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
 162static void 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_maxlen);
 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_inode = inode;
 887	journal->j_bh = bh;
 888
 889	ocfs2_set_journal_params(osb);
 890
 891	journal->j_state = OCFS2_JOURNAL_LOADED;
 892
 893	status = 0;
 894done:
 895	if (status < 0) {
 896		if (inode_lock)
 897			ocfs2_inode_unlock(inode, 1);
 898		brelse(bh);
 899		if (inode) {
 900			OCFS2_I(inode)->ip_open_count--;
 901			iput(inode);
 902		}
 903	}
 904
 905	return status;
 906}
 907
 908static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
 909{
 910	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
 911}
 912
 913static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
 914{
 915	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
 916}
 917
 918static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
 919				      int dirty, int replayed)
 920{
 921	int status;
 922	unsigned int flags;
 923	struct ocfs2_journal *journal = osb->journal;
 924	struct buffer_head *bh = journal->j_bh;
 925	struct ocfs2_dinode *fe;
 926
 927	fe = (struct ocfs2_dinode *)bh->b_data;
 928
 929	/* The journal bh on the osb always comes from ocfs2_journal_init()
 930	 * and was validated there inside ocfs2_inode_lock_full().  It's a
 931	 * code bug if we mess it up. */
 932	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 933
 934	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
 935	if (dirty)
 936		flags |= OCFS2_JOURNAL_DIRTY_FL;
 937	else
 938		flags &= ~OCFS2_JOURNAL_DIRTY_FL;
 939	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
 940
 941	if (replayed)
 942		ocfs2_bump_recovery_generation(fe);
 943
 944	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
 945	status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
 946	if (status < 0)
 947		mlog_errno(status);
 948
 949	return status;
 950}
 951
 952/*
 953 * If the journal has been kmalloc'd it needs to be freed after this
 954 * call.
 955 */
 956void ocfs2_journal_shutdown(struct ocfs2_super *osb)
 957{
 958	struct ocfs2_journal *journal = NULL;
 959	int status = 0;
 960	struct inode *inode = NULL;
 961	int num_running_trans = 0;
 962
 963	BUG_ON(!osb);
 964
 965	journal = osb->journal;
 966	if (!journal)
 967		goto done;
 968
 969	inode = journal->j_inode;
 970
 971	if (journal->j_state != OCFS2_JOURNAL_LOADED)
 972		goto done;
 973
 974	/* need to inc inode use count - jbd2_journal_destroy will iput. */
 975	if (!igrab(inode))
 976		BUG();
 977
 978	num_running_trans = atomic_read(&(osb->journal->j_num_trans));
 979	trace_ocfs2_journal_shutdown(num_running_trans);
 980
 981	/* Do a commit_cache here. It will flush our journal, *and*
 982	 * release any locks that are still held.
 983	 * set the SHUTDOWN flag and release the trans lock.
 984	 * the commit thread will take the trans lock for us below. */
 985	journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
 986
 987	/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
 988	 * drop the trans_lock (which we want to hold until we
 989	 * completely destroy the journal. */
 990	if (osb->commit_task) {
 991		/* Wait for the commit thread */
 992		trace_ocfs2_journal_shutdown_wait(osb->commit_task);
 993		kthread_stop(osb->commit_task);
 994		osb->commit_task = NULL;
 995	}
 996
 997	BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
 998
 999	if (ocfs2_mount_local(osb)) {
 
1000		jbd2_journal_lock_updates(journal->j_journal);
1001		status = jbd2_journal_flush(journal->j_journal);
1002		jbd2_journal_unlock_updates(journal->j_journal);
1003		if (status < 0)
1004			mlog_errno(status);
1005	}
1006
1007	/* Shutdown the kernel journal system */
1008	if (!jbd2_journal_destroy(journal->j_journal) && !status) {
1009		/*
1010		 * Do not toggle if flush was unsuccessful otherwise
1011		 * will leave dirty metadata in a "clean" journal
1012		 */
1013		status = ocfs2_journal_toggle_dirty(osb, 0, 0);
1014		if (status < 0)
1015			mlog_errno(status);
1016	}
1017	journal->j_journal = NULL;
1018
1019	OCFS2_I(inode)->ip_open_count--;
1020
1021	/* unlock our journal */
1022	ocfs2_inode_unlock(inode, 1);
1023
1024	brelse(journal->j_bh);
1025	journal->j_bh = NULL;
1026
1027	journal->j_state = OCFS2_JOURNAL_FREE;
1028
1029//	up_write(&journal->j_trans_barrier);
1030done:
1031	iput(inode);
 
 
1032}
1033
1034static void ocfs2_clear_journal_error(struct super_block *sb,
1035				      journal_t *journal,
1036				      int slot)
1037{
1038	int olderr;
1039
1040	olderr = jbd2_journal_errno(journal);
1041	if (olderr) {
1042		mlog(ML_ERROR, "File system error %d recorded in "
1043		     "journal %u.\n", olderr, slot);
1044		mlog(ML_ERROR, "File system on device %s needs checking.\n",
1045		     sb->s_id);
1046
1047		jbd2_journal_ack_err(journal);
1048		jbd2_journal_clear_err(journal);
1049	}
1050}
1051
1052int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1053{
1054	int status = 0;
1055	struct ocfs2_super *osb;
1056
1057	BUG_ON(!journal);
1058
1059	osb = journal->j_osb;
1060
1061	status = jbd2_journal_load(journal->j_journal);
1062	if (status < 0) {
1063		mlog(ML_ERROR, "Failed to load journal!\n");
1064		goto done;
1065	}
1066
1067	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1068
1069	if (replayed) {
1070		jbd2_journal_lock_updates(journal->j_journal);
1071		status = jbd2_journal_flush(journal->j_journal);
1072		jbd2_journal_unlock_updates(journal->j_journal);
1073		if (status < 0)
1074			mlog_errno(status);
1075	}
1076
1077	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1078	if (status < 0) {
1079		mlog_errno(status);
1080		goto done;
1081	}
1082
1083	/* Launch the commit thread */
1084	if (!local) {
1085		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1086				"ocfs2cmt-%s", osb->uuid_str);
1087		if (IS_ERR(osb->commit_task)) {
1088			status = PTR_ERR(osb->commit_task);
1089			osb->commit_task = NULL;
1090			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1091			     "error=%d", status);
1092			goto done;
1093		}
1094	} else
1095		osb->commit_task = NULL;
1096
1097done:
1098	return status;
1099}
1100
1101
1102/* 'full' flag tells us whether we clear out all blocks or if we just
1103 * mark the journal clean */
1104int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1105{
1106	int status;
1107
1108	BUG_ON(!journal);
1109
1110	status = jbd2_journal_wipe(journal->j_journal, full);
1111	if (status < 0) {
1112		mlog_errno(status);
1113		goto bail;
1114	}
1115
1116	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1117	if (status < 0)
1118		mlog_errno(status);
1119
1120bail:
1121	return status;
1122}
1123
1124static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1125{
1126	int empty;
1127	struct ocfs2_recovery_map *rm = osb->recovery_map;
1128
1129	spin_lock(&osb->osb_lock);
1130	empty = (rm->rm_used == 0);
1131	spin_unlock(&osb->osb_lock);
1132
1133	return empty;
1134}
1135
1136void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1137{
1138	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1139}
1140
1141/*
1142 * JBD Might read a cached version of another nodes journal file. We
1143 * don't want this as this file changes often and we get no
1144 * notification on those changes. The only way to be sure that we've
1145 * got the most up to date version of those blocks then is to force
1146 * read them off disk. Just searching through the buffer cache won't
1147 * work as there may be pages backing this file which are still marked
1148 * up to date. We know things can't change on this file underneath us
1149 * as we have the lock by now :)
1150 */
1151static int ocfs2_force_read_journal(struct inode *inode)
1152{
1153	int status = 0;
1154	int i;
1155	u64 v_blkno, p_blkno, p_blocks, num_blocks;
1156	struct buffer_head *bh = NULL;
1157	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1158
1159	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1160	v_blkno = 0;
1161	while (v_blkno < num_blocks) {
1162		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1163						     &p_blkno, &p_blocks, NULL);
1164		if (status < 0) {
1165			mlog_errno(status);
1166			goto bail;
1167		}
1168
1169		for (i = 0; i < p_blocks; i++, p_blkno++) {
1170			bh = __find_get_block(osb->sb->s_bdev, p_blkno,
1171					osb->sb->s_blocksize);
1172			/* block not cached. */
1173			if (!bh)
1174				continue;
1175
1176			brelse(bh);
1177			bh = NULL;
1178			/* We are reading journal data which should not
1179			 * be put in the uptodate cache.
1180			 */
1181			status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh);
1182			if (status < 0) {
1183				mlog_errno(status);
1184				goto bail;
1185			}
1186
1187			brelse(bh);
1188			bh = NULL;
1189		}
1190
1191		v_blkno += p_blocks;
1192	}
1193
1194bail:
1195	return status;
1196}
1197
1198struct ocfs2_la_recovery_item {
1199	struct list_head	lri_list;
1200	int			lri_slot;
1201	struct ocfs2_dinode	*lri_la_dinode;
1202	struct ocfs2_dinode	*lri_tl_dinode;
1203	struct ocfs2_quota_recovery *lri_qrec;
1204	enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
1205};
1206
1207/* Does the second half of the recovery process. By this point, the
1208 * node is marked clean and can actually be considered recovered,
1209 * hence it's no longer in the recovery map, but there's still some
1210 * cleanup we can do which shouldn't happen within the recovery thread
1211 * as locking in that context becomes very difficult if we are to take
1212 * recovering nodes into account.
1213 *
1214 * NOTE: This function can and will sleep on recovery of other nodes
1215 * during cluster locking, just like any other ocfs2 process.
1216 */
1217void ocfs2_complete_recovery(struct work_struct *work)
1218{
1219	int ret = 0;
1220	struct ocfs2_journal *journal =
1221		container_of(work, struct ocfs2_journal, j_recovery_work);
1222	struct ocfs2_super *osb = journal->j_osb;
1223	struct ocfs2_dinode *la_dinode, *tl_dinode;
1224	struct ocfs2_la_recovery_item *item, *n;
1225	struct ocfs2_quota_recovery *qrec;
1226	enum ocfs2_orphan_reco_type orphan_reco_type;
1227	LIST_HEAD(tmp_la_list);
1228
1229	trace_ocfs2_complete_recovery(
1230		(unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1231
1232	spin_lock(&journal->j_lock);
1233	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1234	spin_unlock(&journal->j_lock);
1235
1236	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1237		list_del_init(&item->lri_list);
1238
1239		ocfs2_wait_on_quotas(osb);
1240
1241		la_dinode = item->lri_la_dinode;
1242		tl_dinode = item->lri_tl_dinode;
1243		qrec = item->lri_qrec;
1244		orphan_reco_type = item->lri_orphan_reco_type;
1245
1246		trace_ocfs2_complete_recovery_slot(item->lri_slot,
1247			la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1248			tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1249			qrec);
1250
1251		if (la_dinode) {
1252			ret = ocfs2_complete_local_alloc_recovery(osb,
1253								  la_dinode);
1254			if (ret < 0)
1255				mlog_errno(ret);
1256
1257			kfree(la_dinode);
1258		}
1259
1260		if (tl_dinode) {
1261			ret = ocfs2_complete_truncate_log_recovery(osb,
1262								   tl_dinode);
1263			if (ret < 0)
1264				mlog_errno(ret);
1265
1266			kfree(tl_dinode);
1267		}
1268
1269		ret = ocfs2_recover_orphans(osb, item->lri_slot,
1270				orphan_reco_type);
1271		if (ret < 0)
1272			mlog_errno(ret);
1273
1274		if (qrec) {
1275			ret = ocfs2_finish_quota_recovery(osb, qrec,
1276							  item->lri_slot);
1277			if (ret < 0)
1278				mlog_errno(ret);
1279			/* Recovery info is already freed now */
1280		}
1281
1282		kfree(item);
1283	}
1284
1285	trace_ocfs2_complete_recovery_end(ret);
1286}
1287
1288/* NOTE: This function always eats your references to la_dinode and
1289 * tl_dinode, either manually on error, or by passing them to
1290 * ocfs2_complete_recovery */
1291static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1292					    int slot_num,
1293					    struct ocfs2_dinode *la_dinode,
1294					    struct ocfs2_dinode *tl_dinode,
1295					    struct ocfs2_quota_recovery *qrec,
1296					    enum ocfs2_orphan_reco_type orphan_reco_type)
1297{
1298	struct ocfs2_la_recovery_item *item;
1299
1300	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1301	if (!item) {
1302		/* Though we wish to avoid it, we are in fact safe in
1303		 * skipping local alloc cleanup as fsck.ocfs2 is more
1304		 * than capable of reclaiming unused space. */
1305		kfree(la_dinode);
1306		kfree(tl_dinode);
1307
1308		if (qrec)
1309			ocfs2_free_quota_recovery(qrec);
1310
1311		mlog_errno(-ENOMEM);
1312		return;
1313	}
1314
1315	INIT_LIST_HEAD(&item->lri_list);
1316	item->lri_la_dinode = la_dinode;
1317	item->lri_slot = slot_num;
1318	item->lri_tl_dinode = tl_dinode;
1319	item->lri_qrec = qrec;
1320	item->lri_orphan_reco_type = orphan_reco_type;
1321
1322	spin_lock(&journal->j_lock);
1323	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1324	queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work);
1325	spin_unlock(&journal->j_lock);
1326}
1327
1328/* Called by the mount code to queue recovery the last part of
1329 * recovery for it's own and offline slot(s). */
1330void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1331{
1332	struct ocfs2_journal *journal = osb->journal;
1333
1334	if (ocfs2_is_hard_readonly(osb))
1335		return;
1336
1337	/* No need to queue up our truncate_log as regular cleanup will catch
1338	 * that */
1339	ocfs2_queue_recovery_completion(journal, osb->slot_num,
1340					osb->local_alloc_copy, NULL, NULL,
1341					ORPHAN_NEED_TRUNCATE);
1342	ocfs2_schedule_truncate_log_flush(osb, 0);
1343
1344	osb->local_alloc_copy = NULL;
1345
1346	/* queue to recover orphan slots for all offline slots */
1347	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1348	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1349	ocfs2_free_replay_slots(osb);
1350}
1351
1352void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1353{
1354	if (osb->quota_rec) {
1355		ocfs2_queue_recovery_completion(osb->journal,
1356						osb->slot_num,
1357						NULL,
1358						NULL,
1359						osb->quota_rec,
1360						ORPHAN_NEED_TRUNCATE);
1361		osb->quota_rec = NULL;
1362	}
1363}
1364
1365static int __ocfs2_recovery_thread(void *arg)
1366{
1367	int status, node_num, slot_num;
1368	struct ocfs2_super *osb = arg;
1369	struct ocfs2_recovery_map *rm = osb->recovery_map;
1370	int *rm_quota = NULL;
1371	int rm_quota_used = 0, i;
1372	struct ocfs2_quota_recovery *qrec;
1373
1374	/* Whether the quota supported. */
1375	int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1376			OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
1377		|| OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1378			OCFS2_FEATURE_RO_COMPAT_GRPQUOTA);
1379
1380	status = ocfs2_wait_on_mount(osb);
1381	if (status < 0) {
1382		goto bail;
1383	}
1384
1385	if (quota_enabled) {
1386		rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS);
1387		if (!rm_quota) {
1388			status = -ENOMEM;
1389			goto bail;
1390		}
1391	}
1392restart:
1393	status = ocfs2_super_lock(osb, 1);
1394	if (status < 0) {
1395		mlog_errno(status);
1396		goto bail;
1397	}
1398
1399	status = ocfs2_compute_replay_slots(osb);
1400	if (status < 0)
1401		mlog_errno(status);
1402
1403	/* queue recovery for our own slot */
1404	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1405					NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1406
1407	spin_lock(&osb->osb_lock);
1408	while (rm->rm_used) {
1409		/* It's always safe to remove entry zero, as we won't
1410		 * clear it until ocfs2_recover_node() has succeeded. */
1411		node_num = rm->rm_entries[0];
1412		spin_unlock(&osb->osb_lock);
1413		slot_num = ocfs2_node_num_to_slot(osb, node_num);
1414		trace_ocfs2_recovery_thread_node(node_num, slot_num);
1415		if (slot_num == -ENOENT) {
1416			status = 0;
1417			goto skip_recovery;
1418		}
1419
1420		/* It is a bit subtle with quota recovery. We cannot do it
1421		 * immediately because we have to obtain cluster locks from
1422		 * quota files and we also don't want to just skip it because
1423		 * then quota usage would be out of sync until some node takes
1424		 * the slot. So we remember which nodes need quota recovery
1425		 * and when everything else is done, we recover quotas. */
1426		if (quota_enabled) {
1427			for (i = 0; i < rm_quota_used
1428					&& rm_quota[i] != slot_num; i++)
1429				;
1430
1431			if (i == rm_quota_used)
1432				rm_quota[rm_quota_used++] = slot_num;
1433		}
1434
1435		status = ocfs2_recover_node(osb, node_num, slot_num);
1436skip_recovery:
1437		if (!status) {
1438			ocfs2_recovery_map_clear(osb, node_num);
1439		} else {
1440			mlog(ML_ERROR,
1441			     "Error %d recovering node %d on device (%u,%u)!\n",
1442			     status, node_num,
1443			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1444			mlog(ML_ERROR, "Volume requires unmount.\n");
1445		}
1446
1447		spin_lock(&osb->osb_lock);
1448	}
1449	spin_unlock(&osb->osb_lock);
1450	trace_ocfs2_recovery_thread_end(status);
1451
1452	/* Refresh all journal recovery generations from disk */
1453	status = ocfs2_check_journals_nolocks(osb);
1454	status = (status == -EROFS) ? 0 : status;
1455	if (status < 0)
1456		mlog_errno(status);
1457
1458	/* Now it is right time to recover quotas... We have to do this under
1459	 * superblock lock so that no one can start using the slot (and crash)
1460	 * before we recover it */
1461	if (quota_enabled) {
1462		for (i = 0; i < rm_quota_used; i++) {
1463			qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1464			if (IS_ERR(qrec)) {
1465				status = PTR_ERR(qrec);
1466				mlog_errno(status);
1467				continue;
1468			}
1469			ocfs2_queue_recovery_completion(osb->journal,
1470					rm_quota[i],
1471					NULL, NULL, qrec,
1472					ORPHAN_NEED_TRUNCATE);
1473		}
1474	}
1475
1476	ocfs2_super_unlock(osb, 1);
1477
1478	/* queue recovery for offline slots */
1479	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1480
1481bail:
1482	mutex_lock(&osb->recovery_lock);
1483	if (!status && !ocfs2_recovery_completed(osb)) {
1484		mutex_unlock(&osb->recovery_lock);
1485		goto restart;
1486	}
1487
1488	ocfs2_free_replay_slots(osb);
1489	osb->recovery_thread_task = NULL;
1490	mb(); /* sync with ocfs2_recovery_thread_running */
1491	wake_up(&osb->recovery_event);
1492
1493	mutex_unlock(&osb->recovery_lock);
1494
1495	if (quota_enabled)
1496		kfree(rm_quota);
1497
1498	/* no one is callint kthread_stop() for us so the kthread() api
1499	 * requires that we call do_exit().  And it isn't exported, but
1500	 * complete_and_exit() seems to be a minimal wrapper around it. */
1501	complete_and_exit(NULL, status);
1502}
1503
1504void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1505{
1506	mutex_lock(&osb->recovery_lock);
1507
1508	trace_ocfs2_recovery_thread(node_num, osb->node_num,
1509		osb->disable_recovery, osb->recovery_thread_task,
1510		osb->disable_recovery ?
1511		-1 : ocfs2_recovery_map_set(osb, node_num));
1512
1513	if (osb->disable_recovery)
1514		goto out;
1515
1516	if (osb->recovery_thread_task)
1517		goto out;
1518
1519	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1520			"ocfs2rec-%s", osb->uuid_str);
1521	if (IS_ERR(osb->recovery_thread_task)) {
1522		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1523		osb->recovery_thread_task = NULL;
1524	}
1525
1526out:
1527	mutex_unlock(&osb->recovery_lock);
1528	wake_up(&osb->recovery_event);
1529}
1530
1531static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1532				    int slot_num,
1533				    struct buffer_head **bh,
1534				    struct inode **ret_inode)
1535{
1536	int status = -EACCES;
1537	struct inode *inode = NULL;
1538
1539	BUG_ON(slot_num >= osb->max_slots);
1540
1541	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1542					    slot_num);
1543	if (!inode || is_bad_inode(inode)) {
1544		mlog_errno(status);
1545		goto bail;
1546	}
1547	SET_INODE_JOURNAL(inode);
1548
1549	status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1550	if (status < 0) {
1551		mlog_errno(status);
1552		goto bail;
1553	}
1554
1555	status = 0;
1556
1557bail:
1558	if (inode) {
1559		if (status || !ret_inode)
1560			iput(inode);
1561		else
1562			*ret_inode = inode;
1563	}
1564	return status;
1565}
1566
1567/* Does the actual journal replay and marks the journal inode as
1568 * clean. Will only replay if the journal inode is marked dirty. */
1569static int ocfs2_replay_journal(struct ocfs2_super *osb,
1570				int node_num,
1571				int slot_num)
1572{
1573	int status;
1574	int got_lock = 0;
1575	unsigned int flags;
1576	struct inode *inode = NULL;
1577	struct ocfs2_dinode *fe;
1578	journal_t *journal = NULL;
1579	struct buffer_head *bh = NULL;
1580	u32 slot_reco_gen;
1581
1582	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1583	if (status) {
1584		mlog_errno(status);
1585		goto done;
1586	}
1587
1588	fe = (struct ocfs2_dinode *)bh->b_data;
1589	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1590	brelse(bh);
1591	bh = NULL;
1592
1593	/*
1594	 * As the fs recovery is asynchronous, there is a small chance that
1595	 * another node mounted (and recovered) the slot before the recovery
1596	 * thread could get the lock. To handle that, we dirty read the journal
1597	 * inode for that slot to get the recovery generation. If it is
1598	 * different than what we expected, the slot has been recovered.
1599	 * If not, it needs recovery.
1600	 */
1601	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1602		trace_ocfs2_replay_journal_recovered(slot_num,
1603		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
1604		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1605		status = -EBUSY;
1606		goto done;
1607	}
1608
1609	/* Continue with recovery as the journal has not yet been recovered */
1610
1611	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1612	if (status < 0) {
1613		trace_ocfs2_replay_journal_lock_err(status);
1614		if (status != -ERESTARTSYS)
1615			mlog(ML_ERROR, "Could not lock journal!\n");
1616		goto done;
1617	}
1618	got_lock = 1;
1619
1620	fe = (struct ocfs2_dinode *) bh->b_data;
1621
1622	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1623	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1624
1625	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1626		trace_ocfs2_replay_journal_skip(node_num);
1627		/* Refresh recovery generation for the slot */
1628		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1629		goto done;
1630	}
1631
1632	/* we need to run complete recovery for offline orphan slots */
1633	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1634
1635	printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1636	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1637	       MINOR(osb->sb->s_dev));
1638
1639	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1640
1641	status = ocfs2_force_read_journal(inode);
1642	if (status < 0) {
1643		mlog_errno(status);
1644		goto done;
1645	}
1646
1647	journal = jbd2_journal_init_inode(inode);
1648	if (journal == NULL) {
1649		mlog(ML_ERROR, "Linux journal layer error\n");
1650		status = -EIO;
1651		goto done;
1652	}
1653
1654	status = jbd2_journal_load(journal);
1655	if (status < 0) {
1656		mlog_errno(status);
1657		if (!igrab(inode))
1658			BUG();
1659		jbd2_journal_destroy(journal);
1660		goto done;
1661	}
1662
1663	ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1664
1665	/* wipe the journal */
1666	jbd2_journal_lock_updates(journal);
1667	status = jbd2_journal_flush(journal);
1668	jbd2_journal_unlock_updates(journal);
1669	if (status < 0)
1670		mlog_errno(status);
1671
1672	/* This will mark the node clean */
1673	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1674	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1675	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1676
1677	/* Increment recovery generation to indicate successful recovery */
1678	ocfs2_bump_recovery_generation(fe);
1679	osb->slot_recovery_generations[slot_num] =
1680					ocfs2_get_recovery_generation(fe);
1681
1682	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1683	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1684	if (status < 0)
1685		mlog_errno(status);
1686
1687	if (!igrab(inode))
1688		BUG();
1689
1690	jbd2_journal_destroy(journal);
1691
1692	printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1693	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1694	       MINOR(osb->sb->s_dev));
1695done:
1696	/* drop the lock on this nodes journal */
1697	if (got_lock)
1698		ocfs2_inode_unlock(inode, 1);
1699
1700	iput(inode);
1701	brelse(bh);
1702
1703	return status;
1704}
1705
1706/*
1707 * Do the most important parts of node recovery:
1708 *  - Replay it's journal
1709 *  - Stamp a clean local allocator file
1710 *  - Stamp a clean truncate log
1711 *  - Mark the node clean
1712 *
1713 * If this function completes without error, a node in OCFS2 can be
1714 * said to have been safely recovered. As a result, failure during the
1715 * second part of a nodes recovery process (local alloc recovery) is
1716 * far less concerning.
1717 */
1718static int ocfs2_recover_node(struct ocfs2_super *osb,
1719			      int node_num, int slot_num)
1720{
1721	int status = 0;
1722	struct ocfs2_dinode *la_copy = NULL;
1723	struct ocfs2_dinode *tl_copy = NULL;
1724
1725	trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1726
1727	/* Should not ever be called to recover ourselves -- in that
1728	 * case we should've called ocfs2_journal_load instead. */
1729	BUG_ON(osb->node_num == node_num);
1730
1731	status = ocfs2_replay_journal(osb, node_num, slot_num);
1732	if (status < 0) {
1733		if (status == -EBUSY) {
1734			trace_ocfs2_recover_node_skip(slot_num, node_num);
1735			status = 0;
1736			goto done;
1737		}
1738		mlog_errno(status);
1739		goto done;
1740	}
1741
1742	/* Stamp a clean local alloc file AFTER recovering the journal... */
1743	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1744	if (status < 0) {
1745		mlog_errno(status);
1746		goto done;
1747	}
1748
1749	/* An error from begin_truncate_log_recovery is not
1750	 * serious enough to warrant halting the rest of
1751	 * recovery. */
1752	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1753	if (status < 0)
1754		mlog_errno(status);
1755
1756	/* Likewise, this would be a strange but ultimately not so
1757	 * harmful place to get an error... */
1758	status = ocfs2_clear_slot(osb, slot_num);
1759	if (status < 0)
1760		mlog_errno(status);
1761
1762	/* This will kfree the memory pointed to by la_copy and tl_copy */
1763	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1764					tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1765
1766	status = 0;
1767done:
1768
1769	return status;
1770}
1771
1772/* Test node liveness by trylocking his journal. If we get the lock,
1773 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1774 * still alive (we couldn't get the lock) and < 0 on error. */
1775static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1776				 int slot_num)
1777{
1778	int status, flags;
1779	struct inode *inode = NULL;
1780
1781	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1782					    slot_num);
1783	if (inode == NULL) {
1784		mlog(ML_ERROR, "access error\n");
1785		status = -EACCES;
1786		goto bail;
1787	}
1788	if (is_bad_inode(inode)) {
1789		mlog(ML_ERROR, "access error (bad inode)\n");
1790		iput(inode);
1791		inode = NULL;
1792		status = -EACCES;
1793		goto bail;
1794	}
1795	SET_INODE_JOURNAL(inode);
1796
1797	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1798	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1799	if (status < 0) {
1800		if (status != -EAGAIN)
1801			mlog_errno(status);
1802		goto bail;
1803	}
1804
1805	ocfs2_inode_unlock(inode, 1);
1806bail:
1807	iput(inode);
1808
1809	return status;
1810}
1811
1812/* Call this underneath ocfs2_super_lock. It also assumes that the
1813 * slot info struct has been updated from disk. */
1814int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1815{
1816	unsigned int node_num;
1817	int status, i;
1818	u32 gen;
1819	struct buffer_head *bh = NULL;
1820	struct ocfs2_dinode *di;
1821
1822	/* This is called with the super block cluster lock, so we
1823	 * know that the slot map can't change underneath us. */
1824
1825	for (i = 0; i < osb->max_slots; i++) {
1826		/* Read journal inode to get the recovery generation */
1827		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1828		if (status) {
1829			mlog_errno(status);
1830			goto bail;
1831		}
1832		di = (struct ocfs2_dinode *)bh->b_data;
1833		gen = ocfs2_get_recovery_generation(di);
1834		brelse(bh);
1835		bh = NULL;
1836
1837		spin_lock(&osb->osb_lock);
1838		osb->slot_recovery_generations[i] = gen;
1839
1840		trace_ocfs2_mark_dead_nodes(i,
1841					    osb->slot_recovery_generations[i]);
1842
1843		if (i == osb->slot_num) {
1844			spin_unlock(&osb->osb_lock);
1845			continue;
1846		}
1847
1848		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1849		if (status == -ENOENT) {
1850			spin_unlock(&osb->osb_lock);
1851			continue;
1852		}
1853
1854		if (__ocfs2_recovery_map_test(osb, node_num)) {
1855			spin_unlock(&osb->osb_lock);
1856			continue;
1857		}
1858		spin_unlock(&osb->osb_lock);
1859
1860		/* Ok, we have a slot occupied by another node which
1861		 * is not in the recovery map. We trylock his journal
1862		 * file here to test if he's alive. */
1863		status = ocfs2_trylock_journal(osb, i);
1864		if (!status) {
1865			/* Since we're called from mount, we know that
1866			 * the recovery thread can't race us on
1867			 * setting / checking the recovery bits. */
1868			ocfs2_recovery_thread(osb, node_num);
1869		} else if ((status < 0) && (status != -EAGAIN)) {
1870			mlog_errno(status);
1871			goto bail;
1872		}
1873	}
1874
1875	status = 0;
1876bail:
1877	return status;
1878}
1879
1880/*
1881 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1882 * randomness to the timeout to minimize multple nodes firing the timer at the
1883 * same time.
1884 */
1885static inline unsigned long ocfs2_orphan_scan_timeout(void)
1886{
1887	unsigned long time;
1888
1889	get_random_bytes(&time, sizeof(time));
1890	time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1891	return msecs_to_jiffies(time);
1892}
1893
1894/*
1895 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1896 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1897 * is done to catch any orphans that are left over in orphan directories.
1898 *
1899 * It scans all slots, even ones that are in use. It does so to handle the
1900 * case described below:
1901 *
1902 *   Node 1 has an inode it was using. The dentry went away due to memory
1903 *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1904 *   has the open lock.
1905 *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1906 *   but node 1 has no dentry and doesn't get the message. It trylocks the
1907 *   open lock, sees that another node has a PR, and does nothing.
1908 *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1909 *   open lock, sees the PR still, and does nothing.
1910 *   Basically, we have to trigger an orphan iput on node 1. The only way
1911 *   for this to happen is if node 1 runs node 2's orphan dir.
1912 *
1913 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1914 * seconds.  It gets an EX lock on os_lockres and checks sequence number
1915 * stored in LVB. If the sequence number has changed, it means some other
1916 * node has done the scan.  This node skips the scan and tracks the
1917 * sequence number.  If the sequence number didn't change, it means a scan
1918 * hasn't happened.  The node queues a scan and increments the
1919 * sequence number in the LVB.
1920 */
1921static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1922{
1923	struct ocfs2_orphan_scan *os;
1924	int status, i;
1925	u32 seqno = 0;
1926
1927	os = &osb->osb_orphan_scan;
1928
1929	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1930		goto out;
1931
1932	trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1933					    atomic_read(&os->os_state));
1934
1935	status = ocfs2_orphan_scan_lock(osb, &seqno);
1936	if (status < 0) {
1937		if (status != -EAGAIN)
1938			mlog_errno(status);
1939		goto out;
1940	}
1941
1942	/* Do no queue the tasks if the volume is being umounted */
1943	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1944		goto unlock;
1945
1946	if (os->os_seqno != seqno) {
1947		os->os_seqno = seqno;
1948		goto unlock;
1949	}
1950
1951	for (i = 0; i < osb->max_slots; i++)
1952		ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1953						NULL, ORPHAN_NO_NEED_TRUNCATE);
1954	/*
1955	 * We queued a recovery on orphan slots, increment the sequence
1956	 * number and update LVB so other node will skip the scan for a while
1957	 */
1958	seqno++;
1959	os->os_count++;
1960	os->os_scantime = ktime_get_seconds();
1961unlock:
1962	ocfs2_orphan_scan_unlock(osb, seqno);
1963out:
1964	trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1965					  atomic_read(&os->os_state));
1966	return;
1967}
1968
1969/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1970static void ocfs2_orphan_scan_work(struct work_struct *work)
1971{
1972	struct ocfs2_orphan_scan *os;
1973	struct ocfs2_super *osb;
1974
1975	os = container_of(work, struct ocfs2_orphan_scan,
1976			  os_orphan_scan_work.work);
1977	osb = os->os_osb;
1978
1979	mutex_lock(&os->os_lock);
1980	ocfs2_queue_orphan_scan(osb);
1981	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1982		queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
1983				      ocfs2_orphan_scan_timeout());
1984	mutex_unlock(&os->os_lock);
1985}
1986
1987void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1988{
1989	struct ocfs2_orphan_scan *os;
1990
1991	os = &osb->osb_orphan_scan;
1992	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1993		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1994		mutex_lock(&os->os_lock);
1995		cancel_delayed_work(&os->os_orphan_scan_work);
1996		mutex_unlock(&os->os_lock);
1997	}
1998}
1999
2000void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
2001{
2002	struct ocfs2_orphan_scan *os;
2003
2004	os = &osb->osb_orphan_scan;
2005	os->os_osb = osb;
2006	os->os_count = 0;
2007	os->os_seqno = 0;
2008	mutex_init(&os->os_lock);
2009	INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
2010}
2011
2012void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2013{
2014	struct ocfs2_orphan_scan *os;
2015
2016	os = &osb->osb_orphan_scan;
2017	os->os_scantime = ktime_get_seconds();
2018	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2019		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2020	else {
2021		atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2022		queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2023				   ocfs2_orphan_scan_timeout());
2024	}
2025}
2026
2027struct ocfs2_orphan_filldir_priv {
2028	struct dir_context	ctx;
2029	struct inode		*head;
2030	struct ocfs2_super	*osb;
2031	enum ocfs2_orphan_reco_type orphan_reco_type;
2032};
2033
2034static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2035				int name_len, loff_t pos, u64 ino,
2036				unsigned type)
2037{
2038	struct ocfs2_orphan_filldir_priv *p =
2039		container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2040	struct inode *iter;
2041
2042	if (name_len == 1 && !strncmp(".", name, 1))
2043		return 0;
2044	if (name_len == 2 && !strncmp("..", name, 2))
2045		return 0;
2046
2047	/* do not include dio entry in case of orphan scan */
2048	if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2049			(!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2050			OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2051		return 0;
2052
2053	/* Skip bad inodes so that recovery can continue */
2054	iter = ocfs2_iget(p->osb, ino,
2055			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2056	if (IS_ERR(iter))
2057		return 0;
2058
2059	if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2060			OCFS2_DIO_ORPHAN_PREFIX_LEN))
2061		OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2062
2063	/* Skip inodes which are already added to recover list, since dio may
2064	 * happen concurrently with unlink/rename */
2065	if (OCFS2_I(iter)->ip_next_orphan) {
2066		iput(iter);
2067		return 0;
2068	}
2069
2070	trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2071	/* No locking is required for the next_orphan queue as there
2072	 * is only ever a single process doing orphan recovery. */
2073	OCFS2_I(iter)->ip_next_orphan = p->head;
2074	p->head = iter;
2075
2076	return 0;
2077}
2078
2079static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2080			       int slot,
2081			       struct inode **head,
2082			       enum ocfs2_orphan_reco_type orphan_reco_type)
2083{
2084	int status;
2085	struct inode *orphan_dir_inode = NULL;
2086	struct ocfs2_orphan_filldir_priv priv = {
2087		.ctx.actor = ocfs2_orphan_filldir,
2088		.osb = osb,
2089		.head = *head,
2090		.orphan_reco_type = orphan_reco_type
2091	};
2092
2093	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2094						       ORPHAN_DIR_SYSTEM_INODE,
2095						       slot);
2096	if  (!orphan_dir_inode) {
2097		status = -ENOENT;
2098		mlog_errno(status);
2099		return status;
2100	}
2101
2102	inode_lock(orphan_dir_inode);
2103	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2104	if (status < 0) {
2105		mlog_errno(status);
2106		goto out;
2107	}
2108
2109	status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2110	if (status) {
2111		mlog_errno(status);
2112		goto out_cluster;
2113	}
2114
2115	*head = priv.head;
2116
2117out_cluster:
2118	ocfs2_inode_unlock(orphan_dir_inode, 0);
2119out:
2120	inode_unlock(orphan_dir_inode);
2121	iput(orphan_dir_inode);
2122	return status;
2123}
2124
2125static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2126					      int slot)
2127{
2128	int ret;
2129
2130	spin_lock(&osb->osb_lock);
2131	ret = !osb->osb_orphan_wipes[slot];
2132	spin_unlock(&osb->osb_lock);
2133	return ret;
2134}
2135
2136static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2137					     int slot)
2138{
2139	spin_lock(&osb->osb_lock);
2140	/* Mark ourselves such that new processes in delete_inode()
2141	 * know to quit early. */
2142	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2143	while (osb->osb_orphan_wipes[slot]) {
2144		/* If any processes are already in the middle of an
2145		 * orphan wipe on this dir, then we need to wait for
2146		 * them. */
2147		spin_unlock(&osb->osb_lock);
2148		wait_event_interruptible(osb->osb_wipe_event,
2149					 ocfs2_orphan_recovery_can_continue(osb, slot));
2150		spin_lock(&osb->osb_lock);
2151	}
2152	spin_unlock(&osb->osb_lock);
2153}
2154
2155static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2156					      int slot)
2157{
2158	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2159}
2160
2161/*
2162 * Orphan recovery. Each mounted node has it's own orphan dir which we
2163 * must run during recovery. Our strategy here is to build a list of
2164 * the inodes in the orphan dir and iget/iput them. The VFS does
2165 * (most) of the rest of the work.
2166 *
2167 * Orphan recovery can happen at any time, not just mount so we have a
2168 * couple of extra considerations.
2169 *
2170 * - We grab as many inodes as we can under the orphan dir lock -
2171 *   doing iget() outside the orphan dir risks getting a reference on
2172 *   an invalid inode.
2173 * - We must be sure not to deadlock with other processes on the
2174 *   system wanting to run delete_inode(). This can happen when they go
2175 *   to lock the orphan dir and the orphan recovery process attempts to
2176 *   iget() inside the orphan dir lock. This can be avoided by
2177 *   advertising our state to ocfs2_delete_inode().
2178 */
2179static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2180				 int slot,
2181				 enum ocfs2_orphan_reco_type orphan_reco_type)
2182{
2183	int ret = 0;
2184	struct inode *inode = NULL;
2185	struct inode *iter;
2186	struct ocfs2_inode_info *oi;
2187	struct buffer_head *di_bh = NULL;
2188	struct ocfs2_dinode *di = NULL;
2189
2190	trace_ocfs2_recover_orphans(slot);
2191
2192	ocfs2_mark_recovering_orphan_dir(osb, slot);
2193	ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2194	ocfs2_clear_recovering_orphan_dir(osb, slot);
2195
2196	/* Error here should be noted, but we want to continue with as
2197	 * many queued inodes as we've got. */
2198	if (ret)
2199		mlog_errno(ret);
2200
2201	while (inode) {
2202		oi = OCFS2_I(inode);
2203		trace_ocfs2_recover_orphans_iput(
2204					(unsigned long long)oi->ip_blkno);
2205
2206		iter = oi->ip_next_orphan;
2207		oi->ip_next_orphan = NULL;
2208
2209		if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2210			inode_lock(inode);
2211			ret = ocfs2_rw_lock(inode, 1);
2212			if (ret < 0) {
2213				mlog_errno(ret);
2214				goto unlock_mutex;
2215			}
2216			/*
2217			 * We need to take and drop the inode lock to
2218			 * force read inode from disk.
2219			 */
2220			ret = ocfs2_inode_lock(inode, &di_bh, 1);
2221			if (ret) {
2222				mlog_errno(ret);
2223				goto unlock_rw;
2224			}
2225
2226			di = (struct ocfs2_dinode *)di_bh->b_data;
2227
2228			if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2229				ret = ocfs2_truncate_file(inode, di_bh,
2230						i_size_read(inode));
2231				if (ret < 0) {
2232					if (ret != -ENOSPC)
2233						mlog_errno(ret);
2234					goto unlock_inode;
2235				}
2236
2237				ret = ocfs2_del_inode_from_orphan(osb, inode,
2238						di_bh, 0, 0);
2239				if (ret)
2240					mlog_errno(ret);
2241			}
2242unlock_inode:
2243			ocfs2_inode_unlock(inode, 1);
2244			brelse(di_bh);
2245			di_bh = NULL;
2246unlock_rw:
2247			ocfs2_rw_unlock(inode, 1);
2248unlock_mutex:
2249			inode_unlock(inode);
2250
2251			/* clear dio flag in ocfs2_inode_info */
2252			oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2253		} else {
2254			spin_lock(&oi->ip_lock);
2255			/* Set the proper information to get us going into
2256			 * ocfs2_delete_inode. */
2257			oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2258			spin_unlock(&oi->ip_lock);
2259		}
2260
2261		iput(inode);
2262		inode = iter;
2263	}
2264
2265	return ret;
2266}
2267
2268static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2269{
2270	/* This check is good because ocfs2 will wait on our recovery
2271	 * thread before changing it to something other than MOUNTED
2272	 * or DISABLED. */
2273	wait_event(osb->osb_mount_event,
2274		  (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2275		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2276		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2277
2278	/* If there's an error on mount, then we may never get to the
2279	 * MOUNTED flag, but this is set right before
2280	 * dismount_volume() so we can trust it. */
2281	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2282		trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2283		mlog(0, "mount error, exiting!\n");
2284		return -EBUSY;
2285	}
2286
2287	return 0;
2288}
2289
2290static int ocfs2_commit_thread(void *arg)
2291{
2292	int status;
2293	struct ocfs2_super *osb = arg;
2294	struct ocfs2_journal *journal = osb->journal;
2295
2296	/* we can trust j_num_trans here because _should_stop() is only set in
2297	 * shutdown and nobody other than ourselves should be able to start
2298	 * transactions.  committing on shutdown might take a few iterations
2299	 * as final transactions put deleted inodes on the list */
2300	while (!(kthread_should_stop() &&
2301		 atomic_read(&journal->j_num_trans) == 0)) {
2302
2303		wait_event_interruptible(osb->checkpoint_event,
2304					 atomic_read(&journal->j_num_trans)
2305					 || kthread_should_stop());
2306
2307		status = ocfs2_commit_cache(osb);
2308		if (status < 0) {
2309			static unsigned long abort_warn_time;
2310
2311			/* Warn about this once per minute */
2312			if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2313				mlog(ML_ERROR, "status = %d, journal is "
2314						"already aborted.\n", status);
2315			/*
2316			 * After ocfs2_commit_cache() fails, j_num_trans has a
2317			 * non-zero value.  Sleep here to avoid a busy-wait
2318			 * loop.
2319			 */
2320			msleep_interruptible(1000);
2321		}
2322
2323		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2324			mlog(ML_KTHREAD,
2325			     "commit_thread: %u transactions pending on "
2326			     "shutdown\n",
2327			     atomic_read(&journal->j_num_trans));
2328		}
2329	}
2330
2331	return 0;
2332}
2333
2334/* Reads all the journal inodes without taking any cluster locks. Used
2335 * for hard readonly access to determine whether any journal requires
2336 * recovery. Also used to refresh the recovery generation numbers after
2337 * a journal has been recovered by another node.
2338 */
2339int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2340{
2341	int ret = 0;
2342	unsigned int slot;
2343	struct buffer_head *di_bh = NULL;
2344	struct ocfs2_dinode *di;
2345	int journal_dirty = 0;
2346
2347	for(slot = 0; slot < osb->max_slots; slot++) {
2348		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2349		if (ret) {
2350			mlog_errno(ret);
2351			goto out;
2352		}
2353
2354		di = (struct ocfs2_dinode *) di_bh->b_data;
2355
2356		osb->slot_recovery_generations[slot] =
2357					ocfs2_get_recovery_generation(di);
2358
2359		if (le32_to_cpu(di->id1.journal1.ij_flags) &
2360		    OCFS2_JOURNAL_DIRTY_FL)
2361			journal_dirty = 1;
2362
2363		brelse(di_bh);
2364		di_bh = NULL;
2365	}
2366
2367out:
2368	if (journal_dirty)
2369		ret = -EROFS;
2370	return ret;
2371}