journal.c 43.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * journal.c
 *
 * Defines functions of journalling api
 *
 * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/fs.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/kthread.h>

#define MLOG_MASK_PREFIX ML_JOURNAL
#include <cluster/masklog.h>

#include "ocfs2.h"

#include "alloc.h"
38
#include "dir.h"
39 40 41 42 43 44 45 46 47 48 49 50
#include "dlmglue.h"
#include "extent_map.h"
#include "heartbeat.h"
#include "inode.h"
#include "journal.h"
#include "localalloc.h"
#include "slot_map.h"
#include "super.h"
#include "sysfile.h"

#include "buffer_head_io.h"

I
Ingo Molnar 已提交
51
DEFINE_SPINLOCK(trans_inc_lock);
52 53 54 55 56 57 58 59

static int ocfs2_force_read_journal(struct inode *inode);
static int ocfs2_recover_node(struct ocfs2_super *osb,
			      int node_num);
static int __ocfs2_recovery_thread(void *arg);
static int ocfs2_commit_cache(struct ocfs2_super *osb);
static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
60
				      int dirty, int replayed);
61 62 63 64 65 66
static int ocfs2_trylock_journal(struct ocfs2_super *osb,
				 int slot_num);
static int ocfs2_recover_orphans(struct ocfs2_super *osb,
				 int slot);
static int ocfs2_commit_thread(void *arg);

67 68 69 70 71 72 73

/*
 * The recovery_list is a simple linked list of node numbers to recover.
 * It is protected by the recovery_lock.
 */

struct ocfs2_recovery_map {
74
	unsigned int rm_used;
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
	unsigned int *rm_entries;
};

int ocfs2_recovery_init(struct ocfs2_super *osb)
{
	struct ocfs2_recovery_map *rm;

	mutex_init(&osb->recovery_lock);
	osb->disable_recovery = 0;
	osb->recovery_thread_task = NULL;
	init_waitqueue_head(&osb->recovery_event);

	rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
		     osb->max_slots * sizeof(unsigned int),
		     GFP_KERNEL);
	if (!rm) {
		mlog_errno(-ENOMEM);
		return -ENOMEM;
	}

	rm->rm_entries = (unsigned int *)((char *)rm +
					  sizeof(struct ocfs2_recovery_map));
	osb->recovery_map = rm;

	return 0;
}

/* we can't grab the goofy sem lock from inside wait_event, so we use
 * memory barriers to make sure that we'll see the null task before
 * being woken up */
static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
{
	mb();
	return osb->recovery_thread_task != NULL;
}

void ocfs2_recovery_exit(struct ocfs2_super *osb)
{
	struct ocfs2_recovery_map *rm;

	/* disable any new recovery threads and wait for any currently
	 * running ones to exit. Do this before setting the vol_state. */
	mutex_lock(&osb->recovery_lock);
	osb->disable_recovery = 1;
	mutex_unlock(&osb->recovery_lock);
	wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));

	/* At this point, we know that no more recovery threads can be
	 * launched, so wait for any recovery completion work to
	 * complete. */
	flush_workqueue(ocfs2_wq);

	/*
	 * Now that recovery is shut down, and the osb is about to be
	 * freed,  the osb_lock is not taken here.
	 */
	rm = osb->recovery_map;
	/* XXX: Should we bug if there are dirty entries? */

	kfree(rm);
}

static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
				     unsigned int node_num)
{
	int i;
	struct ocfs2_recovery_map *rm = osb->recovery_map;

	assert_spin_locked(&osb->osb_lock);

	for (i = 0; i < rm->rm_used; i++) {
		if (rm->rm_entries[i] == node_num)
			return 1;
	}

	return 0;
}

/* Behaves like test-and-set.  Returns the previous value */
static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
				  unsigned int node_num)
{
	struct ocfs2_recovery_map *rm = osb->recovery_map;

	spin_lock(&osb->osb_lock);
	if (__ocfs2_recovery_map_test(osb, node_num)) {
		spin_unlock(&osb->osb_lock);
		return 1;
	}

	/* XXX: Can this be exploited? Not from o2dlm... */
	BUG_ON(rm->rm_used >= osb->max_slots);

	rm->rm_entries[rm->rm_used] = node_num;
	rm->rm_used++;
	spin_unlock(&osb->osb_lock);

	return 0;
}

static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
				     unsigned int node_num)
{
	int i;
	struct ocfs2_recovery_map *rm = osb->recovery_map;

	spin_lock(&osb->osb_lock);

	for (i = 0; i < rm->rm_used; i++) {
		if (rm->rm_entries[i] == node_num)
			break;
	}

	if (i < rm->rm_used) {
		/* XXX: be careful with the pointer math */
		memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
			(rm->rm_used - i - 1) * sizeof(unsigned int));
		rm->rm_used--;
	}

	spin_unlock(&osb->osb_lock);
}

198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235
static int ocfs2_commit_cache(struct ocfs2_super *osb)
{
	int status = 0;
	unsigned int flushed;
	unsigned long old_id;
	struct ocfs2_journal *journal = NULL;

	mlog_entry_void();

	journal = osb->journal;

	/* Flush all pending commits and checkpoint the journal. */
	down_write(&journal->j_trans_barrier);

	if (atomic_read(&journal->j_num_trans) == 0) {
		up_write(&journal->j_trans_barrier);
		mlog(0, "No transactions for me to flush!\n");
		goto finally;
	}

	journal_lock_updates(journal->j_journal);
	status = journal_flush(journal->j_journal);
	journal_unlock_updates(journal->j_journal);
	if (status < 0) {
		up_write(&journal->j_trans_barrier);
		mlog_errno(status);
		goto finally;
	}

	old_id = ocfs2_inc_trans_id(journal);

	flushed = atomic_read(&journal->j_num_trans);
	atomic_set(&journal->j_num_trans, 0);
	up_write(&journal->j_trans_barrier);

	mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
	     journal->j_trans_id, flushed);

M
Mark Fasheh 已提交
236
	ocfs2_wake_downconvert_thread(osb);
237 238 239 240 241 242 243 244 245
	wake_up(&journal->j_checkpointed);
finally:
	mlog_exit(status);
	return status;
}

/* pass it NULL and it will allocate a new handle object for you.  If
 * you pass it a handle however, it may still return error, in which
 * case it has free'd the passed handle for you. */
246
handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
247 248
{
	journal_t *journal = osb->journal->j_journal;
249
	handle_t *handle;
250

251
	BUG_ON(!osb || !osb->journal->j_journal);
252

253 254
	if (ocfs2_is_hard_readonly(osb))
		return ERR_PTR(-EROFS);
255 256 257 258 259 260 261 262 263 264 265 266

	BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
	BUG_ON(max_buffs <= 0);

	/* JBD might support this, but our journalling code doesn't yet. */
	if (journal_current_handle()) {
		mlog(ML_ERROR, "Recursive transaction attempted!\n");
		BUG();
	}

	down_read(&osb->journal->j_trans_barrier);

267 268
	handle = journal_start(journal, max_buffs);
	if (IS_ERR(handle)) {
269 270
		up_read(&osb->journal->j_trans_barrier);

271
		mlog_errno(PTR_ERR(handle));
272 273 274

		if (is_journal_aborted(journal)) {
			ocfs2_abort(osb->sb, "Detected aborted journal");
275
			handle = ERR_PTR(-EROFS);
276
		}
S
Sunil Mushran 已提交
277 278 279 280
	} else {
		if (!ocfs2_mount_local(osb))
			atomic_inc(&(osb->journal->j_num_trans));
	}
281 282 283 284

	return handle;
}

285 286
int ocfs2_commit_trans(struct ocfs2_super *osb,
		       handle_t *handle)
287
{
288
	int ret;
289
	struct ocfs2_journal *journal = osb->journal;
290 291 292

	BUG_ON(!handle);

293 294 295
	ret = journal_stop(handle);
	if (ret < 0)
		mlog_errno(ret);
296 297 298

	up_read(&journal->j_trans_barrier);

299
	return ret;
300 301 302 303 304 305 306
}

/*
 * 'nblocks' is what you want to add to the current
 * transaction. extend_trans will either extend the current handle by
 * nblocks, or commit it and start a new one with nblocks credits.
 *
307 308 309 310 311 312
 * This might call journal_restart() which will commit dirty buffers
 * and then restart the transaction. Before calling
 * ocfs2_extend_trans(), any changed blocks should have been
 * dirtied. After calling it, all blocks which need to be changed must
 * go through another set of journal_access/journal_dirty calls.
 *
313 314 315 316 317 318 319 320
 * WARNING: This will not release any semaphores or disk locks taken
 * during the transaction, so make sure they were taken *before*
 * start_trans or we'll have ordering deadlocks.
 *
 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
 * good because transaction ids haven't yet been recorded on the
 * cluster locks associated with this handle.
 */
321
int ocfs2_extend_trans(handle_t *handle, int nblocks)
322 323 324 325 326 327 328 329 330 331
{
	int status;

	BUG_ON(!handle);
	BUG_ON(!nblocks);

	mlog_entry_void();

	mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);

332
#ifdef CONFIG_OCFS2_DEBUG_FS
333 334
	status = 1;
#else
335
	status = journal_extend(handle, nblocks);
336 337 338 339
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
340
#endif
341 342 343

	if (status > 0) {
		mlog(0, "journal_extend failed, trying journal_restart\n");
344
		status = journal_restart(handle, nblocks);
345 346 347 348
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
349
	}
350 351 352 353 354 355 356 357

	status = 0;
bail:

	mlog_exit(status);
	return status;
}

358
int ocfs2_journal_access(handle_t *handle,
359 360 361 362 363 364 365 366 367 368
			 struct inode *inode,
			 struct buffer_head *bh,
			 int type)
{
	int status;

	BUG_ON(!inode);
	BUG_ON(!handle);
	BUG_ON(!bh);

369
	mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
		   (unsigned long long)bh->b_blocknr, type,
		   (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
		   "OCFS2_JOURNAL_ACCESS_CREATE" :
		   "OCFS2_JOURNAL_ACCESS_WRITE",
		   bh->b_size);

	/* we can safely remove this assertion after testing. */
	if (!buffer_uptodate(bh)) {
		mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
		mlog(ML_ERROR, "b_blocknr=%llu\n",
		     (unsigned long long)bh->b_blocknr);
		BUG();
	}

	/* Set the current transaction information on the inode so
	 * that the locking code knows whether it can drop it's locks
	 * on this inode or not. We're protected from the commit
	 * thread updating the current transaction id until
	 * ocfs2_commit_trans() because ocfs2_start_trans() took
	 * j_trans_barrier for us. */
	ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);

M
Mark Fasheh 已提交
392
	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
393 394 395
	switch (type) {
	case OCFS2_JOURNAL_ACCESS_CREATE:
	case OCFS2_JOURNAL_ACCESS_WRITE:
396
		status = journal_get_write_access(handle, bh);
397 398 399
		break;

	case OCFS2_JOURNAL_ACCESS_UNDO:
400
		status = journal_get_undo_access(handle, bh);
401 402 403 404 405 406
		break;

	default:
		status = -EINVAL;
		mlog(ML_ERROR, "Uknown access type!\n");
	}
M
Mark Fasheh 已提交
407
	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
408 409 410 411 412 413 414 415 416

	if (status < 0)
		mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
		     status, type);

	mlog_exit(status);
	return status;
}

417
int ocfs2_journal_dirty(handle_t *handle,
418 419 420 421 422 423 424
			struct buffer_head *bh)
{
	int status;

	mlog_entry("(bh->b_blocknr=%llu)\n",
		   (unsigned long long)bh->b_blocknr);

425
	status = journal_dirty_metadata(handle, bh);
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446
	if (status < 0)
		mlog(ML_ERROR, "Could not dirty metadata buffer. "
		     "(bh->b_blocknr=%llu)\n",
		     (unsigned long long)bh->b_blocknr);

	mlog_exit(status);
	return status;
}

int ocfs2_journal_dirty_data(handle_t *handle,
			     struct buffer_head *bh)
{
	int err = journal_dirty_data(handle, bh);
	if (err)
		mlog_errno(err);
	/* TODO: When we can handle it, abort the handle and go RO on
	 * error here. */

	return err;
}

447
#define OCFS2_DEFAULT_COMMIT_INTERVAL 	(HZ * JBD_DEFAULT_MAX_COMMIT_AGE)
448 449 450 451

void ocfs2_set_journal_params(struct ocfs2_super *osb)
{
	journal_t *journal = osb->journal->j_journal;
452 453 454 455
	unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;

	if (osb->osb_commit_interval)
		commit_interval = osb->osb_commit_interval;
456 457

	spin_lock(&journal->j_state_lock);
458
	journal->j_commit_interval = commit_interval;
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
	if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
		journal->j_flags |= JFS_BARRIER;
	else
		journal->j_flags &= ~JFS_BARRIER;
	spin_unlock(&journal->j_state_lock);
}

int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
{
	int status = -1;
	struct inode *inode = NULL; /* the journal inode */
	journal_t *j_journal = NULL;
	struct ocfs2_dinode *di = NULL;
	struct buffer_head *bh = NULL;
	struct ocfs2_super *osb;
M
Mark Fasheh 已提交
474
	int inode_lock = 0;
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500

	mlog_entry_void();

	BUG_ON(!journal);

	osb = journal->j_osb;

	/* already have the inode for our journal */
	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
					    osb->slot_num);
	if (inode == NULL) {
		status = -EACCES;
		mlog_errno(status);
		goto done;
	}
	if (is_bad_inode(inode)) {
		mlog(ML_ERROR, "access error (bad inode)\n");
		iput(inode);
		inode = NULL;
		status = -EACCES;
		goto done;
	}

	SET_INODE_JOURNAL(inode);
	OCFS2_I(inode)->ip_open_count++;

501 502 503
	/* Skip recovery waits here - journal inode metadata never
	 * changes in a live cluster so it can be considered an
	 * exception to the rule. */
M
Mark Fasheh 已提交
504
	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
505 506 507 508 509 510
	if (status < 0) {
		if (status != -ERESTARTSYS)
			mlog(ML_ERROR, "Could not get lock on journal!\n");
		goto done;
	}

M
Mark Fasheh 已提交
511
	inode_lock = 1;
512 513 514 515 516 517 518 519 520 521
	di = (struct ocfs2_dinode *)bh->b_data;

	if (inode->i_size <  OCFS2_MIN_JOURNAL_SIZE) {
		mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
		     inode->i_size);
		status = -EINVAL;
		goto done;
	}

	mlog(0, "inode->i_size = %lld\n", inode->i_size);
522 523
	mlog(0, "inode->i_blocks = %llu\n",
			(unsigned long long)inode->i_blocks);
524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550
	mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);

	/* call the kernels journal init function now */
	j_journal = journal_init_inode(inode);
	if (j_journal == NULL) {
		mlog(ML_ERROR, "Linux journal layer error\n");
		status = -EINVAL;
		goto done;
	}

	mlog(0, "Returned from journal_init_inode\n");
	mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);

	*dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
		  OCFS2_JOURNAL_DIRTY_FL);

	journal->j_journal = j_journal;
	journal->j_inode = inode;
	journal->j_bh = bh;

	ocfs2_set_journal_params(osb);

	journal->j_state = OCFS2_JOURNAL_LOADED;

	status = 0;
done:
	if (status < 0) {
M
Mark Fasheh 已提交
551 552
		if (inode_lock)
			ocfs2_inode_unlock(inode, 1);
553 554 555 556 557 558 559 560 561 562 563 564
		if (bh != NULL)
			brelse(bh);
		if (inode) {
			OCFS2_I(inode)->ip_open_count--;
			iput(inode);
		}
	}

	mlog_exit(status);
	return status;
}

565 566 567 568 569 570 571 572 573 574
static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
{
	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
}

static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
{
	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
}

575
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
576
				      int dirty, int replayed)
577 578 579 580 581 582 583 584 585 586 587 588 589 590
{
	int status;
	unsigned int flags;
	struct ocfs2_journal *journal = osb->journal;
	struct buffer_head *bh = journal->j_bh;
	struct ocfs2_dinode *fe;

	mlog_entry_void();

	fe = (struct ocfs2_dinode *)bh->b_data;
	if (!OCFS2_IS_VALID_DINODE(fe)) {
		/* This is called from startup/shutdown which will
		 * handle the errors in a specific manner, so no need
		 * to call ocfs2_error() here. */
591
		mlog(ML_ERROR, "Journal dinode %llu  has invalid "
592 593
		     "signature: %.*s",
		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
594
		     fe->i_signature);
595 596 597 598 599 600 601 602 603 604 605
		status = -EIO;
		goto out;
	}

	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
	if (dirty)
		flags |= OCFS2_JOURNAL_DIRTY_FL;
	else
		flags &= ~OCFS2_JOURNAL_DIRTY_FL;
	fe->id1.journal1.ij_flags = cpu_to_le32(flags);

606 607 608
	if (replayed)
		ocfs2_bump_recovery_generation(fe);

609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630
	status = ocfs2_write_block(osb, bh, journal->j_inode);
	if (status < 0)
		mlog_errno(status);

out:
	mlog_exit(status);
	return status;
}

/*
 * If the journal has been kmalloc'd it needs to be freed after this
 * call.
 */
void ocfs2_journal_shutdown(struct ocfs2_super *osb)
{
	struct ocfs2_journal *journal = NULL;
	int status = 0;
	struct inode *inode = NULL;
	int num_running_trans = 0;

	mlog_entry_void();

631
	BUG_ON(!osb);
632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669

	journal = osb->journal;
	if (!journal)
		goto done;

	inode = journal->j_inode;

	if (journal->j_state != OCFS2_JOURNAL_LOADED)
		goto done;

	/* need to inc inode use count as journal_destroy will iput. */
	if (!igrab(inode))
		BUG();

	num_running_trans = atomic_read(&(osb->journal->j_num_trans));
	if (num_running_trans > 0)
		mlog(0, "Shutting down journal: must wait on %d "
		     "running transactions!\n",
		     num_running_trans);

	/* Do a commit_cache here. It will flush our journal, *and*
	 * release any locks that are still held.
	 * set the SHUTDOWN flag and release the trans lock.
	 * the commit thread will take the trans lock for us below. */
	journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;

	/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
	 * drop the trans_lock (which we want to hold until we
	 * completely destroy the journal. */
	if (osb->commit_task) {
		/* Wait for the commit thread */
		mlog(0, "Waiting for ocfs2commit to exit....\n");
		kthread_stop(osb->commit_task);
		osb->commit_task = NULL;
	}

	BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);

S
Sunil Mushran 已提交
670 671 672 673 674 675 676 677 678 679 680 681 682
	if (ocfs2_mount_local(osb)) {
		journal_lock_updates(journal->j_journal);
		status = journal_flush(journal->j_journal);
		journal_unlock_updates(journal->j_journal);
		if (status < 0)
			mlog_errno(status);
	}

	if (status == 0) {
		/*
		 * Do not toggle if flush was unsuccessful otherwise
		 * will leave dirty metadata in a "clean" journal
		 */
683
		status = ocfs2_journal_toggle_dirty(osb, 0, 0);
S
Sunil Mushran 已提交
684 685 686
		if (status < 0)
			mlog_errno(status);
	}
687 688 689 690 691 692 693

	/* Shutdown the kernel journal system */
	journal_destroy(journal->j_journal);

	OCFS2_I(inode)->ip_open_count--;

	/* unlock our journal */
M
Mark Fasheh 已提交
694
	ocfs2_inode_unlock(inode, 1);
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725

	brelse(journal->j_bh);
	journal->j_bh = NULL;

	journal->j_state = OCFS2_JOURNAL_FREE;

//	up_write(&journal->j_trans_barrier);
done:
	if (inode)
		iput(inode);
	mlog_exit_void();
}

static void ocfs2_clear_journal_error(struct super_block *sb,
				      journal_t *journal,
				      int slot)
{
	int olderr;

	olderr = journal_errno(journal);
	if (olderr) {
		mlog(ML_ERROR, "File system error %d recorded in "
		     "journal %u.\n", olderr, slot);
		mlog(ML_ERROR, "File system on device %s needs checking.\n",
		     sb->s_id);

		journal_ack_err(journal);
		journal_clear_err(journal);
	}
}

726
int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
727 728 729 730 731 732
{
	int status = 0;
	struct ocfs2_super *osb;

	mlog_entry_void();

J
Julia Lawall 已提交
733
	BUG_ON(!journal);
734 735 736 737 738 739 740 741 742 743 744

	osb = journal->j_osb;

	status = journal_load(journal->j_journal);
	if (status < 0) {
		mlog(ML_ERROR, "Failed to load journal!\n");
		goto done;
	}

	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);

745
	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
746 747 748 749 750 751
	if (status < 0) {
		mlog_errno(status);
		goto done;
	}

	/* Launch the commit thread */
S
Sunil Mushran 已提交
752 753 754 755 756 757 758 759 760 761 762
	if (!local) {
		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
					       "ocfs2cmt");
		if (IS_ERR(osb->commit_task)) {
			status = PTR_ERR(osb->commit_task);
			osb->commit_task = NULL;
			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
			     "error=%d", status);
			goto done;
		}
	} else
763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778
		osb->commit_task = NULL;

done:
	mlog_exit(status);
	return status;
}


/* 'full' flag tells us whether we clear out all blocks or if we just
 * mark the journal clean */
int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
{
	int status;

	mlog_entry_void();

779
	BUG_ON(!journal);
780 781 782 783 784 785 786

	status = journal_wipe(journal->j_journal, full);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

787
	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
788 789 790 791 792 793 794 795
	if (status < 0)
		mlog_errno(status);

bail:
	mlog_exit(status);
	return status;
}

796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812
static int ocfs2_recovery_completed(struct ocfs2_super *osb)
{
	int empty;
	struct ocfs2_recovery_map *rm = osb->recovery_map;

	spin_lock(&osb->osb_lock);
	empty = (rm->rm_used == 0);
	spin_unlock(&osb->osb_lock);

	return empty;
}

void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
{
	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
}

813 814 815 816 817 818 819 820 821 822 823 824 825
/*
 * JBD Might read a cached version of another nodes journal file. We
 * don't want this as this file changes often and we get no
 * notification on those changes. The only way to be sure that we've
 * got the most up to date version of those blocks then is to force
 * read them off disk. Just searching through the buffer cache won't
 * work as there may be pages backing this file which are still marked
 * up to date. We know things can't change on this file underneath us
 * as we have the lock by now :)
 */
static int ocfs2_force_read_journal(struct inode *inode)
{
	int status = 0;
826
	int i;
827
	u64 v_blkno, p_blkno, p_blocks, num_blocks;
828
#define CONCURRENT_JOURNAL_FILL 32ULL
829 830 831 832 833 834
	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];

	mlog_entry_void();

	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);

835
	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
836
	v_blkno = 0;
837
	while (v_blkno < num_blocks) {
838
		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
839
						     &p_blkno, &p_blocks, NULL);
840 841 842 843 844 845 846 847
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		if (p_blocks > CONCURRENT_JOURNAL_FILL)
			p_blocks = CONCURRENT_JOURNAL_FILL;

848 849
		/* We are reading journal data which should not
		 * be put in the uptodate cache */
850 851
		status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
					   p_blkno, p_blocks, bhs, 0,
852
					   NULL);
853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		for(i = 0; i < p_blocks; i++) {
			brelse(bhs[i]);
			bhs[i] = NULL;
		}

		v_blkno += p_blocks;
	}

bail:
	for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
		if (bhs[i])
			brelse(bhs[i]);
	mlog_exit(status);
	return status;
}

struct ocfs2_la_recovery_item {
	struct list_head	lri_list;
	int			lri_slot;
	struct ocfs2_dinode	*lri_la_dinode;
	struct ocfs2_dinode	*lri_tl_dinode;
};

/* Does the second half of the recovery process. By this point, the
 * node is marked clean and can actually be considered recovered,
 * hence it's no longer in the recovery map, but there's still some
 * cleanup we can do which shouldn't happen within the recovery thread
 * as locking in that context becomes very difficult if we are to take
 * recovering nodes into account.
 *
 * NOTE: This function can and will sleep on recovery of other nodes
 * during cluster locking, just like any other ocfs2 process.
 */
D
David Howells 已提交
891
void ocfs2_complete_recovery(struct work_struct *work)
892 893
{
	int ret;
D
David Howells 已提交
894 895 896
	struct ocfs2_journal *journal =
		container_of(work, struct ocfs2_journal, j_recovery_work);
	struct ocfs2_super *osb = journal->j_osb;
897
	struct ocfs2_dinode *la_dinode, *tl_dinode;
898
	struct ocfs2_la_recovery_item *item, *n;
899 900 901 902 903 904 905 906 907 908
	LIST_HEAD(tmp_la_list);

	mlog_entry_void();

	mlog(0, "completing recovery from keventd\n");

	spin_lock(&journal->j_lock);
	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
	spin_unlock(&journal->j_lock);

909
	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
910 911 912 913 914 915
		list_del_init(&item->lri_list);

		mlog(0, "Complete recovery for slot %d\n", item->lri_slot);

		la_dinode = item->lri_la_dinode;
		if (la_dinode) {
916
			mlog(0, "Clean up local alloc %llu\n",
917
			     (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
918 919 920 921 922 923 924 925 926 927 928

			ret = ocfs2_complete_local_alloc_recovery(osb,
								  la_dinode);
			if (ret < 0)
				mlog_errno(ret);

			kfree(la_dinode);
		}

		tl_dinode = item->lri_tl_dinode;
		if (tl_dinode) {
929
			mlog(0, "Clean up truncate log %llu\n",
930
			     (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960

			ret = ocfs2_complete_truncate_log_recovery(osb,
								   tl_dinode);
			if (ret < 0)
				mlog_errno(ret);

			kfree(tl_dinode);
		}

		ret = ocfs2_recover_orphans(osb, item->lri_slot);
		if (ret < 0)
			mlog_errno(ret);

		kfree(item);
	}

	mlog(0, "Recovery completion\n");
	mlog_exit_void();
}

/* NOTE: This function always eats your references to la_dinode and
 * tl_dinode, either manually on error, or by passing them to
 * ocfs2_complete_recovery */
static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
					    int slot_num,
					    struct ocfs2_dinode *la_dinode,
					    struct ocfs2_dinode *tl_dinode)
{
	struct ocfs2_la_recovery_item *item;

961
	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
	if (!item) {
		/* Though we wish to avoid it, we are in fact safe in
		 * skipping local alloc cleanup as fsck.ocfs2 is more
		 * than capable of reclaiming unused space. */
		if (la_dinode)
			kfree(la_dinode);

		if (tl_dinode)
			kfree(tl_dinode);

		mlog_errno(-ENOMEM);
		return;
	}

	INIT_LIST_HEAD(&item->lri_list);
	item->lri_la_dinode = la_dinode;
	item->lri_slot = slot_num;
	item->lri_tl_dinode = tl_dinode;

	spin_lock(&journal->j_lock);
	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
	queue_work(ocfs2_wq, &journal->j_recovery_work);
	spin_unlock(&journal->j_lock);
}

/* Called by the mount code to queue recovery the last part of
 * recovery for it's own slot. */
void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
{
	struct ocfs2_journal *journal = osb->journal;

	if (osb->dirty) {
		/* No need to queue up our truncate_log as regular
		 * cleanup will catch that. */
		ocfs2_queue_recovery_completion(journal,
						osb->slot_num,
						osb->local_alloc_copy,
						NULL);
		ocfs2_schedule_truncate_log_flush(osb, 0);

		osb->local_alloc_copy = NULL;
		osb->dirty = 0;
	}
}

static int __ocfs2_recovery_thread(void *arg)
{
	int status, node_num;
	struct ocfs2_super *osb = arg;
1011
	struct ocfs2_recovery_map *rm = osb->recovery_map;
1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026

	mlog_entry_void();

	status = ocfs2_wait_on_mount(osb);
	if (status < 0) {
		goto bail;
	}

restart:
	status = ocfs2_super_lock(osb, 1);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

1027 1028 1029 1030 1031 1032
	spin_lock(&osb->osb_lock);
	while (rm->rm_used) {
		/* It's always safe to remove entry zero, as we won't
		 * clear it until ocfs2_recover_node() has succeeded. */
		node_num = rm->rm_entries[0];
		spin_unlock(&osb->osb_lock);
1033 1034

		status = ocfs2_recover_node(osb, node_num);
1035 1036 1037
		if (!status) {
			ocfs2_recovery_map_clear(osb, node_num);
		} else {
1038 1039 1040 1041 1042 1043 1044
			mlog(ML_ERROR,
			     "Error %d recovering node %d on device (%u,%u)!\n",
			     status, node_num,
			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
			mlog(ML_ERROR, "Volume requires unmount.\n");
		}

1045
		spin_lock(&osb->osb_lock);
1046
	}
1047 1048 1049
	spin_unlock(&osb->osb_lock);
	mlog(0, "All nodes recovered\n");

1050 1051 1052 1053 1054 1055
	/* Refresh all journal recovery generations from disk */
	status = ocfs2_check_journals_nolocks(osb);
	status = (status == -EROFS) ? 0 : status;
	if (status < 0)
		mlog_errno(status);

1056 1057 1058
	ocfs2_super_unlock(osb, 1);

	/* We always run recovery on our own orphan dir - the dead
M
Mark Fasheh 已提交
1059 1060
	 * node(s) may have disallowd a previos inode delete. Re-processing
	 * is therefore required. */
1061 1062 1063 1064
	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
					NULL);

bail:
1065
	mutex_lock(&osb->recovery_lock);
1066
	if (!status && !ocfs2_recovery_completed(osb)) {
1067
		mutex_unlock(&osb->recovery_lock);
1068 1069 1070 1071 1072 1073 1074
		goto restart;
	}

	osb->recovery_thread_task = NULL;
	mb(); /* sync with ocfs2_recovery_thread_running */
	wake_up(&osb->recovery_event);

1075
	mutex_unlock(&osb->recovery_lock);
1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089

	mlog_exit(status);
	/* no one is callint kthread_stop() for us so the kthread() api
	 * requires that we call do_exit().  And it isn't exported, but
	 * complete_and_exit() seems to be a minimal wrapper around it. */
	complete_and_exit(NULL, status);
	return status;
}

void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
{
	mlog_entry("(node_num=%d, osb->node_num = %d)\n",
		   node_num, osb->node_num);

1090
	mutex_lock(&osb->recovery_lock);
1091 1092 1093 1094 1095
	if (osb->disable_recovery)
		goto out;

	/* People waiting on recovery will wait on
	 * the recovery map to empty. */
1096 1097
	if (ocfs2_recovery_map_set(osb, node_num))
		mlog(0, "node %d already in recovery map.\n", node_num);
1098 1099 1100 1101 1102 1103 1104

	mlog(0, "starting recovery thread...\n");

	if (osb->recovery_thread_task)
		goto out;

	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
M
Mark Fasheh 已提交
1105
						 "ocfs2rec");
1106 1107 1108 1109 1110 1111
	if (IS_ERR(osb->recovery_thread_task)) {
		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
		osb->recovery_thread_task = NULL;
	}

out:
1112
	mutex_unlock(&osb->recovery_lock);
1113 1114 1115 1116 1117
	wake_up(&osb->recovery_event);

	mlog_exit_void();
}

1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
				    int slot_num,
				    struct buffer_head **bh,
				    struct inode **ret_inode)
{
	int status = -EACCES;
	struct inode *inode = NULL;

	BUG_ON(slot_num >= osb->max_slots);

	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
					    slot_num);
	if (!inode || is_bad_inode(inode)) {
		mlog_errno(status);
		goto bail;
	}
	SET_INODE_JOURNAL(inode);

	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, bh, 0, inode);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	status = 0;

bail:
	if (inode) {
		if (status || !ret_inode)
			iput(inode);
		else
			*ret_inode = inode;
	}
	return status;
}

1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
/* Does the actual journal replay and marks the journal inode as
 * clean. Will only replay if the journal inode is marked dirty. */
static int ocfs2_replay_journal(struct ocfs2_super *osb,
				int node_num,
				int slot_num)
{
	int status;
	int got_lock = 0;
	unsigned int flags;
	struct inode *inode = NULL;
	struct ocfs2_dinode *fe;
	journal_t *journal = NULL;
	struct buffer_head *bh = NULL;
1167
	u32 slot_reco_gen;
1168

1169 1170
	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
	if (status) {
1171 1172 1173
		mlog_errno(status);
		goto done;
	}
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192

	fe = (struct ocfs2_dinode *)bh->b_data;
	slot_reco_gen = ocfs2_get_recovery_generation(fe);
	brelse(bh);
	bh = NULL;

	/*
	 * As the fs recovery is asynchronous, there is a small chance that
	 * another node mounted (and recovered) the slot before the recovery
	 * thread could get the lock. To handle that, we dirty read the journal
	 * inode for that slot to get the recovery generation. If it is
	 * different than what we expected, the slot has been recovered.
	 * If not, it needs recovery.
	 */
	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
		mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num,
		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
		status = -EBUSY;
1193 1194
		goto done;
	}
1195 1196

	/* Continue with recovery as the journal has not yet been recovered */
1197

M
Mark Fasheh 已提交
1198
	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1199
	if (status < 0) {
M
Mark Fasheh 已提交
1200
		mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
1201 1202 1203 1204 1205 1206 1207 1208 1209
		if (status != -ERESTARTSYS)
			mlog(ML_ERROR, "Could not lock journal!\n");
		goto done;
	}
	got_lock = 1;

	fe = (struct ocfs2_dinode *) bh->b_data;

	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1210
	slot_reco_gen = ocfs2_get_recovery_generation(fe);
1211 1212 1213

	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
		mlog(0, "No recovery required for node %d\n", node_num);
1214 1215
		/* Refresh recovery generation for the slot */
		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
		goto done;
	}

	mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
	     node_num, slot_num,
	     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));

	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);

	status = ocfs2_force_read_journal(inode);
	if (status < 0) {
		mlog_errno(status);
		goto done;
	}

	mlog(0, "calling journal_init_inode\n");
	journal = journal_init_inode(inode);
	if (journal == NULL) {
		mlog(ML_ERROR, "Linux journal layer error\n");
		status = -EIO;
		goto done;
	}

	status = journal_load(journal);
	if (status < 0) {
		mlog_errno(status);
		if (!igrab(inode))
			BUG();
		journal_destroy(journal);
		goto done;
	}

	ocfs2_clear_journal_error(osb->sb, journal, slot_num);

	/* wipe the journal */
	mlog(0, "flushing the journal.\n");
	journal_lock_updates(journal);
	status = journal_flush(journal);
	journal_unlock_updates(journal);
	if (status < 0)
		mlog_errno(status);

	/* This will mark the node clean */
	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
	fe->id1.journal1.ij_flags = cpu_to_le32(flags);

1263 1264 1265 1266 1267
	/* Increment recovery generation to indicate successful recovery */
	ocfs2_bump_recovery_generation(fe);
	osb->slot_recovery_generations[slot_num] =
					ocfs2_get_recovery_generation(fe);

1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279
	status = ocfs2_write_block(osb, bh, inode);
	if (status < 0)
		mlog_errno(status);

	if (!igrab(inode))
		BUG();

	journal_destroy(journal);

done:
	/* drop the lock on this nodes journal */
	if (got_lock)
M
Mark Fasheh 已提交
1280
		ocfs2_inode_unlock(inode, 1);
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318

	if (inode)
		iput(inode);

	if (bh)
		brelse(bh);

	mlog_exit(status);
	return status;
}

/*
 * Do the most important parts of node recovery:
 *  - Replay it's journal
 *  - Stamp a clean local allocator file
 *  - Stamp a clean truncate log
 *  - Mark the node clean
 *
 * If this function completes without error, a node in OCFS2 can be
 * said to have been safely recovered. As a result, failure during the
 * second part of a nodes recovery process (local alloc recovery) is
 * far less concerning.
 */
static int ocfs2_recover_node(struct ocfs2_super *osb,
			      int node_num)
{
	int status = 0;
	int slot_num;
	struct ocfs2_dinode *la_copy = NULL;
	struct ocfs2_dinode *tl_copy = NULL;

	mlog_entry("(node_num=%d, osb->node_num = %d)\n",
		   node_num, osb->node_num);

	mlog(0, "checking node %d\n", node_num);

	/* Should not ever be called to recover ourselves -- in that
	 * case we should've called ocfs2_journal_load instead. */
1319
	BUG_ON(osb->node_num == node_num);
1320

1321 1322
	slot_num = ocfs2_node_num_to_slot(osb, node_num);
	if (slot_num == -ENOENT) {
1323 1324 1325 1326 1327 1328 1329 1330 1331
		status = 0;
		mlog(0, "no slot for this node, so no recovery required.\n");
		goto done;
	}

	mlog(0, "node %d was using slot %d\n", node_num, slot_num);

	status = ocfs2_replay_journal(osb, node_num, slot_num);
	if (status < 0) {
1332 1333 1334 1335 1336 1337 1338
		if (status == -EBUSY) {
			mlog(0, "Skipping recovery for slot %u (node %u) "
			     "as another node has recovered it\n", slot_num,
			     node_num);
			status = 0;
			goto done;
		}
1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358
		mlog_errno(status);
		goto done;
	}

	/* Stamp a clean local alloc file AFTER recovering the journal... */
	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
	if (status < 0) {
		mlog_errno(status);
		goto done;
	}

	/* An error from begin_truncate_log_recovery is not
	 * serious enough to warrant halting the rest of
	 * recovery. */
	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
	if (status < 0)
		mlog_errno(status);

	/* Likewise, this would be a strange but ultimately not so
	 * harmful place to get an error... */
1359
	status = ocfs2_clear_slot(osb, slot_num);
1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399
	if (status < 0)
		mlog_errno(status);

	/* This will kfree the memory pointed to by la_copy and tl_copy */
	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
					tl_copy);

	status = 0;
done:

	mlog_exit(status);
	return status;
}

/* Test node liveness by trylocking his journal. If we get the lock,
 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
 * still alive (we couldn't get the lock) and < 0 on error. */
static int ocfs2_trylock_journal(struct ocfs2_super *osb,
				 int slot_num)
{
	int status, flags;
	struct inode *inode = NULL;

	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
					    slot_num);
	if (inode == NULL) {
		mlog(ML_ERROR, "access error\n");
		status = -EACCES;
		goto bail;
	}
	if (is_bad_inode(inode)) {
		mlog(ML_ERROR, "access error (bad inode)\n");
		iput(inode);
		inode = NULL;
		status = -EACCES;
		goto bail;
	}
	SET_INODE_JOURNAL(inode);

	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
M
Mark Fasheh 已提交
1400
	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1401 1402 1403 1404 1405 1406
	if (status < 0) {
		if (status != -EAGAIN)
			mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
1407
	ocfs2_inode_unlock(inode, 1);
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418
bail:
	if (inode)
		iput(inode);

	return status;
}

/* Call this underneath ocfs2_super_lock. It also assumes that the
 * slot info struct has been updated from disk. */
int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
{
1419 1420
	unsigned int node_num;
	int status, i;
1421 1422
	struct buffer_head *bh = NULL;
	struct ocfs2_dinode *di;
1423 1424 1425 1426

	/* This is called with the super block cluster lock, so we
	 * know that the slot map can't change underneath us. */

1427 1428
	spin_lock(&osb->osb_lock);
	for (i = 0; i < osb->max_slots; i++) {
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
		/* Read journal inode to get the recovery generation */
		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
		if (status) {
			mlog_errno(status);
			goto bail;
		}
		di = (struct ocfs2_dinode *)bh->b_data;
		osb->slot_recovery_generations[i] =
					ocfs2_get_recovery_generation(di);
		brelse(bh);
		bh = NULL;

		mlog(0, "Slot %u recovery generation is %u\n", i,
		     osb->slot_recovery_generations[i]);

1444 1445
		if (i == osb->slot_num)
			continue;
1446 1447 1448

		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
		if (status == -ENOENT)
1449 1450
			continue;

1451
		if (__ocfs2_recovery_map_test(osb, node_num))
1452
			continue;
1453
		spin_unlock(&osb->osb_lock);
1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468

		/* Ok, we have a slot occupied by another node which
		 * is not in the recovery map. We trylock his journal
		 * file here to test if he's alive. */
		status = ocfs2_trylock_journal(osb, i);
		if (!status) {
			/* Since we're called from mount, we know that
			 * the recovery thread can't race us on
			 * setting / checking the recovery bits. */
			ocfs2_recovery_thread(osb, node_num);
		} else if ((status < 0) && (status != -EAGAIN)) {
			mlog_errno(status);
			goto bail;
		}

1469
		spin_lock(&osb->osb_lock);
1470
	}
1471
	spin_unlock(&osb->osb_lock);
1472 1473 1474 1475 1476 1477 1478

	status = 0;
bail:
	mlog_exit(status);
	return status;
}

M
Mark Fasheh 已提交
1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496
struct ocfs2_orphan_filldir_priv {
	struct inode		*head;
	struct ocfs2_super	*osb;
};

static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
				loff_t pos, u64 ino, unsigned type)
{
	struct ocfs2_orphan_filldir_priv *p = priv;
	struct inode *iter;

	if (name_len == 1 && !strncmp(".", name, 1))
		return 0;
	if (name_len == 2 && !strncmp("..", name, 2))
		return 0;

	/* Skip bad inodes so that recovery can continue */
	iter = ocfs2_iget(p->osb, ino,
J
Jan Kara 已提交
1497
			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
M
Mark Fasheh 已提交
1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510
	if (IS_ERR(iter))
		return 0;

	mlog(0, "queue orphan %llu\n",
	     (unsigned long long)OCFS2_I(iter)->ip_blkno);
	/* No locking is required for the next_orphan queue as there
	 * is only ever a single process doing orphan recovery. */
	OCFS2_I(iter)->ip_next_orphan = p->head;
	p->head = iter;

	return 0;
}

1511 1512 1513
static int ocfs2_queue_orphans(struct ocfs2_super *osb,
			       int slot,
			       struct inode **head)
1514
{
1515
	int status;
1516
	struct inode *orphan_dir_inode = NULL;
M
Mark Fasheh 已提交
1517 1518 1519 1520 1521
	struct ocfs2_orphan_filldir_priv priv;
	loff_t pos = 0;

	priv.osb = osb;
	priv.head = *head;
1522 1523 1524 1525 1526 1527 1528

	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
						       ORPHAN_DIR_SYSTEM_INODE,
						       slot);
	if  (!orphan_dir_inode) {
		status = -ENOENT;
		mlog_errno(status);
1529 1530
		return status;
	}	
1531

1532
	mutex_lock(&orphan_dir_inode->i_mutex);
M
Mark Fasheh 已提交
1533
	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
1534 1535 1536 1537 1538
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

M
Mark Fasheh 已提交
1539 1540 1541 1542
	status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
				   ocfs2_orphan_filldir);
	if (status) {
		mlog_errno(status);
1543
		goto out_cluster;
1544 1545
	}

M
Mark Fasheh 已提交
1546 1547
	*head = priv.head;

1548
out_cluster:
M
Mark Fasheh 已提交
1549
	ocfs2_inode_unlock(orphan_dir_inode, 0);
1550 1551
out:
	mutex_unlock(&orphan_dir_inode->i_mutex);
1552
	iput(orphan_dir_inode);
1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627
	return status;
}

static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
					      int slot)
{
	int ret;

	spin_lock(&osb->osb_lock);
	ret = !osb->osb_orphan_wipes[slot];
	spin_unlock(&osb->osb_lock);
	return ret;
}

static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
					     int slot)
{
	spin_lock(&osb->osb_lock);
	/* Mark ourselves such that new processes in delete_inode()
	 * know to quit early. */
	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
	while (osb->osb_orphan_wipes[slot]) {
		/* If any processes are already in the middle of an
		 * orphan wipe on this dir, then we need to wait for
		 * them. */
		spin_unlock(&osb->osb_lock);
		wait_event_interruptible(osb->osb_wipe_event,
					 ocfs2_orphan_recovery_can_continue(osb, slot));
		spin_lock(&osb->osb_lock);
	}
	spin_unlock(&osb->osb_lock);
}

static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
					      int slot)
{
	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
}

/*
 * Orphan recovery. Each mounted node has it's own orphan dir which we
 * must run during recovery. Our strategy here is to build a list of
 * the inodes in the orphan dir and iget/iput them. The VFS does
 * (most) of the rest of the work.
 *
 * Orphan recovery can happen at any time, not just mount so we have a
 * couple of extra considerations.
 *
 * - We grab as many inodes as we can under the orphan dir lock -
 *   doing iget() outside the orphan dir risks getting a reference on
 *   an invalid inode.
 * - We must be sure not to deadlock with other processes on the
 *   system wanting to run delete_inode(). This can happen when they go
 *   to lock the orphan dir and the orphan recovery process attempts to
 *   iget() inside the orphan dir lock. This can be avoided by
 *   advertising our state to ocfs2_delete_inode().
 */
static int ocfs2_recover_orphans(struct ocfs2_super *osb,
				 int slot)
{
	int ret = 0;
	struct inode *inode = NULL;
	struct inode *iter;
	struct ocfs2_inode_info *oi;

	mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);

	ocfs2_mark_recovering_orphan_dir(osb, slot);
	ret = ocfs2_queue_orphans(osb, slot, &inode);
	ocfs2_clear_recovering_orphan_dir(osb, slot);

	/* Error here should be noted, but we want to continue with as
	 * many queued inodes as we've got. */
	if (ret)
		mlog_errno(ret);
1628 1629 1630

	while (inode) {
		oi = OCFS2_I(inode);
1631
		mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
1632 1633 1634 1635

		iter = oi->ip_next_orphan;

		spin_lock(&oi->ip_lock);
M
Mark Fasheh 已提交
1636 1637 1638 1639
		/* The remote delete code may have set these on the
		 * assumption that the other node would wipe them
		 * successfully.  If they are still in the node's
		 * orphan dir, we need to reset that state. */
1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651
		oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);

		/* Set the proper information to get us going into
		 * ocfs2_delete_inode. */
		oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
		spin_unlock(&oi->ip_lock);

		iput(inode);

		inode = iter;
	}

1652
	return ret;
1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687
}

static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
{
	/* This check is good because ocfs2 will wait on our recovery
	 * thread before changing it to something other than MOUNTED
	 * or DISABLED. */
	wait_event(osb->osb_mount_event,
		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);

	/* If there's an error on mount, then we may never get to the
	 * MOUNTED flag, but this is set right before
	 * dismount_volume() so we can trust it. */
	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
		mlog(0, "mount error, exiting!\n");
		return -EBUSY;
	}

	return 0;
}

static int ocfs2_commit_thread(void *arg)
{
	int status;
	struct ocfs2_super *osb = arg;
	struct ocfs2_journal *journal = osb->journal;

	/* we can trust j_num_trans here because _should_stop() is only set in
	 * shutdown and nobody other than ourselves should be able to start
	 * transactions.  committing on shutdown might take a few iterations
	 * as final transactions put deleted inodes on the list */
	while (!(kthread_should_stop() &&
		 atomic_read(&journal->j_num_trans) == 0)) {

1688 1689 1690
		wait_event_interruptible(osb->checkpoint_event,
					 atomic_read(&journal->j_num_trans)
					 || kthread_should_stop());
1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706

		status = ocfs2_commit_cache(osb);
		if (status < 0)
			mlog_errno(status);

		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
			mlog(ML_KTHREAD,
			     "commit_thread: %u transactions pending on "
			     "shutdown\n",
			     atomic_read(&journal->j_num_trans));
		}
	}

	return 0;
}

1707 1708 1709 1710 1711
/* Reads all the journal inodes without taking any cluster locks. Used
 * for hard readonly access to determine whether any journal requires
 * recovery. Also used to refresh the recovery generation numbers after
 * a journal has been recovered by another node.
 */
1712 1713 1714 1715
int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
{
	int ret = 0;
	unsigned int slot;
1716
	struct buffer_head *di_bh = NULL;
1717
	struct ocfs2_dinode *di;
1718
	int journal_dirty = 0;
1719 1720

	for(slot = 0; slot < osb->max_slots; slot++) {
1721 1722
		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
		if (ret) {
1723 1724 1725 1726 1727 1728
			mlog_errno(ret);
			goto out;
		}

		di = (struct ocfs2_dinode *) di_bh->b_data;

1729 1730 1731
		osb->slot_recovery_generations[slot] =
					ocfs2_get_recovery_generation(di);

1732 1733
		if (le32_to_cpu(di->id1.journal1.ij_flags) &
		    OCFS2_JOURNAL_DIRTY_FL)
1734
			journal_dirty = 1;
1735 1736

		brelse(di_bh);
1737
		di_bh = NULL;
1738 1739 1740
	}

out:
1741 1742
	if (journal_dirty)
		ret = -EROFS;
1743 1744
	return ret;
}