Tao Ma | f2c870e | 2009-08-18 11:19:26 +0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * refcounttree.c |
| 5 | * |
| 6 | * Copyright (C) 2009 Oracle. All rights reserved. |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public |
| 10 | * License version 2 as published by the Free Software Foundation. |
| 11 | * |
| 12 | * This program is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * General Public License for more details. |
| 16 | */ |
| 17 | |
| 18 | #define MLOG_MASK_PREFIX ML_REFCOUNT |
| 19 | #include <cluster/masklog.h> |
| 20 | #include "ocfs2.h" |
| 21 | #include "inode.h" |
| 22 | #include "alloc.h" |
| 23 | #include "suballoc.h" |
| 24 | #include "journal.h" |
| 25 | #include "uptodate.h" |
| 26 | #include "super.h" |
| 27 | #include "buffer_head_io.h" |
| 28 | #include "blockcheck.h" |
Tao Ma | c732eb1 | 2009-08-18 11:21:00 +0800 | [diff] [blame] | 29 | #include "refcounttree.h" |
Tao Ma | 8bf396d | 2009-08-24 11:12:02 +0800 | [diff] [blame^] | 30 | #include "sysfile.h" |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 31 | #include "dlmglue.h" |
Tao Ma | c732eb1 | 2009-08-18 11:21:00 +0800 | [diff] [blame] | 32 | |
| 33 | static inline struct ocfs2_refcount_tree * |
| 34 | cache_info_to_refcount(struct ocfs2_caching_info *ci) |
| 35 | { |
| 36 | return container_of(ci, struct ocfs2_refcount_tree, rf_ci); |
| 37 | } |
Tao Ma | f2c870e | 2009-08-18 11:19:26 +0800 | [diff] [blame] | 38 | |
| 39 | static int ocfs2_validate_refcount_block(struct super_block *sb, |
| 40 | struct buffer_head *bh) |
| 41 | { |
| 42 | int rc; |
| 43 | struct ocfs2_refcount_block *rb = |
| 44 | (struct ocfs2_refcount_block *)bh->b_data; |
| 45 | |
| 46 | mlog(0, "Validating refcount block %llu\n", |
| 47 | (unsigned long long)bh->b_blocknr); |
| 48 | |
| 49 | BUG_ON(!buffer_uptodate(bh)); |
| 50 | |
| 51 | /* |
| 52 | * If the ecc fails, we return the error but otherwise |
| 53 | * leave the filesystem running. We know any error is |
| 54 | * local to this block. |
| 55 | */ |
| 56 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check); |
| 57 | if (rc) { |
| 58 | mlog(ML_ERROR, "Checksum failed for refcount block %llu\n", |
| 59 | (unsigned long long)bh->b_blocknr); |
| 60 | return rc; |
| 61 | } |
| 62 | |
| 63 | |
| 64 | if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) { |
| 65 | ocfs2_error(sb, |
| 66 | "Refcount block #%llu has bad signature %.*s", |
| 67 | (unsigned long long)bh->b_blocknr, 7, |
| 68 | rb->rf_signature); |
| 69 | return -EINVAL; |
| 70 | } |
| 71 | |
| 72 | if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) { |
| 73 | ocfs2_error(sb, |
| 74 | "Refcount block #%llu has an invalid rf_blkno " |
| 75 | "of %llu", |
| 76 | (unsigned long long)bh->b_blocknr, |
| 77 | (unsigned long long)le64_to_cpu(rb->rf_blkno)); |
| 78 | return -EINVAL; |
| 79 | } |
| 80 | |
| 81 | if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) { |
| 82 | ocfs2_error(sb, |
| 83 | "Refcount block #%llu has an invalid " |
| 84 | "rf_fs_generation of #%u", |
| 85 | (unsigned long long)bh->b_blocknr, |
| 86 | le32_to_cpu(rb->rf_fs_generation)); |
| 87 | return -EINVAL; |
| 88 | } |
| 89 | |
| 90 | return 0; |
| 91 | } |
| 92 | |
| 93 | static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci, |
| 94 | u64 rb_blkno, |
| 95 | struct buffer_head **bh) |
| 96 | { |
| 97 | int rc; |
| 98 | struct buffer_head *tmp = *bh; |
| 99 | |
| 100 | rc = ocfs2_read_block(ci, rb_blkno, &tmp, |
| 101 | ocfs2_validate_refcount_block); |
| 102 | |
| 103 | /* If ocfs2_read_block() got us a new bh, pass it up. */ |
| 104 | if (!rc && !*bh) |
| 105 | *bh = tmp; |
| 106 | |
| 107 | return rc; |
| 108 | } |
Tao Ma | c732eb1 | 2009-08-18 11:21:00 +0800 | [diff] [blame] | 109 | |
| 110 | static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci) |
| 111 | { |
| 112 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 113 | |
| 114 | return rf->rf_blkno; |
| 115 | } |
| 116 | |
| 117 | static struct super_block * |
| 118 | ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci) |
| 119 | { |
| 120 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 121 | |
| 122 | return rf->rf_sb; |
| 123 | } |
| 124 | |
| 125 | static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci) |
| 126 | { |
| 127 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 128 | |
| 129 | spin_lock(&rf->rf_lock); |
| 130 | } |
| 131 | |
| 132 | static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci) |
| 133 | { |
| 134 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 135 | |
| 136 | spin_unlock(&rf->rf_lock); |
| 137 | } |
| 138 | |
| 139 | static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci) |
| 140 | { |
| 141 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 142 | |
| 143 | mutex_lock(&rf->rf_io_mutex); |
| 144 | } |
| 145 | |
| 146 | static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci) |
| 147 | { |
| 148 | struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci); |
| 149 | |
| 150 | mutex_unlock(&rf->rf_io_mutex); |
| 151 | } |
| 152 | |
| 153 | static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = { |
| 154 | .co_owner = ocfs2_refcount_cache_owner, |
| 155 | .co_get_super = ocfs2_refcount_cache_get_super, |
| 156 | .co_cache_lock = ocfs2_refcount_cache_lock, |
| 157 | .co_cache_unlock = ocfs2_refcount_cache_unlock, |
| 158 | .co_io_lock = ocfs2_refcount_cache_io_lock, |
| 159 | .co_io_unlock = ocfs2_refcount_cache_io_unlock, |
| 160 | }; |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 161 | |
| 162 | static struct ocfs2_refcount_tree * |
| 163 | ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno) |
| 164 | { |
| 165 | struct rb_node *n = osb->osb_rf_lock_tree.rb_node; |
| 166 | struct ocfs2_refcount_tree *tree = NULL; |
| 167 | |
| 168 | while (n) { |
| 169 | tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node); |
| 170 | |
| 171 | if (blkno < tree->rf_blkno) |
| 172 | n = n->rb_left; |
| 173 | else if (blkno > tree->rf_blkno) |
| 174 | n = n->rb_right; |
| 175 | else |
| 176 | return tree; |
| 177 | } |
| 178 | |
| 179 | return NULL; |
| 180 | } |
| 181 | |
| 182 | /* osb_lock is already locked. */ |
| 183 | static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb, |
| 184 | struct ocfs2_refcount_tree *new) |
| 185 | { |
| 186 | u64 rf_blkno = new->rf_blkno; |
| 187 | struct rb_node *parent = NULL; |
| 188 | struct rb_node **p = &osb->osb_rf_lock_tree.rb_node; |
| 189 | struct ocfs2_refcount_tree *tmp; |
| 190 | |
| 191 | while (*p) { |
| 192 | parent = *p; |
| 193 | |
| 194 | tmp = rb_entry(parent, struct ocfs2_refcount_tree, |
| 195 | rf_node); |
| 196 | |
| 197 | if (rf_blkno < tmp->rf_blkno) |
| 198 | p = &(*p)->rb_left; |
| 199 | else if (rf_blkno > tmp->rf_blkno) |
| 200 | p = &(*p)->rb_right; |
| 201 | else { |
| 202 | /* This should never happen! */ |
| 203 | mlog(ML_ERROR, "Duplicate refcount block %llu found!\n", |
| 204 | (unsigned long long)rf_blkno); |
| 205 | BUG(); |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | rb_link_node(&new->rf_node, parent, p); |
| 210 | rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree); |
| 211 | } |
| 212 | |
| 213 | static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree) |
| 214 | { |
| 215 | ocfs2_metadata_cache_exit(&tree->rf_ci); |
| 216 | ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres); |
| 217 | ocfs2_lock_res_free(&tree->rf_lockres); |
| 218 | kfree(tree); |
| 219 | } |
| 220 | |
| 221 | static inline void |
| 222 | ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb, |
| 223 | struct ocfs2_refcount_tree *tree) |
| 224 | { |
| 225 | rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree); |
| 226 | if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree) |
| 227 | osb->osb_ref_tree_lru = NULL; |
| 228 | } |
| 229 | |
| 230 | static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb, |
| 231 | struct ocfs2_refcount_tree *tree) |
| 232 | { |
| 233 | spin_lock(&osb->osb_lock); |
| 234 | ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree); |
| 235 | spin_unlock(&osb->osb_lock); |
| 236 | } |
| 237 | |
| 238 | void ocfs2_kref_remove_refcount_tree(struct kref *kref) |
| 239 | { |
| 240 | struct ocfs2_refcount_tree *tree = |
| 241 | container_of(kref, struct ocfs2_refcount_tree, rf_getcnt); |
| 242 | |
| 243 | ocfs2_free_refcount_tree(tree); |
| 244 | } |
| 245 | |
| 246 | static inline void |
| 247 | ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree) |
| 248 | { |
| 249 | kref_get(&tree->rf_getcnt); |
| 250 | } |
| 251 | |
| 252 | static inline void |
| 253 | ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree) |
| 254 | { |
| 255 | kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree); |
| 256 | } |
| 257 | |
| 258 | static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new, |
| 259 | struct super_block *sb) |
| 260 | { |
| 261 | ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops); |
| 262 | mutex_init(&new->rf_io_mutex); |
| 263 | new->rf_sb = sb; |
| 264 | spin_lock_init(&new->rf_lock); |
| 265 | } |
| 266 | |
| 267 | static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb, |
| 268 | struct ocfs2_refcount_tree *new, |
| 269 | u64 rf_blkno, u32 generation) |
| 270 | { |
| 271 | init_rwsem(&new->rf_sem); |
| 272 | ocfs2_refcount_lock_res_init(&new->rf_lockres, osb, |
| 273 | rf_blkno, generation); |
| 274 | } |
| 275 | |
Tao Ma | 8bf396d | 2009-08-24 11:12:02 +0800 | [diff] [blame^] | 276 | static struct ocfs2_refcount_tree* |
| 277 | ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno) |
| 278 | { |
| 279 | struct ocfs2_refcount_tree *new; |
| 280 | |
| 281 | new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS); |
| 282 | if (!new) |
| 283 | return NULL; |
| 284 | |
| 285 | new->rf_blkno = rf_blkno; |
| 286 | kref_init(&new->rf_getcnt); |
| 287 | ocfs2_init_refcount_tree_ci(new, osb->sb); |
| 288 | |
| 289 | return new; |
| 290 | } |
| 291 | |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 292 | static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno, |
| 293 | struct ocfs2_refcount_tree **ret_tree) |
| 294 | { |
| 295 | int ret = 0; |
| 296 | struct ocfs2_refcount_tree *tree, *new = NULL; |
| 297 | struct buffer_head *ref_root_bh = NULL; |
| 298 | struct ocfs2_refcount_block *ref_rb; |
| 299 | |
| 300 | spin_lock(&osb->osb_lock); |
| 301 | if (osb->osb_ref_tree_lru && |
| 302 | osb->osb_ref_tree_lru->rf_blkno == rf_blkno) |
| 303 | tree = osb->osb_ref_tree_lru; |
| 304 | else |
| 305 | tree = ocfs2_find_refcount_tree(osb, rf_blkno); |
| 306 | if (tree) |
| 307 | goto out; |
| 308 | |
| 309 | spin_unlock(&osb->osb_lock); |
| 310 | |
Tao Ma | 8bf396d | 2009-08-24 11:12:02 +0800 | [diff] [blame^] | 311 | new = ocfs2_allocate_refcount_tree(osb, rf_blkno); |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 312 | if (!new) { |
| 313 | ret = -ENOMEM; |
Tao Ma | 8bf396d | 2009-08-24 11:12:02 +0800 | [diff] [blame^] | 314 | mlog_errno(ret); |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 315 | return ret; |
| 316 | } |
Tao Ma | 374a263 | 2009-08-24 11:13:37 +0800 | [diff] [blame] | 317 | /* |
| 318 | * We need the generation to create the refcount tree lock and since |
| 319 | * it isn't changed during the tree modification, we are safe here to |
| 320 | * read without protection. |
| 321 | * We also have to purge the cache after we create the lock since the |
| 322 | * refcount block may have the stale data. It can only be trusted when |
| 323 | * we hold the refcount lock. |
| 324 | */ |
| 325 | ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh); |
| 326 | if (ret) { |
| 327 | mlog_errno(ret); |
| 328 | ocfs2_metadata_cache_exit(&new->rf_ci); |
| 329 | kfree(new); |
| 330 | return ret; |
| 331 | } |
| 332 | |
| 333 | ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data; |
| 334 | new->rf_generation = le32_to_cpu(ref_rb->rf_generation); |
| 335 | ocfs2_init_refcount_tree_lock(osb, new, rf_blkno, |
| 336 | new->rf_generation); |
| 337 | ocfs2_metadata_cache_purge(&new->rf_ci); |
| 338 | |
| 339 | spin_lock(&osb->osb_lock); |
| 340 | tree = ocfs2_find_refcount_tree(osb, rf_blkno); |
| 341 | if (tree) |
| 342 | goto out; |
| 343 | |
| 344 | ocfs2_insert_refcount_tree(osb, new); |
| 345 | |
| 346 | tree = new; |
| 347 | new = NULL; |
| 348 | |
| 349 | out: |
| 350 | *ret_tree = tree; |
| 351 | |
| 352 | osb->osb_ref_tree_lru = tree; |
| 353 | |
| 354 | spin_unlock(&osb->osb_lock); |
| 355 | |
| 356 | if (new) |
| 357 | ocfs2_free_refcount_tree(new); |
| 358 | |
| 359 | brelse(ref_root_bh); |
| 360 | return ret; |
| 361 | } |
| 362 | |
| 363 | static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno) |
| 364 | { |
| 365 | int ret; |
| 366 | struct buffer_head *di_bh = NULL; |
| 367 | struct ocfs2_dinode *di; |
| 368 | |
| 369 | ret = ocfs2_read_inode_block(inode, &di_bh); |
| 370 | if (ret) { |
| 371 | mlog_errno(ret); |
| 372 | goto out; |
| 373 | } |
| 374 | |
| 375 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)); |
| 376 | |
| 377 | di = (struct ocfs2_dinode *)di_bh->b_data; |
| 378 | *ref_blkno = le64_to_cpu(di->i_refcount_loc); |
| 379 | brelse(di_bh); |
| 380 | out: |
| 381 | return ret; |
| 382 | } |
| 383 | |
| 384 | static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb, |
| 385 | struct ocfs2_refcount_tree *tree, int rw) |
| 386 | { |
| 387 | int ret; |
| 388 | |
| 389 | ret = ocfs2_refcount_lock(tree, rw); |
| 390 | if (ret) { |
| 391 | mlog_errno(ret); |
| 392 | goto out; |
| 393 | } |
| 394 | |
| 395 | if (rw) |
| 396 | down_write(&tree->rf_sem); |
| 397 | else |
| 398 | down_read(&tree->rf_sem); |
| 399 | |
| 400 | out: |
| 401 | return ret; |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * Lock the refcount tree pointed by ref_blkno and return the tree. |
| 406 | * In most case, we lock the tree and read the refcount block. |
| 407 | * So read it here if the caller really needs it. |
| 408 | * |
| 409 | * If the tree has been re-created by other node, it will free the |
| 410 | * old one and re-create it. |
| 411 | */ |
| 412 | int ocfs2_lock_refcount_tree(struct ocfs2_super *osb, |
| 413 | u64 ref_blkno, int rw, |
| 414 | struct ocfs2_refcount_tree **ret_tree, |
| 415 | struct buffer_head **ref_bh) |
| 416 | { |
| 417 | int ret, delete_tree = 0; |
| 418 | struct ocfs2_refcount_tree *tree = NULL; |
| 419 | struct buffer_head *ref_root_bh = NULL; |
| 420 | struct ocfs2_refcount_block *rb; |
| 421 | |
| 422 | again: |
| 423 | ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree); |
| 424 | if (ret) { |
| 425 | mlog_errno(ret); |
| 426 | return ret; |
| 427 | } |
| 428 | |
| 429 | ocfs2_refcount_tree_get(tree); |
| 430 | |
| 431 | ret = __ocfs2_lock_refcount_tree(osb, tree, rw); |
| 432 | if (ret) { |
| 433 | mlog_errno(ret); |
| 434 | ocfs2_refcount_tree_put(tree); |
| 435 | goto out; |
| 436 | } |
| 437 | |
| 438 | ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno, |
| 439 | &ref_root_bh); |
| 440 | if (ret) { |
| 441 | mlog_errno(ret); |
| 442 | ocfs2_unlock_refcount_tree(osb, tree, rw); |
| 443 | ocfs2_refcount_tree_put(tree); |
| 444 | goto out; |
| 445 | } |
| 446 | |
| 447 | rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data; |
| 448 | /* |
| 449 | * If the refcount block has been freed and re-created, we may need |
| 450 | * to recreate the refcount tree also. |
| 451 | * |
| 452 | * Here we just remove the tree from the rb-tree, and the last |
| 453 | * kref holder will unlock and delete this refcount_tree. |
| 454 | * Then we goto "again" and ocfs2_get_refcount_tree will create |
| 455 | * the new refcount tree for us. |
| 456 | */ |
| 457 | if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) { |
| 458 | if (!tree->rf_removed) { |
| 459 | ocfs2_erase_refcount_tree_from_list(osb, tree); |
| 460 | tree->rf_removed = 1; |
| 461 | delete_tree = 1; |
| 462 | } |
| 463 | |
| 464 | ocfs2_unlock_refcount_tree(osb, tree, rw); |
| 465 | /* |
| 466 | * We get an extra reference when we create the refcount |
| 467 | * tree, so another put will destroy it. |
| 468 | */ |
| 469 | if (delete_tree) |
| 470 | ocfs2_refcount_tree_put(tree); |
| 471 | brelse(ref_root_bh); |
| 472 | ref_root_bh = NULL; |
| 473 | goto again; |
| 474 | } |
| 475 | |
| 476 | *ret_tree = tree; |
| 477 | if (ref_bh) { |
| 478 | *ref_bh = ref_root_bh; |
| 479 | ref_root_bh = NULL; |
| 480 | } |
| 481 | out: |
| 482 | brelse(ref_root_bh); |
| 483 | return ret; |
| 484 | } |
| 485 | |
| 486 | int ocfs2_lock_refcount_tree_by_inode(struct inode *inode, int rw, |
| 487 | struct ocfs2_refcount_tree **ret_tree, |
| 488 | struct buffer_head **ref_bh) |
| 489 | { |
| 490 | int ret; |
| 491 | u64 ref_blkno; |
| 492 | |
| 493 | ret = ocfs2_get_refcount_block(inode, &ref_blkno); |
| 494 | if (ret) { |
| 495 | mlog_errno(ret); |
| 496 | return ret; |
| 497 | } |
| 498 | |
| 499 | return ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, |
| 500 | rw, ret_tree, ref_bh); |
| 501 | } |
| 502 | |
| 503 | void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb, |
| 504 | struct ocfs2_refcount_tree *tree, int rw) |
| 505 | { |
| 506 | if (rw) |
| 507 | up_write(&tree->rf_sem); |
| 508 | else |
| 509 | up_read(&tree->rf_sem); |
| 510 | |
| 511 | ocfs2_refcount_unlock(tree, rw); |
| 512 | ocfs2_refcount_tree_put(tree); |
| 513 | } |
| 514 | |
| 515 | void ocfs2_purge_refcount_trees(struct ocfs2_super *osb) |
| 516 | { |
| 517 | struct rb_node *node; |
| 518 | struct ocfs2_refcount_tree *tree; |
| 519 | struct rb_root *root = &osb->osb_rf_lock_tree; |
| 520 | |
| 521 | while ((node = rb_last(root)) != NULL) { |
| 522 | tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node); |
| 523 | |
| 524 | mlog(0, "Purge tree %llu\n", |
| 525 | (unsigned long long) tree->rf_blkno); |
| 526 | |
| 527 | rb_erase(&tree->rf_node, root); |
| 528 | ocfs2_free_refcount_tree(tree); |
| 529 | } |
| 530 | } |
Tao Ma | 8bf396d | 2009-08-24 11:12:02 +0800 | [diff] [blame^] | 531 | |
| 532 | /* |
| 533 | * Create a refcount tree for an inode. |
| 534 | * We take for granted that the inode is already locked. |
| 535 | */ |
| 536 | static int ocfs2_create_refcount_tree(struct inode *inode, |
| 537 | struct buffer_head *di_bh) |
| 538 | { |
| 539 | int ret; |
| 540 | handle_t *handle = NULL; |
| 541 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 542 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
| 543 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 544 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 545 | struct buffer_head *new_bh = NULL; |
| 546 | struct ocfs2_refcount_block *rb; |
| 547 | struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL; |
| 548 | u16 suballoc_bit_start; |
| 549 | u32 num_got; |
| 550 | u64 first_blkno; |
| 551 | |
| 552 | BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL); |
| 553 | |
| 554 | mlog(0, "create tree for inode %lu\n", inode->i_ino); |
| 555 | |
| 556 | ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac); |
| 557 | if (ret) { |
| 558 | mlog_errno(ret); |
| 559 | goto out; |
| 560 | } |
| 561 | |
| 562 | handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS); |
| 563 | if (IS_ERR(handle)) { |
| 564 | ret = PTR_ERR(handle); |
| 565 | mlog_errno(ret); |
| 566 | goto out; |
| 567 | } |
| 568 | |
| 569 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
| 570 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 571 | if (ret) { |
| 572 | mlog_errno(ret); |
| 573 | goto out_commit; |
| 574 | } |
| 575 | |
| 576 | ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1, |
| 577 | &suballoc_bit_start, &num_got, |
| 578 | &first_blkno); |
| 579 | if (ret) { |
| 580 | mlog_errno(ret); |
| 581 | goto out_commit; |
| 582 | } |
| 583 | |
| 584 | new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno); |
| 585 | if (!new_tree) { |
| 586 | ret = -ENOMEM; |
| 587 | mlog_errno(ret); |
| 588 | goto out_commit; |
| 589 | } |
| 590 | |
| 591 | new_bh = sb_getblk(inode->i_sb, first_blkno); |
| 592 | ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh); |
| 593 | |
| 594 | ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh, |
| 595 | OCFS2_JOURNAL_ACCESS_CREATE); |
| 596 | if (ret) { |
| 597 | mlog_errno(ret); |
| 598 | goto out_commit; |
| 599 | } |
| 600 | |
| 601 | /* Initialize ocfs2_refcount_block. */ |
| 602 | rb = (struct ocfs2_refcount_block *)new_bh->b_data; |
| 603 | memset(rb, 0, inode->i_sb->s_blocksize); |
| 604 | strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE); |
| 605 | rb->rf_suballoc_slot = cpu_to_le16(osb->slot_num); |
| 606 | rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
| 607 | rb->rf_fs_generation = cpu_to_le32(osb->fs_generation); |
| 608 | rb->rf_blkno = cpu_to_le64(first_blkno); |
| 609 | rb->rf_count = cpu_to_le32(1); |
| 610 | rb->rf_records.rl_count = |
| 611 | cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb)); |
| 612 | spin_lock(&osb->osb_lock); |
| 613 | rb->rf_generation = osb->s_next_generation++; |
| 614 | spin_unlock(&osb->osb_lock); |
| 615 | |
| 616 | ocfs2_journal_dirty(handle, new_bh); |
| 617 | |
| 618 | spin_lock(&oi->ip_lock); |
| 619 | oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL; |
| 620 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); |
| 621 | di->i_refcount_loc = cpu_to_le64(first_blkno); |
| 622 | spin_unlock(&oi->ip_lock); |
| 623 | |
| 624 | mlog(0, "created tree for inode %lu, refblock %llu\n", |
| 625 | inode->i_ino, (unsigned long long)first_blkno); |
| 626 | |
| 627 | ocfs2_journal_dirty(handle, di_bh); |
| 628 | |
| 629 | /* |
| 630 | * We have to init the tree lock here since it will use |
| 631 | * the generation number to create it. |
| 632 | */ |
| 633 | new_tree->rf_generation = le32_to_cpu(rb->rf_generation); |
| 634 | ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno, |
| 635 | new_tree->rf_generation); |
| 636 | |
| 637 | spin_lock(&osb->osb_lock); |
| 638 | tree = ocfs2_find_refcount_tree(osb, first_blkno); |
| 639 | |
| 640 | /* |
| 641 | * We've just created a new refcount tree in this block. If |
| 642 | * we found a refcount tree on the ocfs2_super, it must be |
| 643 | * one we just deleted. We free the old tree before |
| 644 | * inserting the new tree. |
| 645 | */ |
| 646 | BUG_ON(tree && tree->rf_generation == new_tree->rf_generation); |
| 647 | if (tree) |
| 648 | ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree); |
| 649 | ocfs2_insert_refcount_tree(osb, new_tree); |
| 650 | spin_unlock(&osb->osb_lock); |
| 651 | new_tree = NULL; |
| 652 | if (tree) |
| 653 | ocfs2_refcount_tree_put(tree); |
| 654 | |
| 655 | out_commit: |
| 656 | ocfs2_commit_trans(osb, handle); |
| 657 | |
| 658 | out: |
| 659 | if (new_tree) { |
| 660 | ocfs2_metadata_cache_exit(&new_tree->rf_ci); |
| 661 | kfree(new_tree); |
| 662 | } |
| 663 | |
| 664 | brelse(new_bh); |
| 665 | if (meta_ac) |
| 666 | ocfs2_free_alloc_context(meta_ac); |
| 667 | |
| 668 | return ret; |
| 669 | } |
| 670 | |
| 671 | static int ocfs2_set_refcount_tree(struct inode *inode, |
| 672 | struct buffer_head *di_bh, |
| 673 | u64 refcount_loc) |
| 674 | { |
| 675 | int ret; |
| 676 | handle_t *handle = NULL; |
| 677 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
| 678 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 679 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 680 | struct buffer_head *ref_root_bh = NULL; |
| 681 | struct ocfs2_refcount_block *rb; |
| 682 | struct ocfs2_refcount_tree *ref_tree; |
| 683 | |
| 684 | BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL); |
| 685 | |
| 686 | ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, |
| 687 | &ref_tree, &ref_root_bh); |
| 688 | if (ret) { |
| 689 | mlog_errno(ret); |
| 690 | return ret; |
| 691 | } |
| 692 | |
| 693 | handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS); |
| 694 | if (IS_ERR(handle)) { |
| 695 | ret = PTR_ERR(handle); |
| 696 | mlog_errno(ret); |
| 697 | goto out; |
| 698 | } |
| 699 | |
| 700 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
| 701 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 702 | if (ret) { |
| 703 | mlog_errno(ret); |
| 704 | goto out_commit; |
| 705 | } |
| 706 | |
| 707 | ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh, |
| 708 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 709 | if (ret) { |
| 710 | mlog_errno(ret); |
| 711 | goto out_commit; |
| 712 | } |
| 713 | |
| 714 | rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data; |
| 715 | le32_add_cpu(&rb->rf_count, 1); |
| 716 | |
| 717 | ocfs2_journal_dirty(handle, ref_root_bh); |
| 718 | |
| 719 | spin_lock(&oi->ip_lock); |
| 720 | oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL; |
| 721 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); |
| 722 | di->i_refcount_loc = cpu_to_le64(refcount_loc); |
| 723 | spin_unlock(&oi->ip_lock); |
| 724 | ocfs2_journal_dirty(handle, di_bh); |
| 725 | |
| 726 | out_commit: |
| 727 | ocfs2_commit_trans(osb, handle); |
| 728 | out: |
| 729 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); |
| 730 | brelse(ref_root_bh); |
| 731 | |
| 732 | return ret; |
| 733 | } |
| 734 | |
| 735 | int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh) |
| 736 | { |
| 737 | int ret, delete_tree = 0; |
| 738 | handle_t *handle = NULL; |
| 739 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
| 740 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 741 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 742 | struct ocfs2_refcount_block *rb; |
| 743 | struct inode *alloc_inode = NULL; |
| 744 | struct buffer_head *alloc_bh = NULL; |
| 745 | struct buffer_head *blk_bh = NULL; |
| 746 | struct ocfs2_refcount_tree *ref_tree; |
| 747 | int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS; |
| 748 | u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc); |
| 749 | u16 bit = 0; |
| 750 | |
| 751 | if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) |
| 752 | return 0; |
| 753 | |
| 754 | BUG_ON(!ref_blkno); |
| 755 | ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh); |
| 756 | if (ret) { |
| 757 | mlog_errno(ret); |
| 758 | return ret; |
| 759 | } |
| 760 | |
| 761 | rb = (struct ocfs2_refcount_block *)blk_bh->b_data; |
| 762 | |
| 763 | /* |
| 764 | * If we are the last user, we need to free the block. |
| 765 | * So lock the allocator ahead. |
| 766 | */ |
| 767 | if (le32_to_cpu(rb->rf_count) == 1) { |
| 768 | blk = le64_to_cpu(rb->rf_blkno); |
| 769 | bit = le16_to_cpu(rb->rf_suballoc_bit); |
| 770 | bg_blkno = ocfs2_which_suballoc_group(blk, bit); |
| 771 | |
| 772 | alloc_inode = ocfs2_get_system_file_inode(osb, |
| 773 | EXTENT_ALLOC_SYSTEM_INODE, |
| 774 | le16_to_cpu(rb->rf_suballoc_slot)); |
| 775 | if (!alloc_inode) { |
| 776 | ret = -ENOMEM; |
| 777 | mlog_errno(ret); |
| 778 | goto out; |
| 779 | } |
| 780 | mutex_lock(&alloc_inode->i_mutex); |
| 781 | |
| 782 | ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1); |
| 783 | if (ret) { |
| 784 | mlog_errno(ret); |
| 785 | goto out_mutex; |
| 786 | } |
| 787 | |
| 788 | credits += OCFS2_SUBALLOC_FREE; |
| 789 | } |
| 790 | |
| 791 | handle = ocfs2_start_trans(osb, credits); |
| 792 | if (IS_ERR(handle)) { |
| 793 | ret = PTR_ERR(handle); |
| 794 | mlog_errno(ret); |
| 795 | goto out_unlock; |
| 796 | } |
| 797 | |
| 798 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
| 799 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 800 | if (ret) { |
| 801 | mlog_errno(ret); |
| 802 | goto out_commit; |
| 803 | } |
| 804 | |
| 805 | ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh, |
| 806 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 807 | if (ret) { |
| 808 | mlog_errno(ret); |
| 809 | goto out_commit; |
| 810 | } |
| 811 | |
| 812 | spin_lock(&oi->ip_lock); |
| 813 | oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL; |
| 814 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); |
| 815 | di->i_refcount_loc = 0; |
| 816 | spin_unlock(&oi->ip_lock); |
| 817 | ocfs2_journal_dirty(handle, di_bh); |
| 818 | |
| 819 | le32_add_cpu(&rb->rf_count , -1); |
| 820 | ocfs2_journal_dirty(handle, blk_bh); |
| 821 | |
| 822 | if (!rb->rf_count) { |
| 823 | delete_tree = 1; |
| 824 | ocfs2_erase_refcount_tree_from_list(osb, ref_tree); |
| 825 | ret = ocfs2_free_suballoc_bits(handle, alloc_inode, |
| 826 | alloc_bh, bit, bg_blkno, 1); |
| 827 | if (ret) |
| 828 | mlog_errno(ret); |
| 829 | } |
| 830 | |
| 831 | out_commit: |
| 832 | ocfs2_commit_trans(osb, handle); |
| 833 | out_unlock: |
| 834 | if (alloc_inode) { |
| 835 | ocfs2_inode_unlock(alloc_inode, 1); |
| 836 | brelse(alloc_bh); |
| 837 | } |
| 838 | out_mutex: |
| 839 | if (alloc_inode) { |
| 840 | mutex_unlock(&alloc_inode->i_mutex); |
| 841 | iput(alloc_inode); |
| 842 | } |
| 843 | out: |
| 844 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); |
| 845 | if (delete_tree) |
| 846 | ocfs2_refcount_tree_put(ref_tree); |
| 847 | brelse(blk_bh); |
| 848 | |
| 849 | return ret; |
| 850 | } |