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 void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, 61 struct ocfs2_journal_handle *handle); 62 static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle); 63 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 64 int dirty); 65 static int ocfs2_trylock_journal(struct ocfs2_super *osb, 66 int slot_num); 67 static int ocfs2_recover_orphans(struct ocfs2_super *osb, 68 int slot); 69 static int ocfs2_commit_thread(void *arg); 70 71 static int ocfs2_commit_cache(struct ocfs2_super *osb) 72 { 73 int status = 0; 74 unsigned int flushed; 75 unsigned long old_id; 76 struct ocfs2_journal *journal = NULL; 77 78 mlog_entry_void(); 79 80 journal = osb->journal; 81 82 /* Flush all pending commits and checkpoint the journal. */ 83 down_write(&journal->j_trans_barrier); 84 85 if (atomic_read(&journal->j_num_trans) == 0) { 86 up_write(&journal->j_trans_barrier); 87 mlog(0, "No transactions for me to flush!\n"); 88 goto finally; 89 } 90 91 journal_lock_updates(journal->j_journal); 92 status = journal_flush(journal->j_journal); 93 journal_unlock_updates(journal->j_journal); 94 if (status < 0) { 95 up_write(&journal->j_trans_barrier); 96 mlog_errno(status); 97 goto finally; 98 } 99 100 old_id = ocfs2_inc_trans_id(journal); 101 102 flushed = atomic_read(&journal->j_num_trans); 103 atomic_set(&journal->j_num_trans, 0); 104 up_write(&journal->j_trans_barrier); 105 106 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 107 journal->j_trans_id, flushed); 108 109 ocfs2_kick_vote_thread(osb); 110 wake_up(&journal->j_checkpointed); 111 finally: 112 mlog_exit(status); 113 return status; 114 } 115 116 struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb) 117 { 118 struct ocfs2_journal_handle *retval = NULL; 119 120 retval = kcalloc(1, sizeof(*retval), GFP_NOFS); 121 if (!retval) { 122 mlog(ML_ERROR, "Failed to allocate memory for journal " 123 "handle!\n"); 124 return NULL; 125 } 126 127 retval->num_locks = 0; 128 retval->k_handle = NULL; 129 130 INIT_LIST_HEAD(&retval->locks); 131 INIT_LIST_HEAD(&retval->inode_list); 132 retval->journal = osb->journal; 133 134 return retval; 135 } 136 137 /* pass it NULL and it will allocate a new handle object for you. If 138 * you pass it a handle however, it may still return error, in which 139 * case it has free'd the passed handle for you. */ 140 struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb, 141 struct ocfs2_journal_handle *handle, 142 int max_buffs) 143 { 144 int ret; 145 journal_t *journal = osb->journal->j_journal; 146 147 mlog_entry("(max_buffs = %d)\n", max_buffs); 148 149 BUG_ON(!osb || !osb->journal->j_journal); 150 151 if (ocfs2_is_hard_readonly(osb)) { 152 ret = -EROFS; 153 goto done_free; 154 } 155 156 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 157 BUG_ON(max_buffs <= 0); 158 159 /* JBD might support this, but our journalling code doesn't yet. */ 160 if (journal_current_handle()) { 161 mlog(ML_ERROR, "Recursive transaction attempted!\n"); 162 BUG(); 163 } 164 165 if (!handle) 166 handle = ocfs2_alloc_handle(osb); 167 if (!handle) { 168 ret = -ENOMEM; 169 mlog(ML_ERROR, "Failed to allocate memory for journal " 170 "handle!\n"); 171 goto done_free; 172 } 173 174 down_read(&osb->journal->j_trans_barrier); 175 176 /* actually start the transaction now */ 177 handle->k_handle = journal_start(journal, max_buffs); 178 if (IS_ERR(handle->k_handle)) { 179 up_read(&osb->journal->j_trans_barrier); 180 181 ret = PTR_ERR(handle->k_handle); 182 handle->k_handle = NULL; 183 mlog_errno(ret); 184 185 if (is_journal_aborted(journal)) { 186 ocfs2_abort(osb->sb, "Detected aborted journal"); 187 ret = -EROFS; 188 } 189 goto done_free; 190 } 191 192 atomic_inc(&(osb->journal->j_num_trans)); 193 handle->flags |= OCFS2_HANDLE_STARTED; 194 195 mlog_exit_ptr(handle); 196 return handle; 197 198 done_free: 199 if (handle) 200 ocfs2_commit_unstarted_handle(handle); /* will kfree handle */ 201 202 mlog_exit(ret); 203 return ERR_PTR(ret); 204 } 205 206 void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle, 207 struct inode *inode) 208 { 209 BUG_ON(!handle); 210 BUG_ON(!inode); 211 212 atomic_inc(&inode->i_count); 213 214 /* we're obviously changing it... */ 215 mutex_lock(&inode->i_mutex); 216 217 /* sanity check */ 218 BUG_ON(OCFS2_I(inode)->ip_handle); 219 BUG_ON(!list_empty(&OCFS2_I(inode)->ip_handle_list)); 220 221 OCFS2_I(inode)->ip_handle = handle; 222 list_move_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list)); 223 } 224 225 static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle) 226 { 227 struct list_head *p, *n; 228 struct inode *inode; 229 struct ocfs2_inode_info *oi; 230 231 list_for_each_safe(p, n, &handle->inode_list) { 232 oi = list_entry(p, struct ocfs2_inode_info, 233 ip_handle_list); 234 inode = &oi->vfs_inode; 235 236 OCFS2_I(inode)->ip_handle = NULL; 237 list_del_init(&OCFS2_I(inode)->ip_handle_list); 238 239 mutex_unlock(&inode->i_mutex); 240 iput(inode); 241 } 242 } 243 244 /* This is trivial so we do it out of the main commit 245 * paths. Beware, it can be called from start_trans too! */ 246 static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle) 247 { 248 mlog_entry_void(); 249 250 BUG_ON(handle->flags & OCFS2_HANDLE_STARTED); 251 252 ocfs2_handle_unlock_inodes(handle); 253 /* You are allowed to add journal locks before the transaction 254 * has started. */ 255 ocfs2_handle_cleanup_locks(handle->journal, handle); 256 257 kfree(handle); 258 259 mlog_exit_void(); 260 } 261 262 void ocfs2_commit_trans(struct ocfs2_journal_handle *handle) 263 { 264 handle_t *jbd_handle; 265 int retval; 266 struct ocfs2_journal *journal = handle->journal; 267 268 mlog_entry_void(); 269 270 BUG_ON(!handle); 271 272 if (!(handle->flags & OCFS2_HANDLE_STARTED)) { 273 ocfs2_commit_unstarted_handle(handle); 274 mlog_exit_void(); 275 return; 276 } 277 278 /* release inode semaphores we took during this transaction */ 279 ocfs2_handle_unlock_inodes(handle); 280 281 /* ocfs2_extend_trans may have had to call journal_restart 282 * which will always commit the transaction, but may return 283 * error for any number of reasons. If this is the case, we 284 * clear k_handle as it's not valid any more. */ 285 if (handle->k_handle) { 286 jbd_handle = handle->k_handle; 287 288 if (handle->flags & OCFS2_HANDLE_SYNC) 289 jbd_handle->h_sync = 1; 290 else 291 jbd_handle->h_sync = 0; 292 293 /* actually stop the transaction. if we've set h_sync, 294 * it'll have been committed when we return */ 295 retval = journal_stop(jbd_handle); 296 if (retval < 0) { 297 mlog_errno(retval); 298 mlog(ML_ERROR, "Could not commit transaction\n"); 299 BUG(); 300 } 301 302 handle->k_handle = NULL; /* it's been free'd in journal_stop */ 303 } 304 305 ocfs2_handle_cleanup_locks(journal, handle); 306 307 up_read(&journal->j_trans_barrier); 308 309 kfree(handle); 310 mlog_exit_void(); 311 } 312 313 /* 314 * 'nblocks' is what you want to add to the current 315 * transaction. extend_trans will either extend the current handle by 316 * nblocks, or commit it and start a new one with nblocks credits. 317 * 318 * WARNING: This will not release any semaphores or disk locks taken 319 * during the transaction, so make sure they were taken *before* 320 * start_trans or we'll have ordering deadlocks. 321 * 322 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 323 * good because transaction ids haven't yet been recorded on the 324 * cluster locks associated with this handle. 325 */ 326 int ocfs2_extend_trans(handle_t *handle, int nblocks) 327 { 328 int status; 329 330 BUG_ON(!handle); 331 BUG_ON(!nblocks); 332 333 mlog_entry_void(); 334 335 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 336 337 status = journal_extend(handle, nblocks); 338 if (status < 0) { 339 mlog_errno(status); 340 goto bail; 341 } 342 343 if (status > 0) { 344 mlog(0, "journal_extend failed, trying journal_restart\n"); 345 status = journal_restart(handle, nblocks); 346 if (status < 0) { 347 mlog_errno(status); 348 goto bail; 349 } 350 } 351 352 status = 0; 353 bail: 354 355 mlog_exit(status); 356 return status; 357 } 358 359 int ocfs2_journal_access(struct ocfs2_journal_handle *handle, 360 struct inode *inode, 361 struct buffer_head *bh, 362 int type) 363 { 364 int status; 365 366 BUG_ON(!inode); 367 BUG_ON(!handle); 368 BUG_ON(!bh); 369 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED)); 370 371 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 372 (unsigned long long)bh->b_blocknr, type, 373 (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 374 "OCFS2_JOURNAL_ACCESS_CREATE" : 375 "OCFS2_JOURNAL_ACCESS_WRITE", 376 bh->b_size); 377 378 /* we can safely remove this assertion after testing. */ 379 if (!buffer_uptodate(bh)) { 380 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 381 mlog(ML_ERROR, "b_blocknr=%llu\n", 382 (unsigned long long)bh->b_blocknr); 383 BUG(); 384 } 385 386 /* Set the current transaction information on the inode so 387 * that the locking code knows whether it can drop it's locks 388 * on this inode or not. We're protected from the commit 389 * thread updating the current transaction id until 390 * ocfs2_commit_trans() because ocfs2_start_trans() took 391 * j_trans_barrier for us. */ 392 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); 393 394 mutex_lock(&OCFS2_I(inode)->ip_io_mutex); 395 switch (type) { 396 case OCFS2_JOURNAL_ACCESS_CREATE: 397 case OCFS2_JOURNAL_ACCESS_WRITE: 398 status = journal_get_write_access(handle->k_handle, bh); 399 break; 400 401 case OCFS2_JOURNAL_ACCESS_UNDO: 402 status = journal_get_undo_access(handle->k_handle, bh); 403 break; 404 405 default: 406 status = -EINVAL; 407 mlog(ML_ERROR, "Uknown access type!\n"); 408 } 409 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); 410 411 if (status < 0) 412 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 413 status, type); 414 415 mlog_exit(status); 416 return status; 417 } 418 419 int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle, 420 struct buffer_head *bh) 421 { 422 int status; 423 424 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED)); 425 426 mlog_entry("(bh->b_blocknr=%llu)\n", 427 (unsigned long long)bh->b_blocknr); 428 429 status = journal_dirty_metadata(handle->k_handle, bh); 430 if (status < 0) 431 mlog(ML_ERROR, "Could not dirty metadata buffer. " 432 "(bh->b_blocknr=%llu)\n", 433 (unsigned long long)bh->b_blocknr); 434 435 mlog_exit(status); 436 return status; 437 } 438 439 int ocfs2_journal_dirty_data(handle_t *handle, 440 struct buffer_head *bh) 441 { 442 int err = journal_dirty_data(handle, bh); 443 if (err) 444 mlog_errno(err); 445 /* TODO: When we can handle it, abort the handle and go RO on 446 * error here. */ 447 448 return err; 449 } 450 451 /* We always assume you're adding a metadata lock at level 'ex' */ 452 int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle, 453 struct inode *inode) 454 { 455 int status; 456 struct ocfs2_journal_lock *lock; 457 458 BUG_ON(!inode); 459 460 lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS); 461 if (!lock) { 462 status = -ENOMEM; 463 mlog_errno(-ENOMEM); 464 goto bail; 465 } 466 467 if (!igrab(inode)) 468 BUG(); 469 lock->jl_inode = inode; 470 471 list_add_tail(&(lock->jl_lock_list), &(handle->locks)); 472 handle->num_locks++; 473 474 status = 0; 475 bail: 476 mlog_exit(status); 477 return status; 478 } 479 480 static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, 481 struct ocfs2_journal_handle *handle) 482 { 483 struct list_head *p, *n; 484 struct ocfs2_journal_lock *lock; 485 struct inode *inode; 486 487 list_for_each_safe(p, n, &(handle->locks)) { 488 lock = list_entry(p, struct ocfs2_journal_lock, 489 jl_lock_list); 490 list_del(&lock->jl_lock_list); 491 handle->num_locks--; 492 493 inode = lock->jl_inode; 494 ocfs2_meta_unlock(inode, 1); 495 if (atomic_read(&inode->i_count) == 1) 496 mlog(ML_ERROR, 497 "Inode %llu, I'm doing a last iput for!", 498 (unsigned long long)OCFS2_I(inode)->ip_blkno); 499 iput(inode); 500 kmem_cache_free(ocfs2_lock_cache, lock); 501 } 502 } 503 504 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) 505 506 void ocfs2_set_journal_params(struct ocfs2_super *osb) 507 { 508 journal_t *journal = osb->journal->j_journal; 509 510 spin_lock(&journal->j_state_lock); 511 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 512 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 513 journal->j_flags |= JFS_BARRIER; 514 else 515 journal->j_flags &= ~JFS_BARRIER; 516 spin_unlock(&journal->j_state_lock); 517 } 518 519 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 520 { 521 int status = -1; 522 struct inode *inode = NULL; /* the journal inode */ 523 journal_t *j_journal = NULL; 524 struct ocfs2_dinode *di = NULL; 525 struct buffer_head *bh = NULL; 526 struct ocfs2_super *osb; 527 int meta_lock = 0; 528 529 mlog_entry_void(); 530 531 BUG_ON(!journal); 532 533 osb = journal->j_osb; 534 535 /* already have the inode for our journal */ 536 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 537 osb->slot_num); 538 if (inode == NULL) { 539 status = -EACCES; 540 mlog_errno(status); 541 goto done; 542 } 543 if (is_bad_inode(inode)) { 544 mlog(ML_ERROR, "access error (bad inode)\n"); 545 iput(inode); 546 inode = NULL; 547 status = -EACCES; 548 goto done; 549 } 550 551 SET_INODE_JOURNAL(inode); 552 OCFS2_I(inode)->ip_open_count++; 553 554 /* Skip recovery waits here - journal inode metadata never 555 * changes in a live cluster so it can be considered an 556 * exception to the rule. */ 557 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, 558 OCFS2_META_LOCK_RECOVERY); 559 if (status < 0) { 560 if (status != -ERESTARTSYS) 561 mlog(ML_ERROR, "Could not get lock on journal!\n"); 562 goto done; 563 } 564 565 meta_lock = 1; 566 di = (struct ocfs2_dinode *)bh->b_data; 567 568 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 569 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 570 inode->i_size); 571 status = -EINVAL; 572 goto done; 573 } 574 575 mlog(0, "inode->i_size = %lld\n", inode->i_size); 576 mlog(0, "inode->i_blocks = %llu\n", 577 (unsigned long long)inode->i_blocks); 578 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 579 580 /* call the kernels journal init function now */ 581 j_journal = journal_init_inode(inode); 582 if (j_journal == NULL) { 583 mlog(ML_ERROR, "Linux journal layer error\n"); 584 status = -EINVAL; 585 goto done; 586 } 587 588 mlog(0, "Returned from journal_init_inode\n"); 589 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 590 591 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 592 OCFS2_JOURNAL_DIRTY_FL); 593 594 journal->j_journal = j_journal; 595 journal->j_inode = inode; 596 journal->j_bh = bh; 597 598 ocfs2_set_journal_params(osb); 599 600 journal->j_state = OCFS2_JOURNAL_LOADED; 601 602 status = 0; 603 done: 604 if (status < 0) { 605 if (meta_lock) 606 ocfs2_meta_unlock(inode, 1); 607 if (bh != NULL) 608 brelse(bh); 609 if (inode) { 610 OCFS2_I(inode)->ip_open_count--; 611 iput(inode); 612 } 613 } 614 615 mlog_exit(status); 616 return status; 617 } 618 619 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 620 int dirty) 621 { 622 int status; 623 unsigned int flags; 624 struct ocfs2_journal *journal = osb->journal; 625 struct buffer_head *bh = journal->j_bh; 626 struct ocfs2_dinode *fe; 627 628 mlog_entry_void(); 629 630 fe = (struct ocfs2_dinode *)bh->b_data; 631 if (!OCFS2_IS_VALID_DINODE(fe)) { 632 /* This is called from startup/shutdown which will 633 * handle the errors in a specific manner, so no need 634 * to call ocfs2_error() here. */ 635 mlog(ML_ERROR, "Journal dinode %llu has invalid " 636 "signature: %.*s", (unsigned long long)fe->i_blkno, 7, 637 fe->i_signature); 638 status = -EIO; 639 goto out; 640 } 641 642 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 643 if (dirty) 644 flags |= OCFS2_JOURNAL_DIRTY_FL; 645 else 646 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 647 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 648 649 status = ocfs2_write_block(osb, bh, journal->j_inode); 650 if (status < 0) 651 mlog_errno(status); 652 653 out: 654 mlog_exit(status); 655 return status; 656 } 657 658 /* 659 * If the journal has been kmalloc'd it needs to be freed after this 660 * call. 661 */ 662 void ocfs2_journal_shutdown(struct ocfs2_super *osb) 663 { 664 struct ocfs2_journal *journal = NULL; 665 int status = 0; 666 struct inode *inode = NULL; 667 int num_running_trans = 0; 668 669 mlog_entry_void(); 670 671 BUG_ON(!osb); 672 673 journal = osb->journal; 674 if (!journal) 675 goto done; 676 677 inode = journal->j_inode; 678 679 if (journal->j_state != OCFS2_JOURNAL_LOADED) 680 goto done; 681 682 /* need to inc inode use count as journal_destroy will iput. */ 683 if (!igrab(inode)) 684 BUG(); 685 686 num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 687 if (num_running_trans > 0) 688 mlog(0, "Shutting down journal: must wait on %d " 689 "running transactions!\n", 690 num_running_trans); 691 692 /* Do a commit_cache here. It will flush our journal, *and* 693 * release any locks that are still held. 694 * set the SHUTDOWN flag and release the trans lock. 695 * the commit thread will take the trans lock for us below. */ 696 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 697 698 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 699 * drop the trans_lock (which we want to hold until we 700 * completely destroy the journal. */ 701 if (osb->commit_task) { 702 /* Wait for the commit thread */ 703 mlog(0, "Waiting for ocfs2commit to exit....\n"); 704 kthread_stop(osb->commit_task); 705 osb->commit_task = NULL; 706 } 707 708 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 709 710 status = ocfs2_journal_toggle_dirty(osb, 0); 711 if (status < 0) 712 mlog_errno(status); 713 714 /* Shutdown the kernel journal system */ 715 journal_destroy(journal->j_journal); 716 717 OCFS2_I(inode)->ip_open_count--; 718 719 /* unlock our journal */ 720 ocfs2_meta_unlock(inode, 1); 721 722 brelse(journal->j_bh); 723 journal->j_bh = NULL; 724 725 journal->j_state = OCFS2_JOURNAL_FREE; 726 727 // up_write(&journal->j_trans_barrier); 728 done: 729 if (inode) 730 iput(inode); 731 mlog_exit_void(); 732 } 733 734 static void ocfs2_clear_journal_error(struct super_block *sb, 735 journal_t *journal, 736 int slot) 737 { 738 int olderr; 739 740 olderr = journal_errno(journal); 741 if (olderr) { 742 mlog(ML_ERROR, "File system error %d recorded in " 743 "journal %u.\n", olderr, slot); 744 mlog(ML_ERROR, "File system on device %s needs checking.\n", 745 sb->s_id); 746 747 journal_ack_err(journal); 748 journal_clear_err(journal); 749 } 750 } 751 752 int ocfs2_journal_load(struct ocfs2_journal *journal) 753 { 754 int status = 0; 755 struct ocfs2_super *osb; 756 757 mlog_entry_void(); 758 759 if (!journal) 760 BUG(); 761 762 osb = journal->j_osb; 763 764 status = journal_load(journal->j_journal); 765 if (status < 0) { 766 mlog(ML_ERROR, "Failed to load journal!\n"); 767 goto done; 768 } 769 770 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 771 772 status = ocfs2_journal_toggle_dirty(osb, 1); 773 if (status < 0) { 774 mlog_errno(status); 775 goto done; 776 } 777 778 /* Launch the commit thread */ 779 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt"); 780 if (IS_ERR(osb->commit_task)) { 781 status = PTR_ERR(osb->commit_task); 782 osb->commit_task = NULL; 783 mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d", 784 status); 785 goto done; 786 } 787 788 done: 789 mlog_exit(status); 790 return status; 791 } 792 793 794 /* 'full' flag tells us whether we clear out all blocks or if we just 795 * mark the journal clean */ 796 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 797 { 798 int status; 799 800 mlog_entry_void(); 801 802 BUG_ON(!journal); 803 804 status = journal_wipe(journal->j_journal, full); 805 if (status < 0) { 806 mlog_errno(status); 807 goto bail; 808 } 809 810 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); 811 if (status < 0) 812 mlog_errno(status); 813 814 bail: 815 mlog_exit(status); 816 return status; 817 } 818 819 /* 820 * JBD Might read a cached version of another nodes journal file. We 821 * don't want this as this file changes often and we get no 822 * notification on those changes. The only way to be sure that we've 823 * got the most up to date version of those blocks then is to force 824 * read them off disk. Just searching through the buffer cache won't 825 * work as there may be pages backing this file which are still marked 826 * up to date. We know things can't change on this file underneath us 827 * as we have the lock by now :) 828 */ 829 static int ocfs2_force_read_journal(struct inode *inode) 830 { 831 int status = 0; 832 int i, p_blocks; 833 u64 v_blkno, p_blkno; 834 #define CONCURRENT_JOURNAL_FILL 32 835 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 836 837 mlog_entry_void(); 838 839 BUG_ON(inode->i_blocks != 840 ocfs2_align_bytes_to_sectors(i_size_read(inode))); 841 842 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 843 844 mlog(0, "Force reading %llu blocks\n", 845 (unsigned long long)(inode->i_blocks >> 846 (inode->i_sb->s_blocksize_bits - 9))); 847 848 v_blkno = 0; 849 while (v_blkno < 850 (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) { 851 852 status = ocfs2_extent_map_get_blocks(inode, v_blkno, 853 1, &p_blkno, 854 &p_blocks); 855 if (status < 0) { 856 mlog_errno(status); 857 goto bail; 858 } 859 860 if (p_blocks > CONCURRENT_JOURNAL_FILL) 861 p_blocks = CONCURRENT_JOURNAL_FILL; 862 863 /* We are reading journal data which should not 864 * be put in the uptodate cache */ 865 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), 866 p_blkno, p_blocks, bhs, 0, 867 NULL); 868 if (status < 0) { 869 mlog_errno(status); 870 goto bail; 871 } 872 873 for(i = 0; i < p_blocks; i++) { 874 brelse(bhs[i]); 875 bhs[i] = NULL; 876 } 877 878 v_blkno += p_blocks; 879 } 880 881 bail: 882 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 883 if (bhs[i]) 884 brelse(bhs[i]); 885 mlog_exit(status); 886 return status; 887 } 888 889 struct ocfs2_la_recovery_item { 890 struct list_head lri_list; 891 int lri_slot; 892 struct ocfs2_dinode *lri_la_dinode; 893 struct ocfs2_dinode *lri_tl_dinode; 894 }; 895 896 /* Does the second half of the recovery process. By this point, the 897 * node is marked clean and can actually be considered recovered, 898 * hence it's no longer in the recovery map, but there's still some 899 * cleanup we can do which shouldn't happen within the recovery thread 900 * as locking in that context becomes very difficult if we are to take 901 * recovering nodes into account. 902 * 903 * NOTE: This function can and will sleep on recovery of other nodes 904 * during cluster locking, just like any other ocfs2 process. 905 */ 906 void ocfs2_complete_recovery(void *data) 907 { 908 int ret; 909 struct ocfs2_super *osb = data; 910 struct ocfs2_journal *journal = osb->journal; 911 struct ocfs2_dinode *la_dinode, *tl_dinode; 912 struct ocfs2_la_recovery_item *item; 913 struct list_head *p, *n; 914 LIST_HEAD(tmp_la_list); 915 916 mlog_entry_void(); 917 918 mlog(0, "completing recovery from keventd\n"); 919 920 spin_lock(&journal->j_lock); 921 list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 922 spin_unlock(&journal->j_lock); 923 924 list_for_each_safe(p, n, &tmp_la_list) { 925 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list); 926 list_del_init(&item->lri_list); 927 928 mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 929 930 la_dinode = item->lri_la_dinode; 931 if (la_dinode) { 932 mlog(0, "Clean up local alloc %llu\n", 933 (unsigned long long)la_dinode->i_blkno); 934 935 ret = ocfs2_complete_local_alloc_recovery(osb, 936 la_dinode); 937 if (ret < 0) 938 mlog_errno(ret); 939 940 kfree(la_dinode); 941 } 942 943 tl_dinode = item->lri_tl_dinode; 944 if (tl_dinode) { 945 mlog(0, "Clean up truncate log %llu\n", 946 (unsigned long long)tl_dinode->i_blkno); 947 948 ret = ocfs2_complete_truncate_log_recovery(osb, 949 tl_dinode); 950 if (ret < 0) 951 mlog_errno(ret); 952 953 kfree(tl_dinode); 954 } 955 956 ret = ocfs2_recover_orphans(osb, item->lri_slot); 957 if (ret < 0) 958 mlog_errno(ret); 959 960 kfree(item); 961 } 962 963 mlog(0, "Recovery completion\n"); 964 mlog_exit_void(); 965 } 966 967 /* NOTE: This function always eats your references to la_dinode and 968 * tl_dinode, either manually on error, or by passing them to 969 * ocfs2_complete_recovery */ 970 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 971 int slot_num, 972 struct ocfs2_dinode *la_dinode, 973 struct ocfs2_dinode *tl_dinode) 974 { 975 struct ocfs2_la_recovery_item *item; 976 977 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 978 if (!item) { 979 /* Though we wish to avoid it, we are in fact safe in 980 * skipping local alloc cleanup as fsck.ocfs2 is more 981 * than capable of reclaiming unused space. */ 982 if (la_dinode) 983 kfree(la_dinode); 984 985 if (tl_dinode) 986 kfree(tl_dinode); 987 988 mlog_errno(-ENOMEM); 989 return; 990 } 991 992 INIT_LIST_HEAD(&item->lri_list); 993 item->lri_la_dinode = la_dinode; 994 item->lri_slot = slot_num; 995 item->lri_tl_dinode = tl_dinode; 996 997 spin_lock(&journal->j_lock); 998 list_add_tail(&item->lri_list, &journal->j_la_cleanups); 999 queue_work(ocfs2_wq, &journal->j_recovery_work); 1000 spin_unlock(&journal->j_lock); 1001 } 1002 1003 /* Called by the mount code to queue recovery the last part of 1004 * recovery for it's own slot. */ 1005 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1006 { 1007 struct ocfs2_journal *journal = osb->journal; 1008 1009 if (osb->dirty) { 1010 /* No need to queue up our truncate_log as regular 1011 * cleanup will catch that. */ 1012 ocfs2_queue_recovery_completion(journal, 1013 osb->slot_num, 1014 osb->local_alloc_copy, 1015 NULL); 1016 ocfs2_schedule_truncate_log_flush(osb, 0); 1017 1018 osb->local_alloc_copy = NULL; 1019 osb->dirty = 0; 1020 } 1021 } 1022 1023 static int __ocfs2_recovery_thread(void *arg) 1024 { 1025 int status, node_num; 1026 struct ocfs2_super *osb = arg; 1027 1028 mlog_entry_void(); 1029 1030 status = ocfs2_wait_on_mount(osb); 1031 if (status < 0) { 1032 goto bail; 1033 } 1034 1035 restart: 1036 status = ocfs2_super_lock(osb, 1); 1037 if (status < 0) { 1038 mlog_errno(status); 1039 goto bail; 1040 } 1041 1042 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 1043 node_num = ocfs2_node_map_first_set_bit(osb, 1044 &osb->recovery_map); 1045 if (node_num == O2NM_INVALID_NODE_NUM) { 1046 mlog(0, "Out of nodes to recover.\n"); 1047 break; 1048 } 1049 1050 status = ocfs2_recover_node(osb, node_num); 1051 if (status < 0) { 1052 mlog(ML_ERROR, 1053 "Error %d recovering node %d on device (%u,%u)!\n", 1054 status, node_num, 1055 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1056 mlog(ML_ERROR, "Volume requires unmount.\n"); 1057 continue; 1058 } 1059 1060 ocfs2_recovery_map_clear(osb, node_num); 1061 } 1062 ocfs2_super_unlock(osb, 1); 1063 1064 /* We always run recovery on our own orphan dir - the dead 1065 * node(s) may have voted "no" on an inode delete earlier. A 1066 * revote is therefore required. */ 1067 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1068 NULL); 1069 1070 bail: 1071 mutex_lock(&osb->recovery_lock); 1072 if (!status && 1073 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { 1074 mutex_unlock(&osb->recovery_lock); 1075 goto restart; 1076 } 1077 1078 osb->recovery_thread_task = NULL; 1079 mb(); /* sync with ocfs2_recovery_thread_running */ 1080 wake_up(&osb->recovery_event); 1081 1082 mutex_unlock(&osb->recovery_lock); 1083 1084 mlog_exit(status); 1085 /* no one is callint kthread_stop() for us so the kthread() api 1086 * requires that we call do_exit(). And it isn't exported, but 1087 * complete_and_exit() seems to be a minimal wrapper around it. */ 1088 complete_and_exit(NULL, status); 1089 return status; 1090 } 1091 1092 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1093 { 1094 mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1095 node_num, osb->node_num); 1096 1097 mutex_lock(&osb->recovery_lock); 1098 if (osb->disable_recovery) 1099 goto out; 1100 1101 /* People waiting on recovery will wait on 1102 * the recovery map to empty. */ 1103 if (!ocfs2_recovery_map_set(osb, node_num)) 1104 mlog(0, "node %d already be in recovery.\n", node_num); 1105 1106 mlog(0, "starting recovery thread...\n"); 1107 1108 if (osb->recovery_thread_task) 1109 goto out; 1110 1111 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 1112 "ocfs2rec"); 1113 if (IS_ERR(osb->recovery_thread_task)) { 1114 mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1115 osb->recovery_thread_task = NULL; 1116 } 1117 1118 out: 1119 mutex_unlock(&osb->recovery_lock); 1120 wake_up(&osb->recovery_event); 1121 1122 mlog_exit_void(); 1123 } 1124 1125 /* Does the actual journal replay and marks the journal inode as 1126 * clean. Will only replay if the journal inode is marked dirty. */ 1127 static int ocfs2_replay_journal(struct ocfs2_super *osb, 1128 int node_num, 1129 int slot_num) 1130 { 1131 int status; 1132 int got_lock = 0; 1133 unsigned int flags; 1134 struct inode *inode = NULL; 1135 struct ocfs2_dinode *fe; 1136 journal_t *journal = NULL; 1137 struct buffer_head *bh = NULL; 1138 1139 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1140 slot_num); 1141 if (inode == NULL) { 1142 status = -EACCES; 1143 mlog_errno(status); 1144 goto done; 1145 } 1146 if (is_bad_inode(inode)) { 1147 status = -EACCES; 1148 iput(inode); 1149 inode = NULL; 1150 mlog_errno(status); 1151 goto done; 1152 } 1153 SET_INODE_JOURNAL(inode); 1154 1155 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, 1156 OCFS2_META_LOCK_RECOVERY); 1157 if (status < 0) { 1158 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); 1159 if (status != -ERESTARTSYS) 1160 mlog(ML_ERROR, "Could not lock journal!\n"); 1161 goto done; 1162 } 1163 got_lock = 1; 1164 1165 fe = (struct ocfs2_dinode *) bh->b_data; 1166 1167 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1168 1169 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1170 mlog(0, "No recovery required for node %d\n", node_num); 1171 goto done; 1172 } 1173 1174 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 1175 node_num, slot_num, 1176 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1177 1178 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1179 1180 status = ocfs2_force_read_journal(inode); 1181 if (status < 0) { 1182 mlog_errno(status); 1183 goto done; 1184 } 1185 1186 mlog(0, "calling journal_init_inode\n"); 1187 journal = journal_init_inode(inode); 1188 if (journal == NULL) { 1189 mlog(ML_ERROR, "Linux journal layer error\n"); 1190 status = -EIO; 1191 goto done; 1192 } 1193 1194 status = journal_load(journal); 1195 if (status < 0) { 1196 mlog_errno(status); 1197 if (!igrab(inode)) 1198 BUG(); 1199 journal_destroy(journal); 1200 goto done; 1201 } 1202 1203 ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1204 1205 /* wipe the journal */ 1206 mlog(0, "flushing the journal.\n"); 1207 journal_lock_updates(journal); 1208 status = journal_flush(journal); 1209 journal_unlock_updates(journal); 1210 if (status < 0) 1211 mlog_errno(status); 1212 1213 /* This will mark the node clean */ 1214 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1215 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1216 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1217 1218 status = ocfs2_write_block(osb, bh, inode); 1219 if (status < 0) 1220 mlog_errno(status); 1221 1222 if (!igrab(inode)) 1223 BUG(); 1224 1225 journal_destroy(journal); 1226 1227 done: 1228 /* drop the lock on this nodes journal */ 1229 if (got_lock) 1230 ocfs2_meta_unlock(inode, 1); 1231 1232 if (inode) 1233 iput(inode); 1234 1235 if (bh) 1236 brelse(bh); 1237 1238 mlog_exit(status); 1239 return status; 1240 } 1241 1242 /* 1243 * Do the most important parts of node recovery: 1244 * - Replay it's journal 1245 * - Stamp a clean local allocator file 1246 * - Stamp a clean truncate log 1247 * - Mark the node clean 1248 * 1249 * If this function completes without error, a node in OCFS2 can be 1250 * said to have been safely recovered. As a result, failure during the 1251 * second part of a nodes recovery process (local alloc recovery) is 1252 * far less concerning. 1253 */ 1254 static int ocfs2_recover_node(struct ocfs2_super *osb, 1255 int node_num) 1256 { 1257 int status = 0; 1258 int slot_num; 1259 struct ocfs2_slot_info *si = osb->slot_info; 1260 struct ocfs2_dinode *la_copy = NULL; 1261 struct ocfs2_dinode *tl_copy = NULL; 1262 1263 mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1264 node_num, osb->node_num); 1265 1266 mlog(0, "checking node %d\n", node_num); 1267 1268 /* Should not ever be called to recover ourselves -- in that 1269 * case we should've called ocfs2_journal_load instead. */ 1270 BUG_ON(osb->node_num == node_num); 1271 1272 slot_num = ocfs2_node_num_to_slot(si, node_num); 1273 if (slot_num == OCFS2_INVALID_SLOT) { 1274 status = 0; 1275 mlog(0, "no slot for this node, so no recovery required.\n"); 1276 goto done; 1277 } 1278 1279 mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1280 1281 status = ocfs2_replay_journal(osb, node_num, slot_num); 1282 if (status < 0) { 1283 mlog_errno(status); 1284 goto done; 1285 } 1286 1287 /* Stamp a clean local alloc file AFTER recovering the journal... */ 1288 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1289 if (status < 0) { 1290 mlog_errno(status); 1291 goto done; 1292 } 1293 1294 /* An error from begin_truncate_log_recovery is not 1295 * serious enough to warrant halting the rest of 1296 * recovery. */ 1297 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1298 if (status < 0) 1299 mlog_errno(status); 1300 1301 /* Likewise, this would be a strange but ultimately not so 1302 * harmful place to get an error... */ 1303 ocfs2_clear_slot(si, slot_num); 1304 status = ocfs2_update_disk_slots(osb, si); 1305 if (status < 0) 1306 mlog_errno(status); 1307 1308 /* This will kfree the memory pointed to by la_copy and tl_copy */ 1309 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1310 tl_copy); 1311 1312 status = 0; 1313 done: 1314 1315 mlog_exit(status); 1316 return status; 1317 } 1318 1319 /* Test node liveness by trylocking his journal. If we get the lock, 1320 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1321 * still alive (we couldn't get the lock) and < 0 on error. */ 1322 static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1323 int slot_num) 1324 { 1325 int status, flags; 1326 struct inode *inode = NULL; 1327 1328 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1329 slot_num); 1330 if (inode == NULL) { 1331 mlog(ML_ERROR, "access error\n"); 1332 status = -EACCES; 1333 goto bail; 1334 } 1335 if (is_bad_inode(inode)) { 1336 mlog(ML_ERROR, "access error (bad inode)\n"); 1337 iput(inode); 1338 inode = NULL; 1339 status = -EACCES; 1340 goto bail; 1341 } 1342 SET_INODE_JOURNAL(inode); 1343 1344 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1345 status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags); 1346 if (status < 0) { 1347 if (status != -EAGAIN) 1348 mlog_errno(status); 1349 goto bail; 1350 } 1351 1352 ocfs2_meta_unlock(inode, 1); 1353 bail: 1354 if (inode) 1355 iput(inode); 1356 1357 return status; 1358 } 1359 1360 /* Call this underneath ocfs2_super_lock. It also assumes that the 1361 * slot info struct has been updated from disk. */ 1362 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1363 { 1364 int status, i, node_num; 1365 struct ocfs2_slot_info *si = osb->slot_info; 1366 1367 /* This is called with the super block cluster lock, so we 1368 * know that the slot map can't change underneath us. */ 1369 1370 spin_lock(&si->si_lock); 1371 for(i = 0; i < si->si_num_slots; i++) { 1372 if (i == osb->slot_num) 1373 continue; 1374 if (ocfs2_is_empty_slot(si, i)) 1375 continue; 1376 1377 node_num = si->si_global_node_nums[i]; 1378 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) 1379 continue; 1380 spin_unlock(&si->si_lock); 1381 1382 /* Ok, we have a slot occupied by another node which 1383 * is not in the recovery map. We trylock his journal 1384 * file here to test if he's alive. */ 1385 status = ocfs2_trylock_journal(osb, i); 1386 if (!status) { 1387 /* Since we're called from mount, we know that 1388 * the recovery thread can't race us on 1389 * setting / checking the recovery bits. */ 1390 ocfs2_recovery_thread(osb, node_num); 1391 } else if ((status < 0) && (status != -EAGAIN)) { 1392 mlog_errno(status); 1393 goto bail; 1394 } 1395 1396 spin_lock(&si->si_lock); 1397 } 1398 spin_unlock(&si->si_lock); 1399 1400 status = 0; 1401 bail: 1402 mlog_exit(status); 1403 return status; 1404 } 1405 1406 static int ocfs2_queue_orphans(struct ocfs2_super *osb, 1407 int slot, 1408 struct inode **head) 1409 { 1410 int status; 1411 struct inode *orphan_dir_inode = NULL; 1412 struct inode *iter; 1413 unsigned long offset, blk, local; 1414 struct buffer_head *bh = NULL; 1415 struct ocfs2_dir_entry *de; 1416 struct super_block *sb = osb->sb; 1417 1418 orphan_dir_inode = ocfs2_get_system_file_inode(osb, 1419 ORPHAN_DIR_SYSTEM_INODE, 1420 slot); 1421 if (!orphan_dir_inode) { 1422 status = -ENOENT; 1423 mlog_errno(status); 1424 return status; 1425 } 1426 1427 mutex_lock(&orphan_dir_inode->i_mutex); 1428 status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0); 1429 if (status < 0) { 1430 mlog_errno(status); 1431 goto out; 1432 } 1433 1434 offset = 0; 1435 iter = NULL; 1436 while(offset < i_size_read(orphan_dir_inode)) { 1437 blk = offset >> sb->s_blocksize_bits; 1438 1439 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0); 1440 if (!bh) 1441 status = -EINVAL; 1442 if (status < 0) { 1443 if (bh) 1444 brelse(bh); 1445 mlog_errno(status); 1446 goto out_unlock; 1447 } 1448 1449 local = 0; 1450 while(offset < i_size_read(orphan_dir_inode) 1451 && local < sb->s_blocksize) { 1452 de = (struct ocfs2_dir_entry *) (bh->b_data + local); 1453 1454 if (!ocfs2_check_dir_entry(orphan_dir_inode, 1455 de, bh, local)) { 1456 status = -EINVAL; 1457 mlog_errno(status); 1458 brelse(bh); 1459 goto out_unlock; 1460 } 1461 1462 local += le16_to_cpu(de->rec_len); 1463 offset += le16_to_cpu(de->rec_len); 1464 1465 /* I guess we silently fail on no inode? */ 1466 if (!le64_to_cpu(de->inode)) 1467 continue; 1468 if (de->file_type > OCFS2_FT_MAX) { 1469 mlog(ML_ERROR, 1470 "block %llu contains invalid de: " 1471 "inode = %llu, rec_len = %u, " 1472 "name_len = %u, file_type = %u, " 1473 "name='%.*s'\n", 1474 (unsigned long long)bh->b_blocknr, 1475 (unsigned long long)le64_to_cpu(de->inode), 1476 le16_to_cpu(de->rec_len), 1477 de->name_len, 1478 de->file_type, 1479 de->name_len, 1480 de->name); 1481 continue; 1482 } 1483 if (de->name_len == 1 && !strncmp(".", de->name, 1)) 1484 continue; 1485 if (de->name_len == 2 && !strncmp("..", de->name, 2)) 1486 continue; 1487 1488 iter = ocfs2_iget(osb, le64_to_cpu(de->inode), 1489 OCFS2_FI_FLAG_NOLOCK); 1490 if (IS_ERR(iter)) 1491 continue; 1492 1493 mlog(0, "queue orphan %llu\n", 1494 (unsigned long long)OCFS2_I(iter)->ip_blkno); 1495 /* No locking is required for the next_orphan 1496 * queue as there is only ever a single 1497 * process doing orphan recovery. */ 1498 OCFS2_I(iter)->ip_next_orphan = *head; 1499 *head = iter; 1500 } 1501 brelse(bh); 1502 } 1503 1504 out_unlock: 1505 ocfs2_meta_unlock(orphan_dir_inode, 0); 1506 out: 1507 mutex_unlock(&orphan_dir_inode->i_mutex); 1508 iput(orphan_dir_inode); 1509 return status; 1510 } 1511 1512 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 1513 int slot) 1514 { 1515 int ret; 1516 1517 spin_lock(&osb->osb_lock); 1518 ret = !osb->osb_orphan_wipes[slot]; 1519 spin_unlock(&osb->osb_lock); 1520 return ret; 1521 } 1522 1523 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 1524 int slot) 1525 { 1526 spin_lock(&osb->osb_lock); 1527 /* Mark ourselves such that new processes in delete_inode() 1528 * know to quit early. */ 1529 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1530 while (osb->osb_orphan_wipes[slot]) { 1531 /* If any processes are already in the middle of an 1532 * orphan wipe on this dir, then we need to wait for 1533 * them. */ 1534 spin_unlock(&osb->osb_lock); 1535 wait_event_interruptible(osb->osb_wipe_event, 1536 ocfs2_orphan_recovery_can_continue(osb, slot)); 1537 spin_lock(&osb->osb_lock); 1538 } 1539 spin_unlock(&osb->osb_lock); 1540 } 1541 1542 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 1543 int slot) 1544 { 1545 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 1546 } 1547 1548 /* 1549 * Orphan recovery. Each mounted node has it's own orphan dir which we 1550 * must run during recovery. Our strategy here is to build a list of 1551 * the inodes in the orphan dir and iget/iput them. The VFS does 1552 * (most) of the rest of the work. 1553 * 1554 * Orphan recovery can happen at any time, not just mount so we have a 1555 * couple of extra considerations. 1556 * 1557 * - We grab as many inodes as we can under the orphan dir lock - 1558 * doing iget() outside the orphan dir risks getting a reference on 1559 * an invalid inode. 1560 * - We must be sure not to deadlock with other processes on the 1561 * system wanting to run delete_inode(). This can happen when they go 1562 * to lock the orphan dir and the orphan recovery process attempts to 1563 * iget() inside the orphan dir lock. This can be avoided by 1564 * advertising our state to ocfs2_delete_inode(). 1565 */ 1566 static int ocfs2_recover_orphans(struct ocfs2_super *osb, 1567 int slot) 1568 { 1569 int ret = 0; 1570 struct inode *inode = NULL; 1571 struct inode *iter; 1572 struct ocfs2_inode_info *oi; 1573 1574 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 1575 1576 ocfs2_mark_recovering_orphan_dir(osb, slot); 1577 ret = ocfs2_queue_orphans(osb, slot, &inode); 1578 ocfs2_clear_recovering_orphan_dir(osb, slot); 1579 1580 /* Error here should be noted, but we want to continue with as 1581 * many queued inodes as we've got. */ 1582 if (ret) 1583 mlog_errno(ret); 1584 1585 while (inode) { 1586 oi = OCFS2_I(inode); 1587 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 1588 1589 iter = oi->ip_next_orphan; 1590 1591 spin_lock(&oi->ip_lock); 1592 /* Delete voting may have set these on the assumption 1593 * that the other node would wipe them successfully. 1594 * If they are still in the node's orphan dir, we need 1595 * to reset that state. */ 1596 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 1597 1598 /* Set the proper information to get us going into 1599 * ocfs2_delete_inode. */ 1600 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 1601 oi->ip_orphaned_slot = slot; 1602 spin_unlock(&oi->ip_lock); 1603 1604 iput(inode); 1605 1606 inode = iter; 1607 } 1608 1609 return ret; 1610 } 1611 1612 static int ocfs2_wait_on_mount(struct ocfs2_super *osb) 1613 { 1614 /* This check is good because ocfs2 will wait on our recovery 1615 * thread before changing it to something other than MOUNTED 1616 * or DISABLED. */ 1617 wait_event(osb->osb_mount_event, 1618 atomic_read(&osb->vol_state) == VOLUME_MOUNTED || 1619 atomic_read(&osb->vol_state) == VOLUME_DISABLED); 1620 1621 /* If there's an error on mount, then we may never get to the 1622 * MOUNTED flag, but this is set right before 1623 * dismount_volume() so we can trust it. */ 1624 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 1625 mlog(0, "mount error, exiting!\n"); 1626 return -EBUSY; 1627 } 1628 1629 return 0; 1630 } 1631 1632 static int ocfs2_commit_thread(void *arg) 1633 { 1634 int status; 1635 struct ocfs2_super *osb = arg; 1636 struct ocfs2_journal *journal = osb->journal; 1637 1638 /* we can trust j_num_trans here because _should_stop() is only set in 1639 * shutdown and nobody other than ourselves should be able to start 1640 * transactions. committing on shutdown might take a few iterations 1641 * as final transactions put deleted inodes on the list */ 1642 while (!(kthread_should_stop() && 1643 atomic_read(&journal->j_num_trans) == 0)) { 1644 1645 wait_event_interruptible(osb->checkpoint_event, 1646 atomic_read(&journal->j_num_trans) 1647 || kthread_should_stop()); 1648 1649 status = ocfs2_commit_cache(osb); 1650 if (status < 0) 1651 mlog_errno(status); 1652 1653 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 1654 mlog(ML_KTHREAD, 1655 "commit_thread: %u transactions pending on " 1656 "shutdown\n", 1657 atomic_read(&journal->j_num_trans)); 1658 } 1659 } 1660 1661 return 0; 1662 } 1663 1664 /* Look for a dirty journal without taking any cluster locks. Used for 1665 * hard readonly access to determine whether the file system journals 1666 * require recovery. */ 1667 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 1668 { 1669 int ret = 0; 1670 unsigned int slot; 1671 struct buffer_head *di_bh; 1672 struct ocfs2_dinode *di; 1673 struct inode *journal = NULL; 1674 1675 for(slot = 0; slot < osb->max_slots; slot++) { 1676 journal = ocfs2_get_system_file_inode(osb, 1677 JOURNAL_SYSTEM_INODE, 1678 slot); 1679 if (!journal || is_bad_inode(journal)) { 1680 ret = -EACCES; 1681 mlog_errno(ret); 1682 goto out; 1683 } 1684 1685 di_bh = NULL; 1686 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, 1687 0, journal); 1688 if (ret < 0) { 1689 mlog_errno(ret); 1690 goto out; 1691 } 1692 1693 di = (struct ocfs2_dinode *) di_bh->b_data; 1694 1695 if (le32_to_cpu(di->id1.journal1.ij_flags) & 1696 OCFS2_JOURNAL_DIRTY_FL) 1697 ret = -EROFS; 1698 1699 brelse(di_bh); 1700 if (ret) 1701 break; 1702 } 1703 1704 out: 1705 if (journal) 1706 iput(journal); 1707 1708 return ret; 1709 } 1710