diff options
Diffstat (limited to 'fs/ocfs2/localalloc.c')
| -rw-r--r-- | fs/ocfs2/localalloc.c | 983 |
1 files changed, 983 insertions, 0 deletions
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c new file mode 100644 index 000000000000..149b35181666 --- /dev/null +++ b/fs/ocfs2/localalloc.c | |||
| @@ -0,0 +1,983 @@ | |||
| 1 | /* -*- mode: c; c-basic-offset: 8; -*- | ||
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: | ||
| 3 | * | ||
| 4 | * localalloc.c | ||
| 5 | * | ||
| 6 | * Node local data allocation | ||
| 7 | * | ||
| 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | ||
| 9 | * | ||
| 10 | * This program is free software; you can redistribute it and/or | ||
| 11 | * modify it under the terms of the GNU General Public | ||
| 12 | * License as published by the Free Software Foundation; either | ||
| 13 | * version 2 of the License, or (at your option) any later version. | ||
| 14 | * | ||
| 15 | * This program is distributed in the hope that it will be useful, | ||
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 18 | * General Public License for more details. | ||
| 19 | * | ||
| 20 | * You should have received a copy of the GNU General Public | ||
| 21 | * License along with this program; if not, write to the | ||
| 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | ||
| 23 | * Boston, MA 021110-1307, USA. | ||
| 24 | */ | ||
| 25 | |||
| 26 | #include <linux/fs.h> | ||
| 27 | #include <linux/types.h> | ||
| 28 | #include <linux/slab.h> | ||
| 29 | #include <linux/highmem.h> | ||
| 30 | #include <linux/bitops.h> | ||
| 31 | |||
| 32 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | ||
| 33 | #include <cluster/masklog.h> | ||
| 34 | |||
| 35 | #include "ocfs2.h" | ||
| 36 | |||
| 37 | #include "alloc.h" | ||
| 38 | #include "dlmglue.h" | ||
| 39 | #include "inode.h" | ||
| 40 | #include "journal.h" | ||
| 41 | #include "localalloc.h" | ||
| 42 | #include "suballoc.h" | ||
| 43 | #include "super.h" | ||
| 44 | #include "sysfile.h" | ||
| 45 | |||
| 46 | #include "buffer_head_io.h" | ||
| 47 | |||
| 48 | #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab)) | ||
| 49 | |||
| 50 | static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb); | ||
| 51 | |||
| 52 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc); | ||
| 53 | |||
| 54 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | ||
| 55 | struct ocfs2_dinode *alloc, | ||
| 56 | u32 numbits); | ||
| 57 | |||
| 58 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc); | ||
| 59 | |||
| 60 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | ||
| 61 | struct ocfs2_journal_handle *handle, | ||
| 62 | struct ocfs2_dinode *alloc, | ||
| 63 | struct inode *main_bm_inode, | ||
| 64 | struct buffer_head *main_bm_bh); | ||
| 65 | |||
| 66 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, | ||
| 67 | struct ocfs2_journal_handle *handle, | ||
| 68 | struct ocfs2_alloc_context **ac, | ||
| 69 | struct inode **bitmap_inode, | ||
| 70 | struct buffer_head **bitmap_bh); | ||
| 71 | |||
| 72 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | ||
| 73 | struct ocfs2_journal_handle *handle, | ||
| 74 | struct ocfs2_alloc_context *ac); | ||
| 75 | |||
| 76 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | ||
| 77 | struct inode *local_alloc_inode); | ||
| 78 | |||
| 79 | /* | ||
| 80 | * Determine how large our local alloc window should be, in bits. | ||
| 81 | * | ||
| 82 | * These values (and the behavior in ocfs2_alloc_should_use_local) have | ||
| 83 | * been chosen so that most allocations, including new block groups go | ||
| 84 | * through local alloc. | ||
| 85 | */ | ||
| 86 | static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb) | ||
| 87 | { | ||
| 88 | BUG_ON(osb->s_clustersize_bits < 12); | ||
| 89 | |||
| 90 | return 2048 >> (osb->s_clustersize_bits - 12); | ||
| 91 | } | ||
| 92 | |||
| 93 | /* | ||
| 94 | * Tell us whether a given allocation should use the local alloc | ||
| 95 | * file. Otherwise, it has to go to the main bitmap. | ||
| 96 | */ | ||
| 97 | int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits) | ||
| 98 | { | ||
| 99 | int la_bits = ocfs2_local_alloc_window_bits(osb); | ||
| 100 | |||
| 101 | if (osb->local_alloc_state != OCFS2_LA_ENABLED) | ||
| 102 | return 0; | ||
| 103 | |||
| 104 | /* la_bits should be at least twice the size (in clusters) of | ||
| 105 | * a new block group. We want to be sure block group | ||
| 106 | * allocations go through the local alloc, so allow an | ||
| 107 | * allocation to take up to half the bitmap. */ | ||
| 108 | if (bits > (la_bits / 2)) | ||
| 109 | return 0; | ||
| 110 | |||
| 111 | return 1; | ||
| 112 | } | ||
| 113 | |||
| 114 | int ocfs2_load_local_alloc(struct ocfs2_super *osb) | ||
| 115 | { | ||
| 116 | int status = 0; | ||
| 117 | struct ocfs2_dinode *alloc = NULL; | ||
| 118 | struct buffer_head *alloc_bh = NULL; | ||
| 119 | u32 num_used; | ||
| 120 | struct inode *inode = NULL; | ||
| 121 | struct ocfs2_local_alloc *la; | ||
| 122 | |||
| 123 | mlog_entry_void(); | ||
| 124 | |||
| 125 | /* read the alloc off disk */ | ||
| 126 | inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE, | ||
| 127 | osb->slot_num); | ||
| 128 | if (!inode) { | ||
| 129 | status = -EINVAL; | ||
| 130 | mlog_errno(status); | ||
| 131 | goto bail; | ||
| 132 | } | ||
| 133 | |||
| 134 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, | ||
| 135 | &alloc_bh, 0, inode); | ||
| 136 | if (status < 0) { | ||
| 137 | mlog_errno(status); | ||
| 138 | goto bail; | ||
| 139 | } | ||
| 140 | |||
| 141 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | ||
| 142 | la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 143 | |||
| 144 | if (!(le32_to_cpu(alloc->i_flags) & | ||
| 145 | (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) { | ||
| 146 | mlog(ML_ERROR, "Invalid local alloc inode, %"MLFu64"\n", | ||
| 147 | OCFS2_I(inode)->ip_blkno); | ||
| 148 | status = -EINVAL; | ||
| 149 | goto bail; | ||
| 150 | } | ||
| 151 | |||
| 152 | if ((la->la_size == 0) || | ||
| 153 | (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) { | ||
| 154 | mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n", | ||
| 155 | le16_to_cpu(la->la_size)); | ||
| 156 | status = -EINVAL; | ||
| 157 | goto bail; | ||
| 158 | } | ||
| 159 | |||
| 160 | /* do a little verification. */ | ||
| 161 | num_used = ocfs2_local_alloc_count_bits(alloc); | ||
| 162 | |||
| 163 | /* hopefully the local alloc has always been recovered before | ||
| 164 | * we load it. */ | ||
| 165 | if (num_used | ||
| 166 | || alloc->id1.bitmap1.i_used | ||
| 167 | || alloc->id1.bitmap1.i_total | ||
| 168 | || la->la_bm_off) | ||
| 169 | mlog(ML_ERROR, "Local alloc hasn't been recovered!\n" | ||
| 170 | "found = %u, set = %u, taken = %u, off = %u\n", | ||
| 171 | num_used, le32_to_cpu(alloc->id1.bitmap1.i_used), | ||
| 172 | le32_to_cpu(alloc->id1.bitmap1.i_total), | ||
| 173 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off); | ||
| 174 | |||
| 175 | osb->local_alloc_bh = alloc_bh; | ||
| 176 | osb->local_alloc_state = OCFS2_LA_ENABLED; | ||
| 177 | |||
| 178 | bail: | ||
| 179 | if (status < 0) | ||
| 180 | if (alloc_bh) | ||
| 181 | brelse(alloc_bh); | ||
| 182 | if (inode) | ||
| 183 | iput(inode); | ||
| 184 | |||
| 185 | mlog_exit(status); | ||
| 186 | return status; | ||
| 187 | } | ||
| 188 | |||
| 189 | /* | ||
| 190 | * return any unused bits to the bitmap and write out a clean | ||
| 191 | * local_alloc. | ||
| 192 | * | ||
| 193 | * local_alloc_bh is optional. If not passed, we will simply use the | ||
| 194 | * one off osb. If you do pass it however, be warned that it *will* be | ||
| 195 | * returned brelse'd and NULL'd out.*/ | ||
| 196 | void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb) | ||
| 197 | { | ||
| 198 | int status; | ||
| 199 | struct ocfs2_journal_handle *handle = NULL; | ||
| 200 | struct inode *local_alloc_inode = NULL; | ||
| 201 | struct buffer_head *bh = NULL; | ||
| 202 | struct buffer_head *main_bm_bh = NULL; | ||
| 203 | struct inode *main_bm_inode = NULL; | ||
| 204 | struct ocfs2_dinode *alloc_copy = NULL; | ||
| 205 | struct ocfs2_dinode *alloc = NULL; | ||
| 206 | |||
| 207 | mlog_entry_void(); | ||
| 208 | |||
| 209 | if (osb->local_alloc_state == OCFS2_LA_UNUSED) | ||
| 210 | goto bail; | ||
| 211 | |||
| 212 | local_alloc_inode = | ||
| 213 | ocfs2_get_system_file_inode(osb, | ||
| 214 | LOCAL_ALLOC_SYSTEM_INODE, | ||
| 215 | osb->slot_num); | ||
| 216 | if (!local_alloc_inode) { | ||
| 217 | status = -ENOENT; | ||
| 218 | mlog_errno(status); | ||
| 219 | goto bail; | ||
| 220 | } | ||
| 221 | |||
| 222 | osb->local_alloc_state = OCFS2_LA_DISABLED; | ||
| 223 | |||
| 224 | handle = ocfs2_alloc_handle(osb); | ||
| 225 | if (!handle) { | ||
| 226 | status = -ENOMEM; | ||
| 227 | mlog_errno(status); | ||
| 228 | goto bail; | ||
| 229 | } | ||
| 230 | |||
| 231 | main_bm_inode = ocfs2_get_system_file_inode(osb, | ||
| 232 | GLOBAL_BITMAP_SYSTEM_INODE, | ||
| 233 | OCFS2_INVALID_SLOT); | ||
| 234 | if (!main_bm_inode) { | ||
| 235 | status = -EINVAL; | ||
| 236 | mlog_errno(status); | ||
| 237 | goto bail; | ||
| 238 | } | ||
| 239 | |||
| 240 | ocfs2_handle_add_inode(handle, main_bm_inode); | ||
| 241 | status = ocfs2_meta_lock(main_bm_inode, handle, &main_bm_bh, 1); | ||
| 242 | if (status < 0) { | ||
| 243 | mlog_errno(status); | ||
| 244 | goto bail; | ||
| 245 | } | ||
| 246 | |||
| 247 | /* WINDOW_MOVE_CREDITS is a bit heavy... */ | ||
| 248 | handle = ocfs2_start_trans(osb, handle, OCFS2_WINDOW_MOVE_CREDITS); | ||
| 249 | if (IS_ERR(handle)) { | ||
| 250 | mlog_errno(PTR_ERR(handle)); | ||
| 251 | handle = NULL; | ||
| 252 | goto bail; | ||
| 253 | } | ||
| 254 | |||
| 255 | bh = osb->local_alloc_bh; | ||
| 256 | alloc = (struct ocfs2_dinode *) bh->b_data; | ||
| 257 | |||
| 258 | alloc_copy = kmalloc(bh->b_size, GFP_KERNEL); | ||
| 259 | if (!alloc_copy) { | ||
| 260 | status = -ENOMEM; | ||
| 261 | goto bail; | ||
| 262 | } | ||
| 263 | memcpy(alloc_copy, alloc, bh->b_size); | ||
| 264 | |||
| 265 | status = ocfs2_journal_access(handle, local_alloc_inode, bh, | ||
| 266 | OCFS2_JOURNAL_ACCESS_WRITE); | ||
| 267 | if (status < 0) { | ||
| 268 | mlog_errno(status); | ||
| 269 | goto bail; | ||
| 270 | } | ||
| 271 | |||
| 272 | ocfs2_clear_local_alloc(alloc); | ||
| 273 | |||
| 274 | status = ocfs2_journal_dirty(handle, bh); | ||
| 275 | if (status < 0) { | ||
| 276 | mlog_errno(status); | ||
| 277 | goto bail; | ||
| 278 | } | ||
| 279 | |||
| 280 | brelse(bh); | ||
| 281 | osb->local_alloc_bh = NULL; | ||
| 282 | osb->local_alloc_state = OCFS2_LA_UNUSED; | ||
| 283 | |||
| 284 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | ||
| 285 | main_bm_inode, main_bm_bh); | ||
| 286 | if (status < 0) | ||
| 287 | mlog_errno(status); | ||
| 288 | |||
| 289 | bail: | ||
| 290 | if (handle) | ||
| 291 | ocfs2_commit_trans(handle); | ||
| 292 | |||
| 293 | if (main_bm_bh) | ||
| 294 | brelse(main_bm_bh); | ||
| 295 | |||
| 296 | if (main_bm_inode) | ||
| 297 | iput(main_bm_inode); | ||
| 298 | |||
| 299 | if (local_alloc_inode) | ||
| 300 | iput(local_alloc_inode); | ||
| 301 | |||
| 302 | if (alloc_copy) | ||
| 303 | kfree(alloc_copy); | ||
| 304 | |||
| 305 | mlog_exit_void(); | ||
| 306 | } | ||
| 307 | |||
| 308 | /* | ||
| 309 | * We want to free the bitmap bits outside of any recovery context as | ||
| 310 | * we'll need a cluster lock to do so, but we must clear the local | ||
| 311 | * alloc before giving up the recovered nodes journal. To solve this, | ||
| 312 | * we kmalloc a copy of the local alloc before it's change for the | ||
| 313 | * caller to process with ocfs2_complete_local_alloc_recovery | ||
| 314 | */ | ||
| 315 | int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb, | ||
| 316 | int slot_num, | ||
| 317 | struct ocfs2_dinode **alloc_copy) | ||
| 318 | { | ||
| 319 | int status = 0; | ||
| 320 | struct buffer_head *alloc_bh = NULL; | ||
| 321 | struct inode *inode = NULL; | ||
| 322 | struct ocfs2_dinode *alloc; | ||
| 323 | |||
| 324 | mlog_entry("(slot_num = %d)\n", slot_num); | ||
| 325 | |||
| 326 | *alloc_copy = NULL; | ||
| 327 | |||
| 328 | inode = ocfs2_get_system_file_inode(osb, | ||
| 329 | LOCAL_ALLOC_SYSTEM_INODE, | ||
| 330 | slot_num); | ||
| 331 | if (!inode) { | ||
| 332 | status = -EINVAL; | ||
| 333 | mlog_errno(status); | ||
| 334 | goto bail; | ||
| 335 | } | ||
| 336 | |||
| 337 | mutex_lock(&inode->i_mutex); | ||
| 338 | |||
| 339 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, | ||
| 340 | &alloc_bh, 0, inode); | ||
| 341 | if (status < 0) { | ||
| 342 | mlog_errno(status); | ||
| 343 | goto bail; | ||
| 344 | } | ||
| 345 | |||
| 346 | *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL); | ||
| 347 | if (!(*alloc_copy)) { | ||
| 348 | status = -ENOMEM; | ||
| 349 | goto bail; | ||
| 350 | } | ||
| 351 | memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size); | ||
| 352 | |||
| 353 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | ||
| 354 | ocfs2_clear_local_alloc(alloc); | ||
| 355 | |||
| 356 | status = ocfs2_write_block(osb, alloc_bh, inode); | ||
| 357 | if (status < 0) | ||
| 358 | mlog_errno(status); | ||
| 359 | |||
| 360 | bail: | ||
| 361 | if ((status < 0) && (*alloc_copy)) { | ||
| 362 | kfree(*alloc_copy); | ||
| 363 | *alloc_copy = NULL; | ||
| 364 | } | ||
| 365 | |||
| 366 | if (alloc_bh) | ||
| 367 | brelse(alloc_bh); | ||
| 368 | |||
| 369 | if (inode) { | ||
| 370 | mutex_unlock(&inode->i_mutex); | ||
| 371 | iput(inode); | ||
| 372 | } | ||
| 373 | |||
| 374 | mlog_exit(status); | ||
| 375 | return status; | ||
| 376 | } | ||
| 377 | |||
| 378 | /* | ||
| 379 | * Step 2: By now, we've completed the journal recovery, we've stamped | ||
| 380 | * a clean local alloc on disk and dropped the node out of the | ||
| 381 | * recovery map. Dlm locks will no longer stall, so lets clear out the | ||
| 382 | * main bitmap. | ||
| 383 | */ | ||
| 384 | int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb, | ||
| 385 | struct ocfs2_dinode *alloc) | ||
| 386 | { | ||
| 387 | int status; | ||
| 388 | struct ocfs2_journal_handle *handle = NULL; | ||
| 389 | struct buffer_head *main_bm_bh = NULL; | ||
| 390 | struct inode *main_bm_inode = NULL; | ||
| 391 | |||
| 392 | mlog_entry_void(); | ||
| 393 | |||
| 394 | handle = ocfs2_alloc_handle(osb); | ||
| 395 | if (!handle) { | ||
| 396 | status = -ENOMEM; | ||
| 397 | mlog_errno(status); | ||
| 398 | goto bail; | ||
| 399 | } | ||
| 400 | |||
| 401 | main_bm_inode = ocfs2_get_system_file_inode(osb, | ||
| 402 | GLOBAL_BITMAP_SYSTEM_INODE, | ||
| 403 | OCFS2_INVALID_SLOT); | ||
| 404 | if (!main_bm_inode) { | ||
| 405 | status = -EINVAL; | ||
| 406 | mlog_errno(status); | ||
| 407 | goto bail; | ||
| 408 | } | ||
| 409 | |||
| 410 | ocfs2_handle_add_inode(handle, main_bm_inode); | ||
| 411 | status = ocfs2_meta_lock(main_bm_inode, handle, &main_bm_bh, 1); | ||
| 412 | if (status < 0) { | ||
| 413 | mlog_errno(status); | ||
| 414 | goto bail; | ||
| 415 | } | ||
| 416 | |||
| 417 | handle = ocfs2_start_trans(osb, handle, OCFS2_WINDOW_MOVE_CREDITS); | ||
| 418 | if (IS_ERR(handle)) { | ||
| 419 | status = PTR_ERR(handle); | ||
| 420 | handle = NULL; | ||
| 421 | mlog_errno(status); | ||
| 422 | goto bail; | ||
| 423 | } | ||
| 424 | |||
| 425 | /* we want the bitmap change to be recorded on disk asap */ | ||
| 426 | ocfs2_handle_set_sync(handle, 1); | ||
| 427 | |||
| 428 | status = ocfs2_sync_local_to_main(osb, handle, alloc, | ||
| 429 | main_bm_inode, main_bm_bh); | ||
| 430 | if (status < 0) | ||
| 431 | mlog_errno(status); | ||
| 432 | |||
| 433 | bail: | ||
| 434 | if (handle) | ||
| 435 | ocfs2_commit_trans(handle); | ||
| 436 | |||
| 437 | if (main_bm_bh) | ||
| 438 | brelse(main_bm_bh); | ||
| 439 | |||
| 440 | if (main_bm_inode) | ||
| 441 | iput(main_bm_inode); | ||
| 442 | |||
| 443 | mlog_exit(status); | ||
| 444 | return status; | ||
| 445 | } | ||
| 446 | |||
| 447 | /* | ||
| 448 | * make sure we've got at least bitswanted contiguous bits in the | ||
| 449 | * local alloc. You lose them when you drop i_mutex. | ||
| 450 | * | ||
| 451 | * We will add ourselves to the transaction passed in, but may start | ||
| 452 | * our own in order to shift windows. | ||
| 453 | */ | ||
| 454 | int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb, | ||
| 455 | struct ocfs2_journal_handle *passed_handle, | ||
| 456 | u32 bits_wanted, | ||
| 457 | struct ocfs2_alloc_context *ac) | ||
| 458 | { | ||
| 459 | int status; | ||
| 460 | struct ocfs2_dinode *alloc; | ||
| 461 | struct inode *local_alloc_inode; | ||
| 462 | unsigned int free_bits; | ||
| 463 | |||
| 464 | mlog_entry_void(); | ||
| 465 | |||
| 466 | BUG_ON(!passed_handle); | ||
| 467 | BUG_ON(!ac); | ||
| 468 | BUG_ON(passed_handle->flags & OCFS2_HANDLE_STARTED); | ||
| 469 | |||
| 470 | local_alloc_inode = | ||
| 471 | ocfs2_get_system_file_inode(osb, | ||
| 472 | LOCAL_ALLOC_SYSTEM_INODE, | ||
| 473 | osb->slot_num); | ||
| 474 | if (!local_alloc_inode) { | ||
| 475 | status = -ENOENT; | ||
| 476 | mlog_errno(status); | ||
| 477 | goto bail; | ||
| 478 | } | ||
| 479 | ocfs2_handle_add_inode(passed_handle, local_alloc_inode); | ||
| 480 | |||
| 481 | if (osb->local_alloc_state != OCFS2_LA_ENABLED) { | ||
| 482 | status = -ENOSPC; | ||
| 483 | goto bail; | ||
| 484 | } | ||
| 485 | |||
| 486 | if (bits_wanted > ocfs2_local_alloc_window_bits(osb)) { | ||
| 487 | mlog(0, "Asking for more than my max window size!\n"); | ||
| 488 | status = -ENOSPC; | ||
| 489 | goto bail; | ||
| 490 | } | ||
| 491 | |||
| 492 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | ||
| 493 | |||
| 494 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) != | ||
| 495 | ocfs2_local_alloc_count_bits(alloc)) { | ||
| 496 | ocfs2_error(osb->sb, "local alloc inode %"MLFu64" says it has " | ||
| 497 | "%u free bits, but a count shows %u", | ||
| 498 | le64_to_cpu(alloc->i_blkno), | ||
| 499 | le32_to_cpu(alloc->id1.bitmap1.i_used), | ||
| 500 | ocfs2_local_alloc_count_bits(alloc)); | ||
| 501 | status = -EIO; | ||
| 502 | goto bail; | ||
| 503 | } | ||
| 504 | |||
| 505 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | ||
| 506 | le32_to_cpu(alloc->id1.bitmap1.i_used); | ||
| 507 | if (bits_wanted > free_bits) { | ||
| 508 | /* uhoh, window change time. */ | ||
| 509 | status = | ||
| 510 | ocfs2_local_alloc_slide_window(osb, local_alloc_inode); | ||
| 511 | if (status < 0) { | ||
| 512 | if (status != -ENOSPC) | ||
| 513 | mlog_errno(status); | ||
| 514 | goto bail; | ||
| 515 | } | ||
| 516 | } | ||
| 517 | |||
| 518 | ac->ac_inode = igrab(local_alloc_inode); | ||
| 519 | get_bh(osb->local_alloc_bh); | ||
| 520 | ac->ac_bh = osb->local_alloc_bh; | ||
| 521 | ac->ac_which = OCFS2_AC_USE_LOCAL; | ||
| 522 | status = 0; | ||
| 523 | bail: | ||
| 524 | if (local_alloc_inode) | ||
| 525 | iput(local_alloc_inode); | ||
| 526 | |||
| 527 | mlog_exit(status); | ||
| 528 | return status; | ||
| 529 | } | ||
| 530 | |||
| 531 | int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb, | ||
| 532 | struct ocfs2_journal_handle *handle, | ||
| 533 | struct ocfs2_alloc_context *ac, | ||
| 534 | u32 min_bits, | ||
| 535 | u32 *bit_off, | ||
| 536 | u32 *num_bits) | ||
| 537 | { | ||
| 538 | int status, start; | ||
| 539 | struct inode *local_alloc_inode; | ||
| 540 | u32 bits_wanted; | ||
| 541 | void *bitmap; | ||
| 542 | struct ocfs2_dinode *alloc; | ||
| 543 | struct ocfs2_local_alloc *la; | ||
| 544 | |||
| 545 | mlog_entry_void(); | ||
| 546 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); | ||
| 547 | |||
| 548 | bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given; | ||
| 549 | local_alloc_inode = ac->ac_inode; | ||
| 550 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | ||
| 551 | la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 552 | |||
| 553 | start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted); | ||
| 554 | if (start == -1) { | ||
| 555 | /* TODO: Shouldn't we just BUG here? */ | ||
| 556 | status = -ENOSPC; | ||
| 557 | mlog_errno(status); | ||
| 558 | goto bail; | ||
| 559 | } | ||
| 560 | |||
| 561 | bitmap = la->la_bitmap; | ||
| 562 | *bit_off = le32_to_cpu(la->la_bm_off) + start; | ||
| 563 | /* local alloc is always contiguous by nature -- we never | ||
| 564 | * delete bits from it! */ | ||
| 565 | *num_bits = bits_wanted; | ||
| 566 | |||
| 567 | status = ocfs2_journal_access(handle, local_alloc_inode, | ||
| 568 | osb->local_alloc_bh, | ||
| 569 | OCFS2_JOURNAL_ACCESS_WRITE); | ||
| 570 | if (status < 0) { | ||
| 571 | mlog_errno(status); | ||
| 572 | goto bail; | ||
| 573 | } | ||
| 574 | |||
| 575 | while(bits_wanted--) | ||
| 576 | ocfs2_set_bit(start++, bitmap); | ||
| 577 | |||
| 578 | alloc->id1.bitmap1.i_used = cpu_to_le32(*num_bits + | ||
| 579 | le32_to_cpu(alloc->id1.bitmap1.i_used)); | ||
| 580 | |||
| 581 | status = ocfs2_journal_dirty(handle, osb->local_alloc_bh); | ||
| 582 | if (status < 0) { | ||
| 583 | mlog_errno(status); | ||
| 584 | goto bail; | ||
| 585 | } | ||
| 586 | |||
| 587 | status = 0; | ||
| 588 | bail: | ||
| 589 | mlog_exit(status); | ||
| 590 | return status; | ||
| 591 | } | ||
| 592 | |||
| 593 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc) | ||
| 594 | { | ||
| 595 | int i; | ||
| 596 | u8 *buffer; | ||
| 597 | u32 count = 0; | ||
| 598 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 599 | |||
| 600 | mlog_entry_void(); | ||
| 601 | |||
| 602 | buffer = la->la_bitmap; | ||
| 603 | for (i = 0; i < le16_to_cpu(la->la_size); i++) | ||
| 604 | count += hweight8(buffer[i]); | ||
| 605 | |||
| 606 | mlog_exit(count); | ||
| 607 | return count; | ||
| 608 | } | ||
| 609 | |||
| 610 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | ||
| 611 | struct ocfs2_dinode *alloc, | ||
| 612 | u32 numbits) | ||
| 613 | { | ||
| 614 | int numfound, bitoff, left, startoff, lastzero; | ||
| 615 | void *bitmap = NULL; | ||
| 616 | |||
| 617 | mlog_entry("(numbits wanted = %u)\n", numbits); | ||
| 618 | |||
| 619 | if (!alloc->id1.bitmap1.i_total) { | ||
| 620 | mlog(0, "No bits in my window!\n"); | ||
| 621 | bitoff = -1; | ||
| 622 | goto bail; | ||
| 623 | } | ||
| 624 | |||
| 625 | bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap; | ||
| 626 | |||
| 627 | numfound = bitoff = startoff = 0; | ||
| 628 | lastzero = -1; | ||
| 629 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | ||
| 630 | while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) { | ||
| 631 | if (bitoff == left) { | ||
| 632 | /* mlog(0, "bitoff (%d) == left", bitoff); */ | ||
| 633 | break; | ||
| 634 | } | ||
| 635 | /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, " | ||
| 636 | "numfound = %d\n", bitoff, startoff, numfound);*/ | ||
| 637 | |||
| 638 | /* Ok, we found a zero bit... is it contig. or do we | ||
| 639 | * start over?*/ | ||
| 640 | if (bitoff == startoff) { | ||
| 641 | /* we found a zero */ | ||
| 642 | numfound++; | ||
| 643 | startoff++; | ||
| 644 | } else { | ||
| 645 | /* got a zero after some ones */ | ||
| 646 | numfound = 1; | ||
| 647 | startoff = bitoff+1; | ||
| 648 | } | ||
| 649 | /* we got everything we needed */ | ||
| 650 | if (numfound == numbits) { | ||
| 651 | /* mlog(0, "Found it all!\n"); */ | ||
| 652 | break; | ||
| 653 | } | ||
| 654 | } | ||
| 655 | |||
| 656 | mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff, | ||
| 657 | numfound); | ||
| 658 | |||
| 659 | if (numfound == numbits) | ||
| 660 | bitoff = startoff - numfound; | ||
| 661 | else | ||
| 662 | bitoff = -1; | ||
| 663 | |||
| 664 | bail: | ||
| 665 | mlog_exit(bitoff); | ||
| 666 | return bitoff; | ||
| 667 | } | ||
| 668 | |||
| 669 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc) | ||
| 670 | { | ||
| 671 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 672 | int i; | ||
| 673 | mlog_entry_void(); | ||
| 674 | |||
| 675 | alloc->id1.bitmap1.i_total = 0; | ||
| 676 | alloc->id1.bitmap1.i_used = 0; | ||
| 677 | la->la_bm_off = 0; | ||
| 678 | for(i = 0; i < le16_to_cpu(la->la_size); i++) | ||
| 679 | la->la_bitmap[i] = 0; | ||
| 680 | |||
| 681 | mlog_exit_void(); | ||
| 682 | } | ||
| 683 | |||
| 684 | #if 0 | ||
| 685 | /* turn this on and uncomment below to aid debugging window shifts. */ | ||
| 686 | static void ocfs2_verify_zero_bits(unsigned long *bitmap, | ||
| 687 | unsigned int start, | ||
| 688 | unsigned int count) | ||
| 689 | { | ||
| 690 | unsigned int tmp = count; | ||
| 691 | while(tmp--) { | ||
| 692 | if (ocfs2_test_bit(start + tmp, bitmap)) { | ||
| 693 | printk("ocfs2_verify_zero_bits: start = %u, count = " | ||
| 694 | "%u\n", start, count); | ||
| 695 | printk("ocfs2_verify_zero_bits: bit %u is set!", | ||
| 696 | start + tmp); | ||
| 697 | BUG(); | ||
| 698 | } | ||
| 699 | } | ||
| 700 | } | ||
| 701 | #endif | ||
| 702 | |||
| 703 | /* | ||
| 704 | * sync the local alloc to main bitmap. | ||
| 705 | * | ||
| 706 | * assumes you've already locked the main bitmap -- the bitmap inode | ||
| 707 | * passed is used for caching. | ||
| 708 | */ | ||
| 709 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | ||
| 710 | struct ocfs2_journal_handle *handle, | ||
| 711 | struct ocfs2_dinode *alloc, | ||
| 712 | struct inode *main_bm_inode, | ||
| 713 | struct buffer_head *main_bm_bh) | ||
| 714 | { | ||
| 715 | int status = 0; | ||
| 716 | int bit_off, left, count, start; | ||
| 717 | u64 la_start_blk; | ||
| 718 | u64 blkno; | ||
| 719 | void *bitmap; | ||
| 720 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 721 | |||
| 722 | mlog_entry("total = %u, COUNT = %u, used = %u\n", | ||
| 723 | le32_to_cpu(alloc->id1.bitmap1.i_total), | ||
| 724 | ocfs2_local_alloc_count_bits(alloc), | ||
| 725 | le32_to_cpu(alloc->id1.bitmap1.i_used)); | ||
| 726 | |||
| 727 | if (!alloc->id1.bitmap1.i_total) { | ||
| 728 | mlog(0, "nothing to sync!\n"); | ||
| 729 | goto bail; | ||
| 730 | } | ||
| 731 | |||
| 732 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) == | ||
| 733 | le32_to_cpu(alloc->id1.bitmap1.i_total)) { | ||
| 734 | mlog(0, "all bits were taken!\n"); | ||
| 735 | goto bail; | ||
| 736 | } | ||
| 737 | |||
| 738 | la_start_blk = ocfs2_clusters_to_blocks(osb->sb, | ||
| 739 | le32_to_cpu(la->la_bm_off)); | ||
| 740 | bitmap = la->la_bitmap; | ||
| 741 | start = count = bit_off = 0; | ||
| 742 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | ||
| 743 | |||
| 744 | while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start)) | ||
| 745 | != -1) { | ||
| 746 | if ((bit_off < left) && (bit_off == start)) { | ||
| 747 | count++; | ||
| 748 | start++; | ||
| 749 | continue; | ||
| 750 | } | ||
| 751 | if (count) { | ||
| 752 | blkno = la_start_blk + | ||
| 753 | ocfs2_clusters_to_blocks(osb->sb, | ||
| 754 | start - count); | ||
| 755 | |||
| 756 | mlog(0, "freeing %u bits starting at local " | ||
| 757 | "alloc bit %u (la_start_blk = %"MLFu64", " | ||
| 758 | "blkno = %"MLFu64")\n", count, start - count, | ||
| 759 | la_start_blk, blkno); | ||
| 760 | |||
| 761 | status = ocfs2_free_clusters(handle, main_bm_inode, | ||
| 762 | main_bm_bh, blkno, count); | ||
| 763 | if (status < 0) { | ||
| 764 | mlog_errno(status); | ||
| 765 | goto bail; | ||
| 766 | } | ||
| 767 | } | ||
| 768 | if (bit_off >= left) | ||
| 769 | break; | ||
| 770 | count = 1; | ||
| 771 | start = bit_off + 1; | ||
| 772 | } | ||
| 773 | |||
| 774 | bail: | ||
| 775 | mlog_exit(status); | ||
| 776 | return status; | ||
| 777 | } | ||
| 778 | |||
| 779 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, | ||
| 780 | struct ocfs2_journal_handle *handle, | ||
| 781 | struct ocfs2_alloc_context **ac, | ||
| 782 | struct inode **bitmap_inode, | ||
| 783 | struct buffer_head **bitmap_bh) | ||
| 784 | { | ||
| 785 | int status; | ||
| 786 | |||
| 787 | *ac = kcalloc(1, sizeof(struct ocfs2_alloc_context), GFP_KERNEL); | ||
| 788 | if (!(*ac)) { | ||
| 789 | status = -ENOMEM; | ||
| 790 | mlog_errno(status); | ||
| 791 | goto bail; | ||
| 792 | } | ||
| 793 | |||
| 794 | (*ac)->ac_handle = handle; | ||
| 795 | (*ac)->ac_bits_wanted = ocfs2_local_alloc_window_bits(osb); | ||
| 796 | |||
| 797 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); | ||
| 798 | if (status < 0) { | ||
| 799 | if (status != -ENOSPC) | ||
| 800 | mlog_errno(status); | ||
| 801 | goto bail; | ||
| 802 | } | ||
| 803 | |||
| 804 | *bitmap_inode = (*ac)->ac_inode; | ||
| 805 | igrab(*bitmap_inode); | ||
| 806 | *bitmap_bh = (*ac)->ac_bh; | ||
| 807 | get_bh(*bitmap_bh); | ||
| 808 | status = 0; | ||
| 809 | bail: | ||
| 810 | if ((status < 0) && *ac) { | ||
| 811 | ocfs2_free_alloc_context(*ac); | ||
| 812 | *ac = NULL; | ||
| 813 | } | ||
| 814 | |||
| 815 | mlog_exit(status); | ||
| 816 | return status; | ||
| 817 | } | ||
| 818 | |||
| 819 | /* | ||
| 820 | * pass it the bitmap lock in lock_bh if you have it. | ||
| 821 | */ | ||
| 822 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | ||
| 823 | struct ocfs2_journal_handle *handle, | ||
| 824 | struct ocfs2_alloc_context *ac) | ||
| 825 | { | ||
| 826 | int status = 0; | ||
| 827 | u32 cluster_off, cluster_count; | ||
| 828 | struct ocfs2_dinode *alloc = NULL; | ||
| 829 | struct ocfs2_local_alloc *la; | ||
| 830 | |||
| 831 | mlog_entry_void(); | ||
| 832 | |||
| 833 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | ||
| 834 | la = OCFS2_LOCAL_ALLOC(alloc); | ||
| 835 | |||
| 836 | if (alloc->id1.bitmap1.i_total) | ||
| 837 | mlog(0, "asking me to alloc a new window over a non-empty " | ||
| 838 | "one\n"); | ||
| 839 | |||
| 840 | mlog(0, "Allocating %u clusters for a new window.\n", | ||
| 841 | ocfs2_local_alloc_window_bits(osb)); | ||
| 842 | /* we used the generic suballoc reserve function, but we set | ||
| 843 | * everything up nicely, so there's no reason why we can't use | ||
| 844 | * the more specific cluster api to claim bits. */ | ||
| 845 | status = ocfs2_claim_clusters(osb, handle, ac, | ||
| 846 | ocfs2_local_alloc_window_bits(osb), | ||
| 847 | &cluster_off, &cluster_count); | ||
| 848 | if (status < 0) { | ||
| 849 | if (status != -ENOSPC) | ||
| 850 | mlog_errno(status); | ||
| 851 | goto bail; | ||
| 852 | } | ||
| 853 | |||
| 854 | la->la_bm_off = cpu_to_le32(cluster_off); | ||
| 855 | alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count); | ||
| 856 | /* just in case... In the future when we find space ourselves, | ||
| 857 | * we don't have to get all contiguous -- but we'll have to | ||
| 858 | * set all previously used bits in bitmap and update | ||
| 859 | * la_bits_set before setting the bits in the main bitmap. */ | ||
| 860 | alloc->id1.bitmap1.i_used = 0; | ||
| 861 | memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0, | ||
| 862 | le16_to_cpu(la->la_size)); | ||
| 863 | |||
| 864 | mlog(0, "New window allocated:\n"); | ||
| 865 | mlog(0, "window la_bm_off = %u\n", | ||
| 866 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off); | ||
| 867 | mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total)); | ||
| 868 | |||
| 869 | bail: | ||
| 870 | mlog_exit(status); | ||
| 871 | return status; | ||
| 872 | } | ||
| 873 | |||
| 874 | /* Note that we do *NOT* lock the local alloc inode here as | ||
| 875 | * it's been locked already for us. */ | ||
| 876 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | ||
| 877 | struct inode *local_alloc_inode) | ||
| 878 | { | ||
| 879 | int status = 0; | ||
| 880 | struct buffer_head *main_bm_bh = NULL; | ||
| 881 | struct inode *main_bm_inode = NULL; | ||
| 882 | struct ocfs2_journal_handle *handle = NULL; | ||
| 883 | struct ocfs2_dinode *alloc; | ||
| 884 | struct ocfs2_dinode *alloc_copy = NULL; | ||
| 885 | struct ocfs2_alloc_context *ac = NULL; | ||
| 886 | |||
| 887 | mlog_entry_void(); | ||
| 888 | |||
| 889 | handle = ocfs2_alloc_handle(osb); | ||
| 890 | if (!handle) { | ||
| 891 | status = -ENOMEM; | ||
| 892 | mlog_errno(status); | ||
| 893 | goto bail; | ||
| 894 | } | ||
| 895 | |||
| 896 | /* This will lock the main bitmap for us. */ | ||
| 897 | status = ocfs2_local_alloc_reserve_for_window(osb, | ||
| 898 | handle, | ||
| 899 | &ac, | ||
| 900 | &main_bm_inode, | ||
| 901 | &main_bm_bh); | ||
| 902 | if (status < 0) { | ||
| 903 | if (status != -ENOSPC) | ||
| 904 | mlog_errno(status); | ||
| 905 | goto bail; | ||
| 906 | } | ||
| 907 | |||
| 908 | handle = ocfs2_start_trans(osb, handle, OCFS2_WINDOW_MOVE_CREDITS); | ||
| 909 | if (IS_ERR(handle)) { | ||
| 910 | status = PTR_ERR(handle); | ||
| 911 | handle = NULL; | ||
| 912 | mlog_errno(status); | ||
| 913 | goto bail; | ||
| 914 | } | ||
| 915 | |||
| 916 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | ||
| 917 | |||
| 918 | /* We want to clear the local alloc before doing anything | ||
| 919 | * else, so that if we error later during this operation, | ||
| 920 | * local alloc shutdown won't try to double free main bitmap | ||
| 921 | * bits. Make a copy so the sync function knows which bits to | ||
| 922 | * free. */ | ||
| 923 | alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_KERNEL); | ||
| 924 | if (!alloc_copy) { | ||
| 925 | status = -ENOMEM; | ||
| 926 | mlog_errno(status); | ||
| 927 | goto bail; | ||
| 928 | } | ||
| 929 | memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size); | ||
| 930 | |||
| 931 | status = ocfs2_journal_access(handle, local_alloc_inode, | ||
| 932 | osb->local_alloc_bh, | ||
| 933 | OCFS2_JOURNAL_ACCESS_WRITE); | ||
| 934 | if (status < 0) { | ||
| 935 | mlog_errno(status); | ||
| 936 | goto bail; | ||
| 937 | } | ||
| 938 | |||
| 939 | ocfs2_clear_local_alloc(alloc); | ||
| 940 | |||
| 941 | status = ocfs2_journal_dirty(handle, osb->local_alloc_bh); | ||
| 942 | if (status < 0) { | ||
| 943 | mlog_errno(status); | ||
| 944 | goto bail; | ||
| 945 | } | ||
| 946 | |||
| 947 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | ||
| 948 | main_bm_inode, main_bm_bh); | ||
| 949 | if (status < 0) { | ||
| 950 | mlog_errno(status); | ||
| 951 | goto bail; | ||
| 952 | } | ||
| 953 | |||
| 954 | status = ocfs2_local_alloc_new_window(osb, handle, ac); | ||
| 955 | if (status < 0) { | ||
| 956 | if (status != -ENOSPC) | ||
| 957 | mlog_errno(status); | ||
| 958 | goto bail; | ||
| 959 | } | ||
| 960 | |||
| 961 | atomic_inc(&osb->alloc_stats.moves); | ||
| 962 | |||
| 963 | status = 0; | ||
| 964 | bail: | ||
| 965 | if (handle) | ||
| 966 | ocfs2_commit_trans(handle); | ||
| 967 | |||
| 968 | if (main_bm_bh) | ||
| 969 | brelse(main_bm_bh); | ||
| 970 | |||
| 971 | if (main_bm_inode) | ||
| 972 | iput(main_bm_inode); | ||
| 973 | |||
| 974 | if (alloc_copy) | ||
| 975 | kfree(alloc_copy); | ||
| 976 | |||
| 977 | if (ac) | ||
| 978 | ocfs2_free_alloc_context(ac); | ||
| 979 | |||
| 980 | mlog_exit(status); | ||
| 981 | return status; | ||
| 982 | } | ||
| 983 | |||
