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#include <linux/time.h> 32#include <linux/random.h> 33 34#define MLOG_MASK_PREFIX ML_JOURNAL 35#include <cluster/masklog.h> 36 37#include "ocfs2.h" 38 39#include "alloc.h" 40#include "blockcheck.h" 41#include "dir.h" 42#include "dlmglue.h" 43#include "extent_map.h" 44#include "heartbeat.h" 45#include "inode.h" 46#include "journal.h" 47#include "localalloc.h" 48#include "slot_map.h" 49#include "super.h" 50#include "sysfile.h" 51#include "uptodate.h" 52#include "quota.h" 53 54#include "buffer_head_io.h" 55 56DEFINE_SPINLOCK(trans_inc_lock); 57 58#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 59 60static int ocfs2_force_read_journal(struct inode *inode); 61static int ocfs2_recover_node(struct ocfs2_super *osb, 62 int node_num, int slot_num); 63static int __ocfs2_recovery_thread(void *arg); 64static int ocfs2_commit_cache(struct ocfs2_super *osb); 65static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 66static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 67 int dirty, int replayed); 68static int ocfs2_trylock_journal(struct ocfs2_super *osb, 69 int slot_num); 70static int ocfs2_recover_orphans(struct ocfs2_super *osb, 71 int slot); 72static int ocfs2_commit_thread(void *arg); 73static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 74 int slot_num, 75 struct ocfs2_dinode *la_dinode, 76 struct ocfs2_dinode *tl_dinode, 77 struct ocfs2_quota_recovery *qrec); 78 79static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 80{ 81 return __ocfs2_wait_on_mount(osb, 0); 82} 83 84static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 85{ 86 return __ocfs2_wait_on_mount(osb, 1); 87} 88 89/* 90 * This replay_map is to track online/offline slots, so we could recover 91 * offline slots during recovery and mount 92 */ 93 94enum ocfs2_replay_state { 95 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 96 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 97 REPLAY_DONE /* Replay was already queued */ 98}; 99 100struct ocfs2_replay_map { 101 unsigned int rm_slots; 102 enum ocfs2_replay_state rm_state; 103 unsigned char rm_replay_slots[0]; 104}; 105 106void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 107{ 108 if (!osb->replay_map) 109 return; 110 111 /* If we've already queued the replay, we don't have any more to do */ 112 if (osb->replay_map->rm_state == REPLAY_DONE) 113 return; 114 115 osb->replay_map->rm_state = state; 116} 117 118int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 119{ 120 struct ocfs2_replay_map *replay_map; 121 int i, node_num; 122 123 /* If replay map is already set, we don't do it again */ 124 if (osb->replay_map) 125 return 0; 126 127 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 128 (osb->max_slots * sizeof(char)), GFP_KERNEL); 129 130 if (!replay_map) { 131 mlog_errno(-ENOMEM); 132 return -ENOMEM; 133 } 134 135 spin_lock(&osb->osb_lock); 136 137 replay_map->rm_slots = osb->max_slots; 138 replay_map->rm_state = REPLAY_UNNEEDED; 139 140 /* set rm_replay_slots for offline slot(s) */ 141 for (i = 0; i < replay_map->rm_slots; i++) { 142 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 143 replay_map->rm_replay_slots[i] = 1; 144 } 145 146 osb->replay_map = replay_map; 147 spin_unlock(&osb->osb_lock); 148 return 0; 149} 150 151void ocfs2_queue_replay_slots(struct ocfs2_super *osb) 152{ 153 struct ocfs2_replay_map *replay_map = osb->replay_map; 154 int i; 155 156 if (!replay_map) 157 return; 158 159 if (replay_map->rm_state != REPLAY_NEEDED) 160 return; 161 162 for (i = 0; i < replay_map->rm_slots; i++) 163 if (replay_map->rm_replay_slots[i]) 164 ocfs2_queue_recovery_completion(osb->journal, i, NULL, 165 NULL, NULL); 166 replay_map->rm_state = REPLAY_DONE; 167} 168 169void ocfs2_free_replay_slots(struct ocfs2_super *osb) 170{ 171 struct ocfs2_replay_map *replay_map = osb->replay_map; 172 173 if (!osb->replay_map) 174 return; 175 176 kfree(replay_map); 177 osb->replay_map = NULL; 178} 179 180int ocfs2_recovery_init(struct ocfs2_super *osb) 181{ 182 struct ocfs2_recovery_map *rm; 183 184 mutex_init(&osb->recovery_lock); 185 osb->disable_recovery = 0; 186 osb->recovery_thread_task = NULL; 187 init_waitqueue_head(&osb->recovery_event); 188 189 rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 190 osb->max_slots * sizeof(unsigned int), 191 GFP_KERNEL); 192 if (!rm) { 193 mlog_errno(-ENOMEM); 194 return -ENOMEM; 195 } 196 197 rm->rm_entries = (unsigned int *)((char *)rm + 198 sizeof(struct ocfs2_recovery_map)); 199 osb->recovery_map = rm; 200 201 return 0; 202} 203 204/* we can't grab the goofy sem lock from inside wait_event, so we use 205 * memory barriers to make sure that we'll see the null task before 206 * being woken up */ 207static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 208{ 209 mb(); 210 return osb->recovery_thread_task != NULL; 211} 212 213void ocfs2_recovery_exit(struct ocfs2_super *osb) 214{ 215 struct ocfs2_recovery_map *rm; 216 217 /* disable any new recovery threads and wait for any currently 218 * running ones to exit. Do this before setting the vol_state. */ 219 mutex_lock(&osb->recovery_lock); 220 osb->disable_recovery = 1; 221 mutex_unlock(&osb->recovery_lock); 222 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 223 224 /* At this point, we know that no more recovery threads can be 225 * launched, so wait for any recovery completion work to 226 * complete. */ 227 flush_workqueue(ocfs2_wq); 228 229 /* 230 * Now that recovery is shut down, and the osb is about to be 231 * freed, the osb_lock is not taken here. 232 */ 233 rm = osb->recovery_map; 234 235 kfree(rm); 236} 237 238static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 239 unsigned int node_num) 240{ 241 int i; 242 struct ocfs2_recovery_map *rm = osb->recovery_map; 243 244 assert_spin_locked(&osb->osb_lock); 245 246 for (i = 0; i < rm->rm_used; i++) { 247 if (rm->rm_entries[i] == node_num) 248 return 1; 249 } 250 251 return 0; 252} 253 254/* Behaves like test-and-set. Returns the previous value */ 255static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 256 unsigned int node_num) 257{ 258 struct ocfs2_recovery_map *rm = osb->recovery_map; 259 260 spin_lock(&osb->osb_lock); 261 if (__ocfs2_recovery_map_test(osb, node_num)) { 262 spin_unlock(&osb->osb_lock); 263 return 1; 264 } 265 266 BUG_ON(rm->rm_used >= osb->max_slots); 267 268 rm->rm_entries[rm->rm_used] = node_num; 269 rm->rm_used++; 270 spin_unlock(&osb->osb_lock); 271 272 return 0; 273} 274 275static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 276 unsigned int node_num) 277{ 278 int i; 279 struct ocfs2_recovery_map *rm = osb->recovery_map; 280 281 spin_lock(&osb->osb_lock); 282 283 for (i = 0; i < rm->rm_used; i++) { 284 if (rm->rm_entries[i] == node_num) 285 break; 286 } 287 288 if (i < rm->rm_used) { 289 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 290 (rm->rm_used - i - 1) * sizeof(unsigned int)); 291 rm->rm_used--; 292 } 293 294 spin_unlock(&osb->osb_lock); 295} 296 297static int ocfs2_commit_cache(struct ocfs2_super *osb) 298{ 299 int status = 0; 300 unsigned int flushed; 301 unsigned long old_id; 302 struct ocfs2_journal *journal = NULL; 303 304 mlog_entry_void(); 305 306 journal = osb->journal; 307 308 /* Flush all pending commits and checkpoint the journal. */ 309 down_write(&journal->j_trans_barrier); 310 311 if (atomic_read(&journal->j_num_trans) == 0) { 312 up_write(&journal->j_trans_barrier); 313 mlog(0, "No transactions for me to flush!\n"); 314 goto finally; 315 } 316 317 jbd2_journal_lock_updates(journal->j_journal); 318 status = jbd2_journal_flush(journal->j_journal); 319 jbd2_journal_unlock_updates(journal->j_journal); 320 if (status < 0) { 321 up_write(&journal->j_trans_barrier); 322 mlog_errno(status); 323 goto finally; 324 } 325 326 old_id = ocfs2_inc_trans_id(journal); 327 328 flushed = atomic_read(&journal->j_num_trans); 329 atomic_set(&journal->j_num_trans, 0); 330 up_write(&journal->j_trans_barrier); 331 332 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", 333 journal->j_trans_id, flushed); 334 335 ocfs2_wake_downconvert_thread(osb); 336 wake_up(&journal->j_checkpointed); 337finally: 338 mlog_exit(status); 339 return status; 340} 341 342/* pass it NULL and it will allocate a new handle object for you. If 343 * you pass it a handle however, it may still return error, in which 344 * case it has free'd the passed handle for you. */ 345handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 346{ 347 journal_t *journal = osb->journal->j_journal; 348 handle_t *handle; 349 350 BUG_ON(!osb || !osb->journal->j_journal); 351 352 if (ocfs2_is_hard_readonly(osb)) 353 return ERR_PTR(-EROFS); 354 355 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 356 BUG_ON(max_buffs <= 0); 357 358 /* Nested transaction? Just return the handle... */ 359 if (journal_current_handle()) 360 return jbd2_journal_start(journal, max_buffs); 361 362 down_read(&osb->journal->j_trans_barrier); 363 364 handle = jbd2_journal_start(journal, max_buffs); 365 if (IS_ERR(handle)) { 366 up_read(&osb->journal->j_trans_barrier); 367 368 mlog_errno(PTR_ERR(handle)); 369 370 if (is_journal_aborted(journal)) { 371 ocfs2_abort(osb->sb, "Detected aborted journal"); 372 handle = ERR_PTR(-EROFS); 373 } 374 } else { 375 if (!ocfs2_mount_local(osb)) 376 atomic_inc(&(osb->journal->j_num_trans)); 377 } 378 379 return handle; 380} 381 382int ocfs2_commit_trans(struct ocfs2_super *osb, 383 handle_t *handle) 384{ 385 int ret, nested; 386 struct ocfs2_journal *journal = osb->journal; 387 388 BUG_ON(!handle); 389 390 nested = handle->h_ref > 1; 391 ret = jbd2_journal_stop(handle); 392 if (ret < 0) 393 mlog_errno(ret); 394 395 if (!nested) 396 up_read(&journal->j_trans_barrier); 397 398 return ret; 399} 400 401/* 402 * 'nblocks' is what you want to add to the current transaction. 403 * 404 * This might call jbd2_journal_restart() which will commit dirty buffers 405 * and then restart the transaction. Before calling 406 * ocfs2_extend_trans(), any changed blocks should have been 407 * dirtied. After calling it, all blocks which need to be changed must 408 * go through another set of journal_access/journal_dirty calls. 409 * 410 * WARNING: This will not release any semaphores or disk locks taken 411 * during the transaction, so make sure they were taken *before* 412 * start_trans or we'll have ordering deadlocks. 413 * 414 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 415 * good because transaction ids haven't yet been recorded on the 416 * cluster locks associated with this handle. 417 */ 418int ocfs2_extend_trans(handle_t *handle, int nblocks) 419{ 420 int status, old_nblocks; 421 422 BUG_ON(!handle); 423 BUG_ON(nblocks < 0); 424 425 if (!nblocks) 426 return 0; 427 428 old_nblocks = handle->h_buffer_credits; 429 mlog_entry_void(); 430 431 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); 432 433#ifdef CONFIG_OCFS2_DEBUG_FS 434 status = 1; 435#else 436 status = jbd2_journal_extend(handle, nblocks); 437 if (status < 0) { 438 mlog_errno(status); 439 goto bail; 440 } 441#endif 442 443 if (status > 0) { 444 mlog(0, 445 "jbd2_journal_extend failed, trying " 446 "jbd2_journal_restart\n"); 447 status = jbd2_journal_restart(handle, 448 old_nblocks + nblocks); 449 if (status < 0) { 450 mlog_errno(status); 451 goto bail; 452 } 453 } 454 455 status = 0; 456bail: 457 458 mlog_exit(status); 459 return status; 460} 461 462struct ocfs2_triggers { 463 struct jbd2_buffer_trigger_type ot_triggers; 464 int ot_offset; 465}; 466 467static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 468{ 469 return container_of(triggers, struct ocfs2_triggers, ot_triggers); 470} 471 472static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 473 struct buffer_head *bh, 474 void *data, size_t size) 475{ 476 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 477 478 /* 479 * We aren't guaranteed to have the superblock here, so we 480 * must unconditionally compute the ecc data. 481 * __ocfs2_journal_access() will only set the triggers if 482 * metaecc is enabled. 483 */ 484 ocfs2_block_check_compute(data, size, data + ot->ot_offset); 485} 486 487/* 488 * Quota blocks have their own trigger because the struct ocfs2_block_check 489 * offset depends on the blocksize. 490 */ 491static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 492 struct buffer_head *bh, 493 void *data, size_t size) 494{ 495 struct ocfs2_disk_dqtrailer *dqt = 496 ocfs2_block_dqtrailer(size, data); 497 498 /* 499 * We aren't guaranteed to have the superblock here, so we 500 * must unconditionally compute the ecc data. 501 * __ocfs2_journal_access() will only set the triggers if 502 * metaecc is enabled. 503 */ 504 ocfs2_block_check_compute(data, size, &dqt->dq_check); 505} 506 507/* 508 * Directory blocks also have their own trigger because the 509 * struct ocfs2_block_check offset depends on the blocksize. 510 */ 511static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 512 struct buffer_head *bh, 513 void *data, size_t size) 514{ 515 struct ocfs2_dir_block_trailer *trailer = 516 ocfs2_dir_trailer_from_size(size, data); 517 518 /* 519 * We aren't guaranteed to have the superblock here, so we 520 * must unconditionally compute the ecc data. 521 * __ocfs2_journal_access() will only set the triggers if 522 * metaecc is enabled. 523 */ 524 ocfs2_block_check_compute(data, size, &trailer->db_check); 525} 526 527static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 528 struct buffer_head *bh) 529{ 530 mlog(ML_ERROR, 531 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 532 "bh->b_blocknr = %llu\n", 533 (unsigned long)bh, 534 (unsigned long long)bh->b_blocknr); 535 536 /* We aren't guaranteed to have the superblock here - but if we 537 * don't, it'll just crash. */ 538 ocfs2_error(bh->b_assoc_map->host->i_sb, 539 "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 540} 541 542static struct ocfs2_triggers di_triggers = { 543 .ot_triggers = { 544 .t_frozen = ocfs2_frozen_trigger, 545 .t_abort = ocfs2_abort_trigger, 546 }, 547 .ot_offset = offsetof(struct ocfs2_dinode, i_check), 548}; 549 550static struct ocfs2_triggers eb_triggers = { 551 .ot_triggers = { 552 .t_frozen = ocfs2_frozen_trigger, 553 .t_abort = ocfs2_abort_trigger, 554 }, 555 .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 556}; 557 558static struct ocfs2_triggers rb_triggers = { 559 .ot_triggers = { 560 .t_frozen = ocfs2_frozen_trigger, 561 .t_abort = ocfs2_abort_trigger, 562 }, 563 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 564}; 565 566static struct ocfs2_triggers gd_triggers = { 567 .ot_triggers = { 568 .t_frozen = ocfs2_frozen_trigger, 569 .t_abort = ocfs2_abort_trigger, 570 }, 571 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 572}; 573 574static struct ocfs2_triggers db_triggers = { 575 .ot_triggers = { 576 .t_frozen = ocfs2_db_frozen_trigger, 577 .t_abort = ocfs2_abort_trigger, 578 }, 579}; 580 581static struct ocfs2_triggers xb_triggers = { 582 .ot_triggers = { 583 .t_frozen = ocfs2_frozen_trigger, 584 .t_abort = ocfs2_abort_trigger, 585 }, 586 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 587}; 588 589static struct ocfs2_triggers dq_triggers = { 590 .ot_triggers = { 591 .t_frozen = ocfs2_dq_frozen_trigger, 592 .t_abort = ocfs2_abort_trigger, 593 }, 594}; 595 596static struct ocfs2_triggers dr_triggers = { 597 .ot_triggers = { 598 .t_frozen = ocfs2_frozen_trigger, 599 .t_abort = ocfs2_abort_trigger, 600 }, 601 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 602}; 603 604static struct ocfs2_triggers dl_triggers = { 605 .ot_triggers = { 606 .t_frozen = ocfs2_frozen_trigger, 607 .t_abort = ocfs2_abort_trigger, 608 }, 609 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 610}; 611 612static int __ocfs2_journal_access(handle_t *handle, 613 struct ocfs2_caching_info *ci, 614 struct buffer_head *bh, 615 struct ocfs2_triggers *triggers, 616 int type) 617{ 618 int status; 619 struct ocfs2_super *osb = 620 OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 621 622 BUG_ON(!ci || !ci->ci_ops); 623 BUG_ON(!handle); 624 BUG_ON(!bh); 625 626 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", 627 (unsigned long long)bh->b_blocknr, type, 628 (type == OCFS2_JOURNAL_ACCESS_CREATE) ? 629 "OCFS2_JOURNAL_ACCESS_CREATE" : 630 "OCFS2_JOURNAL_ACCESS_WRITE", 631 bh->b_size); 632 633 /* we can safely remove this assertion after testing. */ 634 if (!buffer_uptodate(bh)) { 635 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 636 mlog(ML_ERROR, "b_blocknr=%llu\n", 637 (unsigned long long)bh->b_blocknr); 638 BUG(); 639 } 640 641 /* Set the current transaction information on the ci so 642 * that the locking code knows whether it can drop it's locks 643 * on this ci or not. We're protected from the commit 644 * thread updating the current transaction id until 645 * ocfs2_commit_trans() because ocfs2_start_trans() took 646 * j_trans_barrier for us. */ 647 ocfs2_set_ci_lock_trans(osb->journal, ci); 648 649 ocfs2_metadata_cache_io_lock(ci); 650 switch (type) { 651 case OCFS2_JOURNAL_ACCESS_CREATE: 652 case OCFS2_JOURNAL_ACCESS_WRITE: 653 status = jbd2_journal_get_write_access(handle, bh); 654 break; 655 656 case OCFS2_JOURNAL_ACCESS_UNDO: 657 status = jbd2_journal_get_undo_access(handle, bh); 658 break; 659 660 default: 661 status = -EINVAL; 662 mlog(ML_ERROR, "Unknown access type!\n"); 663 } 664 if (!status && ocfs2_meta_ecc(osb) && triggers) 665 jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 666 ocfs2_metadata_cache_io_unlock(ci); 667 668 if (status < 0) 669 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 670 status, type); 671 672 mlog_exit(status); 673 return status; 674} 675 676int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 677 struct buffer_head *bh, int type) 678{ 679 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 680} 681 682int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 683 struct buffer_head *bh, int type) 684{ 685 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 686} 687 688int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 689 struct buffer_head *bh, int type) 690{ 691 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 692 type); 693} 694 695int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 696 struct buffer_head *bh, int type) 697{ 698 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 699} 700 701int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 702 struct buffer_head *bh, int type) 703{ 704 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 705} 706 707int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 708 struct buffer_head *bh, int type) 709{ 710 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 711} 712 713int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 714 struct buffer_head *bh, int type) 715{ 716 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 717} 718 719int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 720 struct buffer_head *bh, int type) 721{ 722 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 723} 724 725int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 726 struct buffer_head *bh, int type) 727{ 728 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 729} 730 731int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 732 struct buffer_head *bh, int type) 733{ 734 return __ocfs2_journal_access(handle, ci, bh, NULL, type); 735} 736 737void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 738{ 739 int status; 740 741 mlog_entry("(bh->b_blocknr=%llu)\n", 742 (unsigned long long)bh->b_blocknr); 743 744 status = jbd2_journal_dirty_metadata(handle, bh); 745 BUG_ON(status); 746 747 mlog_exit_void(); 748} 749 750#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 751 752void ocfs2_set_journal_params(struct ocfs2_super *osb) 753{ 754 journal_t *journal = osb->journal->j_journal; 755 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 756 757 if (osb->osb_commit_interval) 758 commit_interval = osb->osb_commit_interval; 759 760 write_lock(&journal->j_state_lock); 761 journal->j_commit_interval = commit_interval; 762 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 763 journal->j_flags |= JBD2_BARRIER; 764 else 765 journal->j_flags &= ~JBD2_BARRIER; 766 write_unlock(&journal->j_state_lock); 767} 768 769int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 770{ 771 int status = -1; 772 struct inode *inode = NULL; /* the journal inode */ 773 journal_t *j_journal = NULL; 774 struct ocfs2_dinode *di = NULL; 775 struct buffer_head *bh = NULL; 776 struct ocfs2_super *osb; 777 int inode_lock = 0; 778 779 mlog_entry_void(); 780 781 BUG_ON(!journal); 782 783 osb = journal->j_osb; 784 785 /* already have the inode for our journal */ 786 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 787 osb->slot_num); 788 if (inode == NULL) { 789 status = -EACCES; 790 mlog_errno(status); 791 goto done; 792 } 793 if (is_bad_inode(inode)) { 794 mlog(ML_ERROR, "access error (bad inode)\n"); 795 iput(inode); 796 inode = NULL; 797 status = -EACCES; 798 goto done; 799 } 800 801 SET_INODE_JOURNAL(inode); 802 OCFS2_I(inode)->ip_open_count++; 803 804 /* Skip recovery waits here - journal inode metadata never 805 * changes in a live cluster so it can be considered an 806 * exception to the rule. */ 807 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 808 if (status < 0) { 809 if (status != -ERESTARTSYS) 810 mlog(ML_ERROR, "Could not get lock on journal!\n"); 811 goto done; 812 } 813 814 inode_lock = 1; 815 di = (struct ocfs2_dinode *)bh->b_data; 816 817 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { 818 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 819 inode->i_size); 820 status = -EINVAL; 821 goto done; 822 } 823 824 mlog(0, "inode->i_size = %lld\n", inode->i_size); 825 mlog(0, "inode->i_blocks = %llu\n", 826 (unsigned long long)inode->i_blocks); 827 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); 828 829 /* call the kernels journal init function now */ 830 j_journal = jbd2_journal_init_inode(inode); 831 if (j_journal == NULL) { 832 mlog(ML_ERROR, "Linux journal layer error\n"); 833 status = -EINVAL; 834 goto done; 835 } 836 837 mlog(0, "Returned from jbd2_journal_init_inode\n"); 838 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); 839 840 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 841 OCFS2_JOURNAL_DIRTY_FL); 842 843 journal->j_journal = j_journal; 844 journal->j_inode = inode; 845 journal->j_bh = bh; 846 847 ocfs2_set_journal_params(osb); 848 849 journal->j_state = OCFS2_JOURNAL_LOADED; 850 851 status = 0; 852done: 853 if (status < 0) { 854 if (inode_lock) 855 ocfs2_inode_unlock(inode, 1); 856 brelse(bh); 857 if (inode) { 858 OCFS2_I(inode)->ip_open_count--; 859 iput(inode); 860 } 861 } 862 863 mlog_exit(status); 864 return status; 865} 866 867static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 868{ 869 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 870} 871 872static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 873{ 874 return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 875} 876 877static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 878 int dirty, int replayed) 879{ 880 int status; 881 unsigned int flags; 882 struct ocfs2_journal *journal = osb->journal; 883 struct buffer_head *bh = journal->j_bh; 884 struct ocfs2_dinode *fe; 885 886 mlog_entry_void(); 887 888 fe = (struct ocfs2_dinode *)bh->b_data; 889 890 /* The journal bh on the osb always comes from ocfs2_journal_init() 891 * and was validated there inside ocfs2_inode_lock_full(). It's a 892 * code bug if we mess it up. */ 893 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 894 895 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 896 if (dirty) 897 flags |= OCFS2_JOURNAL_DIRTY_FL; 898 else 899 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 900 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 901 902 if (replayed) 903 ocfs2_bump_recovery_generation(fe); 904 905 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 906 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 907 if (status < 0) 908 mlog_errno(status); 909 910 mlog_exit(status); 911 return status; 912} 913 914/* 915 * If the journal has been kmalloc'd it needs to be freed after this 916 * call. 917 */ 918void ocfs2_journal_shutdown(struct ocfs2_super *osb) 919{ 920 struct ocfs2_journal *journal = NULL; 921 int status = 0; 922 struct inode *inode = NULL; 923 int num_running_trans = 0; 924 925 mlog_entry_void(); 926 927 BUG_ON(!osb); 928 929 journal = osb->journal; 930 if (!journal) 931 goto done; 932 933 inode = journal->j_inode; 934 935 if (journal->j_state != OCFS2_JOURNAL_LOADED) 936 goto done; 937 938 /* need to inc inode use count - jbd2_journal_destroy will iput. */ 939 if (!igrab(inode)) 940 BUG(); 941 942 num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 943 if (num_running_trans > 0) 944 mlog(0, "Shutting down journal: must wait on %d " 945 "running transactions!\n", 946 num_running_trans); 947 948 /* Do a commit_cache here. It will flush our journal, *and* 949 * release any locks that are still held. 950 * set the SHUTDOWN flag and release the trans lock. 951 * the commit thread will take the trans lock for us below. */ 952 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 953 954 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 955 * drop the trans_lock (which we want to hold until we 956 * completely destroy the journal. */ 957 if (osb->commit_task) { 958 /* Wait for the commit thread */ 959 mlog(0, "Waiting for ocfs2commit to exit....\n"); 960 kthread_stop(osb->commit_task); 961 osb->commit_task = NULL; 962 } 963 964 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 965 966 if (ocfs2_mount_local(osb)) { 967 jbd2_journal_lock_updates(journal->j_journal); 968 status = jbd2_journal_flush(journal->j_journal); 969 jbd2_journal_unlock_updates(journal->j_journal); 970 if (status < 0) 971 mlog_errno(status); 972 } 973 974 if (status == 0) { 975 /* 976 * Do not toggle if flush was unsuccessful otherwise 977 * will leave dirty metadata in a "clean" journal 978 */ 979 status = ocfs2_journal_toggle_dirty(osb, 0, 0); 980 if (status < 0) 981 mlog_errno(status); 982 } 983 984 /* Shutdown the kernel journal system */ 985 jbd2_journal_destroy(journal->j_journal); 986 journal->j_journal = NULL; 987 988 OCFS2_I(inode)->ip_open_count--; 989 990 /* unlock our journal */ 991 ocfs2_inode_unlock(inode, 1); 992 993 brelse(journal->j_bh); 994 journal->j_bh = NULL; 995 996 journal->j_state = OCFS2_JOURNAL_FREE; 997 998// up_write(&journal->j_trans_barrier); 999done: 1000 if (inode) 1001 iput(inode); 1002 mlog_exit_void(); 1003} 1004 1005static void ocfs2_clear_journal_error(struct super_block *sb, 1006 journal_t *journal, 1007 int slot) 1008{ 1009 int olderr; 1010 1011 olderr = jbd2_journal_errno(journal); 1012 if (olderr) { 1013 mlog(ML_ERROR, "File system error %d recorded in " 1014 "journal %u.\n", olderr, slot); 1015 mlog(ML_ERROR, "File system on device %s needs checking.\n", 1016 sb->s_id); 1017 1018 jbd2_journal_ack_err(journal); 1019 jbd2_journal_clear_err(journal); 1020 } 1021} 1022 1023int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1024{ 1025 int status = 0; 1026 struct ocfs2_super *osb; 1027 1028 mlog_entry_void(); 1029 1030 BUG_ON(!journal); 1031 1032 osb = journal->j_osb; 1033 1034 status = jbd2_journal_load(journal->j_journal); 1035 if (status < 0) { 1036 mlog(ML_ERROR, "Failed to load journal!\n"); 1037 goto done; 1038 } 1039 1040 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1041 1042 status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1043 if (status < 0) { 1044 mlog_errno(status); 1045 goto done; 1046 } 1047 1048 /* Launch the commit thread */ 1049 if (!local) { 1050 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 1051 "ocfs2cmt"); 1052 if (IS_ERR(osb->commit_task)) { 1053 status = PTR_ERR(osb->commit_task); 1054 osb->commit_task = NULL; 1055 mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1056 "error=%d", status); 1057 goto done; 1058 } 1059 } else 1060 osb->commit_task = NULL; 1061 1062done: 1063 mlog_exit(status); 1064 return status; 1065} 1066 1067 1068/* 'full' flag tells us whether we clear out all blocks or if we just 1069 * mark the journal clean */ 1070int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1071{ 1072 int status; 1073 1074 mlog_entry_void(); 1075 1076 BUG_ON(!journal); 1077 1078 status = jbd2_journal_wipe(journal->j_journal, full); 1079 if (status < 0) { 1080 mlog_errno(status); 1081 goto bail; 1082 } 1083 1084 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1085 if (status < 0) 1086 mlog_errno(status); 1087 1088bail: 1089 mlog_exit(status); 1090 return status; 1091} 1092 1093static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1094{ 1095 int empty; 1096 struct ocfs2_recovery_map *rm = osb->recovery_map; 1097 1098 spin_lock(&osb->osb_lock); 1099 empty = (rm->rm_used == 0); 1100 spin_unlock(&osb->osb_lock); 1101 1102 return empty; 1103} 1104 1105void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1106{ 1107 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1108} 1109 1110/* 1111 * JBD Might read a cached version of another nodes journal file. We 1112 * don't want this as this file changes often and we get no 1113 * notification on those changes. The only way to be sure that we've 1114 * got the most up to date version of those blocks then is to force 1115 * read them off disk. Just searching through the buffer cache won't 1116 * work as there may be pages backing this file which are still marked 1117 * up to date. We know things can't change on this file underneath us 1118 * as we have the lock by now :) 1119 */ 1120static int ocfs2_force_read_journal(struct inode *inode) 1121{ 1122 int status = 0; 1123 int i; 1124 u64 v_blkno, p_blkno, p_blocks, num_blocks; 1125#define CONCURRENT_JOURNAL_FILL 32ULL 1126 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; 1127 1128 mlog_entry_void(); 1129 1130 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); 1131 1132 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); 1133 v_blkno = 0; 1134 while (v_blkno < num_blocks) { 1135 status = ocfs2_extent_map_get_blocks(inode, v_blkno, 1136 &p_blkno, &p_blocks, NULL); 1137 if (status < 0) { 1138 mlog_errno(status); 1139 goto bail; 1140 } 1141 1142 if (p_blocks > CONCURRENT_JOURNAL_FILL) 1143 p_blocks = CONCURRENT_JOURNAL_FILL; 1144 1145 /* We are reading journal data which should not 1146 * be put in the uptodate cache */ 1147 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), 1148 p_blkno, p_blocks, bhs); 1149 if (status < 0) { 1150 mlog_errno(status); 1151 goto bail; 1152 } 1153 1154 for(i = 0; i < p_blocks; i++) { 1155 brelse(bhs[i]); 1156 bhs[i] = NULL; 1157 } 1158 1159 v_blkno += p_blocks; 1160 } 1161 1162bail: 1163 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) 1164 brelse(bhs[i]); 1165 mlog_exit(status); 1166 return status; 1167} 1168 1169struct ocfs2_la_recovery_item { 1170 struct list_head lri_list; 1171 int lri_slot; 1172 struct ocfs2_dinode *lri_la_dinode; 1173 struct ocfs2_dinode *lri_tl_dinode; 1174 struct ocfs2_quota_recovery *lri_qrec; 1175}; 1176 1177/* Does the second half of the recovery process. By this point, the 1178 * node is marked clean and can actually be considered recovered, 1179 * hence it's no longer in the recovery map, but there's still some 1180 * cleanup we can do which shouldn't happen within the recovery thread 1181 * as locking in that context becomes very difficult if we are to take 1182 * recovering nodes into account. 1183 * 1184 * NOTE: This function can and will sleep on recovery of other nodes 1185 * during cluster locking, just like any other ocfs2 process. 1186 */ 1187void ocfs2_complete_recovery(struct work_struct *work) 1188{ 1189 int ret; 1190 struct ocfs2_journal *journal = 1191 container_of(work, struct ocfs2_journal, j_recovery_work); 1192 struct ocfs2_super *osb = journal->j_osb; 1193 struct ocfs2_dinode *la_dinode, *tl_dinode; 1194 struct ocfs2_la_recovery_item *item, *n; 1195 struct ocfs2_quota_recovery *qrec; 1196 LIST_HEAD(tmp_la_list); 1197 1198 mlog_entry_void(); 1199 1200 mlog(0, "completing recovery from keventd\n"); 1201 1202 spin_lock(&journal->j_lock); 1203 list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1204 spin_unlock(&journal->j_lock); 1205 1206 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1207 list_del_init(&item->lri_list); 1208 1209 mlog(0, "Complete recovery for slot %d\n", item->lri_slot); 1210 1211 ocfs2_wait_on_quotas(osb); 1212 1213 la_dinode = item->lri_la_dinode; 1214 if (la_dinode) { 1215 mlog(0, "Clean up local alloc %llu\n", 1216 (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); 1217 1218 ret = ocfs2_complete_local_alloc_recovery(osb, 1219 la_dinode); 1220 if (ret < 0) 1221 mlog_errno(ret); 1222 1223 kfree(la_dinode); 1224 } 1225 1226 tl_dinode = item->lri_tl_dinode; 1227 if (tl_dinode) { 1228 mlog(0, "Clean up truncate log %llu\n", 1229 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); 1230 1231 ret = ocfs2_complete_truncate_log_recovery(osb, 1232 tl_dinode); 1233 if (ret < 0) 1234 mlog_errno(ret); 1235 1236 kfree(tl_dinode); 1237 } 1238 1239 ret = ocfs2_recover_orphans(osb, item->lri_slot); 1240 if (ret < 0) 1241 mlog_errno(ret); 1242 1243 qrec = item->lri_qrec; 1244 if (qrec) { 1245 mlog(0, "Recovering quota files"); 1246 ret = ocfs2_finish_quota_recovery(osb, qrec, 1247 item->lri_slot); 1248 if (ret < 0) 1249 mlog_errno(ret); 1250 /* Recovery info is already freed now */ 1251 } 1252 1253 kfree(item); 1254 } 1255 1256 mlog(0, "Recovery completion\n"); 1257 mlog_exit_void(); 1258} 1259 1260/* NOTE: This function always eats your references to la_dinode and 1261 * tl_dinode, either manually on error, or by passing them to 1262 * ocfs2_complete_recovery */ 1263static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1264 int slot_num, 1265 struct ocfs2_dinode *la_dinode, 1266 struct ocfs2_dinode *tl_dinode, 1267 struct ocfs2_quota_recovery *qrec) 1268{ 1269 struct ocfs2_la_recovery_item *item; 1270 1271 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1272 if (!item) { 1273 /* Though we wish to avoid it, we are in fact safe in 1274 * skipping local alloc cleanup as fsck.ocfs2 is more 1275 * than capable of reclaiming unused space. */ 1276 if (la_dinode) 1277 kfree(la_dinode); 1278 1279 if (tl_dinode) 1280 kfree(tl_dinode); 1281 1282 if (qrec) 1283 ocfs2_free_quota_recovery(qrec); 1284 1285 mlog_errno(-ENOMEM); 1286 return; 1287 } 1288 1289 INIT_LIST_HEAD(&item->lri_list); 1290 item->lri_la_dinode = la_dinode; 1291 item->lri_slot = slot_num; 1292 item->lri_tl_dinode = tl_dinode; 1293 item->lri_qrec = qrec; 1294 1295 spin_lock(&journal->j_lock); 1296 list_add_tail(&item->lri_list, &journal->j_la_cleanups); 1297 queue_work(ocfs2_wq, &journal->j_recovery_work); 1298 spin_unlock(&journal->j_lock); 1299} 1300 1301/* Called by the mount code to queue recovery the last part of 1302 * recovery for it's own and offline slot(s). */ 1303void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1304{ 1305 struct ocfs2_journal *journal = osb->journal; 1306 1307 /* No need to queue up our truncate_log as regular cleanup will catch 1308 * that */ 1309 ocfs2_queue_recovery_completion(journal, osb->slot_num, 1310 osb->local_alloc_copy, NULL, NULL); 1311 ocfs2_schedule_truncate_log_flush(osb, 0); 1312 1313 osb->local_alloc_copy = NULL; 1314 osb->dirty = 0; 1315 1316 /* queue to recover orphan slots for all offline slots */ 1317 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1318 ocfs2_queue_replay_slots(osb); 1319 ocfs2_free_replay_slots(osb); 1320} 1321 1322void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 1323{ 1324 if (osb->quota_rec) { 1325 ocfs2_queue_recovery_completion(osb->journal, 1326 osb->slot_num, 1327 NULL, 1328 NULL, 1329 osb->quota_rec); 1330 osb->quota_rec = NULL; 1331 } 1332} 1333 1334static int __ocfs2_recovery_thread(void *arg) 1335{ 1336 int status, node_num, slot_num; 1337 struct ocfs2_super *osb = arg; 1338 struct ocfs2_recovery_map *rm = osb->recovery_map; 1339 int *rm_quota = NULL; 1340 int rm_quota_used = 0, i; 1341 struct ocfs2_quota_recovery *qrec; 1342 1343 mlog_entry_void(); 1344 1345 status = ocfs2_wait_on_mount(osb); 1346 if (status < 0) { 1347 goto bail; 1348 } 1349 1350 rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); 1351 if (!rm_quota) { 1352 status = -ENOMEM; 1353 goto bail; 1354 } 1355restart: 1356 status = ocfs2_super_lock(osb, 1); 1357 if (status < 0) { 1358 mlog_errno(status); 1359 goto bail; 1360 } 1361 1362 status = ocfs2_compute_replay_slots(osb); 1363 if (status < 0) 1364 mlog_errno(status); 1365 1366 /* queue recovery for our own slot */ 1367 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1368 NULL, NULL); 1369 1370 spin_lock(&osb->osb_lock); 1371 while (rm->rm_used) { 1372 /* It's always safe to remove entry zero, as we won't 1373 * clear it until ocfs2_recover_node() has succeeded. */ 1374 node_num = rm->rm_entries[0]; 1375 spin_unlock(&osb->osb_lock); 1376 mlog(0, "checking node %d\n", node_num); 1377 slot_num = ocfs2_node_num_to_slot(osb, node_num); 1378 if (slot_num == -ENOENT) { 1379 status = 0; 1380 mlog(0, "no slot for this node, so no recovery" 1381 "required.\n"); 1382 goto skip_recovery; 1383 } 1384 mlog(0, "node %d was using slot %d\n", node_num, slot_num); 1385 1386 /* It is a bit subtle with quota recovery. We cannot do it 1387 * immediately because we have to obtain cluster locks from 1388 * quota files and we also don't want to just skip it because 1389 * then quota usage would be out of sync until some node takes 1390 * the slot. So we remember which nodes need quota recovery 1391 * and when everything else is done, we recover quotas. */ 1392 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); 1393 if (i == rm_quota_used) 1394 rm_quota[rm_quota_used++] = slot_num; 1395 1396 status = ocfs2_recover_node(osb, node_num, slot_num); 1397skip_recovery: 1398 if (!status) { 1399 ocfs2_recovery_map_clear(osb, node_num); 1400 } else { 1401 mlog(ML_ERROR, 1402 "Error %d recovering node %d on device (%u,%u)!\n", 1403 status, node_num, 1404 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1405 mlog(ML_ERROR, "Volume requires unmount.\n"); 1406 } 1407 1408 spin_lock(&osb->osb_lock); 1409 } 1410 spin_unlock(&osb->osb_lock); 1411 mlog(0, "All nodes recovered\n"); 1412 1413 /* Refresh all journal recovery generations from disk */ 1414 status = ocfs2_check_journals_nolocks(osb); 1415 status = (status == -EROFS) ? 0 : status; 1416 if (status < 0) 1417 mlog_errno(status); 1418 1419 /* Now it is right time to recover quotas... We have to do this under 1420 * superblock lock so that noone can start using the slot (and crash) 1421 * before we recover it */ 1422 for (i = 0; i < rm_quota_used; i++) { 1423 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 1424 if (IS_ERR(qrec)) { 1425 status = PTR_ERR(qrec); 1426 mlog_errno(status); 1427 continue; 1428 } 1429 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], 1430 NULL, NULL, qrec); 1431 } 1432 1433 ocfs2_super_unlock(osb, 1); 1434 1435 /* queue recovery for offline slots */ 1436 ocfs2_queue_replay_slots(osb); 1437 1438bail: 1439 mutex_lock(&osb->recovery_lock); 1440 if (!status && !ocfs2_recovery_completed(osb)) { 1441 mutex_unlock(&osb->recovery_lock); 1442 goto restart; 1443 } 1444 1445 ocfs2_free_replay_slots(osb); 1446 osb->recovery_thread_task = NULL; 1447 mb(); /* sync with ocfs2_recovery_thread_running */ 1448 wake_up(&osb->recovery_event); 1449 1450 mutex_unlock(&osb->recovery_lock); 1451 1452 if (rm_quota) 1453 kfree(rm_quota); 1454 1455 mlog_exit(status); 1456 /* no one is callint kthread_stop() for us so the kthread() api 1457 * requires that we call do_exit(). And it isn't exported, but 1458 * complete_and_exit() seems to be a minimal wrapper around it. */ 1459 complete_and_exit(NULL, status); 1460 return status; 1461} 1462 1463void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1464{ 1465 mlog_entry("(node_num=%d, osb->node_num = %d)\n", 1466 node_num, osb->node_num); 1467 1468 mutex_lock(&osb->recovery_lock); 1469 if (osb->disable_recovery) 1470 goto out; 1471 1472 /* People waiting on recovery will wait on 1473 * the recovery map to empty. */ 1474 if (ocfs2_recovery_map_set(osb, node_num)) 1475 mlog(0, "node %d already in recovery map.\n", node_num); 1476 1477 mlog(0, "starting recovery thread...\n"); 1478 1479 if (osb->recovery_thread_task) 1480 goto out; 1481 1482 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 1483 "ocfs2rec"); 1484 if (IS_ERR(osb->recovery_thread_task)) { 1485 mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1486 osb->recovery_thread_task = NULL; 1487 } 1488 1489out: 1490 mutex_unlock(&osb->recovery_lock); 1491 wake_up(&osb->recovery_event); 1492 1493 mlog_exit_void(); 1494} 1495 1496static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1497 int slot_num, 1498 struct buffer_head **bh, 1499 struct inode **ret_inode) 1500{ 1501 int status = -EACCES; 1502 struct inode *inode = NULL; 1503 1504 BUG_ON(slot_num >= osb->max_slots); 1505 1506 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1507 slot_num); 1508 if (!inode || is_bad_inode(inode)) { 1509 mlog_errno(status); 1510 goto bail; 1511 } 1512 SET_INODE_JOURNAL(inode); 1513 1514 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1515 if (status < 0) { 1516 mlog_errno(status); 1517 goto bail; 1518 } 1519 1520 status = 0; 1521 1522bail: 1523 if (inode) { 1524 if (status || !ret_inode) 1525 iput(inode); 1526 else 1527 *ret_inode = inode; 1528 } 1529 return status; 1530} 1531 1532/* Does the actual journal replay and marks the journal inode as 1533 * clean. Will only replay if the journal inode is marked dirty. */ 1534static int ocfs2_replay_journal(struct ocfs2_super *osb, 1535 int node_num, 1536 int slot_num) 1537{ 1538 int status; 1539 int got_lock = 0; 1540 unsigned int flags; 1541 struct inode *inode = NULL; 1542 struct ocfs2_dinode *fe; 1543 journal_t *journal = NULL; 1544 struct buffer_head *bh = NULL; 1545 u32 slot_reco_gen; 1546 1547 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1548 if (status) { 1549 mlog_errno(status); 1550 goto done; 1551 } 1552 1553 fe = (struct ocfs2_dinode *)bh->b_data; 1554 slot_reco_gen = ocfs2_get_recovery_generation(fe); 1555 brelse(bh); 1556 bh = NULL; 1557 1558 /* 1559 * As the fs recovery is asynchronous, there is a small chance that 1560 * another node mounted (and recovered) the slot before the recovery 1561 * thread could get the lock. To handle that, we dirty read the journal 1562 * inode for that slot to get the recovery generation. If it is 1563 * different than what we expected, the slot has been recovered. 1564 * If not, it needs recovery. 1565 */ 1566 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1567 mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num, 1568 osb->slot_recovery_generations[slot_num], slot_reco_gen); 1569 osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1570 status = -EBUSY; 1571 goto done; 1572 } 1573 1574 /* Continue with recovery as the journal has not yet been recovered */ 1575 1576 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1577 if (status < 0) { 1578 mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); 1579 if (status != -ERESTARTSYS) 1580 mlog(ML_ERROR, "Could not lock journal!\n"); 1581 goto done; 1582 } 1583 got_lock = 1; 1584 1585 fe = (struct ocfs2_dinode *) bh->b_data; 1586 1587 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1588 slot_reco_gen = ocfs2_get_recovery_generation(fe); 1589 1590 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1591 mlog(0, "No recovery required for node %d\n", node_num); 1592 /* Refresh recovery generation for the slot */ 1593 osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1594 goto done; 1595 } 1596 1597 /* we need to run complete recovery for offline orphan slots */ 1598 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1599 1600 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", 1601 node_num, slot_num, 1602 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1603 1604 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1605 1606 status = ocfs2_force_read_journal(inode); 1607 if (status < 0) { 1608 mlog_errno(status); 1609 goto done; 1610 } 1611 1612 mlog(0, "calling journal_init_inode\n"); 1613 journal = jbd2_journal_init_inode(inode); 1614 if (journal == NULL) { 1615 mlog(ML_ERROR, "Linux journal layer error\n"); 1616 status = -EIO; 1617 goto done; 1618 } 1619 1620 status = jbd2_journal_load(journal); 1621 if (status < 0) { 1622 mlog_errno(status); 1623 if (!igrab(inode)) 1624 BUG(); 1625 jbd2_journal_destroy(journal); 1626 goto done; 1627 } 1628 1629 ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1630 1631 /* wipe the journal */ 1632 mlog(0, "flushing the journal.\n"); 1633 jbd2_journal_lock_updates(journal); 1634 status = jbd2_journal_flush(journal); 1635 jbd2_journal_unlock_updates(journal); 1636 if (status < 0) 1637 mlog_errno(status); 1638 1639 /* This will mark the node clean */ 1640 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1641 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1642 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1643 1644 /* Increment recovery generation to indicate successful recovery */ 1645 ocfs2_bump_recovery_generation(fe); 1646 osb->slot_recovery_generations[slot_num] = 1647 ocfs2_get_recovery_generation(fe); 1648 1649 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 1650 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1651 if (status < 0) 1652 mlog_errno(status); 1653 1654 if (!igrab(inode)) 1655 BUG(); 1656 1657 jbd2_journal_destroy(journal); 1658 1659done: 1660 /* drop the lock on this nodes journal */ 1661 if (got_lock) 1662 ocfs2_inode_unlock(inode, 1); 1663 1664 if (inode) 1665 iput(inode); 1666 1667 brelse(bh); 1668 1669 mlog_exit(status); 1670 return status; 1671} 1672 1673/* 1674 * Do the most important parts of node recovery: 1675 * - Replay it's journal 1676 * - Stamp a clean local allocator file 1677 * - Stamp a clean truncate log 1678 * - Mark the node clean 1679 * 1680 * If this function completes without error, a node in OCFS2 can be 1681 * said to have been safely recovered. As a result, failure during the 1682 * second part of a nodes recovery process (local alloc recovery) is 1683 * far less concerning. 1684 */ 1685static int ocfs2_recover_node(struct ocfs2_super *osb, 1686 int node_num, int slot_num) 1687{ 1688 int status = 0; 1689 struct ocfs2_dinode *la_copy = NULL; 1690 struct ocfs2_dinode *tl_copy = NULL; 1691 1692 mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n", 1693 node_num, slot_num, osb->node_num); 1694 1695 /* Should not ever be called to recover ourselves -- in that 1696 * case we should've called ocfs2_journal_load instead. */ 1697 BUG_ON(osb->node_num == node_num); 1698 1699 status = ocfs2_replay_journal(osb, node_num, slot_num); 1700 if (status < 0) { 1701 if (status == -EBUSY) { 1702 mlog(0, "Skipping recovery for slot %u (node %u) " 1703 "as another node has recovered it\n", slot_num, 1704 node_num); 1705 status = 0; 1706 goto done; 1707 } 1708 mlog_errno(status); 1709 goto done; 1710 } 1711 1712 /* Stamp a clean local alloc file AFTER recovering the journal... */ 1713 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1714 if (status < 0) { 1715 mlog_errno(status); 1716 goto done; 1717 } 1718 1719 /* An error from begin_truncate_log_recovery is not 1720 * serious enough to warrant halting the rest of 1721 * recovery. */ 1722 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1723 if (status < 0) 1724 mlog_errno(status); 1725 1726 /* Likewise, this would be a strange but ultimately not so 1727 * harmful place to get an error... */ 1728 status = ocfs2_clear_slot(osb, slot_num); 1729 if (status < 0) 1730 mlog_errno(status); 1731 1732 /* This will kfree the memory pointed to by la_copy and tl_copy */ 1733 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1734 tl_copy, NULL); 1735 1736 status = 0; 1737done: 1738 1739 mlog_exit(status); 1740 return status; 1741} 1742 1743/* Test node liveness by trylocking his journal. If we get the lock, 1744 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1745 * still alive (we couldn't get the lock) and < 0 on error. */ 1746static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1747 int slot_num) 1748{ 1749 int status, flags; 1750 struct inode *inode = NULL; 1751 1752 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1753 slot_num); 1754 if (inode == NULL) { 1755 mlog(ML_ERROR, "access error\n"); 1756 status = -EACCES; 1757 goto bail; 1758 } 1759 if (is_bad_inode(inode)) { 1760 mlog(ML_ERROR, "access error (bad inode)\n"); 1761 iput(inode); 1762 inode = NULL; 1763 status = -EACCES; 1764 goto bail; 1765 } 1766 SET_INODE_JOURNAL(inode); 1767 1768 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1769 status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1770 if (status < 0) { 1771 if (status != -EAGAIN) 1772 mlog_errno(status); 1773 goto bail; 1774 } 1775 1776 ocfs2_inode_unlock(inode, 1); 1777bail: 1778 if (inode) 1779 iput(inode); 1780 1781 return status; 1782} 1783 1784/* Call this underneath ocfs2_super_lock. It also assumes that the 1785 * slot info struct has been updated from disk. */ 1786int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1787{ 1788 unsigned int node_num; 1789 int status, i; 1790 u32 gen; 1791 struct buffer_head *bh = NULL; 1792 struct ocfs2_dinode *di; 1793 1794 /* This is called with the super block cluster lock, so we 1795 * know that the slot map can't change underneath us. */ 1796 1797 for (i = 0; i < osb->max_slots; i++) { 1798 /* Read journal inode to get the recovery generation */ 1799 status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1800 if (status) { 1801 mlog_errno(status); 1802 goto bail; 1803 } 1804 di = (struct ocfs2_dinode *)bh->b_data; 1805 gen = ocfs2_get_recovery_generation(di); 1806 brelse(bh); 1807 bh = NULL; 1808 1809 spin_lock(&osb->osb_lock); 1810 osb->slot_recovery_generations[i] = gen; 1811 1812 mlog(0, "Slot %u recovery generation is %u\n", i, 1813 osb->slot_recovery_generations[i]); 1814 1815 if (i == osb->slot_num) { 1816 spin_unlock(&osb->osb_lock); 1817 continue; 1818 } 1819 1820 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1821 if (status == -ENOENT) { 1822 spin_unlock(&osb->osb_lock); 1823 continue; 1824 } 1825 1826 if (__ocfs2_recovery_map_test(osb, node_num)) { 1827 spin_unlock(&osb->osb_lock); 1828 continue; 1829 } 1830 spin_unlock(&osb->osb_lock); 1831 1832 /* Ok, we have a slot occupied by another node which 1833 * is not in the recovery map. We trylock his journal 1834 * file here to test if he's alive. */ 1835 status = ocfs2_trylock_journal(osb, i); 1836 if (!status) { 1837 /* Since we're called from mount, we know that 1838 * the recovery thread can't race us on 1839 * setting / checking the recovery bits. */ 1840 ocfs2_recovery_thread(osb, node_num); 1841 } else if ((status < 0) && (status != -EAGAIN)) { 1842 mlog_errno(status); 1843 goto bail; 1844 } 1845 } 1846 1847 status = 0; 1848bail: 1849 mlog_exit(status); 1850 return status; 1851} 1852 1853/* 1854 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 1855 * randomness to the timeout to minimize multple nodes firing the timer at the 1856 * same time. 1857 */ 1858static inline unsigned long ocfs2_orphan_scan_timeout(void) 1859{ 1860 unsigned long time; 1861 1862 get_random_bytes(&time, sizeof(time)); 1863 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 1864 return msecs_to_jiffies(time); 1865} 1866 1867/* 1868 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 1869 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 1870 * is done to catch any orphans that are left over in orphan directories. 1871 * 1872 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 1873 * seconds. It gets an EX lock on os_lockres and checks sequence number 1874 * stored in LVB. If the sequence number has changed, it means some other 1875 * node has done the scan. This node skips the scan and tracks the 1876 * sequence number. If the sequence number didn't change, it means a scan 1877 * hasn't happened. The node queues a scan and increments the 1878 * sequence number in the LVB. 1879 */ 1880void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 1881{ 1882 struct ocfs2_orphan_scan *os; 1883 int status, i; 1884 u32 seqno = 0; 1885 1886 os = &osb->osb_orphan_scan; 1887 1888 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1889 goto out; 1890 1891 status = ocfs2_orphan_scan_lock(osb, &seqno); 1892 if (status < 0) { 1893 if (status != -EAGAIN) 1894 mlog_errno(status); 1895 goto out; 1896 } 1897 1898 /* Do no queue the tasks if the volume is being umounted */ 1899 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1900 goto unlock; 1901 1902 if (os->os_seqno != seqno) { 1903 os->os_seqno = seqno; 1904 goto unlock; 1905 } 1906 1907 for (i = 0; i < osb->max_slots; i++) 1908 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1909 NULL); 1910 /* 1911 * We queued a recovery on orphan slots, increment the sequence 1912 * number and update LVB so other node will skip the scan for a while 1913 */ 1914 seqno++; 1915 os->os_count++; 1916 os->os_scantime = CURRENT_TIME; 1917unlock: 1918 ocfs2_orphan_scan_unlock(osb, seqno); 1919out: 1920 return; 1921} 1922 1923/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 1924void ocfs2_orphan_scan_work(struct work_struct *work) 1925{ 1926 struct ocfs2_orphan_scan *os; 1927 struct ocfs2_super *osb; 1928 1929 os = container_of(work, struct ocfs2_orphan_scan, 1930 os_orphan_scan_work.work); 1931 osb = os->os_osb; 1932 1933 mutex_lock(&os->os_lock); 1934 ocfs2_queue_orphan_scan(osb); 1935 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 1936 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 1937 ocfs2_orphan_scan_timeout()); 1938 mutex_unlock(&os->os_lock); 1939} 1940 1941void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 1942{ 1943 struct ocfs2_orphan_scan *os; 1944 1945 os = &osb->osb_orphan_scan; 1946 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1947 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 1948 mutex_lock(&os->os_lock); 1949 cancel_delayed_work(&os->os_orphan_scan_work); 1950 mutex_unlock(&os->os_lock); 1951 } 1952} 1953 1954void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 1955{ 1956 struct ocfs2_orphan_scan *os; 1957 1958 os = &osb->osb_orphan_scan; 1959 os->os_osb = osb; 1960 os->os_count = 0; 1961 os->os_seqno = 0; 1962 mutex_init(&os->os_lock); 1963 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 1964} 1965 1966void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 1967{ 1968 struct ocfs2_orphan_scan *os; 1969 1970 os = &osb->osb_orphan_scan; 1971 os->os_scantime = CURRENT_TIME; 1972 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 1973 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 1974 else { 1975 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 1976 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, 1977 ocfs2_orphan_scan_timeout()); 1978 } 1979} 1980 1981struct ocfs2_orphan_filldir_priv { 1982 struct inode *head; 1983 struct ocfs2_super *osb; 1984}; 1985 1986static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, 1987 loff_t pos, u64 ino, unsigned type) 1988{ 1989 struct ocfs2_orphan_filldir_priv *p = priv; 1990 struct inode *iter; 1991 1992 if (name_len == 1 && !strncmp(".", name, 1)) 1993 return 0; 1994 if (name_len == 2 && !strncmp("..", name, 2)) 1995 return 0; 1996 1997 /* Skip bad inodes so that recovery can continue */ 1998 iter = ocfs2_iget(p->osb, ino, 1999 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 2000 if (IS_ERR(iter)) 2001 return 0; 2002 2003 mlog(0, "queue orphan %llu\n", 2004 (unsigned long long)OCFS2_I(iter)->ip_blkno); 2005 /* No locking is required for the next_orphan queue as there 2006 * is only ever a single process doing orphan recovery. */ 2007 OCFS2_I(iter)->ip_next_orphan = p->head; 2008 p->head = iter; 2009 2010 return 0; 2011} 2012 2013static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2014 int slot, 2015 struct inode **head) 2016{ 2017 int status; 2018 struct inode *orphan_dir_inode = NULL; 2019 struct ocfs2_orphan_filldir_priv priv; 2020 loff_t pos = 0; 2021 2022 priv.osb = osb; 2023 priv.head = *head; 2024 2025 orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2026 ORPHAN_DIR_SYSTEM_INODE, 2027 slot); 2028 if (!orphan_dir_inode) { 2029 status = -ENOENT; 2030 mlog_errno(status); 2031 return status; 2032 } 2033 2034 mutex_lock(&orphan_dir_inode->i_mutex); 2035 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2036 if (status < 0) { 2037 mlog_errno(status); 2038 goto out; 2039 } 2040 2041 status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, 2042 ocfs2_orphan_filldir); 2043 if (status) { 2044 mlog_errno(status); 2045 goto out_cluster; 2046 } 2047 2048 *head = priv.head; 2049 2050out_cluster: 2051 ocfs2_inode_unlock(orphan_dir_inode, 0); 2052out: 2053 mutex_unlock(&orphan_dir_inode->i_mutex); 2054 iput(orphan_dir_inode); 2055 return status; 2056} 2057 2058static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2059 int slot) 2060{ 2061 int ret; 2062 2063 spin_lock(&osb->osb_lock); 2064 ret = !osb->osb_orphan_wipes[slot]; 2065 spin_unlock(&osb->osb_lock); 2066 return ret; 2067} 2068 2069static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2070 int slot) 2071{ 2072 spin_lock(&osb->osb_lock); 2073 /* Mark ourselves such that new processes in delete_inode() 2074 * know to quit early. */ 2075 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2076 while (osb->osb_orphan_wipes[slot]) { 2077 /* If any processes are already in the middle of an 2078 * orphan wipe on this dir, then we need to wait for 2079 * them. */ 2080 spin_unlock(&osb->osb_lock); 2081 wait_event_interruptible(osb->osb_wipe_event, 2082 ocfs2_orphan_recovery_can_continue(osb, slot)); 2083 spin_lock(&osb->osb_lock); 2084 } 2085 spin_unlock(&osb->osb_lock); 2086} 2087 2088static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2089 int slot) 2090{ 2091 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2092} 2093 2094/* 2095 * Orphan recovery. Each mounted node has it's own orphan dir which we 2096 * must run during recovery. Our strategy here is to build a list of 2097 * the inodes in the orphan dir and iget/iput them. The VFS does 2098 * (most) of the rest of the work. 2099 * 2100 * Orphan recovery can happen at any time, not just mount so we have a 2101 * couple of extra considerations. 2102 * 2103 * - We grab as many inodes as we can under the orphan dir lock - 2104 * doing iget() outside the orphan dir risks getting a reference on 2105 * an invalid inode. 2106 * - We must be sure not to deadlock with other processes on the 2107 * system wanting to run delete_inode(). This can happen when they go 2108 * to lock the orphan dir and the orphan recovery process attempts to 2109 * iget() inside the orphan dir lock. This can be avoided by 2110 * advertising our state to ocfs2_delete_inode(). 2111 */ 2112static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2113 int slot) 2114{ 2115 int ret = 0; 2116 struct inode *inode = NULL; 2117 struct inode *iter; 2118 struct ocfs2_inode_info *oi; 2119 2120 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); 2121 2122 ocfs2_mark_recovering_orphan_dir(osb, slot); 2123 ret = ocfs2_queue_orphans(osb, slot, &inode); 2124 ocfs2_clear_recovering_orphan_dir(osb, slot); 2125 2126 /* Error here should be noted, but we want to continue with as 2127 * many queued inodes as we've got. */ 2128 if (ret) 2129 mlog_errno(ret); 2130 2131 while (inode) { 2132 oi = OCFS2_I(inode); 2133 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); 2134 2135 iter = oi->ip_next_orphan; 2136 2137 spin_lock(&oi->ip_lock); 2138 /* The remote delete code may have set these on the 2139 * assumption that the other node would wipe them 2140 * successfully. If they are still in the node's 2141 * orphan dir, we need to reset that state. */ 2142 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); 2143 2144 /* Set the proper information to get us going into 2145 * ocfs2_delete_inode. */ 2146 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 2147 spin_unlock(&oi->ip_lock); 2148 2149 iput(inode); 2150 2151 inode = iter; 2152 } 2153 2154 return ret; 2155} 2156 2157static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2158{ 2159 /* This check is good because ocfs2 will wait on our recovery 2160 * thread before changing it to something other than MOUNTED 2161 * or DISABLED. */ 2162 wait_event(osb->osb_mount_event, 2163 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 2164 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2165 atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2166 2167 /* If there's an error on mount, then we may never get to the 2168 * MOUNTED flag, but this is set right before 2169 * dismount_volume() so we can trust it. */ 2170 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2171 mlog(0, "mount error, exiting!\n"); 2172 return -EBUSY; 2173 } 2174 2175 return 0; 2176} 2177 2178static int ocfs2_commit_thread(void *arg) 2179{ 2180 int status; 2181 struct ocfs2_super *osb = arg; 2182 struct ocfs2_journal *journal = osb->journal; 2183 2184 /* we can trust j_num_trans here because _should_stop() is only set in 2185 * shutdown and nobody other than ourselves should be able to start 2186 * transactions. committing on shutdown might take a few iterations 2187 * as final transactions put deleted inodes on the list */ 2188 while (!(kthread_should_stop() && 2189 atomic_read(&journal->j_num_trans) == 0)) { 2190 2191 wait_event_interruptible(osb->checkpoint_event, 2192 atomic_read(&journal->j_num_trans) 2193 || kthread_should_stop()); 2194 2195 status = ocfs2_commit_cache(osb); 2196 if (status < 0) 2197 mlog_errno(status); 2198 2199 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2200 mlog(ML_KTHREAD, 2201 "commit_thread: %u transactions pending on " 2202 "shutdown\n", 2203 atomic_read(&journal->j_num_trans)); 2204 } 2205 } 2206 2207 return 0; 2208} 2209 2210/* Reads all the journal inodes without taking any cluster locks. Used 2211 * for hard readonly access to determine whether any journal requires 2212 * recovery. Also used to refresh the recovery generation numbers after 2213 * a journal has been recovered by another node. 2214 */ 2215int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2216{ 2217 int ret = 0; 2218 unsigned int slot; 2219 struct buffer_head *di_bh = NULL; 2220 struct ocfs2_dinode *di; 2221 int journal_dirty = 0; 2222 2223 for(slot = 0; slot < osb->max_slots; slot++) { 2224 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2225 if (ret) { 2226 mlog_errno(ret); 2227 goto out; 2228 } 2229 2230 di = (struct ocfs2_dinode *) di_bh->b_data; 2231 2232 osb->slot_recovery_generations[slot] = 2233 ocfs2_get_recovery_generation(di); 2234 2235 if (le32_to_cpu(di->id1.journal1.ij_flags) & 2236 OCFS2_JOURNAL_DIRTY_FL) 2237 journal_dirty = 1; 2238 2239 brelse(di_bh); 2240 di_bh = NULL; 2241 } 2242 2243out: 2244 if (journal_dirty) 2245 ret = -EROFS; 2246 return ret; 2247} 2248