1// SPDX-License-Identifier: GPL-2.0-or-later 2/* -*- mode: c; c-basic-offset: 8; -*- 3 * vim: noexpandtab sw=8 ts=8 sts=0: 4 * 5 * journal.c 6 * 7 * Defines functions of journalling api 8 * 9 * Copyright (C) 2003, 2004 Oracle. All rights reserved. 10 */ 11 12#include <linux/fs.h> 13#include <linux/types.h> 14#include <linux/slab.h> 15#include <linux/highmem.h> 16#include <linux/kthread.h> 17#include <linux/time.h> 18#include <linux/random.h> 19#include <linux/delay.h> 20 21#include <cluster/masklog.h> 22 23#include "ocfs2.h" 24 25#include "alloc.h" 26#include "blockcheck.h" 27#include "dir.h" 28#include "dlmglue.h" 29#include "extent_map.h" 30#include "heartbeat.h" 31#include "inode.h" 32#include "journal.h" 33#include "localalloc.h" 34#include "slot_map.h" 35#include "super.h" 36#include "sysfile.h" 37#include "uptodate.h" 38#include "quota.h" 39#include "file.h" 40#include "namei.h" 41 42#include "buffer_head_io.h" 43#include "ocfs2_trace.h" 44 45DEFINE_SPINLOCK(trans_inc_lock); 46 47#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 48 49static int ocfs2_force_read_journal(struct inode *inode); 50static int ocfs2_recover_node(struct ocfs2_super *osb, 51 int node_num, int slot_num); 52static int __ocfs2_recovery_thread(void *arg); 53static int ocfs2_commit_cache(struct ocfs2_super *osb); 54static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); 55static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 56 int dirty, int replayed); 57static int ocfs2_trylock_journal(struct ocfs2_super *osb, 58 int slot_num); 59static int ocfs2_recover_orphans(struct ocfs2_super *osb, 60 int slot, 61 enum ocfs2_orphan_reco_type orphan_reco_type); 62static int ocfs2_commit_thread(void *arg); 63static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 64 int slot_num, 65 struct ocfs2_dinode *la_dinode, 66 struct ocfs2_dinode *tl_dinode, 67 struct ocfs2_quota_recovery *qrec, 68 enum ocfs2_orphan_reco_type orphan_reco_type); 69 70static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) 71{ 72 return __ocfs2_wait_on_mount(osb, 0); 73} 74 75static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) 76{ 77 return __ocfs2_wait_on_mount(osb, 1); 78} 79 80/* 81 * This replay_map is to track online/offline slots, so we could recover 82 * offline slots during recovery and mount 83 */ 84 85enum ocfs2_replay_state { 86 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ 87 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ 88 REPLAY_DONE /* Replay was already queued */ 89}; 90 91struct ocfs2_replay_map { 92 unsigned int rm_slots; 93 enum ocfs2_replay_state rm_state; 94 unsigned char rm_replay_slots[]; 95}; 96 97static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) 98{ 99 if (!osb->replay_map) 100 return; 101 102 /* If we've already queued the replay, we don't have any more to do */ 103 if (osb->replay_map->rm_state == REPLAY_DONE) 104 return; 105 106 osb->replay_map->rm_state = state; 107} 108 109int ocfs2_compute_replay_slots(struct ocfs2_super *osb) 110{ 111 struct ocfs2_replay_map *replay_map; 112 int i, node_num; 113 114 /* If replay map is already set, we don't do it again */ 115 if (osb->replay_map) 116 return 0; 117 118 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + 119 (osb->max_slots * sizeof(char)), GFP_KERNEL); 120 121 if (!replay_map) { 122 mlog_errno(-ENOMEM); 123 return -ENOMEM; 124 } 125 126 spin_lock(&osb->osb_lock); 127 128 replay_map->rm_slots = osb->max_slots; 129 replay_map->rm_state = REPLAY_UNNEEDED; 130 131 /* set rm_replay_slots for offline slot(s) */ 132 for (i = 0; i < replay_map->rm_slots; i++) { 133 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) 134 replay_map->rm_replay_slots[i] = 1; 135 } 136 137 osb->replay_map = replay_map; 138 spin_unlock(&osb->osb_lock); 139 return 0; 140} 141 142static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, 143 enum ocfs2_orphan_reco_type orphan_reco_type) 144{ 145 struct ocfs2_replay_map *replay_map = osb->replay_map; 146 int i; 147 148 if (!replay_map) 149 return; 150 151 if (replay_map->rm_state != REPLAY_NEEDED) 152 return; 153 154 for (i = 0; i < replay_map->rm_slots; i++) 155 if (replay_map->rm_replay_slots[i]) 156 ocfs2_queue_recovery_completion(osb->journal, i, NULL, 157 NULL, NULL, 158 orphan_reco_type); 159 replay_map->rm_state = REPLAY_DONE; 160} 161 162void ocfs2_free_replay_slots(struct ocfs2_super *osb) 163{ 164 struct ocfs2_replay_map *replay_map = osb->replay_map; 165 166 if (!osb->replay_map) 167 return; 168 169 kfree(replay_map); 170 osb->replay_map = NULL; 171} 172 173int ocfs2_recovery_init(struct ocfs2_super *osb) 174{ 175 struct ocfs2_recovery_map *rm; 176 177 mutex_init(&osb->recovery_lock); 178 osb->disable_recovery = 0; 179 osb->recovery_thread_task = NULL; 180 init_waitqueue_head(&osb->recovery_event); 181 182 rm = kzalloc(sizeof(struct ocfs2_recovery_map) + 183 osb->max_slots * sizeof(unsigned int), 184 GFP_KERNEL); 185 if (!rm) { 186 mlog_errno(-ENOMEM); 187 return -ENOMEM; 188 } 189 190 rm->rm_entries = (unsigned int *)((char *)rm + 191 sizeof(struct ocfs2_recovery_map)); 192 osb->recovery_map = rm; 193 194 return 0; 195} 196 197/* we can't grab the goofy sem lock from inside wait_event, so we use 198 * memory barriers to make sure that we'll see the null task before 199 * being woken up */ 200static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) 201{ 202 mb(); 203 return osb->recovery_thread_task != NULL; 204} 205 206void ocfs2_recovery_exit(struct ocfs2_super *osb) 207{ 208 struct ocfs2_recovery_map *rm; 209 210 /* disable any new recovery threads and wait for any currently 211 * running ones to exit. Do this before setting the vol_state. */ 212 mutex_lock(&osb->recovery_lock); 213 osb->disable_recovery = 1; 214 mutex_unlock(&osb->recovery_lock); 215 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); 216 217 /* At this point, we know that no more recovery threads can be 218 * launched, so wait for any recovery completion work to 219 * complete. */ 220 if (osb->ocfs2_wq) 221 flush_workqueue(osb->ocfs2_wq); 222 223 /* 224 * Now that recovery is shut down, and the osb is about to be 225 * freed, the osb_lock is not taken here. 226 */ 227 rm = osb->recovery_map; 228 /* XXX: Should we bug if there are dirty entries? */ 229 230 kfree(rm); 231} 232 233static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, 234 unsigned int node_num) 235{ 236 int i; 237 struct ocfs2_recovery_map *rm = osb->recovery_map; 238 239 assert_spin_locked(&osb->osb_lock); 240 241 for (i = 0; i < rm->rm_used; i++) { 242 if (rm->rm_entries[i] == node_num) 243 return 1; 244 } 245 246 return 0; 247} 248 249/* Behaves like test-and-set. Returns the previous value */ 250static int ocfs2_recovery_map_set(struct ocfs2_super *osb, 251 unsigned int node_num) 252{ 253 struct ocfs2_recovery_map *rm = osb->recovery_map; 254 255 spin_lock(&osb->osb_lock); 256 if (__ocfs2_recovery_map_test(osb, node_num)) { 257 spin_unlock(&osb->osb_lock); 258 return 1; 259 } 260 261 /* XXX: Can this be exploited? Not from o2dlm... */ 262 BUG_ON(rm->rm_used >= osb->max_slots); 263 264 rm->rm_entries[rm->rm_used] = node_num; 265 rm->rm_used++; 266 spin_unlock(&osb->osb_lock); 267 268 return 0; 269} 270 271static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, 272 unsigned int node_num) 273{ 274 int i; 275 struct ocfs2_recovery_map *rm = osb->recovery_map; 276 277 spin_lock(&osb->osb_lock); 278 279 for (i = 0; i < rm->rm_used; i++) { 280 if (rm->rm_entries[i] == node_num) 281 break; 282 } 283 284 if (i < rm->rm_used) { 285 /* XXX: be careful with the pointer math */ 286 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), 287 (rm->rm_used - i - 1) * sizeof(unsigned int)); 288 rm->rm_used--; 289 } 290 291 spin_unlock(&osb->osb_lock); 292} 293 294static int ocfs2_commit_cache(struct ocfs2_super *osb) 295{ 296 int status = 0; 297 unsigned int flushed; 298 struct ocfs2_journal *journal = NULL; 299 300 journal = osb->journal; 301 302 /* Flush all pending commits and checkpoint the journal. */ 303 down_write(&journal->j_trans_barrier); 304 305 flushed = atomic_read(&journal->j_num_trans); 306 trace_ocfs2_commit_cache_begin(flushed); 307 if (flushed == 0) { 308 up_write(&journal->j_trans_barrier); 309 goto finally; 310 } 311 312 jbd2_journal_lock_updates(journal->j_journal); 313 status = jbd2_journal_flush(journal->j_journal); 314 jbd2_journal_unlock_updates(journal->j_journal); 315 if (status < 0) { 316 up_write(&journal->j_trans_barrier); 317 mlog_errno(status); 318 goto finally; 319 } 320 321 ocfs2_inc_trans_id(journal); 322 323 flushed = atomic_read(&journal->j_num_trans); 324 atomic_set(&journal->j_num_trans, 0); 325 up_write(&journal->j_trans_barrier); 326 327 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); 328 329 ocfs2_wake_downconvert_thread(osb); 330 wake_up(&journal->j_checkpointed); 331finally: 332 return status; 333} 334 335handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) 336{ 337 journal_t *journal = osb->journal->j_journal; 338 handle_t *handle; 339 340 BUG_ON(!osb || !osb->journal->j_journal); 341 342 if (ocfs2_is_hard_readonly(osb)) 343 return ERR_PTR(-EROFS); 344 345 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); 346 BUG_ON(max_buffs <= 0); 347 348 /* Nested transaction? Just return the handle... */ 349 if (journal_current_handle()) 350 return jbd2_journal_start(journal, max_buffs); 351 352 sb_start_intwrite(osb->sb); 353 354 down_read(&osb->journal->j_trans_barrier); 355 356 handle = jbd2_journal_start(journal, max_buffs); 357 if (IS_ERR(handle)) { 358 up_read(&osb->journal->j_trans_barrier); 359 sb_end_intwrite(osb->sb); 360 361 mlog_errno(PTR_ERR(handle)); 362 363 if (is_journal_aborted(journal)) { 364 ocfs2_abort(osb->sb, "Detected aborted journal\n"); 365 handle = ERR_PTR(-EROFS); 366 } 367 } else { 368 if (!ocfs2_mount_local(osb)) 369 atomic_inc(&(osb->journal->j_num_trans)); 370 } 371 372 return handle; 373} 374 375int ocfs2_commit_trans(struct ocfs2_super *osb, 376 handle_t *handle) 377{ 378 int ret, nested; 379 struct ocfs2_journal *journal = osb->journal; 380 381 BUG_ON(!handle); 382 383 nested = handle->h_ref > 1; 384 ret = jbd2_journal_stop(handle); 385 if (ret < 0) 386 mlog_errno(ret); 387 388 if (!nested) { 389 up_read(&journal->j_trans_barrier); 390 sb_end_intwrite(osb->sb); 391 } 392 393 return ret; 394} 395 396/* 397 * 'nblocks' is what you want to add to the current transaction. 398 * 399 * This might call jbd2_journal_restart() which will commit dirty buffers 400 * and then restart the transaction. Before calling 401 * ocfs2_extend_trans(), any changed blocks should have been 402 * dirtied. After calling it, all blocks which need to be changed must 403 * go through another set of journal_access/journal_dirty calls. 404 * 405 * WARNING: This will not release any semaphores or disk locks taken 406 * during the transaction, so make sure they were taken *before* 407 * start_trans or we'll have ordering deadlocks. 408 * 409 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is 410 * good because transaction ids haven't yet been recorded on the 411 * cluster locks associated with this handle. 412 */ 413int ocfs2_extend_trans(handle_t *handle, int nblocks) 414{ 415 int status, old_nblocks; 416 417 BUG_ON(!handle); 418 BUG_ON(nblocks < 0); 419 420 if (!nblocks) 421 return 0; 422 423 old_nblocks = jbd2_handle_buffer_credits(handle); 424 425 trace_ocfs2_extend_trans(old_nblocks, nblocks); 426 427#ifdef CONFIG_OCFS2_DEBUG_FS 428 status = 1; 429#else 430 status = jbd2_journal_extend(handle, nblocks, 0); 431 if (status < 0) { 432 mlog_errno(status); 433 goto bail; 434 } 435#endif 436 437 if (status > 0) { 438 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); 439 status = jbd2_journal_restart(handle, 440 old_nblocks + nblocks); 441 if (status < 0) { 442 mlog_errno(status); 443 goto bail; 444 } 445 } 446 447 status = 0; 448bail: 449 return status; 450} 451 452/* 453 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. 454 * If that fails, restart the transaction & regain write access for the 455 * buffer head which is used for metadata modifications. 456 * Taken from Ext4: extend_or_restart_transaction() 457 */ 458int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) 459{ 460 int status, old_nblks; 461 462 BUG_ON(!handle); 463 464 old_nblks = jbd2_handle_buffer_credits(handle); 465 trace_ocfs2_allocate_extend_trans(old_nblks, thresh); 466 467 if (old_nblks < thresh) 468 return 0; 469 470 status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0); 471 if (status < 0) { 472 mlog_errno(status); 473 goto bail; 474 } 475 476 if (status > 0) { 477 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); 478 if (status < 0) 479 mlog_errno(status); 480 } 481 482bail: 483 return status; 484} 485 486 487struct ocfs2_triggers { 488 struct jbd2_buffer_trigger_type ot_triggers; 489 int ot_offset; 490}; 491 492static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) 493{ 494 return container_of(triggers, struct ocfs2_triggers, ot_triggers); 495} 496 497static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 498 struct buffer_head *bh, 499 void *data, size_t size) 500{ 501 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); 502 503 /* 504 * We aren't guaranteed to have the superblock here, so we 505 * must unconditionally compute the ecc data. 506 * __ocfs2_journal_access() will only set the triggers if 507 * metaecc is enabled. 508 */ 509 ocfs2_block_check_compute(data, size, data + ot->ot_offset); 510} 511 512/* 513 * Quota blocks have their own trigger because the struct ocfs2_block_check 514 * offset depends on the blocksize. 515 */ 516static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 517 struct buffer_head *bh, 518 void *data, size_t size) 519{ 520 struct ocfs2_disk_dqtrailer *dqt = 521 ocfs2_block_dqtrailer(size, data); 522 523 /* 524 * We aren't guaranteed to have the superblock here, so we 525 * must unconditionally compute the ecc data. 526 * __ocfs2_journal_access() will only set the triggers if 527 * metaecc is enabled. 528 */ 529 ocfs2_block_check_compute(data, size, &dqt->dq_check); 530} 531 532/* 533 * Directory blocks also have their own trigger because the 534 * struct ocfs2_block_check offset depends on the blocksize. 535 */ 536static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, 537 struct buffer_head *bh, 538 void *data, size_t size) 539{ 540 struct ocfs2_dir_block_trailer *trailer = 541 ocfs2_dir_trailer_from_size(size, data); 542 543 /* 544 * We aren't guaranteed to have the superblock here, so we 545 * must unconditionally compute the ecc data. 546 * __ocfs2_journal_access() will only set the triggers if 547 * metaecc is enabled. 548 */ 549 ocfs2_block_check_compute(data, size, &trailer->db_check); 550} 551 552static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, 553 struct buffer_head *bh) 554{ 555 mlog(ML_ERROR, 556 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " 557 "bh->b_blocknr = %llu\n", 558 (unsigned long)bh, 559 (unsigned long long)bh->b_blocknr); 560 561 ocfs2_error(bh->b_bdev->bd_super, 562 "JBD2 has aborted our journal, ocfs2 cannot continue\n"); 563} 564 565static struct ocfs2_triggers di_triggers = { 566 .ot_triggers = { 567 .t_frozen = ocfs2_frozen_trigger, 568 .t_abort = ocfs2_abort_trigger, 569 }, 570 .ot_offset = offsetof(struct ocfs2_dinode, i_check), 571}; 572 573static struct ocfs2_triggers eb_triggers = { 574 .ot_triggers = { 575 .t_frozen = ocfs2_frozen_trigger, 576 .t_abort = ocfs2_abort_trigger, 577 }, 578 .ot_offset = offsetof(struct ocfs2_extent_block, h_check), 579}; 580 581static struct ocfs2_triggers rb_triggers = { 582 .ot_triggers = { 583 .t_frozen = ocfs2_frozen_trigger, 584 .t_abort = ocfs2_abort_trigger, 585 }, 586 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), 587}; 588 589static struct ocfs2_triggers gd_triggers = { 590 .ot_triggers = { 591 .t_frozen = ocfs2_frozen_trigger, 592 .t_abort = ocfs2_abort_trigger, 593 }, 594 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), 595}; 596 597static struct ocfs2_triggers db_triggers = { 598 .ot_triggers = { 599 .t_frozen = ocfs2_db_frozen_trigger, 600 .t_abort = ocfs2_abort_trigger, 601 }, 602}; 603 604static struct ocfs2_triggers xb_triggers = { 605 .ot_triggers = { 606 .t_frozen = ocfs2_frozen_trigger, 607 .t_abort = ocfs2_abort_trigger, 608 }, 609 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), 610}; 611 612static struct ocfs2_triggers dq_triggers = { 613 .ot_triggers = { 614 .t_frozen = ocfs2_dq_frozen_trigger, 615 .t_abort = ocfs2_abort_trigger, 616 }, 617}; 618 619static struct ocfs2_triggers dr_triggers = { 620 .ot_triggers = { 621 .t_frozen = ocfs2_frozen_trigger, 622 .t_abort = ocfs2_abort_trigger, 623 }, 624 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), 625}; 626 627static struct ocfs2_triggers dl_triggers = { 628 .ot_triggers = { 629 .t_frozen = ocfs2_frozen_trigger, 630 .t_abort = ocfs2_abort_trigger, 631 }, 632 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), 633}; 634 635static int __ocfs2_journal_access(handle_t *handle, 636 struct ocfs2_caching_info *ci, 637 struct buffer_head *bh, 638 struct ocfs2_triggers *triggers, 639 int type) 640{ 641 int status; 642 struct ocfs2_super *osb = 643 OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); 644 645 BUG_ON(!ci || !ci->ci_ops); 646 BUG_ON(!handle); 647 BUG_ON(!bh); 648 649 trace_ocfs2_journal_access( 650 (unsigned long long)ocfs2_metadata_cache_owner(ci), 651 (unsigned long long)bh->b_blocknr, type, bh->b_size); 652 653 /* we can safely remove this assertion after testing. */ 654 if (!buffer_uptodate(bh)) { 655 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); 656 mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", 657 (unsigned long long)bh->b_blocknr, bh->b_state); 658 659 lock_buffer(bh); 660 /* 661 * A previous transaction with a couple of buffer heads fail 662 * to checkpoint, so all the bhs are marked as BH_Write_EIO. 663 * For current transaction, the bh is just among those error 664 * bhs which previous transaction handle. We can't just clear 665 * its BH_Write_EIO and reuse directly, since other bhs are 666 * not written to disk yet and that will cause metadata 667 * inconsistency. So we should set fs read-only to avoid 668 * further damage. 669 */ 670 if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { 671 unlock_buffer(bh); 672 return ocfs2_error(osb->sb, "A previous attempt to " 673 "write this buffer head failed\n"); 674 } 675 unlock_buffer(bh); 676 } 677 678 /* Set the current transaction information on the ci so 679 * that the locking code knows whether it can drop it's locks 680 * on this ci or not. We're protected from the commit 681 * thread updating the current transaction id until 682 * ocfs2_commit_trans() because ocfs2_start_trans() took 683 * j_trans_barrier for us. */ 684 ocfs2_set_ci_lock_trans(osb->journal, ci); 685 686 ocfs2_metadata_cache_io_lock(ci); 687 switch (type) { 688 case OCFS2_JOURNAL_ACCESS_CREATE: 689 case OCFS2_JOURNAL_ACCESS_WRITE: 690 status = jbd2_journal_get_write_access(handle, bh); 691 break; 692 693 case OCFS2_JOURNAL_ACCESS_UNDO: 694 status = jbd2_journal_get_undo_access(handle, bh); 695 break; 696 697 default: 698 status = -EINVAL; 699 mlog(ML_ERROR, "Unknown access type!\n"); 700 } 701 if (!status && ocfs2_meta_ecc(osb) && triggers) 702 jbd2_journal_set_triggers(bh, &triggers->ot_triggers); 703 ocfs2_metadata_cache_io_unlock(ci); 704 705 if (status < 0) 706 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", 707 status, type); 708 709 return status; 710} 711 712int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, 713 struct buffer_head *bh, int type) 714{ 715 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); 716} 717 718int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, 719 struct buffer_head *bh, int type) 720{ 721 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); 722} 723 724int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, 725 struct buffer_head *bh, int type) 726{ 727 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, 728 type); 729} 730 731int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, 732 struct buffer_head *bh, int type) 733{ 734 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); 735} 736 737int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, 738 struct buffer_head *bh, int type) 739{ 740 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); 741} 742 743int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, 744 struct buffer_head *bh, int type) 745{ 746 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); 747} 748 749int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, 750 struct buffer_head *bh, int type) 751{ 752 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); 753} 754 755int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, 756 struct buffer_head *bh, int type) 757{ 758 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); 759} 760 761int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, 762 struct buffer_head *bh, int type) 763{ 764 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); 765} 766 767int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, 768 struct buffer_head *bh, int type) 769{ 770 return __ocfs2_journal_access(handle, ci, bh, NULL, type); 771} 772 773void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) 774{ 775 int status; 776 777 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); 778 779 status = jbd2_journal_dirty_metadata(handle, bh); 780 if (status) { 781 mlog_errno(status); 782 if (!is_handle_aborted(handle)) { 783 journal_t *journal = handle->h_transaction->t_journal; 784 struct super_block *sb = bh->b_bdev->bd_super; 785 786 mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " 787 "Aborting transaction and journal.\n"); 788 handle->h_err = status; 789 jbd2_journal_abort_handle(handle); 790 jbd2_journal_abort(journal, status); 791 ocfs2_abort(sb, "Journal already aborted.\n"); 792 } 793 } 794} 795 796#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) 797 798void ocfs2_set_journal_params(struct ocfs2_super *osb) 799{ 800 journal_t *journal = osb->journal->j_journal; 801 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; 802 803 if (osb->osb_commit_interval) 804 commit_interval = osb->osb_commit_interval; 805 806 write_lock(&journal->j_state_lock); 807 journal->j_commit_interval = commit_interval; 808 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) 809 journal->j_flags |= JBD2_BARRIER; 810 else 811 journal->j_flags &= ~JBD2_BARRIER; 812 write_unlock(&journal->j_state_lock); 813} 814 815int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) 816{ 817 int status = -1; 818 struct inode *inode = NULL; /* the journal inode */ 819 journal_t *j_journal = NULL; 820 struct ocfs2_dinode *di = NULL; 821 struct buffer_head *bh = NULL; 822 struct ocfs2_super *osb; 823 int inode_lock = 0; 824 825 BUG_ON(!journal); 826 827 osb = journal->j_osb; 828 829 /* already have the inode for our journal */ 830 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 831 osb->slot_num); 832 if (inode == NULL) { 833 status = -EACCES; 834 mlog_errno(status); 835 goto done; 836 } 837 if (is_bad_inode(inode)) { 838 mlog(ML_ERROR, "access error (bad inode)\n"); 839 iput(inode); 840 inode = NULL; 841 status = -EACCES; 842 goto done; 843 } 844 845 SET_INODE_JOURNAL(inode); 846 OCFS2_I(inode)->ip_open_count++; 847 848 /* Skip recovery waits here - journal inode metadata never 849 * changes in a live cluster so it can be considered an 850 * exception to the rule. */ 851 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 852 if (status < 0) { 853 if (status != -ERESTARTSYS) 854 mlog(ML_ERROR, "Could not get lock on journal!\n"); 855 goto done; 856 } 857 858 inode_lock = 1; 859 di = (struct ocfs2_dinode *)bh->b_data; 860 861 if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { 862 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", 863 i_size_read(inode)); 864 status = -EINVAL; 865 goto done; 866 } 867 868 trace_ocfs2_journal_init(i_size_read(inode), 869 (unsigned long long)inode->i_blocks, 870 OCFS2_I(inode)->ip_clusters); 871 872 /* call the kernels journal init function now */ 873 j_journal = jbd2_journal_init_inode(inode); 874 if (j_journal == NULL) { 875 mlog(ML_ERROR, "Linux journal layer error\n"); 876 status = -EINVAL; 877 goto done; 878 } 879 880 trace_ocfs2_journal_init_maxlen(j_journal->j_total_len); 881 882 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & 883 OCFS2_JOURNAL_DIRTY_FL); 884 885 journal->j_journal = j_journal; 886 journal->j_journal->j_submit_inode_data_buffers = 887 jbd2_journal_submit_inode_data_buffers; 888 journal->j_journal->j_finish_inode_data_buffers = 889 jbd2_journal_finish_inode_data_buffers; 890 journal->j_inode = inode; 891 journal->j_bh = bh; 892 893 ocfs2_set_journal_params(osb); 894 895 journal->j_state = OCFS2_JOURNAL_LOADED; 896 897 status = 0; 898done: 899 if (status < 0) { 900 if (inode_lock) 901 ocfs2_inode_unlock(inode, 1); 902 brelse(bh); 903 if (inode) { 904 OCFS2_I(inode)->ip_open_count--; 905 iput(inode); 906 } 907 } 908 909 return status; 910} 911 912static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) 913{ 914 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); 915} 916 917static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) 918{ 919 return le32_to_cpu(di->id1.journal1.ij_recovery_generation); 920} 921 922static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, 923 int dirty, int replayed) 924{ 925 int status; 926 unsigned int flags; 927 struct ocfs2_journal *journal = osb->journal; 928 struct buffer_head *bh = journal->j_bh; 929 struct ocfs2_dinode *fe; 930 931 fe = (struct ocfs2_dinode *)bh->b_data; 932 933 /* The journal bh on the osb always comes from ocfs2_journal_init() 934 * and was validated there inside ocfs2_inode_lock_full(). It's a 935 * code bug if we mess it up. */ 936 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); 937 938 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 939 if (dirty) 940 flags |= OCFS2_JOURNAL_DIRTY_FL; 941 else 942 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 943 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 944 945 if (replayed) 946 ocfs2_bump_recovery_generation(fe); 947 948 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 949 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); 950 if (status < 0) 951 mlog_errno(status); 952 953 return status; 954} 955 956/* 957 * If the journal has been kmalloc'd it needs to be freed after this 958 * call. 959 */ 960void ocfs2_journal_shutdown(struct ocfs2_super *osb) 961{ 962 struct ocfs2_journal *journal = NULL; 963 int status = 0; 964 struct inode *inode = NULL; 965 int num_running_trans = 0; 966 967 BUG_ON(!osb); 968 969 journal = osb->journal; 970 if (!journal) 971 goto done; 972 973 inode = journal->j_inode; 974 975 if (journal->j_state != OCFS2_JOURNAL_LOADED) 976 goto done; 977 978 /* need to inc inode use count - jbd2_journal_destroy will iput. */ 979 if (!igrab(inode)) 980 BUG(); 981 982 num_running_trans = atomic_read(&(osb->journal->j_num_trans)); 983 trace_ocfs2_journal_shutdown(num_running_trans); 984 985 /* Do a commit_cache here. It will flush our journal, *and* 986 * release any locks that are still held. 987 * set the SHUTDOWN flag and release the trans lock. 988 * the commit thread will take the trans lock for us below. */ 989 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; 990 991 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not 992 * drop the trans_lock (which we want to hold until we 993 * completely destroy the journal. */ 994 if (osb->commit_task) { 995 /* Wait for the commit thread */ 996 trace_ocfs2_journal_shutdown_wait(osb->commit_task); 997 kthread_stop(osb->commit_task); 998 osb->commit_task = NULL; 999 } 1000 1001 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); 1002 1003 if (ocfs2_mount_local(osb)) { 1004 jbd2_journal_lock_updates(journal->j_journal); 1005 status = jbd2_journal_flush(journal->j_journal); 1006 jbd2_journal_unlock_updates(journal->j_journal); 1007 if (status < 0) 1008 mlog_errno(status); 1009 } 1010 1011 /* Shutdown the kernel journal system */ 1012 if (!jbd2_journal_destroy(journal->j_journal) && !status) { 1013 /* 1014 * Do not toggle if flush was unsuccessful otherwise 1015 * will leave dirty metadata in a "clean" journal 1016 */ 1017 status = ocfs2_journal_toggle_dirty(osb, 0, 0); 1018 if (status < 0) 1019 mlog_errno(status); 1020 } 1021 journal->j_journal = NULL; 1022 1023 OCFS2_I(inode)->ip_open_count--; 1024 1025 /* unlock our journal */ 1026 ocfs2_inode_unlock(inode, 1); 1027 1028 brelse(journal->j_bh); 1029 journal->j_bh = NULL; 1030 1031 journal->j_state = OCFS2_JOURNAL_FREE; 1032 1033// up_write(&journal->j_trans_barrier); 1034done: 1035 iput(inode); 1036} 1037 1038static void ocfs2_clear_journal_error(struct super_block *sb, 1039 journal_t *journal, 1040 int slot) 1041{ 1042 int olderr; 1043 1044 olderr = jbd2_journal_errno(journal); 1045 if (olderr) { 1046 mlog(ML_ERROR, "File system error %d recorded in " 1047 "journal %u.\n", olderr, slot); 1048 mlog(ML_ERROR, "File system on device %s needs checking.\n", 1049 sb->s_id); 1050 1051 jbd2_journal_ack_err(journal); 1052 jbd2_journal_clear_err(journal); 1053 } 1054} 1055 1056int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) 1057{ 1058 int status = 0; 1059 struct ocfs2_super *osb; 1060 1061 BUG_ON(!journal); 1062 1063 osb = journal->j_osb; 1064 1065 status = jbd2_journal_load(journal->j_journal); 1066 if (status < 0) { 1067 mlog(ML_ERROR, "Failed to load journal!\n"); 1068 goto done; 1069 } 1070 1071 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); 1072 1073 if (replayed) { 1074 jbd2_journal_lock_updates(journal->j_journal); 1075 status = jbd2_journal_flush(journal->j_journal); 1076 jbd2_journal_unlock_updates(journal->j_journal); 1077 if (status < 0) 1078 mlog_errno(status); 1079 } 1080 1081 status = ocfs2_journal_toggle_dirty(osb, 1, replayed); 1082 if (status < 0) { 1083 mlog_errno(status); 1084 goto done; 1085 } 1086 1087 /* Launch the commit thread */ 1088 if (!local) { 1089 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, 1090 "ocfs2cmt-%s", osb->uuid_str); 1091 if (IS_ERR(osb->commit_task)) { 1092 status = PTR_ERR(osb->commit_task); 1093 osb->commit_task = NULL; 1094 mlog(ML_ERROR, "unable to launch ocfs2commit thread, " 1095 "error=%d", status); 1096 goto done; 1097 } 1098 } else 1099 osb->commit_task = NULL; 1100 1101done: 1102 return status; 1103} 1104 1105 1106/* 'full' flag tells us whether we clear out all blocks or if we just 1107 * mark the journal clean */ 1108int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) 1109{ 1110 int status; 1111 1112 BUG_ON(!journal); 1113 1114 status = jbd2_journal_wipe(journal->j_journal, full); 1115 if (status < 0) { 1116 mlog_errno(status); 1117 goto bail; 1118 } 1119 1120 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); 1121 if (status < 0) 1122 mlog_errno(status); 1123 1124bail: 1125 return status; 1126} 1127 1128static int ocfs2_recovery_completed(struct ocfs2_super *osb) 1129{ 1130 int empty; 1131 struct ocfs2_recovery_map *rm = osb->recovery_map; 1132 1133 spin_lock(&osb->osb_lock); 1134 empty = (rm->rm_used == 0); 1135 spin_unlock(&osb->osb_lock); 1136 1137 return empty; 1138} 1139 1140void ocfs2_wait_for_recovery(struct ocfs2_super *osb) 1141{ 1142 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); 1143} 1144 1145/* 1146 * JBD Might read a cached version of another nodes journal file. We 1147 * don't want this as this file changes often and we get no 1148 * notification on those changes. The only way to be sure that we've 1149 * got the most up to date version of those blocks then is to force 1150 * read them off disk. Just searching through the buffer cache won't 1151 * work as there may be pages backing this file which are still marked 1152 * up to date. We know things can't change on this file underneath us 1153 * as we have the lock by now :) 1154 */ 1155static int ocfs2_force_read_journal(struct inode *inode) 1156{ 1157 int status = 0; 1158 int i; 1159 u64 v_blkno, p_blkno, p_blocks, num_blocks; 1160 struct buffer_head *bh = NULL; 1161 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1162 1163 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 1164 v_blkno = 0; 1165 while (v_blkno < num_blocks) { 1166 status = ocfs2_extent_map_get_blocks(inode, v_blkno, 1167 &p_blkno, &p_blocks, NULL); 1168 if (status < 0) { 1169 mlog_errno(status); 1170 goto bail; 1171 } 1172 1173 for (i = 0; i < p_blocks; i++, p_blkno++) { 1174 bh = __find_get_block(osb->sb->s_bdev, p_blkno, 1175 osb->sb->s_blocksize); 1176 /* block not cached. */ 1177 if (!bh) 1178 continue; 1179 1180 brelse(bh); 1181 bh = NULL; 1182 /* We are reading journal data which should not 1183 * be put in the uptodate cache. 1184 */ 1185 status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); 1186 if (status < 0) { 1187 mlog_errno(status); 1188 goto bail; 1189 } 1190 1191 brelse(bh); 1192 bh = NULL; 1193 } 1194 1195 v_blkno += p_blocks; 1196 } 1197 1198bail: 1199 return status; 1200} 1201 1202struct ocfs2_la_recovery_item { 1203 struct list_head lri_list; 1204 int lri_slot; 1205 struct ocfs2_dinode *lri_la_dinode; 1206 struct ocfs2_dinode *lri_tl_dinode; 1207 struct ocfs2_quota_recovery *lri_qrec; 1208 enum ocfs2_orphan_reco_type lri_orphan_reco_type; 1209}; 1210 1211/* Does the second half of the recovery process. By this point, the 1212 * node is marked clean and can actually be considered recovered, 1213 * hence it's no longer in the recovery map, but there's still some 1214 * cleanup we can do which shouldn't happen within the recovery thread 1215 * as locking in that context becomes very difficult if we are to take 1216 * recovering nodes into account. 1217 * 1218 * NOTE: This function can and will sleep on recovery of other nodes 1219 * during cluster locking, just like any other ocfs2 process. 1220 */ 1221void ocfs2_complete_recovery(struct work_struct *work) 1222{ 1223 int ret = 0; 1224 struct ocfs2_journal *journal = 1225 container_of(work, struct ocfs2_journal, j_recovery_work); 1226 struct ocfs2_super *osb = journal->j_osb; 1227 struct ocfs2_dinode *la_dinode, *tl_dinode; 1228 struct ocfs2_la_recovery_item *item, *n; 1229 struct ocfs2_quota_recovery *qrec; 1230 enum ocfs2_orphan_reco_type orphan_reco_type; 1231 LIST_HEAD(tmp_la_list); 1232 1233 trace_ocfs2_complete_recovery( 1234 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); 1235 1236 spin_lock(&journal->j_lock); 1237 list_splice_init(&journal->j_la_cleanups, &tmp_la_list); 1238 spin_unlock(&journal->j_lock); 1239 1240 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { 1241 list_del_init(&item->lri_list); 1242 1243 ocfs2_wait_on_quotas(osb); 1244 1245 la_dinode = item->lri_la_dinode; 1246 tl_dinode = item->lri_tl_dinode; 1247 qrec = item->lri_qrec; 1248 orphan_reco_type = item->lri_orphan_reco_type; 1249 1250 trace_ocfs2_complete_recovery_slot(item->lri_slot, 1251 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, 1252 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, 1253 qrec); 1254 1255 if (la_dinode) { 1256 ret = ocfs2_complete_local_alloc_recovery(osb, 1257 la_dinode); 1258 if (ret < 0) 1259 mlog_errno(ret); 1260 1261 kfree(la_dinode); 1262 } 1263 1264 if (tl_dinode) { 1265 ret = ocfs2_complete_truncate_log_recovery(osb, 1266 tl_dinode); 1267 if (ret < 0) 1268 mlog_errno(ret); 1269 1270 kfree(tl_dinode); 1271 } 1272 1273 ret = ocfs2_recover_orphans(osb, item->lri_slot, 1274 orphan_reco_type); 1275 if (ret < 0) 1276 mlog_errno(ret); 1277 1278 if (qrec) { 1279 ret = ocfs2_finish_quota_recovery(osb, qrec, 1280 item->lri_slot); 1281 if (ret < 0) 1282 mlog_errno(ret); 1283 /* Recovery info is already freed now */ 1284 } 1285 1286 kfree(item); 1287 } 1288 1289 trace_ocfs2_complete_recovery_end(ret); 1290} 1291 1292/* NOTE: This function always eats your references to la_dinode and 1293 * tl_dinode, either manually on error, or by passing them to 1294 * ocfs2_complete_recovery */ 1295static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, 1296 int slot_num, 1297 struct ocfs2_dinode *la_dinode, 1298 struct ocfs2_dinode *tl_dinode, 1299 struct ocfs2_quota_recovery *qrec, 1300 enum ocfs2_orphan_reco_type orphan_reco_type) 1301{ 1302 struct ocfs2_la_recovery_item *item; 1303 1304 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); 1305 if (!item) { 1306 /* Though we wish to avoid it, we are in fact safe in 1307 * skipping local alloc cleanup as fsck.ocfs2 is more 1308 * than capable of reclaiming unused space. */ 1309 kfree(la_dinode); 1310 kfree(tl_dinode); 1311 1312 if (qrec) 1313 ocfs2_free_quota_recovery(qrec); 1314 1315 mlog_errno(-ENOMEM); 1316 return; 1317 } 1318 1319 INIT_LIST_HEAD(&item->lri_list); 1320 item->lri_la_dinode = la_dinode; 1321 item->lri_slot = slot_num; 1322 item->lri_tl_dinode = tl_dinode; 1323 item->lri_qrec = qrec; 1324 item->lri_orphan_reco_type = orphan_reco_type; 1325 1326 spin_lock(&journal->j_lock); 1327 list_add_tail(&item->lri_list, &journal->j_la_cleanups); 1328 queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); 1329 spin_unlock(&journal->j_lock); 1330} 1331 1332/* Called by the mount code to queue recovery the last part of 1333 * recovery for it's own and offline slot(s). */ 1334void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) 1335{ 1336 struct ocfs2_journal *journal = osb->journal; 1337 1338 if (ocfs2_is_hard_readonly(osb)) 1339 return; 1340 1341 /* No need to queue up our truncate_log as regular cleanup will catch 1342 * that */ 1343 ocfs2_queue_recovery_completion(journal, osb->slot_num, 1344 osb->local_alloc_copy, NULL, NULL, 1345 ORPHAN_NEED_TRUNCATE); 1346 ocfs2_schedule_truncate_log_flush(osb, 0); 1347 1348 osb->local_alloc_copy = NULL; 1349 1350 /* queue to recover orphan slots for all offline slots */ 1351 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1352 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1353 ocfs2_free_replay_slots(osb); 1354} 1355 1356void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) 1357{ 1358 if (osb->quota_rec) { 1359 ocfs2_queue_recovery_completion(osb->journal, 1360 osb->slot_num, 1361 NULL, 1362 NULL, 1363 osb->quota_rec, 1364 ORPHAN_NEED_TRUNCATE); 1365 osb->quota_rec = NULL; 1366 } 1367} 1368 1369static int __ocfs2_recovery_thread(void *arg) 1370{ 1371 int status, node_num, slot_num; 1372 struct ocfs2_super *osb = arg; 1373 struct ocfs2_recovery_map *rm = osb->recovery_map; 1374 int *rm_quota = NULL; 1375 int rm_quota_used = 0, i; 1376 struct ocfs2_quota_recovery *qrec; 1377 1378 /* Whether the quota supported. */ 1379 int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 1380 OCFS2_FEATURE_RO_COMPAT_USRQUOTA) 1381 || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, 1382 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); 1383 1384 status = ocfs2_wait_on_mount(osb); 1385 if (status < 0) { 1386 goto bail; 1387 } 1388 1389 if (quota_enabled) { 1390 rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); 1391 if (!rm_quota) { 1392 status = -ENOMEM; 1393 goto bail; 1394 } 1395 } 1396restart: 1397 status = ocfs2_super_lock(osb, 1); 1398 if (status < 0) { 1399 mlog_errno(status); 1400 goto bail; 1401 } 1402 1403 status = ocfs2_compute_replay_slots(osb); 1404 if (status < 0) 1405 mlog_errno(status); 1406 1407 /* queue recovery for our own slot */ 1408 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, 1409 NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); 1410 1411 spin_lock(&osb->osb_lock); 1412 while (rm->rm_used) { 1413 /* It's always safe to remove entry zero, as we won't 1414 * clear it until ocfs2_recover_node() has succeeded. */ 1415 node_num = rm->rm_entries[0]; 1416 spin_unlock(&osb->osb_lock); 1417 slot_num = ocfs2_node_num_to_slot(osb, node_num); 1418 trace_ocfs2_recovery_thread_node(node_num, slot_num); 1419 if (slot_num == -ENOENT) { 1420 status = 0; 1421 goto skip_recovery; 1422 } 1423 1424 /* It is a bit subtle with quota recovery. We cannot do it 1425 * immediately because we have to obtain cluster locks from 1426 * quota files and we also don't want to just skip it because 1427 * then quota usage would be out of sync until some node takes 1428 * the slot. So we remember which nodes need quota recovery 1429 * and when everything else is done, we recover quotas. */ 1430 if (quota_enabled) { 1431 for (i = 0; i < rm_quota_used 1432 && rm_quota[i] != slot_num; i++) 1433 ; 1434 1435 if (i == rm_quota_used) 1436 rm_quota[rm_quota_used++] = slot_num; 1437 } 1438 1439 status = ocfs2_recover_node(osb, node_num, slot_num); 1440skip_recovery: 1441 if (!status) { 1442 ocfs2_recovery_map_clear(osb, node_num); 1443 } else { 1444 mlog(ML_ERROR, 1445 "Error %d recovering node %d on device (%u,%u)!\n", 1446 status, node_num, 1447 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); 1448 mlog(ML_ERROR, "Volume requires unmount.\n"); 1449 } 1450 1451 spin_lock(&osb->osb_lock); 1452 } 1453 spin_unlock(&osb->osb_lock); 1454 trace_ocfs2_recovery_thread_end(status); 1455 1456 /* Refresh all journal recovery generations from disk */ 1457 status = ocfs2_check_journals_nolocks(osb); 1458 status = (status == -EROFS) ? 0 : status; 1459 if (status < 0) 1460 mlog_errno(status); 1461 1462 /* Now it is right time to recover quotas... We have to do this under 1463 * superblock lock so that no one can start using the slot (and crash) 1464 * before we recover it */ 1465 if (quota_enabled) { 1466 for (i = 0; i < rm_quota_used; i++) { 1467 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); 1468 if (IS_ERR(qrec)) { 1469 status = PTR_ERR(qrec); 1470 mlog_errno(status); 1471 continue; 1472 } 1473 ocfs2_queue_recovery_completion(osb->journal, 1474 rm_quota[i], 1475 NULL, NULL, qrec, 1476 ORPHAN_NEED_TRUNCATE); 1477 } 1478 } 1479 1480 ocfs2_super_unlock(osb, 1); 1481 1482 /* queue recovery for offline slots */ 1483 ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); 1484 1485bail: 1486 mutex_lock(&osb->recovery_lock); 1487 if (!status && !ocfs2_recovery_completed(osb)) { 1488 mutex_unlock(&osb->recovery_lock); 1489 goto restart; 1490 } 1491 1492 ocfs2_free_replay_slots(osb); 1493 osb->recovery_thread_task = NULL; 1494 mb(); /* sync with ocfs2_recovery_thread_running */ 1495 wake_up(&osb->recovery_event); 1496 1497 mutex_unlock(&osb->recovery_lock); 1498 1499 if (quota_enabled) 1500 kfree(rm_quota); 1501 1502 /* no one is callint kthread_stop() for us so the kthread() api 1503 * requires that we call do_exit(). And it isn't exported, but 1504 * complete_and_exit() seems to be a minimal wrapper around it. */ 1505 complete_and_exit(NULL, status); 1506} 1507 1508void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) 1509{ 1510 mutex_lock(&osb->recovery_lock); 1511 1512 trace_ocfs2_recovery_thread(node_num, osb->node_num, 1513 osb->disable_recovery, osb->recovery_thread_task, 1514 osb->disable_recovery ? 1515 -1 : ocfs2_recovery_map_set(osb, node_num)); 1516 1517 if (osb->disable_recovery) 1518 goto out; 1519 1520 if (osb->recovery_thread_task) 1521 goto out; 1522 1523 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, 1524 "ocfs2rec-%s", osb->uuid_str); 1525 if (IS_ERR(osb->recovery_thread_task)) { 1526 mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); 1527 osb->recovery_thread_task = NULL; 1528 } 1529 1530out: 1531 mutex_unlock(&osb->recovery_lock); 1532 wake_up(&osb->recovery_event); 1533} 1534 1535static int ocfs2_read_journal_inode(struct ocfs2_super *osb, 1536 int slot_num, 1537 struct buffer_head **bh, 1538 struct inode **ret_inode) 1539{ 1540 int status = -EACCES; 1541 struct inode *inode = NULL; 1542 1543 BUG_ON(slot_num >= osb->max_slots); 1544 1545 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1546 slot_num); 1547 if (!inode || is_bad_inode(inode)) { 1548 mlog_errno(status); 1549 goto bail; 1550 } 1551 SET_INODE_JOURNAL(inode); 1552 1553 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); 1554 if (status < 0) { 1555 mlog_errno(status); 1556 goto bail; 1557 } 1558 1559 status = 0; 1560 1561bail: 1562 if (inode) { 1563 if (status || !ret_inode) 1564 iput(inode); 1565 else 1566 *ret_inode = inode; 1567 } 1568 return status; 1569} 1570 1571/* Does the actual journal replay and marks the journal inode as 1572 * clean. Will only replay if the journal inode is marked dirty. */ 1573static int ocfs2_replay_journal(struct ocfs2_super *osb, 1574 int node_num, 1575 int slot_num) 1576{ 1577 int status; 1578 int got_lock = 0; 1579 unsigned int flags; 1580 struct inode *inode = NULL; 1581 struct ocfs2_dinode *fe; 1582 journal_t *journal = NULL; 1583 struct buffer_head *bh = NULL; 1584 u32 slot_reco_gen; 1585 1586 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); 1587 if (status) { 1588 mlog_errno(status); 1589 goto done; 1590 } 1591 1592 fe = (struct ocfs2_dinode *)bh->b_data; 1593 slot_reco_gen = ocfs2_get_recovery_generation(fe); 1594 brelse(bh); 1595 bh = NULL; 1596 1597 /* 1598 * As the fs recovery is asynchronous, there is a small chance that 1599 * another node mounted (and recovered) the slot before the recovery 1600 * thread could get the lock. To handle that, we dirty read the journal 1601 * inode for that slot to get the recovery generation. If it is 1602 * different than what we expected, the slot has been recovered. 1603 * If not, it needs recovery. 1604 */ 1605 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { 1606 trace_ocfs2_replay_journal_recovered(slot_num, 1607 osb->slot_recovery_generations[slot_num], slot_reco_gen); 1608 osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1609 status = -EBUSY; 1610 goto done; 1611 } 1612 1613 /* Continue with recovery as the journal has not yet been recovered */ 1614 1615 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); 1616 if (status < 0) { 1617 trace_ocfs2_replay_journal_lock_err(status); 1618 if (status != -ERESTARTSYS) 1619 mlog(ML_ERROR, "Could not lock journal!\n"); 1620 goto done; 1621 } 1622 got_lock = 1; 1623 1624 fe = (struct ocfs2_dinode *) bh->b_data; 1625 1626 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1627 slot_reco_gen = ocfs2_get_recovery_generation(fe); 1628 1629 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { 1630 trace_ocfs2_replay_journal_skip(node_num); 1631 /* Refresh recovery generation for the slot */ 1632 osb->slot_recovery_generations[slot_num] = slot_reco_gen; 1633 goto done; 1634 } 1635 1636 /* we need to run complete recovery for offline orphan slots */ 1637 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); 1638 1639 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ 1640 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1641 MINOR(osb->sb->s_dev)); 1642 1643 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); 1644 1645 status = ocfs2_force_read_journal(inode); 1646 if (status < 0) { 1647 mlog_errno(status); 1648 goto done; 1649 } 1650 1651 journal = jbd2_journal_init_inode(inode); 1652 if (journal == NULL) { 1653 mlog(ML_ERROR, "Linux journal layer error\n"); 1654 status = -EIO; 1655 goto done; 1656 } 1657 1658 status = jbd2_journal_load(journal); 1659 if (status < 0) { 1660 mlog_errno(status); 1661 if (!igrab(inode)) 1662 BUG(); 1663 jbd2_journal_destroy(journal); 1664 goto done; 1665 } 1666 1667 ocfs2_clear_journal_error(osb->sb, journal, slot_num); 1668 1669 /* wipe the journal */ 1670 jbd2_journal_lock_updates(journal); 1671 status = jbd2_journal_flush(journal); 1672 jbd2_journal_unlock_updates(journal); 1673 if (status < 0) 1674 mlog_errno(status); 1675 1676 /* This will mark the node clean */ 1677 flags = le32_to_cpu(fe->id1.journal1.ij_flags); 1678 flags &= ~OCFS2_JOURNAL_DIRTY_FL; 1679 fe->id1.journal1.ij_flags = cpu_to_le32(flags); 1680 1681 /* Increment recovery generation to indicate successful recovery */ 1682 ocfs2_bump_recovery_generation(fe); 1683 osb->slot_recovery_generations[slot_num] = 1684 ocfs2_get_recovery_generation(fe); 1685 1686 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); 1687 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); 1688 if (status < 0) 1689 mlog_errno(status); 1690 1691 if (!igrab(inode)) 1692 BUG(); 1693 1694 jbd2_journal_destroy(journal); 1695 1696 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ 1697 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), 1698 MINOR(osb->sb->s_dev)); 1699done: 1700 /* drop the lock on this nodes journal */ 1701 if (got_lock) 1702 ocfs2_inode_unlock(inode, 1); 1703 1704 iput(inode); 1705 brelse(bh); 1706 1707 return status; 1708} 1709 1710/* 1711 * Do the most important parts of node recovery: 1712 * - Replay it's journal 1713 * - Stamp a clean local allocator file 1714 * - Stamp a clean truncate log 1715 * - Mark the node clean 1716 * 1717 * If this function completes without error, a node in OCFS2 can be 1718 * said to have been safely recovered. As a result, failure during the 1719 * second part of a nodes recovery process (local alloc recovery) is 1720 * far less concerning. 1721 */ 1722static int ocfs2_recover_node(struct ocfs2_super *osb, 1723 int node_num, int slot_num) 1724{ 1725 int status = 0; 1726 struct ocfs2_dinode *la_copy = NULL; 1727 struct ocfs2_dinode *tl_copy = NULL; 1728 1729 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); 1730 1731 /* Should not ever be called to recover ourselves -- in that 1732 * case we should've called ocfs2_journal_load instead. */ 1733 BUG_ON(osb->node_num == node_num); 1734 1735 status = ocfs2_replay_journal(osb, node_num, slot_num); 1736 if (status < 0) { 1737 if (status == -EBUSY) { 1738 trace_ocfs2_recover_node_skip(slot_num, node_num); 1739 status = 0; 1740 goto done; 1741 } 1742 mlog_errno(status); 1743 goto done; 1744 } 1745 1746 /* Stamp a clean local alloc file AFTER recovering the journal... */ 1747 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); 1748 if (status < 0) { 1749 mlog_errno(status); 1750 goto done; 1751 } 1752 1753 /* An error from begin_truncate_log_recovery is not 1754 * serious enough to warrant halting the rest of 1755 * recovery. */ 1756 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); 1757 if (status < 0) 1758 mlog_errno(status); 1759 1760 /* Likewise, this would be a strange but ultimately not so 1761 * harmful place to get an error... */ 1762 status = ocfs2_clear_slot(osb, slot_num); 1763 if (status < 0) 1764 mlog_errno(status); 1765 1766 /* This will kfree the memory pointed to by la_copy and tl_copy */ 1767 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, 1768 tl_copy, NULL, ORPHAN_NEED_TRUNCATE); 1769 1770 status = 0; 1771done: 1772 1773 return status; 1774} 1775 1776/* Test node liveness by trylocking his journal. If we get the lock, 1777 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is 1778 * still alive (we couldn't get the lock) and < 0 on error. */ 1779static int ocfs2_trylock_journal(struct ocfs2_super *osb, 1780 int slot_num) 1781{ 1782 int status, flags; 1783 struct inode *inode = NULL; 1784 1785 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, 1786 slot_num); 1787 if (inode == NULL) { 1788 mlog(ML_ERROR, "access error\n"); 1789 status = -EACCES; 1790 goto bail; 1791 } 1792 if (is_bad_inode(inode)) { 1793 mlog(ML_ERROR, "access error (bad inode)\n"); 1794 iput(inode); 1795 inode = NULL; 1796 status = -EACCES; 1797 goto bail; 1798 } 1799 SET_INODE_JOURNAL(inode); 1800 1801 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; 1802 status = ocfs2_inode_lock_full(inode, NULL, 1, flags); 1803 if (status < 0) { 1804 if (status != -EAGAIN) 1805 mlog_errno(status); 1806 goto bail; 1807 } 1808 1809 ocfs2_inode_unlock(inode, 1); 1810bail: 1811 iput(inode); 1812 1813 return status; 1814} 1815 1816/* Call this underneath ocfs2_super_lock. It also assumes that the 1817 * slot info struct has been updated from disk. */ 1818int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) 1819{ 1820 unsigned int node_num; 1821 int status, i; 1822 u32 gen; 1823 struct buffer_head *bh = NULL; 1824 struct ocfs2_dinode *di; 1825 1826 /* This is called with the super block cluster lock, so we 1827 * know that the slot map can't change underneath us. */ 1828 1829 for (i = 0; i < osb->max_slots; i++) { 1830 /* Read journal inode to get the recovery generation */ 1831 status = ocfs2_read_journal_inode(osb, i, &bh, NULL); 1832 if (status) { 1833 mlog_errno(status); 1834 goto bail; 1835 } 1836 di = (struct ocfs2_dinode *)bh->b_data; 1837 gen = ocfs2_get_recovery_generation(di); 1838 brelse(bh); 1839 bh = NULL; 1840 1841 spin_lock(&osb->osb_lock); 1842 osb->slot_recovery_generations[i] = gen; 1843 1844 trace_ocfs2_mark_dead_nodes(i, 1845 osb->slot_recovery_generations[i]); 1846 1847 if (i == osb->slot_num) { 1848 spin_unlock(&osb->osb_lock); 1849 continue; 1850 } 1851 1852 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); 1853 if (status == -ENOENT) { 1854 spin_unlock(&osb->osb_lock); 1855 continue; 1856 } 1857 1858 if (__ocfs2_recovery_map_test(osb, node_num)) { 1859 spin_unlock(&osb->osb_lock); 1860 continue; 1861 } 1862 spin_unlock(&osb->osb_lock); 1863 1864 /* Ok, we have a slot occupied by another node which 1865 * is not in the recovery map. We trylock his journal 1866 * file here to test if he's alive. */ 1867 status = ocfs2_trylock_journal(osb, i); 1868 if (!status) { 1869 /* Since we're called from mount, we know that 1870 * the recovery thread can't race us on 1871 * setting / checking the recovery bits. */ 1872 ocfs2_recovery_thread(osb, node_num); 1873 } else if ((status < 0) && (status != -EAGAIN)) { 1874 mlog_errno(status); 1875 goto bail; 1876 } 1877 } 1878 1879 status = 0; 1880bail: 1881 return status; 1882} 1883 1884/* 1885 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some 1886 * randomness to the timeout to minimize multple nodes firing the timer at the 1887 * same time. 1888 */ 1889static inline unsigned long ocfs2_orphan_scan_timeout(void) 1890{ 1891 unsigned long time; 1892 1893 get_random_bytes(&time, sizeof(time)); 1894 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); 1895 return msecs_to_jiffies(time); 1896} 1897 1898/* 1899 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for 1900 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This 1901 * is done to catch any orphans that are left over in orphan directories. 1902 * 1903 * It scans all slots, even ones that are in use. It does so to handle the 1904 * case described below: 1905 * 1906 * Node 1 has an inode it was using. The dentry went away due to memory 1907 * pressure. Node 1 closes the inode, but it's on the free list. The node 1908 * has the open lock. 1909 * Node 2 unlinks the inode. It grabs the dentry lock to notify others, 1910 * but node 1 has no dentry and doesn't get the message. It trylocks the 1911 * open lock, sees that another node has a PR, and does nothing. 1912 * Later node 2 runs its orphan dir. It igets the inode, trylocks the 1913 * open lock, sees the PR still, and does nothing. 1914 * Basically, we have to trigger an orphan iput on node 1. The only way 1915 * for this to happen is if node 1 runs node 2's orphan dir. 1916 * 1917 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT 1918 * seconds. It gets an EX lock on os_lockres and checks sequence number 1919 * stored in LVB. If the sequence number has changed, it means some other 1920 * node has done the scan. This node skips the scan and tracks the 1921 * sequence number. If the sequence number didn't change, it means a scan 1922 * hasn't happened. The node queues a scan and increments the 1923 * sequence number in the LVB. 1924 */ 1925static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) 1926{ 1927 struct ocfs2_orphan_scan *os; 1928 int status, i; 1929 u32 seqno = 0; 1930 1931 os = &osb->osb_orphan_scan; 1932 1933 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1934 goto out; 1935 1936 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, 1937 atomic_read(&os->os_state)); 1938 1939 status = ocfs2_orphan_scan_lock(osb, &seqno); 1940 if (status < 0) { 1941 if (status != -EAGAIN) 1942 mlog_errno(status); 1943 goto out; 1944 } 1945 1946 /* Do no queue the tasks if the volume is being umounted */ 1947 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) 1948 goto unlock; 1949 1950 if (os->os_seqno != seqno) { 1951 os->os_seqno = seqno; 1952 goto unlock; 1953 } 1954 1955 for (i = 0; i < osb->max_slots; i++) 1956 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, 1957 NULL, ORPHAN_NO_NEED_TRUNCATE); 1958 /* 1959 * We queued a recovery on orphan slots, increment the sequence 1960 * number and update LVB so other node will skip the scan for a while 1961 */ 1962 seqno++; 1963 os->os_count++; 1964 os->os_scantime = ktime_get_seconds(); 1965unlock: 1966 ocfs2_orphan_scan_unlock(osb, seqno); 1967out: 1968 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, 1969 atomic_read(&os->os_state)); 1970 return; 1971} 1972 1973/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ 1974static void ocfs2_orphan_scan_work(struct work_struct *work) 1975{ 1976 struct ocfs2_orphan_scan *os; 1977 struct ocfs2_super *osb; 1978 1979 os = container_of(work, struct ocfs2_orphan_scan, 1980 os_orphan_scan_work.work); 1981 osb = os->os_osb; 1982 1983 mutex_lock(&os->os_lock); 1984 ocfs2_queue_orphan_scan(osb); 1985 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) 1986 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 1987 ocfs2_orphan_scan_timeout()); 1988 mutex_unlock(&os->os_lock); 1989} 1990 1991void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) 1992{ 1993 struct ocfs2_orphan_scan *os; 1994 1995 os = &osb->osb_orphan_scan; 1996 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { 1997 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 1998 mutex_lock(&os->os_lock); 1999 cancel_delayed_work(&os->os_orphan_scan_work); 2000 mutex_unlock(&os->os_lock); 2001 } 2002} 2003 2004void ocfs2_orphan_scan_init(struct ocfs2_super *osb) 2005{ 2006 struct ocfs2_orphan_scan *os; 2007 2008 os = &osb->osb_orphan_scan; 2009 os->os_osb = osb; 2010 os->os_count = 0; 2011 os->os_seqno = 0; 2012 mutex_init(&os->os_lock); 2013 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); 2014} 2015 2016void ocfs2_orphan_scan_start(struct ocfs2_super *osb) 2017{ 2018 struct ocfs2_orphan_scan *os; 2019 2020 os = &osb->osb_orphan_scan; 2021 os->os_scantime = ktime_get_seconds(); 2022 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) 2023 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); 2024 else { 2025 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); 2026 queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, 2027 ocfs2_orphan_scan_timeout()); 2028 } 2029} 2030 2031struct ocfs2_orphan_filldir_priv { 2032 struct dir_context ctx; 2033 struct inode *head; 2034 struct ocfs2_super *osb; 2035 enum ocfs2_orphan_reco_type orphan_reco_type; 2036}; 2037 2038static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, 2039 int name_len, loff_t pos, u64 ino, 2040 unsigned type) 2041{ 2042 struct ocfs2_orphan_filldir_priv *p = 2043 container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); 2044 struct inode *iter; 2045 2046 if (name_len == 1 && !strncmp(".", name, 1)) 2047 return 0; 2048 if (name_len == 2 && !strncmp("..", name, 2)) 2049 return 0; 2050 2051 /* do not include dio entry in case of orphan scan */ 2052 if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && 2053 (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 2054 OCFS2_DIO_ORPHAN_PREFIX_LEN))) 2055 return 0; 2056 2057 /* Skip bad inodes so that recovery can continue */ 2058 iter = ocfs2_iget(p->osb, ino, 2059 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); 2060 if (IS_ERR(iter)) 2061 return 0; 2062 2063 if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, 2064 OCFS2_DIO_ORPHAN_PREFIX_LEN)) 2065 OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; 2066 2067 /* Skip inodes which are already added to recover list, since dio may 2068 * happen concurrently with unlink/rename */ 2069 if (OCFS2_I(iter)->ip_next_orphan) { 2070 iput(iter); 2071 return 0; 2072 } 2073 2074 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); 2075 /* No locking is required for the next_orphan queue as there 2076 * is only ever a single process doing orphan recovery. */ 2077 OCFS2_I(iter)->ip_next_orphan = p->head; 2078 p->head = iter; 2079 2080 return 0; 2081} 2082 2083static int ocfs2_queue_orphans(struct ocfs2_super *osb, 2084 int slot, 2085 struct inode **head, 2086 enum ocfs2_orphan_reco_type orphan_reco_type) 2087{ 2088 int status; 2089 struct inode *orphan_dir_inode = NULL; 2090 struct ocfs2_orphan_filldir_priv priv = { 2091 .ctx.actor = ocfs2_orphan_filldir, 2092 .osb = osb, 2093 .head = *head, 2094 .orphan_reco_type = orphan_reco_type 2095 }; 2096 2097 orphan_dir_inode = ocfs2_get_system_file_inode(osb, 2098 ORPHAN_DIR_SYSTEM_INODE, 2099 slot); 2100 if (!orphan_dir_inode) { 2101 status = -ENOENT; 2102 mlog_errno(status); 2103 return status; 2104 } 2105 2106 inode_lock(orphan_dir_inode); 2107 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); 2108 if (status < 0) { 2109 mlog_errno(status); 2110 goto out; 2111 } 2112 2113 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); 2114 if (status) { 2115 mlog_errno(status); 2116 goto out_cluster; 2117 } 2118 2119 *head = priv.head; 2120 2121out_cluster: 2122 ocfs2_inode_unlock(orphan_dir_inode, 0); 2123out: 2124 inode_unlock(orphan_dir_inode); 2125 iput(orphan_dir_inode); 2126 return status; 2127} 2128 2129static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, 2130 int slot) 2131{ 2132 int ret; 2133 2134 spin_lock(&osb->osb_lock); 2135 ret = !osb->osb_orphan_wipes[slot]; 2136 spin_unlock(&osb->osb_lock); 2137 return ret; 2138} 2139 2140static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, 2141 int slot) 2142{ 2143 spin_lock(&osb->osb_lock); 2144 /* Mark ourselves such that new processes in delete_inode() 2145 * know to quit early. */ 2146 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2147 while (osb->osb_orphan_wipes[slot]) { 2148 /* If any processes are already in the middle of an 2149 * orphan wipe on this dir, then we need to wait for 2150 * them. */ 2151 spin_unlock(&osb->osb_lock); 2152 wait_event_interruptible(osb->osb_wipe_event, 2153 ocfs2_orphan_recovery_can_continue(osb, slot)); 2154 spin_lock(&osb->osb_lock); 2155 } 2156 spin_unlock(&osb->osb_lock); 2157} 2158 2159static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, 2160 int slot) 2161{ 2162 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); 2163} 2164 2165/* 2166 * Orphan recovery. Each mounted node has it's own orphan dir which we 2167 * must run during recovery. Our strategy here is to build a list of 2168 * the inodes in the orphan dir and iget/iput them. The VFS does 2169 * (most) of the rest of the work. 2170 * 2171 * Orphan recovery can happen at any time, not just mount so we have a 2172 * couple of extra considerations. 2173 * 2174 * - We grab as many inodes as we can under the orphan dir lock - 2175 * doing iget() outside the orphan dir risks getting a reference on 2176 * an invalid inode. 2177 * - We must be sure not to deadlock with other processes on the 2178 * system wanting to run delete_inode(). This can happen when they go 2179 * to lock the orphan dir and the orphan recovery process attempts to 2180 * iget() inside the orphan dir lock. This can be avoided by 2181 * advertising our state to ocfs2_delete_inode(). 2182 */ 2183static int ocfs2_recover_orphans(struct ocfs2_super *osb, 2184 int slot, 2185 enum ocfs2_orphan_reco_type orphan_reco_type) 2186{ 2187 int ret = 0; 2188 struct inode *inode = NULL; 2189 struct inode *iter; 2190 struct ocfs2_inode_info *oi; 2191 struct buffer_head *di_bh = NULL; 2192 struct ocfs2_dinode *di = NULL; 2193 2194 trace_ocfs2_recover_orphans(slot); 2195 2196 ocfs2_mark_recovering_orphan_dir(osb, slot); 2197 ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); 2198 ocfs2_clear_recovering_orphan_dir(osb, slot); 2199 2200 /* Error here should be noted, but we want to continue with as 2201 * many queued inodes as we've got. */ 2202 if (ret) 2203 mlog_errno(ret); 2204 2205 while (inode) { 2206 oi = OCFS2_I(inode); 2207 trace_ocfs2_recover_orphans_iput( 2208 (unsigned long long)oi->ip_blkno); 2209 2210 iter = oi->ip_next_orphan; 2211 oi->ip_next_orphan = NULL; 2212 2213 if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { 2214 inode_lock(inode); 2215 ret = ocfs2_rw_lock(inode, 1); 2216 if (ret < 0) { 2217 mlog_errno(ret); 2218 goto unlock_mutex; 2219 } 2220 /* 2221 * We need to take and drop the inode lock to 2222 * force read inode from disk. 2223 */ 2224 ret = ocfs2_inode_lock(inode, &di_bh, 1); 2225 if (ret) { 2226 mlog_errno(ret); 2227 goto unlock_rw; 2228 } 2229 2230 di = (struct ocfs2_dinode *)di_bh->b_data; 2231 2232 if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { 2233 ret = ocfs2_truncate_file(inode, di_bh, 2234 i_size_read(inode)); 2235 if (ret < 0) { 2236 if (ret != -ENOSPC) 2237 mlog_errno(ret); 2238 goto unlock_inode; 2239 } 2240 2241 ret = ocfs2_del_inode_from_orphan(osb, inode, 2242 di_bh, 0, 0); 2243 if (ret) 2244 mlog_errno(ret); 2245 } 2246unlock_inode: 2247 ocfs2_inode_unlock(inode, 1); 2248 brelse(di_bh); 2249 di_bh = NULL; 2250unlock_rw: 2251 ocfs2_rw_unlock(inode, 1); 2252unlock_mutex: 2253 inode_unlock(inode); 2254 2255 /* clear dio flag in ocfs2_inode_info */ 2256 oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; 2257 } else { 2258 spin_lock(&oi->ip_lock); 2259 /* Set the proper information to get us going into 2260 * ocfs2_delete_inode. */ 2261 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; 2262 spin_unlock(&oi->ip_lock); 2263 } 2264 2265 iput(inode); 2266 inode = iter; 2267 } 2268 2269 return ret; 2270} 2271 2272static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) 2273{ 2274 /* This check is good because ocfs2 will wait on our recovery 2275 * thread before changing it to something other than MOUNTED 2276 * or DISABLED. */ 2277 wait_event(osb->osb_mount_event, 2278 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || 2279 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || 2280 atomic_read(&osb->vol_state) == VOLUME_DISABLED); 2281 2282 /* If there's an error on mount, then we may never get to the 2283 * MOUNTED flag, but this is set right before 2284 * dismount_volume() so we can trust it. */ 2285 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { 2286 trace_ocfs2_wait_on_mount(VOLUME_DISABLED); 2287 mlog(0, "mount error, exiting!\n"); 2288 return -EBUSY; 2289 } 2290 2291 return 0; 2292} 2293 2294static int ocfs2_commit_thread(void *arg) 2295{ 2296 int status; 2297 struct ocfs2_super *osb = arg; 2298 struct ocfs2_journal *journal = osb->journal; 2299 2300 /* we can trust j_num_trans here because _should_stop() is only set in 2301 * shutdown and nobody other than ourselves should be able to start 2302 * transactions. committing on shutdown might take a few iterations 2303 * as final transactions put deleted inodes on the list */ 2304 while (!(kthread_should_stop() && 2305 atomic_read(&journal->j_num_trans) == 0)) { 2306 2307 wait_event_interruptible(osb->checkpoint_event, 2308 atomic_read(&journal->j_num_trans) 2309 || kthread_should_stop()); 2310 2311 status = ocfs2_commit_cache(osb); 2312 if (status < 0) { 2313 static unsigned long abort_warn_time; 2314 2315 /* Warn about this once per minute */ 2316 if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) 2317 mlog(ML_ERROR, "status = %d, journal is " 2318 "already aborted.\n", status); 2319 /* 2320 * After ocfs2_commit_cache() fails, j_num_trans has a 2321 * non-zero value. Sleep here to avoid a busy-wait 2322 * loop. 2323 */ 2324 msleep_interruptible(1000); 2325 } 2326 2327 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ 2328 mlog(ML_KTHREAD, 2329 "commit_thread: %u transactions pending on " 2330 "shutdown\n", 2331 atomic_read(&journal->j_num_trans)); 2332 } 2333 } 2334 2335 return 0; 2336} 2337 2338/* Reads all the journal inodes without taking any cluster locks. Used 2339 * for hard readonly access to determine whether any journal requires 2340 * recovery. Also used to refresh the recovery generation numbers after 2341 * a journal has been recovered by another node. 2342 */ 2343int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) 2344{ 2345 int ret = 0; 2346 unsigned int slot; 2347 struct buffer_head *di_bh = NULL; 2348 struct ocfs2_dinode *di; 2349 int journal_dirty = 0; 2350 2351 for(slot = 0; slot < osb->max_slots; slot++) { 2352 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); 2353 if (ret) { 2354 mlog_errno(ret); 2355 goto out; 2356 } 2357 2358 di = (struct ocfs2_dinode *) di_bh->b_data; 2359 2360 osb->slot_recovery_generations[slot] = 2361 ocfs2_get_recovery_generation(di); 2362 2363 if (le32_to_cpu(di->id1.journal1.ij_flags) & 2364 OCFS2_JOURNAL_DIRTY_FL) 2365 journal_dirty = 1; 2366 2367 brelse(di_bh); 2368 di_bh = NULL; 2369 } 2370 2371out: 2372 if (journal_dirty) 2373 ret = -EROFS; 2374 return ret; 2375} 2376