1 /* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * journal.c 5 * 6 * Defines functions of journalling api 7 * 8 * Copyright (C) 2003, 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/kthread.h> 31 32 #define MLOG_MASK_PREFIX ML_JOURNAL 33 #include <cluster/masklog.h> 34 35 #include "ocfs2.h" 36 37 #include "alloc.h" 38 #include "dlmglue.h" 39 #include "extent_map.h" 40 #include "heartbeat.h" 41 #include "inode.h" 42 #include "journal.h" 43 #include "localalloc.h" 44 #include "namei.h" 45 #include "slot_map.h" 46 #include "super.h" 47 #include "vote.h" 48 #include "sysfile.h" 49 50 #include "buffer_head_io.h" 51 52 DEFINE_SPINLOCK(trans_inc_lock); 53 54 static int ocfs2_force_read_journal(struct inode *inode); 55 static int ocfs2_recover_node(struct ocfs2_super *osb, 56 int node_num); 57 static int __ocfs2_recovery_thread(void *arg); 58 static int ocfs2_commit_cache(struct ocfs2_super *osb); 59 static int ocfs2_wait_on_mount(struct ocfs2_super *osb); 60 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 61 int dirty); 62 static int ocfs2_trylock_journal(struct ocfs2_super *osb, 63 int slot_num); 64 static int ocfs2_recover_orphans(struct ocfs2_super *osb, 65 int slot); 66 static int ocfs2_commit_thread(void *arg); 67 68 static int ocfs2_commit_cache(struct ocfs2_super *osb) 69 { 70 int status = 0; 71 unsigned int flushed; 72 unsigned long old_id; 73 struct ocfs2_journal *journal = NULL; 74 75 mlog_entry_void(); 76 77 journal = osb->journal; 78 79 /* Flush all pending commits and checkpoint the journal. */ 80 down_write(&journal->j_trans_barrier); 81 82 if (atomic_read(&journal->j_num_trans) == 0) { 83 up_write(&journal->j_trans_barrier); 84 mlog(0, "No transactions for me to flush!\n"); 85 goto finally; 86 } 87 88 journal_lock_updates(journal->j_journal); 89 status = journal_flush(journal->j_journal); 90 journal_unlock_updates(journal->j_journal); 91 if (status < 0) { 92 up_write(&journal->j_trans_barrier); 93 mlog_errno(status); 94 goto finally; 95 } 96 97 old_id = ocfs2_inc_trans_id(journal); 98 99 flushed = atomic_read(&journal->j_num_trans); 100 atomic_set(&journal->j_num_trans, 0); 101 up_write(&journal->j_trans_barrier); 102 103 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 104 journal->j_trans_id, flushed); 105 106 ocfs2_kick_vote_thread(osb); 107 wake_up(&journal->j_checkpointed); 108 finally: 109 mlog_exit(status); 110 return status; 111 } 112 113 /* pass it NULL and it will allocate a new handle object for you. If 114 * you pass it a handle however, it may still return error, in which 115 * case it has free'd the passed handle for you. */ 116 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 117 { 118 journal_t *journal = osb->journal->j_journal; 119 handle_t *handle; 120 121 BUG_ON(!osb || !osb->journal->j_journal); 122 123 if (ocfs2_is_hard_readonly(osb)) 124 return ERR_PTR(-EROFS); 125 126 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 127 BUG_ON(max_buffs <= 0); 128 129 /* JBD might support this, but our journalling code doesn't yet. */ 130 if (journal_current_handle()) { 131 mlog(ML_ERROR, "Recursive transaction attempted!\n"); 132 BUG(); 133 } 134 135 down_read(&osb->journal->j_trans_barrier); 136 137 handle = journal_start(journal, max_buffs); 138 if (IS_ERR(handle)) { 139 up_read(&osb->journal->j_trans_barrier); 140 141 mlog_errno(PTR_ERR(handle)); 142 143 if (is_journal_aborted(journal)) { 144 ocfs2_abort(osb->sb, "Detected aborted journal"); 145 handle = ERR_PTR(-EROFS); 146 } 147 } else { 148 if (!ocfs2_mount_local(osb)) 149 atomic_inc(&(osb->journal->j_num_trans)); 150 } 151 152 return handle; 153 } 154 155 int ocfs2_commit_trans(struct ocfs2_super *osb, 156 handle_t *handle) 157 { 158 int ret; 159 struct ocfs2_journal *journal = osb->journal; 160 161 BUG_ON(!handle); 162 163 ret = journal_stop(handle); 164 if (ret < 0) 165 mlog_errno(ret); 166 167 up_read(&journal->j_trans_barrier); 168 169 return ret; 170 } 171 172 /* 173 * 'nblocks' is what you want to add to the current 174 * transaction. extend_trans will either extend the current handle by 175 * nblocks, or commit it and start a new one with nblocks credits. 176 * 177 * WARNING: This will not release any semaphores or disk locks taken 178 * during the transaction, so make sure they were taken *before* 179 * start_trans or we'll have ordering deadlocks. 180 * 181 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 182 * good because transaction ids haven't yet been recorded on the 183 * cluster locks associated with this handle. 184 */ 185 int ocfs2_extend_trans(handle_t *handle, int nblocks) 186 { 187 int status; 188 189 BUG_ON(!handle); 190 BUG_ON(!nblocks); 191 192 mlog_entry_void(); 193 194 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 195 196 status = journal_extend(handle, nblocks); 197 if (status < 0) { 198 mlog_errno(status); 199 goto bail; 200 } 201 202 if (status > 0) { 203 mlog(0, "journal_extend failed, trying journal_restart\n"); 204 status = journal_restart(handle, nblocks); 205 if (status < 0) { 206 mlog_errno(status); 207 goto bail; 208 } 209 } 210 211 status = 0; 212 bail: 213 214 mlog_exit(status); 215 return status; 216 } 217 218 int ocfs2_journal_access(handle_t *handle, 219 struct inode *inode, 220 struct buffer_head *bh, 221 int type) 222 { 223 int status; 224 225 BUG_ON(!inode); 226 BUG_ON(!handle); 227 BUG_ON(!bh); 228 229 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 230 (unsigned long long)bh->b_blocknr, type, 231 (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 232 "OCFS2_JOURNAL_ACCESS_CREATE" : 233 "OCFS2_JOURNAL_ACCESS_WRITE", 234 bh->b_size); 235 236 /* we can safely remove this assertion after testing. */ 237 if (!buffer_uptodate(bh)) { 238 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 239 mlog(ML_ERROR, "b_blocknr=%llu\n", 240 (unsigned long long)bh->b_blocknr); 241 BUG(); 242 } 243 244 /* Set the current transaction information on the inode so 245 * that the locking code knows whether it can drop it's locks 246 * on this inode or not. We're protected from the commit 247 * thread updating the current transaction id until 248 * ocfs2_commit_trans() because ocfs2_start_trans() took 249 * j_trans_barrier for us. */ 250 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 251 252 mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 253 switch (type) { 254 case OCFS2_JOURNAL_ACCESS_CREATE: 255 case OCFS2_JOURNAL_ACCESS_WRITE: 256 status = journal_get_write_access(handle, bh); 257 break; 258 259 case OCFS2_JOURNAL_ACCESS_UNDO: 260 status = journal_get_undo_access(handle, bh); 261 break; 262 263 default: 264 status = -EINVAL; 265 mlog(ML_ERROR, "Uknown access type!\n"); 266 } 267 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 268 269 if (status < 0) 270 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 271 status, type); 272 273 mlog_exit(status); 274 return status; 275 } 276 277 int ocfs2_journal_dirty(handle_t *handle, 278 struct buffer_head *bh) 279 { 280 int status; 281 282 mlog_entry("(bh->b_blocknr=%llu)\n", 283 (unsigned long long)bh->b_blocknr); 284 285 status = journal_dirty_metadata(handle, bh); 286 if (status < 0) 287 mlog(ML_ERROR, "Could not dirty metadata buffer. " 288 "(bh->b_blocknr=%llu)\n", 289 (unsigned long long)bh->b_blocknr); 290 291 mlog_exit(status); 292 return status; 293 } 294 295 int ocfs2_journal_dirty_data(handle_t *handle, 296 struct buffer_head *bh) 297 { 298 int err = journal_dirty_data(handle, bh); 299 if (err) 300 mlog_errno(err); 301 /* TODO: When we can handle it, abort the handle and go RO on 302 * error here. */ 303 304 return err; 305 } 306 307 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) 308 309 void ocfs2_set_journal_params(struct ocfs2_super *osb) 310 { 311 journal_t *journal = osb->journal->j_journal; 312 313 spin_lock(&journal->j_state_lock); 314 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 315 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 316 journal->j_flags |= JFS_BARRIER; 317 else 318 journal->j_flags &= ~JFS_BARRIER; 319 spin_unlock(&journal->j_state_lock); 320 } 321 322 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 323 { 324 int status = -1; 325 struct inode *inode = NULL; /* the journal inode */ 326 journal_t *j_journal = NULL; 327 struct ocfs2_dinode *di = NULL; 328 struct buffer_head *bh = NULL; 329 struct ocfs2_super *osb; 330 int meta_lock = 0; 331 332 mlog_entry_void(); 333 334 BUG_ON(!journal); 335 336 osb = journal->j_osb; 337 338 /* already have the inode for our journal */ 339 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 340 osb->slot_num); 341 if (inode == NULL) { 342 status = -EACCES; 343 mlog_errno(status); 344 goto done; 345 } 346 if (is_bad_inode(inode)) { 347 mlog(ML_ERROR, "access error (bad inode)\n"); 348 iput(inode); 349 inode = NULL; 350 status = -EACCES; 351 goto done; 352 } 353 354 SET_INODE_JOURNAL(inode); 355 OCFS2_I(inode)->ip_open_count++; 356 357 /* Skip recovery waits here - journal inode metadata never 358 * changes in a live cluster so it can be considered an 359 * exception to the rule. */ 360 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 361 if (status < 0) { 362 if (status != -ERESTARTSYS) 363 mlog(ML_ERROR, "Could not get lock on journal!\n"); 364 goto done; 365 } 366 367 meta_lock = 1; 368 di = (struct ocfs2_dinode *)bh->b_data; 369 370 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 371 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 372 inode->i_size); 373 status = -EINVAL; 374 goto done; 375 } 376 377 mlog(0, "inode->i_size = %lld\n", inode->i_size); 378 mlog(0, "inode->i_blocks = %llu\n", 379 (unsigned long long)inode->i_blocks); 380 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 381 382 /* call the kernels journal init function now */ 383 j_journal = journal_init_inode(inode); 384 if (j_journal == NULL) { 385 mlog(ML_ERROR, "Linux journal layer error\n"); 386 status = -EINVAL; 387 goto done; 388 } 389 390 mlog(0, "Returned from journal_init_inode\n"); 391 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 392 393 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 394 OCFS2_JOURNAL_DIRTY_FL); 395 396 journal->j_journal = j_journal; 397 journal->j_inode = inode; 398 journal->j_bh = bh; 399 400 ocfs2_set_journal_params(osb); 401 402 journal->j_state = OCFS2_JOURNAL_LOADED; 403 404 status = 0; 405 done: 406 if (status < 0) { 407 if (meta_lock) 408 ocfs2_meta_unlock(inode, 1); 409 if (bh != NULL) 410 brelse(bh); 411 if (inode) { 412 OCFS2_I(inode)->ip_open_count--; 413 iput(inode); 414 } 415 } 416 417 mlog_exit(status); 418 return status; 419 } 420 421 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 422 int dirty) 423 { 424 int status; 425 unsigned int flags; 426 struct ocfs2_journal *journal = osb->journal; 427 struct buffer_head *bh = journal->j_bh; 428 struct ocfs2_dinode *fe; 429 430 mlog_entry_void(); 431 432 fe = (struct ocfs2_dinode *)bh->b_data; 433 if (!OCFS2_IS_VALID_DINODE(fe)) { 434 /* This is called from startup/shutdown which will 435 * handle the errors in a specific manner, so no need 436 * to call ocfs2_error() here. */ 437 mlog(ML_ERROR, "Journal dinode %llu has invalid " 438 "signature: %.*s", 439 (unsigned long long)le64_to_cpu(fe->i_blkno), 7, 440 fe->i_signature); 441 status = -EIO; 442 goto out; 443 } 444 445 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 446 if (dirty) 447 flags |= OCFS2_JOURNAL_DIRTY_FL; 448 else 449 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 450 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 451 452 status = ocfs2_write_block(osb, bh, journal->j_inode); 453 if (status < 0) 454 mlog_errno(status); 455 456 out: 457 mlog_exit(status); 458 return status; 459 } 460 461 /* 462 * If the journal has been kmalloc'd it needs to be freed after this 463 * call. 464 */ 465 void ocfs2_journal_shutdown(struct ocfs2_super *osb) 466 { 467 struct ocfs2_journal *journal = NULL; 468 int status = 0; 469 struct inode *inode = NULL; 470 int num_running_trans = 0; 471 472 mlog_entry_void(); 473 474 BUG_ON(!osb); 475 476 journal = osb->journal; 477 if (!journal) 478 goto done; 479 480 inode = journal->j_inode; 481 482 if (journal->j_state != OCFS2_JOURNAL_LOADED) 483 goto done; 484 485 /* need to inc inode use count as journal_destroy will iput. */ 486 if (!igrab(inode)) 487 BUG(); 488 489 num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 490 if (num_running_trans > 0) 491 mlog(0, "Shutting down journal: must wait on %d " 492 "running transactions!\n", 493 num_running_trans); 494 495 /* Do a commit_cache here. It will flush our journal, *and* 496 * release any locks that are still held. 497 * set the SHUTDOWN flag and release the trans lock. 498 * the commit thread will take the trans lock for us below. */ 499 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 500 501 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 502 * drop the trans_lock (which we want to hold until we 503 * completely destroy the journal. */ 504 if (osb->commit_task) { 505 /* Wait for the commit thread */ 506 mlog(0, "Waiting for ocfs2commit to exit....\n"); 507 kthread_stop(osb->commit_task); 508 osb->commit_task = NULL; 509 } 510 511 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 512 513 if (ocfs2_mount_local(osb)) { 514 journal_lock_updates(journal->j_journal); 515 status = journal_flush(journal->j_journal); 516 journal_unlock_updates(journal->j_journal); 517 if (status < 0) 518 mlog_errno(status); 519 } 520 521 if (status == 0) { 522 /* 523 * Do not toggle if flush was unsuccessful otherwise 524 * will leave dirty metadata in a "clean" journal 525 */ 526 status = ocfs2_journal_toggle_dirty(osb, 0); 527 if (status < 0) 528 mlog_errno(status); 529 } 530 531 /* Shutdown the kernel journal system */ 532 journal_destroy(journal->j_journal); 533 534 OCFS2_I(inode)->ip_open_count--; 535 536 /* unlock our journal */ 537 ocfs2_meta_unlock(inode, 1); 538 539 brelse(journal->j_bh); 540 journal->j_bh = NULL; 541 542 journal->j_state = OCFS2_JOURNAL_FREE; 543 544 // up_write(&journal->j_trans_barrier); 545 done: 546 if (inode) 547 iput(inode); 548 mlog_exit_void(); 549 } 550 551 static void ocfs2_clear_journal_error(struct super_block *sb, 552 journal_t *journal, 553 int slot) 554 { 555 int olderr; 556 557 olderr = journal_errno(journal); 558 if (olderr) { 559 mlog(ML_ERROR, "File system error %d recorded in " 560 "journal %u.\n", olderr, slot); 561 mlog(ML_ERROR, "File system on device %s needs checking.\n", 562 sb->s_id); 563 564 journal_ack_err(journal); 565 journal_clear_err(journal); 566 } 567 } 568 569 int ocfs2_journal_load(struct ocfs2_journal *journal, int local) 570 { 571 int status = 0; 572 struct ocfs2_super *osb; 573 574 mlog_entry_void(); 575 576 if (!journal) 577 BUG(); 578 579 osb = journal->j_osb; 580 581 status = journal_load(journal->j_journal); 582 if (status < 0) { 583 mlog(ML_ERROR, "Failed to load journal!\n"); 584 goto done; 585 } 586 587 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 588 589 status = ocfs2_journal_toggle_dirty(osb, 1); 590 if (status < 0) { 591 mlog_errno(status); 592 goto done; 593 } 594 595 /* Launch the commit thread */ 596 if (!local) { 597 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 598 "ocfs2cmt"); 599 if (IS_ERR(osb->commit_task)) { 600 status = PTR_ERR(osb->commit_task); 601 osb->commit_task = NULL; 602 mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 603 "error=%d", status); 604 goto done; 605 } 606 } else 607 osb->commit_task = NULL; 608 609 done: 610 mlog_exit(status); 611 return status; 612 } 613 614 615 /* 'full' flag tells us whether we clear out all blocks or if we just 616 * mark the journal clean */ 617 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 618 { 619 int status; 620 621 mlog_entry_void(); 622 623 BUG_ON(!journal); 624 625 status = journal_wipe(journal->j_journal, full); 626 if (status < 0) { 627 mlog_errno(status); 628 goto bail; 629 } 630 631 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); 632 if (status < 0) 633 mlog_errno(status); 634 635 bail: 636 mlog_exit(status); 637 return status; 638 } 639 640 /* 641 * JBD Might read a cached version of another nodes journal file. We 642 * don't want this as this file changes often and we get no 643 * notification on those changes. The only way to be sure that we've 644 * got the most up to date version of those blocks then is to force 645 * read them off disk. Just searching through the buffer cache won't 646 * work as there may be pages backing this file which are still marked 647 * up to date. We know things can't change on this file underneath us 648 * as we have the lock by now :) 649 */ 650 static int ocfs2_force_read_journal(struct inode *inode) 651 { 652 int status = 0; 653 int i; 654 u64 v_blkno, p_blkno, p_blocks, num_blocks; 655 #define CONCURRENT_JOURNAL_FILL 32ULL 656 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 657 658 mlog_entry_void(); 659 660 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 661 662 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); 663 v_blkno = 0; 664 while (v_blkno < num_blocks) { 665 status = ocfs2_extent_map_get_blocks(inode, v_blkno, 666 &p_blkno, &p_blocks, NULL); 667 if (status < 0) { 668 mlog_errno(status); 669 goto bail; 670 } 671 672 if (p_blocks > CONCURRENT_JOURNAL_FILL) 673 p_blocks = CONCURRENT_JOURNAL_FILL; 674 675 /* We are reading journal data which should not 676 * be put in the uptodate cache */ 677 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 678 p_blkno, p_blocks, bhs, 0, 679 NULL); 680 if (status < 0) { 681 mlog_errno(status); 682 goto bail; 683 } 684 685 for(i = 0; i < p_blocks; i++) { 686 brelse(bhs[i]); 687 bhs[i] = NULL; 688 } 689 690 v_blkno += p_blocks; 691 } 692 693 bail: 694 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 695 if (bhs[i]) 696 brelse(bhs[i]); 697 mlog_exit(status); 698 return status; 699 } 700 701 struct ocfs2_la_recovery_item { 702 struct list_head lri_list; 703 int lri_slot; 704 struct ocfs2_dinode *lri_la_dinode; 705 struct ocfs2_dinode *lri_tl_dinode; 706 }; 707 708 /* Does the second half of the recovery process. By this point, the 709 * node is marked clean and can actually be considered recovered, 710 * hence it's no longer in the recovery map, but there's still some 711 * cleanup we can do which shouldn't happen within the recovery thread 712 * as locking in that context becomes very difficult if we are to take 713 * recovering nodes into account. 714 * 715 * NOTE: This function can and will sleep on recovery of other nodes 716 * during cluster locking, just like any other ocfs2 process. 717 */ 718 void ocfs2_complete_recovery(struct work_struct *work) 719 { 720 int ret; 721 struct ocfs2_journal *journal = 722 container_of(work, struct ocfs2_journal, j_recovery_work); 723 struct ocfs2_super *osb = journal->j_osb; 724 struct ocfs2_dinode *la_dinode, *tl_dinode; 725 struct ocfs2_la_recovery_item *item; 726 struct list_head *p, *n; 727 LIST_HEAD(tmp_la_list); 728 729 mlog_entry_void(); 730 731 mlog(0, "completing recovery from keventd\n"); 732 733 spin_lock(&journal->j_lock); 734 list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 735 spin_unlock(&journal->j_lock); 736 737 list_for_each_safe(p, n, &tmp_la_list) { 738 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list); 739 list_del_init(&item->lri_list); 740 741 mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 742 743 la_dinode = item->lri_la_dinode; 744 if (la_dinode) { 745 mlog(0, "Clean up local alloc %llu\n", 746 (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); 747 748 ret = ocfs2_complete_local_alloc_recovery(osb, 749 la_dinode); 750 if (ret < 0) 751 mlog_errno(ret); 752 753 kfree(la_dinode); 754 } 755 756 tl_dinode = item->lri_tl_dinode; 757 if (tl_dinode) { 758 mlog(0, "Clean up truncate log %llu\n", 759 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); 760 761 ret = ocfs2_complete_truncate_log_recovery(osb, 762 tl_dinode); 763 if (ret < 0) 764 mlog_errno(ret); 765 766 kfree(tl_dinode); 767 } 768 769 ret = ocfs2_recover_orphans(osb, item->lri_slot); 770 if (ret < 0) 771 mlog_errno(ret); 772 773 kfree(item); 774 } 775 776 mlog(0, "Recovery completion\n"); 777 mlog_exit_void(); 778 } 779 780 /* NOTE: This function always eats your references to la_dinode and 781 * tl_dinode, either manually on error, or by passing them to 782 * ocfs2_complete_recovery */ 783 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 784 int slot_num, 785 struct ocfs2_dinode *la_dinode, 786 struct ocfs2_dinode *tl_dinode) 787 { 788 struct ocfs2_la_recovery_item *item; 789 790 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 791 if (!item) { 792 /* Though we wish to avoid it, we are in fact safe in 793 * skipping local alloc cleanup as fsck.ocfs2 is more 794 * than capable of reclaiming unused space. */ 795 if (la_dinode) 796 kfree(la_dinode); 797 798 if (tl_dinode) 799 kfree(tl_dinode); 800 801 mlog_errno(-ENOMEM); 802 return; 803 } 804 805 INIT_LIST_HEAD(&item->lri_list); 806 item->lri_la_dinode = la_dinode; 807 item->lri_slot = slot_num; 808 item->lri_tl_dinode = tl_dinode; 809 810 spin_lock(&journal->j_lock); 811 list_add_tail(&item->lri_list, &journal->j_la_cleanups); 812 queue_work(ocfs2_wq, &journal->j_recovery_work); 813 spin_unlock(&journal->j_lock); 814 } 815 816 /* Called by the mount code to queue recovery the last part of 817 * recovery for it's own slot. */ 818 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 819 { 820 struct ocfs2_journal *journal = osb->journal; 821 822 if (osb->dirty) { 823 /* No need to queue up our truncate_log as regular 824 * cleanup will catch that. */ 825 ocfs2_queue_recovery_completion(journal, 826 osb->slot_num, 827 osb->local_alloc_copy, 828 NULL); 829 ocfs2_schedule_truncate_log_flush(osb, 0); 830 831 osb->local_alloc_copy = NULL; 832 osb->dirty = 0; 833 } 834 } 835 836 static int __ocfs2_recovery_thread(void *arg) 837 { 838 int status, node_num; 839 struct ocfs2_super *osb = arg; 840 841 mlog_entry_void(); 842 843 status = ocfs2_wait_on_mount(osb); 844 if (status < 0) { 845 goto bail; 846 } 847 848 restart: 849 status = ocfs2_super_lock(osb, 1); 850 if (status < 0) { 851 mlog_errno(status); 852 goto bail; 853 } 854 855 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 856 node_num = ocfs2_node_map_first_set_bit(osb, 857 &osb->recovery_map); 858 if (node_num == O2NM_INVALID_NODE_NUM) { 859 mlog(0, "Out of nodes to recover.\n"); 860 break; 861 } 862 863 status = ocfs2_recover_node(osb, node_num); 864 if (status < 0) { 865 mlog(ML_ERROR, 866 "Error %d recovering node %d on device (%u,%u)!\n", 867 status, node_num, 868 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 869 mlog(ML_ERROR, "Volume requires unmount.\n"); 870 continue; 871 } 872 873 ocfs2_recovery_map_clear(osb, node_num); 874 } 875 ocfs2_super_unlock(osb, 1); 876 877 /* We always run recovery on our own orphan dir - the dead 878 * node(s) may have voted "no" on an inode delete earlier. A 879 * revote is therefore required. */ 880 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 881 NULL); 882 883 bail: 884 mutex_lock(&osb->recovery_lock); 885 if (!status && 886 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 887 mutex_unlock(&osb->recovery_lock); 888 goto restart; 889 } 890 891 osb->recovery_thread_task = NULL; 892 mb(); /* sync with ocfs2_recovery_thread_running */ 893 wake_up(&osb->recovery_event); 894 895 mutex_unlock(&osb->recovery_lock); 896 897 mlog_exit(status); 898 /* no one is callint kthread_stop() for us so the kthread() api 899 * requires that we call do_exit(). And it isn't exported, but 900 * complete_and_exit() seems to be a minimal wrapper around it. */ 901 complete_and_exit(NULL, status); 902 return status; 903 } 904 905 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 906 { 907 mlog_entry("(node_num=%d, osb->node_num = %d)\n", 908 node_num, osb->node_num); 909 910 mutex_lock(&osb->recovery_lock); 911 if (osb->disable_recovery) 912 goto out; 913 914 /* People waiting on recovery will wait on 915 * the recovery map to empty. */ 916 if (!ocfs2_recovery_map_set(osb, node_num)) 917 mlog(0, "node %d already be in recovery.\n", node_num); 918 919 mlog(0, "starting recovery thread...\n"); 920 921 if (osb->recovery_thread_task) 922 goto out; 923 924 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 925 "ocfs2rec"); 926 if (IS_ERR(osb->recovery_thread_task)) { 927 mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 928 osb->recovery_thread_task = NULL; 929 } 930 931 out: 932 mutex_unlock(&osb->recovery_lock); 933 wake_up(&osb->recovery_event); 934 935 mlog_exit_void(); 936 } 937 938 /* Does the actual journal replay and marks the journal inode as 939 * clean. Will only replay if the journal inode is marked dirty. */ 940 static int ocfs2_replay_journal(struct ocfs2_super *osb, 941 int node_num, 942 int slot_num) 943 { 944 int status; 945 int got_lock = 0; 946 unsigned int flags; 947 struct inode *inode = NULL; 948 struct ocfs2_dinode *fe; 949 journal_t *journal = NULL; 950 struct buffer_head *bh = NULL; 951 952 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 953 slot_num); 954 if (inode == NULL) { 955 status = -EACCES; 956 mlog_errno(status); 957 goto done; 958 } 959 if (is_bad_inode(inode)) { 960 status = -EACCES; 961 iput(inode); 962 inode = NULL; 963 mlog_errno(status); 964 goto done; 965 } 966 SET_INODE_JOURNAL(inode); 967 968 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 969 if (status < 0) { 970 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); 971 if (status != -ERESTARTSYS) 972 mlog(ML_ERROR, "Could not lock journal!\n"); 973 goto done; 974 } 975 got_lock = 1; 976 977 fe = (struct ocfs2_dinode *) bh->b_data; 978 979 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 980 981 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 982 mlog(0, "No recovery required for node %d\n", node_num); 983 goto done; 984 } 985 986 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 987 node_num, slot_num, 988 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 989 990 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 991 992 status = ocfs2_force_read_journal(inode); 993 if (status < 0) { 994 mlog_errno(status); 995 goto done; 996 } 997 998 mlog(0, "calling journal_init_inode\n"); 999 journal = journal_init_inode(inode); 1000 if (journal == NULL) { 1001 mlog(ML_ERROR, "Linux journal layer error\n"); 1002 status = -EIO; 1003 goto done; 1004 } 1005 1006 status = journal_load(journal); 1007 if (status < 0) { 1008 mlog_errno(status); 1009 if (!igrab(inode)) 1010 BUG(); 1011 journal_destroy(journal); 1012 goto done; 1013 } 1014 1015 ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1016 1017 /* wipe the journal */ 1018 mlog(0, "flushing the journal.\n"); 1019 journal_lock_updates(journal); 1020 status = journal_flush(journal); 1021 journal_unlock_updates(journal); 1022 if (status < 0) 1023 mlog_errno(status); 1024 1025 /* This will mark the node clean */ 1026 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1027 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1028 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1029 1030 status = ocfs2_write_block(osb, bh, inode); 1031 if (status < 0) 1032 mlog_errno(status); 1033 1034 if (!igrab(inode)) 1035 BUG(); 1036 1037 journal_destroy(journal); 1038 1039 done: 1040 /* drop the lock on this nodes journal */ 1041 if (got_lock) 1042 ocfs2_meta_unlock(inode, 1); 1043 1044 if (inode) 1045 iput(inode); 1046 1047 if (bh) 1048 brelse(bh); 1049 1050 mlog_exit(status); 1051 return status; 1052 } 1053 1054 /* 1055 * Do the most important parts of node recovery: 1056 * - Replay it's journal 1057 * - Stamp a clean local allocator file 1058 * - Stamp a clean truncate log 1059 * - Mark the node clean 1060 * 1061 * If this function completes without error, a node in OCFS2 can be 1062 * said to have been safely recovered. As a result, failure during the 1063 * second part of a nodes recovery process (local alloc recovery) is 1064 * far less concerning. 1065 */ 1066 static int ocfs2_recover_node(struct ocfs2_super *osb, 1067 int node_num) 1068 { 1069 int status = 0; 1070 int slot_num; 1071 struct ocfs2_slot_info *si = osb->slot_info; 1072 struct ocfs2_dinode *la_copy = NULL; 1073 struct ocfs2_dinode *tl_copy = NULL; 1074 1075 mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1076 node_num, osb->node_num); 1077 1078 mlog(0, "checking node %d\n", node_num); 1079 1080 /* Should not ever be called to recover ourselves -- in that 1081 * case we should've called ocfs2_journal_load instead. */ 1082 BUG_ON(osb->node_num == node_num); 1083 1084 slot_num = ocfs2_node_num_to_slot(si, node_num); 1085 if (slot_num == OCFS2_INVALID_SLOT) { 1086 status = 0; 1087 mlog(0, "no slot for this node, so no recovery required.\n"); 1088 goto done; 1089 } 1090 1091 mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1092 1093 status = ocfs2_replay_journal(osb, node_num, slot_num); 1094 if (status < 0) { 1095 mlog_errno(status); 1096 goto done; 1097 } 1098 1099 /* Stamp a clean local alloc file AFTER recovering the journal... */ 1100 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1101 if (status < 0) { 1102 mlog_errno(status); 1103 goto done; 1104 } 1105 1106 /* An error from begin_truncate_log_recovery is not 1107 * serious enough to warrant halting the rest of 1108 * recovery. */ 1109 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1110 if (status < 0) 1111 mlog_errno(status); 1112 1113 /* Likewise, this would be a strange but ultimately not so 1114 * harmful place to get an error... */ 1115 ocfs2_clear_slot(si, slot_num); 1116 status = ocfs2_update_disk_slots(osb, si); 1117 if (status < 0) 1118 mlog_errno(status); 1119 1120 /* This will kfree the memory pointed to by la_copy and tl_copy */ 1121 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1122 tl_copy); 1123 1124 status = 0; 1125 done: 1126 1127 mlog_exit(status); 1128 return status; 1129 } 1130 1131 /* Test node liveness by trylocking his journal. If we get the lock, 1132 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1133 * still alive (we couldn't get the lock) and < 0 on error. */ 1134 static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1135 int slot_num) 1136 { 1137 int status, flags; 1138 struct inode *inode = NULL; 1139 1140 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1141 slot_num); 1142 if (inode == NULL) { 1143 mlog(ML_ERROR, "access error\n"); 1144 status = -EACCES; 1145 goto bail; 1146 } 1147 if (is_bad_inode(inode)) { 1148 mlog(ML_ERROR, "access error (bad inode)\n"); 1149 iput(inode); 1150 inode = NULL; 1151 status = -EACCES; 1152 goto bail; 1153 } 1154 SET_INODE_JOURNAL(inode); 1155 1156 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1157 status = ocfs2_meta_lock_full(inode, NULL, 1, flags); 1158 if (status < 0) { 1159 if (status != -EAGAIN) 1160 mlog_errno(status); 1161 goto bail; 1162 } 1163 1164 ocfs2_meta_unlock(inode, 1); 1165 bail: 1166 if (inode) 1167 iput(inode); 1168 1169 return status; 1170 } 1171 1172 /* Call this underneath ocfs2_super_lock. It also assumes that the 1173 * slot info struct has been updated from disk. */ 1174 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1175 { 1176 int status, i, node_num; 1177 struct ocfs2_slot_info *si = osb->slot_info; 1178 1179 /* This is called with the super block cluster lock, so we 1180 * know that the slot map can't change underneath us. */ 1181 1182 spin_lock(&si->si_lock); 1183 for(i = 0; i < si->si_num_slots; i++) { 1184 if (i == osb->slot_num) 1185 continue; 1186 if (ocfs2_is_empty_slot(si, i)) 1187 continue; 1188 1189 node_num = si->si_global_node_nums[i]; 1190 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) 1191 continue; 1192 spin_unlock(&si->si_lock); 1193 1194 /* Ok, we have a slot occupied by another node which 1195 * is not in the recovery map. We trylock his journal 1196 * file here to test if he's alive. */ 1197 status = ocfs2_trylock_journal(osb, i); 1198 if (!status) { 1199 /* Since we're called from mount, we know that 1200 * the recovery thread can't race us on 1201 * setting / checking the recovery bits. */ 1202 ocfs2_recovery_thread(osb, node_num); 1203 } else if ((status < 0) && (status != -EAGAIN)) { 1204 mlog_errno(status); 1205 goto bail; 1206 } 1207 1208 spin_lock(&si->si_lock); 1209 } 1210 spin_unlock(&si->si_lock); 1211 1212 status = 0; 1213 bail: 1214 mlog_exit(status); 1215 return status; 1216 } 1217 1218 static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1219 int slot, 1220 struct inode **head) 1221 { 1222 int status; 1223 struct inode *orphan_dir_inode = NULL; 1224 struct inode *iter; 1225 unsigned long offset, blk, local; 1226 struct buffer_head *bh = NULL; 1227 struct ocfs2_dir_entry *de; 1228 struct super_block *sb = osb->sb; 1229 1230 orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1231 ORPHAN_DIR_SYSTEM_INODE, 1232 slot); 1233 if (!orphan_dir_inode) { 1234 status = -ENOENT; 1235 mlog_errno(status); 1236 return status; 1237 } 1238 1239 mutex_lock(&orphan_dir_inode->i_mutex); 1240 status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0); 1241 if (status < 0) { 1242 mlog_errno(status); 1243 goto out; 1244 } 1245 1246 offset = 0; 1247 iter = NULL; 1248 while(offset < i_size_read(orphan_dir_inode)) { 1249 blk = offset >> sb->s_blocksize_bits; 1250 1251 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0); 1252 if (!bh) 1253 status = -EINVAL; 1254 if (status < 0) { 1255 if (bh) 1256 brelse(bh); 1257 mlog_errno(status); 1258 goto out_unlock; 1259 } 1260 1261 local = 0; 1262 while(offset < i_size_read(orphan_dir_inode) 1263 && local < sb->s_blocksize) { 1264 de = (struct ocfs2_dir_entry *) (bh->b_data + local); 1265 1266 if (!ocfs2_check_dir_entry(orphan_dir_inode, 1267 de, bh, local)) { 1268 status = -EINVAL; 1269 mlog_errno(status); 1270 brelse(bh); 1271 goto out_unlock; 1272 } 1273 1274 local += le16_to_cpu(de->rec_len); 1275 offset += le16_to_cpu(de->rec_len); 1276 1277 /* I guess we silently fail on no inode? */ 1278 if (!le64_to_cpu(de->inode)) 1279 continue; 1280 if (de->file_type > OCFS2_FT_MAX) { 1281 mlog(ML_ERROR, 1282 "block %llu contains invalid de: " 1283 "inode = %llu, rec_len = %u, " 1284 "name_len = %u, file_type = %u, " 1285 "name='%.*s'\n", 1286 (unsigned long long)bh->b_blocknr, 1287 (unsigned long long)le64_to_cpu(de->inode), 1288 le16_to_cpu(de->rec_len), 1289 de->name_len, 1290 de->file_type, 1291 de->name_len, 1292 de->name); 1293 continue; 1294 } 1295 if (de->name_len == 1 && !strncmp(".", de->name, 1)) 1296 continue; 1297 if (de->name_len == 2 && !strncmp("..", de->name, 2)) 1298 continue; 1299 1300 iter = ocfs2_iget(osb, le64_to_cpu(de->inode), 1301 OCFS2_FI_FLAG_ORPHAN_RECOVERY); 1302 if (IS_ERR(iter)) 1303 continue; 1304 1305 mlog(0, "queue orphan %llu\n", 1306 (unsigned long long)OCFS2_I(iter)->ip_blkno); 1307 /* No locking is required for the next_orphan 1308 * queue as there is only ever a single 1309 * process doing orphan recovery. */ 1310 OCFS2_I(iter)->ip_next_orphan = *head; 1311 *head = iter; 1312 } 1313 brelse(bh); 1314 } 1315 1316 out_unlock: 1317 ocfs2_meta_unlock(orphan_dir_inode, 0); 1318 out: 1319 mutex_unlock(&orphan_dir_inode->i_mutex); 1320 iput(orphan_dir_inode); 1321 return status; 1322 } 1323 1324 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1325 int slot) 1326 { 1327 int ret; 1328 1329 spin_lock(&osb->osb_lock); 1330 ret = !osb->osb_orphan_wipes[slot]; 1331 spin_unlock(&osb->osb_lock); 1332 return ret; 1333 } 1334 1335 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1336 int slot) 1337 { 1338 spin_lock(&osb->osb_lock); 1339 /* Mark ourselves such that new processes in delete_inode() 1340 * know to quit early. */ 1341 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1342 while (osb->osb_orphan_wipes[slot]) { 1343 /* If any processes are already in the middle of an 1344 * orphan wipe on this dir, then we need to wait for 1345 * them. */ 1346 spin_unlock(&osb->osb_lock); 1347 wait_event_interruptible(osb->osb_wipe_event, 1348 ocfs2_orphan_recovery_can_continue(osb, slot)); 1349 spin_lock(&osb->osb_lock); 1350 } 1351 spin_unlock(&osb->osb_lock); 1352 } 1353 1354 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1355 int slot) 1356 { 1357 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1358 } 1359 1360 /* 1361 * Orphan recovery. Each mounted node has it's own orphan dir which we 1362 * must run during recovery. Our strategy here is to build a list of 1363 * the inodes in the orphan dir and iget/iput them. The VFS does 1364 * (most) of the rest of the work. 1365 * 1366 * Orphan recovery can happen at any time, not just mount so we have a 1367 * couple of extra considerations. 1368 * 1369 * - We grab as many inodes as we can under the orphan dir lock - 1370 * doing iget() outside the orphan dir risks getting a reference on 1371 * an invalid inode. 1372 * - We must be sure not to deadlock with other processes on the 1373 * system wanting to run delete_inode(). This can happen when they go 1374 * to lock the orphan dir and the orphan recovery process attempts to 1375 * iget() inside the orphan dir lock. This can be avoided by 1376 * advertising our state to ocfs2_delete_inode(). 1377 */ 1378 static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1379 int slot) 1380 { 1381 int ret = 0; 1382 struct inode *inode = NULL; 1383 struct inode *iter; 1384 struct ocfs2_inode_info *oi; 1385 1386 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1387 1388 ocfs2_mark_recovering_orphan_dir(osb, slot); 1389 ret = ocfs2_queue_orphans(osb, slot, &inode); 1390 ocfs2_clear_recovering_orphan_dir(osb, slot); 1391 1392 /* Error here should be noted, but we want to continue with as 1393 * many queued inodes as we've got. */ 1394 if (ret) 1395 mlog_errno(ret); 1396 1397 while (inode) { 1398 oi = OCFS2_I(inode); 1399 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1400 1401 iter = oi->ip_next_orphan; 1402 1403 spin_lock(&oi->ip_lock); 1404 /* Delete voting may have set these on the assumption 1405 * that the other node would wipe them successfully. 1406 * If they are still in the node's orphan dir, we need 1407 * to reset that state. */ 1408 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1409 1410 /* Set the proper information to get us going into 1411 * ocfs2_delete_inode. */ 1412 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1413 spin_unlock(&oi->ip_lock); 1414 1415 iput(inode); 1416 1417 inode = iter; 1418 } 1419 1420 return ret; 1421 } 1422 1423 static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1424 { 1425 /* This check is good because ocfs2 will wait on our recovery 1426 * thread before changing it to something other than MOUNTED 1427 * or DISABLED. */ 1428 wait_event(osb->osb_mount_event, 1429 atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1430 atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1431 1432 /* If there's an error on mount, then we may never get to the 1433 * MOUNTED flag, but this is set right before 1434 * dismount_volume() so we can trust it. */ 1435 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1436 mlog(0, "mount error, exiting!\n"); 1437 return -EBUSY; 1438 } 1439 1440 return 0; 1441 } 1442 1443 static int ocfs2_commit_thread(void *arg) 1444 { 1445 int status; 1446 struct ocfs2_super *osb = arg; 1447 struct ocfs2_journal *journal = osb->journal; 1448 1449 /* we can trust j_num_trans here because _should_stop() is only set in 1450 * shutdown and nobody other than ourselves should be able to start 1451 * transactions. committing on shutdown might take a few iterations 1452 * as final transactions put deleted inodes on the list */ 1453 while (!(kthread_should_stop() && 1454 atomic_read(&journal->j_num_trans) == 0)) { 1455 1456 wait_event_interruptible(osb->checkpoint_event, 1457 atomic_read(&journal->j_num_trans) 1458 || kthread_should_stop()); 1459 1460 status = ocfs2_commit_cache(osb); 1461 if (status < 0) 1462 mlog_errno(status); 1463 1464 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1465 mlog(ML_KTHREAD, 1466 "commit_thread: %u transactions pending on " 1467 "shutdown\n", 1468 atomic_read(&journal->j_num_trans)); 1469 } 1470 } 1471 1472 return 0; 1473 } 1474 1475 /* Look for a dirty journal without taking any cluster locks. Used for 1476 * hard readonly access to determine whether the file system journals 1477 * require recovery. */ 1478 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1479 { 1480 int ret = 0; 1481 unsigned int slot; 1482 struct buffer_head *di_bh; 1483 struct ocfs2_dinode *di; 1484 struct inode *journal = NULL; 1485 1486 for(slot = 0; slot < osb->max_slots; slot++) { 1487 journal = ocfs2_get_system_file_inode(osb, 1488 JOURNAL_SYSTEM_INODE, 1489 slot); 1490 if (!journal || is_bad_inode(journal)) { 1491 ret = -EACCES; 1492 mlog_errno(ret); 1493 goto out; 1494 } 1495 1496 di_bh = NULL; 1497 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, 1498 0, journal); 1499 if (ret < 0) { 1500 mlog_errno(ret); 1501 goto out; 1502 } 1503 1504 di = (struct ocfs2_dinode *) di_bh->b_data; 1505 1506 if (le32_to_cpu(di->id1.journal1.ij_flags) & 1507 OCFS2_JOURNAL_DIRTY_FL) 1508 ret = -EROFS; 1509 1510 brelse(di_bh); 1511 if (ret) 1512 break; 1513 } 1514 1515 out: 1516 if (journal) 1517 iput(journal); 1518 1519 return ret; 1520 } 1521