap_write_and_wait(inode->i_mapping); |
| 433 | |
| 434 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
| 435 | key.offset = block_group->key.objectid; |
| 436 | key.type = 0; |
| 437 | |
| 438 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); |
| 439 | if (ret < 0) { |
| 440 | ret = 0; |
| 441 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
| 442 | EXTENT_DIRTY | EXTENT_DELALLOC | |
| 443 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS); |
| 444 | goto out_free; |
| 445 | } |
| 446 | leaf = path->nodes[0]; |
| 447 | if (ret > 0) { |
| 448 | struct btrfs_key found_key; |
| 449 | BUG_ON(!path->slots[0]); |
| 450 | path->slots[0]--; |
| 451 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 452 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || |
| 453 | found_key.offset != block_group->key.objectid) { |
| 454 | ret = 0; |
| 455 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, |
| 456 | EXTENT_DIRTY | EXTENT_DELALLOC | |
| 457 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, |
| 458 | GFP_NOFS); |
| 459 | btrfs_release_path(root, path); |
| 460 | goto out_free; |
| 461 | } |
| 462 | } |
| 463 | header = btrfs_item_ptr(leaf, path->slots[0], |
| 464 | struct btrfs_free_space_header); |
| 465 | btrfs_set_free_space_entries(leaf, header, entries); |
| 466 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); |
| 467 | btrfs_set_free_space_generation(leaf, header, trans->transid); |
| 468 | btrfs_mark_buffer_dirty(leaf); |
| 469 | btrfs_release_path(root, path); |
| 470 | |
| 471 | ret = 1; |
| 472 | |
| 473 | out_free: |
| 474 | if (ret == 0) { |
| 475 | invalidate_inode_pages2_range(inode->i_mapping, 0, index); |
| 476 | spin_lock(&block_group->lock); |
| 477 | block_group->disk_cache_state = BTRFS_DC_ERROR; |
| 478 | spin_unlock(&block_group->lock); |
| 479 | BTRFS_I(inode)->generation = 0; |
| 480 | } |
| 481 | kfree(checksums); |
| 482 | btrfs_update_inode(trans, root, inode); |
| 483 | iput(inode); |
| 484 | return ret; |
| 485 | } |
| 486 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 487 | static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize, |
| 488 | u64 offset) |
| 489 | { |
| 490 | BUG_ON(offset < bitmap_start); |
| 491 | offset -= bitmap_start; |
| 492 | return (unsigned long)(div64_u64(offset, sectorsize)); |
| 493 | } |
| 494 | |
| 495 | static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize) |
| 496 | { |
| 497 | return (unsigned long)(div64_u64(bytes, sectorsize)); |
| 498 | } |
| 499 | |
| 500 | static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group, |
| 501 | u64 offset) |
| 502 | { |
| 503 | u64 bitmap_start; |
| 504 | u64 bytes_per_bitmap; |
| 505 | |
| 506 | bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize; |
| 507 | bitmap_start = offset - block_group->key.objectid; |
| 508 | bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap); |
| 509 | bitmap_start *= bytes_per_bitmap; |
| 510 | bitmap_start += block_group->key.objectid; |
| 511 | |
| 512 | return bitmap_start; |
| 513 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 514 | |
| 515 | static int tree_insert_offset(struct rb_root *root, u64 offset, |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 516 | struct rb_node *node, int bitmap) |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 517 | { |
| 518 | struct rb_node **p = &root->rb_node; |
| 519 | struct rb_node *parent = NULL; |
| 520 | struct btrfs_free_space *info; |
| 521 | |
| 522 | while (*p) { |
| 523 | parent = *p; |
| 524 | info = rb_entry(parent, struct btrfs_free_space, offset_index); |
| 525 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 526 | if (offset < info->offset) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 527 | p = &(*p)->rb_left; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 528 | } else if (offset > info->offset) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 529 | p = &(*p)->rb_right; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 530 | } else { |
| 531 | /* |
| 532 | * we could have a bitmap entry and an extent entry |
| 533 | * share the same offset. If this is the case, we want |
| 534 | * the extent entry to always be found first if we do a |
| 535 | * linear search through the tree, since we want to have |
| 536 | * the quickest allocation time, and allocating from an |
| 537 | * extent is faster than allocating from a bitmap. So |
| 538 | * if we're inserting a bitmap and we find an entry at |
| 539 | * this offset, we want to go right, or after this entry |
| 540 | * logically. If we are inserting an extent and we've |
| 541 | * found a bitmap, we want to go left, or before |
| 542 | * logically. |
| 543 | */ |
| 544 | if (bitmap) { |
| 545 | WARN_ON(info->bitmap); |
| 546 | p = &(*p)->rb_right; |
| 547 | } else { |
| 548 | WARN_ON(!info->bitmap); |
| 549 | p = &(*p)->rb_left; |
| 550 | } |
| 551 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 552 | } |
| 553 | |
| 554 | rb_link_node(node, parent, p); |
| 555 | rb_insert_color(node, root); |
| 556 | |
| 557 | return 0; |
| 558 | } |
| 559 | |
| 560 | /* |
Josef Bacik | 70cb074 | 2009-04-03 10:14:19 -0400 | [diff] [blame] | 561 | * searches the tree for the given offset. |
| 562 | * |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 563 | * fuzzy - If this is set, then we are trying to make an allocation, and we just |
| 564 | * want a section that has at least bytes size and comes at or after the given |
| 565 | * offset. |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 566 | */ |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 567 | static struct btrfs_free_space * |
| 568 | tree_search_offset(struct btrfs_block_group_cache *block_group, |
| 569 | u64 offset, int bitmap_only, int fuzzy) |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 570 | { |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 571 | struct rb_node *n = block_group->free_space_offset.rb_node; |
| 572 | struct btrfs_free_space *entry, *prev = NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 573 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 574 | /* find entry that is closest to the 'offset' */ |
| 575 | while (1) { |
| 576 | if (!n) { |
| 577 | entry = NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 578 | break; |
| 579 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 580 | |
| 581 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
| 582 | prev = entry; |
| 583 | |
| 584 | if (offset < entry->offset) |
| 585 | n = n->rb_left; |
| 586 | else if (offset > entry->offset) |
| 587 | n = n->rb_right; |
| 588 | else |
| 589 | break; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 590 | } |
| 591 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 592 | if (bitmap_only) { |
| 593 | if (!entry) |
| 594 | return NULL; |
| 595 | if (entry->bitmap) |
| 596 | return entry; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 597 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 598 | /* |
| 599 | * bitmap entry and extent entry may share same offset, |
| 600 | * in that case, bitmap entry comes after extent entry. |
| 601 | */ |
| 602 | n = rb_next(n); |
| 603 | if (!n) |
| 604 | return NULL; |
| 605 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
| 606 | if (entry->offset != offset) |
| 607 | return NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 608 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 609 | WARN_ON(!entry->bitmap); |
| 610 | return entry; |
| 611 | } else if (entry) { |
| 612 | if (entry->bitmap) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 613 | /* |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 614 | * if previous extent entry covers the offset, |
| 615 | * we should return it instead of the bitmap entry |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 616 | */ |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 617 | n = &entry->offset_index; |
| 618 | while (1) { |
| 619 | n = rb_prev(n); |
| 620 | if (!n) |
| 621 | break; |
| 622 | prev = rb_entry(n, struct btrfs_free_space, |
| 623 | offset_index); |
| 624 | if (!prev->bitmap) { |
| 625 | if (prev->offset + prev->bytes > offset) |
| 626 | entry = prev; |
| 627 | break; |
| 628 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 629 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 630 | } |
| 631 | return entry; |
| 632 | } |
| 633 | |
| 634 | if (!prev) |
| 635 | return NULL; |
| 636 | |
| 637 | /* find last entry before the 'offset' */ |
| 638 | entry = prev; |
| 639 | if (entry->offset > offset) { |
| 640 | n = rb_prev(&entry->offset_index); |
| 641 | if (n) { |
| 642 | entry = rb_entry(n, struct btrfs_free_space, |
| 643 | offset_index); |
| 644 | BUG_ON(entry->offset > offset); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 645 | } else { |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 646 | if (fuzzy) |
| 647 | return entry; |
| 648 | else |
| 649 | return NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 650 | } |
| 651 | } |
| 652 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 653 | if (entry->bitmap) { |
| 654 | n = &entry->offset_index; |
| 655 | while (1) { |
| 656 | n = rb_prev(n); |
| 657 | if (!n) |
| 658 | break; |
| 659 | prev = rb_entry(n, struct btrfs_free_space, |
| 660 | offset_index); |
| 661 | if (!prev->bitmap) { |
| 662 | if (prev->offset + prev->bytes > offset) |
| 663 | return prev; |
| 664 | break; |
| 665 | } |
| 666 | } |
| 667 | if (entry->offset + BITS_PER_BITMAP * |
| 668 | block_group->sectorsize > offset) |
| 669 | return entry; |
| 670 | } else if (entry->offset + entry->bytes > offset) |
| 671 | return entry; |
| 672 | |
| 673 | if (!fuzzy) |
| 674 | return NULL; |
| 675 | |
| 676 | while (1) { |
| 677 | if (entry->bitmap) { |
| 678 | if (entry->offset + BITS_PER_BITMAP * |
| 679 | block_group->sectorsize > offset) |
| 680 | break; |
| 681 | } else { |
| 682 | if (entry->offset + entry->bytes > offset) |
| 683 | break; |
| 684 | } |
| 685 | |
| 686 | n = rb_next(&entry->offset_index); |
| 687 | if (!n) |
| 688 | return NULL; |
| 689 | entry = rb_entry(n, struct btrfs_free_space, offset_index); |
| 690 | } |
| 691 | return entry; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 692 | } |
| 693 | |
| 694 | static void unlink_free_space(struct btrfs_block_group_cache *block_group, |
| 695 | struct btrfs_free_space *info) |
| 696 | { |
| 697 | rb_erase(&info->offset_index, &block_group->free_space_offset); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 698 | block_group->free_extents--; |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 699 | block_group->free_space -= info->bytes; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 700 | } |
| 701 | |
| 702 | static int link_free_space(struct btrfs_block_group_cache *block_group, |
| 703 | struct btrfs_free_space *info) |
| 704 | { |
| 705 | int ret = 0; |
| 706 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 707 | BUG_ON(!info->bitmap && !info->bytes); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 708 | ret = tree_insert_offset(&block_group->free_space_offset, info->offset, |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 709 | &info->offset_index, (info->bitmap != NULL)); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 710 | if (ret) |
| 711 | return ret; |
| 712 | |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 713 | block_group->free_space += info->bytes; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 714 | block_group->free_extents++; |
| 715 | return ret; |
| 716 | } |
| 717 | |
| 718 | static void recalculate_thresholds(struct btrfs_block_group_cache *block_group) |
| 719 | { |
Josef Bacik | 25891f7 | 2009-09-11 16:11:20 -0400 | [diff] [blame] | 720 | u64 max_bytes; |
| 721 | u64 bitmap_bytes; |
| 722 | u64 extent_bytes; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 723 | |
| 724 | /* |
| 725 | * The goal is to keep the total amount of memory used per 1gb of space |
| 726 | * at or below 32k, so we need to adjust how much memory we allow to be |
| 727 | * used by extent based free space tracking |
| 728 | */ |
| 729 | max_bytes = MAX_CACHE_BYTES_PER_GIG * |
| 730 | (div64_u64(block_group->key.offset, 1024 * 1024 * 1024)); |
| 731 | |
Josef Bacik | 25891f7 | 2009-09-11 16:11:20 -0400 | [diff] [blame] | 732 | /* |
| 733 | * we want to account for 1 more bitmap than what we have so we can make |
| 734 | * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as |
| 735 | * we add more bitmaps. |
| 736 | */ |
| 737 | bitmap_bytes = (block_group->total_bitmaps + 1) * PAGE_CACHE_SIZE; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 738 | |
Josef Bacik | 25891f7 | 2009-09-11 16:11:20 -0400 | [diff] [blame] | 739 | if (bitmap_bytes >= max_bytes) { |
| 740 | block_group->extents_thresh = 0; |
| 741 | return; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 742 | } |
Josef Bacik | 25891f7 | 2009-09-11 16:11:20 -0400 | [diff] [blame] | 743 | |
| 744 | /* |
| 745 | * we want the extent entry threshold to always be at most 1/2 the maxw |
| 746 | * bytes we can have, or whatever is less than that. |
| 747 | */ |
| 748 | extent_bytes = max_bytes - bitmap_bytes; |
| 749 | extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2)); |
| 750 | |
| 751 | block_group->extents_thresh = |
| 752 | div64_u64(extent_bytes, (sizeof(struct btrfs_free_space))); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 753 | } |
| 754 | |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 755 | static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group, |
| 756 | struct btrfs_free_space *info, u64 offset, |
| 757 | u64 bytes) |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 758 | { |
| 759 | unsigned long start, end; |
| 760 | unsigned long i; |
| 761 | |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 762 | start = offset_to_bit(info->offset, block_group->sectorsize, offset); |
| 763 | end = start + bytes_to_bits(bytes, block_group->sectorsize); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 764 | BUG_ON(end > BITS_PER_BITMAP); |
| 765 | |
| 766 | for (i = start; i < end; i++) |
| 767 | clear_bit(i, info->bitmap); |
| 768 | |
| 769 | info->bytes -= bytes; |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 770 | block_group->free_space -= bytes; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 771 | } |
| 772 | |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 773 | static void bitmap_set_bits(struct btrfs_block_group_cache *block_group, |
| 774 | struct btrfs_free_space *info, u64 offset, |
| 775 | u64 bytes) |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 776 | { |
| 777 | unsigned long start, end; |
| 778 | unsigned long i; |
| 779 | |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 780 | start = offset_to_bit(info->offset, block_group->sectorsize, offset); |
| 781 | end = start + bytes_to_bits(bytes, block_group->sectorsize); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 782 | BUG_ON(end > BITS_PER_BITMAP); |
| 783 | |
| 784 | for (i = start; i < end; i++) |
| 785 | set_bit(i, info->bitmap); |
| 786 | |
| 787 | info->bytes += bytes; |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 788 | block_group->free_space += bytes; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 789 | } |
| 790 | |
| 791 | static int search_bitmap(struct btrfs_block_group_cache *block_group, |
| 792 | struct btrfs_free_space *bitmap_info, u64 *offset, |
| 793 | u64 *bytes) |
| 794 | { |
| 795 | unsigned long found_bits = 0; |
| 796 | unsigned long bits, i; |
| 797 | unsigned long next_zero; |
| 798 | |
| 799 | i = offset_to_bit(bitmap_info->offset, block_group->sectorsize, |
| 800 | max_t(u64, *offset, bitmap_info->offset)); |
| 801 | bits = bytes_to_bits(*bytes, block_group->sectorsize); |
| 802 | |
| 803 | for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i); |
| 804 | i < BITS_PER_BITMAP; |
| 805 | i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) { |
| 806 | next_zero = find_next_zero_bit(bitmap_info->bitmap, |
| 807 | BITS_PER_BITMAP, i); |
| 808 | if ((next_zero - i) >= bits) { |
| 809 | found_bits = next_zero - i; |
| 810 | break; |
| 811 | } |
| 812 | i = next_zero; |
| 813 | } |
| 814 | |
| 815 | if (found_bits) { |
| 816 | *offset = (u64)(i * block_group->sectorsize) + |
| 817 | bitmap_info->offset; |
| 818 | *bytes = (u64)(found_bits) * block_group->sectorsize; |
| 819 | return 0; |
| 820 | } |
| 821 | |
| 822 | return -1; |
| 823 | } |
| 824 | |
| 825 | static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache |
| 826 | *block_group, u64 *offset, |
| 827 | u64 *bytes, int debug) |
| 828 | { |
| 829 | struct btrfs_free_space *entry; |
| 830 | struct rb_node *node; |
| 831 | int ret; |
| 832 | |
| 833 | if (!block_group->free_space_offset.rb_node) |
| 834 | return NULL; |
| 835 | |
| 836 | entry = tree_search_offset(block_group, |
| 837 | offset_to_bitmap(block_group, *offset), |
| 838 | 0, 1); |
| 839 | if (!entry) |
| 840 | return NULL; |
| 841 | |
| 842 | for (node = &entry->offset_index; node; node = rb_next(node)) { |
| 843 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
| 844 | if (entry->bytes < *bytes) |
| 845 | continue; |
| 846 | |
| 847 | if (entry->bitmap) { |
| 848 | ret = search_bitmap(block_group, entry, offset, bytes); |
| 849 | if (!ret) |
| 850 | return entry; |
| 851 | continue; |
| 852 | } |
| 853 | |
| 854 | *offset = entry->offset; |
| 855 | *bytes = entry->bytes; |
| 856 | return entry; |
| 857 | } |
| 858 | |
| 859 | return NULL; |
| 860 | } |
| 861 | |
| 862 | static void add_new_bitmap(struct btrfs_block_group_cache *block_group, |
| 863 | struct btrfs_free_space *info, u64 offset) |
| 864 | { |
| 865 | u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize; |
| 866 | int max_bitmaps = (int)div64_u64(block_group->key.offset + |
| 867 | bytes_per_bg - 1, bytes_per_bg); |
| 868 | BUG_ON(block_group->total_bitmaps >= max_bitmaps); |
| 869 | |
| 870 | info->offset = offset_to_bitmap(block_group, offset); |
Josef Bacik | f019f42 | 2009-09-11 16:11:20 -0400 | [diff] [blame] | 871 | info->bytes = 0; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 872 | link_free_space(block_group, info); |
| 873 | block_group->total_bitmaps++; |
| 874 | |
| 875 | recalculate_thresholds(block_group); |
| 876 | } |
| 877 | |
| 878 | static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group, |
| 879 | struct btrfs_free_space *bitmap_info, |
| 880 | u64 *offset, u64 *bytes) |
| 881 | { |
| 882 | u64 end; |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 883 | u64 search_start, search_bytes; |
| 884 | int ret; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 885 | |
| 886 | again: |
| 887 | end = bitmap_info->offset + |
| 888 | (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1; |
| 889 | |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 890 | /* |
| 891 | * XXX - this can go away after a few releases. |
| 892 | * |
| 893 | * since the only user of btrfs_remove_free_space is the tree logging |
| 894 | * stuff, and the only way to test that is under crash conditions, we |
| 895 | * want to have this debug stuff here just in case somethings not |
| 896 | * working. Search the bitmap for the space we are trying to use to |
| 897 | * make sure its actually there. If its not there then we need to stop |
| 898 | * because something has gone wrong. |
| 899 | */ |
| 900 | search_start = *offset; |
| 901 | search_bytes = *bytes; |
| 902 | ret = search_bitmap(block_group, bitmap_info, &search_start, |
| 903 | &search_bytes); |
| 904 | BUG_ON(ret < 0 || search_start != *offset); |
| 905 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 906 | if (*offset > bitmap_info->offset && *offset + *bytes > end) { |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 907 | bitmap_clear_bits(block_group, bitmap_info, *offset, |
| 908 | end - *offset + 1); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 909 | *bytes -= end - *offset + 1; |
| 910 | *offset = end + 1; |
| 911 | } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) { |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 912 | bitmap_clear_bits(block_group, bitmap_info, *offset, *bytes); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 913 | *bytes = 0; |
| 914 | } |
| 915 | |
| 916 | if (*bytes) { |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 917 | struct rb_node *next = rb_next(&bitmap_info->offset_index); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 918 | if (!bitmap_info->bytes) { |
| 919 | unlink_free_space(block_group, bitmap_info); |
| 920 | kfree(bitmap_info->bitmap); |
| 921 | kfree(bitmap_info); |
| 922 | block_group->total_bitmaps--; |
| 923 | recalculate_thresholds(block_group); |
| 924 | } |
| 925 | |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 926 | /* |
| 927 | * no entry after this bitmap, but we still have bytes to |
| 928 | * remove, so something has gone wrong. |
| 929 | */ |
| 930 | if (!next) |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 931 | return -EINVAL; |
| 932 | |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 933 | bitmap_info = rb_entry(next, struct btrfs_free_space, |
| 934 | offset_index); |
| 935 | |
| 936 | /* |
| 937 | * if the next entry isn't a bitmap we need to return to let the |
| 938 | * extent stuff do its work. |
| 939 | */ |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 940 | if (!bitmap_info->bitmap) |
| 941 | return -EAGAIN; |
| 942 | |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 943 | /* |
| 944 | * Ok the next item is a bitmap, but it may not actually hold |
| 945 | * the information for the rest of this free space stuff, so |
| 946 | * look for it, and if we don't find it return so we can try |
| 947 | * everything over again. |
| 948 | */ |
| 949 | search_start = *offset; |
| 950 | search_bytes = *bytes; |
| 951 | ret = search_bitmap(block_group, bitmap_info, &search_start, |
| 952 | &search_bytes); |
| 953 | if (ret < 0 || search_start != *offset) |
| 954 | return -EAGAIN; |
| 955 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 956 | goto again; |
| 957 | } else if (!bitmap_info->bytes) { |
| 958 | unlink_free_space(block_group, bitmap_info); |
| 959 | kfree(bitmap_info->bitmap); |
| 960 | kfree(bitmap_info); |
| 961 | block_group->total_bitmaps--; |
| 962 | recalculate_thresholds(block_group); |
| 963 | } |
| 964 | |
| 965 | return 0; |
| 966 | } |
| 967 | |
| 968 | static int insert_into_bitmap(struct btrfs_block_group_cache *block_group, |
| 969 | struct btrfs_free_space *info) |
| 970 | { |
| 971 | struct btrfs_free_space *bitmap_info; |
| 972 | int added = 0; |
| 973 | u64 bytes, offset, end; |
| 974 | int ret; |
| 975 | |
| 976 | /* |
| 977 | * If we are below the extents threshold then we can add this as an |
| 978 | * extent, and don't have to deal with the bitmap |
| 979 | */ |
| 980 | if (block_group->free_extents < block_group->extents_thresh && |
| 981 | info->bytes > block_group->sectorsize * 4) |
| 982 | return 0; |
| 983 | |
| 984 | /* |
| 985 | * some block groups are so tiny they can't be enveloped by a bitmap, so |
| 986 | * don't even bother to create a bitmap for this |
| 987 | */ |
| 988 | if (BITS_PER_BITMAP * block_group->sectorsize > |
| 989 | block_group->key.offset) |
| 990 | return 0; |
| 991 | |
| 992 | bytes = info->bytes; |
| 993 | offset = info->offset; |
| 994 | |
| 995 | again: |
| 996 | bitmap_info = tree_search_offset(block_group, |
| 997 | offset_to_bitmap(block_group, offset), |
| 998 | 1, 0); |
| 999 | if (!bitmap_info) { |
| 1000 | BUG_ON(added); |
| 1001 | goto new_bitmap; |
| 1002 | } |
| 1003 | |
| 1004 | end = bitmap_info->offset + |
| 1005 | (u64)(BITS_PER_BITMAP * block_group->sectorsize); |
| 1006 | |
| 1007 | if (offset >= bitmap_info->offset && offset + bytes > end) { |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1008 | bitmap_set_bits(block_group, bitmap_info, offset, |
| 1009 | end - offset); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1010 | bytes -= end - offset; |
| 1011 | offset = end; |
| 1012 | added = 0; |
| 1013 | } else if (offset >= bitmap_info->offset && offset + bytes <= end) { |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1014 | bitmap_set_bits(block_group, bitmap_info, offset, bytes); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1015 | bytes = 0; |
| 1016 | } else { |
| 1017 | BUG(); |
| 1018 | } |
| 1019 | |
| 1020 | if (!bytes) { |
| 1021 | ret = 1; |
| 1022 | goto out; |
| 1023 | } else |
| 1024 | goto again; |
| 1025 | |
| 1026 | new_bitmap: |
| 1027 | if (info && info->bitmap) { |
| 1028 | add_new_bitmap(block_group, info, offset); |
| 1029 | added = 1; |
| 1030 | info = NULL; |
| 1031 | goto again; |
| 1032 | } else { |
| 1033 | spin_unlock(&block_group->tree_lock); |
| 1034 | |
| 1035 | /* no pre-allocated info, allocate a new one */ |
| 1036 | if (!info) { |
| 1037 | info = kzalloc(sizeof(struct btrfs_free_space), |
| 1038 | GFP_NOFS); |
| 1039 | if (!info) { |
| 1040 | spin_lock(&block_group->tree_lock); |
| 1041 | ret = -ENOMEM; |
| 1042 | goto out; |
| 1043 | } |
| 1044 | } |
| 1045 | |
| 1046 | /* allocate the bitmap */ |
| 1047 | info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); |
| 1048 | spin_lock(&block_group->tree_lock); |
| 1049 | if (!info->bitmap) { |
| 1050 | ret = -ENOMEM; |
| 1051 | goto out; |
| 1052 | } |
| 1053 | goto again; |
| 1054 | } |
| 1055 | |
| 1056 | out: |
| 1057 | if (info) { |
| 1058 | if (info->bitmap) |
| 1059 | kfree(info->bitmap); |
| 1060 | kfree(info); |
| 1061 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1062 | |
| 1063 | return ret; |
| 1064 | } |
| 1065 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1066 | int btrfs_add_free_space(struct btrfs_block_group_cache *block_group, |
| 1067 | u64 offset, u64 bytes) |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1068 | { |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1069 | struct btrfs_free_space *right_info = NULL; |
| 1070 | struct btrfs_free_space *left_info = NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1071 | struct btrfs_free_space *info = NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1072 | int ret = 0; |
| 1073 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1074 | info = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS); |
| 1075 | if (!info) |
| 1076 | return -ENOMEM; |
| 1077 | |
| 1078 | info->offset = offset; |
| 1079 | info->bytes = bytes; |
| 1080 | |
| 1081 | spin_lock(&block_group->tree_lock); |
| 1082 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1083 | /* |
| 1084 | * first we want to see if there is free space adjacent to the range we |
| 1085 | * are adding, if there is remove that struct and add a new one to |
| 1086 | * cover the entire range |
| 1087 | */ |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1088 | right_info = tree_search_offset(block_group, offset + bytes, 0, 0); |
| 1089 | if (right_info && rb_prev(&right_info->offset_index)) |
| 1090 | left_info = rb_entry(rb_prev(&right_info->offset_index), |
| 1091 | struct btrfs_free_space, offset_index); |
| 1092 | else |
| 1093 | left_info = tree_search_offset(block_group, offset - 1, 0, 0); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1094 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1095 | /* |
| 1096 | * If there was no extent directly to the left or right of this new |
| 1097 | * extent then we know we're going to have to allocate a new extent, so |
| 1098 | * before we do that see if we need to drop this into a bitmap |
| 1099 | */ |
| 1100 | if ((!left_info || left_info->bitmap) && |
| 1101 | (!right_info || right_info->bitmap)) { |
| 1102 | ret = insert_into_bitmap(block_group, info); |
| 1103 | |
| 1104 | if (ret < 0) { |
| 1105 | goto out; |
| 1106 | } else if (ret) { |
| 1107 | ret = 0; |
| 1108 | goto out; |
| 1109 | } |
| 1110 | } |
| 1111 | |
| 1112 | if (right_info && !right_info->bitmap) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1113 | unlink_free_space(block_group, right_info); |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1114 | info->bytes += right_info->bytes; |
| 1115 | kfree(right_info); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1116 | } |
| 1117 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1118 | if (left_info && !left_info->bitmap && |
| 1119 | left_info->offset + left_info->bytes == offset) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1120 | unlink_free_space(block_group, left_info); |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1121 | info->offset = left_info->offset; |
| 1122 | info->bytes += left_info->bytes; |
| 1123 | kfree(left_info); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1124 | } |
| 1125 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1126 | ret = link_free_space(block_group, info); |
| 1127 | if (ret) |
| 1128 | kfree(info); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1129 | out: |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1130 | spin_unlock(&block_group->tree_lock); |
| 1131 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1132 | if (ret) { |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1133 | printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret); |
Stoyan Gaydarov | c293498 | 2009-04-02 17:05:11 -0400 | [diff] [blame] | 1134 | BUG_ON(ret == -EEXIST); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1135 | } |
| 1136 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1137 | return ret; |
| 1138 | } |
| 1139 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1140 | int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group, |
| 1141 | u64 offset, u64 bytes) |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1142 | { |
| 1143 | struct btrfs_free_space *info; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1144 | struct btrfs_free_space *next_info = NULL; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1145 | int ret = 0; |
| 1146 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1147 | spin_lock(&block_group->tree_lock); |
| 1148 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1149 | again: |
| 1150 | info = tree_search_offset(block_group, offset, 0, 0); |
| 1151 | if (!info) { |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 1152 | /* |
| 1153 | * oops didn't find an extent that matched the space we wanted |
| 1154 | * to remove, look for a bitmap instead |
| 1155 | */ |
| 1156 | info = tree_search_offset(block_group, |
| 1157 | offset_to_bitmap(block_group, offset), |
| 1158 | 1, 0); |
| 1159 | if (!info) { |
| 1160 | WARN_ON(1); |
| 1161 | goto out_lock; |
| 1162 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1163 | } |
| 1164 | |
| 1165 | if (info->bytes < bytes && rb_next(&info->offset_index)) { |
| 1166 | u64 end; |
| 1167 | next_info = rb_entry(rb_next(&info->offset_index), |
| 1168 | struct btrfs_free_space, |
| 1169 | offset_index); |
| 1170 | |
| 1171 | if (next_info->bitmap) |
| 1172 | end = next_info->offset + BITS_PER_BITMAP * |
| 1173 | block_group->sectorsize - 1; |
| 1174 | else |
| 1175 | end = next_info->offset + next_info->bytes; |
| 1176 | |
| 1177 | if (next_info->bytes < bytes || |
| 1178 | next_info->offset > offset || offset > end) { |
| 1179 | printk(KERN_CRIT "Found free space at %llu, size %llu," |
| 1180 | " trying to use %llu\n", |
| 1181 | (unsigned long long)info->offset, |
| 1182 | (unsigned long long)info->bytes, |
| 1183 | (unsigned long long)bytes); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1184 | WARN_ON(1); |
| 1185 | ret = -EINVAL; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1186 | goto out_lock; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1187 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1188 | |
| 1189 | info = next_info; |
| 1190 | } |
| 1191 | |
| 1192 | if (info->bytes == bytes) { |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1193 | unlink_free_space(block_group, info); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1194 | if (info->bitmap) { |
| 1195 | kfree(info->bitmap); |
| 1196 | block_group->total_bitmaps--; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1197 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1198 | kfree(info); |
| 1199 | goto out_lock; |
| 1200 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1201 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1202 | if (!info->bitmap && info->offset == offset) { |
| 1203 | unlink_free_space(block_group, info); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1204 | info->offset += bytes; |
| 1205 | info->bytes -= bytes; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1206 | link_free_space(block_group, info); |
| 1207 | goto out_lock; |
| 1208 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1209 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1210 | if (!info->bitmap && info->offset <= offset && |
| 1211 | info->offset + info->bytes >= offset + bytes) { |
Chris Mason | 9b49c9b | 2008-09-24 11:23:25 -0400 | [diff] [blame] | 1212 | u64 old_start = info->offset; |
| 1213 | /* |
| 1214 | * we're freeing space in the middle of the info, |
| 1215 | * this can happen during tree log replay |
| 1216 | * |
| 1217 | * first unlink the old info and then |
| 1218 | * insert it again after the hole we're creating |
| 1219 | */ |
| 1220 | unlink_free_space(block_group, info); |
| 1221 | if (offset + bytes < info->offset + info->bytes) { |
| 1222 | u64 old_end = info->offset + info->bytes; |
| 1223 | |
| 1224 | info->offset = offset + bytes; |
| 1225 | info->bytes = old_end - info->offset; |
| 1226 | ret = link_free_space(block_group, info); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1227 | WARN_ON(ret); |
| 1228 | if (ret) |
| 1229 | goto out_lock; |
Chris Mason | 9b49c9b | 2008-09-24 11:23:25 -0400 | [diff] [blame] | 1230 | } else { |
| 1231 | /* the hole we're creating ends at the end |
| 1232 | * of the info struct, just free the info |
| 1233 | */ |
| 1234 | kfree(info); |
| 1235 | } |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1236 | spin_unlock(&block_group->tree_lock); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1237 | |
| 1238 | /* step two, insert a new info struct to cover |
| 1239 | * anything before the hole |
Chris Mason | 9b49c9b | 2008-09-24 11:23:25 -0400 | [diff] [blame] | 1240 | */ |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1241 | ret = btrfs_add_free_space(block_group, old_start, |
| 1242 | offset - old_start); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1243 | WARN_ON(ret); |
| 1244 | goto out; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1245 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1246 | |
| 1247 | ret = remove_from_bitmap(block_group, info, &offset, &bytes); |
| 1248 | if (ret == -EAGAIN) |
| 1249 | goto again; |
| 1250 | BUG_ON(ret); |
| 1251 | out_lock: |
| 1252 | spin_unlock(&block_group->tree_lock); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1253 | out: |
Josef Bacik | 2517920 | 2008-10-29 14:49:05 -0400 | [diff] [blame] | 1254 | return ret; |
| 1255 | } |
| 1256 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1257 | void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group, |
| 1258 | u64 bytes) |
| 1259 | { |
| 1260 | struct btrfs_free_space *info; |
| 1261 | struct rb_node *n; |
| 1262 | int count = 0; |
| 1263 | |
| 1264 | for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) { |
| 1265 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
| 1266 | if (info->bytes >= bytes) |
| 1267 | count++; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1268 | printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n", |
Joel Becker | 2138093 | 2009-04-21 12:38:29 -0700 | [diff] [blame] | 1269 | (unsigned long long)info->offset, |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1270 | (unsigned long long)info->bytes, |
| 1271 | (info->bitmap) ? "yes" : "no"); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1272 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1273 | printk(KERN_INFO "block group has cluster?: %s\n", |
| 1274 | list_empty(&block_group->cluster_list) ? "no" : "yes"); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1275 | printk(KERN_INFO "%d blocks of free space at or bigger than bytes is" |
| 1276 | "\n", count); |
| 1277 | } |
| 1278 | |
| 1279 | u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group) |
| 1280 | { |
| 1281 | struct btrfs_free_space *info; |
| 1282 | struct rb_node *n; |
| 1283 | u64 ret = 0; |
| 1284 | |
| 1285 | for (n = rb_first(&block_group->free_space_offset); n; |
| 1286 | n = rb_next(n)) { |
| 1287 | info = rb_entry(n, struct btrfs_free_space, offset_index); |
| 1288 | ret += info->bytes; |
| 1289 | } |
| 1290 | |
| 1291 | return ret; |
| 1292 | } |
| 1293 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1294 | /* |
| 1295 | * for a given cluster, put all of its extents back into the free |
| 1296 | * space cache. If the block group passed doesn't match the block group |
| 1297 | * pointed to by the cluster, someone else raced in and freed the |
| 1298 | * cluster already. In that case, we just return without changing anything |
| 1299 | */ |
| 1300 | static int |
| 1301 | __btrfs_return_cluster_to_free_space( |
| 1302 | struct btrfs_block_group_cache *block_group, |
| 1303 | struct btrfs_free_cluster *cluster) |
| 1304 | { |
| 1305 | struct btrfs_free_space *entry; |
| 1306 | struct rb_node *node; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1307 | bool bitmap; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1308 | |
| 1309 | spin_lock(&cluster->lock); |
| 1310 | if (cluster->block_group != block_group) |
| 1311 | goto out; |
| 1312 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1313 | bitmap = cluster->points_to_bitmap; |
| 1314 | cluster->block_group = NULL; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1315 | cluster->window_start = 0; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1316 | list_del_init(&cluster->block_group_list); |
| 1317 | cluster->points_to_bitmap = false; |
| 1318 | |
| 1319 | if (bitmap) |
| 1320 | goto out; |
| 1321 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1322 | node = rb_first(&cluster->root); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1323 | while (node) { |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1324 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
| 1325 | node = rb_next(&entry->offset_index); |
| 1326 | rb_erase(&entry->offset_index, &cluster->root); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1327 | BUG_ON(entry->bitmap); |
| 1328 | tree_insert_offset(&block_group->free_space_offset, |
| 1329 | entry->offset, &entry->offset_index, 0); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1330 | } |
Eric Paris | 6bef4d3 | 2010-02-23 19:43:04 +0000 | [diff] [blame] | 1331 | cluster->root = RB_ROOT; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1332 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1333 | out: |
| 1334 | spin_unlock(&cluster->lock); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1335 | btrfs_put_block_group(block_group); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1336 | return 0; |
| 1337 | } |
| 1338 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1339 | void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group) |
| 1340 | { |
| 1341 | struct btrfs_free_space *info; |
| 1342 | struct rb_node *node; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1343 | struct btrfs_free_cluster *cluster; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1344 | struct list_head *head; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1345 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1346 | spin_lock(&block_group->tree_lock); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1347 | while ((head = block_group->cluster_list.next) != |
| 1348 | &block_group->cluster_list) { |
| 1349 | cluster = list_entry(head, struct btrfs_free_cluster, |
| 1350 | block_group_list); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1351 | |
| 1352 | WARN_ON(cluster->block_group != block_group); |
| 1353 | __btrfs_return_cluster_to_free_space(block_group, cluster); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1354 | if (need_resched()) { |
| 1355 | spin_unlock(&block_group->tree_lock); |
| 1356 | cond_resched(); |
| 1357 | spin_lock(&block_group->tree_lock); |
| 1358 | } |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1359 | } |
| 1360 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1361 | while ((node = rb_last(&block_group->free_space_offset)) != NULL) { |
| 1362 | info = rb_entry(node, struct btrfs_free_space, offset_index); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1363 | unlink_free_space(block_group, info); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1364 | if (info->bitmap) |
| 1365 | kfree(info->bitmap); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1366 | kfree(info); |
| 1367 | if (need_resched()) { |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1368 | spin_unlock(&block_group->tree_lock); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1369 | cond_resched(); |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1370 | spin_lock(&block_group->tree_lock); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1371 | } |
| 1372 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1373 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1374 | spin_unlock(&block_group->tree_lock); |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1375 | } |
| 1376 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1377 | u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group, |
| 1378 | u64 offset, u64 bytes, u64 empty_size) |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1379 | { |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1380 | struct btrfs_free_space *entry = NULL; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1381 | u64 bytes_search = bytes + empty_size; |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1382 | u64 ret = 0; |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1383 | |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1384 | spin_lock(&block_group->tree_lock); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1385 | entry = find_free_space(block_group, &offset, &bytes_search, 0); |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1386 | if (!entry) |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1387 | goto out; |
| 1388 | |
| 1389 | ret = offset; |
| 1390 | if (entry->bitmap) { |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1391 | bitmap_clear_bits(block_group, entry, offset, bytes); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1392 | if (!entry->bytes) { |
| 1393 | unlink_free_space(block_group, entry); |
| 1394 | kfree(entry->bitmap); |
| 1395 | kfree(entry); |
| 1396 | block_group->total_bitmaps--; |
| 1397 | recalculate_thresholds(block_group); |
| 1398 | } |
| 1399 | } else { |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1400 | unlink_free_space(block_group, entry); |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1401 | entry->offset += bytes; |
| 1402 | entry->bytes -= bytes; |
Josef Bacik | 6226cb0a | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 1403 | if (!entry->bytes) |
| 1404 | kfree(entry); |
| 1405 | else |
| 1406 | link_free_space(block_group, entry); |
| 1407 | } |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1408 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1409 | out: |
| 1410 | spin_unlock(&block_group->tree_lock); |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1411 | |
Josef Bacik | 0f9dd46 | 2008-09-23 13:14:11 -0400 | [diff] [blame] | 1412 | return ret; |
| 1413 | } |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1414 | |
| 1415 | /* |
| 1416 | * given a cluster, put all of its extents back into the free space |
| 1417 | * cache. If a block group is passed, this function will only free |
| 1418 | * a cluster that belongs to the passed block group. |
| 1419 | * |
| 1420 | * Otherwise, it'll get a reference on the block group pointed to by the |
| 1421 | * cluster and remove the cluster from it. |
| 1422 | */ |
| 1423 | int btrfs_return_cluster_to_free_space( |
| 1424 | struct btrfs_block_group_cache *block_group, |
| 1425 | struct btrfs_free_cluster *cluster) |
| 1426 | { |
| 1427 | int ret; |
| 1428 | |
| 1429 | /* first, get a safe pointer to the block group */ |
| 1430 | spin_lock(&cluster->lock); |
| 1431 | if (!block_group) { |
| 1432 | block_group = cluster->block_group; |
| 1433 | if (!block_group) { |
| 1434 | spin_unlock(&cluster->lock); |
| 1435 | return 0; |
| 1436 | } |
| 1437 | } else if (cluster->block_group != block_group) { |
| 1438 | /* someone else has already freed it don't redo their work */ |
| 1439 | spin_unlock(&cluster->lock); |
| 1440 | return 0; |
| 1441 | } |
| 1442 | atomic_inc(&block_group->count); |
| 1443 | spin_unlock(&cluster->lock); |
| 1444 | |
| 1445 | /* now return any extents the cluster had on it */ |
| 1446 | spin_lock(&block_group->tree_lock); |
| 1447 | ret = __btrfs_return_cluster_to_free_space(block_group, cluster); |
| 1448 | spin_unlock(&block_group->tree_lock); |
| 1449 | |
| 1450 | /* finally drop our ref */ |
| 1451 | btrfs_put_block_group(block_group); |
| 1452 | return ret; |
| 1453 | } |
| 1454 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1455 | static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group, |
| 1456 | struct btrfs_free_cluster *cluster, |
| 1457 | u64 bytes, u64 min_start) |
| 1458 | { |
| 1459 | struct btrfs_free_space *entry; |
| 1460 | int err; |
| 1461 | u64 search_start = cluster->window_start; |
| 1462 | u64 search_bytes = bytes; |
| 1463 | u64 ret = 0; |
| 1464 | |
| 1465 | spin_lock(&block_group->tree_lock); |
| 1466 | spin_lock(&cluster->lock); |
| 1467 | |
| 1468 | if (!cluster->points_to_bitmap) |
| 1469 | goto out; |
| 1470 | |
| 1471 | if (cluster->block_group != block_group) |
| 1472 | goto out; |
| 1473 | |
Josef Bacik | 6606bb9 | 2009-07-31 11:03:58 -0400 | [diff] [blame] | 1474 | /* |
| 1475 | * search_start is the beginning of the bitmap, but at some point it may |
| 1476 | * be a good idea to point to the actual start of the free area in the |
| 1477 | * bitmap, so do the offset_to_bitmap trick anyway, and set bitmap_only |
| 1478 | * to 1 to make sure we get the bitmap entry |
| 1479 | */ |
| 1480 | entry = tree_search_offset(block_group, |
| 1481 | offset_to_bitmap(block_group, search_start), |
| 1482 | 1, 0); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1483 | if (!entry || !entry->bitmap) |
| 1484 | goto out; |
| 1485 | |
| 1486 | search_start = min_start; |
| 1487 | search_bytes = bytes; |
| 1488 | |
| 1489 | err = search_bitmap(block_group, entry, &search_start, |
| 1490 | &search_bytes); |
| 1491 | if (err) |
| 1492 | goto out; |
| 1493 | |
| 1494 | ret = search_start; |
Josef Bacik | 817d52f | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1495 | bitmap_clear_bits(block_group, entry, ret, bytes); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1496 | out: |
| 1497 | spin_unlock(&cluster->lock); |
| 1498 | spin_unlock(&block_group->tree_lock); |
| 1499 | |
| 1500 | return ret; |
| 1501 | } |
| 1502 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1503 | /* |
| 1504 | * given a cluster, try to allocate 'bytes' from it, returns 0 |
| 1505 | * if it couldn't find anything suitably large, or a logical disk offset |
| 1506 | * if things worked out |
| 1507 | */ |
| 1508 | u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group, |
| 1509 | struct btrfs_free_cluster *cluster, u64 bytes, |
| 1510 | u64 min_start) |
| 1511 | { |
| 1512 | struct btrfs_free_space *entry = NULL; |
| 1513 | struct rb_node *node; |
| 1514 | u64 ret = 0; |
| 1515 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1516 | if (cluster->points_to_bitmap) |
| 1517 | return btrfs_alloc_from_bitmap(block_group, cluster, bytes, |
| 1518 | min_start); |
| 1519 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1520 | spin_lock(&cluster->lock); |
| 1521 | if (bytes > cluster->max_size) |
| 1522 | goto out; |
| 1523 | |
| 1524 | if (cluster->block_group != block_group) |
| 1525 | goto out; |
| 1526 | |
| 1527 | node = rb_first(&cluster->root); |
| 1528 | if (!node) |
| 1529 | goto out; |
| 1530 | |
| 1531 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
| 1532 | |
| 1533 | while(1) { |
| 1534 | if (entry->bytes < bytes || entry->offset < min_start) { |
| 1535 | struct rb_node *node; |
| 1536 | |
| 1537 | node = rb_next(&entry->offset_index); |
| 1538 | if (!node) |
| 1539 | break; |
| 1540 | entry = rb_entry(node, struct btrfs_free_space, |
| 1541 | offset_index); |
| 1542 | continue; |
| 1543 | } |
| 1544 | ret = entry->offset; |
| 1545 | |
| 1546 | entry->offset += bytes; |
| 1547 | entry->bytes -= bytes; |
| 1548 | |
| 1549 | if (entry->bytes == 0) { |
| 1550 | rb_erase(&entry->offset_index, &cluster->root); |
| 1551 | kfree(entry); |
| 1552 | } |
| 1553 | break; |
| 1554 | } |
| 1555 | out: |
| 1556 | spin_unlock(&cluster->lock); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1557 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1558 | return ret; |
| 1559 | } |
| 1560 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1561 | static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group, |
| 1562 | struct btrfs_free_space *entry, |
| 1563 | struct btrfs_free_cluster *cluster, |
| 1564 | u64 offset, u64 bytes, u64 min_bytes) |
| 1565 | { |
| 1566 | unsigned long next_zero; |
| 1567 | unsigned long i; |
| 1568 | unsigned long search_bits; |
| 1569 | unsigned long total_bits; |
| 1570 | unsigned long found_bits; |
| 1571 | unsigned long start = 0; |
| 1572 | unsigned long total_found = 0; |
| 1573 | bool found = false; |
| 1574 | |
| 1575 | i = offset_to_bit(entry->offset, block_group->sectorsize, |
| 1576 | max_t(u64, offset, entry->offset)); |
| 1577 | search_bits = bytes_to_bits(min_bytes, block_group->sectorsize); |
| 1578 | total_bits = bytes_to_bits(bytes, block_group->sectorsize); |
| 1579 | |
| 1580 | again: |
| 1581 | found_bits = 0; |
| 1582 | for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i); |
| 1583 | i < BITS_PER_BITMAP; |
| 1584 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) { |
| 1585 | next_zero = find_next_zero_bit(entry->bitmap, |
| 1586 | BITS_PER_BITMAP, i); |
| 1587 | if (next_zero - i >= search_bits) { |
| 1588 | found_bits = next_zero - i; |
| 1589 | break; |
| 1590 | } |
| 1591 | i = next_zero; |
| 1592 | } |
| 1593 | |
| 1594 | if (!found_bits) |
| 1595 | return -1; |
| 1596 | |
| 1597 | if (!found) { |
| 1598 | start = i; |
| 1599 | found = true; |
| 1600 | } |
| 1601 | |
| 1602 | total_found += found_bits; |
| 1603 | |
| 1604 | if (cluster->max_size < found_bits * block_group->sectorsize) |
| 1605 | cluster->max_size = found_bits * block_group->sectorsize; |
| 1606 | |
| 1607 | if (total_found < total_bits) { |
| 1608 | i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero); |
| 1609 | if (i - start > total_bits * 2) { |
| 1610 | total_found = 0; |
| 1611 | cluster->max_size = 0; |
| 1612 | found = false; |
| 1613 | } |
| 1614 | goto again; |
| 1615 | } |
| 1616 | |
| 1617 | cluster->window_start = start * block_group->sectorsize + |
| 1618 | entry->offset; |
| 1619 | cluster->points_to_bitmap = true; |
| 1620 | |
| 1621 | return 0; |
| 1622 | } |
| 1623 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1624 | /* |
| 1625 | * here we try to find a cluster of blocks in a block group. The goal |
| 1626 | * is to find at least bytes free and up to empty_size + bytes free. |
| 1627 | * We might not find them all in one contiguous area. |
| 1628 | * |
| 1629 | * returns zero and sets up cluster if things worked out, otherwise |
| 1630 | * it returns -enospc |
| 1631 | */ |
| 1632 | int btrfs_find_space_cluster(struct btrfs_trans_handle *trans, |
Chris Mason | 451d758 | 2009-06-09 20:28:34 -0400 | [diff] [blame] | 1633 | struct btrfs_root *root, |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1634 | struct btrfs_block_group_cache *block_group, |
| 1635 | struct btrfs_free_cluster *cluster, |
| 1636 | u64 offset, u64 bytes, u64 empty_size) |
| 1637 | { |
| 1638 | struct btrfs_free_space *entry = NULL; |
| 1639 | struct rb_node *node; |
| 1640 | struct btrfs_free_space *next; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1641 | struct btrfs_free_space *last = NULL; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1642 | u64 min_bytes; |
| 1643 | u64 window_start; |
| 1644 | u64 window_free; |
| 1645 | u64 max_extent = 0; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1646 | bool found_bitmap = false; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1647 | int ret; |
| 1648 | |
| 1649 | /* for metadata, allow allocates with more holes */ |
Chris Mason | 451d758 | 2009-06-09 20:28:34 -0400 | [diff] [blame] | 1650 | if (btrfs_test_opt(root, SSD_SPREAD)) { |
| 1651 | min_bytes = bytes + empty_size; |
| 1652 | } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) { |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1653 | /* |
| 1654 | * we want to do larger allocations when we are |
| 1655 | * flushing out the delayed refs, it helps prevent |
| 1656 | * making more work as we go along. |
| 1657 | */ |
| 1658 | if (trans->transaction->delayed_refs.flushing) |
| 1659 | min_bytes = max(bytes, (bytes + empty_size) >> 1); |
| 1660 | else |
| 1661 | min_bytes = max(bytes, (bytes + empty_size) >> 4); |
| 1662 | } else |
| 1663 | min_bytes = max(bytes, (bytes + empty_size) >> 2); |
| 1664 | |
| 1665 | spin_lock(&block_group->tree_lock); |
| 1666 | spin_lock(&cluster->lock); |
| 1667 | |
| 1668 | /* someone already found a cluster, hooray */ |
| 1669 | if (cluster->block_group) { |
| 1670 | ret = 0; |
| 1671 | goto out; |
| 1672 | } |
| 1673 | again: |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1674 | entry = tree_search_offset(block_group, offset, found_bitmap, 1); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1675 | if (!entry) { |
| 1676 | ret = -ENOSPC; |
| 1677 | goto out; |
| 1678 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1679 | |
| 1680 | /* |
| 1681 | * If found_bitmap is true, we exhausted our search for extent entries, |
| 1682 | * and we just want to search all of the bitmaps that we can find, and |
| 1683 | * ignore any extent entries we find. |
| 1684 | */ |
| 1685 | while (entry->bitmap || found_bitmap || |
| 1686 | (!entry->bitmap && entry->bytes < min_bytes)) { |
| 1687 | struct rb_node *node = rb_next(&entry->offset_index); |
| 1688 | |
| 1689 | if (entry->bitmap && entry->bytes > bytes + empty_size) { |
| 1690 | ret = btrfs_bitmap_cluster(block_group, entry, cluster, |
| 1691 | offset, bytes + empty_size, |
| 1692 | min_bytes); |
| 1693 | if (!ret) |
| 1694 | goto got_it; |
| 1695 | } |
| 1696 | |
| 1697 | if (!node) { |
| 1698 | ret = -ENOSPC; |
| 1699 | goto out; |
| 1700 | } |
| 1701 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
| 1702 | } |
| 1703 | |
| 1704 | /* |
| 1705 | * We already searched all the extent entries from the passed in offset |
| 1706 | * to the end and didn't find enough space for the cluster, and we also |
| 1707 | * didn't find any bitmaps that met our criteria, just go ahead and exit |
| 1708 | */ |
| 1709 | if (found_bitmap) { |
| 1710 | ret = -ENOSPC; |
| 1711 | goto out; |
| 1712 | } |
| 1713 | |
| 1714 | cluster->points_to_bitmap = false; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1715 | window_start = entry->offset; |
| 1716 | window_free = entry->bytes; |
| 1717 | last = entry; |
| 1718 | max_extent = entry->bytes; |
| 1719 | |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1720 | while (1) { |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1721 | /* out window is just right, lets fill it */ |
| 1722 | if (window_free >= bytes + empty_size) |
| 1723 | break; |
| 1724 | |
| 1725 | node = rb_next(&last->offset_index); |
| 1726 | if (!node) { |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1727 | if (found_bitmap) |
| 1728 | goto again; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1729 | ret = -ENOSPC; |
| 1730 | goto out; |
| 1731 | } |
| 1732 | next = rb_entry(node, struct btrfs_free_space, offset_index); |
| 1733 | |
| 1734 | /* |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1735 | * we found a bitmap, so if this search doesn't result in a |
| 1736 | * cluster, we know to go and search again for the bitmaps and |
| 1737 | * start looking for space there |
| 1738 | */ |
| 1739 | if (next->bitmap) { |
| 1740 | if (!found_bitmap) |
| 1741 | offset = next->offset; |
| 1742 | found_bitmap = true; |
| 1743 | last = next; |
| 1744 | continue; |
| 1745 | } |
| 1746 | |
| 1747 | /* |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1748 | * we haven't filled the empty size and the window is |
| 1749 | * very large. reset and try again |
| 1750 | */ |
Chris Mason | c604480 | 2009-06-09 18:35:15 -0400 | [diff] [blame] | 1751 | if (next->offset - (last->offset + last->bytes) > 128 * 1024 || |
| 1752 | next->offset - window_start > (bytes + empty_size) * 2) { |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1753 | entry = next; |
| 1754 | window_start = entry->offset; |
| 1755 | window_free = entry->bytes; |
| 1756 | last = entry; |
Josef Bacik | 01dea1e | 2009-11-10 21:23:48 -0500 | [diff] [blame] | 1757 | max_extent = entry->bytes; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1758 | } else { |
| 1759 | last = next; |
| 1760 | window_free += next->bytes; |
| 1761 | if (entry->bytes > max_extent) |
| 1762 | max_extent = entry->bytes; |
| 1763 | } |
| 1764 | } |
| 1765 | |
| 1766 | cluster->window_start = entry->offset; |
| 1767 | |
| 1768 | /* |
| 1769 | * now we've found our entries, pull them out of the free space |
| 1770 | * cache and put them into the cluster rbtree |
| 1771 | * |
| 1772 | * The cluster includes an rbtree, but only uses the offset index |
| 1773 | * of each free space cache entry. |
| 1774 | */ |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1775 | while (1) { |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1776 | node = rb_next(&entry->offset_index); |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1777 | if (entry->bitmap && node) { |
| 1778 | entry = rb_entry(node, struct btrfs_free_space, |
| 1779 | offset_index); |
| 1780 | continue; |
| 1781 | } else if (entry->bitmap && !node) { |
| 1782 | break; |
| 1783 | } |
| 1784 | |
| 1785 | rb_erase(&entry->offset_index, &block_group->free_space_offset); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1786 | ret = tree_insert_offset(&cluster->root, entry->offset, |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1787 | &entry->offset_index, 0); |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1788 | BUG_ON(ret); |
| 1789 | |
| 1790 | if (!node || entry == last) |
| 1791 | break; |
| 1792 | |
| 1793 | entry = rb_entry(node, struct btrfs_free_space, offset_index); |
| 1794 | } |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1795 | |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1796 | cluster->max_size = max_extent; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1797 | got_it: |
| 1798 | ret = 0; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1799 | atomic_inc(&block_group->count); |
| 1800 | list_add_tail(&cluster->block_group_list, &block_group->cluster_list); |
| 1801 | cluster->block_group = block_group; |
| 1802 | out: |
| 1803 | spin_unlock(&cluster->lock); |
| 1804 | spin_unlock(&block_group->tree_lock); |
| 1805 | |
| 1806 | return ret; |
| 1807 | } |
| 1808 | |
| 1809 | /* |
| 1810 | * simple code to zero out a cluster |
| 1811 | */ |
| 1812 | void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster) |
| 1813 | { |
| 1814 | spin_lock_init(&cluster->lock); |
| 1815 | spin_lock_init(&cluster->refill_lock); |
Eric Paris | 6bef4d3 | 2010-02-23 19:43:04 +0000 | [diff] [blame] | 1816 | cluster->root = RB_ROOT; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1817 | cluster->max_size = 0; |
Josef Bacik | 9630308 | 2009-07-13 21:29:25 -0400 | [diff] [blame] | 1818 | cluster->points_to_bitmap = false; |
Chris Mason | fa9c0d79 | 2009-04-03 09:47:43 -0400 | [diff] [blame] | 1819 | INIT_LIST_HEAD(&cluster->block_group_list); |
| 1820 | cluster->block_group = NULL; |
| 1821 | } |
| 1822 | |